milvus/internal/datanode/compaction/mix_compactor.go
yihao.dai 7730b910b9
enhance: Decouple compaction from shard (#33138)
Decouple compaction from shard, remove dependencies on shards (e.g.
SyncSegments, injection).

issue: https://github.com/milvus-io/milvus/issues/32809

---------

Signed-off-by: bigsheeper <yihao.dai@zilliz.com>
2024-05-24 09:07:41 +08:00

543 lines
16 KiB
Go

// Licensed to the LF AI & Data foundation under one
// or more contributor license agreements. See the NOTICE file
// distributed with this work for additional information
// regarding copyright ownership. The ASF licenses this file
// to you 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 compaction
import (
"context"
"fmt"
sio "io"
"strconv"
"time"
"github.com/cockroachdb/errors"
"github.com/samber/lo"
"go.opentelemetry.io/otel"
"go.uber.org/zap"
"github.com/milvus-io/milvus-proto/go-api/v2/commonpb"
"github.com/milvus-io/milvus/internal/datanode/allocator"
"github.com/milvus-io/milvus/internal/datanode/io"
iter "github.com/milvus-io/milvus/internal/datanode/iterators"
"github.com/milvus-io/milvus/internal/metastore/kv/binlog"
"github.com/milvus-io/milvus/internal/proto/datapb"
"github.com/milvus-io/milvus/internal/storage"
"github.com/milvus-io/milvus/pkg/log"
"github.com/milvus-io/milvus/pkg/metrics"
"github.com/milvus-io/milvus/pkg/util/funcutil"
"github.com/milvus-io/milvus/pkg/util/paramtable"
"github.com/milvus-io/milvus/pkg/util/timerecord"
"github.com/milvus-io/milvus/pkg/util/tsoutil"
"github.com/milvus-io/milvus/pkg/util/typeutil"
)
// for MixCompaction only
type mixCompactionTask struct {
binlogIO io.BinlogIO
allocator.Allocator
currentTs typeutil.Timestamp
plan *datapb.CompactionPlan
ctx context.Context
cancel context.CancelFunc
done chan struct{}
tr *timerecord.TimeRecorder
}
// make sure compactionTask implements compactor interface
var _ Compactor = (*mixCompactionTask)(nil)
func NewMixCompactionTask(
ctx context.Context,
binlogIO io.BinlogIO,
alloc allocator.Allocator,
plan *datapb.CompactionPlan,
) *mixCompactionTask {
ctx1, cancel := context.WithCancel(ctx)
return &mixCompactionTask{
ctx: ctx1,
cancel: cancel,
binlogIO: binlogIO,
Allocator: alloc,
plan: plan,
tr: timerecord.NewTimeRecorder("mix compaction"),
currentTs: tsoutil.GetCurrentTime(),
done: make(chan struct{}, 1),
}
}
func (t *mixCompactionTask) Complete() {
t.done <- struct{}{}
}
func (t *mixCompactionTask) Stop() {
t.cancel()
<-t.done
}
func (t *mixCompactionTask) GetPlanID() typeutil.UniqueID {
return t.plan.GetPlanID()
}
func (t *mixCompactionTask) GetChannelName() string {
return t.plan.GetChannel()
}
// return num rows of all segment compaction from
func (t *mixCompactionTask) getNumRows() int64 {
numRows := int64(0)
for _, binlog := range t.plan.SegmentBinlogs {
if len(binlog.GetFieldBinlogs()) > 0 {
for _, ct := range binlog.GetFieldBinlogs()[0].GetBinlogs() {
numRows += ct.GetEntriesNum()
}
}
}
return numRows
}
func (t *mixCompactionTask) mergeDeltalogs(ctx context.Context, dpaths map[typeutil.UniqueID][]string) (map[interface{}]typeutil.Timestamp, error) {
t.tr.RecordSpan()
ctx, span := otel.Tracer(typeutil.DataNodeRole).Start(ctx, "mergeDeltalogs")
defer span.End()
log := log.With(zap.Int64("planID", t.GetPlanID()))
pk2ts := make(map[interface{}]typeutil.Timestamp)
if len(dpaths) == 0 {
log.Info("compact with no deltalogs, skip merge deltalogs")
return pk2ts, nil
}
allIters := make([]*iter.DeltalogIterator, 0)
for segID, paths := range dpaths {
if len(paths) == 0 {
continue
}
blobs, err := t.binlogIO.Download(ctx, paths)
if err != nil {
log.Warn("compact wrong, fail to download deltalogs",
zap.Int64("segment", segID),
zap.Strings("path", paths),
zap.Error(err))
return nil, err
}
allIters = append(allIters, iter.NewDeltalogIterator(blobs, nil))
}
for _, deltaIter := range allIters {
for deltaIter.HasNext() {
labeled, _ := deltaIter.Next()
ts := labeled.GetTimestamp()
if lastTs, ok := pk2ts[labeled.GetPk().GetValue()]; ok && lastTs > ts {
ts = lastTs
}
pk2ts[labeled.GetPk().GetValue()] = ts
}
}
log.Info("compact mergeDeltalogs end",
zap.Int("deleted pk counts", len(pk2ts)),
zap.Duration("elapse", t.tr.RecordSpan()))
return pk2ts, nil
}
func (t *mixCompactionTask) statSerializeWrite(ctx context.Context, writer *SegmentWriter, finalRowCount int64) (*datapb.FieldBinlog, error) {
ctx, span := otel.Tracer(typeutil.DataNodeRole).Start(ctx, "statslog serializeWrite")
defer span.End()
sblob, err := writer.Finish(finalRowCount)
if err != nil {
return nil, err
}
logID, err := t.AllocOne()
if err != nil {
return nil, err
}
key, _ := binlog.BuildLogPath(storage.StatsBinlog, writer.GetCollectionID(), writer.GetPartitionID(), writer.GetSegmentID(), writer.GetPkID(), logID)
kvs := map[string][]byte{key: sblob.GetValue()}
statFieldLog := &datapb.FieldBinlog{
FieldID: writer.GetPkID(),
Binlogs: []*datapb.Binlog{
{
LogSize: int64(len(sblob.GetValue())),
MemorySize: int64(len(sblob.GetValue())),
LogPath: key,
EntriesNum: finalRowCount,
},
},
}
if err := t.binlogIO.Upload(ctx, kvs); err != nil {
log.Warn("failed to upload insert log", zap.Error(err))
return nil, err
}
return statFieldLog, nil
}
func (t *mixCompactionTask) serializeWrite(ctx context.Context, writer *SegmentWriter) (kvs map[string][]byte, fieldBinlogs map[int64]*datapb.FieldBinlog, err error) {
_, span := otel.Tracer(typeutil.DataNodeRole).Start(ctx, "serializeWrite")
defer span.End()
blobs, tr, err := writer.SerializeYield()
startID, _, err := t.Alloc(uint32(len(blobs)))
if err != nil {
return nil, nil, err
}
kvs = make(map[string][]byte)
fieldBinlogs = make(map[int64]*datapb.FieldBinlog)
for i := range blobs {
// Blob Key is generated by Serialize from int64 fieldID in collection schema, which won't raise error in ParseInt
fID, _ := strconv.ParseInt(blobs[i].GetKey(), 10, 64)
key, _ := binlog.BuildLogPath(storage.InsertBinlog, writer.GetCollectionID(), writer.GetPartitionID(), writer.GetSegmentID(), fID, startID+int64(i))
kvs[key] = blobs[i].GetValue()
fieldBinlogs[fID] = &datapb.FieldBinlog{
FieldID: fID,
Binlogs: []*datapb.Binlog{
{
LogSize: int64(len(blobs[i].GetValue())),
MemorySize: blobs[i].GetMemorySize(),
LogPath: key,
EntriesNum: blobs[i].RowNum,
TimestampFrom: tr.GetMinTimestamp(),
TimestampTo: tr.GetMaxTimestamp(),
},
},
}
}
return
}
func (t *mixCompactionTask) merge(
ctx context.Context,
binlogPaths [][]string,
delta map[interface{}]typeutil.Timestamp,
writer *SegmentWriter,
) (*datapb.CompactionSegment, error) {
_ = t.tr.RecordSpan()
ctx, span := otel.Tracer(typeutil.DataNodeRole).Start(ctx, "CompactMerge")
defer span.End()
log := log.With(zap.Int64("planID", t.GetPlanID()), zap.Int64("compactTo segment", writer.GetSegmentID()))
var (
syncBatchCount int // binlog batch count
remainingRowCount int64 // the number of remaining entities
expiredRowCount int64 // the number of expired entities
unflushedRowCount int64 = 0
// All binlog meta of a segment
allBinlogs = make(map[typeutil.UniqueID]*datapb.FieldBinlog)
)
isValueDeleted := func(v *storage.Value) bool {
ts, ok := delta[v.PK.GetValue()]
// insert task and delete task has the same ts when upsert
// here should be < instead of <=
// to avoid the upsert data to be deleted after compact
if ok && uint64(v.Timestamp) < ts {
return true
}
return false
}
downloadTimeCost := time.Duration(0)
serWriteTimeCost := time.Duration(0)
uploadTimeCost := time.Duration(0)
for _, paths := range binlogPaths {
log := log.With(zap.Strings("paths", paths))
downloadStart := time.Now()
allValues, err := t.binlogIO.Download(ctx, paths)
if err != nil {
log.Warn("compact wrong, fail to download insertLogs", zap.Error(err))
}
downloadTimeCost += time.Since(downloadStart)
blobs := lo.Map(allValues, func(v []byte, i int) *storage.Blob {
return &storage.Blob{Key: paths[i], Value: v}
})
iter, err := storage.NewBinlogDeserializeReader(blobs, writer.GetPkID())
if err != nil {
log.Warn("compact wrong, failed to new insert binlogs reader", zap.Error(err))
return nil, err
}
for {
err := iter.Next()
if err != nil {
if err == sio.EOF {
break
} else {
log.Warn("compact wrong, failed to iter through data", zap.Error(err))
return nil, err
}
}
v := iter.Value()
if isValueDeleted(v) {
continue
}
// Filtering expired entity
if t.isExpiredEntity(typeutil.Timestamp(v.Timestamp)) {
expiredRowCount++
continue
}
err = writer.Write(v)
if err != nil {
log.Warn("compact wrong, failed to writer row", zap.Error(err))
return nil, err
}
unflushedRowCount++
remainingRowCount++
if (unflushedRowCount+1)%100 == 0 && writer.IsFull() {
serWriteStart := time.Now()
kvs, partialBinlogs, err := t.serializeWrite(ctx, writer)
if err != nil {
log.Warn("compact wrong, failed to serialize writer", zap.Error(err))
return nil, err
}
serWriteTimeCost += time.Since(serWriteStart)
uploadStart := time.Now()
if err := t.binlogIO.Upload(ctx, kvs); err != nil {
log.Warn("compact wrong, failed to upload kvs", zap.Error(err))
}
uploadTimeCost += time.Since(uploadStart)
mergeFieldBinlogs(allBinlogs, partialBinlogs)
syncBatchCount++
unflushedRowCount = 0
}
}
}
if !writer.IsEmpty() {
serWriteStart := time.Now()
kvs, partialBinlogs, err := t.serializeWrite(ctx, writer)
if err != nil {
log.Warn("compact wrong, failed to serialize writer", zap.Error(err))
return nil, err
}
serWriteTimeCost += time.Since(serWriteStart)
uploadStart := time.Now()
if err := t.binlogIO.Upload(ctx, kvs); err != nil {
log.Warn("compact wrong, failed to upload kvs", zap.Error(err))
}
uploadTimeCost += time.Since(uploadStart)
mergeFieldBinlogs(allBinlogs, partialBinlogs)
syncBatchCount++
}
serWriteStart := time.Now()
sPath, err := t.statSerializeWrite(ctx, writer, remainingRowCount)
if err != nil {
log.Warn("compact wrong, failed to serialize write segment stats",
zap.Int64("remaining row count", remainingRowCount), zap.Error(err))
return nil, err
}
serWriteTimeCost += time.Since(serWriteStart)
pack := &datapb.CompactionSegment{
SegmentID: writer.GetSegmentID(),
InsertLogs: lo.Values(allBinlogs),
Field2StatslogPaths: []*datapb.FieldBinlog{sPath},
NumOfRows: remainingRowCount,
Channel: t.plan.GetChannel(),
}
totalElapse := t.tr.RecordSpan()
log.Info("compact merge end",
zap.Int64("remaining row count", remainingRowCount),
zap.Int64("expired entities", expiredRowCount),
zap.Int("binlog batch count", syncBatchCount),
zap.Duration("download binlogs elapse", downloadTimeCost),
zap.Duration("upload binlogs elapse", uploadTimeCost),
zap.Duration("serWrite elapse", serWriteTimeCost),
zap.Duration("deRead elapse", totalElapse-serWriteTimeCost-downloadTimeCost-uploadTimeCost),
zap.Duration("total elapse", totalElapse))
return pack, nil
}
func mergeFieldBinlogs(base, paths map[typeutil.UniqueID]*datapb.FieldBinlog) {
for fID, fpath := range paths {
if _, ok := base[fID]; !ok {
base[fID] = &datapb.FieldBinlog{FieldID: fID, Binlogs: make([]*datapb.Binlog, 0)}
}
base[fID].Binlogs = append(base[fID].Binlogs, fpath.GetBinlogs()...)
}
}
func (t *mixCompactionTask) Compact() (*datapb.CompactionPlanResult, error) {
durInQueue := t.tr.RecordSpan()
compactStart := time.Now()
ctx, span := otel.Tracer(typeutil.DataNodeRole).Start(t.ctx, fmt.Sprintf("MixCompact-%d", t.GetPlanID()))
defer span.End()
if len(t.plan.GetSegmentBinlogs()) < 1 {
log.Warn("compact wrong, there's no segments in segment binlogs", zap.Int64("planID", t.plan.GetPlanID()))
return nil, errors.New("compaction plan is illegal")
}
collectionID := t.plan.GetSegmentBinlogs()[0].GetCollectionID()
partitionID := t.plan.GetSegmentBinlogs()[0].GetPartitionID()
log := log.Ctx(ctx).With(zap.Int64("planID", t.plan.GetPlanID()),
zap.Int64("collectionID", collectionID),
zap.Int64("partitionID", partitionID),
zap.Int32("timeout in seconds", t.plan.GetTimeoutInSeconds()))
if ok := funcutil.CheckCtxValid(ctx); !ok {
log.Warn("compact wrong, task context done or timeout")
return nil, ctx.Err()
}
ctxTimeout, cancelAll := context.WithTimeout(ctx, time.Duration(t.plan.GetTimeoutInSeconds())*time.Second)
defer cancelAll()
log.Info("compact start")
targetSegID, err := t.AllocOne()
if err != nil {
log.Warn("compact wrong, unable to allocate segmentID", zap.Error(err))
return nil, err
}
previousRowCount := t.getNumRows()
writer, err := NewSegmentWriter(t.plan.GetSchema(), previousRowCount, targetSegID, partitionID, collectionID)
if err != nil {
log.Warn("compact wrong, unable to init segment writer", zap.Error(err))
return nil, err
}
segIDs := lo.Map(t.plan.GetSegmentBinlogs(), func(binlogs *datapb.CompactionSegmentBinlogs, _ int) int64 {
return binlogs.GetSegmentID()
})
if err := binlog.DecompressCompactionBinlogs(t.plan.GetSegmentBinlogs()); err != nil {
log.Warn("compact wrong, fail to decompress compaction binlogs", zap.Error(err))
return nil, err
}
deltaPaths := make(map[typeutil.UniqueID][]string) // segmentID to deltalog paths
allPath := make([][]string, 0) // group by binlog batch
for _, s := range t.plan.GetSegmentBinlogs() {
// Get the batch count of field binlog files from non-empty segment
// each segment might contain different batches
var binlogBatchCount int
for _, b := range s.GetFieldBinlogs() {
if b != nil {
binlogBatchCount = len(b.GetBinlogs())
break
}
}
if binlogBatchCount == 0 {
log.Warn("compacting empty segment", zap.Int64("segmentID", s.GetSegmentID()))
continue
}
for idx := 0; idx < binlogBatchCount; idx++ {
var batchPaths []string
for _, f := range s.GetFieldBinlogs() {
batchPaths = append(batchPaths, f.GetBinlogs()[idx].GetLogPath())
}
allPath = append(allPath, batchPaths)
}
deltaPaths[s.GetSegmentID()] = []string{}
for _, d := range s.GetDeltalogs() {
for _, l := range d.GetBinlogs() {
deltaPaths[s.GetSegmentID()] = append(deltaPaths[s.GetSegmentID()], l.GetLogPath())
}
}
}
// Unable to deal with all empty segments cases, so return error
if len(allPath) == 0 {
log.Warn("compact wrong, all segments' binlogs are empty")
return nil, errors.New("illegal compaction plan")
}
deltaPk2Ts, err := t.mergeDeltalogs(ctxTimeout, deltaPaths)
if err != nil {
log.Warn("compact wrong, fail to merge deltalogs", zap.Error(err))
return nil, err
}
compactToSeg, err := t.merge(ctxTimeout, allPath, deltaPk2Ts, writer)
if err != nil {
log.Warn("compact wrong, fail to merge", zap.Error(err))
return nil, err
}
log.Info("compact done",
zap.Int64("compact to segment", targetSegID),
zap.Int64s("compact from segments", segIDs),
zap.Int("num of binlog paths", len(compactToSeg.GetInsertLogs())),
zap.Int("num of stats paths", 1),
zap.Int("num of delta paths", len(compactToSeg.GetDeltalogs())),
zap.Duration("compact elapse", time.Since(compactStart)),
)
metrics.DataNodeCompactionLatency.WithLabelValues(fmt.Sprint(paramtable.GetNodeID()), t.plan.GetType().String()).Observe(float64(t.tr.ElapseSpan().Milliseconds()))
metrics.DataNodeCompactionLatencyInQueue.WithLabelValues(fmt.Sprint(paramtable.GetNodeID())).Observe(float64(durInQueue.Milliseconds()))
planResult := &datapb.CompactionPlanResult{
State: commonpb.CompactionState_Completed,
PlanID: t.GetPlanID(),
Channel: t.GetChannelName(),
Segments: []*datapb.CompactionSegment{compactToSeg},
Type: t.plan.GetType(),
}
return planResult, nil
}
func (t *mixCompactionTask) GetCollection() typeutil.UniqueID {
// The length of SegmentBinlogs is checked before task enqueueing.
return t.plan.GetSegmentBinlogs()[0].GetCollectionID()
}
func (t *mixCompactionTask) isExpiredEntity(ts typeutil.Timestamp) bool {
now := t.currentTs
// entity expire is not enabled if duration <= 0
if t.plan.GetCollectionTtl() <= 0 {
return false
}
entityT, _ := tsoutil.ParseTS(ts)
nowT, _ := tsoutil.ParseTS(now)
return entityT.Add(time.Duration(t.plan.GetCollectionTtl())).Before(nowT)
}