semisync_master.cc 33.9 KB
Newer Older
He Zhenxing's avatar
He Zhenxing committed
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
/* Copyright (C) 2007 Google Inc.
   Copyright (C) 2008 MySQL AB

   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  02110-1301  USA */


#include "semisync_master.h"

#define TIME_THOUSAND 1000
#define TIME_MILLION  1000000
#define TIME_BILLION  1000000000

/* This indicates whether semi-synchronous replication is enabled. */
char rpl_semi_sync_master_enabled;
unsigned long rpl_semi_sync_master_timeout;
unsigned long rpl_semi_sync_master_trace_level;
unsigned long rpl_semi_sync_master_status           = 0;
unsigned long rpl_semi_sync_master_yes_transactions = 0;
unsigned long rpl_semi_sync_master_no_transactions  = 0;
unsigned long rpl_semi_sync_master_off_times        = 0;
unsigned long rpl_semi_sync_master_timefunc_fails   = 0;
unsigned long rpl_semi_sync_master_num_timeouts     = 0;
unsigned long rpl_semi_sync_master_wait_sessions    = 0;
unsigned long rpl_semi_sync_master_back_wait_pos    = 0;
unsigned long rpl_semi_sync_master_trx_wait_time    = 0;
unsigned long long rpl_semi_sync_master_trx_wait_num = 0;
unsigned long rpl_semi_sync_master_net_wait_time    = 0;
unsigned long long rpl_semi_sync_master_net_wait_num = 0;
unsigned long rpl_semi_sync_master_clients          = 0;
unsigned long long rpl_semi_sync_master_net_wait_total_time = 0;
unsigned long long rpl_semi_sync_master_trx_wait_total_time = 0;


static int getWaitTime(const struct timeval& start_tv);

/*******************************************************************************
 *
 * <ActiveTranx> class : manage all active transaction nodes
 *
 ******************************************************************************/

ActiveTranx::ActiveTranx(int max_connections,
			 pthread_mutex_t *lock,
			 unsigned long trace_level)
  : Trace(trace_level), num_transactions_(max_connections),
    num_entries_(max_connections << 1),
    lock_(lock)
{
  /* Allocate the memory for the array */
  node_array_ = new TranxNode[num_transactions_];
  for (int idx = 0; idx < num_transactions_; ++idx)
  {
    node_array_[idx].log_pos_     = 0;
    node_array_[idx].hash_next_   = NULL;
    node_array_[idx].next_        = node_array_ + idx + 1;

    node_array_[idx].log_name_    = new char[FN_REFLEN];
    node_array_[idx].log_name_[0] = '\x0';
  }
  node_array_[num_transactions_-1].next_ = NULL;

  /* All nodes in the array go to the pool initially. */
  free_pool_ = node_array_;

  /* No transactions are in the list initially. */
  trx_front_ = NULL;
  trx_rear_  = NULL;

  /* Create the hash table to find a transaction's ending event. */
  trx_htb_ = new TranxNode *[num_entries_];
  for (int idx = 0; idx < num_entries_; ++idx)
    trx_htb_[idx] = NULL;

  sql_print_information("Semi-sync replication initialized for %d "
                        "transactions.", num_transactions_);
}

ActiveTranx::~ActiveTranx()
{
  for (int idx = 0; idx < num_transactions_; ++idx)
  {
    delete [] node_array_[idx].log_name_;
    node_array_[idx].log_name_ = NULL;
  }

  delete [] node_array_;
  delete [] trx_htb_;

  node_array_       = NULL;
  trx_htb_          = NULL;
  num_transactions_ = 0;
  num_entries_      = 0;
}

unsigned int ActiveTranx::calc_hash(const unsigned char *key,
                                    unsigned int length)
{
  unsigned int nr = 1, nr2 = 4;

  /* The hash implementation comes from calc_hashnr() in mysys/hash.c. */
  while (length--)
  {
    nr  ^= (((nr & 63)+nr2)*((unsigned int) (unsigned char) *key++))+ (nr << 8);
    nr2 += 3;
  }
  return((unsigned int) nr);
}

unsigned int ActiveTranx::get_hash_value(const char *log_file_name,
				 my_off_t    log_file_pos)
{
  unsigned int hash1 = calc_hash((const unsigned char *)log_file_name,
                                 strlen(log_file_name));
  unsigned int hash2 = calc_hash((const unsigned char *)(&log_file_pos),
                                 sizeof(log_file_pos));

  return (hash1 + hash2) % num_entries_;
}

ActiveTranx::TranxNode* ActiveTranx::alloc_tranx_node()
{
  TranxNode *ptr = free_pool_;

  if (free_pool_)
  {
    free_pool_ = free_pool_->next_;
    ptr->next_ = NULL;
    ptr->hash_next_ = NULL;
  }
  else
  {
    /*
      free_pool should never be NULL here, because we have
      max_connections number of pre-allocated nodes.
    */
    sql_print_error("You have encountered a semi-sync bug (free_pool == NULL), "
                    "please report to http://bugs.mysql.com");
    assert(free_pool_);
  }

  return ptr;
}

int ActiveTranx::compare(const char *log_file_name1, my_off_t log_file_pos1,
			 const char *log_file_name2, my_off_t log_file_pos2)
{
  int cmp = strcmp(log_file_name1, log_file_name2);

  if (cmp != 0)
    return cmp;

  if (log_file_pos1 > log_file_pos2)
    return 1;
  else if (log_file_pos1 < log_file_pos2)
    return -1;
  return 0;
}

int ActiveTranx::insert_tranx_node(const char *log_file_name,
				   my_off_t log_file_pos)
{
  const char *kWho = "ActiveTranx:insert_tranx_node";
  TranxNode  *ins_node;
  int         result = 0;
  unsigned int        hash_val;

  function_enter(kWho);

  ins_node = alloc_tranx_node();
  if (!ins_node)
  {
    sql_print_error("%s: transaction node allocation failed for: (%s, %lu)",
                    kWho, log_file_name, (unsigned long)log_file_pos);
    result = -1;
    goto l_end;
  }

  /* insert the binlog position in the active transaction list. */
  strcpy(ins_node->log_name_, log_file_name);
  ins_node->log_pos_ = log_file_pos;

  if (!trx_front_)
  {
    /* The list is empty. */
    trx_front_ = trx_rear_ = ins_node;
  }
  else
  {
    int cmp = compare(ins_node, trx_rear_);
    if (cmp > 0)
    {
      /* Compare with the tail first.  If the transaction happens later in
       * binlog, then make it the new tail.
       */
      trx_rear_->next_ = ins_node;
      trx_rear_        = ins_node;
    }
    else
    {
      /* Otherwise, it is an error because the transaction should hold the
       * mysql_bin_log.LOCK_log when appending events.
       */
      sql_print_error("%s: binlog write out-of-order, tail (%s, %lu), "
                      "new node (%s, %lu)", kWho,
                      trx_rear_->log_name_, (unsigned long)trx_rear_->log_pos_,
                      ins_node->log_name_, (unsigned long)ins_node->log_pos_);
      result = -1;
      goto l_end;
    }
  }

  hash_val = get_hash_value(ins_node->log_name_, ins_node->log_pos_);
  ins_node->hash_next_ = trx_htb_[hash_val];
  trx_htb_[hash_val]   = ins_node;

  if (trace_level_ & kTraceDetail)
    sql_print_information("%s: insert (%s, %lu) in entry(%u)", kWho,
                          ins_node->log_name_, (unsigned long)ins_node->log_pos_,
                          hash_val);

 l_end:
  return function_exit(kWho, result);
}

bool ActiveTranx::is_tranx_end_pos(const char *log_file_name,
				   my_off_t    log_file_pos)
{
  const char *kWho = "ActiveTranx::is_tranx_end_pos";
  function_enter(kWho);

  unsigned int hash_val = get_hash_value(log_file_name, log_file_pos);
  TranxNode *entry = trx_htb_[hash_val];

  while (entry != NULL)
  {
    if (compare(entry, log_file_name, log_file_pos) == 0)
      break;

    entry = entry->hash_next_;
  }

  if (trace_level_ & kTraceDetail)
    sql_print_information("%s: probe (%s, %lu) in entry(%u)", kWho,
                          log_file_name, (unsigned long)log_file_pos, hash_val);

  function_exit(kWho, (entry != NULL));
  return (entry != NULL);
}

int ActiveTranx::clear_active_tranx_nodes(const char *log_file_name,
					  my_off_t log_file_pos)
{
  const char *kWho = "ActiveTranx::::clear_active_tranx_nodes";
  TranxNode *new_front;

  function_enter(kWho);

  if (log_file_name != NULL)
  {
    new_front = trx_front_;

    while (new_front)
    {
      if (compare(new_front, log_file_name, log_file_pos) > 0)
        break;
      new_front = new_front->next_;
    }
  }
  else
  {
    /* If log_file_name is NULL, clear everything. */
    new_front = NULL;
  }

  if (new_front == NULL)
  {
    /* No active transaction nodes after the call. */

    /* Clear the hash table. */
    memset(trx_htb_, 0, num_entries_ * sizeof(TranxNode *));

    /* Clear the active transaction list. */
    if (trx_front_ != NULL)
    {
      trx_rear_->next_ = free_pool_;
      free_pool_ = trx_front_;
      trx_front_ = NULL;
      trx_rear_  = NULL;
    }

    if (trace_level_ & kTraceDetail)
      sql_print_information("%s: free all nodes back to free list", kWho);
  }
  else if (new_front != trx_front_)
  {
    TranxNode *curr_node, *next_node;

    /* Delete all transaction nodes before the confirmation point. */
    int n_frees = 0;
    curr_node = trx_front_;
    while (curr_node != new_front)
    {
      next_node = curr_node->next_;

      /* Put the node in the memory pool. */
      curr_node->next_ = free_pool_;
      free_pool_       = curr_node;
      n_frees++;

      /* Remove the node from the hash table. */
      unsigned int hash_val = get_hash_value(curr_node->log_name_, curr_node->log_pos_);
      TranxNode **hash_ptr = &(trx_htb_[hash_val]);
      while ((*hash_ptr) != NULL)
      {
        if ((*hash_ptr) == curr_node)
	{
          (*hash_ptr) = curr_node->hash_next_;
          break;
        }
        hash_ptr = &((*hash_ptr)->hash_next_);
      }

      curr_node = next_node;
    }

    trx_front_ = new_front;

    if (trace_level_ & kTraceDetail)
      sql_print_information("%s: free %d nodes back until pos (%s, %lu)",
                            kWho, n_frees,
                            trx_front_->log_name_, (unsigned long)trx_front_->log_pos_);
  }

  return function_exit(kWho, 0);
}


/*******************************************************************************
 *
 * <ReplSemiSyncMaster> class: the basic code layer for sync-replication master.
 * <ReplSemiSyncSlave>  class: the basic code layer for sync-replication slave.
 *
 * The most important functions during semi-syn replication listed:
 *
 * Master:
 *  . reportReplyBinlog(): called by the binlog dump thread when it receives
 *                         the slave's status information.
 *  . updateSyncHeader():  based on transaction waiting information, decide
 *                         whether to request the slave to reply.
 *  . writeTraxInBinlog(): called by the transaction thread when it finishes
 *                         writing all transaction events in binlog.
 *  . commitTrx():         transaction thread wait for the slave reply.
 *
 * Slave:
 *  . slaveReadSyncHeader(): read the semi-sync header from the master, get the
 *                         sync status and get the payload for events.
 *  . slaveReply():        reply to the master about the replication progress.
 *
 ******************************************************************************/

ReplSemiSyncMaster::ReplSemiSyncMaster()
  : active_tranxs_(NULL),
    init_done_(false),
    reply_file_name_inited_(false),
    reply_file_pos_(0L),
    wait_file_name_inited_(false),
    wait_file_pos_(0),
    master_enabled_(false),
    wait_timeout_(0L),
    state_(0),
    enabled_transactions_(0),
    disabled_transactions_(0),
    switched_off_times_(0),
    timefunc_fails_(0),
    wait_sessions_(0),
    wait_backtraverse_(0),
    total_trx_wait_num_(0),
    total_trx_wait_time_(0),
    total_net_wait_num_(0),
    total_net_wait_time_(0),
    max_transactions_(0L)
{
  strcpy(reply_file_name_, "");
  strcpy(wait_file_name_, "");
}

int ReplSemiSyncMaster::initObject()
{
  int result;
  const char *kWho = "ReplSemiSyncMaster::initObject";

  if (init_done_)
  {
    fprintf(stderr, "%s called twice\n", kWho);
    return 1;
  }
  init_done_ = true;

  /* References to the parameter works after set_options(). */
  setWaitTimeout(rpl_semi_sync_master_timeout);
  setTraceLevel(rpl_semi_sync_master_trace_level);
  max_transactions_ = (int)max_connections;

  /* Mutex initialization can only be done after MY_INIT(). */
  pthread_mutex_init(&LOCK_binlog_, MY_MUTEX_INIT_FAST);
  pthread_cond_init(&COND_binlog_send_, NULL);

  if (rpl_semi_sync_master_enabled)
    result = enableMaster();
  else
    result = disableMaster();

  return result;
}

int ReplSemiSyncMaster::enableMaster()
{
  int result = 0;

  /* Must have the lock when we do enable of disable. */
  lock();

  if (!getMasterEnabled())
  {
    active_tranxs_ = new ActiveTranx(max_connections,
				     &LOCK_binlog_,
				     trace_level_);
    if (active_tranxs_ != NULL)
    {
      commit_file_name_inited_ = false;
      reply_file_name_inited_  = false;
      wait_file_name_inited_   = false;

      set_master_enabled(true);
      state_ = true;
      sql_print_information("Semi-sync replication enabled on the master.");
    }
    else
    {
      sql_print_error("Cannot allocate memory to enable semi-sync on the master.");
      result = -1;
    }
  }

  unlock();

  return result;
}

int ReplSemiSyncMaster::disableMaster()
{
  /* Must have the lock when we do enable of disable. */
  lock();

  if (getMasterEnabled())
  {
    /* Switch off the semi-sync first so that waiting transaction will be
     * waken up.
     */
    switch_off();

    assert(active_tranxs_ != NULL);
    delete active_tranxs_;
    active_tranxs_ = NULL;

    reply_file_name_inited_ = false;
    wait_file_name_inited_  = false;
    commit_file_name_inited_ = false;

    set_master_enabled(false);
    sql_print_information("Semi-sync replication disabled on the master.");
  }

  unlock();

  return 0;
}

ReplSemiSyncMaster::~ReplSemiSyncMaster()
{
  if (init_done_)
  {
    pthread_mutex_destroy(&LOCK_binlog_);
    pthread_cond_destroy(&COND_binlog_send_);
  }

  delete active_tranxs_;
}

void ReplSemiSyncMaster::lock()
{
  pthread_mutex_lock(&LOCK_binlog_);
}

void ReplSemiSyncMaster::unlock()
{
  pthread_mutex_unlock(&LOCK_binlog_);
}

void ReplSemiSyncMaster::cond_broadcast()
{
  pthread_cond_broadcast(&COND_binlog_send_);
}

int ReplSemiSyncMaster::cond_timewait(struct timespec *wait_time)
{
  const char *kWho = "ReplSemiSyncMaster::cond_timewait()";
  int wait_res;

  function_enter(kWho);
  wait_res = pthread_cond_timedwait(&COND_binlog_send_,
                                    &LOCK_binlog_, wait_time);
  return function_exit(kWho, wait_res);
}

void ReplSemiSyncMaster::add_slave()
{
  lock();
  rpl_semi_sync_master_clients++;
  unlock();
}

void ReplSemiSyncMaster::remove_slave()
{
  lock();
  rpl_semi_sync_master_clients--;
  unlock();
}

bool ReplSemiSyncMaster::is_semi_sync_slave()
{
  int null_value;
  long long val= 0;
  get_user_var_int("rpl_semi_sync_slave", &val, &null_value);
  return val;
}

int ReplSemiSyncMaster::reportReplyBinlog(const char *log_file_pos)
{
  char log_name[FN_REFLEN];
  char *endptr;
  my_off_t log_pos= strtoull(log_file_pos, &endptr, 10);
  if (!log_pos || !endptr || *endptr != ':' )
    return 1;
  endptr++;                                     // skip the ':' seperator
  strncpy(log_name, endptr, FN_REFLEN);
  uint32 server_id= 0;
  return reportReplyBinlog(server_id, log_name, log_pos);
}

int ReplSemiSyncMaster::reportReplyBinlog(uint32 server_id,
					  const char *log_file_name,
					  my_off_t log_file_pos)
{
  const char *kWho = "ReplSemiSyncMaster::reportReplyBinlog";
  int   cmp;
  bool  can_release_threads = false;
  bool  need_copy_send_pos = true;

  if (!(getMasterEnabled()))
    return 0;

  function_enter(kWho);

  lock();

  /* This is the real check inside the mutex. */
  if (!getMasterEnabled())
    goto l_end;

  if (!is_on())
    /* We check to see whether we can switch semi-sync ON. */
    try_switch_on(server_id, log_file_name, log_file_pos);

  /* The position should increase monotonically, if there is only one
   * thread sending the binlog to the slave.
   * In reality, to improve the transaction availability, we allow multiple
   * sync replication slaves.  So, if any one of them get the transaction,
   * the transaction session in the primary can move forward.
   */
  if (reply_file_name_inited_)
  {
    cmp = ActiveTranx::compare(log_file_name, log_file_pos,
                               reply_file_name_, reply_file_pos_);

    /* If the requested position is behind the sending binlog position,
     * would not adjust sending binlog position.
     * We based on the assumption that there are multiple semi-sync slave,
     * and at least one of them shou/ld be up to date.
     * If all semi-sync slaves are behind, at least initially, the primary
     * can find the situation after the waiting timeout.  After that, some
     * slaves should catch up quickly.
     */
    if (cmp < 0)
    {
      /* If the position is behind, do not copy it. */
      need_copy_send_pos = false;
    }
  }

  if (need_copy_send_pos)
  {
    strcpy(reply_file_name_, log_file_name);
    reply_file_pos_ = log_file_pos;
    reply_file_name_inited_ = true;

    /* Remove all active transaction nodes before this point. */
    assert(active_tranxs_ != NULL);
    active_tranxs_->clear_active_tranx_nodes(log_file_name, log_file_pos);

    if (trace_level_ & kTraceDetail)
      sql_print_information("%s: Got reply at (%s, %lu)", kWho,
                            log_file_name, (unsigned long)log_file_pos);
  }

  if (wait_sessions_ > 0)
  {
    /* Let us check if some of the waiting threads doing a trx
     * commit can now proceed.
     */
    cmp = ActiveTranx::compare(reply_file_name_, reply_file_pos_,
                               wait_file_name_, wait_file_pos_);
    if (cmp >= 0)
    {
      /* Yes, at least one waiting thread can now proceed:
       * let us release all waiting threads with a broadcast
       */
      can_release_threads = true;
      wait_file_name_inited_ = false;
    }
  }

 l_end:
  unlock();

  if (can_release_threads)
  {
    if (trace_level_ & kTraceDetail)
      sql_print_information("%s: signal all waiting threads.", kWho);

    cond_broadcast();
  }

  return function_exit(kWho, 0);
}

int ReplSemiSyncMaster::commitTrx(const char* trx_wait_binlog_name,
				  my_off_t trx_wait_binlog_pos)
{
  const char *kWho = "ReplSemiSyncMaster::commitTrx";

  function_enter(kWho);

  if (getMasterEnabled() && trx_wait_binlog_name)
  {
    struct timeval start_tv;
    struct timespec abstime;
    int wait_result, start_time_err;
    const char *old_msg= 0;

    start_time_err = gettimeofday(&start_tv, 0);

    /* Acquire the mutex. */
    lock();

    /* This must be called after acquired the lock */
    old_msg= thd_enter_cond(NULL, &COND_binlog_send_, &LOCK_binlog_,
                            "Waiting for semi-sync ACK from slave");

    /* This is the real check inside the mutex. */
    if (!getMasterEnabled() || !is_on() || !rpl_semi_sync_master_clients)
      goto l_end;

    if (trace_level_ & kTraceDetail)
    {
      sql_print_information("%s: wait pos (%s, %lu), repl(%d)\n", kWho,
                            trx_wait_binlog_name, (unsigned long)trx_wait_binlog_pos,
                            (int)is_on());
    }

    while (is_on())
    {
      int cmp = ActiveTranx::compare(reply_file_name_, reply_file_pos_,
                                     trx_wait_binlog_name, trx_wait_binlog_pos);
      if (cmp >= 0)
      {
        /* We have already sent the relevant binlog to the slave: no need to
         * wait here.
         */
        if (trace_level_ & kTraceDetail)
          sql_print_information("%s: Binlog reply is ahead (%s, %lu),",
                                kWho, reply_file_name_, (unsigned long)reply_file_pos_);
        break;
      }

      /* Let us update the info about the minimum binlog position of waiting
       * threads.
       */
      if (wait_file_name_inited_)
      {
        cmp = ActiveTranx::compare(trx_wait_binlog_name, trx_wait_binlog_pos,
                                   wait_file_name_, wait_file_pos_);
        if (cmp <= 0)
	{
          /* This thd has a lower position, let's update the minimum info. */
          strcpy(wait_file_name_, trx_wait_binlog_name);
          wait_file_pos_ = trx_wait_binlog_pos;

          wait_backtraverse_++;
          if (trace_level_ & kTraceDetail)
            sql_print_information("%s: move back wait position (%s, %lu),",
                                  kWho, wait_file_name_, (unsigned long)wait_file_pos_);
        }
      }
      else
      {
        strcpy(wait_file_name_, trx_wait_binlog_name);
        wait_file_pos_ = trx_wait_binlog_pos;
        wait_file_name_inited_ = true;

        if (trace_level_ & kTraceDetail)
          sql_print_information("%s: init wait position (%s, %lu),",
                                kWho, wait_file_name_, (unsigned long)wait_file_pos_);
      }

      if (start_time_err == 0)
      {
        int diff_usecs = start_tv.tv_usec + wait_timeout_ * TIME_THOUSAND;

        /* Calcuate the waiting period. */
        abstime.tv_sec = start_tv.tv_sec;
        if (diff_usecs < TIME_MILLION)
	{
          abstime.tv_nsec = diff_usecs * TIME_THOUSAND;
        }
	else
	{
          while (diff_usecs >= TIME_MILLION)
	  {
            abstime.tv_sec++;
            diff_usecs -= TIME_MILLION;
          }
          abstime.tv_nsec = diff_usecs * TIME_THOUSAND;
        }

        /* In semi-synchronous replication, we wait until the binlog-dump
         * thread has received the reply on the relevant binlog segment from the
         * replication slave.
         *
         * Let us suspend this thread to wait on the condition;
         * when replication has progressed far enough, we will release
         * these waiting threads.
         */
        wait_sessions_++;

        if (trace_level_ & kTraceDetail)
          sql_print_information("%s: wait %lu ms for binlog sent (%s, %lu)",
                                kWho, wait_timeout_,
                                wait_file_name_, (unsigned long)wait_file_pos_);

        wait_result = cond_timewait(&abstime);
        wait_sessions_--;

        if (wait_result != 0)
	{
          /* This is a real wait timeout. */
          sql_print_warning("Timeout waiting for reply of binlog (file: %s, pos: %lu), "
                            "semi-sync up to file %s, position %lu.",
                            trx_wait_binlog_name, (unsigned long)trx_wait_binlog_pos,
                            reply_file_name_, (unsigned long)reply_file_pos_);
          total_wait_timeouts_++;

          /* switch semi-sync off */
          switch_off();
        }
	else
	{
          int wait_time;

          wait_time = getWaitTime(start_tv);
          if (wait_time < 0)
	  {
            if (trace_level_ & kTraceGeneral)
	    {
              /* This is a time/gettimeofday function call error. */
              sql_print_error("Replication semi-sync gettimeofday fail1 at "
                              "wait position (%s, %lu)",
                              trx_wait_binlog_name, (unsigned long)trx_wait_binlog_pos);
            }
            timefunc_fails_++;
          }
	  else
	  {
            total_trx_wait_num_++;
            total_trx_wait_time_ += wait_time;
          }
        }
      }
      else
      {
        if (trace_level_ & kTraceGeneral)
	{
          /* This is a gettimeofday function call error. */
          sql_print_error("Replication semi-sync gettimeofday fail2 at "
                          "wait position (%s, %lu)",
                          trx_wait_binlog_name, (unsigned long)trx_wait_binlog_pos);
        }
        timefunc_fails_++;

        /* switch semi-sync off */
        switch_off();
      }
    }

  l_end:
    /* Update the status counter. */
    if (is_on() && rpl_semi_sync_master_clients)
      enabled_transactions_++;
    else
      disabled_transactions_++;

    /* The lock held will be released by thd_exit_cond, so no need to
       call unlock() here */
    thd_exit_cond(NULL, old_msg);
  }

  return function_exit(kWho, 0);
}

/* Indicate that semi-sync replication is OFF now.
 * 
 * What should we do when it is disabled?  The problem is that we want
 * the semi-sync replication enabled again when the slave catches up
 * later.  But, it is not that easy to detect that the slave has caught
 * up.  This is caused by the fact that MySQL's replication protocol is
 * asynchronous, meaning that if the master does not use the semi-sync
 * protocol, the slave would not send anything to the master.
 * Still, if the master is sending (N+1)-th event, we assume that it is
 * an indicator that the slave has received N-th event and earlier ones.
 *
 * If semi-sync is disabled, all transactions still update the wait
 * position with the last position in binlog.  But no transactions will
 * wait for confirmations and the active transaction list would not be
 * maintained.  In binlog dump thread, updateSyncHeader() checks whether
 * the current sending event catches up with last wait position.  If it
 * does match, semi-sync will be switched on again.
 */
int ReplSemiSyncMaster::switch_off()
{
  const char *kWho = "ReplSemiSyncMaster::switch_off";
  int result;

  function_enter(kWho);
  state_ = false;

  /* Clear the active transaction list. */
  assert(active_tranxs_ != NULL);
  result = active_tranxs_->clear_active_tranx_nodes(NULL, 0);

  switched_off_times_++;
  wait_file_name_inited_   = false;
  reply_file_name_inited_  = false;
  sql_print_information("Semi-sync replication switched OFF.");
  cond_broadcast();                            /* wake up all waiting threads */

  return function_exit(kWho, result);
}

int ReplSemiSyncMaster::try_switch_on(int server_id,
				      const char *log_file_name,
				      my_off_t log_file_pos)
{
  const char *kWho = "ReplSemiSyncMaster::try_switch_on";
  bool semi_sync_on = false;

  function_enter(kWho);

  /* If the current sending event's position is larger than or equal to the
   * 'largest' commit transaction binlog position, the slave is already
   * catching up now and we can switch semi-sync on here.
   * If commit_file_name_inited_ indicates there are no recent transactions,
   * we can enable semi-sync immediately.
   */
  if (commit_file_name_inited_)
  {
    int cmp = ActiveTranx::compare(log_file_name, log_file_pos,
                                   commit_file_name_, commit_file_pos_);
    semi_sync_on = (cmp >= 0);
  }
  else
  {
    semi_sync_on = true;
  }

  if (semi_sync_on)
  {
    /* Switch semi-sync replication on. */
    state_ = true;

    sql_print_information("Semi-sync replication switched ON with slave (server_id: %d) "
                          "at (%s, %lu)",
                          server_id, log_file_name,
                          (unsigned long)log_file_pos);
  }

  return function_exit(kWho, 0);
}

int ReplSemiSyncMaster::reserveSyncHeader(unsigned char *header,
					  unsigned long size)
{
  const char *kWho = "ReplSemiSyncMaster::reserveSyncHeader";
  function_enter(kWho);

  int hlen=0;
  if (!is_semi_sync_slave())
  {
    hlen= 0;
  }
  else
  {
    /* No enough space for the extra header, disable semi-sync master */
    if (sizeof(kSyncHeader) > size)
    {
      sql_print_warning("No enough space in the packet "
                        "for semi-sync extra header, "
                        "semi-sync replication disabled");
      disableMaster();
      return 0;
    }
    
    /* Set the magic number and the sync status.  By default, no sync
     * is required.
     */
    memcpy(header, kSyncHeader, sizeof(kSyncHeader));
    hlen= sizeof(kSyncHeader);
  }
  return function_exit(kWho, hlen);
}

int ReplSemiSyncMaster::updateSyncHeader(unsigned char *packet,
					 const char *log_file_name,
					 my_off_t log_file_pos,
					 uint32 server_id)
{
  const char *kWho = "ReplSemiSyncMaster::updateSyncHeader";
  int  cmp = 0;
  bool sync = false;

  /* If the semi-sync master is not enabled, or the slave is not a semi-sync
   * target, do not request replies from the slave.
   */
  if (!getMasterEnabled() || !is_semi_sync_slave())
  {
    sync = false;
    return 0;
  }

  function_enter(kWho);

  lock();

  /* This is the real check inside the mutex. */
  if (!getMasterEnabled())
  {
    sync = false;
    goto l_end;
  }

  if (is_on())
  {
    /* semi-sync is ON */
    sync = false;     /* No sync unless a transaction is involved. */

    if (reply_file_name_inited_)
    {
      cmp = ActiveTranx::compare(log_file_name, log_file_pos,
                                 reply_file_name_, reply_file_pos_);
      if (cmp <= 0)
      {
        /* If we have already got the reply for the event, then we do
         * not need to sync the transaction again.
         */
        goto l_end;
      }
    }

    if (wait_file_name_inited_)
    {
      cmp = ActiveTranx::compare(log_file_name, log_file_pos,
                                 wait_file_name_, wait_file_pos_);
    }
    else
    {
      cmp = 1;
    }
    
    /* If we are already waiting for some transaction replies which
     * are later in binlog, do not wait for this one event.
     */
    if (cmp >= 0)
    {
      /* 
       * We only wait if the event is a transaction's ending event.
       */
      assert(active_tranxs_ != NULL);
      sync = active_tranxs_->is_tranx_end_pos(log_file_name,
                                               log_file_pos);
    }
  }
  else
  {
    if (commit_file_name_inited_)
    {
      int cmp = ActiveTranx::compare(log_file_name, log_file_pos,
                                     commit_file_name_, commit_file_pos_);
      sync = (cmp >= 0);
    }
    else
    {
      sync = true;
    }
  }

  if (trace_level_ & kTraceDetail)
    sql_print_information("%s: server(%d), (%s, %lu) sync(%d), repl(%d)",
                          kWho, server_id, log_file_name,
                          (unsigned long)log_file_pos, sync, (int)is_on());

 l_end:
  unlock();

  /* We do not need to clear sync flag because we set it to 0 when we
   * reserve the packet header.
   */
  if (sync)
    (packet)[2] = kPacketFlagSync;

  return function_exit(kWho, 0);
}

int ReplSemiSyncMaster::writeTranxInBinlog(const char* log_file_name,
					   my_off_t log_file_pos)
{
  const char *kWho = "ReplSemiSyncMaster::writeTranxInBinlog";
  int result = 0;

  function_enter(kWho);

  lock();

  /* This is the real check inside the mutex. */
  if (!getMasterEnabled())
    goto l_end;

  /* Update the 'largest' transaction commit position seen so far even
   * though semi-sync is switched off.
   * It is much better that we update commit_file_* here, instead of
   * inside commitTrx().  This is mostly because updateSyncHeader()
   * will watch for commit_file_* to decide whether to switch semi-sync
   * on. The detailed reason is explained in function updateSyncHeader().
   */
  if (commit_file_name_inited_)
  {
    int cmp = ActiveTranx::compare(log_file_name, log_file_pos,
                                   commit_file_name_, commit_file_pos_);
    if (cmp > 0)
    {
      /* This is a larger position, let's update the maximum info. */
      strcpy(commit_file_name_, log_file_name);
      commit_file_pos_ = log_file_pos;
    }
  }
  else
  {
    strcpy(commit_file_name_, log_file_name);
    commit_file_pos_ = log_file_pos;
    commit_file_name_inited_ = true;
  }

  if (is_on() && rpl_semi_sync_master_clients)
  {
    assert(active_tranxs_ != NULL);
    if(active_tranxs_->insert_tranx_node(log_file_name, log_file_pos))
    {
      /*
        if insert tranx_node failed, print a warning message
        and turn off semi-sync
      */
      sql_print_warning("Semi-sync failed to insert tranx_node for binlog file: %s, position: %ul",
                        log_file_name, log_file_pos);
      switch_off();
    }
  }

 l_end:
  unlock();

  return function_exit(kWho, result);
}

int ReplSemiSyncMaster::resetMaster()
{
  const char *kWho = "ReplSemiSyncMaster::resetMaster";
  int result = 0;

  function_enter(kWho);


  lock();

  state_ = getMasterEnabled()? 1 : 0;

  wait_file_name_inited_   = false;
  reply_file_name_inited_  = false;
  commit_file_name_inited_ = false;

  enabled_transactions_ = 0;
  disabled_transactions_ = 0;
  switched_off_times_ = 0;
  timefunc_fails_ = 0;
  wait_sessions_ = 0;
  wait_backtraverse_ = 0;
  total_trx_wait_num_ = 0;
  total_trx_wait_time_ = 0;
  total_net_wait_num_ = 0;
  total_net_wait_time_ = 0;

  unlock();

  return function_exit(kWho, result);
}

void ReplSemiSyncMaster::setExportStats()
{
  lock();

  rpl_semi_sync_master_status           = state_ && rpl_semi_sync_master_clients;
  rpl_semi_sync_master_yes_transactions = enabled_transactions_;
  rpl_semi_sync_master_no_transactions  = disabled_transactions_;
  rpl_semi_sync_master_off_times        = switched_off_times_;
  rpl_semi_sync_master_timefunc_fails   = timefunc_fails_;
  rpl_semi_sync_master_num_timeouts     = total_wait_timeouts_;
  rpl_semi_sync_master_wait_sessions    = wait_sessions_;
  rpl_semi_sync_master_back_wait_pos    = wait_backtraverse_;
  rpl_semi_sync_master_trx_wait_num     = total_trx_wait_num_;
  rpl_semi_sync_master_trx_wait_time    =
    ((total_trx_wait_num_) ?
     (unsigned long)((double)total_trx_wait_time_ /
             ((double)total_trx_wait_num_)) : 0);
  rpl_semi_sync_master_net_wait_num     = total_net_wait_num_;
  rpl_semi_sync_master_net_wait_time    =
    ((total_net_wait_num_) ?
     (unsigned long)((double)total_net_wait_time_ /
             ((double)total_net_wait_num_)) : 0);

  rpl_semi_sync_master_net_wait_total_time = total_net_wait_time_;
  rpl_semi_sync_master_trx_wait_total_time = total_trx_wait_time_;

  unlock();
}

/* Get the waiting time given the wait's staring time.
 * 
 * Return:
 *  >= 0: the waiting time in microsecons(us)
 *   < 0: error in gettimeofday or time back traverse
 */
static int getWaitTime(const struct timeval& start_tv)
{
  unsigned long long start_usecs, end_usecs;
  struct timeval end_tv;
  int end_time_err;

  /* Starting time in microseconds(us). */
  start_usecs = start_tv.tv_sec * TIME_MILLION + start_tv.tv_usec;

  /* Get the wait time interval. */
  end_time_err = gettimeofday(&end_tv, 0);

  /* Ending time in microseconds(us). */
  end_usecs = end_tv.tv_sec * TIME_MILLION + end_tv.tv_usec;

  if (end_time_err != 0 || end_usecs < start_usecs)
    return -1;

  return (int)(end_usecs - start_usecs);
}