2021-08-03 12:14:26 +00:00
// 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 tsdb
import (
2023-11-16 18:54:41 +00:00
"errors"
2021-08-03 12:14:26 +00:00
"fmt"
"math"
2021-08-11 10:13:17 +00:00
"os"
"path/filepath"
"strconv"
"strings"
2021-08-03 12:14:26 +00:00
"sync"
2021-08-03 14:33:54 +00:00
"time"
2021-08-03 12:14:26 +00:00
"github.com/go-kit/log/level"
"go.uber.org/atomic"
2021-11-08 14:23:17 +00:00
"github.com/prometheus/prometheus/model/exemplar"
2022-12-28 08:55:07 +00:00
"github.com/prometheus/prometheus/model/histogram"
2021-11-08 14:23:17 +00:00
"github.com/prometheus/prometheus/model/labels"
2022-07-19 08:58:52 +00:00
"github.com/prometheus/prometheus/model/metadata"
2021-08-03 12:14:26 +00:00
"github.com/prometheus/prometheus/storage"
2021-10-22 08:19:38 +00:00
"github.com/prometheus/prometheus/tsdb/chunkenc"
2021-11-06 10:10:04 +00:00
"github.com/prometheus/prometheus/tsdb/chunks"
2021-10-22 08:19:38 +00:00
"github.com/prometheus/prometheus/tsdb/encoding"
tsdb_errors "github.com/prometheus/prometheus/tsdb/errors"
"github.com/prometheus/prometheus/tsdb/fileutil"
2021-08-03 12:14:26 +00:00
"github.com/prometheus/prometheus/tsdb/record"
"github.com/prometheus/prometheus/tsdb/tombstones"
2022-10-10 15:08:46 +00:00
"github.com/prometheus/prometheus/tsdb/wlog"
2023-03-29 19:34:34 +00:00
"github.com/prometheus/prometheus/util/zeropool"
2021-08-03 12:14:26 +00:00
)
2022-12-28 08:55:07 +00:00
// histogramRecord combines both RefHistogramSample and RefFloatHistogramSample
// to simplify the WAL replay.
type histogramRecord struct {
ref chunks . HeadSeriesRef
t int64
h * histogram . Histogram
fh * histogram . FloatHistogram
}
2022-10-10 15:08:46 +00:00
func ( h * Head ) loadWAL ( r * wlog . Reader , multiRef map [ chunks . HeadSeriesRef ] chunks . HeadSeriesRef , mmappedChunks , oooMmappedChunks map [ chunks . HeadSeriesRef ] [ ] * mmappedChunk ) ( err error ) {
2021-08-03 12:14:26 +00:00
// Track number of samples that referenced a series we don't know about
// for error reporting.
var unknownRefs atomic . Uint64
var unknownExemplarRefs atomic . Uint64
2022-10-05 20:14:49 +00:00
var unknownHistogramRefs atomic . Uint64
2022-07-19 08:58:52 +00:00
var unknownMetadataRefs atomic . Uint64
2021-10-01 11:30:22 +00:00
// Track number of series records that had overlapping m-map chunks.
2022-08-17 13:53:57 +00:00
var mmapOverlappingChunks atomic . Uint64
2021-08-03 12:14:26 +00:00
// Start workers that each process samples for a partition of the series ID space.
var (
wg sync . WaitGroup
2023-03-07 16:41:33 +00:00
concurrency = h . opts . WALReplayConcurrency
processors = make ( [ ] walSubsetProcessor , concurrency )
2021-08-03 12:14:26 +00:00
exemplarsInput chan record . RefExemplar
2021-11-30 16:19:06 +00:00
dec record . Decoder
2023-03-07 16:41:33 +00:00
shards = make ( [ ] [ ] record . RefSample , concurrency )
histogramShards = make ( [ ] [ ] histogramRecord , concurrency )
2021-08-03 12:14:26 +00:00
decoded = make ( chan interface { } , 10 )
decodeErr , seriesCreationErr error
2023-03-29 19:34:34 +00:00
seriesPool zeropool . Pool [ [ ] record . RefSeries ]
samplesPool zeropool . Pool [ [ ] record . RefSample ]
tstonesPool zeropool . Pool [ [ ] tombstones . Stone ]
exemplarsPool zeropool . Pool [ [ ] record . RefExemplar ]
histogramsPool zeropool . Pool [ [ ] record . RefHistogramSample ]
floatHistogramsPool zeropool . Pool [ [ ] record . RefFloatHistogramSample ]
metadataPool zeropool . Pool [ [ ] record . RefMetadata ]
2021-08-03 12:14:26 +00:00
)
defer func ( ) {
// For CorruptionErr ensure to terminate all workers before exiting.
2022-10-10 15:08:46 +00:00
_ , ok := err . ( * wlog . CorruptionErr )
2021-08-03 12:14:26 +00:00
if ok || seriesCreationErr != nil {
2023-03-07 16:41:33 +00:00
for i := 0 ; i < concurrency ; i ++ {
2021-11-25 08:06:14 +00:00
processors [ i ] . closeAndDrain ( )
2021-08-03 12:14:26 +00:00
}
close ( exemplarsInput )
wg . Wait ( )
}
} ( )
2023-03-07 16:41:33 +00:00
wg . Add ( concurrency )
for i := 0 ; i < concurrency ; i ++ {
2021-11-25 08:06:14 +00:00
processors [ i ] . setup ( )
2021-08-03 12:14:26 +00:00
2021-11-25 08:06:14 +00:00
go func ( wp * walSubsetProcessor ) {
2022-10-05 20:14:49 +00:00
unknown , unknownHistograms , overlapping := wp . processWALSamples ( h , mmappedChunks , oooMmappedChunks )
2021-08-03 12:14:26 +00:00
unknownRefs . Add ( unknown )
2022-08-17 13:53:57 +00:00
mmapOverlappingChunks . Add ( overlapping )
2021-11-30 16:19:06 +00:00
unknownHistogramRefs . Add ( unknownHistograms )
2021-08-03 12:14:26 +00:00
wg . Done ( )
2021-11-25 08:06:14 +00:00
} ( & processors [ i ] )
2021-08-03 12:14:26 +00:00
}
wg . Add ( 1 )
exemplarsInput = make ( chan record . RefExemplar , 300 )
go func ( input <- chan record . RefExemplar ) {
2021-08-27 06:19:34 +00:00
var err error
2021-08-03 12:14:26 +00:00
defer wg . Done ( )
for e := range input {
ms := h . series . getByID ( e . Ref )
if ms == nil {
unknownExemplarRefs . Inc ( )
continue
}
if e . T < h . minValidTime . Load ( ) {
continue
}
// At the moment the only possible error here is out of order exemplars, which we shouldn't see when
// replaying the WAL, so lets just log the error if it's not that type.
err = h . exemplars . AddExemplar ( ms . lset , exemplar . Exemplar { Ts : e . T , Value : e . V , Labels : e . Labels } )
2023-11-16 18:54:41 +00:00
if err != nil && errors . Is ( err , storage . ErrOutOfOrderExemplar ) {
2021-08-03 12:14:26 +00:00
level . Warn ( h . logger ) . Log ( "msg" , "Unexpected error when replaying WAL on exemplar record" , "err" , err )
}
}
} ( exemplarsInput )
go func ( ) {
defer close ( decoded )
2022-04-12 10:30:20 +00:00
var err error
2021-08-03 12:14:26 +00:00
for r . Next ( ) {
rec := r . Record ( )
switch dec . Type ( rec ) {
case record . Series :
2023-03-29 19:34:34 +00:00
series := seriesPool . Get ( ) [ : 0 ]
2021-08-03 12:14:26 +00:00
series , err = dec . Series ( rec , series )
if err != nil {
2022-10-10 15:08:46 +00:00
decodeErr = & wlog . CorruptionErr {
2023-11-16 18:54:41 +00:00
Err : fmt . Errorf ( "decode series: %w" , err ) ,
2021-08-03 12:14:26 +00:00
Segment : r . Segment ( ) ,
Offset : r . Offset ( ) ,
}
return
}
decoded <- series
case record . Samples :
2023-03-29 19:34:34 +00:00
samples := samplesPool . Get ( ) [ : 0 ]
2021-08-03 12:14:26 +00:00
samples , err = dec . Samples ( rec , samples )
if err != nil {
2022-10-10 15:08:46 +00:00
decodeErr = & wlog . CorruptionErr {
2023-11-16 18:54:41 +00:00
Err : fmt . Errorf ( "decode samples: %w" , err ) ,
2021-08-03 12:14:26 +00:00
Segment : r . Segment ( ) ,
Offset : r . Offset ( ) ,
}
return
}
decoded <- samples
case record . Tombstones :
2023-03-29 19:34:34 +00:00
tstones := tstonesPool . Get ( ) [ : 0 ]
2021-08-03 12:14:26 +00:00
tstones , err = dec . Tombstones ( rec , tstones )
if err != nil {
2022-10-10 15:08:46 +00:00
decodeErr = & wlog . CorruptionErr {
2023-11-16 18:54:41 +00:00
Err : fmt . Errorf ( "decode tombstones: %w" , err ) ,
2021-08-03 12:14:26 +00:00
Segment : r . Segment ( ) ,
Offset : r . Offset ( ) ,
}
return
}
decoded <- tstones
case record . Exemplars :
2023-03-29 19:34:34 +00:00
exemplars := exemplarsPool . Get ( ) [ : 0 ]
2021-08-03 12:14:26 +00:00
exemplars , err = dec . Exemplars ( rec , exemplars )
if err != nil {
2022-10-10 15:08:46 +00:00
decodeErr = & wlog . CorruptionErr {
2023-11-16 18:54:41 +00:00
Err : fmt . Errorf ( "decode exemplars: %w" , err ) ,
2021-08-03 12:14:26 +00:00
Segment : r . Segment ( ) ,
Offset : r . Offset ( ) ,
}
return
}
decoded <- exemplars
2022-08-29 12:08:36 +00:00
case record . HistogramSamples :
2023-03-29 19:34:34 +00:00
hists := histogramsPool . Get ( ) [ : 0 ]
2022-08-29 12:08:36 +00:00
hists , err = dec . HistogramSamples ( rec , hists )
2021-08-11 12:08:48 +00:00
if err != nil {
2022-10-12 12:29:42 +00:00
decodeErr = & wlog . CorruptionErr {
2023-11-16 18:54:41 +00:00
Err : fmt . Errorf ( "decode histograms: %w" , err ) ,
2021-08-11 12:08:48 +00:00
Segment : r . Segment ( ) ,
Offset : r . Offset ( ) ,
}
return
}
decoded <- hists
2022-12-28 08:55:07 +00:00
case record . FloatHistogramSamples :
2023-03-29 19:34:34 +00:00
hists := floatHistogramsPool . Get ( ) [ : 0 ]
2022-12-28 08:55:07 +00:00
hists , err = dec . FloatHistogramSamples ( rec , hists )
if err != nil {
decodeErr = & wlog . CorruptionErr {
2023-11-16 18:54:41 +00:00
Err : fmt . Errorf ( "decode float histograms: %w" , err ) ,
2022-12-28 08:55:07 +00:00
Segment : r . Segment ( ) ,
Offset : r . Offset ( ) ,
}
return
}
decoded <- hists
2022-07-19 08:58:52 +00:00
case record . Metadata :
2023-03-29 19:34:34 +00:00
meta := metadataPool . Get ( ) [ : 0 ]
2022-07-19 08:58:52 +00:00
meta , err := dec . Metadata ( rec , meta )
if err != nil {
2022-10-10 15:08:46 +00:00
decodeErr = & wlog . CorruptionErr {
2023-11-16 18:54:41 +00:00
Err : fmt . Errorf ( "decode metadata: %w" , err ) ,
2022-07-19 08:58:52 +00:00
Segment : r . Segment ( ) ,
Offset : r . Offset ( ) ,
}
return
}
decoded <- meta
2021-08-03 12:14:26 +00:00
default :
// Noop.
}
}
} ( )
2021-08-03 14:33:54 +00:00
// The records are always replayed from the oldest to the newest.
2021-08-03 12:14:26 +00:00
Outer :
for d := range decoded {
switch v := d . ( type ) {
case [ ] record . RefSeries :
2021-08-03 14:33:54 +00:00
for _ , walSeries := range v {
mSeries , created , err := h . getOrCreateWithID ( walSeries . Ref , walSeries . Labels . Hash ( ) , walSeries . Labels )
2021-08-03 12:14:26 +00:00
if err != nil {
seriesCreationErr = err
break Outer
}
2021-11-06 10:10:04 +00:00
if chunks . HeadSeriesRef ( h . lastSeriesID . Load ( ) ) < walSeries . Ref {
h . lastSeriesID . Store ( uint64 ( walSeries . Ref ) )
2021-08-03 14:33:54 +00:00
}
2022-08-17 13:53:57 +00:00
if ! created {
multiRef [ walSeries . Ref ] = mSeries . ref
2021-08-03 14:33:54 +00:00
}
2021-08-03 12:14:26 +00:00
2023-03-07 16:41:33 +00:00
idx := uint64 ( mSeries . ref ) % uint64 ( concurrency )
2022-08-17 13:53:57 +00:00
processors [ idx ] . input <- walSubsetProcessorInputItem { walSeriesRef : walSeries . Ref , existingSeries : mSeries }
2021-08-03 12:14:26 +00:00
}
seriesPool . Put ( v )
case [ ] record . RefSample :
samples := v
2022-09-15 07:06:57 +00:00
minValidTime := h . minValidTime . Load ( )
2021-08-03 12:14:26 +00:00
// We split up the samples into chunks of 5000 samples or less.
// With O(300 * #cores) in-flight sample batches, large scrapes could otherwise
// cause thousands of very large in flight buffers occupying large amounts
// of unused memory.
for len ( samples ) > 0 {
m := 5000
if len ( samples ) < m {
m = len ( samples )
}
2023-03-07 16:41:33 +00:00
for i := 0 ; i < concurrency ; i ++ {
2022-09-20 14:13:30 +00:00
if shards [ i ] == nil {
shards [ i ] = processors [ i ] . reuseBuf ( )
}
2021-08-03 12:14:26 +00:00
}
for _ , sam := range samples [ : m ] {
2022-09-15 07:06:57 +00:00
if sam . T < minValidTime {
continue // Before minValidTime: discard.
}
2021-08-03 12:14:26 +00:00
if r , ok := multiRef [ sam . Ref ] ; ok {
sam . Ref = r
}
2023-03-07 16:41:33 +00:00
mod := uint64 ( sam . Ref ) % uint64 ( concurrency )
2021-08-03 12:14:26 +00:00
shards [ mod ] = append ( shards [ mod ] , sam )
}
2023-03-07 16:41:33 +00:00
for i := 0 ; i < concurrency ; i ++ {
2022-09-20 14:13:30 +00:00
if len ( shards [ i ] ) > 0 {
processors [ i ] . input <- walSubsetProcessorInputItem { samples : shards [ i ] }
shards [ i ] = nil
}
2021-08-03 12:14:26 +00:00
}
samples = samples [ m : ]
}
samplesPool . Put ( v )
case [ ] tombstones . Stone :
for _ , s := range v {
for _ , itv := range s . Intervals {
if itv . Maxt < h . minValidTime . Load ( ) {
continue
}
2021-11-06 10:10:04 +00:00
if m := h . series . getByID ( chunks . HeadSeriesRef ( s . Ref ) ) ; m == nil {
2021-08-03 12:14:26 +00:00
unknownRefs . Inc ( )
continue
}
2023-04-09 07:08:40 +00:00
h . tombstones . AddInterval ( s . Ref , itv )
2021-08-03 12:14:26 +00:00
}
}
tstonesPool . Put ( v )
case [ ] record . RefExemplar :
for _ , e := range v {
exemplarsInput <- e
}
exemplarsPool . Put ( v )
2022-08-29 12:08:36 +00:00
case [ ] record . RefHistogramSample :
2021-11-30 16:19:06 +00:00
samples := v
2022-10-11 16:35:35 +00:00
minValidTime := h . minValidTime . Load ( )
2021-11-30 16:19:06 +00:00
// We split up the samples into chunks of 5000 samples or less.
// With O(300 * #cores) in-flight sample batches, large scrapes could otherwise
// cause thousands of very large in flight buffers occupying large amounts
// of unused memory.
for len ( samples ) > 0 {
m := 5000
if len ( samples ) < m {
m = len ( samples )
}
2023-03-07 16:41:33 +00:00
for i := 0 ; i < concurrency ; i ++ {
2022-10-05 20:14:49 +00:00
if histogramShards [ i ] == nil {
histogramShards [ i ] = processors [ i ] . reuseHistogramBuf ( )
}
2021-11-30 16:19:06 +00:00
}
for _ , sam := range samples [ : m ] {
2022-10-11 16:35:35 +00:00
if sam . T < minValidTime {
continue // Before minValidTime: discard.
}
2021-11-30 16:19:06 +00:00
if r , ok := multiRef [ sam . Ref ] ; ok {
sam . Ref = r
}
2023-03-07 16:41:33 +00:00
mod := uint64 ( sam . Ref ) % uint64 ( concurrency )
2022-12-28 08:55:07 +00:00
histogramShards [ mod ] = append ( histogramShards [ mod ] , histogramRecord { ref : sam . Ref , t : sam . T , h : sam . H } )
2021-11-30 16:19:06 +00:00
}
2023-03-07 16:41:33 +00:00
for i := 0 ; i < concurrency ; i ++ {
2022-10-11 16:35:35 +00:00
if len ( histogramShards [ i ] ) > 0 {
2022-10-05 20:14:49 +00:00
processors [ i ] . input <- walSubsetProcessorInputItem { histogramSamples : histogramShards [ i ] }
histogramShards [ i ] = nil
}
2021-11-30 16:19:06 +00:00
}
samples = samples [ m : ]
2021-08-11 12:08:48 +00:00
}
histogramsPool . Put ( v )
2022-12-28 08:55:07 +00:00
case [ ] record . RefFloatHistogramSample :
samples := v
minValidTime := h . minValidTime . Load ( )
// We split up the samples into chunks of 5000 samples or less.
// With O(300 * #cores) in-flight sample batches, large scrapes could otherwise
// cause thousands of very large in flight buffers occupying large amounts
// of unused memory.
for len ( samples ) > 0 {
m := 5000
if len ( samples ) < m {
m = len ( samples )
}
2023-03-07 16:41:33 +00:00
for i := 0 ; i < concurrency ; i ++ {
2022-12-28 08:55:07 +00:00
if histogramShards [ i ] == nil {
histogramShards [ i ] = processors [ i ] . reuseHistogramBuf ( )
}
}
for _ , sam := range samples [ : m ] {
if sam . T < minValidTime {
continue // Before minValidTime: discard.
}
if r , ok := multiRef [ sam . Ref ] ; ok {
sam . Ref = r
}
2023-03-07 16:41:33 +00:00
mod := uint64 ( sam . Ref ) % uint64 ( concurrency )
2022-12-28 08:55:07 +00:00
histogramShards [ mod ] = append ( histogramShards [ mod ] , histogramRecord { ref : sam . Ref , t : sam . T , fh : sam . FH } )
}
2023-03-07 16:41:33 +00:00
for i := 0 ; i < concurrency ; i ++ {
2022-12-28 08:55:07 +00:00
if len ( histogramShards [ i ] ) > 0 {
processors [ i ] . input <- walSubsetProcessorInputItem { histogramSamples : histogramShards [ i ] }
histogramShards [ i ] = nil
}
}
samples = samples [ m : ]
}
floatHistogramsPool . Put ( v )
2022-07-19 08:58:52 +00:00
case [ ] record . RefMetadata :
for _ , m := range v {
2023-04-09 07:08:40 +00:00
s := h . series . getByID ( m . Ref )
2022-07-19 08:58:52 +00:00
if s == nil {
unknownMetadataRefs . Inc ( )
continue
}
2022-08-17 10:02:28 +00:00
s . meta = & metadata . Metadata {
2023-11-22 14:39:21 +00:00
Type : record . ToMetricType ( m . Type ) ,
2022-07-19 08:58:52 +00:00
Unit : m . Unit ,
Help : m . Help ,
}
}
metadataPool . Put ( v )
2021-08-03 12:14:26 +00:00
default :
panic ( fmt . Errorf ( "unexpected decoded type: %T" , d ) )
}
}
if decodeErr != nil {
return decodeErr
}
if seriesCreationErr != nil {
// Drain the channel to unblock the goroutine.
for range decoded {
}
return seriesCreationErr
}
// Signal termination to each worker and wait for it to close its output channel.
2023-03-07 16:41:33 +00:00
for i := 0 ; i < concurrency ; i ++ {
2021-11-25 08:06:14 +00:00
processors [ i ] . closeAndDrain ( )
2021-08-03 12:14:26 +00:00
}
close ( exemplarsInput )
wg . Wait ( )
2023-11-16 18:54:41 +00:00
if err := r . Err ( ) ; err != nil {
return fmt . Errorf ( "read records: %w" , err )
2021-08-03 12:14:26 +00:00
}
2022-08-10 15:54:37 +00:00
if unknownRefs . Load ( ) + unknownExemplarRefs . Load ( ) + unknownHistogramRefs . Load ( ) + unknownMetadataRefs . Load ( ) > 0 {
level . Warn ( h . logger ) . Log (
"msg" , "Unknown series references" ,
"samples" , unknownRefs . Load ( ) ,
"exemplars" , unknownExemplarRefs . Load ( ) ,
"histograms" , unknownHistogramRefs . Load ( ) ,
"metadata" , unknownMetadataRefs . Load ( ) ,
)
2021-08-03 12:14:26 +00:00
}
2022-08-17 13:53:57 +00:00
if count := mmapOverlappingChunks . Load ( ) ; count > 0 {
level . Info ( h . logger ) . Log ( "msg" , "Overlapping m-map chunks on duplicate series records" , "count" , count )
2021-10-01 11:30:22 +00:00
}
2021-08-03 12:14:26 +00:00
return nil
}
2021-11-25 08:06:14 +00:00
// resetSeriesWithMMappedChunks is only used during the WAL replay.
2022-09-20 17:05:50 +00:00
func ( h * Head ) resetSeriesWithMMappedChunks ( mSeries * memSeries , mmc , oooMmc [ ] * mmappedChunk , walSeriesRef chunks . HeadSeriesRef ) ( overlapped bool ) {
2022-08-17 13:53:57 +00:00
if mSeries . ref != walSeriesRef {
// Checking if the new m-mapped chunks overlap with the already existing ones.
if len ( mSeries . mmappedChunks ) > 0 && len ( mmc ) > 0 {
if overlapsClosedInterval (
mSeries . mmappedChunks [ 0 ] . minTime ,
mSeries . mmappedChunks [ len ( mSeries . mmappedChunks ) - 1 ] . maxTime ,
mmc [ 0 ] . minTime ,
mmc [ len ( mmc ) - 1 ] . maxTime ,
) {
level . Debug ( h . logger ) . Log (
"msg" , "M-mapped chunks overlap on a duplicate series record" ,
"series" , mSeries . lset . String ( ) ,
"oldref" , mSeries . ref ,
"oldmint" , mSeries . mmappedChunks [ 0 ] . minTime ,
"oldmaxt" , mSeries . mmappedChunks [ len ( mSeries . mmappedChunks ) - 1 ] . maxTime ,
"newref" , walSeriesRef ,
"newmint" , mmc [ 0 ] . minTime ,
"newmaxt" , mmc [ len ( mmc ) - 1 ] . maxTime ,
)
overlapped = true
}
}
}
2022-09-20 17:05:50 +00:00
h . metrics . chunksCreated . Add ( float64 ( len ( mmc ) + len ( oooMmc ) ) )
2021-08-11 10:13:17 +00:00
h . metrics . chunksRemoved . Add ( float64 ( len ( mSeries . mmappedChunks ) ) )
2022-09-20 17:05:50 +00:00
h . metrics . chunks . Add ( float64 ( len ( mmc ) + len ( oooMmc ) - len ( mSeries . mmappedChunks ) ) )
2023-01-17 15:42:19 +00:00
if mSeries . ooo != nil {
h . metrics . chunksRemoved . Add ( float64 ( len ( mSeries . ooo . oooMmappedChunks ) ) )
h . metrics . chunks . Sub ( float64 ( len ( mSeries . ooo . oooMmappedChunks ) ) )
}
2021-08-11 10:13:17 +00:00
mSeries . mmappedChunks = mmc
2022-12-28 10:19:41 +00:00
if len ( oooMmc ) == 0 {
mSeries . ooo = nil
} else {
if mSeries . ooo == nil {
mSeries . ooo = & memSeriesOOOFields { }
}
* mSeries . ooo = memSeriesOOOFields { oooMmappedChunks : oooMmc }
}
2021-08-11 10:13:17 +00:00
// Cache the last mmapped chunk time, so we can skip calling append() for samples it will reject.
if len ( mmc ) == 0 {
mSeries . mmMaxTime = math . MinInt64
} else {
mSeries . mmMaxTime = mmc [ len ( mmc ) - 1 ] . maxTime
h . updateMinMaxTime ( mmc [ 0 ] . minTime , mSeries . mmMaxTime )
}
2022-09-20 17:05:50 +00:00
if len ( oooMmc ) != 0 {
// Mint and maxt can be in any chunk, they are not sorted.
mint , maxt := int64 ( math . MaxInt64 ) , int64 ( math . MinInt64 )
for _ , ch := range oooMmc {
if ch . minTime < mint {
mint = ch . minTime
}
if ch . maxTime > maxt {
maxt = ch . maxTime
}
}
h . updateMinOOOMaxOOOTime ( mint , maxt )
}
2021-11-25 08:06:14 +00:00
// Any samples replayed till now would already be compacted. Resetting the head chunk.
mSeries . nextAt = 0
Use a linked list for memSeries.headChunk (#11818)
Currently memSeries holds a single head chunk in-memory and a slice of mmapped chunks.
When append() is called on memSeries it might decide that a new headChunk is needed to use for given append() call.
If that happens it will first mmap existing head chunk and only after that happens it will create a new empty headChunk and continue appending
our sample to it.
Since appending samples uses write lock on memSeries no other read or write can happen until any append is completed.
When we have an append() that must create a new head chunk the whole memSeries is blocked until mmapping of existing head chunk finishes.
Mmapping itself uses a lock as it needs to be serialised, which means that the more chunks to mmap we have the longer each chunk might wait
for it to be mmapped.
If there's enough chunks that require mmapping some memSeries will be locked for long enough that it will start affecting
queries and scrapes.
Queries might timeout, since by default they have a 2 minute timeout set.
Scrapes will be blocked inside append() call, which means there will be a gap between samples. This will first affect range queries
or calls using rate() and such, since the time range requested in the query might have too few samples to calculate anything.
To avoid this we need to remove mmapping from append path, since mmapping is blocking.
But this means that when we cut a new head chunk we need to keep the old one around, so we can mmap it later.
This change makes memSeries.headChunk a linked list, memSeries.headChunk still points to the 'open' head chunk that receives new samples,
while older, yet to be mmapped, chunks are linked to it.
Mmapping is done on a schedule by iterating all memSeries one by one. Thanks to this we control when mmapping is done, since we trigger
it manually, which reduces the risk that it will have to compete for mmap locks with other chunks.
Signed-off-by: Łukasz Mierzwa <l.mierzwa@gmail.com>
2023-07-31 09:10:24 +00:00
mSeries . headChunks = nil
2021-11-25 08:06:14 +00:00
mSeries . app = nil
2022-08-17 13:53:57 +00:00
return
2021-11-25 08:06:14 +00:00
}
type walSubsetProcessor struct {
2022-10-05 20:14:49 +00:00
input chan walSubsetProcessorInputItem
2021-11-30 16:19:06 +00:00
output chan [ ] record . RefSample
2022-12-28 08:55:07 +00:00
histogramsOutput chan [ ] histogramRecord
2021-11-25 08:06:14 +00:00
}
2022-08-17 13:53:57 +00:00
type walSubsetProcessorInputItem struct {
2022-10-05 20:14:49 +00:00
samples [ ] record . RefSample
2022-12-28 08:55:07 +00:00
histogramSamples [ ] histogramRecord
2022-10-05 20:14:49 +00:00
existingSeries * memSeries
walSeriesRef chunks . HeadSeriesRef
2022-08-17 13:53:57 +00:00
}
2021-11-25 08:06:14 +00:00
func ( wp * walSubsetProcessor ) setup ( ) {
2022-08-17 13:53:57 +00:00
wp . input = make ( chan walSubsetProcessorInputItem , 300 )
2021-11-25 08:06:14 +00:00
wp . output = make ( chan [ ] record . RefSample , 300 )
2022-12-28 08:55:07 +00:00
wp . histogramsOutput = make ( chan [ ] histogramRecord , 300 )
2021-11-25 08:06:14 +00:00
}
func ( wp * walSubsetProcessor ) closeAndDrain ( ) {
close ( wp . input )
for range wp . output {
}
2021-11-30 16:19:06 +00:00
for range wp . histogramsOutput {
}
2021-11-25 08:06:14 +00:00
}
// If there is a buffer in the output chan, return it for reuse, otherwise return nil.
func ( wp * walSubsetProcessor ) reuseBuf ( ) [ ] record . RefSample {
select {
case buf := <- wp . output :
return buf [ : 0 ]
default :
}
return nil
2021-08-10 09:23:31 +00:00
}
2021-11-30 16:19:06 +00:00
// If there is a buffer in the output chan, return it for reuse, otherwise return nil.
2022-12-28 08:55:07 +00:00
func ( wp * walSubsetProcessor ) reuseHistogramBuf ( ) [ ] histogramRecord {
2021-11-30 16:19:06 +00:00
select {
case buf := <- wp . histogramsOutput :
return buf [ : 0 ]
default :
}
return nil
2021-08-11 10:13:17 +00:00
}
// processWALSamples adds the samples it receives to the head and passes
// the buffer received to an output channel for reuse.
// Samples before the minValidTime timestamp are discarded.
2022-10-05 20:14:49 +00:00
func ( wp * walSubsetProcessor ) processWALSamples ( h * Head , mmappedChunks , oooMmappedChunks map [ chunks . HeadSeriesRef ] [ ] * mmappedChunk ) ( unknownRefs , unknownHistogramRefs , mmapOverlappingChunks uint64 ) {
2021-11-25 08:06:14 +00:00
defer close ( wp . output )
2021-11-30 16:19:06 +00:00
defer close ( wp . histogramsOutput )
2021-08-03 12:14:26 +00:00
2021-11-25 08:06:14 +00:00
minValidTime := h . minValidTime . Load ( )
2021-08-03 12:14:26 +00:00
mint , maxt := int64 ( math . MaxInt64 ) , int64 ( math . MinInt64 )
2023-05-25 20:12:32 +00:00
appendChunkOpts := chunkOpts {
chunkDiskMapper : h . chunkDiskMapper ,
chunkRange : h . chunkRange . Load ( ) ,
samplesPerChunk : h . opts . SamplesPerChunk ,
}
2021-08-03 12:14:26 +00:00
2022-08-17 13:53:57 +00:00
for in := range wp . input {
if in . existingSeries != nil {
mmc := mmappedChunks [ in . walSeriesRef ]
2022-09-20 17:05:50 +00:00
oooMmc := oooMmappedChunks [ in . walSeriesRef ]
if h . resetSeriesWithMMappedChunks ( in . existingSeries , mmc , oooMmc , in . walSeriesRef ) {
2022-08-17 13:53:57 +00:00
mmapOverlappingChunks ++
}
continue
}
2021-08-03 12:14:26 +00:00
2022-08-17 13:53:57 +00:00
for _ , s := range in . samples {
2021-08-10 09:23:31 +00:00
ms := h . series . getByID ( s . Ref )
2021-08-03 12:14:26 +00:00
if ms == nil {
2021-08-10 09:23:31 +00:00
unknownRefs ++
continue
}
if s . T <= ms . mmMaxTime {
continue
2021-08-03 12:14:26 +00:00
}
2023-05-25 20:12:32 +00:00
if _ , chunkCreated := ms . append ( s . T , s . V , 0 , appendChunkOpts ) ; chunkCreated {
2021-08-03 12:14:26 +00:00
h . metrics . chunksCreated . Inc ( )
h . metrics . chunks . Inc ( )
Use a linked list for memSeries.headChunk (#11818)
Currently memSeries holds a single head chunk in-memory and a slice of mmapped chunks.
When append() is called on memSeries it might decide that a new headChunk is needed to use for given append() call.
If that happens it will first mmap existing head chunk and only after that happens it will create a new empty headChunk and continue appending
our sample to it.
Since appending samples uses write lock on memSeries no other read or write can happen until any append is completed.
When we have an append() that must create a new head chunk the whole memSeries is blocked until mmapping of existing head chunk finishes.
Mmapping itself uses a lock as it needs to be serialised, which means that the more chunks to mmap we have the longer each chunk might wait
for it to be mmapped.
If there's enough chunks that require mmapping some memSeries will be locked for long enough that it will start affecting
queries and scrapes.
Queries might timeout, since by default they have a 2 minute timeout set.
Scrapes will be blocked inside append() call, which means there will be a gap between samples. This will first affect range queries
or calls using rate() and such, since the time range requested in the query might have too few samples to calculate anything.
To avoid this we need to remove mmapping from append path, since mmapping is blocking.
But this means that when we cut a new head chunk we need to keep the old one around, so we can mmap it later.
This change makes memSeries.headChunk a linked list, memSeries.headChunk still points to the 'open' head chunk that receives new samples,
while older, yet to be mmapped, chunks are linked to it.
Mmapping is done on a schedule by iterating all memSeries one by one. Thanks to this we control when mmapping is done, since we trigger
it manually, which reduces the risk that it will have to compete for mmap locks with other chunks.
Signed-off-by: Łukasz Mierzwa <l.mierzwa@gmail.com>
2023-07-31 09:10:24 +00:00
_ = ms . mmapChunks ( h . chunkDiskMapper )
2021-08-03 12:14:26 +00:00
}
if s . T > maxt {
maxt = s . T
}
if s . T < mint {
mint = s . T
}
}
2021-11-25 08:06:14 +00:00
select {
2022-08-17 13:53:57 +00:00
case wp . output <- in . samples :
default :
2021-11-25 08:06:14 +00:00
}
2022-10-05 20:14:49 +00:00
for _ , s := range in . histogramSamples {
2022-12-28 08:55:07 +00:00
if s . t < minValidTime {
2022-10-05 20:14:49 +00:00
continue
}
2022-12-28 08:55:07 +00:00
ms := h . series . getByID ( s . ref )
2022-10-05 20:14:49 +00:00
if ms == nil {
unknownHistogramRefs ++
continue
}
2022-12-28 08:55:07 +00:00
if s . t <= ms . mmMaxTime {
2022-10-05 20:14:49 +00:00
continue
}
2022-12-28 08:55:07 +00:00
var chunkCreated bool
if s . h != nil {
2023-05-25 20:12:32 +00:00
_ , chunkCreated = ms . appendHistogram ( s . t , s . h , 0 , appendChunkOpts )
2022-12-28 08:55:07 +00:00
} else {
2023-05-25 20:12:32 +00:00
_ , chunkCreated = ms . appendFloatHistogram ( s . t , s . fh , 0 , appendChunkOpts )
2022-12-28 08:55:07 +00:00
}
if chunkCreated {
2022-10-05 20:14:49 +00:00
h . metrics . chunksCreated . Inc ( )
h . metrics . chunks . Inc ( )
}
2022-12-28 08:55:07 +00:00
if s . t > maxt {
maxt = s . t
2022-10-05 20:14:49 +00:00
}
2022-12-28 08:55:07 +00:00
if s . t < mint {
mint = s . t
2022-10-05 20:14:49 +00:00
}
}
select {
case wp . histogramsOutput <- in . histogramSamples :
default :
}
2021-11-25 08:06:14 +00:00
}
2022-08-17 13:53:57 +00:00
h . updateMinMaxTime ( mint , maxt )
2022-10-05 20:14:49 +00:00
return unknownRefs , unknownHistogramRefs , mmapOverlappingChunks
2021-11-25 08:06:14 +00:00
}
2022-10-10 15:08:46 +00:00
func ( h * Head ) loadWBL ( r * wlog . Reader , multiRef map [ chunks . HeadSeriesRef ] chunks . HeadSeriesRef , lastMmapRef chunks . ChunkDiskMapperRef ) ( err error ) {
2022-09-20 17:05:50 +00:00
// Track number of samples, m-map markers, that referenced a series we don't know about
// for error reporting.
var unknownRefs , mmapMarkerUnknownRefs atomic . Uint64
lastSeq , lastOff := lastMmapRef . Unpack ( )
// Start workers that each process samples for a partition of the series ID space.
var (
2023-03-07 16:41:33 +00:00
wg sync . WaitGroup
concurrency = h . opts . WALReplayConcurrency
processors = make ( [ ] wblSubsetProcessor , concurrency )
2022-09-20 17:05:50 +00:00
dec record . Decoder
2023-03-07 16:41:33 +00:00
shards = make ( [ ] [ ] record . RefSample , concurrency )
2022-09-20 17:05:50 +00:00
decodedCh = make ( chan interface { } , 10 )
decodeErr error
samplesPool = sync . Pool {
New : func ( ) interface { } {
return [ ] record . RefSample { }
} ,
}
markersPool = sync . Pool {
New : func ( ) interface { } {
return [ ] record . RefMmapMarker { }
} ,
}
)
defer func ( ) {
// For CorruptionErr ensure to terminate all workers before exiting.
// We also wrap it to identify OOO WBL corruption.
2022-10-10 15:08:46 +00:00
_ , ok := err . ( * wlog . CorruptionErr )
2022-09-20 17:05:50 +00:00
if ok {
err = & errLoadWbl { err : err }
2023-03-07 16:41:33 +00:00
for i := 0 ; i < concurrency ; i ++ {
2022-09-20 17:05:50 +00:00
processors [ i ] . closeAndDrain ( )
}
wg . Wait ( )
}
} ( )
2023-03-07 16:41:33 +00:00
wg . Add ( concurrency )
for i := 0 ; i < concurrency ; i ++ {
2022-09-20 17:05:50 +00:00
processors [ i ] . setup ( )
go func ( wp * wblSubsetProcessor ) {
unknown := wp . processWBLSamples ( h )
unknownRefs . Add ( unknown )
wg . Done ( )
} ( & processors [ i ] )
}
go func ( ) {
defer close ( decodedCh )
for r . Next ( ) {
rec := r . Record ( )
switch dec . Type ( rec ) {
case record . Samples :
samples := samplesPool . Get ( ) . ( [ ] record . RefSample ) [ : 0 ]
samples , err = dec . Samples ( rec , samples )
if err != nil {
2022-10-10 15:08:46 +00:00
decodeErr = & wlog . CorruptionErr {
2023-11-16 18:54:41 +00:00
Err : fmt . Errorf ( "decode samples: %w" , err ) ,
2022-09-20 17:05:50 +00:00
Segment : r . Segment ( ) ,
Offset : r . Offset ( ) ,
}
return
2021-11-30 16:19:06 +00:00
}
2022-09-20 17:05:50 +00:00
decodedCh <- samples
case record . MmapMarkers :
markers := markersPool . Get ( ) . ( [ ] record . RefMmapMarker ) [ : 0 ]
markers , err = dec . MmapMarkers ( rec , markers )
if err != nil {
2022-10-10 15:08:46 +00:00
decodeErr = & wlog . CorruptionErr {
2023-11-16 18:54:41 +00:00
Err : fmt . Errorf ( "decode mmap markers: %w" , err ) ,
2022-09-20 17:05:50 +00:00
Segment : r . Segment ( ) ,
Offset : r . Offset ( ) ,
}
return
2021-11-30 16:19:06 +00:00
}
2022-09-20 17:05:50 +00:00
decodedCh <- markers
default :
// Noop.
}
}
} ( )
// The records are always replayed from the oldest to the newest.
for d := range decodedCh {
switch v := d . ( type ) {
case [ ] record . RefSample :
samples := v
// We split up the samples into parts of 5000 samples or less.
// With O(300 * #cores) in-flight sample batches, large scrapes could otherwise
// cause thousands of very large in flight buffers occupying large amounts
// of unused memory.
for len ( samples ) > 0 {
m := 5000
if len ( samples ) < m {
m = len ( samples )
2021-11-30 16:19:06 +00:00
}
2023-03-07 16:41:33 +00:00
for i := 0 ; i < concurrency ; i ++ {
2023-09-12 14:26:02 +00:00
if shards [ i ] == nil {
shards [ i ] = processors [ i ] . reuseBuf ( )
}
2021-11-30 16:19:06 +00:00
}
2022-09-20 17:05:50 +00:00
for _ , sam := range samples [ : m ] {
if r , ok := multiRef [ sam . Ref ] ; ok {
sam . Ref = r
}
2023-03-07 16:41:33 +00:00
mod := uint64 ( sam . Ref ) % uint64 ( concurrency )
2022-09-20 17:05:50 +00:00
shards [ mod ] = append ( shards [ mod ] , sam )
2021-11-30 16:19:06 +00:00
}
2023-03-07 16:41:33 +00:00
for i := 0 ; i < concurrency ; i ++ {
2023-09-12 14:26:02 +00:00
if len ( shards [ i ] ) > 0 {
2023-09-12 19:31:10 +00:00
processors [ i ] . input <- wblSubsetProcessorInputItem { samples : shards [ i ] }
2023-09-12 14:26:02 +00:00
shards [ i ] = nil
}
2021-11-30 16:19:06 +00:00
}
2022-09-20 17:05:50 +00:00
samples = samples [ m : ]
2021-08-03 12:14:26 +00:00
}
2022-09-20 17:05:50 +00:00
samplesPool . Put ( d )
case [ ] record . RefMmapMarker :
markers := v
for _ , rm := range markers {
seq , off := rm . MmapRef . Unpack ( )
if seq > lastSeq || ( seq == lastSeq && off > lastOff ) {
// This m-map chunk from markers was not present during
// the load of mmapped chunks that happened in the head
// initialization.
2021-11-30 16:19:06 +00:00
continue
}
2022-09-20 17:05:50 +00:00
if r , ok := multiRef [ rm . Ref ] ; ok {
rm . Ref = r
2021-11-30 16:19:06 +00:00
}
2022-09-20 17:05:50 +00:00
ms := h . series . getByID ( rm . Ref )
if ms == nil {
mmapMarkerUnknownRefs . Inc ( )
2021-11-30 16:19:06 +00:00
continue
}
2023-03-07 16:41:33 +00:00
idx := uint64 ( ms . ref ) % uint64 ( concurrency )
2023-09-12 19:31:10 +00:00
processors [ idx ] . input <- wblSubsetProcessorInputItem { mmappedSeries : ms }
2022-09-20 17:05:50 +00:00
}
default :
panic ( fmt . Errorf ( "unexpected decodedCh type: %T" , d ) )
}
}
if decodeErr != nil {
return decodeErr
}
// Signal termination to each worker and wait for it to close its output channel.
2023-03-07 16:41:33 +00:00
for i := 0 ; i < concurrency ; i ++ {
2022-09-20 17:05:50 +00:00
processors [ i ] . closeAndDrain ( )
}
wg . Wait ( )
2023-11-16 18:54:41 +00:00
if err := r . Err ( ) ; err != nil {
return fmt . Errorf ( "read records: %w" , err )
2022-09-20 17:05:50 +00:00
}
if unknownRefs . Load ( ) > 0 || mmapMarkerUnknownRefs . Load ( ) > 0 {
level . Warn ( h . logger ) . Log ( "msg" , "Unknown series references for ooo WAL replay" , "samples" , unknownRefs . Load ( ) , "mmap_markers" , mmapMarkerUnknownRefs . Load ( ) )
}
return nil
}
type errLoadWbl struct {
err error
}
func ( e errLoadWbl ) Error ( ) string {
return e . err . Error ( )
}
func ( e errLoadWbl ) Cause ( ) error {
return e . err
}
func ( e errLoadWbl ) Unwrap ( ) error {
return e . err
}
type wblSubsetProcessor struct {
2023-09-12 19:31:10 +00:00
input chan wblSubsetProcessorInputItem
2022-09-20 17:05:50 +00:00
output chan [ ] record . RefSample
}
2023-09-12 19:31:10 +00:00
type wblSubsetProcessorInputItem struct {
mmappedSeries * memSeries
samples [ ] record . RefSample
}
2022-09-20 17:05:50 +00:00
func ( wp * wblSubsetProcessor ) setup ( ) {
wp . output = make ( chan [ ] record . RefSample , 300 )
2023-09-12 19:31:10 +00:00
wp . input = make ( chan wblSubsetProcessorInputItem , 300 )
2022-09-20 17:05:50 +00:00
}
func ( wp * wblSubsetProcessor ) closeAndDrain ( ) {
close ( wp . input )
for range wp . output {
}
}
// If there is a buffer in the output chan, return it for reuse, otherwise return nil.
func ( wp * wblSubsetProcessor ) reuseBuf ( ) [ ] record . RefSample {
select {
case buf := <- wp . output :
return buf [ : 0 ]
default :
}
return nil
}
// processWBLSamples adds the samples it receives to the head and passes
// the buffer received to an output channel for reuse.
func ( wp * wblSubsetProcessor ) processWBLSamples ( h * Head ) ( unknownRefs uint64 ) {
defer close ( wp . output )
2022-09-27 08:22:22 +00:00
oooCapMax := h . opts . OutOfOrderCapMax . Load ( )
2022-09-20 17:05:50 +00:00
// We don't check for minValidTime for ooo samples.
mint , maxt := int64 ( math . MaxInt64 ) , int64 ( math . MinInt64 )
2023-09-12 19:31:10 +00:00
for in := range wp . input {
if in . mmappedSeries != nil && in . mmappedSeries . ooo != nil {
// All samples till now have been m-mapped. Hence clear out the headChunk.
// In case some samples slipped through and went into m-map chunks because of changed
// chunk size parameters, we are not taking care of that here.
// TODO(codesome): see if there is a way to avoid duplicate m-map chunks if
// the size of ooo chunk was reduced between restart.
in . mmappedSeries . ooo . oooHeadChunk = nil
continue
}
for _ , s := range in . samples {
2022-09-20 17:05:50 +00:00
ms := h . series . getByID ( s . Ref )
if ms == nil {
unknownRefs ++
continue
}
2022-09-27 08:22:22 +00:00
ok , chunkCreated , _ := ms . insert ( s . T , s . V , h . chunkDiskMapper , oooCapMax )
2022-09-20 17:05:50 +00:00
if chunkCreated {
h . metrics . chunksCreated . Inc ( )
h . metrics . chunks . Inc ( )
}
if ok {
if s . T < mint {
mint = s . T
2021-11-30 16:19:06 +00:00
}
if s . T > maxt {
maxt = s . T
}
2021-08-03 12:14:26 +00:00
}
}
2023-09-12 19:31:10 +00:00
select {
case wp . output <- in . samples :
default :
}
2021-08-03 12:14:26 +00:00
}
2022-09-20 17:05:50 +00:00
h . updateMinOOOMaxOOOTime ( mint , maxt )
return unknownRefs
2021-08-03 12:14:26 +00:00
}
2021-08-06 16:51:01 +00:00
2021-08-11 10:13:17 +00:00
const (
chunkSnapshotRecordTypeSeries uint8 = 1
chunkSnapshotRecordTypeTombstones uint8 = 2
2021-08-30 14:04:38 +00:00
chunkSnapshotRecordTypeExemplars uint8 = 3
2021-08-11 10:13:17 +00:00
)
type chunkSnapshotRecord struct {
2023-07-05 09:44:13 +00:00
ref chunks . HeadSeriesRef
lset labels . Labels
mc * memChunk
lastValue float64
lastHistogramValue * histogram . Histogram
lastFloatHistogramValue * histogram . FloatHistogram
2021-08-11 10:13:17 +00:00
}
func ( s * memSeries ) encodeToSnapshotRecord ( b [ ] byte ) [ ] byte {
buf := encoding . Encbuf { B : b }
buf . PutByte ( chunkSnapshotRecordTypeSeries )
2021-11-06 10:10:04 +00:00
buf . PutBE64 ( uint64 ( s . ref ) )
2022-07-26 14:42:00 +00:00
record . EncodeLabels ( & buf , s . lset )
2022-09-27 08:22:22 +00:00
buf . PutBE64int64 ( 0 ) // Backwards-compatibility; was chunkRange but now unused.
2021-08-11 10:13:17 +00:00
s . Lock ( )
Use a linked list for memSeries.headChunk (#11818)
Currently memSeries holds a single head chunk in-memory and a slice of mmapped chunks.
When append() is called on memSeries it might decide that a new headChunk is needed to use for given append() call.
If that happens it will first mmap existing head chunk and only after that happens it will create a new empty headChunk and continue appending
our sample to it.
Since appending samples uses write lock on memSeries no other read or write can happen until any append is completed.
When we have an append() that must create a new head chunk the whole memSeries is blocked until mmapping of existing head chunk finishes.
Mmapping itself uses a lock as it needs to be serialised, which means that the more chunks to mmap we have the longer each chunk might wait
for it to be mmapped.
If there's enough chunks that require mmapping some memSeries will be locked for long enough that it will start affecting
queries and scrapes.
Queries might timeout, since by default they have a 2 minute timeout set.
Scrapes will be blocked inside append() call, which means there will be a gap between samples. This will first affect range queries
or calls using rate() and such, since the time range requested in the query might have too few samples to calculate anything.
To avoid this we need to remove mmapping from append path, since mmapping is blocking.
But this means that when we cut a new head chunk we need to keep the old one around, so we can mmap it later.
This change makes memSeries.headChunk a linked list, memSeries.headChunk still points to the 'open' head chunk that receives new samples,
while older, yet to be mmapped, chunks are linked to it.
Mmapping is done on a schedule by iterating all memSeries one by one. Thanks to this we control when mmapping is done, since we trigger
it manually, which reduces the risk that it will have to compete for mmap locks with other chunks.
Signed-off-by: Łukasz Mierzwa <l.mierzwa@gmail.com>
2023-07-31 09:10:24 +00:00
if s . headChunks == nil {
2021-08-11 10:13:17 +00:00
buf . PutUvarint ( 0 )
} else {
Use a linked list for memSeries.headChunk (#11818)
Currently memSeries holds a single head chunk in-memory and a slice of mmapped chunks.
When append() is called on memSeries it might decide that a new headChunk is needed to use for given append() call.
If that happens it will first mmap existing head chunk and only after that happens it will create a new empty headChunk and continue appending
our sample to it.
Since appending samples uses write lock on memSeries no other read or write can happen until any append is completed.
When we have an append() that must create a new head chunk the whole memSeries is blocked until mmapping of existing head chunk finishes.
Mmapping itself uses a lock as it needs to be serialised, which means that the more chunks to mmap we have the longer each chunk might wait
for it to be mmapped.
If there's enough chunks that require mmapping some memSeries will be locked for long enough that it will start affecting
queries and scrapes.
Queries might timeout, since by default they have a 2 minute timeout set.
Scrapes will be blocked inside append() call, which means there will be a gap between samples. This will first affect range queries
or calls using rate() and such, since the time range requested in the query might have too few samples to calculate anything.
To avoid this we need to remove mmapping from append path, since mmapping is blocking.
But this means that when we cut a new head chunk we need to keep the old one around, so we can mmap it later.
This change makes memSeries.headChunk a linked list, memSeries.headChunk still points to the 'open' head chunk that receives new samples,
while older, yet to be mmapped, chunks are linked to it.
Mmapping is done on a schedule by iterating all memSeries one by one. Thanks to this we control when mmapping is done, since we trigger
it manually, which reduces the risk that it will have to compete for mmap locks with other chunks.
Signed-off-by: Łukasz Mierzwa <l.mierzwa@gmail.com>
2023-07-31 09:10:24 +00:00
enc := s . headChunks . chunk . Encoding ( )
2021-08-11 10:13:17 +00:00
buf . PutUvarint ( 1 )
Use a linked list for memSeries.headChunk (#11818)
Currently memSeries holds a single head chunk in-memory and a slice of mmapped chunks.
When append() is called on memSeries it might decide that a new headChunk is needed to use for given append() call.
If that happens it will first mmap existing head chunk and only after that happens it will create a new empty headChunk and continue appending
our sample to it.
Since appending samples uses write lock on memSeries no other read or write can happen until any append is completed.
When we have an append() that must create a new head chunk the whole memSeries is blocked until mmapping of existing head chunk finishes.
Mmapping itself uses a lock as it needs to be serialised, which means that the more chunks to mmap we have the longer each chunk might wait
for it to be mmapped.
If there's enough chunks that require mmapping some memSeries will be locked for long enough that it will start affecting
queries and scrapes.
Queries might timeout, since by default they have a 2 minute timeout set.
Scrapes will be blocked inside append() call, which means there will be a gap between samples. This will first affect range queries
or calls using rate() and such, since the time range requested in the query might have too few samples to calculate anything.
To avoid this we need to remove mmapping from append path, since mmapping is blocking.
But this means that when we cut a new head chunk we need to keep the old one around, so we can mmap it later.
This change makes memSeries.headChunk a linked list, memSeries.headChunk still points to the 'open' head chunk that receives new samples,
while older, yet to be mmapped, chunks are linked to it.
Mmapping is done on a schedule by iterating all memSeries one by one. Thanks to this we control when mmapping is done, since we trigger
it manually, which reduces the risk that it will have to compete for mmap locks with other chunks.
Signed-off-by: Łukasz Mierzwa <l.mierzwa@gmail.com>
2023-07-31 09:10:24 +00:00
buf . PutBE64int64 ( s . headChunks . minTime )
buf . PutBE64int64 ( s . headChunks . maxTime )
2023-07-05 09:44:13 +00:00
buf . PutByte ( byte ( enc ) )
Use a linked list for memSeries.headChunk (#11818)
Currently memSeries holds a single head chunk in-memory and a slice of mmapped chunks.
When append() is called on memSeries it might decide that a new headChunk is needed to use for given append() call.
If that happens it will first mmap existing head chunk and only after that happens it will create a new empty headChunk and continue appending
our sample to it.
Since appending samples uses write lock on memSeries no other read or write can happen until any append is completed.
When we have an append() that must create a new head chunk the whole memSeries is blocked until mmapping of existing head chunk finishes.
Mmapping itself uses a lock as it needs to be serialised, which means that the more chunks to mmap we have the longer each chunk might wait
for it to be mmapped.
If there's enough chunks that require mmapping some memSeries will be locked for long enough that it will start affecting
queries and scrapes.
Queries might timeout, since by default they have a 2 minute timeout set.
Scrapes will be blocked inside append() call, which means there will be a gap between samples. This will first affect range queries
or calls using rate() and such, since the time range requested in the query might have too few samples to calculate anything.
To avoid this we need to remove mmapping from append path, since mmapping is blocking.
But this means that when we cut a new head chunk we need to keep the old one around, so we can mmap it later.
This change makes memSeries.headChunk a linked list, memSeries.headChunk still points to the 'open' head chunk that receives new samples,
while older, yet to be mmapped, chunks are linked to it.
Mmapping is done on a schedule by iterating all memSeries one by one. Thanks to this we control when mmapping is done, since we trigger
it manually, which reduces the risk that it will have to compete for mmap locks with other chunks.
Signed-off-by: Łukasz Mierzwa <l.mierzwa@gmail.com>
2023-07-31 09:10:24 +00:00
buf . PutUvarintBytes ( s . headChunks . chunk . Bytes ( ) )
2023-07-05 09:44:13 +00:00
switch enc {
case chunkenc . EncXOR :
// Backwards compatibility for old sampleBuf which had last 4 samples.
for i := 0 ; i < 3 ; i ++ {
buf . PutBE64int64 ( 0 )
buf . PutBEFloat64 ( 0 )
}
2022-09-27 14:02:05 +00:00
buf . PutBE64int64 ( 0 )
2023-07-05 09:44:13 +00:00
buf . PutBEFloat64 ( s . lastValue )
case chunkenc . EncHistogram :
record . EncodeHistogram ( & buf , s . lastHistogramValue )
default : // chunkenc.FloatHistogram.
record . EncodeFloatHistogram ( & buf , s . lastFloatHistogramValue )
2021-08-11 10:13:17 +00:00
}
}
s . Unlock ( )
return buf . Get ( )
}
2022-07-26 14:42:00 +00:00
func decodeSeriesFromChunkSnapshot ( d * record . Decoder , b [ ] byte ) ( csr chunkSnapshotRecord , err error ) {
2021-08-11 10:13:17 +00:00
dec := encoding . Decbuf { B : b }
if flag := dec . Byte ( ) ; flag != chunkSnapshotRecordTypeSeries {
2023-11-14 13:04:31 +00:00
return csr , fmt . Errorf ( "invalid record type %x" , flag )
2021-08-11 10:13:17 +00:00
}
2021-11-06 10:10:04 +00:00
csr . ref = chunks . HeadSeriesRef ( dec . Be64 ( ) )
2021-08-11 10:13:17 +00:00
// The label set written to the disk is already sorted.
2022-07-26 14:42:00 +00:00
// TODO: figure out why DecodeLabels calls Sort(), and perhaps remove it.
csr . lset = d . DecodeLabels ( & dec )
2021-08-11 10:13:17 +00:00
2022-09-27 08:22:22 +00:00
_ = dec . Be64int64 ( ) // Was chunkRange but now unused.
2021-08-11 10:13:17 +00:00
if dec . Uvarint ( ) == 0 {
return
}
csr . mc = & memChunk { }
csr . mc . minTime = dec . Be64int64 ( )
csr . mc . maxTime = dec . Be64int64 ( )
enc := chunkenc . Encoding ( dec . Byte ( ) )
// The underlying bytes gets re-used later, so make a copy.
chunkBytes := dec . UvarintBytes ( )
chunkBytesCopy := make ( [ ] byte , len ( chunkBytes ) )
copy ( chunkBytesCopy , chunkBytes )
chk , err := chunkenc . FromData ( enc , chunkBytesCopy )
if err != nil {
2023-11-16 18:54:41 +00:00
return csr , fmt . Errorf ( "chunk from data: %w" , err )
2021-08-11 10:13:17 +00:00
}
csr . mc . chunk = chk
2023-07-05 09:44:13 +00:00
switch enc {
case chunkenc . EncXOR :
// Backwards-compatibility for old sampleBuf which had last 4 samples.
for i := 0 ; i < 3 ; i ++ {
_ = dec . Be64int64 ( )
_ = dec . Be64Float64 ( )
}
2022-09-27 14:02:05 +00:00
_ = dec . Be64int64 ( )
2023-07-05 09:44:13 +00:00
csr . lastValue = dec . Be64Float64 ( )
case chunkenc . EncHistogram :
csr . lastHistogramValue = & histogram . Histogram { }
record . DecodeHistogram ( & dec , csr . lastHistogramValue )
default : // chunkenc.FloatHistogram.
csr . lastFloatHistogramValue = & histogram . FloatHistogram { }
record . DecodeFloatHistogram ( & dec , csr . lastFloatHistogramValue )
2021-08-11 10:13:17 +00:00
}
err = dec . Err ( )
if err != nil && len ( dec . B ) > 0 {
2023-11-14 13:04:31 +00:00
err = fmt . Errorf ( "unexpected %d bytes left in entry" , len ( dec . B ) )
2021-08-11 10:13:17 +00:00
}
return
}
func encodeTombstonesToSnapshotRecord ( tr tombstones . Reader ) ( [ ] byte , error ) {
buf := encoding . Encbuf { }
buf . PutByte ( chunkSnapshotRecordTypeTombstones )
b , err := tombstones . Encode ( tr )
if err != nil {
2023-11-16 18:54:41 +00:00
return nil , fmt . Errorf ( "encode tombstones: %w" , err )
2021-08-11 10:13:17 +00:00
}
buf . PutUvarintBytes ( b )
return buf . Get ( ) , nil
}
func decodeTombstonesSnapshotRecord ( b [ ] byte ) ( tombstones . Reader , error ) {
dec := encoding . Decbuf { B : b }
if flag := dec . Byte ( ) ; flag != chunkSnapshotRecordTypeTombstones {
2023-11-14 13:04:31 +00:00
return nil , fmt . Errorf ( "invalid record type %x" , flag )
2021-08-11 10:13:17 +00:00
}
tr , err := tombstones . Decode ( dec . UvarintBytes ( ) )
2023-11-16 18:54:41 +00:00
if err != nil {
return tr , fmt . Errorf ( "decode tombstones: %w" , err )
}
return tr , nil
2021-08-11 10:13:17 +00:00
}
const chunkSnapshotPrefix = "chunk_snapshot."
// ChunkSnapshot creates a snapshot of all the series and tombstones in the head.
// It deletes the old chunk snapshots if the chunk snapshot creation is successful.
//
// The chunk snapshot is stored in a directory named chunk_snapshot.N.M and is written
// using the WAL package. N is the last WAL segment present during snapshotting and
// M is the offset in segment N upto which data was written.
2021-08-30 14:04:38 +00:00
//
// The snapshot first contains all series (each in individual records and not sorted), followed by
// tombstones (a single record), and finally exemplars (>= 1 record). Exemplars are in the order they
// were written to the circular buffer.
2021-08-11 10:13:17 +00:00
func ( h * Head ) ChunkSnapshot ( ) ( * ChunkSnapshotStats , error ) {
if h . wal == nil {
// If we are not storing any WAL, does not make sense to take a snapshot too.
level . Warn ( h . logger ) . Log ( "msg" , "skipping chunk snapshotting as WAL is disabled" )
return & ChunkSnapshotStats { } , nil
}
h . chunkSnapshotMtx . Lock ( )
defer h . chunkSnapshotMtx . Unlock ( )
stats := & ChunkSnapshotStats { }
wlast , woffset , err := h . wal . LastSegmentAndOffset ( )
2023-11-16 18:54:41 +00:00
if err != nil && ! errors . Is ( err , record . ErrNotFound ) {
return stats , fmt . Errorf ( "get last wal segment and offset: %w" , err )
2021-08-11 10:13:17 +00:00
}
_ , cslast , csoffset , err := LastChunkSnapshot ( h . opts . ChunkDirRoot )
2023-11-16 18:54:41 +00:00
if err != nil && ! errors . Is ( err , record . ErrNotFound ) {
return stats , fmt . Errorf ( "find last chunk snapshot: %w" , err )
2021-08-11 10:13:17 +00:00
}
if wlast == cslast && woffset == csoffset {
// Nothing has been written to the WAL/Head since the last snapshot.
return stats , nil
}
2021-09-08 14:23:44 +00:00
snapshotName := chunkSnapshotDir ( wlast , woffset )
2021-08-11 10:13:17 +00:00
cpdir := filepath . Join ( h . opts . ChunkDirRoot , snapshotName )
cpdirtmp := cpdir + ".tmp"
stats . Dir = cpdir
2021-10-22 08:06:44 +00:00
if err := os . MkdirAll ( cpdirtmp , 0 o777 ) ; err != nil {
2023-11-16 18:54:41 +00:00
return stats , fmt . Errorf ( "create chunk snapshot dir: %w" , err )
2021-08-11 10:13:17 +00:00
}
2023-07-11 12:57:57 +00:00
cp , err := wlog . New ( nil , nil , cpdirtmp , h . wal . CompressionType ( ) )
2021-08-11 10:13:17 +00:00
if err != nil {
2023-11-16 18:54:41 +00:00
return stats , fmt . Errorf ( "open chunk snapshot: %w" , err )
2021-08-11 10:13:17 +00:00
}
// Ensures that an early return caused by an error doesn't leave any tmp files.
defer func ( ) {
cp . Close ( )
os . RemoveAll ( cpdirtmp )
} ( )
var (
buf [ ] byte
recs [ ] [ ] byte
)
2021-08-30 14:04:38 +00:00
// Add all series to the snapshot.
2021-08-11 10:13:17 +00:00
stripeSize := h . series . size
for i := 0 ; i < stripeSize ; i ++ {
h . series . locks [ i ] . RLock ( )
for _ , s := range h . series . series [ i ] {
start := len ( buf )
buf = s . encodeToSnapshotRecord ( buf )
if len ( buf [ start : ] ) == 0 {
continue // All contents discarded.
}
recs = append ( recs , buf [ start : ] )
// Flush records in 10 MB increments.
if len ( buf ) > 10 * 1024 * 1024 {
if err := cp . Log ( recs ... ) ; err != nil {
h . series . locks [ i ] . RUnlock ( )
2023-11-16 18:54:41 +00:00
return stats , fmt . Errorf ( "flush records: %w" , err )
2021-08-11 10:13:17 +00:00
}
buf , recs = buf [ : 0 ] , recs [ : 0 ]
}
}
stats . TotalSeries += len ( h . series . series [ i ] )
h . series . locks [ i ] . RUnlock ( )
}
// Add tombstones to the snapshot.
tombstonesReader , err := h . Tombstones ( )
if err != nil {
2023-11-16 18:54:41 +00:00
return stats , fmt . Errorf ( "get tombstones: %w" , err )
2021-08-11 10:13:17 +00:00
}
rec , err := encodeTombstonesToSnapshotRecord ( tombstonesReader )
if err != nil {
2023-11-16 18:54:41 +00:00
return stats , fmt . Errorf ( "encode tombstones: %w" , err )
2021-08-11 10:13:17 +00:00
}
recs = append ( recs , rec )
2021-08-30 14:04:38 +00:00
// Flush remaining series records and tombstones.
2021-08-11 10:13:17 +00:00
if err := cp . Log ( recs ... ) ; err != nil {
2023-11-16 18:54:41 +00:00
return stats , fmt . Errorf ( "flush records: %w" , err )
2021-08-11 10:13:17 +00:00
}
2021-08-30 14:04:38 +00:00
buf = buf [ : 0 ]
// Add exemplars in the snapshot.
// We log in batches, with each record having upto 10000 exemplars.
// Assuming 100 bytes (overestimate) per exemplar, that's ~1MB.
maxExemplarsPerRecord := 10000
batch := make ( [ ] record . RefExemplar , 0 , maxExemplarsPerRecord )
enc := record . Encoder { }
flushExemplars := func ( ) error {
if len ( batch ) == 0 {
return nil
}
buf = buf [ : 0 ]
encbuf := encoding . Encbuf { B : buf }
encbuf . PutByte ( chunkSnapshotRecordTypeExemplars )
enc . EncodeExemplarsIntoBuffer ( batch , & encbuf )
if err := cp . Log ( encbuf . Get ( ) ) ; err != nil {
2023-11-16 18:54:41 +00:00
return fmt . Errorf ( "log exemplars: %w" , err )
2021-08-30 14:04:38 +00:00
}
buf , batch = buf [ : 0 ] , batch [ : 0 ]
return nil
}
err = h . exemplars . IterateExemplars ( func ( seriesLabels labels . Labels , e exemplar . Exemplar ) error {
if len ( batch ) >= maxExemplarsPerRecord {
if err := flushExemplars ( ) ; err != nil {
2023-11-16 18:54:41 +00:00
return fmt . Errorf ( "flush exemplars: %w" , err )
2021-08-30 14:04:38 +00:00
}
}
ms := h . series . getByHash ( seriesLabels . Hash ( ) , seriesLabels )
if ms == nil {
// It is possible that exemplar refers to some old series. We discard such exemplars.
return nil
}
batch = append ( batch , record . RefExemplar {
Ref : ms . ref ,
T : e . Ts ,
V : e . Value ,
Labels : e . Labels ,
} )
return nil
} )
if err != nil {
2023-11-16 18:54:41 +00:00
return stats , fmt . Errorf ( "iterate exemplars: %w" , err )
2021-08-30 14:04:38 +00:00
}
// Flush remaining exemplars.
if err := flushExemplars ( ) ; err != nil {
2023-11-16 18:54:41 +00:00
return stats , fmt . Errorf ( "flush exemplars at the end: %w" , err )
2021-08-30 14:04:38 +00:00
}
2021-08-11 10:13:17 +00:00
if err := cp . Close ( ) ; err != nil {
2023-11-16 18:54:41 +00:00
return stats , fmt . Errorf ( "close chunk snapshot: %w" , err )
2021-08-11 10:13:17 +00:00
}
if err := fileutil . Replace ( cpdirtmp , cpdir ) ; err != nil {
2023-11-16 18:54:41 +00:00
return stats , fmt . Errorf ( "rename chunk snapshot directory: %w" , err )
2021-08-11 10:13:17 +00:00
}
2021-09-08 14:23:44 +00:00
if err := DeleteChunkSnapshots ( h . opts . ChunkDirRoot , wlast , woffset ) ; err != nil {
2021-08-11 10:13:17 +00:00
// Leftover old chunk snapshots do not cause problems down the line beyond
// occupying disk space.
// They will just be ignored since a higher chunk snapshot exists.
level . Error ( h . logger ) . Log ( "msg" , "delete old chunk snapshots" , "err" , err )
}
return stats , nil
}
2021-09-08 14:23:44 +00:00
func chunkSnapshotDir ( wlast , woffset int ) string {
return fmt . Sprintf ( chunkSnapshotPrefix + "%06d.%010d" , wlast , woffset )
}
2021-08-11 10:13:17 +00:00
func ( h * Head ) performChunkSnapshot ( ) error {
level . Info ( h . logger ) . Log ( "msg" , "creating chunk snapshot" )
startTime := time . Now ( )
stats , err := h . ChunkSnapshot ( )
elapsed := time . Since ( startTime )
if err == nil {
level . Info ( h . logger ) . Log ( "msg" , "chunk snapshot complete" , "duration" , elapsed . String ( ) , "num_series" , stats . TotalSeries , "dir" , stats . Dir )
}
2023-11-16 18:54:41 +00:00
if err != nil {
return fmt . Errorf ( "chunk snapshot: %w" , err )
}
return nil
2021-08-11 10:13:17 +00:00
}
// ChunkSnapshotStats returns stats about a created chunk snapshot.
type ChunkSnapshotStats struct {
TotalSeries int
Dir string
}
// LastChunkSnapshot returns the directory name and index of the most recent chunk snapshot.
// If dir does not contain any chunk snapshots, ErrNotFound is returned.
func LastChunkSnapshot ( dir string ) ( string , int , int , error ) {
2022-04-27 09:24:36 +00:00
files , err := os . ReadDir ( dir )
2021-08-11 10:13:17 +00:00
if err != nil {
return "" , 0 , 0 , err
}
2021-09-08 14:23:44 +00:00
maxIdx , maxOffset := - 1 , - 1
maxFileName := ""
for i := 0 ; i < len ( files ) ; i ++ {
2021-08-11 10:13:17 +00:00
fi := files [ i ]
if ! strings . HasPrefix ( fi . Name ( ) , chunkSnapshotPrefix ) {
continue
}
if ! fi . IsDir ( ) {
2023-11-14 13:04:31 +00:00
return "" , 0 , 0 , fmt . Errorf ( "chunk snapshot %s is not a directory" , fi . Name ( ) )
2021-08-11 10:13:17 +00:00
}
splits := strings . Split ( fi . Name ( ) [ len ( chunkSnapshotPrefix ) : ] , "." )
if len ( splits ) != 2 {
2021-12-08 15:32:14 +00:00
// Chunk snapshots is not in the right format, we do not care about it.
continue
2021-08-11 10:13:17 +00:00
}
idx , err := strconv . Atoi ( splits [ 0 ] )
if err != nil {
continue
}
offset , err := strconv . Atoi ( splits [ 1 ] )
if err != nil {
continue
}
2021-09-08 14:23:44 +00:00
if idx > maxIdx || ( idx == maxIdx && offset > maxOffset ) {
maxIdx , maxOffset = idx , offset
maxFileName = filepath . Join ( dir , fi . Name ( ) )
}
2021-08-11 10:13:17 +00:00
}
2021-09-08 14:23:44 +00:00
if maxFileName == "" {
return "" , 0 , 0 , record . ErrNotFound
2021-08-06 16:51:01 +00:00
}
2021-09-08 14:23:44 +00:00
return maxFileName , maxIdx , maxOffset , nil
2021-08-11 10:13:17 +00:00
}
// DeleteChunkSnapshots deletes all chunk snapshots in a directory below a given index.
func DeleteChunkSnapshots ( dir string , maxIndex , maxOffset int ) error {
2022-04-27 09:24:36 +00:00
files , err := os . ReadDir ( dir )
2021-08-11 10:13:17 +00:00
if err != nil {
return err
}
errs := tsdb_errors . NewMulti ( )
for _ , fi := range files {
if ! strings . HasPrefix ( fi . Name ( ) , chunkSnapshotPrefix ) {
continue
}
splits := strings . Split ( fi . Name ( ) [ len ( chunkSnapshotPrefix ) : ] , "." )
if len ( splits ) != 2 {
continue
}
idx , err := strconv . Atoi ( splits [ 0 ] )
if err != nil {
continue
}
offset , err := strconv . Atoi ( splits [ 1 ] )
if err != nil {
continue
}
2021-09-08 14:23:44 +00:00
if idx < maxIndex || ( idx == maxIndex && offset < maxOffset ) {
2021-08-11 10:13:17 +00:00
if err := os . RemoveAll ( filepath . Join ( dir , fi . Name ( ) ) ) ; err != nil {
errs . Add ( err )
}
}
}
return errs . Err ( )
}
2021-08-17 17:08:16 +00:00
// loadChunkSnapshot replays the chunk snapshot and restores the Head state from it. If there was any error returned,
// it is the responsibility of the caller to clear the contents of the Head.
2021-11-06 10:10:04 +00:00
func ( h * Head ) loadChunkSnapshot ( ) ( int , int , map [ chunks . HeadSeriesRef ] * memSeries , error ) {
2021-08-11 10:13:17 +00:00
dir , snapIdx , snapOffset , err := LastChunkSnapshot ( h . opts . ChunkDirRoot )
if err != nil {
2023-11-16 18:54:41 +00:00
if errors . Is ( err , record . ErrNotFound ) {
2021-08-11 10:13:17 +00:00
return snapIdx , snapOffset , nil , nil
}
2023-11-16 18:54:41 +00:00
return snapIdx , snapOffset , nil , fmt . Errorf ( "find last chunk snapshot: %w" , err )
2021-08-11 10:13:17 +00:00
}
start := time . Now ( )
2022-10-10 15:08:46 +00:00
sr , err := wlog . NewSegmentsReader ( dir )
2021-08-11 10:13:17 +00:00
if err != nil {
2023-11-16 18:54:41 +00:00
return snapIdx , snapOffset , nil , fmt . Errorf ( "open chunk snapshot: %w" , err )
2021-08-11 10:13:17 +00:00
}
defer func ( ) {
if err := sr . Close ( ) ; err != nil {
level . Warn ( h . logger ) . Log ( "msg" , "error while closing the wal segments reader" , "err" , err )
}
} ( )
var (
numSeries = 0
unknownRefs = int64 ( 0 )
2023-03-07 16:41:33 +00:00
concurrency = h . opts . WALReplayConcurrency
2021-08-11 10:13:17 +00:00
wg sync . WaitGroup
2023-03-07 16:41:33 +00:00
recordChan = make ( chan chunkSnapshotRecord , 5 * concurrency )
shardedRefSeries = make ( [ ] map [ chunks . HeadSeriesRef ] * memSeries , concurrency )
errChan = make ( chan error , concurrency )
2021-11-06 10:10:04 +00:00
refSeries map [ chunks . HeadSeriesRef ] * memSeries
2021-08-30 14:04:38 +00:00
exemplarBuf [ ] record . RefExemplar
dec record . Decoder
2021-08-11 10:13:17 +00:00
)
2023-03-07 16:41:33 +00:00
wg . Add ( concurrency )
for i := 0 ; i < concurrency ; i ++ {
2021-08-11 10:13:17 +00:00
go func ( idx int , rc <- chan chunkSnapshotRecord ) {
defer wg . Done ( )
defer func ( ) {
// If there was an error, drain the channel
// to unblock the main thread.
for range rc {
}
} ( )
2021-11-06 10:10:04 +00:00
shardedRefSeries [ idx ] = make ( map [ chunks . HeadSeriesRef ] * memSeries )
2021-08-11 10:13:17 +00:00
localRefSeries := shardedRefSeries [ idx ]
for csr := range rc {
series , _ , err := h . getOrCreateWithID ( csr . ref , csr . lset . Hash ( ) , csr . lset )
if err != nil {
errChan <- err
return
}
localRefSeries [ csr . ref ] = series
2022-08-04 08:09:14 +00:00
for {
seriesID := uint64 ( series . ref )
lastSeriesID := h . lastSeriesID . Load ( )
2022-09-09 11:28:55 +00:00
if lastSeriesID >= seriesID || h . lastSeriesID . CompareAndSwap ( lastSeriesID , seriesID ) {
2022-08-04 08:09:14 +00:00
break
}
2021-08-11 10:13:17 +00:00
}
if csr . mc == nil {
continue
}
series . nextAt = csr . mc . maxTime // This will create a new chunk on append.
Use a linked list for memSeries.headChunk (#11818)
Currently memSeries holds a single head chunk in-memory and a slice of mmapped chunks.
When append() is called on memSeries it might decide that a new headChunk is needed to use for given append() call.
If that happens it will first mmap existing head chunk and only after that happens it will create a new empty headChunk and continue appending
our sample to it.
Since appending samples uses write lock on memSeries no other read or write can happen until any append is completed.
When we have an append() that must create a new head chunk the whole memSeries is blocked until mmapping of existing head chunk finishes.
Mmapping itself uses a lock as it needs to be serialised, which means that the more chunks to mmap we have the longer each chunk might wait
for it to be mmapped.
If there's enough chunks that require mmapping some memSeries will be locked for long enough that it will start affecting
queries and scrapes.
Queries might timeout, since by default they have a 2 minute timeout set.
Scrapes will be blocked inside append() call, which means there will be a gap between samples. This will first affect range queries
or calls using rate() and such, since the time range requested in the query might have too few samples to calculate anything.
To avoid this we need to remove mmapping from append path, since mmapping is blocking.
But this means that when we cut a new head chunk we need to keep the old one around, so we can mmap it later.
This change makes memSeries.headChunk a linked list, memSeries.headChunk still points to the 'open' head chunk that receives new samples,
while older, yet to be mmapped, chunks are linked to it.
Mmapping is done on a schedule by iterating all memSeries one by one. Thanks to this we control when mmapping is done, since we trigger
it manually, which reduces the risk that it will have to compete for mmap locks with other chunks.
Signed-off-by: Łukasz Mierzwa <l.mierzwa@gmail.com>
2023-07-31 09:10:24 +00:00
series . headChunks = csr . mc
2022-09-27 14:02:05 +00:00
series . lastValue = csr . lastValue
2023-07-05 09:44:13 +00:00
series . lastHistogramValue = csr . lastHistogramValue
series . lastFloatHistogramValue = csr . lastFloatHistogramValue
2021-08-11 10:13:17 +00:00
Use a linked list for memSeries.headChunk (#11818)
Currently memSeries holds a single head chunk in-memory and a slice of mmapped chunks.
When append() is called on memSeries it might decide that a new headChunk is needed to use for given append() call.
If that happens it will first mmap existing head chunk and only after that happens it will create a new empty headChunk and continue appending
our sample to it.
Since appending samples uses write lock on memSeries no other read or write can happen until any append is completed.
When we have an append() that must create a new head chunk the whole memSeries is blocked until mmapping of existing head chunk finishes.
Mmapping itself uses a lock as it needs to be serialised, which means that the more chunks to mmap we have the longer each chunk might wait
for it to be mmapped.
If there's enough chunks that require mmapping some memSeries will be locked for long enough that it will start affecting
queries and scrapes.
Queries might timeout, since by default they have a 2 minute timeout set.
Scrapes will be blocked inside append() call, which means there will be a gap between samples. This will first affect range queries
or calls using rate() and such, since the time range requested in the query might have too few samples to calculate anything.
To avoid this we need to remove mmapping from append path, since mmapping is blocking.
But this means that when we cut a new head chunk we need to keep the old one around, so we can mmap it later.
This change makes memSeries.headChunk a linked list, memSeries.headChunk still points to the 'open' head chunk that receives new samples,
while older, yet to be mmapped, chunks are linked to it.
Mmapping is done on a schedule by iterating all memSeries one by one. Thanks to this we control when mmapping is done, since we trigger
it manually, which reduces the risk that it will have to compete for mmap locks with other chunks.
Signed-off-by: Łukasz Mierzwa <l.mierzwa@gmail.com>
2023-07-31 09:10:24 +00:00
app , err := series . headChunks . chunk . Appender ( )
2021-08-11 10:13:17 +00:00
if err != nil {
errChan <- err
return
}
series . app = app
h . updateMinMaxTime ( csr . mc . minTime , csr . mc . maxTime )
}
} ( i , recordChan )
}
2022-10-10 15:08:46 +00:00
r := wlog . NewReader ( sr )
2021-08-11 10:13:17 +00:00
var loopErr error
Outer :
for r . Next ( ) {
select {
case err := <- errChan :
errChan <- err
break Outer
default :
}
rec := r . Record ( )
switch rec [ 0 ] {
case chunkSnapshotRecordTypeSeries :
numSeries ++
2022-07-26 14:42:00 +00:00
csr , err := decodeSeriesFromChunkSnapshot ( & dec , rec )
2021-08-11 10:13:17 +00:00
if err != nil {
2023-11-16 18:54:41 +00:00
loopErr = fmt . Errorf ( "decode series record: %w" , err )
2021-08-11 10:13:17 +00:00
break Outer
}
recordChan <- csr
case chunkSnapshotRecordTypeTombstones :
tr , err := decodeTombstonesSnapshotRecord ( rec )
if err != nil {
2023-11-16 18:54:41 +00:00
loopErr = fmt . Errorf ( "decode tombstones: %w" , err )
2021-08-11 10:13:17 +00:00
break Outer
}
2021-11-06 10:10:04 +00:00
if err = tr . Iter ( func ( ref storage . SeriesRef , ivs tombstones . Intervals ) error {
2021-08-11 10:13:17 +00:00
h . tombstones . AddInterval ( ref , ivs ... )
return nil
} ) ; err != nil {
2023-11-16 18:54:41 +00:00
loopErr = fmt . Errorf ( "iterate tombstones: %w" , err )
2021-08-11 10:13:17 +00:00
break Outer
}
2021-08-30 14:04:38 +00:00
case chunkSnapshotRecordTypeExemplars :
// Exemplars are at the end of snapshot. So all series are loaded at this point.
if len ( refSeries ) == 0 {
close ( recordChan )
wg . Wait ( )
2021-11-06 10:10:04 +00:00
refSeries = make ( map [ chunks . HeadSeriesRef ] * memSeries , numSeries )
2021-08-30 14:04:38 +00:00
for _ , shard := range shardedRefSeries {
for k , v := range shard {
refSeries [ k ] = v
}
}
}
2021-10-05 05:21:25 +00:00
if ! h . opts . EnableExemplarStorage || h . opts . MaxExemplars . Load ( ) <= 0 {
// Exemplar storage is disabled.
continue Outer
}
2021-08-30 14:04:38 +00:00
decbuf := encoding . Decbuf { B : rec [ 1 : ] }
exemplarBuf = exemplarBuf [ : 0 ]
exemplarBuf , err = dec . ExemplarsFromBuffer ( & decbuf , exemplarBuf )
if err != nil {
2023-11-16 18:54:41 +00:00
loopErr = fmt . Errorf ( "exemplars from buffer: %w" , err )
2021-08-30 14:04:38 +00:00
break Outer
}
for _ , e := range exemplarBuf {
ms , ok := refSeries [ e . Ref ]
if ! ok {
unknownRefs ++
continue
}
if err := h . exemplars . AddExemplar ( ms . lset , exemplar . Exemplar {
Labels : e . Labels ,
Value : e . V ,
Ts : e . T ,
} ) ; err != nil {
2023-11-16 18:54:41 +00:00
loopErr = fmt . Errorf ( "add exemplar: %w" , err )
2021-08-30 14:04:38 +00:00
break Outer
}
}
2021-08-17 17:08:16 +00:00
default :
// This is a record type we don't understand. It is either and old format from earlier versions,
// or a new format and the code was rolled back to old version.
2023-11-14 13:04:31 +00:00
loopErr = fmt . Errorf ( "unsupported snapshot record type 0b%b" , rec [ 0 ] )
2021-08-30 14:04:38 +00:00
break Outer
2021-08-06 16:51:01 +00:00
}
}
2021-08-30 14:04:38 +00:00
if len ( refSeries ) == 0 {
close ( recordChan )
wg . Wait ( )
2021-08-11 10:13:17 +00:00
}
close ( errChan )
2023-11-16 18:54:41 +00:00
merr := tsdb_errors . NewMulti ( )
if loopErr != nil {
merr . Add ( fmt . Errorf ( "decode loop: %w" , loopErr ) )
}
2021-08-11 10:13:17 +00:00
for err := range errChan {
2023-11-16 18:54:41 +00:00
merr . Add ( fmt . Errorf ( "record processing: %w" , err ) )
2021-08-11 10:13:17 +00:00
}
if err := merr . Err ( ) ; err != nil {
return - 1 , - 1 , nil , err
}
2023-11-16 18:54:41 +00:00
if err := r . Err ( ) ; err != nil {
return - 1 , - 1 , nil , fmt . Errorf ( "read records: %w" , err )
2021-08-17 17:08:16 +00:00
}
2021-08-30 14:04:38 +00:00
if len ( refSeries ) == 0 {
// We had no exemplar record, so we have to build the map here.
2021-11-06 10:10:04 +00:00
refSeries = make ( map [ chunks . HeadSeriesRef ] * memSeries , numSeries )
2021-08-30 14:04:38 +00:00
for _ , shard := range shardedRefSeries {
for k , v := range shard {
refSeries [ k ] = v
}
2021-08-11 10:13:17 +00:00
}
}
elapsed := time . Since ( start )
level . Info ( h . logger ) . Log ( "msg" , "chunk snapshot loaded" , "dir" , dir , "num_series" , numSeries , "duration" , elapsed . String ( ) )
if unknownRefs > 0 {
level . Warn ( h . logger ) . Log ( "msg" , "unknown series references during chunk snapshot replay" , "count" , unknownRefs )
}
return snapIdx , snapOffset , refSeries , nil
}