Write chunks via queue, predicting the refs (#10051)

* Write chunks via queue, predicting the refs

Our load tests have shown that there is a latency spike in the
remote write handler whenever the head chunks need to be written,
because chunkDiskMapper.WriteChunk() blocks until the chunks are written
to disk.

This adds a queue to the chunk disk mapper which makes the WriteChunk()
method non-blocking unless the queue is full. Reads can still be served
from the queue.

Signed-off-by: Mauro Stettler <mauro.stettler@gmail.com>

* address PR feeddback

Signed-off-by: Mauro Stettler <mauro.stettler@gmail.com>

* initialize metrics without .Add(0)

Signed-off-by: Mauro Stettler <mauro.stettler@gmail.com>

* change isRunningMtx to normal lock

Signed-off-by: Mauro Stettler <mauro.stettler@gmail.com>

* do not re-initialize chunkrefmap

Signed-off-by: Mauro Stettler <mauro.stettler@gmail.com>

* update metric outside of lock scope

Signed-off-by: Mauro Stettler <mauro.stettler@gmail.com>

* add benchmark for adding job to chunk write queue

Signed-off-by: Mauro Stettler <mauro.stettler@gmail.com>

* remove unnecessary "success" var

Signed-off-by: Mauro Stettler <mauro.stettler@gmail.com>

* gofumpt -extra

Signed-off-by: Mauro Stettler <mauro.stettler@gmail.com>

* avoid WithLabelValues call in addJob

Signed-off-by: Mauro Stettler <mauro.stettler@gmail.com>

* format comments

Signed-off-by: Mauro Stettler <mauro.stettler@gmail.com>

* addressing PR feedback

Signed-off-by: Mauro Stettler <mauro.stettler@gmail.com>

* rename cutExpectRef to cutAndExpectRef

Signed-off-by: Mauro Stettler <mauro.stettler@gmail.com>

* use head.Init() instead of .initTime()

Signed-off-by: Mauro Stettler <mauro.stettler@gmail.com>

* address PR feedback

Signed-off-by: Mauro Stettler <mauro.stettler@gmail.com>

* PR feedback

Co-authored-by: Ganesh Vernekar <15064823+codesome@users.noreply.github.com>
Signed-off-by: Mauro Stettler <mauro.stettler@gmail.com>

* update test according to PR feedback

Signed-off-by: Mauro Stettler <mauro.stettler@gmail.com>

* replace callbackWg -> awaitCb

Signed-off-by: Mauro Stettler <mauro.stettler@gmail.com>

* better test of truncation with empty files

Signed-off-by: Mauro Stettler <mauro.stettler@gmail.com>

* replace callbackWg -> awaitCb

Signed-off-by: Mauro Stettler <mauro.stettler@gmail.com>

Co-authored-by: Ganesh Vernekar <15064823+codesome@users.noreply.github.com>

tsdb/chunks/chunk_write_queue.go

@@ -0,0 +1,165 @@
+// 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"
+
+	"github.com/prometheus/client_golang/prometheus"
+
+	"github.com/prometheus/prometheus/tsdb/chunkenc"
+)
+
+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 job is non-blocking as long as the queue isn't full.
+type chunkWriteQueue struct {
+	jobs chan chunkWriteJob
+
+	chunkRefMapMtx sync.RWMutex
+	chunkRefMap    map[ChunkDiskMapperRef]chunkenc.Chunk
+
+	isRunningMtx sync.Mutex // Protects the isRunning property.
+	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 three 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
+}
+
+// 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"},
+	)
+
+	q := &chunkWriteQueue{
+		jobs:        make(chan chunkWriteJob, size),
+		chunkRefMap: make(map[ChunkDiskMapperRef]chunkenc.Chunk, size),
+		writeChunk:  writeChunk,
+
+		adds:      counters.WithLabelValues("add"),
+		gets:      counters.WithLabelValues("get"),
+		completed: counters.WithLabelValues("complete"),
+	}
+
+	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 := range c.jobs {
+			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()
+}
+
+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 started")
+	}
+
+	c.chunkRefMapMtx.Lock()
+	c.chunkRefMap[job.ref] = job.chk
+	c.chunkRefMapMtx.Unlock()
+
+	c.jobs <- job
+
+	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
+
+	close(c.jobs)
+
+	c.workerWg.Wait()
+}

tsdb/chunks/chunk_write_queue_test.go

@@ -0,0 +1,287 @@
+// 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"
+	"fmt"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/require"
+	"go.uber.org/atomic"
+
+	"github.com/prometheus/prometheus/tsdb/chunkenc"
+)
+
+func TestChunkWriteQueue_GettingChunkFromQueue(t *testing.T) {
+	var blockWriterWg sync.WaitGroup
+	blockWriterWg.Add(1)
+
+	// blockingChunkWriter blocks until blockWriterWg is done.
+	blockingChunkWriter := func(_ HeadSeriesRef, _, _ int64, _ chunkenc.Chunk, _ ChunkDiskMapperRef, _ bool) error {
+		blockWriterWg.Wait()
+		return nil
+	}
+
+	q := newChunkWriteQueue(nil, 1000, blockingChunkWriter)
+
+	defer q.stop()
+	defer blockWriterWg.Done()
+
+	testChunk := chunkenc.NewXORChunk()
+	var ref ChunkDiskMapperRef
+	job := chunkWriteJob{
+		chk: testChunk,
+		ref: ref,
+	}
+	require.NoError(t, q.addJob(job))
+
+	// Retrieve chunk from the queue.
+	gotChunk := q.get(ref)
+	require.Equal(t, testChunk, gotChunk)
+}
+
+func TestChunkWriteQueue_WritingThroughQueue(t *testing.T) {
+	var (
+		gotSeriesRef     HeadSeriesRef
+		gotMint, gotMaxt int64
+		gotChunk         chunkenc.Chunk
+		gotRef           ChunkDiskMapperRef
+		gotCutFile       bool
+	)
+
+	blockingChunkWriter := func(seriesRef HeadSeriesRef, mint, maxt int64, chunk chunkenc.Chunk, ref ChunkDiskMapperRef, cutFile bool) error {
+		gotSeriesRef = seriesRef
+		gotMint = mint
+		gotMaxt = maxt
+		gotChunk = chunk
+		gotRef = ref
+		gotCutFile = cutFile
+		return nil
+	}
+
+	q := newChunkWriteQueue(nil, 1000, blockingChunkWriter)
+	defer q.stop()
+
+	seriesRef := HeadSeriesRef(1)
+	var mint, maxt int64 = 2, 3
+	chunk := chunkenc.NewXORChunk()
+	ref := newChunkDiskMapperRef(321, 123)
+	cutFile := true
+	awaitCb := make(chan struct{})
+	require.NoError(t, q.addJob(chunkWriteJob{seriesRef: seriesRef, mint: mint, maxt: maxt, chk: chunk, ref: ref, cutFile: cutFile, callback: func(err error) {
+		close(awaitCb)
+	}}))
+	<-awaitCb
+
+	// Compare whether the write function has received all job attributes correctly.
+	require.Equal(t, seriesRef, gotSeriesRef)
+	require.Equal(t, mint, gotMint)
+	require.Equal(t, maxt, gotMaxt)
+	require.Equal(t, chunk, gotChunk)
+	require.Equal(t, ref, gotRef)
+	require.Equal(t, cutFile, gotCutFile)
+}
+
+func TestChunkWriteQueue_WrappingAroundSizeLimit(t *testing.T) {
+	sizeLimit := 100
+	unblockChunkWriterCh := make(chan struct{}, sizeLimit)
+
+	// blockingChunkWriter blocks until the unblockChunkWriterCh channel returns a value.
+	blockingChunkWriter := func(seriesRef HeadSeriesRef, mint, maxt int64, chunk chunkenc.Chunk, ref ChunkDiskMapperRef, cutFile bool) error {
+		<-unblockChunkWriterCh
+		return nil
+	}
+
+	q := newChunkWriteQueue(nil, sizeLimit, blockingChunkWriter)
+	defer q.stop()
+	// Unblock writers when shutting down.
+	defer close(unblockChunkWriterCh)
+
+	var chunkRef ChunkDiskMapperRef
+	var callbackWg sync.WaitGroup
+	addChunk := func() {
+		callbackWg.Add(1)
+		require.NoError(t, q.addJob(chunkWriteJob{
+			ref: chunkRef,
+			callback: func(err error) {
+				callbackWg.Done()
+			},
+		}))
+		chunkRef++
+	}
+
+	unblockChunkWriter := func() {
+		unblockChunkWriterCh <- struct{}{}
+	}
+
+	// Fill the queue to the middle of the size limit.
+	for job := 0; job < sizeLimit/2; job++ {
+		addChunk()
+	}
+
+	// Consume the jobs.
+	for job := 0; job < sizeLimit/2; job++ {
+		unblockChunkWriter()
+	}
+
+	// Add jobs until the queue is full.
+	// Note that one more job than <sizeLimit> can be added because one will be processed by the worker already
+	// and it will block on the chunk write function.
+	for job := 0; job < sizeLimit+1; job++ {
+		addChunk()
+	}
+
+	// The queue should be full.
+	require.True(t, queueIsFull(q))
+
+	// Adding another job should block as long as no job from the queue gets consumed.
+	addedJob := atomic.NewBool(false)
+	go func() {
+		addChunk()
+		addedJob.Store(true)
+	}()
+
+	// Wait for 10ms while the adding of a new job is blocked.
+	time.Sleep(time.Millisecond * 10)
+	require.False(t, addedJob.Load())
+
+	// Consume one job from the queue.
+	unblockChunkWriter()
+
+	// Wait until the job has been added to the queue.
+	require.Eventually(t, func() bool { return addedJob.Load() }, time.Second, time.Millisecond*10)
+
+	// The queue should be full again.
+	require.True(t, queueIsFull(q))
+
+	// Consume <sizeLimit>+1 jobs from the queue.
+	// To drain the queue we need to consume <sizeLimit>+1 jobs because 1 job
+	// is already in the state of being processed.
+	for job := 0; job < sizeLimit+1; job++ {
+		require.False(t, queueIsEmpty(q))
+		unblockChunkWriter()
+	}
+
+	// Wait until all jobs have been processed.
+	callbackWg.Wait()
+	require.True(t, queueIsEmpty(q))
+}
+
+func TestChunkWriteQueue_HandlerErrorViaCallback(t *testing.T) {
+	testError := errors.New("test error")
+	chunkWriter := func(_ HeadSeriesRef, _, _ int64, _ chunkenc.Chunk, _ ChunkDiskMapperRef, _ bool) error {
+		return testError
+	}
+
+	awaitCb := make(chan struct{})
+	var gotError error
+	callback := func(err error) {
+		gotError = err
+		close(awaitCb)
+	}
+
+	q := newChunkWriteQueue(nil, 1, chunkWriter)
+	defer q.stop()
+
+	job := chunkWriteJob{callback: callback}
+	require.NoError(t, q.addJob(job))
+
+	<-awaitCb
+
+	require.Equal(t, testError, gotError)
+}
+
+func BenchmarkChunkWriteQueue_addJob(b *testing.B) {
+	for _, withReads := range []bool{false, true} {
+		b.Run(fmt.Sprintf("with reads %t", withReads), func(b *testing.B) {
+			for _, concurrentWrites := range []int{1, 10, 100, 1000} {
+				b.Run(fmt.Sprintf("%d concurrent writes", concurrentWrites), func(b *testing.B) {
+					issueReadSignal := make(chan struct{})
+					q := newChunkWriteQueue(nil, 1000, func(ref HeadSeriesRef, i, i2 int64, chunk chunkenc.Chunk, ref2 ChunkDiskMapperRef, b bool) error {
+						if withReads {
+							select {
+							case issueReadSignal <- struct{}{}:
+							default:
+								// Can't write to issueReadSignal, don't block but omit read instead.
+							}
+						}
+						return nil
+					})
+					b.Cleanup(func() {
+						// Stopped already, so no more writes will happen.
+						close(issueReadSignal)
+					})
+					b.Cleanup(q.stop)
+
+					start := sync.WaitGroup{}
+					start.Add(1)
+
+					jobs := make(chan chunkWriteJob, b.N)
+					for i := 0; i < b.N; i++ {
+						jobs <- chunkWriteJob{
+							seriesRef: HeadSeriesRef(i),
+							ref:       ChunkDiskMapperRef(i),
+						}
+					}
+					close(jobs)
+
+					go func() {
+						for range issueReadSignal {
+							// We don't care about the ID we're getting, we just want to grab the lock.
+							_ = q.get(ChunkDiskMapperRef(0))
+						}
+					}()
+
+					done := sync.WaitGroup{}
+					done.Add(concurrentWrites)
+					for w := 0; w < concurrentWrites; w++ {
+						go func() {
+							start.Wait()
+							for j := range jobs {
+								_ = q.addJob(j)
+							}
+							done.Done()
+						}()
+					}
+
+					b.ResetTimer()
+					start.Done()
+					done.Wait()
+				})
+			}
+		})
+	}
+}
+
+func queueIsEmpty(q *chunkWriteQueue) bool {
+	return queueSize(q) == 0
+}
+
+func queueIsFull(q *chunkWriteQueue) 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 queueSize(q) == cap(q.jobs)+1
+}
+
+func queueSize(q *chunkWriteQueue) int {
+	q.chunkRefMapMtx.Lock()
+	defer q.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(q.chunkRefMap)
+}

tsdb/chunks/head_chunks.go

@@ -26,7 +26,9 @@ import (
 	"strconv"
 	"sync"
 
+	"github.com/dennwc/varint"
 	"github.com/pkg/errors"
+	"github.com/prometheus/client_golang/prometheus"
 	"go.uber.org/atomic"
 
 	"github.com/prometheus/prometheus/tsdb/chunkenc"
@@ -66,6 +68,8 @@ const (
 	MaxWriteBufferSize = 8 * 1024 * 1024 // 8 MiB.
 	// DefaultWriteBufferSize is the default write buffer size.
 	DefaultWriteBufferSize = 4 * 1024 * 1024 // 4 MiB.
+	// DefaultWriteQueueSize is the default size of the in-memory queue used before flushing chunks to the disk.
+	DefaultWriteQueueSize = 1000
 )
 
 // ChunkDiskMapperRef represents the location of a head chunk on disk.
@@ -77,10 +81,10 @@ func newChunkDiskMapperRef(seq, offset uint64) ChunkDiskMapperRef {
 	return ChunkDiskMapperRef((seq << 32) | offset)
 }
 
-func (ref ChunkDiskMapperRef) Unpack() (sgmIndex, chkStart int) {
-	sgmIndex = int(ref >> 32)
-	chkStart = int((ref << 32) >> 32)
-	return sgmIndex, chkStart
+func (ref ChunkDiskMapperRef) Unpack() (seq, offset int) {
+	seq = int(ref >> 32)
+	offset = int((ref << 32) >> 32)
+	return seq, offset
 }
 
 // CorruptionErr is an error that's returned when corruption is encountered.
@@ -94,18 +98,97 @@ func (e *CorruptionErr) Error() string {
 	return errors.Wrapf(e.Err, "corruption in head chunk file %s", segmentFile(e.Dir, e.FileIndex)).Error()
 }
 
+// chunkPos keeps track of the position in the head chunk files.
+// chunkPos is not thread-safe, a lock must be used to protect it.
+type chunkPos struct {
+	seq     uint64 // Index of chunk file.
+	offset  uint64 // Offset within chunk file.
+	cutFile bool   // When true then the next chunk will be written to a new file.
+}
+
+// getNextChunkRef takes a chunk and returns the chunk reference which will refer to it once it has been written.
+// getNextChunkRef also decides whether a new file should be cut before writing this chunk, and it returns the decision via the second return value.
+// The order of calling getNextChunkRef must be the order in which chunks are written to the disk.
+func (f *chunkPos) getNextChunkRef(chk chunkenc.Chunk) (chkRef ChunkDiskMapperRef, cutFile bool) {
+	chkLen := uint64(len(chk.Bytes()))
+	bytesToWrite := f.bytesToWriteForChunk(chkLen)
+
+	if f.shouldCutNewFile(chkLen) {
+		f.toNewFile()
+		f.cutFile = false
+		cutFile = true
+	}
+
+	chkOffset := f.offset
+	f.offset += bytesToWrite
+
+	return newChunkDiskMapperRef(f.seq, chkOffset), cutFile
+}
+
+// toNewFile updates the seq/offset position to point to the beginning of a new chunk file.
+func (f *chunkPos) toNewFile() {
+	f.seq++
+	f.offset = SegmentHeaderSize
+}
+
+// cutFileOnNextChunk triggers that the next chunk will be written in to a new file.
+// Not thread safe, a lock must be held when calling this.
+func (f *chunkPos) cutFileOnNextChunk() {
+	f.cutFile = true
+}
+
+// initSeq sets the sequence number of the head chunk file.
+// Should only be used for initialization, after that the sequence number will be managed by chunkPos.
+func (f *chunkPos) initSeq(seq uint64) {
+	f.seq = seq
+}
+
+// shouldCutNewFile returns whether a new file should be cut based on the file size.
+// The read or write lock on chunkPos must be held when calling this.
+func (f *chunkPos) shouldCutNewFile(chunkSize uint64) bool {
+	if f.cutFile {
+		return true
+	}
+
+	return f.offset == 0 || // First head chunk file.
+		f.offset+chunkSize+MaxHeadChunkMetaSize > MaxHeadChunkFileSize // Exceeds the max head chunk file size.
+}
+
+// bytesToWriteForChunk returns the number of bytes that will need to be written for the given chunk size,
+// including all meta data before and after the chunk data.
+// Head chunk format: https://github.com/prometheus/prometheus/blob/main/tsdb/docs/format/head_chunks.md#chunk
+func (f *chunkPos) bytesToWriteForChunk(chkLen uint64) uint64 {
+	// Headers.
+	bytes := uint64(SeriesRefSize) + 2*MintMaxtSize + ChunkEncodingSize
+
+	// Size of chunk length encoded as uvarint.
+	bytes += uint64(varint.UvarintSize(chkLen))
+
+	// Chunk length.
+	bytes += chkLen
+
+	// crc32.
+	bytes += CRCSize
+
+	return bytes
+}
+
 // ChunkDiskMapper is for writing the Head block chunks to the disk
 // and access chunks via mmapped file.
 type ChunkDiskMapper struct {
-	curFileNumBytes atomic.Int64 // Bytes written in current open file.
-
 	/// Writer.
 	dir             *os.File
 	writeBufferSize int
 
-	curFile         *os.File // File being written to.
-	curFileSequence int      // Index of current open file being appended to.
-	curFileMaxt     int64    // Used for the size retention.
+	curFile         *os.File      // File being written to.
+	curFileSequence int           // Index of current open file being appended to.
+	curFileOffset   atomic.Uint64 // Bytes written in current open file.
+	curFileMaxt     int64         // Used for the size retention.
+
+	// The values in evtlPos represent the file position which will eventually be
+	// reached once the content of the write queue has been fully processed.
+	evtlPosMtx sync.Mutex
+	evtlPos    chunkPos
 
 	byteBuf      [MaxHeadChunkMetaSize]byte // Buffer used to write the header of the chunk.
 	chkWriter    *bufio.Writer              // Writer for the current open file.
@@ -128,6 +211,8 @@ type ChunkDiskMapper struct {
 	// This is done after iterating through all the chunks in those files using the IterateAllChunks method.
 	fileMaxtSet bool
 
+	writeQueue *chunkWriteQueue
+
 	closed bool
 }
 
@@ -141,7 +226,7 @@ type mmappedChunkFile struct {
 // using the default head chunk file duration.
 // NOTE: 'IterateAllChunks' method needs to be called at least once after creating ChunkDiskMapper
 // to set the maxt of all the file.
-func NewChunkDiskMapper(dir string, pool chunkenc.Pool, writeBufferSize int) (*ChunkDiskMapper, error) {
+func NewChunkDiskMapper(reg prometheus.Registerer, dir string, pool chunkenc.Pool, writeBufferSize, writeQueueSize int) (*ChunkDiskMapper, error) {
 	// Validate write buffer size.
 	if writeBufferSize < MinWriteBufferSize || writeBufferSize > MaxWriteBufferSize {
 		return nil, errors.Errorf("ChunkDiskMapper write buffer size should be between %d and %d (actual: %d)", MinWriteBufferSize, MaxWriteBufferSize, writeBufferSize)
@@ -165,6 +250,7 @@ func NewChunkDiskMapper(dir string, pool chunkenc.Pool, writeBufferSize int) (*C
 		crc32:           newCRC32(),
 		chunkBuffer:     newChunkBuffer(),
 	}
+	m.writeQueue = newChunkWriteQueue(reg, writeQueueSize, m.writeChunk)
 
 	if m.pool == nil {
 		m.pool = chunkenc.NewPool()
@@ -235,6 +321,8 @@ func (cdm *ChunkDiskMapper) openMMapFiles() (returnErr error) {
 		}
 	}
 
+	cdm.evtlPos.initSeq(uint64(lastSeq))
+
 	return nil
 }
 
@@ -287,17 +375,44 @@ func repairLastChunkFile(files map[int]string) (_ map[int]string, returnErr erro
 
 // WriteChunk writes the chunk to the disk.
 // The returned chunk ref is the reference from where the chunk encoding starts for the chunk.
-func (cdm *ChunkDiskMapper) WriteChunk(seriesRef HeadSeriesRef, mint, maxt int64, chk chunkenc.Chunk) (chkRef ChunkDiskMapperRef, err error) {
+func (cdm *ChunkDiskMapper) WriteChunk(seriesRef HeadSeriesRef, mint, maxt int64, chk chunkenc.Chunk, callback func(err error)) (chkRef ChunkDiskMapperRef) {
+	var err error
+	defer func() {
+		if err != nil && callback != nil {
+			callback(err)
+		}
+	}()
+
+	// cdm.evtlPosMtx must be held to serialize the calls to .getNextChunkRef() and .addJob().
+	cdm.evtlPosMtx.Lock()
+	defer cdm.evtlPosMtx.Unlock()
+
+	ref, cutFile := cdm.evtlPos.getNextChunkRef(chk)
+	err = cdm.writeQueue.addJob(chunkWriteJob{
+		cutFile:   cutFile,
+		seriesRef: seriesRef,
+		mint:      mint,
+		maxt:      maxt,
+		chk:       chk,
+		ref:       ref,
+		callback:  callback,
+	})
+
+	return ref
+}
+
+func (cdm *ChunkDiskMapper) writeChunk(seriesRef HeadSeriesRef, mint, maxt int64, chk chunkenc.Chunk, ref ChunkDiskMapperRef, cutFile bool) (err error) {
 	cdm.writePathMtx.Lock()
 	defer cdm.writePathMtx.Unlock()
 
 	if cdm.closed {
-		return 0, ErrChunkDiskMapperClosed
+		return ErrChunkDiskMapperClosed
 	}
 
-	if cdm.shouldCutNewFile(len(chk.Bytes())) {
-		if err := cdm.cut(); err != nil {
-			return 0, err
+	if cutFile {
+		err := cdm.cutAndExpectRef(ref)
+		if err != nil {
+			return err
 		}
 	}
 
@@ -305,15 +420,13 @@ func (cdm *ChunkDiskMapper) WriteChunk(seriesRef HeadSeriesRef, mint, maxt int64
 	// so no need to flush here, as we have to flush at the end (to not keep partial chunks in buffer).
 	if len(chk.Bytes())+MaxHeadChunkMetaSize < cdm.writeBufferSize && cdm.chkWriter.Available() < MaxHeadChunkMetaSize+len(chk.Bytes()) {
 		if err := cdm.flushBuffer(); err != nil {
-			return 0, err
+			return err
 		}
 	}
 
 	cdm.crc32.Reset()
 	bytesWritten := 0
 
-	chkRef = newChunkDiskMapperRef(uint64(cdm.curFileSequence), uint64(cdm.curFileSize()))
-
 	binary.BigEndian.PutUint64(cdm.byteBuf[bytesWritten:], uint64(seriesRef))
 	bytesWritten += SeriesRefSize
 	binary.BigEndian.PutUint64(cdm.byteBuf[bytesWritten:], uint64(mint))
@@ -326,59 +439,69 @@ func (cdm *ChunkDiskMapper) WriteChunk(seriesRef HeadSeriesRef, mint, maxt int64
 	bytesWritten += n
 
 	if err := cdm.writeAndAppendToCRC32(cdm.byteBuf[:bytesWritten]); err != nil {
-		return 0, err
+		return err
 	}
 	if err := cdm.writeAndAppendToCRC32(chk.Bytes()); err != nil {
-		return 0, err
+		return err
 	}
 	if err := cdm.writeCRC32(); err != nil {
-		return 0, err
+		return err
 	}
 
 	if maxt > cdm.curFileMaxt {
 		cdm.curFileMaxt = maxt
 	}
 
-	cdm.chunkBuffer.put(chkRef, chk)
+	cdm.chunkBuffer.put(ref, chk)
 
 	if len(chk.Bytes())+MaxHeadChunkMetaSize >= cdm.writeBufferSize {
 		// The chunk was bigger than the buffer itself.
 		// Flushing to not keep partial chunks in buffer.
 		if err := cdm.flushBuffer(); err != nil {
-			return 0, err
+			return err
 		}
 	}
 
-	return chkRef, nil
+	return nil
 }
 
-// shouldCutNewFile returns whether a new file should be cut, based on time and size retention.
-// Size retention: because depending on the system architecture, there is a limit on how big of a file we can m-map.
-// Time retention: so that we can delete old chunks with some time guarantee in low load environments.
-func (cdm *ChunkDiskMapper) shouldCutNewFile(chunkSize int) bool {
-	return cdm.curFileSize() == 0 || // First head chunk file.
-		cdm.curFileSize()+int64(chunkSize+MaxHeadChunkMetaSize) > MaxHeadChunkFileSize // Exceeds the max head chunk file size.
+// CutNewFile makes that a new file will be created the next time a chunk is written.
+func (cdm *ChunkDiskMapper) CutNewFile() {
+	cdm.evtlPosMtx.Lock()
+	defer cdm.evtlPosMtx.Unlock()
+
+	cdm.evtlPos.cutFileOnNextChunk()
 }
 
-// CutNewFile creates a new m-mapped file.
-func (cdm *ChunkDiskMapper) CutNewFile() (returnErr error) {
-	cdm.writePathMtx.Lock()
-	defer cdm.writePathMtx.Unlock()
-
-	return cdm.cut()
-}
-
-// cut creates a new m-mapped file. The write lock should be held before calling this.
-func (cdm *ChunkDiskMapper) cut() (returnErr error) {
-	// Sync current tail to disk and close.
-	if err := cdm.finalizeCurFile(); err != nil {
-		return err
-	}
-
-	n, newFile, seq, err := cutSegmentFile(cdm.dir, MagicHeadChunks, headChunksFormatV1, HeadChunkFilePreallocationSize)
+// cutAndExpectRef creates a new m-mapped file.
+// The write lock should be held before calling this.
+// It ensures that the position in the new file matches the given chunk reference, if not then it errors.
+func (cdm *ChunkDiskMapper) cutAndExpectRef(chkRef ChunkDiskMapperRef) (err error) {
+	seq, offset, err := cdm.cut()
 	if err != nil {
 		return err
 	}
+
+	if expSeq, expOffset := chkRef.Unpack(); seq != expSeq || offset != expOffset {
+		return errors.Errorf("expected newly cut file to have sequence:offset %d:%d, got %d:%d", expSeq, expOffset, seq, offset)
+	}
+
+	return nil
+}
+
+// cut creates a new m-mapped file. The write lock should be held before calling this.
+// It returns the file sequence and the offset in that file to start writing chunks.
+func (cdm *ChunkDiskMapper) cut() (seq, offset int, returnErr error) {
+	// Sync current tail to disk and close.
+	if err := cdm.finalizeCurFile(); err != nil {
+		return 0, 0, err
+	}
+
+	offset, newFile, seq, err := cutSegmentFile(cdm.dir, MagicHeadChunks, headChunksFormatV1, HeadChunkFilePreallocationSize)
+	if err != nil {
+		return 0, 0, err
+	}
+
 	defer func() {
 		// The file should not be closed if there is no error,
 		// its kept open in the ChunkDiskMapper.
@@ -387,7 +510,7 @@ func (cdm *ChunkDiskMapper) cut() (returnErr error) {
 		}
 	}()
 
-	cdm.curFileNumBytes.Store(int64(n))
+	cdm.curFileOffset.Store(uint64(offset))
 
 	if cdm.curFile != nil {
 		cdm.readPathMtx.Lock()
@@ -397,7 +520,7 @@ func (cdm *ChunkDiskMapper) cut() (returnErr error) {
 
 	mmapFile, err := fileutil.OpenMmapFileWithSize(newFile.Name(), MaxHeadChunkFileSize)
 	if err != nil {
-		return err
+		return 0, 0, err
 	}
 
 	cdm.readPathMtx.Lock()
@@ -415,7 +538,7 @@ func (cdm *ChunkDiskMapper) cut() (returnErr error) {
 
 	cdm.curFileMaxt = 0
 
-	return nil
+	return seq, offset, nil
 }
 
 // finalizeCurFile writes all pending data to the current tail file,
@@ -438,7 +561,7 @@ func (cdm *ChunkDiskMapper) finalizeCurFile() error {
 
 func (cdm *ChunkDiskMapper) write(b []byte) error {
 	n, err := cdm.chkWriter.Write(b)
-	cdm.curFileNumBytes.Add(int64(n))
+	cdm.curFileOffset.Add(uint64(n))
 	return err
 }
 
@@ -476,6 +599,11 @@ func (cdm *ChunkDiskMapper) Chunk(ref ChunkDiskMapperRef) (chunkenc.Chunk, error
 		return nil, ErrChunkDiskMapperClosed
 	}
 
+	chunk := cdm.writeQueue.get(ref)
+	if chunk != nil {
+		return chunk, nil
+	}
+
 	sgmIndex, chkStart := ref.Unpack()
 	// We skip the series ref and the mint/maxt beforehand.
 	chkStart += SeriesRefSize + (2 * MintMaxtSize)
@@ -732,7 +860,10 @@ func (cdm *ChunkDiskMapper) Truncate(mint int64) error {
 	errs := tsdb_errors.NewMulti()
 	// Cut a new file only if the current file has some chunks.
 	if cdm.curFileSize() > HeadChunkFileHeaderSize {
-		errs.Add(cdm.CutNewFile())
+		// There is a known race condition here because between the check of curFileSize() and the call to CutNewFile()
+		// a new file could already be cut, this is acceptable because it will simply result in an empty file which
+		// won't do any harm.
+		cdm.CutNewFile()
 	}
 	errs.Add(cdm.deleteFiles(removedFiles))
 	return errs.Err()
@@ -787,13 +918,19 @@ func (cdm *ChunkDiskMapper) Size() (int64, error) {
 	return fileutil.DirSize(cdm.dir.Name())
 }
 
-func (cdm *ChunkDiskMapper) curFileSize() int64 {
-	return cdm.curFileNumBytes.Load()
+func (cdm *ChunkDiskMapper) curFileSize() uint64 {
+	return cdm.curFileOffset.Load()
 }
 
 // Close closes all the open files in ChunkDiskMapper.
 // It is not longer safe to access chunks from this struct after calling Close.
 func (cdm *ChunkDiskMapper) Close() error {
+	// Locking the eventual position lock blocks WriteChunk()
+	cdm.evtlPosMtx.Lock()
+	defer cdm.evtlPosMtx.Unlock()
+
+	cdm.writeQueue.stop()
+
 	// 'WriteChunk' locks writePathMtx first and then readPathMtx for cutting head chunk file.
 	// The lock order should not be reversed here else it can cause deadlocks.
 	cdm.writePathMtx.Lock()

tsdb/chunks/head_chunks_test.go

@@ -28,7 +28,7 @@ import (
 )
 
 func TestChunkDiskMapper_WriteChunk_Chunk_IterateChunks(t *testing.T) {
-	hrw := testChunkDiskMapper(t)
+	hrw := createChunkDiskMapper(t, "")
 	defer func() {
 		require.NoError(t, hrw.Close())
 	}()
@@ -129,8 +129,7 @@ func TestChunkDiskMapper_WriteChunk_Chunk_IterateChunks(t *testing.T) {
 	// Testing IterateAllChunks method.
 	dir := hrw.dir.Name()
 	require.NoError(t, hrw.Close())
-	hrw, err = NewChunkDiskMapper(dir, chunkenc.NewPool(), DefaultWriteBufferSize)
-	require.NoError(t, err)
+	hrw = createChunkDiskMapper(t, dir)
 
 	idx := 0
 	require.NoError(t, hrw.IterateAllChunks(func(seriesRef HeadSeriesRef, chunkRef ChunkDiskMapperRef, mint, maxt int64, numSamples uint16) error {
@@ -156,10 +155,9 @@ func TestChunkDiskMapper_WriteChunk_Chunk_IterateChunks(t *testing.T) {
 // TestChunkDiskMapper_Truncate tests
 // * If truncation is happening properly based on the time passed.
 // * The active file is not deleted even if the passed time makes it eligible to be deleted.
-// * Empty current file does not lead to creation of another file after truncation.
 // * Non-empty current file leads to creation of another file after truncation.
 func TestChunkDiskMapper_Truncate(t *testing.T) {
-	hrw := testChunkDiskMapper(t)
+	hrw := createChunkDiskMapper(t, "")
 	defer func() {
 		require.NoError(t, hrw.Close())
 	}()
@@ -167,16 +165,20 @@ func TestChunkDiskMapper_Truncate(t *testing.T) {
 	timeRange := 0
 	fileTimeStep := 100
 	var thirdFileMinT, sixthFileMinT int64
-
 	addChunk := func() int {
-		mint := timeRange + 1                // Just after the new file cut.
-		maxt := timeRange + fileTimeStep - 1 // Just before the next file.
+		t.Helper()
 
-		// Write a chunks to set maxt for the segment.
-		_, err := hrw.WriteChunk(1, int64(mint), int64(maxt), randomChunk(t))
+		step := 100
+		mint, maxt := timeRange+1, timeRange+step-1
+		var err error
+		awaitCb := make(chan struct{})
+		hrw.WriteChunk(1, int64(mint), int64(maxt), randomChunk(t), func(cbErr error) {
+			err = cbErr
+			close(awaitCb)
+		})
+		<-awaitCb
 		require.NoError(t, err)
-
-		timeRange += fileTimeStep
+		timeRange += step
 
 		return mint
 	}
@@ -198,7 +200,7 @@ func TestChunkDiskMapper_Truncate(t *testing.T) {
 
 	// Create segments 1 to 7.
 	for i := 1; i <= 7; i++ {
-		require.NoError(t, hrw.CutNewFile())
+		hrw.CutNewFile()
 		mint := int64(addChunk())
 		if i == 3 {
 			thirdFileMinT = mint
@@ -210,19 +212,17 @@ func TestChunkDiskMapper_Truncate(t *testing.T) {
 
 	// Truncating files.
 	require.NoError(t, hrw.Truncate(thirdFileMinT))
+
+	// Add a chunk to trigger cutting of new file.
+	addChunk()
+
 	verifyFiles([]int{3, 4, 5, 6, 7, 8})
 
 	dir := hrw.dir.Name()
 	require.NoError(t, hrw.Close())
 
 	// Restarted.
-	var err error
-	hrw, err = NewChunkDiskMapper(dir, chunkenc.NewPool(), DefaultWriteBufferSize)
-	require.NoError(t, err)
-
-	require.False(t, hrw.fileMaxtSet)
-	require.NoError(t, hrw.IterateAllChunks(func(_ HeadSeriesRef, _ ChunkDiskMapperRef, _, _ int64, _ uint16) error { return nil }))
-	require.True(t, hrw.fileMaxtSet)
+	hrw = createChunkDiskMapper(t, dir)
 
 	verifyFiles([]int{3, 4, 5, 6, 7, 8})
 	// New file is created after restart even if last file was empty.
@@ -231,15 +231,23 @@ func TestChunkDiskMapper_Truncate(t *testing.T) {
 
 	// Truncating files after restart.
 	require.NoError(t, hrw.Truncate(sixthFileMinT))
-	verifyFiles([]int{6, 7, 8, 9, 10})
+	verifyFiles([]int{6, 7, 8, 9})
 
-	// As the last file was empty, this creates no new files.
+	// Truncating a second time without adding a chunk shouldn't create a new file.
 	require.NoError(t, hrw.Truncate(sixthFileMinT+1))
-	verifyFiles([]int{6, 7, 8, 9, 10})
+	verifyFiles([]int{6, 7, 8, 9})
+
+	// Add a chunk to trigger cutting of new file.
 	addChunk()
 
+	verifyFiles([]int{6, 7, 8, 9, 10})
+
 	// Truncating till current time should not delete the current active file.
 	require.NoError(t, hrw.Truncate(int64(timeRange+(2*fileTimeStep))))
+
+	// Add a chunk to trigger cutting of new file.
+	addChunk()
+
 	verifyFiles([]int{10, 11}) // One file is the previously active file and one currently created.
 }
 
@@ -248,23 +256,40 @@ func TestChunkDiskMapper_Truncate(t *testing.T) {
 // This test exposes https://github.com/prometheus/prometheus/issues/7412 where the truncation
 // simply deleted all empty files instead of stopping once it encountered a non-empty file.
 func TestChunkDiskMapper_Truncate_PreservesFileSequence(t *testing.T) {
-	hrw := testChunkDiskMapper(t)
+	hrw := createChunkDiskMapper(t, "")
 	defer func() {
 		require.NoError(t, hrw.Close())
 	}()
-
 	timeRange := 0
+
 	addChunk := func() {
+		t.Helper()
+
+		awaitCb := make(chan struct{})
+
 		step := 100
 		mint, maxt := timeRange+1, timeRange+step-1
-		_, err := hrw.WriteChunk(1, int64(mint), int64(maxt), randomChunk(t))
-		require.NoError(t, err)
+		hrw.WriteChunk(1, int64(mint), int64(maxt), randomChunk(t), func(err error) {
+			close(awaitCb)
+			require.NoError(t, err)
+		})
+		<-awaitCb
 		timeRange += step
 	}
+
 	emptyFile := func() {
-		require.NoError(t, hrw.CutNewFile())
+		t.Helper()
+
+		_, _, err := hrw.cut()
+		require.NoError(t, err)
+		hrw.evtlPosMtx.Lock()
+		hrw.evtlPos.toNewFile()
+		hrw.evtlPosMtx.Unlock()
 	}
+
 	nonEmptyFile := func() {
+		t.Helper()
+
 		emptyFile()
 		addChunk()
 	}
@@ -297,42 +322,48 @@ func TestChunkDiskMapper_Truncate_PreservesFileSequence(t *testing.T) {
 	// though files 4 and 6 are empty.
 	file2Maxt := hrw.mmappedChunkFiles[2].maxt
 	require.NoError(t, hrw.Truncate(file2Maxt+1))
-	// As 6 was empty, it should not create another file.
 	verifyFiles([]int{3, 4, 5, 6})
 
+	// Add chunk, so file 6 is not empty anymore.
 	addChunk()
-	// Truncate creates another file as 6 is not empty now.
-	require.NoError(t, hrw.Truncate(file2Maxt+1))
-	verifyFiles([]int{3, 4, 5, 6, 7})
+	verifyFiles([]int{3, 4, 5, 6})
+
+	// Truncating till file 3 should also delete file 4, because it is empty.
+	file3Maxt := hrw.mmappedChunkFiles[3].maxt
+	require.NoError(t, hrw.Truncate(file3Maxt+1))
+	addChunk()
+	verifyFiles([]int{5, 6, 7})
 
 	dir := hrw.dir.Name()
 	require.NoError(t, hrw.Close())
-
 	// Restarting checks for unsequential files.
-	var err error
-	hrw, err = NewChunkDiskMapper(dir, chunkenc.NewPool(), DefaultWriteBufferSize)
-	require.NoError(t, err)
-	verifyFiles([]int{3, 4, 5, 6, 7})
+	hrw = createChunkDiskMapper(t, dir)
+	verifyFiles([]int{5, 6, 7})
 }
 
 // TestHeadReadWriter_TruncateAfterIterateChunksError tests for
 // https://github.com/prometheus/prometheus/issues/7753
 func TestHeadReadWriter_TruncateAfterFailedIterateChunks(t *testing.T) {
-	hrw := testChunkDiskMapper(t)
+	hrw := createChunkDiskMapper(t, "")
 	defer func() {
 		require.NoError(t, hrw.Close())
 	}()
 
 	// Write a chunks to iterate on it later.
-	_, err := hrw.WriteChunk(1, 0, 1000, randomChunk(t))
+	var err error
+	awaitCb := make(chan struct{})
+	hrw.WriteChunk(1, 0, 1000, randomChunk(t), func(cbErr error) {
+		err = cbErr
+		close(awaitCb)
+	})
+	<-awaitCb
 	require.NoError(t, err)
 
 	dir := hrw.dir.Name()
 	require.NoError(t, hrw.Close())
 
 	// Restarting to recreate https://github.com/prometheus/prometheus/issues/7753.
-	hrw, err = NewChunkDiskMapper(dir, chunkenc.NewPool(), DefaultWriteBufferSize)
-	require.NoError(t, err)
+	hrw = createChunkDiskMapper(t, dir)
 
 	// Forcefully failing IterateAllChunks.
 	require.Error(t, hrw.IterateAllChunks(func(_ HeadSeriesRef, _ ChunkDiskMapperRef, _, _ int64, _ uint16) error {
@@ -344,21 +375,31 @@ func TestHeadReadWriter_TruncateAfterFailedIterateChunks(t *testing.T) {
 }
 
 func TestHeadReadWriter_ReadRepairOnEmptyLastFile(t *testing.T) {
-	hrw := testChunkDiskMapper(t)
+	hrw := createChunkDiskMapper(t, "")
 	defer func() {
 		require.NoError(t, hrw.Close())
 	}()
 
 	timeRange := 0
 	addChunk := func() {
+		t.Helper()
+
 		step := 100
 		mint, maxt := timeRange+1, timeRange+step-1
-		_, err := hrw.WriteChunk(1, int64(mint), int64(maxt), randomChunk(t))
+		var err error
+		awaitCb := make(chan struct{})
+		hrw.WriteChunk(1, int64(mint), int64(maxt), randomChunk(t), func(cbErr error) {
+			err = cbErr
+			close(awaitCb)
+		})
+		<-awaitCb
 		require.NoError(t, err)
 		timeRange += step
 	}
 	nonEmptyFile := func() {
-		require.NoError(t, hrw.CutNewFile())
+		t.Helper()
+
+		hrw.CutNewFile()
 		addChunk()
 	}
 
@@ -388,11 +429,7 @@ func TestHeadReadWriter_ReadRepairOnEmptyLastFile(t *testing.T) {
 	require.NoError(t, f.Close())
 
 	// Open chunk disk mapper again, corrupt file should be removed.
-	hrw, err = NewChunkDiskMapper(dir, chunkenc.NewPool(), DefaultWriteBufferSize)
-	require.NoError(t, err)
-	require.False(t, hrw.fileMaxtSet)
-	require.NoError(t, hrw.IterateAllChunks(func(_ HeadSeriesRef, _ ChunkDiskMapperRef, _, _ int64, _ uint16) error { return nil }))
-	require.True(t, hrw.fileMaxtSet)
+	hrw = createChunkDiskMapper(t, dir)
 
 	// Removed from memory.
 	require.Equal(t, 3, len(hrw.mmappedChunkFiles))
@@ -411,18 +448,22 @@ func TestHeadReadWriter_ReadRepairOnEmptyLastFile(t *testing.T) {
 	}
 }
 
-func testChunkDiskMapper(t *testing.T) *ChunkDiskMapper {
-	tmpdir, err := ioutil.TempDir("", "data")
-	require.NoError(t, err)
-	t.Cleanup(func() {
-		require.NoError(t, os.RemoveAll(tmpdir))
-	})
+func createChunkDiskMapper(t *testing.T, dir string) *ChunkDiskMapper {
+	if dir == "" {
+		var err error
+		dir, err = ioutil.TempDir("", "data")
+		require.NoError(t, err)
+		t.Cleanup(func() {
+			require.NoError(t, os.RemoveAll(dir))
+		})
+	}
 
-	hrw, err := NewChunkDiskMapper(tmpdir, chunkenc.NewPool(), DefaultWriteBufferSize)
+	hrw, err := NewChunkDiskMapper(nil, dir, chunkenc.NewPool(), DefaultWriteBufferSize, DefaultWriteQueueSize)
 	require.NoError(t, err)
 	require.False(t, hrw.fileMaxtSet)
 	require.NoError(t, hrw.IterateAllChunks(func(_ HeadSeriesRef, _ ChunkDiskMapperRef, _, _ int64, _ uint16) error { return nil }))
 	require.True(t, hrw.fileMaxtSet)
+
 	return hrw
 }
 
@@ -443,7 +484,11 @@ func createChunk(t *testing.T, idx int, hrw *ChunkDiskMapper) (seriesRef HeadSer
 	mint = int64((idx)*1000 + 1)
 	maxt = int64((idx + 1) * 1000)
 	chunk = randomChunk(t)
-	chunkRef, err = hrw.WriteChunk(seriesRef, mint, maxt, chunk)
-	require.NoError(t, err)
+	awaitCb := make(chan struct{})
+	chunkRef = hrw.WriteChunk(seriesRef, mint, maxt, chunk, func(cbErr error) {
+		require.NoError(t, err)
+		close(awaitCb)
+	})
+	<-awaitCb
 	return
 }

tsdb/db.go

@@ -80,6 +80,7 @@ func DefaultOptions() *Options {
 		StripeSize:                DefaultStripeSize,
 		HeadChunksWriteBufferSize: chunks.DefaultWriteBufferSize,
 		IsolationDisabled:         defaultIsolationDisabled,
+		HeadChunksWriteQueueSize:  chunks.DefaultWriteQueueSize,
 	}
 }
 
@@ -135,6 +136,9 @@ type Options struct {
 	// HeadChunksWriteBufferSize configures the write buffer size used by the head chunks mapper.
 	HeadChunksWriteBufferSize int
 
+	// HeadChunksWriteQueueSize configures the size of the chunk write queue used in the head chunks mapper.
+	HeadChunksWriteQueueSize int
+
 	// SeriesLifecycleCallback specifies a list of callbacks that will be called during a lifecycle of a series.
 	// It is always a no-op in Prometheus and mainly meant for external users who import TSDB.
 	SeriesLifecycleCallback SeriesLifecycleCallback
@@ -582,6 +586,9 @@ func validateOpts(opts *Options, rngs []int64) (*Options, []int64) {
 	if opts.HeadChunksWriteBufferSize <= 0 {
 		opts.HeadChunksWriteBufferSize = chunks.DefaultWriteBufferSize
 	}
+	if opts.HeadChunksWriteQueueSize < 0 {
+		opts.HeadChunksWriteQueueSize = chunks.DefaultWriteQueueSize
+	}
 	if opts.MaxBlockChunkSegmentSize <= 0 {
 		opts.MaxBlockChunkSegmentSize = chunks.DefaultChunkSegmentSize
 	}
@@ -704,6 +711,7 @@ func open(dir string, l log.Logger, r prometheus.Registerer, opts *Options, rngs
 	headOpts.ChunkDirRoot = dir
 	headOpts.ChunkPool = db.chunkPool
 	headOpts.ChunkWriteBufferSize = opts.HeadChunksWriteBufferSize
+	headOpts.ChunkWriteQueueSize = opts.HeadChunksWriteQueueSize
 	headOpts.StripeSize = opts.StripeSize
 	headOpts.SeriesCallback = opts.SeriesLifecycleCallback
 	headOpts.EnableExemplarStorage = opts.EnableExemplarStorage

tsdb/db_test.go

@@ -212,17 +212,13 @@ func TestNoPanicAfterWALCorruption(t *testing.T) {
 	var maxt int64
 	ctx := context.Background()
 	{
-		for {
+		// Appending 121 samples because on the 121st a new chunk will be created.
+		for i := 0; i < 121; i++ {
 			app := db.Appender(ctx)
 			_, err := app.Append(0, labels.FromStrings("foo", "bar"), maxt, 0)
 			expSamples = append(expSamples, sample{t: maxt, v: 0})
 			require.NoError(t, err)
 			require.NoError(t, app.Commit())
-			mmapedChunks, err := ioutil.ReadDir(mmappedChunksDir(db.Dir()))
-			require.NoError(t, err)
-			if len(mmapedChunks) > 0 {
-				break
-			}
 			maxt++
 		}
 		require.NoError(t, db.Close())
@@ -2453,7 +2449,7 @@ func TestDBReadOnly_FlushWAL(t *testing.T) {
 			require.NoError(t, err)
 		}
 		require.NoError(t, app.Commit())
-		defer func() { require.NoError(t, db.Close()) }()
+		require.NoError(t, db.Close())
 	}
 
 	// Flush WAL.

tsdb/head.go

@@ -127,6 +127,8 @@ type HeadOptions struct {
 	ChunkDirRoot         string
 	ChunkPool            chunkenc.Pool
 	ChunkWriteBufferSize int
+	ChunkWriteQueueSize  int
+
 	// StripeSize sets the number of entries in the hash map, it must be a power of 2.
 	// A larger StripeSize will allocate more memory up-front, but will increase performance when handling a large number of series.
 	// A smaller StripeSize reduces the memory allocated, but can decrease performance with large number of series.
@@ -144,6 +146,7 @@ func DefaultHeadOptions() *HeadOptions {
 		ChunkDirRoot:         "",
 		ChunkPool:            chunkenc.NewPool(),
 		ChunkWriteBufferSize: chunks.DefaultWriteBufferSize,
+		ChunkWriteQueueSize:  chunks.DefaultWriteQueueSize,
 		StripeSize:           DefaultStripeSize,
 		SeriesCallback:       &noopSeriesLifecycleCallback{},
 		IsolationDisabled:    defaultIsolationDisabled,
@@ -208,9 +211,11 @@ func NewHead(r prometheus.Registerer, l log.Logger, wal *wal.WAL, opts *HeadOpti
 	}
 
 	h.chunkDiskMapper, err = chunks.NewChunkDiskMapper(
+		r,
 		mmappedChunksDir(opts.ChunkDirRoot),
 		opts.ChunkPool,
 		opts.ChunkWriteBufferSize,
+		opts.ChunkWriteQueueSize,
 	)
 	if err != nil {
 		return nil, err

tsdb/head_append.go

@@ -571,12 +571,7 @@ func (s *memSeries) mmapCurrentHeadChunk(chunkDiskMapper *chunks.ChunkDiskMapper
 		return
 	}
 
-	chunkRef, err := chunkDiskMapper.WriteChunk(s.ref, s.headChunk.minTime, s.headChunk.maxTime, s.headChunk.chunk)
-	if err != nil {
-		if err != chunks.ErrChunkDiskMapperClosed {
-			panic(err)
-		}
-	}
+	chunkRef := chunkDiskMapper.WriteChunk(s.ref, s.headChunk.minTime, s.headChunk.maxTime, s.headChunk.chunk, handleChunkWriteError)
 	s.mmappedChunks = append(s.mmappedChunks, &mmappedChunk{
 		ref:        chunkRef,
 		numSamples: uint16(s.headChunk.chunk.NumSamples()),
@@ -585,6 +580,12 @@ func (s *memSeries) mmapCurrentHeadChunk(chunkDiskMapper *chunks.ChunkDiskMapper
 	})
 }
 
+func handleChunkWriteError(err error) {
+	if err != nil && err != chunks.ErrChunkDiskMapperClosed {
+		panic(err)
+	}
+}
+
 // Rollback removes the samples and exemplars from headAppender and writes any series to WAL.
 func (a *headAppender) Rollback() (err error) {
 	if a.closed {

tsdb/head_test.go

@@ -35,6 +35,7 @@ import (
 	prom_testutil "github.com/prometheus/client_golang/prometheus/testutil"
 	"github.com/stretchr/testify/require"
 	"go.uber.org/atomic"
+	"golang.org/x/sync/errgroup"
 
 	"github.com/prometheus/prometheus/config"
 	"github.com/prometheus/prometheus/model/exemplar"
@@ -224,7 +225,7 @@ func BenchmarkLoadWAL(b *testing.B) {
 
 					// Write mmapped chunks.
 					if c.mmappedChunkT != 0 {
-						chunkDiskMapper, err := chunks.NewChunkDiskMapper(mmappedChunksDir(dir), chunkenc.NewPool(), chunks.DefaultWriteBufferSize)
+						chunkDiskMapper, err := chunks.NewChunkDiskMapper(nil, mmappedChunksDir(dir), chunkenc.NewPool(), chunks.DefaultWriteBufferSize, chunks.DefaultWriteQueueSize)
 						require.NoError(b, err)
 						for k := 0; k < c.batches*c.seriesPerBatch; k++ {
 							// Create one mmapped chunk per series, with one sample at the given time.
@@ -270,6 +271,195 @@ func BenchmarkLoadWAL(b *testing.B) {
 	}
 }
 
+// TestHead_HighConcurrencyReadAndWrite generates 1000 series with a step of 15s and fills a whole block with samples,
+// this means in total it generates 4000 chunks because with a step of 15s there are 4 chunks per block per series.
+// While appending the samples to the head it concurrently queries them from multiple go routines and verifies that the
+// returned results are correct.
+func TestHead_HighConcurrencyReadAndWrite(t *testing.T) {
+	head, _ := newTestHead(t, DefaultBlockDuration, false)
+	defer func() {
+		require.NoError(t, head.Close())
+	}()
+
+	seriesCnt := 1000
+	readConcurrency := 2
+	writeConcurrency := 10
+	startTs := uint64(DefaultBlockDuration) // start at the second block relative to the unix epoch.
+	qryRange := uint64(5 * time.Minute.Milliseconds())
+	step := uint64(15 * time.Second / time.Millisecond)
+	endTs := startTs + uint64(DefaultBlockDuration)
+
+	labelSets := make([]labels.Labels, seriesCnt)
+	for i := 0; i < seriesCnt; i++ {
+		labelSets[i] = labels.FromStrings("seriesId", strconv.Itoa(i))
+	}
+
+	head.Init(0)
+
+	g, ctx := errgroup.WithContext(context.Background())
+	whileNotCanceled := func(f func() (bool, error)) error {
+		for ctx.Err() == nil {
+			cont, err := f()
+			if err != nil {
+				return err
+			}
+			if !cont {
+				return nil
+			}
+		}
+		return nil
+	}
+
+	// Create one channel for each write worker, the channels will be used by the coordinator
+	// go routine to coordinate which timestamps each write worker has to write.
+	writerTsCh := make([]chan uint64, writeConcurrency)
+	for writerTsChIdx := range writerTsCh {
+		writerTsCh[writerTsChIdx] = make(chan uint64)
+	}
+
+	// workerReadyWg is used to synchronize the start of the test,
+	// we only start the test once all workers signal that they're ready.
+	var workerReadyWg sync.WaitGroup
+	workerReadyWg.Add(writeConcurrency + readConcurrency)
+
+	// Start the write workers.
+	for wid := 0; wid < writeConcurrency; wid++ {
+		// Create copy of workerID to be used by worker routine.
+		workerID := wid
+
+		g.Go(func() error {
+			// The label sets which this worker will write.
+			workerLabelSets := labelSets[(seriesCnt/writeConcurrency)*workerID : (seriesCnt/writeConcurrency)*(workerID+1)]
+
+			// Signal that this worker is ready.
+			workerReadyWg.Done()
+
+			return whileNotCanceled(func() (bool, error) {
+				ts, ok := <-writerTsCh[workerID]
+				if !ok {
+					return false, nil
+				}
+
+				app := head.Appender(ctx)
+				for i := 0; i < len(workerLabelSets); i++ {
+					// We also use the timestamp as the sample value.
+					_, err := app.Append(0, workerLabelSets[i], int64(ts), float64(ts))
+					if err != nil {
+						return false, fmt.Errorf("Error when appending to head: %w", err)
+					}
+				}
+
+				return true, app.Commit()
+			})
+		})
+	}
+
+	// queryHead is a helper to query the head for a given time range and labelset.
+	queryHead := func(mint, maxt uint64, label labels.Label) (map[string][]tsdbutil.Sample, error) {
+		q, err := NewBlockQuerier(head, int64(mint), int64(maxt))
+		if err != nil {
+			return nil, err
+		}
+		return query(t, q, labels.MustNewMatcher(labels.MatchEqual, label.Name, label.Value)), nil
+	}
+
+	// readerTsCh will be used by the coordinator go routine to coordinate which timestamps the reader should read.
+	readerTsCh := make(chan uint64)
+
+	// Start the read workers.
+	for wid := 0; wid < readConcurrency; wid++ {
+		// Create copy of threadID to be used by worker routine.
+		workerID := wid
+
+		g.Go(func() error {
+			querySeriesRef := (seriesCnt / readConcurrency) * workerID
+
+			// Signal that this worker is ready.
+			workerReadyWg.Done()
+
+			return whileNotCanceled(func() (bool, error) {
+				ts, ok := <-readerTsCh
+				if !ok {
+					return false, nil
+				}
+
+				querySeriesRef = (querySeriesRef + 1) % seriesCnt
+				lbls := labelSets[querySeriesRef]
+				samples, err := queryHead(ts-qryRange, ts, lbls[0])
+				if err != nil {
+					return false, err
+				}
+
+				if len(samples) != 1 {
+					return false, fmt.Errorf("expected 1 series, got %d", len(samples))
+				}
+
+				series := lbls.String()
+				expectSampleCnt := qryRange/step + 1
+				if expectSampleCnt != uint64(len(samples[series])) {
+					return false, fmt.Errorf("expected %d samples, got %d", expectSampleCnt, len(samples[series]))
+				}
+
+				for sampleIdx, sample := range samples[series] {
+					expectedValue := ts - qryRange + (uint64(sampleIdx) * step)
+					if sample.T() != int64(expectedValue) {
+						return false, fmt.Errorf("expected sample %d to have ts %d, got %d", sampleIdx, expectedValue, sample.T())
+					}
+					if sample.V() != float64(expectedValue) {
+						return false, fmt.Errorf("expected sample %d to have value %d, got %f", sampleIdx, expectedValue, sample.V())
+					}
+				}
+
+				return true, nil
+			})
+		})
+	}
+
+	// Start the coordinator go routine.
+	g.Go(func() error {
+		currTs := startTs
+
+		defer func() {
+			// End of the test, close all channels to stop the workers.
+			for _, ch := range writerTsCh {
+				close(ch)
+			}
+			close(readerTsCh)
+		}()
+
+		// Wait until all workers are ready to start the test.
+		workerReadyWg.Wait()
+		return whileNotCanceled(func() (bool, error) {
+			// Send the current timestamp to each of the writers.
+			for _, ch := range writerTsCh {
+				select {
+				case ch <- currTs:
+				case <-ctx.Done():
+					return false, nil
+				}
+			}
+
+			// Once data for at least <qryRange> has been ingested, send the current timestamp to the readers.
+			if currTs > startTs+qryRange {
+				select {
+				case readerTsCh <- currTs - step:
+				case <-ctx.Done():
+					return false, nil
+				}
+			}
+
+			currTs += step
+			if currTs > endTs {
+				return false, nil
+			}
+
+			return true, nil
+		})
+	})
+
+	require.NoError(t, g.Wait())
+}
+
 func TestHead_ReadWAL(t *testing.T) {
 	for _, compress := range []bool{false, true} {
 		t.Run(fmt.Sprintf("compress=%t", compress), func(t *testing.T) {
@@ -540,7 +730,7 @@ func TestMemSeries_truncateChunks(t *testing.T) {
 		require.NoError(t, os.RemoveAll(dir))
 	}()
 	// This is usually taken from the Head, but passing manually here.
-	chunkDiskMapper, err := chunks.NewChunkDiskMapper(dir, chunkenc.NewPool(), chunks.DefaultWriteBufferSize)
+	chunkDiskMapper, err := chunks.NewChunkDiskMapper(nil, dir, chunkenc.NewPool(), chunks.DefaultWriteBufferSize, chunks.DefaultWriteQueueSize)
 	require.NoError(t, err)
 	defer func() {
 		require.NoError(t, chunkDiskMapper.Close())
@@ -1084,7 +1274,7 @@ func TestMemSeries_append(t *testing.T) {
 		require.NoError(t, os.RemoveAll(dir))
 	}()
 	// This is usually taken from the Head, but passing manually here.
-	chunkDiskMapper, err := chunks.NewChunkDiskMapper(dir, chunkenc.NewPool(), chunks.DefaultWriteBufferSize)
+	chunkDiskMapper, err := chunks.NewChunkDiskMapper(nil, dir, chunkenc.NewPool(), chunks.DefaultWriteBufferSize, chunks.DefaultWriteQueueSize)
 	require.NoError(t, err)
 	defer func() {
 		require.NoError(t, chunkDiskMapper.Close())
@@ -1462,14 +1652,16 @@ func TestHeadReadWriterRepair(t *testing.T) {
 			ok, chunkCreated = s.append(int64(i*chunkRange)+chunkRange-1, float64(i*chunkRange), 0, h.chunkDiskMapper)
 			require.True(t, ok, "series append failed")
 			require.False(t, chunkCreated, "chunk was created")
-			require.NoError(t, h.chunkDiskMapper.CutNewFile())
+			h.chunkDiskMapper.CutNewFile()
 		}
 		require.NoError(t, h.Close())
 
-		// Verify that there are 7 segment files.
+		// Verify that there are 6 segment files.
+		// It should only be 6 because the last call to .CutNewFile() won't
+		// take effect without another chunk being written.
 		files, err := ioutil.ReadDir(mmappedChunksDir(dir))
 		require.NoError(t, err)
-		require.Equal(t, 7, len(files))
+		require.Equal(t, 6, len(files))
 
 		// Corrupt the 4th file by writing a random byte to series ref.
 		f, err := os.OpenFile(filepath.Join(mmappedChunksDir(dir), files[3].Name()), os.O_WRONLY, 0o666)
@@ -2270,7 +2462,7 @@ func TestMemSafeIteratorSeekIntoBuffer(t *testing.T) {
 		require.NoError(t, os.RemoveAll(dir))
 	}()
 	// This is usually taken from the Head, but passing manually here.
-	chunkDiskMapper, err := chunks.NewChunkDiskMapper(dir, chunkenc.NewPool(), chunks.DefaultWriteBufferSize)
+	chunkDiskMapper, err := chunks.NewChunkDiskMapper(nil, dir, chunkenc.NewPool(), chunks.DefaultWriteBufferSize, chunks.DefaultWriteQueueSize)
 	require.NoError(t, err)
 	defer func() {
 		require.NoError(t, chunkDiskMapper.Close())