seaweedfs/weed/mq/topic/local_partition.go

214 lines
6.3 KiB
Go
Raw Normal View History

package topic
import (
2024-03-28 13:58:20 +08:00
"context"
2023-08-29 00:02:12 +08:00
"fmt"
2024-01-15 16:20:12 +08:00
"github.com/seaweedfs/seaweedfs/weed/glog"
"github.com/seaweedfs/seaweedfs/weed/pb"
"github.com/seaweedfs/seaweedfs/weed/pb/mq_pb"
"github.com/seaweedfs/seaweedfs/weed/util/log_buffer"
"google.golang.org/grpc"
2024-03-17 14:16:33 +08:00
"sync"
2024-03-11 05:34:28 +08:00
"sync/atomic"
"time"
)
type LocalPartition struct {
2024-03-17 14:16:33 +08:00
ListenersWaits int64
AckTsNs int64
2024-03-17 14:16:33 +08:00
// notifying clients
ListenersLock sync.Mutex
ListenersCond *sync.Cond
Partition
2024-03-17 01:51:47 +08:00
isLeader bool
FollowerBrokers []pb.ServerAddress
LogBuffer *log_buffer.LogBuffer
ConsumerCount int32
Publishers *LocalPartitionPublishers
Subscribers *LocalPartitionSubscribers
FollowerId int32
2024-03-27 12:52:12 +08:00
FollowerStream mq_pb.SeaweedMessaging_PublishFollowMeClient
FollowerGrpcConnection *grpc.ClientConn
}
var TIME_FORMAT = "2006-01-02-15-04-05"
2024-03-01 01:38:52 +08:00
2024-01-15 16:20:12 +08:00
func NewLocalPartition(partition Partition, isLeader bool, followerBrokers []pb.ServerAddress, logFlushFn log_buffer.LogFlushFuncType, readFromDiskFn log_buffer.LogReadFromDiskFuncType) *LocalPartition {
2024-03-17 14:16:33 +08:00
lp := &LocalPartition{
2023-08-29 00:02:12 +08:00
Partition: partition,
isLeader: isLeader,
FollowerBrokers: followerBrokers,
Merge accumulated changes related to message queue (#5098) * balance partitions on brokers * prepare topic partition first and then publish, move partition * purge unused APIs * clean up * adjust logs * add BalanceTopics() grpc API * configure topic * configure topic command * refactor * repair missing partitions * sequence of operations to ensure ordering * proto to close publishers and consumers * rename file * topic partition versioned by unixTimeNs * create local topic partition * close publishers * randomize the client name * wait until no publishers * logs * close stop publisher channel * send last ack * comments * comment * comments * support list of brokers * add cli options * Update .gitignore * logs * return io.eof directly * refactor * optionally create topic * refactoring * detect consumer disconnection * sub client wait for more messages * subscribe by time stamp * rename * rename to sub_balancer * rename * adjust comments * rename * fix compilation * rename * rename * SubscriberToSubCoordinator * sticky rebalance * go fmt * add tests * balance partitions on brokers * prepare topic partition first and then publish, move partition * purge unused APIs * clean up * adjust logs * add BalanceTopics() grpc API * configure topic * configure topic command * refactor * repair missing partitions * sequence of operations to ensure ordering * proto to close publishers and consumers * rename file * topic partition versioned by unixTimeNs * create local topic partition * close publishers * randomize the client name * wait until no publishers * logs * close stop publisher channel * send last ack * comments * comment * comments * support list of brokers * add cli options * Update .gitignore * logs * return io.eof directly * refactor * optionally create topic * refactoring * detect consumer disconnection * sub client wait for more messages * subscribe by time stamp * rename * rename to sub_balancer * rename * adjust comments * rename * fix compilation * rename * rename * SubscriberToSubCoordinator * sticky rebalance * go fmt * add tests * tracking topic=>broker * merge * comment
2023-12-12 04:05:54 +08:00
Publishers: NewLocalPartitionPublishers(),
Subscribers: NewLocalPartitionSubscribers(),
2023-08-29 00:02:12 +08:00
}
2024-03-17 14:16:33 +08:00
lp.ListenersCond = sync.NewCond(&lp.ListenersLock)
lp.LogBuffer = log_buffer.NewLogBuffer(fmt.Sprintf("%d/%04d-%04d", partition.UnixTimeNs, partition.RangeStart, partition.RangeStop),
2*time.Minute, logFlushFn, readFromDiskFn, func() {
if atomic.LoadInt64(&lp.ListenersWaits) > 0 {
lp.ListenersCond.Broadcast()
}
})
return lp
2023-08-29 00:02:12 +08:00
}
Merge accumulated changes related to message queue (#5098) * balance partitions on brokers * prepare topic partition first and then publish, move partition * purge unused APIs * clean up * adjust logs * add BalanceTopics() grpc API * configure topic * configure topic command * refactor * repair missing partitions * sequence of operations to ensure ordering * proto to close publishers and consumers * rename file * topic partition versioned by unixTimeNs * create local topic partition * close publishers * randomize the client name * wait until no publishers * logs * close stop publisher channel * send last ack * comments * comment * comments * support list of brokers * add cli options * Update .gitignore * logs * return io.eof directly * refactor * optionally create topic * refactoring * detect consumer disconnection * sub client wait for more messages * subscribe by time stamp * rename * rename to sub_balancer * rename * adjust comments * rename * fix compilation * rename * rename * SubscriberToSubCoordinator * sticky rebalance * go fmt * add tests * balance partitions on brokers * prepare topic partition first and then publish, move partition * purge unused APIs * clean up * adjust logs * add BalanceTopics() grpc API * configure topic * configure topic command * refactor * repair missing partitions * sequence of operations to ensure ordering * proto to close publishers and consumers * rename file * topic partition versioned by unixTimeNs * create local topic partition * close publishers * randomize the client name * wait until no publishers * logs * close stop publisher channel * send last ack * comments * comment * comments * support list of brokers * add cli options * Update .gitignore * logs * return io.eof directly * refactor * optionally create topic * refactoring * detect consumer disconnection * sub client wait for more messages * subscribe by time stamp * rename * rename to sub_balancer * rename * adjust comments * rename * fix compilation * rename * rename * SubscriberToSubCoordinator * sticky rebalance * go fmt * add tests * tracking topic=>broker * merge * comment
2023-12-12 04:05:54 +08:00
func (p *LocalPartition) Publish(message *mq_pb.DataMessage) {
p.LogBuffer.AddToBuffer(message.Key, message.Value, time.Now().UnixNano())
}
2024-01-15 16:20:12 +08:00
func (p *LocalPartition) Subscribe(clientName string, startPosition log_buffer.MessagePosition,
onNoMessageFn func() bool, eachMessageFn log_buffer.EachLogEntryFuncType) error {
var processedPosition log_buffer.MessagePosition
var readPersistedLogErr error
var readInMemoryLogErr error
var isDone bool
for {
processedPosition, isDone, readPersistedLogErr = p.LogBuffer.ReadFromDiskFn(startPosition, 0, eachMessageFn)
2024-01-15 16:20:12 +08:00
if readPersistedLogErr != nil {
glog.V(0).Infof("%s read %v persisted log: %v", clientName, p.Partition, readPersistedLogErr)
return readPersistedLogErr
}
if isDone {
return nil
}
startPosition = processedPosition
processedPosition, isDone, readInMemoryLogErr = p.LogBuffer.LoopProcessLogData(clientName, startPosition, 0, onNoMessageFn, eachMessageFn)
2024-03-11 05:34:28 +08:00
if isDone {
return nil
}
2024-01-15 16:20:12 +08:00
startPosition = processedPosition
if readInMemoryLogErr == log_buffer.ResumeFromDiskError {
continue
}
if readInMemoryLogErr != nil {
glog.V(0).Infof("%s read %v in memory log: %v", clientName, p.Partition, readInMemoryLogErr)
return readInMemoryLogErr
}
}
2024-01-08 16:03:08 +08:00
}
func (p *LocalPartition) GetEarliestMessageTimeInMemory() time.Time {
return p.LogBuffer.GetEarliestTime()
2024-01-08 16:03:08 +08:00
}
func (p *LocalPartition) HasData() bool {
return !p.LogBuffer.GetEarliestTime().IsZero()
2024-01-08 16:03:08 +08:00
}
func (p *LocalPartition) GetEarliestInMemoryMessagePosition() log_buffer.MessagePosition {
return p.LogBuffer.GetEarliestPosition()
2023-08-28 04:13:14 +08:00
}
2024-01-16 13:22:41 +08:00
func FromPbBrokerPartitionAssignment(self pb.ServerAddress, partition Partition, assignment *mq_pb.BrokerPartitionAssignment, logFlushFn log_buffer.LogFlushFuncType, readFromDiskFn log_buffer.LogReadFromDiskFuncType) *LocalPartition {
Merge accumulated changes related to message queue (#5098) * balance partitions on brokers * prepare topic partition first and then publish, move partition * purge unused APIs * clean up * adjust logs * add BalanceTopics() grpc API * configure topic * configure topic command * refactor * repair missing partitions * sequence of operations to ensure ordering * proto to close publishers and consumers * rename file * topic partition versioned by unixTimeNs * create local topic partition * close publishers * randomize the client name * wait until no publishers * logs * close stop publisher channel * send last ack * comments * comment * comments * support list of brokers * add cli options * Update .gitignore * logs * return io.eof directly * refactor * optionally create topic * refactoring * detect consumer disconnection * sub client wait for more messages * subscribe by time stamp * rename * rename to sub_balancer * rename * adjust comments * rename * fix compilation * rename * rename * SubscriberToSubCoordinator * sticky rebalance * go fmt * add tests * balance partitions on brokers * prepare topic partition first and then publish, move partition * purge unused APIs * clean up * adjust logs * add BalanceTopics() grpc API * configure topic * configure topic command * refactor * repair missing partitions * sequence of operations to ensure ordering * proto to close publishers and consumers * rename file * topic partition versioned by unixTimeNs * create local topic partition * close publishers * randomize the client name * wait until no publishers * logs * close stop publisher channel * send last ack * comments * comment * comments * support list of brokers * add cli options * Update .gitignore * logs * return io.eof directly * refactor * optionally create topic * refactoring * detect consumer disconnection * sub client wait for more messages * subscribe by time stamp * rename * rename to sub_balancer * rename * adjust comments * rename * fix compilation * rename * rename * SubscriberToSubCoordinator * sticky rebalance * go fmt * add tests * tracking topic=>broker * merge * comment
2023-12-12 04:05:54 +08:00
isLeader := assignment.LeaderBroker == string(self)
followers := make([]pb.ServerAddress, len(assignment.FollowerBrokers))
Merge accumulated changes related to message queue (#5098) * balance partitions on brokers * prepare topic partition first and then publish, move partition * purge unused APIs * clean up * adjust logs * add BalanceTopics() grpc API * configure topic * configure topic command * refactor * repair missing partitions * sequence of operations to ensure ordering * proto to close publishers and consumers * rename file * topic partition versioned by unixTimeNs * create local topic partition * close publishers * randomize the client name * wait until no publishers * logs * close stop publisher channel * send last ack * comments * comment * comments * support list of brokers * add cli options * Update .gitignore * logs * return io.eof directly * refactor * optionally create topic * refactoring * detect consumer disconnection * sub client wait for more messages * subscribe by time stamp * rename * rename to sub_balancer * rename * adjust comments * rename * fix compilation * rename * rename * SubscriberToSubCoordinator * sticky rebalance * go fmt * add tests * balance partitions on brokers * prepare topic partition first and then publish, move partition * purge unused APIs * clean up * adjust logs * add BalanceTopics() grpc API * configure topic * configure topic command * refactor * repair missing partitions * sequence of operations to ensure ordering * proto to close publishers and consumers * rename file * topic partition versioned by unixTimeNs * create local topic partition * close publishers * randomize the client name * wait until no publishers * logs * close stop publisher channel * send last ack * comments * comment * comments * support list of brokers * add cli options * Update .gitignore * logs * return io.eof directly * refactor * optionally create topic * refactoring * detect consumer disconnection * sub client wait for more messages * subscribe by time stamp * rename * rename to sub_balancer * rename * adjust comments * rename * fix compilation * rename * rename * SubscriberToSubCoordinator * sticky rebalance * go fmt * add tests * tracking topic=>broker * merge * comment
2023-12-12 04:05:54 +08:00
for i, followerBroker := range assignment.FollowerBrokers {
followers[i] = pb.ServerAddress(followerBroker)
}
2024-01-16 13:22:41 +08:00
return NewLocalPartition(partition, isLeader, followers, logFlushFn, readFromDiskFn)
Merge accumulated changes related to message queue (#5098) * balance partitions on brokers * prepare topic partition first and then publish, move partition * purge unused APIs * clean up * adjust logs * add BalanceTopics() grpc API * configure topic * configure topic command * refactor * repair missing partitions * sequence of operations to ensure ordering * proto to close publishers and consumers * rename file * topic partition versioned by unixTimeNs * create local topic partition * close publishers * randomize the client name * wait until no publishers * logs * close stop publisher channel * send last ack * comments * comment * comments * support list of brokers * add cli options * Update .gitignore * logs * return io.eof directly * refactor * optionally create topic * refactoring * detect consumer disconnection * sub client wait for more messages * subscribe by time stamp * rename * rename to sub_balancer * rename * adjust comments * rename * fix compilation * rename * rename * SubscriberToSubCoordinator * sticky rebalance * go fmt * add tests * balance partitions on brokers * prepare topic partition first and then publish, move partition * purge unused APIs * clean up * adjust logs * add BalanceTopics() grpc API * configure topic * configure topic command * refactor * repair missing partitions * sequence of operations to ensure ordering * proto to close publishers and consumers * rename file * topic partition versioned by unixTimeNs * create local topic partition * close publishers * randomize the client name * wait until no publishers * logs * close stop publisher channel * send last ack * comments * comment * comments * support list of brokers * add cli options * Update .gitignore * logs * return io.eof directly * refactor * optionally create topic * refactoring * detect consumer disconnection * sub client wait for more messages * subscribe by time stamp * rename * rename to sub_balancer * rename * adjust comments * rename * fix compilation * rename * rename * SubscriberToSubCoordinator * sticky rebalance * go fmt * add tests * tracking topic=>broker * merge * comment
2023-12-12 04:05:54 +08:00
}
func (p *LocalPartition) closePublishers() {
p.Publishers.SignalShutdown()
}
func (p *LocalPartition) closeSubscribers() {
p.Subscribers.SignalShutdown()
}
func (p *LocalPartition) WaitUntilNoPublishers() {
for {
if p.Publishers.IsEmpty() {
return
}
time.Sleep(113 * time.Millisecond)
}
}
2024-03-28 13:58:20 +08:00
func (p *LocalPartition) MaybeConnectToFollowers(initMessage *mq_pb.PublishMessageRequest_InitMessage, grpcDialOption grpc.DialOption) (err error) {
if p.FollowerStream != nil {
return nil
}
if len(initMessage.FollowerBrokers) == 0 {
return nil
}
follower := initMessage.FollowerBrokers[0]
ctx := context.Background()
p.FollowerGrpcConnection, err = pb.GrpcDial(ctx, follower, true, grpcDialOption)
if err != nil {
return fmt.Errorf("fail to dial %s: %v", follower, err)
}
followerClient := mq_pb.NewSeaweedMessagingClient(p.FollowerGrpcConnection)
p.FollowerStream, err = followerClient.PublishFollowMe(ctx)
if err != nil {
return fmt.Errorf("fail to create publish client: %v", err)
}
if err = p.FollowerStream.Send(&mq_pb.PublishFollowMeRequest{
Message: &mq_pb.PublishFollowMeRequest_Init{
Init: &mq_pb.PublishFollowMeRequest_InitMessage{
Topic: initMessage.Topic,
Partition: initMessage.Partition,
},
},
}); err != nil {
return err
}
// start receiving ack from follower
go func() {
defer func() {
println("stop receiving ack from follower")
}()
for {
ack, err := p.FollowerStream.Recv()
if err != nil {
glog.Errorf("Error receiving follower ack: %v", err)
return
}
atomic.StoreInt64(&p.AckTsNs, ack.AckTsNs)
println("recv ack", ack.AckTsNs)
}
}()
return nil
}
func (p *LocalPartition) MaybeShutdownLocalPartition() (hasShutdown bool) {
2024-03-28 13:48:37 +08:00
if p.MaybeShutdownLocalPartition() {
if p.FollowerStream != nil {
// send close to the follower
if followErr := p.FollowerStream.Send(&mq_pb.PublishFollowMeRequest{
Message: &mq_pb.PublishFollowMeRequest_Close{
Close: &mq_pb.PublishFollowMeRequest_CloseMessage{},
},
}); followErr != nil {
glog.Errorf("Error closing follower stream: %v", followErr)
}
println("closing grpcConnection to follower")
p.FollowerGrpcConnection.Close()
}
}
return
}
func (p *LocalPartition) canShutdownLocalPartition() (hasShutdown bool) {
if p.Publishers.IsEmpty() && p.Subscribers.IsEmpty() {
p.LogBuffer.ShutdownLogBuffer()
hasShutdown = true
}
return
}
2024-03-11 05:34:28 +08:00
func (p *LocalPartition) Shutdown() {
p.closePublishers()
p.closeSubscribers()
p.LogBuffer.ShutdownLogBuffer()
atomic.StoreInt32(&p.FollowerId, 0)
2024-03-18 13:24:20 +08:00
glog.V(0).Infof("local partition %v shutting down", p.Partition)
2024-03-11 05:34:28 +08:00
}