File: drwmutex.go

package info (click to toggle)
golang-github-minio-dsync 0.0~git20170209.0.b9f7da7-2
  • links: PTS, VCS
  • area: main
  • in suites: bullseye, buster, sid
  • size: 236 kB
  • sloc: makefile: 5
file content (448 lines) | stat: -rw-r--r-- 11,644 bytes parent folder | download
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
/*
 * Minio Cloud Storage, (C) 2016 Minio, Inc.
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

package dsync

import (
	cryptorand "crypto/rand"
	"fmt"
	golog "log"
	"math"
	"math/rand"
	"net"
	"os"
	"sync"
	"time"
)

// Indicator if logging is enabled.
var dsyncLog bool

func init() {
	// Check for DSYNC_LOG env variable, if set logging will be enabled for failed RPC operations.
	dsyncLog = os.Getenv("DSYNC_LOG") == "1"
}

func log(msg ...interface{}) {
	if dsyncLog {
		golog.Println(msg...)
	}
}

// DRWMutexAcquireTimeout - tolerance limit to wait for lock acquisition before.
const DRWMutexAcquireTimeout = 25 * time.Millisecond // 25ms.

// A DRWMutex is a distributed mutual exclusion lock.
type DRWMutex struct {
	Name         string
	writeLocks   []string   // Array of nodes that granted a write lock
	readersLocks [][]string // Array of array of nodes that granted reader locks
	m            sync.Mutex // Mutex to prevent multiple simultaneous locks from this node
}

type Granted struct {
	index   int
	lockUid string // Locked if set with UID string, unlocked if empty
}

func (g *Granted) isLocked() bool {
	return isLocked(g.lockUid)
}

func isLocked(uid string) bool {
	return len(uid) > 0
}

func NewDRWMutex(name string) *DRWMutex {
	return &DRWMutex{
		Name:       name,
		writeLocks: make([]string, dnodeCount),
	}
}

// Lock holds a write lock on dm.
//
// If the lock is already in use, the calling go routine
// blocks until the mutex is available.
func (dm *DRWMutex) Lock() {

	isReadLock := false
	dm.lockBlocking(isReadLock)
}

// RLock holds a read lock on dm.
//
// If one or more read lock are already in use, it will grant another lock.
// Otherwise the calling go routine blocks until the mutex is available.
func (dm *DRWMutex) RLock() {

	isReadLock := true
	dm.lockBlocking(isReadLock)
}

// lockBlocking will acquire either a read or a write lock
//
// The call will block until the lock is granted using a built-in
// timing randomized back-off algorithm to try again until successful
func (dm *DRWMutex) lockBlocking(isReadLock bool) {

	runs, backOff := 1, 1

	for {
		// create temp array on stack
		locks := make([]string, dnodeCount)

		// try to acquire the lock
		success := lock(clnts, &locks, dm.Name, isReadLock)
		if success {
			dm.m.Lock()
			defer dm.m.Unlock()

			// if success, copy array to object
			if isReadLock {
				// append new array of strings at the end
				dm.readersLocks = append(dm.readersLocks, make([]string, dnodeCount))
				// and copy stack array into last spot
				copy(dm.readersLocks[len(dm.readersLocks)-1], locks[:])
			} else {
				copy(dm.writeLocks, locks[:])
			}

			return
		}

		// We timed out on the previous lock, incrementally wait for a longer back-off time,
		// and try again afterwards
		time.Sleep(time.Duration(backOff) * time.Millisecond)

		backOff += int(rand.Float64() * math.Pow(2, float64(runs)))
		if backOff > 1024 {
			backOff = backOff % 64

			runs = 1 // reset runs
		} else if runs < 10 {
			runs++
		}
	}
}

// lock tries to acquire the distributed lock, returning true or false
//
func lock(clnts []NetLocker, locks *[]string, lockName string, isReadLock bool) bool {

	// Create buffered channel of size equal to total number of nodes.
	ch := make(chan Granted, dnodeCount)

	for index, c := range clnts {

		// broadcast lock request to all nodes
		go func(index int, isReadLock bool, c NetLocker) {
			// All client methods issuing RPCs are thread-safe and goroutine-safe,
			// i.e. it is safe to call them from multiple concurrently running go routines.
			bytesUid := [16]byte{}
			cryptorand.Read(bytesUid[:])
			uid := fmt.Sprintf("%X", bytesUid[:])

			args := LockArgs{
				UID:             uid,
				Resource:        lockName,
				ServerAddr:      clnts[ownNode].ServerAddr(),
				ServiceEndpoint: clnts[ownNode].ServiceEndpoint(),
			}

			var locked bool
			var err error
			if isReadLock {
				if locked, err = c.RLock(args); err != nil {
					log("Unable to call RLock", err)
				}
			} else {
				if locked, err = c.Lock(args); err != nil {
					log("Unable to call Lock", err)
				}
			}

			g := Granted{index: index}
			if locked {
				g.lockUid = args.UID
			}
			ch <- g

		}(index, isReadLock, c)
	}

	quorum := false

	var wg sync.WaitGroup
	wg.Add(1)
	go func(isReadLock bool) {

		// Wait until we have either a) received all lock responses, b) received too many 'non-'locks for quorum to be or c) time out
		i, locksFailed := 0, 0
		done := false
		timeout := time.After(DRWMutexAcquireTimeout)

		for ; i < dnodeCount; i++ { // Loop until we acquired all locks

			select {
			case grant := <-ch:
				if grant.isLocked() {
					// Mark that this node has acquired the lock
					(*locks)[grant.index] = grant.lockUid
				} else {
					locksFailed++
					if !isReadLock && locksFailed > dnodeCount-dquorum ||
						isReadLock && locksFailed > dnodeCount-dquorumReads {
						// We know that we are not going to get the lock anymore, so exit out
						// and release any locks that did get acquired
						done = true
						// Increment the number of grants received from the buffered channel.
						i++
						releaseAll(clnts, locks, lockName, isReadLock)
					}
				}

			case <-timeout:
				done = true
				// timeout happened, maybe one of the nodes is slow, count
				// number of locks to check whether we have quorum or not
				if !quorumMet(locks, isReadLock) {
					releaseAll(clnts, locks, lockName, isReadLock)
				}
			}

			if done {
				break
			}
		}

		// Count locks in order to determine whterh we have quorum or not
		quorum = quorumMet(locks, isReadLock)

		// Signal that we have the quorum
		wg.Done()

		// Wait for the other responses and immediately release the locks
		// (do not add them to the locks array because the DRWMutex could
		//  already has been unlocked again by the original calling thread)
		for ; i < dnodeCount; i++ {
			grantToBeReleased := <-ch
			if grantToBeReleased.isLocked() {
				// release lock
				sendRelease(clnts[grantToBeReleased.index], lockName, grantToBeReleased.lockUid, isReadLock)
			}
		}
	}(isReadLock)

	wg.Wait()

	// Verify that localhost server is actively participating in the lock (the lock maintenance relies on this fact)
	if quorum && !isLocked((*locks)[ownNode]) {
		// If not, release lock (and try again later)
		releaseAll(clnts, locks, lockName, isReadLock)
		quorum = false
	}

	return quorum
}

// quorumMet determines whether we have acquired the required quorum of underlying locks or not
func quorumMet(locks *[]string, isReadLock bool) bool {

	count := 0
	for _, uid := range *locks {
		if isLocked(uid) {
			count++
		}
	}

	if isReadLock {
		return count >= dquorumReads
	} else {
		return count >= dquorum
	}
}

// releaseAll releases all locks that are marked as locked
func releaseAll(clnts []NetLocker, locks *[]string, lockName string, isReadLock bool) {
	for lock := 0; lock < dnodeCount; lock++ {
		if isLocked((*locks)[lock]) {
			sendRelease(clnts[lock], lockName, (*locks)[lock], isReadLock)
			(*locks)[lock] = ""
		}
	}
}

// Unlock unlocks the write lock.
//
// It is a run-time error if dm is not locked on entry to Unlock.
func (dm *DRWMutex) Unlock() {

	// create temp array on stack
	locks := make([]string, dnodeCount)

	{
		dm.m.Lock()
		defer dm.m.Unlock()

		// Check if minimally a single bool is set in the writeLocks array
		lockFound := false
		for _, uid := range dm.writeLocks {
			if isLocked(uid) {
				lockFound = true
				break
			}
		}
		if !lockFound {
			panic("Trying to Unlock() while no Lock() is active")
		}

		// Copy write locks to stack array
		copy(locks, dm.writeLocks[:])
		// Clear write locks array
		dm.writeLocks = make([]string, dnodeCount)
	}

	isReadLock := false
	unlock(locks, dm.Name, isReadLock)
}

// RUnlock releases a read lock held on dm.
//
// It is a run-time error if dm is not locked on entry to RUnlock.
func (dm *DRWMutex) RUnlock() {

	// create temp array on stack
	locks := make([]string, dnodeCount)

	{
		dm.m.Lock()
		defer dm.m.Unlock()
		if len(dm.readersLocks) == 0 {
			panic("Trying to RUnlock() while no RLock() is active")
		}
		// Copy out first element to release it first (FIFO)
		copy(locks, dm.readersLocks[0][:])
		// Drop first element from array
		dm.readersLocks = dm.readersLocks[1:]
	}

	isReadLock := true
	unlock(locks, dm.Name, isReadLock)
}

func unlock(locks []string, name string, isReadLock bool) {

	// We don't need to synchronously wait until we have released all the locks (or the quorum)
	// (a subsequent lock will retry automatically in case it would fail to get quorum)

	for index, c := range clnts {

		if isLocked(locks[index]) {
			// broadcast lock release to all nodes that granted the lock
			sendRelease(c, name, locks[index], isReadLock)
		}
	}
}

// ForceUnlock will forcefully clear a write or read lock.
func (dm *DRWMutex) ForceUnlock() {

	{
		dm.m.Lock()
		defer dm.m.Unlock()

		// Clear write locks array
		dm.writeLocks = make([]string, dnodeCount)
		// Clear read locks array
		dm.readersLocks = nil
	}

	for _, c := range clnts {
		// broadcast lock release to all nodes that granted the lock
		sendRelease(c, dm.Name, "", false)
	}
}

// sendRelease sends a release message to a node that previously granted a lock
func sendRelease(c NetLocker, name, uid string, isReadLock bool) {

	backOffArray := []time.Duration{
		30 * time.Second, // 30 secs
		1 * time.Minute,  // 1 min
		3 * time.Minute,  // 3 min
		10 * time.Minute, // 10 min
		30 * time.Minute, // 30 min
		1 * time.Hour,    // 1 hour
	}

	go func(c NetLocker, name string) {

		for _, backOff := range backOffArray {

			// All client methods issuing RPCs are thread-safe and goroutine-safe,
			// i.e. it is safe to call them from multiple concurrently running goroutines.
			args := LockArgs{
				UID:             uid,
				Resource:        name,
				ServerAddr:      clnts[ownNode].ServerAddr(),
				ServiceEndpoint: clnts[ownNode].ServiceEndpoint(),
			}

			var err error
			if len(uid) == 0 {
				if _, err = c.ForceUnlock(args); err != nil {
					log("Unable to call ForceUnlock", err)
				}
			} else if isReadLock {
				if _, err = c.RUnlock(args); err != nil {
					log("Unable to call RUnlock", err)
				}
			} else {
				if _, err = c.Unlock(args); err != nil {
					log("Unable to call Unlock", err)
				}
			}

			if err != nil {
				// Ignore if err is net.Error and it is occurred due to timeout.
				// The cause could have been server timestamp mismatch or server may have restarted.
				// FIXME: This is minio specific behaviour and we would need a way to make it generically.
				if nErr, ok := err.(net.Error); ok && nErr.Timeout() {
					err = nil
				}
			}

			if err == nil {
				return
			}

			// Wait..
			time.Sleep(backOff)
		}
	}(c, name)
}

// DRLocker returns a sync.Locker interface that implements
// the Lock and Unlock methods by calling drw.RLock and drw.RUnlock.
func (dm *DRWMutex) DRLocker() sync.Locker {
	return (*drlocker)(dm)
}

type drlocker DRWMutex

func (dr *drlocker) Lock()   { (*DRWMutex)(dr).RLock() }
func (dr *drlocker) Unlock() { (*DRWMutex)(dr).RUnlock() }