1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
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
292
293
294
295
296
297
298
299
300
301
302
303
304
305
|
package pub_client
import (
"context"
"fmt"
"sort"
"sync"
"sync/atomic"
"time"
"github.com/seaweedfs/seaweedfs/weed/pb"
"github.com/seaweedfs/seaweedfs/weed/pb/mq_pb"
"github.com/seaweedfs/seaweedfs/weed/util/buffered_queue"
"github.com/seaweedfs/seaweedfs/weed/util/log"
"google.golang.org/grpc"
"google.golang.org/grpc/codes"
"google.golang.org/grpc/credentials/insecure"
"google.golang.org/grpc/status"
)
type EachPartitionError struct {
*mq_pb.BrokerPartitionAssignment
Err error
generation int
}
type EachPartitionPublishJob struct {
*mq_pb.BrokerPartitionAssignment
stopChan chan bool
wg sync.WaitGroup
generation int
inputQueue *buffered_queue.BufferedQueue[*mq_pb.DataMessage]
}
func (p *TopicPublisher) startSchedulerThread(wg *sync.WaitGroup) error {
if err := p.doConfigureTopic(); err != nil {
wg.Done()
return fmt.Errorf("configure topic %s: %v", p.config.Topic, err)
}
log.Infof("start scheduler thread for topic %s", p.config.Topic)
generation := 0
var errChan chan EachPartitionError
for {
log.V(3).Infof("lookup partitions gen %d topic %s", generation+1, p.config.Topic)
if assignments, err := p.doLookupTopicPartitions(); err == nil {
generation++
log.V(3).Infof("start generation %d with %d assignments", generation, len(assignments))
if errChan == nil {
errChan = make(chan EachPartitionError, len(assignments))
}
p.onEachAssignments(generation, assignments, errChan)
} else {
log.Errorf("lookup topic %s: %v", p.config.Topic, err)
time.Sleep(5 * time.Second)
continue
}
if generation == 1 {
wg.Done()
}
// wait for any error to happen. If so, consume all remaining errors, and retry
for {
select {
case eachErr := <-errChan:
log.Errorf("gen %d publish to topic %s partition %v: %v", eachErr.generation, p.config.Topic, eachErr.Partition, eachErr.Err)
if eachErr.generation < generation {
continue
}
break
}
}
}
}
func (p *TopicPublisher) onEachAssignments(generation int, assignments []*mq_pb.BrokerPartitionAssignment, errChan chan EachPartitionError) {
// TODO assuming this is not re-configured so the partitions are fixed.
sort.Slice(assignments, func(i, j int) bool {
return assignments[i].Partition.RangeStart < assignments[j].Partition.RangeStart
})
var jobs []*EachPartitionPublishJob
hasExistingJob := len(p.jobs) == len(assignments)
for i, assignment := range assignments {
if assignment.LeaderBroker == "" {
continue
}
if hasExistingJob {
var existingJob *EachPartitionPublishJob
existingJob = p.jobs[i]
if existingJob.BrokerPartitionAssignment.LeaderBroker == assignment.LeaderBroker {
existingJob.generation = generation
jobs = append(jobs, existingJob)
continue
} else {
if existingJob.LeaderBroker != "" {
close(existingJob.stopChan)
existingJob.LeaderBroker = ""
existingJob.wg.Wait()
}
}
}
// start a go routine to publish to this partition
job := &EachPartitionPublishJob{
BrokerPartitionAssignment: assignment,
stopChan: make(chan bool, 1),
generation: generation,
inputQueue: buffered_queue.NewBufferedQueue[*mq_pb.DataMessage](1024),
}
job.wg.Add(1)
go func(job *EachPartitionPublishJob) {
defer job.wg.Done()
if err := p.doPublishToPartition(job); err != nil {
log.Infof("publish to %s partition %v: %v", p.config.Topic, job.Partition, err)
errChan <- EachPartitionError{assignment, err, generation}
}
}(job)
jobs = append(jobs, job)
// TODO assuming this is not re-configured so the partitions are fixed.
// better just re-use the existing job
p.partition2Buffer.Insert(assignment.Partition.RangeStart, assignment.Partition.RangeStop, job.inputQueue)
}
p.jobs = jobs
}
func (p *TopicPublisher) doPublishToPartition(job *EachPartitionPublishJob) error {
log.Infof("connecting to %v for topic partition %+v", job.LeaderBroker, job.Partition)
grpcConnection, err := grpc.NewClient(job.LeaderBroker, grpc.WithTransportCredentials(insecure.NewCredentials()), p.grpcDialOption)
if err != nil {
return fmt.Errorf("dial broker %s: %v", job.LeaderBroker, err)
}
brokerClient := mq_pb.NewSeaweedMessagingClient(grpcConnection)
stream, err := brokerClient.PublishMessage(context.Background())
if err != nil {
return fmt.Errorf("create publish client: %v", err)
}
publishClient := &PublishClient{
SeaweedMessaging_PublishMessageClient: stream,
Broker: job.LeaderBroker,
}
if err = publishClient.Send(&mq_pb.PublishMessageRequest{
Message: &mq_pb.PublishMessageRequest_Init{
Init: &mq_pb.PublishMessageRequest_InitMessage{
Topic: p.config.Topic.ToPbTopic(),
Partition: job.Partition,
AckInterval: 128,
FollowerBroker: job.FollowerBroker,
PublisherName: p.config.PublisherName,
},
},
}); err != nil {
return fmt.Errorf("send init message: %v", err)
}
// process the hello message
resp, err := stream.Recv()
if err != nil {
e, _ := status.FromError(err)
if e.Code() == codes.Unknown && e.Message() == "EOF" {
log.Infof("publish to %s EOF", publishClient.Broker)
return nil
}
publishClient.Err = err
log.Errorf("publish1 to %s error: %v", publishClient.Broker, err)
return err
}
if resp.Error != "" {
publishClient.Err = fmt.Errorf("ack error: %v", resp.Error)
log.Errorf("publish2 to %s error: %v", publishClient.Broker, resp.Error)
return fmt.Errorf("ack error: %v", resp.Error)
}
var publishedTsNs int64
hasMoreData := int32(1)
var wg sync.WaitGroup
wg.Add(1)
go func() {
defer wg.Done()
for {
ackResp, err := publishClient.Recv()
if err != nil {
e, _ := status.FromError(err)
if e.Code() == codes.Unknown && e.Message() == "EOF" {
log.Infof("publish to %s EOF", publishClient.Broker)
return
}
publishClient.Err = err
log.Errorf("publish1 to %s error: %v", publishClient.Broker, err)
return
}
if ackResp.Error != "" {
publishClient.Err = fmt.Errorf("ack error: %v", ackResp.Error)
log.Errorf("publish2 to %s error: %v", publishClient.Broker, ackResp.Error)
return
}
if ackResp.AckSequence > 0 {
log.Infof("ack %d published %d hasMoreData:%d", ackResp.AckSequence, atomic.LoadInt64(&publishedTsNs), atomic.LoadInt32(&hasMoreData))
}
if atomic.LoadInt64(&publishedTsNs) <= ackResp.AckSequence && atomic.LoadInt32(&hasMoreData) == 0 {
return
}
}
}()
publishCounter := 0
for data, hasData := job.inputQueue.Dequeue(); hasData; data, hasData = job.inputQueue.Dequeue() {
if data.Ctrl != nil && data.Ctrl.IsClose {
// need to set this before sending to brokers, to avoid timing issue
atomic.StoreInt32(&hasMoreData, 0)
}
if err := publishClient.Send(&mq_pb.PublishMessageRequest{
Message: &mq_pb.PublishMessageRequest_Data{
Data: data,
},
}); err != nil {
return fmt.Errorf("send publish data: %v", err)
}
publishCounter++
atomic.StoreInt64(&publishedTsNs, data.TsNs)
}
if publishCounter > 0 {
wg.Wait()
} else {
// CloseSend would cancel the context on the server side
if err := publishClient.CloseSend(); err != nil {
return fmt.Errorf("close send: %v", err)
}
}
log.Infof("published %d messages to %v for topic partition %+v", publishCounter, job.LeaderBroker, job.Partition)
return nil
}
func (p *TopicPublisher) doConfigureTopic() (err error) {
if len(p.config.Brokers) == 0 {
return fmt.Errorf("topic configuring found no bootstrap brokers")
}
var lastErr error
for _, brokerAddress := range p.config.Brokers {
err = pb.WithBrokerGrpcClient(false,
brokerAddress,
p.grpcDialOption,
func(client mq_pb.SeaweedMessagingClient) error {
_, err := client.ConfigureTopic(context.Background(), &mq_pb.ConfigureTopicRequest{
Topic: p.config.Topic.ToPbTopic(),
PartitionCount: p.config.PartitionCount,
RecordType: p.config.RecordType, // TODO schema upgrade
})
return err
})
if err == nil {
lastErr = nil
return nil
} else {
lastErr = fmt.Errorf("%s: %v", brokerAddress, err)
}
}
if lastErr != nil {
return fmt.Errorf("doConfigureTopic %s: %v", p.config.Topic, err)
}
return nil
}
func (p *TopicPublisher) doLookupTopicPartitions() (assignments []*mq_pb.BrokerPartitionAssignment, err error) {
if len(p.config.Brokers) == 0 {
return nil, fmt.Errorf("lookup found no bootstrap brokers")
}
var lastErr error
for _, brokerAddress := range p.config.Brokers {
err := pb.WithBrokerGrpcClient(false,
brokerAddress,
p.grpcDialOption,
func(client mq_pb.SeaweedMessagingClient) error {
lookupResp, err := client.LookupTopicBrokers(context.Background(),
&mq_pb.LookupTopicBrokersRequest{
Topic: p.config.Topic.ToPbTopic(),
})
log.V(3).Infof("lookup topic %s: %v", p.config.Topic, lookupResp)
if err != nil {
return err
}
if len(lookupResp.BrokerPartitionAssignments) == 0 {
return fmt.Errorf("no broker partition assignments")
}
assignments = lookupResp.BrokerPartitionAssignments
return nil
})
if err == nil {
return assignments, nil
} else {
lastErr = err
}
}
return nil, fmt.Errorf("lookup topic %s: %v", p.config.Topic, lastErr)
}
|