server.go 23.1 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)
137
	s.segAllocator, err = newSegmentAllocator(s.meta, s.allocator)
S
sunby 已提交
138 139 140
	if err != nil {
		return err
	}
N
neza2017 已提交
141
	s.ddHandler = newDDHandler(s.meta, s.segAllocator)
142 143
	s.initSegmentInfoChannel()
	if err = s.loadMetaFromMaster(); err != nil {
S
sunby 已提交
144 145
		return err
	}
S
sunby 已提交
146
	s.startServerLoop()
147
	s.waitDataNodeRegister()
148
	s.cluster.WatchInsertChannels(s.insertChannels)
149 150 151
	if err = s.initMsgProducer(); err != nil {
		return err
	}
S
sunby 已提交
152
	s.state.Store(internalpb2.StateCode_HEALTHY)
S
sunby 已提交
153 154 155 156
	log.Println("start success")
	return nil
}

S
sunby 已提交
157 158 159 160
func (s *Server) checkStateIsHealthy() bool {
	return s.state.Load().(internalpb2.StateCode) == internalpb2.StateCode_HEALTHY
}

S
sunby 已提交
161 162 163 164 165 166 167
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
168
	s.meta, err = newMeta(etcdKV)
S
sunby 已提交
169 170 171 172 173 174
	if err != nil {
		return err
	}
	return nil
}

175
func (s *Server) initSegmentInfoChannel() {
176 177
	factory := msgstream.ProtoUDFactory{}
	segmentInfoStream := pulsarms.NewPulsarMsgStream(s.ctx, 1024, 1024, factory.NewUnmarshalDispatcher())
178 179 180 181
	segmentInfoStream.SetPulsarClient(Params.PulsarAddress)
	segmentInfoStream.CreatePulsarProducers([]string{Params.SegmentInfoChannelName})
	s.segmentInfoStream = segmentInfoStream
	s.segmentInfoStream.Start()
S
sunby 已提交
182
}
S
sunby 已提交
183
func (s *Server) initMsgProducer() error {
184 185
	factory := msgstream.ProtoUDFactory{}
	ttMsgStream := pulsarms.NewPulsarMsgStream(s.ctx, 1024, 1024, factory.NewUnmarshalDispatcher())
N
neza2017 已提交
186
	ttMsgStream.SetPulsarClient(Params.PulsarAddress)
187
	ttMsgStream.CreatePulsarConsumers([]string{Params.TimeTickChannelName}, Params.DataServiceSubscriptionName)
N
neza2017 已提交
188
	s.ttMsgStream = ttMsgStream
S
sunby 已提交
189 190 191
	s.ttMsgStream.Start()
	timeTickBarrier := timesync.NewHardTimeTickBarrier(s.ttMsgStream, s.cluster.GetNodeIDs())
	dataNodeTTWatcher := newDataNodeTimeTickWatcher(s.meta, s.segAllocator, s.cluster)
192
	k2sStream := pulsarms.NewPulsarMsgStream(s.ctx, 1024, 1024, factory.NewUnmarshalDispatcher())
193 194 195 196 197 198
	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
199 200 201 202 203 204 205
	if err != nil {
		return err
	}
	s.msgProducer = producer
	s.msgProducer.Start(s.ctx)
	return nil
}
S
sunby 已提交
206

S
sunby 已提交
207 208
func (s *Server) loadMetaFromMaster() error {
	log.Println("loading collection meta from master")
S
sunby 已提交
209 210 211
	if err := s.checkMasterIsHealthy(); err != nil {
		return err
	}
N
neza2017 已提交
212 213 214 215 216 217 218
	if s.ddChannelName == "" {
		channel, err := s.masterClient.GetDdChannel()
		if err != nil {
			return err
		}
		s.ddChannelName = channel
	}
S
sunby 已提交
219 220 221 222 223
	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
224
			SourceID:  Params.NodeID,
S
sunby 已提交
225 226 227 228 229 230 231 232 233 234 235 236
		},
		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
237
				SourceID:  Params.NodeID,
S
sunby 已提交
238 239 240 241 242 243 244 245 246 247 248 249 250
			},
			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
251
				SourceID:  Params.NodeID,
S
sunby 已提交
252 253 254 255 256 257 258 259 260 261 262 263
			},
			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
264
			Partitions: partitions.PartitionIDs,
S
sunby 已提交
265 266 267 268 269 270 271 272
		})
		if err != nil {
			log.Println(err.Error())
			continue
		}
	}
	log.Println("load collection meta from master complete")
	return nil
S
sunby 已提交
273
}
S
sunby 已提交
274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303

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
}

304 305
func (s *Server) startServerLoop() {
	s.serverLoopCtx, s.serverLoopCancel = context.WithCancel(s.ctx)
N
neza2017 已提交
306
	s.serverLoopWg.Add(3)
307 308
	go s.startStatsChannel(s.serverLoopCtx)
	go s.startSegmentFlushChannel(s.serverLoopCtx)
N
neza2017 已提交
309
	go s.startDDChannel(s.serverLoopCtx)
310 311 312 313
}

func (s *Server) startStatsChannel(ctx context.Context) {
	defer s.serverLoopWg.Done()
314 315
	factory := msgstream.ProtoUDFactory{}
	statsStream := pulsarms.NewPulsarMsgStream(ctx, 1024, 1024, factory.NewUnmarshalDispatcher())
316
	statsStream.SetPulsarClient(Params.PulsarAddress)
317
	statsStream.CreatePulsarConsumers([]string{Params.StatisticsChannelName}, Params.DataServiceSubscriptionName)
318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340
	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()
341 342
	factory := msgstream.ProtoUDFactory{}
	flushStream := pulsarms.NewPulsarMsgStream(ctx, 1024, 1024, factory.NewUnmarshalDispatcher())
343
	flushStream.SetPulsarClient(Params.PulsarAddress)
344
	flushStream.CreatePulsarConsumers([]string{Params.SegmentInfoChannelName}, Params.DataServiceSubscriptionName)
345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374
	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
375 376
func (s *Server) startDDChannel(ctx context.Context) {
	defer s.serverLoopWg.Done()
377 378
	factory := msgstream.ProtoUDFactory{}
	ddStream := pulsarms.NewPulsarMsgStream(ctx, 1024, 1024, factory.NewUnmarshalDispatcher())
N
neza2017 已提交
379
	ddStream.SetPulsarClient(Params.PulsarAddress)
380
	ddStream.CreatePulsarConsumers([]string{s.ddChannelName}, Params.DataServiceSubscriptionName)
N
neza2017 已提交
381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399
	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
			}
		}
	}
}

400 401 402 403 404
func (s *Server) waitDataNodeRegister() {
	log.Println("waiting data node to register")
	<-s.registerFinishCh
	log.Println("all data nodes register")
}
S
sunby 已提交
405 406

func (s *Server) Stop() error {
S
sunby 已提交
407
	s.cluster.ShutDownClients()
S
sunby 已提交
408
	s.ttMsgStream.Close()
409
	s.k2sMsgStream.Close()
S
sunby 已提交
410
	s.msgProducer.Close()
N
neza2017 已提交
411
	s.segmentInfoStream.Close()
S
sunby 已提交
412
	s.stopServerLoop()
S
sunby 已提交
413 414 415
	return nil
}

S
sunby 已提交
416 417 418 419 420
func (s *Server) stopServerLoop() {
	s.serverLoopCancel()
	s.serverLoopWg.Wait()
}

S
sunby 已提交
421
func (s *Server) GetComponentStates() (*internalpb2.ComponentStates, error) {
S
sunby 已提交
422 423 424 425
	resp := &internalpb2.ComponentStates{
		State: &internalpb2.ComponentInfo{
			NodeID:    Params.NodeID,
			Role:      role,
S
sunby 已提交
426
			StateCode: s.state.Load().(internalpb2.StateCode),
S
sunby 已提交
427 428 429 430 431 432 433 434 435 436 437 438 439
		},
		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
440 441
}

N
neza2017 已提交
442 443
func (s *Server) GetTimeTickChannel() (string, error) {
	return Params.TimeTickChannelName, nil
S
sunby 已提交
444 445
}

N
neza2017 已提交
446 447
func (s *Server) GetStatisticsChannel() (string, error) {
	return Params.StatisticsChannelName, nil
S
sunby 已提交
448 449 450
}

func (s *Server) RegisterNode(req *datapb.RegisterNodeRequest) (*datapb.RegisterNodeResponse, error) {
S
sunby 已提交
451
	ret := &datapb.RegisterNodeResponse{
S
sunby 已提交
452
		Status: &commonpb.Status{
S
sunby 已提交
453
			ErrorCode: commonpb.ErrorCode_UNEXPECTED_ERROR,
S
sunby 已提交
454
		},
S
sunby 已提交
455
	}
S
sunby 已提交
456 457 458 459 460
	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
461
	if s.ddChannelName == "" {
N
neza2017 已提交
462
		resp, err := s.masterClient.GetDdChannel()
S
sunby 已提交
463 464 465 466
		if err != nil {
			ret.Status.Reason = err.Error()
			return ret, err
		}
N
neza2017 已提交
467
		s.ddChannelName = resp
S
sunby 已提交
468 469 470 471 472 473 474 475
	}
	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
476
			{Key: "CompleteFlushChannelName", Value: Params.SegmentInfoChannelName},
S
sunby 已提交
477 478 479
		},
	}
	return ret, nil
S
sunby 已提交
480 481
}

S
sunby 已提交
482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500
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
501
func (s *Server) Flush(req *datapb.FlushRequest) (*commonpb.Status, error) {
S
sunby 已提交
502 503 504 505 506 507
	if !s.checkStateIsHealthy() {
		return &commonpb.Status{
			ErrorCode: commonpb.ErrorCode_UNEXPECTED_ERROR,
			Reason:    "server is initializing",
		}, nil
	}
N
neza2017 已提交
508
	s.segAllocator.SealAllSegments(req.CollectionID)
S
sunby 已提交
509 510 511
	return &commonpb.Status{
		ErrorCode: commonpb.ErrorCode_SUCCESS,
	}, nil
S
sunby 已提交
512 513 514 515 516 517 518 519 520
}

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
521 522 523 524 525
	if !s.checkStateIsHealthy() {
		resp.Status.ErrorCode = commonpb.ErrorCode_UNEXPECTED_ERROR
		resp.Status.Reason = "server is initializing"
		return resp, nil
	}
S
sunby 已提交
526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571
	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
572 573 574 575
	id, err := s.allocator.allocID()
	if err != nil {
		return err
	}
S
sunby 已提交
576
	segmentInfo, err := BuildSegment(collectionID, partitionID, id, channelName)
S
sunby 已提交
577 578 579 580 581 582
	if err != nil {
		return err
	}
	if err = s.meta.AddSegment(segmentInfo); err != nil {
		return err
	}
S
sunby 已提交
583
	if err = s.segAllocator.OpenSegment(segmentInfo); err != nil {
S
sunby 已提交
584 585
		return err
	}
586
	infoMsg := &msgstream.SegmentInfoMsg{
S
sunby 已提交
587 588 589
		BaseMsg: msgstream.BaseMsg{
			HashValues: []uint32{0},
		},
590 591 592 593
		SegmentMsg: datapb.SegmentMsg{
			Base: &commonpb.MsgBase{
				MsgType:   commonpb.MsgType_kSegmentInfo,
				MsgID:     0,
N
neza2017 已提交
594 595
				Timestamp: 0,
				SourceID:  Params.NodeID,
596 597 598 599 600 601 602 603 604 605
			},
			Segment: segmentInfo,
		},
	}
	msgPack := &pulsarms.MsgPack{
		Msgs: []msgstream.TsMsg{infoMsg},
	}
	if err = s.segmentInfoStream.Produce(msgPack); err != nil {
		return err
	}
S
sunby 已提交
606 607 608 609
	return nil
}

func (s *Server) ShowSegments(req *datapb.ShowSegmentRequest) (*datapb.ShowSegmentResponse, error) {
S
sunby 已提交
610 611 612 613 614
	resp := &datapb.ShowSegmentResponse{
		Status: &commonpb.Status{
			ErrorCode: commonpb.ErrorCode_UNEXPECTED_ERROR,
		},
	}
S
sunby 已提交
615
	if !s.checkStateIsHealthy() {
S
sunby 已提交
616 617
		resp.Status.Reason = "server is initializing"
		return resp, nil
S
sunby 已提交
618
	}
619
	ids := s.meta.GetSegmentsOfPartition(req.CollectionID, req.PartitionID)
S
sunby 已提交
620 621 622
	resp.Status.ErrorCode = commonpb.ErrorCode_SUCCESS
	resp.SegmentIDs = ids
	return resp, nil
S
sunby 已提交
623 624 625 626 627 628 629 630
}

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
631 632 633 634
	if !s.checkStateIsHealthy() {
		resp.Status.Reason = "server is initializing"
		return resp, nil
	}
S
sunby 已提交
635

Z
zhenshan.cao 已提交
636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654
	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
			state.StartPositions = segmentInfo.StartPosition
			state.EndPositions = segmentInfo.EndPosition
		}
		resp.States = append(resp.States, state)
S
sunby 已提交
655
	}
S
sunby 已提交
656
	resp.Status.ErrorCode = commonpb.ErrorCode_SUCCESS
Z
zhenshan.cao 已提交
657

S
sunby 已提交
658 659 660 661
	return resp, nil
}

func (s *Server) GetInsertBinlogPaths(req *datapb.InsertBinlogPathRequest) (*datapb.InsertBinlogPathsResponse, error) {
S
sunby 已提交
662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680
	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
681 682 683
	for i, field := range flushMeta.Fields {
		fields[i] = field.FieldID
		paths[i] = &internalpb2.StringList{Values: field.BinlogPaths}
S
sunby 已提交
684
	}
S
sunby 已提交
685
	resp.Status.ErrorCode = commonpb.ErrorCode_SUCCESS
S
sunby 已提交
686 687 688
	resp.FieldIDs = fields
	resp.Paths = paths
	return resp, nil
S
sunby 已提交
689 690
}

N
neza2017 已提交
691
func (s *Server) GetInsertChannels(req *datapb.InsertChannelRequest) ([]string, error) {
692
	return s.insertChannels, nil
S
sunby 已提交
693 694 695
}

func (s *Server) GetCollectionStatistics(req *datapb.CollectionStatsRequest) (*datapb.CollectionStatsResponse, error) {
696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714
	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
715 716 717 718 719
}

func (s *Server) GetPartitionStatistics(req *datapb.PartitionStatsRequest) (*datapb.PartitionStatsResponse, error) {
	// todo implement
	return nil, nil
S
sunby 已提交
720
}
N
neza2017 已提交
721 722 723 724

func (s *Server) GetSegmentInfoChannel() (string, error) {
	return Params.SegmentInfoChannelName, nil
}
725 726 727 728 729 730 731

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
732 733 734 735
	if !s.checkStateIsHealthy() {
		resp.Status.Reason = "data service is not healthy"
		return resp, nil
	}
736 737 738 739 740 741 742 743 744
	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
745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768

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
}