slave.cc 242 KB
Newer Older
1
/* Copyright (c) 2000, 2015, Oracle and/or its affiliates.
2
   Copyright (c) 2008, 2015, MariaDB
3

unknown's avatar
unknown committed
4 5
   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
unknown's avatar
unknown committed
6
   the Free Software Foundation; version 2 of the License.
7

unknown's avatar
unknown committed
8 9 10 11
   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.
12

unknown's avatar
unknown committed
13 14
   You should have received a copy of the GNU General Public License
   along with this program; if not, write to the Free Software
15
   Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301  USA */
unknown's avatar
unknown committed
16

17 18 19 20 21 22 23 24 25 26 27

/**
  @addtogroup Replication
  @{

  @file

  @brief Code to run the io thread and the sql thread on the
  replication slave.
*/

28
#include <my_global.h>
29
#include "sql_priv.h"
30
#include "slave.h"
31 32
#include "sql_parse.h"                         // execute_init_command
#include "sql_table.h"                         // mysql_rm_table
33 34
#include "rpl_mi.h"
#include "rpl_rli.h"
35
#include "sql_repl.h"
unknown's avatar
unknown committed
36
#include "rpl_filter.h"
37
#include "repl_failsafe.h"
Konstantin Osipov's avatar
Konstantin Osipov committed
38
#include "transaction.h"
unknown's avatar
unknown committed
39
#include <thr_alarm.h>
unknown's avatar
unknown committed
40
#include <my_dir.h>
unknown's avatar
unknown committed
41
#include <sql_common.h>
42
#include <errmsg.h>
43
#include <mysqld_error.h>
44
#include <mysys_err.h>
He Zhenxing's avatar
He Zhenxing committed
45
#include "rpl_handler.h"
46 47 48 49 50 51 52 53 54
#include <signal.h>
#include <mysql.h>
#include <myisam.h>

#include "sql_base.h"                           // close_thread_tables
#include "tztime.h"                             // struct Time_zone
#include "log_event.h"                          // Rotate_log_event,
                                                // Create_file_log_event,
                                                // Format_description_log_event
unknown's avatar
SCRUM  
unknown committed
55

56 57 58
#ifdef WITH_WSREP
#include "wsrep_mysqld.h"
#endif
59 60 61
#ifdef HAVE_REPLICATION

#include "rpl_tblmap.h"
62
#include "debug_sync.h"
63 64
#include "rpl_parallel.h"

65

66
#define FLAGSTR(V,F) ((V)&(F)?#F" ":"")
67

68
#define MAX_SLAVE_RETRY_PAUSE 5
69 70 71 72
/*
  a parameter of sql_slave_killed() to defer the killed status
*/
#define SLAVE_WAIT_GROUP_DONE 60
73 74
bool use_slave_mask = 0;
MY_BITMAP slave_error_mask;
75
char slave_skip_error_names[SHOW_VAR_FUNC_BUFF_SIZE];
76

77
char* slave_load_tmpdir = 0;
78
Master_info *active_mi= 0;
79
Master_info_index *master_info_index;
80
my_bool replicate_same_server_id;
81
ulonglong relay_log_space_limit = 0;
82
LEX_STRING default_master_connection_name= { (char*) "", 0 };
unknown's avatar
unknown committed
83 84 85 86 87

/*
  When slave thread exits, we need to remember the temporary tables so we
  can re-use them on slave start.

88
  TODO: move the vars below under Master_info
unknown's avatar
unknown committed
89
*/
90

91
int disconnect_slave_event_count = 0, abort_slave_event_count = 0;
unknown's avatar
unknown committed
92

He Zhenxing's avatar
He Zhenxing committed
93 94
static pthread_key(Master_info*, RPL_MASTER_INFO);

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
enum enum_slave_reconnect_actions
{
  SLAVE_RECON_ACT_REG= 0,
  SLAVE_RECON_ACT_DUMP= 1,
  SLAVE_RECON_ACT_EVENT= 2,
  SLAVE_RECON_ACT_MAX
};

enum enum_slave_reconnect_messages
{
  SLAVE_RECON_MSG_WAIT= 0,
  SLAVE_RECON_MSG_KILLED_WAITING= 1,
  SLAVE_RECON_MSG_AFTER= 2,
  SLAVE_RECON_MSG_FAILED= 3,
  SLAVE_RECON_MSG_COMMAND= 4,
  SLAVE_RECON_MSG_KILLED_AFTER= 5,
  SLAVE_RECON_MSG_MAX
};

static const char *reconnect_messages[SLAVE_RECON_ACT_MAX][SLAVE_RECON_MSG_MAX]=
{
  {
    "Waiting to reconnect after a failed registration on master",
    "Slave I/O thread killed while waitnig to reconnect after a failed \
registration on master",
    "Reconnecting after a failed registration on master",
    "failed registering on master, reconnecting to try again, \
unknown's avatar
unknown committed
122
log '%s' at position %s%s",
123 124 125 126 127 128 129
    "COM_REGISTER_SLAVE",
    "Slave I/O thread killed during or after reconnect"
  },
  {
    "Waiting to reconnect after a failed binlog dump request",
    "Slave I/O thread killed while retrying master dump",
    "Reconnecting after a failed binlog dump request",
unknown's avatar
unknown committed
130
    "failed dump request, reconnecting to try again, log '%s' at position %s%s",
131 132 133 134 135 136 137 138
    "COM_BINLOG_DUMP",
    "Slave I/O thread killed during or after reconnect"
  },
  {
    "Waiting to reconnect after a failed master event read",
    "Slave I/O thread killed while waiting to reconnect after a failed read",
    "Reconnecting after a failed master event read",
    "Slave I/O thread: Failed reading log event, reconnecting to retry, \
unknown's avatar
unknown committed
139
log '%s' at position %s%s",
140 141 142 143 144 145 146
    "",
    "Slave I/O thread killed during or after a reconnect done to recover from \
failed read"
  }
};
 

147
typedef enum { SLAVE_THD_IO, SLAVE_THD_SQL} SLAVE_THD_TYPE;
unknown's avatar
unknown committed
148

149 150
static int process_io_rotate(Master_info* mi, Rotate_log_event* rev);
static int process_io_create_file(Master_info* mi, Create_file_log_event* cev);
151
static bool wait_for_relay_log_space(Relay_log_info* rli);
Michael Widenius's avatar
Michael Widenius committed
152 153
static bool io_slave_killed(Master_info* mi);
static bool sql_slave_killed(rpl_group_info *rgi);
154
static int init_slave_thread(THD*, Master_info *, SLAVE_THD_TYPE);
155
static void print_slave_skip_errors(void);
156
static int safe_connect(THD* thd, MYSQL* mysql, Master_info* mi);
157 158
static int safe_reconnect(THD*, MYSQL*, Master_info*, bool);
static int connect_to_master(THD*, MYSQL*, Master_info*, bool, bool);
unknown's avatar
unknown committed
159
static Log_event* next_event(rpl_group_info* rgi, ulonglong *event_size);
160
static int queue_event(Master_info* mi,const char* buf,ulong event_len);
161 162
static int terminate_slave_thread(THD *, mysql_mutex_t *, mysql_cond_t *,
                                  volatile uint *, bool);
Michael Widenius's avatar
Michael Widenius committed
163
static bool check_io_slave_killed(Master_info *mi, const char *info);
164 165
static bool send_show_master_info_header(THD *, bool, size_t);
static bool send_show_master_info_data(THD *, Master_info *, bool, String *);
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
/*
  Function to set the slave's max_allowed_packet based on the value
  of slave_max_allowed_packet.

    @in_param    thd    Thread handler for slave
    @in_param    mysql  MySQL connection handle
*/

static void set_slave_max_allowed_packet(THD *thd, MYSQL *mysql)
{
  DBUG_ENTER("set_slave_max_allowed_packet");
  // thd and mysql must be valid
  DBUG_ASSERT(thd && mysql);

  thd->variables.max_allowed_packet= slave_max_allowed_packet;
  thd->net.max_packet_size= slave_max_allowed_packet;
  /*
    Adding MAX_LOG_EVENT_HEADER_LEN to the max_packet_size on the I/O
    thread and the mysql->option max_allowed_packet, since a
    replication event can become this much  larger than
    the corresponding packet (query) sent from client to master.
  */
  thd->net.max_packet_size+= MAX_LOG_EVENT_HEADER;
  /*
    Skipping the setting of mysql->net.max_packet size to slave
    max_allowed_packet since this is done during mysql_real_connect.
  */
  mysql->options.max_allowed_packet=
    slave_max_allowed_packet+MAX_LOG_EVENT_HEADER;
  DBUG_VOID_RETURN;
}
197 198

/*
unknown's avatar
unknown committed
199
  Find out which replications threads are running
200

unknown's avatar
unknown committed
201 202
  SYNOPSIS
    init_thread_mask()
unknown's avatar
unknown committed
203 204 205
    mask                Return value here
    mi                  master_info for slave
    inverse             If set, returns which threads are not running
unknown's avatar
unknown committed
206 207 208 209 210 211

  IMPLEMENTATION
    Get a bit mask for which threads are running so that we can later restart
    these threads.

  RETURN
unknown's avatar
unknown committed
212 213
    mask        If inverse == 0, running threads
                If inverse == 1, stopped threads
214 215
*/

216
void init_thread_mask(int* mask,Master_info* mi,bool inverse)
217 218 219
{
  bool set_io = mi->slave_running, set_sql = mi->rli.slave_running;
  register int tmp_mask=0;
220 221
  DBUG_ENTER("init_thread_mask");

222 223 224 225
  if (set_io)
    tmp_mask |= SLAVE_IO;
  if (set_sql)
    tmp_mask |= SLAVE_SQL;
226 227
  if (inverse)
    tmp_mask^= (SLAVE_IO | SLAVE_SQL);
228
  *mask = tmp_mask;
229
  DBUG_VOID_RETURN;
230 231
}

232

unknown's avatar
unknown committed
233
/*
234
  lock_slave_threads()
unknown's avatar
unknown committed
235
*/
236

237
void lock_slave_threads(Master_info* mi)
238
{
239 240
  DBUG_ENTER("lock_slave_threads");

241
  //TODO: see if we can do this without dual mutex
Marc Alff's avatar
Marc Alff committed
242 243
  mysql_mutex_lock(&mi->run_lock);
  mysql_mutex_lock(&mi->rli.run_lock);
244
  DBUG_VOID_RETURN;
245 246
}

247

unknown's avatar
unknown committed
248
/*
249
  unlock_slave_threads()
unknown's avatar
unknown committed
250
*/
251

252
void unlock_slave_threads(Master_info* mi)
253
{
254 255
  DBUG_ENTER("unlock_slave_threads");

256
  //TODO: see if we can do this without dual mutex
Marc Alff's avatar
Marc Alff committed
257 258
  mysql_mutex_unlock(&mi->rli.run_lock);
  mysql_mutex_unlock(&mi->run_lock);
259
  DBUG_VOID_RETURN;
260 261
}

Marc Alff's avatar
Marc Alff committed
262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282
#ifdef HAVE_PSI_INTERFACE
static PSI_thread_key key_thread_slave_io, key_thread_slave_sql;

static PSI_thread_info all_slave_threads[]=
{
  { &key_thread_slave_io, "slave_io", PSI_FLAG_GLOBAL},
  { &key_thread_slave_sql, "slave_sql", PSI_FLAG_GLOBAL}
};

static void init_slave_psi_keys(void)
{
  const char* category= "sql";
  int count;

  if (PSI_server == NULL)
    return;

  count= array_elements(all_slave_threads);
  PSI_server->register_thread(category, all_slave_threads, count);
}
#endif /* HAVE_PSI_INTERFACE */
283

284 285 286 287 288 289 290 291 292 293 294 295 296 297 298

static bool slave_init_thread_running;


pthread_handler_t
handle_slave_init(void *arg __attribute__((unused)))
{
  THD *thd;

  my_thread_init();
  thd= new THD;
  thd->thread_stack= (char*) &thd;           /* Set approximate stack start */
  mysql_mutex_lock(&LOCK_thread_count);
  thd->thread_id= thread_id++;
  mysql_mutex_unlock(&LOCK_thread_count);
299
  thd->system_thread = SYSTEM_THREAD_SLAVE_INIT;
300
  thd->store_globals();
301 302
  thd->security_ctx->skip_grants();
  thd->set_command(COM_DAEMON);
303 304 305 306 307 308

  thd_proc_info(thd, "Loading slave GTID position from table");
  if (rpl_load_gtid_slave_state(thd))
    sql_print_warning("Failed to load slave replication state from table "
                      "%s.%s: %u: %s", "mysql",
                      rpl_gtid_slave_state_table_name.str,
Sergei Golubchik's avatar
Sergei Golubchik committed
309 310
                      thd->get_stmt_da()->sql_errno(),
                      thd->get_stmt_da()->message());
311 312 313 314 315 316

  mysql_mutex_lock(&LOCK_thread_count);
  delete thd;
  mysql_mutex_unlock(&LOCK_thread_count);
  my_thread_end();

317
  mysql_mutex_lock(&LOCK_slave_init);
318
  slave_init_thread_running= false;
319 320
  mysql_cond_broadcast(&COND_slave_init);
  mysql_mutex_unlock(&LOCK_slave_init);
321 322 323 324 325

  return 0;
}


326
/*
327
  Start the slave init thread.
328

329 330 331
  This thread is used to load the GTID state from mysql.gtid_slave_pos at
  server start; reading from table requires valid THD, which is otherwise not
  available during server init.
332
*/
333 334 335 336 337 338
static int
run_slave_init_thread()
{
  pthread_t th;

  slave_init_thread_running= true;
unknown's avatar
unknown committed
339
  if (mysql_thread_create(key_thread_slave_init, &th, &connection_attrib,
340 341 342 343 344 345
                          handle_slave_init, NULL))
  {
    sql_print_error("Failed to create thread while initialising slave");
    return 1;
  }

346
  mysql_mutex_lock(&LOCK_slave_init);
347
  while (slave_init_thread_running)
348 349
    mysql_cond_wait(&COND_slave_init, &LOCK_slave_init);
  mysql_mutex_unlock(&LOCK_slave_init);
350 351 352 353 354

  return 0;
}


unknown's avatar
unknown committed
355
/* Initialize slave structures */
356

357 358
int init_slave()
{
359
  DBUG_ENTER("init_slave");
360
  int error= 0;
361

Marc Alff's avatar
Marc Alff committed
362 363 364 365
#ifdef HAVE_PSI_INTERFACE
  init_slave_psi_keys();
#endif

366 367 368
  if (run_slave_init_thread())
    return 1;

369 370 371
  if (global_rpl_thread_pool.init(opt_slave_parallel_threads))
    return 1;

372 373 374 375 376
  /*
    This is called when mysqld starts. Before client connections are
    accepted. However bootstrap may conflict with us if it does START SLAVE.
    So it's safer to take the lock.
  */
Marc Alff's avatar
Marc Alff committed
377
  mysql_mutex_lock(&LOCK_active_mi);
378

He Zhenxing's avatar
He Zhenxing committed
379 380
  if (pthread_key_create(&RPL_MASTER_INFO, NULL))
    goto err;
381

382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404
  master_info_index= new Master_info_index;
  if (!master_info_index || master_info_index->init_all_master_info())
  {
    sql_print_error("Failed to initialize multi master structures");
    mysql_mutex_unlock(&LOCK_active_mi);
    DBUG_RETURN(1);
  }
  if (!(active_mi= new Master_info(&default_master_connection_name,
                                   relay_log_recovery)) ||
      active_mi->error())
  {
    delete active_mi;
    active_mi= 0;
    goto err;
  }

  if (master_info_index->add_master_info(active_mi, FALSE))
  {
    delete active_mi;
    active_mi= 0;
    goto err;
  }

405 406 407 408 409 410 411 412 413
  /*
    If --slave-skip-errors=... was not used, the string value for the
    system variable has not been set up yet. Do it now.
  */
  if (!use_slave_mask)
  {
    print_slave_skip_errors();
  }

414
  /*
415 416 417
    If master_host is not specified, try to read it from the master_info file.
    If master_host is specified, create the master_info file if it doesn't
    exists.
418
  */
419

unknown's avatar
unknown committed
420
  if (init_master_info(active_mi,master_info_file,relay_log_info_file,
421
                       1, (SLAVE_IO | SLAVE_SQL)))
422
  {
423
    sql_print_error("Failed to initialize the master info structure");
unknown's avatar
unknown committed
424
    goto err;
425
  }
426

427 428
  /* If server id is not set, start_slave_thread() will say it */

429
  if (active_mi->host[0] && !opt_skip_slave_start)
430
  {
431
    if (start_slave_threads(1 /* need mutex */,
unknown's avatar
unknown committed
432 433 434 435 436
                            0 /* no wait for start*/,
                            active_mi,
                            master_info_file,
                            relay_log_info_file,
                            SLAVE_IO | SLAVE_SQL))
unknown's avatar
unknown committed
437
    {
438
      sql_print_error("Failed to create slave threads");
unknown's avatar
unknown committed
439 440
      goto err;
    }
441 442
  }

443
end:
Marc Alff's avatar
Marc Alff committed
444
  mysql_mutex_unlock(&LOCK_active_mi);
445
  DBUG_RETURN(error);
446 447 448 449 450

err:
  sql_print_error("Failed to allocate memory for the Master Info structure");
  error= 1;
  goto end;
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
/*
  Updates the master info based on the information stored in the
  relay info and ignores relay logs previously retrieved by the IO 
  thread, which thus starts fetching again based on to the  
  group_master_log_pos and group_master_log_name. Eventually, the old
  relay logs will be purged by the normal purge mechanism.

  In the feature, we should improve this routine in order to avoid throwing
  away logs that are safely stored in the disk. Note also that this recovery 
  routine relies on the correctness of the relay-log.info and only tolerates 
  coordinate problems in master.info.
  
  In this function, there is no need for a mutex as the caller 
  (i.e. init_slave) already has one acquired.
  
  Specifically, the following structures are updated:
 
  1 - mi->master_log_pos  <-- rli->group_master_log_pos
  2 - mi->master_log_name <-- rli->group_master_log_name
  3 - It moves the relay log to the new relay log file, by
      rli->group_relay_log_pos  <-- BIN_LOG_HEADER_SIZE;
      rli->event_relay_log_pos  <-- BIN_LOG_HEADER_SIZE;
      rli->group_relay_log_name <-- rli->relay_log.get_log_fname();
      rli->event_relay_log_name <-- rli->relay_log.get_log_fname();
  
   If there is an error, it returns (1), otherwise returns (0).
 */
480
int init_recovery(Master_info* mi, const char** errmsg)
481 482 483 484 485 486
{
  DBUG_ENTER("init_recovery");
 
  Relay_log_info *rli= &mi->rli;
  if (rli->group_master_log_name[0])
  {
487
    mi->master_log_pos= MY_MAX(BIN_LOG_HEADER_SIZE,
488
                             rli->group_master_log_pos);
489
    strmake_buf(mi->master_log_name, rli->group_master_log_name);
490 491 492 493
 
    sql_print_warning("Recovery from master pos %ld and file %s.",
                      (ulong) mi->master_log_pos, mi->master_log_name);
 
494 495
    strmake_buf(rli->group_relay_log_name, rli->relay_log.get_log_fname());
    strmake_buf(rli->event_relay_log_name, rli->relay_log.get_log_fname());
496 497 498
 
    rli->group_relay_log_pos= rli->event_relay_log_pos= BIN_LOG_HEADER_SIZE;
  }
499

500 501
  DBUG_RETURN(0);
}
unknown's avatar
unknown committed
502

503
 
504 505 506 507 508 509 510 511 512 513 514 515
/**
  Convert slave skip errors bitmap into a printable string.
*/

static void print_slave_skip_errors(void)
{
  /*
    To be safe, we want 10 characters of room in the buffer for a number
    plus terminators. Also, we need some space for constant strings.
    10 characters must be sufficient for a number plus {',' | '...'}
    plus a NUL terminator. That is a max 6 digit number.
  */
516
  const size_t MIN_ROOM= 10;
517 518 519 520
  DBUG_ENTER("print_slave_skip_errors");
  DBUG_ASSERT(sizeof(slave_skip_error_names) > MIN_ROOM);
  DBUG_ASSERT(MAX_SLAVE_ERROR <= 999999); // 6 digits

521 522 523
  /* Make @@slave_skip_errors show the nice human-readable value.  */
  opt_slave_skip_errors= slave_skip_error_names;

524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541
  if (!use_slave_mask || bitmap_is_clear_all(&slave_error_mask))
  {
    /* purecov: begin tested */
    memcpy(slave_skip_error_names, STRING_WITH_LEN("OFF"));
    /* purecov: end */
  }
  else if (bitmap_is_set_all(&slave_error_mask))
  {
    /* purecov: begin tested */
    memcpy(slave_skip_error_names, STRING_WITH_LEN("ALL"));
    /* purecov: end */
  }
  else
  {
    char *buff= slave_skip_error_names;
    char *bend= buff + sizeof(slave_skip_error_names);
    int  errnum;

542
    for (errnum= 0; errnum < MAX_SLAVE_ERROR; errnum++)
543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564
    {
      if (bitmap_is_set(&slave_error_mask, errnum))
      {
        if (buff + MIN_ROOM >= bend)
          break; /* purecov: tested */
        buff= int10_to_str(errnum, buff, 10);
        *buff++= ',';
      }
    }
    if (buff != slave_skip_error_names)
      buff--; // Remove last ','
    if (errnum < MAX_SLAVE_ERROR)
    {
      /* Couldn't show all errors */
      buff= strmov(buff, "..."); /* purecov: tested */
    }
    *buff=0;
  }
  DBUG_PRINT("init", ("error_names: '%s'", slave_skip_error_names));
  DBUG_VOID_RETURN;
}

unknown's avatar
unknown committed
565
/*
566
  Init function to set up array for errors that should be skipped for slave
unknown's avatar
unknown committed
567

unknown's avatar
unknown committed
568 569
  SYNOPSIS
    init_slave_skip_errors()
unknown's avatar
unknown committed
570
    arg         List of errors numbers to skip, separated with ','
unknown's avatar
unknown committed
571 572 573 574

  NOTES
    Called from get_options() in mysqld.cc on start-up
*/
575

unknown's avatar
unknown committed
576
void init_slave_skip_errors(const char* arg)
577
{
unknown's avatar
unknown committed
578
  const char *p;
579 580
  DBUG_ENTER("init_slave_skip_errors");

581
  if (my_bitmap_init(&slave_error_mask,0,MAX_SLAVE_ERROR,0))
582 583 584 585 586
  {
    fprintf(stderr, "Badly out of memory, please check your system status\n");
    exit(1);
  }
  use_slave_mask = 1;
587
  for (;my_isspace(system_charset_info,*arg);++arg)
588
    /* empty */;
589
  if (!my_strnncoll(system_charset_info,(uchar*)arg,4,(const uchar*)"all",4))
590 591
  {
    bitmap_set_all(&slave_error_mask);
592
    print_slave_skip_errors();
593
    DBUG_VOID_RETURN;
594 595 596 597 598 599 600 601
  }
  for (p= arg ; *p; )
  {
    long err_code;
    if (!(p= str2int(p, 10, 0, LONG_MAX, &err_code)))
      break;
    if (err_code < MAX_SLAVE_ERROR)
       bitmap_set_bit(&slave_error_mask,(uint)err_code);
602
    while (!my_isdigit(system_charset_info,*p) && *p)
603 604
      p++;
  }
605 606
  /* Convert slave skip errors bitmap into a printable string. */
  print_slave_skip_errors();
607
  DBUG_VOID_RETURN;
608 609
}

610
int terminate_slave_threads(Master_info* mi,int thread_mask,bool skip_lock)
611
{
612 613
  DBUG_ENTER("terminate_slave_threads");

614
  if (!mi->inited)
615
    DBUG_RETURN(0); /* successfully do nothing */
616
  int error,force_all = (thread_mask & SLAVE_FORCE_ALL);
Marc Alff's avatar
Marc Alff committed
617
  mysql_mutex_t *sql_lock = &mi->rli.run_lock, *io_lock = &mi->run_lock;
618
  mysql_mutex_t *log_lock= mi->rli.relay_log.get_log_lock();
unknown's avatar
unknown committed
619

620
  if (thread_mask & (SLAVE_SQL|SLAVE_FORCE_ALL))
621
  {
unknown's avatar
unknown committed
622
    DBUG_PRINT("info",("Terminating SQL thread"));
623 624 625 626 627 628 629 630
    if (opt_slave_parallel_threads > 0 &&
        mi->rli.abort_slave && mi->rli.stop_for_until)
    {
      mi->rli.stop_for_until= false;
      mi->rli.parallel.stop_during_until();
    }
    else
      mi->rli.abort_slave=1;
Michael Widenius's avatar
Michael Widenius committed
631
    if ((error=terminate_slave_thread(mi->rli.sql_driver_thd, sql_lock,
unknown's avatar
unknown committed
632
                                      &mi->rli.stop_cond,
633 634
                                      &mi->rli.slave_running,
                                      skip_lock)) &&
unknown's avatar
unknown committed
635
        !force_all)
636
      DBUG_RETURN(error);
637

638
    mysql_mutex_lock(log_lock);
639

Libing Song's avatar
Libing Song committed
640
    DBUG_PRINT("info",("Flushing relay-log info file."));
641
    if (current_thd)
Sergei Golubchik's avatar
Sergei Golubchik committed
642
      THD_STAGE_INFO(current_thd, stage_flushing_relay_log_info_file);
Libing Song's avatar
Libing Song committed
643
    if (flush_relay_log_info(&mi->rli))
644
      DBUG_RETURN(ER_ERROR_DURING_FLUSH_LOGS);
Libing Song's avatar
Libing Song committed
645 646
    
    if (my_sync(mi->rli.info_fd, MYF(MY_WME)))
647 648
      DBUG_RETURN(ER_ERROR_DURING_FLUSH_LOGS);

649
    mysql_mutex_unlock(log_lock);
650
  }
651 652 653
  if (opt_slave_parallel_threads > 0 &&
      !master_info_index->any_slave_sql_running())
    rpl_parallel_inactivate_pool(&global_rpl_thread_pool);
654
  if (thread_mask & (SLAVE_IO|SLAVE_FORCE_ALL))
655
  {
unknown's avatar
unknown committed
656
    DBUG_PRINT("info",("Terminating IO thread"));
657
    mi->abort_slave=1;
658
    if ((error=terminate_slave_thread(mi->io_thd, io_lock,
unknown's avatar
unknown committed
659
                                      &mi->stop_cond,
660 661
                                      &mi->slave_running,
                                      skip_lock)) &&
unknown's avatar
unknown committed
662
        !force_all)
663
      DBUG_RETURN(error);
664

665
    mysql_mutex_lock(log_lock);
666 667 668

    DBUG_PRINT("info",("Flushing relay log and master info file."));
    if (current_thd)
Sergei Golubchik's avatar
Sergei Golubchik committed
669
      THD_STAGE_INFO(current_thd, stage_flushing_relay_log_and_master_info_repository);
670
    if (flush_master_info(mi, TRUE, FALSE))
671 672
      DBUG_RETURN(ER_ERROR_DURING_FLUSH_LOGS);

Luis Soares's avatar
Luis Soares committed
673 674
    if (mi->rli.relay_log.is_open() &&
        my_sync(mi->rli.relay_log.get_log_file()->file, MYF(MY_WME)))
675 676 677 678 679
      DBUG_RETURN(ER_ERROR_DURING_FLUSH_LOGS);

    if (my_sync(mi->fd, MYF(MY_WME)))
      DBUG_RETURN(ER_ERROR_DURING_FLUSH_LOGS);

680
    mysql_mutex_unlock(log_lock);
681
  }
682
  DBUG_RETURN(0); 
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
/**
   Wait for a slave thread to terminate.

   This function is called after requesting the thread to terminate
   (by setting @c abort_slave member of @c Relay_log_info or @c
   Master_info structure to 1). Termination of the thread is
   controlled with the the predicate <code>*slave_running</code>.

   Function will acquire @c term_lock before waiting on the condition
   unless @c skip_lock is true in which case the mutex should be owned
   by the caller of this function and will remain acquired after
   return from the function.

   @param term_lock
          Associated lock to use when waiting for @c term_cond

   @param term_cond
          Condition that is signalled when the thread has terminated

   @param slave_running
          Pointer to predicate to check for slave thread termination

   @param skip_lock
          If @c true the lock will not be acquired before waiting on
          the condition. In this case, it is assumed that the calling
          function acquires the lock before calling this function.

Andrei Elkin's avatar
Andrei Elkin committed
713
   @retval 0 All OK ER_SLAVE_NOT_RUNNING otherwise.
714

Andrei Elkin's avatar
Andrei Elkin committed
715 716 717
   @note  If the executing thread has to acquire term_lock (skip_lock
          is false), the negative running status does not represent
          any issue therefore no error is reported.
718

719 720 721
 */
static int
terminate_slave_thread(THD *thd,
Marc Alff's avatar
Marc Alff committed
722 723
                       mysql_mutex_t *term_lock,
                       mysql_cond_t *term_cond,
724 725
                       volatile uint *slave_running,
                       bool skip_lock)
726
{
727
  DBUG_ENTER("terminate_slave_thread");
728
  if (!skip_lock)
729
  {
Marc Alff's avatar
Marc Alff committed
730
    mysql_mutex_lock(term_lock);
731 732 733
  }
  else
  {
Marc Alff's avatar
Marc Alff committed
734
    mysql_mutex_assert_owner(term_lock);
735
  }
736 737 738
  if (!*slave_running)
  {
    if (!skip_lock)
739
    {
740 741 742 743
      /*
        if run_lock (term_lock) is acquired locally then either
        slave_running status is fine
      */
Marc Alff's avatar
Marc Alff committed
744
      mysql_mutex_unlock(term_lock);
745 746 747 748
      DBUG_RETURN(0);
    }
    else
    {
749
      DBUG_RETURN(ER_SLAVE_NOT_RUNNING);
750 751 752
    }
  }
  DBUG_ASSERT(thd != 0);
753
  THD_CHECK_SENTRY(thd);
754

unknown's avatar
unknown committed
755
  /*
756
    Is is critical to test if the slave is running. Otherwise, we might
unknown's avatar
unknown committed
757
    be referening freed memory trying to kick it
758
  */
unknown's avatar
unknown committed
759

unknown's avatar
unknown committed
760
  while (*slave_running)                        // Should always be true
761
  {
Michael Widenius's avatar
Michael Widenius committed
762
    int error __attribute__((unused));
763
    DBUG_PRINT("loop", ("killing slave thread"));
764

Marc Alff's avatar
Marc Alff committed
765
    mysql_mutex_lock(&thd->LOCK_thd_data);
766 767 768 769 770 771
#ifndef DONT_USE_THR_ALARM
    /*
      Error codes from pthread_kill are:
      EINVAL: invalid signal number (can't happen)
      ESRCH: thread already killed (can happen, should be ignored)
    */
772
    int err __attribute__((unused))= pthread_kill(thd->real_id, thr_client_alarm);
773 774
    DBUG_ASSERT(err != EINVAL);
#endif
Sergei Golubchik's avatar
Sergei Golubchik committed
775 776
    thd->awake(NOT_KILLED);

Marc Alff's avatar
Marc Alff committed
777
    mysql_mutex_unlock(&thd->LOCK_thd_data);
778

unknown's avatar
unknown committed
779 780 781
    /*
      There is a small chance that slave thread might miss the first
      alarm. To protect againts it, resend the signal until it reacts
782 783
    */
    struct timespec abstime;
unknown's avatar
unknown committed
784
    set_timespec(abstime,2);
Marc Alff's avatar
Marc Alff committed
785
    error= mysql_cond_timedwait(term_cond, term_lock, &abstime);
786
    DBUG_ASSERT(error == ETIMEDOUT || error == 0);
787
  }
788 789 790 791

  DBUG_ASSERT(*slave_running == 0);

  if (!skip_lock)
Marc Alff's avatar
Marc Alff committed
792
    mysql_mutex_unlock(term_lock);
793
  DBUG_RETURN(0);
794 795
}

796

Marc Alff's avatar
Marc Alff committed
797 798 799 800 801 802 803
int start_slave_thread(
#ifdef HAVE_PSI_INTERFACE
                       PSI_thread_key thread_key,
#endif
                       pthread_handler h_func, mysql_mutex_t *start_lock,
                       mysql_mutex_t *cond_lock,
                       mysql_cond_t *start_cond,
unknown's avatar
unknown committed
804 805
                       volatile uint *slave_running,
                       volatile ulong *slave_run_id,
Konstantin Osipov's avatar
Konstantin Osipov committed
806
                       Master_info* mi)
807 808
{
  pthread_t th;
unknown's avatar
unknown committed
809
  ulong start_id;
810
  int error;
unknown's avatar
unknown committed
811 812
  DBUG_ENTER("start_slave_thread");

813 814
  DBUG_ASSERT(mi->inited);

815
  if (start_lock)
Marc Alff's avatar
Marc Alff committed
816
    mysql_mutex_lock(start_lock);
817
  if (!global_system_variables.server_id)
818 819
  {
    if (start_cond)
Marc Alff's avatar
Marc Alff committed
820
      mysql_cond_broadcast(start_cond);
821
    if (start_lock)
Marc Alff's avatar
Marc Alff committed
822
      mysql_mutex_unlock(start_lock);
823
    sql_print_error("Server id not set, will not start slave");
unknown's avatar
unknown committed
824
    DBUG_RETURN(ER_BAD_SLAVE);
825
  }
unknown's avatar
unknown committed
826

827
  if (*slave_running)
828 829
  {
    if (start_cond)
Marc Alff's avatar
Marc Alff committed
830
      mysql_cond_broadcast(start_cond);
831
    if (start_lock)
Marc Alff's avatar
Marc Alff committed
832
      mysql_mutex_unlock(start_lock);
unknown's avatar
unknown committed
833
    DBUG_RETURN(ER_SLAVE_MUST_STOP);
834
  }
unknown's avatar
unknown committed
835 836
  start_id= *slave_run_id;
  DBUG_PRINT("info",("Creating new slave thread"));
837 838
  if ((error = mysql_thread_create(thread_key,
                           &th, &connection_attrib, h_func, (void*)mi)))
839
  {
840
    sql_print_error("Can't create slave thread (errno= %d).", error);
841
    if (start_lock)
Marc Alff's avatar
Marc Alff committed
842
      mysql_mutex_unlock(start_lock);
unknown's avatar
unknown committed
843
    DBUG_RETURN(ER_SLAVE_THREAD);
844
  }
unknown's avatar
unknown committed
845
  if (start_cond && cond_lock) // caller has cond_lock
846 847
  {
    THD* thd = current_thd;
unknown's avatar
unknown committed
848
    while (start_id == *slave_run_id)
849
    {
unknown's avatar
unknown committed
850
      DBUG_PRINT("sleep",("Waiting for slave thread to start"));
Sergei Golubchik's avatar
Sergei Golubchik committed
851 852 853 854
      PSI_stage_info saved_stage= {0, "", 0};
      thd->ENTER_COND(start_cond, cond_lock,
                      & stage_waiting_for_slave_thread_to_start,
                      & saved_stage);
855 856 857 858 859 860 861 862 863
      /*
        It is not sufficient to test this at loop bottom. We must test
        it after registering the mutex in enter_cond(). If the kill
        happens after testing of thd->killed and before the mutex is
        registered, we could otherwise go waiting though thd->killed is
        set.
      */
      if (!thd->killed)
        mysql_cond_wait(start_cond, cond_lock);
Sergei Golubchik's avatar
Sergei Golubchik committed
864
      thd->EXIT_COND(& saved_stage);
Marc Alff's avatar
Marc Alff committed
865
      mysql_mutex_lock(cond_lock); // re-acquire it as exit_cond() released
866
      if (thd->killed)
867 868
      {
        if (start_lock)
Marc Alff's avatar
Marc Alff committed
869
          mysql_mutex_unlock(start_lock);
unknown's avatar
unknown committed
870
        DBUG_RETURN(thd->killed_errno());
871
      }
872 873 874
    }
  }
  if (start_lock)
Marc Alff's avatar
Marc Alff committed
875
    mysql_mutex_unlock(start_lock);
unknown's avatar
unknown committed
876
  DBUG_RETURN(0);
877
}
unknown's avatar
unknown committed
878

879

unknown's avatar
unknown committed
880
/*
881
  start_slave_threads()
unknown's avatar
unknown committed
882

unknown's avatar
unknown committed
883 884 885 886
  NOTES
    SLAVE_FORCE_ALL is not implemented here on purpose since it does not make
    sense to do that for starting a slave--we always care if it actually
    started the threads that were not previously running
887
*/
unknown's avatar
unknown committed
888

889
int start_slave_threads(bool need_slave_mutex, bool wait_for_start,
890
                        Master_info* mi, const char* master_info_fname,
unknown's avatar
unknown committed
891
                        const char* slave_info_fname, int thread_mask)
892
{
Marc Alff's avatar
Marc Alff committed
893 894
  mysql_mutex_t *lock_io=0, *lock_sql=0, *lock_cond_io=0, *lock_cond_sql=0;
  mysql_cond_t* cond_io=0, *cond_sql=0;
895
  int error=0;
unknown's avatar
unknown committed
896
  const char *errmsg;
897
  DBUG_ENTER("start_slave_threads");
unknown's avatar
unknown committed
898

899 900 901 902 903 904 905 906 907 908 909 910
  if (need_slave_mutex)
  {
    lock_io = &mi->run_lock;
    lock_sql = &mi->rli.run_lock;
  }
  if (wait_for_start)
  {
    cond_io = &mi->start_cond;
    cond_sql = &mi->rli.start_cond;
    lock_cond_io = &mi->run_lock;
    lock_cond_sql = &mi->rli.run_lock;
  }
911

unknown's avatar
unknown committed
912 913 914 915 916 917 918 919 920
  /*
    If we are using GTID and both SQL and IO threads are stopped, then get
    rid of all relay logs.

    Relay logs are not very useful when using GTID, except as a buffer
    between the fetch in the IO thread and the apply in SQL thread. However
    while one of the threads is running, they are in use and cannot be
    removed.
  */
unknown's avatar
unknown committed
921 922
  if (mi->using_gtid != Master_info::USE_GTID_NO &&
      !mi->slave_running && !mi->rli.slave_running)
unknown's avatar
unknown committed
923
  {
924 925 926 927 928 929 930
    /*
      purge_relay_logs() clears the mi->rli.group_master_log_pos.
      So save and restore them, like we do in CHANGE MASTER.
      (We are not going to use them for GTID, but it might be worth to
      keep them in case connection with GTID fails and user wants to go
      back and continue with previous old-style replication coordinates).
    */
Sergei Golubchik's avatar
Sergei Golubchik committed
931 932
    mi->master_log_pos = MY_MAX(BIN_LOG_HEADER_SIZE,
                                mi->rli.group_master_log_pos);
933 934
    strmake(mi->master_log_name, mi->rli.group_master_log_name,
            sizeof(mi->master_log_name)-1);
unknown's avatar
unknown committed
935
    purge_relay_logs(&mi->rli, NULL, 0, &errmsg);
936 937 938 939
    mi->rli.group_master_log_pos= mi->master_log_pos;
    strmake(mi->rli.group_master_log_name, mi->master_log_name,
            sizeof(mi->rli.group_master_log_name)-1);

unknown's avatar
unknown committed
940 941 942 943
    error= rpl_load_gtid_state(&mi->gtid_current_pos, mi->using_gtid ==
                                             Master_info::USE_GTID_CURRENT_POS);
    mi->events_queued_since_last_gtid= 0;
    mi->gtid_reconnect_event_skip_count= 0;
944 945

    mi->rli.restart_gtid_pos.reset();
unknown's avatar
unknown committed
946 947
  }

unknown's avatar
unknown committed
948
  if (!error && (thread_mask & SLAVE_IO))
Marc Alff's avatar
Marc Alff committed
949 950 951 952 953 954 955 956
    error= start_slave_thread(
#ifdef HAVE_PSI_INTERFACE
                              key_thread_slave_io,
#endif
                              handle_slave_io, lock_io, lock_cond_io,
                              cond_io,
                              &mi->slave_running, &mi->slave_run_id,
                              mi);
957
  if (!error && (thread_mask & SLAVE_SQL))
958
  {
959 960 961 962
    if (opt_slave_parallel_threads > 0)
      error= rpl_parallel_activate_pool(&global_rpl_thread_pool);
    if (!error)
      error= start_slave_thread(
Marc Alff's avatar
Marc Alff committed
963 964 965 966 967 968 969
#ifdef HAVE_PSI_INTERFACE
                              key_thread_slave_sql,
#endif
                              handle_slave_sql, lock_sql, lock_cond_sql,
                              cond_sql,
                              &mi->rli.slave_running, &mi->rli.slave_run_id,
                              mi);
970
    if (error)
971
      terminate_slave_threads(mi, thread_mask & SLAVE_IO, !need_slave_mutex);
972
  }
973
  DBUG_RETURN(error);
974
}
975

976

unknown's avatar
unknown committed
977
/*
978
  Release slave threads at time of executing shutdown.
unknown's avatar
unknown committed
979 980 981 982

  SYNOPSIS
    end_slave()
*/
983

984 985
void end_slave()
{
986 987
  DBUG_ENTER("end_slave");

988 989 990 991 992 993
  /*
    This is called when the server terminates, in close_connections().
    It terminates slave threads. However, some CHANGE MASTER etc may still be
    running presently. If a START SLAVE was in progress, the mutex lock below
    will make us wait until slave threads have started, and START SLAVE
    returns, then we terminate them here.
994 995 996

    We can also be called by cleanup(), which only happens if some
    startup parameter to the server was wrong.
997
  */
Marc Alff's avatar
Marc Alff committed
998
  mysql_mutex_lock(&LOCK_active_mi);
999 1000 1001 1002
  /* This will call terminate_slave_threads() on all connections */
  delete master_info_index;
  master_info_index= 0;
  active_mi= 0;
Marc Alff's avatar
Marc Alff committed
1003
  mysql_mutex_unlock(&LOCK_active_mi);
1004
  global_rpl_thread_pool.destroy();
1005
  free_all_rpl_filters();
1006 1007 1008
  DBUG_VOID_RETURN;
}

Michael Widenius's avatar
Michael Widenius committed
1009
static bool io_slave_killed(Master_info* mi)
unknown's avatar
unknown committed
1010
{
1011 1012
  DBUG_ENTER("io_slave_killed");

1013
  DBUG_ASSERT(mi->slave_running); // tracking buffer overrun
Michael Widenius's avatar
Michael Widenius committed
1014
  DBUG_RETURN(mi->abort_slave || abort_loop || mi->io_thd->killed);
unknown's avatar
unknown committed
1015 1016
}

1017 1018 1019 1020 1021 1022
/**
   The function analyzes a possible killed status and makes
   a decision whether to accept it or not.
   Normally upon accepting the sql thread goes to shutdown.
   In the event of deffering decision @rli->last_event_start_time waiting
   timer is set to force the killed status be accepted upon its expiration.
1023

1024
   @param rgi   pointer to relay_group_info instance
1025

1026 1027 1028
   @return TRUE the killed status is recognized, FALSE a possible killed
           status is deferred.
*/
Michael Widenius's avatar
Michael Widenius committed
1029
static bool sql_slave_killed(rpl_group_info *rgi)
1030
{
1031
  bool ret= FALSE;
Michael Widenius's avatar
Michael Widenius committed
1032 1033
  Relay_log_info *rli= rgi->rli;
  THD *thd= rgi->thd;
1034 1035
  DBUG_ENTER("sql_slave_killed");

Michael Widenius's avatar
Michael Widenius committed
1036
  DBUG_ASSERT(rli->sql_driver_thd == thd);
1037
  DBUG_ASSERT(rli->slave_running == 1);// tracking buffer overrun
Michael Widenius's avatar
Michael Widenius committed
1038
  if (abort_loop || rli->sql_driver_thd->killed || rli->abort_slave)
1039
  {
1040
    /*
Michael Widenius's avatar
Michael Widenius committed
1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053
      The transaction should always be binlogged if OPTION_KEEP_LOG is
      set (it implies that something can not be rolled back). And such
      case should be regarded similarly as modifing a
      non-transactional table because retrying of the transaction will
      lead to an error or inconsistency as well.

      Example: OPTION_KEEP_LOG is set if a temporary table is created
      or dropped.

      Note that transaction.all.modified_non_trans_table may be 1
      if last statement was a single row transaction without begin/end.
      Testing this flag must always be done in connection with
      rli->is_in_group().
1054
    */
Michael Widenius's avatar
Michael Widenius committed
1055

unknown's avatar
unknown committed
1056
    if ((thd->transaction.all.modified_non_trans_table ||
Michael Widenius's avatar
Michael Widenius committed
1057 1058
         (thd->variables.option_bits & OPTION_KEEP_LOG)) &&
        rli->is_in_group())
1059
    {
1060
      char msg_stopped[]=
1061 1062 1063 1064
        "... Slave SQL Thread stopped with incomplete event group "
        "having non-transactional changes. "
        "If the group consists solely of row-based events, you can try "
        "to restart the slave with --slave-exec-mode=IDEMPOTENT, which "
1065 1066 1067
        "ignores duplicate key, key not found, and similar errors (see "
        "documentation for details).";

Michael Widenius's avatar
Michael Widenius committed
1068
      DBUG_PRINT("info", ("modified_non_trans_table: %d  OPTION_BEGIN: %d  "
1069
                          "OPTION_KEEP_LOG: %d  is_in_group: %d",
Michael Widenius's avatar
Michael Widenius committed
1070
                          thd->transaction.all.modified_non_trans_table,
1071 1072
                          MY_TEST(thd->variables.option_bits & OPTION_BEGIN),
                          MY_TEST(thd->variables.option_bits & OPTION_KEEP_LOG),
Michael Widenius's avatar
Michael Widenius committed
1073 1074
                          rli->is_in_group()));

1075 1076
      if (rli->abort_slave)
      {
Michael Widenius's avatar
Michael Widenius committed
1077 1078 1079 1080
        DBUG_PRINT("info",
                   ("Request to stop slave SQL Thread received while "
                    "applying a group that has non-transactional "
                    "changes; waiting for completion of the group ... "));
1081 1082

        /*
Michael Widenius's avatar
Michael Widenius committed
1083 1084 1085 1086 1087 1088 1089 1090
          Slave sql thread shutdown in face of unfinished group
          modified Non-trans table is handled via a timer. The slave
          may eventually give out to complete the current group and in
          that case there might be issues at consequent slave restart,
          see the error message.  WL#2975 offers a robust solution
          requiring to store the last exectuted event's coordinates
          along with the group's coordianates instead of waiting with
          @c last_event_start_time the timer.
1091 1092
        */

1093 1094 1095
        if (rgi->last_event_start_time == 0)
          rgi->last_event_start_time= my_time(0);
        ret= difftime(my_time(0), rgi->last_event_start_time) <=
1096 1097 1098 1099 1100 1101 1102 1103
          SLAVE_WAIT_GROUP_DONE ? FALSE : TRUE;

        DBUG_EXECUTE_IF("stop_slave_middle_group", 
                        DBUG_EXECUTE_IF("incomplete_group_in_relay_log",
                                        ret= TRUE;);); // time is over

        if (ret == 0)
        {
1104
          rli->report(WARNING_LEVEL, 0, rgi->gtid_info(),
1105 1106 1107
                      "Request to stop slave SQL Thread received while "
                      "applying a group that has non-transactional "
                      "changes; waiting for completion of the group ... ");
1108 1109 1110
        }
        else
        {
1111
          rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, rgi->gtid_info(),
1112 1113 1114 1115 1116 1117
                      ER(ER_SLAVE_FATAL_ERROR), msg_stopped);
        }
      }
      else
      {
        ret= TRUE;
1118
        rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, rgi->gtid_info(),
Michael Widenius's avatar
Michael Widenius committed
1119
                    ER(ER_SLAVE_FATAL_ERROR),
1120 1121 1122 1123 1124 1125
                    msg_stopped);
      }
    }
    else
    {
      ret= TRUE;
1126 1127
    }
  }
1128
  if (ret)
1129
    rgi->last_event_start_time= 0;
1130 1131
  
  DBUG_RETURN(ret);
1132 1133
}

1134

unknown's avatar
unknown committed
1135
/*
1136 1137
  skip_load_data_infile()

unknown's avatar
unknown committed
1138 1139 1140
  NOTES
    This is used to tell a 3.23 master to break send_file()
*/
1141 1142 1143

void skip_load_data_infile(NET *net)
{
1144 1145
  DBUG_ENTER("skip_load_data_infile");

1146
  (void)net_request_file(net, "/dev/null");
unknown's avatar
unknown committed
1147
  (void)my_net_read(net);                               // discard response
1148
  (void)net_write_command(net, 0, (uchar*) "", 0, (uchar*) "", 0); // ok
1149
  DBUG_VOID_RETURN;
1150 1151
}

1152

1153
bool net_request_file(NET* net, const char* fname)
1154
{
1155
  DBUG_ENTER("net_request_file");
1156 1157
  DBUG_RETURN(net_write_command(net, 251, (uchar*) fname, strlen(fname),
                                (uchar*) "", 0));
1158 1159
}

1160 1161
/*
  From other comments and tests in code, it looks like
1162
  sometimes Query_log_event and Load_log_event can have db == 0
1163 1164 1165
  (see rewrite_db() above for example)
  (cases where this happens are unclear; it may be when the master is 3.23).
*/
1166 1167

const char *print_slave_db_safe(const char* db)
1168
{
1169 1170 1171
  DBUG_ENTER("*print_slave_db_safe");

  DBUG_RETURN((db ? db : ""));
1172
}
1173

Sergei Golubchik's avatar
Sergei Golubchik committed
1174 1175
#endif /* HAVE_REPLICATION */

1176
int init_strvar_from_file(char *var, int max_size, IO_CACHE *f,
unknown's avatar
unknown committed
1177
                                 const char *default_val)
unknown's avatar
unknown committed
1178
{
unknown's avatar
unknown committed
1179
  uint length;
1180 1181
  DBUG_ENTER("init_strvar_from_file");

unknown's avatar
unknown committed
1182 1183 1184 1185 1186 1187
  if ((length=my_b_gets(f,var, max_size)))
  {
    char* last_p = var + length -1;
    if (*last_p == '\n')
      *last_p = 0; // if we stopped on newline, kill it
    else
unknown's avatar
unknown committed
1188
    {
unknown's avatar
unknown committed
1189
      /*
unknown's avatar
unknown committed
1190 1191
        If we truncated a line or stopped on last char, remove all chars
        up to and including newline.
unknown's avatar
unknown committed
1192
      */
unknown's avatar
unknown committed
1193
      int c;
1194
      while (((c=my_b_get(f)) != '\n' && c != my_b_EOF)) ;
unknown's avatar
unknown committed
1195
    }
1196
    DBUG_RETURN(0);
unknown's avatar
unknown committed
1197 1198 1199
  }
  else if (default_val)
  {
unknown's avatar
unknown committed
1200
    strmake(var,  default_val, max_size-1);
1201
    DBUG_RETURN(0);
unknown's avatar
unknown committed
1202
  }
1203
  DBUG_RETURN(1);
unknown's avatar
unknown committed
1204 1205
}

1206

Sergei Golubchik's avatar
Sergei Golubchik committed
1207 1208 1209 1210
/*
  when moving these functions to mysys, don't forget to
  remove slave.cc from libmysqld/CMakeLists.txt
*/
1211
int init_intvar_from_file(int* var, IO_CACHE* f, int default_val)
unknown's avatar
unknown committed
1212 1213
{
  char buf[32];
1214 1215
  DBUG_ENTER("init_intvar_from_file");

unknown's avatar
unknown committed
1216 1217

  if (my_b_gets(f, buf, sizeof(buf)))
unknown's avatar
unknown committed
1218 1219
  {
    *var = atoi(buf);
1220
    DBUG_RETURN(0);
unknown's avatar
unknown committed
1221
  }
unknown's avatar
unknown committed
1222
  else if (default_val)
unknown's avatar
unknown committed
1223 1224
  {
    *var = default_val;
1225
    DBUG_RETURN(0);
unknown's avatar
unknown committed
1226
  }
1227
  DBUG_RETURN(1);
unknown's avatar
unknown committed
1228 1229
}

Andrei Elkin's avatar
Andrei Elkin committed
1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250
int init_floatvar_from_file(float* var, IO_CACHE* f, float default_val)
{
  char buf[16];
  DBUG_ENTER("init_floatvar_from_file");


  if (my_b_gets(f, buf, sizeof(buf)))
  {
    if (sscanf(buf, "%f", var) != 1)
      DBUG_RETURN(1);
    else
      DBUG_RETURN(0);
  }
  else if (default_val != 0.0)
  {
    *var = default_val;
    DBUG_RETURN(0);
  }
  DBUG_RETURN(1);
}

Andrei Elkin's avatar
Andrei Elkin committed
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 1294 1295 1296 1297 1298 1299 1300 1301 1302

/**
   A master info read method

   This function is called from @c init_master_info() along with
   relatives to restore some of @c active_mi members.
   Particularly, this function is responsible for restoring
   IGNORE_SERVER_IDS list of servers whose events the slave is
   going to ignore (to not log them in the relay log).
   Items being read are supposed to be decimal output of values of a
   type shorter or equal of @c long and separated by the single space.

   @param arr         @c DYNAMIC_ARRAY pointer to storage for servers id
   @param f           @c IO_CACHE pointer to the source file

   @retval 0         All OK
   @retval non-zero  An error
*/

int init_dynarray_intvar_from_file(DYNAMIC_ARRAY* arr, IO_CACHE* f)
{
  int ret= 0;
  char buf[16 * (sizeof(long)*4 + 1)]; // static buffer to use most of times
  char *buf_act= buf; // actual buffer can be dynamic if static is short
  char *token, *last;
  uint num_items;     // number of items of `arr'
  size_t read_size;
  DBUG_ENTER("init_dynarray_intvar_from_file");

  if ((read_size= my_b_gets(f, buf_act, sizeof(buf))) == 0)
  {
    return 0; // no line in master.info
  }
  if (read_size + 1 == sizeof(buf) && buf[sizeof(buf) - 2] != '\n')
  {
    /*
      short read happend; allocate sufficient memory and make the 2nd read
    */
    char buf_work[(sizeof(long)*3 + 1)*16];
    memcpy(buf_work, buf, sizeof(buf_work));
    num_items= atoi(strtok_r(buf_work, " ", &last));
    size_t snd_size;
    /*
      max size lower bound approximate estimation bases on the formula:
      (the items number + items themselves) * 
          (decimal size + space) - 1 + `\n' + '\0'
    */
    size_t max_size= (1 + num_items) * (sizeof(long)*3 + 1) + 1;
    buf_act= (char*) my_malloc(max_size, MYF(MY_WME));
    memcpy(buf_act, buf, read_size);
    snd_size= my_b_gets(f, buf_act + read_size, max_size - read_size);
    if (snd_size == 0 ||
1303
        ((snd_size + 1 == max_size - read_size) &&  buf_act[max_size - 2] != '\n'))
Andrei Elkin's avatar
Andrei Elkin committed
1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334
    {
      /*
        failure to make the 2nd read or short read again
      */
      ret= 1;
      goto err;
    }
  }
  token= strtok_r(buf_act, " ", &last);
  if (token == NULL)
  {
    ret= 1;
    goto err;
  }
  num_items= atoi(token);
  for (uint i=0; i < num_items; i++)
  {
    token= strtok_r(NULL, " ", &last);
    if (token == NULL)
    {
      ret= 1;
      goto err;
    }
    else
    {
      ulong val= atol(token);
      insert_dynamic(arr, (uchar *) &val);
    }
  }
err:
  if (buf_act != buf)
1335
    my_free(buf_act);
Andrei Elkin's avatar
Andrei Elkin committed
1336 1337 1338
  DBUG_RETURN(ret);
}

Sergei Golubchik's avatar
Sergei Golubchik committed
1339 1340
#ifdef HAVE_REPLICATION

1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355
/*
  Check if the error is caused by network.
  @param[in]   errorno   Number of the error.
  RETURNS:
  TRUE         network error
  FALSE        not network error
*/

bool is_network_error(uint errorno)
{ 
  if (errorno == CR_CONNECTION_ERROR || 
      errorno == CR_CONN_HOST_ERROR ||
      errorno == CR_SERVER_GONE_ERROR ||
      errorno == CR_SERVER_LOST ||
      errorno == ER_CON_COUNT_ERROR ||
1356
      errorno == ER_CONNECTION_KILLED ||
1357
      errorno == ER_NEW_ABORTING_CONNECTION ||
1358
      errorno == ER_NET_READ_INTERRUPTED ||
1359 1360 1361 1362 1363 1364 1365
      errorno == ER_SERVER_SHUTDOWN)
    return TRUE;

  return FALSE;   
}


1366 1367 1368 1369 1370 1371 1372 1373
/*
  Note that we rely on the master's version (3.23, 4.0.14 etc) instead of
  relying on the binlog's version. This is not perfect: imagine an upgrade
  of the master without waiting that all slaves are in sync with the master;
  then a slave could be fooled about the binlog's format. This is what happens
  when people upgrade a 3.23 master to 4.0 without doing RESET MASTER: 4.0
  slaves are fooled. So we do this only to distinguish between 3.23 and more
  recent masters (it's too late to change things for 3.23).
unknown's avatar
unknown committed
1374

1375 1376 1377
  RETURNS
  0       ok
  1       error
1378
  2       transient network problem, the caller should try to reconnect
1379
*/
1380

1381
static int get_master_version_and_clock(MYSQL* mysql, Master_info* mi)
1382
{
1383
  char err_buff[MAX_SLAVE_ERRMSG], err_buff2[MAX_SLAVE_ERRMSG];
1384
  const char* errmsg= 0;
1385 1386 1387
  int err_code= 0;
  MYSQL_RES *master_res= 0;
  MYSQL_ROW master_row;
1388
  uint version= mysql_get_server_version(mysql) / 10000;
1389
  DBUG_ENTER("get_master_version_and_clock");
1390 1391 1392 1393 1394 1395 1396

  /*
    Free old description_event_for_queue (that is needed if we are in
    a reconnection).
  */
  delete mi->rli.relay_log.description_event_for_queue;
  mi->rli.relay_log.description_event_for_queue= 0;
unknown's avatar
unknown committed
1397

1398
  if (!my_isdigit(&my_charset_bin,*mysql->server_version))
1399
  {
1400 1401 1402 1403
    errmsg= err_buff2;
    snprintf(err_buff2, sizeof(err_buff2),
             "Master reported unrecognized MySQL version: %s",
             mysql->server_version);
1404
    err_code= ER_SLAVE_FATAL_ERROR;
1405
    sprintf(err_buff, ER(err_code), err_buff2);
1406
  }
1407 1408 1409 1410 1411
  else
  {
    /*
      Note the following switch will bug when we have MySQL branch 30 ;)
    */
1412 1413 1414 1415 1416 1417 1418 1419
    switch (version) {
    case 0:
    case 1:
    case 2:
      errmsg= err_buff2;
      snprintf(err_buff2, sizeof(err_buff2),
               "Master reported unrecognized MySQL version: %s",
               mysql->server_version);
1420
      err_code= ER_SLAVE_FATAL_ERROR;
1421
      sprintf(err_buff, ER(err_code), err_buff2);
1422
      break;
1423
    case 3:
1424
      mi->rli.relay_log.description_event_for_queue= new
unknown's avatar
unknown committed
1425
        Format_description_log_event(1, mysql->server_version);
1426
      break;
1427
    case 4:
1428
      mi->rli.relay_log.description_event_for_queue= new
unknown's avatar
unknown committed
1429
        Format_description_log_event(3, mysql->server_version);
1430
      break;
unknown's avatar
unknown committed
1431
    default:
1432 1433 1434 1435 1436 1437 1438 1439 1440
      /*
        Master is MySQL >=5.0. Give a default Format_desc event, so that we can
        take the early steps (like tests for "is this a 3.23 master") which we
        have to take before we receive the real master's Format_desc which will
        override this one. Note that the Format_desc we create below is garbage
        (it has the format of the *slave*); it's only good to help know if the
        master is 3.23, 4.0, etc.
      */
      mi->rli.relay_log.description_event_for_queue= new
unknown's avatar
unknown committed
1441
        Format_description_log_event(4, mysql->server_version);
1442 1443
      break;
    }
1444
  }
unknown's avatar
unknown committed
1445 1446

  /*
1447 1448 1449 1450 1451
     This does not mean that a 5.0 slave will be able to read a 6.0 master; but
     as we don't know yet, we don't want to forbid this for now. If a 5.0 slave
     can't read a 6.0 master, this will show up when the slave can't read some
     events sent by the master, and there will be error messages.
  */
unknown's avatar
unknown committed
1452

1453
  if (errmsg)
1454
    goto err;
1455 1456 1457 1458

  /* as we are here, we tried to allocate the event */
  if (!mi->rli.relay_log.description_event_for_queue)
  {
1459 1460 1461 1462
    errmsg= "default Format_description_log_event";
    err_code= ER_SLAVE_CREATE_EVENT_FAILURE;
    sprintf(err_buff, ER(err_code), errmsg);
    goto err;
1463 1464
  }

1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506
  /*
    FD_q's (A) is set initially from RL's (A): FD_q.(A) := RL.(A).
    It's necessary to adjust FD_q.(A) at this point because in the following
    course FD_q is going to be dumped to RL.
    Generally FD_q is derived from a received FD_m (roughly FD_q := FD_m) 
    in queue_event and the master's (A) is installed.
    At one step with the assignment the Relay-Log's checksum alg is set to 
    a new value: RL.(A) := FD_q.(A). If the slave service is stopped
    the last time assigned RL.(A) will be passed over to the restarting
    service (to the current execution point).
    RL.A is a "codec" to verify checksum in queue_event() almost all the time
    the first fake Rotate event.
    Starting from this point IO thread will executes the following checksum
    warmup sequence  of actions:

    FD_q.A := RL.A,
    A_m^0 := master.@@global.binlog_checksum,
    {queue_event(R_f): verifies(R_f, A_m^0)},
    {queue_event(FD_m): verifies(FD_m, FD_m.A), dump(FD_q), rotate(RL),
                        FD_q := FD_m, RL.A := FD_q.A)}

    See legends definition on MYSQL_BIN_LOG::relay_log_checksum_alg
    docs lines (binlog.h).
    In above A_m^0 - the value of master's
    @@binlog_checksum determined in the upcoming handshake (stored in
    mi->checksum_alg_before_fd).


    After the warm-up sequence IO gets to "normal" checksum verification mode
    to use RL.A in 
    
    {queue_event(E_m): verifies(E_m, RL.A)}

    until it has received a new FD_m.
  */
  mi->rli.relay_log.description_event_for_queue->checksum_alg=
    mi->rli.relay_log.relay_log_checksum_alg;

  DBUG_ASSERT(mi->rli.relay_log.description_event_for_queue->checksum_alg !=
              BINLOG_CHECKSUM_ALG_UNDEF);
  DBUG_ASSERT(mi->rli.relay_log.relay_log_checksum_alg !=
              BINLOG_CHECKSUM_ALG_UNDEF); 
1507 1508 1509 1510
  /*
    Compare the master and slave's clock. Do not die if master's clock is
    unavailable (very old master not supporting UNIX_TIMESTAMP()?).
  */
unknown's avatar
unknown committed
1511

1512
#ifdef ENABLED_DEBUG_SYNC
1513 1514 1515 1516 1517
  DBUG_EXECUTE_IF("dbug.before_get_UNIX_TIMESTAMP",
                  {
                    const char act[]=
                      "now "
                      "wait_for signal.get_unix_timestamp";
1518
                    DBUG_ASSERT(debug_sync_service);
1519 1520 1521
                    DBUG_ASSERT(!debug_sync_set_action(current_thd,
                                                       STRING_WITH_LEN(act)));
                  };);
1522
#endif
1523

1524
  master_res= NULL;
1525
  if (!mysql_real_query(mysql, STRING_WITH_LEN("SELECT UNIX_TIMESTAMP()")) &&
1526 1527
      (master_res= mysql_store_result(mysql)) &&
      (master_row= mysql_fetch_row(master_res)))
unknown's avatar
unknown committed
1528
  {
unknown's avatar
unknown committed
1529
    mi->clock_diff_with_master=
1530
      (long) (time((time_t*) 0) - strtoul(master_row[0], 0, 10));
unknown's avatar
unknown committed
1531
  }
Michael Widenius's avatar
Michael Widenius committed
1532
  else if (check_io_slave_killed(mi, NULL))
1533
    goto slave_killed_err;
1534 1535
  else if (is_network_error(mysql_errno(mysql)))
  {
1536
    mi->report(WARNING_LEVEL, mysql_errno(mysql), NULL,
1537 1538 1539 1540
               "Get master clock failed with error: %s", mysql_error(mysql));
    goto network_err;
  }
  else 
unknown's avatar
unknown committed
1541
  {
1542
    mi->clock_diff_with_master= 0; /* The "most sensible" value */
1543 1544 1545 1546
    sql_print_warning("\"SELECT UNIX_TIMESTAMP()\" failed on master, "
                      "do not trust column Seconds_Behind_Master of SHOW "
                      "SLAVE STATUS. Error: %s (%d)",
                      mysql_error(mysql), mysql_errno(mysql));
1547 1548
  }
  if (master_res)
1549
  {
unknown's avatar
unknown committed
1550
    mysql_free_result(master_res);
1551 1552
    master_res= NULL;
  }
unknown's avatar
unknown committed
1553

1554 1555 1556 1557 1558 1559 1560 1561 1562 1563
  /*
    Check that the master's server id and ours are different. Because if they
    are equal (which can result from a simple copy of master's datadir to slave,
    thus copying some my.cnf), replication will work but all events will be
    skipped.
    Do not die if SHOW VARIABLES LIKE 'SERVER_ID' fails on master (very old
    master?).
    Note: we could have put a @@SERVER_ID in the previous SELECT
    UNIX_TIMESTAMP() instead, but this would not have worked on 3.23 masters.
  */
1564
#ifdef ENABLED_DEBUG_SYNC
1565 1566 1567 1568 1569
  DBUG_EXECUTE_IF("dbug.before_get_SERVER_ID",
                  {
                    const char act[]=
                      "now "
                      "wait_for signal.get_server_id";
1570
                    DBUG_ASSERT(debug_sync_service);
1571 1572 1573
                    DBUG_ASSERT(!debug_sync_set_action(current_thd, 
                                                       STRING_WITH_LEN(act)));
                  };);
1574
#endif
1575 1576
  master_res= NULL;
  master_row= NULL;
1577 1578
  if (!mysql_real_query(mysql,
                        STRING_WITH_LEN("SHOW VARIABLES LIKE 'SERVER_ID'")) &&
1579 1580
      (master_res= mysql_store_result(mysql)) &&
      (master_row= mysql_fetch_row(master_res)))
1581
  {
1582 1583
    if ((global_system_variables.server_id ==
             (mi->master_id= strtoul(master_row[1], 0, 10))) &&
1584
        !mi->rli.replicate_same_server_id)
1585
    {
1586 1587 1588 1589
      errmsg= "The slave I/O thread stops because master and slave have equal \
MySQL server ids; these ids must be different for replication to work (or \
the --replicate-same-server-id option must be used on slave but this does \
not always make sense; please check the manual before using it).";
1590 1591
      err_code= ER_SLAVE_FATAL_ERROR;
      sprintf(err_buff, ER(err_code), errmsg);
1592
      goto err;
1593
    }
1594 1595 1596
  }
  else if (mysql_errno(mysql))
  {
Michael Widenius's avatar
Michael Widenius committed
1597
    if (check_io_slave_killed(mi, NULL))
1598 1599
      goto slave_killed_err;
    else if (is_network_error(mysql_errno(mysql)))
1600
    {
1601
      mi->report(WARNING_LEVEL, mysql_errno(mysql), NULL,
1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613
                 "Get master SERVER_ID failed with error: %s", mysql_error(mysql));
      goto network_err;
    }
    /* Fatal error */
    errmsg= "The slave I/O thread stops because a fatal error is encountered \
when it try to get the value of SERVER_ID variable from master.";
    err_code= mysql_errno(mysql);
    sprintf(err_buff, "%s Error: %s", errmsg, mysql_error(mysql));
    goto err;
  }
  else if (!master_row && master_res)
  {
1614
    mi->report(WARNING_LEVEL, ER_UNKNOWN_SYSTEM_VARIABLE, NULL,
1615 1616 1617 1618 1619
               "Unknown system variable 'SERVER_ID' on master, \
maybe it is a *VERY OLD MASTER*.");
  }
  if (master_res)
  {
1620
    mysql_free_result(master_res);
1621
    master_res= NULL;
unknown's avatar
unknown committed
1622
  }
Andrei Elkin's avatar
Andrei Elkin committed
1623 1624 1625 1626 1627 1628 1629
  if (mi->master_id == 0 && mi->ignore_server_ids.elements > 0)
  {
    errmsg= "Slave configured with server id filtering could not detect the master server id.";
    err_code= ER_SLAVE_FATAL_ERROR;
    sprintf(err_buff, ER(err_code), errmsg);
    goto err;
  }
unknown's avatar
unknown committed
1630

1631 1632 1633
  /*
    Check that the master's global character_set_server and ours are the same.
    Not fatal if query fails (old master?).
1634 1635 1636 1637 1638 1639
    Note that we don't check for equality of global character_set_client and
    collation_connection (neither do we prevent their setting in
    set_var.cc). That's because from what I (Guilhem) have tested, the global
    values of these 2 are never used (new connections don't use them).
    We don't test equality of global collation_database either as it's is
    going to be deprecated (made read-only) in 4.1 very soon.
1640 1641 1642 1643 1644 1645
    The test is only relevant if master < 5.0.3 (we'll test only if it's older
    than the 5 branch; < 5.0.3 was alpha...), as >= 5.0.3 master stores
    charset info in each binlog event.
    We don't do it for 3.23 because masters <3.23.50 hang on
    SELECT @@unknown_var (BUG#7965 - see changelog of 3.23.50). So finally we
    test only if master is 4.x.
1646
  */
1647 1648

  /* redundant with rest of code but safer against later additions */
1649
  if (version == 3)
1650
    goto err;
1651

1652
  if (version == 4)
unknown's avatar
unknown committed
1653
  {
1654 1655 1656 1657 1658
    master_res= NULL;
    if (!mysql_real_query(mysql,
                          STRING_WITH_LEN("SELECT @@GLOBAL.COLLATION_SERVER")) &&
        (master_res= mysql_store_result(mysql)) &&
        (master_row= mysql_fetch_row(master_res)))
1659
    {
1660 1661 1662
      if (strcmp(master_row[0], global_system_variables.collation_server->name))
      {
        errmsg= "The slave I/O thread stops because master and slave have \
1663
different values for the COLLATION_SERVER global variable. The values must \
1664 1665 1666 1667 1668
be equal for the Statement-format replication to work";
        err_code= ER_SLAVE_FATAL_ERROR;
        sprintf(err_buff, ER(err_code), errmsg);
        goto err;
      }
1669
    }
Michael Widenius's avatar
Michael Widenius committed
1670
    else if (check_io_slave_killed(mi, NULL))
1671
      goto slave_killed_err;
1672 1673
    else if (is_network_error(mysql_errno(mysql)))
    {
1674
      mi->report(WARNING_LEVEL, mysql_errno(mysql), NULL,
1675 1676 1677 1678 1679 1680 1681 1682 1683 1684
                 "Get master COLLATION_SERVER failed with error: %s", mysql_error(mysql));
      goto network_err;
    }
    else if (mysql_errno(mysql) != ER_UNKNOWN_SYSTEM_VARIABLE)
    {
      /* Fatal error */
      errmsg= "The slave I/O thread stops because a fatal error is encountered \
when it try to get the value of COLLATION_SERVER global variable from master.";
      err_code= mysql_errno(mysql);
      sprintf(err_buff, "%s Error: %s", errmsg, mysql_error(mysql));
1685
      goto err;
1686 1687
    }
    else
1688
      mi->report(WARNING_LEVEL, ER_UNKNOWN_SYSTEM_VARIABLE, NULL,
1689 1690 1691 1692 1693 1694 1695 1696 1697
                 "Unknown system variable 'COLLATION_SERVER' on master, \
maybe it is a *VERY OLD MASTER*. *NOTE*: slave may experience \
inconsistency if replicated data deals with collation.");

    if (master_res)
    {
      mysql_free_result(master_res);
      master_res= NULL;
    }
unknown's avatar
unknown committed
1698
  }
1699

1700 1701 1702 1703 1704 1705 1706 1707
  /*
    Perform analogous check for time zone. Theoretically we also should
    perform check here to verify that SYSTEM time zones are the same on
    slave and master, but we can't rely on value of @@system_time_zone
    variable (it is time zone abbreviation) since it determined at start
    time and so could differ for slave and master even if they are really
    in the same system time zone. So we are omiting this check and just
    relying on documentation. Also according to Monty there are many users
unknown's avatar
unknown committed
1708 1709 1710
    who are using replication between servers in various time zones. Hence
    such check will broke everything for them. (And now everything will
    work for them because by default both their master and slave will have
1711
    'SYSTEM' time zone).
1712 1713
    This check is only necessary for 4.x masters (and < 5.0.4 masters but
    those were alpha).
1714
  */
1715
  if (version == 4)
unknown's avatar
unknown committed
1716
  {
1717 1718 1719 1720
    master_res= NULL;
    if (!mysql_real_query(mysql, STRING_WITH_LEN("SELECT @@GLOBAL.TIME_ZONE")) &&
        (master_res= mysql_store_result(mysql)) &&
        (master_row= mysql_fetch_row(master_res)))
1721
    {
1722 1723 1724 1725
      if (strcmp(master_row[0],
                 global_system_variables.time_zone->get_name()->ptr()))
      {
        errmsg= "The slave I/O thread stops because master and slave have \
1726
different values for the TIME_ZONE global variable. The values must \
1727 1728 1729 1730 1731
be equal for the Statement-format replication to work";
        err_code= ER_SLAVE_FATAL_ERROR;
        sprintf(err_buff, ER(err_code), errmsg);
        goto err;
      }
1732
    }
Michael Widenius's avatar
Michael Widenius committed
1733
    else if (check_io_slave_killed(mi, NULL))
1734
      goto slave_killed_err;
1735
    else if (is_network_error(err_code= mysql_errno(mysql)))
1736
    {
1737
      mi->report(ERROR_LEVEL, err_code, NULL,
1738 1739
                 "Get master TIME_ZONE failed with error: %s",
                 mysql_error(mysql));
1740
      goto network_err;
1741 1742 1743 1744
    }
    else if (err_code == ER_UNKNOWN_SYSTEM_VARIABLE)
    {
      /* We use ERROR_LEVEL to get the error logged to file */
1745
      mi->report(ERROR_LEVEL, err_code, NULL,
1746 1747 1748 1749 1750

                 "MySQL master doesn't have a TIME_ZONE variable. Note that"
                 "if your timezone is not same between master and slave, your "
                 "slave may get wrong data into timestamp columns");
    }
1751 1752 1753 1754 1755 1756
    else
    {
      /* Fatal error */
      errmsg= "The slave I/O thread stops because a fatal error is encountered \
when it try to get the value of TIME_ZONE global variable from master.";
      sprintf(err_buff, "%s Error: %s", errmsg, mysql_error(mysql));
1757
      goto err;
1758 1759 1760 1761 1762 1763
    }
    if (master_res)
    {
      mysql_free_result(master_res);
      master_res= NULL;
    }
unknown's avatar
unknown committed
1764 1765
  }

Andrei Elkin's avatar
Andrei Elkin committed
1766 1767 1768 1769 1770 1771 1772 1773 1774
  if (mi->heartbeat_period != 0.0)
  {
    char llbuf[22];
    const char query_format[]= "SET @master_heartbeat_period= %s";
    char query[sizeof(query_format) - 2 + sizeof(llbuf)];
    /* 
       the period is an ulonglong of nano-secs. 
    */
    llstr((ulonglong) (mi->heartbeat_period*1000000000UL), llbuf);
1775
    sprintf(query, query_format, llbuf);
Andrei Elkin's avatar
Andrei Elkin committed
1776

1777 1778 1779 1780 1781 1782
    DBUG_EXECUTE_IF("simulate_slave_heartbeat_network_error",
                    { static ulong dbug_count= 0;
                      if (++dbug_count < 3)
                        goto heartbeat_network_error;
                    });
    if (mysql_real_query(mysql, query, strlen(query)))
Andrei Elkin's avatar
Andrei Elkin committed
1783
    {
Sergei Golubchik's avatar
Sergei Golubchik committed
1784
      if (check_io_slave_killed(mi, NULL))
1785 1786 1787 1788 1789
        goto slave_killed_err;

      if (is_network_error(mysql_errno(mysql)))
      {
      IF_DBUG(heartbeat_network_error: , )
Sergei Golubchik's avatar
Sergei Golubchik committed
1790
        mi->report(WARNING_LEVEL, mysql_errno(mysql), NULL,
1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805
                   "SET @master_heartbeat_period to master failed with error: %s",
                   mysql_error(mysql));
        mysql_free_result(mysql_store_result(mysql));
        goto network_err;
      }
      else
      {
        /* Fatal error */
        errmsg= "The slave I/O thread stops because a fatal error is encountered "
          "when it tries to SET @master_heartbeat_period on master.";
        err_code= ER_SLAVE_FATAL_ERROR;
        sprintf(err_buff, "%s Error: %s", errmsg, mysql_error(mysql));
        mysql_free_result(mysql_store_result(mysql));
        goto err;
      }
Andrei Elkin's avatar
Andrei Elkin committed
1806 1807 1808 1809
    }
    mysql_free_result(mysql_store_result(mysql));
  }
 
1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832
  /*
    Querying if master is capable to checksum and notifying it about own
    CRC-awareness. The master's side instant value of @@global.binlog_checksum 
    is stored in the dump thread's uservar area as well as cached locally
    to become known in consensus by master and slave.
  */
  DBUG_EXECUTE_IF("simulate_slave_unaware_checksum",
                  mi->checksum_alg_before_fd= BINLOG_CHECKSUM_ALG_OFF;
                  goto past_checksum;);
  {
    int rc;
    const char query[]= "SET @master_binlog_checksum= @@global.binlog_checksum";
    master_res= NULL;
    mi->checksum_alg_before_fd= BINLOG_CHECKSUM_ALG_UNDEF; //initially undefined
    /*
      @c checksum_alg_before_fd is queried from master in this block.
      If master is old checksum-unaware the value stays undefined.
      Once the first FD will be received its alg descriptor will replace
      the being queried one.
    */
    rc= mysql_real_query(mysql, query, strlen(query));
    if (rc != 0)
    {
Michael Widenius's avatar
Michael Widenius committed
1833
      if (check_io_slave_killed(mi, NULL))
1834 1835 1836 1837
        goto slave_killed_err;

      if (mysql_errno(mysql) == ER_UNKNOWN_SYSTEM_VARIABLE)
      {
1838 1839 1840 1841
        /* Ignore this expected error if not a high error level */
        if (global_system_variables.log_warnings > 1)
        {
          // this is tolerable as OM -> NS is supported
1842
          mi->report(WARNING_LEVEL, mysql_errno(mysql), NULL,
1843 1844 1845
                     "Notifying master by %s failed with "
                     "error: %s", query, mysql_error(mysql));
        }
1846 1847 1848 1849 1850
      }
      else
      {
        if (is_network_error(mysql_errno(mysql)))
        {
1851
          mi->report(WARNING_LEVEL, mysql_errno(mysql), NULL,
1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882
                     "Notifying master by %s failed with "
                     "error: %s", query, mysql_error(mysql));
          mysql_free_result(mysql_store_result(mysql));
          goto network_err;
        }
        else
        {
          errmsg= "The slave I/O thread stops because a fatal error is encountered "
            "when it tried to SET @master_binlog_checksum on master.";
          err_code= ER_SLAVE_FATAL_ERROR;
          sprintf(err_buff, "%s Error: %s", errmsg, mysql_error(mysql));
          mysql_free_result(mysql_store_result(mysql));
          goto err;
        }
      }
    }
    else
    {
      mysql_free_result(mysql_store_result(mysql));
      if (!mysql_real_query(mysql,
                            STRING_WITH_LEN("SELECT @master_binlog_checksum")) &&
          (master_res= mysql_store_result(mysql)) &&
          (master_row= mysql_fetch_row(master_res)) &&
          (master_row[0] != NULL))
      {
        mi->checksum_alg_before_fd= (uint8)
          find_type(master_row[0], &binlog_checksum_typelib, 1) - 1;
        // valid outcome is either of
        DBUG_ASSERT(mi->checksum_alg_before_fd == BINLOG_CHECKSUM_ALG_OFF ||
                    mi->checksum_alg_before_fd == BINLOG_CHECKSUM_ALG_CRC32);
      }
Michael Widenius's avatar
Michael Widenius committed
1883
      else if (check_io_slave_killed(mi, NULL))
1884 1885 1886
        goto slave_killed_err;
      else if (is_network_error(mysql_errno(mysql)))
      {
1887
        mi->report(WARNING_LEVEL, mysql_errno(mysql), NULL,
1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910
                   "Get master BINLOG_CHECKSUM failed with error: %s", mysql_error(mysql));
        goto network_err;
      }
      else
      {
        errmsg= "The slave I/O thread stops because a fatal error is encountered "
          "when it tried to SELECT @master_binlog_checksum.";
        err_code= ER_SLAVE_FATAL_ERROR;
        sprintf(err_buff, "%s Error: %s", errmsg, mysql_error(mysql));
        mysql_free_result(mysql_store_result(mysql));
        goto err;
      }
    }
    if (master_res)
    {
      mysql_free_result(master_res);
      master_res= NULL;
    }
  }

#ifndef DBUG_OFF
past_checksum:
#endif
Andrei Elkin's avatar
Andrei Elkin committed
1911

1912
  /*
1913
    Request the master to filter away events with the @@skip_replication flag
1914 1915
    set, if we are running with
    --replicate-events-marked-for-skip=FILTER_ON_MASTER.
1916
  */
1917
  if (opt_replicate_events_marked_for_skip == RPL_SKIP_FILTER_ON_MASTER)
1918
  {
1919
    if (mysql_real_query(mysql, STRING_WITH_LEN("SET skip_replication=1")))
1920 1921 1922 1923
    {
      err_code= mysql_errno(mysql);
      if (is_network_error(err_code))
      {
1924
        mi->report(ERROR_LEVEL, err_code, NULL,
1925
                   "Setting master-side filtering of @@skip_replication failed "
1926 1927 1928 1929 1930 1931 1932
                   "with error: %s", mysql_error(mysql));
        goto network_err;
      }
      else if (err_code == ER_UNKNOWN_SYSTEM_VARIABLE)
      {
        /*
          The master is older than the slave and does not support the
1933
          @@skip_replication feature.
1934
          This is not a problem, as such master will not generate events with
1935
          the @@skip_replication flag set in the first place. We will still
1936 1937
          do slave-side filtering of such events though, to handle the (rare)
          case of downgrading a master and receiving old events generated from
1938
          before the downgrade with the @@skip_replication flag set.
1939 1940
        */
        DBUG_PRINT("info", ("Old master does not support master-side filtering "
1941 1942 1943 1944 1945 1946 1947 1948 1949 1950
                            "of @@skip_replication events."));
      }
      else
      {
        /* Fatal error */
        errmsg= "The slave I/O thread stops because a fatal error is "
          "encountered when it tries to request filtering of events marked "
          "with the @@skip_replication flag.";
        sprintf(err_buff, "%s Error: %s", errmsg, mysql_error(mysql));
        goto err;
1951 1952 1953
      }
    }
  }
1954 1955 1956 1957

  /* Announce MariaDB slave capabilities. */
  DBUG_EXECUTE_IF("simulate_slave_capability_none", goto after_set_capability;);
  {
1958 1959 1960 1961 1962 1963
    int rc= DBUG_EVALUATE_IF("simulate_slave_capability_old_53",
        mysql_real_query(mysql, STRING_WITH_LEN("SET @mariadb_slave_capability="
                         STRINGIFY_ARG(MARIA_SLAVE_CAPABILITY_ANNOTATE))),
        mysql_real_query(mysql, STRING_WITH_LEN("SET @mariadb_slave_capability="
                         STRINGIFY_ARG(MARIA_SLAVE_CAPABILITY_MINE))));
    if (rc)
1964 1965 1966 1967
    {
      err_code= mysql_errno(mysql);
      if (is_network_error(err_code))
      {
1968
        mi->report(ERROR_LEVEL, err_code, NULL,
1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982
                   "Setting @mariadb_slave_capability failed with error: %s",
                   mysql_error(mysql));
        goto network_err;
      }
      else
      {
        /* Fatal error */
        errmsg= "The slave I/O thread stops because a fatal error is "
          "encountered when it tries to set @mariadb_slave_capability.";
        sprintf(err_buff, "%s Error: %s", errmsg, mysql_error(mysql));
        goto err;
      }
    }
  }
1983
#ifndef DBUG_OFF
1984
after_set_capability:
1985
#endif
1986

unknown's avatar
unknown committed
1987
  if (mi->using_gtid != Master_info::USE_GTID_NO)
1988
  {
unknown's avatar
unknown committed
1989
    /* Request dump to start from slave replication GTID state. */
1990 1991
    int rc;
    char str_buf[256];
1992 1993
    String query_str(str_buf, sizeof(str_buf), system_charset_info);
    query_str.length(0);
unknown's avatar
unknown committed
1994

unknown's avatar
unknown committed
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006
    /*
      Read the master @@GLOBAL.gtid_domain_id variable.
      This is mostly to check that master is GTID aware, but we could later
      perhaps use it to check that different multi-source masters are correctly
      configured with distinct domain_id.
    */
    if (mysql_real_query(mysql,
                         STRING_WITH_LEN("SELECT @@GLOBAL.gtid_domain_id")) ||
        !(master_res= mysql_store_result(mysql)) ||
        !(master_row= mysql_fetch_row(master_res)))
    {
      err_code= mysql_errno(mysql);
2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021
      if (is_network_error(err_code))
      {
        mi->report(ERROR_LEVEL, err_code, NULL,
                   "Get master @@GLOBAL.gtid_domain_id failed with error: %s",
                   mysql_error(mysql));
        goto network_err;
      }
      else
      {
        errmsg= "The slave I/O thread stops because master does not support "
          "MariaDB global transaction id. A fatal error is encountered when "
          "it tries to SELECT @@GLOBAL.gtid_domain_id.";
        sprintf(err_buff, "%s Error: %s", errmsg, mysql_error(mysql));
        goto err;
      }
unknown's avatar
unknown committed
2022 2023 2024 2025
    }
    mysql_free_result(master_res);
    master_res= NULL;

2026 2027
    query_str.append(STRING_WITH_LEN("SET @slave_connect_state='"),
                     system_charset_info);
unknown's avatar
unknown committed
2028
    if (mi->gtid_current_pos.append_to_string(&query_str))
unknown's avatar
unknown committed
2029 2030 2031 2032 2033 2034 2035
    {
      err_code= ER_OUTOFMEMORY;
      errmsg= "The slave I/O thread stops because a fatal out-of-memory "
        "error is encountered when it tries to compute @slave_connect_state.";
      sprintf(err_buff, "%s Error: Out of memory", errmsg);
      goto err;
    }
2036
    query_str.append(STRING_WITH_LEN("'"), system_charset_info);
unknown's avatar
unknown committed
2037

2038
    rc= mysql_real_query(mysql, query_str.ptr(), query_str.length());
2039 2040 2041 2042 2043
    if (rc)
    {
      err_code= mysql_errno(mysql);
      if (is_network_error(err_code))
      {
2044
        mi->report(ERROR_LEVEL, err_code, NULL,
2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057
                   "Setting @slave_connect_state failed with error: %s",
                   mysql_error(mysql));
        goto network_err;
      }
      else
      {
        /* Fatal error */
        errmsg= "The slave I/O thread stops because a fatal error is "
          "encountered when it tries to set @slave_connect_state.";
        sprintf(err_buff, "%s Error: %s", errmsg, mysql_error(mysql));
        goto err;
      }
    }
unknown's avatar
unknown committed
2058

2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076
    query_str.length(0);
    if (query_str.append(STRING_WITH_LEN("SET @slave_gtid_strict_mode="),
                         system_charset_info) ||
        query_str.append_ulonglong(opt_gtid_strict_mode != false))
    {
      err_code= ER_OUTOFMEMORY;
      errmsg= "The slave I/O thread stops because a fatal out-of-memory "
        "error is encountered when it tries to set @slave_gtid_strict_mode.";
      sprintf(err_buff, "%s Error: Out of memory", errmsg);
      goto err;
    }

    rc= mysql_real_query(mysql, query_str.ptr(), query_str.length());
    if (rc)
    {
      err_code= mysql_errno(mysql);
      if (is_network_error(err_code))
      {
2077
        mi->report(ERROR_LEVEL, err_code, NULL,
2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091
                   "Setting @slave_gtid_strict_mode failed with error: %s",
                   mysql_error(mysql));
        goto network_err;
      }
      else
      {
        /* Fatal error */
        errmsg= "The slave I/O thread stops because a fatal error is "
          "encountered when it tries to set @slave_gtid_strict_mode.";
        sprintf(err_buff, "%s Error: %s", errmsg, mysql_error(mysql));
        goto err;
      }
    }

2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109
    query_str.length(0);
    if (query_str.append(STRING_WITH_LEN("SET @slave_gtid_ignore_duplicates="),
                         system_charset_info) ||
        query_str.append_ulonglong(opt_gtid_ignore_duplicates != false))
    {
      err_code= ER_OUTOFMEMORY;
      errmsg= "The slave I/O thread stops because a fatal out-of-memory error "
        "is encountered when it tries to set @slave_gtid_ignore_duplicates.";
      sprintf(err_buff, "%s Error: Out of memory", errmsg);
      goto err;
    }

    rc= mysql_real_query(mysql, query_str.ptr(), query_str.length());
    if (rc)
    {
      err_code= mysql_errno(mysql);
      if (is_network_error(err_code))
      {
2110
        mi->report(ERROR_LEVEL, err_code, NULL,
2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124
                   "Setting @slave_gtid_ignore_duplicates failed with "
                   "error: %s", mysql_error(mysql));
        goto network_err;
      }
      else
      {
        /* Fatal error */
        errmsg= "The slave I/O thread stops because a fatal error is "
          "encountered when it tries to set @slave_gtid_ignore_duplicates.";
        sprintf(err_buff, "%s Error: %s", errmsg, mysql_error(mysql));
        goto err;
      }
    }

unknown's avatar
unknown committed
2125 2126
    if (mi->rli.until_condition == Relay_log_info::UNTIL_GTID)
    {
2127 2128 2129 2130
      query_str.length(0);
      query_str.append(STRING_WITH_LEN("SET @slave_until_gtid='"),
                       system_charset_info);
      if (mi->rli.until_gtid_pos.append_to_string(&query_str))
unknown's avatar
unknown committed
2131 2132 2133 2134 2135 2136 2137
      {
        err_code= ER_OUTOFMEMORY;
        errmsg= "The slave I/O thread stops because a fatal out-of-memory "
          "error is encountered when it tries to compute @slave_until_gtid.";
        sprintf(err_buff, "%s Error: Out of memory", errmsg);
        goto err;
      }
2138
      query_str.append(STRING_WITH_LEN("'"), system_charset_info);
unknown's avatar
unknown committed
2139

2140
      rc= mysql_real_query(mysql, query_str.ptr(), query_str.length());
unknown's avatar
unknown committed
2141 2142 2143 2144 2145
      if (rc)
      {
        err_code= mysql_errno(mysql);
        if (is_network_error(err_code))
        {
2146
          mi->report(ERROR_LEVEL, err_code, NULL,
unknown's avatar
unknown committed
2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160
                     "Setting @slave_until_gtid failed with error: %s",
                     mysql_error(mysql));
          goto network_err;
        }
        else
        {
          /* Fatal error */
          errmsg= "The slave I/O thread stops because a fatal error is "
            "encountered when it tries to set @slave_until_gtid.";
          sprintf(err_buff, "%s Error: %s", errmsg, mysql_error(mysql));
          goto err;
        }
      }
    }
2161
  }
unknown's avatar
unknown committed
2162
  else
unknown's avatar
unknown committed
2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186
  {
    /*
      If we are not using GTID to connect this time, then instead request
      the corresponding GTID position from the master, so that the user
      can reconnect the next time using MASTER_GTID_POS=AUTO.
    */
    char quote_buf[2*sizeof(mi->master_log_name)+1];
    char str_buf[28+2*sizeof(mi->master_log_name)+10];
    String query(str_buf, sizeof(str_buf), system_charset_info);
    query.length(0);

    query.append("SELECT binlog_gtid_pos('");
    escape_quotes_for_mysql(&my_charset_bin, quote_buf, sizeof(quote_buf),
                            mi->master_log_name, strlen(mi->master_log_name));
    query.append(quote_buf);
    query.append("',");
    query.append_ulonglong(mi->master_log_pos);
    query.append(")");

    if (!mysql_real_query(mysql, query.c_ptr_safe(), query.length()) &&
        (master_res= mysql_store_result(mysql)) &&
        (master_row= mysql_fetch_row(master_res)) &&
        (master_row[0] != NULL))
    {
unknown's avatar
unknown committed
2187
      rpl_global_gtid_slave_state.load(mi->io_thd, master_row[0],
unknown's avatar
unknown committed
2188
                                       strlen(master_row[0]), false, false);
unknown's avatar
unknown committed
2189
    }
Michael Widenius's avatar
Michael Widenius committed
2190
    else if (check_io_slave_killed(mi, NULL))
unknown's avatar
unknown committed
2191 2192 2193
      goto slave_killed_err;
    else if (is_network_error(mysql_errno(mysql)))
    {
2194
      mi->report(WARNING_LEVEL, mysql_errno(mysql), NULL,
unknown's avatar
unknown committed
2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216
                 "Get master GTID position failed with error: %s", mysql_error(mysql));
      goto network_err;
    }
    else
    {
      /*
        ToDo: If the master does not have the binlog_gtid_pos() function, it
        just means that it is an old master with no GTID support, so we should
        do nothing.

        However, if binlog_gtid_pos() exists, but fails or returns NULL, then
        it means that the requested position is not valid. We could use this
        to catch attempts to replicate from within the middle of an event,
        avoiding strange failures or possible corruption.
      */
    }
    if (master_res)
    {
      mysql_free_result(master_res);
      master_res= NULL;
    }
  }
2217

2218
err:
2219 2220
  if (errmsg)
  {
2221 2222
    if (master_res)
      mysql_free_result(master_res);
2223
    DBUG_ASSERT(err_code != 0);
2224
    mi->report(ERROR_LEVEL, err_code, NULL, "%s", err_buff);
2225
    DBUG_RETURN(1);
2226
  }
2227

2228
  DBUG_RETURN(0);
2229 2230 2231 2232 2233

network_err:
  if (master_res)
    mysql_free_result(master_res);
  DBUG_RETURN(2);
unknown's avatar
unknown committed
2234

2235 2236 2237 2238
slave_killed_err:
  if (master_res)
    mysql_free_result(master_res);
  DBUG_RETURN(2);
unknown's avatar
unknown committed
2239
}
unknown's avatar
unknown committed
2240

2241

2242
static bool wait_for_relay_log_space(Relay_log_info* rli)
unknown's avatar
unknown committed
2243
{
2244
  bool slave_killed=0;
2245
  bool ignore_log_space_limit;
2246
  Master_info* mi = rli->mi;
Sergei Golubchik's avatar
Sergei Golubchik committed
2247
  PSI_stage_info old_stage;
unknown's avatar
unknown committed
2248 2249
  THD* thd = mi->io_thd;
  DBUG_ENTER("wait_for_relay_log_space");
2250

Marc Alff's avatar
Marc Alff committed
2251
  mysql_mutex_lock(&rli->log_space_lock);
Sergei Golubchik's avatar
Sergei Golubchik committed
2252 2253 2254 2255
  thd->ENTER_COND(&rli->log_space_cond,
                  &rli->log_space_lock,
                  &stage_waiting_for_relay_log_space,
                  &old_stage);
unknown's avatar
unknown committed
2256
  while (rli->log_space_limit < rli->log_space_total &&
Michael Widenius's avatar
Michael Widenius committed
2257
         !(slave_killed=io_slave_killed(mi)) &&
2258
         !rli->ignore_log_space_limit)
Marc Alff's avatar
Marc Alff committed
2259
    mysql_cond_wait(&rli->log_space_cond, &rli->log_space_lock);
2260

2261 2262 2263 2264 2265
  ignore_log_space_limit= rli->ignore_log_space_limit;
  rli->ignore_log_space_limit= 0;

  thd->EXIT_COND(&old_stage);

2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288
  /* 
    Makes the IO thread read only one event at a time
    until the SQL thread is able to purge the relay 
    logs, freeing some space.

    Therefore, once the SQL thread processes this next 
    event, it goes to sleep (no more events in the queue),
    sets ignore_log_space_limit=true and wakes the IO thread. 
    However, this event may have been enough already for 
    the SQL thread to purge some log files, freeing 
    rli->log_space_total .

    This guarantees that the SQL and IO thread move
    forward only one event at a time (to avoid deadlocks), 
    when the relay space limit is reached. It also 
    guarantees that when the SQL thread is prepared to
    rotate (to be able to purge some logs), the IO thread
    will know about it and will rotate.

    NOTE: The ignore_log_space_limit is only set when the SQL
          thread sleeps waiting for events.

   */
2289 2290

  if (ignore_log_space_limit)
2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306
  {
#ifndef DBUG_OFF
    {
      char llbuf1[22], llbuf2[22];
      DBUG_PRINT("info", ("log_space_limit=%s "
                          "log_space_total=%s "
                          "ignore_log_space_limit=%d "
                          "sql_force_rotate_relay=%d", 
                        llstr(rli->log_space_limit,llbuf1),
                        llstr(rli->log_space_total,llbuf2),
                        (int) rli->ignore_log_space_limit,
                        (int) rli->sql_force_rotate_relay));
    }
#endif
    if (rli->sql_force_rotate_relay)
    {
2307
      mysql_mutex_lock(&mi->data_lock);
2308
      rotate_relay_log(rli->mi);
2309
      mysql_mutex_unlock(&mi->data_lock);
2310 2311 2312 2313
      rli->sql_force_rotate_relay= false;
    }
  }

unknown's avatar
unknown committed
2314 2315
  DBUG_RETURN(slave_killed);
}
2316

2317

2318 2319
/*
  Builds a Rotate from the ignored events' info and writes it to relay log.
2320

2321 2322 2323 2324 2325 2326 2327 2328 2329 2330
  SYNOPSIS
  write_ignored_events_info_to_relay_log()
    thd             pointer to I/O thread's thd
    mi

  DESCRIPTION
    Slave I/O thread, going to die, must leave a durable trace of the
    ignored events' end position for the use of the slave SQL thread, by
    calling this function. Only that thread can call it (see assertion).
 */
2331
static void write_ignored_events_info_to_relay_log(THD *thd, Master_info *mi)
2332
{
2333
  Relay_log_info *rli= &mi->rli;
Marc Alff's avatar
Marc Alff committed
2334
  mysql_mutex_t *log_lock= rli->relay_log.get_log_lock();
2335 2336
  DBUG_ENTER("write_ignored_events_info_to_relay_log");

2337
  DBUG_ASSERT(thd == mi->io_thd);
Marc Alff's avatar
Marc Alff committed
2338
  mysql_mutex_lock(log_lock);
2339 2340
  if (rli->ign_master_log_name_end[0] || rli->ign_gtids.count())
  {
2341 2342
    Rotate_log_event *rev= NULL;
    Gtid_list_log_event *glev= NULL;
2343 2344 2345 2346 2347 2348 2349
    if (rli->ign_master_log_name_end[0])
    {
      rev= new Rotate_log_event(rli->ign_master_log_name_end,
                                0, rli->ign_master_log_pos_end,
                                Rotate_log_event::DUP_NAME);
      rli->ign_master_log_name_end[0]= 0;
      if (unlikely(!(bool)rev))
2350
        mi->report(ERROR_LEVEL, ER_SLAVE_CREATE_EVENT_FAILURE, NULL,
2351 2352 2353 2354 2355 2356 2357 2358 2359 2360
                   ER(ER_SLAVE_CREATE_EVENT_FAILURE),
                   "Rotate_event (out of memory?),"
                   " SHOW SLAVE STATUS may be inaccurate");
    }
    if (rli->ign_gtids.count())
    {
      glev= new Gtid_list_log_event(&rli->ign_gtids,
                                    Gtid_list_log_event::FLAG_IGN_GTIDS);
      rli->ign_gtids.reset();
      if (unlikely(!(bool)glev))
2361
        mi->report(ERROR_LEVEL, ER_SLAVE_CREATE_EVENT_FAILURE, NULL,
2362 2363 2364 2365 2366 2367
                   ER(ER_SLAVE_CREATE_EVENT_FAILURE),
                   "Gtid_list_event (out of memory?),"
                   " gtid_slave_pos may be inaccurate");
    }

    /* Can unlock before writing as slave SQL thd will soon see our event. */
Marc Alff's avatar
Marc Alff committed
2368
    mysql_mutex_unlock(log_lock);
2369
    if (rev)
2370
    {
2371 2372 2373
      DBUG_PRINT("info",("writing a Rotate event to track down ignored events"));
      rev->server_id= 0; // don't be ignored by slave SQL thread
      if (unlikely(rli->relay_log.append(rev)))
2374
        mi->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_WRITE_FAILURE, NULL,
2375 2376 2377 2378
                   ER(ER_SLAVE_RELAY_LOG_WRITE_FAILURE),
                   "failed to write a Rotate event"
                   " to the relay log, SHOW SLAVE STATUS may be"
                   " inaccurate");
2379 2380 2381 2382 2383 2384 2385 2386
      delete rev;
    }
    if (glev)
    {
      DBUG_PRINT("info",("writing a Gtid_list event to track down ignored events"));
      glev->server_id= 0; // don't be ignored by slave SQL thread
      glev->set_artificial_event(); // Don't mess up Exec_Master_Log_Pos
      if (unlikely(rli->relay_log.append(glev)))
2387
        mi->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_WRITE_FAILURE, NULL,
2388 2389 2390 2391 2392 2393 2394
                   ER(ER_SLAVE_RELAY_LOG_WRITE_FAILURE),
                   "failed to write a Gtid_list event to the relay log, "
                   "gtid_slave_pos may be inaccurate");
      delete glev;
    }
    if (likely (rev || glev))
    {
2395
      rli->relay_log.harvest_bytes_written(&rli->log_space_total);
Luis Soares's avatar
Luis Soares committed
2396
      if (flush_master_info(mi, TRUE, TRUE))
2397
        sql_print_error("Failed to flush master info file");
2398
    }
unknown's avatar
unknown committed
2399
  }
2400
  else
Marc Alff's avatar
Marc Alff committed
2401
    mysql_mutex_unlock(log_lock);
2402
  DBUG_VOID_RETURN;
unknown's avatar
unknown committed
2403 2404
}

2405

2406
int register_slave_on_master(MYSQL* mysql, Master_info *mi,
2407
                             bool *suppress_warnings)
2408
{
2409
  uchar buf[1024], *pos= buf;
2410
  uint report_host_len=0, report_user_len=0, report_password_len=0;
2411
  DBUG_ENTER("register_slave_on_master");
2412

2413
  *suppress_warnings= FALSE;
2414 2415 2416 2417 2418 2419 2420 2421
  if (report_host)
    report_host_len= strlen(report_host);
  if (report_host_len > HOSTNAME_LENGTH)
  {
    sql_print_warning("The length of report_host is %d. "
                      "It is larger than the max length(%d), so this "
                      "slave cannot be registered to the master.",
                      report_host_len, HOSTNAME_LENGTH);
2422
    DBUG_RETURN(0);
2423 2424
  }

2425
  if (report_user)
2426
    report_user_len= strlen(report_user);
2427 2428 2429 2430 2431 2432 2433 2434 2435
  if (report_user_len > USERNAME_LENGTH)
  {
    sql_print_warning("The length of report_user is %d. "
                      "It is larger than the max length(%d), so this "
                      "slave cannot be registered to the master.",
                      report_user_len, USERNAME_LENGTH);
    DBUG_RETURN(0);
  }

unknown's avatar
unknown committed
2436
  if (report_password)
2437
    report_password_len= strlen(report_password);
2438 2439 2440 2441 2442 2443 2444 2445
  if (report_password_len > MAX_PASSWORD_LENGTH)
  {
    sql_print_warning("The length of report_password is %d. "
                      "It is larger than the max length(%d), so this "
                      "slave cannot be registered to the master.",
                      report_password_len, MAX_PASSWORD_LENGTH);
    DBUG_RETURN(0);
  }
2446

2447
  int4store(pos, global_system_variables.server_id); pos+= 4;
2448 2449 2450
  pos= net_store_data(pos, (uchar*) report_host, report_host_len);
  pos= net_store_data(pos, (uchar*) report_user, report_user_len);
  pos= net_store_data(pos, (uchar*) report_password, report_password_len);
2451
  int2store(pos, (uint16) report_port); pos+= 2;
2452 2453 2454 2455 2456 2457
  /* 
    Fake rpl_recovery_rank, which was removed in BUG#13963,
    so that this server can register itself on old servers,
    see BUG#49259.
   */
  int4store(pos, /* rpl_recovery_rank */ 0);    pos+= 4;
2458
  /* The master will fill in master_id */
unknown's avatar
unknown committed
2459
  int4store(pos, 0);                    pos+= 4;
2460

2461
  if (simple_command(mysql, COM_REGISTER_SLAVE, buf, (size_t) (pos- buf), 0))
2462
  {
2463 2464 2465 2466
    if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
    {
      *suppress_warnings= TRUE;                 // Suppress reconnect warning
    }
Michael Widenius's avatar
Michael Widenius committed
2467
    else if (!check_io_slave_killed(mi, NULL))
2468 2469 2470 2471
    {
      char buf[256];
      my_snprintf(buf, sizeof(buf), "%s (Errno: %d)", mysql_error(mysql), 
                  mysql_errno(mysql));
2472
      mi->report(ERROR_LEVEL, ER_SLAVE_MASTER_COM_FAILURE, NULL,
2473 2474
                 ER(ER_SLAVE_MASTER_COM_FAILURE), "COM_REGISTER_SLAVE", buf);
    }
2475
    DBUG_RETURN(1);
2476
  }
2477
  DBUG_RETURN(0);
2478 2479
}

2480

2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491
/**
  Execute a SHOW SLAVE STATUS statement.

  @param thd Pointer to THD object for the client thread executing the
  statement.

  @param mi Pointer to Master_info object for the IO thread.

  @retval FALSE success
  @retval TRUE failure
*/
2492 2493 2494 2495

bool show_master_info(THD *thd, Master_info *mi, bool full)
{
  DBUG_ENTER("show_master_info");
unknown's avatar
unknown committed
2496
  String gtid_pos;
2497

unknown's avatar
unknown committed
2498
  if (full && rpl_global_gtid_slave_state.tostring(&gtid_pos, NULL, 0))
2499
    DBUG_RETURN(TRUE);
unknown's avatar
unknown committed
2500 2501 2502
  if (send_show_master_info_header(thd, full, gtid_pos.length()))
    DBUG_RETURN(TRUE);
  if (send_show_master_info_data(thd, mi, full, &gtid_pos))
2503 2504 2505 2506 2507
    DBUG_RETURN(TRUE);
  my_eof(thd);
  DBUG_RETURN(FALSE);
}

unknown's avatar
unknown committed
2508 2509
static bool send_show_master_info_header(THD *thd, bool full,
                                         size_t gtid_pos_length)
unknown's avatar
unknown committed
2510 2511
{
  List<Item> field_list;
2512
  Protocol *protocol= thd->protocol;
2513
  Master_info *mi;
2514 2515 2516
  DBUG_ENTER("show_master_info_header");

  if (full)
2517
  {
2518 2519
    field_list.push_back(new Item_empty_string("Connection_name",
                                               MAX_CONNECTION_NAME));
2520 2521 2522
    field_list.push_back(new Item_empty_string("Slave_SQL_State",
                                               30));
  }
2523

unknown's avatar
unknown committed
2524
  field_list.push_back(new Item_empty_string("Slave_IO_State",
2525
                                                     30));
unknown's avatar
unknown committed
2526
  field_list.push_back(new Item_empty_string("Master_Host",
unknown's avatar
unknown committed
2527
                                                     sizeof(mi->host)));
unknown's avatar
unknown committed
2528
  field_list.push_back(new Item_empty_string("Master_User",
unknown's avatar
unknown committed
2529
                                                     sizeof(mi->user)));
2530
  field_list.push_back(new Item_return_int("Master_Port", 7,
unknown's avatar
unknown committed
2531
                                           MYSQL_TYPE_LONG));
2532
  field_list.push_back(new Item_return_int("Connect_Retry", 10,
unknown's avatar
unknown committed
2533
                                           MYSQL_TYPE_LONG));
2534
  field_list.push_back(new Item_empty_string("Master_Log_File",
unknown's avatar
unknown committed
2535
                                             FN_REFLEN));
2536
  field_list.push_back(new Item_return_int("Read_Master_Log_Pos", 10,
unknown's avatar
unknown committed
2537
                                           MYSQL_TYPE_LONGLONG));
2538
  field_list.push_back(new Item_empty_string("Relay_Log_File",
unknown's avatar
unknown committed
2539
                                             FN_REFLEN));
2540
  field_list.push_back(new Item_return_int("Relay_Log_Pos", 10,
unknown's avatar
unknown committed
2541
                                           MYSQL_TYPE_LONGLONG));
2542
  field_list.push_back(new Item_empty_string("Relay_Master_Log_File",
unknown's avatar
unknown committed
2543
                                             FN_REFLEN));
2544 2545
  field_list.push_back(new Item_empty_string("Slave_IO_Running", 3));
  field_list.push_back(new Item_empty_string("Slave_SQL_Running", 3));
2546 2547 2548 2549 2550 2551
  field_list.push_back(new Item_empty_string("Replicate_Do_DB", 20));
  field_list.push_back(new Item_empty_string("Replicate_Ignore_DB", 20));
  field_list.push_back(new Item_empty_string("Replicate_Do_Table", 20));
  field_list.push_back(new Item_empty_string("Replicate_Ignore_Table", 23));
  field_list.push_back(new Item_empty_string("Replicate_Wild_Do_Table", 24));
  field_list.push_back(new Item_empty_string("Replicate_Wild_Ignore_Table",
unknown's avatar
unknown committed
2552
                                             28));
2553 2554 2555
  field_list.push_back(new Item_return_int("Last_Errno", 4, MYSQL_TYPE_LONG));
  field_list.push_back(new Item_empty_string("Last_Error", 20));
  field_list.push_back(new Item_return_int("Skip_Counter", 10,
unknown's avatar
unknown committed
2556
                                           MYSQL_TYPE_LONG));
2557
  field_list.push_back(new Item_return_int("Exec_Master_Log_Pos", 10,
unknown's avatar
unknown committed
2558
                                           MYSQL_TYPE_LONGLONG));
2559
  field_list.push_back(new Item_return_int("Relay_Log_Space", 10,
unknown's avatar
unknown committed
2560
                                           MYSQL_TYPE_LONGLONG));
2561
  field_list.push_back(new Item_empty_string("Until_Condition", 6));
2562
  field_list.push_back(new Item_empty_string("Until_Log_File", FN_REFLEN));
unknown's avatar
unknown committed
2563
  field_list.push_back(new Item_return_int("Until_Log_Pos", 10,
2564
                                           MYSQL_TYPE_LONGLONG));
unknown's avatar
unknown committed
2565 2566 2567
  field_list.push_back(new Item_empty_string("Master_SSL_Allowed", 7));
  field_list.push_back(new Item_empty_string("Master_SSL_CA_File",
                                             sizeof(mi->ssl_ca)));
unknown's avatar
unknown committed
2568
  field_list.push_back(new Item_empty_string("Master_SSL_CA_Path",
unknown's avatar
unknown committed
2569
                                             sizeof(mi->ssl_capath)));
unknown's avatar
unknown committed
2570
  field_list.push_back(new Item_empty_string("Master_SSL_Cert",
unknown's avatar
unknown committed
2571
                                             sizeof(mi->ssl_cert)));
unknown's avatar
unknown committed
2572
  field_list.push_back(new Item_empty_string("Master_SSL_Cipher",
unknown's avatar
unknown committed
2573
                                             sizeof(mi->ssl_cipher)));
unknown's avatar
unknown committed
2574
  field_list.push_back(new Item_empty_string("Master_SSL_Key",
unknown's avatar
unknown committed
2575
                                             sizeof(mi->ssl_key)));
2576
  field_list.push_back(new Item_return_int("Seconds_Behind_Master", 10,
unknown's avatar
unknown committed
2577
                                           MYSQL_TYPE_LONGLONG));
2578 2579
  field_list.push_back(new Item_empty_string("Master_SSL_Verify_Server_Cert",
                                             3));
2580 2581 2582 2583
  field_list.push_back(new Item_return_int("Last_IO_Errno", 4, MYSQL_TYPE_LONG));
  field_list.push_back(new Item_empty_string("Last_IO_Error", 20));
  field_list.push_back(new Item_return_int("Last_SQL_Errno", 4, MYSQL_TYPE_LONG));
  field_list.push_back(new Item_empty_string("Last_SQL_Error", 20));
Andrei Elkin's avatar
Andrei Elkin committed
2584 2585 2586 2587
  field_list.push_back(new Item_empty_string("Replicate_Ignore_Server_Ids",
                                             FN_REFLEN));
  field_list.push_back(new Item_return_int("Master_Server_Id", sizeof(ulong),
                                           MYSQL_TYPE_LONG));
Michael Widenius's avatar
Michael Widenius committed
2588 2589 2590 2591
  field_list.push_back(new Item_empty_string("Master_SSL_Crl",
                                             sizeof(mi->ssl_crl)));
  field_list.push_back(new Item_empty_string("Master_SSL_Crlpath",
                                             sizeof(mi->ssl_crlpath)));
unknown's avatar
unknown committed
2592 2593
  field_list.push_back(new Item_empty_string("Using_Gtid",
                                             sizeof("Current_Pos")-1));
2594
  field_list.push_back(new Item_empty_string("Gtid_IO_Pos", 30));
2595 2596 2597 2598 2599 2600 2601 2602
  if (full)
  {
    field_list.push_back(new Item_return_int("Retried_transactions",
                                             10, MYSQL_TYPE_LONG));
    field_list.push_back(new Item_return_int("Max_relay_log_size",
                                             10, MYSQL_TYPE_LONGLONG));
    field_list.push_back(new Item_return_int("Executed_log_entries",
                                             10, MYSQL_TYPE_LONG));
2603 2604 2605 2606
    field_list.push_back(new Item_return_int("Slave_received_heartbeats",
                                             10, MYSQL_TYPE_LONG));
    field_list.push_back(new Item_float("Slave_heartbeat_period",
                                        0.0, 3, 10));
unknown's avatar
unknown committed
2607 2608
    field_list.push_back(new Item_empty_string("Gtid_Slave_Pos",
                                               gtid_pos_length));
2609
  }
unknown's avatar
unknown committed
2610

2611
  if (protocol->send_result_set_metadata(&field_list,
2612
                            Protocol::SEND_NUM_ROWS | Protocol::SEND_EOF))
unknown's avatar
unknown committed
2613
    DBUG_RETURN(TRUE);
2614 2615 2616 2617
  DBUG_RETURN(FALSE);
}


unknown's avatar
unknown committed
2618 2619
static bool send_show_master_info_data(THD *thd, Master_info *mi, bool full,
                                       String *gtid_pos)
2620 2621
{
  DBUG_ENTER("send_show_master_info_data");
unknown's avatar
unknown committed
2622

2623 2624
  if (mi->host[0])
  {
2625
    DBUG_PRINT("info",("host is set: '%s'", mi->host));
2626
    String *packet= &thd->packet;
2627
    Protocol *protocol= thd->protocol;
2628
    Rpl_filter *rpl_filter= mi->rpl_filter;
2629 2630
    char buf[256];
    String tmp(buf, sizeof(buf), &my_charset_bin);
2631

2632
    protocol->prepare_for_resend();
unknown's avatar
unknown committed
2633

2634
    /*
2635 2636
      slave_running can be accessed without run_lock but not other
      non-volotile members like mi->io_thd, which is guarded by the mutex.
2637
    */
2638 2639 2640
    if (full)
      protocol->store(mi->connection_name.str, mi->connection_name.length,
                      &my_charset_bin);
Marc Alff's avatar
Marc Alff committed
2641
    mysql_mutex_lock(&mi->run_lock);
2642
    if (full)
Michael Widenius's avatar
Michael Widenius committed
2643 2644 2645 2646 2647 2648
    {
      /*
        Show what the sql driver replication thread is doing
        This is only meaningful if there is only one slave thread.
      */
      protocol->store(mi->rli.sql_driver_thd ?
2649
                      mi->rli.sql_driver_thd->get_proc_info() : "",
2650
                      &my_charset_bin);
Michael Widenius's avatar
Michael Widenius committed
2651
    }
Sergei Golubchik's avatar
Sergei Golubchik committed
2652
    protocol->store(mi->io_thd ? mi->io_thd->get_proc_info() : "", &my_charset_bin);
Marc Alff's avatar
Marc Alff committed
2653
    mysql_mutex_unlock(&mi->run_lock);
2654

Marc Alff's avatar
Marc Alff committed
2655 2656 2657 2658
    mysql_mutex_lock(&mi->data_lock);
    mysql_mutex_lock(&mi->rli.data_lock);
    mysql_mutex_lock(&mi->err_lock);
    mysql_mutex_lock(&mi->rli.err_lock);
2659 2660
    protocol->store(mi->host, &my_charset_bin);
    protocol->store(mi->user, &my_charset_bin);
2661 2662
    protocol->store((uint32) mi->port);
    protocol->store((uint32) mi->connect_retry);
2663
    protocol->store(mi->master_log_name, &my_charset_bin);
2664
    protocol->store((ulonglong) mi->master_log_pos);
2665
    protocol->store(mi->rli.group_relay_log_name +
unknown's avatar
unknown committed
2666 2667
                    dirname_length(mi->rli.group_relay_log_name),
                    &my_charset_bin);
2668 2669
    protocol->store((ulonglong) mi->rli.group_relay_log_pos);
    protocol->store(mi->rli.group_master_log_name, &my_charset_bin);
unknown's avatar
unknown committed
2670
    protocol->store(mi->slave_running == MYSQL_SLAVE_RUN_CONNECT ?
2671 2672
                    "Yes" : (mi->slave_running == MYSQL_SLAVE_RUN_NOT_CONNECT ?
                             "Connecting" : "No"), &my_charset_bin);
2673
    protocol->store(mi->rli.slave_running ? "Yes":"No", &my_charset_bin);
2674 2675 2676 2677
    protocol->store(rpl_filter->get_do_db());
    protocol->store(rpl_filter->get_ignore_db());

    rpl_filter->get_do_table(&tmp);
2678
    protocol->store(&tmp);
2679
    rpl_filter->get_ignore_table(&tmp);
2680
    protocol->store(&tmp);
2681
    rpl_filter->get_wild_do_table(&tmp);
2682
    protocol->store(&tmp);
2683
    rpl_filter->get_wild_ignore_table(&tmp);
2684 2685
    protocol->store(&tmp);

2686 2687
    protocol->store(mi->rli.last_error().number);
    protocol->store(mi->rli.last_error().message, &my_charset_bin);
2688
    protocol->store((uint32) mi->rli.slave_skip_counter);
2689
    protocol->store((ulonglong) mi->rli.group_master_log_pos);
2690
    protocol->store((ulonglong) mi->rli.log_space_total);
2691 2692

    protocol->store(
2693 2694
      mi->rli.until_condition==Relay_log_info::UNTIL_NONE ? "None":
        ( mi->rli.until_condition==Relay_log_info::UNTIL_MASTER_POS? "Master":
unknown's avatar
unknown committed
2695 2696
          ( mi->rli.until_condition==Relay_log_info::UNTIL_RELAY_POS? "Relay":
            "Gtid")), &my_charset_bin);
2697 2698
    protocol->store(mi->rli.until_log_name, &my_charset_bin);
    protocol->store((ulonglong) mi->rli.until_log_pos);
unknown's avatar
unknown committed
2699 2700

#ifdef HAVE_OPENSSL
unknown's avatar
unknown committed
2701 2702 2703 2704 2705 2706 2707 2708 2709
    protocol->store(mi->ssl? "Yes":"No", &my_charset_bin);
#else
    protocol->store(mi->ssl? "Ignored":"No", &my_charset_bin);
#endif
    protocol->store(mi->ssl_ca, &my_charset_bin);
    protocol->store(mi->ssl_capath, &my_charset_bin);
    protocol->store(mi->ssl_cert, &my_charset_bin);
    protocol->store(mi->ssl_cipher, &my_charset_bin);
    protocol->store(mi->ssl_key, &my_charset_bin);
unknown's avatar
unknown committed
2710

2711 2712 2713 2714 2715 2716
    /*
      Seconds_Behind_Master: if SQL thread is running and I/O thread is
      connected, we can compute it otherwise show NULL (i.e. unknown).
    */
    if ((mi->slave_running == MYSQL_SLAVE_RUN_CONNECT) &&
        mi->rli.slave_running)
2717
    {
2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735
      long time_diff;
      bool idle;
      time_t stamp= mi->rli.last_master_timestamp;

      if (!stamp)
        idle= true;
      else
      {
        idle= mi->rli.sql_thread_caught_up;
        if (opt_slave_parallel_threads > 0 && idle &&
            !mi->rli.parallel.workers_idle())
          idle= false;
      }
      if (idle)
        time_diff= 0;
      else
      {
        time_diff= ((long)(time(0) - stamp) - mi->clock_diff_with_master);
2736
      /*
unknown's avatar
unknown committed
2737 2738
        Apparently on some systems time_diff can be <0. Here are possible
        reasons related to MySQL:
2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750
        - the master is itself a slave of another master whose time is ahead.
        - somebody used an explicit SET TIMESTAMP on the master.
        Possible reason related to granularity-to-second of time functions
        (nothing to do with MySQL), which can explain a value of -1:
        assume the master's and slave's time are perfectly synchronized, and
        that at slave's connection time, when the master's timestamp is read,
        it is at the very end of second 1, and (a very short time later) when
        the slave's timestamp is read it is at the very beginning of second
        2. Then the recorded value for master is 1 and the recorded value for
        slave is 2. At SHOW SLAVE STATUS time, assume that the difference
        between timestamp of slave and rli->last_master_timestamp is 0
        (i.e. they are in the same second), then we get 0-(2-1)=-1 as a result.
2751
        This confuses users, so we don't go below 0.
2752 2753 2754

        last_master_timestamp == 0 (an "impossible" timestamp 1970) is a
        special marker to say "consider we have caught up".
2755
      */
2756 2757 2758 2759
        if (time_diff < 0)
          time_diff= 0;
      }
      protocol->store((longlong)time_diff);
2760
    }
unknown's avatar
unknown committed
2761
    else
2762
    {
unknown's avatar
unknown committed
2763
      protocol->store_null();
2764 2765
    }
    protocol->store(mi->ssl_verify_server_cert? "Yes":"No", &my_charset_bin);
unknown's avatar
unknown committed
2766

2767
    // Last_IO_Errno
2768
    protocol->store(mi->last_error().number);
2769
    // Last_IO_Error
2770
    protocol->store(mi->last_error().message, &my_charset_bin);
2771
    // Last_SQL_Errno
2772
    protocol->store(mi->rli.last_error().number);
2773
    // Last_SQL_Error
2774
    protocol->store(mi->rli.last_error().message, &my_charset_bin);
Andrei Elkin's avatar
Andrei Elkin committed
2775 2776 2777 2778 2779 2780 2781 2782 2783 2784
    // Replicate_Ignore_Server_Ids
    {
      char buff[FN_REFLEN];
      ulong i, cur_len;
      for (i= 0, buff[0]= 0, cur_len= 0;
           i < mi->ignore_server_ids.elements; i++)
      {
        ulong s_id, slen;
        char sbuff[FN_REFLEN];
        get_dynamic(&mi->ignore_server_ids, (uchar*) &s_id, i);
2785
        slen= sprintf(sbuff, (i==0? "%lu" : ", %lu"), s_id);
Andrei Elkin's avatar
Andrei Elkin committed
2786 2787 2788 2789 2790 2791
        if (cur_len + slen + 4 > FN_REFLEN)
        {
          /*
            break the loop whenever remained space could not fit
            ellipses on the next cycle
          */
2792
          sprintf(buff + cur_len, "...");
Andrei Elkin's avatar
Andrei Elkin committed
2793 2794
          break;
        }
2795
        cur_len += sprintf(buff + cur_len, "%s", sbuff);
Andrei Elkin's avatar
Andrei Elkin committed
2796 2797 2798 2799 2800
      }
      protocol->store(buff, &my_charset_bin);
    }
    // Master_Server_id
    protocol->store((uint32) mi->master_id);
Michael Widenius's avatar
Michael Widenius committed
2801 2802 2803 2804
    // Master_Ssl_Crl
    protocol->store(mi->ssl_ca, &my_charset_bin);
    // Master_Ssl_Crlpath
    protocol->store(mi->ssl_capath, &my_charset_bin);
unknown's avatar
unknown committed
2805
    protocol->store(mi->using_gtid_astext(mi->using_gtid), &my_charset_bin);
2806 2807 2808 2809 2810 2811
    {
      char buff[30];
      String tmp(buff, sizeof(buff), system_charset_info);
      mi->gtid_current_pos.to_string(&tmp);
      protocol->store(tmp.ptr(), tmp.length(), &my_charset_bin);
    }
2812 2813 2814 2815 2816
    if (full)
    {
      protocol->store((uint32)    mi->rli.retried_trans);
      protocol->store((ulonglong) mi->rli.max_relay_log_size);
      protocol->store((uint32)    mi->rli.executed_entries);
2817 2818
      protocol->store((uint32)    mi->received_heartbeats);
      protocol->store((double)    mi->heartbeat_period, 3, &tmp);
unknown's avatar
unknown committed
2819
      protocol->store(gtid_pos->ptr(), gtid_pos->length(), &my_charset_bin);
2820
    }
2821

Marc Alff's avatar
Marc Alff committed
2822 2823 2824 2825
    mysql_mutex_unlock(&mi->rli.err_lock);
    mysql_mutex_unlock(&mi->err_lock);
    mysql_mutex_unlock(&mi->rli.data_lock);
    mysql_mutex_unlock(&mi->data_lock);
2826

2827
    if (my_net_write(&thd->net, (uchar*) thd->packet.ptr(), packet->length()))
unknown's avatar
unknown committed
2828
      DBUG_RETURN(TRUE);
2829
  }
2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853
  DBUG_RETURN(FALSE);
}


/* Used to sort connections by name */

static int cmp_mi_by_name(const Master_info **arg1,
                          const Master_info **arg2)
{
  return my_strcasecmp(system_charset_info, (*arg1)->connection_name.str,
                       (*arg2)->connection_name.str);
}


/**
  Execute a SHOW FULL SLAVE STATUS statement.

  @param thd Pointer to THD object for the client thread executing the
  statement.

  Elements are sorted according to the original connection_name.

  @retval FALSE success
  @retval TRUE failure
2854 2855 2856

  @note
  master_info_index is protected by LOCK_active_mi.
2857 2858 2859 2860 2861
*/

bool show_all_master_info(THD* thd)
{
  uint i, elements;
unknown's avatar
unknown committed
2862
  String gtid_pos;
2863 2864
  Master_info **tmp;
  DBUG_ENTER("show_master_info");
2865
  mysql_mutex_assert_owner(&LOCK_active_mi);
2866

unknown's avatar
unknown committed
2867 2868 2869 2870 2871 2872
  gtid_pos.length(0);
  if (rpl_append_gtid_state(&gtid_pos, true))
  {
    my_error(ER_OUT_OF_RESOURCES, MYF(0));
    DBUG_RETURN(TRUE);
  }
unknown's avatar
unknown committed
2873 2874

  if (send_show_master_info_header(thd, 1, gtid_pos.length()))
2875 2876
    DBUG_RETURN(TRUE);

2877 2878
  if (!master_info_index ||
      !(elements= master_info_index->master_info_hash.records))
2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896
    goto end;

  /*
    Sort lines to get them into a predicted order
    (needed for test cases and to not confuse users)
  */
  if (!(tmp= (Master_info**) thd->alloc(sizeof(Master_info*) * elements)))
    DBUG_RETURN(TRUE);

  for (i= 0; i < elements; i++)
  {
    tmp[i]= (Master_info *) my_hash_element(&master_info_index->
                                            master_info_hash, i);
  }
  my_qsort(tmp, elements, sizeof(Master_info*), (qsort_cmp) cmp_mi_by_name);

  for (i= 0; i < elements; i++)
  {
unknown's avatar
unknown committed
2897
    if (send_show_master_info_data(thd, tmp[i], 1, &gtid_pos))
2898 2899 2900 2901
      DBUG_RETURN(TRUE);
  }

end:
2902
  my_eof(thd);
unknown's avatar
unknown committed
2903
  DBUG_RETURN(FALSE);
unknown's avatar
unknown committed
2904 2905
}

unknown's avatar
unknown committed
2906

2907 2908
void set_slave_thread_options(THD* thd)
{
2909
  DBUG_ENTER("set_slave_thread_options");
2910 2911 2912 2913 2914 2915 2916 2917 2918
  /*
     It's nonsense to constrain the slave threads with max_join_size; if a
     query succeeded on master, we HAVE to execute it. So set
     OPTION_BIG_SELECTS. Setting max_join_size to HA_POS_ERROR is not enough
     (and it's not needed if we have OPTION_BIG_SELECTS) because an INSERT
     SELECT examining more than 4 billion rows would still fail (yes, because
     when max_join_size is 4G, OPTION_BIG_SELECTS is automatically set, but
     only for client threads.
  */
2919
  ulonglong options= thd->variables.option_bits | OPTION_BIG_SELECTS;
2920 2921 2922 2923
  if (opt_log_slave_updates)
    options|= OPTION_BIN_LOG;
  else
    options&= ~OPTION_BIN_LOG;
2924
  thd->variables.option_bits= options;
2925
  thd->variables.completion_type= 0;
2926
  DBUG_VOID_RETURN;
2927
}
2928

2929
void set_slave_thread_default_charset(THD* thd, rpl_group_info *rgi)
2930
{
2931 2932
  DBUG_ENTER("set_slave_thread_default_charset");

2933 2934 2935 2936 2937 2938 2939
  thd->variables.character_set_client=
    global_system_variables.character_set_client;
  thd->variables.collation_connection=
    global_system_variables.collation_connection;
  thd->variables.collation_server=
    global_system_variables.collation_server;
  thd->update_charset();
2940

2941
  thd->system_thread_info.rpl_sql_info->cached_charset_invalidate();
2942
  DBUG_VOID_RETURN;
2943 2944
}

unknown's avatar
unknown committed
2945
/*
2946
  init_slave_thread()
unknown's avatar
unknown committed
2947
*/
2948

2949 2950
static int init_slave_thread(THD* thd, Master_info *mi,
                             SLAVE_THD_TYPE thd_type)
unknown's avatar
unknown committed
2951 2952
{
  DBUG_ENTER("init_slave_thread");
Sergei Golubchik's avatar
Sergei Golubchik committed
2953
  int simulate_error __attribute__((unused))= 0;
2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966
  DBUG_EXECUTE_IF("simulate_io_slave_error_on_init",
                  simulate_error|= (1 << SLAVE_THD_IO););
  DBUG_EXECUTE_IF("simulate_sql_slave_error_on_init",
                  simulate_error|= (1 << SLAVE_THD_SQL););
  /* We must call store_globals() before doing my_net_init() */
  if (init_thr_lock() || thd->store_globals() ||
      my_net_init(&thd->net, 0, MYF(MY_THREAD_SPECIFIC)) ||
      IF_DBUG(simulate_error & (1<< thd_type), 0))
  {
    thd->cleanup();
    DBUG_RETURN(-1);
  }

2967
  thd->system_thread = (thd_type == SLAVE_THD_SQL) ?
unknown's avatar
unknown committed
2968
    SYSTEM_THREAD_SLAVE_SQL : SYSTEM_THREAD_SLAVE_IO;
2969
  thd->security_ctx->skip_grants();
2970 2971
  thd->slave_thread= 1;
  thd->connection_name= mi->connection_name;
2972
  thd->enable_slow_log= opt_log_slow_slave_statements;
2973
  thd->variables.log_slow_filter= global_system_variables.log_slow_filter;
2974
  set_slave_thread_options(thd);
unknown's avatar
unknown committed
2975
  thd->client_capabilities = CLIENT_LOCAL_FILES;
Marc Alff's avatar
Marc Alff committed
2976
  mysql_mutex_lock(&LOCK_thread_count);
unknown's avatar
unknown committed
2977
  thd->thread_id= thd->variables.pseudo_thread_id= thread_id++;
Marc Alff's avatar
Marc Alff committed
2978
  mysql_mutex_unlock(&LOCK_thread_count);
unknown's avatar
unknown committed
2979

2980
  if (thd_type == SLAVE_THD_SQL)
Sergei Golubchik's avatar
Sergei Golubchik committed
2981
    THD_STAGE_INFO(thd, stage_waiting_for_the_next_event_in_relay_log);
2982
  else
Sergei Golubchik's avatar
Sergei Golubchik committed
2983
    THD_STAGE_INFO(thd, stage_waiting_for_master_update);
unknown's avatar
unknown committed
2984
  thd->set_time();
2985 2986
  /* Do not use user-supplied timeout value for system threads. */
  thd->variables.lock_wait_timeout= LONG_TIMEOUT;
unknown's avatar
unknown committed
2987 2988 2989
  DBUG_RETURN(0);
}

2990 2991 2992 2993 2994 2995 2996
/*
  Sleep for a given amount of time or until killed.

  @param thd        Thread context of the current thread.
  @param seconds    The number of seconds to sleep.
  @param func       Function object to check if the thread has been killed.
  @param info       The Rpl_info object associated with this sleep.
2997

2998 2999 3000
  @retval True if the thread has been killed, false otherwise.
*/
template <typename killed_func, typename rpl_info>
Michael Widenius's avatar
Michael Widenius committed
3001 3002
static bool slave_sleep(THD *thd, time_t seconds,
                        killed_func func, rpl_info info)
unknown's avatar
unknown committed
3003
{
3004

3005 3006
  bool ret;
  struct timespec abstime;
unknown's avatar
unknown committed
3007

3008 3009
  mysql_mutex_t *lock= &info->sleep_lock;
  mysql_cond_t *cond= &info->sleep_cond;
unknown's avatar
unknown committed
3010

3011 3012 3013
  /* Absolute system time at which the sleep time expires. */
  set_timespec(abstime, seconds);
  mysql_mutex_lock(lock);
Sergei Golubchik's avatar
Sergei Golubchik committed
3014
  thd->ENTER_COND(cond, lock, NULL, NULL);
3015

Michael Widenius's avatar
Michael Widenius committed
3016
  while (! (ret= func(info)))
3017 3018 3019 3020
  {
    int error= mysql_cond_timedwait(cond, lock, &abstime);
    if (error == ETIMEDOUT || error == ETIME)
      break;
unknown's avatar
unknown committed
3021
  }
3022
  /* Implicitly unlocks the mutex. */
Sergei Golubchik's avatar
Sergei Golubchik committed
3023
  thd->EXIT_COND(NULL);
3024
  return ret;
unknown's avatar
unknown committed
3025 3026
}

3027

He Zhenxing's avatar
He Zhenxing committed
3028 3029
static int request_dump(THD *thd, MYSQL* mysql, Master_info* mi,
			bool *suppress_warnings)
unknown's avatar
unknown committed
3030
{
3031
  uchar buf[FN_REFLEN + 10];
unknown's avatar
unknown committed
3032
  int len;
He Zhenxing's avatar
He Zhenxing committed
3033
  ushort binlog_flags = 0; // for now
3034
  char* logname = mi->master_log_name;
3035
  DBUG_ENTER("request_dump");
3036 3037
  
  *suppress_warnings= FALSE;
3038

3039
  if (opt_log_slave_updates && opt_replicate_annotate_row_events)
3040 3041
    binlog_flags|= BINLOG_SEND_ANNOTATE_ROWS_EVENT;

He Zhenxing's avatar
He Zhenxing committed
3042 3043 3044 3045 3046
  if (RUN_HOOK(binlog_relay_io,
               before_request_transmit,
               (thd, mi, binlog_flags)))
    DBUG_RETURN(1);
  
unknown's avatar
unknown committed
3047
  // TODO if big log files: Change next to int8store()
unknown's avatar
unknown committed
3048
  int4store(buf, (ulong) mi->master_log_pos);
unknown's avatar
unknown committed
3049
  int2store(buf + 4, binlog_flags);
3050
  int4store(buf + 6, global_system_variables.server_id);
unknown's avatar
unknown committed
3051
  len = (uint) strlen(logname);
3052
  memcpy(buf + 10, logname,len);
unknown's avatar
SCRUM  
unknown committed
3053
  if (simple_command(mysql, COM_BINLOG_DUMP, buf, len + 10, 1))
unknown's avatar
unknown committed
3054
  {
unknown's avatar
unknown committed
3055 3056 3057 3058 3059
    /*
      Something went wrong, so we will just reconnect and retry later
      in the future, we should do a better error analysis, but for
      now we just fill up the error log :-)
    */
unknown's avatar
SCRUM  
unknown committed
3060
    if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
3061
      *suppress_warnings= TRUE;                 // Suppress reconnect warning
unknown's avatar
unknown committed
3062
    else
3063
      sql_print_error("Error on COM_BINLOG_DUMP: %d  %s, will retry in %d secs",
unknown's avatar
unknown committed
3064
                      mysql_errno(mysql), mysql_error(mysql),
3065
                      mi->connect_retry);
3066
    DBUG_RETURN(1);
unknown's avatar
unknown committed
3067
  }
unknown's avatar
unknown committed
3068

3069
  DBUG_RETURN(0);
unknown's avatar
unknown committed
3070 3071
}

3072

unknown's avatar
unknown committed
3073
/*
3074
  Read one event from the master
unknown's avatar
unknown committed
3075

unknown's avatar
unknown committed
3076 3077
  SYNOPSIS
    read_event()
unknown's avatar
unknown committed
3078 3079 3080 3081 3082 3083
    mysql               MySQL connection
    mi                  Master connection information
    suppress_warnings   TRUE when a normal net read timeout has caused us to
                        try a reconnect.  We do not want to print anything to
                        the error log in this case because this a anormal
                        event in an idle server.
3084

unknown's avatar
unknown committed
3085
    RETURN VALUES
unknown's avatar
unknown committed
3086 3087
    'packet_error'      Error
    number              Length of packet
unknown's avatar
unknown committed
3088 3089
*/

3090
static ulong read_event(MYSQL* mysql, Master_info *mi, bool* suppress_warnings)
unknown's avatar
unknown committed
3091
{
3092
  ulong len;
3093
  DBUG_ENTER("read_event");
unknown's avatar
unknown committed
3094

3095
  *suppress_warnings= FALSE;
unknown's avatar
unknown committed
3096 3097 3098 3099
  /*
    my_real_read() will time us out
    We check if we were told to die, and if not, try reading again
  */
unknown's avatar
unknown committed
3100
#ifndef DBUG_OFF
3101
  if (disconnect_slave_event_count && !(mi->events_till_disconnect--))
unknown's avatar
unknown committed
3102
    DBUG_RETURN(packet_error);
unknown's avatar
unknown committed
3103
#endif
unknown's avatar
unknown committed
3104

3105
  len = cli_safe_read(mysql);
3106
  if (len == packet_error || (long) len < 1)
unknown's avatar
unknown committed
3107
  {
unknown's avatar
SCRUM  
unknown committed
3108
    if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
unknown's avatar
unknown committed
3109 3110
    {
      /*
unknown's avatar
unknown committed
3111 3112 3113
        We are trying a normal reconnect after a read timeout;
        we suppress prints to .err file as long as the reconnect
        happens without problems
unknown's avatar
unknown committed
3114 3115 3116 3117
      */
      *suppress_warnings= TRUE;
    }
    else
3118
      sql_print_error("Error reading packet from server: %s ( server_errno=%d)",
unknown's avatar
unknown committed
3119
                      mysql_error(mysql), mysql_errno(mysql));
3120
    DBUG_RETURN(packet_error);
unknown's avatar
unknown committed
3121 3122
  }

3123 3124
  /* Check if eof packet */
  if (len < 8 && mysql->net.read_pos[0] == 254)
unknown's avatar
unknown committed
3125
  {
3126 3127
    sql_print_information("Slave: received end packet from server, apparent "
                          "master shutdown: %s",
unknown's avatar
unknown committed
3128
                     mysql_error(mysql));
3129
     DBUG_RETURN(packet_error);
unknown's avatar
unknown committed
3130
  }
unknown's avatar
unknown committed
3131

unknown's avatar
unknown committed
3132
  DBUG_PRINT("exit", ("len: %lu  net->read_pos[4]: %d",
unknown's avatar
unknown committed
3133 3134
                      len, mysql->net.read_pos[4]));
  DBUG_RETURN(len - 1);
unknown's avatar
unknown committed
3135 3136
}

unknown's avatar
unknown committed
3137 3138 3139 3140 3141 3142 3143
/*
  Check if the current error is of temporary nature of not.
  Some errors are temporary in nature, such as
  ER_LOCK_DEADLOCK and ER_LOCK_WAIT_TIMEOUT.  Ndb also signals
  that the error is temporary by pushing a warning with the error code
  ER_GET_TEMPORARY_ERRMSG, if the originating error is temporary.
*/
3144 3145
int
has_temporary_error(THD *thd)
unknown's avatar
unknown committed
3146
{
3147 3148
  DBUG_ENTER("has_temporary_error");

3149
  DBUG_EXECUTE_IF("all_errors_are_temporary_errors",
3150
                  if (thd->get_stmt_da()->is_error())
3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162
                  {
                    thd->clear_error();
                    my_error(ER_LOCK_DEADLOCK, MYF(0));
                  });

  /*
    If there is no message in THD, we can't say if it's a temporary
    error or not. This is currently the case for Incident_log_event,
    which sets no message. Return FALSE.
  */
  if (!thd->is_error())
    DBUG_RETURN(0);
unknown's avatar
unknown committed
3163 3164 3165 3166 3167 3168

  /*
    Temporary error codes:
    currently, InnoDB deadlock detected by InnoDB or lock
    wait timeout (innodb_lock_wait_timeout exceeded
  */
3169 3170
  if (thd->get_stmt_da()->sql_errno() == ER_LOCK_DEADLOCK ||
      thd->get_stmt_da()->sql_errno() == ER_LOCK_WAIT_TIMEOUT)
3171
    DBUG_RETURN(1);
unknown's avatar
unknown committed
3172 3173 3174 3175 3176

#ifdef HAVE_NDB_BINLOG
  /*
    currently temporary error set in ndbcluster
  */
3177 3178
  List_iterator_fast<Sql_condition> it(thd->warning_info->warn_list());
  Sql_condition *err;
unknown's avatar
unknown committed
3179 3180
  while ((err= it++))
  {
Marc Alff's avatar
Marc Alff committed
3181 3182 3183
    DBUG_PRINT("info", ("has condition %d %s", err->get_sql_errno(),
                        err->get_message_text()));
    switch (err->get_sql_errno())
unknown's avatar
unknown committed
3184 3185
    {
    case ER_GET_TEMPORARY_ERRMSG:
3186
      DBUG_RETURN(1);
unknown's avatar
unknown committed
3187 3188 3189 3190 3191
    default:
      break;
    }
  }
#endif
3192
  DBUG_RETURN(0);
unknown's avatar
unknown committed
3193
}
3194

3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221

/**
  Applies the given event and advances the relay log position.

  In essence, this function does:

  @code
    ev->apply_event(rli);
    ev->update_pos(rli);
  @endcode

  But it also does some maintainance, such as skipping events if
  needed and reporting errors.

  If the @c skip flag is set, then it is tested whether the event
  should be skipped, by looking at the slave_skip_counter and the
  server id.  The skip flag should be set when calling this from a
  replication thread but not set when executing an explicit BINLOG
  statement.

  @retval 0 OK.

  @retval 1 Error calling ev->apply_event().

  @retval 2 No error calling ev->apply_event(), but error calling
  ev->update_pos().
*/
3222
int apply_event_and_update_pos(Log_event* ev, THD* thd,
unknown's avatar
unknown committed
3223
                               rpl_group_info *rgi,
3224
                               rpl_parallel_thread *rpt)
3225 3226
{
  int exec_res= 0;
3227
  Relay_log_info* rli= rgi->rli;
3228 3229 3230 3231 3232
  DBUG_ENTER("apply_event_and_update_pos");

  DBUG_PRINT("exec_event",("%s(type_code: %d; server_id: %d)",
                           ev->get_type_str(), ev->get_type_code(),
                           ev->server_id));
3233
  DBUG_PRINT("info", ("thd->options: '%s%s%s'  rgi->last_event_start_time: %lu",
3234 3235
                      FLAGSTR(thd->variables.option_bits, OPTION_NOT_AUTOCOMMIT),
                      FLAGSTR(thd->variables.option_bits, OPTION_BEGIN),
3236
                      FLAGSTR(thd->variables.option_bits, OPTION_GTID_BEGIN),
3237
                      (ulong) rgi->last_event_start_time));
3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262

  /*
    Execute the event to change the database and update the binary
    log coordinates, but first we set some data that is needed for
    the thread.

    The event will be executed unless it is supposed to be skipped.

    Queries originating from this server must be skipped.  Low-level
    events (Format_description_log_event, Rotate_log_event,
    Stop_log_event) from this server must also be skipped. But for
    those we don't want to modify 'group_master_log_pos', because
    these events did not exist on the master.
    Format_description_log_event is not completely skipped.

    Skip queries specified by the user in 'slave_skip_counter'.  We
    can't however skip events that has something to do with the log
    files themselves.

    Filtering on own server id is extremely important, to ignore
    execution of events created by the creation/rotation of the relay
    log (remember that now the relay log starts with its Format_desc,
    has a Rotate etc).
  */

3263 3264
  /* Use the original server id for logging. */
  thd->variables.server_id = ev->server_id;
3265 3266 3267
  thd->set_time();                            // time the query
  thd->lex->current_select= 0;
  if (!ev->when)
3268 3269
  {
    my_hrtime_t hrtime= my_hrtime();
3270
    ev->when= hrtime_to_my_time(hrtime);
3271 3272
    ev->when_sec_part= hrtime_sec_part(hrtime);
  }
3273 3274
  thd->variables.option_bits=
    (thd->variables.option_bits & ~OPTION_SKIP_REPLICATION) |
3275
    (ev->flags & LOG_EVENT_SKIP_REPLICATION_F ? OPTION_SKIP_REPLICATION : 0);
3276 3277
  ev->thd = thd; // because up to this point, ev->thd == 0

Michael Widenius's avatar
Michael Widenius committed
3278
  int reason= ev->shall_skip(rgi);
3279
  if (reason == Log_event::EVENT_SKIP_COUNT)
3280 3281 3282 3283
  {
    DBUG_ASSERT(rli->slave_skip_counter > 0);
    rli->slave_skip_counter--;
  }
Marc Alff's avatar
Marc Alff committed
3284
  mysql_mutex_unlock(&rli->data_lock);
3285 3286 3287 3288 3289 3290
  DBUG_EXECUTE_IF("inject_slave_sql_before_apply_event",
    {
      DBUG_ASSERT(!debug_sync_set_action
                  (thd, STRING_WITH_LEN("now WAIT_FOR continue")));
      DBUG_SET_INITIAL("-d,inject_slave_sql_before_apply_event");
    };);
3291
  if (reason == Log_event::EVENT_SKIP_NOT)
3292
    exec_res= ev->apply_event(rgi);
3293

3294 3295 3296 3297 3298 3299
#ifdef WITH_WSREP
    if (exec_res && thd->wsrep_conflict_state != NO_CONFLICT)
    {
      WSREP_DEBUG("SQL apply failed, res %d conflict state: %d",
                  exec_res, thd->wsrep_conflict_state);
      rli->abort_slave= 1;
3300 3301
      rli->report(ERROR_LEVEL, ER_UNKNOWN_COM_ERROR, rgi->gtid_info(),
                  "Node has dropped from cluster");
3302 3303 3304
    }
#endif

3305
#ifndef DBUG_OFF
3306 3307
  /*
    This only prints information to the debug trace.
3308

3309 3310 3311 3312 3313 3314 3315 3316 3317 3318
    TODO: Print an informational message to the error log?
  */
  static const char *const explain[] = {
    // EVENT_SKIP_NOT,
    "not skipped",
    // EVENT_SKIP_IGNORE,
    "skipped because event should be ignored",
    // EVENT_SKIP_COUNT
    "skipped because event skip counter was non-zero"
  };
Michael Widenius's avatar
Michael Widenius committed
3319
  DBUG_PRINT("info", ("OPTION_BEGIN: %d  IN_STMT: %d  IN_TRANSACTION: %d",
3320
                      MY_TEST(thd->variables.option_bits & OPTION_BEGIN),
Michael Widenius's avatar
Michael Widenius committed
3321 3322
                      rli->get_flag(Relay_log_info::IN_STMT),
                      rli->get_flag(Relay_log_info::IN_TRANSACTION)));
3323 3324
  DBUG_PRINT("skip_event", ("%s event was %s",
                            ev->get_type_str(), explain[reason]));
3325 3326 3327 3328 3329
#endif

  DBUG_PRINT("info", ("apply_event error = %d", exec_res));
  if (exec_res == 0)
  {
3330
    int error= ev->update_pos(rgi);
3331
#ifdef HAVE_valgrind
3332 3333 3334
    if (!rli->is_fake)
#endif
    {
3335
#ifndef DBUG_OFF
3336
      char buf[22];
3337
#endif
3338 3339 3340 3341 3342 3343 3344 3345
      DBUG_PRINT("info", ("update_pos error = %d", error));
      DBUG_PRINT("info", ("group %s %s",
                          llstr(rli->group_relay_log_pos, buf),
                          rli->group_relay_log_name));
      DBUG_PRINT("info", ("event %s %s",
                          llstr(rli->event_relay_log_pos, buf),
                          rli->event_relay_log_name));
    }
3346 3347 3348 3349 3350 3351 3352 3353 3354
    /*
      The update should not fail, so print an error message and
      return an error code.

      TODO: Replace this with a decent error message when merged
      with BUG#24954 (which adds several new error message).
    */
    if (error)
    {
3355
      char buf[22];
3356
      rli->report(ERROR_LEVEL, ER_UNKNOWN_ERROR, rgi->gtid_info(),
3357 3358 3359 3360 3361 3362 3363 3364 3365
                  "It was not possible to update the positions"
                  " of the relay log information: the slave may"
                  " be in an inconsistent state."
                  " Stopped in %s position %s",
                  rli->group_relay_log_name,
                  llstr(rli->group_relay_log_pos, buf));
      DBUG_RETURN(2);
    }
  }
3366 3367 3368
  else
  {
    /*
3369
      Make sure we do not erroneously update gtid_slave_pos with a lingering
3370 3371
      GTID from this failed event group (MDEV-4906).
    */
3372
    rgi->gtid_pending= false;
3373
  }
3374 3375 3376 3377 3378

  DBUG_RETURN(exec_res ? 1 : 0);
}


Michael Widenius's avatar
Michael Widenius committed
3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452
/**
   Keep the relay log transaction state up to date.

   The state reflects how things are after the given event, that has just been
   read from the relay log, is executed.

   This is only needed to ensure we:
   - Don't abort the sql driver thread in the middle of an event group.
   - Don't rotate the io thread in the middle of a statement or transaction.
     The mechanism is that the io thread, when it needs to rotate the relay
     log, will wait until the sql driver has read all the cached events
     and then continue reading events one by one from the master until
     the sql threads signals that log doesn't have an active group anymore.

     There are two possible cases. We keep them as 2 separate flags mainly
     to make debugging easier.

     - IN_STMT is set when we have read an event that should be used
       together with the next event.  This is for example setting a
       variable that is used when executing the next statement.
     - IN_TRANSACTION is set when we are inside a BEGIN...COMMIT group

     To test the state one should use the is_in_group() function.
*/

inline void update_state_of_relay_log(Relay_log_info *rli, Log_event *ev)
{
  Log_event_type typ= ev->get_type_code();

  /* check if we are in a multi part event */
  if (ev->is_part_of_group())
    rli->set_flag(Relay_log_info::IN_STMT);
  else if (Log_event::is_group_event(typ))
  {
    /*
      If it was not a is_part_of_group() and not a group event (like
      rotate) then we can reset the IN_STMT flag.  We have the above
      if only to allow us to have a rotate element anywhere.
    */
    rli->clear_flag(Relay_log_info::IN_STMT);
  }

  /* Check for an event that starts or stops a transaction */
  if (typ == QUERY_EVENT)
  {
    Query_log_event *qev= (Query_log_event*) ev;
    /*
      Trivial optimization to avoid the following somewhat expensive
      checks.
    */
    if (qev->q_len <= sizeof("ROLLBACK"))
    {
      if (qev->is_begin())
        rli->set_flag(Relay_log_info::IN_TRANSACTION);
      if (qev->is_commit() || qev->is_rollback())
        rli->clear_flag(Relay_log_info::IN_TRANSACTION);
    }
  }
  if (typ == XID_EVENT)
    rli->clear_flag(Relay_log_info::IN_TRANSACTION);
  if (typ == GTID_EVENT &&
      !(((Gtid_log_event*) ev)->flags2 & Gtid_log_event::FL_STANDALONE))
  {
    /* This GTID_EVENT will generate a BEGIN event */
    rli->set_flag(Relay_log_info::IN_TRANSACTION);
  }

  DBUG_PRINT("info", ("event: %u  IN_STMT: %d  IN_TRANSACTION: %d",
                      (uint) typ,
                      rli->get_flag(Relay_log_info::IN_STMT),
                      rli->get_flag(Relay_log_info::IN_TRANSACTION)));
}


3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472
/**
  Top-level function for executing the next event from the relay log.

  This function reads the event from the relay log, executes it, and
  advances the relay log position.  It also handles errors, etc.

  This function may fail to apply the event for the following reasons:

   - The position specfied by the UNTIL condition of the START SLAVE
     command is reached.

   - It was not possible to read the event from the log.

   - The slave is killed.

   - An error occurred when applying the event, and the event has been
     tried slave_trans_retries times.  If the event has been retried
     fewer times, 0 is returned.

   - init_master_info or init_relay_log_pos failed. (These are called
3473
     if a failure occurs when applying the event.)
3474 3475 3476 3477 3478 3479 3480

   - An error occurred when updating the binlog position.

  @retval 0 The event was applied.

  @retval 1 The event was not applied.
*/
Michael Widenius's avatar
Michael Widenius committed
3481

3482 3483
static int exec_relay_log_event(THD* thd, Relay_log_info* rli,
                                rpl_group_info *serial_rgi)
unknown's avatar
unknown committed
3484
{
unknown's avatar
unknown committed
3485
  ulonglong event_size;
3486 3487
  DBUG_ENTER("exec_relay_log_event");

3488
  /*
Michael Widenius's avatar
Michael Widenius committed
3489 3490 3491 3492
    We acquire this mutex since we need it for all operations except
    event execution. But we will release it in places where we will
    wait for something for example inside of next_event().
  */
Marc Alff's avatar
Marc Alff committed
3493
  mysql_mutex_lock(&rli->data_lock);
3494

unknown's avatar
unknown committed
3495
  Log_event *ev= next_event(serial_rgi, &event_size);
3496

Michael Widenius's avatar
Michael Widenius committed
3497
  if (sql_slave_killed(serial_rgi))
3498
  {
Marc Alff's avatar
Marc Alff committed
3499
    mysql_mutex_unlock(&rli->data_lock);
unknown's avatar
unknown committed
3500
    delete ev;
3501
    DBUG_RETURN(1);
3502
  }
3503 3504
  if (ev)
  {
3505
    int exec_res;
3506
    Log_event_type typ= ev->get_type_code();
3507

3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522
    /*
      Even if we don't execute this event, we keep the master timestamp,
      so that seconds behind master shows correct delta (there are events
      that are not replayed, so we keep falling behind).

      If it is an artificial event, or a relay log event (IO thread generated
      event) or ev->when is set to 0, we don't update the
      last_master_timestamp.
     */
    if (!(ev->is_artificial_event() || ev->is_relay_log_event() || (ev->when == 0)))
    {
      rli->last_master_timestamp= ev->when + (time_t) ev->exec_time;
      DBUG_ASSERT(rli->last_master_timestamp >= 0);
    }

3523 3524 3525 3526
    /*
      This tests if the position of the beginning of the current event
      hits the UNTIL barrier.
    */
unknown's avatar
unknown committed
3527 3528
    if ((rli->until_condition == Relay_log_info::UNTIL_MASTER_POS ||
         rli->until_condition == Relay_log_info::UNTIL_RELAY_POS) &&
3529
        rli->is_until_satisfied(thd, ev))
3530 3531 3532 3533 3534
    {
      char buf[22];
      sql_print_information("Slave SQL thread stopped because it reached its"
                            " UNTIL position %s", llstr(rli->until_pos(), buf));
      /*
Michael Widenius's avatar
Michael Widenius committed
3535 3536
        Setting abort_slave flag because we do not want additional
        message about error in query execution to be printed.
3537 3538
      */
      rli->abort_slave= 1;
3539
      rli->stop_for_until= true;
Marc Alff's avatar
Marc Alff committed
3540
      mysql_mutex_unlock(&rli->data_lock);
3541
      delete ev;
3542
      DBUG_RETURN(1);
3543
    }
3544 3545 3546 3547 3548 3549 3550 3551

    { /**
         The following failure injecion works in cooperation with tests 
         setting @@global.debug= 'd,incomplete_group_in_relay_log'.
         Xid or Commit events are not executed to force the slave sql
         read hanging if the realy log does not have any more events.
      */
      DBUG_EXECUTE_IF("incomplete_group_in_relay_log",
3552 3553
                      if ((typ == XID_EVENT) ||
                          ((typ == QUERY_EVENT) &&
3554 3555 3556 3557
                           strcmp("COMMIT", ((Query_log_event *) ev)->query) == 0))
                      {
                        DBUG_ASSERT(thd->transaction.all.modified_non_trans_table);
                        rli->abort_slave= 1;
Marc Alff's avatar
Marc Alff committed
3558
                        mysql_mutex_unlock(&rli->data_lock);
3559
                        delete ev;
3560
                        serial_rgi->inc_event_relay_log_pos();
3561 3562 3563 3564
                        DBUG_RETURN(0);
                      };);
    }

Michael Widenius's avatar
Michael Widenius committed
3565
    update_state_of_relay_log(rli, ev);
3566

3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577
    if (opt_slave_parallel_threads > 0)
    {
      int res= rli->parallel.do_event(serial_rgi, ev, event_size);
      if (res >= 0)
        DBUG_RETURN(res);
      /*
        Else we proceed to execute the event non-parallel.
        This is the case for pre-10.0 events without GTID, and for handling
        slave_skip_counter.
      */
    }
3578

3579
    if (typ == GTID_EVENT)
3580
    {
3581 3582 3583 3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615
      Gtid_log_event *gev= static_cast<Gtid_log_event *>(ev);

      /*
        For GTID, allocate a new sub_id for the given domain_id.
        The sub_id must be allocated in increasing order of binlog order.
      */
      if (event_group_new_gtid(serial_rgi, gev))
      {
        sql_print_error("Error reading relay log event: %s", "slave SQL thread "
                        "aborted because of out-of-memory error");
        mysql_mutex_unlock(&rli->data_lock);
        delete ev;
        DBUG_RETURN(1);
      }

      if (opt_gtid_ignore_duplicates)
      {
        int res= rpl_global_gtid_slave_state.check_duplicate_gtid
          (&serial_rgi->current_gtid, serial_rgi);
        if (res < 0)
        {
          sql_print_error("Error processing GTID event: %s", "slave SQL "
                          "thread aborted because of out-of-memory error");
          mysql_mutex_unlock(&rli->data_lock);
          delete ev;
          DBUG_RETURN(1);
        }
        /*
          If we need to skip this event group (because the GTID was already
          applied), then do it using the code for slave_skip_counter, which
          is able to handle skipping until the end of the event group.
        */
        if (!res)
          rli->slave_skip_counter= 1;
      }
3616 3617
    }

3618
    serial_rgi->future_event_relay_log_pos= rli->future_event_relay_log_pos;
unknown's avatar
unknown committed
3619 3620
    serial_rgi->event_relay_log_name= rli->event_relay_log_name;
    serial_rgi->event_relay_log_pos= rli->event_relay_log_pos;
3621
    exec_res= apply_event_and_update_pos(ev, thd, serial_rgi, NULL);
3622

3623 3624 3625 3626
#ifdef WITH_WSREP
    WSREP_DEBUG("apply_event_and_update_pos() result: %d", exec_res);
#endif /* WITH_WSREP */

3627
    delete_or_keep_event_post_apply(serial_rgi, typ, ev);
3628

3629
    /*
3630 3631
      update_log_pos failed: this should not happen, so we don't
      retry.
3632
    */
3633 3634 3635
    if (exec_res == 2)
      DBUG_RETURN(1);

3636 3637 3638 3639 3640 3641
#ifdef WITH_WSREP
    mysql_mutex_lock(&thd->LOCK_wsrep_thd);
    if (thd->wsrep_conflict_state == NO_CONFLICT)
    {
      mysql_mutex_unlock(&thd->LOCK_wsrep_thd);
#endif /* WITH_WSREP */
3642 3643
    if (slave_trans_retries)
    {
3644
      int temp_err;
Michael Widenius's avatar
Michael Widenius committed
3645
      LINT_INIT(temp_err);
3646
      if (exec_res && (temp_err= has_temporary_error(thd)))
3647 3648
      {
        const char *errmsg;
3649
        rli->clear_error();
3650 3651
        /*
          We were in a transaction which has been rolled back because of a
unknown's avatar
unknown committed
3652 3653
          temporary error;
          let's seek back to BEGIN log event and retry it all again.
3654
	  Note, if lock wait timeout (innodb_lock_wait_timeout exceeded)
3655
	  there is no rollback since 5.0.13 (ref: manual).
3656
          We have to not only seek but also
3657 3658 3659 3660 3661 3662

          a) init_master_info(), to seek back to hot relay log's start
          for later (for when we will come back to this hot log after
          re-processing the possibly existing old logs where BEGIN is:
          check_binlog_magic() will then need the cache to be at
          position 0 (see comments at beginning of
3663 3664 3665
          init_master_info()).
          b) init_relay_log_pos(), because the BEGIN may be an older relay log.
        */
3666
        if (serial_rgi->trans_retries < slave_trans_retries)
3667 3668 3669 3670 3671 3672
        {
          if (init_master_info(rli->mi, 0, 0, 0, SLAVE_SQL))
            sql_print_error("Failed to initialize the master info structure");
          else if (init_relay_log_pos(rli,
                                      rli->group_relay_log_name,
                                      rli->group_relay_log_pos,
3673
                                      1, &errmsg, 1))
3674 3675 3676 3677 3678
            sql_print_error("Error initializing relay log position: %s",
                            errmsg);
          else
          {
            exec_res= 0;
unknown's avatar
unknown committed
3679
            serial_rgi->cleanup_context(thd, 1);
unknown's avatar
unknown committed
3680
            /* chance for concurrent connection to get more locks */
3681 3682 3683
            slave_sleep(thd, MY_MIN(serial_rgi->trans_retries,
                                    MAX_SLAVE_RETRY_PAUSE),
                       sql_slave_killed, serial_rgi);
3684
            serial_rgi->trans_retries++;
Marc Alff's avatar
Marc Alff committed
3685
            mysql_mutex_lock(&rli->data_lock); // because of SHOW STATUS
3686
            rli->retried_trans++;
3687
            statistic_increment(slave_retried_transactions, LOCK_status);
Marc Alff's avatar
Marc Alff committed
3688
            mysql_mutex_unlock(&rli->data_lock);
3689
            DBUG_PRINT("info", ("Slave retries transaction "
3690 3691
                                "rgi->trans_retries: %lu",
                                serial_rgi->trans_retries));
unknown's avatar
unknown committed
3692
          }
3693 3694 3695 3696 3697 3698 3699
        }
        else
          sql_print_error("Slave SQL thread retried transaction %lu time(s) "
                          "in vain, giving up. Consider raising the value of "
                          "the slave_transaction_retries variable.",
                          slave_trans_retries);
      }
3700
      else if ((exec_res && !temp_err) ||
3701 3702
               (opt_using_transactions &&
                rli->group_relay_log_pos == rli->event_relay_log_pos))
3703 3704
      {
        /*
3705 3706 3707
          Only reset the retry counter if the entire group succeeded
          or failed with a non-transient error.  On a successful
          event, the execution will proceed as usual; in the case of a
3708 3709
          non-transient error, the slave will stop with an error.
         */
3710 3711 3712
        serial_rgi->trans_retries= 0; // restart from fresh
        DBUG_PRINT("info", ("Resetting retry counter, rgi->trans_retries: %lu",
                            serial_rgi->trans_retries));
3713
      }
3714
    }
3715 3716 3717 3718 3719 3720
#ifdef WITH_WSREP
    } else {
      mysql_mutex_unlock(&thd->LOCK_wsrep_thd);
    }
#endif /* WITH_WSREP */

3721 3722
    thread_safe_increment64(&rli->executed_entries,
                            &slave_executed_entries_lock);
3723
    DBUG_RETURN(exec_res);
3724
  }
Marc Alff's avatar
Marc Alff committed
3725
  mysql_mutex_unlock(&rli->data_lock);
3726
  rli->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_READ_FAILURE, NULL,
3727
              ER(ER_SLAVE_RELAY_LOG_READ_FAILURE), "\
3728 3729 3730 3731 3732 3733 3734
Could not parse relay log event entry. The possible reasons are: the master's \
binary log is corrupted (you can check this by running 'mysqlbinlog' on the \
binary log), the slave's relay log is corrupted (you can check this by running \
'mysqlbinlog' on the relay log), a network problem, or a bug in the master's \
or slave's MySQL code. If you want to check the master's binary log or slave's \
relay log, you will be able to know their names by issuing 'SHOW SLAVE STATUS' \
on this slave.\
unknown's avatar
unknown committed
3735
");
3736
  DBUG_RETURN(1);
unknown's avatar
unknown committed
3737 3738
}

3739

Michael Widenius's avatar
Michael Widenius committed
3740
static bool check_io_slave_killed(Master_info *mi, const char *info)
3741
{
Michael Widenius's avatar
Michael Widenius committed
3742
  if (io_slave_killed(mi))
3743
  {
3744
    if (info && global_system_variables.log_warnings)
3745
      sql_print_information("%s", info);
3746 3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776
    return TRUE;
  }
  return FALSE;
}

/**
  @brief Try to reconnect slave IO thread.

  @details Terminates current connection to master, sleeps for
  @c mi->connect_retry msecs and initiates new connection with
  @c safe_reconnect(). Variable pointed by @c retry_count is increased -
  if it exceeds @c master_retry_count then connection is not re-established
  and function signals error.
  Unless @c suppres_warnings is TRUE, a warning is put in the server error log
  when reconnecting. The warning message and messages used to report errors
  are taken from @c messages array. In case @c master_retry_count is exceeded,
  no messages are added to the log.

  @param[in]     thd                 Thread context.
  @param[in]     mysql               MySQL connection.
  @param[in]     mi                  Master connection information.
  @param[in,out] retry_count         Number of attempts to reconnect.
  @param[in]     suppress_warnings   TRUE when a normal net read timeout 
                                     has caused to reconnecting.
  @param[in]     messages            Messages to print/log, see 
                                     reconnect_messages[] array.

  @retval        0                   OK.
  @retval        1                   There was an error.
*/

3777
static int try_to_reconnect(THD *thd, MYSQL *mysql, Master_info *mi,
3778 3779 3780 3781 3782 3783 3784 3785 3786 3787 3788 3789 3790
                            uint *retry_count, bool suppress_warnings,
                            const char *messages[SLAVE_RECON_MSG_MAX])
{
  mi->slave_running= MYSQL_SLAVE_RUN_NOT_CONNECT;
  thd->proc_info= messages[SLAVE_RECON_MSG_WAIT];
#ifdef SIGNAL_WITH_VIO_CLOSE  
  thd->clear_active_vio();
#endif
  end_server(mysql);
  if ((*retry_count)++)
  {
    if (*retry_count > master_retry_count)
      return 1;                             // Don't retry forever
3791
    slave_sleep(thd, mi->connect_retry, io_slave_killed, mi);
3792
  }
Michael Widenius's avatar
Michael Widenius committed
3793
  if (check_io_slave_killed(mi, messages[SLAVE_RECON_MSG_KILLED_WAITING]))
3794 3795 3796 3797 3798
    return 1;
  thd->proc_info = messages[SLAVE_RECON_MSG_AFTER];
  if (!suppress_warnings) 
  {
    char buf[256], llbuff[22];
unknown's avatar
unknown committed
3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811
    String tmp;
    if (mi->using_gtid != Master_info::USE_GTID_NO)
    {
      tmp.append(STRING_WITH_LEN("; GTID position '"));
      mi->gtid_current_pos.append_to_string(&tmp);
      if (mi->events_queued_since_last_gtid == 0)
        tmp.append(STRING_WITH_LEN("'"));
      else
      {
        tmp.append(STRING_WITH_LEN("', GTID event skip "));
        tmp.append_ulonglong((ulonglong)mi->events_queued_since_last_gtid);
      }
    }
3812
    my_snprintf(buf, sizeof(buf), messages[SLAVE_RECON_MSG_FAILED], 
unknown's avatar
unknown committed
3813 3814
                IO_RPL_LOG_NAME, llstr(mi->master_log_pos, llbuff),
                tmp.c_ptr_safe());
3815 3816 3817 3818 3819 3820
    /* 
      Raise a warining during registering on master/requesting dump.
      Log a message reading event.
    */
    if (messages[SLAVE_RECON_MSG_COMMAND][0])
    {
3821
      mi->report(WARNING_LEVEL, ER_SLAVE_MASTER_COM_FAILURE, NULL,
3822 3823 3824 3825 3826
                 ER(ER_SLAVE_MASTER_COM_FAILURE), 
                 messages[SLAVE_RECON_MSG_COMMAND], buf);
    }
    else
    {
3827
      sql_print_information("%s", buf);
3828 3829
    }
  }
Michael Widenius's avatar
Michael Widenius committed
3830
  if (safe_reconnect(thd, mysql, mi, 1) || io_slave_killed(mi))
3831 3832
  {
    if (global_system_variables.log_warnings)
3833
      sql_print_information("%s", messages[SLAVE_RECON_MSG_KILLED_AFTER]);
3834 3835 3836 3837 3838 3839
    return 1;
  }
  return 0;
}


3840 3841 3842 3843 3844
/**
  Slave IO thread entry point.

  @param arg Pointer to Master_info struct that holds information for
  the IO thread.
3845

3846 3847
  @return Always 0.
*/
3848
pthread_handler_t handle_slave_io(void *arg)
unknown's avatar
unknown committed
3849
{
unknown's avatar
unknown committed
3850 3851
  THD *thd; // needs to be first for thread_stack
  MYSQL *mysql;
3852
  Master_info *mi = (Master_info*)arg;
3853
  Relay_log_info *rli= &mi->rli;
unknown's avatar
unknown committed
3854 3855
  char llbuff[22];
  uint retry_count;
3856
  bool suppress_warnings;
3857
  int ret;
3858
  rpl_io_thread_info io_info;
3859 3860 3861
#ifndef DBUG_OFF
  uint retry_count_reg= 0, retry_count_dump= 0, retry_count_event= 0;
#endif
unknown's avatar
unknown committed
3862 3863
  // needs to call my_thread_init(), otherwise we get a coredump in DBUG_ stuff
  my_thread_init();
3864
  DBUG_ENTER("handle_slave_io");
unknown's avatar
unknown committed
3865

3866
  DBUG_ASSERT(mi->inited);
unknown's avatar
unknown committed
3867 3868 3869
  mysql= NULL ;
  retry_count= 0;

3870 3871
  thd= new THD; // note that contructor of THD uses DBUG_ !

Marc Alff's avatar
Marc Alff committed
3872
  mysql_mutex_lock(&mi->run_lock);
unknown's avatar
unknown committed
3873 3874 3875
  /* Inform waiting threads that slave has started */
  mi->slave_run_id++;

3876
#ifndef DBUG_OFF
3877
  mi->events_till_disconnect = disconnect_slave_event_count;
3878 3879
#endif

3880
  THD_CHECK_SENTRY(thd);
3881
  mi->io_thd = thd;
unknown's avatar
unknown committed
3882 3883

  pthread_detach_this_thread();
3884
  thd->thread_stack= (char*) &thd; // remember where our stack is
3885
  mi->clear_error();
3886
  if (init_slave_thread(thd, mi, SLAVE_THD_IO))
unknown's avatar
unknown committed
3887
  {
Marc Alff's avatar
Marc Alff committed
3888
    mysql_cond_broadcast(&mi->start_cond);
unknown's avatar
unknown committed
3889
    sql_print_error("Failed during slave I/O thread initialization");
3890
    goto err_during_init;
unknown's avatar
unknown committed
3891
  }
3892
  thd->system_thread_info.rpl_io_info= &io_info;
Marc Alff's avatar
Marc Alff committed
3893
  mysql_mutex_lock(&LOCK_thread_count);
unknown's avatar
unknown committed
3894
  threads.append(thd);
Marc Alff's avatar
Marc Alff committed
3895
  mysql_mutex_unlock(&LOCK_thread_count);
3896
  mi->slave_running = MYSQL_SLAVE_RUN_NOT_CONNECT;
3897
  mi->abort_slave = 0;
Marc Alff's avatar
Marc Alff committed
3898 3899
  mysql_mutex_unlock(&mi->run_lock);
  mysql_cond_broadcast(&mi->start_cond);
3900

3901
  DBUG_PRINT("master_info",("log_file_name: '%s'  position: %s",
unknown's avatar
unknown committed
3902 3903
                            mi->master_log_name,
                            llstr(mi->master_log_pos,llbuff)));
3904

He Zhenxing's avatar
He Zhenxing committed
3905 3906 3907
  /* This must be called before run any binlog_relay_io hooks */
  my_pthread_setspecific_ptr(RPL_MASTER_INFO, mi);

3908 3909 3910
  /* Load the set of seen GTIDs, if we did not already. */
  if (rpl_load_gtid_slave_state(thd))
  {
3911
    mi->report(ERROR_LEVEL, thd->get_stmt_da()->sql_errno(), NULL,
3912
                "Unable to load replication GTID slave state from mysql.%s: %s",
Sergei Golubchik's avatar
Sergei Golubchik committed
3913 3914
                rpl_gtid_slave_state_table_name.str,
                thd->get_stmt_da()->message());
3915 3916 3917 3918 3919 3920 3921
    /*
      If we are using old-style replication, we can continue, even though we
      then will not be able to record the GTIDs we receive. But if using GTID,
      we must give up.
    */
    if (mi->using_gtid != Master_info::USE_GTID_NO || opt_gtid_strict_mode)
      goto err;
3922 3923 3924
  }


He Zhenxing's avatar
He Zhenxing committed
3925 3926
  if (RUN_HOOK(binlog_relay_io, thread_start, (thd, mi)))
  {
3927
    mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, NULL,
He Zhenxing's avatar
He Zhenxing committed
3928 3929 3930 3931
               ER(ER_SLAVE_FATAL_ERROR), "Failed to run 'thread_start' hook");
    goto err;
  }

unknown's avatar
SCRUM  
unknown committed
3932
  if (!(mi->mysql = mysql = mysql_init(NULL)))
unknown's avatar
unknown committed
3933
  {
3934
    mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, NULL,
3935
               ER(ER_SLAVE_FATAL_ERROR), "error in mysql_init()");
unknown's avatar
unknown committed
3936 3937
    goto err;
  }
3938

Sergei Golubchik's avatar
Sergei Golubchik committed
3939
  THD_STAGE_INFO(thd, stage_connecting_to_master);
3940
  // we can get killed during safe_connect
3941
  if (!safe_connect(thd, mysql, mi))
3942
  {
unknown's avatar
unknown committed
3943 3944 3945 3946 3947 3948 3949 3950 3951 3952 3953 3954 3955 3956
    if (mi->using_gtid == Master_info::USE_GTID_NO)
      sql_print_information("Slave I/O thread: connected to master '%s@%s:%d',"
                            "replication started in log '%s' at position %s",
                            mi->user, mi->host, mi->port,
                            IO_RPL_LOG_NAME,
                            llstr(mi->master_log_pos,llbuff));
    else
    {
      String tmp;
      mi->gtid_current_pos.to_string(&tmp);
      sql_print_information("Slave I/O thread: connected to master '%s@%s:%d',"
                            "replication starts at GTID position '%s'",
                            mi->user, mi->host, mi->port, tmp.c_ptr_safe());
    }
3957
  }
3958
  else
unknown's avatar
unknown committed
3959
  {
3960
    sql_print_information("Slave I/O thread killed while connecting to master");
unknown's avatar
unknown committed
3961 3962
    goto err;
  }
3963

3964
connected:
3965

unknown's avatar
unknown committed
3966 3967 3968 3969 3970 3971 3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982 3983 3984
  if (mi->using_gtid != Master_info::USE_GTID_NO)
  {
    /*
      When the IO thread (re)connects to the master using GTID, it will
      connect at the start of an event group. But the IO thread may have
      previously logged part of the following event group to the relay
      log.

      When the IO and SQL thread are started together, we erase any previous
      relay logs, but this is not possible/desirable while the SQL thread is
      running. To avoid duplicating partial event groups in the relay logs in
      this case, we remember the count of events in any partially logged event
      group before the reconnect, and then here at connect we set up a counter
      to skip the already-logged part of the group.
    */
    mi->gtid_reconnect_event_skip_count= mi->events_queued_since_last_gtid;
    mi->gtid_event_seen= false;
  }

3985
#ifdef ENABLED_DEBUG_SYNC
3986 3987 3988 3989 3990
    DBUG_EXECUTE_IF("dbug.before_get_running_status_yes",
                    {
                      const char act[]=
                        "now "
                        "wait_for signal.io_thread_let_running";
3991
                      DBUG_ASSERT(debug_sync_service);
3992 3993 3994
                      DBUG_ASSERT(!debug_sync_set_action(thd, 
                                                         STRING_WITH_LEN(act)));
                    };);
3995
#endif
3996

3997 3998
  // TODO: the assignment below should be under mutex (5.0)
  mi->slave_running= MYSQL_SLAVE_RUN_CONNECT;
3999
  thd->slave_net = &mysql->net;
Sergei Golubchik's avatar
Sergei Golubchik committed
4000
  THD_STAGE_INFO(thd, stage_checking_master_version);
4001 4002 4003
  ret= get_master_version_and_clock(mysql, mi);
  if (ret == 1)
    /* Fatal error */
4004
    goto err;
4005

4006 4007
  if (ret == 2) 
  { 
Michael Widenius's avatar
Michael Widenius committed
4008
    if (check_io_slave_killed(mi, "Slave I/O thread killed"
4009 4010 4011
                              "while calling get_master_version_and_clock(...)"))
      goto err;
    suppress_warnings= FALSE;
Michael Widenius's avatar
Michael Widenius committed
4012 4013 4014 4015
    /*
      Try to reconnect because the error was caused by a transient network
      problem
    */
4016 4017 4018 4019 4020
    if (try_to_reconnect(thd, mysql, mi, &retry_count, suppress_warnings,
                             reconnect_messages[SLAVE_RECON_ACT_REG]))
      goto err;
    goto connected;
  } 
4021 4022

  if (mi->rli.relay_log.description_event_for_queue->binlog_version > 1)
4023
  {
unknown's avatar
unknown committed
4024 4025 4026
    /*
      Register ourselves with the master.
    */
Sergei Golubchik's avatar
Sergei Golubchik committed
4027
    THD_STAGE_INFO(thd, stage_registering_slave_on_master);
4028
    if (register_slave_on_master(mysql, mi, &suppress_warnings))
4029
    {
Michael Widenius's avatar
Michael Widenius committed
4030
      if (!check_io_slave_killed(mi, "Slave I/O thread killed "
4031 4032 4033 4034 4035 4036 4037 4038
                                "while registering slave on master"))
      {
        sql_print_error("Slave I/O thread couldn't register on master");
        if (try_to_reconnect(thd, mysql, mi, &retry_count, suppress_warnings,
                             reconnect_messages[SLAVE_RECON_ACT_REG]))
          goto err;
      }
      else
4039 4040
        goto err;
      goto connected;
4041
    }
4042 4043 4044 4045 4046 4047 4048 4049 4050 4051
    DBUG_EXECUTE_IF("FORCE_SLAVE_TO_RECONNECT_REG", 
      if (!retry_count_reg)
      {
        retry_count_reg++;
        sql_print_information("Forcing to reconnect slave I/O thread");
        if (try_to_reconnect(thd, mysql, mi, &retry_count, suppress_warnings,
                             reconnect_messages[SLAVE_RECON_ACT_REG]))
          goto err;
        goto connected;
      });
4052
  }
4053

4054
  DBUG_PRINT("info",("Starting reading binary log from master"));
Michael Widenius's avatar
Michael Widenius committed
4055
  while (!io_slave_killed(mi))
unknown's avatar
unknown committed
4056
  {
Sergei Golubchik's avatar
Sergei Golubchik committed
4057
    THD_STAGE_INFO(thd, stage_requesting_binlog_dump);
He Zhenxing's avatar
He Zhenxing committed
4058
    if (request_dump(thd, mysql, mi, &suppress_warnings))
unknown's avatar
unknown committed
4059 4060
    {
      sql_print_error("Failed on request_dump()");
Michael Widenius's avatar
Michael Widenius committed
4061
      if (check_io_slave_killed(mi, "Slave I/O thread killed while \
4062 4063 4064
requesting master dump") ||
          try_to_reconnect(thd, mysql, mi, &retry_count, suppress_warnings,
                           reconnect_messages[SLAVE_RECON_ACT_DUMP]))
unknown's avatar
unknown committed
4065
        goto err;
unknown's avatar
unknown committed
4066 4067
      goto connected;
    }
4068 4069 4070 4071 4072 4073 4074 4075 4076 4077
    DBUG_EXECUTE_IF("FORCE_SLAVE_TO_RECONNECT_DUMP", 
      if (!retry_count_dump)
      {
        retry_count_dump++;
        sql_print_information("Forcing to reconnect slave I/O thread");
        if (try_to_reconnect(thd, mysql, mi, &retry_count, suppress_warnings,
                             reconnect_messages[SLAVE_RECON_ACT_DUMP]))
          goto err;
        goto connected;
      });
He Zhenxing's avatar
He Zhenxing committed
4078
    const char *event_buf;
unknown's avatar
unknown committed
4079

4080
    DBUG_ASSERT(mi->last_error().number == 0);
Michael Widenius's avatar
Michael Widenius committed
4081
    while (!io_slave_killed(mi))
unknown's avatar
unknown committed
4082
    {
4083
      ulong event_len;
4084
      /*
4085
         We say "waiting" because read_event() will wait if there's nothing to
4086 4087 4088
         read. But if there's something to read, it will not wait. The
         important thing is to not confuse users by saying "reading" whereas
         we're in fact receiving nothing.
4089
      */
Sergei Golubchik's avatar
Sergei Golubchik committed
4090
      THD_STAGE_INFO(thd, stage_waiting_for_master_to_send_event);
4091
      event_len= read_event(mysql, mi, &suppress_warnings);
Michael Widenius's avatar
Michael Widenius committed
4092
      if (check_io_slave_killed(mi, "Slave I/O thread killed while \
4093
reading event"))
unknown's avatar
unknown committed
4094
        goto err;
4095 4096 4097 4098 4099 4100 4101 4102 4103 4104
      DBUG_EXECUTE_IF("FORCE_SLAVE_TO_RECONNECT_EVENT",
        if (!retry_count_event)
        {
          retry_count_event++;
          sql_print_information("Forcing to reconnect slave I/O thread");
          if (try_to_reconnect(thd, mysql, mi, &retry_count, suppress_warnings,
                               reconnect_messages[SLAVE_RECON_ACT_EVENT]))
            goto err;
          goto connected;
        });
4105

unknown's avatar
unknown committed
4106 4107
      if (event_len == packet_error)
      {
unknown's avatar
unknown committed
4108
        uint mysql_error_number= mysql_errno(mysql);
4109 4110
        switch (mysql_error_number) {
        case CR_NET_PACKET_TOO_LARGE:
unknown's avatar
unknown committed
4111
          sql_print_error("\
4112
Log entry on master is longer than slave_max_allowed_packet (%lu) on \
4113
slave. If the entry is correct, restart the server with a higher value of \
4114 4115
slave_max_allowed_packet",
                         slave_max_allowed_packet);
4116
          mi->report(ERROR_LEVEL, ER_NET_PACKET_TOO_LARGE, NULL,
4117
                     "%s", "Got a packet bigger than 'slave_max_allowed_packet' bytes");
unknown's avatar
unknown committed
4118
          goto err;
4119
        case ER_MASTER_FATAL_ERROR_READING_BINLOG:
4120
          mi->report(ERROR_LEVEL, ER_MASTER_FATAL_ERROR_READING_BINLOG, NULL,
4121 4122
                     ER(ER_MASTER_FATAL_ERROR_READING_BINLOG),
                     mysql_error_number, mysql_error(mysql));
unknown's avatar
unknown committed
4123
          goto err;
4124
        case ER_OUT_OF_RESOURCES:
4125 4126
          sql_print_error("\
Stopping slave I/O thread due to out-of-memory error from master");
4127
          mi->report(ERROR_LEVEL, ER_OUT_OF_RESOURCES, NULL,
4128
                     "%s", ER(ER_OUT_OF_RESOURCES));
unknown's avatar
unknown committed
4129 4130
          goto err;
        }
4131 4132
        if (try_to_reconnect(thd, mysql, mi, &retry_count, suppress_warnings,
                             reconnect_messages[SLAVE_RECON_ACT_EVENT]))
unknown's avatar
unknown committed
4133 4134
          goto err;
        goto connected;
unknown's avatar
unknown committed
4135
      } // if (event_len == packet_error)
4136

unknown's avatar
unknown committed
4137
      retry_count=0;                    // ok event, reset retry counter
Sergei Golubchik's avatar
Sergei Golubchik committed
4138
      THD_STAGE_INFO(thd, stage_queueing_master_event_to_the_relay_log);
He Zhenxing's avatar
He Zhenxing committed
4139 4140 4141 4142 4143
      event_buf= (const char*)mysql->net.read_pos + 1;
      if (RUN_HOOK(binlog_relay_io, after_read_event,
                   (thd, mi,(const char*)mysql->net.read_pos + 1,
                    event_len, &event_buf, &event_len)))
      {
4144
        mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, NULL,
He Zhenxing's avatar
He Zhenxing committed
4145 4146 4147 4148 4149 4150 4151 4152 4153
                   ER(ER_SLAVE_FATAL_ERROR),
                   "Failed to run 'after_read_event' hook");
        goto err;
      }

      /* XXX: 'synced' should be updated by queue_event to indicate
         whether event has been synced to disk */
      bool synced= 0;
      if (queue_event(mi, event_buf, event_len))
unknown's avatar
unknown committed
4154
      {
4155
        mi->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_WRITE_FAILURE, NULL,
4156 4157
                   ER(ER_SLAVE_RELAY_LOG_WRITE_FAILURE),
                   "could not queue event from master");
unknown's avatar
unknown committed
4158
        goto err;
unknown's avatar
unknown committed
4159
      }
He Zhenxing's avatar
He Zhenxing committed
4160 4161 4162 4163

      if (RUN_HOOK(binlog_relay_io, after_queue_event,
                   (thd, mi, event_buf, event_len, synced)))
      {
4164
        mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, NULL,
He Zhenxing's avatar
He Zhenxing committed
4165 4166 4167 4168 4169
                   ER(ER_SLAVE_FATAL_ERROR),
                   "Failed to run 'after_queue_event' hook");
        goto err;
      }

4170
      if (mi->using_gtid == Master_info::USE_GTID_NO &&
unknown's avatar
unknown committed
4171
          flush_master_info(mi, TRUE, TRUE))
4172 4173 4174 4175
      {
        sql_print_error("Failed to flush master info file");
        goto err;
      }
4176 4177 4178 4179 4180 4181 4182
      /*
        See if the relay logs take too much space.
        We don't lock mi->rli.log_space_lock here; this dirty read saves time
        and does not introduce any problem:
        - if mi->rli.ignore_log_space_limit is 1 but becomes 0 just after (so
        the clean value is 0), then we are reading only one more event as we
        should, and we'll block only at the next event. No big deal.
Michael Widenius's avatar
Michael Widenius committed
4183 4184 4185 4186 4187
        - if mi->rli.ignore_log_space_limit is 0 but becomes 1 just
        after (so the clean value is 1), then we are going into
        wait_for_relay_log_space() for no reason, but this function
        will do a clean read, notice the clean value and exit
        immediately.
4188
      */
4189 4190 4191 4192 4193
#ifndef DBUG_OFF
      {
        char llbuf1[22], llbuf2[22];
        DBUG_PRINT("info", ("log_space_limit=%s log_space_total=%s \
ignore_log_space_limit=%d",
4194 4195
                            llstr(rli->log_space_limit,llbuf1),
                            llstr(rli->log_space_total,llbuf2),
unknown's avatar
unknown committed
4196
                            (int) rli->ignore_log_space_limit));
4197 4198 4199
      }
#endif

4200
      if (rli->log_space_limit && rli->log_space_limit <
unknown's avatar
unknown committed
4201
          rli->log_space_total &&
4202
          !rli->ignore_log_space_limit)
unknown's avatar
unknown committed
4203 4204 4205
        if (wait_for_relay_log_space(rli))
        {
          sql_print_error("Slave I/O thread aborted while waiting for relay \
unknown's avatar
unknown committed
4206
log space");
unknown's avatar
unknown committed
4207 4208 4209
          goto err;
        }
    }
4210
  }
unknown's avatar
unknown committed
4211

unknown's avatar
unknown committed
4212
  // error = 0;
unknown's avatar
unknown committed
4213
err:
4214
  // print the current replication position
unknown's avatar
unknown committed
4215 4216 4217 4218 4219 4220 4221 4222 4223 4224 4225 4226 4227
  if (mi->using_gtid == Master_info::USE_GTID_NO)
    sql_print_information("Slave I/O thread exiting, read up to log '%s', "
                          "position %s",
                          IO_RPL_LOG_NAME, llstr(mi->master_log_pos,llbuff));
  else
  {
    String tmp;
    mi->gtid_current_pos.to_string(&tmp);
    sql_print_information("Slave I/O thread exiting, read up to log '%s', "
                          "position %s; GTID position %s",
                          IO_RPL_LOG_NAME, llstr(mi->master_log_pos,llbuff),
                          tmp.c_ptr_safe());
  }
He Zhenxing's avatar
He Zhenxing committed
4228
  RUN_HOOK(binlog_relay_io, thread_stop, (thd, mi));
4229
  thd->reset_query();
unknown's avatar
unknown committed
4230
  thd->reset_db(NULL, 0);
unknown's avatar
unknown committed
4231 4232
  if (mysql)
  {
4233 4234 4235 4236 4237 4238 4239 4240 4241 4242 4243
    /*
      Here we need to clear the active VIO before closing the
      connection with the master.  The reason is that THD::awake()
      might be called from terminate_slave_thread() because somebody
      issued a STOP SLAVE.  If that happends, the close_active_vio()
      can be called in the middle of closing the VIO associated with
      the 'mysql' object, causing a crash.
    */
#ifdef SIGNAL_WITH_VIO_CLOSE
    thd->clear_active_vio();
#endif
unknown's avatar
SCRUM  
unknown committed
4244
    mysql_close(mysql);
unknown's avatar
unknown committed
4245 4246
    mi->mysql=0;
  }
4247
  write_ignored_events_info_to_relay_log(thd, mi);
unknown's avatar
unknown committed
4248 4249
  if (mi->using_gtid != Master_info::USE_GTID_NO)
    flush_master_info(mi, TRUE, TRUE);
Sergei Golubchik's avatar
Sergei Golubchik committed
4250
  THD_STAGE_INFO(thd, stage_waiting_for_slave_mutex_on_exit);
4251
  thd->add_status_to_global();
Marc Alff's avatar
Marc Alff committed
4252
  mysql_mutex_lock(&mi->run_lock);
4253

4254
err_during_init:
4255 4256 4257
  /* Forget the relay log's format */
  delete mi->rli.relay_log.description_event_for_queue;
  mi->rli.relay_log.description_event_for_queue= 0;
4258
  // TODO: make rpl_status part of Master_info
4259
  change_rpl_status(RPL_ACTIVE_SLAVE,RPL_IDLE_SLAVE);
Marc Alff's avatar
Marc Alff committed
4260
  mysql_mutex_lock(&LOCK_thread_count);
4261 4262
  thd->unlink();
  mysql_mutex_unlock(&LOCK_thread_count);
4263
  THD_CHECK_SENTRY(thd);
unknown's avatar
unknown committed
4264
  delete thd;
unknown's avatar
unknown committed
4265
  mi->abort_slave= 0;
4266
  mi->slave_running= MYSQL_SLAVE_NOT_RUN;
unknown's avatar
unknown committed
4267
  mi->io_thd= 0;
4268 4269 4270 4271 4272
  /*
    Note: the order of the two following calls (first broadcast, then unlock)
    is important. Otherwise a killer_thread can execute between the calls and
    delete the mi structure leading to a crash! (see BUG#25306 for details)
   */ 
Marc Alff's avatar
Marc Alff committed
4273
  mysql_cond_broadcast(&mi->stop_cond);       // tell the world we are done
4274
  DBUG_EXECUTE_IF("simulate_slave_delay_at_terminate_bug38694", sleep(5););
Marc Alff's avatar
Marc Alff committed
4275
  mysql_mutex_unlock(&mi->run_lock);
4276 4277

  DBUG_LEAVE;                                   // Must match DBUG_ENTER()
4278
  my_thread_end();
4279
#ifdef HAVE_OPENSSL
4280
  ERR_remove_state(0);
4281
#endif
unknown's avatar
unknown committed
4282
  pthread_exit(0);
4283
  return 0;                                     // Avoid compiler warnings
unknown's avatar
unknown committed
4284 4285
}

4286 4287 4288
/*
  Check the temporary directory used by commands like
  LOAD DATA INFILE.
4289 4290 4291 4292

  As the directory never changes during a mysqld run, we only
  test this once and cache the result. This also resolve a race condition
  when this can be run by multiple threads at the same time.
4293
 */
4294 4295 4296 4297

static bool check_temp_dir_run= 0;
static int check_temp_dir_result= 0;

4298
static 
4299
int check_temp_dir(char* tmp_file)
4300
{
4301 4302
  File fd;
  int result= 1;                                // Assume failure
4303
  MY_DIR *dirp;
4304 4305
  char tmp_dir[FN_REFLEN];
  size_t tmp_dir_size;
4306 4307
  DBUG_ENTER("check_temp_dir");

4308 4309 4310 4311 4312 4313 4314 4315
  mysql_mutex_lock(&LOCK_thread_count);
  if (check_temp_dir_run)
  {
    result= check_temp_dir_result;
    goto end;
  }
  check_temp_dir_run= 1;

4316 4317 4318 4319 4320
  /*
    Get the directory from the temporary file.
  */
  dirname_part(tmp_dir, tmp_file, &tmp_dir_size);

4321 4322 4323 4324
  /*
    Check if the directory exists.
   */
  if (!(dirp=my_dir(tmp_dir,MYF(MY_WME))))
4325
    goto end;
4326 4327 4328
  my_dirend(dirp);

  /*
4329 4330
    Check permissions to create a file. We use O_TRUNC to ensure that
    things works even if we happen to have and old file laying around.
4331
   */
Marc Alff's avatar
Marc Alff committed
4332 4333
  if ((fd= mysql_file_create(key_file_misc,
                             tmp_file, CREATE_MODE,
4334
                             O_WRONLY | O_BINARY | O_TRUNC | O_NOFOLLOW,
Marc Alff's avatar
Marc Alff committed
4335
                             MYF(MY_WME))) < 0)
4336
    goto end;
4337

4338
  result= 0;                                    // Directory name ok
4339 4340 4341
  /*
    Clean up.
   */
Marc Alff's avatar
Marc Alff committed
4342 4343
  mysql_file_close(fd, MYF(0));
  mysql_file_delete(key_file_misc, tmp_file, MYF(0));
4344

4345 4346 4347 4348
end:
  check_temp_dir_result= result;
  mysql_mutex_unlock(&LOCK_thread_count);
  DBUG_RETURN(result);
4349
}
unknown's avatar
unknown committed
4350

4351

unknown's avatar
unknown committed
4352
void
4353
slave_output_error_info(rpl_group_info *rgi, THD *thd)
unknown's avatar
unknown committed
4354 4355 4356 4357 4358 4359
{
  /*
    retrieve as much info as possible from the thd and, error
    codes and warnings and print this to the error log as to
    allow the user to locate the error
  */
4360
  Relay_log_info *rli= rgi->rli;
unknown's avatar
unknown committed
4361 4362 4363 4364 4365
  uint32 const last_errno= rli->last_error().number;
  char llbuff[22];

  if (thd->is_error())
  {
4366
    char const *const errmsg= thd->get_stmt_da()->message();
unknown's avatar
unknown committed
4367 4368

    DBUG_PRINT("info",
4369 4370
               ("thd->get_stmt_da()->sql_errno()=%d; rli->last_error.number=%d",
                thd->get_stmt_da()->sql_errno(), last_errno));
unknown's avatar
unknown committed
4371 4372 4373 4374 4375 4376
    if (last_errno == 0)
    {
      /*
        This function is reporting an error which was not reported
        while executing exec_relay_log_event().
      */ 
4377 4378
      rli->report(ERROR_LEVEL, thd->get_stmt_da()->sql_errno(),
                  rgi->gtid_info(), "%s", errmsg);
unknown's avatar
unknown committed
4379
    }
4380
    else if (last_errno != thd->get_stmt_da()->sql_errno())
unknown's avatar
unknown committed
4381 4382 4383 4384 4385 4386 4387 4388
    {
      /*
       * An error was reported while executing exec_relay_log_event()
       * however the error code differs from what is in the thread.
       * This function prints out more information to help finding
       * what caused the problem.
       */  
      sql_print_error("Slave (additional info): %s Error_code: %d",
4389
                      errmsg, thd->get_stmt_da()->sql_errno());
unknown's avatar
unknown committed
4390 4391 4392 4393
    }
  }

  /* Print any warnings issued */
4394 4395 4396
  Diagnostics_area::Sql_condition_iterator it=
    thd->get_stmt_da()->sql_conditions();
  const Sql_condition *err;
unknown's avatar
unknown committed
4397 4398 4399 4400 4401 4402 4403 4404 4405 4406 4407 4408 4409 4410 4411 4412 4413 4414 4415 4416 4417 4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431 4432 4433 4434 4435 4436 4437 4438 4439 4440
  /*
    Added controlled slave thread cancel for replication
    of user-defined variables.
  */
  bool udf_error = false;
  while ((err= it++))
  {
    if (err->get_sql_errno() == ER_CANT_OPEN_LIBRARY)
      udf_error = true;
    sql_print_warning("Slave: %s Error_code: %d", err->get_message_text(), err->get_sql_errno());
  }
  if (udf_error)
  {
    String tmp;
    if (rli->mi->using_gtid != Master_info::USE_GTID_NO)
    {
      tmp.append(STRING_WITH_LEN("; GTID position '"));
      rpl_append_gtid_state(&tmp, false);
      tmp.append(STRING_WITH_LEN("'"));
    }
    sql_print_error("Error loading user-defined library, slave SQL "
      "thread aborted. Install the missing library, and restart the "
      "slave SQL thread with \"SLAVE START\". We stopped at log '%s' "
      "position %s%s", RPL_LOG_NAME, llstr(rli->group_master_log_pos,
      llbuff), tmp.c_ptr_safe());
  }
  else
  {
    String tmp;
    if (rli->mi->using_gtid != Master_info::USE_GTID_NO)
    {
      tmp.append(STRING_WITH_LEN("; GTID position '"));
      rpl_append_gtid_state(&tmp, false);
      tmp.append(STRING_WITH_LEN("'"));
    }
    sql_print_error("\
Error running query, slave SQL thread aborted. Fix the problem, and restart \
the slave SQL thread with \"SLAVE START\". We stopped at log \
'%s' position %s%s", RPL_LOG_NAME, llstr(rli->group_master_log_pos, llbuff),
                    tmp.c_ptr_safe());
  }
}


4441 4442 4443 4444 4445
/**
  Slave SQL thread entry point.

  @param arg Pointer to Relay_log_info object that holds information
  for the SQL thread.
unknown's avatar
unknown committed
4446

4447 4448
  @return Always 0.
*/
4449
pthread_handler_t handle_slave_sql(void *arg)
4450
{
unknown's avatar
unknown committed
4451
  THD *thd;                     /* needs to be first for thread_stack */
4452
  char llbuff[22],llbuff1[22];
4453 4454
  char saved_log_name[FN_REFLEN];
  char saved_master_log_name[FN_REFLEN];
4455 4456
  my_off_t UNINIT_VAR(saved_log_pos);
  my_off_t UNINIT_VAR(saved_master_log_pos);
4457
  String saved_skip_gtid_pos;
4458
  my_off_t saved_skip= 0;
4459 4460
  Master_info *mi= ((Master_info*)arg);
  Relay_log_info* rli = &mi->rli;
4461 4462 4463
#ifdef WITH_WSREP
  my_bool wsrep_node_dropped= FALSE;
#endif /* WITH_WSREP */
unknown's avatar
unknown committed
4464
  const char *errmsg;
unknown's avatar
unknown committed
4465
  rpl_group_info *serial_rgi;
4466
  rpl_sql_thread_info sql_info(mi->rpl_filter);
unknown's avatar
unknown committed
4467 4468 4469

  // needs to call my_thread_init(), otherwise we get a coredump in DBUG_ stuff
  my_thread_init();
4470
  DBUG_ENTER("handle_slave_sql");
4471 4472 4473
#ifdef WITH_WSREP
 wsrep_restart_point:
#endif /* WITH_WSREP */
unknown's avatar
unknown committed
4474

Michael Widenius's avatar
Michael Widenius committed
4475 4476
  LINT_INIT(saved_master_log_pos);
  LINT_INIT(saved_log_pos);
4477

unknown's avatar
unknown committed
4478
  serial_rgi= new rpl_group_info(rli);
4479 4480
  thd = new THD; // note that contructor of THD uses DBUG_ !
  thd->thread_stack = (char*)&thd; // remember where our stack is
4481
  thd->system_thread_info.rpl_sql_info= &sql_info;
4482

4483
  DBUG_ASSERT(rli->inited);
4484
  DBUG_ASSERT(rli->mi == mi);
Marc Alff's avatar
Marc Alff committed
4485
  mysql_mutex_lock(&rli->run_lock);
4486
  DBUG_ASSERT(!rli->slave_running);
unknown's avatar
unknown committed
4487
  errmsg= 0;
unknown's avatar
unknown committed
4488
#ifndef DBUG_OFF
4489
  rli->events_till_abort = abort_slave_event_count;
unknown's avatar
unknown committed
4490
#endif
4491

Michael Widenius's avatar
Michael Widenius committed
4492 4493 4494 4495 4496 4497 4498 4499 4500
  /*
    THD for the sql driver thd. In parallel replication this is the thread
    that reads things from the relay log and calls rpl_parallel::do_event()
    to execute queries.

    In single thread replication this is the THD for the thread that is
    executing SQL queries too.
  */
  serial_rgi->thd= rli->sql_driver_thd= thd;
4501
  
unknown's avatar
unknown committed
4502 4503
  /* Inform waiting threads that slave has started */
  rli->slave_run_id++;
4504
  rli->slave_running= MYSQL_SLAVE_RUN_NOT_CONNECT;
unknown's avatar
unknown committed
4505

4506
  pthread_detach_this_thread();
4507
  if (init_slave_thread(thd, mi, SLAVE_THD_SQL))
unknown's avatar
unknown committed
4508 4509 4510 4511 4512
  {
    /*
      TODO: this is currently broken - slave start and change master
      will be stuck if we fail here
    */
Marc Alff's avatar
Marc Alff committed
4513
    mysql_cond_broadcast(&rli->start_cond);
4514
    rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, NULL,
4515
                "Failed during slave thread initialization");
4516
    goto err_during_init;
unknown's avatar
unknown committed
4517
  }
4518
  thd->init_for_queries();
unknown's avatar
unknown committed
4519 4520
  thd->rgi_slave= serial_rgi;
  if ((serial_rgi->deferred_events_collecting= mi->rpl_filter->is_on()))
4521
  {
unknown's avatar
unknown committed
4522
    serial_rgi->deferred_events= new Deferred_log_events(rli);
4523 4524
  }

4525
  /*
4526
    binlog_annotate_row_events must be TRUE only after an Annotate_rows event
4527
    has been received and only till the last corresponding rbr event has been
4528 4529
    applied. In all other cases it must be FALSE.
  */
4530
  thd->variables.binlog_annotate_row_events= 0;
Marc Alff's avatar
Marc Alff committed
4531
  mysql_mutex_lock(&LOCK_thread_count);
4532
  threads.append(thd);
Marc Alff's avatar
Marc Alff committed
4533
  mysql_mutex_unlock(&LOCK_thread_count);
4534 4535 4536 4537 4538 4539 4540 4541
  /*
    We are going to set slave_running to 1. Assuming slave I/O thread is
    alive and connected, this is going to make Seconds_Behind_Master be 0
    i.e. "caught up". Even if we're just at start of thread. Well it's ok, at
    the moment we start we can think we are caught up, and the next second we
    start receiving data so we realize we are not caught up and
    Seconds_Behind_Master grows. No big deal.
  */
4542
  rli->abort_slave = 0;
4543
  rli->stop_for_until= false;
Marc Alff's avatar
Marc Alff committed
4544 4545
  mysql_mutex_unlock(&rli->run_lock);
  mysql_cond_broadcast(&rli->start_cond);
4546

unknown's avatar
unknown committed
4547 4548 4549
  /*
    Reset errors for a clean start (otherwise, if the master is idle, the SQL
    thread may execute no Query_log_event, so the error will remain even
unknown's avatar
unknown committed
4550
    though there's no problem anymore). Do not reset the master timestamp
4551 4552 4553 4554
    (imagine the slave has caught everything, the STOP SLAVE and START SLAVE:
    as we are not sure that we are going to receive a query, we want to
    remember the last master timestamp (to say how many seconds behind we are
    now.
unknown's avatar
unknown committed
4555
    But the master timestamp is reset by RESET SLAVE & CHANGE MASTER.
unknown's avatar
unknown committed
4556
  */
4557
  rli->clear_error();
unknown's avatar
unknown committed
4558
  rli->parallel.reset();
4559 4560 4561 4562

  //tell the I/O thread to take relay_log_space_limit into account from now on
  rli->ignore_log_space_limit= 0;

4563
  serial_rgi->gtid_sub_id= 0;
4564
  serial_rgi->gtid_pending= false;
4565 4566 4567 4568 4569 4570 4571 4572 4573 4574
  if (mi->using_gtid != Master_info::USE_GTID_NO)
  {
    /*
      We initialize the relay log state from the know starting position.
      It will then be updated as required by GTID and GTID_LIST events found
      while applying events read from relay logs.
    */
    rli->relay_log_state.load(&rpl_global_gtid_slave_state);
  }
  rli->gtid_skip_flag = GTID_SKIP_NOT;
4575
  if (init_relay_log_pos(rli,
unknown's avatar
unknown committed
4576 4577 4578
                         rli->group_relay_log_name,
                         rli->group_relay_log_pos,
                         1 /*need data lock*/, &errmsg,
4579
                         1 /*look for a description_event*/))
4580
  { 
4581
    rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, NULL,
4582
                "Error initializing relay log position: %s", errmsg);
4583 4584
    goto err;
  }
4585
  rli->reset_inuse_relaylog();
4586 4587 4588
  if (rli->alloc_inuse_relaylog(rli->group_relay_log_name))
    goto err;

unknown's avatar
unknown committed
4589
  strcpy(rli->future_event_master_log_name, rli->group_master_log_name);
4590
  THD_CHECK_SENTRY(thd);
4591 4592 4593 4594
#ifndef DBUG_OFF
  {
    char llbuf1[22], llbuf2[22];
    DBUG_PRINT("info", ("my_b_tell(rli->cur_log)=%s rli->event_relay_log_pos=%s",
unknown's avatar
unknown committed
4595
                        llstr(my_b_tell(rli->cur_log),llbuf1),
4596 4597 4598 4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613 4614
                        llstr(rli->event_relay_log_pos,llbuf2)));
    DBUG_ASSERT(rli->event_relay_log_pos >= BIN_LOG_HEADER_SIZE);
    /*
      Wonder if this is correct. I (Guilhem) wonder if my_b_tell() returns the
      correct position when it's called just after my_b_seek() (the questionable
      stuff is those "seek is done on next read" comments in the my_b_seek()
      source code).
      The crude reality is that this assertion randomly fails whereas
      replication seems to work fine. And there is no easy explanation why it
      fails (as we my_b_seek(rli->event_relay_log_pos) at the very end of
      init_relay_log_pos() called above). Maybe the assertion would be
      meaningful if we held rli->data_lock between the my_b_seek() and the
      DBUG_ASSERT().
    */
#ifdef SHOULD_BE_CHECKED
    DBUG_ASSERT(my_b_tell(rli->cur_log) == rli->event_relay_log_pos);
#endif
  }
#endif
4615

4616 4617 4618 4619 4620
#ifdef WITH_WSREP
  thd->wsrep_exec_mode= LOCAL_STATE;
  /* synchronize with wsrep replication */
  if (WSREP_ON) wsrep_ready_wait();
#endif
4621
  DBUG_PRINT("master_info",("log_file_name: %s  position: %s",
unknown's avatar
unknown committed
4622 4623
                            rli->group_master_log_name,
                            llstr(rli->group_master_log_pos,llbuff)));
4624
  if (global_system_variables.log_warnings)
unknown's avatar
unknown committed
4625 4626 4627 4628 4629 4630 4631 4632 4633
  {
    String tmp;
    if (mi->using_gtid != Master_info::USE_GTID_NO)
    {
      tmp.append(STRING_WITH_LEN("; GTID position '"));
      rpl_append_gtid_state(&tmp,
                            mi->using_gtid==Master_info::USE_GTID_CURRENT_POS);
      tmp.append(STRING_WITH_LEN("'"));
    }
4634
    sql_print_information("Slave SQL thread initialized, starting replication in \
unknown's avatar
unknown committed
4635
log '%s' at position %s, relay log '%s' position: %s%s", RPL_LOG_NAME,
unknown's avatar
unknown committed
4636
                    llstr(rli->group_master_log_pos,llbuff),rli->group_relay_log_name,
unknown's avatar
unknown committed
4637 4638
                    llstr(rli->group_relay_log_pos,llbuff1), tmp.c_ptr_safe());
  }
4639

4640
  if (check_temp_dir(rli->slave_patternload_file))
4641
  {
4642
    rli->report(ERROR_LEVEL, thd->get_stmt_da()->sql_errno(), NULL,
4643
                "Unable to use slave's temporary directory %s - %s", 
4644
                slave_load_tmpdir, thd->get_stmt_da()->message());
4645 4646 4647
    goto err;
  }

4648 4649 4650
  /* Load the set of seen GTIDs, if we did not already. */
  if (rpl_load_gtid_slave_state(thd))
  {
4651
    rli->report(ERROR_LEVEL, thd->get_stmt_da()->sql_errno(), NULL,
4652
                "Unable to load replication GTID slave state from mysql.%s: %s",
Sergei Golubchik's avatar
Sergei Golubchik committed
4653 4654
                rpl_gtid_slave_state_table_name.str,
                thd->get_stmt_da()->message());
4655 4656 4657 4658 4659 4660 4661
    /*
      If we are using old-style replication, we can continue, even though we
      then will not be able to record the GTIDs we receive. But if using GTID,
      we must give up.
    */
    if (mi->using_gtid != Master_info::USE_GTID_NO || opt_gtid_strict_mode)
      goto err;
4662 4663
  }

unknown's avatar
unknown committed
4664
  /* execute init_slave variable */
4665
  if (opt_init_slave.length)
unknown's avatar
unknown committed
4666
  {
4667
    execute_init_command(thd, &opt_init_slave, &LOCK_sys_init_slave);
4668
    if (thd->is_slave_error)
unknown's avatar
unknown committed
4669
    {
4670
      rli->report(ERROR_LEVEL, thd->get_stmt_da()->sql_errno(), NULL,
4671
                  "Slave SQL thread aborted. Can't execute init_slave query");
unknown's avatar
unknown committed
4672 4673 4674 4675
      goto err;
    }
  }

4676 4677 4678 4679
  /*
    First check until condition - probably there is nothing to execute. We
    do not want to wait for next event in this case.
  */
Marc Alff's avatar
Marc Alff committed
4680
  mysql_mutex_lock(&rli->data_lock);
4681 4682
  if (rli->slave_skip_counter)
  {
4683 4684
    strmake_buf(saved_log_name, rli->group_relay_log_name);
    strmake_buf(saved_master_log_name, rli->group_master_log_name);
4685 4686
    saved_log_pos= rli->group_relay_log_pos;
    saved_master_log_pos= rli->group_master_log_pos;
4687 4688 4689 4690 4691 4692
    if (mi->using_gtid != Master_info::USE_GTID_NO)
    {
      saved_skip_gtid_pos.append(STRING_WITH_LEN(", GTID '"));
      rpl_append_gtid_state(&saved_skip_gtid_pos, false);
      saved_skip_gtid_pos.append(STRING_WITH_LEN("'; "));
    }
4693 4694
    saved_skip= rli->slave_skip_counter;
  }
unknown's avatar
unknown committed
4695 4696
  if ((rli->until_condition == Relay_log_info::UNTIL_MASTER_POS ||
       rli->until_condition == Relay_log_info::UNTIL_RELAY_POS) &&
4697
      rli->is_until_satisfied(thd, NULL))
4698 4699 4700 4701
  {
    char buf[22];
    sql_print_information("Slave SQL thread stopped because it reached its"
                          " UNTIL position %s", llstr(rli->until_pos(), buf));
Marc Alff's avatar
Marc Alff committed
4702
    mysql_mutex_unlock(&rli->data_lock);
4703 4704
    goto err;
  }
Marc Alff's avatar
Marc Alff committed
4705
  mysql_mutex_unlock(&rli->data_lock);
4706

4707 4708
  /* Read queries from the IO/THREAD until this thread is killed */

Michael Widenius's avatar
Michael Widenius committed
4709
  while (!sql_slave_killed(serial_rgi))
4710
  {
Sergei Golubchik's avatar
Sergei Golubchik committed
4711
    THD_STAGE_INFO(thd, stage_reading_event_from_the_relay_log);
4712
    THD_CHECK_SENTRY(thd);
4713 4714 4715

    if (saved_skip && rli->slave_skip_counter == 0)
    {
4716 4717 4718 4719 4720 4721 4722 4723
      String tmp;
      if (mi->using_gtid != Master_info::USE_GTID_NO)
      {
        tmp.append(STRING_WITH_LEN(", GTID '"));
        rpl_append_gtid_state(&tmp, false);
        tmp.append(STRING_WITH_LEN("'; "));
      }

4724 4725
      sql_print_information("'SQL_SLAVE_SKIP_COUNTER=%ld' executed at "
        "relay_log_file='%s', relay_log_pos='%ld', master_log_name='%s', "
4726
        "master_log_pos='%ld'%s and new position at "
4727
        "relay_log_file='%s', relay_log_pos='%ld', master_log_name='%s', "
4728
        "master_log_pos='%ld'%s ",
4729 4730
        (ulong) saved_skip, saved_log_name, (ulong) saved_log_pos,
        saved_master_log_name, (ulong) saved_master_log_pos,
4731
        saved_skip_gtid_pos.c_ptr_safe(),
4732
        rli->group_relay_log_name, (ulong) rli->group_relay_log_pos,
4733 4734
        rli->group_master_log_name, (ulong) rli->group_master_log_pos,
        tmp.c_ptr_safe());
4735
      saved_skip= 0;
4736
      saved_skip_gtid_pos.free();
4737 4738
    }
    
unknown's avatar
unknown committed
4739
    if (exec_relay_log_event(thd, rli, serial_rgi))
4740
    {
4741 4742 4743 4744 4745 4746 4747 4748
#ifdef WITH_WSREP
      if (thd->wsrep_conflict_state != NO_CONFLICT)
      {
        wsrep_node_dropped= TRUE;
        rli->abort_slave= TRUE;
      }
#endif /* WITH_WSREP */

4749
      DBUG_PRINT("info", ("exec_relay_log_event() failed"));
4750
      // do not scare the user if SQL thread was simply killed or stopped
Michael Widenius's avatar
Michael Widenius committed
4751
      if (!sql_slave_killed(serial_rgi))
4752
      {
4753
        slave_output_error_info(serial_rgi, thd);
4754
#ifdef WITH_WSREP
Nirbhay Choubey's avatar
Nirbhay Choubey committed
4755
        uint32 const last_errno= rli->last_error().number;
4756 4757
        if (WSREP_ON && last_errno == ER_UNKNOWN_COM_ERROR)
        {
4758 4759
          wsrep_node_dropped= TRUE;
        }
4760 4761
#endif /* WITH_WSREP */
      }
4762 4763
      goto err;
    }
4764
  }
4765

4766
  if (opt_slave_parallel_threads > 0)
4767
    rli->parallel.wait_for_done(thd, rli);
4768

4769
  /* Thread stopped. Print the current replication position to the log */
unknown's avatar
unknown committed
4770 4771 4772 4773 4774 4775 4776 4777 4778 4779 4780 4781 4782 4783
  {
    String tmp;
    if (mi->using_gtid != Master_info::USE_GTID_NO)
    {
      tmp.append(STRING_WITH_LEN("; GTID position '"));
      rpl_append_gtid_state(&tmp, false);
      tmp.append(STRING_WITH_LEN("'"));
    }
    sql_print_information("Slave SQL thread exiting, replication stopped in "
                          "log '%s' at position %s%s",
                          RPL_LOG_NAME,
                          llstr(rli->group_master_log_pos,llbuff),
                          tmp.c_ptr_safe());
  }
4784 4785

 err:
4786

4787 4788 4789 4790 4791 4792
  /*
    Once again, in case we aborted with an error and skipped the first one.
    (We want the first one to be before the printout of stop position to
    get the correct position printed.)
  */
  if (opt_slave_parallel_threads > 0)
4793
    rli->parallel.wait_for_done(thd, rli);
4794

4795 4796 4797 4798 4799 4800
  /*
    Some events set some playgrounds, which won't be cleared because thread
    stops. Stopping of this thread may not be known to these events ("stop"
    request is detected only by the present function, not by events), so we
    must "proactively" clear playgrounds:
  */
4801
  thd->clear_error();
unknown's avatar
unknown committed
4802
  serial_rgi->cleanup_context(thd, 1);
4803 4804 4805 4806 4807
  /*
    Some extra safety, which should not been needed (normally, event deletion
    should already have done these assignments (each event which sets these
    variables is supposed to set them to 0 before terminating)).
  */
4808
  thd->catalog= 0;
4809
  thd->reset_query();
unknown's avatar
unknown committed
4810
  thd->reset_db(NULL, 0);
unknown's avatar
unknown committed
4811
  if (rli->mi->using_gtid != Master_info::USE_GTID_NO)
4812 4813 4814
  {
    ulong domain_count;

unknown's avatar
unknown committed
4815
    flush_relay_log_info(rli);
4816 4817 4818 4819 4820 4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831 4832 4833 4834 4835 4836 4837 4838 4839 4840 4841 4842 4843 4844 4845 4846 4847 4848 4849 4850 4851 4852 4853 4854
    if (opt_slave_parallel_threads > 0)
    {
      /*
        In parallel replication GTID mode, we may stop with different domains
        at different positions in the relay log.

        To handle this when we restart the SQL thread, mark the current
        per-domain position in the Relay_log_info.
      */
      mysql_mutex_lock(&rpl_global_gtid_slave_state.LOCK_slave_state);
      domain_count= rpl_global_gtid_slave_state.count();
      mysql_mutex_unlock(&rpl_global_gtid_slave_state.LOCK_slave_state);
      if (domain_count > 1)
      {
        inuse_relaylog *ir;

        /*
          Load the starting GTID position, so that we can skip already applied
          GTIDs when we restart the SQL thread. And set the start position in
          the relay log back to a known safe place to start (prior to any not
          yet applied transaction in any domain).
        */
        rli->restart_gtid_pos.load(&rpl_global_gtid_slave_state, NULL, 0);
        if ((ir= rli->inuse_relaylog_list))
        {
          rpl_gtid *gtid= ir->relay_log_state;
          uint32 count= ir->relay_log_state_count;
          while (count > 0)
          {
            process_gtid_for_restart_pos(rli, gtid);
            ++gtid;
            --count;
          }
          strmake_buf(rli->group_relay_log_name, ir->name);
          rli->group_relay_log_pos= BIN_LOG_HEADER_SIZE;
        }
      }
    }
  }
Sergei Golubchik's avatar
Sergei Golubchik committed
4855
  THD_STAGE_INFO(thd, stage_waiting_for_slave_mutex_on_exit);
4856
  thd->add_status_to_global();
Marc Alff's avatar
Marc Alff committed
4857
  mysql_mutex_lock(&rli->run_lock);
4858
err_during_init:
4859
  /* We need data_lock, at least to wake up any waiting master_pos_wait() */
Marc Alff's avatar
Marc Alff committed
4860
  mysql_mutex_lock(&rli->data_lock);
4861
  DBUG_ASSERT(rli->slave_running == MYSQL_SLAVE_RUN_NOT_CONNECT); // tracking buffer overrun
4862
  /* When master_pos_wait() wakes up it will check this and terminate */
4863
  rli->slave_running= MYSQL_SLAVE_NOT_RUN;
4864 4865 4866
  /* Forget the relay log's format */
  delete rli->relay_log.description_event_for_exec;
  rli->relay_log.description_event_for_exec= 0;
4867
  rli->reset_inuse_relaylog();
4868
  /* Wake up master_pos_wait() */
Marc Alff's avatar
Marc Alff committed
4869
  mysql_mutex_unlock(&rli->data_lock);
4870
  DBUG_PRINT("info",("Signaling possibly waiting master_pos_wait() functions"));
Marc Alff's avatar
Marc Alff committed
4871
  mysql_cond_broadcast(&rli->data_cond);
unknown's avatar
unknown committed
4872
  rli->ignore_log_space_limit= 0; /* don't need any lock */
4873
  /* we die so won't remember charset - re-update them on next thread start */
4874
  thd->system_thread_info.rpl_sql_info->cached_charset_invalidate();
unknown's avatar
unknown committed
4875 4876 4877 4878 4879

  /*
    TODO: see if we can do this conditionally in next_event() instead
    to avoid unneeded position re-init
  */
4880
  thd->temporary_tables = 0; // remove tempation from destructor to close them
4881
  THD_CHECK_SENTRY(thd);
4882
  rli->sql_driver_thd= 0;
Marc Alff's avatar
Marc Alff committed
4883
  mysql_mutex_lock(&LOCK_thread_count);
Alexander Barkov's avatar
Alexander Barkov committed
4884
  thd->rgi_fake= thd->rgi_slave= NULL;
unknown's avatar
unknown committed
4885
  delete serial_rgi;
Marc Alff's avatar
Marc Alff committed
4886
  mysql_mutex_unlock(&LOCK_thread_count);
4887

4888
#ifdef WITH_WSREP
4889 4890 4891
  /*
    If slave stopped due to node going non primary, we set global flag to
    trigger automatic restart of slave when node joins back to cluster.
4892
  */
4893 4894 4895 4896 4897 4898 4899 4900 4901 4902 4903 4904 4905 4906 4907 4908 4909 4910 4911
  if (wsrep_node_dropped && wsrep_restart_slave)
  {
    if (wsrep_ready)
    {
      WSREP_INFO("Slave error due to node temporarily non-primary"
                 "SQL slave will continue");
      wsrep_node_dropped= FALSE;
      mysql_mutex_unlock(&rli->run_lock);
      WSREP_DEBUG("wsrep_conflict_state now: %d", thd->wsrep_conflict_state);
      WSREP_INFO("slave restart: %d", thd->wsrep_conflict_state);
      thd->wsrep_conflict_state= NO_CONFLICT;
      goto wsrep_restart_point;
    } else {
      WSREP_INFO("Slave error due to node going non-primary");
      WSREP_INFO("wsrep_restart_slave was set and therefore slave will be "
                 "automatically restarted when node joins back to cluster.");
      wsrep_restart_slave_activated= TRUE;
    }
  }
4912
#endif /* WITH_WSREP */
4913

4914 4915 4916 4917 4918
 /*
  Note: the order of the broadcast and unlock calls below (first broadcast, then unlock)
  is important. Otherwise a killer_thread can execute between the calls and
  delete the mi structure leading to a crash! (see BUG#25306 for details)
 */ 
Marc Alff's avatar
Marc Alff committed
4919
  mysql_cond_broadcast(&rli->stop_cond);
4920
  DBUG_EXECUTE_IF("simulate_slave_delay_at_terminate_bug38694", sleep(5););
Marc Alff's avatar
Marc Alff committed
4921
  mysql_mutex_unlock(&rli->run_lock);  // tell the world we are done
4922

4923 4924 4925 4926 4927 4928 4929 4930 4931 4932 4933 4934 4935 4936 4937 4938
  /*
    Deactivate the parallel replication thread pool, if there are now no more
    SQL threads running. Do this here, when we have released all locks, but
    while our THD (and current_thd) is still valid.
  */
  mysql_mutex_lock(&LOCK_active_mi);
  if (opt_slave_parallel_threads > 0 &&
      !master_info_index->any_slave_sql_running())
    rpl_parallel_inactivate_pool(&global_rpl_thread_pool);
  mysql_mutex_unlock(&LOCK_active_mi);

  mysql_mutex_lock(&LOCK_thread_count);
  THD_CHECK_SENTRY(thd);
  delete thd;
  mysql_mutex_unlock(&LOCK_thread_count);

4939
  DBUG_LEAVE;                                   // Must match DBUG_ENTER()
unknown's avatar
unknown committed
4940
  my_thread_end();
4941
#ifdef HAVE_OPENSSL
4942
  ERR_remove_state(0);
4943
#endif
4944
  pthread_exit(0);
4945
  return 0;                                     // Avoid compiler warnings
4946
}
unknown's avatar
unknown committed
4947

4948

unknown's avatar
unknown committed
4949
/*
4950
  process_io_create_file()
unknown's avatar
unknown committed
4951
*/
4952

4953
static int process_io_create_file(Master_info* mi, Create_file_log_event* cev)
unknown's avatar
unknown committed
4954 4955 4956 4957
{
  int error = 1;
  ulong num_bytes;
  bool cev_not_written;
4958 4959
  THD *thd = mi->io_thd;
  NET *net = &mi->mysql->net;
unknown's avatar
unknown committed
4960
  DBUG_ENTER("process_io_create_file");
unknown's avatar
unknown committed
4961 4962

  if (unlikely(!cev->is_valid()))
unknown's avatar
unknown committed
4963
    DBUG_RETURN(1);
4964

4965
  if (!mi->rpl_filter->db_ok(cev->db))
unknown's avatar
unknown committed
4966 4967
  {
    skip_load_data_infile(net);
unknown's avatar
unknown committed
4968
    DBUG_RETURN(0);
unknown's avatar
unknown committed
4969 4970 4971
  }
  DBUG_ASSERT(cev->inited_from_old);
  thd->file_id = cev->file_id = mi->file_id++;
4972
  thd->variables.server_id = cev->server_id;
unknown's avatar
unknown committed
4973
  cev_not_written = 1;
unknown's avatar
unknown committed
4974

unknown's avatar
unknown committed
4975 4976 4977
  if (unlikely(net_request_file(net,cev->fname)))
  {
    sql_print_error("Slave I/O: failed requesting download of '%s'",
unknown's avatar
unknown committed
4978
                    cev->fname);
unknown's avatar
unknown committed
4979 4980 4981
    goto err;
  }

unknown's avatar
unknown committed
4982 4983 4984 4985
  /*
    This dummy block is so we could instantiate Append_block_log_event
    once and then modify it slightly instead of doing it multiple times
    in the loop
unknown's avatar
unknown committed
4986 4987
  */
  {
4988
    Append_block_log_event aev(thd,0,0,0,0);
unknown's avatar
unknown committed
4989

unknown's avatar
unknown committed
4990 4991 4992 4993
    for (;;)
    {
      if (unlikely((num_bytes=my_net_read(net)) == packet_error))
      {
unknown's avatar
unknown committed
4994 4995 4996
        sql_print_error("Network read error downloading '%s' from master",
                        cev->fname);
        goto err;
unknown's avatar
unknown committed
4997 4998 4999
      }
      if (unlikely(!num_bytes)) /* eof */
      {
5000 5001
	/* 3.23 master wants it */
        net_write_command(net, 0, (uchar*) "", 0, (uchar*) "", 0);
unknown's avatar
unknown committed
5002 5003 5004 5005 5006 5007 5008
        /*
          If we wrote Create_file_log_event, then we need to write
          Execute_load_log_event. If we did not write Create_file_log_event,
          then this is an empty file and we can just do as if the LOAD DATA
          INFILE had not existed, i.e. write nothing.
        */
        if (unlikely(cev_not_written))
unknown's avatar
unknown committed
5009 5010 5011 5012 5013
          break;
        Execute_load_log_event xev(thd,0,0);
        xev.log_pos = cev->log_pos;
        if (unlikely(mi->rli.relay_log.append(&xev)))
        {
5014
          mi->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_WRITE_FAILURE, NULL,
5015 5016
                     ER(ER_SLAVE_RELAY_LOG_WRITE_FAILURE),
                     "error writing Exec_load event to relay log");
unknown's avatar
unknown committed
5017 5018 5019 5020
          goto err;
        }
        mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total);
        break;
unknown's avatar
unknown committed
5021 5022 5023
      }
      if (unlikely(cev_not_written))
      {
unknown's avatar
unknown committed
5024
        cev->block = net->read_pos;
unknown's avatar
unknown committed
5025 5026 5027
        cev->block_len = num_bytes;
        if (unlikely(mi->rli.relay_log.append(cev)))
        {
5028
          mi->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_WRITE_FAILURE, NULL,
5029 5030
                     ER(ER_SLAVE_RELAY_LOG_WRITE_FAILURE),
                     "error writing Create_file event to relay log");
unknown's avatar
unknown committed
5031 5032 5033 5034
          goto err;
        }
        cev_not_written=0;
        mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total);
unknown's avatar
unknown committed
5035 5036 5037
      }
      else
      {
unknown's avatar
unknown committed
5038
        aev.block = net->read_pos;
unknown's avatar
unknown committed
5039 5040 5041 5042
        aev.block_len = num_bytes;
        aev.log_pos = cev->log_pos;
        if (unlikely(mi->rli.relay_log.append(&aev)))
        {
5043
          mi->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_WRITE_FAILURE, NULL,
5044 5045
                     ER(ER_SLAVE_RELAY_LOG_WRITE_FAILURE),
                     "error writing Append_block event to relay log");
unknown's avatar
unknown committed
5046 5047 5048
          goto err;
        }
        mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total) ;
unknown's avatar
unknown committed
5049 5050 5051 5052 5053
      }
    }
  }
  error=0;
err:
unknown's avatar
unknown committed
5054
  DBUG_RETURN(error);
unknown's avatar
unknown committed
5055
}
unknown's avatar
unknown committed
5056

5057

unknown's avatar
unknown committed
5058
/*
unknown's avatar
unknown committed
5059 5060 5061 5062
  Start using a new binary log on the master

  SYNOPSIS
    process_io_rotate()
unknown's avatar
unknown committed
5063 5064
    mi                  master_info for the slave
    rev                 The rotate log event read from the binary log
unknown's avatar
unknown committed
5065 5066

  DESCRIPTION
5067
    Updates the master info with the place in the next binary
unknown's avatar
unknown committed
5068
    log where we should start reading.
5069
    Rotate the relay log to avoid mixed-format relay logs.
unknown's avatar
unknown committed
5070 5071 5072 5073 5074

  NOTES
    We assume we already locked mi->data_lock

  RETURN VALUES
unknown's avatar
unknown committed
5075 5076
    0           ok
    1           Log event is illegal
unknown's avatar
unknown committed
5077 5078 5079

*/

5080
static int process_io_rotate(Master_info *mi, Rotate_log_event *rev)
5081
{
5082
  DBUG_ENTER("process_io_rotate");
Marc Alff's avatar
Marc Alff committed
5083
  mysql_mutex_assert_owner(&mi->data_lock);
5084

unknown's avatar
unknown committed
5085
  if (unlikely(!rev->is_valid()))
5086
    DBUG_RETURN(1);
unknown's avatar
unknown committed
5087

5088
  /* Safe copy as 'rev' has been "sanitized" in Rotate_log_event's ctor */
unknown's avatar
unknown committed
5089 5090
  memcpy(mi->master_log_name, rev->new_log_ident, rev->ident_len+1);
  mi->master_log_pos= rev->pos;
unknown's avatar
unknown committed
5091
  DBUG_PRINT("info", ("master_log_pos: '%s' %lu",
unknown's avatar
unknown committed
5092
                      mi->master_log_name, (ulong) mi->master_log_pos));
5093
#ifndef DBUG_OFF
unknown's avatar
unknown committed
5094 5095 5096 5097 5098
  /*
    If we do not do this, we will be getting the first
    rotate event forever, so we need to not disconnect after one.
  */
  if (disconnect_slave_event_count)
5099
    mi->events_till_disconnect++;
5100
#endif
5101

5102 5103 5104 5105 5106 5107 5108 5109 5110
  /*
    If description_event_for_queue is format <4, there is conversion in the
    relay log to the slave's format (4). And Rotate can mean upgrade or
    nothing. If upgrade, it's to 5.0 or newer, so we will get a Format_desc, so
    no need to reset description_event_for_queue now. And if it's nothing (same
    master version as before), no need (still using the slave's format).
  */
  if (mi->rli.relay_log.description_event_for_queue->binlog_version >= 4)
  {
5111 5112 5113
    DBUG_ASSERT(mi->rli.relay_log.description_event_for_queue->checksum_alg ==
                mi->rli.relay_log.relay_log_checksum_alg);
    
5114 5115 5116 5117
    delete mi->rli.relay_log.description_event_for_queue;
    /* start from format 3 (MySQL 4.0) again */
    mi->rli.relay_log.description_event_for_queue= new
      Format_description_log_event(3);
5118 5119
    mi->rli.relay_log.description_event_for_queue->checksum_alg=
      mi->rli.relay_log.relay_log_checksum_alg;    
5120
  }
5121 5122 5123 5124
  /*
    Rotate the relay log makes binlog format detection easier (at next slave
    start or mysqlbinlog)
  */
5125
  DBUG_RETURN(rotate_relay_log(mi) /* will take the right mutexes */);
5126 5127
}

unknown's avatar
unknown committed
5128
/*
5129 5130
  Reads a 3.23 event and converts it to the slave's format. This code was
  copied from MySQL 4.0.
unknown's avatar
unknown committed
5131
*/
5132
static int queue_binlog_ver_1_event(Master_info *mi, const char *buf,
unknown's avatar
unknown committed
5133
                           ulong event_len)
5134
{
unknown's avatar
unknown committed
5135
  const char *errmsg = 0;
unknown's avatar
unknown committed
5136 5137 5138
  ulong inc_pos;
  bool ignore_event= 0;
  char *tmp_buf = 0;
5139
  Relay_log_info *rli= &mi->rli;
5140
  DBUG_ENTER("queue_binlog_ver_1_event");
unknown's avatar
unknown committed
5141

unknown's avatar
unknown committed
5142 5143 5144
  /*
    If we get Load event, we need to pass a non-reusable buffer
    to read_log_event, so we do a trick
unknown's avatar
unknown committed
5145
  */
5146
  if ((uchar)buf[EVENT_TYPE_OFFSET] == LOAD_EVENT)
unknown's avatar
unknown committed
5147 5148 5149
  {
    if (unlikely(!(tmp_buf=(char*)my_malloc(event_len+1,MYF(MY_WME)))))
    {
5150
      mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, NULL,
5151
                 ER(ER_SLAVE_FATAL_ERROR), "Memory allocation failed");
unknown's avatar
unknown committed
5152
      DBUG_RETURN(1);
unknown's avatar
unknown committed
5153 5154
    }
    memcpy(tmp_buf,buf,event_len);
5155 5156 5157 5158 5159 5160 5161 5162
    /*
      Create_file constructor wants a 0 as last char of buffer, this 0 will
      serve as the string-termination char for the file's name (which is at the
      end of the buffer)
      We must increment event_len, otherwise the event constructor will not see
      this end 0, which leads to segfault.
    */
    tmp_buf[event_len++]=0;
unknown's avatar
unknown committed
5163
    int4store(tmp_buf+EVENT_LEN_OFFSET, event_len);
unknown's avatar
unknown committed
5164 5165
    buf = (const char*)tmp_buf;
  }
5166 5167 5168 5169 5170 5171
  /*
    This will transform LOAD_EVENT into CREATE_FILE_EVENT, ask the master to
    send the loaded file, and write it to the relay log in the form of
    Append_block/Exec_load (the SQL thread needs the data, as that thread is not
    connected to the master).
  */
5172 5173 5174
  Log_event *ev=
    Log_event::read_log_event(buf, event_len, &errmsg,
                              mi->rli.relay_log.description_event_for_queue, 0);
5175
  if (unlikely(!ev))
5176 5177
  {
    sql_print_error("Read invalid event from master: '%s',\
unknown's avatar
unknown committed
5178
 master could be corrupt but a more likely cause of this is a bug",
unknown's avatar
unknown committed
5179
                    errmsg);
5180
    my_free(tmp_buf);
unknown's avatar
unknown committed
5181
    DBUG_RETURN(1);
5182
  }
5183

Marc Alff's avatar
Marc Alff committed
5184
  mysql_mutex_lock(&mi->data_lock);
5185
  ev->log_pos= mi->master_log_pos; /* 3.23 events don't contain log_pos */
unknown's avatar
unknown committed
5186
  switch (ev->get_type_code()) {
unknown's avatar
unknown committed
5187
  case STOP_EVENT:
5188
    ignore_event= 1;
unknown's avatar
unknown committed
5189 5190
    inc_pos= event_len;
    break;
5191
  case ROTATE_EVENT:
5192
    if (unlikely(process_io_rotate(mi,(Rotate_log_event*)ev)))
5193 5194
    {
      delete ev;
Marc Alff's avatar
Marc Alff committed
5195
      mysql_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
5196
      DBUG_RETURN(1);
5197
    }
unknown's avatar
unknown committed
5198
    inc_pos= 0;
5199
    break;
unknown's avatar
unknown committed
5200
  case CREATE_FILE_EVENT:
5201 5202 5203 5204 5205 5206
    /*
      Yes it's possible to have CREATE_FILE_EVENT here, even if we're in
      queue_old_event() which is for 3.23 events which don't comprise
      CREATE_FILE_EVENT. This is because read_log_event() above has just
      transformed LOAD_EVENT into CREATE_FILE_EVENT.
    */
unknown's avatar
unknown committed
5207
  {
unknown's avatar
unknown committed
5208
    /* We come here when and only when tmp_buf != 0 */
5209
    DBUG_ASSERT(tmp_buf != 0);
5210 5211
    inc_pos=event_len;
    ev->log_pos+= inc_pos;
unknown's avatar
unknown committed
5212
    int error = process_io_create_file(mi,(Create_file_log_event*)ev);
5213
    delete ev;
5214
    mi->master_log_pos += inc_pos;
unknown's avatar
unknown committed
5215
    DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
Marc Alff's avatar
Marc Alff committed
5216
    mysql_mutex_unlock(&mi->data_lock);
5217
    my_free(tmp_buf);
unknown's avatar
unknown committed
5218
    DBUG_RETURN(error);
unknown's avatar
unknown committed
5219
  }
5220
  default:
unknown's avatar
unknown committed
5221
    inc_pos= event_len;
5222 5223
    break;
  }
unknown's avatar
unknown committed
5224
  if (likely(!ignore_event))
5225
  {
unknown's avatar
unknown committed
5226 5227
    if (ev->log_pos)
      /*
5228 5229 5230 5231
         Don't do it for fake Rotate events (see comment in
      Log_event::Log_event(const char* buf...) in log_event.cc).
      */
      ev->log_pos+= event_len; /* make log_pos be the pos of the end of the event */
unknown's avatar
unknown committed
5232
    if (unlikely(rli->relay_log.append(ev)))
5233 5234
    {
      delete ev;
Marc Alff's avatar
Marc Alff committed
5235
      mysql_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
5236
      DBUG_RETURN(1);
5237
    }
unknown's avatar
unknown committed
5238
    rli->relay_log.harvest_bytes_written(&rli->log_space_total);
5239 5240
  }
  delete ev;
unknown's avatar
unknown committed
5241
  mi->master_log_pos+= inc_pos;
unknown's avatar
unknown committed
5242
  DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
Marc Alff's avatar
Marc Alff committed
5243
  mysql_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
5244
  DBUG_RETURN(0);
5245 5246
}

5247 5248 5249 5250
/*
  Reads a 4.0 event and converts it to the slave's format. This code was copied
  from queue_binlog_ver_1_event(), with some affordable simplifications.
*/
5251
static int queue_binlog_ver_3_event(Master_info *mi, const char *buf,
unknown's avatar
unknown committed
5252
                           ulong event_len)
5253 5254 5255 5256
{
  const char *errmsg = 0;
  ulong inc_pos;
  char *tmp_buf = 0;
5257
  Relay_log_info *rli= &mi->rli;
5258 5259 5260
  DBUG_ENTER("queue_binlog_ver_3_event");

  /* read_log_event() will adjust log_pos to be end_log_pos */
5261 5262 5263
  Log_event *ev=
    Log_event::read_log_event(buf,event_len, &errmsg,
                              mi->rli.relay_log.description_event_for_queue, 0);
5264 5265 5266 5267
  if (unlikely(!ev))
  {
    sql_print_error("Read invalid event from master: '%s',\
 master could be corrupt but a more likely cause of this is a bug",
unknown's avatar
unknown committed
5268
                    errmsg);
5269
    my_free(tmp_buf);
5270 5271
    DBUG_RETURN(1);
  }
Marc Alff's avatar
Marc Alff committed
5272
  mysql_mutex_lock(&mi->data_lock);
5273 5274 5275 5276 5277 5278 5279
  switch (ev->get_type_code()) {
  case STOP_EVENT:
    goto err;
  case ROTATE_EVENT:
    if (unlikely(process_io_rotate(mi,(Rotate_log_event*)ev)))
    {
      delete ev;
Marc Alff's avatar
Marc Alff committed
5280
      mysql_mutex_unlock(&mi->data_lock);
5281 5282 5283 5284 5285 5286 5287 5288
      DBUG_RETURN(1);
    }
    inc_pos= 0;
    break;
  default:
    inc_pos= event_len;
    break;
  }
5289

5290 5291 5292
  if (unlikely(rli->relay_log.append(ev)))
  {
    delete ev;
Marc Alff's avatar
Marc Alff committed
5293
    mysql_mutex_unlock(&mi->data_lock);
5294 5295 5296 5297 5298 5299
    DBUG_RETURN(1);
  }
  rli->relay_log.harvest_bytes_written(&rli->log_space_total);
  delete ev;
  mi->master_log_pos+= inc_pos;
err:
unknown's avatar
unknown committed
5300
  DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
Marc Alff's avatar
Marc Alff committed
5301
  mysql_mutex_unlock(&mi->data_lock);
5302 5303 5304 5305 5306 5307 5308 5309 5310 5311
  DBUG_RETURN(0);
}

/*
  queue_old_event()

  Writes a 3.23 or 4.0 event to the relay log, after converting it to the 5.0
  (exactly, slave's) format. To do the conversion, we create a 5.0 event from
  the 3.23/4.0 bytes, then write this event to the relay log.

unknown's avatar
unknown committed
5312
  TODO:
5313 5314 5315 5316
    Test this code before release - it has to be tested on a separate
    setup with 3.23 master or 4.0 master
*/

5317
static int queue_old_event(Master_info *mi, const char *buf,
unknown's avatar
unknown committed
5318
                           ulong event_len)
5319
{
5320 5321
  DBUG_ENTER("queue_old_event");

5322 5323 5324
  switch (mi->rli.relay_log.description_event_for_queue->binlog_version)
  {
  case 1:
5325
      DBUG_RETURN(queue_binlog_ver_1_event(mi,buf,event_len));
5326
  case 3:
5327
      DBUG_RETURN(queue_binlog_ver_3_event(mi,buf,event_len));
5328 5329
  default: /* unsupported format; eg version 2 */
    DBUG_PRINT("info",("unsupported binlog format %d in queue_old_event()",
unknown's avatar
unknown committed
5330
                       mi->rli.relay_log.description_event_for_queue->binlog_version));
5331
    DBUG_RETURN(1);
5332 5333
  }
}
5334

unknown's avatar
unknown committed
5335
/*
5336 5337
  queue_event()

5338 5339 5340 5341 5342
  If the event is 3.23/4.0, passes it to queue_old_event() which will convert
  it. Otherwise, writes a 5.0 (or newer) event to the relay log. Then there is
  no format conversion, it's pure read/write of bytes.
  So a 5.0.0 slave's relay log can contain events in the slave's format or in
  any >=5.0.0 format.
unknown's avatar
unknown committed
5343 5344
*/

5345
static int queue_event(Master_info* mi,const char* buf, ulong event_len)
5346
{
unknown's avatar
unknown committed
5347
  int error= 0;
Andrei Elkin's avatar
Andrei Elkin committed
5348
  String error_msg;
5349 5350
  ulonglong inc_pos;
  ulonglong event_pos;
5351
  Relay_log_info *rli= &mi->rli;
Marc Alff's avatar
Marc Alff committed
5352
  mysql_mutex_t *log_lock= rli->relay_log.get_log_lock();
Andrei Elkin's avatar
Andrei Elkin committed
5353
  ulong s_id;
5354
  bool unlock_data_lock= TRUE;
unknown's avatar
unknown committed
5355
  bool gtid_skip_enqueue= false;
5356 5357
  bool got_gtid_event= false;
  rpl_gtid event_gtid;
unknown's avatar
unknown committed
5358

5359 5360 5361 5362 5363 5364 5365 5366 5367 5368 5369 5370 5371 5372 5373 5374 5375
  /*
    FD_q must have been prepared for the first R_a event
    inside get_master_version_and_clock()
    Show-up of FD:s affects checksum_alg at once because
    that changes FD_queue.
  */
  uint8 checksum_alg= mi->checksum_alg_before_fd != BINLOG_CHECKSUM_ALG_UNDEF ? 
    mi->checksum_alg_before_fd :
    mi->rli.relay_log.relay_log_checksum_alg;

  char *save_buf= NULL; // needed for checksumming the fake Rotate event
  char rot_buf[LOG_EVENT_HEADER_LEN + ROTATE_HEADER_LEN + FN_REFLEN];

  DBUG_ASSERT(checksum_alg == BINLOG_CHECKSUM_ALG_OFF || 
              checksum_alg == BINLOG_CHECKSUM_ALG_UNDEF || 
              checksum_alg == BINLOG_CHECKSUM_ALG_CRC32); 

unknown's avatar
unknown committed
5376
  DBUG_ENTER("queue_event");
5377 5378 5379 5380 5381 5382 5383 5384 5385 5386 5387 5388 5389 5390 5391 5392 5393 5394 5395 5396 5397 5398 5399 5400 5401 5402 5403 5404 5405 5406 5407 5408 5409 5410 5411 5412 5413 5414 5415 5416 5417 5418 5419
  /*
    FD_queue checksum alg description does not apply in a case of
    FD itself. The one carries both parts of the checksum data.
  */
  if (buf[EVENT_TYPE_OFFSET] == FORMAT_DESCRIPTION_EVENT)
  {
    checksum_alg= get_checksum_alg(buf, event_len);
  }
  else if (buf[EVENT_TYPE_OFFSET] == START_EVENT_V3)
  {
    // checksum behaviour is similar to the pre-checksum FD handling
    mi->checksum_alg_before_fd= BINLOG_CHECKSUM_ALG_UNDEF;
    mi->rli.relay_log.description_event_for_queue->checksum_alg=
      mi->rli.relay_log.relay_log_checksum_alg= checksum_alg=
      BINLOG_CHECKSUM_ALG_OFF;
  }

  // does not hold always because of old binlog can work with NM 
  // DBUG_ASSERT(checksum_alg != BINLOG_CHECKSUM_ALG_UNDEF);

  // should hold unless manipulations with RL. Tests that do that
  // will have to refine the clause.
  DBUG_ASSERT(mi->rli.relay_log.relay_log_checksum_alg !=
              BINLOG_CHECKSUM_ALG_UNDEF);
              
  // Emulate the network corruption
  DBUG_EXECUTE_IF("corrupt_queue_event",
    if (buf[EVENT_TYPE_OFFSET] != FORMAT_DESCRIPTION_EVENT)
    {
      char *debug_event_buf_c = (char*) buf;
      int debug_cor_pos = rand() % (event_len - BINLOG_CHECKSUM_LEN);
      debug_event_buf_c[debug_cor_pos] =~ debug_event_buf_c[debug_cor_pos];
      DBUG_PRINT("info", ("Corrupt the event at queue_event: byte on position %d", debug_cor_pos));
      DBUG_SET("-d,corrupt_queue_event");
    }
  );
                                              
  if (event_checksum_test((uchar *) buf, event_len, checksum_alg))
  {
    error= ER_NETWORK_READ_EVENT_CHECKSUM_FAILURE;
    unlock_data_lock= FALSE;
    goto err;
  }
unknown's avatar
unknown committed
5420

5421
  if (mi->rli.relay_log.description_event_for_queue->binlog_version<4 &&
5422
      (uchar)buf[EVENT_TYPE_OFFSET] != FORMAT_DESCRIPTION_EVENT /* a way to escape */)
unknown's avatar
unknown committed
5423
    DBUG_RETURN(queue_old_event(mi,buf,event_len));
5424

Marc Alff's avatar
Marc Alff committed
5425
  mysql_mutex_lock(&mi->data_lock);
unknown's avatar
unknown committed
5426

5427
  switch ((uchar)buf[EVENT_TYPE_OFFSET]) {
5428
  case STOP_EVENT:
5429 5430
    /*
      We needn't write this event to the relay log. Indeed, it just indicates a
unknown's avatar
unknown committed
5431 5432 5433
      master server shutdown. The only thing this does is cleaning. But
      cleaning is already done on a per-master-thread basis (as the master
      server is shutting down cleanly, it has written all DROP TEMPORARY TABLE
5434
      prepared statements' deletion are TODO only when we binlog prep stmts).
unknown's avatar
unknown committed
5435

unknown's avatar
unknown committed
5436 5437 5438 5439
      We don't even increment mi->master_log_pos, because we may be just after
      a Rotate event. Btw, in a few milliseconds we are going to have a Start
      event from the next binlog (unless the master is presently running
      without --log-bin).
5440 5441
    */
    goto err;
5442 5443
  case ROTATE_EVENT:
  {
5444 5445 5446 5447
    Rotate_log_event rev(buf, checksum_alg != BINLOG_CHECKSUM_ALG_OFF ?
                         event_len - BINLOG_CHECKSUM_LEN : event_len,
                         mi->rli.relay_log.description_event_for_queue);

5448 5449 5450 5451 5452 5453 5454 5455 5456 5457 5458 5459 5460 5461 5462 5463 5464 5465 5466 5467 5468 5469 5470 5471 5472 5473 5474 5475 5476 5477 5478 5479 5480 5481 5482 5483 5484 5485 5486 5487 5488 5489 5490 5491 5492 5493 5494 5495 5496 5497 5498 5499 5500 5501 5502 5503 5504 5505 5506 5507 5508 5509 5510 5511 5512 5513 5514 5515 5516 5517 5518 5519 5520 5521 5522 5523 5524 5525 5526 5527
    if (unlikely(mi->gtid_reconnect_event_skip_count) &&
        unlikely(!mi->gtid_event_seen) &&
        rev.is_artificial_event() &&
        (mi->prev_master_id != mi->master_id ||
         strcmp(rev.new_log_ident, mi->master_log_name) != 0))
    {
      /*
        Artificial Rotate_log_event is the first event we receive at the start
        of each master binlog file. It gives the name of the new binlog file.

        Normally, we already have this name from the real rotate event at the
        end of the previous binlog file (unless we are making a new connection
        using GTID). But if the master server restarted/crashed, there is no
        rotate event at the end of the prior binlog file, so the name is new.

        We use this fact to handle a special case of master crashing. If the
        master crashed while writing the binlog, it might end with a partial
        event group lacking the COMMIT/XID event, which must be rolled
        back. If the slave IO thread happens to get a disconnect in the middle
        of exactly this event group, it will try to reconnect at the same GTID
        and skip already fetched events. However, that GTID did not commit on
        the master before the crash, so it does not really exist, and the
        master will connect the slave at the next following GTID starting in
        the next binlog. This could confuse the slave and make it mix the
        start of one event group with the end of another.

        But we detect this case here, by noticing the change of binlog name
        which detects the missing rotate event at the end of the previous
        binlog file. In this case, we reset the counters to make us not skip
        the next event group, and queue an artificial Format Description
        event. The previously fetched incomplete event group will then be
        rolled back when the Format Description event is executed by the SQL
        thread.

        A similar case is if the reconnect somehow connects to a different
        master server (like due to a network proxy or IP address takeover).
        We detect this case by noticing a change of server_id and in this
        case likewise rollback the partially received event group.
      */
      Format_description_log_event fdle(4);

      if (mi->prev_master_id != mi->master_id)
        sql_print_warning("The server_id of master server changed in the "
                          "middle of GTID %u-%u-%llu. Assuming a change of "
                          "master server, so rolling back the previously "
                          "received partial transaction. Expected: %lu, "
                          "received: %lu", mi->last_queued_gtid.domain_id,
                          mi->last_queued_gtid.server_id,
                          mi->last_queued_gtid.seq_no,
                          mi->prev_master_id, mi->master_id);
      else if (strcmp(rev.new_log_ident, mi->master_log_name) != 0)
        sql_print_warning("Unexpected change of master binlog file name in the "
                          "middle of GTID %u-%u-%llu, assuming that master has "
                          "crashed and rolling back the transaction. Expected: "
                          "'%s', received: '%s'",
                          mi->last_queued_gtid.domain_id,
                          mi->last_queued_gtid.server_id,
                          mi->last_queued_gtid.seq_no,
                          mi->master_log_name, rev.new_log_ident);

      mysql_mutex_lock(log_lock);
      if (likely(!fdle.write(rli->relay_log.get_log_file()) &&
                 !rli->relay_log.flush_and_sync(NULL)))
      {
        rli->relay_log.harvest_bytes_written(&rli->log_space_total);
      }
      else
      {
        error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
        mysql_mutex_unlock(log_lock);
        goto err;
      }
      rli->relay_log.signal_update();
      mysql_mutex_unlock(log_lock);

      mi->gtid_reconnect_event_skip_count= 0;
      mi->events_queued_since_last_gtid= 0;
    }
    mi->prev_master_id= mi->master_id;

5528
    if (unlikely(process_io_rotate(mi, &rev)))
unknown's avatar
unknown committed
5529
    {
Andrei Elkin's avatar
Andrei Elkin committed
5530
      error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
5531
      goto err;
unknown's avatar
unknown committed
5532
    }
5533 5534 5535 5536 5537 5538 5539 5540 5541 5542 5543 5544 5545 5546 5547 5548
    /* 
       Checksum special cases for the fake Rotate (R_f) event caused by the protocol
       of events generation and serialization in RL where Rotate of master is 
       queued right next to FD of slave.
       Since it's only FD that carries the alg desc of FD_s has to apply to R_m.
       Two special rules apply only to the first R_f which comes in before any FD_m.
       The 2nd R_f should be compatible with the FD_s that must have taken over
       the last seen FD_m's (A).
       
       RSC_1: If OM \and fake Rotate \and slave is configured to
              to compute checksum for its first FD event for RL
              the fake Rotate gets checksummed here.
    */
    if (uint4korr(&buf[0]) == 0 && checksum_alg == BINLOG_CHECKSUM_ALG_OFF &&
        mi->rli.relay_log.relay_log_checksum_alg != BINLOG_CHECKSUM_ALG_OFF)
    {
5549
      ha_checksum rot_crc= 0;
5550 5551 5552 5553 5554 5555 5556 5557 5558 5559 5560 5561 5562 5563 5564 5565 5566 5567 5568 5569 5570 5571 5572 5573 5574 5575 5576 5577 5578 5579 5580 5581 5582 5583 5584
      event_len += BINLOG_CHECKSUM_LEN;
      memcpy(rot_buf, buf, event_len - BINLOG_CHECKSUM_LEN);
      int4store(&rot_buf[EVENT_LEN_OFFSET],
                uint4korr(&rot_buf[EVENT_LEN_OFFSET]) + BINLOG_CHECKSUM_LEN);
      rot_crc= my_checksum(rot_crc, (const uchar *) rot_buf,
                           event_len - BINLOG_CHECKSUM_LEN);
      int4store(&rot_buf[event_len - BINLOG_CHECKSUM_LEN], rot_crc);
      DBUG_ASSERT(event_len == uint4korr(&rot_buf[EVENT_LEN_OFFSET]));
      DBUG_ASSERT(mi->rli.relay_log.description_event_for_queue->checksum_alg ==
                  mi->rli.relay_log.relay_log_checksum_alg);
      /* the first one */
      DBUG_ASSERT(mi->checksum_alg_before_fd != BINLOG_CHECKSUM_ALG_UNDEF);
      save_buf= (char *) buf;
      buf= rot_buf;
    }
    else
      /*
        RSC_2: If NM \and fake Rotate \and slave does not compute checksum
        the fake Rotate's checksum is stripped off before relay-logging.
      */
      if (uint4korr(&buf[0]) == 0 && checksum_alg != BINLOG_CHECKSUM_ALG_OFF &&
          mi->rli.relay_log.relay_log_checksum_alg == BINLOG_CHECKSUM_ALG_OFF)
      {
        event_len -= BINLOG_CHECKSUM_LEN;
        memcpy(rot_buf, buf, event_len);
        int4store(&rot_buf[EVENT_LEN_OFFSET],
                  uint4korr(&rot_buf[EVENT_LEN_OFFSET]) - BINLOG_CHECKSUM_LEN);
        DBUG_ASSERT(event_len == uint4korr(&rot_buf[EVENT_LEN_OFFSET]));
        DBUG_ASSERT(mi->rli.relay_log.description_event_for_queue->checksum_alg ==
                    mi->rli.relay_log.relay_log_checksum_alg);
        /* the first one */
        DBUG_ASSERT(mi->checksum_alg_before_fd != BINLOG_CHECKSUM_ALG_UNDEF);
        save_buf= (char *) buf;
        buf= rot_buf;
      }
5585 5586 5587 5588
    /*
      Now the I/O thread has just changed its mi->master_log_name, so
      incrementing mi->master_log_pos is nonsense.
    */
unknown's avatar
unknown committed
5589
    inc_pos= 0;
5590 5591
    break;
  }
5592 5593 5594 5595 5596 5597 5598
  case FORMAT_DESCRIPTION_EVENT:
  {
    /*
      Create an event, and save it (when we rotate the relay log, we will have
      to write this event again).
    */
    /*
5599 5600 5601
      We are the only thread which reads/writes description_event_for_queue.
      The relay_log struct does not move (though some members of it can
      change), so we needn't any lock (no rli->data_lock, no log lock).
5602
    */
5603
    Format_description_log_event* tmp;
5604
    const char* errmsg;
5605 5606
    // mark it as undefined that is irrelevant anymore
    mi->checksum_alg_before_fd= BINLOG_CHECKSUM_ALG_UNDEF;
5607
    if (!(tmp= (Format_description_log_event*)
5608
          Log_event::read_log_event(buf, event_len, &errmsg,
5609 5610
                                    mi->rli.relay_log.description_event_for_queue,
                                    1)))
5611
    {
Andrei Elkin's avatar
Andrei Elkin committed
5612
      error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
5613 5614
      goto err;
    }
5615 5616
    delete mi->rli.relay_log.description_event_for_queue;
    mi->rli.relay_log.description_event_for_queue= tmp;
5617 5618 5619 5620 5621 5622
    if (tmp->checksum_alg == BINLOG_CHECKSUM_ALG_UNDEF)
      tmp->checksum_alg= BINLOG_CHECKSUM_ALG_OFF;

    /* installing new value of checksum Alg for relay log */
    mi->rli.relay_log.relay_log_checksum_alg= tmp->checksum_alg;

unknown's avatar
unknown committed
5623 5624 5625 5626 5627 5628 5629 5630
    /*
      Do not queue any format description event that we receive after a
      reconnect where we are skipping over a partial event group received
      before the reconnect.

      (If we queued such an event, and it was the first format_description
      event after master restart, the slave SQL thread would think that
      the partial event group before it in the relay log was from a
5631
      previous master crash and should be rolled back).
unknown's avatar
unknown committed
5632
    */
5633
    if (unlikely(mi->gtid_reconnect_event_skip_count && !mi->gtid_event_seen))
unknown's avatar
unknown committed
5634 5635
        gtid_skip_enqueue= true;

unknown's avatar
unknown committed
5636
    /*
5637
       Though this does some conversion to the slave's format, this will
unknown's avatar
unknown committed
5638
       preserve the master's binlog format version, and number of event types.
5639
    */
unknown's avatar
unknown committed
5640
    /*
5641
       If the event was not requested by the slave (the slave did not ask for
unknown's avatar
unknown committed
5642
       it), i.e. has end_log_pos=0, we do not increment mi->master_log_pos
5643 5644 5645
    */
    inc_pos= uint4korr(buf+LOG_POS_OFFSET) ? event_len : 0;
    DBUG_PRINT("info",("binlog format is now %d",
unknown's avatar
unknown committed
5646
                       mi->rli.relay_log.description_event_for_queue->binlog_version));
5647 5648 5649

  }
  break;
Andrei Elkin's avatar
Andrei Elkin committed
5650 5651 5652 5653 5654 5655 5656

  case HEARTBEAT_LOG_EVENT:
  {
    /*
      HB (heartbeat) cannot come before RL (Relay)
    */
    char  llbuf[22];
unknown's avatar
unknown committed
5657 5658 5659 5660 5661
    Heartbeat_log_event hb(buf,
                           mi->rli.relay_log.relay_log_checksum_alg
                           != BINLOG_CHECKSUM_ALG_OFF ?
                           event_len - BINLOG_CHECKSUM_LEN : event_len,
                           mi->rli.relay_log.description_event_for_queue);
Andrei Elkin's avatar
Andrei Elkin committed
5662 5663 5664 5665 5666 5667 5668 5669 5670 5671 5672 5673 5674 5675 5676 5677 5678
    if (!hb.is_valid())
    {
      error= ER_SLAVE_HEARTBEAT_FAILURE;
      error_msg.append(STRING_WITH_LEN("inconsistent heartbeat event content;"));
      error_msg.append(STRING_WITH_LEN("the event's data: log_file_name "));
      error_msg.append(hb.get_log_ident(), (uint) strlen(hb.get_log_ident()));
      error_msg.append(STRING_WITH_LEN(" log_pos "));
      llstr(hb.log_pos, llbuf);
      error_msg.append(llbuf, strlen(llbuf));
      goto err;
    }
    mi->received_heartbeats++;
    /* 
       compare local and event's versions of log_file, log_pos.
       
       Heartbeat is sent only after an event corresponding to the corrdinates
       the heartbeat carries.
unknown's avatar
unknown committed
5679
       Slave can not have a higher coordinate except in the only
Andrei Elkin's avatar
Andrei Elkin committed
5680 5681 5682 5683
       special case when mi->master_log_name, master_log_pos have never
       been updated by Rotate event i.e when slave does not have any history
       with the master (and thereafter mi->master_log_pos is NULL).

unknown's avatar
unknown committed
5684 5685
       Slave can have lower coordinates, if some event from master was omitted.

Andrei Elkin's avatar
Andrei Elkin committed
5686 5687 5688 5689
       TODO: handling `when' for SHOW SLAVE STATUS' snds behind
    */
    if ((memcmp(mi->master_log_name, hb.get_log_ident(), hb.get_ident_len())
         && mi->master_log_name != NULL)
unknown's avatar
unknown committed
5690
        || mi->master_log_pos > hb.log_pos)
Andrei Elkin's avatar
Andrei Elkin committed
5691 5692 5693 5694 5695 5696 5697 5698 5699 5700 5701 5702 5703 5704
    {
      /* missed events of heartbeat from the past */
      error= ER_SLAVE_HEARTBEAT_FAILURE;
      error_msg.append(STRING_WITH_LEN("heartbeat is not compatible with local info;"));
      error_msg.append(STRING_WITH_LEN("the event's data: log_file_name "));
      error_msg.append(hb.get_log_ident(), (uint) strlen(hb.get_log_ident()));
      error_msg.append(STRING_WITH_LEN(" log_pos "));
      llstr(hb.log_pos, llbuf);
      error_msg.append(llbuf, strlen(llbuf));
      goto err;
    }
    goto skip_relay_logging;
  }
  break;
5705

unknown's avatar
unknown committed
5706 5707 5708 5709 5710
  case GTID_LIST_EVENT:
  {
    const char *errmsg;
    Gtid_list_log_event *glev;
    Log_event *tmp;
5711
    uint32 flags;
unknown's avatar
unknown committed
5712 5713 5714 5715 5716 5717 5718 5719 5720

    if (!(tmp= Log_event::read_log_event(buf, event_len, &errmsg,
           mi->rli.relay_log.description_event_for_queue,
           opt_slave_sql_verify_checksum)))
    {
      error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
      goto err;
    }
    glev= static_cast<Gtid_list_log_event *>(tmp);
unknown's avatar
unknown committed
5721
    event_pos= glev->log_pos;
5722
    flags= glev->gl_flags;
unknown's avatar
unknown committed
5723 5724 5725
    delete glev;

    /*
unknown's avatar
unknown committed
5726 5727 5728 5729 5730
      We use fake Gtid_list events to update the old-style position (among
      other things).

      Early code created fake Gtid_list events with zero log_pos, those should
      not modify old-style position.
unknown's avatar
unknown committed
5731
    */
unknown's avatar
unknown committed
5732 5733 5734 5735
    if (event_pos == 0 || event_pos <= mi->master_log_pos)
      inc_pos= 0;
    else
      inc_pos= event_pos - mi->master_log_pos;
5736 5737 5738 5739 5740 5741 5742

    if (mi->rli.until_condition == Relay_log_info::UNTIL_GTID &&
        flags & Gtid_list_log_event::FLAG_UNTIL_REACHED)
    {
      char str_buf[128];
      String str(str_buf, sizeof(str_buf), system_charset_info);
      mi->rli.until_gtid_pos.to_string(&str);
5743
      sql_print_information("Slave I/O thread stops because it reached its"
5744 5745 5746
                            " UNTIL master_gtid_pos %s", str.c_ptr_safe());
      mi->abort_slave= true;
    }
unknown's avatar
unknown committed
5747 5748 5749 5750 5751
  }
  break;

  case GTID_EVENT:
  {
5752
    uchar gtid_flag;
unknown's avatar
unknown committed
5753

5754 5755
    if (Gtid_log_event::peek(buf, event_len, checksum_alg,
                             &event_gtid.domain_id, &event_gtid.server_id,
5756
                             &event_gtid.seq_no, &gtid_flag,
5757
                             rli->relay_log.description_event_for_queue))
5758 5759 5760 5761 5762
    {
      error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
      goto err;
    }
    got_gtid_event= true;
unknown's avatar
unknown committed
5763 5764 5765 5766 5767 5768 5769 5770 5771 5772 5773 5774 5775 5776 5777
    if (mi->using_gtid == Master_info::USE_GTID_NO)
      goto default_action;
    if (unlikely(!mi->gtid_event_seen))
    {
      mi->gtid_event_seen= true;
      if (mi->gtid_reconnect_event_skip_count)
      {
        /*
          If we are reconnecting, and we need to skip a partial event group
          already queued to the relay log before the reconnect, then we check
          that we actually get the same event group (same GTID) as before, so
          we do not end up with half of one group and half another.

          The only way we should be able to receive a different GTID than what
          we expect is if the binlog on the master (or more likely the whole
5778
          master server) was replaced with a different one, on the same IP
unknown's avatar
unknown committed
5779 5780 5781 5782
          address, _and_ the new master happens to have domains in a different
          order so we get the GTID from a different domain first. Still, it is
          best to protect against this case.
        */
5783 5784 5785
        if (event_gtid.domain_id != mi->last_queued_gtid.domain_id ||
            event_gtid.server_id != mi->last_queued_gtid.server_id ||
            event_gtid.seq_no != mi->last_queued_gtid.seq_no)
unknown's avatar
unknown committed
5786 5787 5788 5789 5790 5791 5792 5793 5794
        {
          bool first;
          error= ER_SLAVE_UNEXPECTED_MASTER_SWITCH;
          error_msg.append(STRING_WITH_LEN("Expected: "));
          first= true;
          rpl_slave_state_tostring_helper(&error_msg, &mi->last_queued_gtid,
                                          &first);
          error_msg.append(STRING_WITH_LEN(", received: "));
          first= true;
5795
          rpl_slave_state_tostring_helper(&error_msg, &event_gtid, &first);
unknown's avatar
unknown committed
5796 5797 5798 5799 5800 5801 5802 5803 5804 5805 5806 5807 5808
          goto err;
        }
      }
    }

    if (unlikely(mi->gtid_reconnect_event_skip_count))
    {
      goto default_action;
    }

    /*
      We have successfully queued to relay log everything before this GTID, so
      in case of reconnect we can start from after any previous GTID.
5809 5810 5811
      (Normally we would have updated gtid_current_pos earlier at the end of
      the previous event group, but better leave an extra check here for
      safety).
unknown's avatar
unknown committed
5812 5813 5814 5815 5816 5817
    */
    if (mi->events_queued_since_last_gtid)
    {
      mi->gtid_current_pos.update(&mi->last_queued_gtid);
      mi->events_queued_since_last_gtid= 0;
    }
5818
    mi->last_queued_gtid= event_gtid;
5819 5820
    mi->last_queued_gtid_standalone=
      (gtid_flag & Gtid_log_event::FL_STANDALONE) != 0;
unknown's avatar
unknown committed
5821
    ++mi->events_queued_since_last_gtid;
5822
    inc_pos= event_len;
unknown's avatar
unknown committed
5823 5824 5825
  }
  break;

5826 5827 5828 5829 5830 5831 5832 5833 5834 5835 5836 5837
#ifndef DBUG_OFF
  case XID_EVENT:
    DBUG_EXECUTE_IF("slave_discard_xid_for_gtid_0_x_1000",
    {
      /* Inject an event group that is missing its XID commit event. */
      if (mi->last_queued_gtid.domain_id == 0 &&
          mi->last_queued_gtid.seq_no == 1000)
        goto skip_relay_logging;
    });
    /* Fall through to default case ... */
#endif

5838
  default:
unknown's avatar
unknown committed
5839
  default_action:
unknown's avatar
unknown committed
5840 5841 5842 5843 5844 5845 5846 5847 5848 5849 5850
    if (mi->using_gtid != Master_info::USE_GTID_NO && mi->gtid_event_seen)
    {
      if (unlikely(mi->gtid_reconnect_event_skip_count))
      {
        --mi->gtid_reconnect_event_skip_count;
        gtid_skip_enqueue= true;
      }
      else if (mi->events_queued_since_last_gtid)
        ++mi->events_queued_since_last_gtid;
    }

unknown's avatar
unknown committed
5851
    inc_pos= event_len;
5852 5853
    break;
  }
5854

5855
  /*
5856
    If we filter events master-side (eg. @@skip_replication), we will see holes
5857 5858 5859 5860 5861 5862 5863 5864 5865
    in the event positions from the master. If we see such a hole, adjust
    mi->master_log_pos accordingly so we maintain the correct position (for
    reconnect, MASTER_POS_WAIT(), etc.)
  */
  if (inc_pos > 0 &&
      event_len >= LOG_POS_OFFSET+4 &&
      (event_pos= uint4korr(buf+LOG_POS_OFFSET)) > mi->master_log_pos + inc_pos)
  {
    inc_pos= event_pos - mi->master_log_pos;
Sergei Golubchik's avatar
Sergei Golubchik committed
5866
    DBUG_PRINT("info", ("Adjust master_log_pos %llu->%llu to account for "
5867
                        "master-side filtering",
Sergei Golubchik's avatar
Sergei Golubchik committed
5868
                        mi->master_log_pos + inc_pos, event_pos));
5869 5870
  }

unknown's avatar
unknown committed
5871 5872
  /*
     If this event is originating from this server, don't queue it.
5873
     We don't check this for 3.23 events because it's simpler like this; 3.23
unknown's avatar
unknown committed
5874 5875
     will be filtered anyway by the SQL slave thread which also tests the
     server id (we must also keep this test in the SQL thread, in case somebody
5876 5877 5878 5879 5880 5881 5882 5883
     upgrades a 4.0 slave which has a not-filtered relay log).

     ANY event coming from ourselves can be ignored: it is obvious for queries;
     for STOP_EVENT/ROTATE_EVENT/START_EVENT: these cannot come from ourselves
     (--log-slave-updates would not log that) unless this slave is also its
     direct master (an unsupported, useless setup!).
  */

Marc Alff's avatar
Marc Alff committed
5884
  mysql_mutex_lock(log_lock);
Andrei Elkin's avatar
Andrei Elkin committed
5885
  s_id= uint4korr(buf + SERVER_ID_OFFSET);
5886 5887 5888 5889 5890 5891 5892 5893
  /*
    Write the event to the relay log, unless we reconnected in the middle
    of an event group and now need to skip the initial part of the group that
    we already wrote before reconnecting.
  */
  if (unlikely(gtid_skip_enqueue))
  {
    mi->master_log_pos+= inc_pos;
5894 5895 5896 5897 5898 5899 5900 5901 5902 5903 5904 5905 5906 5907 5908 5909 5910 5911 5912 5913
    if ((uchar)buf[EVENT_TYPE_OFFSET] == FORMAT_DESCRIPTION_EVENT &&
        s_id == mi->master_id)
    {
      /*
        If we write this master's description event in the middle of an event
        group due to GTID reconnect, SQL thread will think that master crashed
        in the middle of the group and roll back the first half, so we must not.

        But we still have to write an artificial copy of the masters description
        event, to override the initial slave-version description event so that
        SQL thread has the right information for parsing the events it reads.
      */
      rli->relay_log.description_event_for_queue->created= 0;
      rli->relay_log.description_event_for_queue->set_artificial_event();
      if (rli->relay_log.append_no_lock
          (rli->relay_log.description_event_for_queue))
        error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
      else
        rli->relay_log.harvest_bytes_written(&rli->log_space_total);
    }
5914 5915 5916 5917 5918 5919 5920 5921 5922 5923 5924 5925 5926
    else if (mi->gtid_reconnect_event_skip_count == 0)
    {
      /*
        Add a fake rotate event so that SQL thread can see the old-style
        position where we re-connected in the middle of a GTID event group.
      */
      Rotate_log_event fake_rev(mi->master_log_name, 0, mi->master_log_pos, 0);
      fake_rev.server_id= mi->master_id;
      if (rli->relay_log.append_no_lock(&fake_rev))
        error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
      else
        rli->relay_log.harvest_bytes_written(&rli->log_space_total);
    }
5927 5928
  }
  else
5929 5930
  if ((s_id == global_system_variables.server_id &&
       !mi->rli.replicate_same_server_id) ||
Andrei Elkin's avatar
Andrei Elkin committed
5931 5932 5933 5934 5935 5936 5937 5938 5939 5940
      /*
        the following conjunction deals with IGNORE_SERVER_IDS, if set
        If the master is on the ignore list, execution of
        format description log events and rotate events is necessary.
      */
      (mi->ignore_server_ids.elements > 0 &&
       mi->shall_ignore_server_id(s_id) &&
       /* everything is filtered out from non-master */
       (s_id != mi->master_id ||
        /* for the master meta information is necessary */
5941 5942
        (buf[EVENT_TYPE_OFFSET] != FORMAT_DESCRIPTION_EVENT &&
         buf[EVENT_TYPE_OFFSET] != ROTATE_EVENT))))
5943
  {
5944 5945
    /*
      Do not write it to the relay log.
5946 5947 5948 5949 5950 5951 5952 5953
      a) We still want to increment mi->master_log_pos, so that we won't
      re-read this event from the master if the slave IO thread is now
      stopped/restarted (more efficient if the events we are ignoring are big
      LOAD DATA INFILE).
      b) We want to record that we are skipping events, for the information of
      the slave SQL thread, otherwise that thread may let
      rli->group_relay_log_pos stay too small if the last binlog's event is
      ignored.
5954 5955 5956
      But events which were generated by this slave and which do not exist in
      the master's binlog (i.e. Format_desc, Rotate & Stop) should not increment
      mi->master_log_pos.
Andrei Elkin's avatar
Andrei Elkin committed
5957 5958 5959
      If the event is originated remotely and is being filtered out by
      IGNORE_SERVER_IDS it increments mi->master_log_pos
      as well as rli->group_relay_log_pos.
5960
    */
5961 5962
    if (!(s_id == global_system_variables.server_id &&
          !mi->rli.replicate_same_server_id) ||
5963 5964 5965
        (buf[EVENT_TYPE_OFFSET] != FORMAT_DESCRIPTION_EVENT &&
         buf[EVENT_TYPE_OFFSET] != ROTATE_EVENT &&
         buf[EVENT_TYPE_OFFSET] != STOP_EVENT))
unknown's avatar
unknown committed
5966
    {
5967
      mi->master_log_pos+= inc_pos;
unknown's avatar
unknown committed
5968 5969 5970
      memcpy(rli->ign_master_log_name_end, mi->master_log_name, FN_REFLEN);
      DBUG_ASSERT(rli->ign_master_log_name_end[0]);
      rli->ign_master_log_pos_end= mi->master_log_pos;
5971 5972
      if (got_gtid_event)
        rli->ign_gtids.update(&event_gtid);
unknown's avatar
unknown committed
5973
    }
5974
    rli->relay_log.signal_update(); // the slave SQL thread needs to re-check
Andrei Elkin's avatar
Andrei Elkin committed
5975 5976
    DBUG_PRINT("info", ("master_log_pos: %lu, event originating from %u server, ignored",
                        (ulong) mi->master_log_pos, uint4korr(buf + SERVER_ID_OFFSET)));
unknown's avatar
unknown committed
5977
  }
unknown's avatar
unknown committed
5978 5979
  else
  {
5980
    if (likely(!(rli->relay_log.appendv(buf,event_len,0))))
5981 5982
    {
      mi->master_log_pos+= inc_pos;
unknown's avatar
unknown committed
5983
      DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
5984 5985
      rli->relay_log.harvest_bytes_written(&rli->log_space_total);
    }
5986
    else
Andrei Elkin's avatar
Andrei Elkin committed
5987 5988 5989
    {
      error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
    }
5990
    rli->ign_master_log_name_end[0]= 0; // last event is not ignored
5991 5992
    if (got_gtid_event)
      rli->ign_gtids.remove_if_present(&event_gtid);
5993 5994
    if (save_buf != NULL)
      buf= save_buf;
unknown's avatar
unknown committed
5995
  }
Marc Alff's avatar
Marc Alff committed
5996
  mysql_mutex_unlock(log_lock);
5997

5998 5999 6000 6001 6002 6003 6004 6005 6006 6007 6008 6009 6010 6011 6012 6013 6014 6015 6016 6017
  if (!error &&
      mi->using_gtid != Master_info::USE_GTID_NO &&
      mi->events_queued_since_last_gtid > 0 &&
      ( (mi->last_queued_gtid_standalone &&
         !Log_event::is_part_of_group((Log_event_type)(uchar)
                                      buf[EVENT_TYPE_OFFSET])) ||
        (!mi->last_queued_gtid_standalone &&
         ((uchar)buf[EVENT_TYPE_OFFSET] == XID_EVENT ||
          ((uchar)buf[EVENT_TYPE_OFFSET] == QUERY_EVENT &&
           Query_log_event::peek_is_commit_rollback(buf, event_len,
                                                    checksum_alg))))))
    {
      /*
        The whole of the current event group is queued. So in case of
        reconnect we can start from after the current GTID.
      */
      mi->gtid_current_pos.update(&mi->last_queued_gtid);
      mi->events_queued_since_last_gtid= 0;
    }

Andrei Elkin's avatar
Andrei Elkin committed
6018 6019
skip_relay_logging:
  
6020
err:
6021
  if (unlock_data_lock)
Sergei Golubchik's avatar
Sergei Golubchik committed
6022
    mysql_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
6023
  DBUG_PRINT("info", ("error: %d", error));
Andrei Elkin's avatar
Andrei Elkin committed
6024
  if (error)
6025
    mi->report(ERROR_LEVEL, error, NULL, ER(error),
Andrei Elkin's avatar
Andrei Elkin committed
6026 6027 6028
               (error == ER_SLAVE_RELAY_LOG_WRITE_FAILURE)?
               "could not queue event from master" :
               error_msg.ptr());
unknown's avatar
unknown committed
6029
  DBUG_RETURN(error);
6030 6031
}

6032

6033
void end_relay_log_info(Relay_log_info* rli)
6034
{
6035 6036
  DBUG_ENTER("end_relay_log_info");

6037
  if (!rli->inited)
6038
    DBUG_VOID_RETURN;
6039
  if (rli->info_fd >= 0)
unknown's avatar
unknown committed
6040 6041
  {
    end_io_cache(&rli->info_file);
Marc Alff's avatar
Marc Alff committed
6042
    mysql_file_close(rli->info_fd, MYF(MY_WME));
unknown's avatar
unknown committed
6043 6044
    rli->info_fd = -1;
  }
6045
  if (rli->cur_log_fd >= 0)
unknown's avatar
unknown committed
6046 6047
  {
    end_io_cache(&rli->cache_buf);
Marc Alff's avatar
Marc Alff committed
6048
    mysql_file_close(rli->cur_log_fd, MYF(MY_WME));
unknown's avatar
unknown committed
6049 6050
    rli->cur_log_fd = -1;
  }
6051
  rli->inited = 0;
6052
  rli->relay_log.close(LOG_CLOSE_INDEX | LOG_CLOSE_STOP_EVENT);
6053
  rli->relay_log.harvest_bytes_written(&rli->log_space_total);
unknown's avatar
unknown committed
6054 6055 6056 6057 6058 6059
  /*
    Delete the slave's temporary tables from memory.
    In the future there will be other actions than this, to ensure persistance
    of slave's temp tables after shutdown.
  */
  rli->close_temporary_tables();
6060
  DBUG_VOID_RETURN;
6061 6062
}

6063 6064 6065 6066 6067 6068 6069 6070 6071 6072 6073 6074 6075 6076 6077 6078 6079 6080 6081

/**
  Hook to detach the active VIO before closing a connection handle.

  The client API might close the connection (and associated data)
  in case it encounters a unrecoverable (network) error. This hook
  is called from the client code before the VIO handle is deleted
  allows the thread to detach the active vio so it does not point
  to freed memory.

  Other calls to THD::clear_active_vio throughout this module are
  redundant due to the hook but are left in place for illustrative
  purposes.
*/

extern "C" void slave_io_thread_detach_vio()
{
#ifdef SIGNAL_WITH_VIO_CLOSE
  THD *thd= current_thd;
6082
  if (thd && thd->slave_thread)
6083 6084 6085 6086 6087
    thd->clear_active_vio();
#endif
}


unknown's avatar
unknown committed
6088 6089
/*
  Try to connect until successful or slave killed
6090

unknown's avatar
unknown committed
6091 6092
  SYNPOSIS
    safe_connect()
unknown's avatar
unknown committed
6093 6094 6095
    thd                 Thread handler for slave
    mysql               MySQL connection handle
    mi                  Replication handle
6096

unknown's avatar
unknown committed
6097
  RETURN
unknown's avatar
unknown committed
6098 6099
    0   ok
    #   Error
unknown's avatar
unknown committed
6100
*/
6101

6102
static int safe_connect(THD* thd, MYSQL* mysql, Master_info* mi)
unknown's avatar
unknown committed
6103
{
6104 6105 6106
  DBUG_ENTER("safe_connect");

  DBUG_RETURN(connect_to_master(thd, mysql, mi, 0, 0));
unknown's avatar
unknown committed
6107 6108
}

6109

6110
/*
unknown's avatar
unknown committed
6111 6112
  SYNPOSIS
    connect_to_master()
unknown's avatar
unknown committed
6113

unknown's avatar
unknown committed
6114 6115 6116
  IMPLEMENTATION
    Try to connect until successful or slave killed or we have retried
    master_retry_count times
6117
*/
unknown's avatar
unknown committed
6118

6119
static int connect_to_master(THD* thd, MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
6120
                             bool reconnect, bool suppress_warnings)
unknown's avatar
unknown committed
6121
{
6122
  int slave_was_killed;
unknown's avatar
unknown committed
6123
  int last_errno= -2;                           // impossible error
6124
  ulong err_count=0;
unknown's avatar
unknown committed
6125
  char llbuff[22];
6126
  my_bool my_true= 1;
6127
  DBUG_ENTER("connect_to_master");
6128
  set_slave_max_allowed_packet(thd, mysql);
unknown's avatar
unknown committed
6129
#ifndef DBUG_OFF
6130
  mi->events_till_disconnect = disconnect_slave_event_count;
unknown's avatar
unknown committed
6131
#endif
6132
  ulong client_flag= 0;
6133
  if (opt_slave_compressed_protocol)
unknown's avatar
unknown committed
6134
    client_flag=CLIENT_COMPRESS;                /* We will use compression */
6135

6136 6137
  mysql_options(mysql, MYSQL_OPT_CONNECT_TIMEOUT, (char *) &slave_net_timeout);
  mysql_options(mysql, MYSQL_OPT_READ_TIMEOUT, (char *) &slave_net_timeout);
6138 6139
  mysql_options(mysql, MYSQL_OPT_USE_THREAD_SPECIFIC_MEMORY,
                (char*) &my_true);
unknown's avatar
unknown committed
6140

unknown's avatar
unknown committed
6141 6142
#ifdef HAVE_OPENSSL
  if (mi->ssl)
6143
  {
unknown's avatar
unknown committed
6144
    mysql_ssl_set(mysql,
unknown's avatar
unknown committed
6145
                  mi->ssl_key[0]?mi->ssl_key:0,
unknown's avatar
unknown committed
6146
                  mi->ssl_cert[0]?mi->ssl_cert:0,
unknown's avatar
unknown committed
6147 6148 6149
                  mi->ssl_ca[0]?mi->ssl_ca:0,
                  mi->ssl_capath[0]?mi->ssl_capath:0,
                  mi->ssl_cipher[0]?mi->ssl_cipher:0);
Michael Widenius's avatar
Michael Widenius committed
6150 6151 6152 6153
    mysql_options(mysql, MYSQL_OPT_SSL_VERIFY_SERVER_CERT,
                  &mi->ssl_verify_server_cert);
    mysql_options(mysql, MYSQL_OPT_SSL_CRLPATH, 
                  mi->ssl_crlpath[0] ? mi->ssl_crlpath : 0);
6154 6155 6156
    mysql_options(mysql, MYSQL_OPT_SSL_VERIFY_SERVER_CERT,
                  &mi->ssl_verify_server_cert);
  }
unknown's avatar
unknown committed
6157 6158
#endif

6159 6160 6161 6162 6163 6164 6165 6166 6167 6168 6169 6170 6171 6172 6173 6174 6175
  /*
    If server's default charset is not supported (like utf16, utf32) as client
    charset, then set client charset to 'latin1' (default client charset).
  */
  if (is_supported_parser_charset(default_charset_info))
    mysql_options(mysql, MYSQL_SET_CHARSET_NAME, default_charset_info->csname);
  else
  {
    sql_print_information("'%s' can not be used as client character set. "
                          "'%s' will be used as default client character set "
                          "while connecting to master.",
                          default_charset_info->csname,
                          default_client_charset_info->csname);
    mysql_options(mysql, MYSQL_SET_CHARSET_NAME,
                  default_client_charset_info->csname);
  }

6176 6177 6178
  /* This one is not strictly needed but we have it here for completeness */
  mysql_options(mysql, MYSQL_SET_CHARSET_DIR, (char *) charsets_dir);

6179
  /* Set MYSQL_PLUGIN_DIR in case master asks for an external authentication plugin */
6180 6181
  if (opt_plugin_dir_ptr && *opt_plugin_dir_ptr)
    mysql_options(mysql, MYSQL_PLUGIN_DIR, opt_plugin_dir_ptr);
6182

6183 6184 6185
  /* we disallow empty users */
  if (mi->user == NULL || mi->user[0] == 0)
  {
6186
    mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, NULL,
6187 6188 6189 6190 6191 6192
               ER(ER_SLAVE_FATAL_ERROR),
               "Invalid (empty) username when attempting to "
               "connect to the master server. Connection attempt "
               "terminated.");
    DBUG_RETURN(1);
  }
Michael Widenius's avatar
Michael Widenius committed
6193
  while (!(slave_was_killed = io_slave_killed(mi)) &&
unknown's avatar
unknown committed
6194 6195 6196
         (reconnect ? mysql_reconnect(mysql) != 0 :
          mysql_real_connect(mysql, mi->host, mi->user, mi->password, 0,
                             mi->port, 0, client_flag) == 0))
unknown's avatar
unknown committed
6197
  {
6198
    /* Don't repeat last error */
unknown's avatar
SCRUM  
unknown committed
6199
    if ((int)mysql_errno(mysql) != last_errno)
6200
    {
unknown's avatar
SCRUM  
unknown committed
6201
      last_errno=mysql_errno(mysql);
unknown's avatar
unknown committed
6202
      suppress_warnings= 0;
6203
      mi->report(ERROR_LEVEL, last_errno, NULL,
6204
                 "error %s to master '%s@%s:%d'"
6205
                 " - retry-time: %d  retries: %lu  message: %s",
6206 6207
                 (reconnect ? "reconnecting" : "connecting"),
                 mi->user, mi->host, mi->port,
6208 6209
                 mi->connect_retry, master_retry_count,
                 mysql_error(mysql));
6210
    }
unknown's avatar
unknown committed
6211 6212 6213
    /*
      By default we try forever. The reason is that failure will trigger
      master election, so if the user did not set master_retry_count we
6214
      do not want to have election triggered on the first failure to
unknown's avatar
unknown committed
6215
      connect
6216
    */
6217
    if (++err_count == master_retry_count)
6218 6219
    {
      slave_was_killed=1;
unknown's avatar
unknown committed
6220 6221
      if (reconnect)
        change_rpl_status(RPL_ACTIVE_SLAVE,RPL_LOST_SOLDIER);
6222 6223
      break;
    }
6224
    slave_sleep(thd,mi->connect_retry,io_slave_killed, mi);
unknown's avatar
unknown committed
6225
  }
6226

6227 6228
  if (!slave_was_killed)
  {
6229
    mi->clear_error(); // clear possible left over reconnect error
unknown's avatar
unknown committed
6230
    if (reconnect)
unknown's avatar
unknown committed
6231
    {
6232
      if (!suppress_warnings && global_system_variables.log_warnings)
unknown's avatar
unknown committed
6233
        sql_print_information("Slave: connected to master '%s@%s:%d',\
6234
replication resumed in log '%s' at position %s", mi->user,
unknown's avatar
unknown committed
6235 6236 6237
                        mi->host, mi->port,
                        IO_RPL_LOG_NAME,
                        llstr(mi->master_log_pos,llbuff));
unknown's avatar
unknown committed
6238
    }
unknown's avatar
unknown committed
6239 6240 6241
    else
    {
      change_rpl_status(RPL_IDLE_SLAVE,RPL_ACTIVE_SLAVE);
6242 6243
      general_log_print(thd, COM_CONNECT_OUT, "%s@%s:%d",
                        mi->user, mi->host, mi->port);
unknown's avatar
unknown committed
6244
    }
6245
#ifdef SIGNAL_WITH_VIO_CLOSE
6246
    thd->set_active_vio(mysql->net.vio);
unknown's avatar
unknown committed
6247
#endif
6248
  }
6249
  mysql->reconnect= 1;
6250 6251
  DBUG_PRINT("exit",("slave_was_killed: %d", slave_was_killed));
  DBUG_RETURN(slave_was_killed);
unknown's avatar
unknown committed
6252 6253
}

6254

unknown's avatar
unknown committed
6255
/*
6256
  safe_reconnect()
unknown's avatar
unknown committed
6257

unknown's avatar
unknown committed
6258 6259 6260
  IMPLEMENTATION
    Try to connect until successful or slave killed or we have retried
    master_retry_count times
unknown's avatar
unknown committed
6261 6262
*/

6263
static int safe_reconnect(THD* thd, MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
6264
                          bool suppress_warnings)
unknown's avatar
unknown committed
6265
{
unknown's avatar
unknown committed
6266 6267
  DBUG_ENTER("safe_reconnect");
  DBUG_RETURN(connect_to_master(thd, mysql, mi, 1, suppress_warnings));
unknown's avatar
unknown committed
6268 6269
}

unknown's avatar
unknown committed
6270

Michael Widenius's avatar
Michael Widenius committed
6271
#ifdef NOT_USED
He Zhenxing's avatar
He Zhenxing committed
6272 6273 6274
MYSQL *rpl_connect_master(MYSQL *mysql)
{
  Master_info *mi= my_pthread_getspecific_ptr(Master_info*, RPL_MASTER_INFO);
6275 6276
  bool allocated= false;
  my_bool my_true= 1;
Michael Widenius's avatar
Michael Widenius committed
6277
  THD *thd;
6278

He Zhenxing's avatar
He Zhenxing committed
6279 6280 6281 6282 6283
  if (!mi)
  {
    sql_print_error("'rpl_connect_master' must be called in slave I/O thread context.");
    return NULL;
  }
Michael Widenius's avatar
Michael Widenius committed
6284
  thd= mi->io_thd;
He Zhenxing's avatar
He Zhenxing committed
6285 6286 6287 6288 6289 6290 6291 6292 6293 6294 6295 6296 6297 6298 6299 6300 6301 6302 6303
  if (!mysql)
  {
    if(!(mysql= mysql_init(NULL)))
    {
      sql_print_error("rpl_connect_master: failed in mysql_init()");
      return NULL;
    }
    allocated= true;
  }

  /*
    XXX: copied from connect_to_master, this function should not
    change the slave status, so we cannot use connect_to_master
    directly
    
    TODO: make this part a seperate function to eliminate duplication
  */
  mysql_options(mysql, MYSQL_OPT_CONNECT_TIMEOUT, (char *) &slave_net_timeout);
  mysql_options(mysql, MYSQL_OPT_READ_TIMEOUT, (char *) &slave_net_timeout);
6304 6305
  mysql_options(mysql, MYSQL_OPT_USE_THREAD_SPECIFIC_MEMORY,
                (char*) &my_true);
He Zhenxing's avatar
He Zhenxing committed
6306 6307 6308 6309 6310 6311 6312 6313 6314 6315 6316 6317 6318 6319 6320 6321 6322 6323 6324

#ifdef HAVE_OPENSSL
  if (mi->ssl)
  {
    mysql_ssl_set(mysql,
                  mi->ssl_key[0]?mi->ssl_key:0,
                  mi->ssl_cert[0]?mi->ssl_cert:0,
                  mi->ssl_ca[0]?mi->ssl_ca:0,
                  mi->ssl_capath[0]?mi->ssl_capath:0,
                  mi->ssl_cipher[0]?mi->ssl_cipher:0);
    mysql_options(mysql, MYSQL_OPT_SSL_VERIFY_SERVER_CERT,
                  &mi->ssl_verify_server_cert);
  }
#endif

  mysql_options(mysql, MYSQL_SET_CHARSET_NAME, default_charset_info->csname);
  /* This one is not strictly needed but we have it here for completeness */
  mysql_options(mysql, MYSQL_SET_CHARSET_DIR, (char *) charsets_dir);

6325 6326
  if (mi->user == NULL
      || mi->user[0] == 0
Michael Widenius's avatar
Michael Widenius committed
6327
      || io_slave_killed( mi)
He Zhenxing's avatar
He Zhenxing committed
6328 6329 6330
      || !mysql_real_connect(mysql, mi->host, mi->user, mi->password, 0,
                             mi->port, 0, 0))
  {
Michael Widenius's avatar
Michael Widenius committed
6331
    if (!io_slave_killed( mi))
He Zhenxing's avatar
He Zhenxing committed
6332 6333 6334 6335 6336 6337 6338 6339 6340
      sql_print_error("rpl_connect_master: error connecting to master: %s (server_error: %d)",
                      mysql_error(mysql), mysql_errno(mysql));
    
    if (allocated)
      mysql_close(mysql);                       // this will free the object
    return NULL;
  }
  return mysql;
}
Michael Widenius's avatar
Michael Widenius committed
6341
#endif
He Zhenxing's avatar
He Zhenxing committed
6342

6343 6344 6345 6346 6347 6348
/*
  Store the file and position where the execute-slave thread are in the
  relay log.

  SYNOPSIS
    flush_relay_log_info()
unknown's avatar
unknown committed
6349
    rli                 Relay log information
6350 6351

  NOTES
6352 6353 6354
    - As this is only called by the slave thread or on STOP SLAVE, with the
      log_lock grabbed and the slave thread stopped, we don't need to have 
      a lock here.
6355 6356 6357 6358 6359 6360 6361 6362 6363 6364 6365 6366 6367 6368
    - If there is an active transaction, then we don't update the position
      in the relay log.  This is to ensure that we re-execute statements
      if we die in the middle of an transaction that was rolled back.
    - As a transaction never spans binary logs, we don't have to handle the
      case where we do a relay-log-rotation in the middle of the transaction.
      If this would not be the case, we would have to ensure that we
      don't delete the relay log file where the transaction started when
      we switch to a new relay log file.

  TODO
    - Change the log file information to a binary format to avoid calling
      longlong2str.

  RETURN VALUES
unknown's avatar
unknown committed
6369 6370
    0   ok
    1   write error
6371 6372
*/

6373
bool flush_relay_log_info(Relay_log_info* rli)
6374
{
6375
  bool error=0;
6376
  DBUG_ENTER("flush_relay_log_info");
6377 6378

  if (unlikely(rli->no_storage))
6379
    DBUG_RETURN(0);
6380

6381 6382 6383
  IO_CACHE *file = &rli->info_file;
  char buff[FN_REFLEN*2+22*2+4], *pos;

6384
  my_b_seek(file, 0L);
6385
  pos=strmov(buff, rli->group_relay_log_name);
6386
  *pos++='\n';
6387
  pos= longlong10_to_str(rli->group_relay_log_pos, pos, 10);
6388
  *pos++='\n';
6389
  pos=strmov(pos, rli->group_master_log_name);
6390
  *pos++='\n';
6391
  pos=longlong10_to_str(rli->group_master_log_pos, pos, 10);
6392
  *pos='\n';
6393
  if (my_b_write(file, (uchar*) buff, (size_t) (pos-buff)+1))
6394 6395 6396
    error=1;
  if (flush_io_cache(file))
    error=1;
6397 6398 6399 6400 6401 6402 6403 6404
  if (sync_relayloginfo_period &&
      !error &&
      ++(rli->sync_counter) >= sync_relayloginfo_period)
  {
    if (my_sync(rli->info_fd, MYF(MY_WME)))
      error=1;
    rli->sync_counter= 0;
  }
6405 6406 6407 6408
  /* 
    Flushing the relay log is done by the slave I/O thread 
    or by the user on STOP SLAVE. 
   */
6409
  DBUG_RETURN(error);
6410 6411
}

6412

unknown's avatar
unknown committed
6413
/*
6414
  Called when we notice that the current "hot" log got rotated under our feet.
unknown's avatar
unknown committed
6415 6416
*/

6417
static IO_CACHE *reopen_relay_log(Relay_log_info *rli, const char **errmsg)
6418
{
6419
  DBUG_ENTER("reopen_relay_log");
6420 6421
  DBUG_ASSERT(rli->cur_log != &rli->cache_buf);
  DBUG_ASSERT(rli->cur_log_fd == -1);
unknown's avatar
unknown committed
6422 6423

  IO_CACHE *cur_log = rli->cur_log=&rli->cache_buf;
6424
  if ((rli->cur_log_fd=open_binlog(cur_log,rli->event_relay_log_name,
unknown's avatar
unknown committed
6425
                                   errmsg)) <0)
unknown's avatar
unknown committed
6426
    DBUG_RETURN(0);
6427 6428
  /*
    We want to start exactly where we was before:
unknown's avatar
unknown committed
6429 6430
    relay_log_pos       Current log pos
    pending             Number of bytes already processed from the event
6431
  */
6432
  rli->event_relay_log_pos= MY_MAX(rli->event_relay_log_pos, BIN_LOG_HEADER_SIZE);
6433
  my_b_seek(cur_log,rli->event_relay_log_pos);
unknown's avatar
unknown committed
6434
  DBUG_RETURN(cur_log);
6435 6436
}

unknown's avatar
unknown committed
6437

6438 6439 6440 6441 6442 6443 6444 6445 6446
/**
  Reads next event from the relay log.  Should be called from the
  slave IO thread.

  @param rli Relay_log_info structure for the slave IO thread.

  @return The event read, or NULL on error.  If an error occurs, the
  error is reported through the sql_print_information() or
  sql_print_error() functions.
unknown's avatar
unknown committed
6447 6448

  The size of the read event (in bytes) is returned in *event_size.
6449
*/
unknown's avatar
unknown committed
6450
static Log_event* next_event(rpl_group_info *rgi, ulonglong *event_size)
6451 6452
{
  Log_event* ev;
Michael Widenius's avatar
Michael Widenius committed
6453
  Relay_log_info *rli= rgi->rli;
6454
  IO_CACHE* cur_log = rli->cur_log;
Marc Alff's avatar
Marc Alff committed
6455
  mysql_mutex_t *log_lock = rli->relay_log.get_log_lock();
6456
  const char* errmsg=0;
unknown's avatar
unknown committed
6457
  DBUG_ENTER("next_event");
6458

6459
  DBUG_ASSERT(rgi->thd != 0 && rgi->thd == rli->sql_driver_thd);
unknown's avatar
unknown committed
6460
  *event_size= 0;
6461

6462 6463 6464 6465 6466
#ifndef DBUG_OFF
  if (abort_slave_event_count && !rli->events_till_abort--)
    DBUG_RETURN(0);
#endif

unknown's avatar
unknown committed
6467 6468
  /*
    For most operations we need to protect rli members with data_lock,
6469 6470 6471
    so we assume calling function acquired this mutex for us and we will
    hold it for the most of the loop below However, we will release it
    whenever it is worth the hassle,  and in the cases when we go into a
Marc Alff's avatar
Marc Alff committed
6472
    mysql_cond_wait() with the non-data_lock mutex
unknown's avatar
unknown committed
6473
  */
Marc Alff's avatar
Marc Alff committed
6474
  mysql_mutex_assert_owner(&rli->data_lock);
unknown's avatar
unknown committed
6475

Michael Widenius's avatar
Michael Widenius committed
6476
  while (!sql_slave_killed(rgi))
unknown's avatar
unknown committed
6477 6478 6479
  {
    /*
      We can have two kinds of log reading:
unknown's avatar
unknown committed
6480 6481 6482 6483 6484 6485 6486 6487
      hot_log:
        rli->cur_log points at the IO_CACHE of relay_log, which
        is actively being updated by the I/O thread. We need to be careful
        in this case and make sure that we are not looking at a stale log that
        has already been rotated. If it has been, we reopen the log.

      The other case is much simpler:
        We just have a read only log that nobody else will be updating.
unknown's avatar
unknown committed
6488
    */
unknown's avatar
unknown committed
6489
    ulonglong old_pos;
6490 6491 6492 6493
    bool hot_log;
    if ((hot_log = (cur_log != &rli->cache_buf)))
    {
      DBUG_ASSERT(rli->cur_log_fd == -1); // foreign descriptor
Marc Alff's avatar
Marc Alff committed
6494
      mysql_mutex_lock(log_lock);
unknown's avatar
unknown committed
6495 6496

      /*
unknown's avatar
unknown committed
6497 6498
        Reading xxx_file_id is safe because the log will only
        be rotated when we hold relay_log.LOCK_log
unknown's avatar
unknown committed
6499
      */
unknown's avatar
unknown committed
6500
      if (rli->relay_log.get_open_count() != rli->cur_log_old_open_count)
6501
      {
unknown's avatar
unknown committed
6502 6503
        // The master has switched to a new log file; Reopen the old log file
        cur_log=reopen_relay_log(rli, &errmsg);
Marc Alff's avatar
Marc Alff committed
6504
        mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
6505 6506 6507
        if (!cur_log)                           // No more log files
          goto err;
        hot_log=0;                              // Using old binary log
6508 6509
      }
    }
6510 6511 6512 6513 6514 6515 6516
    /* 
      As there is no guarantee that the relay is open (for example, an I/O
      error during a write by the slave I/O thread may have closed it), we
      have to test it.
    */
    if (!my_b_inited(cur_log))
      goto err;
6517 6518
#ifndef DBUG_OFF
    {
6519
      /* This is an assertion which sometimes fails, let's try to track it */
6520
      char llbuf1[22], llbuf2[22];
6521
      DBUG_PRINT("info", ("my_b_tell(cur_log)=%s rli->event_relay_log_pos=%s",
6522
                          llstr(my_b_tell(cur_log),llbuf1),
6523 6524
                          llstr(rli->event_relay_log_pos,llbuf2)));
      DBUG_ASSERT(my_b_tell(cur_log) >= BIN_LOG_HEADER_SIZE);
6525 6526
      DBUG_ASSERT(opt_slave_parallel_threads > 0 ||
                  my_b_tell(cur_log) == rli->event_relay_log_pos);
6527 6528
    }
#endif
unknown's avatar
unknown committed
6529 6530
    /*
      Relay log is always in new format - if the master is 3.23, the
6531
      I/O thread will convert the format for us.
6532 6533
      A problem: the description event may be in a previous relay log. So if
      the slave has been shutdown meanwhile, we would have to look in old relay
6534 6535
      logs, which may even have been deleted. So we need to write this
      description event at the beginning of the relay log.
6536 6537
      When the relay log is created when the I/O thread starts, easy: the
      master will send the description event and we will queue it.
6538
      But if the relay log is created by new_file(): then the solution is:
6539
      MYSQL_BIN_LOG::open() will write the buffered description event.
unknown's avatar
unknown committed
6540
    */
unknown's avatar
unknown committed
6541
    old_pos= rli->event_relay_log_pos;
6542 6543 6544
    if ((ev= Log_event::read_log_event(cur_log,0,
                                       rli->relay_log.description_event_for_exec,
                                       opt_slave_sql_verify_checksum)))
6545

6546
    {
6547 6548 6549 6550 6551
      /*
        read it while we have a lock, to avoid a mutex lock in
        inc_event_relay_log_pos()
      */
      rli->future_event_relay_log_pos= my_b_tell(cur_log);
unknown's avatar
unknown committed
6552
      *event_size= rli->future_event_relay_log_pos - old_pos;
6553

6554
      if (hot_log)
Marc Alff's avatar
Marc Alff committed
6555
        mysql_mutex_unlock(log_lock);
6556
      rli->sql_thread_caught_up= false;
unknown's avatar
unknown committed
6557
      DBUG_RETURN(ev);
6558
    }
unknown's avatar
unknown committed
6559
    if (opt_reckless_slave)                     // For mysql-test
unknown's avatar
unknown committed
6560
      cur_log->error = 0;
unknown's avatar
unknown committed
6561
    if (cur_log->error < 0)
unknown's avatar
unknown committed
6562 6563
    {
      errmsg = "slave SQL thread aborted because of I/O error";
unknown's avatar
unknown committed
6564
      if (hot_log)
Marc Alff's avatar
Marc Alff committed
6565
        mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
6566 6567
      goto err;
    }
6568 6569
    if (!cur_log->error) /* EOF */
    {
unknown's avatar
unknown committed
6570
      /*
unknown's avatar
unknown committed
6571 6572 6573
        On a hot log, EOF means that there are no more updates to
        process and we must block until I/O thread adds some and
        signals us to continue
unknown's avatar
unknown committed
6574
      */
6575 6576
      if (hot_log)
      {
6577 6578 6579 6580 6581 6582 6583 6584 6585 6586
        /*
          We say in Seconds_Behind_Master that we have "caught up". Note that
          for example if network link is broken but I/O slave thread hasn't
          noticed it (slave_net_timeout not elapsed), then we'll say "caught
          up" whereas we're not really caught up. Fixing that would require
          internally cutting timeout in smaller pieces in network read, no
          thanks. Another example: SQL has caught up on I/O, now I/O has read
          a new event and is queuing it; the false "0" will exist until SQL
          finishes executing the new event; it will be look abnormal only if
          the events have old timestamps (then you get "many", 0, "many").
6587 6588 6589 6590 6591 6592 6593

          Transient phases like this can be fixed with implemeting
          Heartbeat event which provides the slave the status of the
          master at time the master does not have any new update to send.
          Seconds_Behind_Master would be zero only when master has no
          more updates in binlog for slave. The heartbeat can be sent
          in a (small) fraction of slave_net_timeout. Until it's done
6594 6595
          rli->sql_thread_caught_up is temporarely (for time of waiting for
          the following event) set whenever EOF is reached.
6596
        */
6597
        rli->sql_thread_caught_up= true;
6598

unknown's avatar
unknown committed
6599
        DBUG_ASSERT(rli->relay_log.get_open_count() ==
unknown's avatar
unknown committed
6600
                    rli->cur_log_old_open_count);
6601 6602 6603 6604 6605

        if (rli->ign_master_log_name_end[0])
        {
          /* We generate and return a Rotate, to make our positions advance */
          DBUG_PRINT("info",("seeing an ignored end segment"));
6606
          ev= new Rotate_log_event(rli->ign_master_log_name_end,
6607
                                   0, rli->ign_master_log_pos_end,
unknown's avatar
unknown committed
6608
                                   Rotate_log_event::DUP_NAME);
6609
          rli->ign_master_log_name_end[0]= 0;
Marc Alff's avatar
Marc Alff committed
6610
          mysql_mutex_unlock(log_lock);
6611 6612 6613 6614 6615 6616 6617 6618 6619 6620
          if (unlikely(!ev))
          {
            errmsg= "Slave SQL thread failed to create a Rotate event "
              "(out of memory?), SHOW SLAVE STATUS may be inaccurate";
            goto err;
          }
          ev->server_id= 0; // don't be ignored by slave SQL thread
          DBUG_RETURN(ev);
        }

6621 6622 6623 6624 6625 6626 6627 6628 6629 6630 6631 6632 6633 6634 6635 6636 6637 6638 6639
        if (rli->ign_gtids.count())
        {
          /* We generate and return a Gtid_list, to update gtid_slave_pos. */
          DBUG_PRINT("info",("seeing ignored end gtids"));
          ev= new Gtid_list_log_event(&rli->ign_gtids,
                                      Gtid_list_log_event::FLAG_IGN_GTIDS);
          rli->ign_gtids.reset();
          mysql_mutex_unlock(log_lock);
          if (unlikely(!ev))
          {
            errmsg= "Slave SQL thread failed to create a Gtid_list event "
              "(out of memory?), gtid_slave_pos may be inaccurate";
            goto err;
          }
          ev->server_id= 0; // don't be ignored by slave SQL thread
          ev->set_artificial_event(); // Don't mess up Exec_Master_Log_Pos
          DBUG_RETURN(ev);
        }

6640 6641 6642 6643 6644 6645 6646 6647 6648 6649 6650
        /*
          We have to check sql_slave_killed() here an extra time.
          Otherwise we may miss a wakeup, since last check was done
          without holding LOCK_log.
        */
        if (sql_slave_killed(rgi))
        {
          mysql_mutex_unlock(log_lock);
          break;
        }

unknown's avatar
unknown committed
6651 6652 6653 6654
        /*
          We can, and should release data_lock while we are waiting for
          update. If we do not, show slave status will block
        */
Marc Alff's avatar
Marc Alff committed
6655
        mysql_mutex_unlock(&rli->data_lock);
6656 6657

        /*
unknown's avatar
unknown committed
6658
          Possible deadlock :
6659 6660 6661 6662 6663 6664 6665 6666 6667 6668 6669
          - the I/O thread has reached log_space_limit
          - the SQL thread has read all relay logs, but cannot purge for some
          reason:
            * it has already purged all logs except the current one
            * there are other logs than the current one but they're involved in
            a transaction that finishes in the current one (or is not finished)
          Solution :
          Wake up the possibly waiting I/O thread, and set a boolean asking
          the I/O thread to temporarily ignore the log_space_limit
          constraint, because we do not want the I/O thread to block because of
          space (it's ok if it blocks for any other reason (e.g. because the
unknown's avatar
unknown committed
6670
          master does not send anything). Then the I/O thread stops waiting
6671 6672 6673
          and reads one more event and starts honoring log_space_limit again.

          If the SQL thread needs more events to be able to rotate the log (it
Michael Widenius's avatar
Michael Widenius committed
6674 6675
          might need to finish the current group first), then it can ask for
          one more at a time. Thus we don't outgrow the relay log indefinitely,
6676 6677 6678 6679
          but rather in a controlled manner, until the next rotate.

          When the SQL thread starts it sets ignore_log_space_limit to false. 
          We should also reset ignore_log_space_limit to 0 when the user does 
Michael Widenius's avatar
Michael Widenius committed
6680 6681 6682
          RESET SLAVE, but in fact, no need as RESET SLAVE requires that the
          slave be stopped, and the SQL thread sets ignore_log_space_limit
          to 0 when
unknown's avatar
unknown committed
6683
          it stops.
6684
        */
Marc Alff's avatar
Marc Alff committed
6685
        mysql_mutex_lock(&rli->log_space_lock);
6686 6687 6688 6689 6690 6691 6692 6693 6694 6695 6696 6697 6698 6699 6700 6701 6702 6703 6704 6705 6706 6707 6708 6709 6710

        /* 
          If we have reached the limit of the relay space and we
          are going to sleep, waiting for more events:

          1. If outside a group, SQL thread asks the IO thread 
             to force a rotation so that the SQL thread purges 
             logs next time it processes an event (thus space is
             freed).

          2. If in a group, SQL thread asks the IO thread to 
             ignore the limit and queues yet one more event 
             so that the SQL thread finishes the group and 
             is are able to rotate and purge sometime soon.
         */
        if (rli->log_space_limit && 
            rli->log_space_limit < rli->log_space_total)
        {
          /* force rotation if not in an unfinished group */
          rli->sql_force_rotate_relay= !rli->is_in_group();

          /* ask for one more event */
          rli->ignore_log_space_limit= true;
        }

Marc Alff's avatar
Marc Alff committed
6711
        mysql_cond_broadcast(&rli->log_space_cond);
6712
        mysql_mutex_unlock(&rli->log_space_lock);
Andrei Elkin's avatar
Andrei Elkin committed
6713
        // Note that wait_for_update_relay_log unlocks lock_log !
Michael Widenius's avatar
Michael Widenius committed
6714
        rli->relay_log.wait_for_update_relay_log(rli->sql_driver_thd);
6715
        // re-acquire data lock since we released it earlier
Marc Alff's avatar
Marc Alff committed
6716
        mysql_mutex_lock(&rli->data_lock);
6717
        rli->sql_thread_caught_up= false;
unknown's avatar
unknown committed
6718
        continue;
6719
      }
unknown's avatar
unknown committed
6720
      /*
unknown's avatar
unknown committed
6721 6722 6723
        If the log was not hot, we need to move to the next log in
        sequence. The next log could be hot or cold, we deal with both
        cases separately after doing some common initialization
unknown's avatar
unknown committed
6724 6725 6726
      */
      end_io_cache(cur_log);
      DBUG_ASSERT(rli->cur_log_fd >= 0);
Marc Alff's avatar
Marc Alff committed
6727
      mysql_file_close(rli->cur_log_fd, MYF(MY_WME));
unknown's avatar
unknown committed
6728
      rli->cur_log_fd = -1;
6729
      rli->last_inuse_relaylog->completed= true;
unknown's avatar
unknown committed
6730

6731
      if (relay_log_purge)
unknown's avatar
unknown committed
6732
      {
unknown's avatar
unknown committed
6733
        /*
6734 6735 6736 6737 6738 6739 6740 6741 6742 6743
          purge_first_log will properly set up relay log coordinates in rli.
          If the group's coordinates are equal to the event's coordinates
          (i.e. the relay log was not rotated in the middle of a group),
          we can purge this relay log too.
          We do ulonglong and string comparisons, this may be slow but
          - purging the last relay log is nice (it can save 1GB of disk), so we
          like to detect the case where we can do it, and given this,
          - I see no better detection method
          - purge_first_log is not called that often
        */
unknown's avatar
unknown committed
6744
        if (rli->relay_log.purge_first_log
6745 6746 6747
            (rli,
             rli->group_relay_log_pos == rli->event_relay_log_pos
             && !strcmp(rli->group_relay_log_name,rli->event_relay_log_name)))
unknown's avatar
unknown committed
6748 6749 6750 6751
        {
          errmsg = "Error purging processed logs";
          goto err;
        }
unknown's avatar
unknown committed
6752
      }
6753 6754
      else
      {
unknown's avatar
unknown committed
6755 6756 6757 6758 6759 6760 6761 6762 6763 6764 6765 6766 6767
        /*
          If hot_log is set, then we already have a lock on
          LOCK_log.  If not, we have to get the lock.

          According to Sasha, the only time this code will ever be executed
          is if we are recovering from a bug.
        */
        if (rli->relay_log.find_next_log(&rli->linfo, !hot_log))
        {
          errmsg = "error switching to the next log";
          goto err;
        }
        rli->event_relay_log_pos = BIN_LOG_HEADER_SIZE;
6768
        strmake_buf(rli->event_relay_log_name,rli->linfo.log_file_name);
unknown's avatar
unknown committed
6769
        flush_relay_log_info(rli);
unknown's avatar
unknown committed
6770
      }
6771 6772 6773 6774 6775 6776 6777 6778 6779 6780 6781 6782 6783

      /*
        Now we want to open this next log. To know if it's a hot log (the one
        being written by the I/O thread now) or a cold log, we can use
        is_active(); if it is hot, we use the I/O cache; if it's cold we open
        the file normally. But if is_active() reports that the log is hot, this
        may change between the test and the consequence of the test. So we may
        open the I/O cache whereas the log is now cold, which is nonsense.
        To guard against this, we need to have LOCK_log.
      */

      DBUG_PRINT("info",("hot_log: %d",hot_log));
      if (!hot_log) /* if hot_log, we already have this mutex */
Marc Alff's avatar
Marc Alff committed
6784
        mysql_mutex_lock(log_lock);
unknown's avatar
unknown committed
6785 6786
      if (rli->relay_log.is_active(rli->linfo.log_file_name))
      {
unknown's avatar
unknown committed
6787 6788 6789 6790 6791
        rli->cur_log= cur_log= rli->relay_log.get_log_file();
        rli->cur_log_old_open_count= rli->relay_log.get_open_count();
        DBUG_ASSERT(rli->cur_log_fd == -1);

        /*
6792 6793 6794 6795 6796 6797 6798 6799 6800 6801 6802 6803 6804 6805 6806 6807 6808 6809 6810 6811 6812 6813 6814 6815 6816 6817 6818 6819 6820 6821 6822 6823 6824 6825 6826 6827 6828 6829 6830 6831 6832 6833 6834 6835 6836 6837 6838 6839 6840 6841 6842 6843 6844 6845 6846 6847 6848
           When the SQL thread is [stopped and] (re)started the
           following may happen:

           1. Log was hot at stop time and remains hot at restart

              SQL thread reads again from hot_log (SQL thread was
              reading from the active log when it was stopped and the
              very same log is still active on SQL thread restart).

              In this case, my_b_seek is performed on cur_log, while
              cur_log points to relay_log.get_log_file();

           2. Log was hot at stop time but got cold before restart

              The log was hot when SQL thread stopped, but it is not
              anymore when the SQL thread restarts.

              In this case, the SQL thread reopens the log, using
              cache_buf, ie, cur_log points to &cache_buf, and thence
              its coordinates are reset.

           3. Log was already cold at stop time

              The log was not hot when the SQL thread stopped, and, of
              course, it will not be hot when it restarts.

              In this case, the SQL thread opens the cold log again,
              using cache_buf, ie, cur_log points to &cache_buf, and
              thence its coordinates are reset.

           4. Log was hot at stop time, DBA changes to previous cold
              log and restarts SQL thread

              The log was hot when the SQL thread was stopped, but the
              user changed the coordinates of the SQL thread to
              restart from a previous cold log.

              In this case, at start time, cur_log points to a cold
              log, opened using &cache_buf as cache, and coordinates
              are reset. However, as it moves on to the next logs, it
              will eventually reach the hot log. If the hot log is the
              same at the time the SQL thread was stopped, then
              coordinates were not reset - the cur_log will point to
              relay_log.get_log_file(), and not a freshly opened
              IO_CACHE through cache_buf. For this reason we need to
              deploy a my_b_seek before calling check_binlog_magic at
              this point of the code (see: BUG#55263 for more
              details).
          
          NOTES: 
            - We must keep the LOCK_log to read the 4 first bytes, as
              this is a hot log (same as when we call read_log_event()
              above: for a hot log we take the mutex).

            - Because of scenario #4 above, we need to have a
              my_b_seek here. Otherwise, we might hit the assertion
              inside check_binlog_magic.
unknown's avatar
unknown committed
6849
        */
6850 6851

        my_b_seek(cur_log, (my_off_t) 0);
unknown's avatar
unknown committed
6852
        if (check_binlog_magic(cur_log,&errmsg))
6853
        {
Marc Alff's avatar
Marc Alff committed
6854 6855
          if (!hot_log)
            mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
6856
          goto err;
6857
        }
6858 6859 6860 6861 6862 6863
        if (rli->alloc_inuse_relaylog(rli->linfo.log_file_name))
        {
          if (!hot_log)
            mysql_mutex_unlock(log_lock);
          goto err;
        }
Marc Alff's avatar
Marc Alff committed
6864 6865
        if (!hot_log)
          mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
6866
        continue;
6867
      }
Marc Alff's avatar
Marc Alff committed
6868 6869
      if (!hot_log)
        mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
6870
      /*
unknown's avatar
unknown committed
6871 6872 6873
        if we get here, the log was not hot, so we will have to open it
        ourselves. We are sure that the log is still not hot now (a log can get
        from hot to cold, but not from cold to hot). No need for LOCK_log.
unknown's avatar
unknown committed
6874 6875 6876
      */
      // open_binlog() will check the magic header
      if ((rli->cur_log_fd=open_binlog(cur_log,rli->linfo.log_file_name,
unknown's avatar
unknown committed
6877 6878
                                       &errmsg)) <0)
        goto err;
6879 6880
      if (rli->alloc_inuse_relaylog(rli->linfo.log_file_name))
        goto err;
6881
    }
unknown's avatar
unknown committed
6882
    else
6883
    {
unknown's avatar
unknown committed
6884
      /*
unknown's avatar
unknown committed
6885 6886
        Read failed with a non-EOF error.
        TODO: come up with something better to handle this error
unknown's avatar
unknown committed
6887 6888
      */
      if (hot_log)
Marc Alff's avatar
Marc Alff committed
6889
        mysql_mutex_unlock(log_lock);
6890
      sql_print_error("Slave SQL thread: I/O error reading \
unknown's avatar
unknown committed
6891
event(errno: %d  cur_log->error: %d)",
unknown's avatar
unknown committed
6892
                      my_errno,cur_log->error);
unknown's avatar
unknown committed
6893
      // set read position to the beginning of the event
6894
      my_b_seek(cur_log,rli->event_relay_log_pos);
unknown's avatar
unknown committed
6895 6896
      /* otherwise, we have had a partial read */
      errmsg = "Aborting slave SQL thread because of partial event read";
unknown's avatar
unknown committed
6897
      break;                                    // To end of function
6898 6899
    }
  }
6900
  if (!errmsg && global_system_variables.log_warnings)
6901
  {
unknown's avatar
unknown committed
6902
    sql_print_information("Error reading relay log event: %s",
6903 6904 6905
                          "slave SQL thread was killed");
    DBUG_RETURN(0);
  }
unknown's avatar
unknown committed
6906

6907
err:
6908 6909
  if (errmsg)
    sql_print_error("Error reading relay log event: %s", errmsg);
unknown's avatar
unknown committed
6910
  DBUG_RETURN(0);
6911 6912
}

6913 6914 6915
/*
  Rotate a relay log (this is used only by FLUSH LOGS; the automatic rotation
  because of size is simpler because when we do it we already have all relevant
unknown's avatar
unknown committed
6916
  locks; here we don't, so this function is mainly taking locks).
6917 6918
  Returns nothing as we cannot catch any error (MYSQL_BIN_LOG::new_file()
  is void).
6919 6920
*/

6921
int rotate_relay_log(Master_info* mi)
6922 6923
{
  DBUG_ENTER("rotate_relay_log");
6924
  Relay_log_info* rli= &mi->rli;
6925
  int error= 0;
unknown's avatar
unknown committed
6926

Luis Soares's avatar
Luis Soares committed
6927
  DBUG_EXECUTE_IF("crash_before_rotate_relaylog", DBUG_SUICIDE(););
6928

unknown's avatar
unknown committed
6929
  /*
unknown's avatar
unknown committed
6930 6931 6932
     We need to test inited because otherwise, new_file() will attempt to lock
     LOCK_log, which may not be inited (if we're not a slave).
  */
6933 6934
  if (!rli->inited)
  {
unknown's avatar
unknown committed
6935
    DBUG_PRINT("info", ("rli->inited == 0"));
unknown's avatar
unknown committed
6936
    goto end;
6937
  }
unknown's avatar
unknown committed
6938

6939
  /* If the relay log is closed, new_file() will do nothing. */
6940 6941
  if ((error= rli->relay_log.new_file()))
    goto end;
unknown's avatar
unknown committed
6942

6943 6944 6945 6946
  /*
    We harvest now, because otherwise BIN_LOG_HEADER_SIZE will not immediately
    be counted, so imagine a succession of FLUSH LOGS  and assume the slave
    threads are started:
unknown's avatar
unknown committed
6947 6948 6949 6950 6951 6952
    relay_log_space decreases by the size of the deleted relay log, but does
    not increase, so flush-after-flush we may become negative, which is wrong.
    Even if this will be corrected as soon as a query is replicated on the
    slave (because the I/O thread will then call harvest_bytes_written() which
    will harvest all these BIN_LOG_HEADER_SIZE we forgot), it may give strange
    output in SHOW SLAVE STATUS meanwhile. So we harvest now.
6953 6954
    If the log is closed, then this will just harvest the last writes, probably
    0 as they probably have been harvested.
Sergei Golubchik's avatar
Sergei Golubchik committed
6955 6956

    Note that it needs to be protected by mi->data_lock.
6957
  */
Sergei Golubchik's avatar
Sergei Golubchik committed
6958
  mysql_mutex_assert_owner(&mi->data_lock);
6959
  rli->relay_log.harvest_bytes_written(&rli->log_space_total);
unknown's avatar
unknown committed
6960
end:
6961
  DBUG_RETURN(error);
6962 6963
}

unknown's avatar
unknown committed
6964

unknown's avatar
unknown committed
6965 6966 6967
/**
   Detects, based on master's version (as found in the relay log), if master
   has a certain bug.
6968
   @param rli Relay_log_info which tells the master's version
unknown's avatar
unknown committed
6969
   @param bug_id Number of the bug as found in bugs.mysql.com
6970
   @param report bool report error message, default TRUE
6971 6972 6973 6974 6975 6976 6977

   @param pred Predicate function that will be called with @c param to
   check for the bug. If the function return @c true, the bug is present,
   otherwise, it is not.

   @param param  State passed to @c pred function.

unknown's avatar
unknown committed
6978 6979
   @return TRUE if master has the bug, FALSE if it does not.
*/
6980 6981
bool rpl_master_has_bug(const Relay_log_info *rli, uint bug_id, bool report,
                        bool (*pred)(const void *), const void *param)
unknown's avatar
unknown committed
6982 6983 6984 6985 6986 6987 6988 6989
{
  struct st_version_range_for_one_bug {
    uint        bug_id;
    const uchar introduced_in[3]; // first version with bug
    const uchar fixed_in[3];      // first version with fix
  };
  static struct st_version_range_for_one_bug versions_for_all_bugs[]=
  {
unknown's avatar
unknown committed
6990
    {24432, { 5, 0, 24 }, { 5, 0, 38 } },
6991 6992 6993
    {24432, { 5, 1, 12 }, { 5, 1, 17 } },
    {33029, { 5, 0,  0 }, { 5, 0, 58 } },
    {33029, { 5, 1,  0 }, { 5, 1, 12 } },
6994
    {37426, { 5, 1,  0 }, { 5, 1, 26 } },
unknown's avatar
unknown committed
6995 6996
  };
  const uchar *master_ver=
6997
    rli->relay_log.description_event_for_exec->server_version_split.ver;
unknown's avatar
unknown committed
6998

6999
  DBUG_ASSERT(sizeof(rli->relay_log.description_event_for_exec->server_version_split.ver) == 3);
unknown's avatar
unknown committed
7000 7001 7002 7003 7004 7005 7006 7007

  for (uint i= 0;
       i < sizeof(versions_for_all_bugs)/sizeof(*versions_for_all_bugs);i++)
  {
    const uchar *introduced_in= versions_for_all_bugs[i].introduced_in,
      *fixed_in= versions_for_all_bugs[i].fixed_in;
    if ((versions_for_all_bugs[i].bug_id == bug_id) &&
        (memcmp(introduced_in, master_ver, 3) <= 0) &&
7008 7009
        (memcmp(fixed_in,      master_ver, 3) >  0) &&
        (pred == NULL || (*pred)(param)))
unknown's avatar
unknown committed
7010
    {
7011 7012
      if (!report)
	return TRUE;
unknown's avatar
unknown committed
7013 7014 7015 7016 7017
      // a short message for SHOW SLAVE STATUS (message length constraints)
      my_printf_error(ER_UNKNOWN_ERROR, "master may suffer from"
                      " http://bugs.mysql.com/bug.php?id=%u"
                      " so slave stops; check error log on slave"
                      " for more info", MYF(0), bug_id);
7018
      // a verbose message for the error log
7019
      rli->report(ERROR_LEVEL, ER_UNKNOWN_ERROR, NULL,
7020 7021
                  "According to the master's version ('%s'),"
                  " it is probable that master suffers from this bug:"
7022 7023 7024 7025 7026 7027 7028 7029 7030 7031 7032 7033 7034 7035 7036 7037
                      " http://bugs.mysql.com/bug.php?id=%u"
                      " and thus replicating the current binary log event"
                      " may make the slave's data become different from the"
                      " master's data."
                      " To take no risk, slave refuses to replicate"
                      " this event and stops."
                      " We recommend that all updates be stopped on the"
                      " master and slave, that the data of both be"
                      " manually synchronized,"
                      " that master's binary logs be deleted,"
                      " that master be upgraded to a version at least"
                      " equal to '%d.%d.%d'. Then replication can be"
                      " restarted.",
                      rli->relay_log.description_event_for_exec->server_version,
                      bug_id,
                      fixed_in[0], fixed_in[1], fixed_in[2]);
unknown's avatar
unknown committed
7038 7039 7040 7041 7042 7043
      return TRUE;
    }
  }
  return FALSE;
}

7044 7045 7046 7047 7048 7049 7050 7051 7052 7053 7054 7055
/**
   BUG#33029, For all 5.0 up to 5.0.58 exclusive, and 5.1 up to 5.1.12
   exclusive, if one statement in a SP generated AUTO_INCREMENT value
   by the top statement, all statements after it would be considered
   generated AUTO_INCREMENT value by the top statement, and a
   erroneous INSERT_ID value might be associated with these statement,
   which could cause duplicate entry error and stop the slave.

   Detect buggy master to work around.
 */
bool rpl_master_erroneous_autoinc(THD *thd)
{
7056
  if (thd->rgi_slave)
7057
  {
7058
    DBUG_EXECUTE_IF("simulate_bug33029", return TRUE;);
7059
    return rpl_master_has_bug(thd->rgi_slave->rli, 33029, FALSE, NULL, NULL);
7060 7061 7062 7063
  }
  return FALSE;
}

7064 7065 7066 7067
/**
  @} (end of group Replication)
*/

unknown's avatar
SCRUM  
unknown committed
7068
#endif /* HAVE_REPLICATION */