server.go 22.5 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 19 20 21

	"github.com/zilliztech/milvus-distributed/internal/msgstream"
	"github.com/zilliztech/milvus-distributed/internal/msgstream/pulsarms"

S
sunby 已提交
22 23 24 25 26 27 28 29 30 31 32 33 34
	"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
35 36
const role = "dataservice"

S
sunby 已提交
37 38
type DataService interface {
	typeutil.Service
N
neza2017 已提交
39
	typeutil.Component
S
sunby 已提交
40 41 42 43 44 45 46
	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
47 48
	GetSegmentInfoChannel() (string, error)
	GetInsertChannels(req *datapb.InsertChannelRequest) ([]string, error)
S
sunby 已提交
49 50 51
	GetCollectionStatistics(req *datapb.CollectionStatsRequest) (*datapb.CollectionStatsResponse, error)
	GetPartitionStatistics(req *datapb.PartitionStatsRequest) (*datapb.PartitionStatsResponse, error)
	GetComponentStates() (*internalpb2.ComponentStates, error)
52
	GetCount(req *datapb.CollectionCountRequest) (*datapb.CollectionCountResponse, error)
X
XuanYang-cn 已提交
53
	GetSegmentInfo(req *datapb.SegmentInfoRequest) (*datapb.SegmentInfoResponse, error)
S
sunby 已提交
54 55
}

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
		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
		allocator         allocator
		cluster           *dataNodeCluster
		msgProducer       *timesync.MsgProducer
		registerFinishCh  chan struct{}
		masterClient      MasterClient
		ttMsgStream       msgstream.MsgStream
		k2sMsgStream      msgstream.MsgStream
		ddChannelName     string
		segmentInfoStream msgstream.MsgStream
96
		insertChannels    []string
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
		ctx:              ctx,
S
sunby 已提交
105 106
		registerFinishCh: ch,
		cluster:          newDataNodeCluster(ch),
S
sunby 已提交
107
	}
108
	s.insertChannels = s.getInsertChannels()
N
neza2017 已提交
109
	s.state.Store(internalpb2.StateCode_INITIALIZING)
S
sunby 已提交
110 111 112
	return s, nil
}

113 114 115 116 117 118 119 120 121
func (s *Server) getInsertChannels() []string {
	channels := make([]string, Params.InsertChannelNum)
	var i int64 = 0
	for ; i < Params.InsertChannelNum; i++ {
		channels[i] = Params.InsertChannelPrefixName + strconv.FormatInt(i, 10)
	}
	return channels
}

S
sunby 已提交
122 123
func (s *Server) SetMasterClient(masterClient MasterClient) {
	s.masterClient = masterClient
S
sunby 已提交
124 125 126
}

func (s *Server) Init() error {
S
sunby 已提交
127 128 129 130
	return nil
}

func (s *Server) Start() error {
131
	var err error
S
sunby 已提交
132
	s.allocator = newAllocatorImpl(s.masterClient)
133
	if err = s.initMeta(); err != nil {
S
sunby 已提交
134 135 136
		return err
	}
	s.statsHandler = newStatsHandler(s.meta)
Y
yukun 已提交
137
	s.segAllocator = newSegmentAllocator(s.meta, s.allocator)
N
neza2017 已提交
138
	s.ddHandler = newDDHandler(s.meta, s.segAllocator)
139 140
	s.initSegmentInfoChannel()
	if err = s.loadMetaFromMaster(); err != nil {
S
sunby 已提交
141 142
		return err
	}
S
sunby 已提交
143
	s.startServerLoop()
144
	s.waitDataNodeRegister()
145
	s.cluster.WatchInsertChannels(s.insertChannels)
146 147 148
	if err = s.initMsgProducer(); err != nil {
		return err
	}
S
sunby 已提交
149
	s.state.Store(internalpb2.StateCode_HEALTHY)
S
sunby 已提交
150 151 152 153
	log.Println("start success")
	return nil
}

S
sunby 已提交
154 155 156 157
func (s *Server) checkStateIsHealthy() bool {
	return s.state.Load().(internalpb2.StateCode) == internalpb2.StateCode_HEALTHY
}

S
sunby 已提交
158 159 160 161 162 163 164
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
165
	s.meta, err = newMeta(etcdKV)
S
sunby 已提交
166 167 168 169 170 171
	if err != nil {
		return err
	}
	return nil
}

172
func (s *Server) initSegmentInfoChannel() {
Z
zhenshan.cao 已提交
173 174 175
	factory := pulsarms.NewFactory(Params.PulsarAddress, 1024, 1024)
	segmentInfoStream, _ := factory.NewMsgStream(s.ctx)
	segmentInfoStream.AsProducer([]string{Params.SegmentInfoChannelName})
176 177
	s.segmentInfoStream = segmentInfoStream
	s.segmentInfoStream.Start()
S
sunby 已提交
178
}
S
sunby 已提交
179
func (s *Server) initMsgProducer() error {
Z
zhenshan.cao 已提交
180
	factory := pulsarms.NewFactory(Params.PulsarAddress, 1024, 1024)
181 182 183
	ttMsgStream, _ := factory.NewMsgStream(s.ctx)
	ttMsgStream.AsConsumer([]string{Params.TimeTickChannelName}, Params.DataServiceSubscriptionName)
	s.ttMsgStream = ttMsgStream
S
sunby 已提交
184
	s.ttMsgStream.Start()
185
	timeTickBarrier := timesync.NewHardTimeTickBarrier(s.ttMsgStream, s.cluster.GetNodeIDs())
S
sunby 已提交
186
	dataNodeTTWatcher := newDataNodeTimeTickWatcher(s.meta, s.segAllocator, s.cluster)
187 188 189 190
	k2sStream, _ := factory.NewMsgStream(s.ctx)
	k2sStream.AsProducer(Params.K2SChannelNames)
	s.k2sMsgStream = k2sStream
	s.k2sMsgStream.Start()
191
	k2sMsgWatcher := timesync.NewMsgTimeTickWatcher(s.k2sMsgStream)
192 193
	producer, err := timesync.NewTimeSyncMsgProducer(timeTickBarrier, dataNodeTTWatcher, k2sMsgWatcher)
	if err != nil {
S
sunby 已提交
194 195
		return err
	}
196
	s.msgProducer = producer
S
sunby 已提交
197 198 199
	s.msgProducer.Start(s.ctx)
	return nil
}
S
sunby 已提交
200

S
sunby 已提交
201 202
func (s *Server) loadMetaFromMaster() error {
	log.Println("loading collection meta from master")
S
sunby 已提交
203 204 205
	if err := s.checkMasterIsHealthy(); err != nil {
		return err
	}
N
neza2017 已提交
206 207 208 209 210 211 212
	if s.ddChannelName == "" {
		channel, err := s.masterClient.GetDdChannel()
		if err != nil {
			return err
		}
		s.ddChannelName = channel
	}
S
sunby 已提交
213 214 215 216 217
	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
218
			SourceID:  Params.NodeID,
S
sunby 已提交
219 220 221 222 223 224 225 226 227 228 229 230
		},
		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
231
				SourceID:  Params.NodeID,
S
sunby 已提交
232 233 234 235 236 237 238 239 240 241 242 243 244
			},
			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
245
				SourceID:  Params.NodeID,
S
sunby 已提交
246 247 248 249 250 251 252 253 254 255 256 257
			},
			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
258
			Partitions: partitions.PartitionIDs,
S
sunby 已提交
259 260 261 262 263 264 265 266
		})
		if err != nil {
			log.Println(err.Error())
			continue
		}
	}
	log.Println("load collection meta from master complete")
	return nil
S
sunby 已提交
267
}
S
sunby 已提交
268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297

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
}

298 299
func (s *Server) startServerLoop() {
	s.serverLoopCtx, s.serverLoopCancel = context.WithCancel(s.ctx)
300
	s.serverLoopWg.Add(3)
301 302
	go s.startStatsChannel(s.serverLoopCtx)
	go s.startSegmentFlushChannel(s.serverLoopCtx)
N
neza2017 已提交
303
	go s.startDDChannel(s.serverLoopCtx)
304 305 306 307
}

func (s *Server) startStatsChannel(ctx context.Context) {
	defer s.serverLoopWg.Done()
Z
zhenshan.cao 已提交
308 309 310
	factory := pulsarms.NewFactory(Params.PulsarAddress, 1024, 1024)
	statsStream, _ := factory.NewMsgStream(ctx)
	statsStream.AsConsumer([]string{Params.StatisticsChannelName}, Params.DataServiceSubscriptionName)
311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333
	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()
Z
zhenshan.cao 已提交
334 335 336
	factory := pulsarms.NewFactory(Params.PulsarAddress, 1024, 1024)
	flushStream, _ := factory.NewMsgStream(ctx)
	flushStream.AsConsumer([]string{Params.SegmentInfoChannelName}, Params.DataServiceSubscriptionName)
337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366
	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
367 368
func (s *Server) startDDChannel(ctx context.Context) {
	defer s.serverLoopWg.Done()
Z
zhenshan.cao 已提交
369 370 371
	factory := pulsarms.NewFactory(Params.PulsarAddress, 1024, 1024)
	ddStream, _ := factory.NewMsgStream(ctx)
	ddStream.AsConsumer([]string{s.ddChannelName}, Params.DataServiceSubscriptionName)
N
neza2017 已提交
372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390
	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
			}
		}
	}
}

391 392 393 394 395
func (s *Server) waitDataNodeRegister() {
	log.Println("waiting data node to register")
	<-s.registerFinishCh
	log.Println("all data nodes register")
}
S
sunby 已提交
396 397

func (s *Server) Stop() error {
S
sunby 已提交
398
	s.cluster.ShutDownClients()
S
sunby 已提交
399
	s.ttMsgStream.Close()
400
	s.k2sMsgStream.Close()
S
sunby 已提交
401
	s.msgProducer.Close()
N
neza2017 已提交
402
	s.segmentInfoStream.Close()
S
sunby 已提交
403
	s.stopServerLoop()
S
sunby 已提交
404 405 406
	return nil
}

S
sunby 已提交
407 408 409 410 411
func (s *Server) stopServerLoop() {
	s.serverLoopCancel()
	s.serverLoopWg.Wait()
}

S
sunby 已提交
412
func (s *Server) GetComponentStates() (*internalpb2.ComponentStates, error) {
S
sunby 已提交
413 414 415 416
	resp := &internalpb2.ComponentStates{
		State: &internalpb2.ComponentInfo{
			NodeID:    Params.NodeID,
			Role:      role,
S
sunby 已提交
417
			StateCode: s.state.Load().(internalpb2.StateCode),
S
sunby 已提交
418 419 420 421 422 423 424 425 426 427 428 429 430
		},
		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
431 432
}

N
neza2017 已提交
433 434
func (s *Server) GetTimeTickChannel() (string, error) {
	return Params.TimeTickChannelName, nil
S
sunby 已提交
435 436
}

N
neza2017 已提交
437 438
func (s *Server) GetStatisticsChannel() (string, error) {
	return Params.StatisticsChannelName, nil
S
sunby 已提交
439 440 441
}

func (s *Server) RegisterNode(req *datapb.RegisterNodeRequest) (*datapb.RegisterNodeResponse, error) {
S
sunby 已提交
442
	ret := &datapb.RegisterNodeResponse{
S
sunby 已提交
443
		Status: &commonpb.Status{
S
sunby 已提交
444
			ErrorCode: commonpb.ErrorCode_UNEXPECTED_ERROR,
S
sunby 已提交
445
		},
S
sunby 已提交
446
	}
S
sunby 已提交
447 448 449 450 451
	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
452
	if s.ddChannelName == "" {
N
neza2017 已提交
453
		resp, err := s.masterClient.GetDdChannel()
S
sunby 已提交
454 455 456 457
		if err != nil {
			ret.Status.Reason = err.Error()
			return ret, err
		}
N
neza2017 已提交
458
		s.ddChannelName = resp
S
sunby 已提交
459 460 461 462 463 464 465 466
	}
	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
467
			{Key: "CompleteFlushChannelName", Value: Params.SegmentInfoChannelName},
S
sunby 已提交
468 469 470
		},
	}
	return ret, nil
S
sunby 已提交
471 472
}

S
sunby 已提交
473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491
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
492
func (s *Server) Flush(req *datapb.FlushRequest) (*commonpb.Status, error) {
S
sunby 已提交
493 494 495 496 497 498
	if !s.checkStateIsHealthy() {
		return &commonpb.Status{
			ErrorCode: commonpb.ErrorCode_UNEXPECTED_ERROR,
			Reason:    "server is initializing",
		}, nil
	}
N
neza2017 已提交
499
	s.segAllocator.SealAllSegments(req.CollectionID)
S
sunby 已提交
500 501 502
	return &commonpb.Status{
		ErrorCode: commonpb.ErrorCode_SUCCESS,
	}, nil
S
sunby 已提交
503 504 505 506 507 508 509 510 511
}

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
512 513 514 515 516
	if !s.checkStateIsHealthy() {
		resp.Status.ErrorCode = commonpb.ErrorCode_UNEXPECTED_ERROR
		resp.Status.Reason = "server is initializing"
		return resp, nil
	}
S
sunby 已提交
517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562
	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 {
N
neza2017 已提交
563 564 565 566
	id, err := s.allocator.allocID()
	if err != nil {
		return err
	}
S
sunby 已提交
567
	segmentInfo, err := BuildSegment(collectionID, partitionID, id, channelName)
S
sunby 已提交
568 569 570 571 572 573
	if err != nil {
		return err
	}
	if err = s.meta.AddSegment(segmentInfo); err != nil {
		return err
	}
S
sunby 已提交
574
	if err = s.segAllocator.OpenSegment(segmentInfo); err != nil {
S
sunby 已提交
575 576
		return err
	}
577
	infoMsg := &msgstream.SegmentInfoMsg{
S
sunby 已提交
578 579 580
		BaseMsg: msgstream.BaseMsg{
			HashValues: []uint32{0},
		},
581 582 583 584
		SegmentMsg: datapb.SegmentMsg{
			Base: &commonpb.MsgBase{
				MsgType:   commonpb.MsgType_kSegmentInfo,
				MsgID:     0,
N
neza2017 已提交
585 586
				Timestamp: 0,
				SourceID:  Params.NodeID,
587 588 589 590 591 592 593 594 595 596
			},
			Segment: segmentInfo,
		},
	}
	msgPack := &pulsarms.MsgPack{
		Msgs: []msgstream.TsMsg{infoMsg},
	}
	if err = s.segmentInfoStream.Produce(msgPack); err != nil {
		return err
	}
S
sunby 已提交
597 598 599 600
	return nil
}

func (s *Server) ShowSegments(req *datapb.ShowSegmentRequest) (*datapb.ShowSegmentResponse, error) {
S
sunby 已提交
601 602 603 604 605
	resp := &datapb.ShowSegmentResponse{
		Status: &commonpb.Status{
			ErrorCode: commonpb.ErrorCode_UNEXPECTED_ERROR,
		},
	}
S
sunby 已提交
606
	if !s.checkStateIsHealthy() {
S
sunby 已提交
607 608
		resp.Status.Reason = "server is initializing"
		return resp, nil
S
sunby 已提交
609
	}
610
	ids := s.meta.GetSegmentsOfPartition(req.CollectionID, req.PartitionID)
S
sunby 已提交
611 612 613
	resp.Status.ErrorCode = commonpb.ErrorCode_SUCCESS
	resp.SegmentIDs = ids
	return resp, nil
S
sunby 已提交
614 615 616 617 618 619 620 621
}

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
622 623 624 625
	if !s.checkStateIsHealthy() {
		resp.Status.Reason = "server is initializing"
		return resp, nil
	}
S
sunby 已提交
626

Z
zhenshan.cao 已提交
627 628 629 630 631 632 633 634 635 636 637 638 639 640 641
	for _, segmentID := range req.SegmentIDs {
		state := &datapb.SegmentStateInfo{
			Status:    &commonpb.Status{},
			SegmentID: segmentID,
		}
		segmentInfo, err := s.meta.GetSegment(segmentID)
		if err != nil {
			state.Status.ErrorCode = commonpb.ErrorCode_UNEXPECTED_ERROR
			state.Status.Reason = "get segment states error: " + err.Error()
		} else {
			state.Status.ErrorCode = commonpb.ErrorCode_SUCCESS
			state.State = segmentInfo.State
			state.CreateTime = segmentInfo.OpenTime
			state.SealedTime = segmentInfo.SealedTime
			state.FlushedTime = segmentInfo.FlushedTime
X
XuanYang-cn 已提交
642 643
			state.StartPosition = segmentInfo.StartPosition
			state.EndPosition = segmentInfo.EndPosition
Z
zhenshan.cao 已提交
644 645
		}
		resp.States = append(resp.States, state)
S
sunby 已提交
646
	}
S
sunby 已提交
647
	resp.Status.ErrorCode = commonpb.ErrorCode_SUCCESS
Z
zhenshan.cao 已提交
648

S
sunby 已提交
649 650 651 652
	return resp, nil
}

func (s *Server) GetInsertBinlogPaths(req *datapb.InsertBinlogPathRequest) (*datapb.InsertBinlogPathsResponse, error) {
S
sunby 已提交
653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671
	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))
X
Xiangyu Wang 已提交
672 673 674
	for i, field := range flushMeta.Fields {
		fields[i] = field.FieldID
		paths[i] = &internalpb2.StringList{Values: field.BinlogPaths}
S
sunby 已提交
675
	}
S
sunby 已提交
676
	resp.Status.ErrorCode = commonpb.ErrorCode_SUCCESS
S
sunby 已提交
677 678 679
	resp.FieldIDs = fields
	resp.Paths = paths
	return resp, nil
S
sunby 已提交
680 681
}

N
neza2017 已提交
682
func (s *Server) GetInsertChannels(req *datapb.InsertChannelRequest) ([]string, error) {
683
	return s.insertChannels, nil
S
sunby 已提交
684 685 686
}

func (s *Server) GetCollectionStatistics(req *datapb.CollectionStatsRequest) (*datapb.CollectionStatsResponse, error) {
687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705
	resp := &datapb.CollectionStatsResponse{
		Status: &commonpb.Status{
			ErrorCode: commonpb.ErrorCode_UNEXPECTED_ERROR,
		},
	}
	nums, err := s.meta.GetNumRowsOfCollection(req.CollectionID)
	if err != nil {
		resp.Status.Reason = err.Error()
		return resp, nil
	}
	memsize, err := s.meta.GetMemSizeOfCollection(req.CollectionID)
	if err != nil {
		resp.Status.Reason = err.Error()
		return resp, nil
	}
	resp.Status.ErrorCode = commonpb.ErrorCode_SUCCESS
	resp.Stats = append(resp.Stats, &commonpb.KeyValuePair{Key: "nums", Value: strconv.FormatInt(nums, 10)})
	resp.Stats = append(resp.Stats, &commonpb.KeyValuePair{Key: "memsize", Value: strconv.FormatInt(memsize, 10)})
	return resp, nil
S
sunby 已提交
706 707 708 709 710
}

func (s *Server) GetPartitionStatistics(req *datapb.PartitionStatsRequest) (*datapb.PartitionStatsResponse, error) {
	// todo implement
	return nil, nil
S
sunby 已提交
711
}
N
neza2017 已提交
712 713 714 715

func (s *Server) GetSegmentInfoChannel() (string, error) {
	return Params.SegmentInfoChannelName, nil
}
716 717 718 719 720 721 722

func (s *Server) GetCount(req *datapb.CollectionCountRequest) (*datapb.CollectionCountResponse, error) {
	resp := &datapb.CollectionCountResponse{
		Status: &commonpb.Status{
			ErrorCode: commonpb.ErrorCode_UNEXPECTED_ERROR,
		},
	}
X
XuanYang-cn 已提交
723 724 725 726
	if !s.checkStateIsHealthy() {
		resp.Status.Reason = "data service is not healthy"
		return resp, nil
	}
727 728 729 730 731 732 733 734 735
	nums, err := s.meta.GetNumRowsOfCollection(req.CollectionID)
	if err != nil {
		resp.Status.Reason = err.Error()
		return resp, nil
	}
	resp.Count = nums
	resp.Status.ErrorCode = commonpb.ErrorCode_SUCCESS
	return resp, nil
}
X
XuanYang-cn 已提交
736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759

func (s *Server) GetSegmentInfo(req *datapb.SegmentInfoRequest) (*datapb.SegmentInfoResponse, error) {
	resp := &datapb.SegmentInfoResponse{
		Status: &commonpb.Status{
			ErrorCode: commonpb.ErrorCode_UNEXPECTED_ERROR,
		},
	}
	if !s.checkStateIsHealthy() {
		resp.Status.Reason = "data service is not healthy"
		return resp, nil
	}
	infos := make([]*datapb.SegmentInfo, len(req.SegmentIDs))
	for i, id := range req.SegmentIDs {
		segmentInfo, err := s.meta.GetSegment(id)
		if err != nil {
			resp.Status.Reason = err.Error()
			return resp, nil
		}
		infos[i] = segmentInfo
	}
	resp.Status.ErrorCode = commonpb.ErrorCode_SUCCESS
	resp.Infos = infos
	return resp, nil
}