milvus/internal/querycoordv2/task/executor.go
bigsheeper 2146af1fb2
Return insufficient memory error when load failed (#21574)
Signed-off-by: bigsheeper <yihao.dai@zilliz.com>
2023-01-10 20:35:39 +08:00

484 lines
14 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 task
import (
"context"
"errors"
"fmt"
"sync"
"time"
"github.com/milvus-io/milvus-proto/go-api/commonpb"
"github.com/milvus-io/milvus/internal/log"
"github.com/milvus-io/milvus/internal/proto/querypb"
"github.com/milvus-io/milvus/internal/querycoordv2/meta"
. "github.com/milvus-io/milvus/internal/querycoordv2/params"
"github.com/milvus-io/milvus/internal/querycoordv2/session"
"github.com/milvus-io/milvus/internal/querycoordv2/utils"
"github.com/milvus-io/milvus/internal/util/tsoutil"
"go.uber.org/atomic"
"go.uber.org/zap"
)
type Executor struct {
doneCh chan struct{}
wg sync.WaitGroup
meta *meta.Meta
dist *meta.DistributionManager
broker meta.Broker
targetMgr *meta.TargetManager
cluster session.Cluster
nodeMgr *session.NodeManager
// Merge load segment requests
merger *Merger[segmentIndex, *querypb.LoadSegmentsRequest]
executingTasks sync.Map
executingTaskNum atomic.Int32
}
func NewExecutor(meta *meta.Meta,
dist *meta.DistributionManager,
broker meta.Broker,
targetMgr *meta.TargetManager,
cluster session.Cluster,
nodeMgr *session.NodeManager) *Executor {
return &Executor{
doneCh: make(chan struct{}),
meta: meta,
dist: dist,
broker: broker,
targetMgr: targetMgr,
cluster: cluster,
nodeMgr: nodeMgr,
merger: NewMerger[segmentIndex, *querypb.LoadSegmentsRequest](),
executingTasks: sync.Map{},
}
}
func (ex *Executor) Start(ctx context.Context) {
ex.merger.Start(ctx)
ex.scheduleRequests()
}
func (ex *Executor) Stop() {
ex.merger.Stop()
ex.wg.Wait()
}
// Execute executes the given action,
// does nothing and returns false if the action is already committed,
// returns true otherwise.
func (ex *Executor) Execute(task Task, step int) bool {
_, exist := ex.executingTasks.LoadOrStore(task.ID(), struct{}{})
if exist {
return false
}
if ex.executingTaskNum.Inc() > Params.QueryCoordCfg.TaskExecutionCap.GetAsInt32() {
ex.executingTasks.Delete(task.ID())
ex.executingTaskNum.Dec()
return false
}
log := log.With(
zap.Int64("taskID", task.ID()),
zap.Int64("collectionID", task.CollectionID()),
zap.Int64("replicaID", task.ReplicaID()),
zap.Int("step", step),
zap.Int64("source", task.SourceID()),
)
go func() {
log.Info("execute the action of task")
switch task.Actions()[step].(type) {
case *SegmentAction:
ex.executeSegmentAction(task.(*SegmentTask), step)
case *ChannelAction:
ex.executeDmChannelAction(task.(*ChannelTask), step)
}
}()
return true
}
func (ex *Executor) Exist(taskID int64) bool {
_, ok := ex.executingTasks.Load(taskID)
return ok
}
func (ex *Executor) scheduleRequests() {
ex.wg.Add(1)
go func() {
defer ex.wg.Done()
for mergeTask := range ex.merger.Chan() {
task := mergeTask.(*LoadSegmentsTask)
log.Info("get merge task, process it",
zap.Int64("collectionID", task.req.GetCollectionID()),
zap.Int64("replicaID", task.req.GetReplicaID()),
zap.String("shard", task.req.GetInfos()[0].GetInsertChannel()),
zap.Int64("nodeID", task.req.GetDstNodeID()),
zap.Int("taskNum", len(task.tasks)),
)
go ex.processMergeTask(mergeTask.(*LoadSegmentsTask))
}
}()
}
func (ex *Executor) processMergeTask(mergeTask *LoadSegmentsTask) {
startTs := time.Now()
task := mergeTask.tasks[0]
action := task.Actions()[mergeTask.steps[0]]
defer func() {
canceled := task.canceled.Load()
for i := range mergeTask.tasks {
mergeTask.tasks[i].SetErr(task.Err())
if canceled {
mergeTask.tasks[i].Cancel()
}
ex.removeTask(mergeTask.tasks[i], mergeTask.steps[i])
}
}()
taskIDs := make([]int64, 0, len(mergeTask.tasks))
segments := make([]int64, 0, len(mergeTask.tasks))
for _, task := range mergeTask.tasks {
taskIDs = append(taskIDs, task.ID())
segments = append(segments, task.SegmentID())
}
log := log.With(
zap.Int64s("taskIDs", taskIDs),
zap.Int64("collectionID", task.CollectionID()),
zap.Int64("replicaID", task.ReplicaID()),
zap.String("shard", task.Shard()),
zap.Int64s("segmentIDs", segments),
zap.Int64("nodeID", action.Node()),
zap.Int64("source", task.SourceID()),
)
// Get shard leader for the given replica and segment
channel := mergeTask.req.GetInfos()[0].GetInsertChannel()
leader, ok := getShardLeader(ex.meta.ReplicaManager, ex.dist, task.CollectionID(), action.Node(), channel)
if !ok {
msg := "no shard leader for the segment to execute loading"
task.SetErr(utils.WrapError(msg, ErrTaskStale))
log.Warn(msg, zap.String("shard", channel))
return
}
log.Info("load segments...")
status, err := ex.cluster.LoadSegments(task.Context(), leader, mergeTask.req)
if err != nil {
log.Warn("failed to load segment, it may be a false failure", zap.Error(err))
return
}
if status.ErrorCode == commonpb.ErrorCode_InsufficientMemoryToLoad {
log.Warn("insufficient memory to load segment", zap.String("err", status.GetReason()))
task.SetErr(fmt.Errorf("%w, err:%s", ErrInsufficientMemory, status.GetReason()))
task.Cancel()
return
}
if status.ErrorCode != commonpb.ErrorCode_Success {
log.Warn("failed to load segment", zap.String("reason", status.GetReason()))
return
}
elapsed := time.Since(startTs)
log.Info("load segments done", zap.Int64("taskID", task.ID()), zap.Duration("timeTaken", elapsed))
}
func (ex *Executor) removeTask(task Task, step int) {
if task.Err() != nil {
log.Info("excute action done, remove it",
zap.Int64("taskID", task.ID()),
zap.Int("step", step),
zap.Error(task.Err()))
}
ex.executingTasks.Delete(task.ID())
ex.executingTaskNum.Dec()
}
func (ex *Executor) executeSegmentAction(task *SegmentTask, step int) {
switch task.Actions()[step].Type() {
case ActionTypeGrow:
ex.loadSegment(task, step)
case ActionTypeReduce:
ex.releaseSegment(task, step)
}
}
// loadSegment commits the request to merger,
// not really executes the request
func (ex *Executor) loadSegment(task *SegmentTask, step int) error {
action := task.Actions()[step].(*SegmentAction)
log := log.With(
zap.Int64("taskID", task.ID()),
zap.Int64("collectionID", task.CollectionID()),
zap.Int64("replicaID", task.ReplicaID()),
zap.Int64("segmentID", task.segmentID),
zap.Int64("node", action.Node()),
zap.Int64("source", task.SourceID()),
)
var err error
defer func() {
if err != nil {
task.SetErr(err)
task.Cancel()
ex.removeTask(task, step)
}
}()
ctx := task.Context()
schema, err := ex.broker.GetCollectionSchema(ctx, task.CollectionID())
if err != nil {
log.Warn("failed to get schema of collection", zap.Error(err))
task.SetErr(err)
return err
}
partitions, err := utils.GetPartitions(ex.meta.CollectionManager, ex.broker, task.CollectionID())
if err != nil {
log.Warn("failed to get partitions of collection", zap.Error(err))
return err
}
loadMeta := packLoadMeta(
ex.meta.GetLoadType(task.CollectionID()),
task.CollectionID(),
partitions...,
)
resp, err := ex.broker.GetSegmentInfo(ctx, task.SegmentID())
if err != nil || len(resp.GetInfos()) == 0 {
log.Warn("failed to get segment info from DataCoord", zap.Error(err))
return err
}
segment := resp.GetInfos()[0]
indexes, err := ex.broker.GetIndexInfo(ctx, task.CollectionID(), segment.GetID())
if err != nil {
log.Warn("failed to get index of segment", zap.Error(err))
return err
}
loadInfo := utils.PackSegmentLoadInfo(segment, indexes)
// Get shard leader for the given replica and segment
leader, ok := getShardLeader(ex.meta.ReplicaManager, ex.dist, task.CollectionID(), action.Node(), segment.GetInsertChannel())
if !ok {
msg := "no shard leader for the segment to execute loading"
err = utils.WrapError(msg, ErrTaskStale)
log.Warn(msg, zap.String("shard", segment.GetInsertChannel()))
return err
}
log = log.With(zap.Int64("shardLeader", leader))
req := packLoadSegmentRequest(task, action, schema, loadMeta, loadInfo, resp)
loadTask := NewLoadSegmentsTask(task, step, req)
ex.merger.Add(loadTask)
log.Info("load segment task committed")
return nil
}
func (ex *Executor) releaseSegment(task *SegmentTask, step int) {
defer ex.removeTask(task, step)
startTs := time.Now()
action := task.Actions()[step].(*SegmentAction)
defer action.isReleaseCommitted.Store(true)
log := log.With(
zap.Int64("taskID", task.ID()),
zap.Int64("collectionID", task.CollectionID()),
zap.Int64("replicaID", task.ReplicaID()),
zap.Int64("segmentID", task.segmentID),
zap.Int64("node", action.Node()),
zap.Int64("source", task.SourceID()),
)
ctx := task.Context()
dstNode := action.Node()
req := packReleaseSegmentRequest(task, action)
if action.Scope() == querypb.DataScope_Streaming {
// Any modification to the segment distribution have to set NeedTransfer true,
// to protect the version, which serves search/query
req.NeedTransfer = true
} else {
var targetSegment *meta.Segment
segments := ex.dist.SegmentDistManager.GetByNode(action.Node())
for _, segment := range segments {
if segment.GetID() == task.SegmentID() {
targetSegment = segment
break
}
}
if targetSegment == nil {
log.Info("segment to release not found in distribution")
return
}
req.Shard = targetSegment.GetInsertChannel()
if ex.meta.CollectionManager.Exist(task.CollectionID()) {
leader, ok := getShardLeader(ex.meta.ReplicaManager, ex.dist, task.CollectionID(), action.Node(), req.GetShard())
if !ok {
log.Warn("no shard leader for the segment to execute releasing", zap.String("shard", req.GetShard()))
return
}
dstNode = leader
log = log.With(zap.Int64("shardLeader", leader))
req.NeedTransfer = true
}
}
log.Info("release segment...")
status, err := ex.cluster.ReleaseSegments(ctx, dstNode, req)
if err != nil {
log.Warn("failed to release segment, it may be a false failure", zap.Error(err))
return
}
if status.ErrorCode != commonpb.ErrorCode_Success {
log.Warn("failed to release segment", zap.String("reason", status.GetReason()))
return
}
elapsed := time.Since(startTs)
log.Info("release segment done", zap.Int64("taskID", task.ID()), zap.Duration("time taken", elapsed))
}
func (ex *Executor) executeDmChannelAction(task *ChannelTask, step int) {
switch task.Actions()[step].Type() {
case ActionTypeGrow:
ex.subDmChannel(task, step)
case ActionTypeReduce:
ex.unsubDmChannel(task, step)
}
}
func (ex *Executor) subDmChannel(task *ChannelTask, step int) error {
defer ex.removeTask(task, step)
startTs := time.Now()
action := task.Actions()[step].(*ChannelAction)
log := log.With(
zap.Int64("taskID", task.ID()),
zap.Int64("collectionID", task.CollectionID()),
zap.Int64("replicaID", task.ReplicaID()),
zap.String("channel", task.Channel()),
zap.Int64("node", action.Node()),
zap.Int64("source", task.SourceID()),
)
var err error
defer func() {
if err != nil {
task.SetErr(err)
task.Cancel()
}
}()
ctx := task.Context()
schema, err := ex.broker.GetCollectionSchema(ctx, task.CollectionID())
if err != nil {
log.Warn("failed to get schema of collection")
return err
}
partitions, err := utils.GetPartitions(ex.meta.CollectionManager, ex.broker, task.CollectionID())
if err != nil {
log.Warn("failed to get partitions of collection")
return err
}
loadMeta := packLoadMeta(
ex.meta.GetLoadType(task.CollectionID()),
task.CollectionID(),
partitions...,
)
dmChannel := ex.targetMgr.GetDmChannel(task.CollectionID(), action.ChannelName(), meta.NextTarget)
if dmChannel == nil {
msg := "channel does not exist in next target, skip it"
log.Warn(msg, zap.String("channelName", action.ChannelName()))
return errors.New(msg)
}
req := packSubDmChannelRequest(task, action, schema, loadMeta, dmChannel)
err = fillSubDmChannelRequest(ctx, req, ex.broker)
if err != nil {
log.Warn("failed to subscribe DmChannel, failed to fill the request with segments",
zap.Error(err))
return err
}
ts := dmChannel.GetSeekPosition().GetTimestamp()
log.Info("subscribe channel...",
zap.Uint64("checkpoint", ts),
zap.Duration("sinceCheckpoint", time.Since(tsoutil.PhysicalTime(ts))),
)
status, err := ex.cluster.WatchDmChannels(ctx, action.Node(), req)
if err != nil {
log.Warn("failed to subscribe DmChannel, it may be a false failure", zap.Error(err))
return err
}
if status.ErrorCode != commonpb.ErrorCode_Success {
err = utils.WrapError("failed to subscribe DmChannel", ErrFailedResponse)
log.Warn("failed to subscribe DmChannel", zap.String("reason", status.GetReason()))
return err
}
elapsed := time.Since(startTs)
log.Info("subscribe DmChannel done", zap.Int64("taskID", task.ID()), zap.Duration("time taken", elapsed))
return nil
}
func (ex *Executor) unsubDmChannel(task *ChannelTask, step int) error {
defer ex.removeTask(task, step)
startTs := time.Now()
action := task.Actions()[step].(*ChannelAction)
log := log.With(
zap.Int64("taskID", task.ID()),
zap.Int64("collectionID", task.CollectionID()),
zap.Int64("replicaID", task.ReplicaID()),
zap.String("channel", task.Channel()),
zap.Int64("node", action.Node()),
zap.Int64("source", task.SourceID()),
)
var err error
defer func() {
if err != nil {
task.SetErr(err)
task.Cancel()
}
}()
ctx := task.Context()
req := packUnsubDmChannelRequest(task, action)
log.Info("unsubscribe channel...")
status, err := ex.cluster.UnsubDmChannel(ctx, action.Node(), req)
if err != nil {
log.Warn("failed to unsubscribe DmChannel, it may be a false failure", zap.Error(err))
return err
}
if status.ErrorCode != commonpb.ErrorCode_Success {
err = utils.WrapError("failed to unsubscribe DmChannel", ErrFailedResponse)
log.Warn("failed to unsubscribe DmChannel", zap.String("reason", status.GetReason()))
return err
}
elapsed := time.Since(startTs)
log.Info("unsubscribe DmChannel done", zap.Int64("taskID", task.ID()), zap.Duration("time taken", elapsed))
return nil
}