prometheus/tsdb/chunks/chunk_write_queue.go
Peter Štibraný ffc60d8397
Reduce chunk write queue memory usage 2 (#10874)
* Job queue

This PR reimplements chan chunkWriteJob with custom buffered queue that should use less memory, because it doesn't preallocate entire buffer for maximum queue size at once. Instead it allocates individual "segments" with smaller size.

As elements are added to the queue, they fill individual segments. When elements are removed from the queue (and segments), empty segments can be thrown away. This doesn't change memory usage of the queue when it's full, but should decrease its memory footprint when it's empty (queue will keep max 1 segment in such case).

Signed-off-by: Peter Štibraný <pstibrany@gmail.com>

* Modify test to work with low resolution timer.

Signed-off-by: Peter Štibraný <pstibrany@gmail.com>

* Improve comments.

Signed-off-by: Peter Štibraný <pstibrany@gmail.com>
2022-06-29 17:51:27 +05:30

264 lines
7.5 KiB
Go

// Copyright 2021 The Prometheus Authors
// 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 chunks
import (
"errors"
"sync"
"time"
"github.com/prometheus/client_golang/prometheus"
"github.com/prometheus/prometheus/tsdb/chunkenc"
)
const (
// Minimum recorded peak since the last shrinking of chunkWriteQueue.chunkrefMap to shrink it again.
chunkRefMapShrinkThreshold = 1000
// Minimum interval between shrinking of chunkWriteQueue.chunkRefMap.
chunkRefMapMinShrinkInterval = 10 * time.Minute
// Maximum size of segment used by job queue (number of elements). With chunkWriteJob being 64 bytes,
// this will use ~512 KiB for empty queue.
maxChunkQueueSegmentSize = 8192
)
type chunkWriteJob struct {
cutFile bool
seriesRef HeadSeriesRef
mint int64
maxt int64
chk chunkenc.Chunk
ref ChunkDiskMapperRef
callback func(error)
}
// chunkWriteQueue is a queue for writing chunks to disk in a non-blocking fashion.
// Chunks that shall be written get added to the queue, which is consumed asynchronously.
// Adding jobs to the queue is non-blocking as long as the queue isn't full.
type chunkWriteQueue struct {
jobs *writeJobQueue
chunkRefMapMtx sync.RWMutex
chunkRefMap map[ChunkDiskMapperRef]chunkenc.Chunk
chunkRefMapPeakSize int // Largest size that chunkRefMap has grown to since the last time we shrank it.
chunkRefMapLastShrink time.Time // When the chunkRefMap has been shrunk the last time.
// isRunningMtx serves two purposes:
// 1. It protects isRunning field.
// 2. It serializes adding of jobs to the chunkRefMap in addJob() method. If jobs channel is full then addJob() will block
// while holding this mutex, which guarantees that chunkRefMap won't ever grow beyond the queue size + 1.
isRunningMtx sync.Mutex
isRunning bool // Used to prevent that new jobs get added to the queue when the chan is already closed.
workerWg sync.WaitGroup
writeChunk writeChunkF
// Keeping separate counters instead of only a single CounterVec to improve the performance of the critical
// addJob() method which otherwise would need to perform a WithLabelValues call on the CounterVec.
adds prometheus.Counter
gets prometheus.Counter
completed prometheus.Counter
shrink prometheus.Counter
}
// writeChunkF is a function which writes chunks, it is dynamic to allow mocking in tests.
type writeChunkF func(HeadSeriesRef, int64, int64, chunkenc.Chunk, ChunkDiskMapperRef, bool) error
func newChunkWriteQueue(reg prometheus.Registerer, size int, writeChunk writeChunkF) *chunkWriteQueue {
counters := prometheus.NewCounterVec(
prometheus.CounterOpts{
Name: "prometheus_tsdb_chunk_write_queue_operations_total",
Help: "Number of operations on the chunk_write_queue.",
},
[]string{"operation"},
)
segmentSize := size
if segmentSize > maxChunkQueueSegmentSize {
segmentSize = maxChunkQueueSegmentSize
}
q := &chunkWriteQueue{
jobs: newWriteJobQueue(size, segmentSize),
chunkRefMap: make(map[ChunkDiskMapperRef]chunkenc.Chunk),
chunkRefMapLastShrink: time.Now(),
writeChunk: writeChunk,
adds: counters.WithLabelValues("add"),
gets: counters.WithLabelValues("get"),
completed: counters.WithLabelValues("complete"),
shrink: counters.WithLabelValues("shrink"),
}
if reg != nil {
reg.MustRegister(counters)
}
q.start()
return q
}
func (c *chunkWriteQueue) start() {
c.workerWg.Add(1)
go func() {
defer c.workerWg.Done()
for {
job, ok := c.jobs.pop()
if !ok {
return
}
c.processJob(job)
}
}()
c.isRunningMtx.Lock()
c.isRunning = true
c.isRunningMtx.Unlock()
}
func (c *chunkWriteQueue) processJob(job chunkWriteJob) {
err := c.writeChunk(job.seriesRef, job.mint, job.maxt, job.chk, job.ref, job.cutFile)
if job.callback != nil {
job.callback(err)
}
c.chunkRefMapMtx.Lock()
defer c.chunkRefMapMtx.Unlock()
delete(c.chunkRefMap, job.ref)
c.completed.Inc()
c.shrinkChunkRefMap()
}
// shrinkChunkRefMap checks whether the conditions to shrink the chunkRefMap are met,
// if so chunkRefMap is reinitialized. The chunkRefMapMtx must be held when calling this method.
//
// We do this because Go runtime doesn't release internal memory used by map after map has been emptied.
// To achieve that we create new map instead and throw the old one away.
func (c *chunkWriteQueue) shrinkChunkRefMap() {
if len(c.chunkRefMap) > 0 {
// Can't shrink it while there is data in it.
return
}
if c.chunkRefMapPeakSize < chunkRefMapShrinkThreshold {
// Not shrinking it because it has not grown to the minimum threshold yet.
return
}
now := time.Now()
if now.Sub(c.chunkRefMapLastShrink) < chunkRefMapMinShrinkInterval {
// Not shrinking it because the minimum duration between shrink-events has not passed yet.
return
}
// Re-initialize the chunk ref map to half of the peak size that it has grown to since the last re-init event.
// We are trying to hit the sweet spot in the trade-off between initializing it to a very small size
// potentially resulting in many allocations to re-grow it, and initializing it to a large size potentially
// resulting in unused allocated memory.
c.chunkRefMap = make(map[ChunkDiskMapperRef]chunkenc.Chunk, c.chunkRefMapPeakSize/2)
c.chunkRefMapPeakSize = 0
c.chunkRefMapLastShrink = now
c.shrink.Inc()
}
func (c *chunkWriteQueue) addJob(job chunkWriteJob) (err error) {
defer func() {
if err == nil {
c.adds.Inc()
}
}()
c.isRunningMtx.Lock()
defer c.isRunningMtx.Unlock()
if !c.isRunning {
return errors.New("queue is not running")
}
c.chunkRefMapMtx.Lock()
c.chunkRefMap[job.ref] = job.chk
// Keep track of the peak usage of c.chunkRefMap.
if len(c.chunkRefMap) > c.chunkRefMapPeakSize {
c.chunkRefMapPeakSize = len(c.chunkRefMap)
}
c.chunkRefMapMtx.Unlock()
if ok := c.jobs.push(job); !ok {
c.chunkRefMapMtx.Lock()
delete(c.chunkRefMap, job.ref)
c.chunkRefMapMtx.Unlock()
return errors.New("queue is closed")
}
return nil
}
func (c *chunkWriteQueue) get(ref ChunkDiskMapperRef) chunkenc.Chunk {
c.chunkRefMapMtx.RLock()
defer c.chunkRefMapMtx.RUnlock()
chk, ok := c.chunkRefMap[ref]
if ok {
c.gets.Inc()
}
return chk
}
func (c *chunkWriteQueue) stop() {
c.isRunningMtx.Lock()
defer c.isRunningMtx.Unlock()
if !c.isRunning {
return
}
c.isRunning = false
c.jobs.close()
c.workerWg.Wait()
}
func (c *chunkWriteQueue) queueIsEmpty() bool {
return c.queueSize() == 0
}
func (c *chunkWriteQueue) queueIsFull() bool {
// When the queue is full and blocked on the writer the chunkRefMap has one more job than the cap of the jobCh
// because one job is currently being processed and blocked in the writer.
return c.queueSize() == c.jobs.maxSize+1
}
func (c *chunkWriteQueue) queueSize() int {
c.chunkRefMapMtx.Lock()
defer c.chunkRefMapMtx.Unlock()
// Looking at chunkRefMap instead of jobCh because the job is popped from the chan before it has
// been fully processed, it remains in the chunkRefMap until the processing is complete.
return len(c.chunkRefMap)
}