milvus/internal/querycoordv2/meta/channel_dist_manager.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 meta

import (
	"sync"

	"github.com/samber/lo"
	"go.uber.org/zap"
	"google.golang.org/protobuf/proto"

	"github.com/milvus-io/milvus/internal/querycoordv2/session"
	"github.com/milvus-io/milvus/internal/util/metrics"
	"github.com/milvus-io/milvus/pkg/v2/log"
	"github.com/milvus-io/milvus/pkg/v2/proto/datapb"
	"github.com/milvus-io/milvus/pkg/v2/proto/querypb"
	"github.com/milvus-io/milvus/pkg/v2/util/metricsinfo"
	"github.com/milvus-io/milvus/pkg/v2/util/typeutil"
)

type LeaderView struct {
	ID                     int64
	CollectionID           int64
	Channel                string
	Version                int64
	Segments               map[int64]*querypb.SegmentDist
	GrowingSegments        map[int64]*Segment
	TargetVersion          int64
	NumOfGrowingRows       int64
	PartitionStatsVersions map[int64]int64
	Status                 *querypb.LeaderViewStatus
}

func (view *LeaderView) Clone() *LeaderView {
	segments := make(map[int64]*querypb.SegmentDist)
	for k, v := range view.Segments {
		segments[k] = v
	}

	growings := make(map[int64]*Segment)
	for k, v := range view.GrowingSegments {
		growings[k] = v
	}

	return &LeaderView{
		ID:                     view.ID,
		CollectionID:           view.CollectionID,
		Channel:                view.Channel,
		Version:                view.Version,
		Segments:               segments,
		GrowingSegments:        growings,
		TargetVersion:          view.TargetVersion,
		NumOfGrowingRows:       view.NumOfGrowingRows,
		PartitionStatsVersions: view.PartitionStatsVersions,
	}
}

type channelDistCriterion struct {
	nodeIDs        typeutil.Set[int64]
	collectionID   int64
	channelName    string
	hasOtherFilter bool
}

type ChannelDistFilter interface {
	Match(ch *DmChannel) bool
	AddFilter(*channelDistCriterion)
}

type collChannelFilter int64

func (f collChannelFilter) Match(ch *DmChannel) bool {
	return ch.GetCollectionID() == int64(f)
}

func (f collChannelFilter) AddFilter(criterion *channelDistCriterion) {
	criterion.collectionID = int64(f)
}

func WithCollectionID2Channel(collectionID int64) ChannelDistFilter {
	return collChannelFilter(collectionID)
}

type nodeChannelFilter int64

func (f nodeChannelFilter) Match(ch *DmChannel) bool {
	return ch.Node == int64(f)
}

func (f nodeChannelFilter) AddFilter(criterion *channelDistCriterion) {
	set := typeutil.NewSet(int64(f))
	if criterion.nodeIDs == nil {
		criterion.nodeIDs = set
	} else {
		criterion.nodeIDs = criterion.nodeIDs.Intersection(set)
	}
}

func WithNodeID2Channel(nodeID int64) ChannelDistFilter {
	return nodeChannelFilter(nodeID)
}

type replicaChannelFilter struct {
	*Replica
}

func (f replicaChannelFilter) Match(ch *DmChannel) bool {
	return ch.GetCollectionID() == f.GetCollectionID() && f.Contains(ch.Node)
}

func (f replicaChannelFilter) AddFilter(criterion *channelDistCriterion) {
	criterion.collectionID = f.GetCollectionID()

	set := typeutil.NewSet(f.GetNodes()...)
	if criterion.nodeIDs == nil {
		criterion.nodeIDs = set
	} else {
		criterion.nodeIDs = criterion.nodeIDs.Intersection(set)
	}
}

func WithReplica2Channel(replica *Replica) ChannelDistFilter {
	return &replicaChannelFilter{
		Replica: replica,
	}
}

type nameChannelFilter string

func (f nameChannelFilter) Match(ch *DmChannel) bool {
	return ch.GetChannelName() == string(f)
}

func (f nameChannelFilter) AddFilter(criterion *channelDistCriterion) {
	criterion.channelName = string(f)
}

func WithChannelName2Channel(channelName string) ChannelDistFilter {
	return nameChannelFilter(channelName)
}

type DmChannel struct {
	*datapb.VchannelInfo
	Node    int64
	Version int64
	View    *LeaderView
}

func DmChannelFromVChannel(channel *datapb.VchannelInfo) *DmChannel {
	return &DmChannel{
		VchannelInfo: channel,
	}
}

func (channel *DmChannel) Clone() *DmChannel {
	return &DmChannel{
		VchannelInfo: proto.Clone(channel.VchannelInfo).(*datapb.VchannelInfo),
		Node:         channel.Node,
		Version:      channel.Version,
		View: &LeaderView{
			ID:           channel.View.ID,
			CollectionID: channel.View.CollectionID,
			Channel:      channel.View.Channel,
			Version:      channel.View.Version,
			Status:       proto.Clone(channel.View.Status).(*querypb.LeaderViewStatus),
		},
	}
}

func (channel *DmChannel) IsServiceable() bool {
	if channel.View == nil {
		return false
	}
	return channel.View.Status.GetServiceable()
}

func newDmChannelMetricsFrom(channel *DmChannel) *metricsinfo.DmChannel {
	dmChannel := metrics.NewDMChannelFrom(channel.VchannelInfo)
	dmChannel.NodeID = channel.Node
	dmChannel.Version = channel.Version
	return dmChannel
}

type nodeChannels struct {
	channels []*DmChannel
	// collection id => channels
	collChannels map[int64][]*DmChannel
	// channel name => DmChannel
	nameChannel map[string]*DmChannel
}

func (c nodeChannels) Filter(critertion *channelDistCriterion) []*DmChannel {
	var channels []*DmChannel
	switch {
	case critertion.channelName != "":
		if ch, ok := c.nameChannel[critertion.channelName]; ok {
			channels = []*DmChannel{ch}
		}
	case critertion.collectionID != 0:
		channels = c.collChannels[critertion.collectionID]
	default:
		channels = c.channels
	}

	return channels // lo.Filter(channels, func(ch *DmChannel, _ int) bool { return mergedFilters(ch) })
}

func composeNodeChannels(channels ...*DmChannel) nodeChannels {
	return nodeChannels{
		channels:     channels,
		collChannels: lo.GroupBy(channels, func(ch *DmChannel) int64 { return ch.GetCollectionID() }),
		nameChannel:  lo.SliceToMap(channels, func(ch *DmChannel) (string, *DmChannel) { return ch.GetChannelName(), ch }),
	}
}

type ChannelDistManagerInterface interface {
	GetByFilter(filters ...ChannelDistFilter) []*DmChannel
	GetByCollectionAndFilter(collectionID int64, filters ...ChannelDistFilter) []*DmChannel
	Update(nodeID typeutil.UniqueID, channels ...*DmChannel) []*DmChannel
	GetShardLeader(channelName string, replica *Replica) *DmChannel
	GetChannelDist(collectionID int64) []*metricsinfo.DmChannel
	GetLeaderView(collectionID int64) []*metricsinfo.LeaderView
}

type ChannelDistManager struct {
	rwmutex sync.RWMutex

	// NodeID -> Channels
	channels map[typeutil.UniqueID]nodeChannels

	// CollectionID -> Channels
	collectionIndex map[int64][]*DmChannel

	nodeManager *session.NodeManager
}

func NewChannelDistManager(nodeManager *session.NodeManager) *ChannelDistManager {
	return &ChannelDistManager{
		channels:        make(map[typeutil.UniqueID]nodeChannels),
		collectionIndex: make(map[int64][]*DmChannel),
		nodeManager:     nodeManager,
	}
}

// return all channels in list which match all given filters
func (m *ChannelDistManager) GetByFilter(filters ...ChannelDistFilter) []*DmChannel {
	m.rwmutex.RLock()
	defer m.rwmutex.RUnlock()

	criterion := &channelDistCriterion{}
	for _, filter := range filters {
		filter.AddFilter(criterion)
	}

	var candidates []nodeChannels
	if criterion.nodeIDs != nil {
		candidates = lo.Map(criterion.nodeIDs.Collect(), func(nodeID int64, _ int) nodeChannels {
			return m.channels[nodeID]
		})
	} else {
		candidates = lo.Values(m.channels)
	}

	var ret []*DmChannel
	for _, candidate := range candidates {
		ret = append(ret, candidate.Filter(criterion)...)
	}
	return ret
}

func (m *ChannelDistManager) GetByCollectionAndFilter(collectionID int64, filters ...ChannelDistFilter) []*DmChannel {
	m.rwmutex.RLock()
	defer m.rwmutex.RUnlock()

	mergedFilters := func(ch *DmChannel) bool {
		for _, fn := range filters {
			if fn != nil && !fn.Match(ch) {
				return false
			}
		}

		return true
	}

	ret := make([]*DmChannel, 0)

	// If a collection ID is provided, use the collection index
	for _, channel := range m.collectionIndex[collectionID] {
		if mergedFilters(channel) {
			ret = append(ret, channel)
		}
	}
	return ret
}

func (m *ChannelDistManager) Update(nodeID typeutil.UniqueID, channels ...*DmChannel) []*DmChannel {
	m.rwmutex.Lock()
	defer m.rwmutex.Unlock()

	newServiceableChannels := make([]*DmChannel, 0)
	for _, channel := range channels {
		channel.Node = nodeID

		old, ok := m.channels[nodeID].nameChannel[channel.GetChannelName()]
		if channel.IsServiceable() && (!ok || !old.IsServiceable()) {
			newServiceableChannels = append(newServiceableChannels, channel)
		}
	}

	m.channels[nodeID] = composeNodeChannels(channels...)
	m.updateCollectionIndex()
	return newServiceableChannels
}

// update secondary index for channel distribution
func (m *ChannelDistManager) updateCollectionIndex() {
	m.collectionIndex = make(map[int64][]*DmChannel)
	for _, nodeChannels := range m.channels {
		for _, channel := range nodeChannels.channels {
			collectionID := channel.GetCollectionID()
			if channels, ok := m.collectionIndex[collectionID]; !ok {
				m.collectionIndex[collectionID] = []*DmChannel{channel}
			} else {
				m.collectionIndex[collectionID] = append(channels, channel)
			}
		}
	}
}

// GetShardLeader return the only one delegator leader which has the highest version in given replica
// if there is no serviceable leader, return the highest version leader
// With specific channel name and replica, return the only one delegator leader
func (m *ChannelDistManager) GetShardLeader(channelName string, replica *Replica) *DmChannel {
	m.rwmutex.RLock()
	defer m.rwmutex.RUnlock()
	logger := log.With(zap.String("Scope", "ChannelDistManager"), zap.String("channelName", channelName),
		zap.Int64("replicaID", replica.GetID()))
	channels := m.collectionIndex[replica.GetCollectionID()]

	var candidates *DmChannel
	for chIdx, channel := range channels {
		logger := logger.With(zap.Int("channelIdx", chIdx))
		logger.Debug("process", zap.Int64("channelID", channel.Node), zap.Int64("channelVersion", channel.Version),
			zap.String("channel name", channel.GetChannelName()),
			zap.Bool("replicaContains", replica.Contains(channel.Node)))
		if channel.GetChannelName() == channelName && replica.Contains(channel.Node) {
			if candidates == nil {
				candidates = channel
			} else {
				// Prioritize serviceability first, then version number
				candidatesServiceable := candidates.IsServiceable()
				channelServiceable := channel.IsServiceable()

				candidateIsStreamingNode := m.checkIfStreamingNode(candidates.Node)
				channelIsStreamingNode := m.checkIfStreamingNode(channel.Node)
				logger.Debug("check whether stream node is serviceable",
					zap.Bool("candidatesServiceable", candidatesServiceable),
					zap.Bool("channelServiceable", channelServiceable),
					zap.Bool("candidateIsStreamingNode", candidateIsStreamingNode),
					zap.Bool("channelIsStreamingNode", channelIsStreamingNode))

				if channelIsStreamingNode && !candidateIsStreamingNode {
					// When upgrading from 2.5 to 2.6, the delegator leader may not locate at streaming node.
					// We always use the streaming node as the delegator leader to avoid the delete data lost when loading segment.
					logger.Debug("set delegator on stream node to candidate shard leader", zap.Int64("node", channel.Node),
						zap.Int64("channel version", channel.Version))
					candidates = channel
				} else if !channelIsStreamingNode && candidateIsStreamingNode {
					// When downgrading from 2.6 to 2.5, the delegator leader may locate at non-streaming node.
					// We always use the non-streaming node as the delegator leader to avoid the delete data lost when loading segment.
					logger.Debug("found delegator which is not on stream node", zap.Int64("node", channel.Node),
						zap.Int64("channel version", channel.Version))
					continue
				} else {
					updateNeeded := false
					switch {
					case !candidatesServiceable && channelServiceable:
						// Current candidate is not serviceable but new channel is
						updateNeeded = true
						logger.Debug("set serviceable delegator to candidate shard leader", zap.Int64("node", channel.Node),
							zap.Int64("channel version", channel.Version))
					case candidatesServiceable == channelServiceable && channel.Version > candidates.Version:
						// Same service status but higher version
						updateNeeded = true
						logger.Debug("set serviceable delegator with larger version to candidate shard leader", zap.Int64("node", channel.Node),
							zap.Int64("channel version", channel.Version), zap.Int64("candidate version", candidates.Version))
					}
					if updateNeeded {
						candidates = channel
					} else {
						logger.Debug("not set any channel to candidates in this round")
					}
				}
			}
		}
	}
	if candidates != nil {
		logger.Debug("final", zap.Any("candidates", candidates),
			zap.Int64("candidates version", candidates.Version),
			zap.Int64("candidates node", candidates.Node))
	}
	return candidates
}

// checkIfStreamingNode checks if the node is a streaming node.
// Because the session of streaming node and embedded query node are different,
// So we need to check if the node is a streaming node from the query node session but not streaming node session to avoid the wrong check result.
func (m *ChannelDistManager) checkIfStreamingNode(nodeID int64) bool {
	node := m.nodeManager.Get(nodeID)
	if node == nil {
		return false
	}
	return node.IsEmbeddedQueryNodeInStreamingNode() || node.IsInStandalone()
}

func (m *ChannelDistManager) GetChannelDist(collectionID int64) []*metricsinfo.DmChannel {
	m.rwmutex.RLock()
	defer m.rwmutex.RUnlock()

	var ret []*metricsinfo.DmChannel
	if collectionID > 0 {
		if channels, ok := m.collectionIndex[collectionID]; ok {
			for _, channel := range channels {
				ret = append(ret, newDmChannelMetricsFrom(channel))
			}
		}
		return ret
	}

	for _, channels := range m.collectionIndex {
		for _, channel := range channels {
			ret = append(ret, newDmChannelMetricsFrom(channel))
		}
	}
	return ret
}

// GetLeaderView returns a slice of LeaderView objects, each representing the state of a leader node.
// It traverses the views map, converts each LeaderView to a metricsinfo.LeaderView, and collects them into a slice.
// The method locks the views map for reading to ensure thread safety.
func (m *ChannelDistManager) GetLeaderView(collectionID int64) []*metricsinfo.LeaderView {
	m.rwmutex.RLock()
	defer m.rwmutex.RUnlock()

	var ret []*metricsinfo.LeaderView
	if collectionID > 0 {
		if channels, ok := m.collectionIndex[collectionID]; ok {
			for _, channel := range channels {
				ret = append(ret, newMetricsLeaderViewFrom(channel.View))
			}
		}
		return ret
	}

	for _, channels := range m.collectionIndex {
		for _, channel := range channels {
			ret = append(ret, newMetricsLeaderViewFrom(channel.View))
		}
	}
	return ret
}

func newMetricsLeaderViewFrom(lv *LeaderView) *metricsinfo.LeaderView {
	leaderView := &metricsinfo.LeaderView{
		LeaderID:         lv.ID,
		CollectionID:     lv.CollectionID,
		Channel:          lv.Channel,
		Version:          lv.Version,
		SealedSegments:   make([]*metricsinfo.Segment, 0, len(lv.Segments)),
		GrowingSegments:  make([]*metricsinfo.Segment, 0, len(lv.GrowingSegments)),
		TargetVersion:    lv.TargetVersion,
		NumOfGrowingRows: lv.NumOfGrowingRows,
	}

	for segID, seg := range lv.Segments {
		leaderView.SealedSegments = append(leaderView.SealedSegments, &metricsinfo.Segment{
			SegmentID: segID,
			NodeID:    seg.NodeID,
		})
	}

	for _, seg := range lv.GrowingSegments {
		leaderView.GrowingSegments = append(leaderView.GrowingSegments, &metricsinfo.Segment{
			SegmentID: seg.ID,
			NodeID:    seg.Node,
		})
	}
	return leaderView
}