-
Notifications
You must be signed in to change notification settings - Fork 1.9k
/
task_runner.go
1672 lines (1406 loc) · 53.2 KB
/
task_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
992
993
994
995
996
997
998
999
1000
// Copyright (c) HashiCorp, Inc.
// SPDX-License-Identifier: BUSL-1.1
package taskrunner
import (
"context"
"errors"
"fmt"
"slices"
"strings"
"sync"
"time"
metrics "github.com/armon/go-metrics"
log "github.com/hashicorp/go-hclog"
multierror "github.com/hashicorp/go-multierror"
"github.com/hashicorp/hcl/v2/hcldec"
"github.com/hashicorp/nomad/client/allocdir"
"github.com/hashicorp/nomad/client/allocrunner/interfaces"
"github.com/hashicorp/nomad/client/allocrunner/taskrunner/restarts"
"github.com/hashicorp/nomad/client/allocrunner/taskrunner/state"
"github.com/hashicorp/nomad/client/config"
"github.com/hashicorp/nomad/client/consul"
"github.com/hashicorp/nomad/client/devicemanager"
"github.com/hashicorp/nomad/client/dynamicplugins"
cinterfaces "github.com/hashicorp/nomad/client/interfaces"
"github.com/hashicorp/nomad/client/lib/cgroupslib"
"github.com/hashicorp/nomad/client/pluginmanager/csimanager"
"github.com/hashicorp/nomad/client/pluginmanager/drivermanager"
"github.com/hashicorp/nomad/client/serviceregistration"
"github.com/hashicorp/nomad/client/serviceregistration/wrapper"
cstate "github.com/hashicorp/nomad/client/state"
cstructs "github.com/hashicorp/nomad/client/structs"
"github.com/hashicorp/nomad/client/taskenv"
"github.com/hashicorp/nomad/client/vaultclient"
"github.com/hashicorp/nomad/client/widmgr"
"github.com/hashicorp/nomad/helper"
"github.com/hashicorp/nomad/helper/pluginutils/hclspecutils"
"github.com/hashicorp/nomad/helper/pluginutils/hclutils"
"github.com/hashicorp/nomad/helper/users/dynamic"
"github.com/hashicorp/nomad/helper/uuid"
"github.com/hashicorp/nomad/nomad/structs"
bstructs "github.com/hashicorp/nomad/plugins/base/structs"
"github.com/hashicorp/nomad/plugins/drivers"
)
const (
// defaultMaxEvents is the default max capacity for task events on the
// task state. Overrideable for testing.
defaultMaxEvents = 10
// killBackoffBaseline is the baseline time for exponential backoff while
// killing a task.
killBackoffBaseline = 5 * time.Second
// killBackoffLimit is the limit of the exponential backoff for killing
// the task.
killBackoffLimit = 2 * time.Minute
// killFailureLimit is how many times we will attempt to kill a task before
// giving up and potentially leaking resources.
killFailureLimit = 5
// triggerUpdateChCap is the capacity for the triggerUpdateCh used for
// triggering updates. It should be exactly 1 as even if multiple
// updates have come in since the last one was handled, we only need to
// handle the last one.
triggerUpdateChCap = 1
// restartChCap is the capacity for the restartCh used for triggering task
// restarts. It should be exactly 1 as even if multiple restarts have come
// we only need to handle the last one.
restartChCap = 1
)
type TaskRunner struct {
// allocID, taskName, taskLeader, and taskResources are immutable so these fields may
// be accessed without locks
allocID string
taskName string
taskLeader bool
taskResources *structs.AllocatedTaskResources
alloc *structs.Allocation
allocLock sync.Mutex
clientConfig *config.Config
// stateUpdater is used to emit updated task state
stateUpdater interfaces.TaskStateHandler
// state captures the state of the task for updating the allocation
// Must acquire stateLock to access.
state *structs.TaskState
// localState captures the node-local state of the task for when the
// Nomad agent restarts.
// Must acquire stateLock to access.
localState *state.LocalState
// stateLock must be acquired when accessing state or localState.
stateLock sync.RWMutex
// stateDB is for persisting localState and taskState
stateDB cstate.StateDB
// restartCh is used to signal that the task should restart.
restartCh chan struct{}
// shutdownCtx is used to exit the TaskRunner *without* affecting task state.
shutdownCtx context.Context
// shutdownCtxCancel causes the TaskRunner to exit immediately without
// affecting task state. Useful for testing or graceful agent shutdown.
shutdownCtxCancel context.CancelFunc
// killCtx is the task runner's context representing the tasks's lifecycle.
// The context is canceled when the task is killed.
killCtx context.Context
// killCtxCancel is called when killing a task.
killCtxCancel context.CancelFunc
// killErr is populated when killing a task. Access should be done use the
// getter/setter
killErr error
killErrLock sync.Mutex
// shutdownDelayCtx is a context from the alloc runner which will
// tell us to exit early from shutdown_delay
shutdownDelayCtx context.Context
shutdownDelayCancelFn context.CancelFunc
// Logger is the logger for the task runner.
logger log.Logger
// triggerUpdateCh is ticked whenever update hooks need to be run and
// must be created with cap=1 to signal a pending update and prevent
// callers from deadlocking if the receiver has exited.
triggerUpdateCh chan struct{}
// waitCh is closed when the task runner has transitioned to a terminal
// state
waitCh chan struct{}
// driver is the driver for the task.
driver drivers.DriverPlugin
// driverCapabilities is the set capabilities the driver supports
driverCapabilities *drivers.Capabilities
// taskSchema is the hcl spec for the task driver configuration
taskSchema hcldec.Spec
// handleLock guards access to handle and handleResult
handleLock sync.Mutex
// handle to the running driver
handle *DriverHandle
// task is the task being run
task *structs.Task
taskLock sync.RWMutex
// taskDir is the directory structure for this task.
taskDir *allocdir.TaskDir
// envBuilder is used to build the task's environment
envBuilder *taskenv.Builder
// restartTracker is used to decide if the task should be restarted.
restartTracker *restarts.RestartTracker
// runnerHooks are task runner lifecycle hooks that should be run on state
// transistions.
runnerHooks []interfaces.TaskHook
// hookResources captures the resources provided by hooks
hookResources *hookResources
// allocHookResources captures the resources provided by the allocrunner hooks
allocHookResources *cstructs.AllocHookResources
// consulClient is the client used by the consul service hook for
// registering services and checks
consulServiceClient serviceregistration.Handler
// consulProxiesClientFunc gets a client used by the envoy version hook for
// asking consul what version of envoy nomad should inject into the connect
// sidecar or gateway task.
consulProxiesClientFunc consul.SupportedProxiesAPIFunc
// sidsClient is the client used by the service identity hook for managing
// service identity tokens
siClient consul.ServiceIdentityAPI
// vaultClientFunc is the function to get a client to use to derive and
// renew Vault tokens
vaultClientFunc vaultclient.VaultClientFunc
// vaultToken is the current Vault token. It should be accessed with the
// getter.
vaultToken string
vaultTokenLock sync.Mutex
// nomadToken is the current Nomad workload identity token. It
// should be accessed with the getter.
nomadToken string
nomadTokenLock sync.Mutex
// baseLabels are used when emitting tagged metrics. All task runner metrics
// will have these tags, and optionally more.
baseLabels []metrics.Label
// logmonHookConfig is used to get the paths to the stdout and stderr fifos
// to be passed to the driver for task logging
logmonHookConfig *logmonHookConfig
// resourceUsage is written via UpdateStats and read via
// LatestResourceUsage. May be nil at all times.
resourceUsage *cstructs.TaskResourceUsage
resourceUsageLock sync.Mutex
// deviceStatsReporter is used to lookup resource usage for alloc devices
deviceStatsReporter cinterfaces.DeviceStatsReporter
// csiManager is used to manage the mounting of CSI volumes into tasks
csiManager csimanager.Manager
// devicemanager is used to mount devices as well as lookup device
// statistics
devicemanager devicemanager.Manager
// driverManager is used to dispense driver plugins and register event
// handlers
driverManager drivermanager.Manager
// dynamicRegistry is where dynamic plugins should be registered.
dynamicRegistry dynamicplugins.Registry
// maxEvents is the capacity of the TaskEvents on the TaskState.
// Defaults to defaultMaxEvents but overrideable for testing.
maxEvents int
// serversContactedCh is passed to TaskRunners so they can detect when
// GetClientAllocs has been called in case of a failed restore.
serversContactedCh <-chan struct{}
// startConditionMetCh signals the TaskRunner when it should start the task
startConditionMetCh <-chan struct{}
// waitOnServers defaults to false but will be set true if a restore
// fails and the Run method should wait until serversContactedCh is
// closed.
waitOnServers bool
networkIsolationLock sync.Mutex
networkIsolationSpec *drivers.NetworkIsolationSpec
// allocNetworkStatus is provided from the allocrunner and allows us to
// include this information as env vars for the task. When manipulating
// this the allocNetworkStatusLock should be used.
allocNetworkStatusLock sync.Mutex
allocNetworkStatus *structs.AllocNetworkStatus
// serviceRegWrapper is the handler wrapper that is used by service hooks
// to perform service and check registration and deregistration.
serviceRegWrapper *wrapper.HandlerWrapper
// getter is an interface for retrieving artifacts.
getter cinterfaces.ArtifactGetter
// wranglers manage unix/windows processes leveraging operating
// system features like cgroups
wranglers cinterfaces.ProcessWranglers
// widmgr manages workload identities
widmgr widmgr.IdentityManager
// users manages the pool of dynamic workload users
users dynamic.Pool
// pauser controls whether the task should be run or stopped based on a
// schedule. (Enterprise)
pauser *pauseGate
}
type Config struct {
Alloc *structs.Allocation
ClientConfig *config.Config
Task *structs.Task
TaskDir *allocdir.TaskDir
Logger log.Logger
// ConsulServices is used for managing Consul service registrations
ConsulServices serviceregistration.Handler
// ConsulProxiesFunc gets a client to use for looking up supported envoy versions
// from Consul.
ConsulProxiesFunc consul.SupportedProxiesAPIFunc
// ConsulSI is the client to use for managing Consul SI tokens
ConsulSI consul.ServiceIdentityAPI
// DynamicRegistry is where dynamic plugins should be registered.
DynamicRegistry dynamicplugins.Registry
// VaultFunc is function to get the client to use to derive and renew Vault tokens
VaultFunc vaultclient.VaultClientFunc
// StateDB is used to store and restore state.
StateDB cstate.StateDB
// StateUpdater is used to emit updated task state
StateUpdater interfaces.TaskStateHandler
// deviceStatsReporter is used to lookup resource usage for alloc devices
DeviceStatsReporter cinterfaces.DeviceStatsReporter
// CSIManager is used to manage the mounting of CSI volumes into tasks
CSIManager csimanager.Manager
// DeviceManager is used to mount devices as well as lookup device
// statistics
DeviceManager devicemanager.Manager
// DriverManager is used to dispense driver plugins and register event
// handlers
DriverManager drivermanager.Manager
// ServersContactedCh is closed when the first GetClientAllocs call to
// servers succeeds and allocs are synced.
ServersContactedCh chan struct{}
// StartConditionMetCh signals the TaskRunner when it should start the task
StartConditionMetCh <-chan struct{}
// ShutdownDelayCtx is a context from the alloc runner which will
// tell us to exit early from shutdown_delay
ShutdownDelayCtx context.Context
// ShutdownDelayCancelFn should only be used in testing.
ShutdownDelayCancelFn context.CancelFunc
// ServiceRegWrapper is the handler wrapper that is used by service hooks
// to perform service and check registration and deregistration.
ServiceRegWrapper *wrapper.HandlerWrapper
// Getter is an interface for retrieving artifacts.
Getter cinterfaces.ArtifactGetter
// Wranglers is an interface for managing OS processes.
Wranglers cinterfaces.ProcessWranglers
// AllocHookResources is how taskrunner hooks can get state written by
// allocrunner hooks
AllocHookResources *cstructs.AllocHookResources
// WIDMgr manages workload identities
WIDMgr widmgr.IdentityManager
// Users manages a pool of dynamic workload users
Users dynamic.Pool
}
func NewTaskRunner(config *Config) (*TaskRunner, error) {
// Create a context for causing the runner to exit
trCtx, trCancel := context.WithCancel(context.Background())
// Create a context for killing the runner
killCtx, killCancel := context.WithCancel(context.Background())
// Initialize the environment builder
envBuilder := taskenv.NewBuilder(
config.ClientConfig.Node,
config.Alloc,
config.Task,
config.ClientConfig.Region,
)
// Initialize state from alloc if it is set
tstate := structs.NewTaskState()
if ts := config.Alloc.TaskStates[config.Task.Name]; ts != nil {
tstate = ts.Copy()
}
tr := &TaskRunner{
alloc: config.Alloc,
allocID: config.Alloc.ID,
clientConfig: config.ClientConfig,
task: config.Task,
taskDir: config.TaskDir,
taskName: config.Task.Name,
taskLeader: config.Task.Leader,
envBuilder: envBuilder,
dynamicRegistry: config.DynamicRegistry,
consulServiceClient: config.ConsulServices,
consulProxiesClientFunc: config.ConsulProxiesFunc,
siClient: config.ConsulSI,
vaultClientFunc: config.VaultFunc,
state: tstate,
localState: state.NewLocalState(),
allocHookResources: config.AllocHookResources,
stateDB: config.StateDB,
stateUpdater: config.StateUpdater,
deviceStatsReporter: config.DeviceStatsReporter,
killCtx: killCtx,
killCtxCancel: killCancel,
shutdownCtx: trCtx,
shutdownCtxCancel: trCancel,
triggerUpdateCh: make(chan struct{}, triggerUpdateChCap),
restartCh: make(chan struct{}, restartChCap),
waitCh: make(chan struct{}),
csiManager: config.CSIManager,
devicemanager: config.DeviceManager,
driverManager: config.DriverManager,
maxEvents: defaultMaxEvents,
serversContactedCh: config.ServersContactedCh,
startConditionMetCh: config.StartConditionMetCh,
shutdownDelayCtx: config.ShutdownDelayCtx,
shutdownDelayCancelFn: config.ShutdownDelayCancelFn,
serviceRegWrapper: config.ServiceRegWrapper,
getter: config.Getter,
wranglers: config.Wranglers,
widmgr: config.WIDMgr,
users: config.Users,
}
// Create the logger based on the allocation ID
tr.logger = config.Logger.Named("task_runner").With("task", config.Task.Name)
// Create the pauser
tr.pauser = newPauseGate(tr)
// Pull out the task's resources
ares := tr.alloc.AllocatedResources
if ares == nil {
return nil, fmt.Errorf("no task resources found on allocation")
}
tres, ok := ares.Tasks[tr.taskName]
if !ok {
return nil, fmt.Errorf("no task resources found on allocation")
}
// we had to allocate the tmpfs with the memory to get correct scheduling
// and tracking on the node, but now that we're creating the task driver
// config we only care about the memory without the secrets.
tres.Memory.MemoryMB -= int64(tr.task.Resources.SecretsMB)
tr.taskResources = tres
// Build the restart tracker.
rp := config.Task.RestartPolicy
if rp == nil {
tg := tr.alloc.Job.LookupTaskGroup(tr.alloc.TaskGroup)
if tg == nil {
tr.logger.Error("alloc missing task group")
return nil, fmt.Errorf("alloc missing task group")
}
rp = tg.RestartPolicy
}
tr.restartTracker = restarts.NewRestartTracker(rp, tr.alloc.Job.Type, config.Task.Lifecycle)
// Get the driver
if err := tr.initDriver(); err != nil {
tr.logger.Error("failed to create driver", "error", err)
return nil, err
}
// Use the client secret only as the initial value; the identity hook will
// update this with a workload identity if one is available
tr.setNomadToken(config.ClientConfig.Node.SecretID)
// Initialize the runners hooks. Must come after initDriver so hooks
// can use tr.driverCapabilities
tr.initHooks()
// Initialize base labels
tr.initLabels()
// Initialize initial task received event
tr.appendEvent(structs.NewTaskEvent(structs.TaskReceived))
return tr, nil
}
func (tr *TaskRunner) initLabels() {
alloc := tr.Alloc()
tr.baseLabels = []metrics.Label{
{
Name: "job",
Value: alloc.Job.Name,
},
{
Name: "task_group",
Value: alloc.TaskGroup,
},
{
Name: "alloc_id",
Value: tr.allocID,
},
{
Name: "task",
Value: tr.taskName,
},
{
Name: "namespace",
Value: tr.alloc.Namespace,
},
}
if tr.alloc.Job.ParentID != "" {
tr.baseLabels = append(tr.baseLabels, metrics.Label{
Name: "parent_id",
Value: tr.alloc.Job.ParentID,
})
if strings.Contains(tr.alloc.Job.Name, "/dispatch-") {
tr.baseLabels = append(tr.baseLabels, metrics.Label{
Name: "dispatch_id",
Value: strings.Split(tr.alloc.Job.Name, "/dispatch-")[1],
})
}
if strings.Contains(tr.alloc.Job.Name, "/periodic-") {
tr.baseLabels = append(tr.baseLabels, metrics.Label{
Name: "periodic_id",
Value: strings.Split(tr.alloc.Job.Name, "/periodic-")[1],
})
}
}
}
// MarkFailedKill marks a task as failed and should be killed.
// It should be invoked when alloc runner prestart hooks fail.
// Afterwards, Run() will perform any necessary cleanup.
func (tr *TaskRunner) MarkFailedKill(reason string) {
// Emit an event that fails the task and gives reasons for humans.
event := structs.NewTaskEvent(structs.TaskSetupFailure).
SetKillReason(structs.TaskRestoreFailed).
SetDisplayMessage(reason).
SetFailsTask()
tr.EmitEvent(event)
// Cancel kill context, so later when allocRunner runs tr.Run(),
// we'll follow the usual kill path and do all the appropriate cleanup steps.
tr.killCtxCancel()
}
// Run the TaskRunner. Starts the user's task or reattaches to a restored task.
// Run closes WaitCh when it exits. Should be started in a goroutine.
func (tr *TaskRunner) Run() {
defer close(tr.waitCh)
var result *drivers.ExitResult
tr.stateLock.RLock()
dead := tr.state.State == structs.TaskStateDead
runComplete := tr.localState.RunComplete
tr.stateLock.RUnlock()
// If restoring a dead task, ensure the task is cleared and, if the local
// state indicates that the previous Run() call is complete, execute all
// post stop hooks and exit early, otherwise proceed until the
// ALLOC_RESTART loop skipping MAIN since the task is dead.
if dead {
// do cleanup functions without emitting any additional events/work
// to handle cases where we restored a dead task where client terminated
// after task finished before completing post-run actions.
tr.clearDriverHandle()
tr.stateUpdater.TaskStateUpdated()
if runComplete {
if err := tr.stop(); err != nil {
tr.logger.Error("stop failed on terminal task", "error", err)
}
return
}
}
// Updates are handled asynchronously with the other hooks but each
// triggered update - whether due to alloc updates or a new vault token
// - should be handled serially.
go tr.handleUpdates()
// If restore failed wait until servers are contacted before running.
// #1795
if tr.waitOnServers {
tr.logger.Info("task failed to restore; waiting to contact server before restarting")
select {
case <-tr.killCtx.Done():
tr.logger.Info("task killed while waiting for server contact")
case <-tr.shutdownCtx.Done():
return
case <-tr.serversContactedCh:
tr.logger.Info("server contacted; unblocking waiting task")
}
}
// Set the initial task state.
tr.stateUpdater.TaskStateUpdated()
// start with a stopped timer; actual restart delay computed later
timer, stop := helper.NewStoppedTimer()
defer stop()
MAIN:
for !tr.shouldShutdown() {
if dead {
break
}
select {
case <-tr.killCtx.Done():
break MAIN
case <-tr.shutdownCtx.Done():
// TaskRunner was told to exit immediately
return
case <-tr.startConditionMetCh:
tr.logger.Debug("lifecycle start condition has been met, proceeding")
// yay proceed
}
// Run the prestart hooks
if err := tr.prestart(); err != nil {
tr.logger.Error("prestart failed", "error", err)
tr.restartTracker.SetStartError(err)
goto RESTART
}
// Unblocks when the task runner is allowed to continue. (Enterprise)
if err := tr.pauser.Wait(); err != nil {
tr.logger.Error("pause scheduled failed", "error", err)
tr.restartTracker.SetStartError(err)
break MAIN
}
// Check for a terminal allocation once more before proceeding as the
// prestart hooks may have been skipped.
if tr.shouldShutdown() {
break MAIN
}
select {
case <-tr.killCtx.Done():
break MAIN
case <-tr.shutdownCtx.Done():
// TaskRunner was told to exit immediately
return
default:
}
// Run the task
if err := tr.runDriver(); err != nil {
tr.logger.Error("running driver failed", "error", err)
tr.restartTracker.SetStartError(err)
goto RESTART
}
// Run the poststart hooks
if err := tr.poststart(); err != nil {
tr.logger.Error("poststart failed", "error", err)
}
// Grab the result proxy and wait for task to exit
WAIT:
{
handle := tr.getDriverHandle()
result = nil
// Do *not* use tr.killCtx here as it would cause
// Wait() to unblock before the task exits when Kill()
// is called.
if resultCh, err := handle.WaitCh(context.Background()); err != nil {
tr.logger.Error("wait task failed", "error", err)
} else {
select {
case <-tr.killCtx.Done():
// We can go through the normal should restart check since
// the restart tracker knowns it is killed
result = tr.handleKill(resultCh)
case <-tr.shutdownCtx.Done():
// TaskRunner was told to exit immediately
return
case result = <-resultCh:
}
// WaitCh returned a result
if retryWait := tr.handleTaskExitResult(result); retryWait {
goto WAIT
}
}
}
// Clear the handle
tr.clearDriverHandle()
// Store the wait result on the restart tracker
tr.restartTracker.SetExitResult(result)
if err := tr.exited(); err != nil {
tr.logger.Error("exited hooks failed", "error", err)
}
RESTART:
restart, restartDelay := tr.shouldRestart()
if !restart {
break MAIN
}
timer.Reset(restartDelay)
// Actually restart by sleeping and also watching for destroy events
select {
case <-timer.C:
case <-tr.killCtx.Done():
tr.logger.Trace("task killed between restarts", "delay", restartDelay)
break MAIN
case <-tr.shutdownCtx.Done():
// TaskRunner was told to exit immediately
tr.logger.Trace("gracefully shutting down during restart delay")
return
}
}
// Ensure handle is cleaned up. Restore could have recovered a task
// that should be terminal, so if the handle still exists we should
// kill it here.
if tr.getDriverHandle() != nil {
if result = tr.handleKill(nil); result != nil {
tr.emitExitResultEvent(result)
}
tr.clearDriverHandle()
if err := tr.exited(); err != nil {
tr.logger.Error("exited hooks failed while cleaning up terminal task", "error", err)
}
}
// Mark the task as dead
tr.UpdateState(structs.TaskStateDead, nil)
// Wait here in case the allocation is restarted. Poststop tasks will never
// run again so skip them to avoid blocking forever.
if !tr.Task().IsPoststop() {
ALLOC_RESTART:
// Run in a loop to handle cases where restartCh is triggered but the
// task runner doesn't need to restart.
for {
select {
case <-tr.killCtx.Done():
break ALLOC_RESTART
case <-tr.shutdownCtx.Done():
return
case <-tr.restartCh:
// Restart without delay since the task is not running anymore.
restart, _ := tr.shouldRestart()
if restart {
// Set runner as not dead to allow the MAIN loop to run.
dead = false
goto MAIN
}
}
}
}
tr.stateLock.Lock()
tr.localState.RunComplete = true
err := tr.stateDB.PutTaskRunnerLocalState(tr.allocID, tr.taskName, tr.localState)
if err != nil {
tr.logger.Warn("error persisting task state on run loop exit", "error", err)
}
tr.stateLock.Unlock()
// Run the stop hooks
if err := tr.stop(); err != nil {
tr.logger.Error("stop failed", "error", err)
}
tr.logger.Debug("task run loop exiting")
}
func (tr *TaskRunner) shouldShutdown() bool {
alloc := tr.Alloc()
if alloc.ClientTerminalStatus() {
return true
}
if !tr.IsPoststopTask() && alloc.ServerTerminalStatus() {
return true
}
return false
}
// handleTaskExitResult handles the results returned by the task exiting. If
// retryWait is true, the caller should attempt to wait on the task again since
// it has not actually finished running. This can happen if the driver plugin
// has exited.
func (tr *TaskRunner) handleTaskExitResult(result *drivers.ExitResult) (retryWait bool) {
if result == nil {
return false
}
if result.Err == bstructs.ErrPluginShutdown {
dn := tr.Task().Driver
tr.logger.Debug("driver plugin has shutdown; attempting to recover task", "driver", dn)
// Initialize a new driver handle
if err := tr.initDriver(); err != nil {
tr.logger.Error("failed to initialize driver after it exited unexpectedly", "error", err, "driver", dn)
return false
}
// Try to restore the handle
tr.stateLock.RLock()
h := tr.localState.TaskHandle
net := tr.localState.DriverNetwork
tr.stateLock.RUnlock()
if !tr.restoreHandle(h, net) {
tr.logger.Error("failed to restore handle on driver after it exited unexpectedly", "driver", dn)
return false
}
tr.logger.Debug("task successfully recovered on driver", "driver", dn)
return true
}
// Emit Terminated event
tr.emitExitResultEvent(result)
return false
}
// emitExitResultEvent emits a TaskTerminated event for an ExitResult.
func (tr *TaskRunner) emitExitResultEvent(result *drivers.ExitResult) {
event := structs.NewTaskEvent(structs.TaskTerminated).
SetExitCode(result.ExitCode).
SetSignal(result.Signal).
SetOOMKilled(result.OOMKilled).
SetExitMessage(result.Err)
tr.EmitEvent(event)
if result.OOMKilled {
metrics.IncrCounterWithLabels([]string{"client", "allocs", "oom_killed"}, 1, tr.baseLabels)
}
}
// handleUpdates runs update hooks when triggerUpdateCh is ticked and exits
// when Run has returned. Should only be run in a goroutine from Run.
func (tr *TaskRunner) handleUpdates() {
for {
select {
case <-tr.triggerUpdateCh:
case <-tr.waitCh:
return
}
// Non-terminal update; run hooks
tr.updateHooks()
}
}
// shouldRestart determines whether the task should be restarted and updates
// the task state unless the task is killed or terminated.
func (tr *TaskRunner) shouldRestart() (bool, time.Duration) {
// Determine if we should restart
state, when := tr.restartTracker.GetState()
reason := tr.restartTracker.GetReason()
switch state {
case structs.TaskKilled:
// Never restart an explicitly killed task. Kill method handles
// updating the server.
tr.EmitEvent(structs.NewTaskEvent(state))
return false, 0
case structs.TaskNotRestarting, structs.TaskTerminated:
tr.logger.Info("not restarting task", "reason", reason)
if state == structs.TaskNotRestarting {
tr.UpdateState(structs.TaskStateDead, structs.NewTaskEvent(structs.TaskNotRestarting).SetRestartReason(reason).SetFailsTask())
}
return false, 0
case structs.TaskRestarting:
tr.logger.Info("restarting task", "reason", reason, "delay", when)
tr.UpdateState(structs.TaskStatePending, structs.NewTaskEvent(structs.TaskRestarting).SetRestartDelay(when).SetRestartReason(reason))
return true, when
default:
tr.logger.Error("restart tracker returned unknown state", "state", state)
return true, when
}
}
func (tr *TaskRunner) assignCgroup(taskConfig *drivers.TaskConfig) {
reserveCores := len(tr.taskResources.Cpu.ReservedCores) > 0
p := cgroupslib.LinuxResourcesPath(taskConfig.AllocID, taskConfig.Name, reserveCores)
taskConfig.Resources.LinuxResources.CpusetCgroupPath = p
}
// runDriver runs the driver and waits for it to exit
// runDriver emits an appropriate task event on success/failure
func (tr *TaskRunner) runDriver() error {
taskConfig := tr.buildTaskConfig()
tr.assignCgroup(taskConfig)
// Build hcl context variables
vars, errs, err := tr.envBuilder.Build().AllValues()
if err != nil {
return fmt.Errorf("error building environment variables: %v", err)
}
// Handle per-key errors
if len(errs) > 0 {
keys := make([]string, 0, len(errs))
for k, err := range errs {
keys = append(keys, k)
if tr.logger.IsTrace() {
// Verbosely log every diagnostic for debugging
tr.logger.Trace("error building environment variables", "key", k, "error", err)
}
}
tr.logger.Warn("some environment variables not available for rendering", "keys", strings.Join(keys, ", "))
}
val, diag, diagErrs := hclutils.ParseHclInterface(tr.task.Config, tr.taskSchema, vars)
if diag.HasErrors() {
parseErr := multierror.Append(errors.New("failed to parse config: "), diagErrs...)
tr.EmitEvent(structs.NewTaskEvent(structs.TaskFailedValidation).SetValidationError(parseErr))
return parseErr
}
if err := taskConfig.EncodeDriverConfig(val); err != nil {
encodeErr := fmt.Errorf("failed to encode driver config: %v", err)
tr.EmitEvent(structs.NewTaskEvent(structs.TaskFailedValidation).SetValidationError(encodeErr))
return encodeErr
}
// If there's already a task handle (eg from a Restore) there's nothing
// to do except update state.
if tr.getDriverHandle() != nil {
// Ensure running state is persisted but do *not* append a new
// task event as restoring is a client event and not relevant
// to a task's lifecycle.
if err := tr.updateStateImpl(structs.TaskStateRunning); err != nil {
//TODO return error and destroy task to avoid an orphaned task?
tr.logger.Warn("error persisting task state", "error", err)
}
return nil
}
// Start the job if there's no existing handle (or if RecoverTask failed)
handle, net, err := tr.driver.StartTask(taskConfig)
if err != nil {
// The plugin has died, try relaunching it
if err == bstructs.ErrPluginShutdown {
tr.logger.Info("failed to start task because plugin shutdown unexpectedly; attempting to recover")
if err := tr.initDriver(); err != nil {
taskErr := fmt.Errorf("failed to initialize driver after it exited unexpectedly: %v", err)
tr.EmitEvent(structs.NewTaskEvent(structs.TaskDriverFailure).SetDriverError(taskErr))
return taskErr
}
handle, net, err = tr.driver.StartTask(taskConfig)
if err != nil {
taskErr := fmt.Errorf("failed to start task after driver exited unexpectedly: %v", err)
tr.EmitEvent(structs.NewTaskEvent(structs.TaskDriverFailure).SetDriverError(taskErr))
return taskErr
}
} else {
// Do *NOT* wrap the error here without maintaining whether or not is Recoverable.
// You must emit a task event failure to be considered Recoverable
tr.EmitEvent(structs.NewTaskEvent(structs.TaskDriverFailure).SetDriverError(err))
return err
}
}
tr.stateLock.Lock()
tr.localState.TaskHandle = handle
tr.localState.DriverNetwork = net
if err := tr.stateDB.PutTaskRunnerLocalState(tr.allocID, tr.taskName, tr.localState); err != nil {
//TODO Nomad will be unable to restore this task; try to kill
// it now and fail? In general we prefer to leave running
// tasks running even if the agent encounters an error.
tr.logger.Warn("error persisting local task state; may be unable to restore after a Nomad restart",
"error", err, "task_id", handle.Config.ID)
}
tr.stateLock.Unlock()
tr.setDriverHandle(NewDriverHandle(tr.driver, taskConfig.ID, tr.Task(), tr.clientConfig.MaxKillTimeout, net))
// Emit an event that we started
tr.UpdateState(structs.TaskStateRunning, structs.NewTaskEvent(structs.TaskStarted))
return nil
}
// initDriver retrives the DriverPlugin from the plugin loader for this task
func (tr *TaskRunner) initDriver() error {
driver, err := tr.driverManager.Dispense(tr.Task().Driver)
if err != nil {
return err
}