-
Notifications
You must be signed in to change notification settings - Fork 1.9k
/
alloc_runner.go
991 lines (824 loc) · 28.5 KB
/
alloc_runner.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
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
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
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
package allocrunner
import (
"context"
"fmt"
"path/filepath"
"sync"
"time"
log "github.com/hashicorp/go-hclog"
multierror "github.com/hashicorp/go-multierror"
"github.com/hashicorp/nomad/client/allocdir"
"github.com/hashicorp/nomad/client/allocrunner/interfaces"
"github.com/hashicorp/nomad/client/allocrunner/state"
"github.com/hashicorp/nomad/client/allocrunner/taskrunner"
"github.com/hashicorp/nomad/client/allocwatcher"
"github.com/hashicorp/nomad/client/config"
"github.com/hashicorp/nomad/client/consul"
"github.com/hashicorp/nomad/client/devicemanager"
cinterfaces "github.com/hashicorp/nomad/client/interfaces"
"github.com/hashicorp/nomad/client/pluginmanager/drivermanager"
cstate "github.com/hashicorp/nomad/client/state"
cstructs "github.com/hashicorp/nomad/client/structs"
"github.com/hashicorp/nomad/client/vaultclient"
"github.com/hashicorp/nomad/helper"
"github.com/hashicorp/nomad/nomad/structs"
"github.com/hashicorp/nomad/plugins/device"
"github.com/hashicorp/nomad/plugins/drivers"
)
// allocRunner is used to run all the tasks in a given allocation
type allocRunner struct {
// id is the ID of the allocation. Can be accessed without a lock
id string
// Logger is the logger for the alloc runner.
logger log.Logger
clientConfig *config.Config
// stateUpdater is used to emit updated alloc state
stateUpdater cinterfaces.AllocStateHandler
// taskStateUpdatedCh is ticked whenever task state as changed. Must
// have len==1 to allow nonblocking notification of state updates while
// the goroutine is already processing a previous update.
taskStateUpdatedCh chan struct{}
// taskStateUpdateHandlerCh is closed when the task state handling
// goroutine exits. It is unsafe to destroy the local allocation state
// before this goroutine exits.
taskStateUpdateHandlerCh chan struct{}
// allocUpdatedCh is a channel that is used to stream allocation updates into
// the allocUpdate handler. Must have len==1 to allow nonblocking notification
// of new allocation updates while the goroutine is processing a previous
// update.
allocUpdatedCh chan *structs.Allocation
// consulClient is the client used by the consul service hook for
// registering services and checks
consulClient consul.ConsulServiceAPI
// vaultClient is the used to manage Vault tokens
vaultClient vaultclient.VaultClient
// waitCh is closed when the Run loop has exited
waitCh chan struct{}
// destroyed is true when the Run loop has exited, postrun hooks have
// run, and alloc runner has been destroyed. Must acquire destroyedLock
// to access.
destroyed bool
// destroyCh is closed when the Run loop has exited, postrun hooks have
// run, and alloc runner has been destroyed.
destroyCh chan struct{}
// shutdown is true when the Run loop has exited, and shutdown hooks have
// run. Must acquire destroyedLock to access.
shutdown bool
// shutdownCh is closed when the Run loop has exited, and shutdown hooks
// have run.
shutdownCh chan struct{}
// destroyLaunched is true if Destroy has been called. Must acquire
// destroyedLock to access.
destroyLaunched bool
// shutdownLaunched is true if Shutdown has been called. Must acquire
// destroyedLock to access.
shutdownLaunched bool
// destroyedLock guards destroyed, destroyLaunched, shutdownLaunched,
// and serializes Shutdown/Destroy calls.
destroyedLock sync.Mutex
// Alloc captures the allocation being run.
alloc *structs.Allocation
allocLock sync.RWMutex
// state is the alloc runner's state
state *state.State
stateLock sync.RWMutex
stateDB cstate.StateDB
// allocDir is used to build the allocations directory structure.
allocDir *allocdir.AllocDir
// runnerHooks are alloc runner lifecycle hooks that should be run on state
// transistions.
runnerHooks []interfaces.RunnerHook
// tasks are the set of task runners
tasks map[string]*taskrunner.TaskRunner
// deviceStatsReporter is used to lookup resource usage for alloc devices
deviceStatsReporter cinterfaces.DeviceStatsReporter
// allocBroadcaster sends client allocation updates to all listeners
allocBroadcaster *cstructs.AllocBroadcaster
// prevAllocWatcher allows waiting for any previous or preempted allocations
// to exit
prevAllocWatcher allocwatcher.PrevAllocWatcher
// prevAllocMigrator allows the migration of a previous allocations alloc dir.
prevAllocMigrator allocwatcher.PrevAllocMigrator
// devicemanager is used to mount devices as well as lookup device
// statistics
devicemanager devicemanager.Manager
// driverManager is responsible for dispensing driver plugins and registering
// event handlers
driverManager drivermanager.Manager
}
// NewAllocRunner returns a new allocation runner.
func NewAllocRunner(config *Config) (*allocRunner, error) {
alloc := config.Alloc
tg := alloc.Job.LookupTaskGroup(alloc.TaskGroup)
if tg == nil {
return nil, fmt.Errorf("failed to lookup task group %q", alloc.TaskGroup)
}
ar := &allocRunner{
id: alloc.ID,
alloc: alloc,
clientConfig: config.ClientConfig,
consulClient: config.Consul,
vaultClient: config.Vault,
tasks: make(map[string]*taskrunner.TaskRunner, len(tg.Tasks)),
waitCh: make(chan struct{}),
destroyCh: make(chan struct{}),
shutdownCh: make(chan struct{}),
state: &state.State{},
stateDB: config.StateDB,
stateUpdater: config.StateUpdater,
taskStateUpdatedCh: make(chan struct{}, 1),
taskStateUpdateHandlerCh: make(chan struct{}),
allocUpdatedCh: make(chan *structs.Allocation, 1),
deviceStatsReporter: config.DeviceStatsReporter,
prevAllocWatcher: config.PrevAllocWatcher,
prevAllocMigrator: config.PrevAllocMigrator,
devicemanager: config.DeviceManager,
driverManager: config.DriverManager,
}
// Create the logger based on the allocation ID
ar.logger = config.Logger.Named("alloc_runner").With("alloc_id", alloc.ID)
// Create alloc broadcaster
ar.allocBroadcaster = cstructs.NewAllocBroadcaster(ar.logger)
// Create alloc dir
ar.allocDir = allocdir.NewAllocDir(ar.logger, filepath.Join(config.ClientConfig.AllocDir, alloc.ID))
// Initialize the runners hooks.
ar.initRunnerHooks()
// Create the TaskRunners
if err := ar.initTaskRunners(tg.Tasks); err != nil {
return nil, err
}
return ar, nil
}
// initTaskRunners creates task runners but does *not* run them.
func (ar *allocRunner) initTaskRunners(tasks []*structs.Task) error {
for _, task := range tasks {
config := &taskrunner.Config{
Alloc: ar.alloc,
ClientConfig: ar.clientConfig,
Task: task,
TaskDir: ar.allocDir.NewTaskDir(task.Name),
Logger: ar.logger,
StateDB: ar.stateDB,
StateUpdater: ar,
Consul: ar.consulClient,
Vault: ar.vaultClient,
DeviceStatsReporter: ar.deviceStatsReporter,
DeviceManager: ar.devicemanager,
DriverManager: ar.driverManager,
}
// Create, but do not Run, the task runner
tr, err := taskrunner.NewTaskRunner(config)
if err != nil {
return fmt.Errorf("failed creating runner for task %q: %v", task.Name, err)
}
ar.tasks[task.Name] = tr
}
return nil
}
func (ar *allocRunner) WaitCh() <-chan struct{} {
return ar.waitCh
}
// Run the AllocRunner. Starts tasks if the alloc is non-terminal and closes
// WaitCh when it exits. Should be started in a goroutine.
func (ar *allocRunner) Run() {
// Close the wait channel on return
defer close(ar.waitCh)
// Start the task state update handler
go ar.handleTaskStateUpdates()
// Start the alloc update handler
go ar.handleAllocUpdates()
// If task update chan has been closed, that means we've been shutdown.
select {
case <-ar.taskStateUpdateHandlerCh:
return
default:
}
// Run the prestart hooks if non-terminal
if ar.shouldRun() {
if err := ar.prerun(); err != nil {
ar.logger.Error("prerun failed", "error", err)
goto POST
}
}
// Run the runners (blocks until they exit)
ar.runTasks()
POST:
// Run the postrun hooks
if err := ar.postrun(); err != nil {
ar.logger.Error("postrun failed", "error", err)
}
}
// shouldRun returns true if the alloc is in a state that the alloc runner
// should run it.
func (ar *allocRunner) shouldRun() bool {
// Do not run allocs that are terminal
if ar.Alloc().TerminalStatus() {
ar.logger.Trace("alloc terminal; not running",
"desired_status", ar.Alloc().DesiredStatus,
"client_status", ar.Alloc().ClientStatus,
)
return false
}
// It's possible that the alloc local state was marked terminal before
// the server copy of the alloc (checked above) was marked as terminal,
// so check the local state as well.
switch clientStatus := ar.AllocState().ClientStatus; clientStatus {
case structs.AllocClientStatusComplete, structs.AllocClientStatusFailed, structs.AllocClientStatusLost:
ar.logger.Trace("alloc terminal; updating server and not running", "status", clientStatus)
return false
}
return true
}
// runTasks is used to run the task runners and block until they exit.
func (ar *allocRunner) runTasks() {
for _, task := range ar.tasks {
go task.Run()
}
for _, task := range ar.tasks {
<-task.WaitCh()
}
}
// Alloc returns the current allocation being run by this runner as sent by the
// server. This view of the allocation does not have updated task states.
func (ar *allocRunner) Alloc() *structs.Allocation {
ar.allocLock.RLock()
defer ar.allocLock.RUnlock()
return ar.alloc
}
func (ar *allocRunner) setAlloc(updated *structs.Allocation) {
ar.allocLock.Lock()
ar.alloc = updated
ar.allocLock.Unlock()
}
// GetAllocDir returns the alloc dir which is safe for concurrent use.
func (ar *allocRunner) GetAllocDir() *allocdir.AllocDir {
return ar.allocDir
}
// Restore state from database. Must be called after NewAllocRunner but before
// Run.
func (ar *allocRunner) Restore() error {
// Retrieve deployment status to avoid reseting it across agent
// restarts. Once a deployment status is set Nomad no longer monitors
// alloc health, so we must persist deployment state across restarts.
ds, err := ar.stateDB.GetDeploymentStatus(ar.id)
if err != nil {
return err
}
ar.stateLock.Lock()
ar.state.DeploymentStatus = ds
ar.stateLock.Unlock()
// Restore task runners
for _, tr := range ar.tasks {
if err := tr.Restore(); err != nil {
return err
}
}
return nil
}
// persistDeploymentStatus stores AllocDeploymentStatus.
func (ar *allocRunner) persistDeploymentStatus(ds *structs.AllocDeploymentStatus) {
if err := ar.stateDB.PutDeploymentStatus(ar.id, ds); err != nil {
// While any persistence errors are very bad, the worst case
// scenario for failing to persist deployment status is that if
// the agent is restarted it will monitor the deployment status
// again. This could cause a deployment's status to change when
// that shouldn't happen. However, allowing that seems better
// than failing the entire allocation.
ar.logger.Error("error storing deployment status", "error", err)
}
}
// TaskStateUpdated is called by TaskRunner when a task's state has been
// updated. It does not process the update synchronously but instead notifies a
// goroutine the state has change. Since processing the state change may cause
// the task to be killed (thus change its state again) it cannot be done
// synchronously as it would cause a deadlock due to reentrancy.
//
// The goroutine is used to compute changes to the alloc's ClientStatus and to
// update the server with the new state.
func (ar *allocRunner) TaskStateUpdated() {
select {
case ar.taskStateUpdatedCh <- struct{}{}:
default:
// already pending updates
}
}
// handleTaskStateUpdates must be run in goroutine as it monitors
// taskStateUpdatedCh for task state update notifications and processes task
// states.
//
// Processing task state updates must be done in a goroutine as it may have to
// kill tasks which causes further task state updates.
func (ar *allocRunner) handleTaskStateUpdates() {
defer close(ar.taskStateUpdateHandlerCh)
for done := false; !done; {
select {
case <-ar.taskStateUpdatedCh:
case <-ar.waitCh:
// Run has exited, sync once more to ensure final
// states are collected.
done = true
}
ar.logger.Trace("handling task state update", "done", done)
// Set with the appropriate event if task runners should be
// killed.
var killEvent *structs.TaskEvent
// If task runners should be killed, this is set to the task
// name whose fault it is.
killTask := ""
// True if task runners should be killed because a leader
// failed (informational).
leaderFailed := false
// Task state has been updated; gather the state of the other tasks
trNum := len(ar.tasks)
liveRunners := make([]*taskrunner.TaskRunner, 0, trNum)
states := make(map[string]*structs.TaskState, trNum)
for name, tr := range ar.tasks {
state := tr.TaskState()
states[name] = state
// Capture live task runners in case we need to kill them
if state.State != structs.TaskStateDead {
liveRunners = append(liveRunners, tr)
continue
}
// Task is dead, determine if other tasks should be killed
if state.Failed {
// Only set failed event if no event has been
// set yet to give dead leaders priority.
if killEvent == nil {
killTask = name
killEvent = structs.NewTaskEvent(structs.TaskSiblingFailed).
SetFailedSibling(name)
}
} else if tr.IsLeader() {
killEvent = structs.NewTaskEvent(structs.TaskLeaderDead)
leaderFailed = true
killTask = name
}
}
// If there's a kill event set and live runners, kill them
if killEvent != nil && len(liveRunners) > 0 {
// Log kill reason
if leaderFailed {
ar.logger.Debug("leader task dead, destroying all tasks", "leader_task", killTask)
} else {
ar.logger.Debug("task failure, destroying all tasks", "failed_task", killTask)
}
// Emit kill event for live runners
for _, tr := range liveRunners {
tr.EmitEvent(killEvent)
}
// Kill 'em all
states = ar.killTasks()
// Wait for TaskRunners to exit before continuing to
// prevent looping before TaskRunners have transitioned
// to Dead.
for _, tr := range liveRunners {
select {
case <-tr.WaitCh():
case <-ar.waitCh:
}
}
}
// Get the client allocation
calloc := ar.clientAlloc(states)
// Update the server
ar.stateUpdater.AllocStateUpdated(calloc)
// Broadcast client alloc to listeners
ar.allocBroadcaster.Send(calloc)
}
}
// killTasks kills all task runners, leader (if there is one) first. Errors are
// logged except taskrunner.ErrTaskNotRunning which is ignored. Task states
// after Kill has been called are returned.
func (ar *allocRunner) killTasks() map[string]*structs.TaskState {
var mu sync.Mutex
states := make(map[string]*structs.TaskState, len(ar.tasks))
// Kill leader first, synchronously
for name, tr := range ar.tasks {
if !tr.IsLeader() {
continue
}
err := tr.Kill(context.TODO(), structs.NewTaskEvent(structs.TaskKilling))
if err != nil && err != taskrunner.ErrTaskNotRunning {
ar.logger.Warn("error stopping leader task", "error", err, "task_name", name)
}
state := tr.TaskState()
states[name] = state
break
}
// Kill the rest concurrently
wg := sync.WaitGroup{}
for name, tr := range ar.tasks {
if tr.IsLeader() {
continue
}
wg.Add(1)
go func(name string, tr *taskrunner.TaskRunner) {
defer wg.Done()
err := tr.Kill(context.TODO(), structs.NewTaskEvent(structs.TaskKilling))
if err != nil && err != taskrunner.ErrTaskNotRunning {
ar.logger.Warn("error stopping task", "error", err, "task_name", name)
}
state := tr.TaskState()
mu.Lock()
states[name] = state
mu.Unlock()
}(name, tr)
}
wg.Wait()
return states
}
// clientAlloc takes in the task states and returns an Allocation populated
// with Client specific fields
func (ar *allocRunner) clientAlloc(taskStates map[string]*structs.TaskState) *structs.Allocation {
ar.stateLock.Lock()
defer ar.stateLock.Unlock()
// store task states for AllocState to expose
ar.state.TaskStates = taskStates
a := &structs.Allocation{
ID: ar.id,
TaskStates: taskStates,
}
if d := ar.state.DeploymentStatus; d != nil {
a.DeploymentStatus = d.Copy()
}
// Compute the ClientStatus
if ar.state.ClientStatus != "" {
// The client status is being forced
a.ClientStatus, a.ClientDescription = ar.state.ClientStatus, ar.state.ClientDescription
} else {
a.ClientStatus, a.ClientDescription = getClientStatus(taskStates)
}
// If the allocation is terminal, make sure all required fields are properly
// set.
if a.ClientTerminalStatus() {
alloc := ar.Alloc()
// If we are part of a deployment and the task has failed, mark the
// alloc as unhealthy. This guards against the watcher not be started.
if a.ClientStatus == structs.AllocClientStatusFailed &&
alloc.DeploymentID != "" && !a.DeploymentStatus.IsUnhealthy() {
a.DeploymentStatus = &structs.AllocDeploymentStatus{
Healthy: helper.BoolToPtr(false),
}
}
// Make sure we have marked the finished at for every task. This is used
// to calculate the reschedule time for failed allocations.
now := time.Now()
for _, task := range alloc.Job.LookupTaskGroup(alloc.TaskGroup).Tasks {
ts, ok := a.TaskStates[task.Name]
if !ok {
ts = &structs.TaskState{}
a.TaskStates[task.Name] = ts
}
if ts.FinishedAt.IsZero() {
ts.FinishedAt = now
}
}
}
return a
}
// getClientStatus takes in the task states for a given allocation and computes
// the client status and description
func getClientStatus(taskStates map[string]*structs.TaskState) (status, description string) {
var pending, running, dead, failed bool
for _, state := range taskStates {
switch state.State {
case structs.TaskStateRunning:
running = true
case structs.TaskStatePending:
pending = true
case structs.TaskStateDead:
if state.Failed {
failed = true
} else {
dead = true
}
}
}
// Determine the alloc status
if failed {
return structs.AllocClientStatusFailed, "Failed tasks"
} else if running {
return structs.AllocClientStatusRunning, "Tasks are running"
} else if pending {
return structs.AllocClientStatusPending, "No tasks have started"
} else if dead {
return structs.AllocClientStatusComplete, "All tasks have completed"
}
return "", ""
}
// SetClientStatus is a helper for forcing a specific client
// status on the alloc runner. This is used during restore errors
// when the task state can't be restored.
func (ar *allocRunner) SetClientStatus(clientStatus string) {
ar.stateLock.Lock()
defer ar.stateLock.Unlock()
ar.state.ClientStatus = clientStatus
}
// AllocState returns a copy of allocation state including a snapshot of task
// states.
func (ar *allocRunner) AllocState() *state.State {
ar.stateLock.RLock()
state := ar.state.Copy()
ar.stateLock.RUnlock()
// If TaskStateUpdated has not been called yet, ar.state.TaskStates
// won't be set as it is not the canonical source of TaskStates.
if len(state.TaskStates) == 0 {
ar.state.TaskStates = make(map[string]*structs.TaskState, len(ar.tasks))
for k, tr := range ar.tasks {
state.TaskStates[k] = tr.TaskState()
}
}
// Generate alloc to get other state fields
alloc := ar.clientAlloc(state.TaskStates)
state.ClientStatus = alloc.ClientStatus
state.ClientDescription = alloc.ClientDescription
state.DeploymentStatus = alloc.DeploymentStatus
return state
}
// Update asyncronously updates the running allocation with a new version
// received from the server.
// When processing a new update, we will first attempt to drain stale updates
// from the queue, before appending the new one.
func (ar *allocRunner) Update(update *structs.Allocation) {
select {
// Drain queued update from the channel if possible, and check the modify
// index
case oldUpdate := <-ar.allocUpdatedCh:
// If the old update is newer than the replacement, then skip the new one
// and return. This case shouldn't happen, but may in the case of a bug
// elsewhere inside the system.
if oldUpdate.AllocModifyIndex > update.AllocModifyIndex {
ar.logger.Debug("Discarding allocation update due to newer alloc revision in queue",
"old_modify_index", oldUpdate.AllocModifyIndex,
"new_modify_index", update.AllocModifyIndex)
ar.allocUpdatedCh <- oldUpdate
return
} else {
ar.logger.Debug("Discarding allocation update",
"skipped_modify_index", oldUpdate.AllocModifyIndex,
"new_modify_index", update.AllocModifyIndex)
}
case <-ar.waitCh:
ar.logger.Trace("AllocRunner has terminated, skipping alloc update",
"modify_index", update.AllocModifyIndex)
return
default:
}
// Queue the new update
ar.allocUpdatedCh <- update
}
func (ar *allocRunner) handleAllocUpdates() {
for {
select {
case update := <-ar.allocUpdatedCh:
ar.handleAllocUpdate(update)
case <-ar.waitCh:
return
}
}
}
// This method sends the updated alloc to Run for serially processing updates.
// If there is already a pending update it will be discarded and replaced by
// the latest update.
func (ar *allocRunner) handleAllocUpdate(update *structs.Allocation) {
// Detect Stop updates
stopping := !ar.Alloc().TerminalStatus() && update.TerminalStatus()
// Update ar.alloc
ar.setAlloc(update)
// Run update hooks if not stopping or dead
if !update.TerminalStatus() {
if err := ar.update(update); err != nil {
ar.logger.Error("error running update hooks", "error", err)
}
}
// Update task runners
for _, tr := range ar.tasks {
tr.Update(update)
}
// If alloc is being terminated, kill all tasks, leader first
if stopping {
ar.killTasks()
}
}
func (ar *allocRunner) Listener() *cstructs.AllocListener {
return ar.allocBroadcaster.Listen()
}
func (ar *allocRunner) destroyImpl() {
// Stop any running tasks and persist states in case the client is
// shutdown before Destroy finishes.
states := ar.killTasks()
calloc := ar.clientAlloc(states)
ar.stateUpdater.AllocStateUpdated(calloc)
// Wait for tasks to exit and postrun hooks to finish
<-ar.waitCh
// Run destroy hooks
if err := ar.destroy(); err != nil {
ar.logger.Warn("error running destroy hooks", "error", err)
}
// Wait for task state update handler to exit before removing local
// state if Run() ran at all.
<-ar.taskStateUpdateHandlerCh
// Cleanup state db
if err := ar.stateDB.DeleteAllocationBucket(ar.id); err != nil {
ar.logger.Warn("failed to delete allocation state", "error", err)
}
// Mark alloc as destroyed
ar.destroyedLock.Lock()
if !ar.shutdown {
ar.shutdown = true
close(ar.shutdownCh)
}
ar.destroyed = true
close(ar.destroyCh)
ar.destroyedLock.Unlock()
}
// Destroy the alloc runner by stopping it if it is still running and cleaning
// up all of its resources.
//
// This method is safe for calling concurrently with Run() and will cause it to
// exit (thus closing WaitCh).
// When the destroy action is completed, it will close DestroyCh().
func (ar *allocRunner) Destroy() {
ar.destroyedLock.Lock()
defer ar.destroyedLock.Unlock()
if ar.destroyed {
// Only destroy once
return
}
if ar.destroyLaunched {
// Only dispatch a destroy once
return
}
ar.destroyLaunched = true
// Synchronize calls to shutdown/destroy
if ar.shutdownLaunched {
go func() {
ar.logger.Debug("Waiting for shutdown before destroying runner")
<-ar.shutdownCh
ar.destroyImpl()
}()
return
}
go ar.destroyImpl()
}
// IsDestroyed returns true if the alloc runner has been destroyed (stopped and
// garbage collected).
//
// This method is safe for calling concurrently with Run(). Callers must
// receive on WaitCh() to block until alloc runner has stopped and been
// destroyed.
func (ar *allocRunner) IsDestroyed() bool {
ar.destroyedLock.Lock()
defer ar.destroyedLock.Unlock()
return ar.destroyed
}
// IsWaiting returns true if the alloc runner is waiting for its previous
// allocation to terminate.
//
// This method is safe for calling concurrently with Run().
func (ar *allocRunner) IsWaiting() bool {
return ar.prevAllocWatcher.IsWaiting()
}
// DestroyCh is a channel that is closed when an allocrunner is closed due to
// an explicit call to Destroy().
func (ar *allocRunner) DestroyCh() <-chan struct{} {
return ar.destroyCh
}
// ShutdownCh is a channel that is closed when an allocrunner is closed due to
// either an explicit call to Shutdown(), or Destroy().
func (ar *allocRunner) ShutdownCh() <-chan struct{} {
return ar.shutdownCh
}
// Shutdown AllocRunner gracefully. Asynchronously shuts down all TaskRunners.
// Tasks are unaffected and may be restored.
// When the destroy action is completed, it will close ShutdownCh().
func (ar *allocRunner) Shutdown() {
ar.destroyedLock.Lock()
defer ar.destroyedLock.Unlock()
// Destroy is a superset of Shutdown so there's nothing to do if this
// has already been destroyed.
if ar.destroyed {
return
}
// Destroy is a superset of Shutdown so if it's been marked for destruction,
// don't try and shutdown in parallel. If shutdown has been launched, don't
// try again.
if ar.destroyLaunched || ar.shutdownLaunched {
return
}
ar.shutdownLaunched = true
go func() {
ar.logger.Trace("shutting down")
// Shutdown tasks gracefully if they were run
wg := sync.WaitGroup{}
for _, tr := range ar.tasks {
wg.Add(1)
go func(tr *taskrunner.TaskRunner) {
tr.Shutdown()
wg.Done()
}(tr)
}
wg.Wait()
// Wait for Run to exit
<-ar.waitCh
// Run shutdown hooks
ar.shutdownHooks()
// Wait for updater to finish its final run
<-ar.taskStateUpdateHandlerCh
ar.destroyedLock.Lock()
ar.shutdown = true
close(ar.shutdownCh)
ar.destroyedLock.Unlock()
}()
}
// IsMigrating returns true if the alloc runner is migrating data from its
// previous allocation.
//
// This method is safe for calling concurrently with Run().
func (ar *allocRunner) IsMigrating() bool {
return ar.prevAllocMigrator.IsMigrating()
}
func (ar *allocRunner) StatsReporter() interfaces.AllocStatsReporter {
return ar
}
// LatestAllocStats returns the latest stats for an allocation. If taskFilter
// is set, only stats for that task -- if it exists -- are returned.
func (ar *allocRunner) LatestAllocStats(taskFilter string) (*cstructs.AllocResourceUsage, error) {
astat := &cstructs.AllocResourceUsage{
Tasks: make(map[string]*cstructs.TaskResourceUsage, len(ar.tasks)),
ResourceUsage: &cstructs.ResourceUsage{
MemoryStats: &cstructs.MemoryStats{},
CpuStats: &cstructs.CpuStats{},
DeviceStats: []*device.DeviceGroupStats{},
},
}
for name, tr := range ar.tasks {
if taskFilter != "" && taskFilter != name {
// Getting stats for a particular task and its not this one!
continue
}
if usage := tr.LatestResourceUsage(); usage != nil {
astat.Tasks[name] = usage
astat.ResourceUsage.Add(usage.ResourceUsage)
if usage.Timestamp > astat.Timestamp {
astat.Timestamp = usage.Timestamp
}
}
}
return astat, nil
}
func (ar *allocRunner) GetTaskEventHandler(taskName string) drivermanager.EventHandler {
if tr, ok := ar.tasks[taskName]; ok {
return func(ev *drivers.TaskEvent) {
tr.EmitEvent(&structs.TaskEvent{
Type: structs.TaskDriverMessage,
Time: ev.Timestamp.UnixNano(),
Details: ev.Annotations,
DriverMessage: ev.Message,
})
}
}
return nil
}
// RestartTask signalls the task runner for the provided task to restart.
func (ar *allocRunner) RestartTask(taskName string, taskEvent *structs.TaskEvent) error {
tr, ok := ar.tasks[taskName]
if !ok {
return fmt.Errorf("Could not find task runner for task: %s", taskName)
}
return tr.Restart(context.TODO(), taskEvent, false)
}
// RestartAll signalls all task runners in the allocation to restart and passes
// a copy of the task event to each restart event.
// Returns any errors in a concatenated form.
func (ar *allocRunner) RestartAll(taskEvent *structs.TaskEvent) error {
var err *multierror.Error
for tn := range ar.tasks {
rerr := ar.RestartTask(tn, taskEvent.Copy())
if rerr != nil {
err = multierror.Append(err, rerr)
}
}
return err.ErrorOrNil()
}
// Signal sends a signal request to task runners inside an allocation. If the
// taskName is empty, then it is sent to all tasks.
func (ar *allocRunner) Signal(taskName, signal string) error {
event := structs.NewTaskEvent(structs.TaskSignaling).SetSignalText(signal)
if taskName != "" {
tr, ok := ar.tasks[taskName]
if !ok {
return fmt.Errorf("Task not found")
}
return tr.Signal(event, signal)
}
var err *multierror.Error
for tn, tr := range ar.tasks {
rerr := tr.Signal(event.Copy(), signal)
if rerr != nil {
err = multierror.Append(err, fmt.Errorf("Failed to signal task: %s, err: %v", tn, rerr))
}
}
return err.ErrorOrNil()
}