/
state.go
340 lines (293 loc) · 9.75 KB
/
state.go
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
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
package terminal
import (
"context"
"encoding/binary"
"encoding/json"
"os"
"runtime/debug"
"sort"
"time"
"github.com/nats-io/nats-server/v2/server"
"github.com/nats-io/nats.go"
)
type EventDataMarker uint8
const (
_ EventDataMarker = iota
SnapshotEvent
JobEvent
)
const snapshotSize = 8
type StateMatchine struct {
LastJobID uint64
LastMsgSeq uint64
LastEventSeq uint64
LastUpdate time.Time
Jobs map[uint64]Job
}
func (t *Terminal) monitorJobStatus(ctx context.Context, s *server.Server, state **StateMatchine) {
t.logger.Noticef("node %s start checking job status", s.Name())
ticker := time.NewTicker(10 * time.Second)
defer func() {
t.logger.Noticef("node %s stop checking job status", s.Name())
ticker.Stop()
}()
check := func(state **StateMatchine) {
defer func() {
if r := recover(); r != nil {
t.logger.Errorf("node %s checking job status contains err %v", s.Name(), r)
debug.PrintStack()
}
}()
var consumers []*nats.ConsumerInfo
for c := range t.jsc.Consumers(t.dataStream, nats.Context(ctx)) {
consumers = append(consumers, c)
}
t.logger.Noticef("node %s reported %d consumers on data stream", s.Name(), len(consumers))
// make a copy of state
t.lock.Lock()
lastState := **state
t.lock.Unlock()
// note: data consumers should have at least one consumer at any time
if len(consumers) == 0 {
t.logger.Noticef("node %s is elected leader in a cluster without consumers on data stream", s.Name())
var (
jobQueueInfo *nats.StreamInfo
err error
)
for jobQueueInfo == nil {
if jobQueueInfo, err = t.jsc.StreamInfo(t.jobStream, nats.Context(ctx)); err != nil {
t.logger.Warnf("node %s will retry to get job stream info, err: %v", s.Name(), err)
time.Sleep(1 * time.Second)
}
}
if jobQueueInfo.State.Msgs == 0 && time.Since(lastState.LastUpdate) > 10*time.Second {
t.logger.Noticef("node %s will bootstrap job queue, last state updated at %v", s.Name(), lastState.LastUpdate)
if err = t.rotate(s, lastState.LastJobID, lastState.LastMsgSeq); err != nil {
t.logger.Noticef("node %s restart job queue %v failed: %v", s.Name(), lastState, err)
} else {
t.logger.Noticef("node %s restarted job queue %v", s.Name(), lastState)
}
}
} else {
t.logger.Noticef("node %s checked job queue is running", s.Name())
}
jobIDs := make([]uint64, 0, len(lastState.Jobs))
for k := range lastState.Jobs {
jobIDs = append(jobIDs, k)
}
sort.Slice(jobIDs, func(i, j int) bool { return jobIDs[i] < jobIDs[j] })
t.logger.Noticef("node %s is checking jobs %v", s.Name(), jobIDs)
for _, jid := range jobIDs {
job := lastState.Jobs[jid]
if job.Status == Idle || job.Status == Active || job.Status == Build {
t.logger.Noticef("node %s stops at idle/active/build job %v", s.Name(), job)
// TODO: check timestamp and make sure the consumer is still running
}
}
}
check(state)
for {
select {
case <-ticker.C:
check(state)
case <-ctx.Done():
return
}
}
}
func (t *Terminal) monitorStateMachine(ctx context.Context, s *server.Server) (**StateMatchine, error) {
var (
state *StateMatchine
eventStreamInfo *nats.StreamInfo
err error
processData func(inner **StateMatchine, data []byte, seq uint64, cnt int8) (int8, bool)
prevMsgs = make([]*nats.Msg, snapshotSize*2)
)
for err = os.ErrInvalid; err != nil; {
if eventStreamInfo, err = t.jsc.StreamInfo(t.eventStream, nats.Context(ctx)); err == nil {
break
}
t.logger.Errorf("node %s will retry to get event stream info: %v", s.Name(), err)
time.Sleep(time.Second)
}
processData = func(inner **StateMatchine, data []byte, seq uint64, cnt int8) (int8, bool) {
canSnap := false
switch EventDataMarker(data[0]) {
case SnapshotEvent:
var snapshot StateMatchine
if err = json.Unmarshal(data[1:], &snapshot); err != nil {
t.logger.Errorf("node %s unmarshall state machine from event stream err: %v", s.Name(), err)
return cnt, false
}
t.logger.Noticef("node %s %dth event message is state snapshot, catchup last %d: %v", s.Name(), seq, seq-snapshot.LastEventSeq, snapshot)
// no need to be atomic
*inner = &snapshot
for m := snapshot.LastEventSeq; m < seq; m++ {
idx := m % uint64(len(prevMsgs))
if prevMsgs[idx] != nil {
processData(inner, prevMsgs[idx].Data, m, -1)
}
}
case JobEvent:
canSnap = true
var job Job
if err = json.Unmarshal(data[1:], &job); err != nil {
t.logger.Errorf("node %s unmarshall job data first byte %d error: %v", s.Name(), data[0], err)
return cnt, false
}
// ignore if state is nil, the system will start tracking from a snapshot state
if *inner != nil {
if cnt >= 0 {
t.lock.Lock()
defer t.lock.Unlock()
}
if job.Status == Done {
delete((*inner).Jobs, job.Id)
} else {
(*inner).Jobs[job.Id] = job
if job.Status == Idle {
(*inner).LastJobID = job.Id
(*inner).LastMsgSeq = job.StartSeq
// update prev job status
prevID := job.Id - 1
if prevJob, ok := (*inner).Jobs[prevID]; ok {
t.logger.Noticef("node %s Idle Job %d, rotate prev job %d from %v to Build, end seq at %d",
s.Name(), job.Id, prevID, prevJob.Status, job.StartSeq)
if prevJob.Status == Active && prevJob.EndSeq == 0 {
prevJob.Status = Build
prevJob.EndSeq = job.StartSeq
}
}
}
}
// snapshot state after every 6 event messages
cnt = (cnt + 1) % snapshotSize
} else {
t.logger.Errorf("node %s replays event data at startup, skips %d event message, no snapshot state yet", s.Name(), seq)
}
default:
t.logger.Errorf("node %s unknown event data marker byte %d error: %v", s.Name(), data[0], err)
}
return cnt, canSnap
}
handle := func(inner **StateMatchine, msg *nats.Msg, seq uint64, cnt int8) (int8, []byte) {
msg.Term()
meta, err := msg.Metadata()
if err != nil {
t.logger.Errorf("node %s skips %dth event message due to metadata() err: %v", s.Name(), seq, err)
return cnt, nil
} else if meta.Sequence.Stream != seq {
t.logger.Errorf("node %s skips %dth event message due to expected seq %d mismatch", s.Name(), seq, meta.Sequence.Stream)
return cnt, nil
}
data := msg.Data
if len(data) == 0 {
t.logger.Warnf("node %s skips %dth event message due to msg data is empty", s.Name(), seq)
return cnt, nil
}
t.logger.Noticef("node %s processes %dth event msg with type %d", s.Name(), seq, data[0])
canSnap := false
cnt, canSnap = processData(inner, data, seq, cnt)
if canSnap {
prevMsgs[seq%uint64(len(prevMsgs))] = msg
if cnt == 0 {
(*inner).LastEventSeq = seq
t.logger.Noticef("node %s snapshot state %v", s.Name(), *inner)
if data, err = json.Marshal(*inner); err == nil {
return cnt, data
} else {
t.logger.Errorf("node %s marshall state machine err: %v", s.Name(), err)
}
}
} else {
prevMsgs[seq%uint64(len(prevMsgs))] = nil
}
return cnt, nil
}
var (
startSeq uint64
entry nats.KeyValueEntry
)
for err = os.ErrInvalid; err != nil; {
if entry, err = t.kvs.Get(stateSnapshot); err == nil {
startSeq = binary.LittleEndian.Uint64(entry.Value())
break
}
time.Sleep(time.Second)
}
// get state
pullOpts := []nats.SubOpt{
nats.StartSequence(startSeq),
nats.BindStream(t.eventStream),
nats.Context(ctx),
}
var sub *nats.Subscription
for err = os.ErrInvalid; err != nil; {
// ephemeral event stream subscriber
sub, err = t.jsc.SubscribeSync(t.eventStream, pullOpts...)
if err == nil {
break
}
t.logger.Errorf("node %s create event stream ephemral subscriber err: %v", s.Name(), err)
time.Sleep(time.Second)
}
t.logger.Noticef("node %s pulls msg [%d, %d] from event stream", s.Name(), startSeq, eventStreamInfo.State.LastSeq)
for seq := startSeq; seq <= eventStreamInfo.State.LastSeq; seq++ {
msg, err := sub.NextMsgWithContext(ctx)
if err != nil {
if err == ctx.Err() {
t.logger.Errorf("node %s exits init event loop with err: %v", s.Name(), err)
return nil, err
} else if err != context.DeadlineExceeded {
// note: nats uses internal TTL
t.logger.Warnf("node %s init event loop fetch data err: %v", s.Name(), err)
}
seq--
continue
}
handle(&state, msg, seq, -1)
}
go func(lastSeq uint64) {
defer func() {
if r := recover(); r != nil {
t.logger.Errorf("error %v", r)
debug.PrintStack()
}
t.logger.Warnf("node %s event loop exits", s.Name())
sub.Unsubscribe()
}()
var (
rollingCount = int8(1)
data []byte
)
for lastSeq++; true; lastSeq++ {
msg, err := sub.NextMsgWithContext(ctx)
if err != nil {
if err == ctx.Err() {
t.logger.Errorf("node %s exits event loop err: %v", s.Name(), err)
break
} else if err != context.DeadlineExceeded {
// note: nats uses internal TTL
t.logger.Warnf("node %s event loop fetch data err: %v", s.Name(), err)
}
lastSeq--
continue
}
if rollingCount, data = handle(&state, msg, lastSeq, rollingCount); len(data) > 0 {
// post snapshot state machine
if _, err := t.jsc.Publish(t.eventStream, append([]byte{byte(SnapshotEvent)}, data...), nats.Context(ctx)); err != nil {
t.logger.Errorf("node %s publishs snapshot state machine err: %v", s.Name(), err)
} else {
t.logger.Noticef("node %s saves snapshot state machine last event seq = %v to kv store", s.Name(), state.LastEventSeq)
data = make([]byte, 8)
binary.LittleEndian.PutUint64(data, state.LastEventSeq)
if _, err = t.kvs.Put(stateSnapshot, data); err != nil {
t.logger.Errorf("node %s put snapshot state machine kv ack %d err: %v", s.Name(), state.LastEventSeq, err)
}
}
}
}
}(eventStreamInfo.State.LastSeq)
t.logger.Noticef("node %s starts up snapshot state is %v", s.Name(), *state)
return &state, nil
}