slave.cc 219 KB
Newer Older
Sergei Golubchik's avatar
Sergei Golubchik committed
1 2
/* Copyright (c) 2000, 2012, Oracle and/or its affiliates.
   Copyright (c) 2008, 2011, Monty Program Ab
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 29
#include "sql_priv.h"
#include "my_global.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 HAVE_REPLICATION

#include "rpl_tblmap.h"
59
#include "debug_sync.h"
60 61
#include "rpl_parallel.h"

62

63
#define FLAGSTR(V,F) ((V)&(F)?#F" ":"")
64

65
#define MAX_SLAVE_RETRY_PAUSE 5
66 67 68 69
/*
  a parameter of sql_slave_killed() to defer the killed status
*/
#define SLAVE_WAIT_GROUP_DONE 60
70 71
bool use_slave_mask = 0;
MY_BITMAP slave_error_mask;
72
char slave_skip_error_names[SHOW_VAR_FUNC_BUFF_SIZE];
73

74
char* slave_load_tmpdir = 0;
75
Master_info *active_mi= 0;
76
Master_info_index *master_info_index;
77
my_bool replicate_same_server_id;
78
ulonglong relay_log_space_limit = 0;
79
LEX_STRING default_master_connection_name= { (char*) "", 0 };
unknown's avatar
unknown committed
80 81 82 83 84

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

85
  TODO: move the vars below under Master_info
unknown's avatar
unknown committed
86
*/
87

88
int disconnect_slave_event_count = 0, abort_slave_event_count = 0;
unknown's avatar
unknown committed
89

He Zhenxing's avatar
He Zhenxing committed
90 91
static pthread_key(Master_info*, RPL_MASTER_INFO);

92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118
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
119
log '%s' at position %s%s",
120 121 122 123 124 125 126
    "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
127
    "failed dump request, reconnecting to try again, log '%s' at position %s%s",
128 129 130 131 132 133 134 135
    "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
136
log '%s' at position %s%s",
137 138 139 140 141 142 143
    "",
    "Slave I/O thread killed during or after a reconnect done to recover from \
failed read"
  }
};
 

144
typedef enum { SLAVE_THD_IO, SLAVE_THD_SQL} SLAVE_THD_TYPE;
unknown's avatar
unknown committed
145

146 147
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);
148
static bool wait_for_relay_log_space(Relay_log_info* rli);
Michael Widenius's avatar
Michael Widenius committed
149 150
static bool io_slave_killed(Master_info* mi);
static bool sql_slave_killed(rpl_group_info *rgi);
151 152
static int init_slave_thread(THD* thd, Master_info *mi,
                             SLAVE_THD_TYPE thd_type);
153
static void print_slave_skip_errors(void);
154 155
static int safe_connect(THD* thd, MYSQL* mysql, Master_info* mi);
static int safe_reconnect(THD* thd, MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
156
                          bool suppress_warnings);
157
static int connect_to_master(THD* thd, MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
158
                             bool reconnect, bool suppress_warnings);
Michael Widenius's avatar
Michael Widenius committed
159
static Log_event* next_event(rpl_group_info* rgi);
160
static int queue_event(Master_info* mi,const char* buf,ulong event_len);
161
static int terminate_slave_thread(THD *thd,
Marc Alff's avatar
Marc Alff committed
162 163
                                  mysql_mutex_t *term_lock,
                                  mysql_cond_t *term_cond,
164 165
                                  volatile uint *slave_running,
                                  bool skip_lock);
Michael Widenius's avatar
Michael Widenius committed
166
static bool check_io_slave_killed(Master_info *mi, const char *info);
unknown's avatar
unknown committed
167 168 169 170
static bool send_show_master_info_header(THD *thd, bool full,
                                         size_t gtid_pos_length);
static bool send_show_master_info_data(THD *thd, Master_info *mi, bool full,
                                       String *gtid_pos);
171 172

/*
unknown's avatar
unknown committed
173
  Find out which replications threads are running
174

unknown's avatar
unknown committed
175 176
  SYNOPSIS
    init_thread_mask()
unknown's avatar
unknown committed
177 178 179
    mask                Return value here
    mi                  master_info for slave
    inverse             If set, returns which threads are not running
unknown's avatar
unknown committed
180 181 182 183 184 185

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

  RETURN
unknown's avatar
unknown committed
186 187
    mask        If inverse == 0, running threads
                If inverse == 1, stopped threads
188 189
*/

190
void init_thread_mask(int* mask,Master_info* mi,bool inverse)
191 192 193
{
  bool set_io = mi->slave_running, set_sql = mi->rli.slave_running;
  register int tmp_mask=0;
194 195
  DBUG_ENTER("init_thread_mask");

196 197 198 199
  if (set_io)
    tmp_mask |= SLAVE_IO;
  if (set_sql)
    tmp_mask |= SLAVE_SQL;
200 201
  if (inverse)
    tmp_mask^= (SLAVE_IO | SLAVE_SQL);
202
  *mask = tmp_mask;
203
  DBUG_VOID_RETURN;
204 205
}

206

unknown's avatar
unknown committed
207
/*
208
  lock_slave_threads()
unknown's avatar
unknown committed
209
*/
210

211
void lock_slave_threads(Master_info* mi)
212
{
213 214
  DBUG_ENTER("lock_slave_threads");

215
  //TODO: see if we can do this without dual mutex
Marc Alff's avatar
Marc Alff committed
216 217
  mysql_mutex_lock(&mi->run_lock);
  mysql_mutex_lock(&mi->rli.run_lock);
218
  DBUG_VOID_RETURN;
219 220
}

221

unknown's avatar
unknown committed
222
/*
223
  unlock_slave_threads()
unknown's avatar
unknown committed
224
*/
225

226
void unlock_slave_threads(Master_info* mi)
227
{
228 229
  DBUG_ENTER("unlock_slave_threads");

230
  //TODO: see if we can do this without dual mutex
Marc Alff's avatar
Marc Alff committed
231 232
  mysql_mutex_unlock(&mi->rli.run_lock);
  mysql_mutex_unlock(&mi->run_lock);
233
  DBUG_VOID_RETURN;
234 235
}

Marc Alff's avatar
Marc Alff committed
236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256
#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 */
257

258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317

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);
  thd->store_globals();

  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,
                      thd->stmt_da->sql_errno(), thd->stmt_da->message());

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

  mysql_mutex_lock(&LOCK_thread_count);
  slave_init_thread_running= false;
  mysql_cond_signal(&COND_thread_count);
  mysql_mutex_unlock(&LOCK_thread_count);

  return 0;
}


static int
run_slave_init_thread()
{
  pthread_t th;

  slave_init_thread_running= true;
  if (mysql_thread_create(key_thread_slave_init, &th, NULL,
                          handle_slave_init, NULL))
  {
    sql_print_error("Failed to create thread while initialising slave");
    return 1;
  }

  mysql_mutex_lock(&LOCK_thread_count);
  while (slave_init_thread_running)
    mysql_cond_wait(&COND_thread_count, &LOCK_thread_count);
  mysql_mutex_unlock(&LOCK_thread_count);

  return 0;
}


unknown's avatar
unknown committed
318
/* Initialize slave structures */
319

320 321
int init_slave()
{
322
  DBUG_ENTER("init_slave");
323
  int error= 0;
324

Marc Alff's avatar
Marc Alff committed
325 326 327 328
#ifdef HAVE_PSI_INTERFACE
  init_slave_psi_keys();
#endif

329 330 331
  if (run_slave_init_thread())
    return 1;

332 333 334 335 336
  /*
    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
337
  mysql_mutex_lock(&LOCK_active_mi);
338

He Zhenxing's avatar
He Zhenxing committed
339 340
  if (pthread_key_create(&RPL_MASTER_INFO, NULL))
    goto err;
341

342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364
  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;
  }

365 366 367
  if (global_rpl_thread_pool.init(opt_slave_parallel_threads))
    return 1;

368 369 370 371 372 373 374 375 376
  /*
    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();
  }

377
  /*
378 379 380
    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.
381
  */
382

unknown's avatar
unknown committed
383
  if (init_master_info(active_mi,master_info_file,relay_log_info_file,
384
                       1, (SLAVE_IO | SLAVE_SQL)))
385
  {
386
    sql_print_error("Failed to initialize the master info structure");
unknown's avatar
unknown committed
387
    goto err;
388
  }
389

390 391
  /* If server id is not set, start_slave_thread() will say it */

392
  if (active_mi->host[0] && !opt_skip_slave_start)
393
  {
394
    if (start_slave_threads(1 /* need mutex */,
unknown's avatar
unknown committed
395 396 397 398 399
                            0 /* no wait for start*/,
                            active_mi,
                            master_info_file,
                            relay_log_info_file,
                            SLAVE_IO | SLAVE_SQL))
unknown's avatar
unknown committed
400
    {
401
      sql_print_error("Failed to create slave threads");
unknown's avatar
unknown committed
402 403
      goto err;
    }
404 405
  }

406
end:
Marc Alff's avatar
Marc Alff committed
407
  mysql_mutex_unlock(&LOCK_active_mi);
408
  DBUG_RETURN(error);
409 410 411 412 413

err:
  sql_print_error("Failed to allocate memory for the Master Info structure");
  error= 1;
  goto end;
414
}
415

416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442
/*
  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).
 */
443
int init_recovery(Master_info* mi, const char** errmsg)
444 445 446 447 448 449 450 451
{
  DBUG_ENTER("init_recovery");
 
  Relay_log_info *rli= &mi->rli;
  if (rli->group_master_log_name[0])
  {
    mi->master_log_pos= max(BIN_LOG_HEADER_SIZE,
                             rli->group_master_log_pos);
452
    strmake_buf(mi->master_log_name, rli->group_master_log_name);
453 454 455 456
 
    sql_print_warning("Recovery from master pos %ld and file %s.",
                      (ulong) mi->master_log_pos, mi->master_log_name);
 
457 458
    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());
459 460 461
 
    rli->group_relay_log_pos= rli->event_relay_log_pos= BIN_LOG_HEADER_SIZE;
  }
462

463 464
  DBUG_RETURN(0);
}
unknown's avatar
unknown committed
465

466
 
467 468 469 470 471 472 473 474 475 476 477 478
/**
  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.
  */
479
  const size_t MIN_ROOM= 10;
480 481 482 483
  DBUG_ENTER("print_slave_skip_errors");
  DBUG_ASSERT(sizeof(slave_skip_error_names) > MIN_ROOM);
  DBUG_ASSERT(MAX_SLAVE_ERROR <= 999999); // 6 digits

484 485 486
  /* Make @@slave_skip_errors show the nice human-readable value.  */
  opt_slave_skip_errors= slave_skip_error_names;

487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504
  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;

505
    for (errnum= 0; errnum < MAX_SLAVE_ERROR; errnum++)
506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527
    {
      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
528
/*
529
  Init function to set up array for errors that should be skipped for slave
unknown's avatar
unknown committed
530

unknown's avatar
unknown committed
531 532
  SYNOPSIS
    init_slave_skip_errors()
unknown's avatar
unknown committed
533
    arg         List of errors numbers to skip, separated with ','
unknown's avatar
unknown committed
534 535 536 537

  NOTES
    Called from get_options() in mysqld.cc on start-up
*/
538

unknown's avatar
unknown committed
539
void init_slave_skip_errors(const char* arg)
540
{
unknown's avatar
unknown committed
541
  const char *p;
542 543
  DBUG_ENTER("init_slave_skip_errors");

544
  if (bitmap_init(&slave_error_mask,0,MAX_SLAVE_ERROR,0))
545 546 547 548 549
  {
    fprintf(stderr, "Badly out of memory, please check your system status\n");
    exit(1);
  }
  use_slave_mask = 1;
550
  for (;my_isspace(system_charset_info,*arg);++arg)
551
    /* empty */;
552
  if (!my_strnncoll(system_charset_info,(uchar*)arg,4,(const uchar*)"all",4))
553 554
  {
    bitmap_set_all(&slave_error_mask);
555
    print_slave_skip_errors();
556
    DBUG_VOID_RETURN;
557 558 559 560 561 562 563 564
  }
  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);
565
    while (!my_isdigit(system_charset_info,*p) && *p)
566 567
      p++;
  }
568 569
  /* Convert slave skip errors bitmap into a printable string. */
  print_slave_skip_errors();
570
  DBUG_VOID_RETURN;
571 572
}

573

574
int terminate_slave_threads(Master_info* mi,int thread_mask,bool skip_lock)
575
{
576 577
  DBUG_ENTER("terminate_slave_threads");

578
  if (!mi->inited)
579
    DBUG_RETURN(0); /* successfully do nothing */
580
  int error,force_all = (thread_mask & SLAVE_FORCE_ALL);
Marc Alff's avatar
Marc Alff committed
581
  mysql_mutex_t *sql_lock = &mi->rli.run_lock, *io_lock = &mi->run_lock;
582
  mysql_mutex_t *log_lock= mi->rli.relay_log.get_log_lock();
unknown's avatar
unknown committed
583

584
  if (thread_mask & (SLAVE_SQL|SLAVE_FORCE_ALL))
585
  {
unknown's avatar
unknown committed
586
    DBUG_PRINT("info",("Terminating SQL thread"));
587
    mi->rli.abort_slave=1;
Michael Widenius's avatar
Michael Widenius committed
588
    if ((error=terminate_slave_thread(mi->rli.sql_driver_thd, sql_lock,
unknown's avatar
unknown committed
589
                                      &mi->rli.stop_cond,
590 591
                                      &mi->rli.slave_running,
                                      skip_lock)) &&
unknown's avatar
unknown committed
592
        !force_all)
593
      DBUG_RETURN(error);
594

595
    mysql_mutex_lock(log_lock);
596

Libing Song's avatar
Libing Song committed
597
    DBUG_PRINT("info",("Flushing relay-log info file."));
598
    if (current_thd)
Libing Song's avatar
Libing Song committed
599 600
      thd_proc_info(current_thd, "Flushing relay-log info file.");
    if (flush_relay_log_info(&mi->rli))
601
      DBUG_RETURN(ER_ERROR_DURING_FLUSH_LOGS);
Libing Song's avatar
Libing Song committed
602 603
    
    if (my_sync(mi->rli.info_fd, MYF(MY_WME)))
604 605
      DBUG_RETURN(ER_ERROR_DURING_FLUSH_LOGS);

606
    mysql_mutex_unlock(log_lock);
607
  }
608
  if (thread_mask & (SLAVE_IO|SLAVE_FORCE_ALL))
609
  {
unknown's avatar
unknown committed
610
    DBUG_PRINT("info",("Terminating IO thread"));
611
    mi->abort_slave=1;
612
    if ((error=terminate_slave_thread(mi->io_thd, io_lock,
unknown's avatar
unknown committed
613
                                      &mi->stop_cond,
614 615
                                      &mi->slave_running,
                                      skip_lock)) &&
unknown's avatar
unknown committed
616
        !force_all)
617
      DBUG_RETURN(error);
618

619
    mysql_mutex_lock(log_lock);
620 621 622 623

    DBUG_PRINT("info",("Flushing relay log and master info file."));
    if (current_thd)
      thd_proc_info(current_thd, "Flushing relay log and master info files.");
624
    if (flush_master_info(mi, TRUE, FALSE))
625 626
      DBUG_RETURN(ER_ERROR_DURING_FLUSH_LOGS);

Luis Soares's avatar
Luis Soares committed
627 628
    if (mi->rli.relay_log.is_open() &&
        my_sync(mi->rli.relay_log.get_log_file()->file, MYF(MY_WME)))
629 630 631 632 633
      DBUG_RETURN(ER_ERROR_DURING_FLUSH_LOGS);

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

634
    mysql_mutex_unlock(log_lock);
635
  }
636
  DBUG_RETURN(0); 
637 638
}

639

640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666
/**
   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
667
   @retval 0 All OK ER_SLAVE_NOT_RUNNING otherwise.
668

Andrei Elkin's avatar
Andrei Elkin committed
669 670 671
   @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.
672

673 674 675
 */
static int
terminate_slave_thread(THD *thd,
Marc Alff's avatar
Marc Alff committed
676 677
                       mysql_mutex_t *term_lock,
                       mysql_cond_t *term_cond,
678 679
                       volatile uint *slave_running,
                       bool skip_lock)
680
{
681
  DBUG_ENTER("terminate_slave_thread");
682
  if (!skip_lock)
683
  {
Marc Alff's avatar
Marc Alff committed
684
    mysql_mutex_lock(term_lock);
685 686 687
  }
  else
  {
Marc Alff's avatar
Marc Alff committed
688
    mysql_mutex_assert_owner(term_lock);
689
  }
690 691 692
  if (!*slave_running)
  {
    if (!skip_lock)
693
    {
694 695 696 697
      /*
        if run_lock (term_lock) is acquired locally then either
        slave_running status is fine
      */
Marc Alff's avatar
Marc Alff committed
698
      mysql_mutex_unlock(term_lock);
699 700 701 702
      DBUG_RETURN(0);
    }
    else
    {
703
      DBUG_RETURN(ER_SLAVE_NOT_RUNNING);
704 705 706
    }
  }
  DBUG_ASSERT(thd != 0);
707
  THD_CHECK_SENTRY(thd);
708

unknown's avatar
unknown committed
709
  /*
710
    Is is critical to test if the slave is running. Otherwise, we might
unknown's avatar
unknown committed
711
    be referening freed memory trying to kick it
712
  */
unknown's avatar
unknown committed
713

unknown's avatar
unknown committed
714
  while (*slave_running)                        // Should always be true
715
  {
Michael Widenius's avatar
Michael Widenius committed
716
    int error __attribute__((unused));
717
    DBUG_PRINT("loop", ("killing slave thread"));
718

Marc Alff's avatar
Marc Alff committed
719
    mysql_mutex_lock(&thd->LOCK_thd_data);
720 721 722 723 724 725
#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)
    */
726
    int err __attribute__((unused))= pthread_kill(thd->real_id, thr_client_alarm);
727 728
    DBUG_ASSERT(err != EINVAL);
#endif
Sergei Golubchik's avatar
Sergei Golubchik committed
729 730
    thd->awake(NOT_KILLED);

Marc Alff's avatar
Marc Alff committed
731
    mysql_mutex_unlock(&thd->LOCK_thd_data);
732

unknown's avatar
unknown committed
733 734 735
    /*
      There is a small chance that slave thread might miss the first
      alarm. To protect againts it, resend the signal until it reacts
736 737
    */
    struct timespec abstime;
unknown's avatar
unknown committed
738
    set_timespec(abstime,2);
Marc Alff's avatar
Marc Alff committed
739
    error= mysql_cond_timedwait(term_cond, term_lock, &abstime);
740
    DBUG_ASSERT(error == ETIMEDOUT || error == 0);
741
  }
742 743 744 745

  DBUG_ASSERT(*slave_running == 0);

  if (!skip_lock)
Marc Alff's avatar
Marc Alff committed
746
    mysql_mutex_unlock(term_lock);
747
  DBUG_RETURN(0);
748 749
}

750

Marc Alff's avatar
Marc Alff committed
751 752 753 754 755 756 757
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
758 759
                       volatile uint *slave_running,
                       volatile ulong *slave_run_id,
Konstantin Osipov's avatar
Konstantin Osipov committed
760
                       Master_info* mi)
761 762
{
  pthread_t th;
unknown's avatar
unknown committed
763 764 765
  ulong start_id;
  DBUG_ENTER("start_slave_thread");

766 767
  DBUG_ASSERT(mi->inited);

768
  if (start_lock)
Marc Alff's avatar
Marc Alff committed
769
    mysql_mutex_lock(start_lock);
770
  if (!global_system_variables.server_id)
771 772
  {
    if (start_cond)
Marc Alff's avatar
Marc Alff committed
773
      mysql_cond_broadcast(start_cond);
774
    if (start_lock)
Marc Alff's avatar
Marc Alff committed
775
      mysql_mutex_unlock(start_lock);
776
    sql_print_error("Server id not set, will not start slave");
unknown's avatar
unknown committed
777
    DBUG_RETURN(ER_BAD_SLAVE);
778
  }
unknown's avatar
unknown committed
779

780
  if (*slave_running)
781 782
  {
    if (start_cond)
Marc Alff's avatar
Marc Alff committed
783
      mysql_cond_broadcast(start_cond);
784
    if (start_lock)
Marc Alff's avatar
Marc Alff committed
785
      mysql_mutex_unlock(start_lock);
unknown's avatar
unknown committed
786
    DBUG_RETURN(ER_SLAVE_MUST_STOP);
787
  }
unknown's avatar
unknown committed
788 789
  start_id= *slave_run_id;
  DBUG_PRINT("info",("Creating new slave thread"));
Marc Alff's avatar
Marc Alff committed
790 791
  if (mysql_thread_create(thread_key,
                          &th, &connection_attrib, h_func, (void*)mi))
792 793
  {
    if (start_lock)
Marc Alff's avatar
Marc Alff committed
794
      mysql_mutex_unlock(start_lock);
unknown's avatar
unknown committed
795
    DBUG_RETURN(ER_SLAVE_THREAD);
796
  }
unknown's avatar
unknown committed
797
  if (start_cond && cond_lock) // caller has cond_lock
798 799
  {
    THD* thd = current_thd;
unknown's avatar
unknown committed
800
    while (start_id == *slave_run_id)
801
    {
unknown's avatar
unknown committed
802
      DBUG_PRINT("sleep",("Waiting for slave thread to start"));
803 804 805 806 807 808 809 810 811 812 813
      const char *old_msg= thd->enter_cond(start_cond, cond_lock,
                                           "Waiting for slave thread to start");
      /*
        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);
814
      thd->exit_cond(old_msg);
Marc Alff's avatar
Marc Alff committed
815
      mysql_mutex_lock(cond_lock); // re-acquire it as exit_cond() released
816
      if (thd->killed)
817 818
      {
        if (start_lock)
Marc Alff's avatar
Marc Alff committed
819
          mysql_mutex_unlock(start_lock);
unknown's avatar
unknown committed
820
        DBUG_RETURN(thd->killed_errno());
821
      }
822 823 824
    }
  }
  if (start_lock)
Marc Alff's avatar
Marc Alff committed
825
    mysql_mutex_unlock(start_lock);
unknown's avatar
unknown committed
826
  DBUG_RETURN(0);
827
}
unknown's avatar
unknown committed
828

829

unknown's avatar
unknown committed
830
/*
831
  start_slave_threads()
unknown's avatar
unknown committed
832

unknown's avatar
unknown committed
833 834 835 836
  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
837
*/
unknown's avatar
unknown committed
838

839
int start_slave_threads(bool need_slave_mutex, bool wait_for_start,
840
                        Master_info* mi, const char* master_info_fname,
unknown's avatar
unknown committed
841
                        const char* slave_info_fname, int thread_mask)
842
{
Marc Alff's avatar
Marc Alff committed
843 844
  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;
845
  int error=0;
unknown's avatar
unknown committed
846
  const char *errmsg;
847
  DBUG_ENTER("start_slave_threads");
unknown's avatar
unknown committed
848

849 850 851 852 853 854 855 856 857 858 859 860
  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;
  }
861

unknown's avatar
unknown committed
862 863 864 865 866 867 868 869 870
  /*
    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
871 872
  if (mi->using_gtid != Master_info::USE_GTID_NO &&
      !mi->slave_running && !mi->rli.slave_running)
unknown's avatar
unknown committed
873
  {
874 875 876 877 878 879 880 881 882 883
    /*
      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).
    */
    mi->master_log_pos = max(BIN_LOG_HEADER_SIZE, mi->rli.group_master_log_pos);
    strmake(mi->master_log_name, mi->rli.group_master_log_name,
            sizeof(mi->master_log_name)-1);
unknown's avatar
unknown committed
884
    purge_relay_logs(&mi->rli, NULL, 0, &errmsg);
885 886 887 888
    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
889 890 891 892
    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;
unknown's avatar
unknown committed
893 894
  }

unknown's avatar
unknown committed
895
  if (!error && (thread_mask & SLAVE_IO))
Marc Alff's avatar
Marc Alff committed
896 897 898 899 900 901 902 903
    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);
904
  if (!error && (thread_mask & SLAVE_SQL))
905
  {
Marc Alff's avatar
Marc Alff committed
906 907 908 909 910 911 912 913
    error= start_slave_thread(
#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);
914
    if (error)
915
      terminate_slave_threads(mi, thread_mask & SLAVE_IO, !need_slave_mutex);
916
  }
917
  DBUG_RETURN(error);
918
}
919

920

unknown's avatar
unknown committed
921
/*
922
  Release slave threads at time of executing shutdown.
unknown's avatar
unknown committed
923 924 925 926

  SYNOPSIS
    end_slave()
*/
927

928 929
void end_slave()
{
930 931
  DBUG_ENTER("end_slave");

932 933 934 935 936 937
  /*
    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.
938 939 940

    We can also be called by cleanup(), which only happens if some
    startup parameter to the server was wrong.
941
  */
Marc Alff's avatar
Marc Alff committed
942
  mysql_mutex_lock(&LOCK_active_mi);
943 944 945 946
  /* 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
947
  mysql_mutex_unlock(&LOCK_active_mi);
948
  global_rpl_thread_pool.destroy();
949
  free_all_rpl_filters();
950 951 952
  DBUG_VOID_RETURN;
}

Michael Widenius's avatar
Michael Widenius committed
953
static bool io_slave_killed(Master_info* mi)
unknown's avatar
unknown committed
954
{
955 956
  DBUG_ENTER("io_slave_killed");

957
  DBUG_ASSERT(mi->slave_running); // tracking buffer overrun
Michael Widenius's avatar
Michael Widenius committed
958
  DBUG_RETURN(mi->abort_slave || abort_loop || mi->io_thd->killed);
unknown's avatar
unknown committed
959 960
}

961 962 963 964 965 966
/**
   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.
967

968 969
   @param thd   pointer to a THD instance
   @param rli   pointer to Relay_log_info instance
970

971 972 973
   @return TRUE the killed status is recognized, FALSE a possible killed
           status is deferred.
*/
Michael Widenius's avatar
Michael Widenius committed
974
static bool sql_slave_killed(rpl_group_info *rgi)
975
{
976
  bool ret= FALSE;
Michael Widenius's avatar
Michael Widenius committed
977 978
  Relay_log_info *rli= rgi->rli;
  THD *thd= rgi->thd;
979 980
  DBUG_ENTER("sql_slave_killed");

Michael Widenius's avatar
Michael Widenius committed
981
  DBUG_ASSERT(rli->sql_driver_thd == thd);
982
  DBUG_ASSERT(rli->slave_running == 1);// tracking buffer overrun
Michael Widenius's avatar
Michael Widenius committed
983
  if (abort_loop || rli->sql_driver_thd->killed || rli->abort_slave)
984
  {
985
    /*
Michael Widenius's avatar
Michael Widenius committed
986 987 988 989 990 991 992 993 994 995 996 997 998
      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().
999
    */
Michael Widenius's avatar
Michael Widenius committed
1000

unknown's avatar
unknown committed
1001
    if ((thd->transaction.all.modified_non_trans_table ||
Michael Widenius's avatar
Michael Widenius committed
1002 1003
         (thd->variables.option_bits & OPTION_KEEP_LOG)) &&
        rli->is_in_group())
1004
    {
1005
      char msg_stopped[]=
1006 1007 1008 1009
        "... 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 "
1010 1011 1012
        "ignores duplicate key, key not found, and similar errors (see "
        "documentation for details).";

Michael Widenius's avatar
Michael Widenius committed
1013 1014 1015 1016 1017 1018
      DBUG_PRINT("info", ("modified_non_trans_table: %d  OPTION_BEGIN: %d  "
                          "is_in_group: %d",
                          thd->transaction.all.modified_non_trans_table,
                          test(thd->variables.option_bits & OPTION_BEGIN),
                          rli->is_in_group()));

1019 1020
      if (rli->abort_slave)
      {
Michael Widenius's avatar
Michael Widenius committed
1021 1022 1023 1024
        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 ... "));
1025 1026

        /*
Michael Widenius's avatar
Michael Widenius committed
1027 1028 1029 1030 1031 1032 1033 1034
          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.
1035 1036
        */

1037 1038 1039
        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) <=
1040 1041 1042 1043 1044 1045 1046 1047 1048
          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)
        {
          rli->report(WARNING_LEVEL, 0,
1049 1050 1051
                      "Request to stop slave SQL Thread received while "
                      "applying a group that has non-transactional "
                      "changes; waiting for completion of the group ... ");
1052 1053 1054 1055 1056 1057 1058 1059 1060 1061
        }
        else
        {
          rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
                      ER(ER_SLAVE_FATAL_ERROR), msg_stopped);
        }
      }
      else
      {
        ret= TRUE;
Michael Widenius's avatar
Michael Widenius committed
1062 1063
        rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
                    ER(ER_SLAVE_FATAL_ERROR),
1064 1065 1066 1067 1068 1069
                    msg_stopped);
      }
    }
    else
    {
      ret= TRUE;
1070 1071
    }
  }
1072
  if (ret)
1073
    rgi->last_event_start_time= 0;
1074 1075
  
  DBUG_RETURN(ret);
1076 1077
}

1078

unknown's avatar
unknown committed
1079
/*
1080 1081
  skip_load_data_infile()

unknown's avatar
unknown committed
1082 1083 1084
  NOTES
    This is used to tell a 3.23 master to break send_file()
*/
1085 1086 1087

void skip_load_data_infile(NET *net)
{
1088 1089
  DBUG_ENTER("skip_load_data_infile");

1090
  (void)net_request_file(net, "/dev/null");
unknown's avatar
unknown committed
1091
  (void)my_net_read(net);                               // discard response
1092
  (void)net_write_command(net, 0, (uchar*) "", 0, (uchar*) "", 0); // ok
1093
  DBUG_VOID_RETURN;
1094 1095
}

1096

1097
bool net_request_file(NET* net, const char* fname)
1098
{
1099
  DBUG_ENTER("net_request_file");
1100 1101
  DBUG_RETURN(net_write_command(net, 251, (uchar*) fname, strlen(fname),
                                (uchar*) "", 0));
1102 1103
}

1104 1105
/*
  From other comments and tests in code, it looks like
1106
  sometimes Query_log_event and Load_log_event can have db == 0
1107 1108 1109
  (see rewrite_db() above for example)
  (cases where this happens are unclear; it may be when the master is 3.23).
*/
1110 1111

const char *print_slave_db_safe(const char* db)
1112
{
1113 1114 1115
  DBUG_ENTER("*print_slave_db_safe");

  DBUG_RETURN((db ? db : ""));
1116
}
1117

Sergei Golubchik's avatar
Sergei Golubchik committed
1118 1119
#endif /* HAVE_REPLICATION */

1120
int init_strvar_from_file(char *var, int max_size, IO_CACHE *f,
unknown's avatar
unknown committed
1121
                                 const char *default_val)
unknown's avatar
unknown committed
1122
{
unknown's avatar
unknown committed
1123
  uint length;
1124 1125
  DBUG_ENTER("init_strvar_from_file");

unknown's avatar
unknown committed
1126 1127 1128 1129 1130 1131
  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
1132
    {
unknown's avatar
unknown committed
1133
      /*
unknown's avatar
unknown committed
1134 1135
        If we truncated a line or stopped on last char, remove all chars
        up to and including newline.
unknown's avatar
unknown committed
1136
      */
unknown's avatar
unknown committed
1137
      int c;
1138
      while (((c=my_b_get(f)) != '\n' && c != my_b_EOF)) ;
unknown's avatar
unknown committed
1139
    }
1140
    DBUG_RETURN(0);
unknown's avatar
unknown committed
1141 1142 1143
  }
  else if (default_val)
  {
unknown's avatar
unknown committed
1144
    strmake(var,  default_val, max_size-1);
1145
    DBUG_RETURN(0);
unknown's avatar
unknown committed
1146
  }
1147
  DBUG_RETURN(1);
unknown's avatar
unknown committed
1148 1149
}

1150

Sergei Golubchik's avatar
Sergei Golubchik committed
1151 1152 1153 1154
/*
  when moving these functions to mysys, don't forget to
  remove slave.cc from libmysqld/CMakeLists.txt
*/
1155
int init_intvar_from_file(int* var, IO_CACHE* f, int default_val)
unknown's avatar
unknown committed
1156 1157
{
  char buf[32];
1158 1159
  DBUG_ENTER("init_intvar_from_file");

unknown's avatar
unknown committed
1160 1161

  if (my_b_gets(f, buf, sizeof(buf)))
unknown's avatar
unknown committed
1162 1163
  {
    *var = atoi(buf);
1164
    DBUG_RETURN(0);
unknown's avatar
unknown committed
1165
  }
unknown's avatar
unknown committed
1166
  else if (default_val)
unknown's avatar
unknown committed
1167 1168
  {
    *var = default_val;
1169
    DBUG_RETURN(0);
unknown's avatar
unknown committed
1170
  }
1171
  DBUG_RETURN(1);
unknown's avatar
unknown committed
1172 1173
}

Andrei Elkin's avatar
Andrei Elkin committed
1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194
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
1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246

/**
   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 ||
1247
        ((snd_size + 1 == max_size - read_size) &&  buf_act[max_size - 2] != '\n'))
Andrei Elkin's avatar
Andrei Elkin committed
1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278
    {
      /*
        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)
1279
    my_free(buf_act);
Andrei Elkin's avatar
Andrei Elkin committed
1280 1281 1282
  DBUG_RETURN(ret);
}

Sergei Golubchik's avatar
Sergei Golubchik committed
1283 1284
#ifdef HAVE_REPLICATION

1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299
/*
  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 ||
1300
      errorno == ER_CONNECTION_KILLED ||
1301
      errorno == ER_NEW_ABORTING_CONNECTION ||
1302 1303 1304 1305 1306 1307 1308
      errorno == ER_SERVER_SHUTDOWN)
    return TRUE;

  return FALSE;   
}


1309 1310 1311 1312 1313 1314 1315 1316
/*
  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
1317

1318 1319 1320
  RETURNS
  0       ok
  1       error
1321
  2       transient network problem, the caller should try to reconnect
1322
*/
1323

1324
static int get_master_version_and_clock(MYSQL* mysql, Master_info* mi)
1325
{
1326
  char err_buff[MAX_SLAVE_ERRMSG], err_buff2[MAX_SLAVE_ERRMSG];
1327
  const char* errmsg= 0;
1328 1329 1330
  int err_code= 0;
  MYSQL_RES *master_res= 0;
  MYSQL_ROW master_row;
1331
  uint version= mysql_get_server_version(mysql) / 10000;
1332
  DBUG_ENTER("get_master_version_and_clock");
1333 1334 1335 1336 1337 1338 1339

  /*
    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
1340

1341
  if (!my_isdigit(&my_charset_bin,*mysql->server_version))
1342
  {
1343 1344 1345 1346
    errmsg= err_buff2;
    snprintf(err_buff2, sizeof(err_buff2),
             "Master reported unrecognized MySQL version: %s",
             mysql->server_version);
1347
    err_code= ER_SLAVE_FATAL_ERROR;
1348
    sprintf(err_buff, ER(err_code), err_buff2);
1349
  }
1350 1351 1352 1353 1354
  else
  {
    /*
      Note the following switch will bug when we have MySQL branch 30 ;)
    */
1355 1356 1357 1358 1359 1360 1361 1362
    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);
1363
      err_code= ER_SLAVE_FATAL_ERROR;
1364
      sprintf(err_buff, ER(err_code), err_buff2);
1365
      break;
1366
    case 3:
1367
      mi->rli.relay_log.description_event_for_queue= new
unknown's avatar
unknown committed
1368
        Format_description_log_event(1, mysql->server_version);
1369
      break;
1370
    case 4:
1371
      mi->rli.relay_log.description_event_for_queue= new
unknown's avatar
unknown committed
1372
        Format_description_log_event(3, mysql->server_version);
1373
      break;
unknown's avatar
unknown committed
1374
    default:
1375 1376 1377 1378 1379 1380 1381 1382 1383
      /*
        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
1384
        Format_description_log_event(4, mysql->server_version);
1385 1386
      break;
    }
1387
  }
unknown's avatar
unknown committed
1388 1389

  /*
1390 1391 1392 1393 1394
     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
1395

1396
  if (errmsg)
1397
    goto err;
1398 1399 1400 1401

  /* as we are here, we tried to allocate the event */
  if (!mi->rli.relay_log.description_event_for_queue)
  {
1402 1403 1404 1405
    errmsg= "default Format_description_log_event";
    err_code= ER_SLAVE_CREATE_EVENT_FAILURE;
    sprintf(err_buff, ER(err_code), errmsg);
    goto err;
1406 1407
  }

1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449
  /*
    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); 
1450 1451 1452 1453
  /*
    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
1454

1455
#ifdef ENABLED_DEBUG_SYNC
1456 1457 1458 1459 1460
  DBUG_EXECUTE_IF("dbug.before_get_UNIX_TIMESTAMP",
                  {
                    const char act[]=
                      "now "
                      "wait_for signal.get_unix_timestamp";
1461
                    DBUG_ASSERT(debug_sync_service);
1462 1463 1464
                    DBUG_ASSERT(!debug_sync_set_action(current_thd,
                                                       STRING_WITH_LEN(act)));
                  };);
1465
#endif
1466

1467
  master_res= NULL;
1468
  if (!mysql_real_query(mysql, STRING_WITH_LEN("SELECT UNIX_TIMESTAMP()")) &&
1469 1470
      (master_res= mysql_store_result(mysql)) &&
      (master_row= mysql_fetch_row(master_res)))
unknown's avatar
unknown committed
1471
  {
unknown's avatar
unknown committed
1472
    mi->clock_diff_with_master=
1473
      (long) (time((time_t*) 0) - strtoul(master_row[0], 0, 10));
unknown's avatar
unknown committed
1474
  }
Michael Widenius's avatar
Michael Widenius committed
1475
  else if (check_io_slave_killed(mi, NULL))
1476
    goto slave_killed_err;
1477 1478 1479 1480 1481 1482 1483
  else if (is_network_error(mysql_errno(mysql)))
  {
    mi->report(WARNING_LEVEL, mysql_errno(mysql),
               "Get master clock failed with error: %s", mysql_error(mysql));
    goto network_err;
  }
  else 
unknown's avatar
unknown committed
1484
  {
1485
    mi->clock_diff_with_master= 0; /* The "most sensible" value */
1486 1487 1488 1489
    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));
1490 1491
  }
  if (master_res)
1492
  {
unknown's avatar
unknown committed
1493
    mysql_free_result(master_res);
1494 1495
    master_res= NULL;
  }
unknown's avatar
unknown committed
1496

1497 1498 1499 1500 1501 1502 1503 1504 1505 1506
  /*
    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.
  */
1507
#ifdef ENABLED_DEBUG_SYNC
1508 1509 1510 1511 1512
  DBUG_EXECUTE_IF("dbug.before_get_SERVER_ID",
                  {
                    const char act[]=
                      "now "
                      "wait_for signal.get_server_id";
1513
                    DBUG_ASSERT(debug_sync_service);
1514 1515 1516
                    DBUG_ASSERT(!debug_sync_set_action(current_thd, 
                                                       STRING_WITH_LEN(act)));
                  };);
1517
#endif
1518 1519
  master_res= NULL;
  master_row= NULL;
1520 1521
  if (!mysql_real_query(mysql,
                        STRING_WITH_LEN("SHOW VARIABLES LIKE 'SERVER_ID'")) &&
1522 1523
      (master_res= mysql_store_result(mysql)) &&
      (master_row= mysql_fetch_row(master_res)))
1524
  {
1525 1526
    if ((global_system_variables.server_id ==
             (mi->master_id= strtoul(master_row[1], 0, 10))) &&
1527
        !mi->rli.replicate_same_server_id)
1528
    {
1529 1530 1531 1532
      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).";
1533 1534
      err_code= ER_SLAVE_FATAL_ERROR;
      sprintf(err_buff, ER(err_code), errmsg);
1535
      goto err;
1536
    }
1537 1538 1539
  }
  else if (mysql_errno(mysql))
  {
Michael Widenius's avatar
Michael Widenius committed
1540
    if (check_io_slave_killed(mi, NULL))
1541 1542
      goto slave_killed_err;
    else if (is_network_error(mysql_errno(mysql)))
1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562
    {
      mi->report(WARNING_LEVEL, mysql_errno(mysql),
                 "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)
  {
    mi->report(WARNING_LEVEL, ER_UNKNOWN_SYSTEM_VARIABLE,
               "Unknown system variable 'SERVER_ID' on master, \
maybe it is a *VERY OLD MASTER*.");
  }
  if (master_res)
  {
1563
    mysql_free_result(master_res);
1564
    master_res= NULL;
unknown's avatar
unknown committed
1565
  }
Andrei Elkin's avatar
Andrei Elkin committed
1566 1567 1568 1569 1570 1571 1572
  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
1573

1574 1575 1576
  /*
    Check that the master's global character_set_server and ours are the same.
    Not fatal if query fails (old master?).
1577 1578 1579 1580 1581 1582
    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.
1583 1584 1585 1586 1587 1588
    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.
1589
  */
1590 1591

  /* redundant with rest of code but safer against later additions */
1592
  if (version == 3)
1593
    goto err;
1594

1595
  if (version == 4)
unknown's avatar
unknown committed
1596
  {
1597 1598 1599 1600 1601
    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)))
1602
    {
1603 1604 1605
      if (strcmp(master_row[0], global_system_variables.collation_server->name))
      {
        errmsg= "The slave I/O thread stops because master and slave have \
1606
different values for the COLLATION_SERVER global variable. The values must \
1607 1608 1609 1610 1611
be equal for the Statement-format replication to work";
        err_code= ER_SLAVE_FATAL_ERROR;
        sprintf(err_buff, ER(err_code), errmsg);
        goto err;
      }
1612
    }
Michael Widenius's avatar
Michael Widenius committed
1613
    else if (check_io_slave_killed(mi, NULL))
1614
      goto slave_killed_err;
1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627
    else if (is_network_error(mysql_errno(mysql)))
    {
      mi->report(WARNING_LEVEL, mysql_errno(mysql),
                 "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));
1628
      goto err;
1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640
    }
    else
      mi->report(WARNING_LEVEL, ER_UNKNOWN_SYSTEM_VARIABLE,
                 "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
1641
  }
1642

1643 1644 1645 1646 1647 1648 1649 1650
  /*
    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
1651 1652 1653
    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
1654
    'SYSTEM' time zone).
1655 1656
    This check is only necessary for 4.x masters (and < 5.0.4 masters but
    those were alpha).
1657
  */
1658
  if (version == 4)
unknown's avatar
unknown committed
1659
  {
1660 1661 1662 1663
    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)))
1664
    {
1665 1666 1667 1668
      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 \
1669
different values for the TIME_ZONE global variable. The values must \
1670 1671 1672 1673 1674
be equal for the Statement-format replication to work";
        err_code= ER_SLAVE_FATAL_ERROR;
        sprintf(err_buff, ER(err_code), errmsg);
        goto err;
      }
1675
    }
Michael Widenius's avatar
Michael Widenius committed
1676
    else if (check_io_slave_killed(mi, NULL))
1677
      goto slave_killed_err;
1678
    else if (is_network_error(err_code= mysql_errno(mysql)))
1679
    {
1680 1681 1682
      mi->report(ERROR_LEVEL, err_code,
                 "Get master TIME_ZONE failed with error: %s",
                 mysql_error(mysql));
1683
      goto network_err;
1684 1685 1686 1687 1688 1689 1690 1691 1692 1693
    }
    else if (err_code == ER_UNKNOWN_SYSTEM_VARIABLE)
    {
      /* We use ERROR_LEVEL to get the error logged to file */
      mi->report(ERROR_LEVEL, err_code,

                 "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");
    }
1694 1695 1696 1697 1698 1699
    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));
1700
      goto err;
1701 1702 1703 1704 1705 1706
    }
    if (master_res)
    {
      mysql_free_result(master_res);
      master_res= NULL;
    }
unknown's avatar
unknown committed
1707 1708
  }

Andrei Elkin's avatar
Andrei Elkin committed
1709 1710 1711 1712 1713 1714 1715 1716 1717
  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);
1718
    sprintf(query, query_format, llbuf);
Andrei Elkin's avatar
Andrei Elkin committed
1719 1720

    if (mysql_real_query(mysql, query, strlen(query))
Michael Widenius's avatar
Michael Widenius committed
1721
        && !check_io_slave_killed(mi, NULL))
Andrei Elkin's avatar
Andrei Elkin committed
1722
    {
Andrei Elkin's avatar
Andrei Elkin committed
1723 1724
      errmsg= "The slave I/O thread stops because SET @master_heartbeat_period "
        "on master failed.";
Andrei Elkin's avatar
Andrei Elkin committed
1725
      err_code= ER_SLAVE_FATAL_ERROR;
Andrei Elkin's avatar
Andrei Elkin committed
1726
      sprintf(err_buff, "%s Error: %s", errmsg, mysql_error(mysql));
Andrei Elkin's avatar
Andrei Elkin committed
1727 1728 1729 1730 1731 1732
      mysql_free_result(mysql_store_result(mysql));
      goto err;
    }
    mysql_free_result(mysql_store_result(mysql));
  }
 
1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755
  /*
    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
1756
      if (check_io_slave_killed(mi, NULL))
1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801
        goto slave_killed_err;

      if (mysql_errno(mysql) == ER_UNKNOWN_SYSTEM_VARIABLE)
      {
        // this is tolerable as OM -> NS is supported
        mi->report(WARNING_LEVEL, mysql_errno(mysql),
                   "Notifying master by %s failed with "
                   "error: %s", query, mysql_error(mysql));
      }
      else
      {
        if (is_network_error(mysql_errno(mysql)))
        {
          mi->report(WARNING_LEVEL, mysql_errno(mysql),
                     "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
1802
      else if (check_io_slave_killed(mi, NULL))
1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829
        goto slave_killed_err;
      else if (is_network_error(mysql_errno(mysql)))
      {
        mi->report(WARNING_LEVEL, mysql_errno(mysql),
                   "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
1830

1831
  /*
1832
    Request the master to filter away events with the @@skip_replication flag
1833 1834
    set, if we are running with
    --replicate-events-marked-for-skip=FILTER_ON_MASTER.
1835
  */
1836
  if (opt_replicate_events_marked_for_skip == RPL_SKIP_FILTER_ON_MASTER)
1837
  {
1838
    if (mysql_real_query(mysql, STRING_WITH_LEN("SET skip_replication=1")))
1839 1840 1841 1842 1843
    {
      err_code= mysql_errno(mysql);
      if (is_network_error(err_code))
      {
        mi->report(ERROR_LEVEL, err_code,
1844
                   "Setting master-side filtering of @@skip_replication failed "
1845 1846 1847 1848 1849 1850 1851
                   "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
1852
          @@skip_replication feature.
1853
          This is not a problem, as such master will not generate events with
1854
          the @@skip_replication flag set in the first place. We will still
1855 1856
          do slave-side filtering of such events though, to handle the (rare)
          case of downgrading a master and receiving old events generated from
1857
          before the downgrade with the @@skip_replication flag set.
1858 1859
        */
        DBUG_PRINT("info", ("Old master does not support master-side filtering "
1860 1861 1862 1863 1864 1865 1866 1867 1868 1869
                            "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;
1870 1871 1872
      }
    }
  }
1873 1874 1875 1876

  /* Announce MariaDB slave capabilities. */
  DBUG_EXECUTE_IF("simulate_slave_capability_none", goto after_set_capability;);
  {
1877 1878 1879 1880 1881 1882
    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)
1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901
    {
      err_code= mysql_errno(mysql);
      if (is_network_error(err_code))
      {
        mi->report(ERROR_LEVEL, err_code,
                   "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;
      }
    }
  }
1902
#ifndef DBUG_OFF
1903
after_set_capability:
1904
#endif
1905

unknown's avatar
unknown committed
1906
  if (mi->using_gtid != Master_info::USE_GTID_NO)
1907
  {
unknown's avatar
unknown committed
1908
    /* Request dump to start from slave replication GTID state. */
1909 1910
    int rc;
    char str_buf[256];
1911 1912
    String query_str(str_buf, sizeof(str_buf), system_charset_info);
    query_str.length(0);
unknown's avatar
unknown committed
1913

unknown's avatar
unknown committed
1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934
    /*
      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);
      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;
    }
    mysql_free_result(master_res);
    master_res= NULL;

1935 1936
    query_str.append(STRING_WITH_LEN("SET @slave_connect_state='"),
                     system_charset_info);
unknown's avatar
unknown committed
1937
    if (mi->gtid_current_pos.append_to_string(&query_str))
unknown's avatar
unknown committed
1938 1939 1940 1941 1942 1943 1944
    {
      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;
    }
1945
    query_str.append(STRING_WITH_LEN("'"), system_charset_info);
unknown's avatar
unknown committed
1946

1947
    rc= mysql_real_query(mysql, query_str.ptr(), query_str.length());
1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966
    if (rc)
    {
      err_code= mysql_errno(mysql);
      if (is_network_error(err_code))
      {
        mi->report(ERROR_LEVEL, err_code,
                   "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
1967

1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000
    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))
      {
        mi->report(ERROR_LEVEL, err_code,
                   "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;
      }
    }

unknown's avatar
unknown committed
2001 2002
    if (mi->rli.until_condition == Relay_log_info::UNTIL_GTID)
    {
2003 2004 2005 2006
      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
2007 2008 2009 2010 2011 2012 2013
      {
        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;
      }
2014
      query_str.append(STRING_WITH_LEN("'"), system_charset_info);
unknown's avatar
unknown committed
2015

2016
      rc= mysql_real_query(mysql, query_str.ptr(), query_str.length());
unknown's avatar
unknown committed
2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036
      if (rc)
      {
        err_code= mysql_errno(mysql);
        if (is_network_error(err_code))
        {
          mi->report(ERROR_LEVEL, err_code,
                     "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;
        }
      }
    }
2037
  }
unknown's avatar
unknown committed
2038
  else
unknown's avatar
unknown committed
2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062
  {
    /*
      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
2063
      rpl_global_gtid_slave_state.load(mi->io_thd, master_row[0],
unknown's avatar
unknown committed
2064
                                       strlen(master_row[0]), false, false);
unknown's avatar
unknown committed
2065
    }
Michael Widenius's avatar
Michael Widenius committed
2066
    else if (check_io_slave_killed(mi, NULL))
unknown's avatar
unknown committed
2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092
      goto slave_killed_err;
    else if (is_network_error(mysql_errno(mysql)))
    {
      mi->report(WARNING_LEVEL, mysql_errno(mysql),
                 "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;
    }
  }
2093

2094
err:
2095 2096
  if (errmsg)
  {
2097 2098
    if (master_res)
      mysql_free_result(master_res);
2099
    DBUG_ASSERT(err_code != 0);
2100
    mi->report(ERROR_LEVEL, err_code, "%s", err_buff);
2101
    DBUG_RETURN(1);
2102
  }
2103

2104
  DBUG_RETURN(0);
2105 2106 2107 2108 2109

network_err:
  if (master_res)
    mysql_free_result(master_res);
  DBUG_RETURN(2);
unknown's avatar
unknown committed
2110

2111 2112 2113 2114
slave_killed_err:
  if (master_res)
    mysql_free_result(master_res);
  DBUG_RETURN(2);
unknown's avatar
unknown committed
2115
}
unknown's avatar
unknown committed
2116

2117

2118
static bool wait_for_relay_log_space(Relay_log_info* rli)
unknown's avatar
unknown committed
2119
{
2120
  bool slave_killed=0;
2121
  Master_info* mi = rli->mi;
unknown's avatar
unknown committed
2122
  const char *save_proc_info;
unknown's avatar
unknown committed
2123 2124
  THD* thd = mi->io_thd;
  DBUG_ENTER("wait_for_relay_log_space");
2125

Marc Alff's avatar
Marc Alff committed
2126
  mysql_mutex_lock(&rli->log_space_lock);
unknown's avatar
unknown committed
2127
  save_proc_info= thd->enter_cond(&rli->log_space_cond,
unknown's avatar
unknown committed
2128 2129
                                  &rli->log_space_lock,
                                  "\
2130
Waiting for the slave SQL thread to free enough relay log space");
unknown's avatar
unknown committed
2131
  while (rli->log_space_limit < rli->log_space_total &&
Michael Widenius's avatar
Michael Widenius committed
2132
         !(slave_killed=io_slave_killed(mi)) &&
2133
         !rli->ignore_log_space_limit)
Marc Alff's avatar
Marc Alff committed
2134
    mysql_cond_wait(&rli->log_space_cond, &rli->log_space_lock);
2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175

  /* 
    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.

   */
  if (rli->ignore_log_space_limit)
  {
#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)
    {
2176
      mysql_mutex_lock(&mi->data_lock);
2177
      rotate_relay_log(rli->mi);
2178
      mysql_mutex_unlock(&mi->data_lock);
2179 2180 2181 2182 2183 2184
      rli->sql_force_rotate_relay= false;
    }

    rli->ignore_log_space_limit= false;
  }

unknown's avatar
unknown committed
2185
  thd->exit_cond(save_proc_info);
unknown's avatar
unknown committed
2186 2187
  DBUG_RETURN(slave_killed);
}
2188

2189

2190 2191
/*
  Builds a Rotate from the ignored events' info and writes it to relay log.
2192

2193 2194 2195 2196 2197 2198 2199 2200 2201 2202
  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).
 */
2203
static void write_ignored_events_info_to_relay_log(THD *thd, Master_info *mi)
2204
{
2205
  Relay_log_info *rli= &mi->rli;
Marc Alff's avatar
Marc Alff committed
2206
  mysql_mutex_t *log_lock= rli->relay_log.get_log_lock();
2207 2208
  DBUG_ENTER("write_ignored_events_info_to_relay_log");

2209
  DBUG_ASSERT(thd == mi->io_thd);
Marc Alff's avatar
Marc Alff committed
2210
  mysql_mutex_lock(log_lock);
2211
  if (rli->ign_master_log_name_end[0])
unknown's avatar
unknown committed
2212
  {
2213
    DBUG_PRINT("info",("writing a Rotate event to track down ignored events"));
2214
    Rotate_log_event *ev= new Rotate_log_event(rli->ign_master_log_name_end,
2215 2216 2217 2218
                                               0, rli->ign_master_log_pos_end,
                                               Rotate_log_event::DUP_NAME);
    rli->ign_master_log_name_end[0]= 0;
    /* can unlock before writing as slave SQL thd will soon see our Rotate */
Marc Alff's avatar
Marc Alff committed
2219
    mysql_mutex_unlock(log_lock);
2220 2221 2222 2223
    if (likely((bool)ev))
    {
      ev->server_id= 0; // don't be ignored by slave SQL thread
      if (unlikely(rli->relay_log.append(ev)))
2224 2225 2226 2227 2228
        mi->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_WRITE_FAILURE,
                   ER(ER_SLAVE_RELAY_LOG_WRITE_FAILURE),
                   "failed to write a Rotate event"
                   " to the relay log, SHOW SLAVE STATUS may be"
                   " inaccurate");
2229
      rli->relay_log.harvest_bytes_written(&rli->log_space_total);
Luis Soares's avatar
Luis Soares committed
2230
      if (flush_master_info(mi, TRUE, TRUE))
2231
        sql_print_error("Failed to flush master info file");
2232 2233 2234
      delete ev;
    }
    else
2235 2236 2237 2238
      mi->report(ERROR_LEVEL, ER_SLAVE_CREATE_EVENT_FAILURE,
                 ER(ER_SLAVE_CREATE_EVENT_FAILURE),
                 "Rotate_event (out of memory?),"
                 " SHOW SLAVE STATUS may be inaccurate");
unknown's avatar
unknown committed
2239
  }
2240
  else
Marc Alff's avatar
Marc Alff committed
2241
    mysql_mutex_unlock(log_lock);
2242
  DBUG_VOID_RETURN;
unknown's avatar
unknown committed
2243 2244
}

2245

2246
int register_slave_on_master(MYSQL* mysql, Master_info *mi,
2247
                             bool *suppress_warnings)
2248
{
2249
  uchar buf[1024], *pos= buf;
2250
  uint report_host_len=0, report_user_len=0, report_password_len=0;
2251
  DBUG_ENTER("register_slave_on_master");
2252

2253
  *suppress_warnings= FALSE;
2254 2255 2256 2257 2258 2259 2260 2261
  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);
2262
    DBUG_RETURN(0);
2263 2264
  }

2265
  if (report_user)
2266
    report_user_len= strlen(report_user);
2267 2268 2269 2270 2271 2272 2273 2274 2275
  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
2276
  if (report_password)
2277
    report_password_len= strlen(report_password);
2278 2279 2280 2281 2282 2283 2284 2285
  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);
  }
2286

2287
  int4store(pos, global_system_variables.server_id); pos+= 4;
2288 2289 2290
  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);
2291
  int2store(pos, (uint16) report_port); pos+= 2;
2292 2293 2294 2295 2296 2297
  /* 
    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;
2298
  /* The master will fill in master_id */
unknown's avatar
unknown committed
2299
  int4store(pos, 0);                    pos+= 4;
2300

2301
  if (simple_command(mysql, COM_REGISTER_SLAVE, buf, (size_t) (pos- buf), 0))
2302
  {
2303 2304 2305 2306
    if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
    {
      *suppress_warnings= TRUE;                 // Suppress reconnect warning
    }
Michael Widenius's avatar
Michael Widenius committed
2307
    else if (!check_io_slave_killed(mi, NULL))
2308 2309 2310 2311 2312 2313 2314
    {
      char buf[256];
      my_snprintf(buf, sizeof(buf), "%s (Errno: %d)", mysql_error(mysql), 
                  mysql_errno(mysql));
      mi->report(ERROR_LEVEL, ER_SLAVE_MASTER_COM_FAILURE,
                 ER(ER_SLAVE_MASTER_COM_FAILURE), "COM_REGISTER_SLAVE", buf);
    }
2315
    DBUG_RETURN(1);
2316
  }
2317
  DBUG_RETURN(0);
2318 2319
}

2320

2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331
/**
  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
*/
2332 2333 2334 2335

bool show_master_info(THD *thd, Master_info *mi, bool full)
{
  DBUG_ENTER("show_master_info");
unknown's avatar
unknown committed
2336
  String gtid_pos;
2337

unknown's avatar
unknown committed
2338
  if (full && rpl_global_gtid_slave_state.tostring(&gtid_pos, NULL, 0))
2339
    DBUG_RETURN(TRUE);
unknown's avatar
unknown committed
2340 2341 2342
  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))
2343 2344 2345 2346 2347
    DBUG_RETURN(TRUE);
  my_eof(thd);
  DBUG_RETURN(FALSE);
}

unknown's avatar
unknown committed
2348 2349
static bool send_show_master_info_header(THD *thd, bool full,
                                         size_t gtid_pos_length)
unknown's avatar
unknown committed
2350 2351
{
  List<Item> field_list;
2352
  Protocol *protocol= thd->protocol;
2353
  Master_info *mi;
2354 2355 2356
  DBUG_ENTER("show_master_info_header");

  if (full)
2357
  {
2358 2359
    field_list.push_back(new Item_empty_string("Connection_name",
                                               MAX_CONNECTION_NAME));
2360 2361 2362
    field_list.push_back(new Item_empty_string("Slave_SQL_State",
                                               30));
  }
2363

unknown's avatar
unknown committed
2364
  field_list.push_back(new Item_empty_string("Slave_IO_State",
2365
                                                     30));
unknown's avatar
unknown committed
2366
  field_list.push_back(new Item_empty_string("Master_Host",
unknown's avatar
unknown committed
2367
                                                     sizeof(mi->host)));
unknown's avatar
unknown committed
2368
  field_list.push_back(new Item_empty_string("Master_User",
unknown's avatar
unknown committed
2369
                                                     sizeof(mi->user)));
2370
  field_list.push_back(new Item_return_int("Master_Port", 7,
unknown's avatar
unknown committed
2371
                                           MYSQL_TYPE_LONG));
2372
  field_list.push_back(new Item_return_int("Connect_Retry", 10,
unknown's avatar
unknown committed
2373
                                           MYSQL_TYPE_LONG));
2374
  field_list.push_back(new Item_empty_string("Master_Log_File",
unknown's avatar
unknown committed
2375
                                             FN_REFLEN));
2376
  field_list.push_back(new Item_return_int("Read_Master_Log_Pos", 10,
unknown's avatar
unknown committed
2377
                                           MYSQL_TYPE_LONGLONG));
2378
  field_list.push_back(new Item_empty_string("Relay_Log_File",
unknown's avatar
unknown committed
2379
                                             FN_REFLEN));
2380
  field_list.push_back(new Item_return_int("Relay_Log_Pos", 10,
unknown's avatar
unknown committed
2381
                                           MYSQL_TYPE_LONGLONG));
2382
  field_list.push_back(new Item_empty_string("Relay_Master_Log_File",
unknown's avatar
unknown committed
2383
                                             FN_REFLEN));
2384 2385
  field_list.push_back(new Item_empty_string("Slave_IO_Running", 3));
  field_list.push_back(new Item_empty_string("Slave_SQL_Running", 3));
2386 2387 2388 2389 2390 2391
  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
2392
                                             28));
2393 2394 2395
  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
2396
                                           MYSQL_TYPE_LONG));
2397
  field_list.push_back(new Item_return_int("Exec_Master_Log_Pos", 10,
unknown's avatar
unknown committed
2398
                                           MYSQL_TYPE_LONGLONG));
2399
  field_list.push_back(new Item_return_int("Relay_Log_Space", 10,
unknown's avatar
unknown committed
2400
                                           MYSQL_TYPE_LONGLONG));
2401
  field_list.push_back(new Item_empty_string("Until_Condition", 6));
2402
  field_list.push_back(new Item_empty_string("Until_Log_File", FN_REFLEN));
unknown's avatar
unknown committed
2403
  field_list.push_back(new Item_return_int("Until_Log_Pos", 10,
2404
                                           MYSQL_TYPE_LONGLONG));
unknown's avatar
unknown committed
2405 2406 2407
  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
2408
  field_list.push_back(new Item_empty_string("Master_SSL_CA_Path",
unknown's avatar
unknown committed
2409
                                             sizeof(mi->ssl_capath)));
unknown's avatar
unknown committed
2410
  field_list.push_back(new Item_empty_string("Master_SSL_Cert",
unknown's avatar
unknown committed
2411
                                             sizeof(mi->ssl_cert)));
unknown's avatar
unknown committed
2412
  field_list.push_back(new Item_empty_string("Master_SSL_Cipher",
unknown's avatar
unknown committed
2413
                                             sizeof(mi->ssl_cipher)));
unknown's avatar
unknown committed
2414
  field_list.push_back(new Item_empty_string("Master_SSL_Key",
unknown's avatar
unknown committed
2415
                                             sizeof(mi->ssl_key)));
2416
  field_list.push_back(new Item_return_int("Seconds_Behind_Master", 10,
unknown's avatar
unknown committed
2417
                                           MYSQL_TYPE_LONGLONG));
2418 2419
  field_list.push_back(new Item_empty_string("Master_SSL_Verify_Server_Cert",
                                             3));
2420 2421 2422 2423
  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
2424 2425 2426 2427
  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));
unknown's avatar
unknown committed
2428 2429
  field_list.push_back(new Item_empty_string("Using_Gtid",
                                             sizeof("Current_Pos")-1));
2430 2431 2432 2433 2434 2435 2436 2437
  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));
2438 2439 2440 2441
    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
2442 2443
    field_list.push_back(new Item_empty_string("Gtid_Slave_Pos",
                                               gtid_pos_length));
2444
  }
2445
  if (protocol->send_result_set_metadata(&field_list,
2446
                            Protocol::SEND_NUM_ROWS | Protocol::SEND_EOF))
unknown's avatar
unknown committed
2447
    DBUG_RETURN(TRUE);
2448 2449 2450 2451
  DBUG_RETURN(FALSE);
}


unknown's avatar
unknown committed
2452 2453
static bool send_show_master_info_data(THD *thd, Master_info *mi, bool full,
                                       String *gtid_pos)
2454 2455
{
  DBUG_ENTER("send_show_master_info_data");
unknown's avatar
unknown committed
2456

2457 2458
  if (mi->host[0])
  {
2459
    DBUG_PRINT("info",("host is set: '%s'", mi->host));
2460
    String *packet= &thd->packet;
2461
    Protocol *protocol= thd->protocol;
2462
    Rpl_filter *rpl_filter= mi->rpl_filter;
2463 2464
    char buf[256];
    String tmp(buf, sizeof(buf), &my_charset_bin);
2465

2466
    protocol->prepare_for_resend();
unknown's avatar
unknown committed
2467

2468
    /*
2469 2470
      slave_running can be accessed without run_lock but not other
      non-volotile members like mi->io_thd, which is guarded by the mutex.
2471
    */
2472 2473 2474
    if (full)
      protocol->store(mi->connection_name.str, mi->connection_name.length,
                      &my_charset_bin);
Marc Alff's avatar
Marc Alff committed
2475
    mysql_mutex_lock(&mi->run_lock);
2476
    if (full)
Michael Widenius's avatar
Michael Widenius committed
2477 2478 2479 2480 2481 2482 2483
    {
      /*
        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 ?
                      mi->rli.sql_driver_thd->proc_info : "",
2484
                      &my_charset_bin);
Michael Widenius's avatar
Michael Widenius committed
2485
    }
2486
    protocol->store(mi->io_thd ? mi->io_thd->proc_info : "", &my_charset_bin);
Marc Alff's avatar
Marc Alff committed
2487
    mysql_mutex_unlock(&mi->run_lock);
2488

Marc Alff's avatar
Marc Alff committed
2489 2490 2491 2492
    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);
2493 2494
    protocol->store(mi->host, &my_charset_bin);
    protocol->store(mi->user, &my_charset_bin);
2495 2496
    protocol->store((uint32) mi->port);
    protocol->store((uint32) mi->connect_retry);
2497
    protocol->store(mi->master_log_name, &my_charset_bin);
2498
    protocol->store((ulonglong) mi->master_log_pos);
2499
    protocol->store(mi->rli.group_relay_log_name +
unknown's avatar
unknown committed
2500 2501
                    dirname_length(mi->rli.group_relay_log_name),
                    &my_charset_bin);
2502 2503
    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
2504
    protocol->store(mi->slave_running == MYSQL_SLAVE_RUN_CONNECT ?
2505 2506
                    "Yes" : (mi->slave_running == MYSQL_SLAVE_RUN_NOT_CONNECT ?
                             "Connecting" : "No"), &my_charset_bin);
2507
    protocol->store(mi->rli.slave_running ? "Yes":"No", &my_charset_bin);
2508 2509 2510 2511
    protocol->store(rpl_filter->get_do_db());
    protocol->store(rpl_filter->get_ignore_db());

    rpl_filter->get_do_table(&tmp);
2512
    protocol->store(&tmp);
2513
    rpl_filter->get_ignore_table(&tmp);
2514
    protocol->store(&tmp);
2515
    rpl_filter->get_wild_do_table(&tmp);
2516
    protocol->store(&tmp);
2517
    rpl_filter->get_wild_ignore_table(&tmp);
2518 2519
    protocol->store(&tmp);

2520 2521
    protocol->store(mi->rli.last_error().number);
    protocol->store(mi->rli.last_error().message, &my_charset_bin);
2522
    protocol->store((uint32) mi->rli.slave_skip_counter);
2523
    protocol->store((ulonglong) mi->rli.group_master_log_pos);
2524
    protocol->store((ulonglong) mi->rli.log_space_total);
2525 2526

    protocol->store(
2527 2528
      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
2529 2530
          ( mi->rli.until_condition==Relay_log_info::UNTIL_RELAY_POS? "Relay":
            "Gtid")), &my_charset_bin);
2531 2532
    protocol->store(mi->rli.until_log_name, &my_charset_bin);
    protocol->store((ulonglong) mi->rli.until_log_pos);
unknown's avatar
unknown committed
2533 2534

#ifdef HAVE_OPENSSL
unknown's avatar
unknown committed
2535 2536 2537 2538 2539 2540 2541 2542 2543
    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
2544

2545 2546 2547 2548 2549 2550
    /*
      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)
2551
    {
2552
      long time_diff= ((long)(time(0) - mi->rli.last_master_timestamp)
unknown's avatar
unknown committed
2553
                       - mi->clock_diff_with_master);
2554
      /*
unknown's avatar
unknown committed
2555 2556
        Apparently on some systems time_diff can be <0. Here are possible
        reasons related to MySQL:
2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568
        - 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.
2569 2570 2571 2572
        This confuses users, so we don't go below 0: hence the max().

        last_master_timestamp == 0 (an "impossible" timestamp 1970) is a
        special marker to say "consider we have caught up".
2573
      */
unknown's avatar
unknown committed
2574 2575
      protocol->store((longlong)(mi->rli.last_master_timestamp ?
                                 max(0, time_diff) : 0));
2576
    }
unknown's avatar
unknown committed
2577
    else
2578
    {
unknown's avatar
unknown committed
2579
      protocol->store_null();
2580 2581
    }
    protocol->store(mi->ssl_verify_server_cert? "Yes":"No", &my_charset_bin);
unknown's avatar
unknown committed
2582

2583
    // Last_IO_Errno
2584
    protocol->store(mi->last_error().number);
2585
    // Last_IO_Error
2586
    protocol->store(mi->last_error().message, &my_charset_bin);
2587
    // Last_SQL_Errno
2588
    protocol->store(mi->rli.last_error().number);
2589
    // Last_SQL_Error
2590
    protocol->store(mi->rli.last_error().message, &my_charset_bin);
Andrei Elkin's avatar
Andrei Elkin committed
2591 2592 2593 2594 2595 2596 2597 2598 2599 2600
    // 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);
2601
        slen= sprintf(sbuff, (i==0? "%lu" : ", %lu"), s_id);
Andrei Elkin's avatar
Andrei Elkin committed
2602 2603 2604 2605 2606 2607
        if (cur_len + slen + 4 > FN_REFLEN)
        {
          /*
            break the loop whenever remained space could not fit
            ellipses on the next cycle
          */
2608
          sprintf(buff + cur_len, "...");
Andrei Elkin's avatar
Andrei Elkin committed
2609 2610
          break;
        }
2611
        cur_len += sprintf(buff + cur_len, "%s", sbuff);
Andrei Elkin's avatar
Andrei Elkin committed
2612 2613 2614 2615 2616
      }
      protocol->store(buff, &my_charset_bin);
    }
    // Master_Server_id
    protocol->store((uint32) mi->master_id);
unknown's avatar
unknown committed
2617
    protocol->store(mi->using_gtid_astext(mi->using_gtid), &my_charset_bin);
2618 2619 2620 2621 2622
    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);
2623 2624
      protocol->store((uint32)    mi->received_heartbeats);
      protocol->store((double)    mi->heartbeat_period, 3, &tmp);
unknown's avatar
unknown committed
2625
      protocol->store(gtid_pos->ptr(), gtid_pos->length(), &my_charset_bin);
2626
    }
2627

Marc Alff's avatar
Marc Alff committed
2628 2629 2630 2631
    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);
2632

2633
    if (my_net_write(&thd->net, (uchar*) thd->packet.ptr(), packet->length()))
unknown's avatar
unknown committed
2634
      DBUG_RETURN(TRUE);
2635
  }
2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659
  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
2660 2661 2662

  @note
  master_info_index is protected by LOCK_active_mi.
2663 2664 2665 2666 2667
*/

bool show_all_master_info(THD* thd)
{
  uint i, elements;
unknown's avatar
unknown committed
2668
  String gtid_pos;
2669 2670
  Master_info **tmp;
  DBUG_ENTER("show_master_info");
2671
  mysql_mutex_assert_owner(&LOCK_active_mi);
2672

unknown's avatar
unknown committed
2673 2674 2675 2676 2677 2678
  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
2679 2680

  if (send_show_master_info_header(thd, 1, gtid_pos.length()))
2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701
    DBUG_RETURN(TRUE);

  if (!(elements= master_info_index->master_info_hash.records))
    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
2702
    if (send_show_master_info_data(thd, tmp[i], 1, &gtid_pos))
2703 2704 2705 2706
      DBUG_RETURN(TRUE);
  }

end:
2707
  my_eof(thd);
unknown's avatar
unknown committed
2708
  DBUG_RETURN(FALSE);
unknown's avatar
unknown committed
2709 2710
}

unknown's avatar
unknown committed
2711

2712 2713
void set_slave_thread_options(THD* thd)
{
2714
  DBUG_ENTER("set_slave_thread_options");
2715 2716 2717 2718 2719 2720 2721 2722 2723
  /*
     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.
  */
2724
  ulonglong options= thd->variables.option_bits | OPTION_BIG_SELECTS;
2725 2726 2727 2728
  if (opt_log_slave_updates)
    options|= OPTION_BIN_LOG;
  else
    options&= ~OPTION_BIN_LOG;
2729
  thd->variables.option_bits= options;
2730
  thd->variables.completion_type= 0;
2731
  DBUG_VOID_RETURN;
2732
}
2733

2734
void set_slave_thread_default_charset(THD* thd, Relay_log_info const *rli)
2735
{
2736 2737
  DBUG_ENTER("set_slave_thread_default_charset");

2738 2739 2740 2741 2742 2743 2744
  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();
2745 2746 2747 2748 2749 2750 2751

  /*
    We use a const cast here since the conceptual (and externally
    visible) behavior of the function is to set the default charset of
    the thread.  That the cache has to be invalidated is a secondary
    effect.
   */
2752
  const_cast<Relay_log_info*>(rli)->cached_charset_invalidate();
2753
  DBUG_VOID_RETURN;
2754 2755
}

unknown's avatar
unknown committed
2756
/*
2757
  init_slave_thread()
unknown's avatar
unknown committed
2758
*/
2759

2760 2761
static int init_slave_thread(THD* thd, Master_info *mi,
                             SLAVE_THD_TYPE thd_type)
unknown's avatar
unknown committed
2762 2763
{
  DBUG_ENTER("init_slave_thread");
Sergei Golubchik's avatar
Sergei Golubchik committed
2764
  int simulate_error __attribute__((unused))= 0;
2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777
  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);
  }

2778
  thd->system_thread = (thd_type == SLAVE_THD_SQL) ?
unknown's avatar
unknown committed
2779
    SYSTEM_THREAD_SLAVE_SQL : SYSTEM_THREAD_SLAVE_IO;
2780
  thd->security_ctx->skip_grants();
2781 2782 2783 2784 2785
/*
  Adding MAX_LOG_EVENT_HEADER_LEN to the max_allowed_packet on all
  slave threads, since a replication event can become this much larger
  than the corresponding packet (query) sent from client to master.
*/
2786
  thd->variables.max_allowed_packet= slave_max_allowed_packet;
2787 2788
  thd->slave_thread= 1;
  thd->connection_name= mi->connection_name;
2789
  thd->enable_slow_log= opt_log_slow_slave_statements;
2790
  thd->variables.log_slow_filter= global_system_variables.log_slow_filter;
2791
  set_slave_thread_options(thd);
unknown's avatar
unknown committed
2792
  thd->client_capabilities = CLIENT_LOCAL_FILES;
Marc Alff's avatar
Marc Alff committed
2793
  mysql_mutex_lock(&LOCK_thread_count);
unknown's avatar
unknown committed
2794
  thd->thread_id= thd->variables.pseudo_thread_id= thread_id++;
Marc Alff's avatar
Marc Alff committed
2795
  mysql_mutex_unlock(&LOCK_thread_count);
unknown's avatar
unknown committed
2796

2797
  if (thd_type == SLAVE_THD_SQL)
2798
    thd_proc_info(thd, "Waiting for the next event in relay log");
2799
  else
2800
    thd_proc_info(thd, "Waiting for master update");
unknown's avatar
unknown committed
2801
  thd->set_time();
2802 2803
  /* Do not use user-supplied timeout value for system threads. */
  thd->variables.lock_wait_timeout= LONG_TIMEOUT;
unknown's avatar
unknown committed
2804 2805 2806
  DBUG_RETURN(0);
}

2807 2808 2809 2810 2811 2812 2813
/*
  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.
2814

2815 2816 2817
  @retval True if the thread has been killed, false otherwise.
*/
template <typename killed_func, typename rpl_info>
Michael Widenius's avatar
Michael Widenius committed
2818 2819
static bool slave_sleep(THD *thd, time_t seconds,
                        killed_func func, rpl_info info)
unknown's avatar
unknown committed
2820
{
2821

2822 2823 2824
  bool ret;
  struct timespec abstime;
  const char *old_proc_info;
unknown's avatar
unknown committed
2825

2826 2827
  mysql_mutex_t *lock= &info->sleep_lock;
  mysql_cond_t *cond= &info->sleep_cond;
unknown's avatar
unknown committed
2828

2829 2830 2831 2832 2833
  /* Absolute system time at which the sleep time expires. */
  set_timespec(abstime, seconds);
  mysql_mutex_lock(lock);
  old_proc_info= thd->enter_cond(cond, lock, thd->proc_info);

Michael Widenius's avatar
Michael Widenius committed
2834
  while (! (ret= func(info)))
2835 2836 2837 2838
  {
    int error= mysql_cond_timedwait(cond, lock, &abstime);
    if (error == ETIMEDOUT || error == ETIME)
      break;
unknown's avatar
unknown committed
2839
  }
2840 2841 2842
  /* Implicitly unlocks the mutex. */
  thd->exit_cond(old_proc_info);
  return ret;
unknown's avatar
unknown committed
2843 2844
}

2845

He Zhenxing's avatar
He Zhenxing committed
2846 2847
static int request_dump(THD *thd, MYSQL* mysql, Master_info* mi,
			bool *suppress_warnings)
unknown's avatar
unknown committed
2848
{
2849
  uchar buf[FN_REFLEN + 10];
unknown's avatar
unknown committed
2850
  int len;
He Zhenxing's avatar
He Zhenxing committed
2851
  ushort binlog_flags = 0; // for now
2852
  char* logname = mi->master_log_name;
2853
  DBUG_ENTER("request_dump");
2854 2855
  
  *suppress_warnings= FALSE;
2856

2857
  if (opt_log_slave_updates && opt_replicate_annotate_row_events)
2858 2859
    binlog_flags|= BINLOG_SEND_ANNOTATE_ROWS_EVENT;

He Zhenxing's avatar
He Zhenxing committed
2860 2861 2862 2863 2864
  if (RUN_HOOK(binlog_relay_io,
               before_request_transmit,
               (thd, mi, binlog_flags)))
    DBUG_RETURN(1);
  
unknown's avatar
unknown committed
2865
  // TODO if big log files: Change next to int8store()
unknown's avatar
unknown committed
2866
  int4store(buf, (ulong) mi->master_log_pos);
unknown's avatar
unknown committed
2867
  int2store(buf + 4, binlog_flags);
2868
  int4store(buf + 6, global_system_variables.server_id);
unknown's avatar
unknown committed
2869
  len = (uint) strlen(logname);
2870
  memcpy(buf + 10, logname,len);
unknown's avatar
SCRUM  
unknown committed
2871
  if (simple_command(mysql, COM_BINLOG_DUMP, buf, len + 10, 1))
unknown's avatar
unknown committed
2872
  {
unknown's avatar
unknown committed
2873 2874 2875 2876 2877
    /*
      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
2878
    if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
2879
      *suppress_warnings= TRUE;                 // Suppress reconnect warning
unknown's avatar
unknown committed
2880
    else
2881
      sql_print_error("Error on COM_BINLOG_DUMP: %d  %s, will retry in %d secs",
unknown's avatar
unknown committed
2882
                      mysql_errno(mysql), mysql_error(mysql),
2883
                      mi->connect_retry);
2884
    DBUG_RETURN(1);
unknown's avatar
unknown committed
2885
  }
unknown's avatar
unknown committed
2886

2887
  DBUG_RETURN(0);
unknown's avatar
unknown committed
2888 2889
}

2890

unknown's avatar
unknown committed
2891
/*
2892
  Read one event from the master
unknown's avatar
unknown committed
2893

unknown's avatar
unknown committed
2894 2895
  SYNOPSIS
    read_event()
unknown's avatar
unknown committed
2896 2897 2898 2899 2900 2901
    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.
2902

unknown's avatar
unknown committed
2903
    RETURN VALUES
unknown's avatar
unknown committed
2904 2905
    'packet_error'      Error
    number              Length of packet
unknown's avatar
unknown committed
2906 2907
*/

2908
static ulong read_event(MYSQL* mysql, Master_info *mi, bool* suppress_warnings)
unknown's avatar
unknown committed
2909
{
2910
  ulong len;
2911
  DBUG_ENTER("read_event");
unknown's avatar
unknown committed
2912

2913
  *suppress_warnings= FALSE;
unknown's avatar
unknown committed
2914 2915 2916 2917
  /*
    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
2918
#ifndef DBUG_OFF
2919
  if (disconnect_slave_event_count && !(mi->events_till_disconnect--))
unknown's avatar
unknown committed
2920
    DBUG_RETURN(packet_error);
unknown's avatar
unknown committed
2921
#endif
unknown's avatar
unknown committed
2922

2923
  len = cli_safe_read(mysql);
2924
  if (len == packet_error || (long) len < 1)
unknown's avatar
unknown committed
2925
  {
unknown's avatar
SCRUM  
unknown committed
2926
    if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
unknown's avatar
unknown committed
2927 2928
    {
      /*
unknown's avatar
unknown committed
2929 2930 2931
        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
2932 2933 2934 2935
      */
      *suppress_warnings= TRUE;
    }
    else
2936
      sql_print_error("Error reading packet from server: %s ( server_errno=%d)",
unknown's avatar
unknown committed
2937
                      mysql_error(mysql), mysql_errno(mysql));
2938
    DBUG_RETURN(packet_error);
unknown's avatar
unknown committed
2939 2940
  }

2941 2942
  /* Check if eof packet */
  if (len < 8 && mysql->net.read_pos[0] == 254)
unknown's avatar
unknown committed
2943
  {
2944 2945
    sql_print_information("Slave: received end packet from server, apparent "
                          "master shutdown: %s",
unknown's avatar
unknown committed
2946
                     mysql_error(mysql));
2947
     DBUG_RETURN(packet_error);
unknown's avatar
unknown committed
2948
  }
unknown's avatar
unknown committed
2949

unknown's avatar
unknown committed
2950
  DBUG_PRINT("exit", ("len: %lu  net->read_pos[4]: %d",
unknown's avatar
unknown committed
2951 2952
                      len, mysql->net.read_pos[4]));
  DBUG_RETURN(len - 1);
unknown's avatar
unknown committed
2953 2954
}

unknown's avatar
unknown committed
2955 2956 2957 2958 2959 2960 2961 2962 2963
/*
  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.
*/
static int has_temporary_error(THD *thd)
{
2964 2965
  DBUG_ENTER("has_temporary_error");

2966
  DBUG_EXECUTE_IF("all_errors_are_temporary_errors",
Marc Alff's avatar
Marc Alff committed
2967
                  if (thd->stmt_da->is_error())
2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979
                  {
                    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
2980 2981 2982 2983 2984 2985

  /*
    Temporary error codes:
    currently, InnoDB deadlock detected by InnoDB or lock
    wait timeout (innodb_lock_wait_timeout exceeded
  */
Marc Alff's avatar
Marc Alff committed
2986 2987
  if (thd->stmt_da->sql_errno() == ER_LOCK_DEADLOCK ||
      thd->stmt_da->sql_errno() == ER_LOCK_WAIT_TIMEOUT)
2988
    DBUG_RETURN(1);
unknown's avatar
unknown committed
2989 2990 2991 2992 2993

#ifdef HAVE_NDB_BINLOG
  /*
    currently temporary error set in ndbcluster
  */
Marc Alff's avatar
Marc Alff committed
2994
  List_iterator_fast<MYSQL_ERROR> it(thd->warning_info->warn_list());
unknown's avatar
unknown committed
2995 2996 2997
  MYSQL_ERROR *err;
  while ((err= it++))
  {
Marc Alff's avatar
Marc Alff committed
2998 2999 3000
    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
3001 3002
    {
    case ER_GET_TEMPORARY_ERRMSG:
3003
      DBUG_RETURN(1);
unknown's avatar
unknown committed
3004 3005 3006 3007 3008
    default:
      break;
    }
  }
#endif
3009
  DBUG_RETURN(0);
unknown's avatar
unknown committed
3010
}
3011

3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038

/**
  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().
*/
3039
int apply_event_and_update_pos(Log_event* ev, THD* thd,
unknown's avatar
unknown committed
3040
                               rpl_group_info *rgi,
3041
                               rpl_parallel_thread *rpt)
3042 3043
{
  int exec_res= 0;
3044
  Relay_log_info* rli= rgi->rli;
3045 3046 3047 3048 3049
  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));
3050
  DBUG_PRINT("info", ("thd->options: %s%s; rgi->last_event_start_time: %lu",
3051 3052
                      FLAGSTR(thd->variables.option_bits, OPTION_NOT_AUTOCOMMIT),
                      FLAGSTR(thd->variables.option_bits, OPTION_BEGIN),
3053
                      (ulong) rgi->last_event_start_time));
3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078

  /*
    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).
  */

3079 3080
  /* Use the original server id for logging. */
  thd->variables.server_id = ev->server_id;
3081 3082 3083
  thd->set_time();                            // time the query
  thd->lex->current_select= 0;
  if (!ev->when)
3084 3085
  {
    my_hrtime_t hrtime= my_hrtime();
3086
    ev->when= hrtime_to_my_time(hrtime);
3087 3088
    ev->when_sec_part= hrtime_sec_part(hrtime);
  }
3089 3090
  thd->variables.option_bits=
    (thd->variables.option_bits & ~OPTION_SKIP_REPLICATION) |
3091
    (ev->flags & LOG_EVENT_SKIP_REPLICATION_F ? OPTION_SKIP_REPLICATION : 0);
3092 3093
  ev->thd = thd; // because up to this point, ev->thd == 0

Michael Widenius's avatar
Michael Widenius committed
3094
  int reason= ev->shall_skip(rgi);
3095
  if (reason == Log_event::EVENT_SKIP_COUNT)
3096 3097 3098 3099
  {
    DBUG_ASSERT(rli->slave_skip_counter > 0);
    rli->slave_skip_counter--;
  }
Marc Alff's avatar
Marc Alff committed
3100
  mysql_mutex_unlock(&rli->data_lock);
3101
  if (reason == Log_event::EVENT_SKIP_NOT)
3102
    exec_res= ev->apply_event(rgi);
3103

3104
#ifndef DBUG_OFF
3105 3106
  /*
    This only prints information to the debug trace.
3107

3108 3109 3110 3111 3112 3113 3114 3115 3116 3117
    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
3118
  DBUG_PRINT("info", ("OPTION_BEGIN: %d  IN_STMT: %d  IN_TRANSACTION: %d",
3119
                      test(thd->variables.option_bits & OPTION_BEGIN),
Michael Widenius's avatar
Michael Widenius committed
3120 3121
                      rli->get_flag(Relay_log_info::IN_STMT),
                      rli->get_flag(Relay_log_info::IN_TRANSACTION)));
3122 3123
  DBUG_PRINT("skip_event", ("%s event was %s",
                            ev->get_type_str(), explain[reason]));
3124 3125 3126 3127 3128
#endif

  DBUG_PRINT("info", ("apply_event error = %d", exec_res));
  if (exec_res == 0)
  {
3129
    int error= ev->update_pos(rgi);
3130
#ifdef HAVE_valgrind
3131 3132 3133
    if (!rli->is_fake)
#endif
    {
3134
#ifndef DBUG_OFF
3135
      char buf[22];
3136
#endif
3137 3138 3139 3140 3141 3142 3143 3144
      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));
    }
3145 3146 3147 3148 3149 3150 3151 3152 3153
    /*
      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)
    {
3154
      char buf[22];
3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169
      rli->report(ERROR_LEVEL, ER_UNKNOWN_ERROR,
                  "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);
    }
  }

  DBUG_RETURN(exec_res ? 1 : 0);
}


Michael Widenius's avatar
Michael Widenius committed
3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 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 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243
/**
   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)));
}


3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263
/**
  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
3264
     if a failure occurs when applying the event.)
3265 3266 3267 3268 3269 3270 3271

   - 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
3272

3273 3274
static int exec_relay_log_event(THD* thd, Relay_log_info* rli,
                                rpl_group_info *serial_rgi)
unknown's avatar
unknown committed
3275
{
3276 3277
  DBUG_ENTER("exec_relay_log_event");

3278
  /*
Michael Widenius's avatar
Michael Widenius committed
3279 3280 3281 3282
    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
3283
  mysql_mutex_lock(&rli->data_lock);
3284

Michael Widenius's avatar
Michael Widenius committed
3285
  Log_event * ev = next_event(serial_rgi);
3286

Michael Widenius's avatar
Michael Widenius committed
3287
  if (sql_slave_killed(serial_rgi))
3288
  {
Marc Alff's avatar
Marc Alff committed
3289
    mysql_mutex_unlock(&rli->data_lock);
unknown's avatar
unknown committed
3290
    delete ev;
3291
    DBUG_RETURN(1);
3292
  }
3293 3294
  if (ev)
  {
3295
    int exec_res;
3296
    Log_event_type typ= ev->get_type_code();
3297

3298 3299 3300 3301
    /*
      This tests if the position of the beginning of the current event
      hits the UNTIL barrier.
    */
unknown's avatar
unknown committed
3302 3303
    if ((rli->until_condition == Relay_log_info::UNTIL_MASTER_POS ||
         rli->until_condition == Relay_log_info::UNTIL_RELAY_POS) &&
3304
        rli->is_until_satisfied(thd, ev))
3305 3306 3307 3308 3309
    {
      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
3310 3311
        Setting abort_slave flag because we do not want additional
        message about error in query execution to be printed.
3312 3313
      */
      rli->abort_slave= 1;
Marc Alff's avatar
Marc Alff committed
3314
      mysql_mutex_unlock(&rli->data_lock);
3315
      delete ev;
3316
      DBUG_RETURN(1);
3317
    }
3318 3319 3320 3321 3322 3323 3324 3325

    { /**
         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",
3326 3327
                      if ((typ == XID_EVENT) ||
                          ((typ == QUERY_EVENT) &&
3328 3329 3330 3331
                           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
3332
                        mysql_mutex_unlock(&rli->data_lock);
3333 3334 3335 3336 3337 3338
                        delete ev;
                        rli->inc_event_relay_log_pos();
                        DBUG_RETURN(0);
                      };);
    }

Michael Widenius's avatar
Michael Widenius committed
3339 3340 3341 3342 3343 3344 3345 3346
    update_state_of_relay_log(rli, ev);

    /*
      Execute queries in parallel, except if slave_skip_counter is set,
      as it's is easier to skip queries in single threaded mode.
    */

    if (opt_slave_parallel_threads > 0 && rli->slave_skip_counter == 0)
3347
      DBUG_RETURN(rli->parallel.do_event(serial_rgi, ev));
3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361

    /*
      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 (typ == GTID_EVENT &&
        event_group_new_gtid(serial_rgi, static_cast<Gtid_log_event *>(ev)))
    {
      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);
    }
3362

3363
    exec_res= apply_event_and_update_pos(ev, thd, serial_rgi, NULL);
3364

3365
    delete_or_keep_event_post_apply(serial_rgi, typ, ev);
3366

3367
    /*
3368 3369
      update_log_pos failed: this should not happen, so we don't
      retry.
3370
    */
3371 3372 3373
    if (exec_res == 2)
      DBUG_RETURN(1);

3374 3375
    if (slave_trans_retries)
    {
3376
      int temp_err;
Michael Widenius's avatar
Michael Widenius committed
3377
      LINT_INIT(temp_err);
3378
      if (exec_res && (temp_err= has_temporary_error(thd)))
3379 3380 3381 3382
      {
        const char *errmsg;
        /*
          We were in a transaction which has been rolled back because of a
unknown's avatar
unknown committed
3383 3384
          temporary error;
          let's seek back to BEGIN log event and retry it all again.
3385
	  Note, if lock wait timeout (innodb_lock_wait_timeout exceeded)
3386
	  there is no rollback since 5.0.13 (ref: manual).
3387
          We have to not only seek but also
3388 3389 3390 3391 3392 3393

          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
3394 3395 3396
          init_master_info()).
          b) init_relay_log_pos(), because the BEGIN may be an older relay log.
        */
3397
        if (serial_rgi->trans_retries < slave_trans_retries)
3398 3399 3400 3401 3402 3403
        {
          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,
3404
                                      1, &errmsg, 1))
3405 3406 3407 3408 3409
            sql_print_error("Error initializing relay log position: %s",
                            errmsg);
          else
          {
            exec_res= 0;
unknown's avatar
unknown committed
3410
            serial_rgi->cleanup_context(thd, 1);
unknown's avatar
unknown committed
3411
            /* chance for concurrent connection to get more locks */
3412 3413
            slave_sleep(thd, min(serial_rgi->trans_retries,
                                 MAX_SLAVE_RETRY_PAUSE),
Michael Widenius's avatar
Michael Widenius committed
3414
                        sql_slave_killed, serial_rgi);
3415
            serial_rgi->trans_retries++;
Marc Alff's avatar
Marc Alff committed
3416
            mysql_mutex_lock(&rli->data_lock); // because of SHOW STATUS
3417
            rli->retried_trans++;
3418
            statistic_increment(slave_retried_transactions, LOCK_status);
Marc Alff's avatar
Marc Alff committed
3419
            mysql_mutex_unlock(&rli->data_lock);
3420
            DBUG_PRINT("info", ("Slave retries transaction "
3421 3422
                                "rgi->trans_retries: %lu",
                                serial_rgi->trans_retries));
unknown's avatar
unknown committed
3423
          }
3424 3425 3426 3427 3428 3429 3430
        }
        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);
      }
3431
      else if ((exec_res && !temp_err) ||
3432 3433
               (opt_using_transactions &&
                rli->group_relay_log_pos == rli->event_relay_log_pos))
3434 3435
      {
        /*
3436 3437 3438
          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
3439 3440
          non-transient error, the slave will stop with an error.
         */
3441 3442 3443
        serial_rgi->trans_retries= 0; // restart from fresh
        DBUG_PRINT("info", ("Resetting retry counter, rgi->trans_retries: %lu",
                            serial_rgi->trans_retries));
3444
      }
3445
    }
3446 3447
    thread_safe_increment64(&rli->executed_entries,
                            &slave_executed_entries_lock);
3448
    DBUG_RETURN(exec_res);
3449
  }
Marc Alff's avatar
Marc Alff committed
3450
  mysql_mutex_unlock(&rli->data_lock);
3451 3452
  rli->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_READ_FAILURE,
              ER(ER_SLAVE_RELAY_LOG_READ_FAILURE), "\
3453 3454 3455 3456 3457 3458 3459
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
3460
");
3461
  DBUG_RETURN(1);
unknown's avatar
unknown committed
3462 3463
}

3464

Michael Widenius's avatar
Michael Widenius committed
3465
static bool check_io_slave_killed(Master_info *mi, const char *info)
3466
{
Michael Widenius's avatar
Michael Widenius committed
3467
  if (io_slave_killed(mi))
3468
  {
3469
    if (info && global_system_variables.log_warnings)
3470
      sql_print_information("%s", info);
3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501
    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.
*/

3502
static int try_to_reconnect(THD *thd, MYSQL *mysql, Master_info *mi,
3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515
                            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
3516
    slave_sleep(thd, mi->connect_retry, io_slave_killed, mi);
3517
  }
Michael Widenius's avatar
Michael Widenius committed
3518
  if (check_io_slave_killed(mi, messages[SLAVE_RECON_MSG_KILLED_WAITING]))
3519 3520 3521 3522 3523
    return 1;
  thd->proc_info = messages[SLAVE_RECON_MSG_AFTER];
  if (!suppress_warnings) 
  {
    char buf[256], llbuff[22];
unknown's avatar
unknown committed
3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536
    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);
      }
    }
3537
    my_snprintf(buf, sizeof(buf), messages[SLAVE_RECON_MSG_FAILED], 
unknown's avatar
unknown committed
3538 3539
                IO_RPL_LOG_NAME, llstr(mi->master_log_pos, llbuff),
                tmp.c_ptr_safe());
3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551
    /* 
      Raise a warining during registering on master/requesting dump.
      Log a message reading event.
    */
    if (messages[SLAVE_RECON_MSG_COMMAND][0])
    {
      mi->report(WARNING_LEVEL, ER_SLAVE_MASTER_COM_FAILURE,
                 ER(ER_SLAVE_MASTER_COM_FAILURE), 
                 messages[SLAVE_RECON_MSG_COMMAND], buf);
    }
    else
    {
3552
      sql_print_information("%s", buf);
3553 3554
    }
  }
Michael Widenius's avatar
Michael Widenius committed
3555
  if (safe_reconnect(thd, mysql, mi, 1) || io_slave_killed(mi))
3556 3557
  {
    if (global_system_variables.log_warnings)
3558
      sql_print_information("%s", messages[SLAVE_RECON_MSG_KILLED_AFTER]);
3559 3560 3561 3562 3563 3564
    return 1;
  }
  return 0;
}


3565 3566 3567 3568 3569
/**
  Slave IO thread entry point.

  @param arg Pointer to Master_info struct that holds information for
  the IO thread.
3570

3571 3572
  @return Always 0.
*/
3573
pthread_handler_t handle_slave_io(void *arg)
unknown's avatar
unknown committed
3574
{
unknown's avatar
unknown committed
3575 3576
  THD *thd; // needs to be first for thread_stack
  MYSQL *mysql;
3577
  Master_info *mi = (Master_info*)arg;
3578
  Relay_log_info *rli= &mi->rli;
unknown's avatar
unknown committed
3579 3580
  char llbuff[22];
  uint retry_count;
3581
  bool suppress_warnings;
3582
  int ret;
3583 3584 3585
#ifndef DBUG_OFF
  uint retry_count_reg= 0, retry_count_dump= 0, retry_count_event= 0;
#endif
unknown's avatar
unknown committed
3586 3587
  // needs to call my_thread_init(), otherwise we get a coredump in DBUG_ stuff
  my_thread_init();
3588
  DBUG_ENTER("handle_slave_io");
unknown's avatar
unknown committed
3589

3590
  DBUG_ASSERT(mi->inited);
unknown's avatar
unknown committed
3591 3592 3593
  mysql= NULL ;
  retry_count= 0;

3594 3595
  thd= new THD; // note that contructor of THD uses DBUG_ !

Marc Alff's avatar
Marc Alff committed
3596
  mysql_mutex_lock(&mi->run_lock);
unknown's avatar
unknown committed
3597 3598 3599
  /* Inform waiting threads that slave has started */
  mi->slave_run_id++;

3600
#ifndef DBUG_OFF
3601
  mi->events_till_disconnect = disconnect_slave_event_count;
3602 3603
#endif

3604
  THD_CHECK_SENTRY(thd);
3605
  mi->io_thd = thd;
unknown's avatar
unknown committed
3606 3607

  pthread_detach_this_thread();
3608
  thd->thread_stack= (char*) &thd; // remember where our stack is
3609
  mi->clear_error();
3610
  if (init_slave_thread(thd, mi, SLAVE_THD_IO))
unknown's avatar
unknown committed
3611
  {
Marc Alff's avatar
Marc Alff committed
3612
    mysql_cond_broadcast(&mi->start_cond);
unknown's avatar
unknown committed
3613
    sql_print_error("Failed during slave I/O thread initialization");
3614
    goto err_during_init;
unknown's avatar
unknown committed
3615
  }
Marc Alff's avatar
Marc Alff committed
3616
  mysql_mutex_lock(&LOCK_thread_count);
unknown's avatar
unknown committed
3617
  threads.append(thd);
Marc Alff's avatar
Marc Alff committed
3618
  mysql_mutex_unlock(&LOCK_thread_count);
3619
  mi->slave_running = MYSQL_SLAVE_RUN_NOT_CONNECT;
3620
  mi->abort_slave = 0;
Marc Alff's avatar
Marc Alff committed
3621 3622
  mysql_mutex_unlock(&mi->run_lock);
  mysql_cond_broadcast(&mi->start_cond);
3623

3624
  DBUG_PRINT("master_info",("log_file_name: '%s'  position: %s",
unknown's avatar
unknown committed
3625 3626
                            mi->master_log_name,
                            llstr(mi->master_log_pos,llbuff)));
3627

He Zhenxing's avatar
He Zhenxing committed
3628 3629 3630
  /* This must be called before run any binlog_relay_io hooks */
  my_pthread_setspecific_ptr(RPL_MASTER_INFO, mi);

3631 3632 3633 3634 3635 3636
  /* Load the set of seen GTIDs, if we did not already. */
  if (rpl_load_gtid_slave_state(thd))
  {
    mi->report(ERROR_LEVEL, thd->stmt_da->sql_errno(), 
                "Unable to load replication GTID slave state from mysql.%s: %s",
                rpl_gtid_slave_state_table_name.str, thd->stmt_da->message());
3637 3638 3639 3640 3641 3642 3643
    /*
      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;
3644 3645 3646
  }


He Zhenxing's avatar
He Zhenxing committed
3647 3648 3649 3650 3651 3652 3653
  if (RUN_HOOK(binlog_relay_io, thread_start, (thd, mi)))
  {
    mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
               ER(ER_SLAVE_FATAL_ERROR), "Failed to run 'thread_start' hook");
    goto err;
  }

unknown's avatar
SCRUM  
unknown committed
3654
  if (!(mi->mysql = mysql = mysql_init(NULL)))
unknown's avatar
unknown committed
3655
  {
3656 3657
    mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
               ER(ER_SLAVE_FATAL_ERROR), "error in mysql_init()");
unknown's avatar
unknown committed
3658 3659
    goto err;
  }
3660

3661
  thd_proc_info(thd, "Connecting to master");
3662
  // we can get killed during safe_connect
3663
  if (!safe_connect(thd, mysql, mi))
3664
  {
unknown's avatar
unknown committed
3665 3666 3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679
    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());
    }

3680 3681 3682 3683 3684
  /*
    Adding MAX_LOG_EVENT_HEADER_LEN to the max_packet_size on the I/O
    thread, since a replication event can become this much larger than
    the corresponding packet (query) sent from client to master.
  */
3685
    thd->net.max_packet_size= slave_max_allowed_packet;
3686 3687
    mysql->net.max_packet_size= thd->net.max_packet_size+= MAX_LOG_EVENT_HEADER;
  }
3688
  else
unknown's avatar
unknown committed
3689
  {
3690
    sql_print_information("Slave I/O thread killed while connecting to master");
unknown's avatar
unknown committed
3691 3692
    goto err;
  }
3693

3694
connected:
3695

unknown's avatar
unknown committed
3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714
  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;
  }

3715
#ifdef ENABLED_DEBUG_SYNC
3716 3717 3718 3719 3720
    DBUG_EXECUTE_IF("dbug.before_get_running_status_yes",
                    {
                      const char act[]=
                        "now "
                        "wait_for signal.io_thread_let_running";
3721
                      DBUG_ASSERT(debug_sync_service);
3722 3723 3724
                      DBUG_ASSERT(!debug_sync_set_action(thd, 
                                                         STRING_WITH_LEN(act)));
                    };);
3725
#endif
3726

3727 3728
  // TODO: the assignment below should be under mutex (5.0)
  mi->slave_running= MYSQL_SLAVE_RUN_CONNECT;
3729
  thd->slave_net = &mysql->net;
3730
  thd_proc_info(thd, "Checking master version");
3731 3732 3733
  ret= get_master_version_and_clock(mysql, mi);
  if (ret == 1)
    /* Fatal error */
3734
    goto err;
3735

3736 3737
  if (ret == 2) 
  { 
Michael Widenius's avatar
Michael Widenius committed
3738
    if (check_io_slave_killed(mi, "Slave I/O thread killed"
3739 3740 3741
                              "while calling get_master_version_and_clock(...)"))
      goto err;
    suppress_warnings= FALSE;
Michael Widenius's avatar
Michael Widenius committed
3742 3743 3744 3745
    /*
      Try to reconnect because the error was caused by a transient network
      problem
    */
3746 3747 3748 3749 3750
    if (try_to_reconnect(thd, mysql, mi, &retry_count, suppress_warnings,
                             reconnect_messages[SLAVE_RECON_ACT_REG]))
      goto err;
    goto connected;
  } 
3751 3752

  if (mi->rli.relay_log.description_event_for_queue->binlog_version > 1)
3753
  {
unknown's avatar
unknown committed
3754 3755 3756
    /*
      Register ourselves with the master.
    */
3757
    thd_proc_info(thd, "Registering slave on master");
3758
    if (register_slave_on_master(mysql, mi, &suppress_warnings))
3759
    {
Michael Widenius's avatar
Michael Widenius committed
3760
      if (!check_io_slave_killed(mi, "Slave I/O thread killed "
3761 3762 3763 3764 3765 3766 3767 3768
                                "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
3769 3770
        goto err;
      goto connected;
3771
    }
3772 3773 3774 3775 3776 3777 3778 3779 3780 3781
    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;
      });
3782
  }
3783

3784
  DBUG_PRINT("info",("Starting reading binary log from master"));
Michael Widenius's avatar
Michael Widenius committed
3785
  while (!io_slave_killed(mi))
unknown's avatar
unknown committed
3786
  {
3787
    thd_proc_info(thd, "Requesting binlog dump");
He Zhenxing's avatar
He Zhenxing committed
3788
    if (request_dump(thd, mysql, mi, &suppress_warnings))
unknown's avatar
unknown committed
3789 3790
    {
      sql_print_error("Failed on request_dump()");
Michael Widenius's avatar
Michael Widenius committed
3791
      if (check_io_slave_killed(mi, "Slave I/O thread killed while \
3792 3793 3794
requesting master dump") ||
          try_to_reconnect(thd, mysql, mi, &retry_count, suppress_warnings,
                           reconnect_messages[SLAVE_RECON_ACT_DUMP]))
unknown's avatar
unknown committed
3795
        goto err;
unknown's avatar
unknown committed
3796 3797
      goto connected;
    }
3798 3799 3800 3801 3802 3803 3804 3805 3806 3807
    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
3808
    const char *event_buf;
unknown's avatar
unknown committed
3809

3810
    DBUG_ASSERT(mi->last_error().number == 0);
Michael Widenius's avatar
Michael Widenius committed
3811
    while (!io_slave_killed(mi))
unknown's avatar
unknown committed
3812
    {
3813
      ulong event_len;
3814
      /*
3815
         We say "waiting" because read_event() will wait if there's nothing to
3816 3817 3818
         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.
3819
      */
3820
      thd_proc_info(thd, "Waiting for master to send event");
3821
      event_len= read_event(mysql, mi, &suppress_warnings);
Michael Widenius's avatar
Michael Widenius committed
3822
      if (check_io_slave_killed(mi, "Slave I/O thread killed while \
3823
reading event"))
unknown's avatar
unknown committed
3824
        goto err;
3825 3826 3827 3828 3829 3830 3831 3832 3833 3834
      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;
        });
3835

unknown's avatar
unknown committed
3836 3837
      if (event_len == packet_error)
      {
unknown's avatar
unknown committed
3838
        uint mysql_error_number= mysql_errno(mysql);
3839 3840
        switch (mysql_error_number) {
        case CR_NET_PACKET_TOO_LARGE:
unknown's avatar
unknown committed
3841
          sql_print_error("\
3842
Log entry on master is longer than slave_max_allowed_packet (%lu) on \
3843
slave. If the entry is correct, restart the server with a higher value of \
3844 3845
slave_max_allowed_packet",
                         slave_max_allowed_packet);
3846
          mi->report(ERROR_LEVEL, ER_NET_PACKET_TOO_LARGE,
3847
                     "%s", "Got a packet bigger than 'slave_max_allowed_packet' bytes");
unknown's avatar
unknown committed
3848
          goto err;
3849
        case ER_MASTER_FATAL_ERROR_READING_BINLOG:
3850 3851 3852
          mi->report(ERROR_LEVEL, ER_MASTER_FATAL_ERROR_READING_BINLOG,
                     ER(ER_MASTER_FATAL_ERROR_READING_BINLOG),
                     mysql_error_number, mysql_error(mysql));
unknown's avatar
unknown committed
3853
          goto err;
3854
        case ER_OUT_OF_RESOURCES:
3855 3856
          sql_print_error("\
Stopping slave I/O thread due to out-of-memory error from master");
3857
          mi->report(ERROR_LEVEL, ER_OUT_OF_RESOURCES,
3858
                     "%s", ER(ER_OUT_OF_RESOURCES));
unknown's avatar
unknown committed
3859 3860
          goto err;
        }
3861 3862
        if (try_to_reconnect(thd, mysql, mi, &retry_count, suppress_warnings,
                             reconnect_messages[SLAVE_RECON_ACT_EVENT]))
unknown's avatar
unknown committed
3863 3864
          goto err;
        goto connected;
unknown's avatar
unknown committed
3865
      } // if (event_len == packet_error)
3866

unknown's avatar
unknown committed
3867
      retry_count=0;                    // ok event, reset retry counter
3868
      thd_proc_info(thd, "Queueing master event to the relay log");
He Zhenxing's avatar
He Zhenxing committed
3869 3870 3871 3872 3873 3874 3875 3876 3877 3878 3879 3880 3881 3882 3883
      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)))
      {
        mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
                   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
3884
      {
3885 3886 3887
        mi->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_WRITE_FAILURE,
                   ER(ER_SLAVE_RELAY_LOG_WRITE_FAILURE),
                   "could not queue event from master");
unknown's avatar
unknown committed
3888
        goto err;
unknown's avatar
unknown committed
3889
      }
He Zhenxing's avatar
He Zhenxing committed
3890 3891 3892 3893 3894 3895 3896 3897 3898 3899

      if (RUN_HOOK(binlog_relay_io, after_queue_event,
                   (thd, mi, event_buf, event_len, synced)))
      {
        mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
                   ER(ER_SLAVE_FATAL_ERROR),
                   "Failed to run 'after_queue_event' hook");
        goto err;
      }

3900
      if (flush_master_info(mi, TRUE, TRUE))
3901 3902 3903 3904
      {
        sql_print_error("Failed to flush master info file");
        goto err;
      }
3905 3906 3907 3908 3909 3910 3911
      /*
        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
3912 3913 3914 3915 3916
        - 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.
3917
      */
3918 3919 3920 3921 3922
#ifndef DBUG_OFF
      {
        char llbuf1[22], llbuf2[22];
        DBUG_PRINT("info", ("log_space_limit=%s log_space_total=%s \
ignore_log_space_limit=%d",
3923 3924
                            llstr(rli->log_space_limit,llbuf1),
                            llstr(rli->log_space_total,llbuf2),
unknown's avatar
unknown committed
3925
                            (int) rli->ignore_log_space_limit));
3926 3927 3928
      }
#endif

3929
      if (rli->log_space_limit && rli->log_space_limit <
unknown's avatar
unknown committed
3930
          rli->log_space_total &&
3931
          !rli->ignore_log_space_limit)
unknown's avatar
unknown committed
3932 3933 3934
        if (wait_for_relay_log_space(rli))
        {
          sql_print_error("Slave I/O thread aborted while waiting for relay \
unknown's avatar
unknown committed
3935
log space");
unknown's avatar
unknown committed
3936 3937 3938
          goto err;
        }
    }
3939
  }
unknown's avatar
unknown committed
3940

unknown's avatar
unknown committed
3941
  // error = 0;
unknown's avatar
unknown committed
3942
err:
3943
  // print the current replication position
unknown's avatar
unknown committed
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 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
3957
  RUN_HOOK(binlog_relay_io, thread_stop, (thd, mi));
3958
  thd->reset_query();
unknown's avatar
unknown committed
3959
  thd->reset_db(NULL, 0);
unknown's avatar
unknown committed
3960 3961
  if (mysql)
  {
3962 3963 3964 3965 3966 3967 3968 3969 3970 3971 3972
    /*
      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
3973
    mysql_close(mysql);
unknown's avatar
unknown committed
3974 3975
    mi->mysql=0;
  }
3976
  write_ignored_events_info_to_relay_log(thd, mi);
Michael Widenius's avatar
Michael Widenius committed
3977
  thd_proc_info(thd, "Slave io thread waiting for slave mutex on exit");
Marc Alff's avatar
Marc Alff committed
3978
  mysql_mutex_lock(&mi->run_lock);
3979

3980
err_during_init:
3981 3982 3983
  /* Forget the relay log's format */
  delete mi->rli.relay_log.description_event_for_queue;
  mi->rli.relay_log.description_event_for_queue= 0;
3984
  // TODO: make rpl_status part of Master_info
3985
  change_rpl_status(RPL_ACTIVE_SLAVE,RPL_IDLE_SLAVE);
Marc Alff's avatar
Marc Alff committed
3986
  mysql_mutex_lock(&LOCK_thread_count);
3987
  THD_CHECK_SENTRY(thd);
unknown's avatar
unknown committed
3988
  delete thd;
Marc Alff's avatar
Marc Alff committed
3989
  mysql_mutex_unlock(&LOCK_thread_count);
unknown's avatar
unknown committed
3990
  mi->abort_slave= 0;
3991
  mi->slave_running= MYSQL_SLAVE_NOT_RUN;
unknown's avatar
unknown committed
3992
  mi->io_thd= 0;
3993 3994 3995 3996 3997
  /*
    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
3998
  mysql_cond_broadcast(&mi->stop_cond);       // tell the world we are done
3999
  DBUG_EXECUTE_IF("simulate_slave_delay_at_terminate_bug38694", sleep(5););
Marc Alff's avatar
Marc Alff committed
4000
  mysql_mutex_unlock(&mi->run_lock);
4001 4002

  DBUG_LEAVE;                                   // Must match DBUG_ENTER()
4003
  my_thread_end();
unknown's avatar
unknown committed
4004
  pthread_exit(0);
4005
  return 0;                                     // Avoid compiler warnings
unknown's avatar
unknown committed
4006 4007
}

4008 4009 4010
/*
  Check the temporary directory used by commands like
  LOAD DATA INFILE.
4011 4012 4013 4014

  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.
4015
 */
4016 4017 4018 4019

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

4020
static 
4021
int check_temp_dir(char* tmp_file)
4022
{
4023 4024
  File fd;
  int result= 1;                                // Assume failure
4025
  MY_DIR *dirp;
4026 4027
  char tmp_dir[FN_REFLEN];
  size_t tmp_dir_size;
4028 4029
  DBUG_ENTER("check_temp_dir");

4030 4031 4032 4033 4034 4035 4036 4037
  mysql_mutex_lock(&LOCK_thread_count);
  if (check_temp_dir_run)
  {
    result= check_temp_dir_result;
    goto end;
  }
  check_temp_dir_run= 1;

4038 4039 4040 4041 4042
  /*
    Get the directory from the temporary file.
  */
  dirname_part(tmp_dir, tmp_file, &tmp_dir_size);

4043 4044 4045 4046
  /*
    Check if the directory exists.
   */
  if (!(dirp=my_dir(tmp_dir,MYF(MY_WME))))
4047
    goto end;
4048 4049 4050
  my_dirend(dirp);

  /*
4051 4052
    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.
4053
   */
Marc Alff's avatar
Marc Alff committed
4054 4055
  if ((fd= mysql_file_create(key_file_misc,
                             tmp_file, CREATE_MODE,
4056
                             O_WRONLY | O_BINARY | O_TRUNC | O_NOFOLLOW,
Marc Alff's avatar
Marc Alff committed
4057
                             MYF(MY_WME))) < 0)
4058
    goto end;
4059

4060
  result= 0;                                    // Directory name ok
4061 4062 4063
  /*
    Clean up.
   */
Marc Alff's avatar
Marc Alff committed
4064 4065
  mysql_file_close(fd, MYF(0));
  mysql_file_delete(key_file_misc, tmp_file, MYF(0));
4066

4067 4068 4069 4070
end:
  check_temp_dir_result= result;
  mysql_mutex_unlock(&LOCK_thread_count);
  DBUG_RETURN(result);
4071
}
unknown's avatar
unknown committed
4072

4073

4074 4075 4076 4077 4078
/**
  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
4079

4080 4081
  @return Always 0.
*/
4082
pthread_handler_t handle_slave_sql(void *arg)
4083
{
unknown's avatar
unknown committed
4084
  THD *thd;                     /* needs to be first for thread_stack */
4085
  char llbuff[22],llbuff1[22];
4086 4087
  char saved_log_name[FN_REFLEN];
  char saved_master_log_name[FN_REFLEN];
4088 4089
  my_off_t UNINIT_VAR(saved_log_pos);
  my_off_t UNINIT_VAR(saved_master_log_pos);
4090
  my_off_t saved_skip= 0;
4091 4092
  Master_info *mi= ((Master_info*)arg);
  Relay_log_info* rli = &mi->rli;
unknown's avatar
unknown committed
4093
  const char *errmsg;
unknown's avatar
unknown committed
4094
  rpl_group_info *serial_rgi;
unknown's avatar
unknown committed
4095 4096 4097

  // needs to call my_thread_init(), otherwise we get a coredump in DBUG_ stuff
  my_thread_init();
4098
  DBUG_ENTER("handle_slave_sql");
unknown's avatar
unknown committed
4099

Michael Widenius's avatar
Michael Widenius committed
4100 4101
  LINT_INIT(saved_master_log_pos);
  LINT_INIT(saved_log_pos);
4102

unknown's avatar
unknown committed
4103
  serial_rgi= new rpl_group_info(rli);
4104 4105
  thd = new THD; // note that contructor of THD uses DBUG_ !
  thd->thread_stack = (char*)&thd; // remember where our stack is
4106
  thd->rpl_filter = mi->rpl_filter;
4107

4108
  DBUG_ASSERT(rli->inited);
4109
  DBUG_ASSERT(rli->mi == mi);
Marc Alff's avatar
Marc Alff committed
4110
  mysql_mutex_lock(&rli->run_lock);
4111
  DBUG_ASSERT(!rli->slave_running);
unknown's avatar
unknown committed
4112
  errmsg= 0;
unknown's avatar
unknown committed
4113
#ifndef DBUG_OFF
4114
  rli->events_till_abort = abort_slave_event_count;
unknown's avatar
unknown committed
4115
#endif
4116

Michael Widenius's avatar
Michael Widenius committed
4117 4118 4119 4120 4121 4122 4123 4124 4125
  /*
    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;
4126
  
unknown's avatar
unknown committed
4127 4128
  /* Inform waiting threads that slave has started */
  rli->slave_run_id++;
4129
  rli->slave_running= MYSQL_SLAVE_RUN_NOT_CONNECT;
unknown's avatar
unknown committed
4130

4131
  pthread_detach_this_thread();
4132
  if (init_slave_thread(thd, mi, SLAVE_THD_SQL))
unknown's avatar
unknown committed
4133 4134 4135 4136 4137
  {
    /*
      TODO: this is currently broken - slave start and change master
      will be stuck if we fail here
    */
Marc Alff's avatar
Marc Alff committed
4138
    mysql_cond_broadcast(&rli->start_cond);
4139 4140
    rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, 
                "Failed during slave thread initialization");
4141
    goto err_during_init;
unknown's avatar
unknown committed
4142
  }
4143
  thd->init_for_queries();
unknown's avatar
unknown committed
4144 4145
  thd->rgi_slave= serial_rgi;
  if ((serial_rgi->deferred_events_collecting= mi->rpl_filter->is_on()))
4146
  {
unknown's avatar
unknown committed
4147
    serial_rgi->deferred_events= new Deferred_log_events(rli);
4148 4149
  }

4150
  /*
4151
    binlog_annotate_row_events must be TRUE only after an Annotate_rows event
4152 4153 4154
    has been recieved and only till the last corresponding rbr event has been
    applied. In all other cases it must be FALSE.
  */
4155
  thd->variables.binlog_annotate_row_events= 0;
Marc Alff's avatar
Marc Alff committed
4156
  mysql_mutex_lock(&LOCK_thread_count);
4157
  threads.append(thd);
Marc Alff's avatar
Marc Alff committed
4158
  mysql_mutex_unlock(&LOCK_thread_count);
4159 4160 4161 4162 4163 4164 4165 4166
  /*
    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.
  */
4167
  rli->abort_slave = 0;
Marc Alff's avatar
Marc Alff committed
4168 4169
  mysql_mutex_unlock(&rli->run_lock);
  mysql_cond_broadcast(&rli->start_cond);
4170

unknown's avatar
unknown committed
4171 4172 4173
  /*
    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
4174
    though there's no problem anymore). Do not reset the master timestamp
4175 4176 4177 4178
    (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
4179
    But the master timestamp is reset by RESET SLAVE & CHANGE MASTER.
unknown's avatar
unknown committed
4180
  */
4181
  rli->clear_error();
unknown's avatar
unknown committed
4182
  rli->parallel.reset();
4183 4184

  //tell the I/O thread to take relay_log_space_limit into account from now on
Marc Alff's avatar
Marc Alff committed
4185
  mysql_mutex_lock(&rli->log_space_lock);
4186
  rli->ignore_log_space_limit= 0;
Marc Alff's avatar
Marc Alff committed
4187
  mysql_mutex_unlock(&rli->log_space_lock);
4188

4189
  if (init_relay_log_pos(rli,
unknown's avatar
unknown committed
4190 4191 4192
                         rli->group_relay_log_name,
                         rli->group_relay_log_pos,
                         1 /*need data lock*/, &errmsg,
4193
                         1 /*look for a description_event*/))
4194 4195 4196
  { 
    rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, 
                "Error initializing relay log position: %s", errmsg);
4197 4198
    goto err;
  }
4199
  THD_CHECK_SENTRY(thd);
4200 4201 4202 4203
#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
4204
                        llstr(my_b_tell(rli->cur_log),llbuf1),
4205 4206 4207 4208 4209 4210 4211 4212 4213 4214 4215 4216 4217 4218 4219 4220 4221 4222 4223
                        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
4224 4225

  DBUG_PRINT("master_info",("log_file_name: %s  position: %s",
unknown's avatar
unknown committed
4226 4227
                            rli->group_master_log_name,
                            llstr(rli->group_master_log_pos,llbuff)));
4228
  if (global_system_variables.log_warnings)
unknown's avatar
unknown committed
4229 4230 4231 4232 4233 4234 4235 4236 4237
  {
    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("'"));
    }
4238
    sql_print_information("Slave SQL thread initialized, starting replication in \
unknown's avatar
unknown committed
4239
log '%s' at position %s, relay log '%s' position: %s%s", RPL_LOG_NAME,
unknown's avatar
unknown committed
4240
                    llstr(rli->group_master_log_pos,llbuff),rli->group_relay_log_name,
unknown's avatar
unknown committed
4241 4242
                    llstr(rli->group_relay_log_pos,llbuff1), tmp.c_ptr_safe());
  }
4243

4244
  if (check_temp_dir(rli->slave_patternload_file))
4245
  {
Marc Alff's avatar
Marc Alff committed
4246
    rli->report(ERROR_LEVEL, thd->stmt_da->sql_errno(), 
4247
                "Unable to use slave's temporary directory %s - %s", 
Marc Alff's avatar
Marc Alff committed
4248
                slave_load_tmpdir, thd->stmt_da->message());
4249 4250 4251
    goto err;
  }

4252 4253 4254 4255 4256 4257
  /* Load the set of seen GTIDs, if we did not already. */
  if (rpl_load_gtid_slave_state(thd))
  {
    rli->report(ERROR_LEVEL, thd->stmt_da->sql_errno(), 
                "Unable to load replication GTID slave state from mysql.%s: %s",
                rpl_gtid_slave_state_table_name.str, thd->stmt_da->message());
4258 4259 4260 4261 4262 4263 4264
    /*
      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;
4265 4266
  }

unknown's avatar
unknown committed
4267
  /* execute init_slave variable */
4268
  if (opt_init_slave.length)
unknown's avatar
unknown committed
4269
  {
4270
    execute_init_command(thd, &opt_init_slave, &LOCK_sys_init_slave);
4271
    if (thd->is_slave_error)
unknown's avatar
unknown committed
4272
    {
Marc Alff's avatar
Marc Alff committed
4273
      rli->report(ERROR_LEVEL, thd->stmt_da->sql_errno(),
4274
                  "Slave SQL thread aborted. Can't execute init_slave query");
unknown's avatar
unknown committed
4275 4276 4277 4278
      goto err;
    }
  }

4279 4280 4281 4282
  /*
    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
4283
  mysql_mutex_lock(&rli->data_lock);
4284 4285
  if (rli->slave_skip_counter)
  {
4286 4287
    strmake_buf(saved_log_name, rli->group_relay_log_name);
    strmake_buf(saved_master_log_name, rli->group_master_log_name);
4288 4289 4290 4291
    saved_log_pos= rli->group_relay_log_pos;
    saved_master_log_pos= rli->group_master_log_pos;
    saved_skip= rli->slave_skip_counter;
  }
unknown's avatar
unknown committed
4292 4293
  if ((rli->until_condition == Relay_log_info::UNTIL_MASTER_POS ||
       rli->until_condition == Relay_log_info::UNTIL_RELAY_POS) &&
4294
      rli->is_until_satisfied(thd, NULL))
4295 4296 4297 4298
  {
    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
4299
    mysql_mutex_unlock(&rli->data_lock);
4300 4301
    goto err;
  }
Marc Alff's avatar
Marc Alff committed
4302
  mysql_mutex_unlock(&rli->data_lock);
4303

4304 4305
  /* Read queries from the IO/THREAD until this thread is killed */

Michael Widenius's avatar
Michael Widenius committed
4306
  while (!sql_slave_killed(serial_rgi))
4307
  {
4308
    thd_proc_info(thd, "Reading event from the relay log");
4309
    THD_CHECK_SENTRY(thd);
4310 4311 4312 4313 4314 4315 4316 4317 4318 4319 4320 4321 4322 4323 4324

    if (saved_skip && rli->slave_skip_counter == 0)
    {
      sql_print_information("'SQL_SLAVE_SKIP_COUNTER=%ld' executed at "
        "relay_log_file='%s', relay_log_pos='%ld', master_log_name='%s', "
        "master_log_pos='%ld' and new position at "
        "relay_log_file='%s', relay_log_pos='%ld', master_log_name='%s', "
        "master_log_pos='%ld' ",
        (ulong) saved_skip, saved_log_name, (ulong) saved_log_pos,
        saved_master_log_name, (ulong) saved_master_log_pos,
        rli->group_relay_log_name, (ulong) rli->group_relay_log_pos,
        rli->group_master_log_name, (ulong) rli->group_master_log_pos);
      saved_skip= 0;
    }
    
unknown's avatar
unknown committed
4325
    if (exec_relay_log_event(thd, rli, serial_rgi))
4326
    {
4327
      DBUG_PRINT("info", ("exec_relay_log_event() failed"));
4328
      // do not scare the user if SQL thread was simply killed or stopped
Michael Widenius's avatar
Michael Widenius committed
4329
      if (!sql_slave_killed(serial_rgi))
unknown's avatar
unknown committed
4330 4331
      {
        /*
4332 4333 4334
          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
unknown's avatar
unknown committed
4335
        */
4336
        uint32 const last_errno= rli->last_error().number;
4337

4338
        if (thd->is_error())
unknown's avatar
unknown committed
4339
        {
Marc Alff's avatar
Marc Alff committed
4340
          char const *const errmsg= thd->stmt_da->message();
4341 4342

          DBUG_PRINT("info",
Marc Alff's avatar
Marc Alff committed
4343 4344
                     ("thd->stmt_da->sql_errno()=%d; rli->last_error.number=%d",
                      thd->stmt_da->sql_errno(), last_errno));
4345
          if (last_errno == 0)
unknown's avatar
unknown committed
4346
          {
4347 4348 4349 4350
            /*
 	      This function is reporting an error which was not reported
 	      while executing exec_relay_log_event().
 	    */ 
4351
            rli->report(ERROR_LEVEL, thd->stmt_da->sql_errno(), "%s", errmsg);
unknown's avatar
unknown committed
4352
          }
Marc Alff's avatar
Marc Alff committed
4353
          else if (last_errno != thd->stmt_da->sql_errno())
unknown's avatar
unknown committed
4354
          {
4355 4356 4357 4358 4359 4360
            /*
             * 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.
             */  
unknown's avatar
unknown committed
4361
            sql_print_error("Slave (additional info): %s Error_code: %d",
Marc Alff's avatar
Marc Alff committed
4362
                            errmsg, thd->stmt_da->sql_errno());
unknown's avatar
unknown committed
4363 4364 4365 4366
          }
        }

        /* Print any warnings issued */
Marc Alff's avatar
Marc Alff committed
4367
        List_iterator_fast<MYSQL_ERROR> it(thd->warning_info->warn_list());
unknown's avatar
unknown committed
4368
        MYSQL_ERROR *err;
4369 4370 4371 4372 4373
        /*
          Added controlled slave thread cancel for replication
          of user-defined variables.
        */
        bool udf_error = false;
unknown's avatar
unknown committed
4374
        while ((err= it++))
4375
        {
Marc Alff's avatar
Marc Alff committed
4376
          if (err->get_sql_errno() == ER_CANT_OPEN_LIBRARY)
4377
            udf_error = true;
Marc Alff's avatar
Marc Alff committed
4378
          sql_print_warning("Slave: %s Error_code: %d", err->get_message_text(), err->get_sql_errno());
4379 4380
        }
        if (udf_error)
unknown's avatar
unknown committed
4381 4382 4383 4384 4385 4386 4387 4388
        {
          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("'"));
          }
4389 4390 4391
          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' "
unknown's avatar
unknown committed
4392 4393 4394
            "position %s%s", RPL_LOG_NAME, llstr(rli->group_master_log_pos,
            llbuff), tmp.c_ptr_safe());
        }
4395
        else
unknown's avatar
unknown committed
4396 4397 4398 4399 4400 4401 4402 4403
        {
          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("'"));
          }
4404
          sql_print_error("\
4405
Error running query, slave SQL thread aborted. Fix the problem, and restart \
unknown's avatar
unknown committed
4406
the slave SQL thread with \"SLAVE START\". We stopped at log \
unknown's avatar
unknown committed
4407 4408 4409
'%s' position %s%s", RPL_LOG_NAME, llstr(rli->group_master_log_pos, llbuff),
                          tmp.c_ptr_safe());
        }
unknown's avatar
unknown committed
4410
      }
4411 4412
      goto err;
    }
4413
  }
4414

4415 4416 4417
  if (opt_slave_parallel_threads > 0)
    rli->parallel.wait_for_done();

4418
  /* Thread stopped. Print the current replication position to the log */
unknown's avatar
unknown committed
4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431 4432
  {
    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());
  }
4433 4434

 err:
4435

4436 4437 4438 4439 4440 4441 4442 4443
  /*
    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)
    rli->parallel.wait_for_done();

4444 4445 4446 4447 4448 4449
  /*
    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:
  */
4450
  thd->clear_error();
unknown's avatar
unknown committed
4451
  serial_rgi->cleanup_context(thd, 1);
4452 4453 4454 4455 4456
  /*
    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)).
  */
4457
  thd->catalog= 0;
4458
  thd->reset_query();
unknown's avatar
unknown committed
4459
  thd->reset_db(NULL, 0);
Michael Widenius's avatar
Michael Widenius committed
4460
  thd_proc_info(thd, "Sql driver thread waiting for slave mutex on exit");
Marc Alff's avatar
Marc Alff committed
4461
  mysql_mutex_lock(&rli->run_lock);
4462
err_during_init:
4463
  /* We need data_lock, at least to wake up any waiting master_pos_wait() */
Marc Alff's avatar
Marc Alff committed
4464
  mysql_mutex_lock(&rli->data_lock);
4465
  DBUG_ASSERT(rli->slave_running == MYSQL_SLAVE_RUN_NOT_CONNECT); // tracking buffer overrun
4466
  /* When master_pos_wait() wakes up it will check this and terminate */
4467
  rli->slave_running= MYSQL_SLAVE_NOT_RUN;
4468 4469 4470
  /* Forget the relay log's format */
  delete rli->relay_log.description_event_for_exec;
  rli->relay_log.description_event_for_exec= 0;
4471
  /* Wake up master_pos_wait() */
Marc Alff's avatar
Marc Alff committed
4472
  mysql_mutex_unlock(&rli->data_lock);
4473
  DBUG_PRINT("info",("Signaling possibly waiting master_pos_wait() functions"));
Marc Alff's avatar
Marc Alff committed
4474
  mysql_cond_broadcast(&rli->data_cond);
unknown's avatar
unknown committed
4475
  rli->ignore_log_space_limit= 0; /* don't need any lock */
4476 4477
  /* we die so won't remember charset - re-update them on next thread start */
  rli->cached_charset_invalidate();
unknown's avatar
unknown committed
4478 4479 4480 4481 4482

  /*
    TODO: see if we can do this conditionally in next_event() instead
    to avoid unneeded position re-init
  */
4483
  thd->temporary_tables = 0; // remove tempation from destructor to close them
4484
  THD_CHECK_SENTRY(thd);
Michael Widenius's avatar
Michael Widenius committed
4485
  serial_rgi->thd= rli->sql_driver_thd= 0;
Marc Alff's avatar
Marc Alff committed
4486
  mysql_mutex_lock(&LOCK_thread_count);
4487
  THD_CHECK_SENTRY(thd);
4488
  delete thd;
unknown's avatar
unknown committed
4489
  delete serial_rgi;
Marc Alff's avatar
Marc Alff committed
4490
  mysql_mutex_unlock(&LOCK_thread_count);
4491 4492 4493 4494 4495
 /*
  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
4496
  mysql_cond_broadcast(&rli->stop_cond);
4497
  DBUG_EXECUTE_IF("simulate_slave_delay_at_terminate_bug38694", sleep(5););
Marc Alff's avatar
Marc Alff committed
4498
  mysql_mutex_unlock(&rli->run_lock);  // tell the world we are done
4499 4500

  DBUG_LEAVE;                                   // Must match DBUG_ENTER()
unknown's avatar
unknown committed
4501
  my_thread_end();
4502
  pthread_exit(0);
4503
  return 0;                                     // Avoid compiler warnings
4504
}
unknown's avatar
unknown committed
4505

4506

unknown's avatar
unknown committed
4507
/*
4508
  process_io_create_file()
unknown's avatar
unknown committed
4509
*/
4510

4511
static int process_io_create_file(Master_info* mi, Create_file_log_event* cev)
unknown's avatar
unknown committed
4512 4513 4514 4515
{
  int error = 1;
  ulong num_bytes;
  bool cev_not_written;
4516 4517
  THD *thd = mi->io_thd;
  NET *net = &mi->mysql->net;
unknown's avatar
unknown committed
4518
  DBUG_ENTER("process_io_create_file");
unknown's avatar
unknown committed
4519 4520

  if (unlikely(!cev->is_valid()))
unknown's avatar
unknown committed
4521
    DBUG_RETURN(1);
4522

4523
  if (!mi->rpl_filter->db_ok(cev->db))
unknown's avatar
unknown committed
4524 4525
  {
    skip_load_data_infile(net);
unknown's avatar
unknown committed
4526
    DBUG_RETURN(0);
unknown's avatar
unknown committed
4527 4528 4529
  }
  DBUG_ASSERT(cev->inited_from_old);
  thd->file_id = cev->file_id = mi->file_id++;
4530
  thd->variables.server_id = cev->server_id;
unknown's avatar
unknown committed
4531
  cev_not_written = 1;
unknown's avatar
unknown committed
4532

unknown's avatar
unknown committed
4533 4534 4535
  if (unlikely(net_request_file(net,cev->fname)))
  {
    sql_print_error("Slave I/O: failed requesting download of '%s'",
unknown's avatar
unknown committed
4536
                    cev->fname);
unknown's avatar
unknown committed
4537 4538 4539
    goto err;
  }

unknown's avatar
unknown committed
4540 4541 4542 4543
  /*
    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
4544 4545
  */
  {
4546
    Append_block_log_event aev(thd,0,0,0,0);
unknown's avatar
unknown committed
4547

unknown's avatar
unknown committed
4548 4549 4550 4551
    for (;;)
    {
      if (unlikely((num_bytes=my_net_read(net)) == packet_error))
      {
unknown's avatar
unknown committed
4552 4553 4554
        sql_print_error("Network read error downloading '%s' from master",
                        cev->fname);
        goto err;
unknown's avatar
unknown committed
4555 4556 4557
      }
      if (unlikely(!num_bytes)) /* eof */
      {
4558 4559
	/* 3.23 master wants it */
        net_write_command(net, 0, (uchar*) "", 0, (uchar*) "", 0);
unknown's avatar
unknown committed
4560 4561 4562 4563 4564 4565 4566
        /*
          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
4567 4568 4569 4570 4571
          break;
        Execute_load_log_event xev(thd,0,0);
        xev.log_pos = cev->log_pos;
        if (unlikely(mi->rli.relay_log.append(&xev)))
        {
4572 4573 4574
          mi->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_WRITE_FAILURE,
                     ER(ER_SLAVE_RELAY_LOG_WRITE_FAILURE),
                     "error writing Exec_load event to relay log");
unknown's avatar
unknown committed
4575 4576 4577 4578
          goto err;
        }
        mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total);
        break;
unknown's avatar
unknown committed
4579 4580 4581
      }
      if (unlikely(cev_not_written))
      {
unknown's avatar
unknown committed
4582
        cev->block = net->read_pos;
unknown's avatar
unknown committed
4583 4584 4585
        cev->block_len = num_bytes;
        if (unlikely(mi->rli.relay_log.append(cev)))
        {
4586 4587 4588
          mi->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_WRITE_FAILURE,
                     ER(ER_SLAVE_RELAY_LOG_WRITE_FAILURE),
                     "error writing Create_file event to relay log");
unknown's avatar
unknown committed
4589 4590 4591 4592
          goto err;
        }
        cev_not_written=0;
        mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total);
unknown's avatar
unknown committed
4593 4594 4595
      }
      else
      {
unknown's avatar
unknown committed
4596
        aev.block = net->read_pos;
unknown's avatar
unknown committed
4597 4598 4599 4600
        aev.block_len = num_bytes;
        aev.log_pos = cev->log_pos;
        if (unlikely(mi->rli.relay_log.append(&aev)))
        {
4601 4602 4603
          mi->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_WRITE_FAILURE,
                     ER(ER_SLAVE_RELAY_LOG_WRITE_FAILURE),
                     "error writing Append_block event to relay log");
unknown's avatar
unknown committed
4604 4605 4606
          goto err;
        }
        mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total) ;
unknown's avatar
unknown committed
4607 4608 4609 4610 4611
      }
    }
  }
  error=0;
err:
unknown's avatar
unknown committed
4612
  DBUG_RETURN(error);
unknown's avatar
unknown committed
4613
}
unknown's avatar
unknown committed
4614

4615

unknown's avatar
unknown committed
4616
/*
unknown's avatar
unknown committed
4617 4618 4619 4620
  Start using a new binary log on the master

  SYNOPSIS
    process_io_rotate()
unknown's avatar
unknown committed
4621 4622
    mi                  master_info for the slave
    rev                 The rotate log event read from the binary log
unknown's avatar
unknown committed
4623 4624

  DESCRIPTION
4625
    Updates the master info with the place in the next binary
unknown's avatar
unknown committed
4626
    log where we should start reading.
4627
    Rotate the relay log to avoid mixed-format relay logs.
unknown's avatar
unknown committed
4628 4629 4630 4631 4632

  NOTES
    We assume we already locked mi->data_lock

  RETURN VALUES
unknown's avatar
unknown committed
4633 4634
    0           ok
    1           Log event is illegal
unknown's avatar
unknown committed
4635 4636 4637

*/

4638
static int process_io_rotate(Master_info *mi, Rotate_log_event *rev)
4639
{
4640
  DBUG_ENTER("process_io_rotate");
Marc Alff's avatar
Marc Alff committed
4641
  mysql_mutex_assert_owner(&mi->data_lock);
4642

unknown's avatar
unknown committed
4643
  if (unlikely(!rev->is_valid()))
4644
    DBUG_RETURN(1);
unknown's avatar
unknown committed
4645

4646
  /* Safe copy as 'rev' has been "sanitized" in Rotate_log_event's ctor */
unknown's avatar
unknown committed
4647 4648
  memcpy(mi->master_log_name, rev->new_log_ident, rev->ident_len+1);
  mi->master_log_pos= rev->pos;
unknown's avatar
unknown committed
4649
  DBUG_PRINT("info", ("master_log_pos: '%s' %lu",
unknown's avatar
unknown committed
4650
                      mi->master_log_name, (ulong) mi->master_log_pos));
4651
#ifndef DBUG_OFF
unknown's avatar
unknown committed
4652 4653 4654 4655 4656
  /*
    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)
4657
    mi->events_till_disconnect++;
4658
#endif
4659

4660 4661 4662 4663 4664 4665 4666 4667 4668
  /*
    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)
  {
4669 4670 4671
    DBUG_ASSERT(mi->rli.relay_log.description_event_for_queue->checksum_alg ==
                mi->rli.relay_log.relay_log_checksum_alg);
    
4672 4673 4674 4675
    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);
4676 4677
    mi->rli.relay_log.description_event_for_queue->checksum_alg=
      mi->rli.relay_log.relay_log_checksum_alg;    
4678
  }
4679 4680 4681 4682
  /*
    Rotate the relay log makes binlog format detection easier (at next slave
    start or mysqlbinlog)
  */
4683
  DBUG_RETURN(rotate_relay_log(mi) /* will take the right mutexes */);
4684 4685
}

unknown's avatar
unknown committed
4686
/*
4687 4688
  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
4689
*/
4690
static int queue_binlog_ver_1_event(Master_info *mi, const char *buf,
unknown's avatar
unknown committed
4691
                           ulong event_len)
4692
{
unknown's avatar
unknown committed
4693
  const char *errmsg = 0;
unknown's avatar
unknown committed
4694 4695 4696
  ulong inc_pos;
  bool ignore_event= 0;
  char *tmp_buf = 0;
4697
  Relay_log_info *rli= &mi->rli;
4698
  DBUG_ENTER("queue_binlog_ver_1_event");
unknown's avatar
unknown committed
4699

unknown's avatar
unknown committed
4700 4701 4702
  /*
    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
4703
  */
4704
  if ((uchar)buf[EVENT_TYPE_OFFSET] == LOAD_EVENT)
unknown's avatar
unknown committed
4705 4706 4707
  {
    if (unlikely(!(tmp_buf=(char*)my_malloc(event_len+1,MYF(MY_WME)))))
    {
4708 4709
      mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
                 ER(ER_SLAVE_FATAL_ERROR), "Memory allocation failed");
unknown's avatar
unknown committed
4710
      DBUG_RETURN(1);
unknown's avatar
unknown committed
4711 4712
    }
    memcpy(tmp_buf,buf,event_len);
4713 4714 4715 4716 4717 4718 4719 4720
    /*
      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
4721
    int4store(tmp_buf+EVENT_LEN_OFFSET, event_len);
unknown's avatar
unknown committed
4722 4723
    buf = (const char*)tmp_buf;
  }
4724 4725 4726 4727 4728 4729
  /*
    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).
  */
4730 4731 4732
  Log_event *ev=
    Log_event::read_log_event(buf, event_len, &errmsg,
                              mi->rli.relay_log.description_event_for_queue, 0);
4733
  if (unlikely(!ev))
4734 4735
  {
    sql_print_error("Read invalid event from master: '%s',\
unknown's avatar
unknown committed
4736
 master could be corrupt but a more likely cause of this is a bug",
unknown's avatar
unknown committed
4737
                    errmsg);
4738
    my_free(tmp_buf);
unknown's avatar
unknown committed
4739
    DBUG_RETURN(1);
4740
  }
4741

Marc Alff's avatar
Marc Alff committed
4742
  mysql_mutex_lock(&mi->data_lock);
4743
  ev->log_pos= mi->master_log_pos; /* 3.23 events don't contain log_pos */
unknown's avatar
unknown committed
4744
  switch (ev->get_type_code()) {
unknown's avatar
unknown committed
4745
  case STOP_EVENT:
4746
    ignore_event= 1;
unknown's avatar
unknown committed
4747 4748
    inc_pos= event_len;
    break;
4749
  case ROTATE_EVENT:
4750
    if (unlikely(process_io_rotate(mi,(Rotate_log_event*)ev)))
4751 4752
    {
      delete ev;
Marc Alff's avatar
Marc Alff committed
4753
      mysql_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
4754
      DBUG_RETURN(1);
4755
    }
unknown's avatar
unknown committed
4756
    inc_pos= 0;
4757
    break;
unknown's avatar
unknown committed
4758
  case CREATE_FILE_EVENT:
4759 4760 4761 4762 4763 4764
    /*
      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
4765
  {
unknown's avatar
unknown committed
4766
    /* We come here when and only when tmp_buf != 0 */
4767
    DBUG_ASSERT(tmp_buf != 0);
4768 4769
    inc_pos=event_len;
    ev->log_pos+= inc_pos;
unknown's avatar
unknown committed
4770
    int error = process_io_create_file(mi,(Create_file_log_event*)ev);
4771
    delete ev;
4772
    mi->master_log_pos += inc_pos;
unknown's avatar
unknown committed
4773
    DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
Marc Alff's avatar
Marc Alff committed
4774
    mysql_mutex_unlock(&mi->data_lock);
4775
    my_free(tmp_buf);
unknown's avatar
unknown committed
4776
    DBUG_RETURN(error);
unknown's avatar
unknown committed
4777
  }
4778
  default:
unknown's avatar
unknown committed
4779
    inc_pos= event_len;
4780 4781
    break;
  }
unknown's avatar
unknown committed
4782
  if (likely(!ignore_event))
4783
  {
unknown's avatar
unknown committed
4784 4785
    if (ev->log_pos)
      /*
4786 4787 4788 4789
         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
4790
    if (unlikely(rli->relay_log.append(ev)))
4791 4792
    {
      delete ev;
Marc Alff's avatar
Marc Alff committed
4793
      mysql_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
4794
      DBUG_RETURN(1);
4795
    }
unknown's avatar
unknown committed
4796
    rli->relay_log.harvest_bytes_written(&rli->log_space_total);
4797 4798
  }
  delete ev;
unknown's avatar
unknown committed
4799
  mi->master_log_pos+= inc_pos;
unknown's avatar
unknown committed
4800
  DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
Marc Alff's avatar
Marc Alff committed
4801
  mysql_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
4802
  DBUG_RETURN(0);
4803 4804
}

4805 4806 4807 4808
/*
  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.
*/
4809
static int queue_binlog_ver_3_event(Master_info *mi, const char *buf,
unknown's avatar
unknown committed
4810
                           ulong event_len)
4811 4812 4813 4814
{
  const char *errmsg = 0;
  ulong inc_pos;
  char *tmp_buf = 0;
4815
  Relay_log_info *rli= &mi->rli;
4816 4817 4818
  DBUG_ENTER("queue_binlog_ver_3_event");

  /* read_log_event() will adjust log_pos to be end_log_pos */
4819 4820 4821
  Log_event *ev=
    Log_event::read_log_event(buf,event_len, &errmsg,
                              mi->rli.relay_log.description_event_for_queue, 0);
4822 4823 4824 4825
  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
4826
                    errmsg);
4827
    my_free(tmp_buf);
4828 4829
    DBUG_RETURN(1);
  }
Marc Alff's avatar
Marc Alff committed
4830
  mysql_mutex_lock(&mi->data_lock);
4831 4832 4833 4834 4835 4836 4837
  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
4838
      mysql_mutex_unlock(&mi->data_lock);
4839 4840 4841 4842 4843 4844 4845 4846
      DBUG_RETURN(1);
    }
    inc_pos= 0;
    break;
  default:
    inc_pos= event_len;
    break;
  }
4847

4848 4849 4850
  if (unlikely(rli->relay_log.append(ev)))
  {
    delete ev;
Marc Alff's avatar
Marc Alff committed
4851
    mysql_mutex_unlock(&mi->data_lock);
4852 4853 4854 4855 4856 4857
    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
4858
  DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
Marc Alff's avatar
Marc Alff committed
4859
  mysql_mutex_unlock(&mi->data_lock);
4860 4861 4862 4863 4864 4865 4866 4867 4868 4869
  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
4870
  TODO:
4871 4872 4873 4874
    Test this code before release - it has to be tested on a separate
    setup with 3.23 master or 4.0 master
*/

4875
static int queue_old_event(Master_info *mi, const char *buf,
unknown's avatar
unknown committed
4876
                           ulong event_len)
4877
{
4878 4879
  DBUG_ENTER("queue_old_event");

4880 4881 4882
  switch (mi->rli.relay_log.description_event_for_queue->binlog_version)
  {
  case 1:
4883
      DBUG_RETURN(queue_binlog_ver_1_event(mi,buf,event_len));
4884
  case 3:
4885
      DBUG_RETURN(queue_binlog_ver_3_event(mi,buf,event_len));
4886 4887
  default: /* unsupported format; eg version 2 */
    DBUG_PRINT("info",("unsupported binlog format %d in queue_old_event()",
unknown's avatar
unknown committed
4888
                       mi->rli.relay_log.description_event_for_queue->binlog_version));
4889
    DBUG_RETURN(1);
4890 4891
  }
}
4892

unknown's avatar
unknown committed
4893
/*
4894 4895
  queue_event()

4896 4897 4898 4899 4900
  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
4901 4902
*/

4903
static int queue_event(Master_info* mi,const char* buf, ulong event_len)
4904
{
unknown's avatar
unknown committed
4905
  int error= 0;
Andrei Elkin's avatar
Andrei Elkin committed
4906
  String error_msg;
4907 4908
  ulonglong inc_pos;
  ulonglong event_pos;
4909
  Relay_log_info *rli= &mi->rli;
Marc Alff's avatar
Marc Alff committed
4910
  mysql_mutex_t *log_lock= rli->relay_log.get_log_lock();
Andrei Elkin's avatar
Andrei Elkin committed
4911
  ulong s_id;
4912
  bool unlock_data_lock= TRUE;
unknown's avatar
unknown committed
4913 4914
  bool gtid_skip_enqueue= false;

4915 4916 4917 4918 4919 4920 4921 4922 4923 4924 4925 4926 4927 4928 4929 4930 4931
  /*
    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
4932
  DBUG_ENTER("queue_event");
4933 4934 4935 4936 4937 4938 4939 4940 4941 4942 4943 4944 4945 4946 4947 4948 4949 4950 4951 4952 4953 4954 4955 4956 4957 4958 4959 4960 4961 4962 4963 4964 4965 4966 4967 4968 4969 4970 4971 4972 4973 4974 4975
  /*
    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
4976

unknown's avatar
unknown committed
4977 4978
  LINT_INIT(inc_pos);

4979
  if (mi->rli.relay_log.description_event_for_queue->binlog_version<4 &&
4980
      (uchar)buf[EVENT_TYPE_OFFSET] != FORMAT_DESCRIPTION_EVENT /* a way to escape */)
unknown's avatar
unknown committed
4981
    DBUG_RETURN(queue_old_event(mi,buf,event_len));
4982

Marc Alff's avatar
Marc Alff committed
4983
  mysql_mutex_lock(&mi->data_lock);
unknown's avatar
unknown committed
4984

4985
  switch ((uchar)buf[EVENT_TYPE_OFFSET]) {
4986
  case STOP_EVENT:
4987 4988
    /*
      We needn't write this event to the relay log. Indeed, it just indicates a
unknown's avatar
unknown committed
4989 4990 4991
      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
4992
      prepared statements' deletion are TODO only when we binlog prep stmts).
unknown's avatar
unknown committed
4993

unknown's avatar
unknown committed
4994 4995 4996 4997
      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).
4998 4999
    */
    goto err;
5000 5001
  case ROTATE_EVENT:
  {
5002 5003 5004 5005 5006
    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);

    if (unlikely(process_io_rotate(mi, &rev)))
unknown's avatar
unknown committed
5007
    {
Andrei Elkin's avatar
Andrei Elkin committed
5008
      error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
5009
      goto err;
unknown's avatar
unknown committed
5010
    }
5011 5012 5013 5014 5015 5016 5017 5018 5019 5020 5021 5022 5023 5024 5025 5026 5027 5028 5029 5030 5031 5032 5033 5034 5035 5036 5037 5038 5039 5040 5041 5042 5043 5044 5045 5046 5047 5048 5049 5050 5051 5052 5053 5054 5055 5056 5057 5058 5059 5060 5061 5062
    /* 
       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)
    {
      ha_checksum rot_crc= my_checksum(0L, NULL, 0);
      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;
      }
5063 5064 5065 5066
    /*
      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
5067
    inc_pos= 0;
5068 5069
    break;
  }
5070 5071 5072 5073 5074 5075 5076
  case FORMAT_DESCRIPTION_EVENT:
  {
    /*
      Create an event, and save it (when we rotate the relay log, we will have
      to write this event again).
    */
    /*
5077 5078 5079
      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).
5080
    */
5081
    Format_description_log_event* tmp;
5082
    const char* errmsg;
5083 5084
    // mark it as undefined that is irrelevant anymore
    mi->checksum_alg_before_fd= BINLOG_CHECKSUM_ALG_UNDEF;
5085
    if (!(tmp= (Format_description_log_event*)
5086
          Log_event::read_log_event(buf, event_len, &errmsg,
5087 5088
                                    mi->rli.relay_log.description_event_for_queue,
                                    1)))
5089
    {
Andrei Elkin's avatar
Andrei Elkin committed
5090
      error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
5091 5092
      goto err;
    }
5093 5094
    delete mi->rli.relay_log.description_event_for_queue;
    mi->rli.relay_log.description_event_for_queue= tmp;
5095 5096 5097 5098 5099 5100
    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
5101 5102 5103 5104 5105 5106 5107 5108
    /*
      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
5109
      previous master crash and should be rolled back).
unknown's avatar
unknown committed
5110
    */
5111
    if (unlikely(mi->gtid_reconnect_event_skip_count && !mi->gtid_event_seen))
unknown's avatar
unknown committed
5112 5113
        gtid_skip_enqueue= true;

unknown's avatar
unknown committed
5114
    /*
5115
       Though this does some conversion to the slave's format, this will
unknown's avatar
unknown committed
5116
       preserve the master's binlog format version, and number of event types.
5117
    */
unknown's avatar
unknown committed
5118
    /*
5119
       If the event was not requested by the slave (the slave did not ask for
unknown's avatar
unknown committed
5120
       it), i.e. has end_log_pos=0, we do not increment mi->master_log_pos
5121 5122 5123
    */
    inc_pos= uint4korr(buf+LOG_POS_OFFSET) ? event_len : 0;
    DBUG_PRINT("info",("binlog format is now %d",
unknown's avatar
unknown committed
5124
                       mi->rli.relay_log.description_event_for_queue->binlog_version));
5125 5126 5127

  }
  break;
Andrei Elkin's avatar
Andrei Elkin committed
5128 5129 5130 5131 5132 5133 5134

  case HEARTBEAT_LOG_EVENT:
  {
    /*
      HB (heartbeat) cannot come before RL (Relay)
    */
    char  llbuf[22];
unknown's avatar
unknown committed
5135 5136 5137 5138 5139
    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
5140 5141 5142 5143 5144 5145 5146 5147 5148 5149 5150 5151 5152 5153 5154 5155 5156
    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
5157
       Slave can not have a higher coordinate except in the only
Andrei Elkin's avatar
Andrei Elkin committed
5158 5159 5160 5161
       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
5162 5163
       Slave can have lower coordinates, if some event from master was omitted.

Andrei Elkin's avatar
Andrei Elkin committed
5164 5165 5166 5167
       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
5168
        || mi->master_log_pos > hb.log_pos)
Andrei Elkin's avatar
Andrei Elkin committed
5169 5170 5171 5172 5173 5174 5175 5176 5177 5178 5179 5180 5181 5182
    {
      /* 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;
5183

unknown's avatar
unknown committed
5184 5185 5186 5187 5188
  case GTID_LIST_EVENT:
  {
    const char *errmsg;
    Gtid_list_log_event *glev;
    Log_event *tmp;
5189
    uint32 flags;
unknown's avatar
unknown committed
5190 5191 5192 5193 5194 5195 5196 5197 5198

    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
5199
    event_pos= glev->log_pos;
5200
    flags= glev->gl_flags;
unknown's avatar
unknown committed
5201 5202 5203
    delete glev;

    /*
unknown's avatar
unknown committed
5204 5205 5206 5207 5208
      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
5209
    */
unknown's avatar
unknown committed
5210 5211 5212 5213
    if (event_pos == 0 || event_pos <= mi->master_log_pos)
      inc_pos= 0;
    else
      inc_pos= event_pos - mi->master_log_pos;
5214 5215 5216 5217 5218 5219 5220 5221 5222 5223 5224

    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);
      sql_print_information("Slave IO thread stops because it reached its"
                            " UNTIL master_gtid_pos %s", str.c_ptr_safe());
      mi->abort_slave= true;
    }
unknown's avatar
unknown committed
5225 5226 5227 5228 5229 5230 5231 5232 5233 5234 5235 5236 5237 5238 5239 5240 5241 5242 5243 5244 5245 5246 5247 5248 5249 5250 5251 5252 5253 5254 5255 5256 5257 5258 5259 5260 5261 5262 5263 5264 5265 5266 5267 5268 5269 5270 5271 5272 5273 5274 5275 5276 5277 5278 5279 5280 5281 5282 5283 5284 5285 5286 5287 5288 5289 5290 5291 5292 5293 5294 5295 5296 5297 5298 5299 5300 5301
  }
  break;

  case GTID_EVENT:
  {
    uchar dummy_flag;

    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)
      {
        rpl_gtid gtid;

        /*
          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
          master server) was replaced with a different one, one the same IP
          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.
        */
        if (Gtid_log_event::peek(buf, event_len, checksum_alg,
                                 &gtid.domain_id, &gtid.server_id,
                                 &gtid.seq_no, &dummy_flag))
        {
          error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
          goto err;
        }
        if (gtid.domain_id != mi->last_queued_gtid.domain_id ||
            gtid.server_id != mi->last_queued_gtid.server_id ||
            gtid.seq_no != mi->last_queued_gtid.seq_no)
        {
          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;
          rpl_slave_state_tostring_helper(&error_msg, &gtid, &first);
          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.
    */
    if (mi->events_queued_since_last_gtid)
    {
      mi->gtid_current_pos.update(&mi->last_queued_gtid);
      mi->events_queued_since_last_gtid= 0;
    }
    if (Gtid_log_event::peek(buf, event_len, checksum_alg,
                             &mi->last_queued_gtid.domain_id,
                             &mi->last_queued_gtid.server_id,
                             &mi->last_queued_gtid.seq_no, &dummy_flag))
    {
      error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
      goto err;
    }
    ++mi->events_queued_since_last_gtid;
unknown's avatar
unknown committed
5302 5303 5304
  }
  break;

5305
  default:
unknown's avatar
unknown committed
5306
  default_action:
unknown's avatar
unknown committed
5307 5308 5309 5310 5311 5312 5313 5314 5315 5316 5317
    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
5318
    inc_pos= event_len;
5319 5320
    break;
  }
5321

5322
  /*
5323
    If we filter events master-side (eg. @@skip_replication), we will see holes
5324 5325 5326 5327 5328 5329 5330 5331 5332
    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
5333
    DBUG_PRINT("info", ("Adjust master_log_pos %llu->%llu to account for "
5334
                        "master-side filtering",
Sergei Golubchik's avatar
Sergei Golubchik committed
5335
                        mi->master_log_pos + inc_pos, event_pos));
5336 5337
  }

unknown's avatar
unknown committed
5338 5339
  /*
     If this event is originating from this server, don't queue it.
5340
     We don't check this for 3.23 events because it's simpler like this; 3.23
unknown's avatar
unknown committed
5341 5342
     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
5343 5344 5345 5346 5347 5348 5349 5350
     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
5351
  mysql_mutex_lock(log_lock);
Andrei Elkin's avatar
Andrei Elkin committed
5352
  s_id= uint4korr(buf + SERVER_ID_OFFSET);
5353 5354
  if ((s_id == global_system_variables.server_id &&
       !mi->rli.replicate_same_server_id) ||
Andrei Elkin's avatar
Andrei Elkin committed
5355 5356 5357 5358 5359 5360 5361 5362 5363 5364
      /*
        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 */
5365 5366
        (buf[EVENT_TYPE_OFFSET] != FORMAT_DESCRIPTION_EVENT &&
         buf[EVENT_TYPE_OFFSET] != ROTATE_EVENT))))
5367
  {
5368 5369
    /*
      Do not write it to the relay log.
5370 5371 5372 5373 5374 5375 5376 5377
      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.
5378 5379 5380
      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
5381 5382 5383
      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.
5384
    */
5385 5386
    if (!(s_id == global_system_variables.server_id &&
          !mi->rli.replicate_same_server_id) ||
5387 5388 5389
        (buf[EVENT_TYPE_OFFSET] != FORMAT_DESCRIPTION_EVENT &&
         buf[EVENT_TYPE_OFFSET] != ROTATE_EVENT &&
         buf[EVENT_TYPE_OFFSET] != STOP_EVENT))
unknown's avatar
unknown committed
5390
    {
5391
      mi->master_log_pos+= inc_pos;
unknown's avatar
unknown committed
5392 5393 5394 5395
      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;
    }
5396
    rli->relay_log.signal_update(); // the slave SQL thread needs to re-check
Andrei Elkin's avatar
Andrei Elkin committed
5397 5398
    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
5399
  }
unknown's avatar
unknown committed
5400 5401
  else
  {
unknown's avatar
unknown committed
5402 5403 5404 5405 5406 5407 5408 5409 5410 5411
    /*
      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;
    }
    else if (likely(!(rli->relay_log.appendv(buf,event_len,0))))
5412 5413
    {
      mi->master_log_pos+= inc_pos;
unknown's avatar
unknown committed
5414
      DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
5415 5416
      rli->relay_log.harvest_bytes_written(&rli->log_space_total);
    }
5417
    else
Andrei Elkin's avatar
Andrei Elkin committed
5418 5419 5420
    {
      error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
    }
5421
    rli->ign_master_log_name_end[0]= 0; // last event is not ignored
5422 5423
    if (save_buf != NULL)
      buf= save_buf;
unknown's avatar
unknown committed
5424
  }
Marc Alff's avatar
Marc Alff committed
5425
  mysql_mutex_unlock(log_lock);
5426

Andrei Elkin's avatar
Andrei Elkin committed
5427 5428
skip_relay_logging:
  
5429
err:
5430
  if (unlock_data_lock)
Sergei Golubchik's avatar
Sergei Golubchik committed
5431
    mysql_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
5432
  DBUG_PRINT("info", ("error: %d", error));
Andrei Elkin's avatar
Andrei Elkin committed
5433 5434 5435 5436 5437
  if (error)
    mi->report(ERROR_LEVEL, error, ER(error), 
               (error == ER_SLAVE_RELAY_LOG_WRITE_FAILURE)?
               "could not queue event from master" :
               error_msg.ptr());
unknown's avatar
unknown committed
5438
  DBUG_RETURN(error);
5439 5440
}

5441

5442
void end_relay_log_info(Relay_log_info* rli)
5443
{
5444 5445
  DBUG_ENTER("end_relay_log_info");

5446
  if (!rli->inited)
5447
    DBUG_VOID_RETURN;
5448
  if (rli->info_fd >= 0)
unknown's avatar
unknown committed
5449 5450
  {
    end_io_cache(&rli->info_file);
Marc Alff's avatar
Marc Alff committed
5451
    mysql_file_close(rli->info_fd, MYF(MY_WME));
unknown's avatar
unknown committed
5452 5453
    rli->info_fd = -1;
  }
5454
  if (rli->cur_log_fd >= 0)
unknown's avatar
unknown committed
5455 5456
  {
    end_io_cache(&rli->cache_buf);
Marc Alff's avatar
Marc Alff committed
5457
    mysql_file_close(rli->cur_log_fd, MYF(MY_WME));
unknown's avatar
unknown committed
5458 5459
    rli->cur_log_fd = -1;
  }
5460
  rli->inited = 0;
5461
  rli->relay_log.close(LOG_CLOSE_INDEX | LOG_CLOSE_STOP_EVENT);
5462
  rli->relay_log.harvest_bytes_written(&rli->log_space_total);
unknown's avatar
unknown committed
5463 5464 5465 5466 5467 5468
  /*
    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();
5469
  DBUG_VOID_RETURN;
5470 5471
}

5472 5473 5474 5475 5476 5477 5478 5479 5480 5481 5482 5483 5484 5485 5486 5487 5488 5489 5490

/**
  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;
5491
  if (thd && thd->slave_thread)
5492 5493 5494 5495 5496
    thd->clear_active_vio();
#endif
}


unknown's avatar
unknown committed
5497 5498
/*
  Try to connect until successful or slave killed
5499

unknown's avatar
unknown committed
5500 5501
  SYNPOSIS
    safe_connect()
unknown's avatar
unknown committed
5502 5503 5504
    thd                 Thread handler for slave
    mysql               MySQL connection handle
    mi                  Replication handle
5505

unknown's avatar
unknown committed
5506
  RETURN
unknown's avatar
unknown committed
5507 5508
    0   ok
    #   Error
unknown's avatar
unknown committed
5509
*/
5510

5511
static int safe_connect(THD* thd, MYSQL* mysql, Master_info* mi)
unknown's avatar
unknown committed
5512
{
5513 5514 5515
  DBUG_ENTER("safe_connect");

  DBUG_RETURN(connect_to_master(thd, mysql, mi, 0, 0));
unknown's avatar
unknown committed
5516 5517
}

5518

5519
/*
unknown's avatar
unknown committed
5520 5521
  SYNPOSIS
    connect_to_master()
unknown's avatar
unknown committed
5522

unknown's avatar
unknown committed
5523 5524 5525
  IMPLEMENTATION
    Try to connect until successful or slave killed or we have retried
    master_retry_count times
5526
*/
unknown's avatar
unknown committed
5527

5528
static int connect_to_master(THD* thd, MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
5529
                             bool reconnect, bool suppress_warnings)
unknown's avatar
unknown committed
5530
{
5531
  int slave_was_killed;
unknown's avatar
unknown committed
5532
  int last_errno= -2;                           // impossible error
5533
  ulong err_count=0;
unknown's avatar
unknown committed
5534
  char llbuff[22];
5535
  my_bool my_true= 1;
5536
  DBUG_ENTER("connect_to_master");
unknown's avatar
unknown committed
5537

unknown's avatar
unknown committed
5538
#ifndef DBUG_OFF
5539
  mi->events_till_disconnect = disconnect_slave_event_count;
unknown's avatar
unknown committed
5540
#endif
5541
  ulong client_flag= CLIENT_REMEMBER_OPTIONS;
5542
  if (opt_slave_compressed_protocol)
unknown's avatar
unknown committed
5543
    client_flag=CLIENT_COMPRESS;                /* We will use compression */
5544

5545 5546
  mysql_options(mysql, MYSQL_OPT_CONNECT_TIMEOUT, (char *) &slave_net_timeout);
  mysql_options(mysql, MYSQL_OPT_READ_TIMEOUT, (char *) &slave_net_timeout);
5547 5548
  mysql_options(mysql, MYSQL_OPT_USE_THREAD_SPECIFIC_MEMORY,
                (char*) &my_true);
unknown's avatar
unknown committed
5549

unknown's avatar
unknown committed
5550 5551
#ifdef HAVE_OPENSSL
  if (mi->ssl)
5552
  {
unknown's avatar
unknown committed
5553
    mysql_ssl_set(mysql,
unknown's avatar
unknown committed
5554
                  mi->ssl_key[0]?mi->ssl_key:0,
unknown's avatar
unknown committed
5555
                  mi->ssl_cert[0]?mi->ssl_cert:0,
unknown's avatar
unknown committed
5556 5557 5558
                  mi->ssl_ca[0]?mi->ssl_ca:0,
                  mi->ssl_capath[0]?mi->ssl_capath:0,
                  mi->ssl_cipher[0]?mi->ssl_cipher:0);
5559 5560 5561
    mysql_options(mysql, MYSQL_OPT_SSL_VERIFY_SERVER_CERT,
                  &mi->ssl_verify_server_cert);
  }
unknown's avatar
unknown committed
5562 5563
#endif

5564 5565 5566 5567
  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);

5568 5569 5570 5571
  /* Set MYSQL_PLUGIN_DIR in case master asks for an external authentication plugin */
  if (opt_plugin_dir_ptr && *opt_plugin_dir_ptr)
    mysql_options(mysql, MYSQL_PLUGIN_DIR, opt_plugin_dir_ptr);

5572 5573 5574 5575 5576 5577 5578 5579 5580 5581
  /* we disallow empty users */
  if (mi->user == NULL || mi->user[0] == 0)
  {
    mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
               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
5582
  while (!(slave_was_killed = io_slave_killed(mi)) &&
unknown's avatar
unknown committed
5583 5584 5585
         (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
5586
  {
5587
    /* Don't repeat last error */
unknown's avatar
SCRUM  
unknown committed
5588
    if ((int)mysql_errno(mysql) != last_errno)
5589
    {
unknown's avatar
SCRUM  
unknown committed
5590
      last_errno=mysql_errno(mysql);
unknown's avatar
unknown committed
5591
      suppress_warnings= 0;
5592 5593
      mi->report(ERROR_LEVEL, last_errno,
                 "error %s to master '%s@%s:%d'"
5594
                 " - retry-time: %d  retries: %lu  message: %s",
5595 5596
                 (reconnect ? "reconnecting" : "connecting"),
                 mi->user, mi->host, mi->port,
5597 5598
                 mi->connect_retry, master_retry_count,
                 mysql_error(mysql));
5599
    }
unknown's avatar
unknown committed
5600 5601 5602
    /*
      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
5603
      do not want to have election triggered on the first failure to
unknown's avatar
unknown committed
5604
      connect
5605
    */
5606
    if (++err_count == master_retry_count)
5607 5608
    {
      slave_was_killed=1;
unknown's avatar
unknown committed
5609 5610
      if (reconnect)
        change_rpl_status(RPL_ACTIVE_SLAVE,RPL_LOST_SOLDIER);
5611 5612
      break;
    }
5613
    slave_sleep(thd,mi->connect_retry,io_slave_killed, mi);
unknown's avatar
unknown committed
5614
  }
5615

5616 5617
  if (!slave_was_killed)
  {
5618
    mi->clear_error(); // clear possible left over reconnect error
unknown's avatar
unknown committed
5619
    if (reconnect)
unknown's avatar
unknown committed
5620
    {
5621
      if (!suppress_warnings && global_system_variables.log_warnings)
unknown's avatar
unknown committed
5622
        sql_print_information("Slave: connected to master '%s@%s:%d',\
5623
replication resumed in log '%s' at position %s", mi->user,
unknown's avatar
unknown committed
5624 5625 5626
                        mi->host, mi->port,
                        IO_RPL_LOG_NAME,
                        llstr(mi->master_log_pos,llbuff));
unknown's avatar
unknown committed
5627
    }
unknown's avatar
unknown committed
5628 5629 5630
    else
    {
      change_rpl_status(RPL_IDLE_SLAVE,RPL_ACTIVE_SLAVE);
5631 5632
      general_log_print(thd, COM_CONNECT_OUT, "%s@%s:%d",
                        mi->user, mi->host, mi->port);
unknown's avatar
unknown committed
5633
    }
5634
#ifdef SIGNAL_WITH_VIO_CLOSE
5635
    thd->set_active_vio(mysql->net.vio);
unknown's avatar
unknown committed
5636
#endif
5637
  }
5638
  mysql->reconnect= 1;
5639 5640
  DBUG_PRINT("exit",("slave_was_killed: %d", slave_was_killed));
  DBUG_RETURN(slave_was_killed);
unknown's avatar
unknown committed
5641 5642
}

5643

unknown's avatar
unknown committed
5644
/*
5645
  safe_reconnect()
unknown's avatar
unknown committed
5646

unknown's avatar
unknown committed
5647 5648 5649
  IMPLEMENTATION
    Try to connect until successful or slave killed or we have retried
    master_retry_count times
unknown's avatar
unknown committed
5650 5651
*/

5652
static int safe_reconnect(THD* thd, MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
5653
                          bool suppress_warnings)
unknown's avatar
unknown committed
5654
{
unknown's avatar
unknown committed
5655 5656
  DBUG_ENTER("safe_reconnect");
  DBUG_RETURN(connect_to_master(thd, mysql, mi, 1, suppress_warnings));
unknown's avatar
unknown committed
5657 5658
}

unknown's avatar
unknown committed
5659

Michael Widenius's avatar
Michael Widenius committed
5660
#ifdef NOT_USED
He Zhenxing's avatar
He Zhenxing committed
5661 5662 5663
MYSQL *rpl_connect_master(MYSQL *mysql)
{
  Master_info *mi= my_pthread_getspecific_ptr(Master_info*, RPL_MASTER_INFO);
5664 5665
  bool allocated= false;
  my_bool my_true= 1;
Michael Widenius's avatar
Michael Widenius committed
5666
  THD *thd;
5667

He Zhenxing's avatar
He Zhenxing committed
5668 5669 5670 5671 5672
  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
5673
  thd= mi->io_thd;
He Zhenxing's avatar
He Zhenxing committed
5674 5675 5676 5677 5678 5679 5680 5681 5682 5683 5684 5685 5686 5687 5688 5689 5690 5691 5692
  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);
5693 5694
  mysql_options(mysql, MYSQL_OPT_USE_THREAD_SPECIFIC_MEMORY,
                (char*) &my_true);
He Zhenxing's avatar
He Zhenxing committed
5695 5696 5697 5698 5699 5700 5701 5702 5703 5704 5705 5706 5707 5708 5709 5710 5711 5712 5713

#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);

5714 5715
  if (mi->user == NULL
      || mi->user[0] == 0
Michael Widenius's avatar
Michael Widenius committed
5716
      || io_slave_killed( mi)
He Zhenxing's avatar
He Zhenxing committed
5717 5718 5719
      || !mysql_real_connect(mysql, mi->host, mi->user, mi->password, 0,
                             mi->port, 0, 0))
  {
Michael Widenius's avatar
Michael Widenius committed
5720
    if (!io_slave_killed( mi))
He Zhenxing's avatar
He Zhenxing committed
5721 5722 5723 5724 5725 5726 5727 5728 5729
      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
5730
#endif
He Zhenxing's avatar
He Zhenxing committed
5731

5732 5733 5734 5735 5736 5737
/*
  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
5738
    rli                 Relay log information
5739 5740

  NOTES
5741 5742 5743
    - 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.
5744 5745 5746 5747 5748 5749 5750 5751 5752 5753 5754 5755 5756 5757
    - 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
5758 5759
    0   ok
    1   write error
5760 5761
*/

5762
bool flush_relay_log_info(Relay_log_info* rli)
5763
{
5764
  bool error=0;
5765
  DBUG_ENTER("flush_relay_log_info");
5766 5767

  if (unlikely(rli->no_storage))
5768
    DBUG_RETURN(0);
5769

5770 5771 5772
  IO_CACHE *file = &rli->info_file;
  char buff[FN_REFLEN*2+22*2+4], *pos;

5773
  my_b_seek(file, 0L);
5774
  pos=strmov(buff, rli->group_relay_log_name);
5775
  *pos++='\n';
5776
  pos= longlong10_to_str(rli->group_relay_log_pos, pos, 10);
5777
  *pos++='\n';
5778
  pos=strmov(pos, rli->group_master_log_name);
5779
  *pos++='\n';
5780
  pos=longlong10_to_str(rli->group_master_log_pos, pos, 10);
5781
  *pos='\n';
5782
  if (my_b_write(file, (uchar*) buff, (size_t) (pos-buff)+1))
5783 5784 5785
    error=1;
  if (flush_io_cache(file))
    error=1;
5786 5787 5788 5789 5790 5791 5792 5793
  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;
  }
5794 5795 5796 5797
  /* 
    Flushing the relay log is done by the slave I/O thread 
    or by the user on STOP SLAVE. 
   */
5798
  DBUG_RETURN(error);
5799 5800
}

5801

unknown's avatar
unknown committed
5802
/*
5803
  Called when we notice that the current "hot" log got rotated under our feet.
unknown's avatar
unknown committed
5804 5805
*/

5806
static IO_CACHE *reopen_relay_log(Relay_log_info *rli, const char **errmsg)
5807
{
5808
  DBUG_ENTER("reopen_relay_log");
5809 5810
  DBUG_ASSERT(rli->cur_log != &rli->cache_buf);
  DBUG_ASSERT(rli->cur_log_fd == -1);
unknown's avatar
unknown committed
5811 5812

  IO_CACHE *cur_log = rli->cur_log=&rli->cache_buf;
5813
  if ((rli->cur_log_fd=open_binlog(cur_log,rli->event_relay_log_name,
unknown's avatar
unknown committed
5814
                                   errmsg)) <0)
unknown's avatar
unknown committed
5815
    DBUG_RETURN(0);
5816 5817
  /*
    We want to start exactly where we was before:
unknown's avatar
unknown committed
5818 5819
    relay_log_pos       Current log pos
    pending             Number of bytes already processed from the event
5820
  */
5821
  rli->event_relay_log_pos= max(rli->event_relay_log_pos, BIN_LOG_HEADER_SIZE);
5822
  my_b_seek(cur_log,rli->event_relay_log_pos);
unknown's avatar
unknown committed
5823
  DBUG_RETURN(cur_log);
5824 5825
}

unknown's avatar
unknown committed
5826

5827 5828 5829 5830 5831 5832 5833 5834 5835 5836
/**
  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.
*/
Michael Widenius's avatar
Michael Widenius committed
5837
static Log_event* next_event(rpl_group_info *rgi)
5838 5839
{
  Log_event* ev;
Michael Widenius's avatar
Michael Widenius committed
5840
  Relay_log_info *rli= rgi->rli;
5841
  IO_CACHE* cur_log = rli->cur_log;
Marc Alff's avatar
Marc Alff committed
5842
  mysql_mutex_t *log_lock = rli->relay_log.get_log_lock();
5843
  const char* errmsg=0;
Michael Widenius's avatar
Michael Widenius committed
5844
  THD *thd = rgi->thd;
unknown's avatar
unknown committed
5845
  DBUG_ENTER("next_event");
5846

Michael Widenius's avatar
Michael Widenius committed
5847
  DBUG_ASSERT(thd != 0 && thd == rli->sql_driver_thd);
5848

5849 5850 5851 5852 5853
#ifndef DBUG_OFF
  if (abort_slave_event_count && !rli->events_till_abort--)
    DBUG_RETURN(0);
#endif

unknown's avatar
unknown committed
5854 5855
  /*
    For most operations we need to protect rli members with data_lock,
5856 5857 5858
    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
5859
    mysql_cond_wait() with the non-data_lock mutex
unknown's avatar
unknown committed
5860
  */
Marc Alff's avatar
Marc Alff committed
5861
  mysql_mutex_assert_owner(&rli->data_lock);
unknown's avatar
unknown committed
5862

Michael Widenius's avatar
Michael Widenius committed
5863
  while (!sql_slave_killed(rgi))
unknown's avatar
unknown committed
5864 5865 5866
  {
    /*
      We can have two kinds of log reading:
unknown's avatar
unknown committed
5867 5868 5869 5870 5871 5872 5873 5874
      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
5875
    */
5876 5877 5878 5879
    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
5880
      mysql_mutex_lock(log_lock);
unknown's avatar
unknown committed
5881 5882

      /*
unknown's avatar
unknown committed
5883 5884
        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
5885
      */
unknown's avatar
unknown committed
5886
      if (rli->relay_log.get_open_count() != rli->cur_log_old_open_count)
5887
      {
unknown's avatar
unknown committed
5888 5889
        // 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
5890
        mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
5891 5892 5893
        if (!cur_log)                           // No more log files
          goto err;
        hot_log=0;                              // Using old binary log
5894 5895
      }
    }
5896 5897 5898 5899 5900 5901 5902
    /* 
      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;
5903 5904
#ifndef DBUG_OFF
    {
5905
      /* This is an assertion which sometimes fails, let's try to track it */
5906
      char llbuf1[22], llbuf2[22];
5907
      DBUG_PRINT("info", ("my_b_tell(cur_log)=%s rli->event_relay_log_pos=%s",
5908
                          llstr(my_b_tell(cur_log),llbuf1),
5909 5910
                          llstr(rli->event_relay_log_pos,llbuf2)));
      DBUG_ASSERT(my_b_tell(cur_log) >= BIN_LOG_HEADER_SIZE);
5911 5912
      DBUG_ASSERT(opt_slave_parallel_threads > 0 ||
                  my_b_tell(cur_log) == rli->event_relay_log_pos);
5913 5914
    }
#endif
unknown's avatar
unknown committed
5915 5916
    /*
      Relay log is always in new format - if the master is 3.23, the
5917
      I/O thread will convert the format for us.
5918 5919
      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
5920 5921
      logs, which may even have been deleted. So we need to write this
      description event at the beginning of the relay log.
5922 5923
      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.
5924
      But if the relay log is created by new_file(): then the solution is:
5925
      MYSQL_BIN_LOG::open() will write the buffered description event.
unknown's avatar
unknown committed
5926
    */
5927 5928 5929
    if ((ev= Log_event::read_log_event(cur_log,0,
                                       rli->relay_log.description_event_for_exec,
                                       opt_slave_sql_verify_checksum)))
5930

5931
    {
5932 5933 5934 5935 5936
      /*
        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);
5937

5938
      if (hot_log)
Marc Alff's avatar
Marc Alff committed
5939
        mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
5940
      DBUG_RETURN(ev);
5941
    }
unknown's avatar
unknown committed
5942
    if (opt_reckless_slave)                     // For mysql-test
unknown's avatar
unknown committed
5943
      cur_log->error = 0;
unknown's avatar
unknown committed
5944
    if (cur_log->error < 0)
unknown's avatar
unknown committed
5945 5946
    {
      errmsg = "slave SQL thread aborted because of I/O error";
unknown's avatar
unknown committed
5947
      if (hot_log)
Marc Alff's avatar
Marc Alff committed
5948
        mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
5949 5950
      goto err;
    }
5951 5952
    if (!cur_log->error) /* EOF */
    {
unknown's avatar
unknown committed
5953
      /*
unknown's avatar
unknown committed
5954 5955 5956
        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
5957
      */
5958 5959
      if (hot_log)
      {
5960 5961 5962 5963 5964 5965 5966 5967 5968 5969
        /*
          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").
5970 5971 5972 5973 5974 5975 5976 5977 5978 5979

          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
          rli->last_master_timestamp is temporarely (for time of
          waiting for the following event) reset whenever EOF is
          reached.
5980
        */
5981
        time_t save_timestamp= rli->last_master_timestamp;
5982 5983
        rli->last_master_timestamp= 0;

unknown's avatar
unknown committed
5984
        DBUG_ASSERT(rli->relay_log.get_open_count() ==
unknown's avatar
unknown committed
5985
                    rli->cur_log_old_open_count);
5986 5987 5988 5989 5990

        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"));
5991
          ev= new Rotate_log_event(rli->ign_master_log_name_end,
5992
                                   0, rli->ign_master_log_pos_end,
unknown's avatar
unknown committed
5993
                                   Rotate_log_event::DUP_NAME);
5994
          rli->ign_master_log_name_end[0]= 0;
Marc Alff's avatar
Marc Alff committed
5995
          mysql_mutex_unlock(log_lock);
5996 5997 5998 5999 6000 6001 6002 6003 6004 6005
          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);
        }

unknown's avatar
unknown committed
6006 6007 6008 6009
        /*
          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
6010
        mysql_mutex_unlock(&rli->data_lock);
6011 6012

        /*
unknown's avatar
unknown committed
6013
          Possible deadlock :
6014 6015 6016 6017 6018 6019 6020 6021 6022 6023 6024
          - 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
6025
          master does not send anything). Then the I/O thread stops waiting
6026 6027 6028
          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
6029 6030
          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,
6031 6032 6033 6034
          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
6035 6036 6037
          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
6038
          it stops.
6039
        */
Marc Alff's avatar
Marc Alff committed
6040
        mysql_mutex_lock(&rli->log_space_lock);
6041 6042 6043 6044 6045 6046 6047 6048 6049 6050 6051 6052 6053 6054 6055 6056 6057 6058 6059 6060 6061 6062 6063 6064 6065

        /* 
          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;
        }

6066
        /*
6067 6068 6069 6070
          If the I/O thread is blocked, unblock it.  Ok to broadcast
          after unlock, because the mutex is only destroyed in
          ~Relay_log_info(), i.e. when rli is destroyed, and rli will
          not be destroyed before we exit the present function.
6071
        */
Marc Alff's avatar
Marc Alff committed
6072 6073
        mysql_mutex_unlock(&rli->log_space_lock);
        mysql_cond_broadcast(&rli->log_space_cond);
Andrei Elkin's avatar
Andrei Elkin committed
6074
        // Note that wait_for_update_relay_log unlocks lock_log !
Michael Widenius's avatar
Michael Widenius committed
6075
        rli->relay_log.wait_for_update_relay_log(rli->sql_driver_thd);
6076
        // re-acquire data lock since we released it earlier
Marc Alff's avatar
Marc Alff committed
6077
        mysql_mutex_lock(&rli->data_lock);
6078
        rli->last_master_timestamp= save_timestamp;
unknown's avatar
unknown committed
6079
        continue;
6080
      }
unknown's avatar
unknown committed
6081
      /*
unknown's avatar
unknown committed
6082 6083 6084
        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
6085 6086 6087
      */
      end_io_cache(cur_log);
      DBUG_ASSERT(rli->cur_log_fd >= 0);
Marc Alff's avatar
Marc Alff committed
6088
      mysql_file_close(rli->cur_log_fd, MYF(MY_WME));
unknown's avatar
unknown committed
6089
      rli->cur_log_fd = -1;
unknown's avatar
unknown committed
6090

6091
      if (relay_log_purge)
unknown's avatar
unknown committed
6092
      {
unknown's avatar
unknown committed
6093
        /*
6094 6095 6096 6097 6098 6099 6100 6101 6102 6103
          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
6104
        if (rli->relay_log.purge_first_log
6105 6106 6107
            (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
6108 6109 6110 6111
        {
          errmsg = "Error purging processed logs";
          goto err;
        }
unknown's avatar
unknown committed
6112
      }
6113 6114
      else
      {
unknown's avatar
unknown committed
6115 6116 6117 6118 6119 6120 6121 6122 6123 6124 6125 6126 6127
        /*
          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;
6128
        strmake_buf(rli->event_relay_log_name,rli->linfo.log_file_name);
unknown's avatar
unknown committed
6129
        flush_relay_log_info(rli);
unknown's avatar
unknown committed
6130
      }
6131 6132 6133 6134 6135 6136 6137 6138 6139 6140 6141 6142 6143

      /*
        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
6144
        mysql_mutex_lock(log_lock);
unknown's avatar
unknown committed
6145 6146
      if (rli->relay_log.is_active(rli->linfo.log_file_name))
      {
unknown's avatar
unknown committed
6147 6148 6149 6150 6151
        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);

        /*
6152 6153 6154 6155 6156 6157 6158 6159 6160 6161 6162 6163 6164 6165 6166 6167 6168 6169 6170 6171 6172 6173 6174 6175 6176 6177 6178 6179 6180 6181 6182 6183 6184 6185 6186 6187 6188 6189 6190 6191 6192 6193 6194 6195 6196 6197 6198 6199 6200 6201 6202 6203 6204 6205 6206 6207 6208
           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
6209
        */
6210 6211

        my_b_seek(cur_log, (my_off_t) 0);
unknown's avatar
unknown committed
6212
        if (check_binlog_magic(cur_log,&errmsg))
6213
        {
Marc Alff's avatar
Marc Alff committed
6214 6215
          if (!hot_log)
            mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
6216
          goto err;
6217
        }
Marc Alff's avatar
Marc Alff committed
6218 6219
        if (!hot_log)
          mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
6220
        continue;
6221
      }
Marc Alff's avatar
Marc Alff committed
6222 6223
      if (!hot_log)
        mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
6224
      /*
unknown's avatar
unknown committed
6225 6226 6227
        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
6228 6229 6230
      */
      // 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
6231 6232
                                       &errmsg)) <0)
        goto err;
6233
    }
unknown's avatar
unknown committed
6234
    else
6235
    {
unknown's avatar
unknown committed
6236
      /*
unknown's avatar
unknown committed
6237 6238
        Read failed with a non-EOF error.
        TODO: come up with something better to handle this error
unknown's avatar
unknown committed
6239 6240
      */
      if (hot_log)
Marc Alff's avatar
Marc Alff committed
6241
        mysql_mutex_unlock(log_lock);
6242
      sql_print_error("Slave SQL thread: I/O error reading \
unknown's avatar
unknown committed
6243
event(errno: %d  cur_log->error: %d)",
unknown's avatar
unknown committed
6244
                      my_errno,cur_log->error);
unknown's avatar
unknown committed
6245
      // set read position to the beginning of the event
6246
      my_b_seek(cur_log,rli->event_relay_log_pos);
unknown's avatar
unknown committed
6247 6248
      /* otherwise, we have had a partial read */
      errmsg = "Aborting slave SQL thread because of partial event read";
unknown's avatar
unknown committed
6249
      break;                                    // To end of function
6250 6251
    }
  }
6252
  if (!errmsg && global_system_variables.log_warnings)
6253
  {
unknown's avatar
unknown committed
6254
    sql_print_information("Error reading relay log event: %s",
6255 6256 6257
                          "slave SQL thread was killed");
    DBUG_RETURN(0);
  }
unknown's avatar
unknown committed
6258

6259
err:
6260 6261
  if (errmsg)
    sql_print_error("Error reading relay log event: %s", errmsg);
unknown's avatar
unknown committed
6262
  DBUG_RETURN(0);
6263 6264
}

6265 6266 6267
/*
  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
6268
  locks; here we don't, so this function is mainly taking locks).
6269 6270
  Returns nothing as we cannot catch any error (MYSQL_BIN_LOG::new_file()
  is void).
6271 6272
*/

6273
int rotate_relay_log(Master_info* mi)
6274 6275
{
  DBUG_ENTER("rotate_relay_log");
6276
  Relay_log_info* rli= &mi->rli;
6277
  int error= 0;
unknown's avatar
unknown committed
6278

Luis Soares's avatar
Luis Soares committed
6279
  DBUG_EXECUTE_IF("crash_before_rotate_relaylog", DBUG_SUICIDE(););
6280

unknown's avatar
unknown committed
6281
  /*
unknown's avatar
unknown committed
6282 6283 6284
     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).
  */
6285 6286
  if (!rli->inited)
  {
unknown's avatar
unknown committed
6287
    DBUG_PRINT("info", ("rli->inited == 0"));
unknown's avatar
unknown committed
6288
    goto end;
6289
  }
unknown's avatar
unknown committed
6290

6291
  /* If the relay log is closed, new_file() will do nothing. */
6292 6293
  if ((error= rli->relay_log.new_file()))
    goto end;
unknown's avatar
unknown committed
6294

6295 6296 6297 6298
  /*
    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
6299 6300 6301 6302 6303 6304
    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.
6305 6306
    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
6307 6308

    Note that it needs to be protected by mi->data_lock.
6309
  */
Sergei Golubchik's avatar
Sergei Golubchik committed
6310
  mysql_mutex_assert_owner(&mi->data_lock);
6311
  rli->relay_log.harvest_bytes_written(&rli->log_space_total);
unknown's avatar
unknown committed
6312
end:
6313
  DBUG_RETURN(error);
6314 6315
}

unknown's avatar
unknown committed
6316

unknown's avatar
unknown committed
6317 6318 6319
/**
   Detects, based on master's version (as found in the relay log), if master
   has a certain bug.
6320
   @param rli Relay_log_info which tells the master's version
unknown's avatar
unknown committed
6321
   @param bug_id Number of the bug as found in bugs.mysql.com
6322
   @param report bool report error message, default TRUE
6323 6324 6325 6326 6327 6328 6329

   @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
6330 6331
   @return TRUE if master has the bug, FALSE if it does not.
*/
6332 6333
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
6334 6335 6336 6337 6338 6339 6340 6341
{
  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
6342
    {24432, { 5, 0, 24 }, { 5, 0, 38 } },
6343 6344 6345
    {24432, { 5, 1, 12 }, { 5, 1, 17 } },
    {33029, { 5, 0,  0 }, { 5, 0, 58 } },
    {33029, { 5, 1,  0 }, { 5, 1, 12 } },
6346
    {37426, { 5, 1,  0 }, { 5, 1, 26 } },
unknown's avatar
unknown committed
6347 6348
  };
  const uchar *master_ver=
6349
    rli->relay_log.description_event_for_exec->server_version_split.ver;
unknown's avatar
unknown committed
6350

6351
  DBUG_ASSERT(sizeof(rli->relay_log.description_event_for_exec->server_version_split.ver) == 3);
unknown's avatar
unknown committed
6352 6353 6354 6355 6356 6357 6358 6359

  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) &&
6360 6361
        (memcmp(fixed_in,      master_ver, 3) >  0) &&
        (pred == NULL || (*pred)(param)))
unknown's avatar
unknown committed
6362
    {
6363 6364
      if (!report)
	return TRUE;
unknown's avatar
unknown committed
6365 6366 6367 6368 6369
      // 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);
6370
      // a verbose message for the error log
6371 6372 6373
      rli->report(ERROR_LEVEL, ER_UNKNOWN_ERROR,
                  "According to the master's version ('%s'),"
                  " it is probable that master suffers from this bug:"
6374 6375 6376 6377 6378 6379 6380 6381 6382 6383 6384 6385 6386 6387 6388 6389
                      " 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
6390 6391 6392 6393 6394 6395
      return TRUE;
    }
  }
  return FALSE;
}

6396 6397 6398 6399 6400 6401 6402 6403 6404 6405 6406 6407
/**
   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)
{
6408
  if (thd->rgi_slave)
6409
  {
6410
    DBUG_EXECUTE_IF("simulate_bug33029", return TRUE;);
6411
    return rpl_master_has_bug(thd->rgi_slave->rli, 33029, FALSE, NULL, NULL);
6412 6413 6414 6415
  }
  return FALSE;
}

6416 6417 6418 6419
/**
  @} (end of group Replication)
*/

unknown's avatar
SCRUM  
unknown committed
6420
#endif /* HAVE_REPLICATION */