File: rpl_replica_commit_order_manager.cc

package info (click to toggle)
mysql-8.0 8.0.43-3
  • links: PTS, VCS
  • area: main
  • in suites: sid
  • size: 1,273,924 kB
  • sloc: cpp: 4,684,605; ansic: 412,450; pascal: 108,398; java: 83,641; perl: 30,221; cs: 27,067; sql: 26,594; sh: 24,181; python: 21,816; yacc: 17,169; php: 11,522; xml: 7,388; javascript: 7,076; makefile: 2,194; lex: 1,075; awk: 670; asm: 520; objc: 183; ruby: 97; lisp: 86
file content (692 lines) | stat: -rw-r--r-- 26,233 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
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
/* Copyright (c) 2014, 2025, Oracle and/or its affiliates.

   This program is free software; you can redistribute it and/or modify
   it under the terms of the GNU General Public License, version 2.0,
   as published by the Free Software Foundation.

   This program is designed to work with certain software (including
   but not limited to OpenSSL) that is licensed under separate terms,
   as designated in a particular file or component or in included license
   documentation.  The authors of MySQL hereby grant you an additional
   permission to link the program and your derivative works with the
   separately licensed software that they have either included with
   the program or referenced in the documentation.

   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, version 2.0, 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 02110-1301  USA */

#include "sql/rpl_replica_commit_order_manager.h"

#include <array>

#include "debug_sync.h"  // debug_sync_set_action
#include "my_compiler.h"
#include "my_dbug.h"
#include "my_sys.h"
#include "mysql/components/services/bits/psi_stage_bits.h"
#include "mysql/components/services/log_builtins.h"  // LogErr
#include "mysql/psi/mysql_cond.h"
#include "mysql/psi/mysql_mutex.h"
#include "mysqld_error.h"
#include "sql/binlog.h"
#include "sql/handler.h"  // ha_flush_logs
#include "sql/mdl.h"
#include "sql/mysqld.h"       // key_commit_order_manager_mutex ..
#include "sql/raii/sentry.h"  // raii::Sentry<F>
#include "sql/rpl_rli_pdb.h"  // Slave_worker
#include "sql/sql_class.h"
#include "sql/sql_error.h"
#include "sql/sql_lex.h"

Commit_order_manager::Commit_order_manager(uint32 worker_numbers)
    : m_workers(worker_numbers) {
  unset_rollback_status();
}

Commit_order_manager::~Commit_order_manager() = default;

void Commit_order_manager::init_worker_context(Slave_worker &worker) {
  this->m_workers[worker.id].m_mdl_context = &worker.info_thd->mdl_context;
}

void Commit_order_manager::register_trx(Slave_worker *worker) {
  DBUG_TRACE;

  DBUG_PRINT("info", ("Worker %d added to the commit order queue",
                      (int)worker->info_thd->thread_id()));

  /* only transition allowed: FINISHED -> REGISTERED */
  assert(this->m_workers[worker->id].m_stage ==
         cs::apply::Commit_order_queue::enum_worker_stage::FINISHED);
  this->m_workers[worker->id].m_stage =
      cs::apply::Commit_order_queue::enum_worker_stage::REGISTERED;
  this->m_workers.push(worker->id);
}

bool Commit_order_manager::wait_on_graph(Slave_worker *worker) {
  auto worker_thd = worker->info_thd;
  bool rollback_status{false};
  bool allow_commit_out_of_order{false};
  raii::Sentry<> wait_status_guard{[&]() -> void {
    worker_thd->mdl_context.m_wait.reset_status();
    if (rollback_status)
      this->m_workers[worker->id].m_stage =
          cs::apply::Commit_order_queue::enum_worker_stage::REGISTERED;
    else if (allow_commit_out_of_order)
      this->m_workers[worker->id].m_stage =
          cs::apply::Commit_order_queue::enum_worker_stage::WAITED_OVERTAKE;
    else
      this->m_workers[worker->id].m_stage =
          cs::apply::Commit_order_queue::enum_worker_stage::WAITED;
  }};

  worker_thd->mdl_context.m_wait.reset_status();
  this->m_workers[worker->id].m_stage =
      cs::apply::Commit_order_queue::enum_worker_stage::FINISHED_APPLYING;

  if (this->m_workers.front() != worker->id) {
    if (worker->found_commit_order_deadlock()) {
      /* purecov: begin inspected */
      rollback_status = true;
      return true;
      /* purecov: end */
    }
    this->m_workers[worker->id].m_stage =
        cs::apply::Commit_order_queue::enum_worker_stage::REQUESTED_GRANT;

    Commit_order_lock_graph ticket{worker_thd->mdl_context, *this,
                                   static_cast<std::uint32_t>(worker->id)};
    worker_thd->mdl_context.will_wait_for(&ticket);
    worker_thd->mdl_context.find_deadlock();
    raii::Sentry<> ticket_guard{
        [&]() -> void { worker_thd->mdl_context.done_waiting_for(); }};

    /*
      Wait up to a year for a MDL grant, but do it for 1 second
      periods.
    */
    const unsigned long wait_time{1};
    MDL_wait::enum_wait_status wait_status{MDL_wait::WS_EMPTY};
    struct timespec abs_timeout;

    for (unsigned long total_waited_time = 0; total_waited_time < LONG_TIMEOUT;
         total_waited_time += wait_time) {
      set_timespec(&abs_timeout, wait_time);
      wait_status = worker_thd->mdl_context.m_wait.timed_wait(
          worker_thd, &abs_timeout, false,
          &stage_worker_waiting_for_its_turn_to_commit);

      if (wait_status != MDL_wait::WS_EMPTY) {
        break;
      }

      /*
        We reach this point when abs_timeout elapses without a MDL_wait outcome.
        If this worker session's BGC ticket is active and the BGC back ticket
        was incremented, it means that there is a following
        view_change_log_event. Then, we are in a situation where, unless we take
        action here, a deadlock can potentially occur (but not necessarily; see
        below).

        * Description of deadlock situation

        1) This server is the primary of Group Replication;
        2) This is a inbound channel from a source outside of the group;
        3) We had a commit order: T1, T2
        4) Transactions were delivered to the group in order T2, T1.
           This occurs because T2 "overtakes" T1 at any time from when the
           transactions are assigned to worker until they are delivered to the
           group. The most common reason is that T2 executes faster, perhaps
           because T1 is big. However, note that there is no guarantee that
           delivery order equals broadcast order, so the scenario is possible
           even if T1 was broadcast before T2.)
        5) A member joined between T2 and T1.

        * Example of deadlock situation

        Suppose the inbound channel receives transactions T1, T2, T3, T4 in this
        order. They are run and broadcast in the order T3, T4, T2, T1.
        Even when transactions are run and broadcast in T1, T2, T3, T4 order,
        the group delivery may reorder them. A member joined in the middle.
        Thus, the transactions are certified in the following order, and
        assigned the tickets accordingly:
          T3           ticket=1
          T4           ticket=1
          View_change  ticket=2
          T2           ticket=3
          T1           ticket=3

        The ticket manager enforces the global order, the channel enforces the
        channel order:
          global order:  T3, T4, T2, T1
          channel order: T1, T2, T3, T4

        In order to unblock this situation we need to allow transactions ordered
        before the member join to commit in a order that does not respect its
        source. Which will not cause harm since these transactions can only run
        concurrently when they do not have dependencies among them.

        * False positives

        Note that `is_ticket_on_its_turn_and_back_ticket_incremented` may give
        false positives. It does not check if the delivery order was different
        from the order required by replica-preserve-commit-order. Therefore, it
        will allow transaction to commit out of order in cases where it would
        have been possible to wait for the correct commit order without the risk
        of a deadlock. Here is an example:

        Example:
          T2           ticket=1
          T1           ticket=1
          View_change  ticket=2

        If T2 observes the MDL wait timeout and reaches the current point in the
        code, `is_ticket_on_its_turn_and_back_ticket_incremented` will return
        true and unblock it. Then T2 may commit before T1. But it would have
        been safe for T2 to wait for T1.

        A precise check could have checked if the ticket number for the current
        transaction is lower than the ticket number for the preceding
        transaction. However, it is hard to read the ticket number from the
        preceding transaction without race conditions, so we use this coarser
        condition.

        * User impact

        replica-preserve-commit-order mainly has two use cases: (1) to maintain
        application-defined constraints, and (2) to avoid performance
        bottlenecks in handling GTID gaps. The logic here may violate the
        application constraints, but will not introduce significant performance
        bottlenecks:

         1. Suppose the user executes the following transactions on the upstream
            source:

            T1: INSERT INTO table1 ...
            T2: INSERT INTO table2 SELECT ... FROM table1 WHERE ...

            If T1 and T2 execute in parallel on the inbound channel of the group
            primary, and T2 finishes first, and T2 is delivered to the group
            before T1, and there is a view change before T1 is delivered, then
            T2 may commit before T1. In case the user expects that table2 only
            contains rows that exist in table1, then that constraint is
            violated. It is now a limitation that replica-preserve-commit-order
            on inbound channels allows such inconsistencies near view changes.

         2. When transactions commit out-of-order, it results in temporary gaps
            in gtid_executed. The gaps (or rather, the intervals) are stored in
            a data structure. There are frequent operations on the data
            structure which have CPU and memory costs that are linear in the
            number of intervals. For that reason, enabling
            replica-preserve-commit-order can increase the throughput for some
            workloads. Now that we violate replica-preserve-commit-order, it may
            introduce temporary gaps in gtid_executed. However, since the gaps
            only occur for a very brief time near a view change, it is unlikely
            that this has any noticeable impact on the overall throughput.
      */
      if (Commit_stage_manager::get_instance()
              .is_ticket_on_its_turn_and_back_ticket_incremented(worker_thd)) {
        char gtid_buf[Gtid::MAX_TEXT_LENGTH + 1];
        worker_thd->variables.gtid_next.gtid.to_string(global_sid_map, gtid_buf,
                                                       true);
        LogErr(WARNING_LEVEL, ER_RPL_MTA_ALLOW_COMMIT_OUT_OF_ORDER, gtid_buf);
        allow_commit_out_of_order = true;
        return false;
      }
    }

    switch (wait_status) {
      case MDL_wait::GRANTED:
        return false;
      case MDL_wait::WS_EMPTY:
      case MDL_wait::TIMEOUT:
        /* purecov: begin inspected */
        my_error(ER_LOCK_WAIT_TIMEOUT, MYF(0));
        break;
        /* purecov: end */
      case MDL_wait::KILLED:
        /* purecov: begin inspected */
        if (worker_thd->is_killed() == ER_QUERY_TIMEOUT)
          my_error(ER_QUERY_TIMEOUT, MYF(0));
        else
          my_error(ER_QUERY_INTERRUPTED, MYF(0));
        break;
        /* purecov: end */
      case MDL_wait::VICTIM:
        my_error(ER_LOCK_DEADLOCK, MYF(0));
        break;
    }
    worker->report_commit_order_deadlock();
    rollback_status = true;
    return true;
  }
  return false;
}

bool Commit_order_manager::wait(Slave_worker *worker) {
  DBUG_TRACE;

  /*
    When prior transaction fail, current trx should stop and wait for signal
    to rollback itself
  */
  if (this->m_workers[worker->id].m_stage ==
      cs::apply::Commit_order_queue::enum_worker_stage::REGISTERED) {
    CONDITIONAL_SYNC_POINT_FOR_TIMESTAMP("commit_order_manager_before_wait");

    if (this->wait_on_graph(worker)) return true;

    THD *worker_thd = worker->info_thd;
    bool rollback_status = m_rollback_trx.load();

    DBUG_EXECUTE_IF("rpl_fake_commit_order_deadlock_for_timestamp_100", {
      if (worker_thd->start_time.tv_sec == 100) {
        my_error(ER_UNKNOWN_ERROR, MYF(0));
        return true;
      }

      if (worker_thd->start_time.tv_sec == 200 && !rollback_status) {
        my_error(ER_UNKNOWN_ERROR, MYF(0));
        return true;
      }
    });

    CONDITIONAL_SYNC_POINT_FOR_TIMESTAMP("commit_order_manager_after_wait");

    if (rollback_status) {
      finish_one(worker);

      DBUG_PRINT("info", ("thd has seen an error signal from old thread"));
      worker_thd->get_stmt_da()->set_overwrite_status(true);
      my_error(ER_REPLICA_WORKER_STOPPED_PREVIOUS_THD_ERROR, MYF(0));
    }
    /*
      Set HA_IGNORE_DURABILITY so that transaction is not flushed to the
      storage engine immediately, instead we hold all the applier worker
      threads and flush them together in group.
      The tx_commit_pending variable determines if transaction commit is
      pending, and next_to_commit is used to maintain the commit queue for
      group commit. The tx_commit_pending and next_to_commit variables are
      reset before thread enters group commit later.
    */
    else if (worker_thd->is_current_stmt_binlog_disabled()) {
      worker_thd->durability_property = HA_IGNORE_DURABILITY;
      worker_thd->tx_commit_pending = true;
      worker_thd->next_to_commit = nullptr;
    }

    return rollback_status;
  }

  return false;
}

void Commit_order_manager::flush_engine_and_signal_threads(
    Slave_worker *worker) {
  DBUG_TRACE;

  /*
    Enroll the session in FLUSH stage queue so that transactions can be flushed
    together in group with binlog queued sessions. enroll_for will return false
    in case this thread became a follower; in that case the leader has completed
    the flush and signal. If this thread is leader, enroll_for returns true and
    holds the queue lock for COMMIT_ORDER_FLUSH_STAGE as well as the stage lock
    for the the flush stage. In the rest of this function, this thread is a
    leader and will flush to engine and update gtid state, on behalf of all the
    followers.
  */
  if (!Commit_stage_manager::get_instance().enroll_for(
          Commit_stage_manager::COMMIT_ORDER_FLUSH_STAGE, worker->info_thd,
          nullptr, mysql_bin_log.get_log_lock())) {
    m_workers[worker->id].m_stage =
        cs::apply::Commit_order_queue::enum_worker_stage::FINISHED;
    return;
  }

  /* Fetch leader thread from commit order flush queue */
  THD *first =
      Commit_stage_manager::get_instance().fetch_queue_skip_acquire_lock(
          Commit_stage_manager::COMMIT_ORDER_FLUSH_STAGE);

  Commit_stage_manager::get_instance().unlock_queue(
      Commit_stage_manager::COMMIT_ORDER_FLUSH_STAGE);

  mysql_mutex_lock(mysql_bin_log.get_commit_lock());
  mysql_mutex_unlock(mysql_bin_log.get_log_lock());

  CONDITIONAL_SYNC_POINT_FOR_TIMESTAMP(
      "commit_order_leader_before_ha_flush_logs");

  /* flush transactions to the storage engine in a group */
  ha_flush_logs(true);

  reset_server_status(first);

  /* add to @@global.gtid_executed */
  gtid_state->update_commit_group(first);

  mysql_mutex_unlock(mysql_bin_log.get_commit_lock());

  /*
    awake all waiting threads for leader to flush transactions
    to the storage engine
  */
  Commit_stage_manager::get_instance().signal_done(
      first, Commit_stage_manager::COMMIT_ORDER_FLUSH_STAGE);
}

void Commit_order_manager::reset_server_status(THD *first_thd) {
  DBUG_TRACE;

  for (THD *thd = first_thd; thd != nullptr; thd = thd->next_to_commit) {
    thd->server_status &= ~SERVER_STATUS_IN_TRANS;
  }
}

void Commit_order_manager::finish_one(Slave_worker *worker) {
  DBUG_TRACE;
  const cs::apply::Commit_order_queue::enum_worker_stage worker_stage =
      this->m_workers[worker->id].m_stage;

  if (worker_stage ==
          cs::apply::Commit_order_queue::enum_worker_stage::WAITED ||
      worker_stage ==
          cs::apply::Commit_order_queue::enum_worker_stage::WAITED_OVERTAKE) {
    auto this_worker{cs::apply::Commit_order_queue::NO_WORKER};
    cs::apply::Commit_order_queue::sequence_type next_seq_nr{0};

    if (worker_stage ==
        cs::apply::Commit_order_queue::enum_worker_stage::WAITED) {
      assert(this->m_workers.front() == worker->id);
      std::tie(this_worker, next_seq_nr) = this->m_workers.pop();
    }

    else if (worker_stage == cs::apply::Commit_order_queue::enum_worker_stage::
                                 WAITED_OVERTAKE) {
      std::tie(this_worker, next_seq_nr) = this->m_workers.remove(worker->id);
    }

    assert(worker->id == this_worker);
    auto next_worker = this->m_workers.front();
    if (next_worker !=
            cs::apply::Commit_order_queue::NO_WORKER &&  // There is a worker to
                                                         // unblock
        (this->m_workers[next_worker].m_stage ==
             cs::apply::Commit_order_queue::enum_worker_stage::
                 FINISHED_APPLYING ||             // but only if that worker
         this->m_workers[next_worker].m_stage ==  // stage implies the need
             cs::apply::Commit_order_queue::enum_worker_stage::
                 REQUESTED_GRANT) &&
        this->m_workers[next_worker].freeze_commit_sequence_nr(
            next_seq_nr)) {  // and this worker is the one that has
                             // to do it
      this->m_workers[next_worker].m_mdl_context->m_wait.set_status(
          MDL_wait::GRANTED);
      this->m_workers[next_worker].unfreeze_commit_sequence_nr(next_seq_nr);
    }

    this->m_workers[this_worker].m_mdl_context->m_wait.reset_status();
    this->m_workers[this_worker].m_stage =
        cs::apply::Commit_order_queue::enum_worker_stage::FINISHED;
  }
}

void Commit_order_manager::finish(Slave_worker *worker) {
  DBUG_TRACE;
  const cs::apply::Commit_order_queue::enum_worker_stage worker_stage =
      this->m_workers[worker->id].m_stage;

  if (worker_stage ==
          cs::apply::Commit_order_queue::enum_worker_stage::WAITED ||
      worker_stage ==
          cs::apply::Commit_order_queue::enum_worker_stage::WAITED_OVERTAKE) {
    DBUG_PRINT("info",
               ("Worker %lu is signalling next transaction", worker->id));

    if (!get_rollback_status() &&
        worker->info_thd->is_current_stmt_binlog_disabled()) {
      /*
        If flush queue contains only threads executing slave preserve commit
        order, then flush committed transactions of all those waiting threads
        to the storage engine and awake them from wait. But if flush queue
        also contains threads writing to binlog, then it changes leader, so
        that the first BGC thread becomes leader. It waits until the new leader
        has committed and signalled all waiting commit order threads.
      */
      flush_engine_and_signal_threads(worker);

    } else {
      /*
        signal top worker of commit order queue to come out of wait
        and continue processing.
      */
      finish_one(worker);
    }
  }
}

void Commit_order_manager::check_and_report_deadlock(THD *thd_self,
                                                     THD *thd_wait_for) {
  DBUG_TRACE;

  Slave_worker *self_w = get_thd_worker(thd_self);
  Slave_worker *wait_for_w = get_thd_worker(thd_wait_for);
  Commit_order_manager *mngr = self_w->get_commit_order_manager();

  /* Check if both workers are working for the same channel */
  if (mngr != nullptr && self_w->c_rli == wait_for_w->c_rli &&
      wait_for_w->sequence_number() > self_w->sequence_number()) {
    DBUG_PRINT("info", ("Found replica order commit deadlock"));
    mngr->report_deadlock(wait_for_w);
  }
}

void Commit_order_manager::report_deadlock(Slave_worker *worker) {
  DBUG_TRACE;
  worker->report_commit_order_deadlock();
  DBUG_EXECUTE_IF("rpl_fake_cod_deadlock", {
    const char act[] = "now signal reported_deadlock";
    assert(!debug_sync_set_action(current_thd, STRING_WITH_LEN(act)));
  });
  this->m_workers[worker->id].m_mdl_context->m_wait.set_status(
      MDL_wait::VICTIM);
}

bool Commit_order_manager::wait(THD *thd) {
  DBUG_TRACE;
  assert(thd);

  if (has_commit_order_manager(thd)) {
    /*
      We only care about read/write transactions and those that
      have been registered in the commit order manager.
     */
    Slave_worker *worker = dynamic_cast<Slave_worker *>(thd->rli_slave);
    Commit_order_manager *mngr = worker->get_commit_order_manager();

    if (mngr->wait(worker)) {
      thd->commit_error = THD::CE_COMMIT_ERROR;
      return true;
    }
  }
  return false;
}

void Commit_order_manager::wait_and_finish(THD *thd, bool error) {
  DBUG_TRACE;
  assert(thd);

  if (has_commit_order_manager(thd)) {
    /*
      We only care about read/write transactions and those that
      have been registered in the commit order manager.
     */
    Slave_worker *worker = dynamic_cast<Slave_worker *>(thd->rli_slave);
    Commit_order_manager *mngr = worker->get_commit_order_manager();

    if (error || worker->found_commit_order_deadlock()) {
      // Error or deadlock: if not retryable, release next worker
      bool ret;
      std::tie(ret, std::ignore, std::ignore) =
          worker->check_and_report_end_of_retries(thd);
      if (ret) {
        /*
          worker can set m_rollback_trx when it is its turn to commit,
          so need to call wait() before updating m_rollback_trx.
        */
        mngr->wait(worker);
        mngr->set_rollback_status();
        mngr->finish(worker);
      }
    } else {
      // No error or deadlock: release next worker.
      mngr->wait(worker);
      mngr->finish(worker);
    }
  }
}

bool Commit_order_manager::get_rollback_status() {
  return m_rollback_trx.load();
}

void Commit_order_manager::set_rollback_status() { m_rollback_trx.store(true); }

void Commit_order_manager::unset_rollback_status() {
  m_rollback_trx.store(false);
}

bool Commit_order_manager::get_rollback_status(THD *thd) {
  DBUG_TRACE;
  assert(thd);
  if (has_commit_order_manager(thd)) {
    /*
      We only care about read/write transactions and those that
      have been registered in the commit order manager.
     */
    Slave_worker *worker = dynamic_cast<Slave_worker *>(thd->rli_slave);
    Commit_order_manager *mngr = worker->get_commit_order_manager();

    return mngr->get_rollback_status();
  }
  return false;
}

void Commit_order_manager::finish_one(THD *thd) {
  DBUG_TRACE;
  assert(thd);
  if (has_commit_order_manager(thd)) {
    /*
      We only care about read/write transactions and those that
      have been registered in the commit order manager.
     */
    Slave_worker *worker = dynamic_cast<Slave_worker *>(thd->rli_slave);
    Commit_order_manager *mngr = worker->get_commit_order_manager();

    mngr->finish_one(worker);
  }
}

bool Commit_order_manager::visit_lock_graph(
    Commit_order_lock_graph &wait_for_commit,
    MDL_wait_for_graph_visitor &visitor) {
  DBUG_TRACE;

  auto src_ctx = wait_for_commit.get_ctx();
  if (src_ctx->m_wait.get_status() != MDL_wait::WS_EMPTY)
    return false;  // Wait status changed in the meanwhile
  if (visitor.enter_node(src_ctx)) return true;

  raii::Sentry<> visitor_guard{[&]() -> void { visitor.leave_node(src_ctx); }};

  static const std::array<
      std::function<bool(cs::apply::Commit_order_queue::Node *,
                         MDL_wait_for_graph_visitor &)>,
      2>
      validators{// Inspect if the worker is a directly dependent node of the
                 // visitor
                 [](cs::apply::Commit_order_queue::Node *wkr,
                    MDL_wait_for_graph_visitor &vstr) -> bool {
                   return vstr.inspect_edge(wkr->m_mdl_context);
                 },
                 // Inspect if the visitor is a recursevly dependent node of the
                 // worker
                 [](cs::apply::Commit_order_queue::Node *wkr,
                    MDL_wait_for_graph_visitor &vstr) -> bool {
                   return wkr->m_mdl_context->visit_subgraph(&vstr);
                 }};

  auto src_worker_id = wait_for_commit.get_worker_id();
  for (auto validate : validators) {
    raii::Sentry<> freeze_guard{[&]() -> void { this->m_workers.unfreeze(); }};
    this->m_workers.freeze();

    for (auto w : this->m_workers) {
      assert(w != nullptr);
      if (w->m_worker_id == src_worker_id) break;
      if (w->m_stage ==
          cs::apply::Commit_order_queue::enum_worker_stage::REQUESTED_GRANT)
        continue;
      if (validate(w, visitor)) {
        return true;
      }
    }
  }

  return false;
}

bool has_commit_order_manager(const THD *thd) {
  return is_mts_worker(thd) &&
         thd->rli_slave->get_commit_order_manager() != nullptr;
}

bool Commit_order_manager::wait_for_its_turn_before_flush_stage(THD *thd) {
  switch (thd->lex->sql_command) {
    case SQLCOM_ALTER_TABLE:
    case SQLCOM_ANALYZE:
    case SQLCOM_DROP_DB:
    case SQLCOM_DROP_EVENT:
    case SQLCOM_DROP_FUNCTION:
    case SQLCOM_DROP_PROCEDURE:
    case SQLCOM_DROP_TRIGGER:
    case SQLCOM_DROP_TABLE:
    case SQLCOM_DROP_VIEW:
    case SQLCOM_OPTIMIZE:
    case SQLCOM_REPAIR:
      return has_commit_order_manager(thd);
    default:
      break;
  }
  return false;
}

Commit_order_lock_graph::Commit_order_lock_graph(MDL_context &ctx,
                                                 Commit_order_manager &mngr,
                                                 uint32 worker_id)
    : m_ctx{ctx}, m_mngr{mngr}, m_worker_id{worker_id} {}

MDL_context *Commit_order_lock_graph::get_ctx() const { return &(this->m_ctx); }

uint32 Commit_order_lock_graph::get_worker_id() const {
  return this->m_worker_id;
}

bool Commit_order_lock_graph::accept_visitor(
    MDL_wait_for_graph_visitor *visitor) {
  DBUG_TRACE;
  return this->m_mngr.visit_lock_graph(*this, *visitor);
}

uint Commit_order_lock_graph::get_deadlock_weight() const {
  return DEADLOCK_WEIGHT_CO;  // Return the lowest weight so that workers are
                              // the threads to back-off because of more
                              // advanced and automated transaction retry
                              // capabilities.
}