-
Notifications
You must be signed in to change notification settings - Fork 776
/
processing_queue.go
362 lines (310 loc) · 11.4 KB
/
processing_queue.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
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
// Copyright (c) 2020 Uber Technologies, Inc.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
//
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
package queue
import (
"errors"
"fmt"
"sort"
"github.com/uber/cadence/common/log"
"github.com/uber/cadence/common/log/tag"
"github.com/uber/cadence/common/metrics"
t "github.com/uber/cadence/common/task"
"github.com/uber/cadence/service/history/task"
)
var (
errTaskNotFound = errors.New("task not found")
)
type (
processingQueueImpl struct {
state *processingQueueStateImpl
outstandingTasks map[task.Key]task.Task
logger log.Logger
metricsClient metrics.Client // TODO: emit metrics
}
)
// NewProcessingQueue creates a new processing queue based on its state
func NewProcessingQueue(
state ProcessingQueueState,
logger log.Logger,
metricsClient metrics.Client,
) ProcessingQueue {
return newProcessingQueue(
state,
nil,
logger,
metricsClient,
)
}
func newProcessingQueue(
state ProcessingQueueState,
outstandingTasks map[task.Key]task.Task,
logger log.Logger,
metricsClient metrics.Client,
) *processingQueueImpl {
if outstandingTasks == nil {
outstandingTasks = make(map[task.Key]task.Task)
}
queue := &processingQueueImpl{
outstandingTasks: outstandingTasks,
logger: logger,
metricsClient: metricsClient,
}
// convert state to *processingQueueStateImpl type so that
// queue implementation can change the state value
if stateImpl, ok := state.(*processingQueueStateImpl); ok {
queue.state = stateImpl
} else {
queue.state = copyQueueState(state)
}
if queue.state.readLevel.Less(queue.state.ackLevel) {
logger.Fatal("ack level larger than readlevel when creating processing queue", tag.Error(
fmt.Errorf("ack level: %v, read level: %v", queue.state.ackLevel, queue.state.readLevel),
))
}
return queue
}
func (q *processingQueueImpl) State() ProcessingQueueState {
return q.state
}
func (q *processingQueueImpl) Split(policy ProcessingQueueSplitPolicy) []ProcessingQueue {
newQueueStates := policy.Evaluate(q)
if len(newQueueStates) == 0 {
// no need to split, return self
return []ProcessingQueue{q}
}
return splitProcessingQueue([]*processingQueueImpl{q}, newQueueStates, q.logger, q.metricsClient)
}
func (q *processingQueueImpl) Merge(queue ProcessingQueue) []ProcessingQueue {
q1, q2 := q, queue.(*processingQueueImpl)
if q1.State().Level() != q2.State().Level() {
errMsg := "Processing queue encountered a queue from different level during merge"
q.logger.Error(errMsg, tag.Error(
fmt.Errorf("current queue level: %v, incoming queue level: %v", q1.state.level, q2.state.level),
))
panic(errMsg)
}
if !q1.state.ackLevel.Less(q2.state.maxLevel) ||
!q2.state.ackLevel.Less(q1.state.maxLevel) {
// one queue's ackLevel is larger or equal than the other one's maxLevel
// this means there's no overlap between two queues
return []ProcessingQueue{q1, q2}
}
// generate new queue states for merged queues
newQueueStates := []ProcessingQueueState{}
if !taskKeyEquals(q1.state.ackLevel, q2.state.ackLevel) {
if q2.state.ackLevel.Less(q1.state.ackLevel) {
q1, q2 = q2, q1
}
newQueueStates = append(newQueueStates, newProcessingQueueState(
q1.state.level,
q1.state.ackLevel,
minTaskKey(q1.state.readLevel, q2.state.ackLevel),
q2.state.ackLevel,
q1.state.domainFilter.copy(),
))
}
if !taskKeyEquals(q1.state.maxLevel, q2.state.maxLevel) {
if q1.state.maxLevel.Less(q2.state.maxLevel) {
q1, q2 = q2, q1
}
newQueueStates = append(newQueueStates, newProcessingQueueState(
q1.state.level,
q2.state.maxLevel,
maxTaskKey(q1.state.readLevel, q2.state.maxLevel),
q1.state.maxLevel,
q1.state.domainFilter.copy(),
))
}
overlappingQueueAckLevel := maxTaskKey(q1.state.ackLevel, q2.state.ackLevel)
newQueueStates = append(newQueueStates, newProcessingQueueState(
q1.state.level,
overlappingQueueAckLevel,
maxTaskKey(minTaskKey(q1.state.readLevel, q2.state.readLevel), overlappingQueueAckLevel),
minTaskKey(q1.state.maxLevel, q2.state.maxLevel),
q1.state.domainFilter.Merge(q2.state.domainFilter),
))
for _, state := range newQueueStates {
if state.ReadLevel().Less(state.AckLevel()) || state.MaxLevel().Less(state.ReadLevel()) {
q.logger.Fatal("invalid processing queue merge result", tag.Error(
fmt.Errorf("q1: %v, q2: %v, merge result: %v", q1.state, q2.state, newQueueStates),
))
}
}
return splitProcessingQueue([]*processingQueueImpl{q1, q2}, newQueueStates, q.logger, q.metricsClient)
}
func (q *processingQueueImpl) AddTasks(tasks map[task.Key]task.Task, newReadLevel task.Key) {
if newReadLevel.Less(q.state.readLevel) {
q.logger.Fatal("processing queue read level moved backward", tag.Error(
fmt.Errorf("current read level: %v, new read level: %v", q.state.readLevel, newReadLevel),
))
}
for key, task := range tasks {
if _, loaded := q.outstandingTasks[key]; loaded {
// TODO: this means the task has been submitted before, we should mark the task state accordingly and
// do not submit this task again in transfer/timer queue processor base
q.logger.Debug(fmt.Sprintf("Skipping task: %+v. DomainID: %v, WorkflowID: %v, RunID: %v, Type: %v",
key, task.GetDomainID(), task.GetWorkflowID(), task.GetRunID(), task.GetTaskType()))
continue
}
if !taskBelongsToProcessQueue(q.state, key, task) {
errMsg := "Processing queue encountered a task doesn't belong to its scope"
q.logger.Error(errMsg, tag.Error(
fmt.Errorf("processing queue state: %+v, task: %+v", q.state, key),
))
panic(errMsg)
}
q.outstandingTasks[key] = task
}
q.state.readLevel = newReadLevel
}
func (q *processingQueueImpl) GetTask(key task.Key) (task.Task, error) {
if task, ok := q.outstandingTasks[key]; ok {
return task, nil
}
return nil, errTaskNotFound
}
func (q *processingQueueImpl) GetTasks() []task.Task {
var outstandingTask []task.Task
for _, task := range q.outstandingTasks {
outstandingTask = append(outstandingTask, task)
}
return outstandingTask
}
func (q *processingQueueImpl) UpdateAckLevel() (task.Key, int) {
keys := make([]task.Key, 0, len(q.outstandingTasks))
for key := range q.outstandingTasks {
keys = append(keys, key)
}
sort.Slice(keys, func(i, j int) bool {
return keys[i].Less(keys[j])
})
var idx int
var key task.Key
for idx, key = range keys {
if q.state.readLevel.Less(key) {
// this can happen as during merge read level can move backward.
// besides that, for timer task key, readLevel is expected to be less than task key
// as the taskID for read level is always 0. This means we can potentially buffer
// more timer tasks in memory. If this becomes a problem, we can change this logic.
break
}
if q.outstandingTasks[key].State() != t.TaskStateAcked {
break
}
q.state.ackLevel = key
delete(q.outstandingTasks, key)
}
// The following loop attempts to delete tasks beyond ack level but has already been acked.
// To ensure ack level can be advanced as quick as possible, for a series of acked tasks,
// we still keep the last one in memory so that when previous pending tasks are acked, ack level
// can be advanced without wait for other tasks.
// Also only delete tasks less than read level as the sequence beyond read level may change.
//
// As an example, say currently we have 9 tasks: 1 2 3 4 5 6 7 9 10. Ack level is 0, read level is 7,
// task 1 2 6 10 is pending and the rest have been acked.
// We can delete task 3 4 but not 5 as otherwise even if task 1 and 2 were acked later, ack level would
// at most be 2 until task 6 is acked, while ideally it should be 5.
// We also can't delete task 7, because it's possible that task 8 will be loaded later. If task 7 got deleted,
// ack level can only be advanced to 6 instead of 7.
for idx < len(keys)-1 && keys[idx].Less(q.state.readLevel) {
if q.outstandingTasks[keys[idx]].State() == t.TaskStateAcked && q.outstandingTasks[keys[idx+1]].State() == t.TaskStateAcked {
delete(q.outstandingTasks, keys[idx])
}
idx++
}
if len(q.outstandingTasks) == 0 {
q.state.ackLevel = q.state.readLevel
}
if timerKey, ok := q.state.ackLevel.(timerTaskKey); ok {
q.state.ackLevel = newTimerTaskKey(timerKey.visibilityTimestamp, 0)
}
if q.state.readLevel.Less(q.state.ackLevel) {
q.logger.Fatal("ack level moved beyond read level", tag.Error(
fmt.Errorf("processing queue state: %v", q.state),
))
}
return q.state.ackLevel, len(q.outstandingTasks)
}
func splitProcessingQueue(
queues []*processingQueueImpl,
newQueueStates []ProcessingQueueState,
logger log.Logger,
metricsClient metrics.Client,
) []ProcessingQueue {
newQueueTasks := make([]map[task.Key]task.Task, 0, len(newQueueStates))
for i := 0; i != len(newQueueStates); i++ {
newQueueTasks = append(newQueueTasks, make(map[task.Key]task.Task))
}
for _, queue := range queues {
SplitTaskLoop:
for key, task := range queue.outstandingTasks {
for i, state := range newQueueStates {
if taskBelongsToProcessQueue(state, key, task) {
newQueueTasks[i][key] = task
continue SplitTaskLoop
}
}
// if code reaches there it means the task doesn't belongs to any new queue.
// there's must be a bug in the code for generating the newQueueStates
// log error, skip the split and return current queues as result
currentQueues := make([]ProcessingQueue, 0, len(newQueueStates))
currentQueueStates := make([]ProcessingQueueState, 0, len(newQueueStates))
for _, q := range queues {
currentQueues = append(currentQueues, q)
currentQueueStates = append(currentQueueStates, queue.State())
}
logger.Error("Processing queue encountered an error during split or merge.", tag.Error(
fmt.Errorf("current queue state: %+v, new queue state: %+v", currentQueueStates, newQueueStates),
))
return currentQueues
}
}
newQueues := make([]ProcessingQueue, 0, len(newQueueStates))
for i, state := range newQueueStates {
queue := newProcessingQueue(
state,
newQueueTasks[i],
logger,
metricsClient,
)
newQueues = append(newQueues, queue)
}
return newQueues
}
func taskBelongsToProcessQueue(state ProcessingQueueState, key task.Key, task task.Task) bool {
return state.DomainFilter().Filter(task.GetDomainID()) &&
state.AckLevel().Less(key) &&
!state.MaxLevel().Less(key)
}
func taskKeyEquals(key1 task.Key, key2 task.Key) bool {
return !key1.Less(key2) && !key2.Less(key1)
}
func minTaskKey(key1 task.Key, key2 task.Key) task.Key {
if key1.Less(key2) {
return key1
}
return key2
}
func maxTaskKey(key1 task.Key, key2 task.Key) task.Key {
if key1.Less(key2) {
return key2
}
return key1
}