server.go 20.7 KB
Newer Older
S
sunby 已提交
1 2
package dataservice

S
sunby 已提交
3 4
import (
	"context"
S
sunby 已提交
5
	"errors"
S
sunby 已提交
6 7
	"fmt"
	"log"
S
sunby 已提交
8 9
	"path"
	"strconv"
S
sunby 已提交
10
	"sync"
S
sunby 已提交
11 12 13
	"sync/atomic"
	"time"

S
sunby 已提交
14 15
	"github.com/zilliztech/milvus-distributed/internal/distributed/datanode"

S
sunby 已提交
16 17
	"github.com/golang/protobuf/proto"
	"github.com/zilliztech/milvus-distributed/internal/proto/masterpb"
S
sunby 已提交
18

N
neza2017 已提交
19 20
	"github.com/zilliztech/milvus-distributed/internal/msgstream/util"

S
sunby 已提交
21 22 23
	"github.com/zilliztech/milvus-distributed/internal/msgstream"
	"github.com/zilliztech/milvus-distributed/internal/msgstream/pulsarms"

S
sunby 已提交
24 25 26 27 28 29 30 31 32 33 34 35 36
	"github.com/zilliztech/milvus-distributed/internal/proto/milvuspb"

	"github.com/zilliztech/milvus-distributed/internal/timesync"

	etcdkv "github.com/zilliztech/milvus-distributed/internal/kv/etcd"
	"go.etcd.io/etcd/clientv3"

	"github.com/zilliztech/milvus-distributed/internal/proto/commonpb"
	"github.com/zilliztech/milvus-distributed/internal/proto/datapb"
	"github.com/zilliztech/milvus-distributed/internal/proto/internalpb2"
	"github.com/zilliztech/milvus-distributed/internal/util/typeutil"
)

S
sunby 已提交
37 38
const role = "dataservice"

S
sunby 已提交
39 40
type DataService interface {
	typeutil.Service
N
neza2017 已提交
41
	typeutil.Component
S
sunby 已提交
42 43 44 45 46 47 48
	RegisterNode(req *datapb.RegisterNodeRequest) (*datapb.RegisterNodeResponse, error)
	Flush(req *datapb.FlushRequest) (*commonpb.Status, error)

	AssignSegmentID(req *datapb.AssignSegIDRequest) (*datapb.AssignSegIDResponse, error)
	ShowSegments(req *datapb.ShowSegmentRequest) (*datapb.ShowSegmentResponse, error)
	GetSegmentStates(req *datapb.SegmentStatesRequest) (*datapb.SegmentStatesResponse, error)
	GetInsertBinlogPaths(req *datapb.InsertBinlogPathRequest) (*datapb.InsertBinlogPathsResponse, error)
N
neza2017 已提交
49 50
	GetSegmentInfoChannel() (string, error)
	GetInsertChannels(req *datapb.InsertChannelRequest) ([]string, error)
S
sunby 已提交
51 52 53 54 55
	GetCollectionStatistics(req *datapb.CollectionStatsRequest) (*datapb.CollectionStatsResponse, error)
	GetPartitionStatistics(req *datapb.PartitionStatsRequest) (*datapb.PartitionStatsResponse, error)
	GetComponentStates() (*internalpb2.ComponentStates, error)
}

S
sunby 已提交
56 57 58 59 60 61 62 63 64 65
type MasterClient interface {
	ShowCollections(in *milvuspb.ShowCollectionRequest) (*milvuspb.ShowCollectionResponse, error)
	DescribeCollection(in *milvuspb.DescribeCollectionRequest) (*milvuspb.DescribeCollectionResponse, error)
	ShowPartitions(in *milvuspb.ShowPartitionRequest) (*milvuspb.ShowPartitionResponse, error)
	GetDdChannel() (string, error)
	AllocTimestamp(in *masterpb.TsoRequest) (*masterpb.TsoResponse, error)
	AllocID(in *masterpb.IDRequest) (*masterpb.IDResponse, error)
	GetComponentStates() (*internalpb2.ComponentStates, error)
}

S
sunby 已提交
66 67 68 69 70 71 72
type DataNodeClient interface {
	WatchDmChannels(in *datapb.WatchDmChannelRequest) (*commonpb.Status, error)
	GetComponentStates(empty *commonpb.Empty) (*internalpb2.ComponentStates, error)
	FlushSegments(in *datapb.FlushSegRequest) (*commonpb.Status, error)
	Stop() error
}

S
sunby 已提交
73
type (
S
sunby 已提交
74 75 76
	UniqueID  = typeutil.UniqueID
	Timestamp = typeutil.Timestamp
	Server    struct {
N
neza2017 已提交
77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96
		ctx               context.Context
		serverLoopCtx     context.Context
		serverLoopCancel  context.CancelFunc
		serverLoopWg      sync.WaitGroup
		state             atomic.Value
		client            *etcdkv.EtcdKV
		meta              *meta
		segAllocator      segmentAllocator
		statsHandler      *statsHandler
		ddHandler         *ddHandler
		insertChannelMgr  *insertChannelManager
		allocator         allocator
		cluster           *dataNodeCluster
		msgProducer       *timesync.MsgProducer
		registerFinishCh  chan struct{}
		masterClient      MasterClient
		ttMsgStream       msgstream.MsgStream
		k2sMsgStream      msgstream.MsgStream
		ddChannelName     string
		segmentInfoStream msgstream.MsgStream
S
sunby 已提交
97 98 99
	}
)

S
sunby 已提交
100
func CreateServer(ctx context.Context) (*Server, error) {
101
	Params.Init()
S
sunby 已提交
102
	ch := make(chan struct{})
S
sunby 已提交
103
	s := &Server{
S
sunby 已提交
104 105
		ctx:              ctx,
		insertChannelMgr: newInsertChannelManager(),
S
sunby 已提交
106 107
		registerFinishCh: ch,
		cluster:          newDataNodeCluster(ch),
S
sunby 已提交
108
	}
S
sunby 已提交
109
	s.state.Store(internalpb2.StateCode_INITIALIZING)
S
sunby 已提交
110 111 112 113 114
	return s, nil
}

func (s *Server) SetMasterClient(masterClient MasterClient) {
	s.masterClient = masterClient
S
sunby 已提交
115 116 117
}

func (s *Server) Init() error {
S
sunby 已提交
118 119 120 121
	return nil
}

func (s *Server) Start() error {
122
	var err error
S
sunby 已提交
123
	s.allocator = newAllocatorImpl(s.masterClient)
124
	if err = s.initMeta(); err != nil {
S
sunby 已提交
125 126 127
		return err
	}
	s.statsHandler = newStatsHandler(s.meta)
128
	s.segAllocator, err = newSegmentAllocator(s.meta, s.allocator)
S
sunby 已提交
129 130 131
	if err != nil {
		return err
	}
N
neza2017 已提交
132
	s.ddHandler = newDDHandler(s.meta, s.segAllocator)
133 134
	s.initSegmentInfoChannel()
	if err = s.loadMetaFromMaster(); err != nil {
S
sunby 已提交
135 136
		return err
	}
S
sunby 已提交
137
	s.startServerLoop()
138
	s.waitDataNodeRegister()
139 140 141
	if err = s.initMsgProducer(); err != nil {
		return err
	}
S
sunby 已提交
142
	s.state.Store(internalpb2.StateCode_HEALTHY)
S
sunby 已提交
143 144 145 146
	log.Println("start success")
	return nil
}

S
sunby 已提交
147 148 149 150
func (s *Server) checkStateIsHealthy() bool {
	return s.state.Load().(internalpb2.StateCode) == internalpb2.StateCode_HEALTHY
}

S
sunby 已提交
151 152 153 154 155 156 157
func (s *Server) initMeta() error {
	etcdClient, err := clientv3.New(clientv3.Config{Endpoints: []string{Params.EtcdAddress}})
	if err != nil {
		return err
	}
	etcdKV := etcdkv.NewEtcdKV(etcdClient, Params.MetaRootPath)
	s.client = etcdKV
N
neza2017 已提交
158
	s.meta, err = newMeta(etcdKV)
S
sunby 已提交
159 160 161 162 163 164
	if err != nil {
		return err
	}
	return nil
}

165 166 167 168 169 170
func (s *Server) initSegmentInfoChannel() {
	segmentInfoStream := pulsarms.NewPulsarMsgStream(s.ctx, 1024)
	segmentInfoStream.SetPulsarClient(Params.PulsarAddress)
	segmentInfoStream.CreatePulsarProducers([]string{Params.SegmentInfoChannelName})
	s.segmentInfoStream = segmentInfoStream
	s.segmentInfoStream.Start()
S
sunby 已提交
171
}
S
sunby 已提交
172
func (s *Server) initMsgProducer() error {
173
	ttMsgStream := pulsarms.NewPulsarMsgStream(s.ctx, 1024)
N
neza2017 已提交
174 175 176
	ttMsgStream.SetPulsarClient(Params.PulsarAddress)
	ttMsgStream.CreatePulsarConsumers([]string{Params.TimeTickChannelName}, Params.DataServiceSubscriptionName, util.NewUnmarshalDispatcher(), 1024)
	s.ttMsgStream = ttMsgStream
S
sunby 已提交
177 178 179
	s.ttMsgStream.Start()
	timeTickBarrier := timesync.NewHardTimeTickBarrier(s.ttMsgStream, s.cluster.GetNodeIDs())
	dataNodeTTWatcher := newDataNodeTimeTickWatcher(s.meta, s.segAllocator, s.cluster)
180 181 182 183 184 185 186
	k2sStream := pulsarms.NewPulsarMsgStream(s.ctx, 1024)
	k2sStream.SetPulsarClient(Params.PulsarAddress)
	k2sStream.CreatePulsarProducers(Params.K2SChannelNames)
	s.k2sMsgStream = k2sStream
	s.k2sMsgStream.Start()
	k2sMsgWatcher := timesync.NewMsgTimeTickWatcher(s.k2sMsgStream)
	producer, err := timesync.NewTimeSyncMsgProducer(timeTickBarrier, dataNodeTTWatcher, k2sMsgWatcher)
S
sunby 已提交
187 188 189 190 191 192 193
	if err != nil {
		return err
	}
	s.msgProducer = producer
	s.msgProducer.Start(s.ctx)
	return nil
}
S
sunby 已提交
194

S
sunby 已提交
195 196
func (s *Server) loadMetaFromMaster() error {
	log.Println("loading collection meta from master")
S
sunby 已提交
197 198 199
	if err := s.checkMasterIsHealthy(); err != nil {
		return err
	}
N
neza2017 已提交
200 201 202 203 204 205 206
	if s.ddChannelName == "" {
		channel, err := s.masterClient.GetDdChannel()
		if err != nil {
			return err
		}
		s.ddChannelName = channel
	}
S
sunby 已提交
207 208 209 210 211
	collections, err := s.masterClient.ShowCollections(&milvuspb.ShowCollectionRequest{
		Base: &commonpb.MsgBase{
			MsgType:   commonpb.MsgType_kShowCollections,
			MsgID:     -1, // todo add msg id
			Timestamp: 0,  // todo
S
sunby 已提交
212
			SourceID:  Params.NodeID,
S
sunby 已提交
213 214 215 216 217 218 219 220 221 222 223 224
		},
		DbName: "",
	})
	if err != nil {
		return err
	}
	for _, collectionName := range collections.CollectionNames {
		collection, err := s.masterClient.DescribeCollection(&milvuspb.DescribeCollectionRequest{
			Base: &commonpb.MsgBase{
				MsgType:   commonpb.MsgType_kDescribeCollection,
				MsgID:     -1, // todo
				Timestamp: 0,  // todo
S
sunby 已提交
225
				SourceID:  Params.NodeID,
S
sunby 已提交
226 227 228 229 230 231 232 233 234 235 236 237 238
			},
			DbName:         "",
			CollectionName: collectionName,
		})
		if err != nil {
			log.Println(err.Error())
			continue
		}
		partitions, err := s.masterClient.ShowPartitions(&milvuspb.ShowPartitionRequest{
			Base: &commonpb.MsgBase{
				MsgType:   commonpb.MsgType_kShowPartitions,
				MsgID:     -1, // todo
				Timestamp: 0,  // todo
S
sunby 已提交
239
				SourceID:  Params.NodeID,
S
sunby 已提交
240 241 242 243 244 245 246 247 248 249 250 251
			},
			DbName:         "",
			CollectionName: collectionName,
			CollectionID:   collection.CollectionID,
		})
		if err != nil {
			log.Println(err.Error())
			continue
		}
		err = s.meta.AddCollection(&collectionInfo{
			ID:         collection.CollectionID,
			Schema:     collection.Schema,
S
sunby 已提交
252
			Partitions: partitions.PartitionIDs,
S
sunby 已提交
253 254 255 256 257 258 259 260
		})
		if err != nil {
			log.Println(err.Error())
			continue
		}
	}
	log.Println("load collection meta from master complete")
	return nil
S
sunby 已提交
261
}
S
sunby 已提交
262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291

func (s *Server) checkMasterIsHealthy() error {
	ticker := time.NewTicker(300 * time.Millisecond)
	ctx, cancel := context.WithTimeout(s.ctx, 30*time.Second)
	defer func() {
		ticker.Stop()
		cancel()
	}()
	for {
		var resp *internalpb2.ComponentStates
		var err error
		select {
		case <-ctx.Done():
			return fmt.Errorf("master is not healthy")
		case <-ticker.C:
			resp, err = s.masterClient.GetComponentStates()
			if err != nil {
				return err
			}
			if resp.Status.ErrorCode != commonpb.ErrorCode_SUCCESS {
				return errors.New(resp.Status.Reason)
			}
		}
		if resp.State.StateCode == internalpb2.StateCode_HEALTHY {
			break
		}
	}
	return nil
}

292 293
func (s *Server) startServerLoop() {
	s.serverLoopCtx, s.serverLoopCancel = context.WithCancel(s.ctx)
N
neza2017 已提交
294
	s.serverLoopWg.Add(3)
295 296
	go s.startStatsChannel(s.serverLoopCtx)
	go s.startSegmentFlushChannel(s.serverLoopCtx)
N
neza2017 已提交
297
	go s.startDDChannel(s.serverLoopCtx)
298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360
}

func (s *Server) startStatsChannel(ctx context.Context) {
	defer s.serverLoopWg.Done()
	statsStream := pulsarms.NewPulsarMsgStream(ctx, 1024)
	statsStream.SetPulsarClient(Params.PulsarAddress)
	statsStream.CreatePulsarConsumers([]string{Params.StatisticsChannelName}, Params.DataServiceSubscriptionName, util.NewUnmarshalDispatcher(), 1024)
	statsStream.Start()
	defer statsStream.Close()
	for {
		select {
		case <-ctx.Done():
			return
		default:
		}
		msgPack := statsStream.Consume()
		for _, msg := range msgPack.Msgs {
			statistics := msg.(*msgstream.SegmentStatisticsMsg)
			for _, stat := range statistics.SegStats {
				if err := s.statsHandler.HandleSegmentStat(stat); err != nil {
					log.Println(err.Error())
					continue
				}
			}
		}
	}
}

func (s *Server) startSegmentFlushChannel(ctx context.Context) {
	defer s.serverLoopWg.Done()
	flushStream := pulsarms.NewPulsarMsgStream(ctx, 1024)
	flushStream.SetPulsarClient(Params.PulsarAddress)
	flushStream.CreatePulsarConsumers([]string{Params.SegmentInfoChannelName}, Params.DataServiceSubscriptionName, util.NewUnmarshalDispatcher(), 1024)
	flushStream.Start()
	defer flushStream.Close()
	for {
		select {
		case <-ctx.Done():
			log.Println("segment flush channel shut down")
			return
		default:
		}
		msgPack := flushStream.Consume()
		for _, msg := range msgPack.Msgs {
			if msg.Type() != commonpb.MsgType_kSegmentFlushDone {
				continue
			}
			realMsg := msg.(*msgstream.FlushCompletedMsg)

			segmentInfo, err := s.meta.GetSegment(realMsg.SegmentID)
			if err != nil {
				log.Println(err.Error())
				continue
			}
			segmentInfo.FlushedTime = realMsg.BeginTimestamp
			if err = s.meta.UpdateSegment(segmentInfo); err != nil {
				log.Println(err.Error())
				continue
			}
		}
	}
}

N
neza2017 已提交
361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384
func (s *Server) startDDChannel(ctx context.Context) {
	defer s.serverLoopWg.Done()
	ddStream := pulsarms.NewPulsarMsgStream(ctx, 1024)
	ddStream.SetPulsarClient(Params.PulsarAddress)
	ddStream.CreatePulsarConsumers([]string{s.ddChannelName}, Params.DataServiceSubscriptionName, util.NewUnmarshalDispatcher(), 1024)
	ddStream.Start()
	defer ddStream.Close()
	for {
		select {
		case <-ctx.Done():
			log.Println("dd channel shut down")
			return
		default:
		}
		msgPack := ddStream.Consume()
		for _, msg := range msgPack.Msgs {
			if err := s.ddHandler.HandleDDMsg(msg); err != nil {
				log.Println(err.Error())
				continue
			}
		}
	}
}

385 386 387 388 389
func (s *Server) waitDataNodeRegister() {
	log.Println("waiting data node to register")
	<-s.registerFinishCh
	log.Println("all data nodes register")
}
S
sunby 已提交
390 391

func (s *Server) Stop() error {
S
sunby 已提交
392
	s.cluster.ShutDownClients()
S
sunby 已提交
393
	s.ttMsgStream.Close()
394
	s.k2sMsgStream.Close()
S
sunby 已提交
395
	s.msgProducer.Close()
N
neza2017 已提交
396
	s.segmentInfoStream.Close()
S
sunby 已提交
397
	s.stopServerLoop()
S
sunby 已提交
398 399 400
	return nil
}

S
sunby 已提交
401 402 403 404 405
func (s *Server) stopServerLoop() {
	s.serverLoopCancel()
	s.serverLoopWg.Wait()
}

S
sunby 已提交
406
func (s *Server) GetComponentStates() (*internalpb2.ComponentStates, error) {
S
sunby 已提交
407 408 409 410
	resp := &internalpb2.ComponentStates{
		State: &internalpb2.ComponentInfo{
			NodeID:    Params.NodeID,
			Role:      role,
S
sunby 已提交
411
			StateCode: s.state.Load().(internalpb2.StateCode),
S
sunby 已提交
412 413 414 415 416 417 418 419 420 421 422 423 424
		},
		Status: &commonpb.Status{
			ErrorCode: commonpb.ErrorCode_UNEXPECTED_ERROR,
		},
	}
	dataNodeStates, err := s.cluster.GetDataNodeStates()
	if err != nil {
		resp.Status.Reason = err.Error()
		return resp, nil
	}
	resp.SubcomponentStates = dataNodeStates
	resp.Status.ErrorCode = commonpb.ErrorCode_SUCCESS
	return resp, nil
S
sunby 已提交
425 426
}

N
neza2017 已提交
427 428
func (s *Server) GetTimeTickChannel() (string, error) {
	return Params.TimeTickChannelName, nil
S
sunby 已提交
429 430
}

N
neza2017 已提交
431 432
func (s *Server) GetStatisticsChannel() (string, error) {
	return Params.StatisticsChannelName, nil
S
sunby 已提交
433 434 435
}

func (s *Server) RegisterNode(req *datapb.RegisterNodeRequest) (*datapb.RegisterNodeResponse, error) {
S
sunby 已提交
436
	ret := &datapb.RegisterNodeResponse{
S
sunby 已提交
437
		Status: &commonpb.Status{
S
sunby 已提交
438
			ErrorCode: commonpb.ErrorCode_UNEXPECTED_ERROR,
S
sunby 已提交
439
		},
S
sunby 已提交
440
	}
S
sunby 已提交
441 442 443 444 445
	node, err := s.newDataNode(req.Address.Ip, req.Address.Port, req.Base.SourceID)
	if err != nil {
		return nil, err
	}
	s.cluster.Register(node)
N
neza2017 已提交
446
	if s.ddChannelName == "" {
N
neza2017 已提交
447
		resp, err := s.masterClient.GetDdChannel()
S
sunby 已提交
448 449 450 451
		if err != nil {
			ret.Status.Reason = err.Error()
			return ret, err
		}
N
neza2017 已提交
452
		s.ddChannelName = resp
S
sunby 已提交
453 454 455 456 457 458 459 460
	}
	ret.Status.ErrorCode = commonpb.ErrorCode_SUCCESS
	ret.InitParams = &internalpb2.InitParams{
		NodeID: Params.NodeID,
		StartParams: []*commonpb.KeyValuePair{
			{Key: "DDChannelName", Value: s.ddChannelName},
			{Key: "SegmentStatisticsChannelName", Value: Params.StatisticsChannelName},
			{Key: "TimeTickChannelName", Value: Params.TimeTickChannelName},
N
neza2017 已提交
461
			{Key: "CompleteFlushChannelName", Value: Params.SegmentInfoChannelName},
S
sunby 已提交
462 463 464
		},
	}
	return ret, nil
S
sunby 已提交
465 466
}

S
sunby 已提交
467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485
func (s *Server) newDataNode(ip string, port int64, id UniqueID) (*dataNode, error) {
	client := datanode.NewClient(fmt.Sprintf("%s:%d", ip, port))
	if err := client.Init(); err != nil {
		return nil, err
	}
	if err := client.Start(); err != nil {
		return nil, err
	}
	return &dataNode{
		id: id,
		address: struct {
			ip   string
			port int64
		}{ip: ip, port: port},
		client:     client,
		channelNum: 0,
	}, nil
}

S
sunby 已提交
486
func (s *Server) Flush(req *datapb.FlushRequest) (*commonpb.Status, error) {
S
sunby 已提交
487 488 489 490 491 492
	if !s.checkStateIsHealthy() {
		return &commonpb.Status{
			ErrorCode: commonpb.ErrorCode_UNEXPECTED_ERROR,
			Reason:    "server is initializing",
		}, nil
	}
N
neza2017 已提交
493
	s.segAllocator.SealAllSegments(req.CollectionID)
S
sunby 已提交
494 495 496
	return &commonpb.Status{
		ErrorCode: commonpb.ErrorCode_SUCCESS,
	}, nil
S
sunby 已提交
497 498 499 500 501 502 503 504 505
}

func (s *Server) AssignSegmentID(req *datapb.AssignSegIDRequest) (*datapb.AssignSegIDResponse, error) {
	resp := &datapb.AssignSegIDResponse{
		Status: &commonpb.Status{
			ErrorCode: commonpb.ErrorCode_SUCCESS,
		},
		SegIDAssignments: make([]*datapb.SegIDAssignment, 0),
	}
S
sunby 已提交
506 507 508 509 510
	if !s.checkStateIsHealthy() {
		resp.Status.ErrorCode = commonpb.ErrorCode_UNEXPECTED_ERROR
		resp.Status.Reason = "server is initializing"
		return resp, nil
	}
S
sunby 已提交
511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560
	for _, r := range req.SegIDRequests {
		result := &datapb.SegIDAssignment{
			Status: &commonpb.Status{
				ErrorCode: commonpb.ErrorCode_UNEXPECTED_ERROR,
			},
		}
		segmentID, retCount, expireTs, err := s.segAllocator.AllocSegment(r.CollectionID, r.PartitionID, r.ChannelName, int(r.Count))
		if err != nil {
			if _, ok := err.(errRemainInSufficient); !ok {
				result.Status.Reason = fmt.Sprintf("allocation of Collection %d, Partition %d, Channel %s, Count %d error:  %s",
					r.CollectionID, r.PartitionID, r.ChannelName, r.Count, err.Error())
				resp.SegIDAssignments = append(resp.SegIDAssignments, result)
				continue
			}

			log.Printf("no enough space for allocation of Collection %d, Partition %d, Channel %s, Count %d",
				r.CollectionID, r.PartitionID, r.ChannelName, r.Count)
			if err = s.openNewSegment(r.CollectionID, r.PartitionID, r.ChannelName); err != nil {
				result.Status.Reason = fmt.Sprintf("open new segment of Collection %d, Partition %d, Channel %s, Count %d error:  %s",
					r.CollectionID, r.PartitionID, r.ChannelName, r.Count, err.Error())
				resp.SegIDAssignments = append(resp.SegIDAssignments, result)
				continue
			}

			segmentID, retCount, expireTs, err = s.segAllocator.AllocSegment(r.CollectionID, r.PartitionID, r.ChannelName, int(r.Count))
			if err != nil {
				result.Status.Reason = fmt.Sprintf("retry allocation of Collection %d, Partition %d, Channel %s, Count %d error:  %s",
					r.CollectionID, r.PartitionID, r.ChannelName, r.Count, err.Error())
				resp.SegIDAssignments = append(resp.SegIDAssignments, result)
				continue
			}
		}

		result.Status.ErrorCode = commonpb.ErrorCode_SUCCESS
		result.CollectionID = r.CollectionID
		result.SegID = segmentID
		result.PartitionID = r.PartitionID
		result.Count = uint32(retCount)
		result.ExpireTime = expireTs
		result.ChannelName = r.ChannelName
		resp.SegIDAssignments = append(resp.SegIDAssignments, result)
	}
	return resp, nil
}

func (s *Server) openNewSegment(collectionID UniqueID, partitionID UniqueID, channelName string) error {
	group, err := s.insertChannelMgr.GetChannelGroup(collectionID, channelName)
	if err != nil {
		return err
	}
N
neza2017 已提交
561 562 563 564 565 566

	id, err := s.allocator.allocID()
	if err != nil {
		return err
	}
	segmentInfo, err := BuildSegment(collectionID, partitionID, id, group)
S
sunby 已提交
567 568 569 570 571 572
	if err != nil {
		return err
	}
	if err = s.meta.AddSegment(segmentInfo); err != nil {
		return err
	}
S
sunby 已提交
573
	if err = s.segAllocator.OpenSegment(segmentInfo); err != nil {
S
sunby 已提交
574 575
		return err
	}
576 577 578 579 580
	infoMsg := &msgstream.SegmentInfoMsg{
		SegmentMsg: datapb.SegmentMsg{
			Base: &commonpb.MsgBase{
				MsgType:   commonpb.MsgType_kSegmentInfo,
				MsgID:     0,
N
neza2017 已提交
581 582
				Timestamp: 0,
				SourceID:  Params.NodeID,
583 584 585 586 587 588 589 590 591 592
			},
			Segment: segmentInfo,
		},
	}
	msgPack := &pulsarms.MsgPack{
		Msgs: []msgstream.TsMsg{infoMsg},
	}
	if err = s.segmentInfoStream.Produce(msgPack); err != nil {
		return err
	}
S
sunby 已提交
593 594 595 596
	return nil
}

func (s *Server) ShowSegments(req *datapb.ShowSegmentRequest) (*datapb.ShowSegmentResponse, error) {
S
sunby 已提交
597 598 599 600 601 602 603 604
	if !s.checkStateIsHealthy() {
		return &datapb.ShowSegmentResponse{
			Status: &commonpb.Status{
				ErrorCode: commonpb.ErrorCode_UNEXPECTED_ERROR,
				Reason:    "server is initializing",
			},
		}, nil
	}
S
sunby 已提交
605 606 607 608 609 610 611 612 613 614
	ids := s.meta.GetSegmentsByCollectionAndPartitionID(req.CollectionID, req.PartitionID)
	return &datapb.ShowSegmentResponse{SegmentIDs: ids}, nil
}

func (s *Server) GetSegmentStates(req *datapb.SegmentStatesRequest) (*datapb.SegmentStatesResponse, error) {
	resp := &datapb.SegmentStatesResponse{
		Status: &commonpb.Status{
			ErrorCode: commonpb.ErrorCode_UNEXPECTED_ERROR,
		},
	}
S
sunby 已提交
615 616 617 618
	if !s.checkStateIsHealthy() {
		resp.Status.Reason = "server is initializing"
		return resp, nil
	}
S
sunby 已提交
619 620 621 622 623 624 625 626 627 628

	segmentInfo, err := s.meta.GetSegment(req.SegmentID)
	if err != nil {
		resp.Status.Reason = "get segment states error: " + err.Error()
		return resp, nil
	}
	resp.State = segmentInfo.State
	resp.CreateTime = segmentInfo.OpenTime
	resp.SealedTime = segmentInfo.SealedTime
	resp.FlushedTime = segmentInfo.FlushedTime
629 630
	resp.StartPositions = segmentInfo.StartPosition
	resp.EndPositions = segmentInfo.EndPosition
S
sunby 已提交
631 632 633 634
	return resp, nil
}

func (s *Server) GetInsertBinlogPaths(req *datapb.InsertBinlogPathRequest) (*datapb.InsertBinlogPathsResponse, error) {
S
sunby 已提交
635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661
	// todo
	resp := &datapb.InsertBinlogPathsResponse{
		Status: &commonpb.Status{
			ErrorCode: commonpb.ErrorCode_UNEXPECTED_ERROR,
		},
	}
	p := path.Join(Params.SegmentFlushMetaPath, strconv.FormatInt(req.SegmentID, 10))
	value, err := s.client.Load(p)
	if err != nil {
		resp.Status.Reason = err.Error()
		return resp, nil
	}
	flushMeta := &datapb.SegmentFlushMeta{}
	err = proto.UnmarshalText(value, flushMeta)
	if err != nil {
		resp.Status.Reason = err.Error()
		return resp, nil
	}
	fields := make([]UniqueID, len(flushMeta.Fields))
	paths := make([]*internalpb2.StringList, len(flushMeta.Fields))
	for _, field := range flushMeta.Fields {
		fields = append(fields, field.FieldID)
		paths = append(paths, &internalpb2.StringList{Values: field.BinlogPaths})
	}
	resp.FieldIDs = fields
	resp.Paths = paths
	return resp, nil
S
sunby 已提交
662 663
}

N
neza2017 已提交
664
func (s *Server) GetInsertChannels(req *datapb.InsertChannelRequest) ([]string, error) {
S
sunby 已提交
665 666 667
	if !s.checkStateIsHealthy() {
		return nil, errors.New("server is initializing")
	}
S
sunby 已提交
668 669
	contains, ret := s.insertChannelMgr.ContainsCollection(req.CollectionID)
	if contains {
N
neza2017 已提交
670
		return ret, nil
S
sunby 已提交
671
	}
S
sunby 已提交
672
	channelGroups, err := s.insertChannelMgr.AllocChannels(req.CollectionID, s.cluster.GetNumOfNodes())
S
sunby 已提交
673
	if err != nil {
N
neza2017 已提交
674
		return nil, err
S
sunby 已提交
675
	}
S
sunby 已提交
676

S
sunby 已提交
677 678
	channels := make([]string, Params.InsertChannelNumPerCollection)
	for _, group := range channelGroups {
S
sunby 已提交
679
		channels = append(channels, group...)
S
sunby 已提交
680
	}
S
sunby 已提交
681 682
	s.cluster.WatchInsertChannels(channelGroups)

N
neza2017 已提交
683
	return channels, nil
S
sunby 已提交
684 685 686 687 688 689 690 691 692 693
}

func (s *Server) GetCollectionStatistics(req *datapb.CollectionStatsRequest) (*datapb.CollectionStatsResponse, error) {
	// todo implement
	return nil, nil
}

func (s *Server) GetPartitionStatistics(req *datapb.PartitionStatsRequest) (*datapb.PartitionStatsResponse, error) {
	// todo implement
	return nil, nil
S
sunby 已提交
694
}
N
neza2017 已提交
695 696 697 698

func (s *Server) GetSegmentInfoChannel() (string, error) {
	return Params.SegmentInfoChannelName, nil
}