forked from Snapchat/KeyDB
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathdefrag.cpp
More file actions
1258 lines (1154 loc) · 46.3 KB
/
defrag.cpp
File metadata and controls
1258 lines (1154 loc) · 46.3 KB
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
/*
* Active memory defragmentation
* Try to find key / value allocations that need to be re-allocated in order
* to reduce external fragmentation.
* We do that by scanning the keyspace and for each pointer we have, we can try to
* ask the allocator if moving it to a new address will help reduce fragmentation.
*
* Copyright (c) 2020, Redis Labs, Inc
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* * Redistributions of source code must retain the above copyright notice,
* this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of Redis nor the names of its contributors may be used
* to endorse or promote products derived from this software without
* specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
* LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*/
#include "server.h"
#include <time.h>
#include <assert.h>
#include <stddef.h>
#ifdef HAVE_DEFRAG
/* this method was added to jemalloc in order to help us understand which
* pointers are worthwhile moving and which aren't */
extern "C" int je_get_defrag_hint(void* ptr);
/* forward declarations*/
void defragDictBucketCallback(void *privdata, dictEntry **bucketref);
dictEntry* replaceSatelliteDictKeyPtrAndOrDefragDictEntry(dict *d, sds oldkey, sds newkey, uint64_t hash, long *defragged);
bool replaceSatelliteOSetKeyPtr(expireset &set, sds oldkey, sds newkey);
/* Defrag helper for generic allocations.
*
* returns NULL in case the allocation wasn't moved.
* when it returns a non-null value, the old pointer was already released
* and should NOT be accessed. */
template<typename TPTR>
TPTR* activeDefragAlloc(TPTR *ptr) {
size_t size;
void *newptr;
if(!je_get_defrag_hint(ptr)) {
g_pserver->stat_active_defrag_misses++;
return NULL;
}
/* move this allocation to a new allocation.
* make sure not to use the thread cache. so that we don't get back the same
* pointers we try to free */
size = zmalloc_size(ptr);
newptr = zmalloc_no_tcache(size);
memcpy(newptr, ptr, size);
zfree_no_tcache(ptr);
return (TPTR*)newptr;
}
template<>
robj* activeDefragAlloc(robj *o) {
void *pvSrc = allocPtrFromObj(o);
void *pvDst = activeDefragAlloc(pvSrc);
return objFromAllocPtr(pvDst);
}
void* activeDefragAlloc(void *ptr) {
size_t size;
void *newptr;
if(!je_get_defrag_hint(ptr)) {
g_pserver->stat_active_defrag_misses++;
return NULL;
}
/* move this allocation to a new allocation.
* make sure not to use the thread cache. so that we don't get back the same
* pointers we try to free */
size = zmalloc_size(ptr);
newptr = zmalloc_no_tcache(size);
memcpy(newptr, ptr, size);
zfree_no_tcache(ptr);
return newptr;
}
/*Defrag helper for sds strings
*
* returns NULL in case the allocation wasn't moved.
* when it returns a non-null value, the old pointer was already released
* and should NOT be accessed. */
sds activeDefragSds(sds sdsptr) {
void* ptr = sdsAllocPtr(sdsptr);
void* newptr = activeDefragAlloc(ptr);
if (newptr) {
size_t offset = sdsptr - (char*)ptr;
sdsptr = (char*)newptr + offset;
return sdsptr;
}
return NULL;
}
/* Defrag helper for robj and/or string objects
*
* returns NULL in case the allocation wasn't moved.
* when it returns a non-null value, the old pointer was already released
* and should NOT be accessed. */
robj *activeDefragStringOb(robj* ob, long *defragged) {
robj *ret = NULL;
if (ob->getrefcount(std::memory_order_relaxed)!=1)
return NULL;
/* try to defrag robj (only if not an EMBSTR type (handled below). */
if (ob->type!=OBJ_STRING || ob->encoding!=OBJ_ENCODING_EMBSTR) {
if ((ret = (robj*)activeDefragAlloc(ob))) {
ob = ret;
(*defragged)++;
}
}
/* try to defrag string object */
if (ob->type == OBJ_STRING) {
if(ob->encoding==OBJ_ENCODING_RAW) {
sds newsds = activeDefragSds((sds)ptrFromObj(ob));
if (newsds) {
ob->m_ptr = newsds;
(*defragged)++;
}
} else if (ob->encoding==OBJ_ENCODING_EMBSTR) {
/* The sds is embedded in the object allocation, calculate the
* offset and update the pointer in the new allocation. */
if ((ret = (robj*)activeDefragAlloc(ob))) {
(*defragged)++;
}
} else if (ob->encoding!=OBJ_ENCODING_INT) {
serverPanic("Unknown string encoding");
}
}
return ret;
}
/* Defrag helper for dictEntries to be used during dict iteration (called on
* each step). Returns a stat of how many pointers were moved. */
long dictIterDefragEntry(dictIterator *iter) {
/* This function is a little bit dirty since it messes with the internals
* of the dict and it's iterator, but the benefit is that it is very easy
* to use, and require no other changes in the dict. */
long defragged = 0;
dictht *ht;
/* Handle the next entry (if there is one), and update the pointer in the
* current entry. */
if (iter->nextEntry) {
dictEntry *newde = (dictEntry*)activeDefragAlloc(iter->nextEntry);
if (newde) {
defragged++;
iter->nextEntry = newde;
iter->entry->next = newde;
}
}
/* handle the case of the first entry in the hash bucket. */
ht = &iter->d->ht[iter->table];
if (ht->table[iter->index] == iter->entry) {
dictEntry *newde = (dictEntry*)activeDefragAlloc(iter->entry);
if (newde) {
iter->entry = newde;
ht->table[iter->index] = newde;
defragged++;
}
}
return defragged;
}
/* Defrag helper for dict main allocations (dict struct, and hash tables).
* receives a pointer to the dict* and implicitly updates it when the dict
* struct itself was moved. Returns a stat of how many pointers were moved. */
long dictDefragTables(dict* d) {
dictEntry **newtable;
long defragged = 0;
/* handle the first hash table */
newtable = (dictEntry**)activeDefragAlloc(d->ht[0].table);
if (newtable)
defragged++, d->ht[0].table = newtable;
/* handle the second hash table */
if (d->ht[1].table) {
newtable = (dictEntry**)activeDefragAlloc(d->ht[1].table);
if (newtable)
defragged++, d->ht[1].table = newtable;
}
return defragged;
}
/* Internal function used by zslDefrag */
void zslUpdateNode(zskiplist *zsl, zskiplistNode *oldnode, zskiplistNode *newnode, zskiplistNode **update) {
int i;
for (i = 0; i < zsl->level; i++) {
if (update[i]->level(i)->forward == oldnode)
update[i]->level(i)->forward = newnode;
}
serverAssert(zsl->header!=oldnode);
if (newnode->level(0)->forward) {
serverAssert(newnode->level(0)->forward->backward==oldnode);
newnode->level(0)->forward->backward = newnode;
} else {
serverAssert(zsl->tail==oldnode);
zsl->tail = newnode;
}
}
/* Defrag helper for sorted set.
* Update the robj pointer, defrag the skiplist struct and return the new score
* reference. We may not access oldele pointer (not even the pointer stored in
* the skiplist), as it was already freed. Newele may be null, in which case we
* only need to defrag the skiplist, but not update the obj pointer.
* When return value is non-NULL, it is the score reference that must be updated
* in the dict record. */
double *zslDefrag(zskiplist *zsl, double score, sds oldele, sds newele) {
zskiplistNode *update[ZSKIPLIST_MAXLEVEL], *x, *newx;
int i;
sds ele = newele? newele: oldele;
/* find the skiplist node referring to the object that was moved,
* and all pointers that need to be updated if we'll end up moving the skiplist node. */
x = zsl->header;
for (i = zsl->level-1; i >= 0; i--) {
while (x->level(i)->forward &&
x->level(i)->forward->ele != oldele && /* make sure not to access the
->obj pointer if it matches
oldele */
(x->level(i)->forward->score < score ||
(x->level(i)->forward->score == score &&
sdscmp(x->level(i)->forward->ele,ele) < 0)))
x = x->level(i)->forward;
update[i] = x;
}
/* update the robj pointer inside the skip list record. */
x = x->level(0)->forward;
serverAssert(x && score == x->score && x->ele==oldele);
if (newele)
x->ele = newele;
/* try to defrag the skiplist record itself */
newx = (zskiplistNode*)activeDefragAlloc(x);
if (newx) {
zslUpdateNode(zsl, x, newx, update);
return &newx->score;
}
return NULL;
}
/* Defrag helper for sorted set.
* Defrag a single dict entry key name, and corresponding skiplist struct */
long activeDefragZsetEntry(zset *zs, dictEntry *de) {
sds newsds;
double* newscore;
long defragged = 0;
sds sdsele = (sds)dictGetKey(de);
if ((newsds = activeDefragSds(sdsele)))
defragged++, de->key = newsds;
newscore = zslDefrag(zs->zsl, *(double*)dictGetVal(de), sdsele, newsds);
if (newscore) {
dictSetVal(zs->dict, de, newscore);
defragged++;
}
return defragged;
}
#define DEFRAG_SDS_DICT_NO_VAL 0
#define DEFRAG_SDS_DICT_VAL_IS_SDS 1
#define DEFRAG_SDS_DICT_VAL_IS_STROB 2
#define DEFRAG_SDS_DICT_VAL_VOID_PTR 3
/* Defrag a dict with sds key and optional value (either ptr, sds or robj string) */
long activeDefragSdsDict(dict* d, int val_type) {
dictIterator *di;
dictEntry *de;
long defragged = 0;
di = dictGetIterator(d);
while((de = dictNext(di)) != NULL) {
sds sdsele = (sds)dictGetKey(de), newsds;
if ((newsds = activeDefragSds(sdsele)))
de->key = newsds, defragged++;
/* defrag the value */
if (val_type == DEFRAG_SDS_DICT_VAL_IS_SDS) {
sdsele = (sds)dictGetVal(de);
if ((newsds = activeDefragSds(sdsele)))
de->v.val = newsds, defragged++;
} else if (val_type == DEFRAG_SDS_DICT_VAL_IS_STROB) {
robj *newele, *ele = (robj*)dictGetVal(de);
if ((newele = activeDefragStringOb(ele, &defragged)))
de->v.val = newele;
} else if (val_type == DEFRAG_SDS_DICT_VAL_VOID_PTR) {
void *newptr, *ptr = dictGetVal(de);
if ((newptr = activeDefragAlloc(ptr)))
de->v.val = newptr, defragged++;
}
defragged += dictIterDefragEntry(di);
}
dictReleaseIterator(di);
return defragged;
}
/* Defrag a list of ptr, sds or robj string values */
long activeDefragList(list *l, int val_type) {
long defragged = 0;
listNode *ln, *newln;
for (ln = l->head; ln; ln = ln->next) {
if ((newln = (listNode*)activeDefragAlloc(ln))) {
if (newln->prev)
newln->prev->next = newln;
else
l->head = newln;
if (newln->next)
newln->next->prev = newln;
else
l->tail = newln;
ln = newln;
defragged++;
}
if (val_type == DEFRAG_SDS_DICT_VAL_IS_SDS) {
sds newsds, sdsele = (sds)ln->value;
if ((newsds = activeDefragSds(sdsele)))
ln->value = newsds, defragged++;
} else if (val_type == DEFRAG_SDS_DICT_VAL_IS_STROB) {
robj *newele, *ele = (robj*)ln->value;
if ((newele = activeDefragStringOb(ele, &defragged)))
ln->value = newele;
} else if (val_type == DEFRAG_SDS_DICT_VAL_VOID_PTR) {
void *newptr, *ptr = ln->value;
if ((newptr = activeDefragAlloc(ptr)))
ln->value = newptr, defragged++;
}
}
return defragged;
}
/* Defrag a list of sds values and a dict with the same sds keys */
long activeDefragSdsListAndDict(list *l, dict *d, int dict_val_type) {
long defragged = 0;
sds newsds, sdsele;
listNode *ln, *newln;
dictIterator *di;
dictEntry *de;
/* Defrag the list and it's sds values */
for (ln = l->head; ln; ln = ln->next) {
if ((newln = (listNode*)activeDefragAlloc(ln))) {
if (newln->prev)
newln->prev->next = newln;
else
l->head = newln;
if (newln->next)
newln->next->prev = newln;
else
l->tail = newln;
ln = newln;
defragged++;
}
sdsele = (sds)ln->value;
if ((newsds = activeDefragSds(sdsele))) {
/* When defragging an sds value, we need to update the dict key */
uint64_t hash = dictGetHash(d, newsds);
dictEntry **deref = dictFindEntryRefByPtrAndHash(d, sdsele, hash);
if (deref)
(*deref)->key = newsds;
ln->value = newsds;
defragged++;
}
}
/* Defrag the dict values (keys were already handled) */
di = dictGetIterator(d);
while((de = dictNext(di)) != NULL) {
if (dict_val_type == DEFRAG_SDS_DICT_VAL_IS_SDS) {
sds newsds, sdsele = (sds)dictGetVal(de);
if ((newsds = activeDefragSds(sdsele)))
de->v.val = newsds, defragged++;
} else if (dict_val_type == DEFRAG_SDS_DICT_VAL_IS_STROB) {
robj *newele, *ele = (robj*)dictGetVal(de);
if ((newele = activeDefragStringOb(ele, &defragged)))
de->v.val = newele, defragged++;
} else if (dict_val_type == DEFRAG_SDS_DICT_VAL_VOID_PTR) {
void *newptr, *ptr = dictGetVal(de);
if ((newptr = activeDefragAlloc(ptr)))
de->v.val = newptr, defragged++;
}
defragged += dictIterDefragEntry(di);
}
dictReleaseIterator(di);
return defragged;
}
/* Utility function that replaces an old key pointer in the dictionary with a
* new pointer. Additionally, we try to defrag the dictEntry in that dict.
* Oldkey mey be a dead pointer and should not be accessed (we get a
* pre-calculated hash value). Newkey may be null if the key pointer wasn't
* moved. Return value is the the dictEntry if found, or NULL if not found.
* NOTE: this is very ugly code, but it let's us avoid the complication of
* doing a scan on another dict. */
dictEntry* replaceSatelliteDictKeyPtrAndOrDefragDictEntry(dict *d, sds oldkey, sds newkey, uint64_t hash, long *defragged) {
dictEntry **deref = dictFindEntryRefByPtrAndHash(d, oldkey, hash);
if (deref) {
dictEntry *de = *deref;
dictEntry *newde = (dictEntry*)activeDefragAlloc(de);
if (newde) {
de = *deref = newde;
(*defragged)++;
}
if (newkey)
de->key = newkey;
return de;
}
return NULL;
}
bool replaceSatelliteOSetKeyPtr(expireset &set, sds oldkey, sds newkey) {
auto itr = set.find(oldkey);
if (itr != set.end())
{
expireEntry eNew(std::move(*itr));
eNew.setKeyUnsafe(newkey);
set.erase(itr);
set.insert(eNew);
serverAssert(set.find(newkey) != set.end());
return true;
}
return false;
}
long activeDefragQuickListNode(quicklist *ql, quicklistNode **node_ref) {
quicklistNode *newnode, *node = *node_ref;
long defragged = 0;
unsigned char *newzl;
if ((newnode = (quicklistNode*)activeDefragAlloc(node))) {
if (newnode->prev)
newnode->prev->next = newnode;
else
ql->head = newnode;
if (newnode->next)
newnode->next->prev = newnode;
else
ql->tail = newnode;
*node_ref = node = newnode;
defragged++;
}
if ((newzl = (unsigned char*)activeDefragAlloc(node->zl)))
defragged++, node->zl = newzl;
return defragged;
}
long activeDefragQuickListNodes(quicklist *ql) {
quicklistNode *node = ql->head;
long defragged = 0;
while (node) {
defragged += activeDefragQuickListNode(ql, &node);
node = node->next;
}
return defragged;
}
/* when the value has lots of elements, we want to handle it later and not as
* part of the main dictionary scan. this is needed in order to prevent latency
* spikes when handling large items */
void defragLater(redisDb *db, dictEntry *kde) {
sds key = sdsdup((sds)dictGetKey(kde));
listAddNodeTail(db->defrag_later, key);
}
/* returns 0 if no more work needs to be been done, and 1 if time is up and more work is needed. */
long scanLaterList(robj *ob, unsigned long *cursor, long long endtime, long long *defragged) {
quicklist *ql = (quicklist*)ptrFromObj(ob);
quicklistNode *node;
long iterations = 0;
int bookmark_failed = 0;
if (ob->type != OBJ_LIST || ob->encoding != OBJ_ENCODING_QUICKLIST)
return 0;
if (*cursor == 0) {
/* if cursor is 0, we start new iteration */
node = ql->head;
} else {
node = quicklistBookmarkFind(ql, "_AD");
if (!node) {
/* if the bookmark was deleted, it means we reached the end. */
*cursor = 0;
return 0;
}
node = node->next;
}
(*cursor)++;
while (node) {
(*defragged) += activeDefragQuickListNode(ql, &node);
g_pserver->stat_active_defrag_scanned++;
if (++iterations > 128 && !bookmark_failed) {
if (ustime() > endtime) {
if (!quicklistBookmarkCreate(&ql, "_AD", node)) {
bookmark_failed = 1;
} else {
ob->m_ptr = ql; /* bookmark creation may have re-allocated the quicklist */
return 1;
}
}
iterations = 0;
}
node = node->next;
}
quicklistBookmarkDelete(ql, "_AD");
*cursor = 0;
return bookmark_failed? 1: 0;
}
typedef struct {
zset *zs;
long defragged;
} scanLaterZsetData;
void scanLaterZsetCallback(void *privdata, const dictEntry *_de) {
dictEntry *de = (dictEntry*)_de;
scanLaterZsetData *data = (scanLaterZsetData*)privdata;
data->defragged += activeDefragZsetEntry(data->zs, de);
g_pserver->stat_active_defrag_scanned++;
}
long scanLaterZset(robj *ob, unsigned long *cursor) {
if (ob->type != OBJ_ZSET || ob->encoding != OBJ_ENCODING_SKIPLIST)
return 0;
zset *zs = (zset*)ptrFromObj(ob);
dict *d = zs->dict;
scanLaterZsetData data = {zs, 0};
*cursor = dictScan(d, *cursor, scanLaterZsetCallback, defragDictBucketCallback, &data);
return data.defragged;
}
void scanLaterSetCallback(void *privdata, const dictEntry *_de) {
dictEntry *de = (dictEntry*)_de;
long *defragged = (long*)privdata;
sds sdsele = (sds)dictGetKey(de), newsds;
if ((newsds = activeDefragSds(sdsele)))
(*defragged)++, de->key = newsds;
g_pserver->stat_active_defrag_scanned++;
}
long scanLaterSet(robj *ob, unsigned long *cursor) {
long defragged = 0;
if (ob->type != OBJ_SET || ob->encoding != OBJ_ENCODING_HT)
return 0;
dict *d = (dict*)ptrFromObj(ob);
*cursor = dictScan(d, *cursor, scanLaterSetCallback, defragDictBucketCallback, &defragged);
return defragged;
}
void scanLaterHashCallback(void *privdata, const dictEntry *_de) {
dictEntry *de = (dictEntry*)_de;
long *defragged = (long*)privdata;
sds sdsele = (sds)dictGetKey(de), newsds;
if ((newsds = activeDefragSds(sdsele)))
(*defragged)++, de->key = newsds;
sdsele = (sds)dictGetVal(de);
if ((newsds = activeDefragSds(sdsele)))
(*defragged)++, de->v.val = newsds;
g_pserver->stat_active_defrag_scanned++;
}
long scanLaterHash(robj *ob, unsigned long *cursor) {
long defragged = 0;
if (ob->type != OBJ_HASH || ob->encoding != OBJ_ENCODING_HT)
return 0;
dict *d = (dict*)ptrFromObj(ob);
*cursor = dictScan(d, *cursor, scanLaterHashCallback, defragDictBucketCallback, &defragged);
return defragged;
}
long defragQuicklist(redisDb *db, dictEntry *kde) {
robj *ob = (robj*)dictGetVal(kde);
long defragged = 0;
quicklist *ql = (quicklist*)ptrFromObj(ob), *newql;
serverAssert(ob->type == OBJ_LIST && ob->encoding == OBJ_ENCODING_QUICKLIST);
if ((newql = (quicklist*)activeDefragAlloc(ql)))
defragged++, ob->m_ptr = ql = newql;
if (ql->len > cserver.active_defrag_max_scan_fields)
defragLater(db, kde);
else
defragged += activeDefragQuickListNodes(ql);
return defragged;
}
long defragZsetSkiplist(redisDb *db, dictEntry *kde) {
robj *ob = (robj*)dictGetVal(kde);
long defragged = 0;
zset *zs = (zset*)ptrFromObj(ob);
zset *newzs;
zskiplist *newzsl;
dict *newdict;
dictEntry *de;
struct zskiplistNode *newheader;
serverAssert(ob->type == OBJ_ZSET && ob->encoding == OBJ_ENCODING_SKIPLIST);
if ((newzs = (zset*)activeDefragAlloc(zs)))
defragged++, ob->m_ptr = zs = newzs;
if ((newzsl = (zskiplist*)activeDefragAlloc(zs->zsl)))
defragged++, zs->zsl = newzsl;
if ((newheader = (zskiplistNode*)activeDefragAlloc(zs->zsl->header)))
defragged++, zs->zsl->header = newheader;
if (dictSize(zs->dict) > cserver.active_defrag_max_scan_fields)
defragLater(db, kde);
else {
dictIterator *di = dictGetIterator(zs->dict);
while((de = dictNext(di)) != NULL) {
defragged += activeDefragZsetEntry(zs, de);
}
dictReleaseIterator(di);
}
/* handle the dict struct */
if ((newdict = (dict*)activeDefragAlloc(zs->dict)))
defragged++, zs->dict = newdict;
/* defrag the dict tables */
defragged += dictDefragTables(zs->dict);
return defragged;
}
long defragHash(redisDb *db, dictEntry *kde) {
long defragged = 0;
robj *ob = (robj*)dictGetVal(kde);
dict *d, *newd;
serverAssert(ob->type == OBJ_HASH && ob->encoding == OBJ_ENCODING_HT);
d = (dict*)ptrFromObj(ob);
if (dictSize(d) > cserver.active_defrag_max_scan_fields)
defragLater(db, kde);
else
defragged += activeDefragSdsDict(d, DEFRAG_SDS_DICT_VAL_IS_SDS);
/* handle the dict struct */
if ((newd = (dict*)activeDefragAlloc(ptrFromObj(ob))))
defragged++, ob->m_ptr = newd;
/* defrag the dict tables */
defragged += dictDefragTables((dict*)ptrFromObj(ob));
return defragged;
}
long defragSet(redisDb *db, dictEntry *kde) {
long defragged = 0;
robj *ob = (robj*)dictGetVal(kde);
dict *d, *newd;
serverAssert(ob->type == OBJ_SET && ob->encoding == OBJ_ENCODING_HT);
d = (dict*)ptrFromObj(ob);
if (dictSize(d) > cserver.active_defrag_max_scan_fields)
defragLater(db, kde);
else
defragged += activeDefragSdsDict(d, DEFRAG_SDS_DICT_NO_VAL);
/* handle the dict struct */
if ((newd = (dict*)activeDefragAlloc(ptrFromObj(ob))))
defragged++, ob->m_ptr = newd;
/* defrag the dict tables */
defragged += dictDefragTables((dict*)ptrFromObj(ob));
return defragged;
}
/* Defrag callback for radix tree iterator, called for each node,
* used in order to defrag the nodes allocations. */
int defragRaxNode(raxNode **noderef) {
raxNode *newnode = (raxNode*)activeDefragAlloc(*noderef);
if (newnode) {
*noderef = newnode;
return 1;
}
return 0;
}
/* returns 0 if no more work needs to be been done, and 1 if time is up and more work is needed. */
int scanLaterStreamListpacks(robj *ob, unsigned long *cursor, long long endtime, long long *defragged) {
static unsigned char last[sizeof(streamID)];
raxIterator ri;
long iterations = 0;
if (ob->type != OBJ_STREAM || ob->encoding != OBJ_ENCODING_STREAM) {
*cursor = 0;
return 0;
}
stream *s = (stream*)ptrFromObj(ob);
raxStart(&ri,s->rax);
if (*cursor == 0) {
/* if cursor is 0, we start new iteration */
defragRaxNode(&s->rax->head);
/* assign the iterator node callback before the seek, so that the
* initial nodes that are processed till the first item are covered */
ri.node_cb = defragRaxNode;
raxSeek(&ri,"^",NULL,0);
} else {
/* if cursor is non-zero, we seek to the static 'last' */
if (!raxSeek(&ri,">", last, sizeof(last))) {
*cursor = 0;
raxStop(&ri);
return 0;
}
/* assign the iterator node callback after the seek, so that the
* initial nodes that are processed till now aren't covered */
ri.node_cb = defragRaxNode;
}
(*cursor)++;
while (raxNext(&ri)) {
void *newdata = activeDefragAlloc(ri.data);
if (newdata)
raxSetData(ri.node, ri.data=newdata), (*defragged)++;
g_pserver->stat_active_defrag_scanned++;
if (++iterations > 128) {
if (ustime() > endtime) {
serverAssert(ri.key_len==sizeof(last));
memcpy(last,ri.key,ri.key_len);
raxStop(&ri);
return 1;
}
iterations = 0;
}
}
raxStop(&ri);
*cursor = 0;
return 0;
}
/* optional callback used defrag each rax element (not including the element pointer itself) */
typedef void *(raxDefragFunction)(raxIterator *ri, void *privdata, long *defragged);
/* defrag radix tree including:
* 1) rax struct
* 2) rax nodes
* 3) rax entry data (only if defrag_data is specified)
* 4) call a callback per element, and allow the callback to return a new pointer for the element */
long defragRadixTree(rax **raxref, int defrag_data, raxDefragFunction *element_cb, void *element_cb_data) {
long defragged = 0;
raxIterator ri;
::rax* rax;
if ((rax = (::rax*)activeDefragAlloc(*raxref)))
defragged++, *raxref = rax;
rax = *raxref;
raxStart(&ri,rax);
ri.node_cb = defragRaxNode;
defragRaxNode(&rax->head);
raxSeek(&ri,"^",NULL,0);
while (raxNext(&ri)) {
void *newdata = NULL;
if (element_cb)
newdata = element_cb(&ri, element_cb_data, &defragged);
if (defrag_data && !newdata)
newdata = activeDefragAlloc(ri.data);
if (newdata)
raxSetData(ri.node, ri.data=newdata), defragged++;
}
raxStop(&ri);
return defragged;
}
typedef struct {
streamCG *cg;
streamConsumer *c;
} PendingEntryContext;
void* defragStreamConsumerPendingEntry(raxIterator *ri, void *privdata, long *defragged) {
UNUSED(defragged);
PendingEntryContext *ctx = (PendingEntryContext*)privdata;
streamNACK *nack = (streamNACK*)ri->data, *newnack;
nack->consumer = ctx->c; /* update nack pointer to consumer */
newnack = (streamNACK*)activeDefragAlloc(nack);
if (newnack) {
/* update consumer group pointer to the nack */
void *prev;
raxInsert(ctx->cg->pel, ri->key, ri->key_len, newnack, &prev);
serverAssert(prev==nack);
/* note: we don't increment 'defragged' that's done by the caller */
}
return newnack;
}
void* defragStreamConsumer(raxIterator *ri, void *privdata, long *defragged) {
streamConsumer *c = (streamConsumer*)ri->data;
streamCG *cg = (streamCG*)privdata;
void *newc = activeDefragAlloc(c);
if (newc) {
/* note: we don't increment 'defragged' that's done by the caller */
c = (streamConsumer*)newc;
}
sds newsds = activeDefragSds(c->name);
if (newsds)
(*defragged)++, c->name = newsds;
if (c->pel) {
PendingEntryContext pel_ctx = {cg, c};
*defragged += defragRadixTree(&c->pel, 0, defragStreamConsumerPendingEntry, &pel_ctx);
}
return newc; /* returns NULL if c was not defragged */
}
void* defragStreamConsumerGroup(raxIterator *ri, void *privdata, long *defragged) {
streamCG *cg = (streamCG*)ri->data;
UNUSED(privdata);
if (cg->consumers)
*defragged += defragRadixTree(&cg->consumers, 0, defragStreamConsumer, cg);
if (cg->pel)
*defragged += defragRadixTree(&cg->pel, 0, NULL, NULL);
return NULL;
}
long defragStream(redisDb *db, dictEntry *kde) {
long defragged = 0;
robj *ob = (robj*)dictGetVal(kde);
serverAssert(ob->type == OBJ_STREAM && ob->encoding == OBJ_ENCODING_STREAM);
stream *s = (stream*)ptrFromObj(ob), *news;
/* handle the main struct */
if ((news = (stream*)activeDefragAlloc(s)))
defragged++, ob->m_ptr = s = news;
if (raxSize(s->rax) > cserver.active_defrag_max_scan_fields) {
rax *newrax = (rax*)activeDefragAlloc(s->rax);
if (newrax)
defragged++, s->rax = newrax;
defragLater(db, kde);
} else
defragged += defragRadixTree(&s->rax, 1, NULL, NULL);
if (s->cgroups)
defragged += defragRadixTree(&s->cgroups, 1, defragStreamConsumerGroup, NULL);
return defragged;
}
/* Defrag a module key. This is either done immediately or scheduled
* for later. Returns then number of pointers defragged.
*/
long defragModule(redisDb *db, dictEntry *kde) {
robj *obj = (robj*)dictGetVal(kde);
serverAssert(obj->type == OBJ_MODULE);
long defragged = 0;
if (!moduleDefragValue((robj*)dictGetKey(kde), obj, &defragged))
defragLater(db, kde);
return defragged;
}
/* for each key we scan in the main dict, this function will attempt to defrag
* all the various pointers it has. Returns a stat of how many pointers were
* moved. */
long defragKey(redisDb *db, dictEntry *de) {
std::unique_lock<fastlock> ul(g_expireLock);
sds keysds = (sds)dictGetKey(de);
robj *newob, *ob;
unsigned char *newzl;
long defragged = 0;
sds newsds;
ob = (robj*)dictGetVal(de);
/* Try to defrag the key name. */
newsds = activeDefragSds(keysds);
if (newsds)
{
defragged++, de->key = newsds;
if (!db->setexpire()->empty()) {
bool fReplaced = replaceSatelliteOSetKeyPtr(*const_cast<expireset*>(db->setexpire()), keysds, newsds);
serverAssert(fReplaced == ob->FExpires());
} else {
serverAssert(!ob->FExpires());
}
}
if ((newob = activeDefragStringOb(ob, &defragged))) {
de->v.val = newob;
ob = newob;
}
if (ob->type == OBJ_STRING) {
/* Already handled in activeDefragStringOb. */
} else if (ob->type == OBJ_LIST) {
if (ob->encoding == OBJ_ENCODING_QUICKLIST) {
defragged += defragQuicklist(db, de);
} else if (ob->encoding == OBJ_ENCODING_ZIPLIST) {
if ((newzl = (unsigned char*)activeDefragAlloc(ptrFromObj(ob))))
defragged++, ob->m_ptr = newzl;
} else {
serverPanic("Unknown list encoding");
}
} else if (ob->type == OBJ_SET) {
if (ob->encoding == OBJ_ENCODING_HT) {
defragged += defragSet(db, de);
} else if (ob->encoding == OBJ_ENCODING_INTSET) {
intset *newis, *is = (intset*)ptrFromObj(ob);
if ((newis = (intset*)activeDefragAlloc(is)))
defragged++, ob->m_ptr = newis;
} else {
serverPanic("Unknown set encoding");
}
} else if (ob->type == OBJ_ZSET) {
if (ob->encoding == OBJ_ENCODING_ZIPLIST) {
if ((newzl = (unsigned char*)activeDefragAlloc(ptrFromObj(ob))))
defragged++, ob->m_ptr = newzl;
} else if (ob->encoding == OBJ_ENCODING_SKIPLIST) {
defragged += defragZsetSkiplist(db, de);
} else {
serverPanic("Unknown sorted set encoding");
}
} else if (ob->type == OBJ_HASH) {
if (ob->encoding == OBJ_ENCODING_ZIPLIST) {
if ((newzl = (unsigned char*)activeDefragAlloc(ptrFromObj(ob))))
defragged++, ob->m_ptr = newzl;
} else if (ob->encoding == OBJ_ENCODING_HT) {
defragged += defragHash(db, de);
} else {
serverPanic("Unknown hash encoding");
}
} else if (ob->type == OBJ_STREAM) {
defragged += defragStream(db, de);
} else if (ob->type == OBJ_MODULE) {
defragged += defragModule(db, de);
} else {
serverPanic("Unknown object type");
}
return defragged;
}
/* Defrag scan callback for the main db dictionary. */
void defragScanCallback(void *privdata, const dictEntry *de) {
long defragged = defragKey((redisDb*)privdata, (dictEntry*)de);
g_pserver->stat_active_defrag_hits += defragged;
if(defragged)
g_pserver->stat_active_defrag_key_hits++;
else
g_pserver->stat_active_defrag_key_misses++;
g_pserver->stat_active_defrag_scanned++;
}
/* Defrag scan callback for each hash table bucket,
* used in order to defrag the dictEntry allocations. */
void defragDictBucketCallback(void *privdata, dictEntry **bucketref) {
UNUSED(privdata); /* NOTE: this function is also used by both activeDefragCycle and scanLaterHash, etc. don't use privdata */
while(*bucketref) {
dictEntry *de = *bucketref, *newde;
if ((newde = (dictEntry*)activeDefragAlloc(de))) {
*bucketref = newde;
}
bucketref = &(*bucketref)->next;
}
}
/* Utility function to get the fragmentation ratio from jemalloc.
* It is critical to do that by comparing only heap maps that belong to
* jemalloc, and skip ones the jemalloc keeps as spare. Since we use this
* fragmentation ratio in order to decide if a defrag action should be taken
* or not, a false detection can cause the defragmenter to waste a lot of CPU
* without the possibility of getting any results. */
float getAllocatorFragmentation(size_t *out_frag_bytes) {
size_t resident, active, allocated;
zmalloc_get_allocator_info(&allocated, &active, &resident);
float frag_pct = ((float)active / allocated)*100 - 100;
size_t frag_bytes = active - allocated;
float rss_pct = ((float)resident / allocated)*100 - 100;
size_t rss_bytes = resident - allocated;
if(out_frag_bytes)
*out_frag_bytes = frag_bytes;
serverLog(LL_DEBUG,
"allocated=%zu, active=%zu, resident=%zu, frag=%.0f%% (%.0f%% rss), frag_bytes=%zu (%zu rss)",
allocated, active, resident, frag_pct, rss_pct, frag_bytes, rss_bytes);
return frag_pct;
}
/* We may need to defrag other globals, one small allocation can hold a full allocator run.
* so although small, it is still important to defrag these */
long defragOtherGlobals() {
long defragged = 0;
/* there are many more pointers to defrag (e.g. client argv, output / aof buffers, etc.
* but we assume most of these are short lived, we only need to defrag allocations
* that remain static for a long time */
defragged += activeDefragSdsDict(g_pserver->lua_scripts, DEFRAG_SDS_DICT_VAL_IS_STROB);
defragged += activeDefragSdsListAndDict(g_pserver->repl_scriptcache_fifo, g_pserver->repl_scriptcache_dict, DEFRAG_SDS_DICT_NO_VAL);
defragged += moduleDefragGlobals();
return defragged;
}
/* returns 0 more work may or may not be needed (see non-zero cursor),
* and 1 if time is up and more work is needed. */
int defragLaterItem(sds key, robj *ob, unsigned long *cursor, long long endtime) {
if (ob) {
if (ob->type == OBJ_LIST) {
return scanLaterList(ob, cursor, endtime, &g_pserver->stat_active_defrag_hits);
} else if (ob->type == OBJ_SET) {
g_pserver->stat_active_defrag_hits += scanLaterSet(ob, cursor);
} else if (ob->type == OBJ_ZSET) {
g_pserver->stat_active_defrag_hits += scanLaterZset(ob, cursor);
} else if (ob->type == OBJ_HASH) {