summaryrefslogtreecommitdiffstats
path: root/modules/graceful/manager.go
blob: 8766cfca0efb410e78375f1f9aebb7e29f7859ae (plain)
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
// Copyright 2019 The Gitea Authors. All rights reserved.
// Use of this source code is governed by a MIT-style
// license that can be found in the LICENSE file.

package graceful

import (
	"context"
	"runtime/pprof"
	"sync"
	"time"

	"code.gitea.io/gitea/modules/log"
	"code.gitea.io/gitea/modules/process"
	"code.gitea.io/gitea/modules/setting"
)

type state uint8

const (
	stateInit state = iota
	stateRunning
	stateShuttingDown
	stateTerminate
)

// There are three places that could inherit sockets:
//
// * HTTP or HTTPS main listener
// * HTTP redirection fallback
// * SSH
//
// If you add an additional place you must increment this number
// and add a function to call manager.InformCleanup if it's not going to be used
const numberOfServersToCreate = 4

// Manager represents the graceful server manager interface
var manager *Manager

var initOnce = sync.Once{}

// GetManager returns the Manager
func GetManager() *Manager {
	InitManager(context.Background())
	return manager
}

// InitManager creates the graceful manager in the provided context
func InitManager(ctx context.Context) {
	initOnce.Do(func() {
		manager = newGracefulManager(ctx)

		// Set the process default context to the HammerContext
		process.DefaultContext = manager.HammerContext()
	})
}

// WithCallback is a runnable to call when the caller has finished
type WithCallback func(callback func())

// RunnableWithShutdownFns is a runnable with functions to run at shutdown and terminate
// After the callback to atShutdown is called and is complete, the main function must return.
// Similarly the callback function provided to atTerminate must return once termination is complete.
// Please note that use of the atShutdown and atTerminate callbacks will create go-routines that will wait till their respective signals
// - users must therefore be careful to only call these as necessary.
type RunnableWithShutdownFns func(atShutdown, atTerminate func(func()))

// RunWithShutdownFns takes a function that has both atShutdown and atTerminate callbacks
// After the callback to atShutdown is called and is complete, the main function must return.
// Similarly the callback function provided to atTerminate must return once termination is complete.
// Please note that use of the atShutdown and atTerminate callbacks will create go-routines that will wait till their respective signals
// - users must therefore be careful to only call these as necessary.
func (g *Manager) RunWithShutdownFns(run RunnableWithShutdownFns) {
	g.runningServerWaitGroup.Add(1)
	defer g.runningServerWaitGroup.Done()
	defer func() {
		if err := recover(); err != nil {
			log.Critical("PANIC during RunWithShutdownFns: %v\nStacktrace: %s", err, log.Stack(2))
			g.doShutdown()
		}
	}()
	run(func(atShutdown func()) {
		g.lock.Lock()
		defer g.lock.Unlock()
		g.toRunAtShutdown = append(g.toRunAtShutdown,
			func() {
				defer func() {
					if err := recover(); err != nil {
						log.Critical("PANIC during RunWithShutdownFns: %v\nStacktrace: %s", err, log.Stack(2))
						g.doShutdown()
					}
				}()
				atShutdown()
			})
	}, func(atTerminate func()) {
		g.RunAtTerminate(atTerminate)
	})
}

// RunWithShutdownContext takes a function that has a context to watch for shutdown.
// After the provided context is Done(), the main function must return once shutdown is complete.
// (Optionally the HammerContext may be obtained and waited for however, this should be avoided if possible.)
func (g *Manager) RunWithShutdownContext(run func(context.Context)) {
	g.runningServerWaitGroup.Add(1)
	defer g.runningServerWaitGroup.Done()
	defer func() {
		if err := recover(); err != nil {
			log.Critical("PANIC during RunWithShutdownContext: %v\nStacktrace: %s", err, log.Stack(2))
			g.doShutdown()
		}
	}()
	ctx := g.ShutdownContext()
	pprof.SetGoroutineLabels(ctx) // We don't have a label to restore back to but I think this is fine
	run(ctx)
}

// RunAtTerminate adds to the terminate wait group and creates a go-routine to run the provided function at termination
func (g *Manager) RunAtTerminate(terminate func()) {
	g.terminateWaitGroup.Add(1)
	g.lock.Lock()
	defer g.lock.Unlock()
	g.toRunAtTerminate = append(g.toRunAtTerminate,
		func() {
			defer g.terminateWaitGroup.Done()
			defer func() {
				if err := recover(); err != nil {
					log.Critical("PANIC during RunAtTerminate: %v\nStacktrace: %s", err, log.Stack(2))
				}
			}()
			terminate()
		})
}

// RunAtShutdown creates a go-routine to run the provided function at shutdown
func (g *Manager) RunAtShutdown(ctx context.Context, shutdown func()) {
	g.lock.Lock()
	defer g.lock.Unlock()
	g.toRunAtShutdown = append(g.toRunAtShutdown,
		func() {
			defer func() {
				if err := recover(); err != nil {
					log.Critical("PANIC during RunAtShutdown: %v\nStacktrace: %s", err, log.Stack(2))
				}
			}()
			select {
			case <-ctx.Done():
				return
			default:
				shutdown()
			}
		})
}

// RunAtHammer creates a go-routine to run the provided function at shutdown
func (g *Manager) RunAtHammer(hammer func()) {
	g.lock.Lock()
	defer g.lock.Unlock()
	g.toRunAtHammer = append(g.toRunAtHammer,
		func() {
			defer func() {
				if err := recover(); err != nil {
					log.Critical("PANIC during RunAtHammer: %v\nStacktrace: %s", err, log.Stack(2))
				}
			}()
			hammer()
		})
}

func (g *Manager) doShutdown() {
	if !g.setStateTransition(stateRunning, stateShuttingDown) {
		g.DoImmediateHammer()
		return
	}
	g.lock.Lock()
	g.shutdownCtxCancel()
	atShutdownCtx := pprof.WithLabels(g.hammerCtx, pprof.Labels("graceful-lifecycle", "post-shutdown"))
	pprof.SetGoroutineLabels(atShutdownCtx)
	for _, fn := range g.toRunAtShutdown {
		go fn()
	}
	g.lock.Unlock()

	if setting.GracefulHammerTime >= 0 {
		go g.doHammerTime(setting.GracefulHammerTime)
	}
	go func() {
		g.WaitForServers()
		// Mop up any remaining unclosed events.
		g.doHammerTime(0)
		<-time.After(1 * time.Second)
		g.doTerminate()
		g.WaitForTerminate()
		g.lock.Lock()
		g.managerCtxCancel()
		g.lock.Unlock()
	}()
}

func (g *Manager) doHammerTime(d time.Duration) {
	time.Sleep(d)
	g.lock.Lock()
	select {
	case <-g.hammerCtx.Done():
	default:
		log.Warn("Setting Hammer condition")
		g.hammerCtxCancel()
		atHammerCtx := pprof.WithLabels(g.terminateCtx, pprof.Labels("graceful-lifecycle", "post-hammer"))
		pprof.SetGoroutineLabels(atHammerCtx)
		for _, fn := range g.toRunAtHammer {
			go fn()
		}
	}
	g.lock.Unlock()
}

func (g *Manager) doTerminate() {
	if !g.setStateTransition(stateShuttingDown, stateTerminate) {
		return
	}
	g.lock.Lock()
	select {
	case <-g.terminateCtx.Done():
	default:
		log.Warn("Terminating")
		g.terminateCtxCancel()
		atTerminateCtx := pprof.WithLabels(g.managerCtx, pprof.Labels("graceful-lifecycle", "post-terminate"))
		pprof.SetGoroutineLabels(atTerminateCtx)

		for _, fn := range g.toRunAtTerminate {
			go fn()
		}
	}
	g.lock.Unlock()
}

// IsChild returns if the current process is a child of previous Gitea process
func (g *Manager) IsChild() bool {
	return g.isChild
}

// IsShutdown returns a channel which will be closed at shutdown.
// The order of closure is IsShutdown, IsHammer (potentially), IsTerminate
func (g *Manager) IsShutdown() <-chan struct{} {
	return g.shutdownCtx.Done()
}

// IsHammer returns a channel which will be closed at hammer
// The order of closure is IsShutdown, IsHammer (potentially), IsTerminate
// Servers running within the running server wait group should respond to IsHammer
// if not shutdown already
func (g *Manager) IsHammer() <-chan struct{} {
	return g.hammerCtx.Done()
}

// IsTerminate returns a channel which will be closed at terminate
// The order of closure is IsShutdown, IsHammer (potentially), IsTerminate
// IsTerminate will only close once all running servers have stopped
func (g *Manager) IsTerminate() <-chan struct{} {
	return g.terminateCtx.Done()
}

// ServerDone declares a running server done and subtracts one from the
// running server wait group. Users probably do not want to call this
// and should use one of the RunWithShutdown* functions
func (g *Manager) ServerDone() {
	g.runningServerWaitGroup.Done()
}

// WaitForServers waits for all running servers to finish. Users should probably
// instead use AtTerminate or IsTerminate
func (g *Manager) WaitForServers() {
	g.runningServerWaitGroup.Wait()
}

// WaitForTerminate waits for all terminating actions to finish.
// Only the main go-routine should use this
func (g *Manager) WaitForTerminate() {
	g.terminateWaitGroup.Wait()
}

func (g *Manager) getState() state {
	g.lock.RLock()
	defer g.lock.RUnlock()
	return g.state
}

func (g *Manager) setStateTransition(old, new state) bool {
	if old != g.getState() {
		return false
	}
	g.lock.Lock()
	if g.state != old {
		g.lock.Unlock()
		return false
	}
	g.state = new
	g.lock.Unlock()
	return true
}

func (g *Manager) setState(st state) {
	g.lock.Lock()
	defer g.lock.Unlock()

	g.state = st
}

// InformCleanup tells the cleanup wait group that we have either taken a listener
// or will not be taking a listener
func (g *Manager) InformCleanup() {
	g.createServerWaitGroup.Done()
}

// Done allows the manager to be viewed as a context.Context, it returns a channel that is closed when the server is finished terminating
func (g *Manager) Done() <-chan struct{} {
	return g.managerCtx.Done()
}

// Err allows the manager to be viewed as a context.Context done at Terminate
func (g *Manager) Err() error {
	return g.managerCtx.Err()
}

// Value allows the manager to be viewed as a context.Context done at Terminate
func (g *Manager) Value(key interface{}) interface{} {
	return g.managerCtx.Value(key)
}

// Deadline returns nil as there is no fixed Deadline for the manager, it allows the manager to be viewed as a context.Context
func (g *Manager) Deadline() (deadline time.Time, ok bool) {
	return g.managerCtx.Deadline()
}