fix: avoid timer leaks in dsync/lsync (#9781)

At a customer setup with lots of concurrent calls
it can be observed that in newRetryTimer there
were lots of tiny alloations which are not
relinquished upon retries, in this codepath
we were only interested in re-using the timer
and use it wisely for each locker.

```
(pprof) top
Showing nodes accounting for 8.68TB, 97.02% of 8.95TB total
Dropped 1198 nodes (cum <= 0.04TB)
Showing top 10 nodes out of 79
      flat  flat%   sum%        cum   cum%
    5.95TB 66.50% 66.50%     5.95TB 66.50%  time.NewTimer
    1.16TB 13.02% 79.51%     1.16TB 13.02%  github.com/ncw/directio.AlignedBlock
    0.67TB  7.53% 87.04%     0.70TB  7.78%  github.com/minio/minio/cmd.xlObjects.putObject
    0.21TB  2.36% 89.40%     0.21TB  2.36%  github.com/minio/minio/cmd.(*posix).Walk
    0.19TB  2.08% 91.49%     0.27TB  2.99%  os.statNolog
    0.14TB  1.59% 93.08%     0.14TB  1.60%  os.(*File).readdirnames
    0.10TB  1.09% 94.17%     0.11TB  1.25%  github.com/minio/minio/cmd.readDirN
    0.10TB  1.07% 95.23%     0.10TB  1.07%  syscall.ByteSliceFromString
    0.09TB  1.03% 96.27%     0.09TB  1.03%  strings.(*Builder).grow
    0.07TB  0.75% 97.02%     0.07TB  0.75%  path.(*lazybuf).append
```
This commit is contained in:
Harshavardhana
2020-06-08 11:28:40 -07:00
committed by GitHub
parent 2ce2e88adf
commit febe9cc26a
16 changed files with 783 additions and 675 deletions
+24 -30
View File
@@ -24,6 +24,8 @@ import (
"os"
"sync"
"time"
"github.com/minio/minio/pkg/retry"
)
// Indicator if logging is enabled.
@@ -128,49 +130,41 @@ func (dm *DRWMutex) GetRLock(id, source string, timeout time.Duration) (locked b
// algorithm until either the lock is acquired successfully or more
// time has elapsed than the timeout value.
func (dm *DRWMutex) lockBlocking(timeout time.Duration, id, source string, isReadLock bool) (locked bool) {
start := time.Now().UTC()
restClnts := dm.clnt.GetLockersFn()
retryCtx, cancel := context.WithCancel(dm.ctx)
retryCtx, cancel := context.WithTimeout(dm.ctx, timeout)
defer cancel()
// Use incremental back-off algorithm for repeated attempts to acquire the lock
for range newRetryTimerSimple(retryCtx) {
select {
case <-dm.ctx.Done():
return
default:
}
for range retry.NewTimer(retryCtx) {
// Create temp array on stack.
locks := make([]string, len(restClnts))
// Try to acquire the lock.
success := lock(dm.clnt, &locks, id, source, isReadLock, dm.Names...)
if success {
dm.m.Lock()
// If success, copy array to object
if isReadLock {
// Append new array of strings at the end
dm.readersLocks = append(dm.readersLocks, make([]string, len(restClnts)))
// and copy stack array into last spot
copy(dm.readersLocks[len(dm.readersLocks)-1], locks[:])
} else {
copy(dm.writeLocks, locks[:])
}
dm.m.Unlock()
return true
if !success {
continue
}
if time.Now().UTC().Sub(start) >= timeout { // Are we past the timeout?
break
dm.m.Lock()
// If success, copy array to object
if isReadLock {
// Append new array of strings at the end
dm.readersLocks = append(dm.readersLocks, make([]string, len(restClnts)))
// and copy stack array into last spot
copy(dm.readersLocks[len(dm.readersLocks)-1], locks[:])
} else {
copy(dm.writeLocks, locks[:])
}
// Failed to acquire the lock on this attempt, incrementally wait
// for a longer back-off time and try again afterwards.
dm.m.Unlock()
return true
}
// Failed to acquire the lock on this attempt, incrementally wait
// for a longer back-off time and try again afterwards.
return false
}
@@ -233,7 +227,7 @@ func lock(ds *Dsync, locks *[]string, id, source string, isReadLock bool, lockNa
//
// a) received all lock responses
// b) received too many 'non-'locks for quorum to be still possible
// c) time out
// c) timedout
//
i, locksFailed := 0, 0
done := false
-136
View File
@@ -1,136 +0,0 @@
/*
* Minio Cloud Storage, (C) 2017 Minio, Inc.
*
* 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 dsync
import (
"context"
"math/rand"
"sync"
"time"
)
// lockedRandSource provides protected rand source, implements rand.Source interface.
type lockedRandSource struct {
lk sync.Mutex
src rand.Source
}
// Int63 returns a non-negative pseudo-random 63-bit integer as an
// int64.
func (r *lockedRandSource) Int63() (n int64) {
r.lk.Lock()
n = r.src.Int63()
r.lk.Unlock()
return
}
// Seed uses the provided seed value to initialize the generator to a
// deterministic state.
func (r *lockedRandSource) Seed(seed int64) {
r.lk.Lock()
r.src.Seed(seed)
r.lk.Unlock()
}
// MaxJitter will randomize over the full exponential backoff time
const MaxJitter = 1.0
// NoJitter disables the use of jitter for randomizing the
// exponential backoff time
const NoJitter = 0.0
// Global random source for fetching random values.
var globalRandomSource = rand.New(&lockedRandSource{
src: rand.NewSource(time.Now().UTC().UnixNano()),
})
// newRetryTimerJitter creates a timer with exponentially increasing delays
// until the maximum retry attempts are reached. - this function is a fully
// configurable version, meant for only advanced use cases. For the most part
// one should use newRetryTimerSimple and newRetryTimer.
func newRetryTimerWithJitter(ctx context.Context, unit time.Duration, cap time.Duration, jitter float64) <-chan int {
attemptCh := make(chan int)
// normalize jitter to the range [0, 1.0]
if jitter < NoJitter {
jitter = NoJitter
}
if jitter > MaxJitter {
jitter = MaxJitter
}
// computes the exponential backoff duration according to
// https://www.awsarchitectureblog.com/2015/03/backoff.html
exponentialBackoffWait := func(attempt int) time.Duration {
// 1<<uint(attempt) below could overflow, so limit the value of attempt
maxAttempt := 30
if attempt > maxAttempt {
attempt = maxAttempt
}
//sleep = random_between(0, min(cap, base * 2 ** attempt))
sleep := unit * time.Duration(1<<uint(attempt))
if sleep > cap {
sleep = cap
}
if jitter != NoJitter {
sleep -= time.Duration(globalRandomSource.Float64() * float64(sleep) * jitter)
}
return sleep
}
go func() {
defer close(attemptCh)
nextBackoff := 0
// Channel used to signal after the expiry of backoff wait seconds.
var timer *time.Timer
for {
select { // Attempts starts.
case attemptCh <- nextBackoff:
nextBackoff++
case <-ctx.Done():
// Stop the routine.
return
}
timer = time.NewTimer(exponentialBackoffWait(nextBackoff))
// wait till next backoff time or till doneCh gets a message.
select {
case <-timer.C:
case <-ctx.Done():
// stop the timer and return.
timer.Stop()
return
}
}
}()
// Start reading..
return attemptCh
}
// Default retry constants.
const (
defaultRetryUnit = time.Second // 1 second.
defaultRetryCap = 1 * time.Second // 1 second.
)
// newRetryTimerSimple creates a timer with exponentially increasing delays
// until the maximum retry attempts are reached. - this function is a
// simpler version with all default values.
func newRetryTimerSimple(ctx context.Context) <-chan int {
return newRetryTimerWithJitter(ctx, defaultRetryUnit, defaultRetryCap, MaxJitter)
}
-87
View File
@@ -1,87 +0,0 @@
/*
* Minio Cloud Storage, (C) 2017 Minio, Inc.
*
* 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 dsync
import (
"context"
"testing"
"time"
)
// Tests for retry timer.
func TestRetryTimerSimple(t *testing.T) {
rctx, cancel := context.WithCancel(context.Background())
attemptCh := newRetryTimerSimple(rctx)
i := <-attemptCh
if i != 0 {
cancel()
t.Fatalf("Invalid attempt counter returned should be 0, found %d instead", i)
}
i = <-attemptCh
if i <= 0 {
cancel()
t.Fatalf("Invalid attempt counter returned should be greater than 0, found %d instead", i)
}
cancel()
_, ok := <-attemptCh
if ok {
t.Fatal("Attempt counter should be closed")
}
}
// Test retry time with no jitter.
func TestRetryTimerWithNoJitter(t *testing.T) {
rctx, cancel := context.WithCancel(context.Background())
// No jitter
attemptCh := newRetryTimerWithJitter(rctx, time.Millisecond, 5*time.Millisecond, NoJitter)
i := <-attemptCh
if i != 0 {
cancel()
t.Fatalf("Invalid attempt counter returned should be 0, found %d instead", i)
}
// Loop through the maximum possible attempt.
for i = range attemptCh {
if i == 30 {
break
}
}
cancel()
_, ok := <-attemptCh
if ok {
t.Fatal("Attempt counter should be closed")
}
}
// Test retry time with Jitter greater than MaxJitter.
func TestRetryTimerWithJitter(t *testing.T) {
rctx, cancel := context.WithCancel(context.Background())
// Jitter will be set back to 1.0
attemptCh := newRetryTimerWithJitter(rctx, time.Second, 30*time.Second, 2.0)
i := <-attemptCh
if i != 0 {
cancel()
t.Fatalf("Invalid attempt counter returned should be 0, found %d instead", i)
}
cancel()
_, ok := <-attemptCh
if ok {
t.Fatal("Attempt counter should be closed")
}
}