summaryrefslogtreecommitdiff
path: root/sql/table_cache.cc
blob: 62ccfba7e7dbbee9dd7167a100859115b960ad94 (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
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
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
/* Copyright (c) 2000, 2012, Oracle and/or its affiliates.
   Copyright (c) 2010, 2011 Monty Program Ab
   Copyright (C) 2013 Sergey Vojtovich and MariaDB Foundation

   This program is free software; you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
   the Free Software Foundation; version 2 of the License.

   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, write to the Free Software
   Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1335  USA */

/**
  @file
  Table definition cache and table cache implementation.

  Table definition cache actions:
  - add new TABLE_SHARE object to cache (tdc_acquire_share())
  - acquire TABLE_SHARE object from cache (tdc_acquire_share())
  - release TABLE_SHARE object to cache (tdc_release_share())
  - purge unused TABLE_SHARE objects from cache (tdc_purge())
  - remove TABLE_SHARE object from cache (tdc_remove_table())
  - get number of TABLE_SHARE objects in cache (tdc_records())

  Table cache actions:
  - add new TABLE object to cache (tc_add_table())
  - acquire TABLE object from cache (tc_acquire_table())
  - release TABLE object to cache (tc_release_table())
  - purge unused TABLE objects from cache (tc_purge())
  - purge unused TABLE objects of a table from cache (tdc_remove_table())
  - get number of TABLE objects in cache (tc_records())

  Dependencies:
  - close_cached_tables(): flush tables on shutdown
  - alloc_table_share()
  - free_table_share()

  Table cache invariants:
  - TABLE_SHARE::free_tables shall not contain objects with TABLE::in_use != 0
  - TABLE_SHARE::free_tables shall not receive new objects if
    TABLE_SHARE::tdc.flushed is true
*/

#include "mariadb.h"
#include "lf.h"
#include "table.h"
#include "sql_base.h"


/** Configuration. */
ulong tdc_size; /**< Table definition cache threshold for LRU eviction. */
ulong tc_size; /**< Table cache threshold for LRU eviction. */
uint32 tc_instances;
static std::atomic<uint32_t> tc_active_instances(1);
static std::atomic<bool> tc_contention_warning_reported;

/** Data collections. */
static LF_HASH tdc_hash; /**< Collection of TABLE_SHARE objects. */
/** Collection of unused TABLE_SHARE objects. */
static
I_P_List <TDC_element,
          I_P_List_adapter<TDC_element, &TDC_element::next, &TDC_element::prev>,
          I_P_List_null_counter,
          I_P_List_fast_push_back<TDC_element> > unused_shares;

static bool tdc_inited;


/**
  Protects unused shares list.

  TDC_element::prev
  TDC_element::next
  unused_shares
*/

static mysql_mutex_t LOCK_unused_shares;

#ifdef HAVE_PSI_INTERFACE
static PSI_mutex_key key_LOCK_unused_shares, key_TABLE_SHARE_LOCK_table_share,
                     key_LOCK_table_cache;
static PSI_mutex_info all_tc_mutexes[]=
{
  { &key_LOCK_unused_shares, "LOCK_unused_shares", PSI_FLAG_GLOBAL },
  { &key_TABLE_SHARE_LOCK_table_share, "TABLE_SHARE::tdc.LOCK_table_share", 0 },
  { &key_LOCK_table_cache, "LOCK_table_cache", 0 }
};

static PSI_cond_key key_TABLE_SHARE_COND_release;
static PSI_cond_info all_tc_conds[]=
{
  { &key_TABLE_SHARE_COND_release, "TABLE_SHARE::tdc.COND_release", 0 }
};
#endif


static int fix_thd_pins(THD *thd)
{
  return thd->tdc_hash_pins ? 0 :
         (thd->tdc_hash_pins= lf_hash_get_pins(&tdc_hash)) == 0;
}


/*
  Auxiliary routines for manipulating with per-share all/unused lists
  and tc_count counter.
  Responsible for preserving invariants between those lists, counter
  and TABLE::in_use member.
  In fact those routines implement sort of implicit table cache as
  part of table definition cache.
*/

struct Table_cache_instance
{
  /**
    Protects free_tables (TABLE::global_free_next and TABLE::global_free_prev),
    records, Share_free_tables::List (TABLE::prev and TABLE::next),
    TABLE::in_use.
  */
  mysql_mutex_t LOCK_table_cache;
  I_P_List <TABLE, I_P_List_adapter<TABLE, &TABLE::global_free_next,
                                    &TABLE::global_free_prev>,
            I_P_List_null_counter, I_P_List_fast_push_back<TABLE> >
    free_tables;
  ulong records;
  uint mutex_waits;
  uint mutex_nowaits;
  /** Avoid false sharing between instances */
  char pad[CPU_LEVEL1_DCACHE_LINESIZE];

  Table_cache_instance(): records(0), mutex_waits(0), mutex_nowaits(0)
  {
    mysql_mutex_init(key_LOCK_table_cache, &LOCK_table_cache,
                     MY_MUTEX_INIT_FAST);
  }

  ~Table_cache_instance()
  {
    mysql_mutex_destroy(&LOCK_table_cache);
    DBUG_ASSERT(free_tables.is_empty());
    DBUG_ASSERT(records == 0);
  }

  /**
    Lock table cache mutex and check contention.

    Instance is considered contested if more than 20% of mutex acquisiotions
    can't be served immediately. Up to 100 000 probes may be performed to avoid
    instance activation on short sporadic peaks. 100 000 is estimated maximum
    number of queries one instance can serve in one second.

    These numbers work well on a 2 socket / 20 core / 40 threads Intel Broadwell
    system, that is expected number of instances is activated within reasonable
    warmup time. It may have to be adjusted for other systems.

    Only TABLE object acquistion is instrumented. We intentionally avoid this
    overhead on TABLE object release. All other table cache mutex acquistions
    are considered out of hot path and are not instrumented either.
  */
  void lock_and_check_contention(uint32_t n_instances, uint32_t instance)
  {
    if (mysql_mutex_trylock(&LOCK_table_cache))
    {
      mysql_mutex_lock(&LOCK_table_cache);
      if (++mutex_waits == 20000)
      {
        if (n_instances < tc_instances)
        {
          if (tc_active_instances.
              compare_exchange_weak(n_instances, n_instances + 1,
                                    std::memory_order_relaxed,
                                    std::memory_order_relaxed))
          {
            sql_print_information("Detected table cache mutex contention at instance %d: "
                                  "%d%% waits. Additional table cache instance "
                                  "activated. Number of instances after "
                                  "activation: %d.",
                                  instance + 1,
                                  mutex_waits * 100 / (mutex_nowaits + mutex_waits),
                                  n_instances + 1);
          }
        }
        else if (!tc_contention_warning_reported.exchange(true,
                                                 std::memory_order_relaxed))
        {
          sql_print_warning("Detected table cache mutex contention at instance %d: "
                            "%d%% waits. Additional table cache instance "
                            "cannot be activated: consider raising "
                            "table_open_cache_instances. Number of active "
                            "instances: %d.",
                            instance + 1,
                            mutex_waits * 100 / (mutex_nowaits + mutex_waits),
                            n_instances);
        }
        mutex_waits= 0;
        mutex_nowaits= 0;
      }
    }
    else if (++mutex_nowaits == 80000)
    {
      mutex_waits= 0;
      mutex_nowaits= 0;
    }
  }
};


static Table_cache_instance *tc;


static void intern_close_table(TABLE *table)
{
  delete table->triggers;
  DBUG_ASSERT(table->file);
  closefrm(table);
  tdc_release_share(table->s);
  my_free(table);
}


/**
  Get number of TABLE objects (used and unused) in table cache.
*/

uint tc_records(void)
{
  ulong total= 0;
  for (uint32 i= 0; i < tc_instances; i++)
  {
    mysql_mutex_lock(&tc[i].LOCK_table_cache);
    total+= tc[i].records;
    mysql_mutex_unlock(&tc[i].LOCK_table_cache);
  }
  return total;
}


/**
  Remove TABLE object from table cache.
*/

static void tc_remove_table(TABLE *table)
{
  TDC_element *element= table->s->tdc;

  mysql_mutex_lock(&element->LOCK_table_share);
  /* Wait for MDL deadlock detector to complete traversing tdc.all_tables. */
  while (element->all_tables_refs)
    mysql_cond_wait(&element->COND_release, &element->LOCK_table_share);
  element->all_tables.remove(table);
  mysql_mutex_unlock(&element->LOCK_table_share);

  intern_close_table(table);
}


static void tc_remove_all_unused_tables(TDC_element *element,
                                        Share_free_tables::List *purge_tables)
{
  for (uint32 i= 0; i < tc_instances; i++)
  {
    mysql_mutex_lock(&tc[i].LOCK_table_cache);
    while (auto table= element->free_tables[i].list.pop_front())
    {
      tc[i].records--;
      tc[i].free_tables.remove(table);
      DBUG_ASSERT(element->all_tables_refs == 0);
      element->all_tables.remove(table);
      purge_tables->push_front(table);
    }
    mysql_mutex_unlock(&tc[i].LOCK_table_cache);
  }
}


/**
  Free all unused TABLE objects.

  While locked:
  - remove unused objects from TABLE_SHARE::tdc.free_tables and
    TABLE_SHARE::tdc.all_tables
  - decrement tc_count

  While unlocked:
  - free resources related to unused objects

  @note This is called by 'handle_manager' when one wants to
        periodicly flush all not used tables.
*/

static my_bool tc_purge_callback(TDC_element *element,
                                 Share_free_tables::List *purge_tables)
{
  mysql_mutex_lock(&element->LOCK_table_share);
  tc_remove_all_unused_tables(element, purge_tables);
  mysql_mutex_unlock(&element->LOCK_table_share);
  return FALSE;
}


void tc_purge()
{
  Share_free_tables::List purge_tables;

  tdc_iterate(0, (my_hash_walk_action) tc_purge_callback, &purge_tables);
  while (auto table= purge_tables.pop_front())
    intern_close_table(table);
}


/**
  Add new TABLE object to table cache.

  @pre TABLE object is used by caller.

  Added object cannot be evicted or acquired.

  While locked:
  - add object to TABLE_SHARE::tdc.all_tables
  - increment tc_count
  - evict LRU object from table cache if we reached threshold

  While unlocked:
  - free evicted object
*/

void tc_add_table(THD *thd, TABLE *table)
{
  uint32_t i=
    thd->thread_id % tc_active_instances.load(std::memory_order_relaxed);
  TABLE *LRU_table= 0;
  TDC_element *element= table->s->tdc;

  DBUG_ASSERT(table->in_use == thd);
  table->instance= i;
  mysql_mutex_lock(&element->LOCK_table_share);
  /* Wait for MDL deadlock detector to complete traversing tdc.all_tables. */
  while (element->all_tables_refs)
    mysql_cond_wait(&element->COND_release, &element->LOCK_table_share);
  element->all_tables.push_front(table);
  mysql_mutex_unlock(&element->LOCK_table_share);

  mysql_mutex_lock(&tc[i].LOCK_table_cache);
  if (tc[i].records == tc_size)
  {
    if ((LRU_table= tc[i].free_tables.pop_front()))
    {
      LRU_table->s->tdc->free_tables[i].list.remove(LRU_table);
      /* Needed if MDL deadlock detector chimes in before tc_remove_table() */
      LRU_table->in_use= thd;
      mysql_mutex_unlock(&tc[i].LOCK_table_cache);
      /* Keep out of locked LOCK_table_cache */
      tc_remove_table(LRU_table);
    }
    else
    {
      tc[i].records++;
      mysql_mutex_unlock(&tc[i].LOCK_table_cache);
    }
    /* Keep out of locked LOCK_table_cache */
    status_var_increment(thd->status_var.table_open_cache_overflows);
  }
  else
  {
    tc[i].records++;
    mysql_mutex_unlock(&tc[i].LOCK_table_cache);
  }
}


/**
  Acquire TABLE object from table cache.

  @pre share must be protected against removal.

  Acquired object cannot be evicted or acquired again.

  @return TABLE object, or NULL if no unused objects.
*/

TABLE *tc_acquire_table(THD *thd, TDC_element *element)
{
  uint32_t n_instances= tc_active_instances.load(std::memory_order_relaxed);
  uint32_t i= thd->thread_id % n_instances;
  TABLE *table;

  tc[i].lock_and_check_contention(n_instances, i);
  table= element->free_tables[i].list.pop_front();
  if (table)
  {
    DBUG_ASSERT(!table->in_use);
    table->in_use= thd;
    /* The ex-unused table must be fully functional. */
    DBUG_ASSERT(table->db_stat && table->file);
    /* The children must be detached from the table. */
    DBUG_ASSERT(!table->file->extra(HA_EXTRA_IS_ATTACHED_CHILDREN));
    tc[i].free_tables.remove(table);
  }
  mysql_mutex_unlock(&tc[i].LOCK_table_cache);
  return table;
}


/**
  Release TABLE object to table cache.

  @pre object is used by caller.

  Released object may be evicted or acquired again.

  While locked:
  - if object is marked for purge, decrement tc_count
  - add object to TABLE_SHARE::tdc.free_tables
  - evict LRU object from table cache if we reached threshold

  While unlocked:
  - mark object not in use by any thread
  - free evicted/purged object

  @note Another thread may mark share for purge any moment (even
  after version check). It means to-be-purged object may go to
  unused lists. This other thread is expected to call tc_purge(),
  which is synchronized with us on TABLE_SHARE::tdc.LOCK_table_share.

  @return
    @retval true  object purged
    @retval false object released
*/

void tc_release_table(TABLE *table)
{
  uint32 i= table->instance;
  DBUG_ENTER("tc_release_table");
  DBUG_ASSERT(table->in_use);
  DBUG_ASSERT(table->file);
  DBUG_ASSERT(!table->pos_in_locked_tables);

  mysql_mutex_lock(&tc[i].LOCK_table_cache);
  if (table->needs_reopen() || table->s->tdc->flushed ||
      tc[i].records > tc_size)
  {
    tc[i].records--;
    mysql_mutex_unlock(&tc[i].LOCK_table_cache);
    tc_remove_table(table);
  }
  else
  {
    table->in_use= 0;
    table->s->tdc->free_tables[i].list.push_front(table);
    tc[i].free_tables.push_back(table);
    mysql_mutex_unlock(&tc[i].LOCK_table_cache);
  }
  DBUG_VOID_RETURN;
}


static void tdc_assert_clean_share(TDC_element *element)
{
  DBUG_ASSERT(element->share == 0);
  DBUG_ASSERT(element->ref_count == 0);
  DBUG_ASSERT(element->m_flush_tickets.is_empty());
  DBUG_ASSERT(element->all_tables.is_empty());
#ifndef DBUG_OFF
  for (uint32 i= 0; i < tc_instances; i++)
    DBUG_ASSERT(element->free_tables[i].list.is_empty());
#endif
  DBUG_ASSERT(element->all_tables_refs == 0);
  DBUG_ASSERT(element->next == 0);
  DBUG_ASSERT(element->prev == 0);
}


/**
  Delete share from hash and free share object.
*/

static void tdc_delete_share_from_hash(TDC_element *element)
{
  THD *thd= current_thd;
  LF_PINS *pins;
  TABLE_SHARE *share;
  DBUG_ENTER("tdc_delete_share_from_hash");

  mysql_mutex_assert_owner(&element->LOCK_table_share);
  share= element->share;
  DBUG_ASSERT(share);
  element->share= 0;
  PSI_CALL_release_table_share(share->m_psi);
  share->m_psi= 0;

  if (!element->m_flush_tickets.is_empty())
  {
    Wait_for_flush_list::Iterator it(element->m_flush_tickets);
    Wait_for_flush *ticket;
    while ((ticket= it++))
      (void) ticket->get_ctx()->m_wait.set_status(MDL_wait::GRANTED);

    do
    {
      mysql_cond_wait(&element->COND_release, &element->LOCK_table_share);
    } while (!element->m_flush_tickets.is_empty());
  }

  mysql_mutex_unlock(&element->LOCK_table_share);

  if (thd)
  {
    fix_thd_pins(thd);
    pins= thd->tdc_hash_pins;
  }
  else
    pins= lf_hash_get_pins(&tdc_hash);

  DBUG_ASSERT(pins); // What can we do about it?
  tdc_assert_clean_share(element);
  lf_hash_delete(&tdc_hash, pins, element->m_key, element->m_key_length);
  if (!thd)
    lf_hash_put_pins(pins);
  free_table_share(share);
  DBUG_VOID_RETURN;
}


/**
  Prepeare table share for use with table definition cache.
*/

static void lf_alloc_constructor(uchar *arg)
{
  TDC_element *element= (TDC_element*) (arg + LF_HASH_OVERHEAD);
  DBUG_ENTER("lf_alloc_constructor");
  mysql_mutex_init(key_TABLE_SHARE_LOCK_table_share,
                   &element->LOCK_table_share, MY_MUTEX_INIT_FAST);
  mysql_cond_init(key_TABLE_SHARE_COND_release, &element->COND_release, 0);
  element->m_flush_tickets.empty();
  element->all_tables.empty();
  for (uint32 i= 0; i < tc_instances; i++)
    element->free_tables[i].list.empty();
  element->all_tables_refs= 0;
  element->share= 0;
  element->ref_count= 0;
  element->next= 0;
  element->prev= 0;
  DBUG_VOID_RETURN;
}


/**
  Release table definition cache specific resources of table share.
*/

static void lf_alloc_destructor(uchar *arg)
{
  TDC_element *element= (TDC_element*) (arg + LF_HASH_OVERHEAD);
  DBUG_ENTER("lf_alloc_destructor");
  tdc_assert_clean_share(element);
  mysql_cond_destroy(&element->COND_release);
  mysql_mutex_destroy(&element->LOCK_table_share);
  DBUG_VOID_RETURN;
}


static void tdc_hash_initializer(LF_HASH *,
                                 TDC_element *element, LEX_STRING *key)
{
  memcpy(element->m_key, key->str, key->length);
  element->m_key_length= (uint)key->length;
  tdc_assert_clean_share(element);
}


static uchar *tdc_hash_key(const TDC_element *element, size_t *length,
                           my_bool)
{
  *length= element->m_key_length;
  return (uchar*) element->m_key;
}


/**
  Initialize table definition cache.
*/

bool tdc_init(void)
{
  DBUG_ENTER("tdc_init");
#ifdef HAVE_PSI_INTERFACE
  mysql_mutex_register("sql", all_tc_mutexes, array_elements(all_tc_mutexes));
  mysql_cond_register("sql", all_tc_conds, array_elements(all_tc_conds));
#endif
  /* Extra instance is allocated to avoid false sharing */
  if (!(tc= new Table_cache_instance[tc_instances + 1]))
    DBUG_RETURN(true);
  tdc_inited= true;
  mysql_mutex_init(key_LOCK_unused_shares, &LOCK_unused_shares,
                   MY_MUTEX_INIT_FAST);
  lf_hash_init(&tdc_hash, sizeof(TDC_element) +
                          sizeof(Share_free_tables) * (tc_instances - 1),
               LF_HASH_UNIQUE, 0, 0,
               (my_hash_get_key) tdc_hash_key,
               &my_charset_bin);
  tdc_hash.alloc.constructor= lf_alloc_constructor;
  tdc_hash.alloc.destructor= lf_alloc_destructor;
  tdc_hash.initializer= (lf_hash_initializer) tdc_hash_initializer;
  DBUG_RETURN(false);
}


/**
  Notify table definition cache that process of shutting down server
  has started so it has to keep number of TABLE and TABLE_SHARE objects
  minimal in order to reduce number of references to pluggable engines.
*/

void tdc_start_shutdown(void)
{
  DBUG_ENTER("tdc_start_shutdown");
  if (tdc_inited)
  {
    /*
      Ensure that TABLE and TABLE_SHARE objects which are created for
      tables that are open during process of plugins' shutdown are
      immediately released. This keeps number of references to engine
      plugins minimal and allows shutdown to proceed smoothly.
    */
    tdc_size= 0;
    tc_size= 0;
    /* Free all cached but unused TABLEs and TABLE_SHAREs. */
    purge_tables();
  }
  DBUG_VOID_RETURN;
}


/**
  Deinitialize table definition cache.
*/

void tdc_deinit(void)
{
  DBUG_ENTER("tdc_deinit");
  if (tdc_inited)
  {
    tdc_inited= false;
    lf_hash_destroy(&tdc_hash);
    mysql_mutex_destroy(&LOCK_unused_shares);
    delete [] tc;
  }
  DBUG_VOID_RETURN;
}


/**
  Get number of cached table definitions.

  @return Number of cached table definitions
*/

ulong tdc_records(void)
{
  return lf_hash_size(&tdc_hash);
}


void tdc_purge(bool all)
{
  DBUG_ENTER("tdc_purge");
  while (all || tdc_records() > tdc_size)
  {
    TDC_element *element;

    mysql_mutex_lock(&LOCK_unused_shares);
    if (!(element= unused_shares.pop_front()))
    {
      mysql_mutex_unlock(&LOCK_unused_shares);
      break;
    }

    /* Concurrent thread may start using share again, reset prev and next. */
    element->prev= 0;
    element->next= 0;
    mysql_mutex_lock(&element->LOCK_table_share);
    if (element->ref_count)
    {
      mysql_mutex_unlock(&element->LOCK_table_share);
      mysql_mutex_unlock(&LOCK_unused_shares);
      continue;
    }
    mysql_mutex_unlock(&LOCK_unused_shares);

    tdc_delete_share_from_hash(element);
  }
  DBUG_VOID_RETURN;
}


/**
  Lock table share.

  Find table share with given db.table_name in table definition cache. Return
  locked table share if found.

  Locked table share means:
  - table share is protected against removal from table definition cache
  - no other thread can acquire/release table share

  Caller is expected to unlock table share with tdc_unlock_share().

  @retval 0 Share not found
  @retval MY_ERRPTR OOM
  @retval ptr Pointer to locked table share
*/

TDC_element *tdc_lock_share(THD *thd, const char *db, const char *table_name)
{
  TDC_element *element;
  char key[MAX_DBKEY_LENGTH];

  DBUG_ENTER("tdc_lock_share");
  if (unlikely(fix_thd_pins(thd)))
    DBUG_RETURN((TDC_element*) MY_ERRPTR);

  element= (TDC_element *) lf_hash_search(&tdc_hash, thd->tdc_hash_pins,
                                          (uchar*) key,
                                          tdc_create_key(key, db, table_name));
  if (element)
  {
    mysql_mutex_lock(&element->LOCK_table_share);
    if (unlikely(!element->share || element->share->error))
    {
      mysql_mutex_unlock(&element->LOCK_table_share);
      element= 0;
    }
    lf_hash_search_unpin(thd->tdc_hash_pins);
  }

  DBUG_RETURN(element);
}


/**
  Unlock share locked by tdc_lock_share().
*/

void tdc_unlock_share(TDC_element *element)
{
  DBUG_ENTER("tdc_unlock_share");
  mysql_mutex_unlock(&element->LOCK_table_share);
  DBUG_VOID_RETURN;
}


int tdc_share_is_cached(THD *thd, const char *db, const char *table_name)
{
  char key[MAX_DBKEY_LENGTH];

  if (unlikely(fix_thd_pins(thd)))
    return -1;

  if (lf_hash_search(&tdc_hash, thd->tdc_hash_pins, (uchar*) key,
                     tdc_create_key(key, db, table_name)))
  {
    lf_hash_search_unpin(thd->tdc_hash_pins);
    return 1;
  }
  return 0;
}


/*
  Get TABLE_SHARE for a table.

  tdc_acquire_share()
  thd                   Thread handle
  tl                    Table that should be opened
  flags                 operation: what to open table or view
  out_table             TABLE for the requested table

  IMPLEMENTATION
    Get a table definition from the table definition cache.
    If it doesn't exist, create a new from the table definition file.

  RETURN
   0  Error
   #  Share for table
*/

TABLE_SHARE *tdc_acquire_share(THD *thd, TABLE_LIST *tl, uint flags,
                               TABLE **out_table)
{
  TABLE_SHARE *share;
  TDC_element *element;
  const char *key;
  uint key_length= get_table_def_key(tl, &key);
  my_hash_value_type hash_value= tl->mdl_request.key.tc_hash_value();
  bool was_unused;
  DBUG_ENTER("tdc_acquire_share");

  if (fix_thd_pins(thd))
    DBUG_RETURN(0);

retry:
  while (!(element= (TDC_element*) lf_hash_search_using_hash_value(&tdc_hash,
                    thd->tdc_hash_pins, hash_value, (uchar*) key, key_length)))
  {
    LEX_STRING tmp= { const_cast<char*>(key), key_length };
    int res= lf_hash_insert(&tdc_hash, thd->tdc_hash_pins, (uchar*) &tmp);

    if (res == -1)
      DBUG_RETURN(0);
    else if (res == 1)
      continue;

    element= (TDC_element*) lf_hash_search_using_hash_value(&tdc_hash,
             thd->tdc_hash_pins, hash_value, (uchar*) key, key_length);
    lf_hash_search_unpin(thd->tdc_hash_pins);
    DBUG_ASSERT(element);

    if (!(share= alloc_table_share(tl->db.str, tl->table_name.str, key, key_length)))
    {
      lf_hash_delete(&tdc_hash, thd->tdc_hash_pins, key, key_length);
      DBUG_RETURN(0);
    }

    /* note that tdc_acquire_share() *always* uses discovery */
    open_table_def(thd, share, flags | GTS_USE_DISCOVERY);

    if (checked_unlikely(share->error))
    {
      free_table_share(share);
      lf_hash_delete(&tdc_hash, thd->tdc_hash_pins, key, key_length);
      DBUG_RETURN(0);
    }

    mysql_mutex_lock(&element->LOCK_table_share);
    element->share= share;
    share->tdc= element;
    element->ref_count++;
    element->flushed= false;
    mysql_mutex_unlock(&element->LOCK_table_share);

    tdc_purge(false);
    if (out_table)
    {
      status_var_increment(thd->status_var.table_open_cache_misses);
      *out_table= 0;
    }
    share->m_psi= PSI_CALL_get_table_share(false, share);
    goto end;
  }

  /* cannot force discovery of a cached share */
  DBUG_ASSERT(!(flags & GTS_FORCE_DISCOVERY));

  if (out_table && (flags & GTS_TABLE))
  {
    if ((*out_table= tc_acquire_table(thd, element)))
    {
      lf_hash_search_unpin(thd->tdc_hash_pins);
      DBUG_ASSERT(!(flags & GTS_NOLOCK));
      DBUG_ASSERT(element->share);
      DBUG_ASSERT(!element->share->error);
      DBUG_ASSERT(!element->share->is_view);
      status_var_increment(thd->status_var.table_open_cache_hits);
      DBUG_RETURN(element->share);
    }
    status_var_increment(thd->status_var.table_open_cache_misses);
  }

  mysql_mutex_lock(&element->LOCK_table_share);
  if (!(share= element->share))
  {
    mysql_mutex_unlock(&element->LOCK_table_share);
    lf_hash_search_unpin(thd->tdc_hash_pins);
    goto retry;
  }
  lf_hash_search_unpin(thd->tdc_hash_pins);

  /*
     We found an existing table definition. Return it if we didn't get
     an error when reading the table definition from file.
  */
  if (unlikely(share->error))
  {
    open_table_error(share, share->error, share->open_errno);
    goto err;
  }

  if (share->is_view && !(flags & GTS_VIEW))
  {
    open_table_error(share, OPEN_FRM_NOT_A_TABLE, ENOENT);
    goto err;
  }
  if (!share->is_view && !(flags & GTS_TABLE))
  {
    open_table_error(share, OPEN_FRM_NOT_A_VIEW, ENOENT);
    goto err;
  }

  was_unused= !element->ref_count;
  element->ref_count++;
  mysql_mutex_unlock(&element->LOCK_table_share);
  if (was_unused)
  {
    mysql_mutex_lock(&LOCK_unused_shares);
    if (element->prev)
    {
      /*
        Share was not used before and it was in the old_unused_share list
        Unlink share from this list
      */
      DBUG_PRINT("info", ("Unlinking from not used list"));
      unused_shares.remove(element);
      element->next= 0;
      element->prev= 0;
    }
    mysql_mutex_unlock(&LOCK_unused_shares);
  }

end:
  DBUG_PRINT("exit", ("share: %p  ref_count: %u",
                      share, share->tdc->ref_count));
  if (flags & GTS_NOLOCK)
  {
    tdc_release_share(share);
    /*
      if GTS_NOLOCK is requested, the returned share pointer cannot be used,
      the share it points to may go away any moment.
      But perhaps the caller is only interested to know whether a share or
      table existed?
      Let's return an invalid pointer here to catch dereferencing attempts.
    */
    share= UNUSABLE_TABLE_SHARE;
  }
  DBUG_RETURN(share);

err:
  mysql_mutex_unlock(&element->LOCK_table_share);
  DBUG_RETURN(0);
}


/**
  Release table share acquired by tdc_acquire_share().
*/

void tdc_release_share(TABLE_SHARE *share)
{
  DBUG_ENTER("tdc_release_share");

  mysql_mutex_lock(&share->tdc->LOCK_table_share);
  DBUG_PRINT("enter",
             ("share: %p  table: %s.%s  ref_count: %u",
              share, share->db.str, share->table_name.str,
              share->tdc->ref_count));
  DBUG_ASSERT(share->tdc->ref_count);

  if (share->tdc->ref_count > 1)
  {
    share->tdc->ref_count--;
    if (!share->is_view)
      mysql_cond_broadcast(&share->tdc->COND_release);
    mysql_mutex_unlock(&share->tdc->LOCK_table_share);
    DBUG_VOID_RETURN;
  }
  mysql_mutex_unlock(&share->tdc->LOCK_table_share);

  mysql_mutex_lock(&LOCK_unused_shares);
  mysql_mutex_lock(&share->tdc->LOCK_table_share);
  if (--share->tdc->ref_count)
  {
    if (!share->is_view)
      mysql_cond_broadcast(&share->tdc->COND_release);
    mysql_mutex_unlock(&share->tdc->LOCK_table_share);
    mysql_mutex_unlock(&LOCK_unused_shares);
    DBUG_VOID_RETURN;
  }
  if (share->tdc->flushed || tdc_records() > tdc_size)
  {
    mysql_mutex_unlock(&LOCK_unused_shares);
    tdc_delete_share_from_hash(share->tdc);
    DBUG_VOID_RETURN;
  }
  /* Link share last in used_table_share list */
  DBUG_PRINT("info", ("moving share to unused list"));
  DBUG_ASSERT(share->tdc->next == 0);
  unused_shares.push_back(share->tdc);
  mysql_mutex_unlock(&share->tdc->LOCK_table_share);
  mysql_mutex_unlock(&LOCK_unused_shares);
  DBUG_VOID_RETURN;
}


void tdc_remove_referenced_share(THD *thd, TABLE_SHARE *share)
{
  DBUG_ASSERT(thd->mdl_context.is_lock_owner(MDL_key::TABLE, share->db.str,
                                             share->table_name.str,
                                             MDL_EXCLUSIVE));
  share->tdc->flush_unused(false);
  mysql_mutex_lock(&share->tdc->LOCK_table_share);
  share->tdc->wait_for_refs(1);
  DBUG_ASSERT(share->tdc->all_tables.is_empty());
  share->tdc->ref_count--;
  tdc_delete_share_from_hash(share->tdc);
}


/**
   Removes all TABLE instances and corresponding TABLE_SHARE

   @param  thd          Thread context
   @param  db           Name of database
   @param  table_name   Name of table

   @note It assumes that table instances are already not used by any
   (other) thread (this should be achieved by using meta-data locks).
*/

void tdc_remove_table(THD *thd, const char *db, const char *table_name)
{
  TDC_element *element;
  DBUG_ENTER("tdc_remove_table");
  DBUG_PRINT("enter", ("name: %s", table_name));

  DBUG_ASSERT(thd->mdl_context.is_lock_owner(MDL_key::TABLE, db, table_name,
                                             MDL_EXCLUSIVE));

  mysql_mutex_lock(&LOCK_unused_shares);
  if (!(element= tdc_lock_share(thd, db, table_name)))
  {
    mysql_mutex_unlock(&LOCK_unused_shares);
    DBUG_VOID_RETURN;
  }

  DBUG_ASSERT(element != MY_ERRPTR); // What can we do about it?

  if (!element->ref_count)
  {
    if (element->prev)
    {
      unused_shares.remove(element);
      element->prev= 0;
      element->next= 0;
    }
    mysql_mutex_unlock(&LOCK_unused_shares);

    tdc_delete_share_from_hash(element);
    DBUG_VOID_RETURN;
  }
  mysql_mutex_unlock(&LOCK_unused_shares);

  element->ref_count++;
  mysql_mutex_unlock(&element->LOCK_table_share);

  /* We have to relock the mutex to avoid code duplication. Sigh. */
  tdc_remove_referenced_share(thd, element->share);
  DBUG_VOID_RETURN;
}


/**
  Check if table's share is being removed from the table definition
  cache and, if yes, wait until the flush is complete.

  @param thd             Thread context.
  @param table_list      Table which share should be checked.
  @param timeout         Timeout for waiting.
  @param deadlock_weight Weight of this wait for deadlock detector.

  @retval 0       Success. Share is up to date or has been flushed.
  @retval 1       Error (OOM, was killed, the wait resulted
                  in a deadlock or timeout). Reported.
*/

int tdc_wait_for_old_version(THD *thd, const char *db, const char *table_name,
                             ulong wait_timeout, uint deadlock_weight)
{
  TDC_element *element;

  if (!(element= tdc_lock_share(thd, db, table_name)))
    return FALSE;
  else if (element == MY_ERRPTR)
    return TRUE;
  else if (element->flushed)
  {
    struct timespec abstime;
    set_timespec(abstime, wait_timeout);
    return element->share->wait_for_old_version(thd, &abstime, deadlock_weight);
  }
  tdc_unlock_share(element);
  return FALSE;
}


/**
  Iterate table definition cache.

  Object is protected against removal from table definition cache.

  @note Returned TABLE_SHARE is not guaranteed to be fully initialized:
  tdc_acquire_share() added new share, but didn't open it yet. If caller
  needs fully initializer share, it must lock table share mutex.
*/

struct eliminate_duplicates_arg
{
  HASH hash;
  MEM_ROOT root;
  my_hash_walk_action action;
  void *argument;
};


static uchar *eliminate_duplicates_get_key(const uchar *element, size_t *length,
                                       my_bool not_used __attribute__((unused)))
{
  LEX_STRING *key= (LEX_STRING *) element;
  *length= key->length;
  return (uchar *) key->str;
}


static my_bool eliminate_duplicates(TDC_element *element,
                                    eliminate_duplicates_arg *arg)
{
  LEX_STRING *key= (LEX_STRING *) alloc_root(&arg->root, sizeof(LEX_STRING));

  if (!key || !(key->str= (char*) memdup_root(&arg->root, element->m_key,
                                              element->m_key_length)))
    return TRUE;

  key->length= element->m_key_length;

  if (my_hash_insert(&arg->hash, (uchar *) key))
    return FALSE;

  return arg->action(element, arg->argument);
}


int tdc_iterate(THD *thd, my_hash_walk_action action, void *argument,
                bool no_dups)
{
  eliminate_duplicates_arg no_dups_argument;
  LF_PINS *pins;
  myf alloc_flags= 0;
  uint hash_flags= HASH_UNIQUE;
  int res;

  if (thd)
  {
    fix_thd_pins(thd);
    pins= thd->tdc_hash_pins;
    alloc_flags= MY_THREAD_SPECIFIC;
    hash_flags|= HASH_THREAD_SPECIFIC;
  }
  else
    pins= lf_hash_get_pins(&tdc_hash);

  if (!pins)
    return ER_OUTOFMEMORY;

  if (no_dups)
  {
    init_alloc_root(PSI_INSTRUMENT_ME, &no_dups_argument.root, 4096, 4096, MYF(alloc_flags));
    my_hash_init(PSI_INSTRUMENT_ME, &no_dups_argument.hash, &my_charset_bin,
                 tdc_records(), 0, 0, eliminate_duplicates_get_key, 0,
                 hash_flags);
    no_dups_argument.action= action;
    no_dups_argument.argument= argument;
    action= (my_hash_walk_action) eliminate_duplicates;
    argument= &no_dups_argument;
  }

  res= lf_hash_iterate(&tdc_hash, pins, action, argument);

  if (!thd)
    lf_hash_put_pins(pins);

  if (no_dups)
  {
    my_hash_free(&no_dups_argument.hash);
    free_root(&no_dups_argument.root, MYF(0));
  }
  return res;
}


int show_tc_active_instances(THD *thd, SHOW_VAR *var, char *buff,
                             enum enum_var_type scope)
{
  var->type= SHOW_UINT;
  var->value= buff;
  *(reinterpret_cast<uint32_t*>(buff))=
    tc_active_instances.load(std::memory_order_relaxed);
  return 0;
}


/**
  Waits until ref_count goes down to given number

  @param  my_refs  Number of references owned by the caller

  Caller must own at least one TABLE_SHARE reference.

  Even though current thread holds exclusive metadata lock on this share,
  concurrent FLUSH TABLES threads may be in process of closing unused table
  instances belonging to this share. E.g.:
  thr1 (FLUSH TABLES): table= share->tdc.free_tables.pop_front();
  thr1 (FLUSH TABLES): share->tdc.all_tables.remove(table);
  thr2 (ALTER TABLE): tdc_remove_table();
  thr1 (FLUSH TABLES): intern_close_table(table);

  Current remove type assumes that all table instances (except for those
  that are owned by current thread) must be closed before
  thd_remove_table() returns. Wait for such tables now.

  intern_close_table() decrements ref_count and signals COND_release. When
  ref_count drops down to number of references owned by current thread
  waiting is completed.

  Unfortunately TABLE_SHARE::wait_for_old_version() cannot be used here
  because it waits for all table instances, whereas we have to wait only
  for those that are not owned by current thread.
*/

void TDC_element::wait_for_refs(uint my_refs)
{
  while (ref_count > my_refs)
    mysql_cond_wait(&COND_release, &LOCK_table_share);
}


/**
  Flushes unused TABLE instances

  @param  thd          Thread context
  @param  mark_flushed Whether to destroy TABLE_SHARE when released

  Caller is allowed to own used TABLE instances.
  There must be no TABLE objects used by other threads and caller must own
  exclusive metadata lock on the table.
*/

void TDC_element::flush(THD *thd, bool mark_flushed)
{
  DBUG_ASSERT(thd->mdl_context.is_lock_owner(MDL_key::TABLE, share->db.str,
                                             share->table_name.str,
                                             MDL_EXCLUSIVE));

  flush_unused(mark_flushed);

  mysql_mutex_lock(&LOCK_table_share);
  All_share_tables_list::Iterator it(all_tables);
  uint my_refs= 0;
  while (auto table= it++)
  {
    if (table->in_use == thd)
      my_refs++;
  }
  wait_for_refs(my_refs);
#ifndef DBUG_OFF
  it.rewind();
  while (auto table= it++)
    DBUG_ASSERT(table->in_use == thd);
#endif
  mysql_mutex_unlock(&LOCK_table_share);
}


/**
  Flushes unused TABLE instances
*/

void TDC_element::flush_unused(bool mark_flushed)
{
  Share_free_tables::List purge_tables;

  mysql_mutex_lock(&LOCK_table_share);
  if (mark_flushed)
    flushed= true;
  tc_remove_all_unused_tables(this, &purge_tables);
  mysql_mutex_unlock(&LOCK_table_share);

  while (auto table= purge_tables.pop_front())
    intern_close_table(table);
}