go-ethereum/swarm/storage/netstore.go
zelig 075eb6ad99 SWORM - swarm poc 0.2 stability RC3
rebase on current develop branch 2016-07-09

swarm/services/ens:
* Rework ENS functionality to use the current spec.
* Refactor ens.go to set up sessions as needed
* Refactoring to support independent registrar and personal resolver contracts
* fixed tests

swarm/examples/album: improve upload UX
* No more unhandled errors and loose ends in the UI.
* Forward/Back buttons work as undo/redo
* all long operations display an informative modal dialog
* minor problems with cursor movements fixed
* reorder labels
* throbber as default image
* change browser history after delete or upload photo

swarm/storage:
* optimise splitter
* simplify reader
* benchmarks, tests improved
* simplify IO, remove chunkreader.go and io.SectionReader clone
* simplify error/timeout handling
* port pyramid splitter by karalabe, adapt to Splitter/DPA interface, rework params etc
* make NetStore.Put -> cloud.Store synchronous
* chunker join fix process leak and keep parallelisation limited to depth 1
* reset the base hash to SHA3. !!hard fork = no backward compatibility :)
* SectionReader -> LazySectionReader: Size method signature change
* introduce abort channel to joiner to fix process leak
* add abort channel   context to all manifest retrieval methods
* global waitgroup fixes intermittent upload test failures due to unfinished storage of chunks
* streamline joiner logic, contexts now unique to each readAt call
* LazyChunkReader seeker complains about missing size only if whence=2
* chunker join fix process leak and keep parallelisation limited to depth 1

swarm/network/kademlia:
* simplify code
* now really fix prox limit adjustment + tests and comments
* simplify and make readable findclosest algo code + comments
* reset initial time interval settings for kaddb findbest
* fix bucket replace scheme to optimise stability and availability
* absolute idle peers are disconnected after maxIdleInterval
* fix index out of range when deleting last idle peer from kaddb
* unwanted peers (due to full kad bucket) are now properly dropped with ErrUnwanted
* improve logging and reorg in kademlia
* simplify and fix code that deletes expired/unconnectable nodes
* kaddb.findBest does not get stuck on empty row but finds an actual missing node
* must replace node if bucket is full otherwise nodes will get stuck on empty rows
* kademlia: timer fix
* remove logging the state when syncing (resulting in deadlock)
* improved logs, fix potential deadlock in String()
* native go time marshalling
* tests fixed

swarm/network:
* log node address consistently in syncdb
* syncer logs queue cardinalities properly in syncUnSyncedKeys loop
* fix hive stop issue leading to send on closed chan + minor logging fixes
* better logging
* hive interface change
* fix process leak by adding select to state.synced <- false

swarm/cmd: swarm control cli improvements
* no more alias, swarm  is executable so it can be called via ssh
* environment vars now set to default, no need to preconfigure
* local and remote hive monitoring
* gethup.sh script now merged into swarm script and nice modularised
* simplify bash code and fix e2e tests in swarm/test
* stop method falls back to kill -9 after 10s
* enode method now supplies ip addr via ipecho request
* execute, options, rawoptions, setup, create-account addpeers and hive subcommands
* update-src -> update
* remote-update-scripts, remote-update-bin and remote-run
* local and remote monitoring of kademlia
* extensive documentation in  swarm/cmd/README.md
* add cleanlog, cleanbzz, update-src, remote-update-scripts, remote-update-bin, remote-run
* for remote binary update, take scripts from swarm/cmd/swarm
* raise default maxpeers to 40
* add latency loggingo to upload and download
* add checkdownload and checkaccess subcommands
* add mem/cpu/disk-info commands
* include randomfile cmd from test
* add enode, connect subcommands, improve startup (only one round needed)
* TODO: simplify tests using new checks

swarm/api/http: server handler: check protocol substring length to fix slice out of bounds crash

swarm/api:
* fix filesystem API - upload/Modify tests
* simplify downloader code and fix process leak
* change KAD defaults bucketsize 4, minproxbinsize 2
* add back final slash to paths in manifest matching
* fix filesystem api tests
2016-07-09 10:28:35 +02:00

118 lines
3.6 KiB
Go

package storage
import (
"path/filepath"
"sync"
"time"
"github.com/ethereum/go-ethereum/logger"
"github.com/ethereum/go-ethereum/logger/glog"
)
/*
NetStore is a cloud storage access abstaction layer for swarm
it contains the shared logic of network served chunk store/retrieval requests
both local (coming from DPA api) and remote (coming from peers via bzz protocol)
it implements the ChunkStore interface and embeds LocalStore
It is called by the bzz protocol instances via Depo (the store/retrieve request handler)
a protocol instance is running on each peer, so this is heavily parallelised.
NetStore falls back to a backend (CloudStorage interface)
implemented by bzz/network/forwarder. forwarder or IPFS or IPΞS
*/
type NetStore struct {
hashfunc Hasher
localStore *LocalStore
cloud CloudStore
lock sync.Mutex
}
// backend engine for cloud store
// It can be aggregate dispatching to several parallel implementations:
// bzz/network/forwarder. forwarder or IPFS or IPΞS
type CloudStore interface {
Store(*Chunk)
Deliver(*Chunk)
Retrieve(*Chunk)
}
type StoreParams struct {
ChunkDbPath string
DbCapacity uint64
CacheCapacity uint
Radius int
}
func NewStoreParams(path string) (self *StoreParams) {
return &StoreParams{
ChunkDbPath: filepath.Join(path, "chunks"),
DbCapacity: defaultDbCapacity,
CacheCapacity: defaultCacheCapacity,
Radius: defaultRadius,
}
}
// netstore contructor, takes path argument that is used to initialise dbStore,
// the persistent (disk) storage component of LocalStore
// the second argument is the hive, the connection/logistics manager for the node
func NewNetStore(hash Hasher, lstore *LocalStore, cloud CloudStore, params *StoreParams) *NetStore {
return &NetStore{
hashfunc: hash,
localStore: lstore,
cloud: cloud,
}
}
const (
// maximum number of peers that a retrieved message is delivered to
requesterCount = 3
)
var (
// timeout interval before retrieval is timed out
searchTimeout = 3 * time.Second
)
// store logic common to local and network chunk store requests
// ~ unsafe put in localdb no check if exists no extra copy no hash validation
// the chunk is forced to propagate (Cloud.Store) even if locally found!
// caller needs to make sure if that is wanted
func (self *NetStore) Put(entry *Chunk) {
self.localStore.Put(entry)
// handle deliveries
if entry.Req != nil {
glog.V(logger.Detail).Infof("[BZZ] NetStore.Put: localStore.Put %v hit existing request...delivering", entry.Key.Log())
// closing C singals to other routines (local requests)
// that the chunk is has been retrieved
close(entry.Req.C)
// deliver the chunk to requesters upstream
self.cloud.Deliver(entry)
} else {
glog.V(logger.Detail).Infof("[BZZ] NetStore.Put: localStore.Put %v stored locally", entry.Key.Log())
// handle propagating store requests
// go self.cloud.Store(entry)
self.cloud.Store(entry)
}
}
// retrieve logic common for local and network chunk retrieval requests
func (self *NetStore) Get(key Key) (*Chunk, error) {
var err error
chunk, err := self.localStore.Get(key)
if err == nil {
if chunk.Req == nil {
glog.V(logger.Detail).Infof("[BZZ] NetStore.Get: %v found locally", key)
} else {
glog.V(logger.Detail).Infof("[BZZ] NetStore.Get: %v hit on an existing request", key)
// no need to launch again
}
return chunk, err
}
// no data and no request status
glog.V(logger.Detail).Infof("[BZZ] NetStore.Get: %v not found locally. open new request", key)
chunk = NewChunk(key, newRequestStatus(key))
self.localStore.memStore.Put(chunk)
go self.cloud.Retrieve(chunk)
return chunk, nil
}