-
Notifications
You must be signed in to change notification settings - Fork 50
/
overlord.go
585 lines (505 loc) · 15.7 KB
/
overlord.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
// Copyright (c) 2014-2020 Canonical Ltd
//
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License version 3 as
// published by the Free Software Foundation.
//
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU General Public License
// along with this program. If not, see <http://www.gnu.org/licenses/>.
// Package overlord is the central control base, and ruler of all things.
package overlord
import (
"errors"
"fmt"
"io"
"os"
"path/filepath"
"sync"
"sync/atomic"
"time"
"github.com/canonical/x-go/randutil"
"gopkg.in/tomb.v2"
"github.com/canonical/pebble/internals/osutil"
"github.com/canonical/pebble/internals/overlord/checkstate"
"github.com/canonical/pebble/internals/overlord/cmdstate"
"github.com/canonical/pebble/internals/overlord/logstate"
"github.com/canonical/pebble/internals/overlord/patch"
"github.com/canonical/pebble/internals/overlord/planstate"
"github.com/canonical/pebble/internals/overlord/restart"
"github.com/canonical/pebble/internals/overlord/servstate"
"github.com/canonical/pebble/internals/overlord/state"
"github.com/canonical/pebble/internals/timing"
)
var (
ensureInterval = 5 * time.Minute
pruneInterval = 10 * time.Minute
// In snapd this is 24h, but that's too short in the context of Pebble.
pruneWait = 24 * time.Hour * 7
// In snapd this is 7d, but also increase that in the context of Pebble.
abortWait = 24 * time.Hour * 14
pruneMaxChanges = 500
)
var pruneTickerC = func(t *time.Ticker) <-chan time.Time {
return t.C
}
// Extension represents an extension of the Overlord.
type Extension interface {
// ExtraManagers allows additional StateManagers to be used.
ExtraManagers(o *Overlord) ([]StateManager, error)
}
// Options is the arguments passed to construct an Overlord.
type Options struct {
// PebbleDir is the path to the pebble directory. It must be provided.
PebbleDir string
// RestartHandler is an optional structure to handle restart requests.
RestartHandler restart.Handler
// ServiceOutput is an optional output for the logging manager.
ServiceOutput io.Writer
// Extension allows extending the overlord with externally defined features.
Extension Extension
}
// Overlord is the central manager of the system, keeping track
// of all available state managers and related helpers.
type Overlord struct {
pebbleDir string
stateEng *StateEngine
// ensure loop
loopTomb *tomb.Tomb
ensureLock sync.Mutex
ensureTimer *time.Timer
ensureNext time.Time
ensureRun int32
pruneTicker *time.Ticker
startOfOperationTime time.Time
// managers
inited bool
startedUp bool
runner *state.TaskRunner
planMgr *planstate.PlanManager
serviceMgr *servstate.ServiceManager
commandMgr *cmdstate.CommandManager
checkMgr *checkstate.CheckManager
logMgr *logstate.LogManager
extension Extension
}
// New creates an Overlord with all its state managers.
func New(opts *Options) (*Overlord, error) {
o := &Overlord{
pebbleDir: opts.PebbleDir,
loopTomb: new(tomb.Tomb),
inited: true,
extension: opts.Extension,
}
if !filepath.IsAbs(o.pebbleDir) {
return nil, fmt.Errorf("directory %q must be absolute", o.pebbleDir)
}
if !osutil.IsDir(o.pebbleDir) {
return nil, fmt.Errorf("directory %q does not exist", o.pebbleDir)
}
statePath := filepath.Join(o.pebbleDir, ".pebble.state")
backend := &overlordStateBackend{
path: statePath,
ensureBefore: o.ensureBefore,
}
s, err := loadState(statePath, opts.RestartHandler, backend)
if err != nil {
return nil, err
}
o.stateEng = NewStateEngine(s)
o.runner = state.NewTaskRunner(s)
// any unknown task should be ignored and succeed
matchAnyUnknownTask := func(_ *state.Task) bool {
return true
}
o.runner.AddOptionalHandler(matchAnyUnknownTask, handleUnknownTask, nil)
o.planMgr, err = planstate.NewManager(s, o.runner, o.pebbleDir)
if err != nil {
return nil, fmt.Errorf("cannot create plan manager: %w", err)
}
o.stateEng.AddManager(o.planMgr)
o.logMgr = logstate.NewLogManager()
o.serviceMgr, err = servstate.NewManager(
s,
o.runner,
opts.ServiceOutput,
opts.RestartHandler,
o.logMgr)
if err != nil {
return nil, fmt.Errorf("cannot create service manager: %w", err)
}
// Tell service manager about plan updates.
o.planMgr.AddChangeListener(o.serviceMgr.PlanChanged)
o.stateEng.AddManager(o.serviceMgr)
// The log manager should be stopped after the service manager, because
// ServiceManager.Stop closes the service ring buffers, which signals to the
// log manager that it's okay to stop log forwarding.
o.stateEng.AddManager(o.logMgr)
o.commandMgr = cmdstate.NewManager(o.runner)
o.stateEng.AddManager(o.commandMgr)
o.checkMgr = checkstate.NewManager(s, o.runner)
o.stateEng.AddManager(o.checkMgr)
// Tell check manager about plan updates.
o.planMgr.AddChangeListener(o.checkMgr.PlanChanged)
// Tell log manager about plan updates.
o.planMgr.AddChangeListener(o.logMgr.PlanChanged)
// Tell service manager about check failures.
o.checkMgr.NotifyCheckFailed(o.serviceMgr.CheckFailed)
if o.extension != nil {
extraManagers, err := o.extension.ExtraManagers(o)
if err != nil {
return nil, fmt.Errorf("cannot add extra managers: %w", err)
}
for _, manager := range extraManagers {
o.stateEng.AddManager(manager)
}
}
// TaskRunner must be the last manager added to the StateEngine,
// because TaskRunner runs all the tasks required by the managers that ran
// before it.
o.stateEng.AddManager(o.runner)
// Load the plan from the Pebble layers directory (which may be missing
// or have no layers, resulting in an empty plan), and propagate PlanChanged
// notifications to all notification subscribers.
err = o.planMgr.Load()
if err != nil {
return nil, fmt.Errorf("cannot load plan: %w", err)
}
return o, nil
}
func (o *Overlord) Extension() Extension {
return o.extension
}
func loadState(statePath string, restartHandler restart.Handler, backend state.Backend) (*state.State, error) {
timings := timing.Start("", "", map[string]string{"startup": "load-state"})
curBootID, err := osutil.BootID()
if err != nil {
return nil, fmt.Errorf("fatal: cannot find current boot ID: %v", err)
}
// If pebble is PID 1 we don't care about /proc/sys/kernel/random/boot_id
// as we are most likely running in a container. LXD mounts it's own boot_id
// to correctly emulate the boot_id behaviour of non-containerized systems.
// Within containerd/docker, boot_id is consistent with the host, which provides
// us no context of restarts, so instead fallback to /proc/sys/kernel/random/uuid.
if os.Getpid() == 1 {
curBootID, err = randutil.RandomKernelUUID()
if err != nil {
return nil, fmt.Errorf("fatal: cannot generate psuedo boot-id: %v", err)
}
}
if !osutil.CanStat(statePath) {
// fail fast, mostly interesting for tests, this dir is set up by pebble
stateDir := filepath.Dir(statePath)
if !osutil.IsDir(stateDir) {
return nil, fmt.Errorf("fatal: directory %q must be present", stateDir)
}
s := state.New(backend)
initRestart(s, curBootID, restartHandler)
patch.Init(s)
return s, nil
}
r, err := os.Open(statePath)
if err != nil {
return nil, fmt.Errorf("cannot read the state file: %s", err)
}
defer r.Close()
var s *state.State
span := timings.StartNested("read-state", "read state from disk")
s, err = state.ReadState(backend, r)
span.Stop()
if err != nil {
return nil, err
}
timings.Stop()
// TODO Implement function to save timings.
//s.Lock()
//perfTimings.Save(s)
//s.Unlock()
err = initRestart(s, curBootID, restartHandler)
if err != nil {
return nil, err
}
// one-shot migrations
err = patch.Apply(s)
if err != nil {
return nil, err
}
return s, nil
}
func initRestart(s *state.State, curBootID string, restartHandler restart.Handler) error {
s.Lock()
defer s.Unlock()
return restart.Init(s, curBootID, restartHandler)
}
func (o *Overlord) StartUp() error {
if o.startedUp {
return nil
}
o.startedUp = true
var err error
st := o.State()
st.Lock()
o.startOfOperationTime, err = o.StartOfOperationTime()
st.Unlock()
if err != nil {
return fmt.Errorf("cannot get start of operation time: %s", err)
}
return o.stateEng.StartUp()
}
func (o *Overlord) ensureTimerSetup() {
o.ensureLock.Lock()
defer o.ensureLock.Unlock()
o.ensureTimer = time.NewTimer(ensureInterval)
o.ensureNext = time.Now().Add(ensureInterval)
o.pruneTicker = time.NewTicker(pruneInterval)
}
func (o *Overlord) ensureTimerReset() time.Time {
o.ensureLock.Lock()
defer o.ensureLock.Unlock()
now := time.Now()
o.ensureTimer.Reset(ensureInterval)
o.ensureNext = now.Add(ensureInterval)
return o.ensureNext
}
func (o *Overlord) ensureBefore(d time.Duration) {
o.ensureLock.Lock()
defer o.ensureLock.Unlock()
if o.ensureTimer == nil {
panic("cannot use EnsureBefore before Overlord.Loop")
}
now := time.Now()
next := now.Add(d)
if next.Before(o.ensureNext) {
o.ensureTimer.Reset(d)
o.ensureNext = next
return
}
if o.ensureNext.Before(now) {
// timer already expired, it will be reset in Loop() and
// next Ensure() will be called shortly.
if !o.ensureTimer.Stop() {
return
}
o.ensureTimer.Reset(0)
o.ensureNext = now
}
}
// Loop runs a loop in a goroutine to ensure the current state regularly through StateEngine Ensure.
func (o *Overlord) Loop() {
o.ensureTimerSetup()
o.loopTomb.Go(func() error {
for {
// TODO: pass a proper context into Ensure
o.ensureTimerReset()
// in case of errors engine logs them,
// continue to the next Ensure() try for now
o.stateEng.Ensure()
o.ensureDidRun()
pruneC := pruneTickerC(o.pruneTicker)
select {
case <-o.loopTomb.Dying():
return nil
case <-o.ensureTimer.C:
case <-pruneC:
st := o.State()
st.Lock()
st.Prune(o.startOfOperationTime, pruneWait, abortWait, pruneMaxChanges)
st.Unlock()
}
}
})
}
func (o *Overlord) ensureDidRun() {
atomic.StoreInt32(&o.ensureRun, 1)
}
func (o *Overlord) CanStandby() bool {
run := atomic.LoadInt32(&o.ensureRun)
return run != 0
}
// Stop stops the ensure loop and the managers under the StateEngine.
func (o *Overlord) Stop() error {
o.loopTomb.Kill(nil)
err := o.loopTomb.Wait()
o.stateEng.Stop()
return err
}
func (o *Overlord) settle(timeout time.Duration, beforeCleanups func()) error {
if err := o.StartUp(); err != nil {
return err
}
func() {
o.ensureLock.Lock()
defer o.ensureLock.Unlock()
if o.ensureTimer != nil {
panic("cannot use Settle concurrently with other Settle or Loop calls")
}
o.ensureTimer = time.NewTimer(0)
}()
defer func() {
o.ensureLock.Lock()
defer o.ensureLock.Unlock()
o.ensureTimer.Stop()
o.ensureTimer = nil
}()
t0 := time.Now()
done := false
var errs []error
for !done {
if timeout > 0 && time.Since(t0) > timeout {
err := fmt.Errorf("Settle is not converging")
if len(errs) != 0 {
return &ensureError{append(errs, err)}
}
return err
}
next := o.ensureTimerReset()
err := o.stateEng.Ensure()
switch ee := err.(type) {
case nil:
case *ensureError:
errs = append(errs, ee.errs...)
default:
errs = append(errs, err)
}
o.stateEng.Wait()
o.ensureLock.Lock()
done = o.ensureNext.Equal(next)
o.ensureLock.Unlock()
if done {
if beforeCleanups != nil {
beforeCleanups()
beforeCleanups = nil
}
// we should wait also for cleanup handlers
st := o.State()
st.Lock()
for _, chg := range st.Changes() {
if chg.IsReady() && !chg.IsClean() {
done = false
break
}
}
st.Unlock()
}
}
if len(errs) != 0 {
return &ensureError{errs}
}
return nil
}
// Settle runs first a state engine Ensure and then wait for
// activities to settle. That's done by waiting for all managers'
// activities to settle while making sure no immediate further Ensure
// is scheduled. It then waits similarly for all ready changes to
// reach the clean state. Chiefly for tests. Cannot be used in
// conjunction with Loop. If timeout is non-zero and settling takes
// longer than timeout, returns an error. Calls StartUp as well.
func (o *Overlord) Settle(timeout time.Duration) error {
return o.settle(timeout, nil)
}
// SettleObserveBeforeCleanups runs first a state engine Ensure and
// then wait for activities to settle. That's done by waiting for all
// managers' activities to settle while making sure no immediate
// further Ensure is scheduled. It then waits similarly for all ready
// changes to reach the clean state, but calls once the provided
// callback before doing that. Chiefly for tests. Cannot be used in
// conjunction with Loop. If timeout is non-zero and settling takes
// longer than timeout, returns an error. Calls StartUp as well.
func (o *Overlord) SettleObserveBeforeCleanups(timeout time.Duration, beforeCleanups func()) error {
return o.settle(timeout, beforeCleanups)
}
// State returns the system state managed by the overlord.
func (o *Overlord) State() *state.State {
return o.stateEng.State()
}
// StateEngine returns the state engine used by overlord.
func (o *Overlord) StateEngine() *StateEngine {
return o.stateEng
}
// TaskRunner returns the shared task runner responsible for running
// tasks for all managers under the overlord.
func (o *Overlord) TaskRunner() *state.TaskRunner {
return o.runner
}
// ServiceManager returns the service manager responsible for services
// under the overlord.
func (o *Overlord) ServiceManager() *servstate.ServiceManager {
return o.serviceMgr
}
// CommandManager returns the command manager responsible for executing
// commands under the overlord.
func (o *Overlord) CommandManager() *cmdstate.CommandManager {
return o.commandMgr
}
// CheckManager returns the check manager responsible for running health
// checks under the overlord.
func (o *Overlord) CheckManager() *checkstate.CheckManager {
return o.checkMgr
}
// PlanManager returns the plan manager responsible for managing the global
// system configuration
func (o *Overlord) PlanManager() *planstate.PlanManager {
return o.planMgr
}
// Fake creates an Overlord without any managers and with a backend
// not using disk. Managers can be added with AddManager. For testing.
func Fake() *Overlord {
return FakeWithState(nil)
}
// FakeWithState creates an Overlord without any managers and
// with a backend not using disk. Managers can be added with AddManager. For
// testing.
func FakeWithState(handleRestart func(restart.RestartType)) *Overlord {
o := &Overlord{
loopTomb: new(tomb.Tomb),
inited: false,
}
s := state.New(fakeBackend{o: o})
o.stateEng = NewStateEngine(s)
o.runner = state.NewTaskRunner(s)
return o
}
// AddManager adds a manager to a fake overlord. It cannot be used for
// a normally initialized overlord those are already fully populated.
func (o *Overlord) AddManager(mgr StateManager) {
if o.inited {
panic("internal error: cannot add managers to a fully initialized Overlord")
}
o.stateEng.AddManager(mgr)
}
var timeNow = time.Now
func (m *Overlord) StartOfOperationTime() (time.Time, error) {
var opTime time.Time
err := m.State().Get("start-of-operation-time", &opTime)
if err == nil {
return opTime, nil
}
if err != nil && !errors.Is(err, state.ErrNoState) {
return opTime, err
}
opTime = timeNow()
m.State().Set("start-of-operation-time", opTime)
return opTime, nil
}
type fakeBackend struct {
o *Overlord
}
func (mb fakeBackend) Checkpoint(data []byte) error {
return nil
}
func (mb fakeBackend) EnsureBefore(d time.Duration) {
mb.o.ensureLock.Lock()
timer := mb.o.ensureTimer
mb.o.ensureLock.Unlock()
if timer == nil {
return
}
mb.o.ensureBefore(d)
}
func (mb fakeBackend) RequestRestart(t restart.RestartType) {
panic("SHOULD NOT BE REACHED")
}