Fix repairing run (#523)

* Fix repairing run
* Fix concurrency bugs
* Add sync2.Limiter concurrency primitive
This commit is contained in:
Egon Elbre 2018-10-24 15:35:59 +03:00 committed by GitHub
parent 842ebc9546
commit 8efb4f0e89
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
4 changed files with 159 additions and 74 deletions

50
internal/sync2/limiter.go Normal file
View File

@ -0,0 +1,50 @@
// Copyright (C) 2018 Storj Labs, Inc.
// See LICENSE for copying information
package sync2
import (
"context"
"sync"
)
// Limiter implements concurrent goroutine limiting
type Limiter struct {
limit chan struct{}
working sync.WaitGroup
}
// NewLimiter creates a new limiter with limit set to n
func NewLimiter(n int) *Limiter {
limiter := &Limiter{}
limiter.limit = make(chan struct{}, n)
return limiter
}
// Go tries to starts fn as a goroutine.
// When the limit is reached it will wait until it can run it
// or the context is canceled.
func (limiter *Limiter) Go(ctx context.Context, fn func()) bool {
select {
case limiter.limit <- struct{}{}:
case <-ctx.Done():
return false
}
limiter.working.Add(1)
go func() {
defer func() {
<-limiter.limit
limiter.working.Done()
}()
fn()
}()
return true
}
// Wait waits for all running goroutines to finish
func (limiter *Limiter) Wait() {
limiter.working.Wait()
}

View File

@ -0,0 +1,67 @@
// Copyright (C) 2018 Storj Labs, Inc.
// See LICENSE for copying information
package sync2
import (
"context"
"sync/atomic"
"testing"
"time"
)
func TestLimiterLimiting(t *testing.T) {
const N, Limit = 1000, 10
ctx := context.Background()
limiter := NewLimiter(Limit)
counter := int32(0)
for i := 0; i < N; i++ {
limiter.Go(ctx, func() {
if atomic.AddInt32(&counter, 1) > Limit {
panic("limit exceeded")
}
time.Sleep(time.Millisecond)
atomic.AddInt32(&counter, -1)
})
}
limiter.Wait()
}
func TestLimiterCancelling(t *testing.T) {
const N, Limit = 1000, 10
limiter := NewLimiter(Limit)
ctx, cancel := context.WithCancel(context.Background())
counter := int32(0)
waitForCancel := make(chan struct{}, N)
block := make(chan struct{})
allreturned := make(chan struct{})
go func() {
for i := 0; i < N; i++ {
limiter.Go(ctx, func() {
if atomic.AddInt32(&counter, 1) > Limit {
panic("limit exceeded")
}
waitForCancel <- struct{}{}
<-block
})
}
close(allreturned)
}()
for i := 0; i < Limit; i++ {
<-waitForCancel
}
cancel()
<-allreturned
close(block)
limiter.Wait()
if counter > Limit {
t.Fatal("too many times run")
}
}

View File

@ -19,29 +19,15 @@ type Config struct {
Interval time.Duration `help:"how frequently checker should audit segments" default:"3600s"`
}
// Initialize a repairer struct
func (c Config) initialize(ctx context.Context) (Repairer, error) {
var r repairer
r.ctx, r.cancel = context.WithCancel(ctx)
client, err := redis.NewClientFrom(c.QueueAddress)
if err != nil {
return nil, Error.Wrap(err)
}
r.queue = q.NewQueue(client)
r.cond.L = &r.mu
r.maxRepair = c.MaxRepair
r.interval = c.Interval
return &r, nil
}
// Run runs the repairer with configured values
func (c Config) Run(ctx context.Context, server *provider.Provider) (err error) {
r, err := c.initialize(ctx)
client, err := redis.NewClientFrom(c.QueueAddress)
if err != nil {
return err
return Error.Wrap(err)
}
return r.Run()
queue := q.NewQueue(client)
repairer := newRepairer(ctx, queue, c.Interval, c.MaxRepair)
return repairer.Run()
}

View File

@ -5,90 +5,72 @@ package repairer
import (
"context"
"fmt"
"sync"
"time"
"go.uber.org/zap"
"storj.io/storj/internal/sync2"
q "storj.io/storj/pkg/datarepair/queue"
"storj.io/storj/pkg/pb"
"storj.io/storj/pkg/utils"
)
// Repairer is the interface for the data repair queue
type Repairer interface {
Repair(seg *pb.InjuredSegment) error
Run() error
Stop() error
}
// repairer holds important values for data repair
type repairer struct {
ctx context.Context
cancel context.CancelFunc
queue q.RepairQueue
errs []error
mu sync.Mutex
cond sync.Cond
maxRepair int
inProgress int
interval time.Duration
limiter *sync2.Limiter
ticker *time.Ticker
}
func newRepairer(ctx context.Context, queue q.RepairQueue, interval time.Duration, concurrency int) *repairer {
return &repairer{
ctx: ctx,
queue: queue,
limiter: sync2.NewLimiter(concurrency),
ticker: time.NewTicker(interval),
}
}
// Run the repairer loop
func (r *repairer) Run() (err error) {
zap.S().Info("Repairer is starting up")
defer mon.Task()(&r.ctx)(&err)
c := make(chan *pb.InjuredSegment)
ticker := time.NewTicker(r.interval)
defer ticker.Stop()
go func() {
for range ticker.C {
for r.inProgress >= r.maxRepair {
r.cond.Wait()
}
// GetNext should lock until there is an actual next item in the queue
seg, err := r.queue.Dequeue()
if err != nil {
r.errs = append(r.errs, err)
r.cancel()
}
c <- &seg
}
}()
// wait for all repairs to complete
defer r.limiter.Wait()
for {
select {
case <-r.ctx.Done():
return utils.CombineErrors(r.errs...)
case seg := <-c:
go func() {
err := r.Repair(seg)
case <-r.ticker.C: // wait for the next interval to happen
case <-r.ctx.Done(): // or the repairer is canceled via context
return r.ctx.Err()
}
seg, err := r.queue.Dequeue()
if err != nil {
r.errs = append(r.errs, err)
r.cancel()
// TODO: only log when err != ErrQueueEmpty
zap.L().Error("dequeue", zap.Error(err))
continue
}
}()
r.limiter.Go(r.ctx, func() {
err := r.Repair(&seg)
if err != nil {
zap.L().Error("Repair failed", zap.Error(err))
}
})
}
}
// Repair starts repair of the segment
func (r *repairer) Repair(seg *pb.InjuredSegment) (err error) {
defer mon.Task()(&r.ctx)(&err)
r.inProgress++
fmt.Println(seg)
r.inProgress--
r.cond.Signal()
// TODO:
zap.L().Debug("Repairing", zap.Any("segment", seg))
return err
}
// Stop the repairer loop
func (r *repairer) Stop() (err error) {
r.cancel()
return nil
}