From acdd9939ea8f686918ef247da3983b4a7d57a988 Mon Sep 17 00:00:00 2001 From: ArchetypicalDude Date: Tue, 20 Feb 2018 11:22:12 +0100 Subject: [PATCH 1/5] - experimental badger-support. Full-sync/import and staying in sync working, not much else tested --- cmd/geth/chaincmd.go | 5 +- cmd/geth/main.go | 3 +- eth/db_upgrade.go | 12 +- ethdb/badger.go | 491 ++++++ ethdb/database.go | 3 +- node/node.go | 2 +- node/service.go | 2 +- tests/testdata | 1 - .../github.com/AndreasBriese/bbloom/LICENSE | 35 + .../github.com/AndreasBriese/bbloom/README.md | 129 ++ .../github.com/AndreasBriese/bbloom/bbloom.go | 270 ++++ .../AndreasBriese/bbloom/sipHash.go | 225 +++ .../github.com/dgraph-io/badger/CHANGELOG.md | 40 + vendor/github.com/dgraph-io/badger/LICENSE | 176 +++ vendor/github.com/dgraph-io/badger/README.md | 606 ++++++++ .../github.com/dgraph-io/badger/appveyor.yml | 48 + vendor/github.com/dgraph-io/badger/backup.go | 150 ++ .../github.com/dgraph-io/badger/compaction.go | 208 +++ vendor/github.com/dgraph-io/badger/db.go | 1324 +++++++++++++++++ .../github.com/dgraph-io/badger/dir_unix.go | 87 ++ .../dgraph-io/badger/dir_windows.go | 90 ++ vendor/github.com/dgraph-io/badger/doc.go | 28 + vendor/github.com/dgraph-io/badger/errors.go | 97 ++ .../github.com/dgraph-io/badger/iterator.go | 518 +++++++ .../dgraph-io/badger/level_handler.go | 294 ++++ vendor/github.com/dgraph-io/badger/levels.go | 698 +++++++++ .../github.com/dgraph-io/badger/managed_db.go | 86 ++ .../github.com/dgraph-io/badger/manifest.go | 425 ++++++ vendor/github.com/dgraph-io/badger/options.go | 112 ++ .../dgraph-io/badger/options/options.go | 30 + .../dgraph-io/badger/protos/backup.pb.go | 497 +++++++ .../dgraph-io/badger/protos/backup.proto | 28 + .../github.com/dgraph-io/badger/protos/gen.sh | 7 + .../dgraph-io/badger/protos/manifest.pb.go | 534 +++++++ .../dgraph-io/badger/protos/manifest.proto | 35 + .../github.com/dgraph-io/badger/skl/README.md | 113 ++ .../github.com/dgraph-io/badger/skl/arena.go | 131 ++ vendor/github.com/dgraph-io/badger/skl/skl.go | 524 +++++++ vendor/github.com/dgraph-io/badger/structs.go | 132 ++ .../dgraph-io/badger/table/README.md | 51 + .../dgraph-io/badger/table/builder.go | 235 +++ .../dgraph-io/badger/table/iterator.go | 539 +++++++ .../dgraph-io/badger/table/table.go | 359 +++++ vendor/github.com/dgraph-io/badger/test.sh | 5 + .../dgraph-io/badger/transaction.go | 546 +++++++ vendor/github.com/dgraph-io/badger/util.go | 139 ++ vendor/github.com/dgraph-io/badger/value.go | 1070 +++++++++++++ vendor/github.com/dgraph-io/badger/y/error.go | 83 ++ .../dgraph-io/badger/y/file_dsync.go | 25 + .../dgraph-io/badger/y/file_nodsync.go | 25 + .../github.com/dgraph-io/badger/y/iterator.go | 264 ++++ .../github.com/dgraph-io/badger/y/metrics.go | 68 + .../dgraph-io/badger/y/mmap_unix.go | 63 + .../dgraph-io/badger/y/mmap_windows.go | 90 ++ vendor/github.com/dgraph-io/badger/y/y.go | 196 +++ vendor/github.com/dgryski/go-farm/LICENSE | 23 + vendor/github.com/dgryski/go-farm/Makefile | 203 +++ vendor/github.com/dgryski/go-farm/README.md | 41 + vendor/github.com/dgryski/go-farm/VERSION | 1 + vendor/github.com/dgryski/go-farm/basics.go | 30 + .../github.com/dgryski/go-farm/farmhashcc.go | 199 +++ .../github.com/dgryski/go-farm/farmhashmk.go | 102 ++ .../github.com/dgryski/go-farm/farmhashna.go | 156 ++ .../github.com/dgryski/go-farm/farmhashuo.go | 117 ++ vendor/github.com/dgryski/go-farm/platform.go | 18 + .../x/net/internal/timeseries/timeseries.go | 525 +++++++ vendor/golang.org/x/net/trace/events.go | 532 +++++++ vendor/golang.org/x/net/trace/histogram.go | 365 +++++ vendor/golang.org/x/net/trace/trace.go | 1082 ++++++++++++++ vendor/golang.org/x/net/trace/trace_go16.go | 21 + vendor/golang.org/x/net/trace/trace_go17.go | 21 + 71 files changed, 15378 insertions(+), 12 deletions(-) create mode 100644 ethdb/badger.go delete mode 160000 tests/testdata create mode 100644 vendor/github.com/AndreasBriese/bbloom/LICENSE create mode 100644 vendor/github.com/AndreasBriese/bbloom/README.md create mode 100644 vendor/github.com/AndreasBriese/bbloom/bbloom.go create mode 100644 vendor/github.com/AndreasBriese/bbloom/sipHash.go create mode 100644 vendor/github.com/dgraph-io/badger/CHANGELOG.md create mode 100644 vendor/github.com/dgraph-io/badger/LICENSE create mode 100644 vendor/github.com/dgraph-io/badger/README.md create mode 100644 vendor/github.com/dgraph-io/badger/appveyor.yml create mode 100644 vendor/github.com/dgraph-io/badger/backup.go create mode 100644 vendor/github.com/dgraph-io/badger/compaction.go create mode 100644 vendor/github.com/dgraph-io/badger/db.go create mode 100644 vendor/github.com/dgraph-io/badger/dir_unix.go create mode 100644 vendor/github.com/dgraph-io/badger/dir_windows.go create mode 100644 vendor/github.com/dgraph-io/badger/doc.go create mode 100644 vendor/github.com/dgraph-io/badger/errors.go create mode 100644 vendor/github.com/dgraph-io/badger/iterator.go create mode 100644 vendor/github.com/dgraph-io/badger/level_handler.go create mode 100644 vendor/github.com/dgraph-io/badger/levels.go create mode 100644 vendor/github.com/dgraph-io/badger/managed_db.go create mode 100644 vendor/github.com/dgraph-io/badger/manifest.go create mode 100644 vendor/github.com/dgraph-io/badger/options.go create mode 100644 vendor/github.com/dgraph-io/badger/options/options.go create mode 100644 vendor/github.com/dgraph-io/badger/protos/backup.pb.go create mode 100644 vendor/github.com/dgraph-io/badger/protos/backup.proto create mode 100755 vendor/github.com/dgraph-io/badger/protos/gen.sh create mode 100644 vendor/github.com/dgraph-io/badger/protos/manifest.pb.go create mode 100644 vendor/github.com/dgraph-io/badger/protos/manifest.proto create mode 100644 vendor/github.com/dgraph-io/badger/skl/README.md create mode 100644 vendor/github.com/dgraph-io/badger/skl/arena.go create mode 100644 vendor/github.com/dgraph-io/badger/skl/skl.go create mode 100644 vendor/github.com/dgraph-io/badger/structs.go create mode 100644 vendor/github.com/dgraph-io/badger/table/README.md create mode 100644 vendor/github.com/dgraph-io/badger/table/builder.go create mode 100644 vendor/github.com/dgraph-io/badger/table/iterator.go create mode 100644 vendor/github.com/dgraph-io/badger/table/table.go create mode 100755 vendor/github.com/dgraph-io/badger/test.sh create mode 100644 vendor/github.com/dgraph-io/badger/transaction.go create mode 100644 vendor/github.com/dgraph-io/badger/util.go create mode 100644 vendor/github.com/dgraph-io/badger/value.go create mode 100644 vendor/github.com/dgraph-io/badger/y/error.go create mode 100644 vendor/github.com/dgraph-io/badger/y/file_dsync.go create mode 100644 vendor/github.com/dgraph-io/badger/y/file_nodsync.go create mode 100644 vendor/github.com/dgraph-io/badger/y/iterator.go create mode 100644 vendor/github.com/dgraph-io/badger/y/metrics.go create mode 100644 vendor/github.com/dgraph-io/badger/y/mmap_unix.go create mode 100644 vendor/github.com/dgraph-io/badger/y/mmap_windows.go create mode 100644 vendor/github.com/dgraph-io/badger/y/y.go create mode 100644 vendor/github.com/dgryski/go-farm/LICENSE create mode 100644 vendor/github.com/dgryski/go-farm/Makefile create mode 100644 vendor/github.com/dgryski/go-farm/README.md create mode 100644 vendor/github.com/dgryski/go-farm/VERSION create mode 100644 vendor/github.com/dgryski/go-farm/basics.go create mode 100644 vendor/github.com/dgryski/go-farm/farmhashcc.go create mode 100644 vendor/github.com/dgryski/go-farm/farmhashmk.go create mode 100644 vendor/github.com/dgryski/go-farm/farmhashna.go create mode 100644 vendor/github.com/dgryski/go-farm/farmhashuo.go create mode 100644 vendor/github.com/dgryski/go-farm/platform.go create mode 100644 vendor/golang.org/x/net/internal/timeseries/timeseries.go create mode 100644 vendor/golang.org/x/net/trace/events.go create mode 100644 vendor/golang.org/x/net/trace/histogram.go create mode 100644 vendor/golang.org/x/net/trace/trace.go create mode 100644 vendor/golang.org/x/net/trace/trace_go16.go create mode 100644 vendor/golang.org/x/net/trace/trace_go17.go diff --git a/cmd/geth/chaincmd.go b/cmd/geth/chaincmd.go index 85d0c3acaa..e7e10219d5 100644 --- a/cmd/geth/chaincmd.go +++ b/cmd/geth/chaincmd.go @@ -216,7 +216,10 @@ func importChain(ctx *cli.Context) error { } chain.Stop() fmt.Printf("Import done in %v.\n\n", time.Since(start)) - + + fmt.Printf("We are using badger, cancel compaction and stat printing for now") + return nil + // Output pre-compaction stats mostly to see the import trashing db := chainDb.(*ethdb.LDBDatabase) diff --git a/cmd/geth/main.go b/cmd/geth/main.go index a82e5c89cf..5e9207c3cd 100644 --- a/cmd/geth/main.go +++ b/cmd/geth/main.go @@ -186,7 +186,8 @@ func init() { app.Flags = append(app.Flags, whisperFlags...) app.Before = func(ctx *cli.Context) error { - runtime.GOMAXPROCS(runtime.NumCPU()) + // Recommended by badger-team + runtime.GOMAXPROCS(128) if err := debug.Setup(ctx); err != nil { return err } diff --git a/eth/db_upgrade.go b/eth/db_upgrade.go index d41afa17c0..6c81b8ba69 100644 --- a/eth/db_upgrade.go +++ b/eth/db_upgrade.go @@ -33,6 +33,7 @@ var deduplicateData = []byte("dbUpgrade_20170714deduplicateData") // upgradeDeduplicateData checks the chain database version and // starts a background process to make upgrades if necessary. // Returns a stop function that blocks until the process has + // been safely stopped. func upgradeDeduplicateData(db ethdb.Database) func() error { // If the database is already converted or empty, bail out @@ -49,10 +50,10 @@ func upgradeDeduplicateData(db ethdb.Database) func() error { stop := make(chan chan error) go func() { - // Create an iterator to read the entire database and covert old lookup entires - it := db.(*ethdb.LDBDatabase).NewIterator() + // Create an iterator to read the entire database and convert old lookup entires + it := db.(*ethdb.BadgerDatabase).NewIterator() defer func() { - if it != nil { + if it.Released() != true { it.Release() } }() @@ -62,7 +63,7 @@ func upgradeDeduplicateData(db ethdb.Database) func() error { failed error ) for failed == nil && it.Next() { - // Skip any entries that don't look like old transaction meta entires (0x01) + // Skip any entries that don't look like old transaction meta entries (0x01)) key := it.Key() if len(key) != common.HashLength+1 || key[common.HashLength] != 0x01 { continue @@ -100,7 +101,7 @@ func upgradeDeduplicateData(db ethdb.Database) func() error { converted++ if converted%100000 == 0 { it.Release() - it = db.(*ethdb.LDBDatabase).NewIterator() + it = db.(*ethdb.BadgerDatabase).NewIterator() it.Seek(key) log.Info("Deduplicating database entries", "deduped", converted) @@ -121,7 +122,6 @@ func upgradeDeduplicateData(db ethdb.Database) func() error { log.Error("Database deduplication failed", "deduped", converted, "err", failed) } it.Release() - it = nil errc := <-stop errc <- failed diff --git a/ethdb/badger.go b/ethdb/badger.go new file mode 100644 index 0000000000..4289618af4 --- /dev/null +++ b/ethdb/badger.go @@ -0,0 +1,491 @@ +// Copyright 2014 The go-ethereum Authors +// This file is part of the go-ethereum library. +// +// The go-ethereum library is free software: you can redistribute it and/or modify +// it under the terms of the GNU Lesser General Public License as published by +// the Free Software Foundation, either version 3 of the License, or +// (at your option) any later version. +// +// The go-ethereum library is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU Lesser General Public License for more details. +// +// You should have received a copy of the GNU Lesser General Public License +// along with the go-ethereum library. If not, see . + +package ethdb + +import ( + //"strconv" + //"strings" + "sync" + "time" + + "github.com/ethereum/go-ethereum/log" + //"github.com/ethereum/go-ethereum/metrics" + "github.com/dgraph-io/badger" + "github.com/dgraph-io/badger/options" + "github.com/ethereum/go-ethereum/common" + + gometrics "github.com/rcrowley/go-metrics" +) + + + +type BadgerDatabase struct { + fn string // filename for reporting + db *badger.DB + badgerCache *BadgerCache + getTimer gometrics.Timer // Timer for measuring the database get request counts and latencies + putTimer gometrics.Timer // Timer for measuring the database put request counts and latencies + delTimer gometrics.Timer // Timer for measuring the database delete request counts and latencies + missMeter gometrics.Meter // Meter for measuring the missed database get requests + readMeter gometrics.Meter // Meter for measuring the database get request data usage + writeMeter gometrics.Meter // Meter for measuring the database put request data usage + compTimeMeter gometrics.Meter // Meter for measuring the total time spent in database compaction + compReadMeter gometrics.Meter // Meter for measuring the data read during compaction + compWriteMeter gometrics.Meter // Meter for measuring the data written during compaction + + quitLock sync.Mutex // Mutex protecting the quit channel access + quitChan chan chan error // Quit channel to stop the metrics collection before closing the database + + log log.Logger // Contextual logger tracking the database path +} + +// NewLDBDatabase returns a LevelDB wrapped object. +func NewBadgerDatabase(file string) (*BadgerDatabase, error) { + logger := log.New("database", file) + + opts := badger.DefaultOptions + opts.Dir = file + opts.ValueDir = file + opts.SyncWrites = false + opts.ValueLogFileSize = 1 << 30 + opts.TableLoadingMode = options.MemoryMap + db, err := badger.Open(opts) + + // (Re)check for errors and abort if opening of the db failed + if err != nil { + return nil, err + } + ret := &BadgerDatabase{ + fn: file, + db: db, + log: logger, + } + + ret.badgerCache = &BadgerCache{db: ret, c: make(map[string][]byte), size: 0, limit: 100000000} + return ret, nil +} + +// Path returns the path to the database directory. +func (db *BadgerDatabase) Path() string { + return db.fn +} + +// Put puts the given key / value to the queue +func (db *BadgerDatabase) Put(key []byte, value []byte) error { + // Measure the database put latency, if requested + if db.putTimer != nil { + defer db.putTimer.UpdateSince(time.Now()) + } + // Generate the data to write to disk, update the meter and write + //value = rle.Compress(value) + + if db.writeMeter != nil { + db.writeMeter.Mark(int64(len(value))) + } + + db.badgerCache.lock.Lock() + db.badgerCache.c[string(key)] = common.CopyBytes(value) + db.badgerCache.size += len(value)+len(key) + db.badgerCache.lock.Unlock() + + if db.badgerCache.size >= db.badgerCache.limit { + return db.badgerCache.Flush() + } + + return nil +} + +func (db *BadgerDatabase) Has(key []byte) (ret bool, err error) { + db.badgerCache.lock.RLock() + defer db.badgerCache.lock.RUnlock() + if db.badgerCache.c[string(key)] != nil { + return true, nil + } + + err = db.db.View(func(txn *badger.Txn) error { + item, err := txn.Get(key) + if item != nil { + ret = true + } + if err == badger.ErrKeyNotFound { + ret = false + err = nil + } + return err + }) + return ret, err +} + +type BadgerCache struct { + db *BadgerDatabase + c map[string][]byte + size int + limit int + lock sync.RWMutex +} + +func (badgerCache *BadgerCache) Flush() (err error) { + badgerCache.lock.Lock() + defer badgerCache.lock.Unlock() + + txn := badgerCache.db.db.NewTransaction(true) + + for key, value := range badgerCache.c { + err = txn.Set([]byte(key), value) + if err == badger.ErrTxnTooBig { + txn.Commit(nil) + txn = badgerCache.db.db.NewTransaction(true) + err = txn.Set([]byte(key), value) + } + } + err = txn.Commit(nil) + log.Info("Badger flushed to disk", "badgerCache size", badgerCache.size) + badgerCache.size = 0 + badgerCache.c = make(map[string][]byte) + return err +} + +// Get returns the given key if it's present. +func (db *BadgerDatabase) Get(key []byte) (dat []byte, err error) { + // Measure the database get latency, if requested + if db.getTimer != nil { + defer db.getTimer.UpdateSince(time.Now()) + } + + + db.badgerCache.lock.RLock() + dat = db.badgerCache.c[string(key)] + db.badgerCache.lock.RUnlock() + if dat == nil { + err = db.db.View(func(txn *badger.Txn) error { + item, err := txn.Get(key) + if err != nil { + return err + } + val, err := item.Value() + if err != nil { + return err + } + dat = common.CopyBytes(val) + return nil + }) + } + if err != nil { + if db.missMeter != nil { + db.missMeter.Mark(1) + } + return nil, err + } + //Update the actually retrieved amount of data + if db.readMeter != nil { + db.readMeter.Mark(int64(len(dat))) + } + return dat, nil + //return rle.Decompress(dat) +} + +// Delete deletes the key from the queue and database +func (db *BadgerDatabase) Delete(key []byte) error { + // Measure the database delete latency, if requested + if db.delTimer != nil { + defer db.delTimer.UpdateSince(time.Now()) + } + // Execute the actual operation + db.badgerCache.lock.Lock() + delete(db.badgerCache.c, string(key)) + + //TODO: also subtract len(value) + db.badgerCache.size-=len(key) + db.badgerCache.lock.Unlock() + return db.db.Update(func(txn *badger.Txn) error { + err := txn.Delete(key) + if err == badger.ErrKeyNotFound { + err = nil + } + return err + }) +} + +type badgerIterator struct { + txn *badger.Txn + internIterator *badger.Iterator + released bool + initialised bool +} + +func (it *badgerIterator) Release() { + it.internIterator.Close() + it.txn.Discard() + it.released = true +} + +func (it *badgerIterator) Released() bool { + return it.released +} + +func (it *badgerIterator) Next() bool { + if(!it.initialised) { + it.internIterator.Rewind() + it.initialised = true + } else { + it.internIterator.Next() + } + return it.internIterator.Valid() +} + +func (it *badgerIterator) Seek(key []byte) { + it.internIterator.Seek(key) +} + +func (it *badgerIterator) Key() []byte { + return it.internIterator.Item().Key() +} + +func (it *badgerIterator) Value() []byte { + value, err := it.internIterator.Item().Value() + if err != nil { + return nil + } + return value +} + +func (db *BadgerDatabase) NewIterator() badgerIterator { + txn := db.db.NewTransaction(false) + opts := badger.DefaultIteratorOptions + internIterator := txn.NewIterator(opts) + return badgerIterator{txn: txn, internIterator: internIterator, released: false, initialised: false} +} + +func (db *BadgerDatabase) Close() { + // Stop the metrics collection to avoid internal database races + db.quitLock.Lock() + defer db.quitLock.Unlock() + + if db.quitChan != nil { + errc := make(chan error) + db.quitChan <- errc + if err := <-errc; err != nil { + db.log.Error("Metrics collection failed", "err", err) + } + } + db.badgerCache.Flush() + err := db.db.Close() + if err == nil { + db.log.Info("Database closed") + } else { + db.log.Error("Failed to close database", "err", err) + } +} + +/* +// Meter configures the database metrics collectors and +func (db *BadgerDatabase) Meter(prefix string) { + // Short circuit metering if the metrics system is disabled + if !metrics.Enabled { + return + } + // Initialize all the metrics collector at the requested prefix + db.getTimer = metrics.NewTimer(prefix + "user/gets") + db.putTimer = metrics.NewTimer(prefix + "user/puts") + db.delTimer = metrics.NewTimer(prefix + "user/dels") + db.missMeter = metrics.NewMeter(prefix + "user/misses") + db.readMeter = metrics.NewMeter(prefix + "user/reads") + db.writeMeter = metrics.NewMeter(prefix + "user/writes") + db.compTimeMeter = metrics.NewMeter(prefix + "compact/time") + db.compReadMeter = metrics.NewMeter(prefix + "compact/input") + db.compWriteMeter = metrics.NewMeter(prefix + "compact/output") + + // Create a quit channel for the periodic collector and run it + db.quitLock.Lock() + db.quitChan = make(chan chan error) + db.quitLock.Unlock() + + go db.meter(3 * time.Second) +} +*/ + +/* +func (db *BadgerDatabase) meter(refresh time.Duration) { + // Create the counters to store current and previous values + counters := make([][]float64, 2) + for i := 0; i < 2; i++ { + counters[i] = make([]float64, 3) + } + // Iterate ad infinitum and collect the stats + for i := 1; ; i++ { + // Retrieve the database stats + stats, err := db.db.GetProperty("leveldb.stats") + if err != nil { + db.log.Error("Failed to read database stats", "err", err) + return + } + // Find the compaction table, skip the header + lines := strings.Split(stats, "\n") + for len(lines) > 0 && strings.TrimSpace(lines[0]) != "Compactions" { + lines = lines[1:] + } + if len(lines) <= 3 { + db.log.Error("Compaction table not found") + return + } + lines = lines[3:] + + // Iterate over all the table rows, and accumulate the entries + for j := 0; j < len(counters[i%2]); j++ { + counters[i%2][j] = 0 + } + for _, line := range lines { + parts := strings.Split(line, "|") + if len(parts) != 6 { + break + } + for idx, counter := range parts[3:] { + value, err := strconv.ParseFloat(strings.TrimSpace(counter), 64) + if err != nil { + db.log.Error("Compaction entry parsing failed", "err", err) + return + } + counters[i%2][idx] += value + } + } + // Update all the requested meters + if db.compTimeMeter != nil { + db.compTimeMeter.Mark(int64((counters[i%2][0] - counters[(i-1)%2][0]) * 1000 * 1000 * 1000)) + } + if db.compReadMeter != nil { + db.compReadMeter.Mark(int64((counters[i%2][1] - counters[(i-1)%2][1]) * 1024 * 1024)) + } + if db.compWriteMeter != nil { + db.compWriteMeter.Mark(int64((counters[i%2][2] - counters[(i-1)%2][2]) * 1024 * 1024)) + } + // Sleep a bit, then repeat the stats collection + select { + case errc := <-db.quitChan: + // Quit requesting, stop hammering the database + errc <- nil + return + + case <-time.After(refresh): + // Timeout, gather a new set of stats + } + } +} +*/ +func (db *BadgerDatabase) NewBatch() Batch { + return &badgerBatch{db: db} +} + +type badgerBatch struct { + db *BadgerDatabase + size int +} + +func (b *badgerBatch) Put(key, value []byte) error { + b.db.badgerCache.lock.Lock() + b.db.badgerCache.c[string(key)] = common.CopyBytes(value) + b.db.badgerCache.size += len(value)+len(key) + b.db.badgerCache.lock.Unlock() + if b.db.badgerCache.size >= b.db.badgerCache.limit { + b.db.badgerCache.Flush() + } + b.size += len(value) + return nil +} + +func (b *badgerBatch) Write() error { + b.size = 0 + if b.db.badgerCache.size >= b.db.badgerCache.limit { + return b.db.badgerCache.Flush() + } + return nil +} + +func (b *badgerBatch) Discard() { + b.size = 0 +} + +func (b *badgerBatch) ValueSize() int { + return b.size +} + +func (b *badgerBatch) Reset() { + b.size = 0 +} + +type table struct { + db Database + prefix string +} + +// NewTable returns a Database object that prefixes all keys with a given +// string. +func NewTable(db Database, prefix string) Database { + return &table{ + db: db, + prefix: prefix, + } +} + +func (dt *table) Put(key []byte, value []byte) error { + return dt.db.Put(append([]byte(dt.prefix), key...), value) +} + +func (dt *table) Has(key []byte) (bool, error) { + return dt.db.Has(append([]byte(dt.prefix), key...)) +} + +func (dt *table) Get(key []byte) ([]byte, error) { + return dt.db.Get(append([]byte(dt.prefix), key...)) +} + +func (dt *table) Delete(key []byte) error { + return dt.db.Delete(append([]byte(dt.prefix), key...)) +} + +func (dt *table) Close() { + // Do nothing; don't close the underlying DB. +} + +type tableBatch struct { + batch Batch + prefix string +} + +// NewTableBatch returns a Batch object which prefixes all keys with a given string. +func NewTableBatch(db Database, prefix string) Batch { + return &tableBatch{db.NewBatch(), prefix} +} + +func (dt *table) NewBatch() Batch { + return &tableBatch{dt.db.NewBatch(), dt.prefix} +} + +func (tb *tableBatch) Put(key, value []byte) error { + return tb.batch.Put(append([]byte(tb.prefix), key...), value) +} + +func (tb *tableBatch) Write() error { + return tb.batch.Write() +} + +func (tb *tableBatch) ValueSize() int { + return tb.batch.ValueSize() +} + +func (tb *tableBatch) Reset() { + tb.batch.Reset() +} \ No newline at end of file diff --git a/ethdb/database.go b/ethdb/database.go index d86585f079..49e425ab44 100644 --- a/ethdb/database.go +++ b/ethdb/database.go @@ -303,7 +303,7 @@ func (b *ldbBatch) Reset() { b.b.Reset() b.size = 0 } - +/* type table struct { db Database prefix string @@ -367,3 +367,4 @@ func (tb *tableBatch) ValueSize() int { func (tb *tableBatch) Reset() { tb.batch.Reset() } +*/ \ No newline at end of file diff --git a/node/node.go b/node/node.go index b02aecfad1..2f5ceb53bc 100644 --- a/node/node.go +++ b/node/node.go @@ -645,7 +645,7 @@ func (n *Node) OpenDatabase(name string, cache, handles int) (ethdb.Database, er if n.config.DataDir == "" { return ethdb.NewMemDatabase() } - return ethdb.NewLDBDatabase(n.config.resolvePath(name), cache, handles) + return ethdb.NewBadgerDatabase(n.config.resolvePath(name)) } // ResolvePath returns the absolute path of a resource in the instance directory. diff --git a/node/service.go b/node/service.go index 55062a5004..f5a8fb777e 100644 --- a/node/service.go +++ b/node/service.go @@ -43,7 +43,7 @@ func (ctx *ServiceContext) OpenDatabase(name string, cache int, handles int) (et if ctx.config.DataDir == "" { return ethdb.NewMemDatabase() } - db, err := ethdb.NewLDBDatabase(ctx.config.resolvePath(name), cache, handles) + db, err := ethdb.NewBadgerDatabase(ctx.config.resolvePath(name)) if err != nil { return nil, err } diff --git a/tests/testdata b/tests/testdata deleted file mode 160000 index 2bb0c3da3b..0000000000 --- a/tests/testdata +++ /dev/null @@ -1 +0,0 @@ -Subproject commit 2bb0c3da3bbb15c528bcef2a7e5ac4bd73f81f87 diff --git a/vendor/github.com/AndreasBriese/bbloom/LICENSE b/vendor/github.com/AndreasBriese/bbloom/LICENSE new file mode 100644 index 0000000000..4b20050e84 --- /dev/null +++ b/vendor/github.com/AndreasBriese/bbloom/LICENSE @@ -0,0 +1,35 @@ +bbloom.go + +// The MIT License (MIT) +// Copyright (c) 2014 Andreas Briese, eduToolbox@Bri-C GmbH, Sarstedt + +// Permission is hereby granted, free of charge, to any person obtaining a copy of +// this software and associated documentation files (the "Software"), to deal in +// the Software without restriction, including without limitation the rights to +// use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of +// the Software, and to permit persons to whom the Software is furnished to do so, +// subject to the following conditions: + +// The above copyright notice and this permission notice shall be included in all +// copies or substantial portions of the Software. + +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS +// FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR +// COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER +// IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN +// CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + +siphash.go + +// https://github.com/dchest/siphash +// +// Written in 2012 by Dmitry Chestnykh. +// +// To the extent possible under law, the author have dedicated all copyright +// and related and neighboring rights to this software to the public domain +// worldwide. This software is distributed without any warranty. +// http://creativecommons.org/publicdomain/zero/1.0/ +// +// Package siphash implements SipHash-2-4, a fast short-input PRF +// created by Jean-Philippe Aumasson and Daniel J. Bernstein. diff --git a/vendor/github.com/AndreasBriese/bbloom/README.md b/vendor/github.com/AndreasBriese/bbloom/README.md new file mode 100644 index 0000000000..6d77e146eb --- /dev/null +++ b/vendor/github.com/AndreasBriese/bbloom/README.md @@ -0,0 +1,129 @@ +## bbloom: a bitset Bloom filter for go/golang +=== + +package implements a fast bloom filter with real 'bitset' and JSONMarshal/JSONUnmarshal to store/reload the Bloom filter. + +NOTE: the package uses unsafe.Pointer to set and read the bits from the bitset. If you're uncomfortable with using the unsafe package, please consider using my bloom filter package at github.com/AndreasBriese/bloom + +=== + +changelog 11/2015: new thread safe methods AddTS(), HasTS(), AddIfNotHasTS() following a suggestion from Srdjan Marinovic (github @a-little-srdjan), who used this to code a bloomfilter cache. + +This bloom filter was developed to strengthen a website-log database and was tested and optimized for this log-entry mask: "2014/%02i/%02i %02i:%02i:%02i /info.html". +Nonetheless bbloom should work with any other form of entries. + +~~Hash function is a modified Berkeley DB sdbm hash (to optimize for smaller strings). sdbm http://www.cse.yorku.ca/~oz/hash.html~~ + +Found sipHash (SipHash-2-4, a fast short-input PRF created by Jean-Philippe Aumasson and Daniel J. Bernstein.) to be about as fast. sipHash had been ported by Dimtry Chestnyk to Go (github.com/dchest/siphash ) + +Minimum hashset size is: 512 ([4]uint64; will be set automatically). + +###install + +```sh +go get github.com/AndreasBriese/bbloom +``` + +###test ++ change to folder ../bbloom ++ create wordlist in file "words.txt" (you might use `python permut.py`) ++ run 'go test -bench=.' within the folder + +```go +go test -bench=. +``` + +~~If you've installed the GOCONVEY TDD-framework http://goconvey.co/ you can run the tests automatically.~~ + +using go's testing framework now (have in mind that the op timing is related to 65536 operations of Add, Has, AddIfNotHas respectively) + +### usage + +after installation add + +```go +import ( + ... + "github.com/AndreasBriese/bbloom" + ... + ) +``` + +at your header. In the program use + +```go +// create a bloom filter for 65536 items and 1 % wrong-positive ratio +bf := bbloom.New(float64(1<<16), float64(0.01)) + +// or +// create a bloom filter with 650000 for 65536 items and 7 locs per hash explicitly +// bf = bbloom.New(float64(650000), float64(7)) +// or +bf = bbloom.New(650000.0, 7.0) + +// add one item +bf.Add([]byte("butter")) + +// Number of elements added is exposed now +// Note: ElemNum will not be included in JSON export (for compatability to older version) +nOfElementsInFilter := bf.ElemNum + +// check if item is in the filter +isIn := bf.Has([]byte("butter")) // should be true +isNotIn := bf.Has([]byte("Butter")) // should be false + +// 'add only if item is new' to the bloomfilter +added := bf.AddIfNotHas([]byte("butter")) // should be false because 'butter' is already in the set +added = bf.AddIfNotHas([]byte("buTTer")) // should be true because 'buTTer' is new + +// thread safe versions for concurrent use: AddTS, HasTS, AddIfNotHasTS +// add one item +bf.AddTS([]byte("peanutbutter")) +// check if item is in the filter +isIn = bf.HasTS([]byte("peanutbutter")) // should be true +isNotIn = bf.HasTS([]byte("peanutButter")) // should be false +// 'add only if item is new' to the bloomfilter +added = bf.AddIfNotHasTS([]byte("butter")) // should be false because 'peanutbutter' is already in the set +added = bf.AddIfNotHasTS([]byte("peanutbuTTer")) // should be true because 'penutbuTTer' is new + +// convert to JSON ([]byte) +Json := bf.JSONMarshal() + +// bloomfilters Mutex is exposed for external un-/locking +// i.e. mutex lock while doing JSON conversion +bf.Mtx.Lock() +Json = bf.JSONMarshal() +bf.Mtx.Unlock() + +// restore a bloom filter from storage +bfNew := bbloom.JSONUnmarshal(Json) + +isInNew := bfNew.Has([]byte("butter")) // should be true +isNotInNew := bfNew.Has([]byte("Butter")) // should be false + +``` + +to work with the bloom filter. + +### why 'fast'? + +It's about 3 times faster than William Fitzgeralds bitset bloom filter https://github.com/willf/bloom . And it is about so fast as my []bool set variant for Boom filters (see https://github.com/AndreasBriese/bloom ) but having a 8times smaller memory footprint: + + + Bloom filter (filter size 524288, 7 hashlocs) + github.com/AndreasBriese/bbloom 'Add' 65536 items (10 repetitions): 6595800 ns (100 ns/op) + github.com/AndreasBriese/bbloom 'Has' 65536 items (10 repetitions): 5986600 ns (91 ns/op) + github.com/AndreasBriese/bloom 'Add' 65536 items (10 repetitions): 6304684 ns (96 ns/op) + github.com/AndreasBriese/bloom 'Has' 65536 items (10 repetitions): 6568663 ns (100 ns/op) + + github.com/willf/bloom 'Add' 65536 items (10 repetitions): 24367224 ns (371 ns/op) + github.com/willf/bloom 'Test' 65536 items (10 repetitions): 21881142 ns (333 ns/op) + github.com/dataence/bloom/standard 'Add' 65536 items (10 repetitions): 23041644 ns (351 ns/op) + github.com/dataence/bloom/standard 'Check' 65536 items (10 repetitions): 19153133 ns (292 ns/op) + github.com/cabello/bloom 'Add' 65536 items (10 repetitions): 131921507 ns (2012 ns/op) + github.com/cabello/bloom 'Contains' 65536 items (10 repetitions): 131108962 ns (2000 ns/op) + +(on MBPro15 OSX10.8.5 i7 4Core 2.4Ghz) + + +With 32bit bloom filters (bloom32) using modified sdbm, bloom32 does hashing with only 2 bit shifts, one xor and one substraction per byte. smdb is about as fast as fnv64a but gives less collisions with the dataset (see mask above). bloom.New(float64(10 * 1<<16),float64(7)) populated with 1<<16 random items from the dataset (see above) and tested against the rest results in less than 0.05% collisions. diff --git a/vendor/github.com/AndreasBriese/bbloom/bbloom.go b/vendor/github.com/AndreasBriese/bbloom/bbloom.go new file mode 100644 index 0000000000..3d45740667 --- /dev/null +++ b/vendor/github.com/AndreasBriese/bbloom/bbloom.go @@ -0,0 +1,270 @@ +// The MIT License (MIT) +// Copyright (c) 2014 Andreas Briese, eduToolbox@Bri-C GmbH, Sarstedt + +// Permission is hereby granted, free of charge, to any person obtaining a copy of +// this software and associated documentation files (the "Software"), to deal in +// the Software without restriction, including without limitation the rights to +// use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of +// the Software, and to permit persons to whom the Software is furnished to do so, +// subject to the following conditions: + +// The above copyright notice and this permission notice shall be included in all +// copies or substantial portions of the Software. + +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS +// FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR +// COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER +// IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN +// CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + +package bbloom + +import ( + "bytes" + "encoding/json" + "log" + "math" + "sync" + "unsafe" +) + +// helper +var mask = []uint8{1, 2, 4, 8, 16, 32, 64, 128} + +func getSize(ui64 uint64) (size uint64, exponent uint64) { + if ui64 < uint64(512) { + ui64 = uint64(512) + } + size = uint64(1) + for size < ui64 { + size <<= 1 + exponent++ + } + return size, exponent +} + +func calcSizeByWrongPositives(numEntries, wrongs float64) (uint64, uint64) { + size := -1 * numEntries * math.Log(wrongs) / math.Pow(float64(0.69314718056), 2) + locs := math.Ceil(float64(0.69314718056) * size / numEntries) + return uint64(size), uint64(locs) +} + +// New +// returns a new bloomfilter +func New(params ...float64) (bloomfilter Bloom) { + var entries, locs uint64 + if len(params) == 2 { + if params[1] < 1 { + entries, locs = calcSizeByWrongPositives(params[0], params[1]) + } else { + entries, locs = uint64(params[0]), uint64(params[1]) + } + } else { + log.Fatal("usage: New(float64(number_of_entries), float64(number_of_hashlocations)) i.e. New(float64(1000), float64(3)) or New(float64(number_of_entries), float64(number_of_hashlocations)) i.e. New(float64(1000), float64(0.03))") + } + size, exponent := getSize(uint64(entries)) + bloomfilter = Bloom{ + sizeExp: exponent, + size: size - 1, + setLocs: locs, + shift: 64 - exponent, + } + bloomfilter.Size(size) + return bloomfilter +} + +// NewWithBoolset +// takes a []byte slice and number of locs per entry +// returns the bloomfilter with a bitset populated according to the input []byte +func NewWithBoolset(bs *[]byte, locs uint64) (bloomfilter Bloom) { + bloomfilter = New(float64(len(*bs)<<3), float64(locs)) + ptr := uintptr(unsafe.Pointer(&bloomfilter.bitset[0])) + for _, b := range *bs { + *(*uint8)(unsafe.Pointer(ptr)) = b + ptr++ + } + return bloomfilter +} + +// bloomJSONImExport +// Im/Export structure used by JSONMarshal / JSONUnmarshal +type bloomJSONImExport struct { + FilterSet []byte + SetLocs uint64 +} + +// JSONUnmarshal +// takes JSON-Object (type bloomJSONImExport) as []bytes +// returns bloom32 / bloom64 object +func JSONUnmarshal(dbData []byte) Bloom { + bloomImEx := bloomJSONImExport{} + json.Unmarshal(dbData, &bloomImEx) + buf := bytes.NewBuffer(bloomImEx.FilterSet) + bs := buf.Bytes() + bf := NewWithBoolset(&bs, bloomImEx.SetLocs) + return bf +} + +// +// Bloom filter +type Bloom struct { + Mtx sync.Mutex + ElemNum uint64 + bitset []uint64 + sizeExp uint64 + size uint64 + setLocs uint64 + shift uint64 +} + +// <--- http://www.cse.yorku.ca/~oz/hash.html +// modified Berkeley DB Hash (32bit) +// hash is casted to l, h = 16bit fragments +// func (bl Bloom) absdbm(b *[]byte) (l, h uint64) { +// hash := uint64(len(*b)) +// for _, c := range *b { +// hash = uint64(c) + (hash << 6) + (hash << bl.sizeExp) - hash +// } +// h = hash >> bl.shift +// l = hash << bl.shift >> bl.shift +// return l, h +// } + +// Update: found sipHash of Jean-Philippe Aumasson & Daniel J. Bernstein to be even faster than absdbm() +// https://131002.net/siphash/ +// siphash was implemented for Go by Dmitry Chestnykh https://github.com/dchest/siphash + +// Add +// set the bit(s) for entry; Adds an entry to the Bloom filter +func (bl *Bloom) Add(entry []byte) { + l, h := bl.sipHash(entry) + for i := uint64(0); i < (*bl).setLocs; i++ { + (*bl).Set((h + i*l) & (*bl).size) + (*bl).ElemNum++ + } +} + +// AddTS +// Thread safe: Mutex.Lock the bloomfilter for the time of processing the entry +func (bl *Bloom) AddTS(entry []byte) { + bl.Mtx.Lock() + defer bl.Mtx.Unlock() + bl.Add(entry[:]) +} + +// Has +// check if bit(s) for entry is/are set +// returns true if the entry was added to the Bloom Filter +func (bl Bloom) Has(entry []byte) bool { + l, h := bl.sipHash(entry) + for i := uint64(0); i < bl.setLocs; i++ { + switch bl.IsSet((h + i*l) & bl.size) { + case false: + return false + } + } + return true +} + +// HasTS +// Thread safe: Mutex.Lock the bloomfilter for the time of processing the entry +func (bl *Bloom) HasTS(entry []byte) bool { + bl.Mtx.Lock() + defer bl.Mtx.Unlock() + return bl.Has(entry[:]) +} + +// AddIfNotHas +// Only Add entry if it's not present in the bloomfilter +// returns true if entry was added +// returns false if entry was allready registered in the bloomfilter +func (bl Bloom) AddIfNotHas(entry []byte) (added bool) { + if bl.Has(entry[:]) { + return added + } + bl.Add(entry[:]) + return true +} + +// AddIfNotHasTS +// Tread safe: Only Add entry if it's not present in the bloomfilter +// returns true if entry was added +// returns false if entry was allready registered in the bloomfilter +func (bl *Bloom) AddIfNotHasTS(entry []byte) (added bool) { + bl.Mtx.Lock() + defer bl.Mtx.Unlock() + return bl.AddIfNotHas(entry[:]) +} + +// Size +// make Bloom filter with as bitset of size sz +func (bl *Bloom) Size(sz uint64) { + (*bl).bitset = make([]uint64, sz>>6) +} + +// Clear +// resets the Bloom filter +func (bl *Bloom) Clear() { + for i, _ := range (*bl).bitset { + (*bl).bitset[i] = 0 + } +} + +// Set +// set the bit[idx] of bitsit +func (bl *Bloom) Set(idx uint64) { + ptr := unsafe.Pointer(uintptr(unsafe.Pointer(&bl.bitset[idx>>6])) + uintptr((idx%64)>>3)) + *(*uint8)(ptr) |= mask[idx%8] +} + +// IsSet +// check if bit[idx] of bitset is set +// returns true/false +func (bl *Bloom) IsSet(idx uint64) bool { + ptr := unsafe.Pointer(uintptr(unsafe.Pointer(&bl.bitset[idx>>6])) + uintptr((idx%64)>>3)) + r := ((*(*uint8)(ptr)) >> (idx % 8)) & 1 + return r == 1 +} + +// JSONMarshal +// returns JSON-object (type bloomJSONImExport) as []byte +func (bl Bloom) JSONMarshal() []byte { + bloomImEx := bloomJSONImExport{} + bloomImEx.SetLocs = uint64(bl.setLocs) + bloomImEx.FilterSet = make([]byte, len(bl.bitset)<<3) + ptr := uintptr(unsafe.Pointer(&bl.bitset[0])) + for i := range bloomImEx.FilterSet { + bloomImEx.FilterSet[i] = *(*byte)(unsafe.Pointer(ptr)) + ptr++ + } + data, err := json.Marshal(bloomImEx) + if err != nil { + log.Fatal("json.Marshal failed: ", err) + } + return data +} + +// // alternative hashFn +// func (bl Bloom) fnv64a(b *[]byte) (l, h uint64) { +// h64 := fnv.New64a() +// h64.Write(*b) +// hash := h64.Sum64() +// h = hash >> 32 +// l = hash << 32 >> 32 +// return l, h +// } +// +// // <-- http://partow.net/programming/hashfunctions/index.html +// // citation: An algorithm proposed by Donald E. Knuth in The Art Of Computer Programming Volume 3, +// // under the topic of sorting and search chapter 6.4. +// // modified to fit with boolset-length +// func (bl Bloom) DEKHash(b *[]byte) (l, h uint64) { +// hash := uint64(len(*b)) +// for _, c := range *b { +// hash = ((hash << 5) ^ (hash >> bl.shift)) ^ uint64(c) +// } +// h = hash >> bl.shift +// l = hash << bl.sizeExp >> bl.sizeExp +// return l, h +// } diff --git a/vendor/github.com/AndreasBriese/bbloom/sipHash.go b/vendor/github.com/AndreasBriese/bbloom/sipHash.go new file mode 100644 index 0000000000..a91d8199b2 --- /dev/null +++ b/vendor/github.com/AndreasBriese/bbloom/sipHash.go @@ -0,0 +1,225 @@ +// Written in 2012 by Dmitry Chestnykh. +// +// To the extent possible under law, the author have dedicated all copyright +// and related and neighboring rights to this software to the public domain +// worldwide. This software is distributed without any warranty. +// http://creativecommons.org/publicdomain/zero/1.0/ +// +// Package siphash implements SipHash-2-4, a fast short-input PRF +// created by Jean-Philippe Aumasson and Daniel J. Bernstein. + +package bbloom + +// Hash returns the 64-bit SipHash-2-4 of the given byte slice with two 64-bit +// parts of 128-bit key: k0 and k1. +func (bl Bloom) sipHash(p []byte) (l, h uint64) { + // Initialization. + v0 := uint64(8317987320269560794) // k0 ^ 0x736f6d6570736575 + v1 := uint64(7237128889637516672) // k1 ^ 0x646f72616e646f6d + v2 := uint64(7816392314733513934) // k0 ^ 0x6c7967656e657261 + v3 := uint64(8387220255325274014) // k1 ^ 0x7465646279746573 + t := uint64(len(p)) << 56 + + // Compression. + for len(p) >= 8 { + + m := uint64(p[0]) | uint64(p[1])<<8 | uint64(p[2])<<16 | uint64(p[3])<<24 | + uint64(p[4])<<32 | uint64(p[5])<<40 | uint64(p[6])<<48 | uint64(p[7])<<56 + + v3 ^= m + + // Round 1. + v0 += v1 + v1 = v1<<13 | v1>>51 + v1 ^= v0 + v0 = v0<<32 | v0>>32 + + v2 += v3 + v3 = v3<<16 | v3>>48 + v3 ^= v2 + + v0 += v3 + v3 = v3<<21 | v3>>43 + v3 ^= v0 + + v2 += v1 + v1 = v1<<17 | v1>>47 + v1 ^= v2 + v2 = v2<<32 | v2>>32 + + // Round 2. + v0 += v1 + v1 = v1<<13 | v1>>51 + v1 ^= v0 + v0 = v0<<32 | v0>>32 + + v2 += v3 + v3 = v3<<16 | v3>>48 + v3 ^= v2 + + v0 += v3 + v3 = v3<<21 | v3>>43 + v3 ^= v0 + + v2 += v1 + v1 = v1<<17 | v1>>47 + v1 ^= v2 + v2 = v2<<32 | v2>>32 + + v0 ^= m + p = p[8:] + } + + // Compress last block. + switch len(p) { + case 7: + t |= uint64(p[6]) << 48 + fallthrough + case 6: + t |= uint64(p[5]) << 40 + fallthrough + case 5: + t |= uint64(p[4]) << 32 + fallthrough + case 4: + t |= uint64(p[3]) << 24 + fallthrough + case 3: + t |= uint64(p[2]) << 16 + fallthrough + case 2: + t |= uint64(p[1]) << 8 + fallthrough + case 1: + t |= uint64(p[0]) + } + + v3 ^= t + + // Round 1. + v0 += v1 + v1 = v1<<13 | v1>>51 + v1 ^= v0 + v0 = v0<<32 | v0>>32 + + v2 += v3 + v3 = v3<<16 | v3>>48 + v3 ^= v2 + + v0 += v3 + v3 = v3<<21 | v3>>43 + v3 ^= v0 + + v2 += v1 + v1 = v1<<17 | v1>>47 + v1 ^= v2 + v2 = v2<<32 | v2>>32 + + // Round 2. + v0 += v1 + v1 = v1<<13 | v1>>51 + v1 ^= v0 + v0 = v0<<32 | v0>>32 + + v2 += v3 + v3 = v3<<16 | v3>>48 + v3 ^= v2 + + v0 += v3 + v3 = v3<<21 | v3>>43 + v3 ^= v0 + + v2 += v1 + v1 = v1<<17 | v1>>47 + v1 ^= v2 + v2 = v2<<32 | v2>>32 + + v0 ^= t + + // Finalization. + v2 ^= 0xff + + // Round 1. + v0 += v1 + v1 = v1<<13 | v1>>51 + v1 ^= v0 + v0 = v0<<32 | v0>>32 + + v2 += v3 + v3 = v3<<16 | v3>>48 + v3 ^= v2 + + v0 += v3 + v3 = v3<<21 | v3>>43 + v3 ^= v0 + + v2 += v1 + v1 = v1<<17 | v1>>47 + v1 ^= v2 + v2 = v2<<32 | v2>>32 + + // Round 2. + v0 += v1 + v1 = v1<<13 | v1>>51 + v1 ^= v0 + v0 = v0<<32 | v0>>32 + + v2 += v3 + v3 = v3<<16 | v3>>48 + v3 ^= v2 + + v0 += v3 + v3 = v3<<21 | v3>>43 + v3 ^= v0 + + v2 += v1 + v1 = v1<<17 | v1>>47 + v1 ^= v2 + v2 = v2<<32 | v2>>32 + + // Round 3. + v0 += v1 + v1 = v1<<13 | v1>>51 + v1 ^= v0 + v0 = v0<<32 | v0>>32 + + v2 += v3 + v3 = v3<<16 | v3>>48 + v3 ^= v2 + + v0 += v3 + v3 = v3<<21 | v3>>43 + v3 ^= v0 + + v2 += v1 + v1 = v1<<17 | v1>>47 + v1 ^= v2 + v2 = v2<<32 | v2>>32 + + // Round 4. + v0 += v1 + v1 = v1<<13 | v1>>51 + v1 ^= v0 + v0 = v0<<32 | v0>>32 + + v2 += v3 + v3 = v3<<16 | v3>>48 + v3 ^= v2 + + v0 += v3 + v3 = v3<<21 | v3>>43 + v3 ^= v0 + + v2 += v1 + v1 = v1<<17 | v1>>47 + v1 ^= v2 + v2 = v2<<32 | v2>>32 + + // return v0 ^ v1 ^ v2 ^ v3 + + hash := v0 ^ v1 ^ v2 ^ v3 + h = hash >> bl.shift + l = hash << bl.shift >> bl.shift + return l, h + +} diff --git a/vendor/github.com/dgraph-io/badger/CHANGELOG.md b/vendor/github.com/dgraph-io/badger/CHANGELOG.md new file mode 100644 index 0000000000..0fa6c3356c --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/CHANGELOG.md @@ -0,0 +1,40 @@ +# Changelog +All notable changes to this project will be documented in this file. + +The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/) +and this project adheres to [Semantic Versioning](http://semver.org/spec/v2.0.0.html). + +## [Unreleased] + +## [1.3.0] - 2017-12-12 +* Add `DB.NextSequence()` method to generate monotonically increasing integer + sequences. +* Add `DB.Size()` method to return the size of LSM and value log files. +* Tweaked mmap code to make Windows 32-bit builds work. +* Tweaked build tags on some files to make iOS builds work. +* Fix `DB.PurgeOlderVersions()` to not violate some constraints. + +## [1.2.0] - 2017-11-30 +* Expose a `Txn.SetEntry()` method to allow setting the key-value pair + and all the metadata at the same time. + +## [1.1.1] - 2017-11-28 +* Fix bug where txn.Get was returing key deleted in same transaction. +* Fix race condition while decrementing reference in oracle. +* Update doneCommit in the callback for CommitAsync. +* Iterator see writes of current txn. + +## [1.1.0] - 2017-11-13 +* Create Badger directory if it does not exist when `badger.Open` is called. +* Added `Item.ValueCopy()` to avoid deadlocks in long-running iterations +* Fixed 64-bit alignment issues to make Badger run on Arm v7 + +## [1.0.1] - 2017-11-06 +* Fix an uint16 overflow when resizing key slice + +[Unreleased]: https://github.com/dgraph-io/badger/compare/v1.3.0...HEAD +[1.3.0]: https://github.com/dgraph-io/badger/compare/v1.2.0...v1.3.0 +[1.2.0]: https://github.com/dgraph-io/badger/compare/v1.1.1...v1.2.0 +[1.1.1]: https://github.com/dgraph-io/badger/compare/v1.1.0...v1.1.1 +[1.1.0]: https://github.com/dgraph-io/badger/compare/v1.0.1...v1.1.0 +[1.0.1]: https://github.com/dgraph-io/badger/compare/v1.0.0...v1.0.1 diff --git a/vendor/github.com/dgraph-io/badger/LICENSE b/vendor/github.com/dgraph-io/badger/LICENSE new file mode 100644 index 0000000000..d9a10c0d8e --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/LICENSE @@ -0,0 +1,176 @@ + Apache License + Version 2.0, January 2004 + http://www.apache.org/licenses/ + + TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION + + 1. Definitions. + + "License" shall mean the terms and conditions for use, reproduction, + and distribution as defined by Sections 1 through 9 of this document. + + "Licensor" shall mean the copyright owner or entity authorized by + the copyright owner that is granting the License. + + "Legal Entity" shall mean the union of the acting entity and all + other entities that control, are controlled by, or are under common + control with that entity. For the purposes of this definition, + "control" means (i) the power, direct or indirect, to cause the + direction or management of such entity, whether by contract or + otherwise, or (ii) ownership of fifty percent (50%) or more of the + outstanding shares, or (iii) beneficial ownership of such entity. + + "You" (or "Your") shall mean an individual or Legal Entity + exercising permissions granted by this License. + + "Source" form shall mean the preferred form for making modifications, + including but not limited to software source code, documentation + source, and configuration files. + + "Object" form shall mean any form resulting from mechanical + transformation or translation of a Source form, including but + not limited to compiled object code, generated documentation, + and conversions to other media types. + + "Work" shall mean the work of authorship, whether in Source or + Object form, made available under the License, as indicated by a + copyright notice that is included in or attached to the work + (an example is provided in the Appendix below). + + "Derivative Works" shall mean any work, whether in Source or Object + form, that is based on (or derived from) the Work and for which the + editorial revisions, annotations, elaborations, or other modifications + represent, as a whole, an original work of authorship. For the purposes + of this License, Derivative Works shall not include works that remain + separable from, or merely link (or bind by name) to the interfaces of, + the Work and Derivative Works thereof. + + "Contribution" shall mean any work of authorship, including + the original version of the Work and any modifications or additions + to that Work or Derivative Works thereof, that is intentionally + submitted to Licensor for inclusion in the Work by the copyright owner + or by an individual or Legal Entity authorized to submit on behalf of + the copyright owner. For the purposes of this definition, "submitted" + means any form of electronic, verbal, or written communication sent + to the Licensor or its representatives, including but not limited to + communication on electronic mailing lists, source code control systems, + and issue tracking systems that are managed by, or on behalf of, the + Licensor for the purpose of discussing and improving the Work, but + excluding communication that is conspicuously marked or otherwise + designated in writing by the copyright owner as "Not a Contribution." + + "Contributor" shall mean Licensor and any individual or Legal Entity + on behalf of whom a Contribution has been received by Licensor and + subsequently incorporated within the Work. + + 2. Grant of Copyright License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + copyright license to reproduce, prepare Derivative Works of, + publicly display, publicly perform, sublicense, and distribute the + Work and such Derivative Works in Source or Object form. + + 3. Grant of Patent License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + (except as stated in this section) patent license to make, have made, + use, offer to sell, sell, import, and otherwise transfer the Work, + where such license applies only to those patent claims licensable + by such Contributor that are necessarily infringed by their + Contribution(s) alone or by combination of their Contribution(s) + with the Work to which such Contribution(s) was submitted. If You + institute patent litigation against any entity (including a + cross-claim or counterclaim in a lawsuit) alleging that the Work + or a Contribution incorporated within the Work constitutes direct + or contributory patent infringement, then any patent licenses + granted to You under this License for that Work shall terminate + as of the date such litigation is filed. + + 4. Redistribution. You may reproduce and distribute copies of the + Work or Derivative Works thereof in any medium, with or without + modifications, and in Source or Object form, provided that You + meet the following conditions: + + (a) You must give any other recipients of the Work or + Derivative Works a copy of this License; and + + (b) You must cause any modified files to carry prominent notices + stating that You changed the files; and + + (c) You must retain, in the Source form of any Derivative Works + that You distribute, all copyright, patent, trademark, and + attribution notices from the Source form of the Work, + excluding those notices that do not pertain to any part of + the Derivative Works; and + + (d) If the Work includes a "NOTICE" text file as part of its + distribution, then any Derivative Works that You distribute must + include a readable copy of the attribution notices contained + within such NOTICE file, excluding those notices that do not + pertain to any part of the Derivative Works, in at least one + of the following places: within a NOTICE text file distributed + as part of the Derivative Works; within the Source form or + documentation, if provided along with the Derivative Works; or, + within a display generated by the Derivative Works, if and + wherever such third-party notices normally appear. The contents + of the NOTICE file are for informational purposes only and + do not modify the License. You may add Your own attribution + notices within Derivative Works that You distribute, alongside + or as an addendum to the NOTICE text from the Work, provided + that such additional attribution notices cannot be construed + as modifying the License. + + You may add Your own copyright statement to Your modifications and + may provide additional or different license terms and conditions + for use, reproduction, or distribution of Your modifications, or + for any such Derivative Works as a whole, provided Your use, + reproduction, and distribution of the Work otherwise complies with + the conditions stated in this License. + + 5. Submission of Contributions. Unless You explicitly state otherwise, + any Contribution intentionally submitted for inclusion in the Work + by You to the Licensor shall be under the terms and conditions of + this License, without any additional terms or conditions. + Notwithstanding the above, nothing herein shall supersede or modify + the terms of any separate license agreement you may have executed + with Licensor regarding such Contributions. + + 6. Trademarks. This License does not grant permission to use the trade + names, trademarks, service marks, or product names of the Licensor, + except as required for reasonable and customary use in describing the + origin of the Work and reproducing the content of the NOTICE file. + + 7. Disclaimer of Warranty. Unless required by applicable law or + agreed to in writing, Licensor provides the Work (and each + Contributor provides its Contributions) on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or + implied, including, without limitation, any warranties or conditions + of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A + PARTICULAR PURPOSE. You are solely responsible for determining the + appropriateness of using or redistributing the Work and assume any + risks associated with Your exercise of permissions under this License. + + 8. Limitation of Liability. In no event and under no legal theory, + whether in tort (including negligence), contract, or otherwise, + unless required by applicable law (such as deliberate and grossly + negligent acts) or agreed to in writing, shall any Contributor be + liable to You for damages, including any direct, indirect, special, + incidental, or consequential damages of any character arising as a + result of this License or out of the use or inability to use the + Work (including but not limited to damages for loss of goodwill, + work stoppage, computer failure or malfunction, or any and all + other commercial damages or losses), even if such Contributor + has been advised of the possibility of such damages. + + 9. Accepting Warranty or Additional Liability. While redistributing + the Work or Derivative Works thereof, You may choose to offer, + and charge a fee for, acceptance of support, warranty, indemnity, + or other liability obligations and/or rights consistent with this + License. However, in accepting such obligations, You may act only + on Your own behalf and on Your sole responsibility, not on behalf + of any other Contributor, and only if You agree to indemnify, + defend, and hold each Contributor harmless for any liability + incurred by, or claims asserted against, such Contributor by reason + of your accepting any such warranty or additional liability. + + END OF TERMS AND CONDITIONS diff --git a/vendor/github.com/dgraph-io/badger/README.md b/vendor/github.com/dgraph-io/badger/README.md new file mode 100644 index 0000000000..bb2840840c --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/README.md @@ -0,0 +1,606 @@ +# BadgerDB [![GoDoc](https://godoc.org/github.com/dgraph-io/badger?status.svg)](https://godoc.org/github.com/dgraph-io/badger) [![Go Report Card](https://goreportcard.com/badge/github.com/dgraph-io/badger)](https://goreportcard.com/report/github.com/dgraph-io/badger) [![Build Status](https://teamcity.dgraph.io/guestAuth/app/rest/builds/buildType:(id:Badger_UnitTests)/statusIcon.svg)](https://teamcity.dgraph.io/viewLog.html?buildTypeId=Badger_UnitTests&buildId=lastFinished&guest=1) ![Appveyor](https://ci.appveyor.com/api/projects/status/github/dgraph-io/badger?branch=master&svg=true) [![Coverage Status](https://coveralls.io/repos/github/dgraph-io/badger/badge.svg?branch=master)](https://coveralls.io/github/dgraph-io/badger?branch=master) + +![Badger mascot](images/diggy-shadow.png) + +BadgerDB is an embeddable, persistent, simple and fast key-value (KV) database +written in pure Go. It's meant to be a performant alternative to non-Go-based +key-value stores like [RocksDB](https://github.com/facebook/rocksdb). + +## Project Status +Badger v1.0 was released in Nov 2017. Check the [Changelog] for the full details. + +[Changelog]:https://github.com/dgraph-io/badger/blob/master/CHANGELOG.md + +We introduced transactions in [v0.9.0] which involved a major API change. If you have a Badger +datastore prior to that, please use [v0.8.1], but we strongly urge you to upgrade. Upgrading from +both v0.8 and v0.9 will require you to [take backups](#database-backup) and restore using the new +version. + +[v1.0.1]: //github.com/dgraph-io/badger/tree/v1.0.1 +[v0.8.1]: //github.com/dgraph-io/badger/tree/v0.8.1 +[v0.9.0]: //github.com/dgraph-io/badger/tree/v0.9.0 + +## Table of Contents + * [Getting Started](#getting-started) + + [Installing](#installing) + + [Opening a database](#opening-a-database) + + [Transactions](#transactions) + - [Read-only transactions](#read-only-transactions) + - [Read-write transactions](#read-write-transactions) + - [Managing transactions manually](#managing-transactions-manually) + + [Using key/value pairs](#using-keyvalue-pairs) + + [Monotonically increasing integers](#monotonically-increasing-integers) + * [Merge Operations](#merge-operations) + + [Setting Time To Live(TTL) and User Metadata on Keys](#setting-time-to-livettl-and-user-metadata-on-keys) + + [Iterating over keys](#iterating-over-keys) + - [Prefix scans](#prefix-scans) + - [Key-only iteration](#key-only-iteration) + + [Garbage Collection](#garbage-collection) + + [Database backup](#database-backup) + + [Memory usage](#memory-usage) + + [Statistics](#statistics) + * [Resources](#resources) + + [Blog Posts](#blog-posts) + * [Contact](#contact) + * [Design](#design) + + [Comparisons](#comparisons) + + [Benchmarks](#benchmarks) + * [Other Projects Using Badger](#other-projects-using-badger) + * [Frequently Asked Questions](#frequently-asked-questions) + +## Getting Started + +### Installing +To start using Badger, install Go 1.8 or above and run `go get`: + +```sh +$ go get github.com/dgraph-io/badger/... +``` + +This will retrieve the library and install the `badger_info` command line +utility into your `$GOBIN` path. + + +### Opening a database +The top-level object in Badger is a `DB`. It represents multiple files on disk +in specific directories, which contain the data for a single database. + +To open your database, use the `badger.Open()` function, with the appropriate +options. The `Dir` and `ValueDir` options are mandatory and must be +specified by the client. They can be set to the same value to simplify things. + +```go +package main + +import ( + "log" + + "github.com/dgraph-io/badger" +) + +func main() { + // Open the Badger database located in the /tmp/badger directory. + // It will be created if it doesn't exist. + opts := badger.DefaultOptions + opts.Dir = "/tmp/badger" + opts.ValueDir = "/tmp/badger" + db, err := badger.Open(opts) + if err != nil { + log.Fatal(err) + } + defer db.Close() +  // Your code here… +} +``` + +Please note that Badger obtains a lock on the directories so multiple processes +cannot open the same database at the same time. + +### Transactions + +#### Read-only transactions +To start a read-only transaction, you can use the `DB.View()` method: + +```go +err := db.View(func(txn *badger.Txn) error { +  // Your code here… +  return nil +}) +``` + +You cannot perform any writes or deletes within this transaction. Badger +ensures that you get a consistent view of the database within this closure. Any +writes that happen elsewhere after the transaction has started, will not be +seen by calls made within the closure. + +#### Read-write transactions +To start a read-write transaction, you can use the `DB.Update()` method: + +```go +err := db.Update(func(txn *badger.Txn) error { +  // Your code here… +  return nil +}) +``` + +All database operations are allowed inside a read-write transaction. + +Always check the returned error value. If you return an error +within your closure it will be passed through. + +An `ErrConflict` error will be reported in case of a conflict. Depending on the state +of your application, you have the option to retry the operation if you receive +this error. + +An `ErrTxnTooBig` will be reported in case the number of pending writes/deletes in +the transaction exceed a certain limit. In that case, it is best to commit the +transaction and start a new transaction immediately. Here is an example (we are +not checking for errors in some places for simplicity): + +```go +updates := make(map[string]string) +txn := db.NewTransaction(true) +for k,v := range updates { + if err := txn.Set(byte[](k),byte[](v)); err == ErrTxnTooBig { + _ = txn.Commit() + txn = db.NewTransaction(..) + _ = txn.Set(k,v) + } +} +_ = txn.Commit() +``` + +#### Managing transactions manually +The `DB.View()` and `DB.Update()` methods are wrappers around the +`DB.NewTransaction()` and `Txn.Commit()` methods (or `Txn.Discard()` in case of +read-only transactions). These helper methods will start the transaction, +execute a function, and then safely discard your transaction if an error is +returned. This is the recommended way to use Badger transactions. + +However, sometimes you may want to manually create and commit your +transactions. You can use the `DB.NewTransaction()` function directly, which +takes in a boolean argument to specify whether a read-write transaction is +required. For read-write transactions, it is necessary to call `Txn.Commit()` +to ensure the transaction is committed. For read-only transactions, calling +`Txn.Discard()` is sufficient. `Txn.Commit()` also calls `Txn.Discard()` +internally to cleanup the transaction, so just calling `Txn.Commit()` is +sufficient for read-write transaction. However, if your code doesn’t call +`Txn.Commit()` for some reason (for e.g it returns prematurely with an error), +then please make sure you call `Txn.Discard()` in a `defer` block. Refer to the +code below. + +```go +// Start a writable transaction. +txn, err := db.NewTransaction(true) +if err != nil { + return err +} +defer txn.Discard() + +// Use the transaction... +err := txn.Set([]byte("answer"), []byte("42")) +if err != nil { + return err +} + +// Commit the transaction and check for error. +if err := txn.Commit(nil); err != nil { + return err +} +``` + +The first argument to `DB.NewTransaction()` is a boolean stating if the transaction +should be writable. + +Badger allows an optional callback to the `Txn.Commit()` method. Normally, the +callback can be set to `nil`, and the method will return after all the writes +have succeeded. However, if this callback is provided, the `Txn.Commit()` +method returns as soon as it has checked for any conflicts. The actual writing +to the disk happens asynchronously, and the callback is invoked once the +writing has finished, or an error has occurred. This can improve the throughput +of the application in some cases. But it also means that a transaction is not +durable until the callback has been invoked with a `nil` error value. + +### Using key/value pairs +To save a key/value pair, use the `Txn.Set()` method: + +```go +err := db.Update(func(txn *badger.Txn) error { + err := txn.Set([]byte("answer"), []byte("42")) + return err +}) +``` + +This will set the value of the `"answer"` key to `"42"`. To retrieve this +value, we can use the `Txn.Get()` method: + +```go +err := db.View(func(txn *badger.Txn) error { + item, err := txn.Get([]byte("answer")) + if err != nil { + return err + } + val, err := item.Value() + if err != nil { + return err + } + fmt.Printf("The answer is: %s\n", val) + return nil +}) +``` + +`Txn.Get()` returns `ErrKeyNotFound` if the value is not found. + +Please note that values returned from `Get()` are only valid while the +transaction is open. If you need to use a value outside of the transaction +then you must use `copy()` to copy it to another byte slice. + +Use the `Txn.Delete()` method to delete a key. + +### Monotonically increasing integers + +To get unique monotonically increasing integers with strong durability, you can +use the `DB.GetSequence` method. This method returns a `Sequence` object, which +is thread-safe and can be used concurrently via various goroutines. + +Badger would lease a range of integers to hand out from memory, with the +bandwidth provided to `DB.GetSequence`. The frequency at which disk writes are +done is determined by this lease bandwidth and the frequency of `Next` +invocations. Setting a bandwith too low would do more disk writes, setting it +too high would result in wasted integers if Badger is closed or crashes. +To avoid wasted integers, call `Release` before closing Badger. + +```go +seq, err := db.GetSequence(key, 1000) +defer seq.Release() +for { + num, err := seq.Next() +} +``` + +### Merge Operations +Badger provides support for unordered merge operations. You can define a func +of type `MergeFunc` which takes in an existing value, and a value to be +_merged_ with it. It returns a new value which is the result of the _merge_ +operation. All values are specified in byte arrays. For e.g., here is a merge +function (`add`) which adds a `uint64` value to an existing `uint64` value. + +```Go +uint64ToBytes(i uint64) []byte { + var buf [8]byte + binary.BigEndian.PutUint64(buf[:], i) + return buf[:] +} + +func bytesToUint64(b []byte) uint64 { + return binary.BigEndian.Uint64(b) +} + +// Merge function to add two uint64 numbers +func add(existing, new []byte) []byte { + return uint64ToBytes(bytesToUint64(existing) + bytesToUint64(new)) +} +``` + +This function can then be passed to the `DB.GetMergeOperator()` method, along +with a key, and a duration value. The duration specifies how often the merge +function is run on values that have been added using the `MergeOperator.Add()` +method. + +`MergeOperator.Get()` method can be used to retrieve the cumulative value of the key +associated with the merge operation. + +```Go +key := []byte("merge") +m := db.GetMergeOperator(key, add, 200*time.Millisecond) +defer m.Stop() + +m.Add(uint64ToBytes(1)) +m.Add(uint64ToBytes(2)) +m.Add(uint64ToBytes(3)) + +res, err := m.Get() // res should have value 6 encoded +fmt.Println(bytesToUint64(res)) +``` + +### Setting Time To Live(TTL) and User Metadata on Keys +Badger allows setting an optional Time to Live (TTL) value on keys. Once the TTL has +elapsed, the key will no longer be retrievable and will be eligible for garbage +collection. A TTL can be set as a `time.Duration` value using the `Txn.SetWithTTL()` +API method. + +An optional user metadata value can be set on each key. A user metadata value +is represented by a single byte. It can be used to set certain bits along +with the key to aid in interpreting or decoding the key-value pair. User +metadata can be set using the `Txn.SetWithMeta()` API method. + +`Txn.SetEntry()` can be used to set the key, value, user metatadata and TTL, +all at once. + +### Iterating over keys +To iterate over keys, we can use an `Iterator`, which can be obtained using the +`Txn.NewIterator()` method. Iteration happens in byte-wise lexicographical sorting +order. + + +```go +err := db.View(func(txn *badger.Txn) error { + opts := badger.DefaultIteratorOptions + opts.PrefetchSize = 10 + it := txn.NewIterator(opts) + for it.Rewind(); it.Valid(); it.Next() { + item := it.Item() + k := item.Key() + v, err := item.Value() + if err != nil { + return err + } + fmt.Printf("key=%s, value=%s\n", k, v) + } + return nil +}) +``` + +The iterator allows you to move to a specific point in the list of keys and move +forward or backward through the keys one at a time. + +By default, Badger prefetches the values of the next 100 items. You can adjust +that with the `IteratorOptions.PrefetchSize` field. However, setting it to +a value higher than GOMAXPROCS (which we recommend to be 128 or higher) +shouldn’t give any additional benefits. You can also turn off the fetching of +values altogether. See section below on key-only iteration. + +#### Prefix scans +To iterate over a key prefix, you can combine `Seek()` and `ValidForPrefix()`: + +```go +db.View(func(txn *badger.Txn) error { + it := txn.NewIterator(badger.DefaultIteratorOptions) + prefix := []byte("1234") + for it.Seek(prefix); it.ValidForPrefix(prefix); it.Next() { + item := it.Item() + k := item.Key() + v, err := item.Value() + if err != nil { + return err + } + fmt.Printf("key=%s, value=%s\n", k, v) + } + return nil +}) +``` + +#### Key-only iteration +Badger supports a unique mode of iteration called _key-only_ iteration. It is +several order of magnitudes faster than regular iteration, because it involves +access to the LSM-tree only, which is usually resident entirely in RAM. To +enable key-only iteration, you need to set the `IteratorOptions.PrefetchValues` +field to `false`. This can also be used to do sparse reads for selected keys +during an iteration, by calling `item.Value()` only when required. + +```go +err := db.View(func(txn *badger.Txn) error { + opts := badger.DefaultIteratorOptions + opts.PrefetchValues = false + it := txn.NewIterator(opts) + for it.Rewind(); it.Valid(); it.Next() { + item := it.Item() + k := item.Key() + fmt.Printf("key=%s\n", k) + } + return nil +}) +``` + +### Garbage Collection +Badger values need to be garbage collected, because of two reasons: + +* Badger keeps values separately from the LSM tree. This means that the compaction operations +that clean up the LSM tree do not touch the values at all. Values need to be cleaned up +separately. + +* Concurrent read/write transactions could leave behind multiple values for a single key, because they +are stored with different versions. These could accumulate, and take up unneeded space beyond the +time these older versions are needed. + +Badger relies on the client to perform garbage collection at a time of their choosing. It provides +the following methods, which can be invoked at an appropriate time: + +* `DB.PurgeOlderVersions()`: This method iterates over the database, and cleans up all but the latest +versions of the key-value pairs. It marks the older versions as deleted, which makes them eligible for +garbage collection. +* `DB.PurgeVersionsBelow(key, ts)`: This method is useful to do a more targeted clean up of older versions +of key-value pairs. You can specify a key, and a timestamp. All versions of the key older than the timestamp +are marked as deleted, making them eligible for garbage collection. +* `DB.RunValueLogGC()`: This method is designed to do garbage collection while + Badger is online. Please ensure that you call the `DB.Purge…()` methods first + before invoking this method. It uses any statistics generated by the + `DB.Purge(…)` methods to pick files that are likely to lead to maximum space + reclamation. It loops until it encounters a file which does not lead to any + garbage collection. + + It could lead to increased I/O if `DB.RunValueLogGC()` hasn’t been called for + a long time, and many deletes have happened in the meanwhile. So it is recommended + that this method be called regularly. + +### Database backup +There are two public API methods `DB.Backup()` and `DB.Load()` which can be +used to do online backups and restores. Badger v0.9 provides a CLI tool +`badger`, which can do offline backup/restore. Make sure you have `$GOPATH/bin` +in your PATH to use this tool. + +The command below will create a version-agnostic backup of the database, to a +file `badger.bak` in the current working directory + +``` +badger backup --dir +``` + +To restore `badger.bak` in the current working directory to a new database: + +``` +badger restore --dir +``` + +See `badger --help` for more details. + +If you have a Badger database that was created using v0.8 (or below), you can +use the `badger_backup` tool provided in v0.8.1, and then restore it using the +command above to upgrade your database to work with the latest version. + +``` +badger_backup --dir --backup-file badger.bak +``` + +### Memory usage +Badger's memory usage can be managed by tweaking several options available in +the `Options` struct that is passed in when opening the database using +`DB.Open`. + +- `Options.ValueLogLoadingMode` can be set to `options.FileIO` (instead of the + default `options.MemoryMap`) to avoid memory-mapping log files. This can be + useful in environments with low RAM. +- Number of memtables (`Options.NumMemtables`) + - If you modify `Options.NumMemtables`, also adjust `Options.NumLevelZeroTables` and + `Options.NumLevelZeroTablesStall` accordingly. +- Number of concurrent compactions (`Options.NumCompactors`) +- Mode in which LSM tree is loaded (`Options.TableLoadingMode`) +- Size of table (`Options.MaxTableSize`) +- Size of value log file (`Options.ValueLogFileSize`) + +If you want to decrease the memory usage of Badger instance, tweak these +options (ideally one at a time) until you achieve the desired +memory usage. + +### Statistics +Badger records metrics using the [expvar] package, which is included in the Go +standard library. All the metrics are documented in [y/metrics.go][metrics] +file. + +`expvar` package adds a handler in to the default HTTP server (which has to be +started explicitly), and serves up the metrics at the `/debug/vars` endpoint. +These metrics can then be collected by a system like [Prometheus], to get +better visibility into what Badger is doing. + +[expvar]: https://golang.org/pkg/expvar/ +[metrics]: https://github.com/dgraph-io/badger/blob/master/y/metrics.go +[Prometheus]: https://prometheus.io/ + +## Resources + +### Blog Posts +1. [Introducing Badger: A fast key-value store written natively in +Go](https://open.dgraph.io/post/badger/) +2. [Make Badger crash resilient with ALICE](https://blog.dgraph.io/post/alice/) +3. [Badger vs LMDB vs BoltDB: Benchmarking key-value databases in Go](https://blog.dgraph.io/post/badger-lmdb-boltdb/) +4. [Concurrent ACID Transactions in Badger](https://blog.dgraph.io/post/badger-txn/) + +## Design +Badger was written with these design goals in mind: + +- Write a key-value database in pure Go. +- Use latest research to build the fastest KV database for data sets spanning terabytes. +- Optimize for SSDs. + +Badger’s design is based on a paper titled _[WiscKey: Separating Keys from +Values in SSD-conscious Storage][wisckey]_. + +[wisckey]: https://www.usenix.org/system/files/conference/fast16/fast16-papers-lu.pdf + +### Comparisons +| Feature | Badger | RocksDB | BoltDB | +| ------- | ------ | ------- | ------ | +| Design | LSM tree with value log | LSM tree only | B+ tree | +| High Read throughput | Yes | No | Yes | +| High Write throughput | Yes | Yes | No | +| Designed for SSDs | Yes (with latest research 1) | Not specifically 2 | No | +| Embeddable | Yes | Yes | Yes | +| Sorted KV access | Yes | Yes | Yes | +| Pure Go (no Cgo) | Yes | No | Yes | +| Transactions | Yes, ACID, concurrent with SSI3 | Yes (but non-ACID) | Yes, ACID | +| Snapshots | Yes | Yes | Yes | +| TTL support | Yes | Yes | No | + +1 The [WISCKEY paper][wisckey] (on which Badger is based) saw big +wins with separating values from keys, significantly reducing the write +amplification compared to a typical LSM tree. + +2 RocksDB is an SSD optimized version of LevelDB, which was designed specifically for rotating disks. +As such RocksDB's design isn't aimed at SSDs. + +3 SSI: Serializable Snapshot Isolation. For more details, see the blog post [Concurrent ACID Transactions in Badger](https://blog.dgraph.io/post/badger-txn/) + +### Benchmarks +We have run comprehensive benchmarks against RocksDB, Bolt and LMDB. The +benchmarking code, and the detailed logs for the benchmarks can be found in the +[badger-bench] repo. More explanation, including graphs can be found the blog posts (linked +above). + +[badger-bench]: https://github.com/dgraph-io/badger-bench + +## Other Projects Using Badger +Below is a list of public, open source projects that use Badger: + +* [Dgraph](https://github.com/dgraph-io/dgraph) - Distributed graph database. +* [go-ipfs](https://github.com/ipfs/go-ipfs) - Go client for the InterPlanetary File System (IPFS), a new hypermedia distribution protocol. +* [0-stor](https://github.com/zero-os/0-stor) - Single device object store. +* [Sandglass](https://github.com/celrenheit/sandglass) - distributed, horizontally scalable, persistent, time sorted message queue. + +If you are using Badger in a project please send a pull request to add it to the list. + +## Frequently Asked Questions +- **My writes are getting stuck. Why?** + +This can happen if a long running iteration with `Prefetch` is set to false, but +a `Item::Value` call is made internally in the loop. That causes Badger to +acquire read locks over the value log files to avoid value log GC removing the +file from underneath. As a side effect, this also blocks a new value log GC +file from being created, when the value log file boundary is hit. + +Please see Github issues [#293](https://github.com/dgraph-io/badger/issues/293) +and [#315](https://github.com/dgraph-io/badger/issues/315). + +There are multiple workarounds during iteration: + +1. Use `Item::ValueCopy` instead of `Item::Value` when retrieving value. +1. Set `Prefetch` to true. Badger would then copy over the value and release the + file lock immediately. +1. When `Prefetch` is false, don't call `Item::Value` and do a pure key-only + iteration. This might be useful if you just want to delete a lot of keys. +1. Do the writes in a separate transaction after the reads. + +- **My writes are really slow. Why?** + +Are you creating a new transaction for every single key update? This will lead +to very low throughput. To get best write performance, batch up multiple writes +inside a transaction using single `DB.Update()` call. You could also have +multiple such `DB.Update()` calls being made concurrently from multiple +goroutines. + +- **I don't see any disk write. Why?** + +If you're using Badger with `SyncWrites=false`, then your writes might not be written to value log +and won't get synced to disk immediately. Writes to LSM tree are done inmemory first, before they +get compacted to disk. The compaction would only happen once `MaxTableSize` has been reached. So, if +you're doing a few writes and then checking, you might not see anything on disk. Once you `Close` +the database, you'll see these writes on disk. + +- **Which instances should I use for Badger?** + +We recommend using instances which provide local SSD storage, without any limit +on the maximum IOPS. In AWS, these are storage optimized instances like i3. They +provide local SSDs which clock 100K IOPS over 4KB blocks easily. + +- **Are there any Go specific settings that I should use?** + +We *highly* recommend setting a high number for GOMAXPROCS, which allows Go to +observe the full IOPS throughput provided by modern SSDs. In Dgraph, we have set +it to 128. For more details, [see this +thread](https://groups.google.com/d/topic/golang-nuts/jPb_h3TvlKE/discussion). + +## Contact +- Please use [discuss.dgraph.io](https://discuss.dgraph.io) for questions, feature requests and discussions. +- Please use [Github issue tracker](https://github.com/dgraph-io/badger/issues) for filing bugs or feature requests. +- Join [![Slack Status](http://slack.dgraph.io/badge.svg)](http://slack.dgraph.io). +- Follow us on Twitter [@dgraphlabs](https://twitter.com/dgraphlabs). + diff --git a/vendor/github.com/dgraph-io/badger/appveyor.yml b/vendor/github.com/dgraph-io/badger/appveyor.yml new file mode 100644 index 0000000000..79dac338e7 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/appveyor.yml @@ -0,0 +1,48 @@ +# version format +version: "{build}" + +# Operating system (build VM template) +os: Windows Server 2012 R2 + +# Platform. +platform: x64 + +clone_folder: c:\gopath\src\github.com\dgraph-io\badger + +# Environment variables +environment: + GOVERSION: 1.8.3 + GOPATH: c:\gopath + +# scripts that run after cloning repository +install: + - set PATH=%GOPATH%\bin;c:\go\bin;%PATH% + - go version + - go env + - python --version + +# To run your custom scripts instead of automatic MSBuild +build_script: + # We need to disable firewall - https://github.com/appveyor/ci/issues/1579#issuecomment-309830648 + - ps: Disable-NetFirewallRule -DisplayName 'File and Printer Sharing (SMB-Out)' + - cd c:\gopath\src\github.com\dgraph-io\badger + - git branch + - go get -t ./... + +# To run your custom scripts instead of automatic tests +test_script: + # Unit tests + - ps: Add-AppveyorTest "Unit Tests" -Outcome Running + - go test -v github.com/dgraph-io/badger/... + - go test -v -vlog_mmap=false github.com/dgraph-io/badger/... + - ps: Update-AppveyorTest "Unit Tests" -Outcome Passed + +notifications: + - provider: Email + to: + - pawan@dgraph.io + on_build_failure: true + on_build_status_changed: true +# to disable deployment +deploy: off + diff --git a/vendor/github.com/dgraph-io/badger/backup.go b/vendor/github.com/dgraph-io/badger/backup.go new file mode 100644 index 0000000000..375b5a3162 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/backup.go @@ -0,0 +1,150 @@ +package badger + +import ( + "bufio" + "encoding/binary" + "io" + "sync" + + "github.com/dgraph-io/badger/y" + + "github.com/dgraph-io/badger/protos" +) + +func writeTo(entry *protos.KVPair, w io.Writer) error { + if err := binary.Write(w, binary.LittleEndian, uint64(entry.Size())); err != nil { + return err + } + buf, err := entry.Marshal() + if err != nil { + return err + } + _, err = w.Write(buf) + return err +} + +// Backup dumps a protobuf-encoded list of all entries in the database into the +// given writer, that are newer than the specified version. It returns a +// timestamp indicating when the entries were dumped which can be passed into a +// later invocation to generate an incremental dump, of entries that have been +// added/modified since the last invocation of DB.Backup() +// +// This can be used to backup the data in a database at a given point in time. +func (db *DB) Backup(w io.Writer, since uint64) (uint64, error) { + var tsNew uint64 + err := db.View(func(txn *Txn) error { + opts := DefaultIteratorOptions + opts.AllVersions = true + it := txn.NewIterator(opts) + for it.Rewind(); it.Valid(); it.Next() { + item := it.Item() + if item.Version() < since { + // Ignore versions less than given timestamp + continue + } + val, err := item.Value() + if err != nil { + return err + } + + entry := &protos.KVPair{ + Key: y.Copy(item.Key()), + Value: y.Copy(val), + UserMeta: []byte{item.UserMeta()}, + Version: item.Version(), + ExpiresAt: item.ExpiresAt(), + } + + // Write entries to disk + if err := writeTo(entry, w); err != nil { + return err + } + } + tsNew = txn.readTs + return nil + }) + return tsNew, err +} + +// Load reads a protobuf-encoded list of all entries from a reader and writes +// them to the database. This can be used to restore the database from a backup +// made by calling DB.Dump(). +// +// DB.Load() should be called on a database that is not running any other +// concurrent transactions while it is running. +func (db *DB) Load(r io.Reader) error { + br := bufio.NewReaderSize(r, 16<<10) + unmarshalBuf := make([]byte, 1<<10) + var entries []*Entry + var wg sync.WaitGroup + errChan := make(chan error, 1) + + // func to check for pending error before sending off a batch for writing + batchSetAsyncIfNoErr := func(entries []*Entry) error { + select { + case err := <-errChan: + return err + default: + wg.Add(1) + return db.batchSetAsync(entries, func(err error) { + defer wg.Done() + if err != nil { + select { + case errChan <- err: + default: + } + } + }) + } + } + + for { + var sz uint64 + err := binary.Read(br, binary.LittleEndian, &sz) + if err == io.EOF { + break + } else if err != nil { + return err + } + + if cap(unmarshalBuf) < int(sz) { + unmarshalBuf = make([]byte, sz) + } + + e := &protos.KVPair{} + if _, err = io.ReadFull(br, unmarshalBuf[:sz]); err != nil { + return err + } + if err = e.Unmarshal(unmarshalBuf[:sz]); err != nil { + return err + } + entries = append(entries, &Entry{ + Key: y.KeyWithTs(e.Key, e.Version), + Value: e.Value, + UserMeta: e.UserMeta[0], + ExpiresAt: e.ExpiresAt, + }) + + if len(entries) == 1000 { + if err := batchSetAsyncIfNoErr(entries); err != nil { + return err + } + entries = entries[:0] + } + } + + if len(entries) > 0 { + if err := batchSetAsyncIfNoErr(entries); err != nil { + return err + } + } + + wg.Wait() + + select { + case err := <-errChan: + return err + default: + return nil + } +} diff --git a/vendor/github.com/dgraph-io/badger/compaction.go b/vendor/github.com/dgraph-io/badger/compaction.go new file mode 100644 index 0000000000..23824ae175 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/compaction.go @@ -0,0 +1,208 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "bytes" + "fmt" + "log" + "sync" + + "golang.org/x/net/trace" + + "github.com/dgraph-io/badger/table" + "github.com/dgraph-io/badger/y" +) + +type keyRange struct { + left []byte + right []byte + inf bool +} + +var infRange = keyRange{inf: true} + +func (r keyRange) String() string { + return fmt.Sprintf("[left=%q, right=%q, inf=%v]", r.left, r.right, r.inf) +} + +func (r keyRange) equals(dst keyRange) bool { + return bytes.Equal(r.left, dst.left) && + bytes.Equal(r.right, dst.right) && + r.inf == dst.inf +} + +func (r keyRange) overlapsWith(dst keyRange) bool { + if r.inf || dst.inf { + return true + } + + // If my left is greater than dst right, we have no overlap. + if y.CompareKeys(r.left, dst.right) > 0 { + return false + } + // If my right is less than dst left, we have no overlap. + if y.CompareKeys(r.right, dst.left) < 0 { + return false + } + // We have overlap. + return true +} + +func getKeyRange(tables []*table.Table) keyRange { + y.AssertTrue(len(tables) > 0) + smallest := tables[0].Smallest() + biggest := tables[0].Biggest() + for i := 1; i < len(tables); i++ { + if y.CompareKeys(tables[i].Smallest(), smallest) < 0 { + smallest = tables[i].Smallest() + } + if y.CompareKeys(tables[i].Biggest(), biggest) > 0 { + biggest = tables[i].Biggest() + } + } + return keyRange{left: smallest, right: biggest} +} + +type levelCompactStatus struct { + ranges []keyRange + delSize int64 +} + +func (lcs *levelCompactStatus) debug() string { + var b bytes.Buffer + for _, r := range lcs.ranges { + b.WriteString(r.String()) + } + return b.String() +} + +func (lcs *levelCompactStatus) overlapsWith(dst keyRange) bool { + for _, r := range lcs.ranges { + if r.overlapsWith(dst) { + return true + } + } + return false +} + +func (lcs *levelCompactStatus) remove(dst keyRange) bool { + final := lcs.ranges[:0] + var found bool + for _, r := range lcs.ranges { + if !r.equals(dst) { + final = append(final, r) + } else { + found = true + } + } + lcs.ranges = final + return found +} + +type compactStatus struct { + sync.RWMutex + levels []*levelCompactStatus +} + +func (cs *compactStatus) toLog(tr trace.Trace) { + cs.RLock() + defer cs.RUnlock() + + tr.LazyPrintf("Compaction status:") + for i, l := range cs.levels { + if len(l.debug()) == 0 { + continue + } + tr.LazyPrintf("[%d] %s", i, l.debug()) + } +} + +func (cs *compactStatus) overlapsWith(level int, this keyRange) bool { + cs.RLock() + defer cs.RUnlock() + + thisLevel := cs.levels[level] + return thisLevel.overlapsWith(this) +} + +func (cs *compactStatus) delSize(l int) int64 { + cs.RLock() + defer cs.RUnlock() + return cs.levels[l].delSize +} + +type thisAndNextLevelRLocked struct{} + +// compareAndAdd will check whether we can run this compactDef. That it doesn't overlap with any +// other running compaction. If it can be run, it would store this run in the compactStatus state. +func (cs *compactStatus) compareAndAdd(_ thisAndNextLevelRLocked, cd compactDef) bool { + cs.Lock() + defer cs.Unlock() + + level := cd.thisLevel.level + + y.AssertTruef(level < len(cs.levels)-1, "Got level %d. Max levels: %d", level, len(cs.levels)) + thisLevel := cs.levels[level] + nextLevel := cs.levels[level+1] + + if thisLevel.overlapsWith(cd.thisRange) { + return false + } + if nextLevel.overlapsWith(cd.nextRange) { + return false + } + // Check whether this level really needs compaction or not. Otherwise, we'll end up + // running parallel compactions for the same level. + // NOTE: We can directly call thisLevel.totalSize, because we already have acquire a read lock + // over this and the next level. + if cd.thisLevel.totalSize-thisLevel.delSize < cd.thisLevel.maxTotalSize { + return false + } + + thisLevel.ranges = append(thisLevel.ranges, cd.thisRange) + nextLevel.ranges = append(nextLevel.ranges, cd.nextRange) + thisLevel.delSize += cd.thisSize + + return true +} + +func (cs *compactStatus) delete(cd compactDef) { + cs.Lock() + defer cs.Unlock() + + level := cd.thisLevel.level + y.AssertTruef(level < len(cs.levels)-1, "Got level %d. Max levels: %d", level, len(cs.levels)) + + thisLevel := cs.levels[level] + nextLevel := cs.levels[level+1] + + thisLevel.delSize -= cd.thisSize + found := thisLevel.remove(cd.thisRange) + found = nextLevel.remove(cd.nextRange) && found + + if !found { + this := cd.thisRange + next := cd.nextRange + fmt.Printf("Looking for: [%q, %q, %v] in this level.\n", this.left, this.right, this.inf) + fmt.Printf("This Level:\n%s\n", thisLevel.debug()) + fmt.Println() + fmt.Printf("Looking for: [%q, %q, %v] in next level.\n", next.left, next.right, next.inf) + fmt.Printf("Next Level:\n%s\n", nextLevel.debug()) + log.Fatal("keyRange not found") + } +} diff --git a/vendor/github.com/dgraph-io/badger/db.go b/vendor/github.com/dgraph-io/badger/db.go new file mode 100644 index 0000000000..1b68b3d553 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/db.go @@ -0,0 +1,1324 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "bytes" + "encoding/binary" + "expvar" + "log" + "math" + "os" + "path/filepath" + "strconv" + "sync" + "time" + + "github.com/dgraph-io/badger/options" + + "golang.org/x/net/trace" + + "github.com/dgraph-io/badger/skl" + "github.com/dgraph-io/badger/table" + "github.com/dgraph-io/badger/y" + "github.com/pkg/errors" +) + +var ( + badgerPrefix = []byte("!badger!") // Prefix for internal keys used by badger. + head = []byte("!badger!head") // For storing value offset for replay. + txnKey = []byte("!badger!txn") // For indicating end of entries in txn. +) + +type closers struct { + updateSize *y.Closer + compactors *y.Closer + memtable *y.Closer + writes *y.Closer + valueGC *y.Closer +} + +// DB provides the various functions required to interact with Badger. +// DB is thread-safe. +type DB struct { + sync.RWMutex // Guards list of inmemory tables, not individual reads and writes. + + dirLockGuard *directoryLockGuard + // nil if Dir and ValueDir are the same + valueDirGuard *directoryLockGuard + + closers closers + elog trace.EventLog + mt *skl.Skiplist // Our latest (actively written) in-memory table + imm []*skl.Skiplist // Add here only AFTER pushing to flushChan. + opt Options + manifest *manifestFile + lc *levelsController + vlog valueLog + vptr valuePointer // less than or equal to a pointer to the last vlog value put into mt + writeCh chan *request + flushChan chan flushTask // For flushing memtables. + + orc *oracle +} + +const ( + kvWriteChCapacity = 1000 +) + +func replayFunction(out *DB) func(Entry, valuePointer) error { + type txnEntry struct { + nk []byte + v y.ValueStruct + } + + var txn []txnEntry + var lastCommit uint64 + + toLSM := func(nk []byte, vs y.ValueStruct) { + for err := out.ensureRoomForWrite(); err != nil; err = out.ensureRoomForWrite() { + out.elog.Printf("Replay: Making room for writes") + time.Sleep(10 * time.Millisecond) + } + out.mt.Put(nk, vs) + } + + first := true + return func(e Entry, vp valuePointer) error { // Function for replaying. + if first { + out.elog.Printf("First key=%s\n", e.Key) + } + first = false + + if out.orc.curRead < y.ParseTs(e.Key) { + out.orc.curRead = y.ParseTs(e.Key) + } + + nk := make([]byte, len(e.Key)) + copy(nk, e.Key) + var nv []byte + meta := e.meta + if out.shouldWriteValueToLSM(e) { + nv = make([]byte, len(e.Value)) + copy(nv, e.Value) + } else { + nv = make([]byte, vptrSize) + vp.Encode(nv) + meta = meta | bitValuePointer + } + + v := y.ValueStruct{ + Value: nv, + Meta: meta, + UserMeta: e.UserMeta, + } + + if e.meta&bitFinTxn > 0 { + txnTs, err := strconv.ParseUint(string(e.Value), 10, 64) + if err != nil { + return errors.Wrapf(err, "Unable to parse txn fin: %q", e.Value) + } + y.AssertTrue(lastCommit == txnTs) + y.AssertTrue(len(txn) > 0) + // Got the end of txn. Now we can store them. + for _, t := range txn { + toLSM(t.nk, t.v) + } + txn = txn[:0] + lastCommit = 0 + + } else if e.meta&bitTxn == 0 { + // This entry is from a rewrite. + toLSM(nk, v) + + // We shouldn't get this entry in the middle of a transaction. + y.AssertTrue(lastCommit == 0) + y.AssertTrue(len(txn) == 0) + + } else { + txnTs := y.ParseTs(nk) + if lastCommit == 0 { + lastCommit = txnTs + } + y.AssertTrue(lastCommit == txnTs) + te := txnEntry{nk: nk, v: v} + txn = append(txn, te) + } + return nil + } +} + +// Open returns a new DB object. +func Open(opt Options) (db *DB, err error) { + opt.maxBatchSize = (15 * opt.MaxTableSize) / 100 + opt.maxBatchCount = opt.maxBatchSize / int64(skl.MaxNodeSize) + + for _, path := range []string{opt.Dir, opt.ValueDir} { + dirExists, err := exists(path) + if err != nil { + return nil, y.Wrapf(err, "Invalid Dir: %q", path) + } + if !dirExists { + // Try to create the directory + err = os.Mkdir(path, 0700) + if err != nil { + return nil, y.Wrapf(err, "Error Creating Dir: %q", path) + } + } + } + absDir, err := filepath.Abs(opt.Dir) + if err != nil { + return nil, err + } + absValueDir, err := filepath.Abs(opt.ValueDir) + if err != nil { + return nil, err + } + + dirLockGuard, err := acquireDirectoryLock(opt.Dir, lockFile) + if err != nil { + return nil, err + } + defer func() { + if dirLockGuard != nil { + _ = dirLockGuard.release() + } + }() + var valueDirLockGuard *directoryLockGuard + if absValueDir != absDir { + valueDirLockGuard, err = acquireDirectoryLock(opt.ValueDir, lockFile) + if err != nil { + return nil, err + } + } + defer func() { + if valueDirLockGuard != nil { + _ = valueDirLockGuard.release() + } + }() + if !(opt.ValueLogFileSize <= 2<<30 && opt.ValueLogFileSize >= 1<<20) { + return nil, ErrValueLogSize + } + if !(opt.ValueLogLoadingMode == options.FileIO || + opt.ValueLogLoadingMode == options.MemoryMap) { + return nil, ErrInvalidLoadingMode + } + manifestFile, manifest, err := openOrCreateManifestFile(opt.Dir) + if err != nil { + return nil, err + } + defer func() { + if manifestFile != nil { + _ = manifestFile.close() + } + }() + + orc := &oracle{ + isManaged: opt.managedTxns, + nextCommit: 1, + commits: make(map[uint64]uint64), + } + + db = &DB{ + imm: make([]*skl.Skiplist, 0, opt.NumMemtables), + flushChan: make(chan flushTask, opt.NumMemtables), + writeCh: make(chan *request, kvWriteChCapacity), + opt: opt, + manifest: manifestFile, + elog: trace.NewEventLog("Badger", "DB"), + dirLockGuard: dirLockGuard, + valueDirGuard: valueDirLockGuard, + orc: orc, + } + + // Calculate initial size. + db.calculateSize() + db.closers.updateSize = y.NewCloser(1) + go db.updateSize(db.closers.updateSize) + db.mt = skl.NewSkiplist(arenaSize(opt)) + + // newLevelsController potentially loads files in directory. + if db.lc, err = newLevelsController(db, &manifest); err != nil { + return nil, err + } + + db.closers.compactors = y.NewCloser(1) + db.lc.startCompact(db.closers.compactors) + + db.closers.memtable = y.NewCloser(1) + go db.flushMemtable(db.closers.memtable) // Need levels controller to be up. + + if err = db.vlog.Open(db, opt); err != nil { + return nil, err + } + + headKey := y.KeyWithTs(head, math.MaxUint64) + // Need to pass with timestamp, lsm get removes the last 8 bytes and compares key + vs, err := db.get(headKey) + if err != nil { + return nil, errors.Wrap(err, "Retrieving head") + } + db.orc.curRead = vs.Version + var vptr valuePointer + if len(vs.Value) > 0 { + vptr.Decode(vs.Value) + } + + // lastUsedCasCounter will either be the value stored in !badger!head, or some subsequently + // written value log entry that we replay. (Subsequent value log entries might be _less_ + // than lastUsedCasCounter, if there was value log gc so we have to max() values while + // replaying.) + // out.lastUsedCasCounter = item.casCounter + // TODO: Figure this out. This would update the read timestamp, and set nextCommitTs. + + replayCloser := y.NewCloser(1) + go db.doWrites(replayCloser) + + if err = db.vlog.Replay(vptr, replayFunction(db)); err != nil { + return db, err + } + + replayCloser.SignalAndWait() // Wait for replay to be applied first. + // Now that we have the curRead, we can update the nextCommit. + db.orc.nextCommit = db.orc.curRead + 1 + + // Mmap writable log + lf := db.vlog.filesMap[db.vlog.maxFid] + if err = lf.mmap(2 * db.vlog.opt.ValueLogFileSize); err != nil { + return db, errors.Wrapf(err, "Unable to mmap RDWR log file") + } + + db.writeCh = make(chan *request, kvWriteChCapacity) + db.closers.writes = y.NewCloser(1) + go db.doWrites(db.closers.writes) + + db.closers.valueGC = y.NewCloser(1) + go db.vlog.waitOnGC(db.closers.valueGC) + + valueDirLockGuard = nil + dirLockGuard = nil + manifestFile = nil + return db, nil +} + +// Close closes a DB. It's crucial to call it to ensure all the pending updates +// make their way to disk. +func (db *DB) Close() (err error) { + db.elog.Printf("Closing database") + // Stop value GC first. + db.closers.valueGC.SignalAndWait() + + // Stop writes next. + db.closers.writes.SignalAndWait() + + // Now close the value log. + if vlogErr := db.vlog.Close(); err == nil { + err = errors.Wrap(vlogErr, "DB.Close") + } + + // Make sure that block writer is done pushing stuff into memtable! + // Otherwise, you will have a race condition: we are trying to flush memtables + // and remove them completely, while the block / memtable writer is still + // trying to push stuff into the memtable. This will also resolve the value + // offset problem: as we push into memtable, we update value offsets there. + if !db.mt.Empty() { + db.elog.Printf("Flushing memtable") + for { + pushedFlushTask := func() bool { + db.Lock() + defer db.Unlock() + y.AssertTrue(db.mt != nil) + select { + case db.flushChan <- flushTask{db.mt, db.vptr}: + db.imm = append(db.imm, db.mt) // Flusher will attempt to remove this from s.imm. + db.mt = nil // Will segfault if we try writing! + db.elog.Printf("pushed to flush chan\n") + return true + default: + // If we fail to push, we need to unlock and wait for a short while. + // The flushing operation needs to update s.imm. Otherwise, we have a deadlock. + // TODO: Think about how to do this more cleanly, maybe without any locks. + } + return false + }() + if pushedFlushTask { + break + } + time.Sleep(10 * time.Millisecond) + } + } + db.flushChan <- flushTask{nil, valuePointer{}} // Tell flusher to quit. + + db.closers.memtable.Wait() + db.elog.Printf("Memtable flushed") + + db.closers.compactors.SignalAndWait() + db.elog.Printf("Compaction finished") + + if lcErr := db.lc.close(); err == nil { + err = errors.Wrap(lcErr, "DB.Close") + } + db.elog.Printf("Waiting for closer") + db.closers.updateSize.SignalAndWait() + + db.elog.Finish() + + if guardErr := db.dirLockGuard.release(); err == nil { + err = errors.Wrap(guardErr, "DB.Close") + } + if db.valueDirGuard != nil { + if guardErr := db.valueDirGuard.release(); err == nil { + err = errors.Wrap(guardErr, "DB.Close") + } + } + if manifestErr := db.manifest.close(); err == nil { + err = errors.Wrap(manifestErr, "DB.Close") + } + + // Fsync directories to ensure that lock file, and any other removed files whose directory + // we haven't specifically fsynced, are guaranteed to have their directory entry removal + // persisted to disk. + if syncErr := syncDir(db.opt.Dir); err == nil { + err = errors.Wrap(syncErr, "DB.Close") + } + if syncErr := syncDir(db.opt.ValueDir); err == nil { + err = errors.Wrap(syncErr, "DB.Close") + } + + return err +} + +const ( + lockFile = "LOCK" +) + +// When you create or delete a file, you have to ensure the directory entry for the file is synced +// in order to guarantee the file is visible (if the system crashes). (See the man page for fsync, +// or see https://github.com/coreos/etcd/issues/6368 for an example.) +func syncDir(dir string) error { + f, err := openDir(dir) + if err != nil { + return errors.Wrapf(err, "While opening directory: %s.", dir) + } + err = f.Sync() + closeErr := f.Close() + if err != nil { + return errors.Wrapf(err, "While syncing directory: %s.", dir) + } + return errors.Wrapf(closeErr, "While closing directory: %s.", dir) +} + +// getMemtables returns the current memtables and get references. +func (db *DB) getMemTables() ([]*skl.Skiplist, func()) { + db.RLock() + defer db.RUnlock() + + tables := make([]*skl.Skiplist, len(db.imm)+1) + + // Get mutable memtable. + tables[0] = db.mt + tables[0].IncrRef() + + // Get immutable memtables. + last := len(db.imm) - 1 + for i := range db.imm { + tables[i+1] = db.imm[last-i] + tables[i+1].IncrRef() + } + return tables, func() { + for _, tbl := range tables { + tbl.DecrRef() + } + } +} + +// get returns the value in memtable or disk for given key. +// Note that value will include meta byte. +func (db *DB) get(key []byte) (y.ValueStruct, error) { + tables, decr := db.getMemTables() // Lock should be released. + defer decr() + + y.NumGets.Add(1) + version := y.ParseTs(key) + var maxVs y.ValueStruct + // Need to search for values in all tables, with managed db + // latest value needn't be present in the latest table. + // Even without managed db, purging can cause this constraint + // to be violated. + // Search until required version is found or iterate over all + // tables and return max version. + for i := 0; i < len(tables); i++ { + vs := tables[i].Get(key) + y.NumMemtableGets.Add(1) + if vs.Meta == 0 && vs.Value == nil { + continue + } + if vs.Version == version { + return vs, nil + } + if maxVs.Version < vs.Version { + maxVs = vs + } + } + return db.lc.get(key, maxVs) +} + +func (db *DB) updateOffset(ptrs []valuePointer) { + var ptr valuePointer + for i := len(ptrs) - 1; i >= 0; i-- { + p := ptrs[i] + if !p.IsZero() { + ptr = p + break + } + } + if ptr.IsZero() { + return + } + + db.Lock() + defer db.Unlock() + y.AssertTrue(!ptr.Less(db.vptr)) + db.vptr = ptr +} + +var requestPool = sync.Pool{ + New: func() interface{} { + return new(request) + }, +} + +func (db *DB) shouldWriteValueToLSM(e Entry) bool { + return len(e.Value) < db.opt.ValueThreshold +} + +func (db *DB) writeToLSM(b *request) error { + if len(b.Ptrs) != len(b.Entries) { + return errors.Errorf("Ptrs and Entries don't match: %+v", b) + } + + for i, entry := range b.Entries { + if entry.meta&bitFinTxn != 0 { + continue + } + if db.shouldWriteValueToLSM(*entry) { // Will include deletion / tombstone case. + db.mt.Put(entry.Key, + y.ValueStruct{ + Value: entry.Value, + Meta: entry.meta, + UserMeta: entry.UserMeta, + ExpiresAt: entry.ExpiresAt, + }) + } else { + var offsetBuf [vptrSize]byte + db.mt.Put(entry.Key, + y.ValueStruct{ + Value: b.Ptrs[i].Encode(offsetBuf[:]), + Meta: entry.meta | bitValuePointer, + UserMeta: entry.UserMeta, + ExpiresAt: entry.ExpiresAt, + }) + } + } + return nil +} + +// writeRequests is called serially by only one goroutine. +func (db *DB) writeRequests(reqs []*request) error { + if len(reqs) == 0 { + return nil + } + + done := func(err error) { + for _, r := range reqs { + r.Err = err + r.Wg.Done() + } + } + + db.elog.Printf("writeRequests called. Writing to value log") + + err := db.vlog.write(reqs) + if err != nil { + done(err) + return err + } + + db.elog.Printf("Writing to memtable") + var count int + for _, b := range reqs { + if len(b.Entries) == 0 { + continue + } + count += len(b.Entries) + for err := db.ensureRoomForWrite(); err != nil; err = db.ensureRoomForWrite() { + db.elog.Printf("Making room for writes") + // We need to poll a bit because both hasRoomForWrite and the flusher need access to s.imm. + // When flushChan is full and you are blocked there, and the flusher is trying to update s.imm, + // you will get a deadlock. + time.Sleep(10 * time.Millisecond) + } + if err != nil { + done(err) + return errors.Wrap(err, "writeRequests") + } + if err := db.writeToLSM(b); err != nil { + done(err) + return errors.Wrap(err, "writeRequests") + } + db.updateOffset(b.Ptrs) + } + done(nil) + db.elog.Printf("%d entries written", count) + return nil +} + +func (db *DB) sendToWriteCh(entries []*Entry) (*request, error) { + var count, size int64 + for _, e := range entries { + size += int64(e.estimateSize(db.opt.ValueThreshold)) + count++ + } + if count >= db.opt.maxBatchCount || size >= db.opt.maxBatchSize { + return nil, ErrTxnTooBig + } + + // We can only service one request because we need each txn to be stored in a contigous section. + // Txns should not interleave among other txns or rewrites. + req := requestPool.Get().(*request) + req.Entries = entries + req.Wg = sync.WaitGroup{} + req.Wg.Add(1) + db.writeCh <- req // Handled in doWrites. + y.NumPuts.Add(int64(len(entries))) + + return req, nil +} + +func (db *DB) doWrites(lc *y.Closer) { + defer lc.Done() + pendingCh := make(chan struct{}, 1) + + writeRequests := func(reqs []*request) { + if err := db.writeRequests(reqs); err != nil { + log.Printf("ERROR in Badger::writeRequests: %v", err) + } + <-pendingCh + } + + // This variable tracks the number of pending writes. + reqLen := new(expvar.Int) + y.PendingWrites.Set(db.opt.Dir, reqLen) + + reqs := make([]*request, 0, 10) + for { + var r *request + select { + case r = <-db.writeCh: + case <-lc.HasBeenClosed(): + goto closedCase + } + + for { + reqs = append(reqs, r) + reqLen.Set(int64(len(reqs))) + + if len(reqs) >= 3*kvWriteChCapacity { + pendingCh <- struct{}{} // blocking. + goto writeCase + } + + select { + // Either push to pending, or continue to pick from writeCh. + case r = <-db.writeCh: + case pendingCh <- struct{}{}: + goto writeCase + case <-lc.HasBeenClosed(): + goto closedCase + } + } + + closedCase: + close(db.writeCh) + for r := range db.writeCh { // Flush the channel. + reqs = append(reqs, r) + } + + pendingCh <- struct{}{} // Push to pending before doing a write. + writeRequests(reqs) + return + + writeCase: + go writeRequests(reqs) + reqs = make([]*request, 0, 10) + reqLen.Set(0) + } +} + +// batchSet applies a list of badger.Entry. If a request level error occurs it +// will be returned. +// Check(kv.BatchSet(entries)) +func (db *DB) batchSet(entries []*Entry) error { + req, err := db.sendToWriteCh(entries) + if err != nil { + return err + } + + return req.Wait() +} + +// batchSetAsync is the asynchronous version of batchSet. It accepts a callback +// function which is called when all the sets are complete. If a request level +// error occurs, it will be passed back via the callback. +// err := kv.BatchSetAsync(entries, func(err error)) { +// Check(err) +// } +func (db *DB) batchSetAsync(entries []*Entry, f func(error)) error { + req, err := db.sendToWriteCh(entries) + if err != nil { + return err + } + go func() { + err := req.Wait() + // Write is complete. Let's call the callback function now. + f(err) + }() + return nil +} + +var errNoRoom = errors.New("No room for write") + +// ensureRoomForWrite is always called serially. +func (db *DB) ensureRoomForWrite() error { + var err error + db.Lock() + defer db.Unlock() + if db.mt.MemSize() < db.opt.MaxTableSize { + return nil + } + + y.AssertTrue(db.mt != nil) // A nil mt indicates that DB is being closed. + select { + case db.flushChan <- flushTask{db.mt, db.vptr}: + db.elog.Printf("Flushing value log to disk if async mode.") + // Ensure value log is synced to disk so this memtable's contents wouldn't be lost. + err = db.vlog.sync() + if err != nil { + return err + } + + db.elog.Printf("Flushing memtable, mt.size=%d size of flushChan: %d\n", + db.mt.MemSize(), len(db.flushChan)) + // We manage to push this task. Let's modify imm. + db.imm = append(db.imm, db.mt) + db.mt = skl.NewSkiplist(arenaSize(db.opt)) + // New memtable is empty. We certainly have room. + return nil + default: + // We need to do this to unlock and allow the flusher to modify imm. + return errNoRoom + } +} + +func arenaSize(opt Options) int64 { + return opt.MaxTableSize + opt.maxBatchSize + opt.maxBatchCount*int64(skl.MaxNodeSize) +} + +// WriteLevel0Table flushes memtable. It drops deleteValues. +func writeLevel0Table(s *skl.Skiplist, f *os.File) error { + iter := s.NewIterator() + defer iter.Close() + b := table.NewTableBuilder() + defer b.Close() + for iter.SeekToFirst(); iter.Valid(); iter.Next() { + if err := b.Add(iter.Key(), iter.Value()); err != nil { + return err + } + } + _, err := f.Write(b.Finish()) + return err +} + +type flushTask struct { + mt *skl.Skiplist + vptr valuePointer +} + +func (db *DB) flushMemtable(lc *y.Closer) error { + defer lc.Done() + + for ft := range db.flushChan { + if ft.mt == nil { + return nil + } + + if !ft.mt.Empty() { + // Store badger head even if vptr is zero, need it for readTs + db.elog.Printf("Storing offset: %+v\n", ft.vptr) + offset := make([]byte, vptrSize) + ft.vptr.Encode(offset) + + // Pick the max commit ts, so in case of crash, our read ts would be higher than all the + // commits. + headTs := y.KeyWithTs(head, db.orc.commitTs()) + ft.mt.Put(headTs, y.ValueStruct{Value: offset}) + } + + fileID := db.lc.reserveFileID() + fd, err := y.CreateSyncedFile(table.NewFilename(fileID, db.opt.Dir), true) + if err != nil { + return y.Wrap(err) + } + + // Don't block just to sync the directory entry. + dirSyncCh := make(chan error) + go func() { dirSyncCh <- syncDir(db.opt.Dir) }() + + err = writeLevel0Table(ft.mt, fd) + dirSyncErr := <-dirSyncCh + + if err != nil { + db.elog.Errorf("ERROR while writing to level 0: %v", err) + return err + } + if dirSyncErr != nil { + db.elog.Errorf("ERROR while syncing level directory: %v", dirSyncErr) + return err + } + + tbl, err := table.OpenTable(fd, db.opt.TableLoadingMode) + if err != nil { + db.elog.Printf("ERROR while opening table: %v", err) + return err + } + // We own a ref on tbl. + err = db.lc.addLevel0Table(tbl) // This will incrRef (if we don't error, sure) + tbl.DecrRef() // Releases our ref. + if err != nil { + return err + } + + // Update s.imm. Need a lock. + db.Lock() + y.AssertTrue(ft.mt == db.imm[0]) //For now, single threaded. + db.imm = db.imm[1:] + ft.mt.DecrRef() // Return memory. + db.Unlock() + } + return nil +} + +func exists(path string) (bool, error) { + _, err := os.Stat(path) + if err == nil { + return true, nil + } + if os.IsNotExist(err) { + return false, nil + } + return true, err +} + +// This function does a filewalk, calculates the size of vlog and sst files and stores it in +// y.LSMSize and y.VlogSize. +func (db *DB) calculateSize() { + newInt := func(val int64) *expvar.Int { + v := new(expvar.Int) + v.Add(val) + return v + } + + totalSize := func(dir string) (int64, int64) { + var lsmSize, vlogSize int64 + err := filepath.Walk(dir, func(path string, info os.FileInfo, err error) error { + if err != nil { + return err + } + ext := filepath.Ext(path) + if ext == ".sst" { + lsmSize += info.Size() + } else if ext == ".vlog" { + vlogSize += info.Size() + } + return nil + }) + if err != nil { + db.elog.Printf("Got error while calculating total size of directory: %s", dir) + } + return lsmSize, vlogSize + } + + lsmSize, vlogSize := totalSize(db.opt.Dir) + y.LSMSize.Set(db.opt.Dir, newInt(lsmSize)) + // If valueDir is different from dir, we'd have to do another walk. + if db.opt.ValueDir != db.opt.Dir { + _, vlogSize = totalSize(db.opt.ValueDir) + } + y.VlogSize.Set(db.opt.Dir, newInt(vlogSize)) + +} + +func (db *DB) updateSize(lc *y.Closer) { + defer lc.Done() + + metricsTicker := time.NewTicker(time.Minute) + defer metricsTicker.Stop() + + for { + select { + case <-metricsTicker.C: + db.calculateSize() + case <-lc.HasBeenClosed(): + return + } + } +} + +// PurgeVersionsBelow will delete all versions of a key below the specified version +func (db *DB) PurgeVersionsBelow(key []byte, ts uint64) error { + txn := db.NewTransaction(false) + defer txn.Discard() + return db.purgeVersionsBelow(txn, key, ts) +} + +func (db *DB) purgeVersionsBelow(txn *Txn, key []byte, ts uint64) error { + opts := DefaultIteratorOptions + opts.AllVersions = true + opts.PrefetchValues = false + it := txn.NewIterator(opts) + defer it.Close() + + var entries []*Entry + + for it.Seek(key); it.ValidForPrefix(key); it.Next() { + item := it.Item() + if !bytes.Equal(key, item.Key()) || item.Version() >= ts { + continue + } + if isDeletedOrExpired(item.meta, item.ExpiresAt()) { + continue + } + + // Found an older version. Mark for deletion + entries = append(entries, + &Entry{ + Key: y.KeyWithTs(key, item.version), + meta: bitDelete, + }) + db.vlog.updateGCStats(item) + } + return db.batchSet(entries) +} + +// PurgeOlderVersions deletes older versions of all keys. +// +// This function could be called prior to doing garbage collection to clean up +// older versions that are no longer needed. The caller must make sure that +// there are no long-running read transactions running before this function is +// called, otherwise they will not work as expected. +func (db *DB) PurgeOlderVersions() error { + return db.View(func(txn *Txn) error { + opts := DefaultIteratorOptions + opts.AllVersions = true + opts.PrefetchValues = false + it := txn.NewIterator(opts) + defer it.Close() + + var entries []*Entry + var lastKey []byte + var count, size int + var wg sync.WaitGroup + errChan := make(chan error, 1) + + // func to check for pending error before sending off a batch for writing + batchSetAsyncIfNoErr := func(entries []*Entry) error { + select { + case err := <-errChan: + return err + default: + wg.Add(1) + return txn.db.batchSetAsync(entries, func(err error) { + defer wg.Done() + if err != nil { + select { + case errChan <- err: + default: + } + } + }) + } + } + + for it.Rewind(); it.Valid(); it.Next() { + item := it.Item() + if !bytes.Equal(lastKey, item.Key()) { + lastKey = y.SafeCopy(lastKey, item.Key()) + continue + } + if isDeletedOrExpired(item.meta, item.ExpiresAt()) { + continue + } + // Found an older version. Mark for deletion + e := &Entry{ + Key: y.KeyWithTs(lastKey, item.version), + meta: bitDelete, + } + db.vlog.updateGCStats(item) + curSize := e.estimateSize(db.opt.ValueThreshold) + + // Batch up min(1000, maxBatchCount) entries at a time and write + // Ensure that total batch size doesn't exceed maxBatchSize + if count == 1000 || count+1 >= int(db.opt.maxBatchCount) || + size+curSize >= int(db.opt.maxBatchSize) { + if err := batchSetAsyncIfNoErr(entries); err != nil { + return err + } + count = 0 + size = 0 + entries = []*Entry{} + } + size += curSize + count++ + entries = append(entries, e) + } + + // Write last batch pending deletes + if count > 0 { + if err := batchSetAsyncIfNoErr(entries); err != nil { + return err + } + } + + wg.Wait() + + select { + case err := <-errChan: + return err + default: + return nil + } + }) +} + +// RunValueLogGC triggers a value log garbage collection. +// +// It picks value log files to perform GC based on statistics that are collected +// duing the session, when DB.PurgeOlderVersions() and DB.PurgeVersions() is +// called. If no such statistics are available, then log files are picked in +// random order. The process stops as soon as the first log file is encountered +// which does not result in garbage collection. +// +// When a log file is picked, it is first sampled If the sample shows that we +// can discard at least discardRatio space of that file, it would be rewritten. +// +// If a call to RunValueLogGC results in no rewrites, then an ErrNoRewrite is +// thrown indicating that the call resulted in no file rewrites. +// +// We recommend setting discardRatio to 0.5, thus indicating that a file be +// rewritten if half the space can be discarded. This results in a lifetime +// value log write amplification of 2 (1 from original write + 0.5 rewrite + +// 0.25 + 0.125 + ... = 2). Setting it to higher value would result in fewer +// space reclaims, while setting it to a lower value would result in more space +// reclaims at the cost of increased activity on the LSM tree. discardRatio +// must be in the range (0.0, 1.0), both endpoints excluded, otherwise an +// ErrInvalidRequest is returned. +// +// Only one GC is allowed at a time. If another value log GC is running, or DB +// has been closed, this would return an ErrRejected. +// +// Note: Every time GC is run, it would produce a spike of activity on the LSM +// tree. +func (db *DB) RunValueLogGC(discardRatio float64) error { + if discardRatio >= 1.0 || discardRatio <= 0.0 { + return ErrInvalidRequest + } + + // Find head on disk + headKey := y.KeyWithTs(head, math.MaxUint64) + // Need to pass with timestamp, lsm get removes the last 8 bytes and compares key + var maxVs y.ValueStruct + val, err := db.lc.get(headKey, maxVs) + if err != nil { + return errors.Wrap(err, "Retrieving head from on-disk LSM") + } + + var head valuePointer + if len(val.Value) > 0 { + head.Decode(val.Value) + } + + // Pick a log file and run GC + return db.vlog.runGC(discardRatio, head) +} + +// Size returns the size of lsm and value log files in bytes. It can be used to decide how often to +// call RunValueLogGC. +func (db *DB) Size() (lsm int64, vlog int64) { + if y.LSMSize.Get(db.opt.Dir) == nil { + lsm, vlog = 0, 0 + return + } + lsm = y.LSMSize.Get(db.opt.Dir).(*expvar.Int).Value() + vlog = y.VlogSize.Get(db.opt.Dir).(*expvar.Int).Value() + return +} + +// Sequence represents a Badger sequence. +type Sequence struct { + sync.Mutex + db *DB + key []byte + next uint64 + leased uint64 + bandwidth uint64 +} + +// Next would return the next integer in the sequence, updating the lease by running a transaction +// if needed. +func (seq *Sequence) Next() (uint64, error) { + seq.Lock() + defer seq.Unlock() + if seq.next >= seq.leased { + if err := seq.updateLease(); err != nil { + return 0, err + } + } + val := seq.next + seq.next++ + return val, nil +} + +// Release the leased sequence to avoid wasted integers. This should be done right +// before closing the associated DB. However it is valid to use the sequence after +// it was released, causing a new lease with full bandwidth. +func (seq *Sequence) Release() error { + seq.Lock() + defer seq.Unlock() + err := seq.db.Update(func(txn *Txn) error { + var buf [8]byte + binary.BigEndian.PutUint64(buf[:], seq.next) + return txn.Set(seq.key, buf[:]) + }) + if err != nil { + return err + } + seq.leased = seq.next + return nil +} + +func (seq *Sequence) updateLease() error { + return seq.db.Update(func(txn *Txn) error { + item, err := txn.Get(seq.key) + if err == ErrKeyNotFound { + seq.next = 0 + } else if err != nil { + return err + } else { + val, err := item.Value() + if err != nil { + return err + } + num := binary.BigEndian.Uint64(val) + seq.next = num + } + + lease := seq.next + seq.bandwidth + var buf [8]byte + binary.BigEndian.PutUint64(buf[:], lease) + if err = txn.Set(seq.key, buf[:]); err != nil { + return err + } + seq.leased = lease + return nil + }) +} + +// GetSequence would initiate a new sequence object, generating it from the stored lease, if +// available, in the database. Sequence can be used to get a list of monotonically increasing +// integers. Multiple sequences can be created by providing different keys. Bandwidth sets the +// size of the lease, determining how many Next() requests can be served from memory. +func (db *DB) GetSequence(key []byte, bandwidth uint64) (*Sequence, error) { + switch { + case len(key) == 0: + return nil, ErrEmptyKey + case bandwidth == 0: + return nil, ErrZeroBandwidth + } + seq := &Sequence{ + db: db, + key: key, + next: 0, + leased: 0, + bandwidth: bandwidth, + } + err := seq.updateLease() + return seq, err +} + +// MergeOperator represents a Badger merge operator. +type MergeOperator struct { + sync.RWMutex + f MergeFunc + db *DB + key []byte + skipAtOrBelow uint64 + closer *y.Closer +} + +// MergeFunc accepts two byte slices, one representing an existing value, and +// another representing a new value that needs to be ‘merged’ into it. MergeFunc +// contains the logic to perform the ‘merge’ and return an updated value. +// MergeFunc could perform operations like integer addition, list appends etc. +// Note that the ordering of the operands is unspecified, so the merge func +// should either be agnostic to ordering or do additional handling if ordering +// is required. +type MergeFunc func(existing, val []byte) []byte + +// GetMergeOperator creates a new MergeOperator for a given key and returns a +// pointer to it. It also fires off a goroutine that performs a compaction using +// the merge function that runs periodically, as specified by dur. +func (db *DB) GetMergeOperator(key []byte, + f MergeFunc, dur time.Duration) *MergeOperator { + op := &MergeOperator{ + f: f, + db: db, + key: key, + closer: y.NewCloser(1), + } + + go op.runCompactions(dur) + return op +} + +func (op *MergeOperator) iterateAndMerge(txn *Txn) (maxVersion uint64, val []byte, err error) { + opt := DefaultIteratorOptions + opt.AllVersions = true + it := txn.NewIterator(opt) + var first bool + for it.Rewind(); it.ValidForPrefix(op.key); it.Next() { + item := it.Item() + if item.Version() <= op.skipAtOrBelow { + continue + } + if item.Version() > maxVersion { + maxVersion = item.Version() + } + if !first { + first = true + val, err = item.ValueCopy(val) + if err != nil { + return 0, nil, err + } + } else { + newVal, err := item.Value() + if err != nil { + return 0, nil, err + } + val = op.f(val, newVal) + } + } + if !first { + return 0, nil, ErrKeyNotFound + } + return maxVersion, val, nil +} + +func (op *MergeOperator) compact() error { + op.Lock() + defer op.Unlock() + var maxVersion uint64 + err := op.db.Update(func(txn *Txn) error { + var ( + val []byte + err error + ) + maxVersion, val, err = op.iterateAndMerge(txn) + if err != nil { + return err + } + + // Write value back to db + if maxVersion > op.skipAtOrBelow { + if err := txn.Set(op.key, val); err != nil { + return err + } + } + return nil + }) + if err != nil && err != ErrKeyNotFound { // Ignore ErrKeyNotFound errors during compaction + return err + } + // Update version + op.skipAtOrBelow = maxVersion + return nil +} + +func (op *MergeOperator) runCompactions(dur time.Duration) { + ticker := time.NewTicker(dur) + defer op.closer.Done() + var stop bool + for { + select { + case <-op.closer.HasBeenClosed(): + stop = true + case <-ticker.C: // wait for tick + } + oldSkipVersion := op.skipAtOrBelow + if err := op.compact(); err != nil { + log.Printf("Error while running merge operation: %s", err) + } + // Purge older versions if version has updated + if op.skipAtOrBelow > oldSkipVersion { + if err := op.db.PurgeVersionsBelow(op.key, op.skipAtOrBelow+1); err != nil { + log.Printf("Error purging merged keys: %s", err) + } + } + if stop { + ticker.Stop() + break + } + } +} + +// Add records a value in Badger which will eventually be merged by a background +// routine into the values that were recorded by previous invocations to Add(). +func (op *MergeOperator) Add(val []byte) error { + return op.db.Update(func(txn *Txn) error { + return txn.Set(op.key, val) + }) +} + +// Get returns the latest value for the merge operator, which is derived by +// applying the merge function to all the values added so far. +// +// If Add has not been called even once, Get will return ErrKeyNotFound +func (op *MergeOperator) Get() ([]byte, error) { + op.RLock() + defer op.RUnlock() + var existing []byte + err := op.db.View(func(txn *Txn) (err error) { + _, existing, err = op.iterateAndMerge(txn) + return err + }) + return existing, err +} + +// Stop waits for any pending merge to complete and then stops the background +// goroutine. +func (op *MergeOperator) Stop() { + op.closer.SignalAndWait() +} diff --git a/vendor/github.com/dgraph-io/badger/dir_unix.go b/vendor/github.com/dgraph-io/badger/dir_unix.go new file mode 100644 index 0000000000..a058f45438 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/dir_unix.go @@ -0,0 +1,87 @@ +// +build !windows + +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "fmt" + "io/ioutil" + "os" + "path/filepath" + + "github.com/pkg/errors" + "golang.org/x/sys/unix" +) + +// directoryLockGuard holds a lock on a directory and a pid file inside. The pid file isn't part +// of the locking mechanism, it's just advisory. +type directoryLockGuard struct { + // File handle on the directory, which we've flocked. + f *os.File + // The absolute path to our pid file. + path string +} + +// acquireDirectoryLock gets an exclusive lock on the directory (using flock). It writes our pid +// to dirPath/pidFileName for convenience. +func acquireDirectoryLock(dirPath string, pidFileName string) (*directoryLockGuard, error) { + // Convert to absolute path so that Release still works even if we do an unbalanced + // chdir in the meantime. + absPidFilePath, err := filepath.Abs(filepath.Join(dirPath, pidFileName)) + if err != nil { + return nil, errors.Wrap(err, "cannot get absolute path for pid lock file") + } + f, err := os.Open(dirPath) + if err != nil { + return nil, errors.Wrapf(err, "cannot open directory %q", dirPath) + } + err = unix.Flock(int(f.Fd()), unix.LOCK_EX|unix.LOCK_NB) + if err != nil { + f.Close() + return nil, errors.Wrapf(err, + "Cannot acquire directory lock on %q. Another process is using this Badger database.", + dirPath) + } + + // Yes, we happily overwrite a pre-existing pid file. We're the only badger process using this + // directory. + err = ioutil.WriteFile(absPidFilePath, []byte(fmt.Sprintf("%d\n", os.Getpid())), 0666) + if err != nil { + f.Close() + return nil, errors.Wrapf(err, + "Cannot write pid file %q", absPidFilePath) + } + + return &directoryLockGuard{f, absPidFilePath}, nil +} + +// Release deletes the pid file and releases our lock on the directory. +func (guard *directoryLockGuard) release() error { + // It's important that we remove the pid file first. + err := os.Remove(guard.path) + if closeErr := guard.f.Close(); err == nil { + err = closeErr + } + guard.path = "" + guard.f = nil + + return err +} + +// openDir opens a directory for syncing. +func openDir(path string) (*os.File, error) { return os.Open(path) } diff --git a/vendor/github.com/dgraph-io/badger/dir_windows.go b/vendor/github.com/dgraph-io/badger/dir_windows.go new file mode 100644 index 0000000000..36d599fd03 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/dir_windows.go @@ -0,0 +1,90 @@ +// +build windows + +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +// OpenDir opens a directory in windows with write access for syncing. +import ( + "fmt" + "os" + "path/filepath" + "syscall" + + "github.com/pkg/errors" +) + +func openDir(path string) (*os.File, error) { + fd, err := openDirWin(path) + if err != nil { + return nil, err + } + return os.NewFile(uintptr(fd), path), nil +} + +func openDirWin(path string) (fd syscall.Handle, err error) { + if len(path) == 0 { + return syscall.InvalidHandle, syscall.ERROR_FILE_NOT_FOUND + } + pathp, err := syscall.UTF16PtrFromString(path) + if err != nil { + return syscall.InvalidHandle, err + } + access := uint32(syscall.GENERIC_READ | syscall.GENERIC_WRITE) + sharemode := uint32(syscall.FILE_SHARE_READ | syscall.FILE_SHARE_WRITE) + createmode := uint32(syscall.OPEN_EXISTING) + fl := uint32(syscall.FILE_FLAG_BACKUP_SEMANTICS) + return syscall.CreateFile(pathp, access, sharemode, nil, createmode, fl, 0) +} + +// DirectoryLockGuard holds a lock on the directory. +type directoryLockGuard struct { + path string +} + +// AcquireDirectoryLock acquires exclusive access to a directory. +func acquireDirectoryLock(dirPath string, pidFileName string) (*directoryLockGuard, error) { + // Convert to absolute path so that Release still works even if we do an unbalanced + // chdir in the meantime. + absLockFilePath, err := filepath.Abs(filepath.Join(dirPath, pidFileName)) + if err != nil { + return nil, errors.Wrap(err, "Cannot get absolute path for pid lock file") + } + + f, err := os.OpenFile(absLockFilePath, os.O_RDWR|os.O_CREATE|os.O_EXCL, 0666) + if err != nil { + return nil, errors.Wrapf(err, + "Cannot create pid lock file %q. Another process is using this Badger database", + absLockFilePath) + } + _, err = fmt.Fprintf(f, "%d\n", os.Getpid()) + closeErr := f.Close() + if err != nil { + return nil, errors.Wrap(err, "Cannot write to pid lock file") + } + if closeErr != nil { + return nil, errors.Wrap(closeErr, "Cannot close pid lock file") + } + return &directoryLockGuard{path: absLockFilePath}, nil +} + +// Release removes the directory lock. +func (g *directoryLockGuard) release() error { + path := g.path + g.path = "" + return os.Remove(path) +} diff --git a/vendor/github.com/dgraph-io/badger/doc.go b/vendor/github.com/dgraph-io/badger/doc.go new file mode 100644 index 0000000000..83dc9a28ac --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/doc.go @@ -0,0 +1,28 @@ +/* +Package badger implements an embeddable, simple and fast key-value database, +written in pure Go. It is designed to be highly performant for both reads and +writes simultaneously. Badger uses Multi-Version Concurrency Control (MVCC), and +supports transactions. It runs transactions concurrently, with serializable +snapshot isolation guarantees. + +Badger uses an LSM tree along with a value log to separate keys from values, +hence reducing both write amplification and the size of the LSM tree. This +allows LSM tree to be served entirely from RAM, while the values are served +from SSD. + + +Usage + +Badger has the following main types: DB, Txn, Item and Iterator. DB contains +keys that are associated with values. It must be opened with the appropriate +options before it can be accessed. + +All operations happen inside a Txn. Txn represents a transaction, which can +be read-only or read-write. Read-only transactions can read values for a +given key (which are returned inside an Item), or iterate over a set of +key-value pairs using an Iterator (which are returned as Item type values as +well). Read-write transactions can also update and delete keys from the DB. + +See the examples for more usage details. +*/ +package badger diff --git a/vendor/github.com/dgraph-io/badger/errors.go b/vendor/github.com/dgraph-io/badger/errors.go new file mode 100644 index 0000000000..ccf37b3fac --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/errors.go @@ -0,0 +1,97 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "encoding/hex" + + "github.com/pkg/errors" +) + +var ( + // ErrValueLogSize is returned when opt.ValueLogFileSize option is not within the valid + // range. + ErrValueLogSize = errors.New("Invalid ValueLogFileSize, must be between 1MB and 2GB") + + // ErrKeyNotFound is returned when key isn't found on a txn.Get. + ErrKeyNotFound = errors.New("Key not found") + + // ErrTxnTooBig is returned if too many writes are fit into a single transaction. + ErrTxnTooBig = errors.New("Txn is too big to fit into one request") + + // ErrConflict is returned when a transaction conflicts with another transaction. This can happen if + // the read rows had been updated concurrently by another transaction. + ErrConflict = errors.New("Transaction Conflict. Please retry") + + // ErrReadOnlyTxn is returned if an update function is called on a read-only transaction. + ErrReadOnlyTxn = errors.New("No sets or deletes are allowed in a read-only transaction") + + // ErrDiscardedTxn is returned if a previously discarded transaction is re-used. + ErrDiscardedTxn = errors.New("This transaction has been discarded. Create a new one") + + // ErrEmptyKey is returned if an empty key is passed on an update function. + ErrEmptyKey = errors.New("Key cannot be empty") + + // ErrRetry is returned when a log file containing the value is not found. + // This usually indicates that it may have been garbage collected, and the + // operation needs to be retried. + ErrRetry = errors.New("Unable to find log file. Please retry") + + // ErrThresholdZero is returned if threshold is set to zero, and value log GC is called. + // In such a case, GC can't be run. + ErrThresholdZero = errors.New( + "Value log GC can't run because threshold is set to zero") + + // ErrNoRewrite is returned if a call for value log GC doesn't result in a log file rewrite. + ErrNoRewrite = errors.New( + "Value log GC attempt didn't result in any cleanup") + + // ErrRejected is returned if a value log GC is called either while another GC is running, or + // after DB::Close has been called. + ErrRejected = errors.New("Value log GC request rejected") + + // ErrInvalidRequest is returned if the user request is invalid. + ErrInvalidRequest = errors.New("Invalid request") + + // ErrManagedTxn is returned if the user tries to use an API which isn't + // allowed due to external management of transactions, when using ManagedDB. + ErrManagedTxn = errors.New( + "Invalid API request. Not allowed to perform this action using ManagedDB") + + // ErrInvalidDump if a data dump made previously cannot be loaded into the database. + ErrInvalidDump = errors.New("Data dump cannot be read") + + // ErrZeroBandwidth is returned if the user passes in zero bandwidth for sequence. + ErrZeroBandwidth = errors.New("Bandwidth must be greater than zero") + + // ErrInvalidLoadingMode is returned when opt.ValueLogLoadingMode option is not + // within the valid range + ErrInvalidLoadingMode = errors.New("Invalid ValueLogLoadingMode, must be FileIO or MemoryMap") +) + +// Key length can't be more than uint16, as determined by table::header. +const maxKeySize = 1<<16 - 8 // 8 bytes are for storing timestamp + +func exceedsMaxKeySizeError(key []byte) error { + return errors.Errorf("Key with size %d exceeded %d limit. Key:\n%s", + len(key), maxKeySize, hex.Dump(key[:1<<10])) +} + +func exceedsMaxValueSizeError(value []byte, maxValueSize int64) error { + return errors.Errorf("Value with size %d exceeded ValueLogFileSize (%d). Key:\n%s", + len(value), maxValueSize, hex.Dump(value[:1<<10])) +} diff --git a/vendor/github.com/dgraph-io/badger/iterator.go b/vendor/github.com/dgraph-io/badger/iterator.go new file mode 100644 index 0000000000..c7ce634655 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/iterator.go @@ -0,0 +1,518 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "bytes" + "fmt" + "sync" + "time" + + "github.com/dgraph-io/badger/options" + + "github.com/dgraph-io/badger/y" + farm "github.com/dgryski/go-farm" +) + +type prefetchStatus uint8 + +const ( + prefetched prefetchStatus = iota + 1 +) + +// Item is returned during iteration. Both the Key() and Value() output is only valid until +// iterator.Next() is called. +type Item struct { + status prefetchStatus + err error + wg sync.WaitGroup + db *DB + key []byte + vptr []byte + meta byte // We need to store meta to know about bitValuePointer. + userMeta byte + expiresAt uint64 + val []byte + slice *y.Slice // Used only during prefetching. + next *Item + version uint64 + txn *Txn +} + +// ToString returns a string representation of Item +func (item *Item) ToString() string { + return fmt.Sprintf("key=%q, version=%d, meta=%x", item.Key(), item.Version(), item.meta) + +} + +// Key returns the key. +// +// Key is only valid as long as item is valid, or transaction is valid. If you need to use it +// outside its validity, please copy it. +func (item *Item) Key() []byte { + return item.key +} + +// Version returns the commit timestamp of the item. +func (item *Item) Version() uint64 { + return item.version +} + +// Value retrieves the value of the item from the value log. +// +// This method must be called within a transaction. Calling it outside a +// transaction is considered undefined behavior. If an iterator is being used, +// then Item.Value() is defined in the current iteration only, because items are +// reused. +// +// If you need to use a value outside a transaction, please use Item.ValueCopy +// instead, or copy it yourself. +func (item *Item) Value() ([]byte, error) { + item.wg.Wait() + if item.status == prefetched { + return item.val, item.err + } + buf, cb, err := item.yieldItemValue() + if cb != nil { + item.txn.callbacks = append(item.txn.callbacks, cb) + } + return buf, err +} + +// ValueCopy returns a copy of the value of the item from the value log, writing it to dst slice. +// If nil is passed, or capacity of dst isn't sufficient, a new slice would be allocated and +// returned. Tip: It might make sense to reuse the returned slice as dst argument for the next call. +// +// This function is useful in long running iterate/update transactions to avoid a write deadlock. +// See Github issue: https://github.com/dgraph-io/badger/issues/315 +func (item *Item) ValueCopy(dst []byte) ([]byte, error) { + item.wg.Wait() + if item.status == prefetched { + return y.SafeCopy(dst, item.val), item.err + } + buf, cb, err := item.yieldItemValue() + defer runCallback(cb) + return y.SafeCopy(dst, buf), err +} + +func (item *Item) hasValue() bool { + if item.meta == 0 && item.vptr == nil { + // key not found + return false + } + return true +} + +func (item *Item) yieldItemValue() ([]byte, func(), error) { + if !item.hasValue() { + return nil, nil, nil + } + + if item.slice == nil { + item.slice = new(y.Slice) + } + + if (item.meta & bitValuePointer) == 0 { + val := item.slice.Resize(len(item.vptr)) + copy(val, item.vptr) + return val, nil, nil + } + + var vp valuePointer + vp.Decode(item.vptr) + return item.db.vlog.Read(vp, item.slice) +} + +func runCallback(cb func()) { + if cb != nil { + cb() + } +} + +func (item *Item) prefetchValue() { + val, cb, err := item.yieldItemValue() + defer runCallback(cb) + + item.err = err + item.status = prefetched + if val == nil { + return + } + if item.db.opt.ValueLogLoadingMode == options.MemoryMap { + buf := item.slice.Resize(len(val)) + copy(buf, val) + item.val = buf + } else { + item.val = val + } +} + +// EstimatedSize returns approximate size of the key-value pair. +// +// This can be called while iterating through a store to quickly estimate the +// size of a range of key-value pairs (without fetching the corresponding +// values). +func (item *Item) EstimatedSize() int64 { + if !item.hasValue() { + return 0 + } + if (item.meta & bitValuePointer) == 0 { + return int64(len(item.key) + len(item.vptr)) + } + var vp valuePointer + vp.Decode(item.vptr) + return int64(vp.Len) // includes key length. +} + +// UserMeta returns the userMeta set by the user. Typically, this byte, optionally set by the user +// is used to interpret the value. +func (item *Item) UserMeta() byte { + return item.userMeta +} + +// ExpiresAt returns a Unix time value indicating when the item will be +// considered expired. 0 indicates that the item will never expire. +func (item *Item) ExpiresAt() uint64 { + return item.expiresAt +} + +// TODO: Switch this to use linked list container in Go. +type list struct { + head *Item + tail *Item +} + +func (l *list) push(i *Item) { + i.next = nil + if l.tail == nil { + l.head = i + l.tail = i + return + } + l.tail.next = i + l.tail = i +} + +func (l *list) pop() *Item { + if l.head == nil { + return nil + } + i := l.head + if l.head == l.tail { + l.tail = nil + l.head = nil + } else { + l.head = i.next + } + i.next = nil + return i +} + +// IteratorOptions is used to set options when iterating over Badger key-value +// stores. +// +// This package provides DefaultIteratorOptions which contains options that +// should work for most applications. Consider using that as a starting point +// before customizing it for your own needs. +type IteratorOptions struct { + // Indicates whether we should prefetch values during iteration and store them. + PrefetchValues bool + // How many KV pairs to prefetch while iterating. Valid only if PrefetchValues is true. + PrefetchSize int + Reverse bool // Direction of iteration. False is forward, true is backward. + AllVersions bool // Fetch all valid versions of the same key. +} + +// DefaultIteratorOptions contains default options when iterating over Badger key-value stores. +var DefaultIteratorOptions = IteratorOptions{ + PrefetchValues: true, + PrefetchSize: 100, + Reverse: false, + AllVersions: false, +} + +// Iterator helps iterating over the KV pairs in a lexicographically sorted order. +type Iterator struct { + iitr *y.MergeIterator + txn *Txn + readTs uint64 + + opt IteratorOptions + item *Item + data list + waste list + + lastKey []byte // Used to skip over multiple versions of the same key. +} + +// NewIterator returns a new iterator. Depending upon the options, either only keys, or both +// key-value pairs would be fetched. The keys are returned in lexicographically sorted order. +// Using prefetch is highly recommended if you're doing a long running iteration. +// Avoid long running iterations in update transactions. +func (txn *Txn) NewIterator(opt IteratorOptions) *Iterator { + tables, decr := txn.db.getMemTables() + defer decr() + txn.db.vlog.incrIteratorCount() + var iters []y.Iterator + if itr := txn.newPendingWritesIterator(opt.Reverse); itr != nil { + iters = append(iters, itr) + } + for i := 0; i < len(tables); i++ { + iters = append(iters, tables[i].NewUniIterator(opt.Reverse)) + } + iters = txn.db.lc.appendIterators(iters, opt.Reverse) // This will increment references. + res := &Iterator{ + txn: txn, + iitr: y.NewMergeIterator(iters, opt.Reverse), + opt: opt, + readTs: txn.readTs, + } + return res +} + +func (it *Iterator) newItem() *Item { + item := it.waste.pop() + if item == nil { + item = &Item{slice: new(y.Slice), db: it.txn.db, txn: it.txn} + } + return item +} + +// Item returns pointer to the current key-value pair. +// This item is only valid until it.Next() gets called. +func (it *Iterator) Item() *Item { + tx := it.txn + if tx.update { + // Track reads if this is an update txn. + tx.reads = append(tx.reads, farm.Fingerprint64(it.item.Key())) + } + return it.item +} + +// Valid returns false when iteration is done. +func (it *Iterator) Valid() bool { return it.item != nil } + +// ValidForPrefix returns false when iteration is done +// or when the current key is not prefixed by the specified prefix. +func (it *Iterator) ValidForPrefix(prefix []byte) bool { + return it.item != nil && bytes.HasPrefix(it.item.key, prefix) +} + +// Close would close the iterator. It is important to call this when you're done with iteration. +func (it *Iterator) Close() { + it.iitr.Close() + // TODO: We could handle this error. + _ = it.txn.db.vlog.decrIteratorCount() +} + +// Next would advance the iterator by one. Always check it.Valid() after a Next() +// to ensure you have access to a valid it.Item(). +func (it *Iterator) Next() { + // Reuse current item + it.item.wg.Wait() // Just cleaner to wait before pushing to avoid doing ref counting. + it.waste.push(it.item) + + // Set next item to current + it.item = it.data.pop() + + for it.iitr.Valid() { + if it.parseItem() { + // parseItem calls one extra next. + // This is used to deal with the complexity of reverse iteration. + break + } + } +} + +func isDeletedOrExpired(meta byte, expiresAt uint64) bool { + if meta&bitDelete > 0 { + return true + } + if expiresAt == 0 { + return false + } + return expiresAt <= uint64(time.Now().Unix()) +} + +// parseItem is a complex function because it needs to handle both forward and reverse iteration +// implementation. We store keys such that their versions are sorted in descending order. This makes +// forward iteration efficient, but revese iteration complicated. This tradeoff is better because +// forward iteration is more common than reverse. +// +// This function advances the iterator. +func (it *Iterator) parseItem() bool { + mi := it.iitr + key := mi.Key() + + setItem := func(item *Item) { + if it.item == nil { + it.item = item + } else { + it.data.push(item) + } + } + + // Skip badger keys. + if bytes.HasPrefix(key, badgerPrefix) { + mi.Next() + return false + } + + // Skip any versions which are beyond the readTs. + version := y.ParseTs(key) + if version > it.readTs { + mi.Next() + return false + } + + if it.opt.AllVersions { + // Return deleted or expired values also, otherwise user can't figure out + // whether the key was deleted. + item := it.newItem() + it.fill(item) + setItem(item) + mi.Next() + return true + } + + // If iterating in forward direction, then just checking the last key against current key would + // be sufficient. + if !it.opt.Reverse { + if y.SameKey(it.lastKey, key) { + mi.Next() + return false + } + // Only track in forward direction. + // We should update lastKey as soon as we find a different key in our snapshot. + // Consider keys: a 5, b 7 (del), b 5. When iterating, lastKey = a. + // Then we see b 7, which is deleted. If we don't store lastKey = b, we'll then return b 5, + // which is wrong. Therefore, update lastKey here. + it.lastKey = y.SafeCopy(it.lastKey, mi.Key()) + } + +FILL: + // If deleted, advance and return. + vs := mi.Value() + if isDeletedOrExpired(vs.Meta, vs.ExpiresAt) { + mi.Next() + return false + } + + item := it.newItem() + it.fill(item) + // fill item based on current cursor position. All Next calls have returned, so reaching here + // means no Next was called. + + mi.Next() // Advance but no fill item yet. + if !it.opt.Reverse || !mi.Valid() { // Forward direction, or invalid. + setItem(item) + return true + } + + // Reverse direction. + nextTs := y.ParseTs(mi.Key()) + mik := y.ParseKey(mi.Key()) + if nextTs <= it.readTs && bytes.Equal(mik, item.key) { + // This is a valid potential candidate. + goto FILL + } + // Ignore the next candidate. Return the current one. + setItem(item) + return true +} + +func (it *Iterator) fill(item *Item) { + vs := it.iitr.Value() + item.meta = vs.Meta + item.userMeta = vs.UserMeta + item.expiresAt = vs.ExpiresAt + + item.version = y.ParseTs(it.iitr.Key()) + item.key = y.SafeCopy(item.key, y.ParseKey(it.iitr.Key())) + + item.vptr = y.SafeCopy(item.vptr, vs.Value) + item.val = nil + if it.opt.PrefetchValues { + item.wg.Add(1) + go func() { + // FIXME we are not handling errors here. + item.prefetchValue() + item.wg.Done() + }() + } +} + +func (it *Iterator) prefetch() { + prefetchSize := 2 + if it.opt.PrefetchValues && it.opt.PrefetchSize > 1 { + prefetchSize = it.opt.PrefetchSize + } + + i := it.iitr + var count int + it.item = nil + for i.Valid() { + if !it.parseItem() { + continue + } + count++ + if count == prefetchSize { + break + } + } +} + +// Seek would seek to the provided key if present. If absent, it would seek to the next smallest key +// greater than provided if iterating in the forward direction. Behavior would be reversed is +// iterating backwards. +func (it *Iterator) Seek(key []byte) { + for i := it.data.pop(); i != nil; i = it.data.pop() { + i.wg.Wait() + it.waste.push(i) + } + + it.lastKey = it.lastKey[:0] + if len(key) == 0 { + it.iitr.Rewind() + it.prefetch() + return + } + + if !it.opt.Reverse { + key = y.KeyWithTs(key, it.txn.readTs) + } else { + key = y.KeyWithTs(key, 0) + } + it.iitr.Seek(key) + it.prefetch() +} + +// Rewind would rewind the iterator cursor all the way to zero-th position, which would be the +// smallest key if iterating forward, and largest if iterating backward. It does not keep track of +// whether the cursor started with a Seek(). +func (it *Iterator) Rewind() { + i := it.data.pop() + for i != nil { + i.wg.Wait() // Just cleaner to wait before pushing. No ref counting needed. + it.waste.push(i) + i = it.data.pop() + } + + it.lastKey = it.lastKey[:0] + it.iitr.Rewind() + it.prefetch() +} diff --git a/vendor/github.com/dgraph-io/badger/level_handler.go b/vendor/github.com/dgraph-io/badger/level_handler.go new file mode 100644 index 0000000000..d7295c4acc --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/level_handler.go @@ -0,0 +1,294 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "fmt" + "sort" + "sync" + + "github.com/dgraph-io/badger/table" + "github.com/dgraph-io/badger/y" + "github.com/pkg/errors" +) + +type levelHandler struct { + // Guards tables, totalSize. + sync.RWMutex + + // For level >= 1, tables are sorted by key ranges, which do not overlap. + // For level 0, tables are sorted by time. + // For level 0, newest table are at the back. Compact the oldest one first, which is at the front. + tables []*table.Table + totalSize int64 + + // The following are initialized once and const. + level int + strLevel string + maxTotalSize int64 + db *DB +} + +func (s *levelHandler) getTotalSize() int64 { + s.RLock() + defer s.RUnlock() + return s.totalSize +} + +// initTables replaces s.tables with given tables. This is done during loading. +func (s *levelHandler) initTables(tables []*table.Table) { + s.Lock() + defer s.Unlock() + + s.tables = tables + s.totalSize = 0 + for _, t := range tables { + s.totalSize += t.Size() + } + + if s.level == 0 { + // Key range will overlap. Just sort by fileID in ascending order + // because newer tables are at the end of level 0. + sort.Slice(s.tables, func(i, j int) bool { + return s.tables[i].ID() < s.tables[j].ID() + }) + } else { + // Sort tables by keys. + sort.Slice(s.tables, func(i, j int) bool { + return y.CompareKeys(s.tables[i].Smallest(), s.tables[j].Smallest()) < 0 + }) + } +} + +// deleteTables remove tables idx0, ..., idx1-1. +func (s *levelHandler) deleteTables(toDel []*table.Table) error { + s.Lock() // s.Unlock() below + + toDelMap := make(map[uint64]struct{}) + for _, t := range toDel { + toDelMap[t.ID()] = struct{}{} + } + + // Make a copy as iterators might be keeping a slice of tables. + var newTables []*table.Table + for _, t := range s.tables { + _, found := toDelMap[t.ID()] + if !found { + newTables = append(newTables, t) + continue + } + s.totalSize -= t.Size() + } + s.tables = newTables + + s.Unlock() // Unlock s _before_ we DecrRef our tables, which can be slow. + + return decrRefs(toDel) +} + +// replaceTables will replace tables[left:right] with newTables. Note this EXCLUDES tables[right]. +// You must call decr() to delete the old tables _after_ writing the update to the manifest. +func (s *levelHandler) replaceTables(newTables []*table.Table) error { + // Need to re-search the range of tables in this level to be replaced as other goroutines might + // be changing it as well. (They can't touch our tables, but if they add/remove other tables, + // the indices get shifted around.) + if len(newTables) == 0 { + return nil + } + + s.Lock() // We s.Unlock() below. + + // Increase totalSize first. + for _, tbl := range newTables { + s.totalSize += tbl.Size() + tbl.IncrRef() + } + + kr := keyRange{ + left: newTables[0].Smallest(), + right: newTables[len(newTables)-1].Biggest(), + } + left, right := s.overlappingTables(levelHandlerRLocked{}, kr) + + toDecr := make([]*table.Table, right-left) + // Update totalSize and reference counts. + for i := left; i < right; i++ { + tbl := s.tables[i] + s.totalSize -= tbl.Size() + toDecr[i-left] = tbl + } + + // To be safe, just make a copy. TODO: Be more careful and avoid copying. + numDeleted := right - left + numAdded := len(newTables) + tables := make([]*table.Table, len(s.tables)-numDeleted+numAdded) + y.AssertTrue(left == copy(tables, s.tables[:left])) + t := tables[left:] + y.AssertTrue(numAdded == copy(t, newTables)) + t = t[numAdded:] + y.AssertTrue(len(s.tables[right:]) == copy(t, s.tables[right:])) + s.tables = tables + s.Unlock() // s.Unlock before we DecrRef tables -- that can be slow. + return decrRefs(toDecr) +} + +func decrRefs(tables []*table.Table) error { + for _, table := range tables { + if err := table.DecrRef(); err != nil { + return err + } + } + return nil +} + +func newLevelHandler(db *DB, level int) *levelHandler { + return &levelHandler{ + level: level, + strLevel: fmt.Sprintf("l%d", level), + db: db, + } +} + +// tryAddLevel0Table returns true if ok and no stalling. +func (s *levelHandler) tryAddLevel0Table(t *table.Table) bool { + y.AssertTrue(s.level == 0) + // Need lock as we may be deleting the first table during a level 0 compaction. + s.Lock() + defer s.Unlock() + if len(s.tables) >= s.db.opt.NumLevelZeroTablesStall { + return false + } + + s.tables = append(s.tables, t) + t.IncrRef() + s.totalSize += t.Size() + + return true +} + +func (s *levelHandler) numTables() int { + s.RLock() + defer s.RUnlock() + return len(s.tables) +} + +func (s *levelHandler) close() error { + s.RLock() + defer s.RUnlock() + var err error + for _, t := range s.tables { + if closeErr := t.Close(); closeErr != nil && err == nil { + err = closeErr + } + } + return errors.Wrap(err, "levelHandler.close") +} + +// getTableForKey acquires a read-lock to access s.tables. It returns a list of tableHandlers. +func (s *levelHandler) getTableForKey(key []byte) ([]*table.Table, func() error) { + s.RLock() + defer s.RUnlock() + + if s.level == 0 { + // For level 0, we need to check every table. Remember to make a copy as s.tables may change + // once we exit this function, and we don't want to lock s.tables while seeking in tables. + // CAUTION: Reverse the tables. + out := make([]*table.Table, 0, len(s.tables)) + for i := len(s.tables) - 1; i >= 0; i-- { + out = append(out, s.tables[i]) + s.tables[i].IncrRef() + } + return out, func() error { + for _, t := range out { + if err := t.DecrRef(); err != nil { + return err + } + } + return nil + } + } + // For level >= 1, we can do a binary search as key range does not overlap. + idx := sort.Search(len(s.tables), func(i int) bool { + return y.CompareKeys(s.tables[i].Biggest(), key) >= 0 + }) + if idx >= len(s.tables) { + // Given key is strictly > than every element we have. + return nil, func() error { return nil } + } + tbl := s.tables[idx] + tbl.IncrRef() + return []*table.Table{tbl}, tbl.DecrRef +} + +// get returns value for a given key or the key after that. If not found, return nil. +func (s *levelHandler) get(key []byte) (y.ValueStruct, error) { + tables, decr := s.getTableForKey(key) + keyNoTs := y.ParseKey(key) + + var maxVs y.ValueStruct + for _, th := range tables { + if th.DoesNotHave(keyNoTs) { + y.NumLSMBloomHits.Add(s.strLevel, 1) + continue + } + + it := th.NewIterator(false) + defer it.Close() + + y.NumLSMGets.Add(s.strLevel, 1) + it.Seek(key) + if !it.Valid() { + continue + } + if y.SameKey(key, it.Key()) { + if version := y.ParseTs(it.Key()); maxVs.Version < version { + maxVs = it.Value() + maxVs.Version = version + } + } + } + return maxVs, decr() +} + +// appendIterators appends iterators to an array of iterators, for merging. +// Note: This obtains references for the table handlers. Remember to close these iterators. +func (s *levelHandler) appendIterators(iters []y.Iterator, reversed bool) []y.Iterator { + s.RLock() + defer s.RUnlock() + + if s.level == 0 { + // Remember to add in reverse order! + // The newer table at the end of s.tables should be added first as it takes precedence. + return appendIteratorsReversed(iters, s.tables, reversed) + } + return append(iters, table.NewConcatIterator(s.tables, reversed)) +} + +type levelHandlerRLocked struct{} + +// overlappingTables returns the tables that intersect with key range. Returns a half-interval. +// This function should already have acquired a read lock, and this is so important the caller must +// pass an empty parameter declaring such. +func (s *levelHandler) overlappingTables(_ levelHandlerRLocked, kr keyRange) (int, int) { + left := sort.Search(len(s.tables), func(i int) bool { + return y.CompareKeys(kr.left, s.tables[i].Biggest()) <= 0 + }) + right := sort.Search(len(s.tables), func(i int) bool { + return y.CompareKeys(kr.right, s.tables[i].Smallest()) < 0 + }) + return left, right +} diff --git a/vendor/github.com/dgraph-io/badger/levels.go b/vendor/github.com/dgraph-io/badger/levels.go new file mode 100644 index 0000000000..55002b4fa6 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/levels.go @@ -0,0 +1,698 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "fmt" + "math/rand" + "os" + "sort" + "time" + + "golang.org/x/net/trace" + + "github.com/dgraph-io/badger/protos" + "github.com/dgraph-io/badger/table" + "github.com/dgraph-io/badger/y" + "github.com/pkg/errors" +) + +type levelsController struct { + nextFileID uint64 // Atomic + elog trace.EventLog + + // The following are initialized once and const. + levels []*levelHandler + kv *DB + + cstatus compactStatus +} + +var ( + // This is for getting timings between stalls. + lastUnstalled time.Time +) + +// revertToManifest checks that all necessary table files exist and removes all table files not +// referenced by the manifest. idMap is a set of table file id's that were read from the directory +// listing. +func revertToManifest(kv *DB, mf *Manifest, idMap map[uint64]struct{}) error { + // 1. Check all files in manifest exist. + for id := range mf.Tables { + if _, ok := idMap[id]; !ok { + return fmt.Errorf("file does not exist for table %d", id) + } + } + + // 2. Delete files that shouldn't exist. + for id := range idMap { + if _, ok := mf.Tables[id]; !ok { + kv.elog.Printf("Table file %d not referenced in MANIFEST\n", id) + filename := table.NewFilename(id, kv.opt.Dir) + if err := os.Remove(filename); err != nil { + return y.Wrapf(err, "While removing table %d", id) + } + } + } + + return nil +} + +func newLevelsController(kv *DB, mf *Manifest) (*levelsController, error) { + y.AssertTrue(kv.opt.NumLevelZeroTablesStall > kv.opt.NumLevelZeroTables) + s := &levelsController{ + kv: kv, + elog: kv.elog, + levels: make([]*levelHandler, kv.opt.MaxLevels), + } + s.cstatus.levels = make([]*levelCompactStatus, kv.opt.MaxLevels) + + for i := 0; i < kv.opt.MaxLevels; i++ { + s.levels[i] = newLevelHandler(kv, i) + if i == 0 { + // Do nothing. + } else if i == 1 { + // Level 1 probably shouldn't be too much bigger than level 0. + s.levels[i].maxTotalSize = kv.opt.LevelOneSize + } else { + s.levels[i].maxTotalSize = s.levels[i-1].maxTotalSize * int64(kv.opt.LevelSizeMultiplier) + } + s.cstatus.levels[i] = new(levelCompactStatus) + } + + // Compare manifest against directory, check for existent/non-existent files, and remove. + if err := revertToManifest(kv, mf, getIDMap(kv.opt.Dir)); err != nil { + return nil, err + } + + // Some files may be deleted. Let's reload. + tables := make([][]*table.Table, kv.opt.MaxLevels) + var maxFileID uint64 + for fileID, tableManifest := range mf.Tables { + fname := table.NewFilename(fileID, kv.opt.Dir) + fd, err := y.OpenExistingSyncedFile(fname, true) + if err != nil { + closeAllTables(tables) + return nil, errors.Wrapf(err, "Opening file: %q", fname) + } + + t, err := table.OpenTable(fd, kv.opt.TableLoadingMode) + if err != nil { + closeAllTables(tables) + return nil, errors.Wrapf(err, "Opening table: %q", fname) + } + + level := tableManifest.Level + tables[level] = append(tables[level], t) + + if fileID > maxFileID { + maxFileID = fileID + } + } + s.nextFileID = maxFileID + 1 + for i, tbls := range tables { + s.levels[i].initTables(tbls) + } + + // Make sure key ranges do not overlap etc. + if err := s.validate(); err != nil { + _ = s.cleanupLevels() + return nil, errors.Wrap(err, "Level validation") + } + + // Sync directory (because we have at least removed some files, or previously created the + // manifest file). + if err := syncDir(kv.opt.Dir); err != nil { + _ = s.close() + return nil, err + } + + return s, nil +} + +// Closes the tables, for cleanup in newLevelsController. (We Close() instead of using DecrRef() +// because that would delete the underlying files.) We ignore errors, which is OK because tables +// are read-only. +func closeAllTables(tables [][]*table.Table) { + for _, tableSlice := range tables { + for _, table := range tableSlice { + _ = table.Close() + } + } +} + +func (s *levelsController) cleanupLevels() error { + var firstErr error + for _, l := range s.levels { + if err := l.close(); err != nil && firstErr == nil { + firstErr = err + } + } + return firstErr +} + +func (s *levelsController) startCompact(lc *y.Closer) { + n := s.kv.opt.NumCompactors + lc.AddRunning(n - 1) + for i := 0; i < n; i++ { + go s.runWorker(lc) + } +} + +func (s *levelsController) runWorker(lc *y.Closer) { + defer lc.Done() + if s.kv.opt.DoNotCompact { + return + } + + time.Sleep(time.Duration(rand.Int31n(1000)) * time.Millisecond) + ticker := time.NewTicker(time.Second) + defer ticker.Stop() + + for { + select { + // Can add a done channel or other stuff. + case <-ticker.C: + prios := s.pickCompactLevels() + for _, p := range prios { + // TODO: Handle error. + didCompact, _ := s.doCompact(p) + if didCompact { + break + } + } + case <-lc.HasBeenClosed(): + return + } + } +} + +// Returns true if level zero may be compacted, without accounting for compactions that already +// might be happening. +func (s *levelsController) isLevel0Compactable() bool { + return s.levels[0].numTables() >= s.kv.opt.NumLevelZeroTables +} + +// Returns true if the non-zero level may be compacted. delSize provides the size of the tables +// which are currently being compacted so that we treat them as already having started being +// compacted (because they have been, yet their size is already counted in getTotalSize). +func (l *levelHandler) isCompactable(delSize int64) bool { + return l.getTotalSize()-delSize >= l.maxTotalSize +} + +type compactionPriority struct { + level int + score float64 +} + +// pickCompactLevel determines which level to compact. +// Based on: https://github.com/facebook/rocksdb/wiki/Leveled-Compaction +func (s *levelsController) pickCompactLevels() (prios []compactionPriority) { + // This function must use identical criteria for guaranteeing compaction's progress that + // addLevel0Table uses. + + // cstatus is checked to see if level 0's tables are already being compacted + if !s.cstatus.overlapsWith(0, infRange) && s.isLevel0Compactable() { + pri := compactionPriority{ + level: 0, + score: float64(s.levels[0].numTables()) / float64(s.kv.opt.NumLevelZeroTables), + } + prios = append(prios, pri) + } + + for i, l := range s.levels[1:] { + // Don't consider those tables that are already being compacted right now. + delSize := s.cstatus.delSize(i + 1) + + if l.isCompactable(delSize) { + pri := compactionPriority{ + level: i + 1, + score: float64(l.getTotalSize()-delSize) / float64(l.maxTotalSize), + } + prios = append(prios, pri) + } + } + sort.Slice(prios, func(i, j int) bool { + return prios[i].score > prios[j].score + }) + return prios +} + +// compactBuildTables merge topTables and botTables to form a list of new tables. +func (s *levelsController) compactBuildTables( + l int, cd compactDef) ([]*table.Table, func() error, error) { + topTables := cd.top + botTables := cd.bot + + // Create iterators across all the tables involved first. + var iters []y.Iterator + if l == 0 { + iters = appendIteratorsReversed(iters, topTables, false) + } else { + y.AssertTrue(len(topTables) == 1) + iters = []y.Iterator{topTables[0].NewIterator(false)} + } + + // Next level has level>=1 and we can use ConcatIterator as key ranges do not overlap. + iters = append(iters, table.NewConcatIterator(botTables, false)) + it := y.NewMergeIterator(iters, false) + defer it.Close() // Important to close the iterator to do ref counting. + + it.Rewind() + + // Start generating new tables. + type newTableResult struct { + table *table.Table + err error + } + resultCh := make(chan newTableResult) + var i int + for ; it.Valid(); i++ { + timeStart := time.Now() + builder := table.NewTableBuilder() + for ; it.Valid(); it.Next() { + if builder.ReachedCapacity(s.kv.opt.MaxTableSize) { + break + } + y.Check(builder.Add(it.Key(), it.Value())) + } + // It was true that it.Valid() at least once in the loop above, which means we + // called Add() at least once, and builder is not Empty(). + y.AssertTrue(!builder.Empty()) + + cd.elog.LazyPrintf("LOG Compact. Iteration to generate one table took: %v\n", time.Since(timeStart)) + + fileID := s.reserveFileID() + go func(builder *table.Builder) { + defer builder.Close() + + fd, err := y.CreateSyncedFile(table.NewFilename(fileID, s.kv.opt.Dir), true) + if err != nil { + resultCh <- newTableResult{nil, errors.Wrapf(err, "While opening new table: %d", fileID)} + return + } + + if _, err := fd.Write(builder.Finish()); err != nil { + resultCh <- newTableResult{nil, errors.Wrapf(err, "Unable to write to file: %d", fileID)} + return + } + + tbl, err := table.OpenTable(fd, s.kv.opt.TableLoadingMode) + // decrRef is added below. + resultCh <- newTableResult{tbl, errors.Wrapf(err, "Unable to open table: %q", fd.Name())} + }(builder) + } + + newTables := make([]*table.Table, 0, 20) + + // Wait for all table builders to finish. + var firstErr error + for x := 0; x < i; x++ { + res := <-resultCh + newTables = append(newTables, res.table) + if firstErr == nil { + firstErr = res.err + } + } + + if firstErr == nil { + // Ensure created files' directory entries are visible. We don't mind the extra latency + // from not doing this ASAP after all file creation has finished because this is a + // background operation. + firstErr = syncDir(s.kv.opt.Dir) + } + + if firstErr != nil { + // An error happened. Delete all the newly created table files (by calling DecrRef + // -- we're the only holders of a ref). + for j := 0; j < i; j++ { + if newTables[j] != nil { + newTables[j].DecrRef() + } + } + errorReturn := errors.Wrapf(firstErr, "While running compaction for: %+v", cd) + return nil, nil, errorReturn + } + + sort.Slice(newTables, func(i, j int) bool { + return y.CompareKeys(newTables[i].Biggest(), newTables[j].Biggest()) < 0 + }) + + return newTables, func() error { return decrRefs(newTables) }, nil +} + +func buildChangeSet(cd *compactDef, newTables []*table.Table) protos.ManifestChangeSet { + changes := []*protos.ManifestChange{} + for _, table := range newTables { + changes = append(changes, makeTableCreateChange(table.ID(), cd.nextLevel.level)) + } + for _, table := range cd.top { + changes = append(changes, makeTableDeleteChange(table.ID())) + } + for _, table := range cd.bot { + changes = append(changes, makeTableDeleteChange(table.ID())) + } + return protos.ManifestChangeSet{Changes: changes} +} + +type compactDef struct { + elog trace.Trace + + thisLevel *levelHandler + nextLevel *levelHandler + + top []*table.Table + bot []*table.Table + + thisRange keyRange + nextRange keyRange + + thisSize int64 +} + +func (cd *compactDef) lockLevels() { + cd.thisLevel.RLock() + cd.nextLevel.RLock() +} + +func (cd *compactDef) unlockLevels() { + cd.nextLevel.RUnlock() + cd.thisLevel.RUnlock() +} + +func (s *levelsController) fillTablesL0(cd *compactDef) bool { + cd.lockLevels() + defer cd.unlockLevels() + + cd.top = make([]*table.Table, len(cd.thisLevel.tables)) + copy(cd.top, cd.thisLevel.tables) + if len(cd.top) == 0 { + return false + } + cd.thisRange = infRange + + kr := getKeyRange(cd.top) + left, right := cd.nextLevel.overlappingTables(levelHandlerRLocked{}, kr) + cd.bot = make([]*table.Table, right-left) + copy(cd.bot, cd.nextLevel.tables[left:right]) + + if len(cd.bot) == 0 { + cd.nextRange = kr + } else { + cd.nextRange = getKeyRange(cd.bot) + } + + if !s.cstatus.compareAndAdd(thisAndNextLevelRLocked{}, *cd) { + return false + } + + return true +} + +func (s *levelsController) fillTables(cd *compactDef) bool { + cd.lockLevels() + defer cd.unlockLevels() + + tbls := make([]*table.Table, len(cd.thisLevel.tables)) + copy(tbls, cd.thisLevel.tables) + if len(tbls) == 0 { + return false + } + + // Find the biggest table, and compact that first. + // TODO: Try other table picking strategies. + sort.Slice(tbls, func(i, j int) bool { + return tbls[i].Size() > tbls[j].Size() + }) + + for _, t := range tbls { + cd.thisSize = t.Size() + cd.thisRange = keyRange{ + left: t.Smallest(), + right: t.Biggest(), + } + if s.cstatus.overlapsWith(cd.thisLevel.level, cd.thisRange) { + continue + } + cd.top = []*table.Table{t} + left, right := cd.nextLevel.overlappingTables(levelHandlerRLocked{}, cd.thisRange) + + cd.bot = make([]*table.Table, right-left) + copy(cd.bot, cd.nextLevel.tables[left:right]) + + if len(cd.bot) == 0 { + cd.bot = []*table.Table{} + cd.nextRange = cd.thisRange + if !s.cstatus.compareAndAdd(thisAndNextLevelRLocked{}, *cd) { + continue + } + return true + } + cd.nextRange = getKeyRange(cd.bot) + + if s.cstatus.overlapsWith(cd.nextLevel.level, cd.nextRange) { + continue + } + + if !s.cstatus.compareAndAdd(thisAndNextLevelRLocked{}, *cd) { + continue + } + return true + } + return false +} + +func (s *levelsController) runCompactDef(l int, cd compactDef) (err error) { + timeStart := time.Now() + + thisLevel := cd.thisLevel + nextLevel := cd.nextLevel + + if thisLevel.level >= 1 && len(cd.bot) == 0 { + y.AssertTrue(len(cd.top) == 1) + tbl := cd.top[0] + + // We write to the manifest _before_ we delete files (and after we created files). + changes := []*protos.ManifestChange{ + // The order matters here -- you can't temporarily have two copies of the same + // table id when reloading the manifest. + makeTableDeleteChange(tbl.ID()), + makeTableCreateChange(tbl.ID(), nextLevel.level), + } + if err := s.kv.manifest.addChanges(changes); err != nil { + return err + } + + // We have to add to nextLevel before we remove from thisLevel, not after. This way, we + // don't have a bug where reads would see keys missing from both levels. + + // Note: It's critical that we add tables (replace them) in nextLevel before deleting them + // in thisLevel. (We could finagle it atomically somehow.) Also, when reading we must + // read, or at least acquire s.RLock(), in increasing order by level, so that we don't skip + // a compaction. + + if err := nextLevel.replaceTables(cd.top); err != nil { + return err + } + if err := thisLevel.deleteTables(cd.top); err != nil { + return err + } + + cd.elog.LazyPrintf("\tLOG Compact-Move %d->%d smallest:%s biggest:%s took %v\n", + l, l+1, string(tbl.Smallest()), string(tbl.Biggest()), time.Since(timeStart)) + return nil + } + + newTables, decr, err := s.compactBuildTables(l, cd) + if err != nil { + return err + } + defer func() { + // Only assign to err, if it's not already nil. + if decErr := decr(); err == nil { + err = decErr + } + }() + changeSet := buildChangeSet(&cd, newTables) + + // We write to the manifest _before_ we delete files (and after we created files) + if err := s.kv.manifest.addChanges(changeSet.Changes); err != nil { + return err + } + + // See comment earlier in this function about the ordering of these ops, and the order in which + // we access levels when reading. + if err := nextLevel.replaceTables(newTables); err != nil { + return err + } + if err := thisLevel.deleteTables(cd.top); err != nil { + return err + } + + // Note: For level 0, while doCompact is running, it is possible that new tables are added. + // However, the tables are added only to the end, so it is ok to just delete the first table. + + cd.elog.LazyPrintf("LOG Compact %d->%d, del %d tables, add %d tables, took %v\n", + l, l+1, len(cd.top)+len(cd.bot), len(newTables), time.Since(timeStart)) + return nil +} + +// doCompact picks some table on level l and compacts it away to the next level. +func (s *levelsController) doCompact(p compactionPriority) (bool, error) { + l := p.level + y.AssertTrue(l+1 < s.kv.opt.MaxLevels) // Sanity check. + + cd := compactDef{ + elog: trace.New("Badger", "Compact"), + thisLevel: s.levels[l], + nextLevel: s.levels[l+1], + } + cd.elog.SetMaxEvents(100) + defer cd.elog.Finish() + + cd.elog.LazyPrintf("Got compaction priority: %+v", p) + + // While picking tables to be compacted, both levels' tables are expected to + // remain unchanged. + if l == 0 { + if !s.fillTablesL0(&cd) { + cd.elog.LazyPrintf("fillTables failed for level: %d\n", l) + return false, nil + } + + } else { + if !s.fillTables(&cd) { + cd.elog.LazyPrintf("fillTables failed for level: %d\n", l) + return false, nil + } + } + + cd.elog.LazyPrintf("Running for level: %d\n", cd.thisLevel.level) + s.cstatus.toLog(cd.elog) + if err := s.runCompactDef(l, cd); err != nil { + // This compaction couldn't be done successfully. + cd.elog.LazyPrintf("\tLOG Compact FAILED with error: %+v: %+v", err, cd) + return false, err + } + + // Done with compaction. So, remove the ranges from compaction status. + s.cstatus.delete(cd) + s.cstatus.toLog(cd.elog) + cd.elog.LazyPrintf("Compaction for level: %d DONE", cd.thisLevel.level) + return true, nil +} + +func (s *levelsController) addLevel0Table(t *table.Table) error { + // We update the manifest _before_ the table becomes part of a levelHandler, because at that + // point it could get used in some compaction. This ensures the manifest file gets updated in + // the proper order. (That means this update happens before that of some compaction which + // deletes the table.) + err := s.kv.manifest.addChanges([]*protos.ManifestChange{ + makeTableCreateChange(t.ID(), 0), + }) + if err != nil { + return err + } + + for !s.levels[0].tryAddLevel0Table(t) { + // Stall. Make sure all levels are healthy before we unstall. + var timeStart time.Time + { + s.elog.Printf("STALLED STALLED STALLED STALLED STALLED STALLED STALLED STALLED: %v\n", + time.Since(lastUnstalled)) + s.cstatus.RLock() + for i := 0; i < s.kv.opt.MaxLevels; i++ { + s.elog.Printf("level=%d. Status=%s Size=%d\n", + i, s.cstatus.levels[i].debug(), s.levels[i].getTotalSize()) + } + s.cstatus.RUnlock() + timeStart = time.Now() + } + // Before we unstall, we need to make sure that level 0 and 1 are healthy. Otherwise, we + // will very quickly fill up level 0 again and if the compaction strategy favors level 0, + // then level 1 is going to super full. + for { + // Passing 0 for delSize to compactable means we're treating incomplete compactions as + // not having finished -- we wait for them to finish. Also, it's crucial this behavior + // replicates pickCompactLevels' behavior in computing compactability in order to + // guarantee progress. + if !s.isLevel0Compactable() && !s.levels[1].isCompactable(0) { + break + } + time.Sleep(10 * time.Millisecond) + } + { + s.elog.Printf("UNSTALLED UNSTALLED UNSTALLED UNSTALLED UNSTALLED UNSTALLED: %v\n", + time.Since(timeStart)) + lastUnstalled = time.Now() + } + } + + return nil +} + +func (s *levelsController) close() error { + err := s.cleanupLevels() + return errors.Wrap(err, "levelsController.Close") +} + +// get returns the found value if any. If not found, we return nil. +func (s *levelsController) get(key []byte, maxVs y.ValueStruct) (y.ValueStruct, error) { + // It's important that we iterate the levels from 0 on upward. The reason is, if we iterated + // in opposite order, or in parallel (naively calling all the h.RLock() in some order) we could + // read level L's tables post-compaction and level L+1's tables pre-compaction. (If we do + // parallelize this, we will need to call the h.RLock() function by increasing order of level + // number.) + + version := y.ParseTs(key) + for _, h := range s.levels { + vs, err := h.get(key) // Calls h.RLock() and h.RUnlock(). + if err != nil { + return y.ValueStruct{}, errors.Wrapf(err, "get key: %q", key) + } + if vs.Value == nil && vs.Meta == 0 { + continue + } + if vs.Version == version { + return vs, nil + } + if maxVs.Version < vs.Version { + maxVs = vs + } + } + return maxVs, nil +} + +func appendIteratorsReversed(out []y.Iterator, th []*table.Table, reversed bool) []y.Iterator { + for i := len(th) - 1; i >= 0; i-- { + // This will increment the reference of the table handler. + out = append(out, th[i].NewIterator(reversed)) + } + return out +} + +// appendIterators appends iterators to an array of iterators, for merging. +// Note: This obtains references for the table handlers. Remember to close these iterators. +func (s *levelsController) appendIterators( + iters []y.Iterator, reversed bool) []y.Iterator { + // Just like with get, it's important we iterate the levels from 0 on upward, to avoid missing + // data when there's a compaction. + for _, level := range s.levels { + iters = level.appendIterators(iters, reversed) + } + return iters +} diff --git a/vendor/github.com/dgraph-io/badger/managed_db.go b/vendor/github.com/dgraph-io/badger/managed_db.go new file mode 100644 index 0000000000..3b52c1e784 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/managed_db.go @@ -0,0 +1,86 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +// ManagedDB allows end users to manage the transactions themselves. Transaction +// start and commit timestamps are set by end-user. +// +// This is only useful for databases built on top of Badger (like Dgraph), and +// can be ignored by most users. +// +// WARNING: This is an experimental feature and may be changed significantly in +// a future release. So please proceed with caution. +type ManagedDB struct { + *DB +} + +// OpenManaged returns a new ManagedDB, which allows more control over setting +// transaction timestamps. +// +// This is only useful for databases built on top of Badger (like Dgraph), and +// can be ignored by most users. +func OpenManaged(opts Options) (*ManagedDB, error) { + opts.managedTxns = true + db, err := Open(opts) + if err != nil { + return nil, err + } + return &ManagedDB{db}, nil +} + +// NewTransaction overrides DB.NewTransaction() and panics when invoked. Use +// NewTransactionAt() instead. +func (db *ManagedDB) NewTransaction(update bool) { + panic("Cannot use NewTransaction() for ManagedDB. Use NewTransactionAt() instead.") +} + +// NewTransactionAt follows the same logic as DB.NewTransaction(), but uses the +// provided read timestamp. +// +// This is only useful for databases built on top of Badger (like Dgraph), and +// can be ignored by most users. +func (db *ManagedDB) NewTransactionAt(readTs uint64, update bool) *Txn { + txn := db.DB.NewTransaction(update) + txn.readTs = readTs + return txn +} + +// CommitAt commits the transaction, following the same logic as Commit(), but +// at the given commit timestamp. This will panic if not used with ManagedDB. +// +// This is only useful for databases built on top of Badger (like Dgraph), and +// can be ignored by most users. +func (txn *Txn) CommitAt(commitTs uint64, callback func(error)) error { + if !txn.db.opt.managedTxns { + return ErrManagedTxn + } + txn.commitTs = commitTs + return txn.Commit(callback) +} + +// PurgeVersionsBelow will delete all versions of a key below the specified version +func (db *ManagedDB) PurgeVersionsBelow(key []byte, ts uint64) error { + txn := db.NewTransactionAt(ts, false) + defer txn.Discard() + return db.purgeVersionsBelow(txn, key, ts) +} + +// GetSequence is not supported on ManagedDB. Calling this would result +// in a panic. +func (db *ManagedDB) GetSequence(_ []byte, _ uint64) (*Sequence, error) { + panic("Cannot use GetSequence for ManagedDB.") +} diff --git a/vendor/github.com/dgraph-io/badger/manifest.go b/vendor/github.com/dgraph-io/badger/manifest.go new file mode 100644 index 0000000000..f2e57d9b06 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/manifest.go @@ -0,0 +1,425 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "bufio" + "bytes" + "encoding/binary" + "fmt" + "hash/crc32" + "io" + "os" + "path/filepath" + "sync" + + "github.com/dgraph-io/badger/protos" + "github.com/dgraph-io/badger/y" + "github.com/pkg/errors" +) + +// Manifest represnts the contents of the MANIFEST file in a Badger store. +// +// The MANIFEST file describes the startup state of the db -- all LSM files and what level they're +// at. +// +// It consists of a sequence of ManifestChangeSet objects. Each of these is treated atomically, +// and contains a sequence of ManifestChange's (file creations/deletions) which we use to +// reconstruct the manifest at startup. +type Manifest struct { + Levels []levelManifest + Tables map[uint64]tableManifest + + // Contains total number of creation and deletion changes in the manifest -- used to compute + // whether it'd be useful to rewrite the manifest. + Creations int + Deletions int +} + +func createManifest() Manifest { + levels := make([]levelManifest, 0) + return Manifest{ + Levels: levels, + Tables: make(map[uint64]tableManifest), + } +} + +// levelManifest contains information about LSM tree levels +// in the MANIFEST file. +type levelManifest struct { + Tables map[uint64]struct{} // Set of table id's +} + +// tableManifest contains information about a specific level +// in the LSM tree. +type tableManifest struct { + Level uint8 +} + +// manifestFile holds the file pointer (and other info) about the manifest file, which is a log +// file we append to. +type manifestFile struct { + fp *os.File + directory string + // We make this configurable so that unit tests can hit rewrite() code quickly + deletionsRewriteThreshold int + + // Guards appends, which includes access to the manifest field. + appendLock sync.Mutex + + // Used to track the current state of the manifest, used when rewriting. + manifest Manifest +} + +const ( + // ManifestFilename is the filename for the manifest file. + ManifestFilename = "MANIFEST" + manifestRewriteFilename = "MANIFEST-REWRITE" + manifestDeletionsRewriteThreshold = 10000 + manifestDeletionsRatio = 10 +) + +// asChanges returns a sequence of changes that could be used to recreate the Manifest in its +// present state. +func (m *Manifest) asChanges() []*protos.ManifestChange { + changes := make([]*protos.ManifestChange, 0, len(m.Tables)) + for id, tm := range m.Tables { + changes = append(changes, makeTableCreateChange(id, int(tm.Level))) + } + return changes +} + +func (m *Manifest) clone() Manifest { + changeSet := protos.ManifestChangeSet{Changes: m.asChanges()} + ret := createManifest() + y.Check(applyChangeSet(&ret, &changeSet)) + return ret +} + +// openOrCreateManifestFile opens a Badger manifest file if it exists, or creates on if +// one doesn’t. +func openOrCreateManifestFile(dir string) (ret *manifestFile, result Manifest, err error) { + return helpOpenOrCreateManifestFile(dir, manifestDeletionsRewriteThreshold) +} + +func helpOpenOrCreateManifestFile(dir string, deletionsThreshold int) (ret *manifestFile, result Manifest, err error) { + path := filepath.Join(dir, ManifestFilename) + fp, err := y.OpenExistingSyncedFile(path, false) // We explicitly sync in addChanges, outside the lock. + if err != nil { + if !os.IsNotExist(err) { + return nil, Manifest{}, err + } + m := createManifest() + fp, netCreations, err := helpRewrite(dir, &m) + if err != nil { + return nil, Manifest{}, err + } + y.AssertTrue(netCreations == 0) + mf := &manifestFile{ + fp: fp, + directory: dir, + manifest: m.clone(), + deletionsRewriteThreshold: deletionsThreshold, + } + return mf, m, nil + } + + manifest, truncOffset, err := ReplayManifestFile(fp) + if err != nil { + _ = fp.Close() + return nil, Manifest{}, err + } + + // Truncate file so we don't have a half-written entry at the end. + if err := fp.Truncate(truncOffset); err != nil { + _ = fp.Close() + return nil, Manifest{}, err + } + + if _, err = fp.Seek(0, io.SeekEnd); err != nil { + _ = fp.Close() + return nil, Manifest{}, err + } + + mf := &manifestFile{ + fp: fp, + directory: dir, + manifest: manifest.clone(), + deletionsRewriteThreshold: deletionsThreshold, + } + return mf, manifest, nil +} + +func (mf *manifestFile) close() error { + return mf.fp.Close() +} + +// addChanges writes a batch of changes, atomically, to the file. By "atomically" that means when +// we replay the MANIFEST file, we'll either replay all the changes or none of them. (The truth of +// this depends on the filesystem -- some might append garbage data if a system crash happens at +// the wrong time.) +func (mf *manifestFile) addChanges(changesParam []*protos.ManifestChange) error { + changes := protos.ManifestChangeSet{Changes: changesParam} + buf, err := changes.Marshal() + if err != nil { + return err + } + + // Maybe we could use O_APPEND instead (on certain file systems) + mf.appendLock.Lock() + if err := applyChangeSet(&mf.manifest, &changes); err != nil { + mf.appendLock.Unlock() + return err + } + // Rewrite manifest if it'd shrink by 1/10 and it's big enough to care + if mf.manifest.Deletions > mf.deletionsRewriteThreshold && + mf.manifest.Deletions > manifestDeletionsRatio*(mf.manifest.Creations-mf.manifest.Deletions) { + if err := mf.rewrite(); err != nil { + mf.appendLock.Unlock() + return err + } + } else { + var lenCrcBuf [8]byte + binary.BigEndian.PutUint32(lenCrcBuf[0:4], uint32(len(buf))) + binary.BigEndian.PutUint32(lenCrcBuf[4:8], crc32.Checksum(buf, y.CastagnoliCrcTable)) + buf = append(lenCrcBuf[:], buf...) + if _, err := mf.fp.Write(buf); err != nil { + mf.appendLock.Unlock() + return err + } + } + + mf.appendLock.Unlock() + return mf.fp.Sync() +} + +// Has to be 4 bytes. The value can never change, ever, anyway. +var magicText = [4]byte{'B', 'd', 'g', 'r'} + +// The magic version number. +const magicVersion = 4 + +func helpRewrite(dir string, m *Manifest) (*os.File, int, error) { + rewritePath := filepath.Join(dir, manifestRewriteFilename) + // We explicitly sync. + fp, err := y.OpenTruncFile(rewritePath, false) + if err != nil { + return nil, 0, err + } + + buf := make([]byte, 8) + copy(buf[0:4], magicText[:]) + binary.BigEndian.PutUint32(buf[4:8], magicVersion) + + netCreations := len(m.Tables) + changes := m.asChanges() + set := protos.ManifestChangeSet{Changes: changes} + + changeBuf, err := set.Marshal() + if err != nil { + fp.Close() + return nil, 0, err + } + var lenCrcBuf [8]byte + binary.BigEndian.PutUint32(lenCrcBuf[0:4], uint32(len(changeBuf))) + binary.BigEndian.PutUint32(lenCrcBuf[4:8], crc32.Checksum(changeBuf, y.CastagnoliCrcTable)) + buf = append(buf, lenCrcBuf[:]...) + buf = append(buf, changeBuf...) + if _, err := fp.Write(buf); err != nil { + fp.Close() + return nil, 0, err + } + if err := fp.Sync(); err != nil { + fp.Close() + return nil, 0, err + } + + // In Windows the files should be closed before doing a Rename. + if err = fp.Close(); err != nil { + return nil, 0, err + } + manifestPath := filepath.Join(dir, ManifestFilename) + if err := os.Rename(rewritePath, manifestPath); err != nil { + return nil, 0, err + } + fp, err = y.OpenExistingSyncedFile(manifestPath, false) + if err != nil { + return nil, 0, err + } + if _, err := fp.Seek(0, io.SeekEnd); err != nil { + fp.Close() + return nil, 0, err + } + if err := syncDir(dir); err != nil { + fp.Close() + return nil, 0, err + } + + return fp, netCreations, nil +} + +// Must be called while appendLock is held. +func (mf *manifestFile) rewrite() error { + // In Windows the files should be closed before doing a Rename. + if err := mf.fp.Close(); err != nil { + return err + } + fp, netCreations, err := helpRewrite(mf.directory, &mf.manifest) + if err != nil { + return err + } + mf.fp = fp + mf.manifest.Creations = netCreations + mf.manifest.Deletions = 0 + + return nil +} + +type countingReader struct { + wrapped *bufio.Reader + count int64 +} + +func (r *countingReader) Read(p []byte) (n int, err error) { + n, err = r.wrapped.Read(p) + r.count += int64(n) + return +} + +func (r *countingReader) ReadByte() (b byte, err error) { + b, err = r.wrapped.ReadByte() + if err == nil { + r.count++ + } + return +} + +var ( + errBadMagic = errors.New("manifest has bad magic") +) + +// ReplayManifestFile reads the manifest file and constructs two manifest objects. (We need one +// immutable copy and one mutable copy of the manifest. Easiest way is to construct two of them.) +// Also, returns the last offset after a completely read manifest entry -- the file must be +// truncated at that point before further appends are made (if there is a partial entry after +// that). In normal conditions, truncOffset is the file size. +func ReplayManifestFile(fp *os.File) (ret Manifest, truncOffset int64, err error) { + r := countingReader{wrapped: bufio.NewReader(fp)} + + var magicBuf [8]byte + if _, err := io.ReadFull(&r, magicBuf[:]); err != nil { + return Manifest{}, 0, errBadMagic + } + if !bytes.Equal(magicBuf[0:4], magicText[:]) { + return Manifest{}, 0, errBadMagic + } + version := binary.BigEndian.Uint32(magicBuf[4:8]) + if version != magicVersion { + return Manifest{}, 0, + fmt.Errorf("manifest has unsupported version: %d (we support %d)", version, magicVersion) + } + + build := createManifest() + var offset int64 + for { + offset = r.count + var lenCrcBuf [8]byte + _, err := io.ReadFull(&r, lenCrcBuf[:]) + if err != nil { + if err == io.EOF || err == io.ErrUnexpectedEOF { + break + } + return Manifest{}, 0, err + } + length := binary.BigEndian.Uint32(lenCrcBuf[0:4]) + var buf = make([]byte, length) + if _, err := io.ReadFull(&r, buf); err != nil { + if err == io.EOF || err == io.ErrUnexpectedEOF { + break + } + return Manifest{}, 0, err + } + if crc32.Checksum(buf, y.CastagnoliCrcTable) != binary.BigEndian.Uint32(lenCrcBuf[4:8]) { + break + } + + var changeSet protos.ManifestChangeSet + if err := changeSet.Unmarshal(buf); err != nil { + return Manifest{}, 0, err + } + + if err := applyChangeSet(&build, &changeSet); err != nil { + return Manifest{}, 0, err + } + } + + return build, offset, err +} + +func applyManifestChange(build *Manifest, tc *protos.ManifestChange) error { + switch tc.Op { + case protos.ManifestChange_CREATE: + if _, ok := build.Tables[tc.Id]; ok { + return fmt.Errorf("MANIFEST invalid, table %d exists", tc.Id) + } + build.Tables[tc.Id] = tableManifest{ + Level: uint8(tc.Level), + } + for len(build.Levels) <= int(tc.Level) { + build.Levels = append(build.Levels, levelManifest{make(map[uint64]struct{})}) + } + build.Levels[tc.Level].Tables[tc.Id] = struct{}{} + build.Creations++ + case protos.ManifestChange_DELETE: + tm, ok := build.Tables[tc.Id] + if !ok { + return fmt.Errorf("MANIFEST removes non-existing table %d", tc.Id) + } + delete(build.Levels[tm.Level].Tables, tc.Id) + delete(build.Tables, tc.Id) + build.Deletions++ + default: + return fmt.Errorf("MANIFEST file has invalid manifestChange op") + } + return nil +} + +// This is not a "recoverable" error -- opening the KV store fails because the MANIFEST file is +// just plain broken. +func applyChangeSet(build *Manifest, changeSet *protos.ManifestChangeSet) error { + for _, change := range changeSet.Changes { + if err := applyManifestChange(build, change); err != nil { + return err + } + } + return nil +} + +func makeTableCreateChange(id uint64, level int) *protos.ManifestChange { + return &protos.ManifestChange{ + Id: id, + Op: protos.ManifestChange_CREATE, + Level: uint32(level), + } +} + +func makeTableDeleteChange(id uint64) *protos.ManifestChange { + return &protos.ManifestChange{ + Id: id, + Op: protos.ManifestChange_DELETE, + } +} diff --git a/vendor/github.com/dgraph-io/badger/options.go b/vendor/github.com/dgraph-io/badger/options.go new file mode 100644 index 0000000000..9b8bc6ee99 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/options.go @@ -0,0 +1,112 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "github.com/dgraph-io/badger/options" +) + +// NOTE: Keep the comments in the following to 75 chars width, so they +// format nicely in godoc. + +// Options are params for creating DB object. +// +// This package provides DefaultOptions which contains options that should +// work for most applications. Consider using that as a starting point before +// customizing it for your own needs. +type Options struct { + // 1. Mandatory flags + // ------------------- + // Directory to store the data in. Should exist and be writable. + Dir string + // Directory to store the value log in. Can be the same as Dir. Should + // exist and be writable. + ValueDir string + + // 2. Frequently modified flags + // ----------------------------- + // Sync all writes to disk. Setting this to true would slow down data + // loading significantly. + SyncWrites bool + + // How should LSM tree be accessed. + TableLoadingMode options.FileLoadingMode + + // How should value log be accessed + ValueLogLoadingMode options.FileLoadingMode + + // 3. Flags that user might want to review + // ---------------------------------------- + // The following affect all levels of LSM tree. + MaxTableSize int64 // Each table (or file) is at most this size. + LevelSizeMultiplier int // Equals SizeOf(Li+1)/SizeOf(Li). + MaxLevels int // Maximum number of levels of compaction. + // If value size >= this threshold, only store value offsets in tree. + ValueThreshold int + // Maximum number of tables to keep in memory, before stalling. + NumMemtables int + // The following affect how we handle LSM tree L0. + // Maximum number of Level 0 tables before we start compacting. + NumLevelZeroTables int + + // If we hit this number of Level 0 tables, we will stall until L0 is + // compacted away. + NumLevelZeroTablesStall int + + // Maximum total size for L1. + LevelOneSize int64 + + // Size of single value log file. + ValueLogFileSize int64 + + // Number of compaction workers to run concurrently. + NumCompactors int + + // Transaction start and commit timestamps are manaVgedTxns by end-user. This + // is a private option used by ManagedDB. + managedTxns bool + + // 4. Flags for testing purposes + // ------------------------------ + DoNotCompact bool // Stops LSM tree from compactions. + + maxBatchCount int64 // max entries in batch + maxBatchSize int64 // max batch size in bytes +} + +// DefaultOptions sets a list of recommended options for good performance. +// Feel free to modify these to suit your needs. +var DefaultOptions = Options{ + DoNotCompact: false, + LevelOneSize: 256 << 20, + LevelSizeMultiplier: 10, + TableLoadingMode: options.LoadToRAM, + ValueLogLoadingMode: options.MemoryMap, + // table.MemoryMap to mmap() the tables. + // table.Nothing to not preload the tables. + MaxLevels: 7, + MaxTableSize: 64 << 20, + NumCompactors: 3, + NumLevelZeroTables: 5, + NumLevelZeroTablesStall: 10, + NumMemtables: 5, + SyncWrites: true, + // Nothing to read/write value log using standard File I/O + // MemoryMap to mmap() the value log files + ValueLogFileSize: 1 << 30, + ValueThreshold: 20, +} diff --git a/vendor/github.com/dgraph-io/badger/options/options.go b/vendor/github.com/dgraph-io/badger/options/options.go new file mode 100644 index 0000000000..06c8b1b7f0 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/options/options.go @@ -0,0 +1,30 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package options + +// FileLoadingMode specifies how data in LSM table files and value log files should +// be loaded. +type FileLoadingMode int + +const ( + // FileIO indicates that files must be loaded using standard I/O + FileIO FileLoadingMode = iota + // LoadToRAM indicates that file must be loaded into RAM + LoadToRAM + // MemoryMap indicates that that the file must be memory-mapped + MemoryMap +) diff --git a/vendor/github.com/dgraph-io/badger/protos/backup.pb.go b/vendor/github.com/dgraph-io/badger/protos/backup.pb.go new file mode 100644 index 0000000000..13a9f61999 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/protos/backup.pb.go @@ -0,0 +1,497 @@ +// Code generated by protoc-gen-gogo. DO NOT EDIT. +// source: backup.proto + +/* + Package protos is a generated protocol buffer package. + + It is generated from these files: + backup.proto + manifest.proto + + It has these top-level messages: + KVPair + ManifestChangeSet + ManifestChange +*/ +package protos + +import proto "github.com/golang/protobuf/proto" +import fmt "fmt" +import math "math" + +import io "io" + +// Reference imports to suppress errors if they are not otherwise used. +var _ = proto.Marshal +var _ = fmt.Errorf +var _ = math.Inf + +// This is a compile-time assertion to ensure that this generated file +// is compatible with the proto package it is being compiled against. +// A compilation error at this line likely means your copy of the +// proto package needs to be updated. +const _ = proto.ProtoPackageIsVersion2 // please upgrade the proto package + +type KVPair struct { + Key []byte `protobuf:"bytes,1,opt,name=key,proto3" json:"key,omitempty"` + Value []byte `protobuf:"bytes,2,opt,name=value,proto3" json:"value,omitempty"` + UserMeta []byte `protobuf:"bytes,3,opt,name=userMeta,proto3" json:"userMeta,omitempty"` + Version uint64 `protobuf:"varint,4,opt,name=version,proto3" json:"version,omitempty"` + ExpiresAt uint64 `protobuf:"varint,5,opt,name=expires_at,json=expiresAt,proto3" json:"expires_at,omitempty"` +} + +func (m *KVPair) Reset() { *m = KVPair{} } +func (m *KVPair) String() string { return proto.CompactTextString(m) } +func (*KVPair) ProtoMessage() {} +func (*KVPair) Descriptor() ([]byte, []int) { return fileDescriptorBackup, []int{0} } + +func (m *KVPair) GetKey() []byte { + if m != nil { + return m.Key + } + return nil +} + +func (m *KVPair) GetValue() []byte { + if m != nil { + return m.Value + } + return nil +} + +func (m *KVPair) GetUserMeta() []byte { + if m != nil { + return m.UserMeta + } + return nil +} + +func (m *KVPair) GetVersion() uint64 { + if m != nil { + return m.Version + } + return 0 +} + +func (m *KVPair) GetExpiresAt() uint64 { + if m != nil { + return m.ExpiresAt + } + return 0 +} + +func init() { + proto.RegisterType((*KVPair)(nil), "protos.KVPair") +} +func (m *KVPair) Marshal() (dAtA []byte, err error) { + size := m.Size() + dAtA = make([]byte, size) + n, err := m.MarshalTo(dAtA) + if err != nil { + return nil, err + } + return dAtA[:n], nil +} + +func (m *KVPair) MarshalTo(dAtA []byte) (int, error) { + var i int + _ = i + var l int + _ = l + if len(m.Key) > 0 { + dAtA[i] = 0xa + i++ + i = encodeVarintBackup(dAtA, i, uint64(len(m.Key))) + i += copy(dAtA[i:], m.Key) + } + if len(m.Value) > 0 { + dAtA[i] = 0x12 + i++ + i = encodeVarintBackup(dAtA, i, uint64(len(m.Value))) + i += copy(dAtA[i:], m.Value) + } + if len(m.UserMeta) > 0 { + dAtA[i] = 0x1a + i++ + i = encodeVarintBackup(dAtA, i, uint64(len(m.UserMeta))) + i += copy(dAtA[i:], m.UserMeta) + } + if m.Version != 0 { + dAtA[i] = 0x20 + i++ + i = encodeVarintBackup(dAtA, i, uint64(m.Version)) + } + if m.ExpiresAt != 0 { + dAtA[i] = 0x28 + i++ + i = encodeVarintBackup(dAtA, i, uint64(m.ExpiresAt)) + } + return i, nil +} + +func encodeFixed64Backup(dAtA []byte, offset int, v uint64) int { + dAtA[offset] = uint8(v) + dAtA[offset+1] = uint8(v >> 8) + dAtA[offset+2] = uint8(v >> 16) + dAtA[offset+3] = uint8(v >> 24) + dAtA[offset+4] = uint8(v >> 32) + dAtA[offset+5] = uint8(v >> 40) + dAtA[offset+6] = uint8(v >> 48) + dAtA[offset+7] = uint8(v >> 56) + return offset + 8 +} +func encodeFixed32Backup(dAtA []byte, offset int, v uint32) int { + dAtA[offset] = uint8(v) + dAtA[offset+1] = uint8(v >> 8) + dAtA[offset+2] = uint8(v >> 16) + dAtA[offset+3] = uint8(v >> 24) + return offset + 4 +} +func encodeVarintBackup(dAtA []byte, offset int, v uint64) int { + for v >= 1<<7 { + dAtA[offset] = uint8(v&0x7f | 0x80) + v >>= 7 + offset++ + } + dAtA[offset] = uint8(v) + return offset + 1 +} +func (m *KVPair) Size() (n int) { + var l int + _ = l + l = len(m.Key) + if l > 0 { + n += 1 + l + sovBackup(uint64(l)) + } + l = len(m.Value) + if l > 0 { + n += 1 + l + sovBackup(uint64(l)) + } + l = len(m.UserMeta) + if l > 0 { + n += 1 + l + sovBackup(uint64(l)) + } + if m.Version != 0 { + n += 1 + sovBackup(uint64(m.Version)) + } + if m.ExpiresAt != 0 { + n += 1 + sovBackup(uint64(m.ExpiresAt)) + } + return n +} + +func sovBackup(x uint64) (n int) { + for { + n++ + x >>= 7 + if x == 0 { + break + } + } + return n +} +func sozBackup(x uint64) (n int) { + return sovBackup(uint64((x << 1) ^ uint64((int64(x) >> 63)))) +} +func (m *KVPair) Unmarshal(dAtA []byte) error { + l := len(dAtA) + iNdEx := 0 + for iNdEx < l { + preIndex := iNdEx + var wire uint64 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBackup + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + wire |= (uint64(b) & 0x7F) << shift + if b < 0x80 { + break + } + } + fieldNum := int32(wire >> 3) + wireType := int(wire & 0x7) + if wireType == 4 { + return fmt.Errorf("proto: KVPair: wiretype end group for non-group") + } + if fieldNum <= 0 { + return fmt.Errorf("proto: KVPair: illegal tag %d (wire type %d)", fieldNum, wire) + } + switch fieldNum { + case 1: + if wireType != 2 { + return fmt.Errorf("proto: wrong wireType = %d for field Key", wireType) + } + var byteLen int + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBackup + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + byteLen |= (int(b) & 0x7F) << shift + if b < 0x80 { + break + } + } + if byteLen < 0 { + return ErrInvalidLengthBackup + } + postIndex := iNdEx + byteLen + if postIndex > l { + return io.ErrUnexpectedEOF + } + m.Key = append(m.Key[:0], dAtA[iNdEx:postIndex]...) + if m.Key == nil { + m.Key = []byte{} + } + iNdEx = postIndex + case 2: + if wireType != 2 { + return fmt.Errorf("proto: wrong wireType = %d for field Value", wireType) + } + var byteLen int + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBackup + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + byteLen |= (int(b) & 0x7F) << shift + if b < 0x80 { + break + } + } + if byteLen < 0 { + return ErrInvalidLengthBackup + } + postIndex := iNdEx + byteLen + if postIndex > l { + return io.ErrUnexpectedEOF + } + m.Value = append(m.Value[:0], dAtA[iNdEx:postIndex]...) + if m.Value == nil { + m.Value = []byte{} + } + iNdEx = postIndex + case 3: + if wireType != 2 { + return fmt.Errorf("proto: wrong wireType = %d for field UserMeta", wireType) + } + var byteLen int + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBackup + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + byteLen |= (int(b) & 0x7F) << shift + if b < 0x80 { + break + } + } + if byteLen < 0 { + return ErrInvalidLengthBackup + } + postIndex := iNdEx + byteLen + if postIndex > l { + return io.ErrUnexpectedEOF + } + m.UserMeta = append(m.UserMeta[:0], dAtA[iNdEx:postIndex]...) + if m.UserMeta == nil { + m.UserMeta = []byte{} + } + iNdEx = postIndex + case 4: + if wireType != 0 { + return fmt.Errorf("proto: wrong wireType = %d for field Version", wireType) + } + m.Version = 0 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBackup + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + m.Version |= (uint64(b) & 0x7F) << shift + if b < 0x80 { + break + } + } + case 5: + if wireType != 0 { + return fmt.Errorf("proto: wrong wireType = %d for field ExpiresAt", wireType) + } + m.ExpiresAt = 0 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBackup + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + m.ExpiresAt |= (uint64(b) & 0x7F) << shift + if b < 0x80 { + break + } + } + default: + iNdEx = preIndex + skippy, err := skipBackup(dAtA[iNdEx:]) + if err != nil { + return err + } + if skippy < 0 { + return ErrInvalidLengthBackup + } + if (iNdEx + skippy) > l { + return io.ErrUnexpectedEOF + } + iNdEx += skippy + } + } + + if iNdEx > l { + return io.ErrUnexpectedEOF + } + return nil +} +func skipBackup(dAtA []byte) (n int, err error) { + l := len(dAtA) + iNdEx := 0 + for iNdEx < l { + var wire uint64 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return 0, ErrIntOverflowBackup + } + if iNdEx >= l { + return 0, io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + wire |= (uint64(b) & 0x7F) << shift + if b < 0x80 { + break + } + } + wireType := int(wire & 0x7) + switch wireType { + case 0: + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return 0, ErrIntOverflowBackup + } + if iNdEx >= l { + return 0, io.ErrUnexpectedEOF + } + iNdEx++ + if dAtA[iNdEx-1] < 0x80 { + break + } + } + return iNdEx, nil + case 1: + iNdEx += 8 + return iNdEx, nil + case 2: + var length int + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return 0, ErrIntOverflowBackup + } + if iNdEx >= l { + return 0, io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + length |= (int(b) & 0x7F) << shift + if b < 0x80 { + break + } + } + iNdEx += length + if length < 0 { + return 0, ErrInvalidLengthBackup + } + return iNdEx, nil + case 3: + for { + var innerWire uint64 + var start int = iNdEx + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return 0, ErrIntOverflowBackup + } + if iNdEx >= l { + return 0, io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + innerWire |= (uint64(b) & 0x7F) << shift + if b < 0x80 { + break + } + } + innerWireType := int(innerWire & 0x7) + if innerWireType == 4 { + break + } + next, err := skipBackup(dAtA[start:]) + if err != nil { + return 0, err + } + iNdEx = start + next + } + return iNdEx, nil + case 4: + return iNdEx, nil + case 5: + iNdEx += 4 + return iNdEx, nil + default: + return 0, fmt.Errorf("proto: illegal wireType %d", wireType) + } + } + panic("unreachable") +} + +var ( + ErrInvalidLengthBackup = fmt.Errorf("proto: negative length found during unmarshaling") + ErrIntOverflowBackup = fmt.Errorf("proto: integer overflow") +) + +func init() { proto.RegisterFile("backup.proto", fileDescriptorBackup) } + +var fileDescriptorBackup = []byte{ + // 167 bytes of a gzipped FileDescriptorProto + 0x1f, 0x8b, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0xff, 0xe2, 0xe2, 0x49, 0x4a, 0x4c, 0xce, + 0x2e, 0x2d, 0xd0, 0x2b, 0x28, 0xca, 0x2f, 0xc9, 0x17, 0x62, 0x03, 0x53, 0xc5, 0x4a, 0xad, 0x8c, + 0x5c, 0x6c, 0xde, 0x61, 0x01, 0x89, 0x99, 0x45, 0x42, 0x02, 0x5c, 0xcc, 0xd9, 0xa9, 0x95, 0x12, + 0x8c, 0x0a, 0x8c, 0x1a, 0x3c, 0x41, 0x20, 0xa6, 0x90, 0x08, 0x17, 0x6b, 0x59, 0x62, 0x4e, 0x69, + 0xaa, 0x04, 0x13, 0x58, 0x0c, 0xc2, 0x11, 0x92, 0xe2, 0xe2, 0x28, 0x2d, 0x4e, 0x2d, 0xf2, 0x4d, + 0x2d, 0x49, 0x94, 0x60, 0x06, 0x4b, 0xc0, 0xf9, 0x42, 0x12, 0x5c, 0xec, 0x65, 0xa9, 0x45, 0xc5, + 0x99, 0xf9, 0x79, 0x12, 0x2c, 0x0a, 0x8c, 0x1a, 0x2c, 0x41, 0x30, 0xae, 0x90, 0x2c, 0x17, 0x57, + 0x6a, 0x45, 0x41, 0x66, 0x51, 0x6a, 0x71, 0x7c, 0x62, 0x89, 0x04, 0x2b, 0x58, 0x92, 0x13, 0x2a, + 0xe2, 0x58, 0xe2, 0x24, 0x70, 0xe2, 0x91, 0x1c, 0xe3, 0x85, 0x47, 0x72, 0x8c, 0x0f, 0x1e, 0xc9, + 0x31, 0xce, 0x78, 0x2c, 0xc7, 0x90, 0x04, 0x71, 0xa1, 0x31, 0x20, 0x00, 0x00, 0xff, 0xff, 0xe7, + 0x3f, 0x3f, 0x95, 0xb8, 0x00, 0x00, 0x00, +} diff --git a/vendor/github.com/dgraph-io/badger/protos/backup.proto b/vendor/github.com/dgraph-io/badger/protos/backup.proto new file mode 100644 index 0000000000..0f4e3d61ee --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/protos/backup.proto @@ -0,0 +1,28 @@ +/* + * Copyright (C) 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +// Use protos/gen.sh to generate .pb.go files. +syntax = "proto3"; + +package protos; + +message KVPair { + bytes key = 1; + bytes value = 2; + bytes userMeta = 3; + uint64 version = 4; + uint64 expires_at = 5; +} \ No newline at end of file diff --git a/vendor/github.com/dgraph-io/badger/protos/gen.sh b/vendor/github.com/dgraph-io/badger/protos/gen.sh new file mode 100755 index 0000000000..15bb38eb97 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/protos/gen.sh @@ -0,0 +1,7 @@ +#!/bin/bash + +# You might need to go get -v github.com/gogo/protobuf/... + +protos=${GOPATH-$HOME/go}/src/github.com/dgraph-io/badger/protos +pushd $protos > /dev/null +protoc --gofast_out=plugins=grpc:. -I=. *.proto diff --git a/vendor/github.com/dgraph-io/badger/protos/manifest.pb.go b/vendor/github.com/dgraph-io/badger/protos/manifest.pb.go new file mode 100644 index 0000000000..d8db55f99d --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/protos/manifest.pb.go @@ -0,0 +1,534 @@ +// Code generated by protoc-gen-gogo. DO NOT EDIT. +// source: manifest.proto + +package protos + +import proto "github.com/golang/protobuf/proto" +import fmt "fmt" +import math "math" + +import io "io" + +// Reference imports to suppress errors if they are not otherwise used. +var _ = proto.Marshal +var _ = fmt.Errorf +var _ = math.Inf + +type ManifestChange_Operation int32 + +const ( + ManifestChange_CREATE ManifestChange_Operation = 0 + ManifestChange_DELETE ManifestChange_Operation = 1 +) + +var ManifestChange_Operation_name = map[int32]string{ + 0: "CREATE", + 1: "DELETE", +} +var ManifestChange_Operation_value = map[string]int32{ + "CREATE": 0, + "DELETE": 1, +} + +func (x ManifestChange_Operation) String() string { + return proto.EnumName(ManifestChange_Operation_name, int32(x)) +} +func (ManifestChange_Operation) EnumDescriptor() ([]byte, []int) { + return fileDescriptorManifest, []int{1, 0} +} + +type ManifestChangeSet struct { + // A set of changes that are applied atomically. + Changes []*ManifestChange `protobuf:"bytes,1,rep,name=changes" json:"changes,omitempty"` +} + +func (m *ManifestChangeSet) Reset() { *m = ManifestChangeSet{} } +func (m *ManifestChangeSet) String() string { return proto.CompactTextString(m) } +func (*ManifestChangeSet) ProtoMessage() {} +func (*ManifestChangeSet) Descriptor() ([]byte, []int) { return fileDescriptorManifest, []int{0} } + +func (m *ManifestChangeSet) GetChanges() []*ManifestChange { + if m != nil { + return m.Changes + } + return nil +} + +type ManifestChange struct { + Id uint64 `protobuf:"varint,1,opt,name=Id,proto3" json:"Id,omitempty"` + Op ManifestChange_Operation `protobuf:"varint,2,opt,name=Op,proto3,enum=protos.ManifestChange_Operation" json:"Op,omitempty"` + Level uint32 `protobuf:"varint,3,opt,name=Level,proto3" json:"Level,omitempty"` +} + +func (m *ManifestChange) Reset() { *m = ManifestChange{} } +func (m *ManifestChange) String() string { return proto.CompactTextString(m) } +func (*ManifestChange) ProtoMessage() {} +func (*ManifestChange) Descriptor() ([]byte, []int) { return fileDescriptorManifest, []int{1} } + +func (m *ManifestChange) GetId() uint64 { + if m != nil { + return m.Id + } + return 0 +} + +func (m *ManifestChange) GetOp() ManifestChange_Operation { + if m != nil { + return m.Op + } + return ManifestChange_CREATE +} + +func (m *ManifestChange) GetLevel() uint32 { + if m != nil { + return m.Level + } + return 0 +} + +func init() { + proto.RegisterType((*ManifestChangeSet)(nil), "protos.ManifestChangeSet") + proto.RegisterType((*ManifestChange)(nil), "protos.ManifestChange") + proto.RegisterEnum("protos.ManifestChange_Operation", ManifestChange_Operation_name, ManifestChange_Operation_value) +} +func (m *ManifestChangeSet) Marshal() (dAtA []byte, err error) { + size := m.Size() + dAtA = make([]byte, size) + n, err := m.MarshalTo(dAtA) + if err != nil { + return nil, err + } + return dAtA[:n], nil +} + +func (m *ManifestChangeSet) MarshalTo(dAtA []byte) (int, error) { + var i int + _ = i + var l int + _ = l + if len(m.Changes) > 0 { + for _, msg := range m.Changes { + dAtA[i] = 0xa + i++ + i = encodeVarintManifest(dAtA, i, uint64(msg.Size())) + n, err := msg.MarshalTo(dAtA[i:]) + if err != nil { + return 0, err + } + i += n + } + } + return i, nil +} + +func (m *ManifestChange) Marshal() (dAtA []byte, err error) { + size := m.Size() + dAtA = make([]byte, size) + n, err := m.MarshalTo(dAtA) + if err != nil { + return nil, err + } + return dAtA[:n], nil +} + +func (m *ManifestChange) MarshalTo(dAtA []byte) (int, error) { + var i int + _ = i + var l int + _ = l + if m.Id != 0 { + dAtA[i] = 0x8 + i++ + i = encodeVarintManifest(dAtA, i, uint64(m.Id)) + } + if m.Op != 0 { + dAtA[i] = 0x10 + i++ + i = encodeVarintManifest(dAtA, i, uint64(m.Op)) + } + if m.Level != 0 { + dAtA[i] = 0x18 + i++ + i = encodeVarintManifest(dAtA, i, uint64(m.Level)) + } + return i, nil +} + +func encodeFixed64Manifest(dAtA []byte, offset int, v uint64) int { + dAtA[offset] = uint8(v) + dAtA[offset+1] = uint8(v >> 8) + dAtA[offset+2] = uint8(v >> 16) + dAtA[offset+3] = uint8(v >> 24) + dAtA[offset+4] = uint8(v >> 32) + dAtA[offset+5] = uint8(v >> 40) + dAtA[offset+6] = uint8(v >> 48) + dAtA[offset+7] = uint8(v >> 56) + return offset + 8 +} +func encodeFixed32Manifest(dAtA []byte, offset int, v uint32) int { + dAtA[offset] = uint8(v) + dAtA[offset+1] = uint8(v >> 8) + dAtA[offset+2] = uint8(v >> 16) + dAtA[offset+3] = uint8(v >> 24) + return offset + 4 +} +func encodeVarintManifest(dAtA []byte, offset int, v uint64) int { + for v >= 1<<7 { + dAtA[offset] = uint8(v&0x7f | 0x80) + v >>= 7 + offset++ + } + dAtA[offset] = uint8(v) + return offset + 1 +} +func (m *ManifestChangeSet) Size() (n int) { + var l int + _ = l + if len(m.Changes) > 0 { + for _, e := range m.Changes { + l = e.Size() + n += 1 + l + sovManifest(uint64(l)) + } + } + return n +} + +func (m *ManifestChange) Size() (n int) { + var l int + _ = l + if m.Id != 0 { + n += 1 + sovManifest(uint64(m.Id)) + } + if m.Op != 0 { + n += 1 + sovManifest(uint64(m.Op)) + } + if m.Level != 0 { + n += 1 + sovManifest(uint64(m.Level)) + } + return n +} + +func sovManifest(x uint64) (n int) { + for { + n++ + x >>= 7 + if x == 0 { + break + } + } + return n +} +func sozManifest(x uint64) (n int) { + return sovManifest(uint64((x << 1) ^ uint64((int64(x) >> 63)))) +} +func (m *ManifestChangeSet) Unmarshal(dAtA []byte) error { + l := len(dAtA) + iNdEx := 0 + for iNdEx < l { + preIndex := iNdEx + var wire uint64 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowManifest + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + wire |= (uint64(b) & 0x7F) << shift + if b < 0x80 { + break + } + } + fieldNum := int32(wire >> 3) + wireType := int(wire & 0x7) + if wireType == 4 { + return fmt.Errorf("proto: ManifestChangeSet: wiretype end group for non-group") + } + if fieldNum <= 0 { + return fmt.Errorf("proto: ManifestChangeSet: illegal tag %d (wire type %d)", fieldNum, wire) + } + switch fieldNum { + case 1: + if wireType != 2 { + return fmt.Errorf("proto: wrong wireType = %d for field Changes", wireType) + } + var msglen int + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowManifest + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + msglen |= (int(b) & 0x7F) << shift + if b < 0x80 { + break + } + } + if msglen < 0 { + return ErrInvalidLengthManifest + } + postIndex := iNdEx + msglen + if postIndex > l { + return io.ErrUnexpectedEOF + } + m.Changes = append(m.Changes, &ManifestChange{}) + if err := m.Changes[len(m.Changes)-1].Unmarshal(dAtA[iNdEx:postIndex]); err != nil { + return err + } + iNdEx = postIndex + default: + iNdEx = preIndex + skippy, err := skipManifest(dAtA[iNdEx:]) + if err != nil { + return err + } + if skippy < 0 { + return ErrInvalidLengthManifest + } + if (iNdEx + skippy) > l { + return io.ErrUnexpectedEOF + } + iNdEx += skippy + } + } + + if iNdEx > l { + return io.ErrUnexpectedEOF + } + return nil +} +func (m *ManifestChange) Unmarshal(dAtA []byte) error { + l := len(dAtA) + iNdEx := 0 + for iNdEx < l { + preIndex := iNdEx + var wire uint64 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowManifest + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + wire |= (uint64(b) & 0x7F) << shift + if b < 0x80 { + break + } + } + fieldNum := int32(wire >> 3) + wireType := int(wire & 0x7) + if wireType == 4 { + return fmt.Errorf("proto: ManifestChange: wiretype end group for non-group") + } + if fieldNum <= 0 { + return fmt.Errorf("proto: ManifestChange: illegal tag %d (wire type %d)", fieldNum, wire) + } + switch fieldNum { + case 1: + if wireType != 0 { + return fmt.Errorf("proto: wrong wireType = %d for field Id", wireType) + } + m.Id = 0 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowManifest + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + m.Id |= (uint64(b) & 0x7F) << shift + if b < 0x80 { + break + } + } + case 2: + if wireType != 0 { + return fmt.Errorf("proto: wrong wireType = %d for field Op", wireType) + } + m.Op = 0 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowManifest + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + m.Op |= (ManifestChange_Operation(b) & 0x7F) << shift + if b < 0x80 { + break + } + } + case 3: + if wireType != 0 { + return fmt.Errorf("proto: wrong wireType = %d for field Level", wireType) + } + m.Level = 0 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowManifest + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + m.Level |= (uint32(b) & 0x7F) << shift + if b < 0x80 { + break + } + } + default: + iNdEx = preIndex + skippy, err := skipManifest(dAtA[iNdEx:]) + if err != nil { + return err + } + if skippy < 0 { + return ErrInvalidLengthManifest + } + if (iNdEx + skippy) > l { + return io.ErrUnexpectedEOF + } + iNdEx += skippy + } + } + + if iNdEx > l { + return io.ErrUnexpectedEOF + } + return nil +} +func skipManifest(dAtA []byte) (n int, err error) { + l := len(dAtA) + iNdEx := 0 + for iNdEx < l { + var wire uint64 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return 0, ErrIntOverflowManifest + } + if iNdEx >= l { + return 0, io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + wire |= (uint64(b) & 0x7F) << shift + if b < 0x80 { + break + } + } + wireType := int(wire & 0x7) + switch wireType { + case 0: + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return 0, ErrIntOverflowManifest + } + if iNdEx >= l { + return 0, io.ErrUnexpectedEOF + } + iNdEx++ + if dAtA[iNdEx-1] < 0x80 { + break + } + } + return iNdEx, nil + case 1: + iNdEx += 8 + return iNdEx, nil + case 2: + var length int + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return 0, ErrIntOverflowManifest + } + if iNdEx >= l { + return 0, io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + length |= (int(b) & 0x7F) << shift + if b < 0x80 { + break + } + } + iNdEx += length + if length < 0 { + return 0, ErrInvalidLengthManifest + } + return iNdEx, nil + case 3: + for { + var innerWire uint64 + var start int = iNdEx + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return 0, ErrIntOverflowManifest + } + if iNdEx >= l { + return 0, io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + innerWire |= (uint64(b) & 0x7F) << shift + if b < 0x80 { + break + } + } + innerWireType := int(innerWire & 0x7) + if innerWireType == 4 { + break + } + next, err := skipManifest(dAtA[start:]) + if err != nil { + return 0, err + } + iNdEx = start + next + } + return iNdEx, nil + case 4: + return iNdEx, nil + case 5: + iNdEx += 4 + return iNdEx, nil + default: + return 0, fmt.Errorf("proto: illegal wireType %d", wireType) + } + } + panic("unreachable") +} + +var ( + ErrInvalidLengthManifest = fmt.Errorf("proto: negative length found during unmarshaling") + ErrIntOverflowManifest = fmt.Errorf("proto: integer overflow") +) + +func init() { proto.RegisterFile("manifest.proto", fileDescriptorManifest) } + +var fileDescriptorManifest = []byte{ + // 208 bytes of a gzipped FileDescriptorProto + 0x1f, 0x8b, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0xff, 0xe2, 0xe2, 0xcb, 0x4d, 0xcc, 0xcb, + 0x4c, 0x4b, 0x2d, 0x2e, 0xd1, 0x2b, 0x28, 0xca, 0x2f, 0xc9, 0x17, 0x62, 0x03, 0x53, 0xc5, 0x4a, + 0xae, 0x5c, 0x82, 0xbe, 0x50, 0x19, 0xe7, 0x8c, 0xc4, 0xbc, 0xf4, 0xd4, 0xe0, 0xd4, 0x12, 0x21, + 0x03, 0x2e, 0xf6, 0x64, 0x30, 0xa7, 0x58, 0x82, 0x51, 0x81, 0x59, 0x83, 0xdb, 0x48, 0x0c, 0xa2, + 0xab, 0x58, 0x0f, 0x55, 0x6d, 0x10, 0x4c, 0x99, 0x52, 0x2f, 0x23, 0x17, 0x1f, 0xaa, 0x9c, 0x10, + 0x1f, 0x17, 0x93, 0x67, 0x8a, 0x04, 0xa3, 0x02, 0xa3, 0x06, 0x4b, 0x10, 0x93, 0x67, 0x8a, 0x90, + 0x01, 0x17, 0x93, 0x7f, 0x81, 0x04, 0x93, 0x02, 0xa3, 0x06, 0x9f, 0x91, 0x02, 0x76, 0xf3, 0xf4, + 0xfc, 0x0b, 0x52, 0x8b, 0x12, 0x4b, 0x32, 0xf3, 0xf3, 0x82, 0x98, 0xfc, 0x0b, 0x84, 0x44, 0xb8, + 0x58, 0x7d, 0x52, 0xcb, 0x52, 0x73, 0x24, 0x98, 0x15, 0x18, 0x35, 0x78, 0x83, 0x20, 0x1c, 0x25, + 0x65, 0x2e, 0x4e, 0xb8, 0x32, 0x21, 0x2e, 0x2e, 0x36, 0xe7, 0x20, 0x57, 0xc7, 0x10, 0x57, 0x01, + 0x06, 0x10, 0xdb, 0xc5, 0xd5, 0xc7, 0x35, 0xc4, 0x55, 0x80, 0xd1, 0x49, 0xe0, 0xc4, 0x23, 0x39, + 0xc6, 0x0b, 0x8f, 0xe4, 0x18, 0x1f, 0x3c, 0x92, 0x63, 0x9c, 0xf1, 0x58, 0x8e, 0x21, 0x09, 0xe2, + 0x61, 0x63, 0x40, 0x00, 0x00, 0x00, 0xff, 0xff, 0x42, 0x6f, 0x23, 0xc9, 0x09, 0x01, 0x00, 0x00, +} diff --git a/vendor/github.com/dgraph-io/badger/protos/manifest.proto b/vendor/github.com/dgraph-io/badger/protos/manifest.proto new file mode 100644 index 0000000000..295c63a482 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/protos/manifest.proto @@ -0,0 +1,35 @@ +/* + * Copyright (C) 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +// Use protos/gen.sh to generate .pb.go files. +syntax = "proto3"; + +package protos; + +message ManifestChangeSet { + // A set of changes that are applied atomically. + repeated ManifestChange changes = 1; +} + +message ManifestChange { + uint64 Id = 1; + enum Operation { + CREATE = 0; + DELETE = 1; + } + Operation Op = 2; + uint32 Level = 3; // Only used for CREATE +} diff --git a/vendor/github.com/dgraph-io/badger/skl/README.md b/vendor/github.com/dgraph-io/badger/skl/README.md new file mode 100644 index 0000000000..92fa68bb53 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/skl/README.md @@ -0,0 +1,113 @@ +This is much better than `skiplist` and `slist`. + +``` +BenchmarkReadWrite/frac_0-8 3000000 537 ns/op +BenchmarkReadWrite/frac_1-8 3000000 503 ns/op +BenchmarkReadWrite/frac_2-8 3000000 492 ns/op +BenchmarkReadWrite/frac_3-8 3000000 475 ns/op +BenchmarkReadWrite/frac_4-8 3000000 440 ns/op +BenchmarkReadWrite/frac_5-8 5000000 442 ns/op +BenchmarkReadWrite/frac_6-8 5000000 380 ns/op +BenchmarkReadWrite/frac_7-8 5000000 338 ns/op +BenchmarkReadWrite/frac_8-8 5000000 294 ns/op +BenchmarkReadWrite/frac_9-8 10000000 268 ns/op +BenchmarkReadWrite/frac_10-8 100000000 26.3 ns/op +``` + +And even better than a simple map with read-write lock: + +``` +BenchmarkReadWriteMap/frac_0-8 2000000 774 ns/op +BenchmarkReadWriteMap/frac_1-8 2000000 647 ns/op +BenchmarkReadWriteMap/frac_2-8 3000000 605 ns/op +BenchmarkReadWriteMap/frac_3-8 3000000 603 ns/op +BenchmarkReadWriteMap/frac_4-8 3000000 556 ns/op +BenchmarkReadWriteMap/frac_5-8 3000000 472 ns/op +BenchmarkReadWriteMap/frac_6-8 3000000 476 ns/op +BenchmarkReadWriteMap/frac_7-8 3000000 457 ns/op +BenchmarkReadWriteMap/frac_8-8 5000000 444 ns/op +BenchmarkReadWriteMap/frac_9-8 5000000 361 ns/op +BenchmarkReadWriteMap/frac_10-8 10000000 212 ns/op +``` + +# Node Pooling + +Command used + +``` +rm -Rf tmp && /usr/bin/time -l ./populate -keys_mil 10 +``` + +For pprof results, we run without using /usr/bin/time. There are four runs below. + +Results seem to vary quite a bit between runs. + +## Before node pooling + +``` +1311.53MB of 1338.69MB total (97.97%) +Dropped 30 nodes (cum <= 6.69MB) +Showing top 10 nodes out of 37 (cum >= 12.50MB) + flat flat% sum% cum cum% + 523.04MB 39.07% 39.07% 523.04MB 39.07% github.com/dgraph-io/badger/skl.(*Skiplist).Put + 184.51MB 13.78% 52.85% 184.51MB 13.78% runtime.stringtoslicebyte + 166.01MB 12.40% 65.25% 689.04MB 51.47% github.com/dgraph-io/badger/mem.(*Table).Put + 165MB 12.33% 77.58% 165MB 12.33% runtime.convT2E + 116.92MB 8.73% 86.31% 116.92MB 8.73% bytes.makeSlice + 62.50MB 4.67% 90.98% 62.50MB 4.67% main.newValue + 34.50MB 2.58% 93.56% 34.50MB 2.58% github.com/dgraph-io/badger/table.(*BlockIterator).parseKV + 25.50MB 1.90% 95.46% 100.06MB 7.47% github.com/dgraph-io/badger/y.(*MergeIterator).Next + 21.06MB 1.57% 97.04% 21.06MB 1.57% github.com/dgraph-io/badger/table.(*Table).read + 12.50MB 0.93% 97.97% 12.50MB 0.93% github.com/dgraph-io/badger/table.header.Encode + + 128.31 real 329.37 user 17.11 sys +3355660288 maximum resident set size + 0 average shared memory size + 0 average unshared data size + 0 average unshared stack size + 2203080 page reclaims + 764 page faults + 0 swaps + 275 block input operations + 76 block output operations + 0 messages sent + 0 messages received + 0 signals received + 49173 voluntary context switches + 599922 involuntary context switches +``` + +## After node pooling + +``` +1963.13MB of 2026.09MB total (96.89%) +Dropped 29 nodes (cum <= 10.13MB) +Showing top 10 nodes out of 41 (cum >= 185.62MB) + flat flat% sum% cum cum% + 658.05MB 32.48% 32.48% 658.05MB 32.48% github.com/dgraph-io/badger/skl.glob..func1 + 297.51MB 14.68% 47.16% 297.51MB 14.68% runtime.convT2E + 257.51MB 12.71% 59.87% 257.51MB 12.71% runtime.stringtoslicebyte + 249.01MB 12.29% 72.16% 1007.06MB 49.70% github.com/dgraph-io/badger/mem.(*Table).Put + 142.43MB 7.03% 79.19% 142.43MB 7.03% bytes.makeSlice + 100MB 4.94% 84.13% 758.05MB 37.41% github.com/dgraph-io/badger/skl.newNode + 99.50MB 4.91% 89.04% 99.50MB 4.91% main.newValue + 75MB 3.70% 92.74% 75MB 3.70% github.com/dgraph-io/badger/table.(*BlockIterator).parseKV + 44.62MB 2.20% 94.94% 44.62MB 2.20% github.com/dgraph-io/badger/table.(*Table).read + 39.50MB 1.95% 96.89% 185.62MB 9.16% github.com/dgraph-io/badger/y.(*MergeIterator).Next + + 135.58 real 374.29 user 17.65 sys +3740614656 maximum resident set size + 0 average shared memory size + 0 average unshared data size + 0 average unshared stack size + 2276566 page reclaims + 770 page faults + 0 swaps + 128 block input operations + 90 block output operations + 0 messages sent + 0 messages received + 0 signals received + 46434 voluntary context switches + 597049 involuntary context switches +``` \ No newline at end of file diff --git a/vendor/github.com/dgraph-io/badger/skl/arena.go b/vendor/github.com/dgraph-io/badger/skl/arena.go new file mode 100644 index 0000000000..849e5ee564 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/skl/arena.go @@ -0,0 +1,131 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package skl + +import ( + "sync/atomic" + "unsafe" + + "github.com/dgraph-io/badger/y" +) + +const ( + offsetSize = int(unsafe.Sizeof(uint32(0))) + ptrAlign = int(unsafe.Sizeof(uintptr(0))) - 1 +) + +// Arena should be lock-free. +type Arena struct { + n uint32 + buf []byte +} + +// newArena returns a new arena. +func newArena(n int64) *Arena { + // Don't store data at position 0 in order to reserve offset=0 as a kind + // of nil pointer. + out := &Arena{ + n: 1, + buf: make([]byte, n), + } + return out +} + +func (s *Arena) size() int64 { + return int64(atomic.LoadUint32(&s.n)) +} + +func (s *Arena) reset() { + atomic.StoreUint32(&s.n, 0) +} + +// putNode allocates a node in the arena. The node is aligned on a pointer-sized +// boundary. The arena offset of the node is returned. +func (s *Arena) putNode(height int) uint32 { + // Compute the amount of the tower that will never be used, since the height + // is less than maxHeight. + unusedSize := (maxHeight - height) * offsetSize + + // Pad the allocation with enough bytes to ensure pointer alignment. + l := uint32(MaxNodeSize - unusedSize + ptrAlign) + n := atomic.AddUint32(&s.n, l) + y.AssertTruef(int(n) <= len(s.buf), + "Arena too small, toWrite:%d newTotal:%d limit:%d", + l, n, len(s.buf)) + + // Return the aligned offset. + m := (n - l + uint32(ptrAlign)) & ^uint32(ptrAlign) + return m +} + +// Put will *copy* val into arena. To make better use of this, reuse your input +// val buffer. Returns an offset into buf. User is responsible for remembering +// size of val. We could also store this size inside arena but the encoding and +// decoding will incur some overhead. +func (s *Arena) putVal(v y.ValueStruct) uint32 { + l := uint32(v.EncodedSize()) + n := atomic.AddUint32(&s.n, l) + y.AssertTruef(int(n) <= len(s.buf), + "Arena too small, toWrite:%d newTotal:%d limit:%d", + l, n, len(s.buf)) + m := n - l + v.Encode(s.buf[m:]) + return m +} + +func (s *Arena) putKey(key []byte) uint32 { + l := uint32(len(key)) + n := atomic.AddUint32(&s.n, l) + y.AssertTruef(int(n) <= len(s.buf), + "Arena too small, toWrite:%d newTotal:%d limit:%d", + l, n, len(s.buf)) + m := n - l + y.AssertTrue(len(key) == copy(s.buf[m:n], key)) + return m +} + +// getNode returns a pointer to the node located at offset. If the offset is +// zero, then the nil node pointer is returned. +func (s *Arena) getNode(offset uint32) *node { + if offset == 0 { + return nil + } + + return (*node)(unsafe.Pointer(&s.buf[offset])) +} + +// getKey returns byte slice at offset. +func (s *Arena) getKey(offset uint32, size uint16) []byte { + return s.buf[offset : offset+uint32(size)] +} + +// getVal returns byte slice at offset. The given size should be just the value +// size and should NOT include the meta bytes. +func (s *Arena) getVal(offset uint32, size uint16) (ret y.ValueStruct) { + ret.Decode(s.buf[offset : offset+uint32(size)]) + return +} + +// getNodeOffset returns the offset of node in the arena. If the node pointer is +// nil, then the zero offset is returned. +func (s *Arena) getNodeOffset(nd *node) uint32 { + if nd == nil { + return 0 + } + + return uint32(uintptr(unsafe.Pointer(nd)) - uintptr(unsafe.Pointer(&s.buf[0]))) +} diff --git a/vendor/github.com/dgraph-io/badger/skl/skl.go b/vendor/github.com/dgraph-io/badger/skl/skl.go new file mode 100644 index 0000000000..7361751a50 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/skl/skl.go @@ -0,0 +1,524 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +/* +Adapted from RocksDB inline skiplist. + +Key differences: +- No optimization for sequential inserts (no "prev"). +- No custom comparator. +- Support overwrites. This requires care when we see the same key when inserting. + For RocksDB or LevelDB, overwrites are implemented as a newer sequence number in the key, so + there is no need for values. We don't intend to support versioning. In-place updates of values + would be more efficient. +- We discard all non-concurrent code. +- We do not support Splices. This simplifies the code a lot. +- No AllocateNode or other pointer arithmetic. +- We combine the findLessThan, findGreaterOrEqual, etc into one function. +*/ + +package skl + +import ( + "math" + "math/rand" + "sync/atomic" + "unsafe" + + "github.com/dgraph-io/badger/y" +) + +const ( + maxHeight = 20 + heightIncrease = math.MaxUint32 / 3 +) + +// MaxNodeSize is the memory footprint of a node of maximum height. +const MaxNodeSize = int(unsafe.Sizeof(node{})) + +type node struct { + // Multiple parts of the value are encoded as a single uint64 so that it + // can be atomically loaded and stored: + // value offset: uint32 (bits 0-31) + // value size : uint16 (bits 32-47) + // 12 bytes are allocated to ensure 8 byte alignment also on 32bit systems. + value [12]byte + + // A byte slice is 24 bytes. We are trying to save space here. + keyOffset uint32 // Immutable. No need to lock to access key. + keySize uint16 // Immutable. No need to lock to access key. + + // Height of the tower. + height uint16 + + // Most nodes do not need to use the full height of the tower, since the + // probability of each successive level decreases exponentially. Because + // these elements are never accessed, they do not need to be allocated. + // Therefore, when a node is allocated in the arena, its memory footprint + // is deliberately truncated to not include unneeded tower elements. + // + // All accesses to elements should use CAS operations, with no need to lock. + tower [maxHeight]uint32 +} + +// Skiplist maps keys to values (in memory) +type Skiplist struct { + height int32 // Current height. 1 <= height <= kMaxHeight. CAS. + head *node + ref int32 + arena *Arena +} + +// IncrRef increases the refcount +func (s *Skiplist) IncrRef() { + atomic.AddInt32(&s.ref, 1) +} + +// DecrRef decrements the refcount, deallocating the Skiplist when done using it +func (s *Skiplist) DecrRef() { + newRef := atomic.AddInt32(&s.ref, -1) + if newRef > 0 { + return + } + + s.arena.reset() + // Indicate we are closed. Good for testing. Also, lets GC reclaim memory. Race condition + // here would suggest we are accessing skiplist when we are supposed to have no reference! + s.arena = nil +} + +func (s *Skiplist) valid() bool { return s.arena != nil } + +func newNode(arena *Arena, key []byte, v y.ValueStruct, height int) *node { + // The base level is already allocated in the node struct. + offset := arena.putNode(height) + node := arena.getNode(offset) + node.keyOffset = arena.putKey(key) + node.keySize = uint16(len(key)) + node.height = uint16(height) + *node.value64BitAlignedPtr() = encodeValue(arena.putVal(v), v.EncodedSize()) + return node +} + +func encodeValue(valOffset uint32, valSize uint16) uint64 { + return uint64(valSize)<<32 | uint64(valOffset) +} + +func decodeValue(value uint64) (valOffset uint32, valSize uint16) { + valOffset = uint32(value) + valSize = uint16(value >> 32) + return +} + +// NewSkiplist makes a new empty skiplist, with a given arena size +func NewSkiplist(arenaSize int64) *Skiplist { + arena := newArena(arenaSize) + head := newNode(arena, nil, y.ValueStruct{}, maxHeight) + return &Skiplist{ + height: 1, + head: head, + arena: arena, + ref: 1, + } +} + +func (s *node) value64BitAlignedPtr() *uint64 { + if uintptr(unsafe.Pointer(&s.value))%8 == 0 { + return (*uint64)(unsafe.Pointer(&s.value)) + } + return (*uint64)(unsafe.Pointer(&s.value[4])) +} + +func (s *node) getValueOffset() (uint32, uint16) { + value := atomic.LoadUint64(s.value64BitAlignedPtr()) + return decodeValue(value) +} + +func (s *node) key(arena *Arena) []byte { + return arena.getKey(s.keyOffset, s.keySize) +} + +func (s *node) setValue(arena *Arena, v y.ValueStruct) { + valOffset := arena.putVal(v) + value := encodeValue(valOffset, v.EncodedSize()) + atomic.StoreUint64(s.value64BitAlignedPtr(), value) +} + +func (s *node) getNextOffset(h int) uint32 { + return atomic.LoadUint32(&s.tower[h]) +} + +func (s *node) casNextOffset(h int, old, val uint32) bool { + return atomic.CompareAndSwapUint32(&s.tower[h], old, val) +} + +// Returns true if key is strictly > n.key. +// If n is nil, this is an "end" marker and we return false. +//func (s *Skiplist) keyIsAfterNode(key []byte, n *node) bool { +// y.AssertTrue(n != s.head) +// return n != nil && y.CompareKeys(key, n.key) > 0 +//} + +func randomHeight() int { + h := 1 + for h < maxHeight && rand.Uint32() <= heightIncrease { + h++ + } + return h +} + +func (s *Skiplist) getNext(nd *node, height int) *node { + return s.arena.getNode(nd.getNextOffset(height)) +} + +// findNear finds the node near to key. +// If less=true, it finds rightmost node such that node.key < key (if allowEqual=false) or +// node.key <= key (if allowEqual=true). +// If less=false, it finds leftmost node such that node.key > key (if allowEqual=false) or +// node.key >= key (if allowEqual=true). +// Returns the node found. The bool returned is true if the node has key equal to given key. +func (s *Skiplist) findNear(key []byte, less bool, allowEqual bool) (*node, bool) { + x := s.head + level := int(s.getHeight() - 1) + for { + // Assume x.key < key. + next := s.getNext(x, level) + if next == nil { + // x.key < key < END OF LIST + if level > 0 { + // Can descend further to iterate closer to the end. + level-- + continue + } + // Level=0. Cannot descend further. Let's return something that makes sense. + if !less { + return nil, false + } + // Try to return x. Make sure it is not a head node. + if x == s.head { + return nil, false + } + return x, false + } + + nextKey := next.key(s.arena) + cmp := y.CompareKeys(key, nextKey) + if cmp > 0 { + // x.key < next.key < key. We can continue to move right. + x = next + continue + } + if cmp == 0 { + // x.key < key == next.key. + if allowEqual { + return next, true + } + if !less { + // We want >, so go to base level to grab the next bigger note. + return s.getNext(next, 0), false + } + // We want <. If not base level, we should go closer in the next level. + if level > 0 { + level-- + continue + } + // On base level. Return x. + if x == s.head { + return nil, false + } + return x, false + } + // cmp < 0. In other words, x.key < key < next. + if level > 0 { + level-- + continue + } + // At base level. Need to return something. + if !less { + return next, false + } + // Try to return x. Make sure it is not a head node. + if x == s.head { + return nil, false + } + return x, false + } +} + +// findSpliceForLevel returns (outBefore, outAfter) with outBefore.key <= key <= outAfter.key. +// The input "before" tells us where to start looking. +// If we found a node with the same key, then we return outBefore = outAfter. +// Otherwise, outBefore.key < key < outAfter.key. +func (s *Skiplist) findSpliceForLevel(key []byte, before *node, level int) (*node, *node) { + for { + // Assume before.key < key. + next := s.getNext(before, level) + if next == nil { + return before, next + } + nextKey := next.key(s.arena) + cmp := y.CompareKeys(key, nextKey) + if cmp == 0 { + // Equality case. + return next, next + } + if cmp < 0 { + // before.key < key < next.key. We are done for this level. + return before, next + } + before = next // Keep moving right on this level. + } +} + +func (s *Skiplist) getHeight() int32 { + return atomic.LoadInt32(&s.height) +} + +// Put inserts the key-value pair. +func (s *Skiplist) Put(key []byte, v y.ValueStruct) { + // Since we allow overwrite, we may not need to create a new node. We might not even need to + // increase the height. Let's defer these actions. + + listHeight := s.getHeight() + var prev [maxHeight + 1]*node + var next [maxHeight + 1]*node + prev[listHeight] = s.head + next[listHeight] = nil + for i := int(listHeight) - 1; i >= 0; i-- { + // Use higher level to speed up for current level. + prev[i], next[i] = s.findSpliceForLevel(key, prev[i+1], i) + if prev[i] == next[i] { + prev[i].setValue(s.arena, v) + return + } + } + + // We do need to create a new node. + height := randomHeight() + x := newNode(s.arena, key, v, height) + + // Try to increase s.height via CAS. + listHeight = s.getHeight() + for height > int(listHeight) { + if atomic.CompareAndSwapInt32(&s.height, listHeight, int32(height)) { + // Successfully increased skiplist.height. + break + } + listHeight = s.getHeight() + } + + // We always insert from the base level and up. After you add a node in base level, we cannot + // create a node in the level above because it would have discovered the node in the base level. + for i := 0; i < height; i++ { + for { + if prev[i] == nil { + y.AssertTrue(i > 1) // This cannot happen in base level. + // We haven't computed prev, next for this level because height exceeds old listHeight. + // For these levels, we expect the lists to be sparse, so we can just search from head. + prev[i], next[i] = s.findSpliceForLevel(key, s.head, i) + // Someone adds the exact same key before we are able to do so. This can only happen on + // the base level. But we know we are not on the base level. + y.AssertTrue(prev[i] != next[i]) + } + nextOffset := s.arena.getNodeOffset(next[i]) + x.tower[i] = nextOffset + if prev[i].casNextOffset(i, nextOffset, s.arena.getNodeOffset(x)) { + // Managed to insert x between prev[i] and next[i]. Go to the next level. + break + } + // CAS failed. We need to recompute prev and next. + // It is unlikely to be helpful to try to use a different level as we redo the search, + // because it is unlikely that lots of nodes are inserted between prev[i] and next[i]. + prev[i], next[i] = s.findSpliceForLevel(key, prev[i], i) + if prev[i] == next[i] { + y.AssertTruef(i == 0, "Equality can happen only on base level: %d", i) + prev[i].setValue(s.arena, v) + return + } + } + } +} + +// Empty returns if the Skiplist is empty. +func (s *Skiplist) Empty() bool { + return s.findLast() == nil +} + +// findLast returns the last element. If head (empty list), we return nil. All the find functions +// will NEVER return the head nodes. +func (s *Skiplist) findLast() *node { + n := s.head + level := int(s.getHeight()) - 1 + for { + next := s.getNext(n, level) + if next != nil { + n = next + continue + } + if level == 0 { + if n == s.head { + return nil + } + return n + } + level-- + } +} + +// Get gets the value associated with the key. It returns a valid value if it finds equal or earlier +// version of the same key. +func (s *Skiplist) Get(key []byte) y.ValueStruct { + n, _ := s.findNear(key, false, true) // findGreaterOrEqual. + if n == nil { + return y.ValueStruct{} + } + + nextKey := s.arena.getKey(n.keyOffset, n.keySize) + if !y.SameKey(key, nextKey) { + return y.ValueStruct{} + } + + valOffset, valSize := n.getValueOffset() + vs := s.arena.getVal(valOffset, valSize) + vs.Version = y.ParseTs(nextKey) + return vs +} + +// NewIterator returns a skiplist iterator. You have to Close() the iterator. +func (s *Skiplist) NewIterator() *Iterator { + s.IncrRef() + return &Iterator{list: s} +} + +// MemSize returns the size of the Skiplist in terms of how much memory is used within its internal +// arena. +func (s *Skiplist) MemSize() int64 { return s.arena.size() } + +// Iterator is an iterator over skiplist object. For new objects, you just +// need to initialize Iterator.list. +type Iterator struct { + list *Skiplist + n *node +} + +// Close frees the resources held by the iterator +func (s *Iterator) Close() error { + s.list.DecrRef() + return nil +} + +// Valid returns true iff the iterator is positioned at a valid node. +func (s *Iterator) Valid() bool { return s.n != nil } + +// Key returns the key at the current position. +func (s *Iterator) Key() []byte { + return s.list.arena.getKey(s.n.keyOffset, s.n.keySize) +} + +// Value returns value. +func (s *Iterator) Value() y.ValueStruct { + valOffset, valSize := s.n.getValueOffset() + return s.list.arena.getVal(valOffset, valSize) +} + +// Next advances to the next position. +func (s *Iterator) Next() { + y.AssertTrue(s.Valid()) + s.n = s.list.getNext(s.n, 0) +} + +// Prev advances to the previous position. +func (s *Iterator) Prev() { + y.AssertTrue(s.Valid()) + s.n, _ = s.list.findNear(s.Key(), true, false) // find <. No equality allowed. +} + +// Seek advances to the first entry with a key >= target. +func (s *Iterator) Seek(target []byte) { + s.n, _ = s.list.findNear(target, false, true) // find >=. +} + +// SeekForPrev finds an entry with key <= target. +func (s *Iterator) SeekForPrev(target []byte) { + s.n, _ = s.list.findNear(target, true, true) // find <=. +} + +// SeekToFirst seeks position at the first entry in list. +// Final state of iterator is Valid() iff list is not empty. +func (s *Iterator) SeekToFirst() { + s.n = s.list.getNext(s.list.head, 0) +} + +// SeekToLast seeks position at the last entry in list. +// Final state of iterator is Valid() iff list is not empty. +func (s *Iterator) SeekToLast() { + s.n = s.list.findLast() +} + +// UniIterator is a unidirectional memtable iterator. It is a thin wrapper around +// Iterator. We like to keep Iterator as before, because it is more powerful and +// we might support bidirectional iterators in the future. +type UniIterator struct { + iter *Iterator + reversed bool +} + +// NewUniIterator returns a UniIterator. +func (s *Skiplist) NewUniIterator(reversed bool) *UniIterator { + return &UniIterator{ + iter: s.NewIterator(), + reversed: reversed, + } +} + +// Next implements y.Interface +func (s *UniIterator) Next() { + if !s.reversed { + s.iter.Next() + } else { + s.iter.Prev() + } +} + +// Rewind implements y.Interface +func (s *UniIterator) Rewind() { + if !s.reversed { + s.iter.SeekToFirst() + } else { + s.iter.SeekToLast() + } +} + +// Seek implements y.Interface +func (s *UniIterator) Seek(key []byte) { + if !s.reversed { + s.iter.Seek(key) + } else { + s.iter.SeekForPrev(key) + } +} + +// Key implements y.Interface +func (s *UniIterator) Key() []byte { return s.iter.Key() } + +// Value implements y.Interface +func (s *UniIterator) Value() y.ValueStruct { return s.iter.Value() } + +// Valid implements y.Interface +func (s *UniIterator) Valid() bool { return s.iter.Valid() } + +// Close implements y.Interface (and frees up the iter's resources) +func (s *UniIterator) Close() error { return s.iter.Close() } diff --git a/vendor/github.com/dgraph-io/badger/structs.go b/vendor/github.com/dgraph-io/badger/structs.go new file mode 100644 index 0000000000..09547a42f6 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/structs.go @@ -0,0 +1,132 @@ +package badger + +import ( + "bytes" + "encoding/binary" + "fmt" + "hash/crc32" + + "github.com/dgraph-io/badger/y" +) + +type valuePointer struct { + Fid uint32 + Len uint32 + Offset uint32 +} + +func (p valuePointer) Less(o valuePointer) bool { + if p.Fid != o.Fid { + return p.Fid < o.Fid + } + if p.Offset != o.Offset { + return p.Offset < o.Offset + } + return p.Len < o.Len +} + +func (p valuePointer) IsZero() bool { + return p.Fid == 0 && p.Offset == 0 && p.Len == 0 +} + +const vptrSize = 12 + +// Encode encodes Pointer into byte buffer. +func (p valuePointer) Encode(b []byte) []byte { + binary.BigEndian.PutUint32(b[:4], p.Fid) + binary.BigEndian.PutUint32(b[4:8], p.Len) + binary.BigEndian.PutUint32(b[8:12], p.Offset) + return b[:vptrSize] +} + +func (p *valuePointer) Decode(b []byte) { + p.Fid = binary.BigEndian.Uint32(b[:4]) + p.Len = binary.BigEndian.Uint32(b[4:8]) + p.Offset = binary.BigEndian.Uint32(b[8:12]) +} + +// header is used in value log as a header before Entry. +type header struct { + klen uint32 + vlen uint32 + expiresAt uint64 + meta byte + userMeta byte +} + +const ( + headerBufSize = 18 +) + +func (h header) Encode(out []byte) { + y.AssertTrue(len(out) >= headerBufSize) + binary.BigEndian.PutUint32(out[0:4], h.klen) + binary.BigEndian.PutUint32(out[4:8], h.vlen) + binary.BigEndian.PutUint64(out[8:16], h.expiresAt) + out[16] = h.meta + out[17] = h.userMeta +} + +// Decodes h from buf. +func (h *header) Decode(buf []byte) { + h.klen = binary.BigEndian.Uint32(buf[0:4]) + h.vlen = binary.BigEndian.Uint32(buf[4:8]) + h.expiresAt = binary.BigEndian.Uint64(buf[8:16]) + h.meta = buf[16] + h.userMeta = buf[17] +} + +// Entry provides Key, Value, UserMeta and ExpiresAt. This struct can be used by the user to set data. +type Entry struct { + Key []byte + Value []byte + UserMeta byte + ExpiresAt uint64 // time.Unix + meta byte + + // Fields maintained internally. + offset uint32 +} + +func (e *Entry) estimateSize(threshold int) int { + if len(e.Value) < threshold { + return len(e.Key) + len(e.Value) + 2 // Meta, UserMeta + } + return len(e.Key) + 12 + 2 // 12 for ValuePointer, 2 for metas. +} + +// Encodes e to buf. Returns number of bytes written. +func encodeEntry(e *Entry, buf *bytes.Buffer) (int, error) { + h := header{ + klen: uint32(len(e.Key)), + vlen: uint32(len(e.Value)), + expiresAt: e.ExpiresAt, + meta: e.meta, + userMeta: e.UserMeta, + } + + var headerEnc [headerBufSize]byte + h.Encode(headerEnc[:]) + + hash := crc32.New(y.CastagnoliCrcTable) + + buf.Write(headerEnc[:]) + hash.Write(headerEnc[:]) + + buf.Write(e.Key) + hash.Write(e.Key) + + buf.Write(e.Value) + hash.Write(e.Value) + + var crcBuf [4]byte + binary.BigEndian.PutUint32(crcBuf[:], hash.Sum32()) + buf.Write(crcBuf[:]) + + return len(headerEnc) + len(e.Key) + len(e.Value) + len(crcBuf), nil +} + +func (e Entry) print(prefix string) { + fmt.Printf("%s Key: %s Meta: %d UserMeta: %d Offset: %d len(val)=%d", + prefix, e.Key, e.meta, e.UserMeta, e.offset, len(e.Value)) +} diff --git a/vendor/github.com/dgraph-io/badger/table/README.md b/vendor/github.com/dgraph-io/badger/table/README.md new file mode 100644 index 0000000000..5d33e96ab5 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/table/README.md @@ -0,0 +1,51 @@ +# BenchmarkRead + +``` +$ go test -bench Read$ -count 3 + +Size of table: 105843444 +BenchmarkRead-8 3 343846914 ns/op +BenchmarkRead-8 3 351790907 ns/op +BenchmarkRead-8 3 351762823 ns/op +``` + +Size of table is 105,843,444 bytes, which is ~101M. + +The rate is ~287M/s which matches our read speed. This is using mmap. + +To read a 64M table, this would take ~0.22s, which is negligible. + +``` +$ go test -bench BenchmarkReadAndBuild -count 3 + +BenchmarkReadAndBuild-8 1 2341034225 ns/op +BenchmarkReadAndBuild-8 1 2346349671 ns/op +BenchmarkReadAndBuild-8 1 2364064576 ns/op +``` + +The rate is ~43M/s. To build a ~64M table, this would take ~1.5s. Note that this +does NOT include the flushing of the table to disk. All we are doing above is +to read one table (mmaped) and write one table in memory. + +The table building takes 1.5-0.22 ~ 1.3s. + +If we are writing out up to 10 tables, this would take 1.5*10 ~ 15s, and ~13s +is spent building the tables. + +When running populate, building one table in memory tends to take ~1.5s to ~2.5s +on my system. Where does this overhead come from? Let's investigate the merging. + +Below, we merge 5 tables. The total size remains unchanged at ~101M. + +``` +$ go test -bench ReadMerged -count 3 +BenchmarkReadMerged-8 1 1321190264 ns/op +BenchmarkReadMerged-8 1 1296958737 ns/op +BenchmarkReadMerged-8 1 1314381178 ns/op +``` + +The rate is ~76M/s. To build a 64M table, this would take ~0.84s. The writing +takes ~1.3s as we saw above. So in total, we expect around 0.84+1.3 ~ 2.1s. +This roughly matches what we observe when running populate. There might be +some additional overhead due to the concurrent writes going on, in flushing the +table to disk. Also, the tables tend to be slightly bigger than 64M/s. \ No newline at end of file diff --git a/vendor/github.com/dgraph-io/badger/table/builder.go b/vendor/github.com/dgraph-io/badger/table/builder.go new file mode 100644 index 0000000000..43e6562239 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/table/builder.go @@ -0,0 +1,235 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package table + +import ( + "bytes" + "encoding/binary" + "io" + "math" + + "github.com/AndreasBriese/bbloom" + "github.com/dgraph-io/badger/y" +) + +var ( + restartInterval = 100 // Might want to change this to be based on total size instead of numKeys. +) + +func newBuffer(sz int) *bytes.Buffer { + b := new(bytes.Buffer) + b.Grow(sz) + return b +} + +type header struct { + plen uint16 // Overlap with base key. + klen uint16 // Length of the diff. + vlen uint16 // Length of value. + prev uint32 // Offset for the previous key-value pair. The offset is relative to block base offset. +} + +// Encode encodes the header. +func (h header) Encode(b []byte) { + binary.BigEndian.PutUint16(b[0:2], h.plen) + binary.BigEndian.PutUint16(b[2:4], h.klen) + binary.BigEndian.PutUint16(b[4:6], h.vlen) + binary.BigEndian.PutUint32(b[6:10], h.prev) +} + +// Decode decodes the header. +func (h *header) Decode(buf []byte) int { + h.plen = binary.BigEndian.Uint16(buf[0:2]) + h.klen = binary.BigEndian.Uint16(buf[2:4]) + h.vlen = binary.BigEndian.Uint16(buf[4:6]) + h.prev = binary.BigEndian.Uint32(buf[6:10]) + return h.Size() +} + +// Size returns size of the header. Currently it's just a constant. +func (h header) Size() int { return 10 } + +// Builder is used in building a table. +type Builder struct { + counter int // Number of keys written for the current block. + + // Typically tens or hundreds of meg. This is for one single file. + buf *bytes.Buffer + + baseKey []byte // Base key for the current block. + baseOffset uint32 // Offset for the current block. + + restarts []uint32 // Base offsets of every block. + + // Tracks offset for the previous key-value pair. Offset is relative to block base offset. + prevOffset uint32 + + keyBuf *bytes.Buffer + keyCount int +} + +// NewTableBuilder makes a new TableBuilder. +func NewTableBuilder() *Builder { + return &Builder{ + keyBuf: newBuffer(1 << 20), + buf: newBuffer(1 << 20), + prevOffset: math.MaxUint32, // Used for the first element! + } +} + +// Close closes the TableBuilder. +func (b *Builder) Close() {} + +// Empty returns whether it's empty. +func (b *Builder) Empty() bool { return b.buf.Len() == 0 } + +// keyDiff returns a suffix of newKey that is different from b.baseKey. +func (b Builder) keyDiff(newKey []byte) []byte { + var i int + for i = 0; i < len(newKey) && i < len(b.baseKey); i++ { + if newKey[i] != b.baseKey[i] { + break + } + } + return newKey[i:] +} + +func (b *Builder) addHelper(key []byte, v y.ValueStruct) { + // Add key to bloom filter. + if len(key) > 0 { + var klen [2]byte + keyNoTs := y.ParseKey(key) + binary.BigEndian.PutUint16(klen[:], uint16(len(keyNoTs))) + b.keyBuf.Write(klen[:]) + b.keyBuf.Write(keyNoTs) + b.keyCount++ + } + + // diffKey stores the difference of key with baseKey. + var diffKey []byte + if len(b.baseKey) == 0 { + // Make a copy. Builder should not keep references. Otherwise, caller has to be very careful + // and will have to make copies of keys every time they add to builder, which is even worse. + b.baseKey = append(b.baseKey[:0], key...) + diffKey = key + } else { + diffKey = b.keyDiff(key) + } + + h := header{ + plen: uint16(len(key) - len(diffKey)), + klen: uint16(len(diffKey)), + vlen: uint16(v.EncodedSize()), + prev: b.prevOffset, // prevOffset is the location of the last key-value added. + } + b.prevOffset = uint32(b.buf.Len()) - b.baseOffset // Remember current offset for the next Add call. + + // Layout: header, diffKey, value. + var hbuf [10]byte + h.Encode(hbuf[:]) + b.buf.Write(hbuf[:]) + b.buf.Write(diffKey) // We only need to store the key difference. + + v.EncodeTo(b.buf) + b.counter++ // Increment number of keys added for this current block. +} + +func (b *Builder) finishBlock() { + // When we are at the end of the block and Valid=false, and the user wants to do a Prev, + // we need a dummy header to tell us the offset of the previous key-value pair. + b.addHelper([]byte{}, y.ValueStruct{}) +} + +// Add adds a key-value pair to the block. +// If doNotRestart is true, we will not restart even if b.counter >= restartInterval. +func (b *Builder) Add(key []byte, value y.ValueStruct) error { + if b.counter >= restartInterval { + b.finishBlock() + // Start a new block. Initialize the block. + b.restarts = append(b.restarts, uint32(b.buf.Len())) + b.counter = 0 + b.baseKey = []byte{} + b.baseOffset = uint32(b.buf.Len()) + b.prevOffset = math.MaxUint32 // First key-value pair of block has header.prev=MaxInt. + } + b.addHelper(key, value) + return nil // Currently, there is no meaningful error. +} + +// TODO: vvv this was the comment on ReachedCapacity. +// FinalSize returns the *rough* final size of the array, counting the header which is not yet written. +// TODO: Look into why there is a discrepancy. I suspect it is because of Write(empty, empty) +// at the end. The diff can vary. + +// ReachedCapacity returns true if we... roughly (?) reached capacity? +func (b *Builder) ReachedCapacity(cap int64) bool { + estimateSz := b.buf.Len() + 8 /* empty header */ + 4*len(b.restarts) + 8 // 8 = end of buf offset + len(restarts). + return int64(estimateSz) > cap +} + +// blockIndex generates the block index for the table. +// It is mainly a list of all the block base offsets. +func (b *Builder) blockIndex() []byte { + // Store the end offset, so we know the length of the final block. + b.restarts = append(b.restarts, uint32(b.buf.Len())) + + // Add 4 because we want to write out number of restarts at the end. + sz := 4*len(b.restarts) + 4 + out := make([]byte, sz) + buf := out + for _, r := range b.restarts { + binary.BigEndian.PutUint32(buf[:4], r) + buf = buf[4:] + } + binary.BigEndian.PutUint32(buf[:4], uint32(len(b.restarts))) + return out +} + +// Finish finishes the table by appending the index. +func (b *Builder) Finish() []byte { + bf := bbloom.New(float64(b.keyCount), 0.01) + var klen [2]byte + key := make([]byte, 1024) + for { + if _, err := b.keyBuf.Read(klen[:]); err == io.EOF { + break + } else if err != nil { + y.Check(err) + } + kl := int(binary.BigEndian.Uint16(klen[:])) + if cap(key) < kl { + key = make([]byte, 2*int(kl)) // 2 * uint16 will overflow + } + key = key[:kl] + y.Check2(b.keyBuf.Read(key)) + bf.Add(key) + } + + b.finishBlock() // This will never start a new block. + index := b.blockIndex() + b.buf.Write(index) + + // Write bloom filter. + bdata := bf.JSONMarshal() + n, err := b.buf.Write(bdata) + y.Check(err) + var buf [4]byte + binary.BigEndian.PutUint32(buf[:], uint32(n)) + b.buf.Write(buf[:]) + + return b.buf.Bytes() +} diff --git a/vendor/github.com/dgraph-io/badger/table/iterator.go b/vendor/github.com/dgraph-io/badger/table/iterator.go new file mode 100644 index 0000000000..0eb5ed01a9 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/table/iterator.go @@ -0,0 +1,539 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package table + +import ( + "bytes" + "io" + "math" + "sort" + + "github.com/dgraph-io/badger/y" + "github.com/pkg/errors" +) + +type blockIterator struct { + data []byte + pos uint32 + err error + baseKey []byte + + key []byte + val []byte + init bool + + last header // The last header we saw. +} + +func (itr *blockIterator) Reset() { + itr.pos = 0 + itr.err = nil + itr.baseKey = []byte{} + itr.key = []byte{} + itr.val = []byte{} + itr.init = false + itr.last = header{} +} + +func (itr *blockIterator) Init() { + if !itr.init { + itr.Next() + } +} + +func (itr *blockIterator) Valid() bool { + return itr != nil && itr.err == nil +} + +func (itr *blockIterator) Error() error { + return itr.err +} + +func (itr *blockIterator) Close() {} + +var ( + origin = 0 + current = 1 +) + +// Seek brings us to the first block element that is >= input key. +func (itr *blockIterator) Seek(key []byte, whence int) { + itr.err = nil + + switch whence { + case origin: + itr.Reset() + case current: + } + + var done bool + for itr.Init(); itr.Valid(); itr.Next() { + k := itr.Key() + if y.CompareKeys(k, key) >= 0 { + // We are done as k is >= key. + done = true + break + } + } + if !done { + itr.err = io.EOF + } +} + +func (itr *blockIterator) SeekToFirst() { + itr.err = nil + itr.Init() +} + +// SeekToLast brings us to the last element. Valid should return true. +func (itr *blockIterator) SeekToLast() { + itr.err = nil + for itr.Init(); itr.Valid(); itr.Next() { + } + itr.Prev() +} + +// parseKV would allocate a new byte slice for key and for value. +func (itr *blockIterator) parseKV(h header) { + if cap(itr.key) < int(h.plen+h.klen) { + sz := int(h.plen) + int(h.klen) // Convert to int before adding to avoid uint16 overflow. + itr.key = make([]byte, 2*sz) + } + itr.key = itr.key[:h.plen+h.klen] + copy(itr.key, itr.baseKey[:h.plen]) + copy(itr.key[h.plen:], itr.data[itr.pos:itr.pos+uint32(h.klen)]) + itr.pos += uint32(h.klen) + + if itr.pos+uint32(h.vlen) > uint32(len(itr.data)) { + itr.err = errors.Errorf("Value exceeded size of block: %d %d %d %d %v", + itr.pos, h.klen, h.vlen, len(itr.data), h) + return + } + itr.val = y.SafeCopy(itr.val, itr.data[itr.pos:itr.pos+uint32(h.vlen)]) + itr.pos += uint32(h.vlen) +} + +func (itr *blockIterator) Next() { + itr.init = true + itr.err = nil + if itr.pos >= uint32(len(itr.data)) { + itr.err = io.EOF + return + } + + var h header + itr.pos += uint32(h.Decode(itr.data[itr.pos:])) + itr.last = h // Store the last header. + + if h.klen == 0 && h.plen == 0 { + // Last entry in the table. + itr.err = io.EOF + return + } + + // Populate baseKey if it isn't set yet. This would only happen for the first Next. + if len(itr.baseKey) == 0 { + // This should be the first Next() for this block. Hence, prefix length should be zero. + y.AssertTrue(h.plen == 0) + itr.baseKey = itr.data[itr.pos : itr.pos+uint32(h.klen)] + } + itr.parseKV(h) +} + +func (itr *blockIterator) Prev() { + if !itr.init { + return + } + itr.err = nil + if itr.last.prev == math.MaxUint32 { + // This is the first element of the block! + itr.err = io.EOF + itr.pos = 0 + return + } + + // Move back using current header's prev. + itr.pos = itr.last.prev + + var h header + y.AssertTruef(itr.pos < uint32(len(itr.data)), "%d %d", itr.pos, len(itr.data)) + itr.pos += uint32(h.Decode(itr.data[itr.pos:])) + itr.parseKV(h) + itr.last = h +} + +func (itr *blockIterator) Key() []byte { + if itr.err != nil { + return nil + } + return itr.key +} + +func (itr *blockIterator) Value() []byte { + if itr.err != nil { + return nil + } + return itr.val +} + +// Iterator is an iterator for a Table. +type Iterator struct { + t *Table + bpos int + bi *blockIterator + err error + + // Internally, Iterator is bidirectional. However, we only expose the + // unidirectional functionality for now. + reversed bool +} + +// NewIterator returns a new iterator of the Table +func (t *Table) NewIterator(reversed bool) *Iterator { + t.IncrRef() // Important. + ti := &Iterator{t: t, reversed: reversed} + ti.next() + return ti +} + +// Close closes the iterator (and it must be called). +func (itr *Iterator) Close() error { + return itr.t.DecrRef() +} + +func (itr *Iterator) reset() { + itr.bpos = 0 + itr.err = nil +} + +// Valid follows the y.Iterator interface +func (itr *Iterator) Valid() bool { + return itr.err == nil +} + +func (itr *Iterator) seekToFirst() { + numBlocks := len(itr.t.blockIndex) + if numBlocks == 0 { + itr.err = io.EOF + return + } + itr.bpos = 0 + block, err := itr.t.block(itr.bpos) + if err != nil { + itr.err = err + return + } + itr.bi = block.NewIterator() + itr.bi.SeekToFirst() + itr.err = itr.bi.Error() +} + +func (itr *Iterator) seekToLast() { + numBlocks := len(itr.t.blockIndex) + if numBlocks == 0 { + itr.err = io.EOF + return + } + itr.bpos = numBlocks - 1 + block, err := itr.t.block(itr.bpos) + if err != nil { + itr.err = err + return + } + itr.bi = block.NewIterator() + itr.bi.SeekToLast() + itr.err = itr.bi.Error() +} + +func (itr *Iterator) seekHelper(blockIdx int, key []byte) { + itr.bpos = blockIdx + block, err := itr.t.block(blockIdx) + if err != nil { + itr.err = err + return + } + itr.bi = block.NewIterator() + itr.bi.Seek(key, origin) + itr.err = itr.bi.Error() +} + +// seekFrom brings us to a key that is >= input key. +func (itr *Iterator) seekFrom(key []byte, whence int) { + itr.err = nil + switch whence { + case origin: + itr.reset() + case current: + } + + idx := sort.Search(len(itr.t.blockIndex), func(idx int) bool { + ko := itr.t.blockIndex[idx] + return y.CompareKeys(ko.key, key) > 0 + }) + if idx == 0 { + // The smallest key in our table is already strictly > key. We can return that. + // This is like a SeekToFirst. + itr.seekHelper(0, key) + return + } + + // block[idx].smallest is > key. + // Since idx>0, we know block[idx-1].smallest is <= key. + // There are two cases. + // 1) Everything in block[idx-1] is strictly < key. In this case, we should go to the first + // element of block[idx]. + // 2) Some element in block[idx-1] is >= key. We should go to that element. + itr.seekHelper(idx-1, key) + if itr.err == io.EOF { + // Case 1. Need to visit block[idx]. + if idx == len(itr.t.blockIndex) { + // If idx == len(itr.t.blockIndex), then input key is greater than ANY element of table. + // There's nothing we can do. Valid() should return false as we seek to end of table. + return + } + // Since block[idx].smallest is > key. This is essentially a block[idx].SeekToFirst. + itr.seekHelper(idx, key) + } + // Case 2: No need to do anything. We already did the seek in block[idx-1]. +} + +// seek will reset iterator and seek to >= key. +func (itr *Iterator) seek(key []byte) { + itr.seekFrom(key, origin) +} + +// seekForPrev will reset iterator and seek to <= key. +func (itr *Iterator) seekForPrev(key []byte) { + // TODO: Optimize this. We shouldn't have to take a Prev step. + itr.seekFrom(key, origin) + if !bytes.Equal(itr.Key(), key) { + itr.prev() + } +} + +func (itr *Iterator) next() { + itr.err = nil + + if itr.bpos >= len(itr.t.blockIndex) { + itr.err = io.EOF + return + } + + if itr.bi == nil { + block, err := itr.t.block(itr.bpos) + if err != nil { + itr.err = err + return + } + itr.bi = block.NewIterator() + itr.bi.SeekToFirst() + itr.err = itr.bi.Error() + return + } + + itr.bi.Next() + if !itr.bi.Valid() { + itr.bpos++ + itr.bi = nil + itr.next() + return + } +} + +func (itr *Iterator) prev() { + itr.err = nil + if itr.bpos < 0 { + itr.err = io.EOF + return + } + + if itr.bi == nil { + block, err := itr.t.block(itr.bpos) + if err != nil { + itr.err = err + return + } + itr.bi = block.NewIterator() + itr.bi.SeekToLast() + itr.err = itr.bi.Error() + return + } + + itr.bi.Prev() + if !itr.bi.Valid() { + itr.bpos-- + itr.bi = nil + itr.prev() + return + } +} + +// Key follows the y.Iterator interface +func (itr *Iterator) Key() []byte { + return itr.bi.Key() +} + +// Value follows the y.Iterator interface +func (itr *Iterator) Value() (ret y.ValueStruct) { + ret.Decode(itr.bi.Value()) + return +} + +// Next follows the y.Iterator interface +func (itr *Iterator) Next() { + if !itr.reversed { + itr.next() + } else { + itr.prev() + } +} + +// Rewind follows the y.Iterator interface +func (itr *Iterator) Rewind() { + if !itr.reversed { + itr.seekToFirst() + } else { + itr.seekToLast() + } +} + +// Seek follows the y.Iterator interface +func (itr *Iterator) Seek(key []byte) { + if !itr.reversed { + itr.seek(key) + } else { + itr.seekForPrev(key) + } +} + +// ConcatIterator concatenates the sequences defined by several iterators. (It only works with +// TableIterators, probably just because it's faster to not be so generic.) +type ConcatIterator struct { + idx int // Which iterator is active now. + cur *Iterator + iters []*Iterator // Corresponds to tables. + tables []*Table // Disregarding reversed, this is in ascending order. + reversed bool +} + +// NewConcatIterator creates a new concatenated iterator +func NewConcatIterator(tbls []*Table, reversed bool) *ConcatIterator { + iters := make([]*Iterator, len(tbls)) + for i := 0; i < len(tbls); i++ { + iters[i] = tbls[i].NewIterator(reversed) + } + return &ConcatIterator{ + reversed: reversed, + iters: iters, + tables: tbls, + idx: -1, // Not really necessary because s.it.Valid()=false, but good to have. + } +} + +func (s *ConcatIterator) setIdx(idx int) { + s.idx = idx + if idx < 0 || idx >= len(s.iters) { + s.cur = nil + } else { + s.cur = s.iters[s.idx] + } +} + +// Rewind implements y.Interface +func (s *ConcatIterator) Rewind() { + if len(s.iters) == 0 { + return + } + if !s.reversed { + s.setIdx(0) + } else { + s.setIdx(len(s.iters) - 1) + } + s.cur.Rewind() +} + +// Valid implements y.Interface +func (s *ConcatIterator) Valid() bool { + return s.cur != nil && s.cur.Valid() +} + +// Key implements y.Interface +func (s *ConcatIterator) Key() []byte { + return s.cur.Key() +} + +// Value implements y.Interface +func (s *ConcatIterator) Value() y.ValueStruct { + return s.cur.Value() +} + +// Seek brings us to element >= key if reversed is false. Otherwise, <= key. +func (s *ConcatIterator) Seek(key []byte) { + var idx int + if !s.reversed { + idx = sort.Search(len(s.tables), func(i int) bool { + return y.CompareKeys(s.tables[i].Biggest(), key) >= 0 + }) + } else { + n := len(s.tables) + idx = n - 1 - sort.Search(n, func(i int) bool { + return y.CompareKeys(s.tables[n-1-i].Smallest(), key) <= 0 + }) + } + if idx >= len(s.tables) || idx < 0 { + s.setIdx(-1) + return + } + // For reversed=false, we know s.tables[i-1].Biggest() < key. Thus, the + // previous table cannot possibly contain key. + s.setIdx(idx) + s.cur.Seek(key) +} + +// Next advances our concat iterator. +func (s *ConcatIterator) Next() { + s.cur.Next() + if s.cur.Valid() { + // Nothing to do. Just stay with the current table. + return + } + for { // In case there are empty tables. + if !s.reversed { + s.setIdx(s.idx + 1) + } else { + s.setIdx(s.idx - 1) + } + if s.cur == nil { + // End of list. Valid will become false. + return + } + s.cur.Rewind() + if s.cur.Valid() { + break + } + } +} + +// Close implements y.Interface. +func (s *ConcatIterator) Close() error { + for _, it := range s.iters { + if err := it.Close(); err != nil { + return errors.Wrap(err, "ConcatIterator") + } + } + return nil +} diff --git a/vendor/github.com/dgraph-io/badger/table/table.go b/vendor/github.com/dgraph-io/badger/table/table.go new file mode 100644 index 0000000000..9804fa1764 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/table/table.go @@ -0,0 +1,359 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package table + +import ( + "encoding/binary" + "fmt" + "os" + "path" + "path/filepath" + "strconv" + "strings" + "sync" + "sync/atomic" + + "github.com/AndreasBriese/bbloom" + "github.com/dgraph-io/badger/options" + "github.com/dgraph-io/badger/y" + "github.com/pkg/errors" +) + +const fileSuffix = ".sst" + +type keyOffset struct { + key []byte + offset int + len int +} + +// Table represents a loaded table file with the info we have about it +type Table struct { + sync.Mutex + + fd *os.File // Own fd. + tableSize int // Initialized in OpenTable, using fd.Stat(). + + blockIndex []keyOffset + ref int32 // For file garbage collection. Atomic. + + loadingMode options.FileLoadingMode + mmap []byte // Memory mapped. + + // The following are initialized once and const. + smallest, biggest []byte // Smallest and largest keys. + id uint64 // file id, part of filename + + bf bbloom.Bloom +} + +// IncrRef increments the refcount (having to do with whether the file should be deleted) +func (t *Table) IncrRef() { + atomic.AddInt32(&t.ref, 1) +} + +// DecrRef decrements the refcount and possibly deletes the table +func (t *Table) DecrRef() error { + newRef := atomic.AddInt32(&t.ref, -1) + if newRef == 0 { + // We can safely delete this file, because for all the current files, we always have + // at least one reference pointing to them. + + // It's necessary to delete windows files + if t.loadingMode == options.MemoryMap { + y.Munmap(t.mmap) + } + if err := t.fd.Truncate(0); err != nil { + // This is very important to let the FS know that the file is deleted. + return err + } + filename := t.fd.Name() + if err := t.fd.Close(); err != nil { + return err + } + if err := os.Remove(filename); err != nil { + return err + } + } + return nil +} + +type block struct { + offset int + data []byte +} + +func (b block) NewIterator() *blockIterator { + return &blockIterator{data: b.data} +} + +// OpenTable assumes file has only one table and opens it. Takes ownership of fd upon function +// entry. Returns a table with one reference count on it (decrementing which may delete the file! +// -- consider t.Close() instead). The fd has to writeable because we call Truncate on it before +// deleting. +func OpenTable(fd *os.File, loadingMode options.FileLoadingMode) (*Table, error) { + fileInfo, err := fd.Stat() + if err != nil { + // It's OK to ignore fd.Close() errs in this function because we have only read + // from the file. + _ = fd.Close() + return nil, y.Wrap(err) + } + + filename := fileInfo.Name() + id, ok := ParseFileID(filename) + if !ok { + _ = fd.Close() + return nil, errors.Errorf("Invalid filename: %s", filename) + } + t := &Table{ + fd: fd, + ref: 1, // Caller is given one reference. + id: id, + loadingMode: loadingMode, + } + + t.tableSize = int(fileInfo.Size()) + + if loadingMode == options.MemoryMap { + t.mmap, err = y.Mmap(fd, false, fileInfo.Size()) + if err != nil { + _ = fd.Close() + return nil, y.Wrapf(err, "Unable to map file") + } + } else if loadingMode == options.LoadToRAM { + err = t.loadToRAM() + if err != nil { + _ = fd.Close() + return nil, y.Wrap(err) + } + } + + if err := t.readIndex(); err != nil { + return nil, y.Wrap(err) + } + + it := t.NewIterator(false) + defer it.Close() + it.Rewind() + if it.Valid() { + t.smallest = it.Key() + } + + it2 := t.NewIterator(true) + defer it2.Close() + it2.Rewind() + if it2.Valid() { + t.biggest = it2.Key() + } + return t, nil +} + +// Close closes the open table. (Releases resources back to the OS.) +func (t *Table) Close() error { + if t.loadingMode == options.MemoryMap { + y.Munmap(t.mmap) + } + + return t.fd.Close() +} + +func (t *Table) read(off int, sz int) ([]byte, error) { + if len(t.mmap) > 0 { + if len(t.mmap[off:]) < sz { + return nil, y.ErrEOF + } + return t.mmap[off : off+sz], nil + } + + res := make([]byte, sz) + nbr, err := t.fd.ReadAt(res, int64(off)) + y.NumReads.Add(1) + y.NumBytesRead.Add(int64(nbr)) + return res, err +} + +func (t *Table) readNoFail(off int, sz int) []byte { + res, err := t.read(off, sz) + y.Check(err) + return res +} + +func (t *Table) readIndex() error { + readPos := t.tableSize + + // Read bloom filter. + readPos -= 4 + buf := t.readNoFail(readPos, 4) + bloomLen := int(binary.BigEndian.Uint32(buf)) + readPos -= bloomLen + data := t.readNoFail(readPos, bloomLen) + t.bf = bbloom.JSONUnmarshal(data) + + readPos -= 4 + buf = t.readNoFail(readPos, 4) + restartsLen := int(binary.BigEndian.Uint32(buf)) + + readPos -= 4 * restartsLen + buf = t.readNoFail(readPos, 4*restartsLen) + + offsets := make([]int, restartsLen) + for i := 0; i < restartsLen; i++ { + offsets[i] = int(binary.BigEndian.Uint32(buf[:4])) + buf = buf[4:] + } + + // The last offset stores the end of the last block. + for i := 0; i < len(offsets); i++ { + var o int + if i == 0 { + o = 0 + } else { + o = offsets[i-1] + } + + ko := keyOffset{ + offset: o, + len: offsets[i] - o, + } + t.blockIndex = append(t.blockIndex, ko) + } + + che := make(chan error, len(t.blockIndex)) + blocks := make(chan int, len(t.blockIndex)) + + for i := 0; i < len(t.blockIndex); i++ { + blocks <- i + } + + for i := 0; i < 64; i++ { // Run 64 goroutines. + go func() { + var h header + + for index := range blocks { + ko := &t.blockIndex[index] + + offset := ko.offset + buf, err := t.read(offset, h.Size()) + if err != nil { + che <- errors.Wrap(err, "While reading first header in block") + continue + } + + h.Decode(buf) + y.AssertTruef(h.plen == 0, "Key offset: %+v, h.plen = %d", *ko, h.plen) + + offset += h.Size() + buf = make([]byte, h.klen) + var out []byte + if out, err = t.read(offset, int(h.klen)); err != nil { + che <- errors.Wrap(err, "While reading first key in block") + continue + } + y.AssertTrue(len(buf) == copy(buf, out)) + + ko.key = buf + che <- nil + } + }() + } + close(blocks) // to stop reading goroutines + + var readError error + for i := 0; i < len(t.blockIndex); i++ { + if err := <-che; err != nil && readError == nil { + readError = err + } + } + if readError != nil { + return readError + } + + return nil +} + +func (t *Table) block(idx int) (block, error) { + y.AssertTruef(idx >= 0, "idx=%d", idx) + if idx >= len(t.blockIndex) { + return block{}, errors.New("block out of index") + } + + ko := t.blockIndex[idx] + blk := block{ + offset: ko.offset, + } + var err error + blk.data, err = t.read(blk.offset, ko.len) + return blk, err +} + +// Size is its file size in bytes +func (t *Table) Size() int64 { return int64(t.tableSize) } + +// Smallest is its smallest key, or nil if there are none +func (t *Table) Smallest() []byte { return t.smallest } + +// Biggest is its biggest key, or nil if there are none +func (t *Table) Biggest() []byte { return t.biggest } + +// Filename is NOT the file name. Just kidding, it is. +func (t *Table) Filename() string { return t.fd.Name() } + +// ID is the table's ID number (used to make the file name). +func (t *Table) ID() uint64 { return t.id } + +// DoesNotHave returns true if (but not "only if") the table does not have the key. It does a +// bloom filter lookup. +func (t *Table) DoesNotHave(key []byte) bool { return !t.bf.Has(key) } + +// ParseFileID reads the file id out of a filename. +func ParseFileID(name string) (uint64, bool) { + name = path.Base(name) + if !strings.HasSuffix(name, fileSuffix) { + return 0, false + } + // suffix := name[len(fileSuffix):] + name = strings.TrimSuffix(name, fileSuffix) + id, err := strconv.Atoi(name) + if err != nil { + return 0, false + } + y.AssertTrue(id >= 0) + return uint64(id), true +} + +// IDToFilename does the inverse of ParseFileID +func IDToFilename(id uint64) string { + return fmt.Sprintf("%06d", id) + fileSuffix +} + +// NewFilename should be named TableFilepath -- it combines the dir with the ID to make a table +// filepath. +func NewFilename(id uint64, dir string) string { + return filepath.Join(dir, IDToFilename(id)) +} + +func (t *Table) loadToRAM() error { + t.mmap = make([]byte, t.tableSize) + read, err := t.fd.ReadAt(t.mmap, 0) + if err != nil || read != t.tableSize { + return y.Wrapf(err, "Unable to load file in memory. Table file: %s", t.Filename()) + } + y.NumReads.Add(1) + y.NumBytesRead.Add(int64(read)) + return nil +} diff --git a/vendor/github.com/dgraph-io/badger/test.sh b/vendor/github.com/dgraph-io/badger/test.sh new file mode 100755 index 0000000000..a0c7f0944f --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/test.sh @@ -0,0 +1,5 @@ +l=$(go list ./...) +for x in $l; do + echo "Testing package $x" + go test -v $x +done diff --git a/vendor/github.com/dgraph-io/badger/transaction.go b/vendor/github.com/dgraph-io/badger/transaction.go new file mode 100644 index 0000000000..fcd96ca8b9 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/transaction.go @@ -0,0 +1,546 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "bytes" + "math" + "sort" + "strconv" + "sync" + "sync/atomic" + "time" + + "github.com/dgraph-io/badger/y" + farm "github.com/dgryski/go-farm" + "github.com/pkg/errors" +) + +type uint64Heap []uint64 + +func (u uint64Heap) Len() int { return len(u) } +func (u uint64Heap) Less(i int, j int) bool { return u[i] < u[j] } +func (u uint64Heap) Swap(i int, j int) { u[i], u[j] = u[j], u[i] } +func (u *uint64Heap) Push(x interface{}) { *u = append(*u, x.(uint64)) } +func (u *uint64Heap) Pop() interface{} { + old := *u + n := len(old) + x := old[n-1] + *u = old[0 : n-1] + return x +} + +type oracle struct { + curRead uint64 // Managed by the mutex. + refCount int64 + isManaged bool // Does not change value, so no locking required. + + sync.Mutex + writeLock sync.Mutex + nextCommit uint64 + + // commits stores a key fingerprint and latest commit counter for it. + // refCount is used to clear out commits map to avoid a memory blowup. + commits map[uint64]uint64 +} + +func (o *oracle) addRef() { + atomic.AddInt64(&o.refCount, 1) +} + +func (o *oracle) decrRef() { + if count := atomic.AddInt64(&o.refCount, -1); count == 0 { + // Clear out commits maps to release memory. + o.Lock() + // Avoids the race where something new is added to commitsMap + // after we check refCount and before we take Lock. + if atomic.LoadInt64(&o.refCount) != 0 { + o.Unlock() + return + } + if len(o.commits) >= 1000 { // If the map is still small, let it slide. + o.commits = make(map[uint64]uint64) + } + o.Unlock() + } +} + +func (o *oracle) readTs() uint64 { + if o.isManaged { + return math.MaxUint64 + } + return atomic.LoadUint64(&o.curRead) +} + +func (o *oracle) commitTs() uint64 { + o.Lock() + defer o.Unlock() + return o.nextCommit +} + +// hasConflict must be called while having a lock. +func (o *oracle) hasConflict(txn *Txn) bool { + if len(txn.reads) == 0 { + return false + } + for _, ro := range txn.reads { + if ts, has := o.commits[ro]; has && ts > txn.readTs { + return true + } + } + return false +} + +func (o *oracle) newCommitTs(txn *Txn) uint64 { + o.Lock() + defer o.Unlock() + + if o.hasConflict(txn) { + return 0 + } + + var ts uint64 + if !o.isManaged { + // This is the general case, when user doesn't specify the read and commit ts. + ts = o.nextCommit + o.nextCommit++ + + } else { + // If commitTs is set, use it instead. + ts = txn.commitTs + } + + for _, w := range txn.writes { + o.commits[w] = ts // Update the commitTs. + } + return ts +} + +func (o *oracle) doneCommit(cts uint64) { + if o.isManaged { + // No need to update anything. + return + } + + for { + curRead := atomic.LoadUint64(&o.curRead) + if cts <= curRead { + return + } + atomic.CompareAndSwapUint64(&o.curRead, curRead, cts) + } +} + +// Txn represents a Badger transaction. +type Txn struct { + readTs uint64 + commitTs uint64 + + update bool // update is used to conditionally keep track of reads. + reads []uint64 // contains fingerprints of keys read. + writes []uint64 // contains fingerprints of keys written. + + pendingWrites map[string]*Entry // cache stores any writes done by txn. + + db *DB + callbacks []func() + discarded bool + + size int64 + count int64 +} + +type pendingWritesIterator struct { + entries []*Entry + nextIdx int + readTs uint64 + reversed bool +} + +func (pi *pendingWritesIterator) Next() { + pi.nextIdx++ +} + +func (pi *pendingWritesIterator) Rewind() { + pi.nextIdx = 0 +} + +func (pi *pendingWritesIterator) Seek(key []byte) { + key = y.ParseKey(key) + pi.nextIdx = sort.Search(len(pi.entries), func(idx int) bool { + cmp := bytes.Compare(pi.entries[idx].Key, key) + if !pi.reversed { + return cmp >= 0 + } + return cmp <= 0 + }) +} + +func (pi *pendingWritesIterator) Key() []byte { + y.AssertTrue(pi.Valid()) + entry := pi.entries[pi.nextIdx] + return y.KeyWithTs(entry.Key, pi.readTs) +} + +func (pi *pendingWritesIterator) Value() y.ValueStruct { + y.AssertTrue(pi.Valid()) + entry := pi.entries[pi.nextIdx] + return y.ValueStruct{ + Value: entry.Value, + Meta: entry.meta, + UserMeta: entry.UserMeta, + ExpiresAt: entry.ExpiresAt, + Version: pi.readTs, + } +} + +func (pi *pendingWritesIterator) Valid() bool { + return pi.nextIdx < len(pi.entries) +} + +func (pi *pendingWritesIterator) Close() error { + return nil +} + +func (txn *Txn) newPendingWritesIterator(reversed bool) *pendingWritesIterator { + if !txn.update || len(txn.pendingWrites) == 0 { + return nil + } + entries := make([]*Entry, 0, len(txn.pendingWrites)) + for _, e := range txn.pendingWrites { + entries = append(entries, e) + } + // Number of pending writes per transaction shouldn't be too big in general. + sort.Slice(entries, func(i, j int) bool { + cmp := bytes.Compare(entries[i].Key, entries[j].Key) + if !reversed { + return cmp < 0 + } + return cmp > 0 + }) + return &pendingWritesIterator{ + readTs: txn.readTs, + entries: entries, + reversed: reversed, + } +} + +func (txn *Txn) checkSize(e *Entry) error { + count := txn.count + 1 + // Extra bytes for version in key. + size := txn.size + int64(e.estimateSize(txn.db.opt.ValueThreshold)) + 10 + if count >= txn.db.opt.maxBatchCount || size >= txn.db.opt.maxBatchSize { + return ErrTxnTooBig + } + txn.count, txn.size = count, size + return nil +} + +// Set adds a key-value pair to the database. +// +// It will return ErrReadOnlyTxn if update flag was set to false when creating the +// transaction. +func (txn *Txn) Set(key, val []byte) error { + e := &Entry{ + Key: key, + Value: val, + } + return txn.SetEntry(e) +} + +// SetWithMeta adds a key-value pair to the database, along with a metadata +// byte. This byte is stored alongside the key, and can be used as an aid to +// interpret the value or store other contextual bits corresponding to the +// key-value pair. +func (txn *Txn) SetWithMeta(key, val []byte, meta byte) error { + e := &Entry{Key: key, Value: val, UserMeta: meta} + return txn.SetEntry(e) +} + +// SetWithTTL adds a key-value pair to the database, along with a time-to-live +// (TTL) setting. A key stored with with a TTL would automatically expire after +// the time has elapsed , and be eligible for garbage collection. +func (txn *Txn) SetWithTTL(key, val []byte, dur time.Duration) error { + expire := time.Now().Add(dur).Unix() + e := &Entry{Key: key, Value: val, ExpiresAt: uint64(expire)} + return txn.SetEntry(e) +} + +// SetEntry takes an Entry struct and adds the key-value pair in the struct, along +// with other metadata to the database. +func (txn *Txn) SetEntry(e *Entry) error { + switch { + case !txn.update: + return ErrReadOnlyTxn + case txn.discarded: + return ErrDiscardedTxn + case len(e.Key) == 0: + return ErrEmptyKey + case len(e.Key) > maxKeySize: + return exceedsMaxKeySizeError(e.Key) + case int64(len(e.Value)) > txn.db.opt.ValueLogFileSize: + return exceedsMaxValueSizeError(e.Value, txn.db.opt.ValueLogFileSize) + } + if err := txn.checkSize(e); err != nil { + return err + } + + fp := farm.Fingerprint64(e.Key) // Avoid dealing with byte arrays. + txn.writes = append(txn.writes, fp) + txn.pendingWrites[string(e.Key)] = e + return nil +} + +// Delete deletes a key. This is done by adding a delete marker for the key at commit timestamp. +// Any reads happening before this timestamp would be unaffected. Any reads after this commit would +// see the deletion. +func (txn *Txn) Delete(key []byte) error { + if !txn.update { + return ErrReadOnlyTxn + } else if txn.discarded { + return ErrDiscardedTxn + } else if len(key) == 0 { + return ErrEmptyKey + } else if len(key) > maxKeySize { + return exceedsMaxKeySizeError(key) + } + + e := &Entry{ + Key: key, + meta: bitDelete, + } + if err := txn.checkSize(e); err != nil { + return err + } + + fp := farm.Fingerprint64(key) // Avoid dealing with byte arrays. + txn.writes = append(txn.writes, fp) + + txn.pendingWrites[string(key)] = e + return nil +} + +// Get looks for key and returns corresponding Item. +// If key is not found, ErrKeyNotFound is returned. +func (txn *Txn) Get(key []byte) (item *Item, rerr error) { + if len(key) == 0 { + return nil, ErrEmptyKey + } else if txn.discarded { + return nil, ErrDiscardedTxn + } + + item = new(Item) + if txn.update { + if e, has := txn.pendingWrites[string(key)]; has && bytes.Equal(key, e.Key) { + if isDeletedOrExpired(e.meta, e.ExpiresAt) { + return nil, ErrKeyNotFound + } + // Fulfill from cache. + item.meta = e.meta + item.val = e.Value + item.userMeta = e.UserMeta + item.key = key + item.status = prefetched + item.version = txn.readTs + // We probably don't need to set db on item here. + return item, nil + } + // Only track reads if this is update txn. No need to track read if txn serviced it + // internally. + fp := farm.Fingerprint64(key) + txn.reads = append(txn.reads, fp) + } + + seek := y.KeyWithTs(key, txn.readTs) + vs, err := txn.db.get(seek) + if err != nil { + return nil, errors.Wrapf(err, "DB::Get key: %q", key) + } + if vs.Value == nil && vs.Meta == 0 { + return nil, ErrKeyNotFound + } + if isDeletedOrExpired(vs.Meta, vs.ExpiresAt) { + return nil, ErrKeyNotFound + } + + item.key = key + item.version = vs.Version + item.meta = vs.Meta + item.userMeta = vs.UserMeta + item.db = txn.db + item.vptr = vs.Value + item.txn = txn + return item, nil +} + +// Discard discards a created transaction. This method is very important and must be called. Commit +// method calls this internally, however, calling this multiple times doesn't cause any issues. So, +// this can safely be called via a defer right when transaction is created. +// +// NOTE: If any operations are run on a discarded transaction, ErrDiscardedTxn is returned. +func (txn *Txn) Discard() { + if txn.discarded { // Avoid a re-run. + return + } + txn.discarded = true + + for _, cb := range txn.callbacks { + cb() + } + if txn.update { + txn.db.orc.decrRef() + } +} + +// Commit commits the transaction, following these steps: +// +// 1. If there are no writes, return immediately. +// +// 2. Check if read rows were updated since txn started. If so, return ErrConflict. +// +// 3. If no conflict, generate a commit timestamp and update written rows' commit ts. +// +// 4. Batch up all writes, write them to value log and LSM tree. +// +// 5. If callback is provided, Badger will return immediately after checking +// for conflicts. Writes to the database will happen in the background. If +// there is a conflict, an error will be returned and the callback will not +// run. If there are no conflicts, the callback will be called in the +// background upon successful completion of writes or any error during write. +// +// If error is nil, the transaction is successfully committed. In case of a non-nil error, the LSM +// tree won't be updated, so there's no need for any rollback. +func (txn *Txn) Commit(callback func(error)) error { + if txn.commitTs == 0 && txn.db.opt.managedTxns { + return ErrManagedTxn + } + if txn.discarded { + return ErrDiscardedTxn + } + defer txn.Discard() + if len(txn.writes) == 0 { + return nil // Nothing to do. + } + + state := txn.db.orc + state.writeLock.Lock() + commitTs := state.newCommitTs(txn) + if commitTs == 0 { + state.writeLock.Unlock() + return ErrConflict + } + + entries := make([]*Entry, 0, len(txn.pendingWrites)+1) + for _, e := range txn.pendingWrites { + // Suffix the keys with commit ts, so the key versions are sorted in + // descending order of commit timestamp. + e.Key = y.KeyWithTs(e.Key, commitTs) + e.meta |= bitTxn + entries = append(entries, e) + } + e := &Entry{ + Key: y.KeyWithTs(txnKey, commitTs), + Value: []byte(strconv.FormatUint(commitTs, 10)), + meta: bitFinTxn, + } + entries = append(entries, e) + + req, err := txn.db.sendToWriteCh(entries) + state.writeLock.Unlock() + if err != nil { + return err + } + + if callback == nil { + // If batchSet failed, LSM would not have been updated. So, no need to rollback anything. + + // TODO: What if some of the txns successfully make it to value log, but others fail. + // Nothing gets updated to LSM, until a restart happens. + defer state.doneCommit(commitTs) + return req.Wait() + } + go func() { + err := req.Wait() + // Write is complete. Let's call the callback function now. + state.doneCommit(commitTs) + callback(err) + }() + return nil +} + +// NewTransaction creates a new transaction. Badger supports concurrent execution of transactions, +// providing serializable snapshot isolation, avoiding write skews. Badger achieves this by tracking +// the keys read and at Commit time, ensuring that these read keys weren't concurrently modified by +// another transaction. +// +// For read-only transactions, set update to false. In this mode, we don't track the rows read for +// any changes. Thus, any long running iterations done in this mode wouldn't pay this overhead. +// +// Running transactions concurrently is OK. However, a transaction itself isn't thread safe, and +// should only be run serially. It doesn't matter if a transaction is created by one goroutine and +// passed down to other, as long as the Txn APIs are called serially. +// +// When you create a new transaction, it is absolutely essential to call +// Discard(). This should be done irrespective of what the update param is set +// to. Commit API internally runs Discard, but running it twice wouldn't cause +// any issues. +// +// txn := db.NewTransaction(false) +// defer txn.Discard() +// // Call various APIs. +func (db *DB) NewTransaction(update bool) *Txn { + txn := &Txn{ + update: update, + db: db, + readTs: db.orc.readTs(), + count: 1, // One extra entry for BitFin. + size: int64(len(txnKey) + 10), // Some buffer for the extra entry. + } + if update { + txn.pendingWrites = make(map[string]*Entry) + txn.db.orc.addRef() + } + return txn +} + +// View executes a function creating and managing a read-only transaction for the user. Error +// returned by the function is relayed by the View method. +func (db *DB) View(fn func(txn *Txn) error) error { + if db.opt.managedTxns { + return ErrManagedTxn + } + txn := db.NewTransaction(false) + defer txn.Discard() + + return fn(txn) +} + +// Update executes a function, creating and managing a read-write transaction +// for the user. Error returned by the function is relayed by the Update method. +func (db *DB) Update(fn func(txn *Txn) error) error { + if db.opt.managedTxns { + return ErrManagedTxn + } + txn := db.NewTransaction(true) + defer txn.Discard() + + if err := fn(txn); err != nil { + return err + } + + return txn.Commit(nil) +} diff --git a/vendor/github.com/dgraph-io/badger/util.go b/vendor/github.com/dgraph-io/badger/util.go new file mode 100644 index 0000000000..88fd74dccb --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/util.go @@ -0,0 +1,139 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "io/ioutil" + "math/rand" + "sync/atomic" + "time" + + "github.com/dgraph-io/badger/table" + "github.com/dgraph-io/badger/y" + "github.com/pkg/errors" +) + +// summary is produced when DB is closed. Currently it is used only for testing. +type summary struct { + fileIDs map[uint64]bool +} + +func (s *levelsController) getSummary() *summary { + out := &summary{ + fileIDs: make(map[uint64]bool), + } + for _, l := range s.levels { + l.getSummary(out) + } + return out +} + +func (s *levelHandler) getSummary(sum *summary) { + s.RLock() + defer s.RUnlock() + for _, t := range s.tables { + sum.fileIDs[t.ID()] = true + } +} + +func (s *DB) validate() error { return s.lc.validate() } + +func (s *levelsController) validate() error { + for _, l := range s.levels { + if err := l.validate(); err != nil { + return errors.Wrap(err, "Levels Controller") + } + } + return nil +} + +// Check does some sanity check on one level of data or in-memory index. +func (s *levelHandler) validate() error { + if s.level == 0 { + return nil + } + + s.RLock() + defer s.RUnlock() + numTables := len(s.tables) + for j := 1; j < numTables; j++ { + if j >= len(s.tables) { + return errors.Errorf("Level %d, j=%d numTables=%d", s.level, j, numTables) + } + + if y.CompareKeys(s.tables[j-1].Biggest(), s.tables[j].Smallest()) >= 0 { + return errors.Errorf( + "Inter: %q vs %q: level=%d j=%d numTables=%d", + string(s.tables[j-1].Biggest()), string(s.tables[j].Smallest()), s.level, j, numTables) + } + + if y.CompareKeys(s.tables[j].Smallest(), s.tables[j].Biggest()) > 0 { + return errors.Errorf( + "Intra: %q vs %q: level=%d j=%d numTables=%d", + s.tables[j].Smallest(), s.tables[j].Biggest(), s.level, j, numTables) + } + } + return nil +} + +// func (s *KV) debugPrintMore() { s.lc.debugPrintMore() } + +// // debugPrintMore shows key ranges of each level. +// func (s *levelsController) debugPrintMore() { +// s.Lock() +// defer s.Unlock() +// for i := 0; i < s.kv.opt.MaxLevels; i++ { +// s.levels[i].debugPrintMore() +// } +// } + +// func (s *levelHandler) debugPrintMore() { +// s.RLock() +// defer s.RUnlock() +// s.elog.Printf("Level %d:", s.level) +// for _, t := range s.tables { +// y.Printf(" [%s, %s]", t.Smallest(), t.Biggest()) +// } +// y.Printf("\n") +// } + +// reserveFileID reserves a unique file id. +func (s *levelsController) reserveFileID() uint64 { + id := atomic.AddUint64(&s.nextFileID, 1) + return id - 1 +} + +func getIDMap(dir string) map[uint64]struct{} { + fileInfos, err := ioutil.ReadDir(dir) + y.Check(err) + idMap := make(map[uint64]struct{}) + for _, info := range fileInfos { + if info.IsDir() { + continue + } + fileID, ok := table.ParseFileID(info.Name()) + if !ok { + continue + } + idMap[fileID] = struct{}{} + } + return idMap +} + +func init() { + rand.Seed(time.Now().UnixNano()) +} diff --git a/vendor/github.com/dgraph-io/badger/value.go b/vendor/github.com/dgraph-io/badger/value.go new file mode 100644 index 0000000000..211b6ff8ab --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/value.go @@ -0,0 +1,1070 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "bufio" + "bytes" + "encoding/binary" + "fmt" + "hash/crc32" + "io" + "io/ioutil" + "log" + "math" + "math/rand" + "os" + "sort" + "strconv" + "strings" + "sync" + "sync/atomic" + "time" + + "github.com/dgraph-io/badger/options" + + "github.com/dgraph-io/badger/y" + "github.com/pkg/errors" + "golang.org/x/net/trace" +) + +// Values have their first byte being byteData or byteDelete. This helps us distinguish between +// a key that has never been seen and a key that has been explicitly deleted. +const ( + bitDelete byte = 1 << 0 // Set if the key has been deleted. + bitValuePointer byte = 1 << 1 // Set if the value is NOT stored directly next to key. + + // The MSB 2 bits are for transactions. + bitTxn byte = 1 << 6 // Set if the entry is part of a txn. + bitFinTxn byte = 1 << 7 // Set if the entry is to indicate end of txn in value log. + + mi int64 = 1 << 20 +) + +type logFile struct { + path string + // This is a lock on the log file. It guards the fd’s value, the file’s + // existence and the file’s memory map. + // + // Use shared ownership when reading/writing the file or memory map, use + // exclusive ownership to open/close the descriptor, unmap or remove the file. + lock sync.RWMutex + fd *os.File + fid uint32 + fmap []byte + size uint32 + loadingMode options.FileLoadingMode +} + +// openReadOnly assumes that we have a write lock on logFile. +func (lf *logFile) openReadOnly() error { + var err error + lf.fd, err = os.OpenFile(lf.path, os.O_RDONLY, 0666) + if err != nil { + return errors.Wrapf(err, "Unable to open %q as RDONLY.", lf.path) + } + + fi, err := lf.fd.Stat() + if err != nil { + return errors.Wrapf(err, "Unable to check stat for %q", lf.path) + } + lf.size = uint32(fi.Size()) + + if err = lf.mmap(fi.Size()); err != nil { + _ = lf.fd.Close() + return y.Wrapf(err, "Unable to map file") + } + + return nil +} + +func (lf *logFile) mmap(size int64) (err error) { + if lf.loadingMode != options.MemoryMap { + // Nothing to do + return nil + } + lf.fmap, err = y.Mmap(lf.fd, false, size) + if err == nil { + err = y.Madvise(lf.fmap, false) // Disable readahead + } + return err +} + +func (lf *logFile) munmap() (err error) { + if lf.loadingMode != options.MemoryMap { + // Nothing to do + return nil + } + if err := y.Munmap(lf.fmap); err != nil { + return errors.Wrapf(err, "Unable to munmap value log: %q", lf.path) + } + return nil +} + +// Acquire lock on mmap/file if you are calling this +func (lf *logFile) read(p valuePointer, s *y.Slice) (buf []byte, err error) { + var nbr int64 + offset := p.Offset + if lf.loadingMode == options.FileIO { + buf = s.Resize(int(p.Len)) + var n int + n, err = lf.fd.ReadAt(buf, int64(offset)) + nbr = int64(n) + } else { + size := uint32(len(lf.fmap)) + valsz := p.Len + if offset >= size || offset+valsz > size { + err = y.ErrEOF + } else { + buf = lf.fmap[offset : offset+valsz] + nbr = int64(valsz) + } + } + y.NumReads.Add(1) + y.NumBytesRead.Add(nbr) + return buf, err +} + +func (lf *logFile) doneWriting(offset uint32) error { + // Sync before acquiring lock. (We call this from write() and thus know we have shared access + // to the fd.) + if err := lf.fd.Sync(); err != nil { + return errors.Wrapf(err, "Unable to sync value log: %q", lf.path) + } + // Close and reopen the file read-only. Acquire lock because fd will become invalid for a bit. + // Acquiring the lock is bad because, while we don't hold the lock for a long time, it forces + // one batch of readers wait for the preceding batch of readers to finish. + // + // If there's a benefit to reopening the file read-only, it might be on Windows. I don't know + // what the benefit is. Consider keeping the file read-write, or use fcntl to change + // permissions. + lf.lock.Lock() + defer lf.lock.Unlock() + if err := lf.munmap(); err != nil { + return err + } + // TODO: Confirm if we need to run a file sync after truncation. + // Truncation must run after unmapping, otherwise Windows would crap itself. + if err := lf.fd.Truncate(int64(offset)); err != nil { + return errors.Wrapf(err, "Unable to truncate file: %q", lf.path) + } + if err := lf.fd.Close(); err != nil { + return errors.Wrapf(err, "Unable to close value log: %q", lf.path) + } + + return lf.openReadOnly() +} + +// You must hold lf.lock to sync() +func (lf *logFile) sync() error { + return lf.fd.Sync() +} + +var errStop = errors.New("Stop iteration") + +type logEntry func(e Entry, vp valuePointer) error + +// iterate iterates over log file. It doesn't not allocate new memory for every kv pair. +// Therefore, the kv pair is only valid for the duration of fn call. +func (vlog *valueLog) iterate(lf *logFile, offset uint32, fn logEntry) error { + _, err := lf.fd.Seek(int64(offset), io.SeekStart) + if err != nil { + return y.Wrap(err) + } + + reader := bufio.NewReader(lf.fd) + var hbuf [headerBufSize]byte + var h header + k := make([]byte, 1<<10) + v := make([]byte, 1<<20) + + truncate := false + recordOffset := offset + for { + hash := crc32.New(y.CastagnoliCrcTable) + tee := io.TeeReader(reader, hash) + + // TODO: Move this entry decode into structs.go + if _, err = io.ReadFull(tee, hbuf[:]); err != nil { + if err == io.EOF { + break + } else if err == io.ErrUnexpectedEOF { + truncate = true + break + } + return err + } + + var e Entry + e.offset = recordOffset + h.Decode(hbuf[:]) + if h.klen > maxKeySize { + truncate = true + break + } + vl := int(h.vlen) + if cap(v) < vl { + v = make([]byte, 2*vl) + } + + kl := int(h.klen) + if cap(k) < kl { + k = make([]byte, 2*kl) + } + e.Key = k[:kl] + e.Value = v[:vl] + + if _, err = io.ReadFull(tee, e.Key); err != nil { + if err == io.EOF || err == io.ErrUnexpectedEOF { + truncate = true + break + } + return err + } + if _, err = io.ReadFull(tee, e.Value); err != nil { + if err == io.EOF || err == io.ErrUnexpectedEOF { + truncate = true + break + } + return err + } + + var crcBuf [4]byte + if _, err = io.ReadFull(reader, crcBuf[:]); err != nil { + if err == io.EOF || err == io.ErrUnexpectedEOF { + truncate = true + break + } + return err + } + crc := binary.BigEndian.Uint32(crcBuf[:]) + if crc != hash.Sum32() { + truncate = true + break + } + e.meta = h.meta + e.UserMeta = h.userMeta + e.ExpiresAt = h.expiresAt + + var vp valuePointer + vp.Len = headerBufSize + h.klen + h.vlen + uint32(len(crcBuf)) + recordOffset += vp.Len + + vp.Offset = e.offset + vp.Fid = lf.fid + + if err := fn(e, vp); err != nil { + if err == errStop { + break + } + return y.Wrap(err) + } + } + + if truncate && len(lf.fmap) == 0 { + // Only truncate if the file isn't mmaped. Otherwise, Windows would puke. + if err := lf.fd.Truncate(int64(recordOffset)); err != nil { + return err + } + } + + return nil +} + +func (vlog *valueLog) rewrite(f *logFile) error { + maxFid := atomic.LoadUint32(&vlog.maxFid) + y.AssertTruef(uint32(f.fid) < maxFid, "fid to move: %d. Current max fid: %d", f.fid, maxFid) + + elog := trace.NewEventLog("badger", "vlog-rewrite") + defer elog.Finish() + elog.Printf("Rewriting fid: %d", f.fid) + + wb := make([]*Entry, 0, 1000) + var size int64 + + y.AssertTrue(vlog.kv != nil) + var count int + fe := func(e Entry) error { + count++ + if count%10000 == 0 { + elog.Printf("Processing entry %d", count) + } + + vs, err := vlog.kv.get(e.Key) + if err != nil { + return err + } + if discardEntry(e, vs) { + return nil + } + + // Value is still present in value log. + if len(vs.Value) == 0 { + return errors.Errorf("Empty value: %+v", vs) + } + var vp valuePointer + vp.Decode(vs.Value) + + if vp.Fid > f.fid { + return nil + } + if vp.Offset > e.offset { + return nil + } + if vp.Fid == f.fid && vp.Offset == e.offset { + // This new entry only contains the key, and a pointer to the value. + ne := new(Entry) + ne.meta = 0 // Remove all bits. + ne.UserMeta = e.UserMeta + ne.Key = make([]byte, len(e.Key)) + copy(ne.Key, e.Key) + ne.Value = make([]byte, len(e.Value)) + copy(ne.Value, e.Value) + wb = append(wb, ne) + size += int64(e.estimateSize(vlog.opt.ValueThreshold)) + if size >= 64*mi { + elog.Printf("request has %d entries, size %d", len(wb), size) + if err := vlog.kv.batchSet(wb); err != nil { + return err + } + size = 0 + wb = wb[:0] + } + } else { + log.Printf("WARNING: This entry should have been caught. %+v\n", e) + } + return nil + } + + err := vlog.iterate(f, 0, func(e Entry, vp valuePointer) error { + return fe(e) + }) + if err != nil { + return err + } + + elog.Printf("request has %d entries, size %d", len(wb), size) + batchSize := 1024 + var loops int + for i := 0; i < len(wb); { + loops++ + if batchSize == 0 { + log.Printf("WARNING: We shouldn't reach batch size of zero.") + return ErrNoRewrite + } + end := i + batchSize + if end > len(wb) { + end = len(wb) + } + if err := vlog.kv.batchSet(wb[i:end]); err != nil { + if err == ErrTxnTooBig { + // Decrease the batch size to half. + batchSize = batchSize / 2 + elog.Printf("Dropped batch size to %d", batchSize) + continue + } + return err + } + i += batchSize + } + elog.Printf("Processed %d entries in %d loops", len(wb), loops) + + elog.Printf("Removing fid: %d", f.fid) + var deleteFileNow bool + // Entries written to LSM. Remove the older file now. + { + vlog.filesLock.Lock() + // Just a sanity-check. + if _, ok := vlog.filesMap[f.fid]; !ok { + vlog.filesLock.Unlock() + return errors.Errorf("Unable to find fid: %d", f.fid) + } + if vlog.numActiveIterators == 0 { + delete(vlog.filesMap, f.fid) + deleteFileNow = true + } else { + vlog.filesToBeDeleted = append(vlog.filesToBeDeleted, f.fid) + } + vlog.filesLock.Unlock() + } + + if deleteFileNow { + vlog.deleteLogFile(f) + } + + return nil +} + +func (vlog *valueLog) incrIteratorCount() { + atomic.AddInt32(&vlog.numActiveIterators, 1) +} + +func (vlog *valueLog) decrIteratorCount() error { + num := atomic.AddInt32(&vlog.numActiveIterators, -1) + if num != 0 { + return nil + } + + vlog.filesLock.Lock() + lfs := make([]*logFile, 0, len(vlog.filesToBeDeleted)) + for _, id := range vlog.filesToBeDeleted { + lfs = append(lfs, vlog.filesMap[id]) + delete(vlog.filesMap, id) + } + vlog.filesToBeDeleted = nil + vlog.filesLock.Unlock() + + for _, lf := range lfs { + if err := vlog.deleteLogFile(lf); err != nil { + return err + } + } + return nil +} + +func (vlog *valueLog) deleteLogFile(lf *logFile) error { + path := vlog.fpath(lf.fid) + if err := lf.munmap(); err != nil { + _ = lf.fd.Close() + return err + } + if err := lf.fd.Close(); err != nil { + return err + } + return os.Remove(path) +} + +// lfDiscardStats keeps track of the amount of data that could be discarded for +// a given logfile. +type lfDiscardStats struct { + sync.Mutex + m map[uint32]int64 +} + +type valueLog struct { + buf bytes.Buffer + dirPath string + elog trace.EventLog + + // guards our view of which files exist, which to be deleted, how many active iterators + filesLock sync.RWMutex + filesMap map[uint32]*logFile + filesToBeDeleted []uint32 + // A refcount of iterators -- when this hits zero, we can delete the filesToBeDeleted. + numActiveIterators int32 + + kv *DB + maxFid uint32 + writableLogOffset uint32 + opt Options + + garbageCh chan struct{} + lfDiscardStats *lfDiscardStats +} + +func vlogFilePath(dirPath string, fid uint32) string { + return fmt.Sprintf("%s%s%06d.vlog", dirPath, string(os.PathSeparator), fid) +} + +func (vlog *valueLog) fpath(fid uint32) string { + return vlogFilePath(vlog.dirPath, fid) +} + +func (vlog *valueLog) openOrCreateFiles() error { + files, err := ioutil.ReadDir(vlog.dirPath) + if err != nil { + return errors.Wrapf(err, "Error while opening value log") + } + + found := make(map[uint64]struct{}) + var maxFid uint32 // Beware len(files) == 0 case, this starts at 0. + for _, file := range files { + if !strings.HasSuffix(file.Name(), ".vlog") { + continue + } + fsz := len(file.Name()) + fid, err := strconv.ParseUint(file.Name()[:fsz-5], 10, 32) + if err != nil { + return errors.Wrapf(err, "Error while parsing value log id for file: %q", file.Name()) + } + if _, ok := found[fid]; ok { + return errors.Errorf("Found the same value log file twice: %d", fid) + } + found[fid] = struct{}{} + + lf := &logFile{ + fid: uint32(fid), + path: vlog.fpath(uint32(fid)), + loadingMode: vlog.opt.ValueLogLoadingMode, + } + vlog.filesMap[uint32(fid)] = lf + if uint32(fid) > maxFid { + maxFid = uint32(fid) + } + } + vlog.maxFid = uint32(maxFid) + + // Open all previous log files as read only. Open the last log file + // as read write. + for fid, lf := range vlog.filesMap { + if fid == maxFid { + if lf.fd, err = y.OpenExistingSyncedFile(vlog.fpath(fid), + vlog.opt.SyncWrites); err != nil { + return errors.Wrapf(err, "Unable to open value log file as RDWR") + } + } else { + if err := lf.openReadOnly(); err != nil { + return err + } + } + } + + // If no files are found, then create a new file. + if len(vlog.filesMap) == 0 { + // We already set vlog.maxFid above + _, err := vlog.createVlogFile(0) + if err != nil { + return err + } + } + return nil +} + +func (vlog *valueLog) createVlogFile(fid uint32) (*logFile, error) { + path := vlog.fpath(fid) + lf := &logFile{fid: fid, path: path, loadingMode: vlog.opt.ValueLogLoadingMode} + vlog.writableLogOffset = 0 + + var err error + if lf.fd, err = y.CreateSyncedFile(path, vlog.opt.SyncWrites); err != nil { + return nil, errors.Wrapf(err, "Unable to create value log file") + } + + if err = syncDir(vlog.dirPath); err != nil { + return nil, errors.Wrapf(err, "Unable to sync value log file dir") + } + + vlog.filesLock.Lock() + vlog.filesMap[fid] = lf + vlog.filesLock.Unlock() + + return lf, nil +} + +func (vlog *valueLog) Open(kv *DB, opt Options) error { + vlog.dirPath = opt.ValueDir + vlog.opt = opt + vlog.kv = kv + vlog.filesMap = make(map[uint32]*logFile) + if err := vlog.openOrCreateFiles(); err != nil { + return errors.Wrapf(err, "Unable to open value log") + } + + vlog.elog = trace.NewEventLog("Badger", "Valuelog") + vlog.garbageCh = make(chan struct{}, 1) // Only allow one GC at a time. + vlog.lfDiscardStats = &lfDiscardStats{m: make(map[uint32]int64)} + return nil +} + +func (vlog *valueLog) Close() error { + vlog.elog.Printf("Stopping garbage collection of values.") + defer vlog.elog.Finish() + + var err error + for id, f := range vlog.filesMap { + + f.lock.Lock() // We won’t release the lock. + if munmapErr := f.munmap(); munmapErr != nil && err == nil { + err = munmapErr + } + + if id == vlog.maxFid { + // truncate writable log file to correct offset. + if truncErr := f.fd.Truncate( + int64(vlog.writableLogOffset)); truncErr != nil && err == nil { + err = truncErr + } + } + + if closeErr := f.fd.Close(); closeErr != nil && err == nil { + err = closeErr + } + + } + return err +} + +// sortedFids returns the file id's not pending deletion, sorted. Assumes we have shared access to +// filesMap. +func (vlog *valueLog) sortedFids() []uint32 { + toBeDeleted := make(map[uint32]struct{}) + for _, fid := range vlog.filesToBeDeleted { + toBeDeleted[fid] = struct{}{} + } + ret := make([]uint32, 0, len(vlog.filesMap)) + for fid := range vlog.filesMap { + if _, ok := toBeDeleted[fid]; !ok { + ret = append(ret, fid) + } + } + sort.Slice(ret, func(i, j int) bool { + return ret[i] < ret[j] + }) + return ret +} + +// Replay replays the value log. The kv provided is only valid for the lifetime of function call. +func (vlog *valueLog) Replay(ptr valuePointer, fn logEntry) error { + fid := ptr.Fid + offset := ptr.Offset + ptr.Len + vlog.elog.Printf("Seeking at value pointer: %+v\n", ptr) + + fids := vlog.sortedFids() + + for _, id := range fids { + if id < fid { + continue + } + of := offset + if id > fid { + of = 0 + } + f := vlog.filesMap[id] + err := vlog.iterate(f, of, fn) + if err != nil { + return errors.Wrapf(err, "Unable to replay value log: %q", f.path) + } + } + + // Seek to the end to start writing. + var err error + last := vlog.filesMap[vlog.maxFid] + lastOffset, err := last.fd.Seek(0, io.SeekEnd) + atomic.AddUint32(&vlog.writableLogOffset, uint32(lastOffset)) + return errors.Wrapf(err, "Unable to seek to end of value log: %q", last.path) +} + +type request struct { + // Input values + Entries []*Entry + // Output values and wait group stuff below + Ptrs []valuePointer + Wg sync.WaitGroup + Err error +} + +func (req *request) Wait() error { + req.Wg.Wait() + req.Entries = nil + err := req.Err + requestPool.Put(req) + return err +} + +// sync is thread-unsafe and should not be called concurrently with write. +func (vlog *valueLog) sync() error { + if vlog.opt.SyncWrites { + return nil + } + + vlog.filesLock.RLock() + if len(vlog.filesMap) == 0 { + vlog.filesLock.RUnlock() + return nil + } + curlf := vlog.filesMap[vlog.maxFid] + curlf.lock.RLock() + vlog.filesLock.RUnlock() + + dirSyncCh := make(chan error) + go func() { dirSyncCh <- syncDir(vlog.opt.ValueDir) }() + err := curlf.sync() + curlf.lock.RUnlock() + dirSyncErr := <-dirSyncCh + if err != nil { + err = dirSyncErr + } + return err +} + +func (vlog *valueLog) writableOffset() uint32 { + return atomic.LoadUint32(&vlog.writableLogOffset) +} + +// write is thread-unsafe by design and should not be called concurrently. +func (vlog *valueLog) write(reqs []*request) error { + vlog.filesLock.RLock() + curlf := vlog.filesMap[vlog.maxFid] + vlog.filesLock.RUnlock() + + toDisk := func() error { + if vlog.buf.Len() == 0 { + return nil + } + vlog.elog.Printf("Flushing %d blocks of total size: %d", len(reqs), vlog.buf.Len()) + n, err := curlf.fd.Write(vlog.buf.Bytes()) + if err != nil { + return errors.Wrapf(err, "Unable to write to value log file: %q", curlf.path) + } + y.NumWrites.Add(1) + y.NumBytesWritten.Add(int64(n)) + vlog.elog.Printf("Done") + atomic.AddUint32(&vlog.writableLogOffset, uint32(n)) + vlog.buf.Reset() + + if vlog.writableOffset() > uint32(vlog.opt.ValueLogFileSize) { + var err error + if err = curlf.doneWriting(vlog.writableLogOffset); err != nil { + return err + } + + newid := atomic.AddUint32(&vlog.maxFid, 1) + y.AssertTruef(newid <= math.MaxUint32, "newid will overflow uint32: %v", newid) + newlf, err := vlog.createVlogFile(newid) + if err != nil { + return err + } + + if err = newlf.mmap(2 * vlog.opt.ValueLogFileSize); err != nil { + return err + } + + curlf = newlf + } + return nil + } + + for i := range reqs { + b := reqs[i] + b.Ptrs = b.Ptrs[:0] + for j := range b.Entries { + e := b.Entries[j] + var p valuePointer + + p.Fid = curlf.fid + // Use the offset including buffer length so far. + p.Offset = vlog.writableOffset() + uint32(vlog.buf.Len()) + plen, err := encodeEntry(e, &vlog.buf) // Now encode the entry into buffer. + if err != nil { + return err + } + p.Len = uint32(plen) + b.Ptrs = append(b.Ptrs, p) + + if p.Offset > uint32(vlog.opt.ValueLogFileSize) { + if err := toDisk(); err != nil { + return err + } + } + } + } + return toDisk() + + // Acquire mutex locks around this manipulation, so that the reads don't try to use + // an invalid file descriptor. +} + +// Gets the logFile and acquires and RLock() for the mmap. You must call RUnlock on the file +// (if non-nil) +func (vlog *valueLog) getFileRLocked(fid uint32) (*logFile, error) { + vlog.filesLock.RLock() + defer vlog.filesLock.RUnlock() + ret, ok := vlog.filesMap[fid] + if !ok { + // log file has gone away, will need to retry the operation. + return nil, ErrRetry + } + ret.lock.RLock() + return ret, nil +} + +// Read reads the value log at a given location. +// TODO: Make this read private. +func (vlog *valueLog) Read(vp valuePointer, s *y.Slice) ([]byte, func(), error) { + // Check for valid offset if we are reading to writable log. + if vp.Fid == vlog.maxFid && vp.Offset >= vlog.writableOffset() { + return nil, nil, errors.Errorf( + "Invalid value pointer offset: %d greater than current offset: %d", + vp.Offset, vlog.writableOffset()) + } + + buf, cb, err := vlog.readValueBytes(vp, s) + if err != nil { + return nil, cb, err + } + var h header + h.Decode(buf) + n := uint32(headerBufSize) + h.klen + return buf[n : n+h.vlen], cb, nil +} + +func (vlog *valueLog) readValueBytes(vp valuePointer, s *y.Slice) ([]byte, func(), error) { + lf, err := vlog.getFileRLocked(vp.Fid) + if err != nil { + return nil, nil, errors.Wrapf(err, "Unable to read from value log: %+v", vp) + } + + buf, err := lf.read(vp, s) + if vlog.opt.ValueLogLoadingMode == options.MemoryMap { + return buf, lf.lock.RUnlock, err + } + // If we are using File I/O we unlock the file immediately + // and return an empty function as callback. + lf.lock.RUnlock() + return buf, nil, err +} + +// Test helper +func valueBytesToEntry(buf []byte) (e Entry) { + var h header + h.Decode(buf) + n := uint32(headerBufSize) + + e.Key = buf[n : n+h.klen] + n += h.klen + e.meta = h.meta + e.UserMeta = h.userMeta + e.Value = buf[n : n+h.vlen] + return +} + +func (vlog *valueLog) pickLog(head valuePointer) *logFile { + vlog.filesLock.RLock() + defer vlog.filesLock.RUnlock() + fids := vlog.sortedFids() + if len(fids) <= 1 || head.Fid == 0 { + return nil + } + + // Pick a candidate that contains the largest amount of discardable data + candidate := struct { + fid uint32 + discard int64 + }{math.MaxUint32, 0} + vlog.lfDiscardStats.Lock() + for _, fid := range fids { + if fid >= head.Fid { + break + } + if vlog.lfDiscardStats.m[fid] > candidate.discard { + candidate.fid = fid + candidate.discard = vlog.lfDiscardStats.m[fid] + } + } + vlog.lfDiscardStats.Unlock() + + if candidate.fid != math.MaxUint32 { // Found a candidate + return vlog.filesMap[candidate.fid] + } + + // Fallback to randomly picking a log file + var idxHead int + for i, fid := range fids { + if fid == head.Fid { + idxHead = i + break + } + } + if idxHead == 0 { // Not found or first file + return nil + } + idx := rand.Intn(idxHead) // Don’t include head.Fid. We pick a random file before it. + if idx > 0 { + idx = rand.Intn(idx + 1) // Another level of rand to favor smaller fids. + } + return vlog.filesMap[fids[idx]] +} + +func discardEntry(e Entry, vs y.ValueStruct) bool { + if vs.Version != y.ParseTs(e.Key) { + // Version not found. Discard. + return true + } + if isDeletedOrExpired(vs.Meta, vs.ExpiresAt) { + return true + } + if (vs.Meta & bitValuePointer) == 0 { + // Key also stores the value in LSM. Discard. + return true + } + if (vs.Meta & bitFinTxn) > 0 { + // Just a txn finish entry. Discard. + return true + } + return false +} + +func (vlog *valueLog) doRunGC(gcThreshold float64, head valuePointer) (err error) { + // Pick a log file for GC + lf := vlog.pickLog(head) + if lf == nil { + return ErrNoRewrite + } + + // Update stats before exiting + defer func() { + if err == nil { + vlog.lfDiscardStats.Lock() + delete(vlog.lfDiscardStats.m, lf.fid) + vlog.lfDiscardStats.Unlock() + } + }() + + type reason struct { + total float64 + keep float64 + discard float64 + } + + var r reason + var window = 100.0 + count := 0 + + // Pick a random start point for the log. + skipFirstM := float64(rand.Intn(int(vlog.opt.ValueLogFileSize/mi))) - window + var skipped float64 + + start := time.Now() + y.AssertTrue(vlog.kv != nil) + s := new(y.Slice) + err = vlog.iterate(lf, 0, func(e Entry, vp valuePointer) error { + esz := float64(vp.Len) / (1 << 20) // in MBs. +4 for the CAS stuff. + skipped += esz + if skipped < skipFirstM { + return nil + } + + count++ + if count%100 == 0 { + time.Sleep(time.Millisecond) + } + r.total += esz + if r.total > window { + return errStop + } + if time.Since(start) > 10*time.Second { + return errStop + } + + vs, err := vlog.kv.get(e.Key) + if err != nil { + return err + } + if discardEntry(e, vs) { + r.discard += esz + return nil + } + + // Value is still present in value log. + y.AssertTrue(len(vs.Value) > 0) + vp.Decode(vs.Value) + + if vp.Fid > lf.fid { + // Value is present in a later log. Discard. + r.discard += esz + return nil + } + if vp.Offset > e.offset { + // Value is present in a later offset, but in the same log. + r.discard += esz + return nil + } + if vp.Fid == lf.fid && vp.Offset == e.offset { + // This is still the active entry. This would need to be rewritten. + r.keep += esz + + } else { + vlog.elog.Printf("Reason=%+v\n", r) + + buf, cb, err := vlog.readValueBytes(vp, s) + if err != nil { + return errStop + } + ne := valueBytesToEntry(buf) + ne.offset = vp.Offset + ne.print("Latest Entry Header in LSM") + e.print("Latest Entry in Log") + runCallback(cb) + return errors.Errorf("This shouldn't happen. Latest Pointer:%+v. Meta:%v.", + vp, vs.Meta) + } + return nil + }) + + if err != nil { + vlog.elog.Errorf("Error while iterating for RunGC: %v", err) + return err + } + vlog.elog.Printf("Fid: %d Data status=%+v\n", lf.fid, r) + + if r.total < 10.0 || r.discard < gcThreshold*r.total { + vlog.elog.Printf("Skipping GC on fid: %d\n\n", lf.fid) + return ErrNoRewrite + } + + vlog.elog.Printf("REWRITING VLOG %d\n", lf.fid) + if err = vlog.rewrite(lf); err != nil { + return err + } + vlog.elog.Printf("Done rewriting.") + return nil +} + +func (vlog *valueLog) waitOnGC(lc *y.Closer) { + defer lc.Done() + + <-lc.HasBeenClosed() // Wait for lc to be closed. + + // Block any GC in progress to finish, and don't allow any more writes to runGC by filling up + // the channel of size 1. + vlog.garbageCh <- struct{}{} +} + +func (vlog *valueLog) runGC(gcThreshold float64, head valuePointer) error { + select { + case vlog.garbageCh <- struct{}{}: + // Run GC + var ( + err error + count int + ) + for { + err = vlog.doRunGC(gcThreshold, head) + if err != nil { + break + } + count++ + } + <-vlog.garbageCh + if err == ErrNoRewrite && count > 0 { + return nil + } + return err + default: + return ErrRejected + } +} + +func (vlog *valueLog) updateGCStats(item *Item) { + if item.meta&bitValuePointer > 0 { + var vp valuePointer + vp.Decode(item.vptr) + vlog.lfDiscardStats.Lock() + vlog.lfDiscardStats.m[vp.Fid] += int64(vp.Len) + vlog.lfDiscardStats.Unlock() + } +} diff --git a/vendor/github.com/dgraph-io/badger/y/error.go b/vendor/github.com/dgraph-io/badger/y/error.go new file mode 100644 index 0000000000..59bb283584 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/y/error.go @@ -0,0 +1,83 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package y + +// This file contains some functions for error handling. Note that we are moving +// towards using x.Trace, i.e., rpc tracing using net/tracer. But for now, these +// functions are useful for simple checks logged on one machine. +// Some common use cases are: +// (1) You receive an error from external lib, and would like to check/log fatal. +// For this, use x.Check, x.Checkf. These will check for err != nil, which is +// more common in Go. If you want to check for boolean being true, use +// x.Assert, x.Assertf. +// (2) You receive an error from external lib, and would like to pass on with some +// stack trace information. In this case, use x.Wrap or x.Wrapf. +// (3) You want to generate a new error with stack trace info. Use x.Errorf. + +import ( + "fmt" + "log" + + "github.com/pkg/errors" +) + +var debugMode = true + +// Check logs fatal if err != nil. +func Check(err error) { + if err != nil { + log.Fatalf("%+v", Wrap(err)) + } +} + +// Check2 acts as convenience wrapper around Check, using the 2nd argument as error. +func Check2(_ interface{}, err error) { + Check(err) +} + +// AssertTrue asserts that b is true. Otherwise, it would log fatal. +func AssertTrue(b bool) { + if !b { + log.Fatalf("%+v", errors.Errorf("Assert failed")) + } +} + +// AssertTruef is AssertTrue with extra info. +func AssertTruef(b bool, format string, args ...interface{}) { + if !b { + log.Fatalf("%+v", errors.Errorf(format, args...)) + } +} + +// Wrap wraps errors from external lib. +func Wrap(err error) error { + if !debugMode { + return err + } + return errors.Wrap(err, "") +} + +// Wrapf is Wrap with extra info. +func Wrapf(err error, format string, args ...interface{}) error { + if !debugMode { + if err == nil { + return nil + } + return fmt.Errorf(format+" error: %+v", append(args, err)...) + } + return errors.Wrapf(err, format, args...) +} diff --git a/vendor/github.com/dgraph-io/badger/y/file_dsync.go b/vendor/github.com/dgraph-io/badger/y/file_dsync.go new file mode 100644 index 0000000000..bd25c9e9a3 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/y/file_dsync.go @@ -0,0 +1,25 @@ +// +build !dragonfly,!freebsd,!windows,!darwin + +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package y + +import "syscall" + +func init() { + datasyncFileFlag = syscall.O_DSYNC +} diff --git a/vendor/github.com/dgraph-io/badger/y/file_nodsync.go b/vendor/github.com/dgraph-io/badger/y/file_nodsync.go new file mode 100644 index 0000000000..be295f2e02 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/y/file_nodsync.go @@ -0,0 +1,25 @@ +// +build dragonfly freebsd windows darwin + +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package y + +import "syscall" + +func init() { + datasyncFileFlag = syscall.O_SYNC +} diff --git a/vendor/github.com/dgraph-io/badger/y/iterator.go b/vendor/github.com/dgraph-io/badger/y/iterator.go new file mode 100644 index 0000000000..719e8ec8ea --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/y/iterator.go @@ -0,0 +1,264 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package y + +import ( + "bytes" + "container/heap" + "encoding/binary" + + "github.com/pkg/errors" +) + +// ValueStruct represents the value info that can be associated with a key, but also the internal +// Meta field. +type ValueStruct struct { + Meta byte + UserMeta byte + ExpiresAt uint64 + Value []byte + + Version uint64 // This field is not serialized. Only for internal usage. +} + +func sizeVarint(x uint64) (n int) { + for { + n++ + x >>= 7 + if x == 0 { + break + } + } + return n +} + +// EncodedSize is the size of the ValueStruct when encoded +func (v *ValueStruct) EncodedSize() uint16 { + sz := len(v.Value) + 2 // meta, usermeta. + if v.ExpiresAt == 0 { + return uint16(sz + 1) + } + + enc := sizeVarint(v.ExpiresAt) + return uint16(sz + enc) +} + +// Decode uses the length of the slice to infer the length of the Value field. +func (v *ValueStruct) Decode(b []byte) { + v.Meta = b[0] + v.UserMeta = b[1] + var sz int + v.ExpiresAt, sz = binary.Uvarint(b[2:]) + v.Value = b[2+sz:] +} + +// Encode expects a slice of length at least v.EncodedSize(). +func (v *ValueStruct) Encode(b []byte) { + b[0] = v.Meta + b[1] = v.UserMeta + sz := binary.PutUvarint(b[2:], v.ExpiresAt) + copy(b[2+sz:], v.Value) +} + +// EncodeTo should be kept in sync with the Encode function above. The reason +// this function exists is to avoid creating byte arrays per key-value pair in +// table/builder.go. +func (v *ValueStruct) EncodeTo(buf *bytes.Buffer) { + buf.WriteByte(v.Meta) + buf.WriteByte(v.UserMeta) + var enc [binary.MaxVarintLen64]byte + sz := binary.PutUvarint(enc[:], v.ExpiresAt) + buf.Write(enc[:sz]) + buf.Write(v.Value) +} + +// Iterator is an interface for a basic iterator. +type Iterator interface { + Next() + Rewind() + Seek(key []byte) + Key() []byte + Value() ValueStruct + Valid() bool + + // All iterators should be closed so that file garbage collection works. + Close() error +} + +type elem struct { + itr Iterator + nice int + reversed bool +} + +type elemHeap []*elem + +func (eh elemHeap) Len() int { return len(eh) } +func (eh elemHeap) Swap(i, j int) { eh[i], eh[j] = eh[j], eh[i] } +func (eh *elemHeap) Push(x interface{}) { *eh = append(*eh, x.(*elem)) } +func (eh *elemHeap) Pop() interface{} { + // Remove the last element, because Go has already swapped 0th elem <-> last. + old := *eh + n := len(old) + x := old[n-1] + *eh = old[0 : n-1] + return x +} +func (eh elemHeap) Less(i, j int) bool { + cmp := CompareKeys(eh[i].itr.Key(), eh[j].itr.Key()) + if cmp < 0 { + return !eh[i].reversed + } + if cmp > 0 { + return eh[i].reversed + } + // The keys are equal. In this case, lower nice take precedence. This is important. + return eh[i].nice < eh[j].nice +} + +// MergeIterator merges multiple iterators. +// NOTE: MergeIterator owns the array of iterators and is responsible for closing them. +type MergeIterator struct { + h elemHeap + curKey []byte + reversed bool + + all []Iterator +} + +// NewMergeIterator returns a new MergeIterator from a list of Iterators. +func NewMergeIterator(iters []Iterator, reversed bool) *MergeIterator { + m := &MergeIterator{all: iters, reversed: reversed} + m.h = make(elemHeap, 0, len(iters)) + m.initHeap() + return m +} + +func (s *MergeIterator) storeKey(smallest Iterator) { + if cap(s.curKey) < len(smallest.Key()) { + s.curKey = make([]byte, 2*len(smallest.Key())) + } + s.curKey = s.curKey[:len(smallest.Key())] + copy(s.curKey, smallest.Key()) +} + +// initHeap checks all iterators and initializes our heap and array of keys. +// Whenever we reverse direction, we need to run this. +func (s *MergeIterator) initHeap() { + s.h = s.h[:0] + for idx, itr := range s.all { + if !itr.Valid() { + continue + } + e := &elem{itr: itr, nice: idx, reversed: s.reversed} + s.h = append(s.h, e) + } + heap.Init(&s.h) + for len(s.h) > 0 { + it := s.h[0].itr + if it == nil || !it.Valid() { + heap.Pop(&s.h) + continue + } + s.storeKey(s.h[0].itr) + break + } +} + +// Valid returns whether the MergeIterator is at a valid element. +func (s *MergeIterator) Valid() bool { + if s == nil { + return false + } + if len(s.h) == 0 { + return false + } + return s.h[0].itr.Valid() +} + +// Key returns the key associated with the current iterator +func (s *MergeIterator) Key() []byte { + if len(s.h) == 0 { + return nil + } + return s.h[0].itr.Key() +} + +// Value returns the value associated with the iterator. +func (s *MergeIterator) Value() ValueStruct { + if len(s.h) == 0 { + return ValueStruct{} + } + return s.h[0].itr.Value() +} + +// Next returns the next element. If it is the same as the current key, ignore it. +func (s *MergeIterator) Next() { + if len(s.h) == 0 { + return + } + + smallest := s.h[0].itr + smallest.Next() + + for len(s.h) > 0 { + smallest = s.h[0].itr + if !smallest.Valid() { + heap.Pop(&s.h) + continue + } + + heap.Fix(&s.h, 0) + smallest = s.h[0].itr + if smallest.Valid() { + if !bytes.Equal(smallest.Key(), s.curKey) { + break + } + smallest.Next() + } + } + if !smallest.Valid() { + return + } + s.storeKey(smallest) +} + +// Rewind seeks to first element (or last element for reverse iterator). +func (s *MergeIterator) Rewind() { + for _, itr := range s.all { + itr.Rewind() + } + s.initHeap() +} + +// Seek brings us to element with key >= given key. +func (s *MergeIterator) Seek(key []byte) { + for _, itr := range s.all { + itr.Seek(key) + } + s.initHeap() +} + +// Close implements y.Iterator +func (s *MergeIterator) Close() error { + for _, itr := range s.all { + if err := itr.Close(); err != nil { + return errors.Wrap(err, "MergeIterator") + } + } + return nil +} diff --git a/vendor/github.com/dgraph-io/badger/y/metrics.go b/vendor/github.com/dgraph-io/badger/y/metrics.go new file mode 100644 index 0000000000..2de17d1004 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/y/metrics.go @@ -0,0 +1,68 @@ +/* + * Copyright (C) 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package y + +import "expvar" + +var ( + // LSMSize has size of the LSM in bytes + LSMSize *expvar.Map + // VlogSize has size of the value log in bytes + VlogSize *expvar.Map + // PendingWrites tracks the number of pending writes. + PendingWrites *expvar.Map + + // These are cumulative + + // NumReads has cumulative number of reads + NumReads *expvar.Int + // NumWrites has cumulative number of writes + NumWrites *expvar.Int + // NumBytesRead has cumulative number of bytes read + NumBytesRead *expvar.Int + // NumBytesWritten has cumulative number of bytes written + NumBytesWritten *expvar.Int + // NumLSMGets is number of LMS gets + NumLSMGets *expvar.Map + // NumLSMBloomHits is number of LMS bloom hits + NumLSMBloomHits *expvar.Map + // NumGets is number of gets + NumGets *expvar.Int + // NumPuts is number of puts + NumPuts *expvar.Int + // NumBlockedPuts is number of blocked puts + NumBlockedPuts *expvar.Int + // NumMemtableGets is number of memtable gets + NumMemtableGets *expvar.Int +) + +// These variables are global and have cumulative values for all kv stores. +func init() { + NumReads = expvar.NewInt("badger_disk_reads_total") + NumWrites = expvar.NewInt("badger_disk_writes_total") + NumBytesRead = expvar.NewInt("badger_read_bytes") + NumBytesWritten = expvar.NewInt("badger_written_bytes") + NumLSMGets = expvar.NewMap("badger_lsm_level_gets_total") + NumLSMBloomHits = expvar.NewMap("badger_lsm_bloom_hits_total") + NumGets = expvar.NewInt("badger_gets_total") + NumPuts = expvar.NewInt("badger_puts_total") + NumBlockedPuts = expvar.NewInt("badger_blocked_puts_total") + NumMemtableGets = expvar.NewInt("badger_memtable_gets_total") + LSMSize = expvar.NewMap("badger_lsm_size_bytes") + VlogSize = expvar.NewMap("badger_vlog_size_bytes") + PendingWrites = expvar.NewMap("badger_pending_writes_total") +} diff --git a/vendor/github.com/dgraph-io/badger/y/mmap_unix.go b/vendor/github.com/dgraph-io/badger/y/mmap_unix.go new file mode 100644 index 0000000000..f9203a0139 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/y/mmap_unix.go @@ -0,0 +1,63 @@ +// +build !windows + +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package y + +import ( + "os" + "syscall" + "unsafe" + + "golang.org/x/sys/unix" +) + +// Mmap uses the mmap system call to memory-map a file. If writable is true, +// memory protection of the pages is set so that they may be written to as well. +func Mmap(fd *os.File, writable bool, size int64) ([]byte, error) { + mtype := unix.PROT_READ + if writable { + mtype |= unix.PROT_WRITE + } + return unix.Mmap(int(fd.Fd()), 0, int(size), mtype, unix.MAP_SHARED) +} + +// Munmap unmaps a previously mapped slice. +func Munmap(b []byte) error { + return unix.Munmap(b) +} + +// Madvise uses the madvise system call to give advise about the use of memory +// when using a slice that is memory-mapped to a file. Set the readahead flag to +// false if page references are expected in random order. +func Madvise(b []byte, readahead bool) error { + flags := unix.MADV_NORMAL + if !readahead { + flags = unix.MADV_RANDOM + } + return madvise(b, flags) +} + +// This is required because the unix package does not support the madvise system call on OS X. +func madvise(b []byte, advice int) (err error) { + _, _, e1 := syscall.Syscall(syscall.SYS_MADVISE, uintptr(unsafe.Pointer(&b[0])), + uintptr(len(b)), uintptr(advice)) + if e1 != 0 { + err = e1 + } + return +} diff --git a/vendor/github.com/dgraph-io/badger/y/mmap_windows.go b/vendor/github.com/dgraph-io/badger/y/mmap_windows.go new file mode 100644 index 0000000000..0efb2d0f8d --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/y/mmap_windows.go @@ -0,0 +1,90 @@ +// +build windows + +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package y + +import ( + "fmt" + "os" + "syscall" + "unsafe" +) + +func Mmap(fd *os.File, write bool, size int64) ([]byte, error) { + protect := syscall.PAGE_READONLY + access := syscall.FILE_MAP_READ + + if write { + protect = syscall.PAGE_READWRITE + access = syscall.FILE_MAP_WRITE + } + fi, err := fd.Stat() + if err != nil { + return nil, err + } + + // Truncate the database to the size of the mmap. + if fi.Size() < size { + if err := fd.Truncate(size); err != nil { + return nil, fmt.Errorf("truncate: %s", err) + } + } + + // Open a file mapping handle. + sizelo := uint32(size >> 32) + sizehi := uint32(size) & 0xffffffff + + handler, err := syscall.CreateFileMapping(syscall.Handle(fd.Fd()), nil, + uint32(protect), sizelo, sizehi, nil) + if err != nil { + return nil, os.NewSyscallError("CreateFileMapping", err) + } + + // Create the memory map. + addr, err := syscall.MapViewOfFile(handler, uint32(access), 0, 0, uintptr(size)) + if addr == 0 { + return nil, os.NewSyscallError("MapViewOfFile", err) + } + + // Close mapping handle. + if err := syscall.CloseHandle(syscall.Handle(handler)); err != nil { + return nil, os.NewSyscallError("CloseHandle", err) + } + + // Slice memory layout + // Copied this snippet from golang/sys package + var sl = struct { + addr uintptr + len int + cap int + }{addr, int(size), int(size)} + + // Use unsafe to turn sl into a []byte. + data := *(*[]byte)(unsafe.Pointer(&sl)) + + return data, nil +} + +func Munmap(b []byte) error { + return syscall.UnmapViewOfFile(uintptr(unsafe.Pointer(&b[0]))) +} + +func Madvise(b []byte, readahead bool) error { + // Do Nothing. We don’t care about this setting on Windows + return nil +} diff --git a/vendor/github.com/dgraph-io/badger/y/y.go b/vendor/github.com/dgraph-io/badger/y/y.go new file mode 100644 index 0000000000..20a8ea55e4 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/y/y.go @@ -0,0 +1,196 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package y + +import ( + "bytes" + "encoding/binary" + "hash/crc32" + "math" + "os" + "sync" + + "github.com/pkg/errors" +) + +// ErrEOF indicates an end of file when trying to read from a memory mapped file +// and encountering the end of slice. +var ErrEOF = errors.New("End of mapped region") + +var ( + // This is O_DSYNC (datasync) on platforms that support it -- see file_unix.go + datasyncFileFlag = 0x0 + + // CastagnoliCrcTable is a CRC32 polynomial table + CastagnoliCrcTable = crc32.MakeTable(crc32.Castagnoli) +) + +// OpenExistingSyncedFile opens an existing file, errors if it doesn't exist. +func OpenExistingSyncedFile(filename string, sync bool) (*os.File, error) { + flags := os.O_RDWR + if sync { + flags |= datasyncFileFlag + } + return os.OpenFile(filename, flags, 0) +} + +// CreateSyncedFile creates a new file (using O_EXCL), errors if it already existed. +func CreateSyncedFile(filename string, sync bool) (*os.File, error) { + flags := os.O_RDWR | os.O_CREATE | os.O_EXCL + if sync { + flags |= datasyncFileFlag + } + return os.OpenFile(filename, flags, 0666) +} + +// OpenSyncedFile creates the file if one doesn't exist. +func OpenSyncedFile(filename string, sync bool) (*os.File, error) { + flags := os.O_RDWR | os.O_CREATE + if sync { + flags |= datasyncFileFlag + } + return os.OpenFile(filename, flags, 0666) +} + +// OpenTruncFile opens the file with O_RDWR | O_CREATE | O_TRUNC +func OpenTruncFile(filename string, sync bool) (*os.File, error) { + flags := os.O_RDWR | os.O_CREATE | os.O_TRUNC + if sync { + flags |= datasyncFileFlag + } + return os.OpenFile(filename, flags, 0666) +} + +// SafeCopy does append(a[:0], src...). +func SafeCopy(a []byte, src []byte) []byte { + return append(a[:0], src...) +} + +// Copy copies a byte slice and returns the copied slice. +func Copy(a []byte) []byte { + b := make([]byte, len(a)) + copy(b, a) + return b +} + +// KeyWithTs generates a new key by appending ts to key. +func KeyWithTs(key []byte, ts uint64) []byte { + out := make([]byte, len(key)+8) + copy(out, key) + binary.BigEndian.PutUint64(out[len(key):], math.MaxUint64-ts) + return out +} + +// ParseTs parses the timestamp from the key bytes. +func ParseTs(key []byte) uint64 { + if len(key) <= 8 { + return 0 + } + return math.MaxUint64 - binary.BigEndian.Uint64(key[len(key)-8:]) +} + +// CompareKeys checks the key without timestamp and checks the timestamp if keyNoTs +// is same. +// a would be sorted higher than aa if we use bytes.compare +// All keys should have timestamp. +func CompareKeys(key1 []byte, key2 []byte) int { + AssertTrue(len(key1) > 8 && len(key2) > 8) + if cmp := bytes.Compare(key1[:len(key1)-8], key2[:len(key2)-8]); cmp != 0 { + return cmp + } + return bytes.Compare(key1[len(key1)-8:], key2[len(key2)-8:]) +} + +// ParseKey parses the actual key from the key bytes. +func ParseKey(key []byte) []byte { + if key == nil { + return nil + } + + AssertTruef(len(key) > 8, "key=%q", key) + return key[:len(key)-8] +} + +// SameKey checks for key equality ignoring the version timestamp suffix. +func SameKey(src, dst []byte) bool { + if len(src) != len(dst) { + return false + } + return bytes.Equal(ParseKey(src), ParseKey(dst)) +} + +// Slice holds a reusable buf, will reallocate if you request a larger size than ever before. +// One problem is with n distinct sizes in random order it'll reallocate log(n) times. +type Slice struct { + buf []byte +} + +// Resize reuses the Slice's buffer (or makes a new one) and returns a slice in that buffer of +// length sz. +func (s *Slice) Resize(sz int) []byte { + if cap(s.buf) < sz { + s.buf = make([]byte, sz) + } + return s.buf[0:sz] +} + +// Closer holds the two things we need to close a goroutine and wait for it to finish: a chan +// to tell the goroutine to shut down, and a WaitGroup with which to wait for it to finish shutting +// down. +type Closer struct { + closed chan struct{} + waiting sync.WaitGroup +} + +// NewCloser constructs a new Closer, with an initial count on the WaitGroup. +func NewCloser(initial int) *Closer { + ret := &Closer{closed: make(chan struct{})} + ret.waiting.Add(initial) + return ret +} + +// AddRunning Add()'s delta to the WaitGroup. +func (lc *Closer) AddRunning(delta int) { + lc.waiting.Add(delta) +} + +// Signal signals the HasBeenClosed signal. +func (lc *Closer) Signal() { + close(lc.closed) +} + +// HasBeenClosed gets signaled when Signal() is called. +func (lc *Closer) HasBeenClosed() <-chan struct{} { + return lc.closed +} + +// Done calls Done() on the WaitGroup. +func (lc *Closer) Done() { + lc.waiting.Done() +} + +// Wait waits on the WaitGroup. (It waits for NewCloser's initial value, AddRunning, and Done +// calls to balance out.) +func (lc *Closer) Wait() { + lc.waiting.Wait() +} + +// SignalAndWait calls Signal(), then Wait(). +func (lc *Closer) SignalAndWait() { + lc.Signal() + lc.Wait() +} diff --git a/vendor/github.com/dgryski/go-farm/LICENSE b/vendor/github.com/dgryski/go-farm/LICENSE new file mode 100644 index 0000000000..3d07f6662d --- /dev/null +++ b/vendor/github.com/dgryski/go-farm/LICENSE @@ -0,0 +1,23 @@ +As this is a highly derivative work, I have placed it under the same license as the original implementation: + +Copyright (c) 2014-2017 Damian Gryski +Copyright (c) 2016-2017 Nicola Asuni - Tecnick.com + +Permission is hereby granted, free of charge, to any person obtaining a copy +of this software and associated documentation files (the "Software"), to deal +in the Software without restriction, including without limitation the rights +to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +copies of the Software, and to permit persons to whom the Software is +furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice shall be included in +all copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +THE SOFTWARE. + diff --git a/vendor/github.com/dgryski/go-farm/Makefile b/vendor/github.com/dgryski/go-farm/Makefile new file mode 100644 index 0000000000..f01244b191 --- /dev/null +++ b/vendor/github.com/dgryski/go-farm/Makefile @@ -0,0 +1,203 @@ +# MAKEFILE +# +# @author Nicola Asuni +# @link https://github.com/dgryski/go-farm +# +# This file is intended to be executed in a Linux-compatible system. +# It also assumes that the project has been cloned in the right path under GOPATH: +# $GOPATH/src/github.com/dgryski/go-farm +# +# ------------------------------------------------------------------------------ + +# List special make targets that are not associated with files +.PHONY: help all test format fmtcheck vet lint coverage cyclo ineffassign misspell structcheck varcheck errcheck gosimple astscan qa deps clean nuke + +# Use bash as shell (Note: Ubuntu now uses dash which doesn't support PIPESTATUS). +SHELL=/bin/bash + +# CVS path (path to the parent dir containing the project) +CVSPATH=github.com/dgryski + +# Project owner +OWNER=dgryski + +# Project vendor +VENDOR=dgryski + +# Project name +PROJECT=go-farm + +# Project version +VERSION=$(shell cat VERSION) + +# Name of RPM or DEB package +PKGNAME=${VENDOR}-${PROJECT} + +# Current directory +CURRENTDIR=$(shell pwd) + +# GO lang path +ifneq ($(GOPATH),) + ifeq ($(findstring $(GOPATH),$(CURRENTDIR)),) + # the defined GOPATH is not valid + GOPATH= + endif +endif +ifeq ($(GOPATH),) + # extract the GOPATH + GOPATH=$(firstword $(subst /src/, ,$(CURRENTDIR))) +endif + +# --- MAKE TARGETS --- + +# Display general help about this command +help: + @echo "" + @echo "$(PROJECT) Makefile." + @echo "GOPATH=$(GOPATH)" + @echo "The following commands are available:" + @echo "" + @echo " make qa : Run all the tests" + @echo " make test : Run the unit tests" + @echo "" + @echo " make format : Format the source code" + @echo " make fmtcheck : Check if the source code has been formatted" + @echo " make vet : Check for suspicious constructs" + @echo " make lint : Check for style errors" + @echo " make coverage : Generate the coverage report" + @echo " make cyclo : Generate the cyclomatic complexity report" + @echo " make ineffassign : Detect ineffectual assignments" + @echo " make misspell : Detect commonly misspelled words in source files" + @echo " make structcheck : Find unused struct fields" + @echo " make varcheck : Find unused global variables and constants" + @echo " make errcheck : Check that error return values are used" + @echo " make gosimple : Suggest code simplifications" + @echo " make astscan : GO AST scanner" + @echo "" + @echo " make docs : Generate source code documentation" + @echo "" + @echo " make deps : Get the dependencies" + @echo " make clean : Remove any build artifact" + @echo " make nuke : Deletes any intermediate file" + @echo "" + + +# Alias for help target +all: help + +# Run the unit tests +test: + @mkdir -p target/test + @mkdir -p target/report + GOPATH=$(GOPATH) \ + go test \ + -covermode=atomic \ + -bench=. \ + -race \ + -cpuprofile=target/report/cpu.out \ + -memprofile=target/report/mem.out \ + -mutexprofile=target/report/mutex.out \ + -coverprofile=target/report/coverage.out \ + -v ./... | \ + tee >(PATH=$(GOPATH)/bin:$(PATH) go-junit-report > target/test/report.xml); \ + test $${PIPESTATUS[0]} -eq 0 + +# Format the source code +format: + @find . -type f -name "*.go" -exec gofmt -s -w {} \; + +# Check if the source code has been formatted +fmtcheck: + @mkdir -p target + @find . -type f -name "*.go" -exec gofmt -s -d {} \; | tee target/format.diff + @test ! -s target/format.diff || { echo "ERROR: the source code has not been formatted - please use 'make format' or 'gofmt'"; exit 1; } + +# Check for syntax errors +vet: + GOPATH=$(GOPATH) go vet . + +# Check for style errors +lint: + GOPATH=$(GOPATH) PATH=$(GOPATH)/bin:$(PATH) golint . + +# Generate the coverage report +coverage: + @mkdir -p target/report + GOPATH=$(GOPATH) \ + go tool cover -html=target/report/coverage.out -o target/report/coverage.html + +# Report cyclomatic complexity +cyclo: + @mkdir -p target/report + GOPATH=$(GOPATH) gocyclo -avg ./ | tee target/report/cyclo.txt ; test $${PIPESTATUS[0]} -eq 0 + +# Detect ineffectual assignments +ineffassign: + @mkdir -p target/report + GOPATH=$(GOPATH) ineffassign ./ | tee target/report/ineffassign.txt ; test $${PIPESTATUS[0]} -eq 0 + +# Detect commonly misspelled words in source files +misspell: + @mkdir -p target/report + GOPATH=$(GOPATH) misspell -error ./ | tee target/report/misspell.txt ; test $${PIPESTATUS[0]} -eq 0 + +# Find unused struct fields +structcheck: + @mkdir -p target/report + GOPATH=$(GOPATH) structcheck -a ./ | tee target/report/structcheck.txt + +# Find unused global variables and constants +varcheck: + @mkdir -p target/report + GOPATH=$(GOPATH) varcheck -e ./ | tee target/report/varcheck.txt + +# Check that error return values are used +errcheck: + @mkdir -p target/report + GOPATH=$(GOPATH) errcheck ./ | tee target/report/errcheck.txt + +# Suggest code simplifications +gosimple: + @mkdir -p target/report + GOPATH=$(GOPATH) gosimple ./ | tee target/report/gosimple.txt + +# AST scanner +astscan: + @mkdir -p target/report + GOPATH=$(GOPATH) gas .//*.go | tee target/report/astscan.txt + +# Generate source docs +docs: + @mkdir -p target/docs + nohup sh -c 'GOPATH=$(GOPATH) godoc -http=127.0.0.1:6060' > target/godoc_server.log 2>&1 & + wget --directory-prefix=target/docs/ --execute robots=off --retry-connrefused --recursive --no-parent --adjust-extension --page-requisites --convert-links http://127.0.0.1:6060/pkg/github.com/${VENDOR}/${PROJECT}/ ; kill -9 `lsof -ti :6060` + @echo ''${PKGNAME}' Documentation ...' > target/docs/index.html + +# Alias to run all quality-assurance checks +qa: fmtcheck test vet lint coverage cyclo ineffassign misspell structcheck varcheck errcheck gosimple astscan + +# --- INSTALL --- + +# Get the dependencies +deps: + GOPATH=$(GOPATH) go get ./... + GOPATH=$(GOPATH) go get github.com/golang/lint/golint + GOPATH=$(GOPATH) go get github.com/jstemmer/go-junit-report + GOPATH=$(GOPATH) go get github.com/axw/gocov/gocov + GOPATH=$(GOPATH) go get github.com/fzipp/gocyclo + GOPATH=$(GOPATH) go get github.com/gordonklaus/ineffassign + GOPATH=$(GOPATH) go get github.com/client9/misspell/cmd/misspell + GOPATH=$(GOPATH) go get github.com/opennota/check/cmd/structcheck + GOPATH=$(GOPATH) go get github.com/opennota/check/cmd/varcheck + GOPATH=$(GOPATH) go get github.com/kisielk/errcheck + GOPATH=$(GOPATH) go get honnef.co/go/tools/cmd/gosimple + GOPATH=$(GOPATH) go get github.com/GoASTScanner/gas + +# Remove any build artifact +clean: + GOPATH=$(GOPATH) go clean ./... + +# Deletes any intermediate file +nuke: + rm -rf ./target + GOPATH=$(GOPATH) go clean -i ./... diff --git a/vendor/github.com/dgryski/go-farm/README.md b/vendor/github.com/dgryski/go-farm/README.md new file mode 100644 index 0000000000..dd07d6f991 --- /dev/null +++ b/vendor/github.com/dgryski/go-farm/README.md @@ -0,0 +1,41 @@ +# go-farm + +*Google's FarmHash hash functions implemented in Go* + +[![Master Branch](https://img.shields.io/badge/-master:-gray.svg)](https://github.com/dgryski/go-farm/tree/master) +[![Master Build Status](https://secure.travis-ci.org/dgryski/go-farm.png?branch=master)](https://travis-ci.org/dgryski/go-farm?branch=master) +[![Master Coverage Status](https://coveralls.io/repos/dgryski/go-farm/badge.svg?branch=master&service=github)](https://coveralls.io/github/dgryski/go-farm?branch=master) +[![Go Report Card](https://goreportcard.com/badge/github.com/dgryski/go-farm)](https://goreportcard.com/report/github.com/dgryski/go-farm) +[![GoDoc](https://godoc.org/github.com/dgryski/go-farm?status.svg)](http://godoc.org/github.com/dgryski/go-farm) + +## Description + +FarmHash, a family of hash functions. + +This is a (mechanical) translation of the non-SSE4/non-AESNI hash functions from Google's FarmHash (https://github.com/google/farmhash). + + +FarmHash provides hash functions for strings and other data. +The functions mix the input bits thoroughly but are not suitable for cryptography. + +All members of the FarmHash family were designed with heavy reliance on previous work by Jyrki Alakuijala, Austin Appleby, Bob Jenkins, and others. + +For more information please consult https://github.com/google/farmhash + + +## Getting started + +This application is written in Go language, please refer to the guides in https://golang.org for getting started. + +This project include a Makefile that allows you to test and build the project with simple commands. +To see all available options: +```bash +make help +``` + +## Running all tests + +Before committing the code, please check if it passes all tests using +```bash +make qa +``` diff --git a/vendor/github.com/dgryski/go-farm/VERSION b/vendor/github.com/dgryski/go-farm/VERSION new file mode 100644 index 0000000000..38f77a65b3 --- /dev/null +++ b/vendor/github.com/dgryski/go-farm/VERSION @@ -0,0 +1 @@ +2.0.1 diff --git a/vendor/github.com/dgryski/go-farm/basics.go b/vendor/github.com/dgryski/go-farm/basics.go new file mode 100644 index 0000000000..f27a422081 --- /dev/null +++ b/vendor/github.com/dgryski/go-farm/basics.go @@ -0,0 +1,30 @@ +package farm + +// Some primes between 2^63 and 2^64 for various uses. +const k0 uint64 = 0xc3a5c85c97cb3127 +const k1 uint64 = 0xb492b66fbe98f273 +const k2 uint64 = 0x9ae16a3b2f90404f + +// Magic numbers for 32-bit hashing. Copied from Murmur3. +const c1 uint32 = 0xcc9e2d51 +const c2 uint32 = 0x1b873593 + +// A 32-bit to 32-bit integer hash copied from Murmur3. +func fmix(h uint32) uint32 { + h ^= h >> 16 + h *= 0x85ebca6b + h ^= h >> 13 + h *= 0xc2b2ae35 + h ^= h >> 16 + return h +} + +func mur(a, h uint32) uint32 { + // Helper from Murmur3 for combining two 32-bit values. + a *= c1 + a = rotate32(a, 17) + a *= c2 + h ^= a + h = rotate32(h, 19) + return h*5 + 0xe6546b64 +} diff --git a/vendor/github.com/dgryski/go-farm/farmhashcc.go b/vendor/github.com/dgryski/go-farm/farmhashcc.go new file mode 100644 index 0000000000..12e5c04151 --- /dev/null +++ b/vendor/github.com/dgryski/go-farm/farmhashcc.go @@ -0,0 +1,199 @@ +package farm + +// This file provides a 32-bit hash equivalent to CityHash32 (v1.1.1) +// and a 128-bit hash equivalent to CityHash128 (v1.1.1). It also provides +// a seeded 32-bit hash function similar to CityHash32. + +func hash32Len13to24Seed(s []byte, seed uint32) uint32 { + slen := len(s) + a := fetch32(s, -4+(slen>>1)) + b := fetch32(s, 4) + c := fetch32(s, slen-8) + d := fetch32(s, (slen >> 1)) + e := fetch32(s, 0) + f := fetch32(s, slen-4) + h := d*c1 + uint32(slen) + seed + a = rotate32(a, 12) + f + h = mur(c, h) + a + a = rotate32(a, 3) + c + h = mur(e, h) + a + a = rotate32(a+f, 12) + d + h = mur(b^seed, h) + a + return fmix(h) +} + +func hash32Len0to4(s []byte, seed uint32) uint32 { + slen := len(s) + b := seed + c := uint32(9) + for i := 0; i < slen; i++ { + v := int8(s[i]) + b = (b * c1) + uint32(v) + c ^= b + } + return fmix(mur(b, mur(uint32(slen), c))) +} + +func hash128to64(x uint128) uint64 { + // Murmur-inspired hashing. + const mul uint64 = 0x9ddfea08eb382d69 + a := (x.lo ^ x.hi) * mul + a ^= (a >> 47) + b := (x.hi ^ a) * mul + b ^= (b >> 47) + b *= mul + return b +} + +type uint128 struct { + lo uint64 + hi uint64 +} + +// A subroutine for CityHash128(). Returns a decent 128-bit hash for strings +// of any length representable in signed long. Based on City and Murmur. +func cityMurmur(s []byte, seed uint128) uint128 { + slen := len(s) + a := seed.lo + b := seed.hi + var c uint64 + var d uint64 + l := slen - 16 + if l <= 0 { // len <= 16 + a = shiftMix(a*k1) * k1 + c = b*k1 + hashLen0to16(s) + if slen >= 8 { + d = shiftMix(a + fetch64(s, 0)) + } else { + d = shiftMix(a + c) + } + } else { // len > 16 + c = hashLen16(fetch64(s, slen-8)+k1, a) + d = hashLen16(b+uint64(slen), c+fetch64(s, slen-16)) + a += d + for { + a ^= shiftMix(fetch64(s, 0)*k1) * k1 + a *= k1 + b ^= a + c ^= shiftMix(fetch64(s, 8)*k1) * k1 + c *= k1 + d ^= c + s = s[16:] + l -= 16 + if l <= 0 { + break + } + } + } + a = hashLen16(a, c) + b = hashLen16(d, b) + return uint128{a ^ b, hashLen16(b, a)} +} + +func cityHash128WithSeed(s []byte, seed uint128) uint128 { + slen := len(s) + if slen < 128 { + return cityMurmur(s, seed) + } + + endIdx := ((slen - 1) / 128) * 128 + lastBlockIdx := endIdx + ((slen - 1) & 127) - 127 + last := s[lastBlockIdx:] + + // We expect len >= 128 to be the common case. Keep 56 bytes of state: + // v, w, x, y, and z. + var v1, v2 uint64 + var w1, w2 uint64 + x := seed.lo + y := seed.hi + z := uint64(slen) * k1 + v1 = rotate64(y^k1, 49)*k1 + fetch64(s, 0) + v2 = rotate64(v1, 42)*k1 + fetch64(s, 8) + w1 = rotate64(y+z, 35)*k1 + x + w2 = rotate64(x+fetch64(s, 88), 53) * k1 + + // This is the same inner loop as CityHash64(), manually unrolled. + for { + x = rotate64(x+y+v1+fetch64(s, 8), 37) * k1 + y = rotate64(y+v2+fetch64(s, 48), 42) * k1 + x ^= w2 + y += v1 + fetch64(s, 40) + z = rotate64(z+w1, 33) * k1 + v1, v2 = weakHashLen32WithSeeds(s, v2*k1, x+w1) + w1, w2 = weakHashLen32WithSeeds(s[32:], z+w2, y+fetch64(s, 16)) + z, x = x, z + s = s[64:] + x = rotate64(x+y+v1+fetch64(s, 8), 37) * k1 + y = rotate64(y+v2+fetch64(s, 48), 42) * k1 + x ^= w2 + y += v1 + fetch64(s, 40) + z = rotate64(z+w1, 33) * k1 + v1, v2 = weakHashLen32WithSeeds(s, v2*k1, x+w1) + w1, w2 = weakHashLen32WithSeeds(s[32:], z+w2, y+fetch64(s, 16)) + z, x = x, z + s = s[64:] + slen -= 128 + if slen < 128 { + break + } + } + x += rotate64(v1+z, 49) * k0 + y = y*k0 + rotate64(w2, 37) + z = z*k0 + rotate64(w1, 27) + w1 *= 9 + v1 *= k0 + // If 0 < len < 128, hash up to 4 chunks of 32 bytes each from the end of s. + for tailDone := 0; tailDone < slen; { + tailDone += 32 + y = rotate64(x+y, 42)*k0 + v2 + w1 += fetch64(last, 128-tailDone+16) + x = x*k0 + w1 + z += w2 + fetch64(last, 128-tailDone) + w2 += v1 + v1, v2 = weakHashLen32WithSeeds(last[128-tailDone:], v1+z, v2) + v1 *= k0 + } + + // At this point our 56 bytes of state should contain more than + // enough information for a strong 128-bit hash. We use two + // different 56-byte-to-8-byte hashes to get a 16-byte final result. + x = hashLen16(x, v1) + y = hashLen16(y+z, w1) + return uint128{hashLen16(x+v2, w2) + y, + hashLen16(x+w2, y+v2)} +} + +func cityHash128(s []byte) uint128 { + slen := len(s) + if slen >= 16 { + return cityHash128WithSeed(s[16:], uint128{fetch64(s, 0), fetch64(s, 8) + k0}) + } + return cityHash128WithSeed(s, uint128{k0, k1}) +} + +// Fingerprint128 is a 128-bit fingerprint function for byte-slices +func Fingerprint128(s []byte) (lo, hi uint64) { + h := cityHash128(s) + return h.lo, h.hi +} + +// Fingerprint64 is a 64-bit fingerprint function for byte-slices +func Fingerprint64(s []byte) uint64 { + return naHash64(s) +} + +// Fingerprint32 is a 32-bit fingerprint function for byte-slices +func Fingerprint32(s []byte) uint32 { + return Hash32(s) +} + +// Hash128 is a 128-bit hash function for byte-slices +func Hash128(s []byte) (lo, hi uint64) { + return Fingerprint128(s) +} + +// Hash128WithSeed is a 128-bit hash function for byte-slices and a 128-bit seed +func Hash128WithSeed(s []byte, seed0, seed1 uint64) (lo, hi uint64) { + h := cityHash128WithSeed(s, uint128{seed0, seed1}) + return h.lo, h.hi +} diff --git a/vendor/github.com/dgryski/go-farm/farmhashmk.go b/vendor/github.com/dgryski/go-farm/farmhashmk.go new file mode 100644 index 0000000000..ee41bd7b83 --- /dev/null +++ b/vendor/github.com/dgryski/go-farm/farmhashmk.go @@ -0,0 +1,102 @@ +package farm + +func hash32Len5to12(s []byte, seed uint32) uint32 { + slen := len(s) + a := uint32(len(s)) + b := uint32(len(s) * 5) + c := uint32(9) + d := b + seed + a += fetch32(s, 0) + b += fetch32(s, slen-4) + c += fetch32(s, ((slen >> 1) & 4)) + return fmix(seed ^ mur(c, mur(b, mur(a, d)))) +} + +// Hash32 hashes a byte slice and returns a uint32 hash value +func Hash32(s []byte) uint32 { + + slen := len(s) + + if slen <= 24 { + if slen <= 12 { + if slen <= 4 { + return hash32Len0to4(s, 0) + } + return hash32Len5to12(s, 0) + } + return hash32Len13to24Seed(s, 0) + } + + // len > 24 + h := uint32(slen) + g := c1 * uint32(slen) + f := g + a0 := rotate32(fetch32(s, slen-4)*c1, 17) * c2 + a1 := rotate32(fetch32(s, slen-8)*c1, 17) * c2 + a2 := rotate32(fetch32(s, slen-16)*c1, 17) * c2 + a3 := rotate32(fetch32(s, slen-12)*c1, 17) * c2 + a4 := rotate32(fetch32(s, slen-20)*c1, 17) * c2 + h ^= a0 + h = rotate32(h, 19) + h = h*5 + 0xe6546b64 + h ^= a2 + h = rotate32(h, 19) + h = h*5 + 0xe6546b64 + g ^= a1 + g = rotate32(g, 19) + g = g*5 + 0xe6546b64 + g ^= a3 + g = rotate32(g, 19) + g = g*5 + 0xe6546b64 + f += a4 + f = rotate32(f, 19) + 113 + iters := (slen - 1) / 20 + for { + a := fetch32(s, 0) + b := fetch32(s, 4) + c := fetch32(s, 8) + d := fetch32(s, 12) + e := fetch32(s, 16) + h += a + g += b + f += c + h = mur(d, h) + e + g = mur(c, g) + a + f = mur(b+e*c1, f) + d + f += g + g += f + s = s[20:] + iters-- + if iters == 0 { + break + } + } + g = rotate32(g, 11) * c1 + g = rotate32(g, 17) * c1 + f = rotate32(f, 11) * c1 + f = rotate32(f, 17) * c1 + h = rotate32(h+g, 19) + h = h*5 + 0xe6546b64 + h = rotate32(h, 17) * c1 + h = rotate32(h+f, 19) + h = h*5 + 0xe6546b64 + h = rotate32(h, 17) * c1 + return h +} + +// Hash32WithSeed hashes a byte slice and a uint32 seed and returns a uint32 hash value +func Hash32WithSeed(s []byte, seed uint32) uint32 { + slen := len(s) + + if slen <= 24 { + if slen >= 13 { + return hash32Len13to24Seed(s, seed*c1) + } + if slen >= 5 { + return hash32Len5to12(s, seed) + } + return hash32Len0to4(s, seed) + } + h := hash32Len13to24Seed(s[:24], seed^uint32(slen)) + return mur(Hash32(s[24:])+seed, h) +} diff --git a/vendor/github.com/dgryski/go-farm/farmhashna.go b/vendor/github.com/dgryski/go-farm/farmhashna.go new file mode 100644 index 0000000000..574925e01a --- /dev/null +++ b/vendor/github.com/dgryski/go-farm/farmhashna.go @@ -0,0 +1,156 @@ +package farm + +func shiftMix(val uint64) uint64 { + return val ^ (val >> 47) +} + +func hashLen16(u, v uint64) uint64 { + return hash128to64(uint128{u, v}) +} + +func hashLen16Mul(u, v, mul uint64) uint64 { + // Murmur-inspired hashing. + a := (u ^ v) * mul + a ^= (a >> 47) + b := (v ^ a) * mul + b ^= (b >> 47) + b *= mul + return b +} + +func hashLen0to16(s []byte) uint64 { + slen := uint64(len(s)) + if slen >= 8 { + mul := k2 + slen*2 + a := fetch64(s, 0) + k2 + b := fetch64(s, int(slen-8)) + c := rotate64(b, 37)*mul + a + d := (rotate64(a, 25) + b) * mul + return hashLen16Mul(c, d, mul) + } + + if slen >= 4 { + mul := k2 + slen*2 + a := fetch32(s, 0) + return hashLen16Mul(slen+(uint64(a)<<3), uint64(fetch32(s, int(slen-4))), mul) + } + if slen > 0 { + a := s[0] + b := s[slen>>1] + c := s[slen-1] + y := uint32(a) + (uint32(b) << 8) + z := uint32(slen) + (uint32(c) << 2) + return shiftMix(uint64(y)*k2^uint64(z)*k0) * k2 + } + return k2 +} + +// This probably works well for 16-byte strings as well, but it may be overkill +// in that case. +func hashLen17to32(s []byte) uint64 { + slen := len(s) + mul := k2 + uint64(slen*2) + a := fetch64(s, 0) * k1 + b := fetch64(s, 8) + c := fetch64(s, slen-8) * mul + d := fetch64(s, slen-16) * k2 + return hashLen16Mul(rotate64(a+b, 43)+rotate64(c, 30)+d, a+rotate64(b+k2, 18)+c, mul) +} + +// Return a 16-byte hash for 48 bytes. Quick and dirty. +// Callers do best to use "random-looking" values for a and b. +func weakHashLen32WithSeedsWords(w, x, y, z, a, b uint64) (uint64, uint64) { + a += w + b = rotate64(b+a+z, 21) + c := a + a += x + a += y + b += rotate64(a, 44) + return a + z, b + c +} + +// Return a 16-byte hash for s[0] ... s[31], a, and b. Quick and dirty. +func weakHashLen32WithSeeds(s []byte, a, b uint64) (uint64, uint64) { + return weakHashLen32WithSeedsWords(fetch64(s, 0), + fetch64(s, 8), + fetch64(s, 16), + fetch64(s, 24), + a, + b) +} + +// Return an 8-byte hash for 33 to 64 bytes. +func hashLen33to64(s []byte) uint64 { + slen := len(s) + mul := k2 + uint64(slen)*2 + a := fetch64(s, 0) * k2 + b := fetch64(s, 8) + c := fetch64(s, slen-8) * mul + d := fetch64(s, slen-16) * k2 + y := rotate64(a+b, 43) + rotate64(c, 30) + d + z := hashLen16Mul(y, a+rotate64(b+k2, 18)+c, mul) + e := fetch64(s, 16) * mul + f := fetch64(s, 24) + g := (y + fetch64(s, slen-32)) * mul + h := (z + fetch64(s, slen-24)) * mul + return hashLen16Mul(rotate64(e+f, 43)+rotate64(g, 30)+h, e+rotate64(f+a, 18)+g, mul) +} + +func naHash64(s []byte) uint64 { + slen := len(s) + var seed uint64 = 81 + if slen <= 32 { + if slen <= 16 { + return hashLen0to16(s) + } + return hashLen17to32(s) + } + if slen <= 64 { + return hashLen33to64(s) + } + // For strings over 64 bytes we loop. + // Internal state consists of 56 bytes: v, w, x, y, and z. + v := uint128{0, 0} + w := uint128{0, 0} + x := seed*k2 + fetch64(s, 0) + y := seed*k1 + 113 + z := shiftMix(y*k2+113) * k2 + // Set end so that after the loop we have 1 to 64 bytes left to process. + endIdx := ((slen - 1) / 64) * 64 + last64Idx := endIdx + ((slen - 1) & 63) - 63 + last64 := s[last64Idx:] + for len(s) > 64 { + x = rotate64(x+y+v.lo+fetch64(s, 8), 37) * k1 + y = rotate64(y+v.hi+fetch64(s, 48), 42) * k1 + x ^= w.hi + y += v.lo + fetch64(s, 40) + z = rotate64(z+w.lo, 33) * k1 + v.lo, v.hi = weakHashLen32WithSeeds(s, v.hi*k1, x+w.lo) + w.lo, w.hi = weakHashLen32WithSeeds(s[32:], z+w.hi, y+fetch64(s, 16)) + x, z = z, x + s = s[64:] + } + mul := k1 + ((z & 0xff) << 1) + // Make s point to the last 64 bytes of input. + s = last64 + w.lo += (uint64(slen-1) & 63) + v.lo += w.lo + w.lo += v.lo + x = rotate64(x+y+v.lo+fetch64(s, 8), 37) * mul + y = rotate64(y+v.hi+fetch64(s, 48), 42) * mul + x ^= w.hi * 9 + y += v.lo*9 + fetch64(s, 40) + z = rotate64(z+w.lo, 33) * mul + v.lo, v.hi = weakHashLen32WithSeeds(s, v.hi*mul, x+w.lo) + w.lo, w.hi = weakHashLen32WithSeeds(s[32:], z+w.hi, y+fetch64(s, 16)) + x, z = z, x + return hashLen16Mul(hashLen16Mul(v.lo, w.lo, mul)+shiftMix(y)*k0+z, hashLen16Mul(v.hi, w.hi, mul)+x, mul) +} + +func naHash64WithSeed(s []byte, seed uint64) uint64 { + return naHash64WithSeeds(s, k2, seed) +} + +func naHash64WithSeeds(s []byte, seed0, seed1 uint64) uint64 { + return hashLen16(naHash64(s)-seed0, seed1) +} diff --git a/vendor/github.com/dgryski/go-farm/farmhashuo.go b/vendor/github.com/dgryski/go-farm/farmhashuo.go new file mode 100644 index 0000000000..7bf3d4259b --- /dev/null +++ b/vendor/github.com/dgryski/go-farm/farmhashuo.go @@ -0,0 +1,117 @@ +package farm + +func uoH(x, y, mul uint64, r uint) uint64 { + a := (x ^ y) * mul + a ^= (a >> 47) + b := (y ^ a) * mul + return rotate64(b, r) * mul +} + +// Hash64WithSeeds hashes a byte slice and two uint64 seeds and returns a uint64 hash value +func Hash64WithSeeds(s []byte, seed0, seed1 uint64) uint64 { + slen := len(s) + if slen <= 64 { + return naHash64WithSeeds(s, seed0, seed1) + } + + // For strings over 64 bytes we loop. + // Internal state consists of 64 bytes: u, v, w, x, y, and z. + x := seed0 + y := seed1*k2 + 113 + z := shiftMix(y*k2) * k2 + v := uint128{seed0, seed1} + var w uint128 + u := x - z + x *= k2 + mul := k2 + (u & 0x82) + + // Set end so that after the loop we have 1 to 64 bytes left to process. + endIdx := ((slen - 1) / 64) * 64 + last64Idx := endIdx + ((slen - 1) & 63) - 63 + last64 := s[last64Idx:] + + for len(s) > 64 { + a0 := fetch64(s, 0) + a1 := fetch64(s, 8) + a2 := fetch64(s, 16) + a3 := fetch64(s, 24) + a4 := fetch64(s, 32) + a5 := fetch64(s, 40) + a6 := fetch64(s, 48) + a7 := fetch64(s, 56) + x += a0 + a1 + y += a2 + z += a3 + v.lo += a4 + v.hi += a5 + a1 + w.lo += a6 + w.hi += a7 + + x = rotate64(x, 26) + x *= 9 + y = rotate64(y, 29) + z *= mul + v.lo = rotate64(v.lo, 33) + v.hi = rotate64(v.hi, 30) + w.lo ^= x + w.lo *= 9 + z = rotate64(z, 32) + z += w.hi + w.hi += z + z *= 9 + u, y = y, u + + z += a0 + a6 + v.lo += a2 + v.hi += a3 + w.lo += a4 + w.hi += a5 + a6 + x += a1 + y += a7 + + y += v.lo + v.lo += x - y + v.hi += w.lo + w.lo += v.hi + w.hi += x - y + x += w.hi + w.hi = rotate64(w.hi, 34) + u, z = z, u + s = s[64:] + } + // Make s point to the last 64 bytes of input. + s = last64 + u *= 9 + v.hi = rotate64(v.hi, 28) + v.lo = rotate64(v.lo, 20) + w.lo += (uint64(slen-1) & 63) + u += y + y += u + x = rotate64(y-x+v.lo+fetch64(s, 8), 37) * mul + y = rotate64(y^v.hi^fetch64(s, 48), 42) * mul + x ^= w.hi * 9 + y += v.lo + fetch64(s, 40) + z = rotate64(z+w.lo, 33) * mul + v.lo, v.hi = weakHashLen32WithSeeds(s, v.hi*mul, x+w.lo) + w.lo, w.hi = weakHashLen32WithSeeds(s[32:], z+w.hi, y+fetch64(s, 16)) + return uoH(hashLen16Mul(v.lo+x, w.lo^y, mul)+z-u, + uoH(v.hi+y, w.hi+z, k2, 30)^x, + k2, + 31) +} + +// Hash64WithSeed hashes a byte slice and a uint64 seed and returns a uint64 hash value +func Hash64WithSeed(s []byte, seed uint64) uint64 { + if len(s) <= 64 { + return naHash64WithSeed(s, seed) + } + return Hash64WithSeeds(s, 0, seed) +} + +// Hash64 hashes a byte slice and returns a uint64 hash value +func Hash64(s []byte) uint64 { + if len(s) <= 64 { + return naHash64(s) + } + return Hash64WithSeeds(s, 81, 0) +} diff --git a/vendor/github.com/dgryski/go-farm/platform.go b/vendor/github.com/dgryski/go-farm/platform.go new file mode 100644 index 0000000000..6ddf45811f --- /dev/null +++ b/vendor/github.com/dgryski/go-farm/platform.go @@ -0,0 +1,18 @@ +package farm + +func rotate32(val uint32, shift uint) uint32 { + return ((val >> shift) | (val << (32 - shift))) +} + +func rotate64(val uint64, shift uint) uint64 { + return ((val >> shift) | (val << (64 - shift))) +} + +func fetch32(s []byte, idx int) uint32 { + return uint32(s[idx+0]) | uint32(s[idx+1])<<8 | uint32(s[idx+2])<<16 | uint32(s[idx+3])<<24 +} + +func fetch64(s []byte, idx int) uint64 { + return uint64(s[idx+0]) | uint64(s[idx+1])<<8 | uint64(s[idx+2])<<16 | uint64(s[idx+3])<<24 | + uint64(s[idx+4])<<32 | uint64(s[idx+5])<<40 | uint64(s[idx+6])<<48 | uint64(s[idx+7])<<56 +} diff --git a/vendor/golang.org/x/net/internal/timeseries/timeseries.go b/vendor/golang.org/x/net/internal/timeseries/timeseries.go new file mode 100644 index 0000000000..685f0e7ea2 --- /dev/null +++ b/vendor/golang.org/x/net/internal/timeseries/timeseries.go @@ -0,0 +1,525 @@ +// Copyright 2015 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// Package timeseries implements a time series structure for stats collection. +package timeseries // import "golang.org/x/net/internal/timeseries" + +import ( + "fmt" + "log" + "time" +) + +const ( + timeSeriesNumBuckets = 64 + minuteHourSeriesNumBuckets = 60 +) + +var timeSeriesResolutions = []time.Duration{ + 1 * time.Second, + 10 * time.Second, + 1 * time.Minute, + 10 * time.Minute, + 1 * time.Hour, + 6 * time.Hour, + 24 * time.Hour, // 1 day + 7 * 24 * time.Hour, // 1 week + 4 * 7 * 24 * time.Hour, // 4 weeks + 16 * 7 * 24 * time.Hour, // 16 weeks +} + +var minuteHourSeriesResolutions = []time.Duration{ + 1 * time.Second, + 1 * time.Minute, +} + +// An Observable is a kind of data that can be aggregated in a time series. +type Observable interface { + Multiply(ratio float64) // Multiplies the data in self by a given ratio + Add(other Observable) // Adds the data from a different observation to self + Clear() // Clears the observation so it can be reused. + CopyFrom(other Observable) // Copies the contents of a given observation to self +} + +// Float attaches the methods of Observable to a float64. +type Float float64 + +// NewFloat returns a Float. +func NewFloat() Observable { + f := Float(0) + return &f +} + +// String returns the float as a string. +func (f *Float) String() string { return fmt.Sprintf("%g", f.Value()) } + +// Value returns the float's value. +func (f *Float) Value() float64 { return float64(*f) } + +func (f *Float) Multiply(ratio float64) { *f *= Float(ratio) } + +func (f *Float) Add(other Observable) { + o := other.(*Float) + *f += *o +} + +func (f *Float) Clear() { *f = 0 } + +func (f *Float) CopyFrom(other Observable) { + o := other.(*Float) + *f = *o +} + +// A Clock tells the current time. +type Clock interface { + Time() time.Time +} + +type defaultClock int + +var defaultClockInstance defaultClock + +func (defaultClock) Time() time.Time { return time.Now() } + +// Information kept per level. Each level consists of a circular list of +// observations. The start of the level may be derived from end and the +// len(buckets) * sizeInMillis. +type tsLevel struct { + oldest int // index to oldest bucketed Observable + newest int // index to newest bucketed Observable + end time.Time // end timestamp for this level + size time.Duration // duration of the bucketed Observable + buckets []Observable // collections of observations + provider func() Observable // used for creating new Observable +} + +func (l *tsLevel) Clear() { + l.oldest = 0 + l.newest = len(l.buckets) - 1 + l.end = time.Time{} + for i := range l.buckets { + if l.buckets[i] != nil { + l.buckets[i].Clear() + l.buckets[i] = nil + } + } +} + +func (l *tsLevel) InitLevel(size time.Duration, numBuckets int, f func() Observable) { + l.size = size + l.provider = f + l.buckets = make([]Observable, numBuckets) +} + +// Keeps a sequence of levels. Each level is responsible for storing data at +// a given resolution. For example, the first level stores data at a one +// minute resolution while the second level stores data at a one hour +// resolution. + +// Each level is represented by a sequence of buckets. Each bucket spans an +// interval equal to the resolution of the level. New observations are added +// to the last bucket. +type timeSeries struct { + provider func() Observable // make more Observable + numBuckets int // number of buckets in each level + levels []*tsLevel // levels of bucketed Observable + lastAdd time.Time // time of last Observable tracked + total Observable // convenient aggregation of all Observable + clock Clock // Clock for getting current time + pending Observable // observations not yet bucketed + pendingTime time.Time // what time are we keeping in pending + dirty bool // if there are pending observations +} + +// init initializes a level according to the supplied criteria. +func (ts *timeSeries) init(resolutions []time.Duration, f func() Observable, numBuckets int, clock Clock) { + ts.provider = f + ts.numBuckets = numBuckets + ts.clock = clock + ts.levels = make([]*tsLevel, len(resolutions)) + + for i := range resolutions { + if i > 0 && resolutions[i-1] >= resolutions[i] { + log.Print("timeseries: resolutions must be monotonically increasing") + break + } + newLevel := new(tsLevel) + newLevel.InitLevel(resolutions[i], ts.numBuckets, ts.provider) + ts.levels[i] = newLevel + } + + ts.Clear() +} + +// Clear removes all observations from the time series. +func (ts *timeSeries) Clear() { + ts.lastAdd = time.Time{} + ts.total = ts.resetObservation(ts.total) + ts.pending = ts.resetObservation(ts.pending) + ts.pendingTime = time.Time{} + ts.dirty = false + + for i := range ts.levels { + ts.levels[i].Clear() + } +} + +// Add records an observation at the current time. +func (ts *timeSeries) Add(observation Observable) { + ts.AddWithTime(observation, ts.clock.Time()) +} + +// AddWithTime records an observation at the specified time. +func (ts *timeSeries) AddWithTime(observation Observable, t time.Time) { + + smallBucketDuration := ts.levels[0].size + + if t.After(ts.lastAdd) { + ts.lastAdd = t + } + + if t.After(ts.pendingTime) { + ts.advance(t) + ts.mergePendingUpdates() + ts.pendingTime = ts.levels[0].end + ts.pending.CopyFrom(observation) + ts.dirty = true + } else if t.After(ts.pendingTime.Add(-1 * smallBucketDuration)) { + // The observation is close enough to go into the pending bucket. + // This compensates for clock skewing and small scheduling delays + // by letting the update stay in the fast path. + ts.pending.Add(observation) + ts.dirty = true + } else { + ts.mergeValue(observation, t) + } +} + +// mergeValue inserts the observation at the specified time in the past into all levels. +func (ts *timeSeries) mergeValue(observation Observable, t time.Time) { + for _, level := range ts.levels { + index := (ts.numBuckets - 1) - int(level.end.Sub(t)/level.size) + if 0 <= index && index < ts.numBuckets { + bucketNumber := (level.oldest + index) % ts.numBuckets + if level.buckets[bucketNumber] == nil { + level.buckets[bucketNumber] = level.provider() + } + level.buckets[bucketNumber].Add(observation) + } + } + ts.total.Add(observation) +} + +// mergePendingUpdates applies the pending updates into all levels. +func (ts *timeSeries) mergePendingUpdates() { + if ts.dirty { + ts.mergeValue(ts.pending, ts.pendingTime) + ts.pending = ts.resetObservation(ts.pending) + ts.dirty = false + } +} + +// advance cycles the buckets at each level until the latest bucket in +// each level can hold the time specified. +func (ts *timeSeries) advance(t time.Time) { + if !t.After(ts.levels[0].end) { + return + } + for i := 0; i < len(ts.levels); i++ { + level := ts.levels[i] + if !level.end.Before(t) { + break + } + + // If the time is sufficiently far, just clear the level and advance + // directly. + if !t.Before(level.end.Add(level.size * time.Duration(ts.numBuckets))) { + for _, b := range level.buckets { + ts.resetObservation(b) + } + level.end = time.Unix(0, (t.UnixNano()/level.size.Nanoseconds())*level.size.Nanoseconds()) + } + + for t.After(level.end) { + level.end = level.end.Add(level.size) + level.newest = level.oldest + level.oldest = (level.oldest + 1) % ts.numBuckets + ts.resetObservation(level.buckets[level.newest]) + } + + t = level.end + } +} + +// Latest returns the sum of the num latest buckets from the level. +func (ts *timeSeries) Latest(level, num int) Observable { + now := ts.clock.Time() + if ts.levels[0].end.Before(now) { + ts.advance(now) + } + + ts.mergePendingUpdates() + + result := ts.provider() + l := ts.levels[level] + index := l.newest + + for i := 0; i < num; i++ { + if l.buckets[index] != nil { + result.Add(l.buckets[index]) + } + if index == 0 { + index = ts.numBuckets + } + index-- + } + + return result +} + +// LatestBuckets returns a copy of the num latest buckets from level. +func (ts *timeSeries) LatestBuckets(level, num int) []Observable { + if level < 0 || level > len(ts.levels) { + log.Print("timeseries: bad level argument: ", level) + return nil + } + if num < 0 || num >= ts.numBuckets { + log.Print("timeseries: bad num argument: ", num) + return nil + } + + results := make([]Observable, num) + now := ts.clock.Time() + if ts.levels[0].end.Before(now) { + ts.advance(now) + } + + ts.mergePendingUpdates() + + l := ts.levels[level] + index := l.newest + + for i := 0; i < num; i++ { + result := ts.provider() + results[i] = result + if l.buckets[index] != nil { + result.CopyFrom(l.buckets[index]) + } + + if index == 0 { + index = ts.numBuckets + } + index -= 1 + } + return results +} + +// ScaleBy updates observations by scaling by factor. +func (ts *timeSeries) ScaleBy(factor float64) { + for _, l := range ts.levels { + for i := 0; i < ts.numBuckets; i++ { + l.buckets[i].Multiply(factor) + } + } + + ts.total.Multiply(factor) + ts.pending.Multiply(factor) +} + +// Range returns the sum of observations added over the specified time range. +// If start or finish times don't fall on bucket boundaries of the same +// level, then return values are approximate answers. +func (ts *timeSeries) Range(start, finish time.Time) Observable { + return ts.ComputeRange(start, finish, 1)[0] +} + +// Recent returns the sum of observations from the last delta. +func (ts *timeSeries) Recent(delta time.Duration) Observable { + now := ts.clock.Time() + return ts.Range(now.Add(-delta), now) +} + +// Total returns the total of all observations. +func (ts *timeSeries) Total() Observable { + ts.mergePendingUpdates() + return ts.total +} + +// ComputeRange computes a specified number of values into a slice using +// the observations recorded over the specified time period. The return +// values are approximate if the start or finish times don't fall on the +// bucket boundaries at the same level or if the number of buckets spanning +// the range is not an integral multiple of num. +func (ts *timeSeries) ComputeRange(start, finish time.Time, num int) []Observable { + if start.After(finish) { + log.Printf("timeseries: start > finish, %v>%v", start, finish) + return nil + } + + if num < 0 { + log.Printf("timeseries: num < 0, %v", num) + return nil + } + + results := make([]Observable, num) + + for _, l := range ts.levels { + if !start.Before(l.end.Add(-l.size * time.Duration(ts.numBuckets))) { + ts.extract(l, start, finish, num, results) + return results + } + } + + // Failed to find a level that covers the desired range. So just + // extract from the last level, even if it doesn't cover the entire + // desired range. + ts.extract(ts.levels[len(ts.levels)-1], start, finish, num, results) + + return results +} + +// RecentList returns the specified number of values in slice over the most +// recent time period of the specified range. +func (ts *timeSeries) RecentList(delta time.Duration, num int) []Observable { + if delta < 0 { + return nil + } + now := ts.clock.Time() + return ts.ComputeRange(now.Add(-delta), now, num) +} + +// extract returns a slice of specified number of observations from a given +// level over a given range. +func (ts *timeSeries) extract(l *tsLevel, start, finish time.Time, num int, results []Observable) { + ts.mergePendingUpdates() + + srcInterval := l.size + dstInterval := finish.Sub(start) / time.Duration(num) + dstStart := start + srcStart := l.end.Add(-srcInterval * time.Duration(ts.numBuckets)) + + srcIndex := 0 + + // Where should scanning start? + if dstStart.After(srcStart) { + advance := dstStart.Sub(srcStart) / srcInterval + srcIndex += int(advance) + srcStart = srcStart.Add(advance * srcInterval) + } + + // The i'th value is computed as show below. + // interval = (finish/start)/num + // i'th value = sum of observation in range + // [ start + i * interval, + // start + (i + 1) * interval ) + for i := 0; i < num; i++ { + results[i] = ts.resetObservation(results[i]) + dstEnd := dstStart.Add(dstInterval) + for srcIndex < ts.numBuckets && srcStart.Before(dstEnd) { + srcEnd := srcStart.Add(srcInterval) + if srcEnd.After(ts.lastAdd) { + srcEnd = ts.lastAdd + } + + if !srcEnd.Before(dstStart) { + srcValue := l.buckets[(srcIndex+l.oldest)%ts.numBuckets] + if !srcStart.Before(dstStart) && !srcEnd.After(dstEnd) { + // dst completely contains src. + if srcValue != nil { + results[i].Add(srcValue) + } + } else { + // dst partially overlaps src. + overlapStart := maxTime(srcStart, dstStart) + overlapEnd := minTime(srcEnd, dstEnd) + base := srcEnd.Sub(srcStart) + fraction := overlapEnd.Sub(overlapStart).Seconds() / base.Seconds() + + used := ts.provider() + if srcValue != nil { + used.CopyFrom(srcValue) + } + used.Multiply(fraction) + results[i].Add(used) + } + + if srcEnd.After(dstEnd) { + break + } + } + srcIndex++ + srcStart = srcStart.Add(srcInterval) + } + dstStart = dstStart.Add(dstInterval) + } +} + +// resetObservation clears the content so the struct may be reused. +func (ts *timeSeries) resetObservation(observation Observable) Observable { + if observation == nil { + observation = ts.provider() + } else { + observation.Clear() + } + return observation +} + +// TimeSeries tracks data at granularities from 1 second to 16 weeks. +type TimeSeries struct { + timeSeries +} + +// NewTimeSeries creates a new TimeSeries using the function provided for creating new Observable. +func NewTimeSeries(f func() Observable) *TimeSeries { + return NewTimeSeriesWithClock(f, defaultClockInstance) +} + +// NewTimeSeriesWithClock creates a new TimeSeries using the function provided for creating new Observable and the clock for +// assigning timestamps. +func NewTimeSeriesWithClock(f func() Observable, clock Clock) *TimeSeries { + ts := new(TimeSeries) + ts.timeSeries.init(timeSeriesResolutions, f, timeSeriesNumBuckets, clock) + return ts +} + +// MinuteHourSeries tracks data at granularities of 1 minute and 1 hour. +type MinuteHourSeries struct { + timeSeries +} + +// NewMinuteHourSeries creates a new MinuteHourSeries using the function provided for creating new Observable. +func NewMinuteHourSeries(f func() Observable) *MinuteHourSeries { + return NewMinuteHourSeriesWithClock(f, defaultClockInstance) +} + +// NewMinuteHourSeriesWithClock creates a new MinuteHourSeries using the function provided for creating new Observable and the clock for +// assigning timestamps. +func NewMinuteHourSeriesWithClock(f func() Observable, clock Clock) *MinuteHourSeries { + ts := new(MinuteHourSeries) + ts.timeSeries.init(minuteHourSeriesResolutions, f, + minuteHourSeriesNumBuckets, clock) + return ts +} + +func (ts *MinuteHourSeries) Minute() Observable { + return ts.timeSeries.Latest(0, 60) +} + +func (ts *MinuteHourSeries) Hour() Observable { + return ts.timeSeries.Latest(1, 60) +} + +func minTime(a, b time.Time) time.Time { + if a.Before(b) { + return a + } + return b +} + +func maxTime(a, b time.Time) time.Time { + if a.After(b) { + return a + } + return b +} diff --git a/vendor/golang.org/x/net/trace/events.go b/vendor/golang.org/x/net/trace/events.go new file mode 100644 index 0000000000..c646a6952e --- /dev/null +++ b/vendor/golang.org/x/net/trace/events.go @@ -0,0 +1,532 @@ +// Copyright 2015 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package trace + +import ( + "bytes" + "fmt" + "html/template" + "io" + "log" + "net/http" + "runtime" + "sort" + "strconv" + "strings" + "sync" + "sync/atomic" + "text/tabwriter" + "time" +) + +const maxEventsPerLog = 100 + +type bucket struct { + MaxErrAge time.Duration + String string +} + +var buckets = []bucket{ + {0, "total"}, + {10 * time.Second, "errs<10s"}, + {1 * time.Minute, "errs<1m"}, + {10 * time.Minute, "errs<10m"}, + {1 * time.Hour, "errs<1h"}, + {10 * time.Hour, "errs<10h"}, + {24000 * time.Hour, "errors"}, +} + +// RenderEvents renders the HTML page typically served at /debug/events. +// It does not do any auth checking. The request may be nil. +// +// Most users will use the Events handler. +func RenderEvents(w http.ResponseWriter, req *http.Request, sensitive bool) { + now := time.Now() + data := &struct { + Families []string // family names + Buckets []bucket + Counts [][]int // eventLog count per family/bucket + + // Set when a bucket has been selected. + Family string + Bucket int + EventLogs eventLogs + Expanded bool + }{ + Buckets: buckets, + } + + data.Families = make([]string, 0, len(families)) + famMu.RLock() + for name := range families { + data.Families = append(data.Families, name) + } + famMu.RUnlock() + sort.Strings(data.Families) + + // Count the number of eventLogs in each family for each error age. + data.Counts = make([][]int, len(data.Families)) + for i, name := range data.Families { + // TODO(sameer): move this loop under the family lock. + f := getEventFamily(name) + data.Counts[i] = make([]int, len(data.Buckets)) + for j, b := range data.Buckets { + data.Counts[i][j] = f.Count(now, b.MaxErrAge) + } + } + + if req != nil { + var ok bool + data.Family, data.Bucket, ok = parseEventsArgs(req) + if !ok { + // No-op + } else { + data.EventLogs = getEventFamily(data.Family).Copy(now, buckets[data.Bucket].MaxErrAge) + } + if data.EventLogs != nil { + defer data.EventLogs.Free() + sort.Sort(data.EventLogs) + } + if exp, err := strconv.ParseBool(req.FormValue("exp")); err == nil { + data.Expanded = exp + } + } + + famMu.RLock() + defer famMu.RUnlock() + if err := eventsTmpl().Execute(w, data); err != nil { + log.Printf("net/trace: Failed executing template: %v", err) + } +} + +func parseEventsArgs(req *http.Request) (fam string, b int, ok bool) { + fam, bStr := req.FormValue("fam"), req.FormValue("b") + if fam == "" || bStr == "" { + return "", 0, false + } + b, err := strconv.Atoi(bStr) + if err != nil || b < 0 || b >= len(buckets) { + return "", 0, false + } + return fam, b, true +} + +// An EventLog provides a log of events associated with a specific object. +type EventLog interface { + // Printf formats its arguments with fmt.Sprintf and adds the + // result to the event log. + Printf(format string, a ...interface{}) + + // Errorf is like Printf, but it marks this event as an error. + Errorf(format string, a ...interface{}) + + // Finish declares that this event log is complete. + // The event log should not be used after calling this method. + Finish() +} + +// NewEventLog returns a new EventLog with the specified family name +// and title. +func NewEventLog(family, title string) EventLog { + el := newEventLog() + el.ref() + el.Family, el.Title = family, title + el.Start = time.Now() + el.events = make([]logEntry, 0, maxEventsPerLog) + el.stack = make([]uintptr, 32) + n := runtime.Callers(2, el.stack) + el.stack = el.stack[:n] + + getEventFamily(family).add(el) + return el +} + +func (el *eventLog) Finish() { + getEventFamily(el.Family).remove(el) + el.unref() // matches ref in New +} + +var ( + famMu sync.RWMutex + families = make(map[string]*eventFamily) // family name => family +) + +func getEventFamily(fam string) *eventFamily { + famMu.Lock() + defer famMu.Unlock() + f := families[fam] + if f == nil { + f = &eventFamily{} + families[fam] = f + } + return f +} + +type eventFamily struct { + mu sync.RWMutex + eventLogs eventLogs +} + +func (f *eventFamily) add(el *eventLog) { + f.mu.Lock() + f.eventLogs = append(f.eventLogs, el) + f.mu.Unlock() +} + +func (f *eventFamily) remove(el *eventLog) { + f.mu.Lock() + defer f.mu.Unlock() + for i, el0 := range f.eventLogs { + if el == el0 { + copy(f.eventLogs[i:], f.eventLogs[i+1:]) + f.eventLogs = f.eventLogs[:len(f.eventLogs)-1] + return + } + } +} + +func (f *eventFamily) Count(now time.Time, maxErrAge time.Duration) (n int) { + f.mu.RLock() + defer f.mu.RUnlock() + for _, el := range f.eventLogs { + if el.hasRecentError(now, maxErrAge) { + n++ + } + } + return +} + +func (f *eventFamily) Copy(now time.Time, maxErrAge time.Duration) (els eventLogs) { + f.mu.RLock() + defer f.mu.RUnlock() + els = make(eventLogs, 0, len(f.eventLogs)) + for _, el := range f.eventLogs { + if el.hasRecentError(now, maxErrAge) { + el.ref() + els = append(els, el) + } + } + return +} + +type eventLogs []*eventLog + +// Free calls unref on each element of the list. +func (els eventLogs) Free() { + for _, el := range els { + el.unref() + } +} + +// eventLogs may be sorted in reverse chronological order. +func (els eventLogs) Len() int { return len(els) } +func (els eventLogs) Less(i, j int) bool { return els[i].Start.After(els[j].Start) } +func (els eventLogs) Swap(i, j int) { els[i], els[j] = els[j], els[i] } + +// A logEntry is a timestamped log entry in an event log. +type logEntry struct { + When time.Time + Elapsed time.Duration // since previous event in log + NewDay bool // whether this event is on a different day to the previous event + What string + IsErr bool +} + +// WhenString returns a string representation of the elapsed time of the event. +// It will include the date if midnight was crossed. +func (e logEntry) WhenString() string { + if e.NewDay { + return e.When.Format("2006/01/02 15:04:05.000000") + } + return e.When.Format("15:04:05.000000") +} + +// An eventLog represents an active event log. +type eventLog struct { + // Family is the top-level grouping of event logs to which this belongs. + Family string + + // Title is the title of this event log. + Title string + + // Timing information. + Start time.Time + + // Call stack where this event log was created. + stack []uintptr + + // Append-only sequence of events. + // + // TODO(sameer): change this to a ring buffer to avoid the array copy + // when we hit maxEventsPerLog. + mu sync.RWMutex + events []logEntry + LastErrorTime time.Time + discarded int + + refs int32 // how many buckets this is in +} + +func (el *eventLog) reset() { + // Clear all but the mutex. Mutexes may not be copied, even when unlocked. + el.Family = "" + el.Title = "" + el.Start = time.Time{} + el.stack = nil + el.events = nil + el.LastErrorTime = time.Time{} + el.discarded = 0 + el.refs = 0 +} + +func (el *eventLog) hasRecentError(now time.Time, maxErrAge time.Duration) bool { + if maxErrAge == 0 { + return true + } + el.mu.RLock() + defer el.mu.RUnlock() + return now.Sub(el.LastErrorTime) < maxErrAge +} + +// delta returns the elapsed time since the last event or the log start, +// and whether it spans midnight. +// L >= el.mu +func (el *eventLog) delta(t time.Time) (time.Duration, bool) { + if len(el.events) == 0 { + return t.Sub(el.Start), false + } + prev := el.events[len(el.events)-1].When + return t.Sub(prev), prev.Day() != t.Day() + +} + +func (el *eventLog) Printf(format string, a ...interface{}) { + el.printf(false, format, a...) +} + +func (el *eventLog) Errorf(format string, a ...interface{}) { + el.printf(true, format, a...) +} + +func (el *eventLog) printf(isErr bool, format string, a ...interface{}) { + e := logEntry{When: time.Now(), IsErr: isErr, What: fmt.Sprintf(format, a...)} + el.mu.Lock() + e.Elapsed, e.NewDay = el.delta(e.When) + if len(el.events) < maxEventsPerLog { + el.events = append(el.events, e) + } else { + // Discard the oldest event. + if el.discarded == 0 { + // el.discarded starts at two to count for the event it + // is replacing, plus the next one that we are about to + // drop. + el.discarded = 2 + } else { + el.discarded++ + } + // TODO(sameer): if this causes allocations on a critical path, + // change eventLog.What to be a fmt.Stringer, as in trace.go. + el.events[0].What = fmt.Sprintf("(%d events discarded)", el.discarded) + // The timestamp of the discarded meta-event should be + // the time of the last event it is representing. + el.events[0].When = el.events[1].When + copy(el.events[1:], el.events[2:]) + el.events[maxEventsPerLog-1] = e + } + if e.IsErr { + el.LastErrorTime = e.When + } + el.mu.Unlock() +} + +func (el *eventLog) ref() { + atomic.AddInt32(&el.refs, 1) +} + +func (el *eventLog) unref() { + if atomic.AddInt32(&el.refs, -1) == 0 { + freeEventLog(el) + } +} + +func (el *eventLog) When() string { + return el.Start.Format("2006/01/02 15:04:05.000000") +} + +func (el *eventLog) ElapsedTime() string { + elapsed := time.Since(el.Start) + return fmt.Sprintf("%.6f", elapsed.Seconds()) +} + +func (el *eventLog) Stack() string { + buf := new(bytes.Buffer) + tw := tabwriter.NewWriter(buf, 1, 8, 1, '\t', 0) + printStackRecord(tw, el.stack) + tw.Flush() + return buf.String() +} + +// printStackRecord prints the function + source line information +// for a single stack trace. +// Adapted from runtime/pprof/pprof.go. +func printStackRecord(w io.Writer, stk []uintptr) { + for _, pc := range stk { + f := runtime.FuncForPC(pc) + if f == nil { + continue + } + file, line := f.FileLine(pc) + name := f.Name() + // Hide runtime.goexit and any runtime functions at the beginning. + if strings.HasPrefix(name, "runtime.") { + continue + } + fmt.Fprintf(w, "# %s\t%s:%d\n", name, file, line) + } +} + +func (el *eventLog) Events() []logEntry { + el.mu.RLock() + defer el.mu.RUnlock() + return el.events +} + +// freeEventLogs is a freelist of *eventLog +var freeEventLogs = make(chan *eventLog, 1000) + +// newEventLog returns a event log ready to use. +func newEventLog() *eventLog { + select { + case el := <-freeEventLogs: + return el + default: + return new(eventLog) + } +} + +// freeEventLog adds el to freeEventLogs if there's room. +// This is non-blocking. +func freeEventLog(el *eventLog) { + el.reset() + select { + case freeEventLogs <- el: + default: + } +} + +var eventsTmplCache *template.Template +var eventsTmplOnce sync.Once + +func eventsTmpl() *template.Template { + eventsTmplOnce.Do(func() { + eventsTmplCache = template.Must(template.New("events").Funcs(template.FuncMap{ + "elapsed": elapsed, + "trimSpace": strings.TrimSpace, + }).Parse(eventsHTML)) + }) + return eventsTmplCache +} + +const eventsHTML = ` + + + events + + + + +

/debug/events

+ + + {{range $i, $fam := .Families}} + + + + {{range $j, $bucket := $.Buckets}} + {{$n := index $.Counts $i $j}} + + {{end}} + + {{end}} +
{{$fam}} + {{if $n}}{{end}} + [{{$n}} {{$bucket.String}}] + {{if $n}}{{end}} +
+ +{{if $.EventLogs}} +
+

Family: {{$.Family}}

+ +{{if $.Expanded}}{{end}} +[Summary]{{if $.Expanded}}{{end}} + +{{if not $.Expanded}}{{end}} +[Expanded]{{if not $.Expanded}}{{end}} + + + + {{range $el := $.EventLogs}} + + + + + {{if $.Expanded}} + + + + + + {{range $el.Events}} + + + + + + {{end}} + {{end}} + {{end}} +
WhenElapsed
{{$el.When}}{{$el.ElapsedTime}}{{$el.Title}} + 
{{$el.Stack|trimSpace}}
{{.WhenString}}{{elapsed .Elapsed}}.{{if .IsErr}}E{{else}}.{{end}}. {{.What}}
+{{end}} + + +` diff --git a/vendor/golang.org/x/net/trace/histogram.go b/vendor/golang.org/x/net/trace/histogram.go new file mode 100644 index 0000000000..9bf4286c79 --- /dev/null +++ b/vendor/golang.org/x/net/trace/histogram.go @@ -0,0 +1,365 @@ +// Copyright 2015 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package trace + +// This file implements histogramming for RPC statistics collection. + +import ( + "bytes" + "fmt" + "html/template" + "log" + "math" + "sync" + + "golang.org/x/net/internal/timeseries" +) + +const ( + bucketCount = 38 +) + +// histogram keeps counts of values in buckets that are spaced +// out in powers of 2: 0-1, 2-3, 4-7... +// histogram implements timeseries.Observable +type histogram struct { + sum int64 // running total of measurements + sumOfSquares float64 // square of running total + buckets []int64 // bucketed values for histogram + value int // holds a single value as an optimization + valueCount int64 // number of values recorded for single value +} + +// AddMeasurement records a value measurement observation to the histogram. +func (h *histogram) addMeasurement(value int64) { + // TODO: assert invariant + h.sum += value + h.sumOfSquares += float64(value) * float64(value) + + bucketIndex := getBucket(value) + + if h.valueCount == 0 || (h.valueCount > 0 && h.value == bucketIndex) { + h.value = bucketIndex + h.valueCount++ + } else { + h.allocateBuckets() + h.buckets[bucketIndex]++ + } +} + +func (h *histogram) allocateBuckets() { + if h.buckets == nil { + h.buckets = make([]int64, bucketCount) + h.buckets[h.value] = h.valueCount + h.value = 0 + h.valueCount = -1 + } +} + +func log2(i int64) int { + n := 0 + for ; i >= 0x100; i >>= 8 { + n += 8 + } + for ; i > 0; i >>= 1 { + n += 1 + } + return n +} + +func getBucket(i int64) (index int) { + index = log2(i) - 1 + if index < 0 { + index = 0 + } + if index >= bucketCount { + index = bucketCount - 1 + } + return +} + +// Total returns the number of recorded observations. +func (h *histogram) total() (total int64) { + if h.valueCount >= 0 { + total = h.valueCount + } + for _, val := range h.buckets { + total += int64(val) + } + return +} + +// Average returns the average value of recorded observations. +func (h *histogram) average() float64 { + t := h.total() + if t == 0 { + return 0 + } + return float64(h.sum) / float64(t) +} + +// Variance returns the variance of recorded observations. +func (h *histogram) variance() float64 { + t := float64(h.total()) + if t == 0 { + return 0 + } + s := float64(h.sum) / t + return h.sumOfSquares/t - s*s +} + +// StandardDeviation returns the standard deviation of recorded observations. +func (h *histogram) standardDeviation() float64 { + return math.Sqrt(h.variance()) +} + +// PercentileBoundary estimates the value that the given fraction of recorded +// observations are less than. +func (h *histogram) percentileBoundary(percentile float64) int64 { + total := h.total() + + // Corner cases (make sure result is strictly less than Total()) + if total == 0 { + return 0 + } else if total == 1 { + return int64(h.average()) + } + + percentOfTotal := round(float64(total) * percentile) + var runningTotal int64 + + for i := range h.buckets { + value := h.buckets[i] + runningTotal += value + if runningTotal == percentOfTotal { + // We hit an exact bucket boundary. If the next bucket has data, it is a + // good estimate of the value. If the bucket is empty, we interpolate the + // midpoint between the next bucket's boundary and the next non-zero + // bucket. If the remaining buckets are all empty, then we use the + // boundary for the next bucket as the estimate. + j := uint8(i + 1) + min := bucketBoundary(j) + if runningTotal < total { + for h.buckets[j] == 0 { + j++ + } + } + max := bucketBoundary(j) + return min + round(float64(max-min)/2) + } else if runningTotal > percentOfTotal { + // The value is in this bucket. Interpolate the value. + delta := runningTotal - percentOfTotal + percentBucket := float64(value-delta) / float64(value) + bucketMin := bucketBoundary(uint8(i)) + nextBucketMin := bucketBoundary(uint8(i + 1)) + bucketSize := nextBucketMin - bucketMin + return bucketMin + round(percentBucket*float64(bucketSize)) + } + } + return bucketBoundary(bucketCount - 1) +} + +// Median returns the estimated median of the observed values. +func (h *histogram) median() int64 { + return h.percentileBoundary(0.5) +} + +// Add adds other to h. +func (h *histogram) Add(other timeseries.Observable) { + o := other.(*histogram) + if o.valueCount == 0 { + // Other histogram is empty + } else if h.valueCount >= 0 && o.valueCount > 0 && h.value == o.value { + // Both have a single bucketed value, aggregate them + h.valueCount += o.valueCount + } else { + // Two different values necessitate buckets in this histogram + h.allocateBuckets() + if o.valueCount >= 0 { + h.buckets[o.value] += o.valueCount + } else { + for i := range h.buckets { + h.buckets[i] += o.buckets[i] + } + } + } + h.sumOfSquares += o.sumOfSquares + h.sum += o.sum +} + +// Clear resets the histogram to an empty state, removing all observed values. +func (h *histogram) Clear() { + h.buckets = nil + h.value = 0 + h.valueCount = 0 + h.sum = 0 + h.sumOfSquares = 0 +} + +// CopyFrom copies from other, which must be a *histogram, into h. +func (h *histogram) CopyFrom(other timeseries.Observable) { + o := other.(*histogram) + if o.valueCount == -1 { + h.allocateBuckets() + copy(h.buckets, o.buckets) + } + h.sum = o.sum + h.sumOfSquares = o.sumOfSquares + h.value = o.value + h.valueCount = o.valueCount +} + +// Multiply scales the histogram by the specified ratio. +func (h *histogram) Multiply(ratio float64) { + if h.valueCount == -1 { + for i := range h.buckets { + h.buckets[i] = int64(float64(h.buckets[i]) * ratio) + } + } else { + h.valueCount = int64(float64(h.valueCount) * ratio) + } + h.sum = int64(float64(h.sum) * ratio) + h.sumOfSquares = h.sumOfSquares * ratio +} + +// New creates a new histogram. +func (h *histogram) New() timeseries.Observable { + r := new(histogram) + r.Clear() + return r +} + +func (h *histogram) String() string { + return fmt.Sprintf("%d, %f, %d, %d, %v", + h.sum, h.sumOfSquares, h.value, h.valueCount, h.buckets) +} + +// round returns the closest int64 to the argument +func round(in float64) int64 { + return int64(math.Floor(in + 0.5)) +} + +// bucketBoundary returns the first value in the bucket. +func bucketBoundary(bucket uint8) int64 { + if bucket == 0 { + return 0 + } + return 1 << bucket +} + +// bucketData holds data about a specific bucket for use in distTmpl. +type bucketData struct { + Lower, Upper int64 + N int64 + Pct, CumulativePct float64 + GraphWidth int +} + +// data holds data about a Distribution for use in distTmpl. +type data struct { + Buckets []*bucketData + Count, Median int64 + Mean, StandardDeviation float64 +} + +// maxHTMLBarWidth is the maximum width of the HTML bar for visualizing buckets. +const maxHTMLBarWidth = 350.0 + +// newData returns data representing h for use in distTmpl. +func (h *histogram) newData() *data { + // Force the allocation of buckets to simplify the rendering implementation + h.allocateBuckets() + // We scale the bars on the right so that the largest bar is + // maxHTMLBarWidth pixels in width. + maxBucket := int64(0) + for _, n := range h.buckets { + if n > maxBucket { + maxBucket = n + } + } + total := h.total() + barsizeMult := maxHTMLBarWidth / float64(maxBucket) + var pctMult float64 + if total == 0 { + pctMult = 1.0 + } else { + pctMult = 100.0 / float64(total) + } + + buckets := make([]*bucketData, len(h.buckets)) + runningTotal := int64(0) + for i, n := range h.buckets { + if n == 0 { + continue + } + runningTotal += n + var upperBound int64 + if i < bucketCount-1 { + upperBound = bucketBoundary(uint8(i + 1)) + } else { + upperBound = math.MaxInt64 + } + buckets[i] = &bucketData{ + Lower: bucketBoundary(uint8(i)), + Upper: upperBound, + N: n, + Pct: float64(n) * pctMult, + CumulativePct: float64(runningTotal) * pctMult, + GraphWidth: int(float64(n) * barsizeMult), + } + } + return &data{ + Buckets: buckets, + Count: total, + Median: h.median(), + Mean: h.average(), + StandardDeviation: h.standardDeviation(), + } +} + +func (h *histogram) html() template.HTML { + buf := new(bytes.Buffer) + if err := distTmpl().Execute(buf, h.newData()); err != nil { + buf.Reset() + log.Printf("net/trace: couldn't execute template: %v", err) + } + return template.HTML(buf.String()) +} + +var distTmplCache *template.Template +var distTmplOnce sync.Once + +func distTmpl() *template.Template { + distTmplOnce.Do(func() { + // Input: data + distTmplCache = template.Must(template.New("distTmpl").Parse(` + + + + + + + +
Count: {{.Count}}Mean: {{printf "%.0f" .Mean}}StdDev: {{printf "%.0f" .StandardDeviation}}Median: {{.Median}}
+
+ +{{range $b := .Buckets}} +{{if $b}} + + + + + + + + + +{{end}} +{{end}} +
[{{.Lower}},{{.Upper}}){{.N}}{{printf "%#.3f" .Pct}}%{{printf "%#.3f" .CumulativePct}}%
+`)) + }) + return distTmplCache +} diff --git a/vendor/golang.org/x/net/trace/trace.go b/vendor/golang.org/x/net/trace/trace.go new file mode 100644 index 0000000000..bb72a527e8 --- /dev/null +++ b/vendor/golang.org/x/net/trace/trace.go @@ -0,0 +1,1082 @@ +// Copyright 2015 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +/* +Package trace implements tracing of requests and long-lived objects. +It exports HTTP interfaces on /debug/requests and /debug/events. + +A trace.Trace provides tracing for short-lived objects, usually requests. +A request handler might be implemented like this: + + func fooHandler(w http.ResponseWriter, req *http.Request) { + tr := trace.New("mypkg.Foo", req.URL.Path) + defer tr.Finish() + ... + tr.LazyPrintf("some event %q happened", str) + ... + if err := somethingImportant(); err != nil { + tr.LazyPrintf("somethingImportant failed: %v", err) + tr.SetError() + } + } + +The /debug/requests HTTP endpoint organizes the traces by family, +errors, and duration. It also provides histogram of request duration +for each family. + +A trace.EventLog provides tracing for long-lived objects, such as RPC +connections. + + // A Fetcher fetches URL paths for a single domain. + type Fetcher struct { + domain string + events trace.EventLog + } + + func NewFetcher(domain string) *Fetcher { + return &Fetcher{ + domain, + trace.NewEventLog("mypkg.Fetcher", domain), + } + } + + func (f *Fetcher) Fetch(path string) (string, error) { + resp, err := http.Get("http://" + f.domain + "/" + path) + if err != nil { + f.events.Errorf("Get(%q) = %v", path, err) + return "", err + } + f.events.Printf("Get(%q) = %s", path, resp.Status) + ... + } + + func (f *Fetcher) Close() error { + f.events.Finish() + return nil + } + +The /debug/events HTTP endpoint organizes the event logs by family and +by time since the last error. The expanded view displays recent log +entries and the log's call stack. +*/ +package trace // import "golang.org/x/net/trace" + +import ( + "bytes" + "fmt" + "html/template" + "io" + "log" + "net" + "net/http" + "runtime" + "sort" + "strconv" + "sync" + "sync/atomic" + "time" + + "golang.org/x/net/internal/timeseries" +) + +// DebugUseAfterFinish controls whether to debug uses of Trace values after finishing. +// FOR DEBUGGING ONLY. This will slow down the program. +var DebugUseAfterFinish = false + +// AuthRequest determines whether a specific request is permitted to load the +// /debug/requests or /debug/events pages. +// +// It returns two bools; the first indicates whether the page may be viewed at all, +// and the second indicates whether sensitive events will be shown. +// +// AuthRequest may be replaced by a program to customize its authorization requirements. +// +// The default AuthRequest function returns (true, true) if and only if the request +// comes from localhost/127.0.0.1/[::1]. +var AuthRequest = func(req *http.Request) (any, sensitive bool) { + // RemoteAddr is commonly in the form "IP" or "IP:port". + // If it is in the form "IP:port", split off the port. + host, _, err := net.SplitHostPort(req.RemoteAddr) + if err != nil { + host = req.RemoteAddr + } + switch host { + case "localhost", "127.0.0.1", "::1": + return true, true + default: + return false, false + } +} + +func init() { + // TODO(jbd): Serve Traces from /debug/traces in the future? + // There is no requirement for a request to be present to have traces. + http.HandleFunc("/debug/requests", Traces) + http.HandleFunc("/debug/events", Events) +} + +// Traces responds with traces from the program. +// The package initialization registers it in http.DefaultServeMux +// at /debug/requests. +// +// It performs authorization by running AuthRequest. +func Traces(w http.ResponseWriter, req *http.Request) { + any, sensitive := AuthRequest(req) + if !any { + http.Error(w, "not allowed", http.StatusUnauthorized) + return + } + w.Header().Set("Content-Type", "text/html; charset=utf-8") + Render(w, req, sensitive) +} + +// Events responds with a page of events collected by EventLogs. +// The package initialization registers it in http.DefaultServeMux +// at /debug/events. +// +// It performs authorization by running AuthRequest. +func Events(w http.ResponseWriter, req *http.Request) { + any, sensitive := AuthRequest(req) + if !any { + http.Error(w, "not allowed", http.StatusUnauthorized) + return + } + w.Header().Set("Content-Type", "text/html; charset=utf-8") + RenderEvents(w, req, sensitive) +} + +// Render renders the HTML page typically served at /debug/requests. +// It does not do any auth checking. The request may be nil. +// +// Most users will use the Traces handler. +func Render(w io.Writer, req *http.Request, sensitive bool) { + data := &struct { + Families []string + ActiveTraceCount map[string]int + CompletedTraces map[string]*family + + // Set when a bucket has been selected. + Traces traceList + Family string + Bucket int + Expanded bool + Traced bool + Active bool + ShowSensitive bool // whether to show sensitive events + + Histogram template.HTML + HistogramWindow string // e.g. "last minute", "last hour", "all time" + + // If non-zero, the set of traces is a partial set, + // and this is the total number. + Total int + }{ + CompletedTraces: completedTraces, + } + + data.ShowSensitive = sensitive + if req != nil { + // Allow show_sensitive=0 to force hiding of sensitive data for testing. + // This only goes one way; you can't use show_sensitive=1 to see things. + if req.FormValue("show_sensitive") == "0" { + data.ShowSensitive = false + } + + if exp, err := strconv.ParseBool(req.FormValue("exp")); err == nil { + data.Expanded = exp + } + if exp, err := strconv.ParseBool(req.FormValue("rtraced")); err == nil { + data.Traced = exp + } + } + + completedMu.RLock() + data.Families = make([]string, 0, len(completedTraces)) + for fam := range completedTraces { + data.Families = append(data.Families, fam) + } + completedMu.RUnlock() + sort.Strings(data.Families) + + // We are careful here to minimize the time spent locking activeMu, + // since that lock is required every time an RPC starts and finishes. + data.ActiveTraceCount = make(map[string]int, len(data.Families)) + activeMu.RLock() + for fam, s := range activeTraces { + data.ActiveTraceCount[fam] = s.Len() + } + activeMu.RUnlock() + + var ok bool + data.Family, data.Bucket, ok = parseArgs(req) + switch { + case !ok: + // No-op + case data.Bucket == -1: + data.Active = true + n := data.ActiveTraceCount[data.Family] + data.Traces = getActiveTraces(data.Family) + if len(data.Traces) < n { + data.Total = n + } + case data.Bucket < bucketsPerFamily: + if b := lookupBucket(data.Family, data.Bucket); b != nil { + data.Traces = b.Copy(data.Traced) + } + default: + if f := getFamily(data.Family, false); f != nil { + var obs timeseries.Observable + f.LatencyMu.RLock() + switch o := data.Bucket - bucketsPerFamily; o { + case 0: + obs = f.Latency.Minute() + data.HistogramWindow = "last minute" + case 1: + obs = f.Latency.Hour() + data.HistogramWindow = "last hour" + case 2: + obs = f.Latency.Total() + data.HistogramWindow = "all time" + } + f.LatencyMu.RUnlock() + if obs != nil { + data.Histogram = obs.(*histogram).html() + } + } + } + + if data.Traces != nil { + defer data.Traces.Free() + sort.Sort(data.Traces) + } + + completedMu.RLock() + defer completedMu.RUnlock() + if err := pageTmpl().ExecuteTemplate(w, "Page", data); err != nil { + log.Printf("net/trace: Failed executing template: %v", err) + } +} + +func parseArgs(req *http.Request) (fam string, b int, ok bool) { + if req == nil { + return "", 0, false + } + fam, bStr := req.FormValue("fam"), req.FormValue("b") + if fam == "" || bStr == "" { + return "", 0, false + } + b, err := strconv.Atoi(bStr) + if err != nil || b < -1 { + return "", 0, false + } + + return fam, b, true +} + +func lookupBucket(fam string, b int) *traceBucket { + f := getFamily(fam, false) + if f == nil || b < 0 || b >= len(f.Buckets) { + return nil + } + return f.Buckets[b] +} + +type contextKeyT string + +var contextKey = contextKeyT("golang.org/x/net/trace.Trace") + +// Trace represents an active request. +type Trace interface { + // LazyLog adds x to the event log. It will be evaluated each time the + // /debug/requests page is rendered. Any memory referenced by x will be + // pinned until the trace is finished and later discarded. + LazyLog(x fmt.Stringer, sensitive bool) + + // LazyPrintf evaluates its arguments with fmt.Sprintf each time the + // /debug/requests page is rendered. Any memory referenced by a will be + // pinned until the trace is finished and later discarded. + LazyPrintf(format string, a ...interface{}) + + // SetError declares that this trace resulted in an error. + SetError() + + // SetRecycler sets a recycler for the trace. + // f will be called for each event passed to LazyLog at a time when + // it is no longer required, whether while the trace is still active + // and the event is discarded, or when a completed trace is discarded. + SetRecycler(f func(interface{})) + + // SetTraceInfo sets the trace info for the trace. + // This is currently unused. + SetTraceInfo(traceID, spanID uint64) + + // SetMaxEvents sets the maximum number of events that will be stored + // in the trace. This has no effect if any events have already been + // added to the trace. + SetMaxEvents(m int) + + // Finish declares that this trace is complete. + // The trace should not be used after calling this method. + Finish() +} + +type lazySprintf struct { + format string + a []interface{} +} + +func (l *lazySprintf) String() string { + return fmt.Sprintf(l.format, l.a...) +} + +// New returns a new Trace with the specified family and title. +func New(family, title string) Trace { + tr := newTrace() + tr.ref() + tr.Family, tr.Title = family, title + tr.Start = time.Now() + tr.maxEvents = maxEventsPerTrace + tr.events = tr.eventsBuf[:0] + + activeMu.RLock() + s := activeTraces[tr.Family] + activeMu.RUnlock() + if s == nil { + activeMu.Lock() + s = activeTraces[tr.Family] // check again + if s == nil { + s = new(traceSet) + activeTraces[tr.Family] = s + } + activeMu.Unlock() + } + s.Add(tr) + + // Trigger allocation of the completed trace structure for this family. + // This will cause the family to be present in the request page during + // the first trace of this family. We don't care about the return value, + // nor is there any need for this to run inline, so we execute it in its + // own goroutine, but only if the family isn't allocated yet. + completedMu.RLock() + if _, ok := completedTraces[tr.Family]; !ok { + go allocFamily(tr.Family) + } + completedMu.RUnlock() + + return tr +} + +func (tr *trace) Finish() { + tr.Elapsed = time.Now().Sub(tr.Start) + if DebugUseAfterFinish { + buf := make([]byte, 4<<10) // 4 KB should be enough + n := runtime.Stack(buf, false) + tr.finishStack = buf[:n] + } + + activeMu.RLock() + m := activeTraces[tr.Family] + activeMu.RUnlock() + m.Remove(tr) + + f := getFamily(tr.Family, true) + for _, b := range f.Buckets { + if b.Cond.match(tr) { + b.Add(tr) + } + } + // Add a sample of elapsed time as microseconds to the family's timeseries + h := new(histogram) + h.addMeasurement(tr.Elapsed.Nanoseconds() / 1e3) + f.LatencyMu.Lock() + f.Latency.Add(h) + f.LatencyMu.Unlock() + + tr.unref() // matches ref in New +} + +const ( + bucketsPerFamily = 9 + tracesPerBucket = 10 + maxActiveTraces = 20 // Maximum number of active traces to show. + maxEventsPerTrace = 10 + numHistogramBuckets = 38 +) + +var ( + // The active traces. + activeMu sync.RWMutex + activeTraces = make(map[string]*traceSet) // family -> traces + + // Families of completed traces. + completedMu sync.RWMutex + completedTraces = make(map[string]*family) // family -> traces +) + +type traceSet struct { + mu sync.RWMutex + m map[*trace]bool + + // We could avoid the entire map scan in FirstN by having a slice of all the traces + // ordered by start time, and an index into that from the trace struct, with a periodic + // repack of the slice after enough traces finish; we could also use a skip list or similar. + // However, that would shift some of the expense from /debug/requests time to RPC time, + // which is probably the wrong trade-off. +} + +func (ts *traceSet) Len() int { + ts.mu.RLock() + defer ts.mu.RUnlock() + return len(ts.m) +} + +func (ts *traceSet) Add(tr *trace) { + ts.mu.Lock() + if ts.m == nil { + ts.m = make(map[*trace]bool) + } + ts.m[tr] = true + ts.mu.Unlock() +} + +func (ts *traceSet) Remove(tr *trace) { + ts.mu.Lock() + delete(ts.m, tr) + ts.mu.Unlock() +} + +// FirstN returns the first n traces ordered by time. +func (ts *traceSet) FirstN(n int) traceList { + ts.mu.RLock() + defer ts.mu.RUnlock() + + if n > len(ts.m) { + n = len(ts.m) + } + trl := make(traceList, 0, n) + + // Fast path for when no selectivity is needed. + if n == len(ts.m) { + for tr := range ts.m { + tr.ref() + trl = append(trl, tr) + } + sort.Sort(trl) + return trl + } + + // Pick the oldest n traces. + // This is inefficient. See the comment in the traceSet struct. + for tr := range ts.m { + // Put the first n traces into trl in the order they occur. + // When we have n, sort trl, and thereafter maintain its order. + if len(trl) < n { + tr.ref() + trl = append(trl, tr) + if len(trl) == n { + // This is guaranteed to happen exactly once during this loop. + sort.Sort(trl) + } + continue + } + if tr.Start.After(trl[n-1].Start) { + continue + } + + // Find where to insert this one. + tr.ref() + i := sort.Search(n, func(i int) bool { return trl[i].Start.After(tr.Start) }) + trl[n-1].unref() + copy(trl[i+1:], trl[i:]) + trl[i] = tr + } + + return trl +} + +func getActiveTraces(fam string) traceList { + activeMu.RLock() + s := activeTraces[fam] + activeMu.RUnlock() + if s == nil { + return nil + } + return s.FirstN(maxActiveTraces) +} + +func getFamily(fam string, allocNew bool) *family { + completedMu.RLock() + f := completedTraces[fam] + completedMu.RUnlock() + if f == nil && allocNew { + f = allocFamily(fam) + } + return f +} + +func allocFamily(fam string) *family { + completedMu.Lock() + defer completedMu.Unlock() + f := completedTraces[fam] + if f == nil { + f = newFamily() + completedTraces[fam] = f + } + return f +} + +// family represents a set of trace buckets and associated latency information. +type family struct { + // traces may occur in multiple buckets. + Buckets [bucketsPerFamily]*traceBucket + + // latency time series + LatencyMu sync.RWMutex + Latency *timeseries.MinuteHourSeries +} + +func newFamily() *family { + return &family{ + Buckets: [bucketsPerFamily]*traceBucket{ + {Cond: minCond(0)}, + {Cond: minCond(50 * time.Millisecond)}, + {Cond: minCond(100 * time.Millisecond)}, + {Cond: minCond(200 * time.Millisecond)}, + {Cond: minCond(500 * time.Millisecond)}, + {Cond: minCond(1 * time.Second)}, + {Cond: minCond(10 * time.Second)}, + {Cond: minCond(100 * time.Second)}, + {Cond: errorCond{}}, + }, + Latency: timeseries.NewMinuteHourSeries(func() timeseries.Observable { return new(histogram) }), + } +} + +// traceBucket represents a size-capped bucket of historic traces, +// along with a condition for a trace to belong to the bucket. +type traceBucket struct { + Cond cond + + // Ring buffer implementation of a fixed-size FIFO queue. + mu sync.RWMutex + buf [tracesPerBucket]*trace + start int // < tracesPerBucket + length int // <= tracesPerBucket +} + +func (b *traceBucket) Add(tr *trace) { + b.mu.Lock() + defer b.mu.Unlock() + + i := b.start + b.length + if i >= tracesPerBucket { + i -= tracesPerBucket + } + if b.length == tracesPerBucket { + // "Remove" an element from the bucket. + b.buf[i].unref() + b.start++ + if b.start == tracesPerBucket { + b.start = 0 + } + } + b.buf[i] = tr + if b.length < tracesPerBucket { + b.length++ + } + tr.ref() +} + +// Copy returns a copy of the traces in the bucket. +// If tracedOnly is true, only the traces with trace information will be returned. +// The logs will be ref'd before returning; the caller should call +// the Free method when it is done with them. +// TODO(dsymonds): keep track of traced requests in separate buckets. +func (b *traceBucket) Copy(tracedOnly bool) traceList { + b.mu.RLock() + defer b.mu.RUnlock() + + trl := make(traceList, 0, b.length) + for i, x := 0, b.start; i < b.length; i++ { + tr := b.buf[x] + if !tracedOnly || tr.spanID != 0 { + tr.ref() + trl = append(trl, tr) + } + x++ + if x == b.length { + x = 0 + } + } + return trl +} + +func (b *traceBucket) Empty() bool { + b.mu.RLock() + defer b.mu.RUnlock() + return b.length == 0 +} + +// cond represents a condition on a trace. +type cond interface { + match(t *trace) bool + String() string +} + +type minCond time.Duration + +func (m minCond) match(t *trace) bool { return t.Elapsed >= time.Duration(m) } +func (m minCond) String() string { return fmt.Sprintf("≥%gs", time.Duration(m).Seconds()) } + +type errorCond struct{} + +func (e errorCond) match(t *trace) bool { return t.IsError } +func (e errorCond) String() string { return "errors" } + +type traceList []*trace + +// Free calls unref on each element of the list. +func (trl traceList) Free() { + for _, t := range trl { + t.unref() + } +} + +// traceList may be sorted in reverse chronological order. +func (trl traceList) Len() int { return len(trl) } +func (trl traceList) Less(i, j int) bool { return trl[i].Start.After(trl[j].Start) } +func (trl traceList) Swap(i, j int) { trl[i], trl[j] = trl[j], trl[i] } + +// An event is a timestamped log entry in a trace. +type event struct { + When time.Time + Elapsed time.Duration // since previous event in trace + NewDay bool // whether this event is on a different day to the previous event + Recyclable bool // whether this event was passed via LazyLog + Sensitive bool // whether this event contains sensitive information + What interface{} // string or fmt.Stringer +} + +// WhenString returns a string representation of the elapsed time of the event. +// It will include the date if midnight was crossed. +func (e event) WhenString() string { + if e.NewDay { + return e.When.Format("2006/01/02 15:04:05.000000") + } + return e.When.Format("15:04:05.000000") +} + +// discarded represents a number of discarded events. +// It is stored as *discarded to make it easier to update in-place. +type discarded int + +func (d *discarded) String() string { + return fmt.Sprintf("(%d events discarded)", int(*d)) +} + +// trace represents an active or complete request, +// either sent or received by this program. +type trace struct { + // Family is the top-level grouping of traces to which this belongs. + Family string + + // Title is the title of this trace. + Title string + + // Timing information. + Start time.Time + Elapsed time.Duration // zero while active + + // Trace information if non-zero. + traceID uint64 + spanID uint64 + + // Whether this trace resulted in an error. + IsError bool + + // Append-only sequence of events (modulo discards). + mu sync.RWMutex + events []event + maxEvents int + + refs int32 // how many buckets this is in + recycler func(interface{}) + disc discarded // scratch space to avoid allocation + + finishStack []byte // where finish was called, if DebugUseAfterFinish is set + + eventsBuf [4]event // preallocated buffer in case we only log a few events +} + +func (tr *trace) reset() { + // Clear all but the mutex. Mutexes may not be copied, even when unlocked. + tr.Family = "" + tr.Title = "" + tr.Start = time.Time{} + tr.Elapsed = 0 + tr.traceID = 0 + tr.spanID = 0 + tr.IsError = false + tr.maxEvents = 0 + tr.events = nil + tr.refs = 0 + tr.recycler = nil + tr.disc = 0 + tr.finishStack = nil + for i := range tr.eventsBuf { + tr.eventsBuf[i] = event{} + } +} + +// delta returns the elapsed time since the last event or the trace start, +// and whether it spans midnight. +// L >= tr.mu +func (tr *trace) delta(t time.Time) (time.Duration, bool) { + if len(tr.events) == 0 { + return t.Sub(tr.Start), false + } + prev := tr.events[len(tr.events)-1].When + return t.Sub(prev), prev.Day() != t.Day() +} + +func (tr *trace) addEvent(x interface{}, recyclable, sensitive bool) { + if DebugUseAfterFinish && tr.finishStack != nil { + buf := make([]byte, 4<<10) // 4 KB should be enough + n := runtime.Stack(buf, false) + log.Printf("net/trace: trace used after finish:\nFinished at:\n%s\nUsed at:\n%s", tr.finishStack, buf[:n]) + } + + /* + NOTE TO DEBUGGERS + + If you are here because your program panicked in this code, + it is almost definitely the fault of code using this package, + and very unlikely to be the fault of this code. + + The most likely scenario is that some code elsewhere is using + a trace.Trace after its Finish method is called. + You can temporarily set the DebugUseAfterFinish var + to help discover where that is; do not leave that var set, + since it makes this package much less efficient. + */ + + e := event{When: time.Now(), What: x, Recyclable: recyclable, Sensitive: sensitive} + tr.mu.Lock() + e.Elapsed, e.NewDay = tr.delta(e.When) + if len(tr.events) < tr.maxEvents { + tr.events = append(tr.events, e) + } else { + // Discard the middle events. + di := int((tr.maxEvents - 1) / 2) + if d, ok := tr.events[di].What.(*discarded); ok { + (*d)++ + } else { + // disc starts at two to count for the event it is replacing, + // plus the next one that we are about to drop. + tr.disc = 2 + if tr.recycler != nil && tr.events[di].Recyclable { + go tr.recycler(tr.events[di].What) + } + tr.events[di].What = &tr.disc + } + // The timestamp of the discarded meta-event should be + // the time of the last event it is representing. + tr.events[di].When = tr.events[di+1].When + + if tr.recycler != nil && tr.events[di+1].Recyclable { + go tr.recycler(tr.events[di+1].What) + } + copy(tr.events[di+1:], tr.events[di+2:]) + tr.events[tr.maxEvents-1] = e + } + tr.mu.Unlock() +} + +func (tr *trace) LazyLog(x fmt.Stringer, sensitive bool) { + tr.addEvent(x, true, sensitive) +} + +func (tr *trace) LazyPrintf(format string, a ...interface{}) { + tr.addEvent(&lazySprintf{format, a}, false, false) +} + +func (tr *trace) SetError() { tr.IsError = true } + +func (tr *trace) SetRecycler(f func(interface{})) { + tr.recycler = f +} + +func (tr *trace) SetTraceInfo(traceID, spanID uint64) { + tr.traceID, tr.spanID = traceID, spanID +} + +func (tr *trace) SetMaxEvents(m int) { + // Always keep at least three events: first, discarded count, last. + if len(tr.events) == 0 && m > 3 { + tr.maxEvents = m + } +} + +func (tr *trace) ref() { + atomic.AddInt32(&tr.refs, 1) +} + +func (tr *trace) unref() { + if atomic.AddInt32(&tr.refs, -1) == 0 { + if tr.recycler != nil { + // freeTrace clears tr, so we hold tr.recycler and tr.events here. + go func(f func(interface{}), es []event) { + for _, e := range es { + if e.Recyclable { + f(e.What) + } + } + }(tr.recycler, tr.events) + } + + freeTrace(tr) + } +} + +func (tr *trace) When() string { + return tr.Start.Format("2006/01/02 15:04:05.000000") +} + +func (tr *trace) ElapsedTime() string { + t := tr.Elapsed + if t == 0 { + // Active trace. + t = time.Since(tr.Start) + } + return fmt.Sprintf("%.6f", t.Seconds()) +} + +func (tr *trace) Events() []event { + tr.mu.RLock() + defer tr.mu.RUnlock() + return tr.events +} + +var traceFreeList = make(chan *trace, 1000) // TODO(dsymonds): Use sync.Pool? + +// newTrace returns a trace ready to use. +func newTrace() *trace { + select { + case tr := <-traceFreeList: + return tr + default: + return new(trace) + } +} + +// freeTrace adds tr to traceFreeList if there's room. +// This is non-blocking. +func freeTrace(tr *trace) { + if DebugUseAfterFinish { + return // never reuse + } + tr.reset() + select { + case traceFreeList <- tr: + default: + } +} + +func elapsed(d time.Duration) string { + b := []byte(fmt.Sprintf("%.6f", d.Seconds())) + + // For subsecond durations, blank all zeros before decimal point, + // and all zeros between the decimal point and the first non-zero digit. + if d < time.Second { + dot := bytes.IndexByte(b, '.') + for i := 0; i < dot; i++ { + b[i] = ' ' + } + for i := dot + 1; i < len(b); i++ { + if b[i] == '0' { + b[i] = ' ' + } else { + break + } + } + } + + return string(b) +} + +var pageTmplCache *template.Template +var pageTmplOnce sync.Once + +func pageTmpl() *template.Template { + pageTmplOnce.Do(func() { + pageTmplCache = template.Must(template.New("Page").Funcs(template.FuncMap{ + "elapsed": elapsed, + "add": func(a, b int) int { return a + b }, + }).Parse(pageHTML)) + }) + return pageTmplCache +} + +const pageHTML = ` +{{template "Prolog" .}} +{{template "StatusTable" .}} +{{template "Epilog" .}} + +{{define "Prolog"}} + + + /debug/requests + + + + +

/debug/requests

+{{end}} {{/* end of Prolog */}} + +{{define "StatusTable"}} + + {{range $fam := .Families}} + + + + {{$n := index $.ActiveTraceCount $fam}} + + + {{$f := index $.CompletedTraces $fam}} + {{range $i, $b := $f.Buckets}} + {{$empty := $b.Empty}} + + {{end}} + + {{$nb := len $f.Buckets}} + + + + + + {{end}} +
{{$fam}} + {{if $n}}{{end}} + [{{$n}} active] + {{if $n}}{{end}} + + {{if not $empty}}{{end}} + [{{.Cond}}] + {{if not $empty}}{{end}} + + [minute] + + [hour] + + [total] +
+{{end}} {{/* end of StatusTable */}} + +{{define "Epilog"}} +{{if $.Traces}} +
+

Family: {{$.Family}}

+ +{{if or $.Expanded $.Traced}} + [Normal/Summary] +{{else}} + [Normal/Summary] +{{end}} + +{{if or (not $.Expanded) $.Traced}} + [Normal/Expanded] +{{else}} + [Normal/Expanded] +{{end}} + +{{if not $.Active}} + {{if or $.Expanded (not $.Traced)}} + [Traced/Summary] + {{else}} + [Traced/Summary] + {{end}} + {{if or (not $.Expanded) (not $.Traced)}} + [Traced/Expanded] + {{else}} + [Traced/Expanded] + {{end}} +{{end}} + +{{if $.Total}} +

Showing {{len $.Traces}} of {{$.Total}} traces.

+{{end}} + + + + + {{range $tr := $.Traces}} + + + + + {{/* TODO: include traceID/spanID */}} + + {{if $.Expanded}} + {{range $tr.Events}} + + + + + + {{end}} + {{end}} + {{end}} +
+ {{if $.Active}}Active{{else}}Completed{{end}} Requests +
WhenElapsed (s)
{{$tr.When}}{{$tr.ElapsedTime}}{{$tr.Title}}
{{.WhenString}}{{elapsed .Elapsed}}{{if or $.ShowSensitive (not .Sensitive)}}... {{.What}}{{else}}[redacted]{{end}}
+{{end}} {{/* if $.Traces */}} + +{{if $.Histogram}} +

Latency (µs) of {{$.Family}} over {{$.HistogramWindow}}

+{{$.Histogram}} +{{end}} {{/* if $.Histogram */}} + + + +{{end}} {{/* end of Epilog */}} +` diff --git a/vendor/golang.org/x/net/trace/trace_go16.go b/vendor/golang.org/x/net/trace/trace_go16.go new file mode 100644 index 0000000000..d608191185 --- /dev/null +++ b/vendor/golang.org/x/net/trace/trace_go16.go @@ -0,0 +1,21 @@ +// Copyright 2017 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// +build !go1.7 + +package trace + +import "golang.org/x/net/context" + +// NewContext returns a copy of the parent context +// and associates it with a Trace. +func NewContext(ctx context.Context, tr Trace) context.Context { + return context.WithValue(ctx, contextKey, tr) +} + +// FromContext returns the Trace bound to the context, if any. +func FromContext(ctx context.Context) (tr Trace, ok bool) { + tr, ok = ctx.Value(contextKey).(Trace) + return +} diff --git a/vendor/golang.org/x/net/trace/trace_go17.go b/vendor/golang.org/x/net/trace/trace_go17.go new file mode 100644 index 0000000000..df6e1fba7c --- /dev/null +++ b/vendor/golang.org/x/net/trace/trace_go17.go @@ -0,0 +1,21 @@ +// Copyright 2017 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// +build go1.7 + +package trace + +import "context" + +// NewContext returns a copy of the parent context +// and associates it with a Trace. +func NewContext(ctx context.Context, tr Trace) context.Context { + return context.WithValue(ctx, contextKey, tr) +} + +// FromContext returns the Trace bound to the context, if any. +func FromContext(ctx context.Context) (tr Trace, ok bool) { + tr, ok = ctx.Value(contextKey).(Trace) + return +} From 47d060e1e88742154c12d0b20e6974e9218368fe Mon Sep 17 00:00:00 2001 From: ArchetypicalDude Date: Tue, 6 Mar 2018 11:22:10 +0100 Subject: [PATCH 2/5] - some fiddling on tx_pool --- core/blockchain.go | 4 +++- core/tx_pool.go | 16 ++++++++++++---- 2 files changed, 15 insertions(+), 5 deletions(-) diff --git a/core/blockchain.go b/core/blockchain.go index 4ae0e4f4ec..eca9f63094 100644 --- a/core/blockchain.go +++ b/core/blockchain.go @@ -920,9 +920,10 @@ func (bc *BlockChain) WriteBlockWithState(block *types.Block, receipts []*types. size = triedb.Size() limit = common.StorageSize(bc.cacheConfig.TrieNodeLimit) * 1024 * 1024 ) - if size > limit || bc.gcproc > bc.cacheConfig.TrieTimeLimit { + if size > 8*limit { // If we're exceeding limits but haven't reached a large enough memory gap, // warn the user that the system is becoming unstable. + /* if chosen < lastWrite+triesInMemory { switch { case size >= 2*limit: @@ -931,6 +932,7 @@ func (bc *BlockChain) WriteBlockWithState(block *types.Block, receipts []*types. log.Info("State in memory for too long, committing", "time", bc.gcproc, "allowance", bc.cacheConfig.TrieTimeLimit, "optimum", float64(chosen-lastWrite)/triesInMemory) } } + */ // If optimum or critical limits reached, write to disk if chosen >= lastWrite+triesInMemory || size >= 2*limit || bc.gcproc >= 2*bc.cacheConfig.TrieTimeLimit { triedb.Commit(header.Root, true) diff --git a/core/tx_pool.go b/core/tx_pool.go index dc3ddc4232..35f0511b8b 100644 --- a/core/tx_pool.go +++ b/core/tx_pool.go @@ -144,15 +144,15 @@ type TxPoolConfig struct { // pool. var DefaultTxPoolConfig = TxPoolConfig{ Journal: "transactions.rlp", - Rejournal: time.Hour, + Rejournal: 24*time.Hour, - PriceLimit: 1, + PriceLimit: 50, PriceBump: 10, AccountSlots: 16, - GlobalSlots: 4096, + GlobalSlots: 200, AccountQueue: 64, - GlobalQueue: 1024, + GlobalQueue: 100, Lifetime: 3 * time.Hour, } @@ -351,6 +351,14 @@ func (pool *TxPool) lockedReset(oldHead, newHead *types.Header) { // reset retrieves the current state of the blockchain and ensures the content // of the transaction pool is valid with regard to the chain state. func (pool *TxPool) reset(oldHead, newHead *types.Header) { + log.Info("TxPool reset exec") + /* + pool.pending = make(map[common.Address]*txList) // All currently processable transactions + pool.queue = make(map[common.Address]*txList) // Queued but non-processable transactions + pool.beats = make(map[common.Address]time.Time) // Last heartbeat from each known account + pool.all = make(map[common.Hash]*types.Transaction) // All transactions to allow lookups + pool.priced = nil // All transactions sorted by price*/ + return // If we're reorging an old state, reinject all dropped transactions var reinject types.Transactions From 816ce69308b4e2a4a0e1649d02560700177d3d89 Mon Sep 17 00:00:00 2001 From: ArchetypicalDude Date: Tue, 6 Mar 2018 14:00:57 +0100 Subject: [PATCH 3/5] - bring changes in line with official repository --- core/tx_pool.go | 16 ++++------------ ethdb/badger.go | 23 +++++++++++------------ 2 files changed, 15 insertions(+), 24 deletions(-) diff --git a/core/tx_pool.go b/core/tx_pool.go index 10951bc322..0534fe57a1 100644 --- a/core/tx_pool.go +++ b/core/tx_pool.go @@ -144,15 +144,15 @@ type TxPoolConfig struct { // pool. var DefaultTxPoolConfig = TxPoolConfig{ Journal: "transactions.rlp", - Rejournal: 24*time.Hour, + Rejournal: time.Hour, - PriceLimit: 50, + PriceLimit: 1, PriceBump: 10, AccountSlots: 16, - GlobalSlots: 200, + GlobalSlots: 4096, AccountQueue: 64, - GlobalQueue: 100, + GlobalQueue: 1024, Lifetime: 3 * time.Hour, } @@ -351,14 +351,6 @@ func (pool *TxPool) lockedReset(oldHead, newHead *types.Header) { // reset retrieves the current state of the blockchain and ensures the content // of the transaction pool is valid with regard to the chain state. func (pool *TxPool) reset(oldHead, newHead *types.Header) { - log.Info("TxPool reset exec") - /* - pool.pending = make(map[common.Address]*txList) // All currently processable transactions - pool.queue = make(map[common.Address]*txList) // Queued but non-processable transactions - pool.beats = make(map[common.Address]time.Time) // Last heartbeat from each known account - pool.all = make(map[common.Hash]*types.Transaction) // All transactions to allow lookups - pool.priced = nil // All transactions sorted by price*/ - return // If we're reorging an old state, reinject all dropped transactions var reinject types.Transactions diff --git a/ethdb/badger.go b/ethdb/badger.go index 4289618af4..df59d7bd28 100644 --- a/ethdb/badger.go +++ b/ethdb/badger.go @@ -23,12 +23,11 @@ import ( "time" "github.com/ethereum/go-ethereum/log" - //"github.com/ethereum/go-ethereum/metrics" + "github.com/ethereum/go-ethereum/metrics" "github.com/dgraph-io/badger" "github.com/dgraph-io/badger/options" "github.com/ethereum/go-ethereum/common" - gometrics "github.com/rcrowley/go-metrics" ) @@ -37,15 +36,15 @@ type BadgerDatabase struct { fn string // filename for reporting db *badger.DB badgerCache *BadgerCache - getTimer gometrics.Timer // Timer for measuring the database get request counts and latencies - putTimer gometrics.Timer // Timer for measuring the database put request counts and latencies - delTimer gometrics.Timer // Timer for measuring the database delete request counts and latencies - missMeter gometrics.Meter // Meter for measuring the missed database get requests - readMeter gometrics.Meter // Meter for measuring the database get request data usage - writeMeter gometrics.Meter // Meter for measuring the database put request data usage - compTimeMeter gometrics.Meter // Meter for measuring the total time spent in database compaction - compReadMeter gometrics.Meter // Meter for measuring the data read during compaction - compWriteMeter gometrics.Meter // Meter for measuring the data written during compaction + getTimer metrics.Timer // Timer for measuring the database get request counts and latencies + putTimer metrics.Timer // Timer for measuring the database put request counts and latencies + delTimer metrics.Timer // Timer for measuring the database delete request counts and latencies + missMeter metrics.Meter // Meter for measuring the missed database get requests + readMeter metrics.Meter // Meter for measuring the database get request data usage + writeMeter metrics.Meter // Meter for measuring the database put request data usage + compTimeMeter metrics.Meter // Meter for measuring the total time spent in database compaction + compReadMeter metrics.Meter // Meter for measuring the data read during compaction + compWriteMeter metrics.Meter // Meter for measuring the data written during compaction quitLock sync.Mutex // Mutex protecting the quit channel access quitChan chan chan error // Quit channel to stop the metrics collection before closing the database @@ -75,7 +74,7 @@ func NewBadgerDatabase(file string) (*BadgerDatabase, error) { log: logger, } - ret.badgerCache = &BadgerCache{db: ret, c: make(map[string][]byte), size: 0, limit: 100000000} + ret.badgerCache = &BadgerCache{db: ret, c: make(map[string][]byte), size: 0, limit: 500000000} return ret, nil } From f13096e3c184f882b1825b42a178830da6e3c2dc Mon Sep 17 00:00:00 2001 From: ArchetypicalDude Date: Tue, 13 Mar 2018 15:39:28 +0100 Subject: [PATCH 4/5] - enabled metrics for badger: eth.db.chaindata.user. Write-metrics not yet comparable to leveldb because everything gets cached first --- eth/backend.go | 2 +- ethdb/badger.go | 140 +++++++++++------------------------------------- 2 files changed, 32 insertions(+), 110 deletions(-) diff --git a/eth/backend.go b/eth/backend.go index 94aad23101..307164854f 100644 --- a/eth/backend.go +++ b/eth/backend.go @@ -204,7 +204,7 @@ func CreateDB(ctx *node.ServiceContext, config *Config, name string) (ethdb.Data if err != nil { return nil, err } - if db, ok := db.(*ethdb.LDBDatabase); ok { + if db, ok := db.(*ethdb.BadgerDatabase); ok { db.Meter("eth/db/chaindata/") } return db, nil diff --git a/ethdb/badger.go b/ethdb/badger.go index df59d7bd28..184cc72202 100644 --- a/ethdb/badger.go +++ b/ethdb/badger.go @@ -42,9 +42,6 @@ type BadgerDatabase struct { missMeter metrics.Meter // Meter for measuring the missed database get requests readMeter metrics.Meter // Meter for measuring the database get request data usage writeMeter metrics.Meter // Meter for measuring the database put request data usage - compTimeMeter metrics.Meter // Meter for measuring the total time spent in database compaction - compReadMeter metrics.Meter // Meter for measuring the data read during compaction - compWriteMeter metrics.Meter // Meter for measuring the data written during compaction quitLock sync.Mutex // Mutex protecting the quit channel access quitChan chan chan error // Quit channel to stop the metrics collection before closing the database @@ -85,17 +82,7 @@ func (db *BadgerDatabase) Path() string { // Put puts the given key / value to the queue func (db *BadgerDatabase) Put(key []byte, value []byte) error { - // Measure the database put latency, if requested - if db.putTimer != nil { - defer db.putTimer.UpdateSince(time.Now()) - } - // Generate the data to write to disk, update the meter and write - //value = rle.Compress(value) - if db.writeMeter != nil { - db.writeMeter.Mark(int64(len(value))) - } - db.badgerCache.lock.Lock() db.badgerCache.c[string(key)] = common.CopyBytes(value) db.badgerCache.size += len(value)+len(key) @@ -142,14 +129,20 @@ func (badgerCache *BadgerCache) Flush() (err error) { defer badgerCache.lock.Unlock() txn := badgerCache.db.db.NewTransaction(true) - for key, value := range badgerCache.c { + putStartTime := time.Now() + if badgerCache.db.writeMeter != nil { + badgerCache.db.writeMeter.Mark(int64(len(value))) + } err = txn.Set([]byte(key), value) if err == badger.ErrTxnTooBig { txn.Commit(nil) txn = badgerCache.db.db.NewTransaction(true) err = txn.Set([]byte(key), value) } + if badgerCache.db.putTimer != nil { + badgerCache.db.putTimer.UpdateSince(putStartTime) + } } err = txn.Commit(nil) log.Info("Badger flushed to disk", "badgerCache size", badgerCache.size) @@ -160,16 +153,14 @@ func (badgerCache *BadgerCache) Flush() (err error) { // Get returns the given key if it's present. func (db *BadgerDatabase) Get(key []byte) (dat []byte, err error) { - // Measure the database get latency, if requested - if db.getTimer != nil { - defer db.getTimer.UpdateSince(time.Now()) - } - - db.badgerCache.lock.RLock() dat = db.badgerCache.c[string(key)] db.badgerCache.lock.RUnlock() if dat == nil { + // Measure the database get latency, if requested + if db.getTimer != nil { + defer db.getTimer.UpdateSince(time.Now()) + } err = db.db.View(func(txn *badger.Txn) error { item, err := txn.Get(key) if err != nil { @@ -182,6 +173,10 @@ func (db *BadgerDatabase) Get(key []byte) (dat []byte, err error) { dat = common.CopyBytes(val) return nil }) + //Update the actually retrieved amount of data + if db.readMeter != nil { + db.readMeter.Mark(int64(len(dat))) + } } if err != nil { if db.missMeter != nil { @@ -189,20 +184,13 @@ func (db *BadgerDatabase) Get(key []byte) (dat []byte, err error) { } return nil, err } - //Update the actually retrieved amount of data - if db.readMeter != nil { - db.readMeter.Mark(int64(len(dat))) - } + return dat, nil - //return rle.Decompress(dat) } // Delete deletes the key from the queue and database func (db *BadgerDatabase) Delete(key []byte) error { - // Measure the database delete latency, if requested - if db.delTimer != nil { - defer db.delTimer.UpdateSince(time.Now()) - } + // Execute the actual operation db.badgerCache.lock.Lock() delete(db.badgerCache.c, string(key)) @@ -210,6 +198,11 @@ func (db *BadgerDatabase) Delete(key []byte) error { //TODO: also subtract len(value) db.badgerCache.size-=len(key) db.badgerCache.lock.Unlock() + + // Measure the database delete latency, if requested + if db.delTimer != nil { + defer db.delTimer.UpdateSince(time.Now()) + } return db.db.Update(func(txn *badger.Txn) error { err := txn.Delete(key) if err == badger.ErrKeyNotFound { @@ -270,6 +263,8 @@ func (db *BadgerDatabase) NewIterator() badgerIterator { } func (db *BadgerDatabase) Close() { + + db.badgerCache.Flush() // Stop the metrics collection to avoid internal database races db.quitLock.Lock() defer db.quitLock.Unlock() @@ -281,7 +276,6 @@ func (db *BadgerDatabase) Close() { db.log.Error("Metrics collection failed", "err", err) } } - db.badgerCache.Flush() err := db.db.Close() if err == nil { db.log.Info("Database closed") @@ -290,7 +284,6 @@ func (db *BadgerDatabase) Close() { } } -/* // Meter configures the database metrics collectors and func (db *BadgerDatabase) Meter(prefix string) { // Short circuit metering if the metrics system is disabled @@ -298,92 +291,21 @@ func (db *BadgerDatabase) Meter(prefix string) { return } // Initialize all the metrics collector at the requested prefix - db.getTimer = metrics.NewTimer(prefix + "user/gets") - db.putTimer = metrics.NewTimer(prefix + "user/puts") - db.delTimer = metrics.NewTimer(prefix + "user/dels") - db.missMeter = metrics.NewMeter(prefix + "user/misses") - db.readMeter = metrics.NewMeter(prefix + "user/reads") - db.writeMeter = metrics.NewMeter(prefix + "user/writes") - db.compTimeMeter = metrics.NewMeter(prefix + "compact/time") - db.compReadMeter = metrics.NewMeter(prefix + "compact/input") - db.compWriteMeter = metrics.NewMeter(prefix + "compact/output") + db.getTimer = metrics.NewRegisteredTimer(prefix+"user/gets", nil) + db.putTimer = metrics.NewRegisteredTimer(prefix+"user/puts", nil) + db.delTimer = metrics.NewRegisteredTimer(prefix+"user/dels", nil) + db.missMeter = metrics.NewRegisteredMeter(prefix+"user/misses", nil) + db.readMeter = metrics.NewRegisteredMeter(prefix+"user/reads", nil) + db.writeMeter = metrics.NewRegisteredMeter(prefix+"user/writes", nil) // Create a quit channel for the periodic collector and run it db.quitLock.Lock() db.quitChan = make(chan chan error) db.quitLock.Unlock() - go db.meter(3 * time.Second) + //go db.meter(3 * time.Second) } -*/ -/* -func (db *BadgerDatabase) meter(refresh time.Duration) { - // Create the counters to store current and previous values - counters := make([][]float64, 2) - for i := 0; i < 2; i++ { - counters[i] = make([]float64, 3) - } - // Iterate ad infinitum and collect the stats - for i := 1; ; i++ { - // Retrieve the database stats - stats, err := db.db.GetProperty("leveldb.stats") - if err != nil { - db.log.Error("Failed to read database stats", "err", err) - return - } - // Find the compaction table, skip the header - lines := strings.Split(stats, "\n") - for len(lines) > 0 && strings.TrimSpace(lines[0]) != "Compactions" { - lines = lines[1:] - } - if len(lines) <= 3 { - db.log.Error("Compaction table not found") - return - } - lines = lines[3:] - - // Iterate over all the table rows, and accumulate the entries - for j := 0; j < len(counters[i%2]); j++ { - counters[i%2][j] = 0 - } - for _, line := range lines { - parts := strings.Split(line, "|") - if len(parts) != 6 { - break - } - for idx, counter := range parts[3:] { - value, err := strconv.ParseFloat(strings.TrimSpace(counter), 64) - if err != nil { - db.log.Error("Compaction entry parsing failed", "err", err) - return - } - counters[i%2][idx] += value - } - } - // Update all the requested meters - if db.compTimeMeter != nil { - db.compTimeMeter.Mark(int64((counters[i%2][0] - counters[(i-1)%2][0]) * 1000 * 1000 * 1000)) - } - if db.compReadMeter != nil { - db.compReadMeter.Mark(int64((counters[i%2][1] - counters[(i-1)%2][1]) * 1024 * 1024)) - } - if db.compWriteMeter != nil { - db.compWriteMeter.Mark(int64((counters[i%2][2] - counters[(i-1)%2][2]) * 1024 * 1024)) - } - // Sleep a bit, then repeat the stats collection - select { - case errc := <-db.quitChan: - // Quit requesting, stop hammering the database - errc <- nil - return - - case <-time.After(refresh): - // Timeout, gather a new set of stats - } - } -} -*/ func (db *BadgerDatabase) NewBatch() Batch { return &badgerBatch{db: db} } From ca6585dc6f737b742f5e323a01664f840eec4f23 Mon Sep 17 00:00:00 2001 From: ArchetypicalDude Date: Tue, 20 Mar 2018 09:06:33 +0100 Subject: [PATCH 5/5] - removed additional badger-cache - implemented "github.com/ethereum/go-ethereum/metrics" - added batchPutTimer, batchWriteTimer and batchWriteMeter to metrics to allow more accurate statistics for database performance ("user/batchPuts", "user/batchWriteTimes", "user/batchWrites") --- eth/backend.go | 4 +- ethdb/badger.go | 179 +++++++++++++++++------------------------------- 2 files changed, 65 insertions(+), 118 deletions(-) diff --git a/eth/backend.go b/eth/backend.go index 307164854f..a1574c108d 100644 --- a/eth/backend.go +++ b/eth/backend.go @@ -158,7 +158,9 @@ func New(ctx *node.ServiceContext, config *Config) (*Ethereum, error) { eth.blockchain.SetHead(compat.RewindTo) core.WriteChainConfig(chainDb, genesisHash, chainConfig) } - eth.bloomIndexer.Start(eth.blockchain) + + // Disable for tests + //eth.bloomIndexer.Start(eth.blockchain) if config.TxPool.Journal != "" { config.TxPool.Journal = ctx.ResolvePath(config.TxPool.Journal) diff --git a/ethdb/badger.go b/ethdb/badger.go index 184cc72202..e8b9604aa7 100644 --- a/ethdb/badger.go +++ b/ethdb/badger.go @@ -35,21 +35,22 @@ import ( type BadgerDatabase struct { fn string // filename for reporting db *badger.DB - badgerCache *BadgerCache getTimer metrics.Timer // Timer for measuring the database get request counts and latencies putTimer metrics.Timer // Timer for measuring the database put request counts and latencies delTimer metrics.Timer // Timer for measuring the database delete request counts and latencies missMeter metrics.Meter // Meter for measuring the missed database get requests readMeter metrics.Meter // Meter for measuring the database get request data usage writeMeter metrics.Meter // Meter for measuring the database put request data usage + batchPutTimer metrics.Timer + batchWriteTimer metrics.Timer + batchWriteMeter metrics.Meter quitLock sync.Mutex // Mutex protecting the quit channel access - quitChan chan chan error // Quit channel to stop the metrics collection before closing the database log log.Logger // Contextual logger tracking the database path } -// NewLDBDatabase returns a LevelDB wrapped object. +// NewBadgerDatabase returns a BadgerDB wrapped object. func NewBadgerDatabase(file string) (*BadgerDatabase, error) { logger := log.New("database", file) @@ -71,7 +72,6 @@ func NewBadgerDatabase(file string) (*BadgerDatabase, error) { log: logger, } - ret.badgerCache = &BadgerCache{db: ret, c: make(map[string][]byte), size: 0, limit: 500000000} return ret, nil } @@ -81,27 +81,23 @@ func (db *BadgerDatabase) Path() string { } // Put puts the given key / value to the queue -func (db *BadgerDatabase) Put(key []byte, value []byte) error { +func (db *BadgerDatabase) Put(key []byte, value []byte) error { + if db.putTimer != nil { + defer db.putTimer.UpdateSince(time.Now()) + } - db.badgerCache.lock.Lock() - db.badgerCache.c[string(key)] = common.CopyBytes(value) - db.badgerCache.size += len(value)+len(key) - db.badgerCache.lock.Unlock() - - if db.badgerCache.size >= db.badgerCache.limit { - return db.badgerCache.Flush() + if db.writeMeter != nil { + db.writeMeter.Mark(int64(len(value))) } - return nil + + return db.db.Update(func(txn *badger.Txn) error { + err := txn.Set(key, value) + return err + }) } func (db *BadgerDatabase) Has(key []byte) (ret bool, err error) { - db.badgerCache.lock.RLock() - defer db.badgerCache.lock.RUnlock() - if db.badgerCache.c[string(key)] != nil { - return true, nil - } - err = db.db.View(func(txn *badger.Txn) error { item, err := txn.Get(key) if item != nil { @@ -116,67 +112,27 @@ func (db *BadgerDatabase) Has(key []byte) (ret bool, err error) { return ret, err } -type BadgerCache struct { - db *BadgerDatabase - c map[string][]byte - size int - limit int - lock sync.RWMutex -} - -func (badgerCache *BadgerCache) Flush() (err error) { - badgerCache.lock.Lock() - defer badgerCache.lock.Unlock() - - txn := badgerCache.db.db.NewTransaction(true) - for key, value := range badgerCache.c { - putStartTime := time.Now() - if badgerCache.db.writeMeter != nil { - badgerCache.db.writeMeter.Mark(int64(len(value))) - } - err = txn.Set([]byte(key), value) - if err == badger.ErrTxnTooBig { - txn.Commit(nil) - txn = badgerCache.db.db.NewTransaction(true) - err = txn.Set([]byte(key), value) - } - if badgerCache.db.putTimer != nil { - badgerCache.db.putTimer.UpdateSince(putStartTime) - } - } - err = txn.Commit(nil) - log.Info("Badger flushed to disk", "badgerCache size", badgerCache.size) - badgerCache.size = 0 - badgerCache.c = make(map[string][]byte) - return err -} - // Get returns the given key if it's present. func (db *BadgerDatabase) Get(key []byte) (dat []byte, err error) { - db.badgerCache.lock.RLock() - dat = db.badgerCache.c[string(key)] - db.badgerCache.lock.RUnlock() - if dat == nil { - // Measure the database get latency, if requested - if db.getTimer != nil { - defer db.getTimer.UpdateSince(time.Now()) + // Measure the database get latency, if requested + if db.getTimer != nil { + defer db.getTimer.UpdateSince(time.Now()) + } + err = db.db.View(func(txn *badger.Txn) error { + item, err := txn.Get(key) + if err != nil { + return err } - err = db.db.View(func(txn *badger.Txn) error { - item, err := txn.Get(key) - if err != nil { - return err - } - val, err := item.Value() - if err != nil { - return err - } - dat = common.CopyBytes(val) - return nil - }) - //Update the actually retrieved amount of data - if db.readMeter != nil { - db.readMeter.Mark(int64(len(dat))) + val, err := item.Value() + if err != nil { + return err } + dat = common.CopyBytes(val) + return nil + }) + //Update the actually retrieved amount of data + if db.readMeter != nil { + db.readMeter.Mark(int64(len(dat))) } if err != nil { if db.missMeter != nil { @@ -190,15 +146,6 @@ func (db *BadgerDatabase) Get(key []byte) (dat []byte, err error) { // Delete deletes the key from the queue and database func (db *BadgerDatabase) Delete(key []byte) error { - - // Execute the actual operation - db.badgerCache.lock.Lock() - delete(db.badgerCache.c, string(key)) - - //TODO: also subtract len(value) - db.badgerCache.size-=len(key) - db.badgerCache.lock.Unlock() - // Measure the database delete latency, if requested if db.delTimer != nil { defer db.delTimer.UpdateSince(time.Now()) @@ -263,19 +210,8 @@ func (db *BadgerDatabase) NewIterator() badgerIterator { } func (db *BadgerDatabase) Close() { - - db.badgerCache.Flush() - // Stop the metrics collection to avoid internal database races - db.quitLock.Lock() - defer db.quitLock.Unlock() - - if db.quitChan != nil { - errc := make(chan error) - db.quitChan <- errc - if err := <-errc; err != nil { - db.log.Error("Metrics collection failed", "err", err) - } - } + // hacky way around https://github.com/dgraph-io/badger/pull/437 + time.Sleep(time.Second * 15) err := db.db.Close() if err == nil { db.log.Info("Database closed") @@ -297,45 +233,53 @@ func (db *BadgerDatabase) Meter(prefix string) { db.missMeter = metrics.NewRegisteredMeter(prefix+"user/misses", nil) db.readMeter = metrics.NewRegisteredMeter(prefix+"user/reads", nil) db.writeMeter = metrics.NewRegisteredMeter(prefix+"user/writes", nil) - - // Create a quit channel for the periodic collector and run it - db.quitLock.Lock() - db.quitChan = make(chan chan error) - db.quitLock.Unlock() - - //go db.meter(3 * time.Second) + db.batchPutTimer = metrics.NewRegisteredTimer(prefix+"user/batchPuts", nil) + db.batchWriteTimer = metrics.NewRegisteredTimer(prefix+"user/batchWriteTimes", nil) + db.batchWriteMeter = metrics.NewRegisteredMeter(prefix+"user/batchWrites", nil) } func (db *BadgerDatabase) NewBatch() Batch { - return &badgerBatch{db: db} + return &badgerBatch{db: db, b: make(map[string][]byte)} } type badgerBatch struct { db *BadgerDatabase + b map[string][]byte size int } func (b *badgerBatch) Put(key, value []byte) error { - b.db.badgerCache.lock.Lock() - b.db.badgerCache.c[string(key)] = common.CopyBytes(value) - b.db.badgerCache.size += len(value)+len(key) - b.db.badgerCache.lock.Unlock() - if b.db.badgerCache.size >= b.db.badgerCache.limit { - b.db.badgerCache.Flush() + if b.db.batchPutTimer != nil { + defer b.db.batchPutTimer.UpdateSince(time.Now()) } + + b.b[string(key)] = common.CopyBytes(value) b.size += len(value) return nil } -func (b *badgerBatch) Write() error { - b.size = 0 - if b.db.badgerCache.size >= b.db.badgerCache.limit { - return b.db.badgerCache.Flush() +func (b *badgerBatch) Write() (err error) { + if b.db.batchWriteTimer != nil { + defer b.db.batchWriteTimer.UpdateSince(time.Now()) } - return nil + + if b.db.batchWriteMeter != nil { + b.db.batchWriteMeter.Mark(int64(b.size)) + } + + err = b.db.db.Update(func(txn *badger.Txn) error { + for key, value := range b.b { + err = txn.Set([]byte(key), value) + } + return err + }) + b.size = 0 + b.b = make(map[string][]byte) + return err } func (b *badgerBatch) Discard() { + b.b = make(map[string][]byte) b.size = 0 } @@ -344,6 +288,7 @@ func (b *badgerBatch) ValueSize() int { } func (b *badgerBatch) Reset() { + b.b = make(map[string][]byte) b.size = 0 }