milvus/internal/datanode/flow_graph_dd_node.go

316 lines
10 KiB
Go
Raw Normal View History

// 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 datanode
import (
"context"
"fmt"
"reflect"
"sync/atomic"
"github.com/golang/protobuf/proto"
"go.opentelemetry.io/otel/trace"
"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/internal/proto/datapb"
"github.com/milvus-io/milvus/internal/util/flowgraph"
"github.com/milvus-io/milvus/pkg/log"
"github.com/milvus-io/milvus/pkg/metrics"
"github.com/milvus-io/milvus/pkg/mq/msgstream"
"github.com/milvus-io/milvus/pkg/util/funcutil"
"github.com/milvus-io/milvus/pkg/util/metricsinfo"
"github.com/milvus-io/milvus/pkg/util/paramtable"
)
// make sure ddNode implements flowgraph.Node
var _ flowgraph.Node = (*ddNode)(nil)
// ddNode filters messages from message streams.
//
// ddNode recives all the messages from message stream dml channels, including insert messages,
//
// delete messages and ddl messages like CreateCollectionMsg and DropCollectionMsg.
//
// ddNode filters insert messages according to the `sealedSegment`.
// ddNode will filter out the insert message for those who belong to `sealedSegment`
//
// When receiving a `DropCollection` message, ddNode will send a signal to DataNode `BackgroundGC`
//
// goroutinue, telling DataNode to release the resources of this particular flow graph.
//
// After the filtering process, ddNode passes all the valid insert messages and delete message
//
// to the following flow graph node, which in DataNode is `insertBufferNode`
type ddNode struct {
BaseNode
ctx context.Context
collectionID UniqueID
vChannelName string
dropMode atomic.Value
compactionExecutor *compactionExecutor
// for recovery
growingSegInfo map[UniqueID]*datapb.SegmentInfo // segmentID
sealedSegInfo map[UniqueID]*datapb.SegmentInfo // segmentID
droppedSegmentIDs []int64
}
// Name returns node name, implementing flowgraph.Node
func (ddn *ddNode) Name() string {
return fmt.Sprintf("ddNode-%d-%s", ddn.collectionID, ddn.vChannelName)
}
func (ddn *ddNode) IsValidInMsg(in []Msg) bool {
if !ddn.BaseNode.IsValidInMsg(in) {
return false
}
_, ok := in[0].(*MsgStreamMsg)
if !ok {
log.Warn("type assertion failed for MsgStreamMsg", zap.String("name", reflect.TypeOf(in[0]).Name()))
return false
}
return true
}
// Operate handles input messages, implementing flowgrpah.Node
func (ddn *ddNode) Operate(in []Msg) []Msg {
msMsg, ok := in[0].(*MsgStreamMsg)
if !ok {
log.Warn("type assertion failed for MsgStreamMsg", zap.String("name", reflect.TypeOf(in[0]).Name()))
return []Msg{}
}
if msMsg.IsCloseMsg() {
fgMsg := flowGraphMsg{
BaseMsg: flowgraph.NewBaseMsg(true),
insertMessages: make([]*msgstream.InsertMsg, 0),
timeRange: TimeRange{
timestampMin: msMsg.TimestampMin(),
timestampMax: msMsg.TimestampMax(),
},
startPositions: msMsg.StartPositions(),
endPositions: msMsg.EndPositions(),
dropCollection: false,
}
log.Warn("MsgStream closed", zap.Any("ddNode node", ddn.Name()), zap.Int64("collection", ddn.collectionID), zap.String("channel", ddn.vChannelName))
return []Msg{&fgMsg}
}
if load := ddn.dropMode.Load(); load != nil && load.(bool) {
log.Info("ddNode in dropMode",
zap.String("vChannelName", ddn.vChannelName),
zap.Int64("collectionID", ddn.collectionID))
return []Msg{}
}
var spans []trace.Span
for _, msg := range msMsg.TsMessages() {
ctx, sp := startTracer(msg, "DDNode-Operate")
spans = append(spans, sp)
msg.SetTraceCtx(ctx)
}
defer func() {
for _, sp := range spans {
sp.End()
}
}()
fgMsg := flowGraphMsg{
insertMessages: make([]*msgstream.InsertMsg, 0),
timeRange: TimeRange{
timestampMin: msMsg.TimestampMin(),
timestampMax: msMsg.TimestampMax(),
},
startPositions: make([]*msgpb.MsgPosition, 0),
endPositions: make([]*msgpb.MsgPosition, 0),
dropCollection: false,
}
for _, msg := range msMsg.TsMessages() {
switch msg.Type() {
case commonpb.MsgType_DropCollection:
if msg.(*msgstream.DropCollectionMsg).GetCollectionID() == ddn.collectionID {
log.Info("Receiving DropCollection msg",
zap.Int64("collectionID", ddn.collectionID),
zap.String("vChannelName", ddn.vChannelName))
ddn.dropMode.Store(true)
log.Info("Stop compaction of vChannel", zap.String("vChannelName", ddn.vChannelName))
ddn.compactionExecutor.stopExecutingtaskByVChannelName(ddn.vChannelName)
fgMsg.dropCollection = true
pChan := funcutil.ToPhysicalChannel(ddn.vChannelName)
metrics.CleanupDataNodeCollectionMetrics(paramtable.GetNodeID(), ddn.collectionID, pChan)
}
case commonpb.MsgType_DropPartition:
dpMsg := msg.(*msgstream.DropPartitionMsg)
if dpMsg.GetCollectionID() == ddn.collectionID {
log.Info("drop partition msg received",
zap.Int64("collectionID", dpMsg.GetCollectionID()),
zap.Int64("partitionID", dpMsg.GetPartitionID()),
zap.String("vChanneName", ddn.vChannelName))
fgMsg.dropPartitions = append(fgMsg.dropPartitions, dpMsg.PartitionID)
}
case commonpb.MsgType_Insert:
imsg := msg.(*msgstream.InsertMsg)
if imsg.CollectionID != ddn.collectionID {
log.Info("filter invalid insert message, collection mis-match",
zap.Int64("Get collID", imsg.CollectionID),
zap.Int64("Expected collID", ddn.collectionID))
continue
}
if ddn.tryToFilterSegmentInsertMessages(imsg) {
log.Info("filter insert messages",
zap.Int64("filter segmentID", imsg.GetSegmentID()),
zap.Uint64("message timestamp", msg.EndTs()),
zap.String("segment's vChannel", imsg.GetShardName()),
zap.String("current vChannel", ddn.vChannelName))
continue
}
rateCol.Add(metricsinfo.InsertConsumeThroughput, float64(proto.Size(&imsg.InsertRequest)))
metrics.DataNodeConsumeBytesCount.
WithLabelValues(fmt.Sprint(paramtable.GetNodeID()), metrics.InsertLabel).
Add(float64(proto.Size(&imsg.InsertRequest)))
metrics.DataNodeConsumeMsgCount.
WithLabelValues(fmt.Sprint(paramtable.GetNodeID()), metrics.InsertLabel, fmt.Sprint(ddn.collectionID)).
Inc()
log.Debug("DDNode receive insert messages",
zap.Int("numRows", len(imsg.GetRowIDs())),
zap.String("vChannelName", ddn.vChannelName))
fgMsg.insertMessages = append(fgMsg.insertMessages, imsg)
case commonpb.MsgType_Delete:
dmsg := msg.(*msgstream.DeleteMsg)
log.Debug("DDNode receive delete messages",
zap.Int64("numRows", dmsg.NumRows),
zap.String("vChannelName", ddn.vChannelName))
if dmsg.CollectionID != ddn.collectionID {
log.Warn("filter invalid DeleteMsg, collection mis-match",
zap.Int64("Get collID", dmsg.CollectionID),
zap.Int64("Expected collID", ddn.collectionID))
continue
}
rateCol.Add(metricsinfo.DeleteConsumeThroughput, float64(proto.Size(&dmsg.DeleteRequest)))
metrics.DataNodeConsumeBytesCount.
WithLabelValues(fmt.Sprint(paramtable.GetNodeID()), metrics.DeleteLabel).
Add(float64(proto.Size(&dmsg.DeleteRequest)))
metrics.DataNodeConsumeMsgCount.
WithLabelValues(fmt.Sprint(paramtable.GetNodeID()), metrics.DeleteLabel, fmt.Sprint(ddn.collectionID)).
Inc()
fgMsg.deleteMessages = append(fgMsg.deleteMessages, dmsg)
}
}
fgMsg.startPositions = append(fgMsg.startPositions, msMsg.StartPositions()...)
fgMsg.endPositions = append(fgMsg.endPositions, msMsg.EndPositions()...)
return []Msg{&fgMsg}
}
func (ddn *ddNode) tryToFilterSegmentInsertMessages(msg *msgstream.InsertMsg) bool {
if msg.GetShardName() != ddn.vChannelName {
return true
}
// Filter all dropped segments
if ddn.isDropped(msg.GetSegmentID()) {
return true
}
// Filter all sealed segments until current Ts > Sealed segment cp
for segID, segInfo := range ddn.sealedSegInfo {
if msg.EndTs() > segInfo.GetDmlPosition().GetTimestamp() {
delete(ddn.sealedSegInfo, segID)
}
}
if _, ok := ddn.sealedSegInfo[msg.GetSegmentID()]; ok {
return true
}
// Filter all growing segments until current Ts > growing segment dmlPosition
if si, ok := ddn.growingSegInfo[msg.GetSegmentID()]; ok {
if msg.EndTs() <= si.GetDmlPosition().GetTimestamp() {
return true
}
delete(ddn.growingSegInfo, msg.GetSegmentID())
}
return false
}
func (ddn *ddNode) isDropped(segID UniqueID) bool {
for _, droppedSegmentID := range ddn.droppedSegmentIDs {
if droppedSegmentID == segID {
return true
}
}
return false
}
func (ddn *ddNode) Close() {}
func newDDNode(ctx context.Context, collID UniqueID, vChannelName string, droppedSegmentIDs []UniqueID,
sealedSegments []*datapb.SegmentInfo, growingSegments []*datapb.SegmentInfo, compactor *compactionExecutor,
) (*ddNode, error) {
baseNode := BaseNode{}
baseNode.SetMaxQueueLength(Params.DataNodeCfg.FlowGraphMaxQueueLength.GetAsInt32())
baseNode.SetMaxParallelism(Params.DataNodeCfg.FlowGraphMaxParallelism.GetAsInt32())
dd := &ddNode{
ctx: ctx,
BaseNode: baseNode,
collectionID: collID,
sealedSegInfo: make(map[UniqueID]*datapb.SegmentInfo, len(sealedSegments)),
growingSegInfo: make(map[UniqueID]*datapb.SegmentInfo, len(growingSegments)),
droppedSegmentIDs: droppedSegmentIDs,
vChannelName: vChannelName,
compactionExecutor: compactor,
}
dd.dropMode.Store(false)
for _, s := range sealedSegments {
dd.sealedSegInfo[s.GetID()] = s
}
for _, s := range growingSegments {
dd.growingSegInfo[s.GetID()] = s
}
log.Info("ddNode add sealed and growing segments",
zap.Int64("collectionID", collID),
zap.Int("No. sealed segments", len(sealedSegments)),
zap.Int("No. growing segments", len(growingSegments)),
)
return dd, nil
}