p2p/protocols: graceful shutdown of accounting metrics and fixes

This commit is contained in:
Fabio Barone 2018-11-19 10:38:32 -05:00
parent 798fbd054f
commit d009c63fc3
5 changed files with 93 additions and 69 deletions

View file

@ -20,7 +20,6 @@ import (
"time"
"github.com/ethereum/go-ethereum/metrics"
"github.com/syndtr/goleveldb/leveldb"
)
//define some metrics
@ -113,7 +112,7 @@ func NewAccounting(balance Balance, po Prices) *Accounting {
//this registry should be independent of any other metrics as it persists at different endpoints.
//It also instantiates the given metrics and starts the persisting go-routine which
//at the passed interval writes the metrics to a LevelDB
func SetupAccountingMetrics(reportInterval time.Duration, path string) *leveldb.DB {
func SetupAccountingMetrics(reportInterval time.Duration, path string) *AccountingMetrics {
//create an empty registry
registry := metrics.NewRegistry()
//instantiate the metrics
@ -126,7 +125,7 @@ func SetupAccountingMetrics(reportInterval time.Duration, path string) *leveldb.
mPeerDrops = metrics.NewRegisteredCounterForced("account.peerdrops", registry)
mSelfDrops = metrics.NewRegisteredCounterForced("account.selfdrops", registry)
//create the DB and start persisting
return NewMetricsDB(registry, reportInterval, path)
return NewAccountingMetrics(registry, reportInterval, path)
}
//Implement Hook.Send

View file

@ -20,7 +20,10 @@ import (
"context"
"flag"
"fmt"
"io/ioutil"
"math/rand"
"os"
"path/filepath"
"reflect"
"sync"
"testing"
@ -66,6 +69,13 @@ func init() {
func TestAccountingSimulation(t *testing.T) {
//setup the balances objects for every node
bal := newBalances(*nodes)
//setup the metrics system or tests will fail trying to write metrics
dir, err := ioutil.TempDir("", "account-sim")
if err != nil {
t.Fatal(err)
}
defer os.RemoveAll(dir)
SetupAccountingMetrics(1*time.Second, filepath.Join(dir, "metrics.db"))
//define the node.Service for this test
services := adapters.Services{
"accounting": func(ctx *adapters.ServiceContext) (node.Service, error) {

View file

@ -26,17 +26,32 @@ import (
"github.com/syndtr/goleveldb/leveldb"
)
//AccountMetrics abstracts away the metrics DB and
//the reporter to persist metrics
type AccountingMetrics struct {
metricsStore *leveldb.DB
reporter *reporter
}
//Close will be called when the node is being shutdown
//for a graceful cleanup
func (am *AccountingMetrics) Close() {
am.reporter.quit <- struct{}{}
am.metricsStore.Close()
}
//reporter is an internal structure used to write p2p accounting related
//metrics to a LevelDB. It will periodically write the accrued metrics to the DB.
type reporter struct {
reg metrics.Registry //the registry for these metrics (independent of other metrics)
interval time.Duration //duration at which the reporter will persist metrics
db *leveldb.DB //the actual DB
quit chan struct{} //quit the reporter loop
}
//NewMetricsDB creates a new LevelDB instance used to persist metrics defined
//inside p2p/protocols/accounting.go
func NewMetricsDB(r metrics.Registry, d time.Duration, path string) *leveldb.DB {
func NewAccountingMetrics(r metrics.Registry, d time.Duration, path string) *AccountingMetrics {
var val = make([]byte, 8)
var err error
@ -50,64 +65,64 @@ func NewMetricsDB(r metrics.Registry, d time.Duration, path string) *leveldb.DB
//Check for all defined metrics that there is a value in the DB
//If there is, assign it to the metric. This means that the node
//has been running before and that metrics have been persisted.
val, err = db.Get([]byte("account.balance.credit"), nil)
if err == nil {
mBalanceCredit.Inc(int64(binary.BigEndian.Uint64(val)))
metricsMap := map[string]metrics.Counter{
"account.balance.credit": mBalanceCredit,
"account.balance.debit": mBalanceDebit,
"account.bytes.credit": mBytesCredit,
"account.bytes.debit": mBytesDebit,
"account.msg.credit": mMsgCredit,
"account.msg.debit": mMsgDebit,
"account.peerdrops": mPeerDrops,
"account.selfdrops": mSelfDrops,
}
val, err = db.Get([]byte("account.balance.debit"), nil)
if err == nil {
mBalanceDebit.Inc(int64(binary.BigEndian.Uint64(val)))
}
val, err = db.Get([]byte("account.bytes.credit"), nil)
if err == nil {
mBytesCredit.Inc(int64(binary.BigEndian.Uint64(val)))
}
val, err = db.Get([]byte("account.bytes.debit"), nil)
if err == nil {
mBytesDebit.Inc(int64(binary.BigEndian.Uint64(val)))
}
val, err = db.Get([]byte("account.msg.credit"), nil)
if err == nil {
mMsgCredit.Inc(int64(binary.BigEndian.Uint64(val)))
}
val, err = db.Get([]byte("account.msg.debit"), nil)
if err == nil {
mMsgDebit.Inc(int64(binary.BigEndian.Uint64(val)))
}
val, err = db.Get([]byte("account.peerdrops"), nil)
if err == nil {
mPeerDrops.Inc(int64(binary.BigEndian.Uint64(val)))
}
val, err = db.Get([]byte("account.selfdrops"), nil)
if err == nil {
mSelfDrops.Inc(int64(binary.BigEndian.Uint64(val)))
//iterate the map and get the values
for key, metric := range metricsMap {
val, err = db.Get([]byte(key), nil)
//until the first time a value is being written,
//this will return an error.
//it could be beneficial though to log errors later,
//but that would require a different logic
if err == nil {
metric.Inc(int64(binary.BigEndian.Uint64(val)))
}
}
//create the reporter
reg := &reporter{
rep := &reporter{
reg: r,
interval: d,
db: db,
quit: make(chan struct{}),
}
//run the go routine
go reg.run()
go rep.run()
return db
m := &AccountingMetrics{
metricsStore: db,
reporter: rep,
}
return m
}
//run is the go routine which periodically sends the metrics to the configued LevelDB
func (r *reporter) run() {
intervalTicker := time.NewTicker(r.interval)
for _ = range intervalTicker.C {
//at each tick send the metrics
if err := r.send(); err != nil {
log.Error("unable to send metrics to LevelDB. err=%v", "err", err)
//If there is an error in writing, exit the routine; we assume here that the error is
//severe and don't attempt to write again.
//Also, this should prevent leaking when the node is stopped
for {
select {
case <-intervalTicker.C:
//at each tick send the metrics
if err := r.send(); err != nil {
log.Error("unable to send metrics to LevelDB", "err", err)
//If there is an error in writing, exit the routine; we assume here that the error is
//severe and don't attempt to write again.
//Also, this should prevent leaking when the node is stopped
return
}
case <-r.quit:
//graceful shutdown
return
}
}

View file

@ -39,8 +39,8 @@ func TestReporter(t *testing.T) {
//setup the metrics
log.Debug("Setting up metrics first time")
reportInterval := 100 * time.Millisecond
db := SetupAccountingMetrics(reportInterval, filepath.Join(dir, "test.db"))
reportInterval := 5 * time.Millisecond
metrics := SetupAccountingMetrics(reportInterval, filepath.Join(dir, "test.db"))
log.Debug("Done.")
//do some metrics
@ -49,18 +49,19 @@ func TestReporter(t *testing.T) {
mMsgDebit.Inc(9)
//give the reporter time to write the metrics to DB
time.Sleep(500 * time.Millisecond)
time.Sleep(20 * time.Millisecond)
//set the metrics to nil - this effectively simulates the node having shut down...
mBalanceCredit = nil
mBytesCredit = nil
mMsgDebit = nil
//close the DB also, or we can't create a new one
db.Close()
metrics.Close()
//setup the metrics again
log.Debug("Setting up metrics second time")
SetupAccountingMetrics(reportInterval, filepath.Join(dir, "test.db"))
metrics = SetupAccountingMetrics(reportInterval, filepath.Join(dir, "test.db"))
defer metrics.Close()
log.Debug("Done.")
//now check the metrics, they should have the same value as before "shutdown"

View file

@ -53,7 +53,6 @@ import (
"github.com/ethereum/go-ethereum/swarm/storage/mock"
"github.com/ethereum/go-ethereum/swarm/swap"
"github.com/ethereum/go-ethereum/swarm/tracing"
"github.com/syndtr/goleveldb/leveldb"
)
var (
@ -67,22 +66,22 @@ var (
// the swarm stack
type Swarm struct {
config *api.Config // swarm configuration
api *api.API // high level api layer (fs/manifest)
dns api.Resolver // DNS registrar
fileStore *storage.FileStore // distributed preimage archive, the local API to the storage with document level storage/retrieval support
streamer *stream.Registry
bzz *network.Bzz // the logistic manager
backend chequebook.Backend // simple blockchain Backend
privateKey *ecdsa.PrivateKey
corsString string
swapEnabled bool
netStore *storage.NetStore
sfs *fuse.SwarmFS // need this to cleanup all the active mounts on node exit
ps *pss.Pss
swap *swap.Swap
stateStore *state.DBStore
metricsStore *leveldb.DB
config *api.Config // swarm configuration
api *api.API // high level api layer (fs/manifest)
dns api.Resolver // DNS registrar
fileStore *storage.FileStore // distributed preimage archive, the local API to the storage with document level storage/retrieval support
streamer *stream.Registry
bzz *network.Bzz // the logistic manager
backend chequebook.Backend // simple blockchain Backend
privateKey *ecdsa.PrivateKey
corsString string
swapEnabled bool
netStore *storage.NetStore
sfs *fuse.SwarmFS // need this to cleanup all the active mounts on node exit
ps *pss.Pss
swap *swap.Swap
stateStore *state.DBStore
accountingMetrics *protocols.AccountingMetrics
tracerClose io.Closer
}
@ -182,7 +181,7 @@ func NewSwarm(config *api.Config, mockStore *mock.NodeStore) (self *Swarm, err e
return nil, err
}
self.swap = swap.New(balancesStore)
self.metricsStore = protocols.SetupAccountingMetrics(10*time.Second, filepath.Join(config.Path, "metrics.db"))
self.accountingMetrics = protocols.SetupAccountingMetrics(10*time.Second, filepath.Join(config.Path, "metrics.db"))
}
var nodeID enode.ID
@ -453,8 +452,8 @@ func (self *Swarm) Stop() error {
if self.swap != nil {
self.swap.Close()
}
if self.metricsStore != nil {
self.metricsStore.Close()
if self.accountingMetrics != nil {
self.accountingMetrics.Close()
}
if self.stateStore != nil {
self.stateStore.Close()