Zhen Ye 7c575a18b0
enhance: support AckSyncUp for broadcaster, and enable it in truncate api (#46313)
issue: #43897
also for issue: #46166

add ack_sync_up flag into broadcast message header, which indicates that
whether the broadcast operation is need to be synced up between the
streaming node and the coordinator.
If the ack_sync_up is false, the broadcast operation will be acked once
the recovery storage see the message at current vchannel, the fast ack
operation can be applied to speed up the broadcast operation.
If the ack_sync_up is true, the broadcast operation will be acked after
the checkpoint of current vchannel reach current message.
The fast ack operation can not be applied to speed up the broadcast
operation, because the ack operation need to be synced up with streaming
node.
e.g. if truncate collection operation want to call ack once callback
after the all segment are flushed at current vchannel, it should set the
ack_sync_up to be true.

TODO: current implementation doesn't promise the ack sync up semantic,
it only promise FastAck operation will not be applied, wait for 3.0 to
implement the ack sync up semantic. only for truncate api now.

---------

Signed-off-by: chyezh <chyezh@outlook.com>
2025-12-17 16:55:17 +08:00

597 lines
21 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 datacoord
import (
"context"
"time"
"github.com/samber/lo"
"go.uber.org/zap"
"github.com/milvus-io/milvus-proto/go-api/v2/commonpb"
"github.com/milvus-io/milvus-proto/go-api/v2/msgpb"
"github.com/milvus-io/milvus/pkg/v2/common"
"github.com/milvus-io/milvus/pkg/v2/log"
"github.com/milvus-io/milvus/pkg/v2/proto/datapb"
"github.com/milvus-io/milvus/pkg/v2/util/funcutil"
"github.com/milvus-io/milvus/pkg/v2/util/retry"
"github.com/milvus-io/milvus/pkg/v2/util/tsoutil"
"github.com/milvus-io/milvus/pkg/v2/util/typeutil"
)
// Handler handles some channel method for ChannelManager
type Handler interface {
// GetQueryVChanPositions gets the information recovery needed of a channel for QueryCoord
GetQueryVChanPositions(ch RWChannel, partitionIDs ...UniqueID) *datapb.VchannelInfo
// GetDataVChanPositions gets the information recovery needed of a channel for DataNode
GetDataVChanPositions(ch RWChannel, partitionID UniqueID) *datapb.VchannelInfo
CheckShouldDropChannel(ch string) bool
FinishDropChannel(ch string, collectionID int64) error
GetCollection(ctx context.Context, collectionID UniqueID) (*collectionInfo, error)
GetCurrentSegmentsView(ctx context.Context, channel RWChannel, partitionIDs ...UniqueID) *SegmentsView
ListLoadedSegments(ctx context.Context) ([]int64, error)
}
type SegmentsView struct {
FlushedSegmentIDs []int64
GrowingSegmentIDs []int64
DroppedSegmentIDs []int64
L0SegmentIDs []int64
ImportingSegmentIDs []int64
}
// ServerHandler is a helper of Server
type ServerHandler struct {
s *Server
}
// newServerHandler creates a new ServerHandler
func newServerHandler(s *Server) *ServerHandler {
return &ServerHandler{s: s}
}
// GetDataVChanPositions gets vchannel latest positions with provided dml channel names for DataNode.
func (h *ServerHandler) GetDataVChanPositions(channel RWChannel, partitionID UniqueID) *datapb.VchannelInfo {
segments := h.s.meta.GetRealSegmentsForChannel(channel.GetName())
log.Info("GetDataVChanPositions",
zap.Int64("collectionID", channel.GetCollectionID()),
zap.String("channel", channel.GetName()),
zap.Int("numOfSegments", len(segments)),
)
var (
flushedIDs = make(typeutil.UniqueSet)
unflushedIDs = make(typeutil.UniqueSet)
droppedIDs = make(typeutil.UniqueSet)
)
for _, s := range segments {
if (partitionID > allPartitionID && s.PartitionID != partitionID) ||
((s.GetState() != commonpb.SegmentState_Growing && s.GetState() != commonpb.SegmentState_Sealed) && s.GetStartPosition() == nil && s.GetDmlPosition() == nil) {
// empty growing and sealed segment don't have dml position and start position
// and it should be recovered for streamingnode, so we add the state-filter here.
continue
}
if s.GetIsImporting() {
// Skip bulk insert segments.
continue
}
if s.GetState() == commonpb.SegmentState_Dropped {
droppedIDs.Insert(s.GetID())
continue
} else if s.GetState() == commonpb.SegmentState_Flushing || s.GetState() == commonpb.SegmentState_Flushed {
flushedIDs.Insert(s.GetID())
} else {
unflushedIDs.Insert(s.GetID())
}
}
return &datapb.VchannelInfo{
CollectionID: channel.GetCollectionID(),
ChannelName: channel.GetName(),
SeekPosition: h.GetChannelSeekPosition(channel, partitionID),
FlushedSegmentIds: flushedIDs.Collect(),
UnflushedSegmentIds: unflushedIDs.Collect(),
DroppedSegmentIds: droppedIDs.Collect(),
}
}
// GetQueryVChanPositions gets vchannel latest positions with provided dml channel names for QueryCoord.
// unflushend segmentIDs ---> L1, growing segments
// flushend segmentIDs ---> L1&L2, flushed segments, including indexed or unindexed
// dropped segmentIDs ---> dropped segments
// level zero segmentIDs ---> L0 segments
func (h *ServerHandler) GetQueryVChanPositions(channel RWChannel, partitionIDs ...UniqueID) *datapb.VchannelInfo {
validPartitions := lo.Filter(partitionIDs, func(partitionID int64, _ int) bool { return partitionID > allPartitionID })
filterWithPartition := len(validPartitions) > 0
validPartitionsMap := make(map[int64]bool)
partStatsVersions := h.s.meta.partitionStatsMeta.GetChannelPartitionsStatsVersion(channel.GetCollectionID(), channel.GetName())
partStatsVersionsMap := make(map[int64]int64)
if filterWithPartition {
for _, partitionID := range validPartitions {
partStatsVersionsMap[partitionID] = partStatsVersions[partitionID]
validPartitionsMap[partitionID] = true
}
validPartitionsMap[common.AllPartitionsID] = true
} else {
partStatsVersionsMap = partStatsVersions
}
var (
flushedIDs = make(typeutil.UniqueSet)
droppedIDs = make(typeutil.UniqueSet)
growingIDs = make(typeutil.UniqueSet)
levelZeroIDs = make(typeutil.UniqueSet)
deleteCheckPoint *msgpb.MsgPosition
)
// cannot use GetSegmentsByChannel since dropped segments are needed here
segments := h.s.meta.GetRealSegmentsForChannel(channel.GetName())
validSegmentInfos := make(map[int64]*SegmentInfo)
indexedSegments := FilterInIndexedSegments(context.Background(), h, h.s.meta, false, segments...)
indexed := typeutil.NewUniqueSet(lo.Map(indexedSegments, func(segment *SegmentInfo, _ int) int64 { return segment.GetID() })...)
for _, s := range segments {
if filterWithPartition && !validPartitionsMap[s.GetPartitionID()] {
continue
}
if s.GetStartPosition() == nil && s.GetDmlPosition() == nil {
continue
}
if s.GetIsImporting() {
// Skip bulk insert segments.
continue
}
validSegmentInfos[s.GetID()] = s
if s.GetIsInvisible() && s.GetCreatedByCompaction() {
// skip invisible compaction segments
continue
}
switch {
case s.GetState() == commonpb.SegmentState_Dropped:
droppedIDs.Insert(s.GetID())
case !isFlushState(s.GetState()) || s.GetIsInvisible():
growingIDs.Insert(s.GetID())
case s.GetLevel() == datapb.SegmentLevel_L0:
levelZeroIDs.Insert(s.GetID())
// use smallest start position of l0 segments as deleteCheckPoint, so query coord will only maintain stream delete record after this ts
if deleteCheckPoint == nil || s.GetStartPosition().GetTimestamp() < deleteCheckPoint.GetTimestamp() {
deleteCheckPoint = s.GetStartPosition()
}
default:
flushedIDs.Insert(s.GetID())
}
}
// ================================================
// Segments blood relationship:
// a b
// \ /
// c d
// \ /
// / \
// e f
//
// GC: a, b
// Indexed: c, d, e, f
// ||
// || (Index dropped and creating new index and not finished)
// \/
// UnIndexed: c, d, e
//
// Retrieve unIndexed expected result:
// unIndexed: c, d
// ================================================
segmentIndexed := func(segID UniqueID) bool {
return indexed.Contain(segID) || (validSegmentInfos[segID].GetIsSorted() && validSegmentInfos[segID].GetNumOfRows() < Params.DataCoordCfg.MinSegmentNumRowsToEnableIndex.GetAsInt64())
}
flushedIDs, droppedIDs = retrieveSegment(validSegmentInfos, flushedIDs, droppedIDs, segmentIndexed)
log.Info("GetQueryVChanPositions",
zap.Int64("collectionID", channel.GetCollectionID()),
zap.String("channel", channel.GetName()),
zap.Int("numOfSegments", len(segments)),
zap.Int("result flushed", len(flushedIDs)),
zap.Int("result growing", len(growingIDs)),
zap.Int("result L0", len(levelZeroIDs)),
zap.Any("partition stats", partStatsVersionsMap),
)
seekPosition := h.GetChannelSeekPosition(channel, partitionIDs...)
// if no l0 segment exist, use checkpoint as delete checkpoint
if len(levelZeroIDs) == 0 {
deleteCheckPoint = seekPosition
}
return &datapb.VchannelInfo{
CollectionID: channel.GetCollectionID(),
ChannelName: channel.GetName(),
SeekPosition: seekPosition,
FlushedSegmentIds: flushedIDs.Collect(),
UnflushedSegmentIds: growingIDs.Collect(),
DroppedSegmentIds: droppedIDs.Collect(),
LevelZeroSegmentIds: levelZeroIDs.Collect(),
PartitionStatsVersions: partStatsVersionsMap,
DeleteCheckpoint: deleteCheckPoint,
}
}
func retrieveSegment(validSegmentInfos map[int64]*SegmentInfo,
flushedIDs, droppedIDs typeutil.UniqueSet,
segmentIndexed func(segID UniqueID) bool,
) (typeutil.UniqueSet, typeutil.UniqueSet) {
newFlushedIDs := make(typeutil.UniqueSet)
isConditionMet := func(condition func(seg *SegmentInfo) bool, ids ...UniqueID) bool {
for _, id := range ids {
if seg, ok := validSegmentInfos[id]; !ok || seg == nil || !condition(seg) {
return false
}
}
return true
}
isValid := func(ids ...UniqueID) bool {
return isConditionMet(func(seg *SegmentInfo) bool {
return true
}, ids...)
}
isVisible := func(ids ...UniqueID) bool {
return isConditionMet(func(seg *SegmentInfo) bool {
return !seg.GetIsInvisible()
}, ids...)
}
var compactionFromExistWithCache func(segID UniqueID) bool
compactionFromExistWithCache = func(segID UniqueID) bool {
var compactionFromExist func(segID UniqueID) bool
compactionFromExistMap := make(map[UniqueID]bool)
compactionFromExist = func(segID UniqueID) bool {
if exist, ok := compactionFromExistMap[segID]; ok {
return exist
}
compactionFrom := validSegmentInfos[segID].GetCompactionFrom()
if len(compactionFrom) == 0 || !isValid(compactionFrom...) {
compactionFromExistMap[segID] = false
return false
}
for _, fromID := range compactionFrom {
if flushedIDs.Contain(fromID) || newFlushedIDs.Contain(fromID) {
compactionFromExistMap[segID] = true
return true
}
if compactionFromExist(fromID) {
compactionFromExistMap[segID] = true
return true
}
}
compactionFromExistMap[segID] = false
return false
}
return compactionFromExist(segID)
}
retrieve := func() bool {
continueRetrieve := false
for id := range flushedIDs {
compactionFrom := validSegmentInfos[id].GetCompactionFrom()
if len(compactionFrom) == 0 {
newFlushedIDs.Insert(id)
} else if !compactionFromExistWithCache(id) && (segmentIndexed(id) || !isVisible(compactionFrom...)) {
newFlushedIDs.Insert(id)
} else {
for _, fromID := range compactionFrom {
newFlushedIDs.Insert(fromID)
continueRetrieve = true
droppedIDs.Remove(fromID)
}
}
}
return continueRetrieve
}
for retrieve() {
flushedIDs = newFlushedIDs
newFlushedIDs = make(typeutil.UniqueSet)
}
return newFlushedIDs, droppedIDs
}
func (h *ServerHandler) GetCurrentSegmentsView(ctx context.Context, channel RWChannel, partitionIDs ...UniqueID) *SegmentsView {
validPartitions := lo.Filter(partitionIDs, func(partitionID int64, _ int) bool { return partitionID > allPartitionID })
filterWithPartition := len(validPartitions) > 0
validPartitionsMap := make(map[int64]bool)
validPartitionsMap[common.AllPartitionsID] = true
for _, partitionID := range validPartitions {
validPartitionsMap[partitionID] = true
}
var (
flushedIDs = make(typeutil.UniqueSet)
droppedIDs = make(typeutil.UniqueSet)
growingIDs = make(typeutil.UniqueSet)
importingIDs = make(typeutil.UniqueSet)
levelZeroIDs = make(typeutil.UniqueSet)
)
// cannot use GetSegmentsByChannel since dropped segments are needed here
segments := h.s.meta.GetRealSegmentsForChannel(channel.GetName())
validSegmentInfos := make(map[int64]*SegmentInfo)
for _, s := range segments {
if filterWithPartition && !validPartitionsMap[s.GetPartitionID()] {
continue
}
if s.GetStartPosition() == nil && s.GetDmlPosition() == nil {
continue
}
validSegmentInfos[s.GetID()] = s
switch {
case s.GetState() == commonpb.SegmentState_Dropped:
droppedIDs.Insert(s.GetID())
case s.GetState() == commonpb.SegmentState_Importing:
importingIDs.Insert(s.GetID())
case s.GetLevel() == datapb.SegmentLevel_L0:
levelZeroIDs.Insert(s.GetID())
case s.GetState() == commonpb.SegmentState_Growing:
growingIDs.Insert(s.GetID())
default:
flushedIDs.Insert(s.GetID())
}
}
flushedIDs, droppedIDs = retrieveSegment(validSegmentInfos, flushedIDs, droppedIDs, func(segID UniqueID) bool {
return true
})
log.Ctx(ctx).Info("GetCurrentSegmentsView",
zap.Int64("collectionID", channel.GetCollectionID()),
zap.String("channel", channel.GetName()),
zap.Int("numOfSegments", len(segments)),
zap.Int("result flushed", len(flushedIDs)),
zap.Int("result growing", len(growingIDs)),
zap.Int("result importing", len(importingIDs)),
zap.Int("result L0", len(levelZeroIDs)),
)
return &SegmentsView{
FlushedSegmentIDs: flushedIDs.Collect(),
GrowingSegmentIDs: growingIDs.Collect(),
DroppedSegmentIDs: droppedIDs.Collect(),
L0SegmentIDs: levelZeroIDs.Collect(),
ImportingSegmentIDs: importingIDs.Collect(),
}
}
// getEarliestSegmentDMLPos returns the earliest dml position of segments,
// this is mainly for COMPATIBILITY with old version <=2.1.x
func (h *ServerHandler) getEarliestSegmentDMLPos(channel string, partitionIDs ...UniqueID) *msgpb.MsgPosition {
var minPos *msgpb.MsgPosition
var minPosSegID int64
var minPosTs uint64
segments := h.s.meta.SelectSegments(context.TODO(), WithChannel(channel))
validPartitions := lo.Filter(partitionIDs, func(partitionID int64, _ int) bool { return partitionID > allPartitionID })
partitionSet := typeutil.NewUniqueSet(validPartitions...)
for _, s := range segments {
if (partitionSet.Len() > 0 && !partitionSet.Contain(s.PartitionID)) ||
(s.GetStartPosition() == nil && s.GetDmlPosition() == nil) {
continue
}
if s.GetIsImporting() {
// Skip bulk insert segments.
continue
}
if s.GetState() == commonpb.SegmentState_Dropped {
continue
}
var segmentPosition *msgpb.MsgPosition
if s.GetDmlPosition() != nil {
segmentPosition = s.GetDmlPosition()
} else {
segmentPosition = s.GetStartPosition()
}
if minPos == nil || segmentPosition.Timestamp < minPos.Timestamp {
minPosSegID = s.GetID()
minPosTs = segmentPosition.GetTimestamp()
minPos = segmentPosition
}
}
if minPos != nil {
log.Info("getEarliestSegmentDMLPos done",
zap.Int64("segmentID", minPosSegID),
zap.Uint64("posTs", minPosTs),
zap.Time("posTime", tsoutil.PhysicalTime(minPosTs)))
}
return minPos
}
// getCollectionStartPos returns collection start position.
func (h *ServerHandler) getCollectionStartPos(channel RWChannel) *msgpb.MsgPosition {
log := log.With(zap.String("channel", channel.GetName()))
if channel.GetStartPosition() != nil {
return channel.GetStartPosition()
}
// use collection start position when segment position is not found
collection, err := h.GetCollection(h.s.ctx, channel.GetCollectionID())
if collection != nil && err == nil {
startPosition := toMsgPosition(channel.GetName(), collection.StartPositions)
// We should not set the timestamp to collectionInfo.CreatedAt
// because after enabling streaming arch, every shard has its own timetick, no comparison can be applied cross shards timetick.
// because when using the collection start position, we don't perform any sync operation of data,
// so we can just use 0 here without introducing any repeated data to avoid filtering some DML whose timetick is less than collectionInfo.CreatedAt.
// And after enabling new DDL framework, the collection start position will have its own timestamp, so we can use it directly.
log.Info("NEITHER segment position or channel start position are found, setting channel seek position to collection start position",
zap.Uint64("posTs", startPosition.GetTimestamp()),
zap.Time("posTime", tsoutil.PhysicalTime(startPosition.GetTimestamp())),
)
return startPosition
}
return nil
}
// GetChannelSeekPosition gets channel seek position from:
// 1. Channel checkpoint meta;
// 2. Segments earliest dml position;
// 3. Collection start position;
// And would return if any position is valid.
func (h *ServerHandler) GetChannelSeekPosition(channel RWChannel, partitionIDs ...UniqueID) *msgpb.MsgPosition {
log := log.With(zap.String("channel", channel.GetName()))
var seekPosition *msgpb.MsgPosition
seekPosition = h.s.meta.GetChannelCheckpoint(channel.GetName())
if seekPosition != nil {
log.Info("channel seek position set from channel checkpoint meta",
zap.Uint64("posTs", seekPosition.Timestamp),
zap.Time("posTime", tsoutil.PhysicalTime(seekPosition.GetTimestamp())))
return seekPosition
}
seekPosition = h.getEarliestSegmentDMLPos(channel.GetName(), partitionIDs...)
if seekPosition != nil {
log.Info("channel seek position set from earliest segment dml position",
zap.Uint64("posTs", seekPosition.Timestamp),
zap.Time("posTime", tsoutil.PhysicalTime(seekPosition.GetTimestamp())))
return seekPosition
}
seekPosition = h.getCollectionStartPos(channel)
if seekPosition != nil {
log.Info("channel seek position set from collection start position",
zap.Uint64("posTs", seekPosition.Timestamp),
zap.Time("posTime", tsoutil.PhysicalTime(seekPosition.GetTimestamp())))
return seekPosition
}
log.Warn("get channel checkpoint failed, channelCPMeta and earliestSegDMLPos and collStartPos are all invalid")
return nil
}
func toMsgPosition(channel string, startPositions []*commonpb.KeyDataPair) *msgpb.MsgPosition {
for _, sp := range startPositions {
if sp.GetKey() != funcutil.ToPhysicalChannel(channel) {
continue
}
return &msgpb.MsgPosition{
ChannelName: channel,
MsgID: sp.GetData(),
}
}
return nil
}
// trimSegmentInfo returns a shallow copy of datapb.SegmentInfo and sets ALL binlog info to nil
func trimSegmentInfo(info *datapb.SegmentInfo) *datapb.SegmentInfo {
return &datapb.SegmentInfo{
ID: info.ID,
CollectionID: info.CollectionID,
PartitionID: info.PartitionID,
InsertChannel: info.InsertChannel,
NumOfRows: info.NumOfRows,
State: info.State,
MaxRowNum: info.MaxRowNum,
LastExpireTime: info.LastExpireTime,
StartPosition: info.StartPosition,
DmlPosition: info.DmlPosition,
}
}
// HasCollection returns whether the collection exist from user's perspective.
func (h *ServerHandler) HasCollection(ctx context.Context, collectionID UniqueID) (bool, error) {
var hasCollection bool
ctx2, cancel := context.WithTimeout(ctx, time.Second*10)
defer cancel()
if err := retry.Do(ctx2, func() error {
has, err := h.s.broker.HasCollection(ctx2, collectionID)
if err != nil {
log.RatedInfo(60, "datacoord ServerHandler HasCollection retry failed", zap.Error(err))
return err
}
hasCollection = has
return nil
}, retry.Attempts(5)); err != nil {
log.Ctx(ctx2).Error("datacoord ServerHandler HasCollection finally failed",
zap.Int64("collectionID", collectionID),
zap.Error(err))
// A workaround for https://github.com/milvus-io/milvus/issues/26863. The collection may be considered as not
// dropped when any exception happened, but there are chances that finally the collection will be cleaned.
return true, nil
}
return hasCollection, nil
}
// GetCollection returns collection info with specified collection id
func (h *ServerHandler) GetCollection(ctx context.Context, collectionID UniqueID) (*collectionInfo, error) {
coll := h.s.meta.GetCollection(collectionID)
if coll != nil {
return coll, nil
}
ctx2, cancel := context.WithTimeout(ctx, time.Second*10)
defer cancel()
if err := retry.Do(ctx2, func() error {
err := h.s.loadCollectionFromRootCoord(ctx2, collectionID)
if err != nil {
log.Warn("failed to load collection from rootcoord", zap.Int64("collectionID", collectionID), zap.Error(err))
return err
}
return nil
}, retry.Attempts(5)); err != nil {
log.Ctx(ctx2).Warn("datacoord ServerHandler GetCollection finally failed",
zap.Int64("collectionID", collectionID),
zap.Error(err))
return nil, err
}
// TODO: the cache should be removed in next step.
return h.s.meta.GetCollection(collectionID), nil
}
// CheckShouldDropChannel returns whether specified channel is marked to be removed
func (h *ServerHandler) CheckShouldDropChannel(channel string) bool {
return h.s.meta.catalog.ShouldDropChannel(h.s.ctx, channel)
}
// FinishDropChannel cleans up the remove flag for channels
// this function is a wrapper of server.meta.FinishDropChannel
func (h *ServerHandler) FinishDropChannel(channel string, collectionID int64) error {
err := h.s.meta.catalog.DropChannel(h.s.ctx, channel)
if err != nil {
log.Warn("DropChannel failed", zap.String("vChannel", channel), zap.Error(err))
return err
}
log.Info("DropChannel succeeded", zap.String("channel", channel))
// Channel checkpoints are cleaned up during garbage collection.
// clean collection info cache when meet drop collection info
h.s.meta.DropCollection(collectionID)
return nil
}
func (h *ServerHandler) ListLoadedSegments(ctx context.Context) ([]int64, error) {
return h.s.listLoadedSegments(ctx)
}