slave.cc 200 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
#define FLAGSTR(V,F) ((V)&(F)?#F" ":"")
62

63
#define MAX_SLAVE_RETRY_PAUSE 5
64 65 66 67
/*
  a parameter of sql_slave_killed() to defer the killed status
*/
#define SLAVE_WAIT_GROUP_DONE 60
68 69
bool use_slave_mask = 0;
MY_BITMAP slave_error_mask;
70
char slave_skip_error_names[SHOW_VAR_FUNC_BUFF_SIZE];
71

72
char* slave_load_tmpdir = 0;
73
Master_info *active_mi= 0;
74
Master_info_index *master_info_index;
75
my_bool replicate_same_server_id;
76
ulonglong relay_log_space_limit = 0;
77
LEX_STRING default_master_connection_name= { (char*) "", 0 };
unknown's avatar
unknown committed
78 79 80 81 82

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

83
  TODO: move the vars below under Master_info
unknown's avatar
unknown committed
84
*/
85

86
int disconnect_slave_event_count = 0, abort_slave_event_count = 0;
unknown's avatar
unknown committed
87

He Zhenxing's avatar
He Zhenxing committed
88 89
static pthread_key(Master_info*, RPL_MASTER_INFO);

90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116
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, \
117
log '%s' at position %s",
118 119 120 121 122 123 124
    "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",
125
    "failed dump request, reconnecting to try again, log '%s' at position %s",
126 127 128 129 130 131 132 133
    "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, \
134
log '%s' at position %s",
135 136 137 138 139 140 141
    "",
    "Slave I/O thread killed during or after a reconnect done to recover from \
failed read"
  }
};
 

142
typedef enum { SLAVE_THD_IO, SLAVE_THD_SQL} SLAVE_THD_TYPE;
unknown's avatar
unknown committed
143

144 145
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);
146
static bool wait_for_relay_log_space(Relay_log_info* rli);
147
static inline bool io_slave_killed(THD* thd,Master_info* mi);
148
static inline bool sql_slave_killed(THD* thd,Relay_log_info* rli);
149 150
static int init_slave_thread(THD* thd, Master_info *mi,
                             SLAVE_THD_TYPE thd_type);
151
static void print_slave_skip_errors(void);
152 153
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
154
                          bool suppress_warnings);
155
static int connect_to_master(THD* thd, MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
156
                             bool reconnect, bool suppress_warnings);
157
static Log_event* next_event(Relay_log_info* rli);
158
static int queue_event(Master_info* mi,const char* buf,ulong event_len);
159
static int terminate_slave_thread(THD *thd,
Marc Alff's avatar
Marc Alff committed
160 161
                                  mysql_mutex_t *term_lock,
                                  mysql_cond_t *term_cond,
162 163
                                  volatile uint *slave_running,
                                  bool skip_lock);
164
static bool check_io_slave_killed(THD *thd, Master_info *mi, const char *info);
unknown's avatar
unknown committed
165 166 167 168
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);
169 170

/*
unknown's avatar
unknown committed
171
  Find out which replications threads are running
172

unknown's avatar
unknown committed
173 174
  SYNOPSIS
    init_thread_mask()
unknown's avatar
unknown committed
175 176 177
    mask                Return value here
    mi                  master_info for slave
    inverse             If set, returns which threads are not running
unknown's avatar
unknown committed
178 179 180 181 182 183

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

  RETURN
unknown's avatar
unknown committed
184 185
    mask        If inverse == 0, running threads
                If inverse == 1, stopped threads
186 187
*/

188
void init_thread_mask(int* mask,Master_info* mi,bool inverse)
189 190 191
{
  bool set_io = mi->slave_running, set_sql = mi->rli.slave_running;
  register int tmp_mask=0;
192 193
  DBUG_ENTER("init_thread_mask");

194 195 196 197
  if (set_io)
    tmp_mask |= SLAVE_IO;
  if (set_sql)
    tmp_mask |= SLAVE_SQL;
198 199
  if (inverse)
    tmp_mask^= (SLAVE_IO | SLAVE_SQL);
200
  *mask = tmp_mask;
201
  DBUG_VOID_RETURN;
202 203
}

204

unknown's avatar
unknown committed
205
/*
206
  lock_slave_threads()
unknown's avatar
unknown committed
207
*/
208

209
void lock_slave_threads(Master_info* mi)
210
{
211 212
  DBUG_ENTER("lock_slave_threads");

213
  //TODO: see if we can do this without dual mutex
Marc Alff's avatar
Marc Alff committed
214 215
  mysql_mutex_lock(&mi->run_lock);
  mysql_mutex_lock(&mi->rli.run_lock);
216
  DBUG_VOID_RETURN;
217 218
}

219

unknown's avatar
unknown committed
220
/*
221
  unlock_slave_threads()
unknown's avatar
unknown committed
222
*/
223

224
void unlock_slave_threads(Master_info* mi)
225
{
226 227
  DBUG_ENTER("unlock_slave_threads");

228
  //TODO: see if we can do this without dual mutex
Marc Alff's avatar
Marc Alff committed
229 230
  mysql_mutex_unlock(&mi->rli.run_lock);
  mysql_mutex_unlock(&mi->run_lock);
231
  DBUG_VOID_RETURN;
232 233
}

Marc Alff's avatar
Marc Alff committed
234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254
#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 */
255

unknown's avatar
unknown committed
256
/* Initialize slave structures */
257

258 259
int init_slave()
{
260
  DBUG_ENTER("init_slave");
261
  int error= 0;
262

Marc Alff's avatar
Marc Alff committed
263 264 265 266
#ifdef HAVE_PSI_INTERFACE
  init_slave_psi_keys();
#endif

267 268 269 270 271
  /*
    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
272
  mysql_mutex_lock(&LOCK_active_mi);
273

He Zhenxing's avatar
He Zhenxing committed
274 275
  if (pthread_key_create(&RPL_MASTER_INFO, NULL))
    goto err;
276

277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299
  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;
  }

300 301 302 303 304 305 306 307 308
  /*
    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();
  }

309
  /*
310 311 312
    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.
313
  */
314

unknown's avatar
unknown committed
315
  if (init_master_info(active_mi,master_info_file,relay_log_info_file,
316
                       1, (SLAVE_IO | SLAVE_SQL)))
317
  {
318
    sql_print_error("Failed to initialize the master info structure");
unknown's avatar
unknown committed
319
    goto err;
320
  }
321

322 323
  /* If server id is not set, start_slave_thread() will say it */

324
  if (active_mi->host[0] && !opt_skip_slave_start)
325
  {
326
    if (start_slave_threads(1 /* need mutex */,
unknown's avatar
unknown committed
327 328 329 330 331
                            0 /* no wait for start*/,
                            active_mi,
                            master_info_file,
                            relay_log_info_file,
                            SLAVE_IO | SLAVE_SQL))
unknown's avatar
unknown committed
332
    {
333
      sql_print_error("Failed to create slave threads");
unknown's avatar
unknown committed
334 335
      goto err;
    }
336 337
  }

338
end:
Marc Alff's avatar
Marc Alff committed
339
  mysql_mutex_unlock(&LOCK_active_mi);
340
  DBUG_RETURN(error);
341 342 343 344 345

err:
  sql_print_error("Failed to allocate memory for the Master Info structure");
  error= 1;
  goto end;
346
}
347

348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374
/*
  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).
 */
375
int init_recovery(Master_info* mi, const char** errmsg)
376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396
{
  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);
    strmake(mi->master_log_name, rli->group_master_log_name,
            sizeof(mi->master_log_name)-1);
 
    sql_print_warning("Recovery from master pos %ld and file %s.",
                      (ulong) mi->master_log_pos, mi->master_log_name);
 
    strmake(rli->group_relay_log_name, rli->relay_log.get_log_fname(),
            sizeof(rli->group_relay_log_name)-1);
    strmake(rli->event_relay_log_name, rli->relay_log.get_log_fname(),
            sizeof(mi->rli.event_relay_log_name)-1);
 
    rli->group_relay_log_pos= rli->event_relay_log_pos= BIN_LOG_HEADER_SIZE;
  }
397

398 399
  DBUG_RETURN(0);
}
unknown's avatar
unknown committed
400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426


/*
  When connecting a slave to a master with GTID, we reset the relay log
  coordinates of the SQL thread and clear the master coordinates of SQL and IO
  threads.

  This way we ensure that we start from the correct place even after a change
  to new master or a crash where relay log coordinates may be wrong (GTID
  state is crash safe but master.info is not). And we get the correct master
  coordinates set upon reading the initial fake rotate event sent from master.
*/
static void
reset_coordinates_for_gtid(Master_info *mi, Relay_log_info *rli)
{
  mi->master_log_pos= 0;
  mi->master_log_name[0]= 0;
  rli->group_master_log_pos= 0;
  rli->group_master_log_name[0]= 0;
  rli->group_relay_log_pos= BIN_LOG_HEADER_SIZE;
  strmake(rli->group_relay_log_name, rli->relay_log.get_log_fname(),
          sizeof(rli->group_relay_log_name)-1);
  rli->event_relay_log_pos= BIN_LOG_HEADER_SIZE;
  strmake(rli->event_relay_log_name, rli->relay_log.get_log_fname(),
          sizeof(mi->rli.event_relay_log_name)-1);
}

427
 
428 429 430 431 432 433 434 435 436 437 438 439
/**
  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.
  */
440
  const size_t MIN_ROOM= 10;
441 442 443 444
  DBUG_ENTER("print_slave_skip_errors");
  DBUG_ASSERT(sizeof(slave_skip_error_names) > MIN_ROOM);
  DBUG_ASSERT(MAX_SLAVE_ERROR <= 999999); // 6 digits

445 446 447
  /* Make @@slave_skip_errors show the nice human-readable value.  */
  opt_slave_skip_errors= slave_skip_error_names;

448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465
  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;

466
    for (errnum= 0; errnum < MAX_SLAVE_ERROR; errnum++)
467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488
    {
      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
489
/*
490
  Init function to set up array for errors that should be skipped for slave
unknown's avatar
unknown committed
491

unknown's avatar
unknown committed
492 493
  SYNOPSIS
    init_slave_skip_errors()
unknown's avatar
unknown committed
494
    arg         List of errors numbers to skip, separated with ','
unknown's avatar
unknown committed
495 496 497 498

  NOTES
    Called from get_options() in mysqld.cc on start-up
*/
499

unknown's avatar
unknown committed
500
void init_slave_skip_errors(const char* arg)
501
{
unknown's avatar
unknown committed
502
  const char *p;
503 504
  DBUG_ENTER("init_slave_skip_errors");

505
  if (bitmap_init(&slave_error_mask,0,MAX_SLAVE_ERROR,0))
506 507 508 509 510
  {
    fprintf(stderr, "Badly out of memory, please check your system status\n");
    exit(1);
  }
  use_slave_mask = 1;
511
  for (;my_isspace(system_charset_info,*arg);++arg)
512
    /* empty */;
513
  if (!my_strnncoll(system_charset_info,(uchar*)arg,4,(const uchar*)"all",4))
514 515
  {
    bitmap_set_all(&slave_error_mask);
516
    print_slave_skip_errors();
517
    DBUG_VOID_RETURN;
518 519 520 521 522 523 524 525
  }
  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);
526
    while (!my_isdigit(system_charset_info,*p) && *p)
527 528
      p++;
  }
529 530
  /* Convert slave skip errors bitmap into a printable string. */
  print_slave_skip_errors();
531
  DBUG_VOID_RETURN;
532 533
}

534
static void set_thd_in_use_temporary_tables(Relay_log_info *rli)
535 536 537 538 539 540
{
  TABLE *table;

  for (table= rli->save_temporary_tables ; table ; table= table->next)
    table->in_use= rli->sql_thd;
}
541

542
int terminate_slave_threads(Master_info* mi,int thread_mask,bool skip_lock)
543
{
544 545
  DBUG_ENTER("terminate_slave_threads");

546
  if (!mi->inited)
547
    DBUG_RETURN(0); /* successfully do nothing */
548
  int error,force_all = (thread_mask & SLAVE_FORCE_ALL);
Marc Alff's avatar
Marc Alff committed
549
  mysql_mutex_t *sql_lock = &mi->rli.run_lock, *io_lock = &mi->run_lock;
550
  mysql_mutex_t *log_lock= mi->rli.relay_log.get_log_lock();
unknown's avatar
unknown committed
551

552
  if (thread_mask & (SLAVE_SQL|SLAVE_FORCE_ALL))
553
  {
unknown's avatar
unknown committed
554
    DBUG_PRINT("info",("Terminating SQL thread"));
555
    mi->rli.abort_slave=1;
556
    if ((error=terminate_slave_thread(mi->rli.sql_thd, sql_lock,
unknown's avatar
unknown committed
557
                                      &mi->rli.stop_cond,
558 559
                                      &mi->rli.slave_running,
                                      skip_lock)) &&
unknown's avatar
unknown committed
560
        !force_all)
561
      DBUG_RETURN(error);
562

563
    mysql_mutex_lock(log_lock);
564

Libing Song's avatar
Libing Song committed
565
    DBUG_PRINT("info",("Flushing relay-log info file."));
566
    if (current_thd)
Libing Song's avatar
Libing Song committed
567 568
      thd_proc_info(current_thd, "Flushing relay-log info file.");
    if (flush_relay_log_info(&mi->rli))
569
      DBUG_RETURN(ER_ERROR_DURING_FLUSH_LOGS);
Libing Song's avatar
Libing Song committed
570 571
    
    if (my_sync(mi->rli.info_fd, MYF(MY_WME)))
572 573
      DBUG_RETURN(ER_ERROR_DURING_FLUSH_LOGS);

574
    mysql_mutex_unlock(log_lock);
575
  }
576
  if (thread_mask & (SLAVE_IO|SLAVE_FORCE_ALL))
577
  {
unknown's avatar
unknown committed
578
    DBUG_PRINT("info",("Terminating IO thread"));
579
    mi->abort_slave=1;
580
    if ((error=terminate_slave_thread(mi->io_thd, io_lock,
unknown's avatar
unknown committed
581
                                      &mi->stop_cond,
582 583
                                      &mi->slave_running,
                                      skip_lock)) &&
unknown's avatar
unknown committed
584
        !force_all)
585
      DBUG_RETURN(error);
586

587
    mysql_mutex_lock(log_lock);
588 589 590 591

    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.");
592
    if (flush_master_info(mi, TRUE, FALSE))
593 594
      DBUG_RETURN(ER_ERROR_DURING_FLUSH_LOGS);

Luis Soares's avatar
Luis Soares committed
595 596
    if (mi->rli.relay_log.is_open() &&
        my_sync(mi->rli.relay_log.get_log_file()->file, MYF(MY_WME)))
597 598 599 600 601
      DBUG_RETURN(ER_ERROR_DURING_FLUSH_LOGS);

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

602
    mysql_mutex_unlock(log_lock);
603
  }
604
  DBUG_RETURN(0); 
605 606
}

607

608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634
/**
   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
635
   @retval 0 All OK ER_SLAVE_NOT_RUNNING otherwise.
636

Andrei Elkin's avatar
Andrei Elkin committed
637 638 639
   @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.
640

641 642 643
 */
static int
terminate_slave_thread(THD *thd,
Marc Alff's avatar
Marc Alff committed
644 645
                       mysql_mutex_t *term_lock,
                       mysql_cond_t *term_cond,
646 647
                       volatile uint *slave_running,
                       bool skip_lock)
648
{
649
  DBUG_ENTER("terminate_slave_thread");
650
  if (!skip_lock)
651
  {
Marc Alff's avatar
Marc Alff committed
652
    mysql_mutex_lock(term_lock);
653 654 655
  }
  else
  {
Marc Alff's avatar
Marc Alff committed
656
    mysql_mutex_assert_owner(term_lock);
657
  }
658 659 660
  if (!*slave_running)
  {
    if (!skip_lock)
661
    {
662 663 664 665
      /*
        if run_lock (term_lock) is acquired locally then either
        slave_running status is fine
      */
Marc Alff's avatar
Marc Alff committed
666
      mysql_mutex_unlock(term_lock);
667 668 669 670
      DBUG_RETURN(0);
    }
    else
    {
671
      DBUG_RETURN(ER_SLAVE_NOT_RUNNING);
672 673 674
    }
  }
  DBUG_ASSERT(thd != 0);
675
  THD_CHECK_SENTRY(thd);
676

unknown's avatar
unknown committed
677
  /*
678
    Is is critical to test if the slave is running. Otherwise, we might
unknown's avatar
unknown committed
679
    be referening freed memory trying to kick it
680
  */
unknown's avatar
unknown committed
681

unknown's avatar
unknown committed
682
  while (*slave_running)                        // Should always be true
683
  {
Michael Widenius's avatar
Michael Widenius committed
684
    int error __attribute__((unused));
685
    DBUG_PRINT("loop", ("killing slave thread"));
686

Marc Alff's avatar
Marc Alff committed
687
    mysql_mutex_lock(&thd->LOCK_thd_data);
688 689 690 691 692 693
#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)
    */
694
    int err __attribute__((unused))= pthread_kill(thd->real_id, thr_client_alarm);
695 696
    DBUG_ASSERT(err != EINVAL);
#endif
Sergei Golubchik's avatar
Sergei Golubchik committed
697 698
    thd->awake(NOT_KILLED);

Marc Alff's avatar
Marc Alff committed
699
    mysql_mutex_unlock(&thd->LOCK_thd_data);
700

unknown's avatar
unknown committed
701 702 703
    /*
      There is a small chance that slave thread might miss the first
      alarm. To protect againts it, resend the signal until it reacts
704 705
    */
    struct timespec abstime;
unknown's avatar
unknown committed
706
    set_timespec(abstime,2);
Marc Alff's avatar
Marc Alff committed
707
    error= mysql_cond_timedwait(term_cond, term_lock, &abstime);
708
    DBUG_ASSERT(error == ETIMEDOUT || error == 0);
709
  }
710 711 712 713

  DBUG_ASSERT(*slave_running == 0);

  if (!skip_lock)
Marc Alff's avatar
Marc Alff committed
714
    mysql_mutex_unlock(term_lock);
715
  DBUG_RETURN(0);
716 717
}

718

Marc Alff's avatar
Marc Alff committed
719 720 721 722 723 724 725
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
726 727
                       volatile uint *slave_running,
                       volatile ulong *slave_run_id,
Konstantin Osipov's avatar
Konstantin Osipov committed
728
                       Master_info* mi)
729 730
{
  pthread_t th;
unknown's avatar
unknown committed
731 732 733
  ulong start_id;
  DBUG_ENTER("start_slave_thread");

734 735
  DBUG_ASSERT(mi->inited);

736
  if (start_lock)
Marc Alff's avatar
Marc Alff committed
737
    mysql_mutex_lock(start_lock);
738
  if (!global_system_variables.server_id)
739 740
  {
    if (start_cond)
Marc Alff's avatar
Marc Alff committed
741
      mysql_cond_broadcast(start_cond);
742
    if (start_lock)
Marc Alff's avatar
Marc Alff committed
743
      mysql_mutex_unlock(start_lock);
744
    sql_print_error("Server id not set, will not start slave");
unknown's avatar
unknown committed
745
    DBUG_RETURN(ER_BAD_SLAVE);
746
  }
unknown's avatar
unknown committed
747

748
  if (*slave_running)
749 750
  {
    if (start_cond)
Marc Alff's avatar
Marc Alff committed
751
      mysql_cond_broadcast(start_cond);
752
    if (start_lock)
Marc Alff's avatar
Marc Alff committed
753
      mysql_mutex_unlock(start_lock);
unknown's avatar
unknown committed
754
    DBUG_RETURN(ER_SLAVE_MUST_STOP);
755
  }
unknown's avatar
unknown committed
756 757
  start_id= *slave_run_id;
  DBUG_PRINT("info",("Creating new slave thread"));
Marc Alff's avatar
Marc Alff committed
758 759
  if (mysql_thread_create(thread_key,
                          &th, &connection_attrib, h_func, (void*)mi))
760 761
  {
    if (start_lock)
Marc Alff's avatar
Marc Alff committed
762
      mysql_mutex_unlock(start_lock);
unknown's avatar
unknown committed
763
    DBUG_RETURN(ER_SLAVE_THREAD);
764
  }
unknown's avatar
unknown committed
765
  if (start_cond && cond_lock) // caller has cond_lock
766 767
  {
    THD* thd = current_thd;
unknown's avatar
unknown committed
768
    while (start_id == *slave_run_id)
769
    {
unknown's avatar
unknown committed
770
      DBUG_PRINT("sleep",("Waiting for slave thread to start"));
771 772 773 774 775 776 777 778 779 780 781
      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);
782
      thd->exit_cond(old_msg);
Marc Alff's avatar
Marc Alff committed
783
      mysql_mutex_lock(cond_lock); // re-acquire it as exit_cond() released
784
      if (thd->killed)
785 786
      {
        if (start_lock)
Marc Alff's avatar
Marc Alff committed
787
          mysql_mutex_unlock(start_lock);
unknown's avatar
unknown committed
788
        DBUG_RETURN(thd->killed_errno());
789
      }
790 791 792
    }
  }
  if (start_lock)
Marc Alff's avatar
Marc Alff committed
793
    mysql_mutex_unlock(start_lock);
unknown's avatar
unknown committed
794
  DBUG_RETURN(0);
795
}
unknown's avatar
unknown committed
796

797

unknown's avatar
unknown committed
798
/*
799
  start_slave_threads()
unknown's avatar
unknown committed
800

unknown's avatar
unknown committed
801 802 803 804
  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
805
*/
unknown's avatar
unknown committed
806

807
int start_slave_threads(bool need_slave_mutex, bool wait_for_start,
808
                        Master_info* mi, const char* master_info_fname,
unknown's avatar
unknown committed
809
                        const char* slave_info_fname, int thread_mask)
810
{
Marc Alff's avatar
Marc Alff committed
811 812
  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;
813
  int error=0;
814
  DBUG_ENTER("start_slave_threads");
unknown's avatar
unknown committed
815

816 817 818 819 820 821 822 823 824 825 826 827
  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;
  }
828 829

  if (thread_mask & SLAVE_IO)
Marc Alff's avatar
Marc Alff committed
830 831 832 833 834 835 836 837
    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);
838
  if (!error && (thread_mask & SLAVE_SQL))
839
  {
Marc Alff's avatar
Marc Alff committed
840 841 842 843 844 845 846 847
    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);
848
    if (error)
849
      terminate_slave_threads(mi, thread_mask & SLAVE_IO, !need_slave_mutex);
850
  }
851
  DBUG_RETURN(error);
852
}
853

854

unknown's avatar
unknown committed
855
/*
856
  Release slave threads at time of executing shutdown.
unknown's avatar
unknown committed
857 858 859 860

  SYNOPSIS
    end_slave()
*/
861

862 863
void end_slave()
{
864 865
  DBUG_ENTER("end_slave");

866 867 868 869 870 871
  /*
    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.
872 873 874

    We can also be called by cleanup(), which only happens if some
    startup parameter to the server was wrong.
875
  */
Marc Alff's avatar
Marc Alff committed
876
  mysql_mutex_lock(&LOCK_active_mi);
877 878 879 880
  /* 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
881
  mysql_mutex_unlock(&LOCK_active_mi);
882 883 884
  DBUG_VOID_RETURN;
}

885
static bool io_slave_killed(THD* thd, Master_info* mi)
unknown's avatar
unknown committed
886
{
887 888
  DBUG_ENTER("io_slave_killed");

889
  DBUG_ASSERT(mi->io_thd == thd);
890
  DBUG_ASSERT(mi->slave_running); // tracking buffer overrun
891
  DBUG_RETURN(mi->abort_slave || abort_loop || thd->killed);
unknown's avatar
unknown committed
892 893
}

894 895 896 897 898 899
/**
   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.
900

901 902
   @param thd   pointer to a THD instance
   @param rli   pointer to Relay_log_info instance
903

904 905 906
   @return TRUE the killed status is recognized, FALSE a possible killed
           status is deferred.
*/
907
static bool sql_slave_killed(THD* thd, Relay_log_info* rli)
908
{
909
  bool ret= FALSE;
910 911
  DBUG_ENTER("sql_slave_killed");

912 913
  DBUG_ASSERT(rli->sql_thd == thd);
  DBUG_ASSERT(rli->slave_running == 1);// tracking buffer overrun
914 915
  if (abort_loop || thd->killed || rli->abort_slave)
  {
916 917 918 919 920 921 922 923
    /*
      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.
    */
unknown's avatar
unknown committed
924 925 926
    if ((thd->transaction.all.modified_non_trans_table ||
         (thd->variables.option_bits & OPTION_KEEP_LOG))
        && rli->is_in_group())
927
    {
928
      char msg_stopped[]=
929 930 931 932
        "... 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 "
933 934 935 936 937
        "ignores duplicate key, key not found, and similar errors (see "
        "documentation for details).";

      if (rli->abort_slave)
      {
938 939 940
        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 ... "));
941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963

        /*
          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.
        */

        if (rli->last_event_start_time == 0)
          rli->last_event_start_time= my_time(0);
        ret= difftime(my_time(0), rli->last_event_start_time) <=
          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,
964 965 966
                      "Request to stop slave SQL Thread received while "
                      "applying a group that has non-transactional "
                      "changes; waiting for completion of the group ... ");
967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983
        }
        else
        {
          rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
                      ER(ER_SLAVE_FATAL_ERROR), msg_stopped);
        }
      }
      else
      {
        ret= TRUE;
        rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, ER(ER_SLAVE_FATAL_ERROR),
                    msg_stopped);
      }
    }
    else
    {
      ret= TRUE;
984 985
    }
  }
986 987 988 989
  if (ret)
    rli->last_event_start_time= 0;
  
  DBUG_RETURN(ret);
990 991
}

992

unknown's avatar
unknown committed
993
/*
994 995
  skip_load_data_infile()

unknown's avatar
unknown committed
996 997 998
  NOTES
    This is used to tell a 3.23 master to break send_file()
*/
999 1000 1001

void skip_load_data_infile(NET *net)
{
1002 1003
  DBUG_ENTER("skip_load_data_infile");

1004
  (void)net_request_file(net, "/dev/null");
unknown's avatar
unknown committed
1005
  (void)my_net_read(net);                               // discard response
1006
  (void)net_write_command(net, 0, (uchar*) "", 0, (uchar*) "", 0); // ok
1007
  DBUG_VOID_RETURN;
1008 1009
}

1010

1011
bool net_request_file(NET* net, const char* fname)
1012
{
1013
  DBUG_ENTER("net_request_file");
1014 1015
  DBUG_RETURN(net_write_command(net, 251, (uchar*) fname, strlen(fname),
                                (uchar*) "", 0));
1016 1017
}

1018 1019
/*
  From other comments and tests in code, it looks like
1020
  sometimes Query_log_event and Load_log_event can have db == 0
1021 1022 1023
  (see rewrite_db() above for example)
  (cases where this happens are unclear; it may be when the master is 3.23).
*/
1024 1025

const char *print_slave_db_safe(const char* db)
1026
{
1027 1028 1029
  DBUG_ENTER("*print_slave_db_safe");

  DBUG_RETURN((db ? db : ""));
1030
}
1031

Sergei Golubchik's avatar
Sergei Golubchik committed
1032 1033
#endif /* HAVE_REPLICATION */

1034
int init_strvar_from_file(char *var, int max_size, IO_CACHE *f,
unknown's avatar
unknown committed
1035
                                 const char *default_val)
unknown's avatar
unknown committed
1036
{
unknown's avatar
unknown committed
1037
  uint length;
1038 1039
  DBUG_ENTER("init_strvar_from_file");

unknown's avatar
unknown committed
1040 1041 1042 1043 1044 1045
  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
1046
    {
unknown's avatar
unknown committed
1047
      /*
unknown's avatar
unknown committed
1048 1049
        If we truncated a line or stopped on last char, remove all chars
        up to and including newline.
unknown's avatar
unknown committed
1050
      */
unknown's avatar
unknown committed
1051
      int c;
1052
      while (((c=my_b_get(f)) != '\n' && c != my_b_EOF)) ;
unknown's avatar
unknown committed
1053
    }
1054
    DBUG_RETURN(0);
unknown's avatar
unknown committed
1055 1056 1057
  }
  else if (default_val)
  {
unknown's avatar
unknown committed
1058
    strmake(var,  default_val, max_size-1);
1059
    DBUG_RETURN(0);
unknown's avatar
unknown committed
1060
  }
1061
  DBUG_RETURN(1);
unknown's avatar
unknown committed
1062 1063
}

1064

Sergei Golubchik's avatar
Sergei Golubchik committed
1065 1066 1067 1068
/*
  when moving these functions to mysys, don't forget to
  remove slave.cc from libmysqld/CMakeLists.txt
*/
1069
int init_intvar_from_file(int* var, IO_CACHE* f, int default_val)
unknown's avatar
unknown committed
1070 1071
{
  char buf[32];
1072 1073
  DBUG_ENTER("init_intvar_from_file");

unknown's avatar
unknown committed
1074 1075

  if (my_b_gets(f, buf, sizeof(buf)))
unknown's avatar
unknown committed
1076 1077
  {
    *var = atoi(buf);
1078
    DBUG_RETURN(0);
unknown's avatar
unknown committed
1079
  }
unknown's avatar
unknown committed
1080
  else if (default_val)
unknown's avatar
unknown committed
1081 1082
  {
    *var = default_val;
1083
    DBUG_RETURN(0);
unknown's avatar
unknown committed
1084
  }
1085
  DBUG_RETURN(1);
unknown's avatar
unknown committed
1086 1087
}

Andrei Elkin's avatar
Andrei Elkin committed
1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108
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
1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160

/**
   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 ||
1161
        ((snd_size + 1 == max_size - read_size) &&  buf_act[max_size - 2] != '\n'))
Andrei Elkin's avatar
Andrei Elkin committed
1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192
    {
      /*
        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)
1193
    my_free(buf_act);
Andrei Elkin's avatar
Andrei Elkin committed
1194 1195 1196
  DBUG_RETURN(ret);
}

Sergei Golubchik's avatar
Sergei Golubchik committed
1197 1198
#ifdef HAVE_REPLICATION

1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213
/*
  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 ||
1214
      errorno == ER_CONNECTION_KILLED ||
1215
      errorno == ER_NEW_ABORTING_CONNECTION ||
1216 1217 1218 1219 1220 1221 1222
      errorno == ER_SERVER_SHUTDOWN)
    return TRUE;

  return FALSE;   
}


1223 1224 1225 1226 1227 1228 1229 1230
/*
  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
1231

1232 1233 1234
  RETURNS
  0       ok
  1       error
1235
  2       transient network problem, the caller should try to reconnect
1236
*/
1237

1238
static int get_master_version_and_clock(MYSQL* mysql, Master_info* mi)
1239
{
1240
  char err_buff[MAX_SLAVE_ERRMSG], err_buff2[MAX_SLAVE_ERRMSG];
1241
  const char* errmsg= 0;
1242 1243 1244
  int err_code= 0;
  MYSQL_RES *master_res= 0;
  MYSQL_ROW master_row;
1245
  uint version= mysql_get_server_version(mysql) / 10000;
1246
  DBUG_ENTER("get_master_version_and_clock");
1247 1248 1249 1250 1251 1252 1253

  /*
    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
1254

1255
  if (!my_isdigit(&my_charset_bin,*mysql->server_version))
1256
  {
1257 1258 1259 1260
    errmsg= err_buff2;
    snprintf(err_buff2, sizeof(err_buff2),
             "Master reported unrecognized MySQL version: %s",
             mysql->server_version);
1261
    err_code= ER_SLAVE_FATAL_ERROR;
1262
    sprintf(err_buff, ER(err_code), err_buff2);
1263
  }
1264 1265 1266 1267 1268
  else
  {
    /*
      Note the following switch will bug when we have MySQL branch 30 ;)
    */
1269 1270 1271 1272 1273 1274 1275 1276
    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);
1277
      err_code= ER_SLAVE_FATAL_ERROR;
1278
      sprintf(err_buff, ER(err_code), err_buff2);
1279
      break;
1280
    case 3:
1281
      mi->rli.relay_log.description_event_for_queue= new
unknown's avatar
unknown committed
1282
        Format_description_log_event(1, mysql->server_version);
1283
      break;
1284
    case 4:
1285
      mi->rli.relay_log.description_event_for_queue= new
unknown's avatar
unknown committed
1286
        Format_description_log_event(3, mysql->server_version);
1287
      break;
unknown's avatar
unknown committed
1288
    default:
1289 1290 1291 1292 1293 1294 1295 1296 1297
      /*
        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
1298
        Format_description_log_event(4, mysql->server_version);
1299 1300
      break;
    }
1301
  }
unknown's avatar
unknown committed
1302 1303

  /*
1304 1305 1306 1307 1308
     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
1309

1310
  if (errmsg)
1311
    goto err;
1312 1313 1314 1315

  /* as we are here, we tried to allocate the event */
  if (!mi->rli.relay_log.description_event_for_queue)
  {
1316 1317 1318 1319
    errmsg= "default Format_description_log_event";
    err_code= ER_SLAVE_CREATE_EVENT_FAILURE;
    sprintf(err_buff, ER(err_code), errmsg);
    goto err;
1320 1321
  }

1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363
  /*
    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); 
1364 1365 1366 1367
  /*
    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
1368

1369 1370 1371 1372 1373
  DBUG_EXECUTE_IF("dbug.before_get_UNIX_TIMESTAMP",
                  {
                    const char act[]=
                      "now "
                      "wait_for signal.get_unix_timestamp";
1374
                    DBUG_ASSERT(debug_sync_service);
1375 1376 1377 1378
                    DBUG_ASSERT(!debug_sync_set_action(current_thd,
                                                       STRING_WITH_LEN(act)));
                  };);

1379
  master_res= NULL;
1380
  if (!mysql_real_query(mysql, STRING_WITH_LEN("SELECT UNIX_TIMESTAMP()")) &&
1381 1382
      (master_res= mysql_store_result(mysql)) &&
      (master_row= mysql_fetch_row(master_res)))
unknown's avatar
unknown committed
1383
  {
unknown's avatar
unknown committed
1384
    mi->clock_diff_with_master=
1385
      (long) (time((time_t*) 0) - strtoul(master_row[0], 0, 10));
unknown's avatar
unknown committed
1386
  }
1387 1388
  else if (check_io_slave_killed(mi->io_thd, mi, NULL))
    goto slave_killed_err;
1389 1390 1391 1392 1393 1394 1395
  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
1396
  {
1397
    mi->clock_diff_with_master= 0; /* The "most sensible" value */
1398 1399 1400 1401
    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));
1402 1403
  }
  if (master_res)
1404
  {
unknown's avatar
unknown committed
1405
    mysql_free_result(master_res);
1406 1407
    master_res= NULL;
  }
unknown's avatar
unknown committed
1408

1409 1410 1411 1412 1413 1414 1415 1416 1417 1418
  /*
    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.
  */
1419 1420 1421 1422 1423
  DBUG_EXECUTE_IF("dbug.before_get_SERVER_ID",
                  {
                    const char act[]=
                      "now "
                      "wait_for signal.get_server_id";
1424
                    DBUG_ASSERT(debug_sync_service);
1425 1426 1427
                    DBUG_ASSERT(!debug_sync_set_action(current_thd, 
                                                       STRING_WITH_LEN(act)));
                  };);
1428 1429
  master_res= NULL;
  master_row= NULL;
1430 1431
  if (!mysql_real_query(mysql,
                        STRING_WITH_LEN("SHOW VARIABLES LIKE 'SERVER_ID'")) &&
1432 1433
      (master_res= mysql_store_result(mysql)) &&
      (master_row= mysql_fetch_row(master_res)))
1434
  {
1435 1436
    if ((global_system_variables.server_id ==
             (mi->master_id= strtoul(master_row[1], 0, 10))) &&
1437
        !mi->rli.replicate_same_server_id)
1438
    {
1439 1440 1441 1442
      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).";
1443 1444
      err_code= ER_SLAVE_FATAL_ERROR;
      sprintf(err_buff, ER(err_code), errmsg);
1445
      goto err;
1446
    }
1447 1448 1449
  }
  else if (mysql_errno(mysql))
  {
1450 1451 1452
    if (check_io_slave_killed(mi->io_thd, mi, NULL))
      goto slave_killed_err;
    else if (is_network_error(mysql_errno(mysql)))
1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472
    {
      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)
  {
1473
    mysql_free_result(master_res);
1474
    master_res= NULL;
unknown's avatar
unknown committed
1475
  }
Andrei Elkin's avatar
Andrei Elkin committed
1476 1477 1478 1479 1480 1481 1482
  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
1483

1484 1485 1486
  /*
    Check that the master's global character_set_server and ours are the same.
    Not fatal if query fails (old master?).
1487 1488 1489 1490 1491 1492
    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.
1493 1494 1495 1496 1497 1498
    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.
1499
  */
1500 1501

  /* redundant with rest of code but safer against later additions */
1502
  if (version == 3)
1503
    goto err;
1504

1505
  if (version == 4)
unknown's avatar
unknown committed
1506
  {
1507 1508 1509 1510 1511
    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)))
1512
    {
1513 1514 1515
      if (strcmp(master_row[0], global_system_variables.collation_server->name))
      {
        errmsg= "The slave I/O thread stops because master and slave have \
1516
different values for the COLLATION_SERVER global variable. The values must \
1517 1518 1519 1520 1521
be equal for the Statement-format replication to work";
        err_code= ER_SLAVE_FATAL_ERROR;
        sprintf(err_buff, ER(err_code), errmsg);
        goto err;
      }
1522
    }
1523 1524
    else if (check_io_slave_killed(mi->io_thd, mi, NULL))
      goto slave_killed_err;
1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537
    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));
1538
      goto err;
1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550
    }
    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
1551
  }
1552

1553 1554 1555 1556 1557 1558 1559 1560
  /*
    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
1561 1562 1563
    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
1564
    'SYSTEM' time zone).
1565 1566
    This check is only necessary for 4.x masters (and < 5.0.4 masters but
    those were alpha).
1567
  */
1568
  if (version == 4)
unknown's avatar
unknown committed
1569
  {
1570 1571 1572 1573
    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)))
1574
    {
1575 1576 1577 1578
      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 \
1579
different values for the TIME_ZONE global variable. The values must \
1580 1581 1582 1583 1584
be equal for the Statement-format replication to work";
        err_code= ER_SLAVE_FATAL_ERROR;
        sprintf(err_buff, ER(err_code), errmsg);
        goto err;
      }
1585
    }
1586 1587
    else if (check_io_slave_killed(mi->io_thd, mi, NULL))
      goto slave_killed_err;
1588
    else if (is_network_error(err_code= mysql_errno(mysql)))
1589
    {
1590 1591 1592
      mi->report(ERROR_LEVEL, err_code,
                 "Get master TIME_ZONE failed with error: %s",
                 mysql_error(mysql));
1593
      goto network_err;
1594 1595 1596 1597 1598 1599 1600 1601 1602 1603
    }
    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");
    }
1604 1605 1606 1607 1608 1609
    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));
1610
      goto err;
1611 1612 1613 1614 1615 1616
    }
    if (master_res)
    {
      mysql_free_result(master_res);
      master_res= NULL;
    }
unknown's avatar
unknown committed
1617 1618
  }

Andrei Elkin's avatar
Andrei Elkin committed
1619 1620 1621 1622 1623 1624 1625 1626 1627
  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);
1628
    sprintf(query, query_format, llbuf);
Andrei Elkin's avatar
Andrei Elkin committed
1629 1630 1631 1632

    if (mysql_real_query(mysql, query, strlen(query))
        && !check_io_slave_killed(mi->io_thd, mi, NULL))
    {
Andrei Elkin's avatar
Andrei Elkin committed
1633 1634
      errmsg= "The slave I/O thread stops because SET @master_heartbeat_period "
        "on master failed.";
Andrei Elkin's avatar
Andrei Elkin committed
1635
      err_code= ER_SLAVE_FATAL_ERROR;
Andrei Elkin's avatar
Andrei Elkin committed
1636
      sprintf(err_buff, "%s Error: %s", errmsg, mysql_error(mysql));
Andrei Elkin's avatar
Andrei Elkin committed
1637 1638 1639 1640 1641 1642
      mysql_free_result(mysql_store_result(mysql));
      goto err;
    }
    mysql_free_result(mysql_store_result(mysql));
  }
 
1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739
  /*
    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)
    {
      if (check_io_slave_killed(mi->io_thd, mi, NULL))
        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);
      }
      else if (check_io_slave_killed(mi->io_thd, mi, NULL))
        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
1740

1741
  /*
1742
    Request the master to filter away events with the @@skip_replication flag
1743 1744
    set, if we are running with
    --replicate-events-marked-for-skip=FILTER_ON_MASTER.
1745
  */
1746
  if (opt_replicate_events_marked_for_skip == RPL_SKIP_FILTER_ON_MASTER)
1747
  {
1748
    if (mysql_real_query(mysql, STRING_WITH_LEN("SET skip_replication=1")))
1749 1750 1751 1752 1753
    {
      err_code= mysql_errno(mysql);
      if (is_network_error(err_code))
      {
        mi->report(ERROR_LEVEL, err_code,
1754
                   "Setting master-side filtering of @@skip_replication failed "
1755 1756 1757 1758 1759 1760 1761
                   "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
1762
          @@skip_replication feature.
1763
          This is not a problem, as such master will not generate events with
1764
          the @@skip_replication flag set in the first place. We will still
1765 1766
          do slave-side filtering of such events though, to handle the (rare)
          case of downgrading a master and receiving old events generated from
1767
          before the downgrade with the @@skip_replication flag set.
1768 1769
        */
        DBUG_PRINT("info", ("Old master does not support master-side filtering "
1770 1771 1772 1773 1774 1775 1776 1777 1778 1779
                            "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;
1780 1781 1782
      }
    }
  }
1783 1784 1785 1786

  /* Announce MariaDB slave capabilities. */
  DBUG_EXECUTE_IF("simulate_slave_capability_none", goto after_set_capability;);
  {
1787 1788 1789 1790 1791 1792
    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)
1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811
    {
      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;
      }
    }
  }
1812
#ifndef DBUG_OFF
1813
after_set_capability:
1814
#endif
1815

1816 1817
  /* Request dump start from slave replication GTID state. */

unknown's avatar
unknown committed
1818
  if (mi->gtid_pos_auto)
1819 1820 1821 1822 1823
  {
    int rc;
    char str_buf[256];
    String connect_state(str_buf, sizeof(str_buf), system_charset_info);
    connect_state.length(0);
unknown's avatar
unknown committed
1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839
    rpl_gtid *binlog_gtid_list= NULL;
    uint32 num_binlog_gtids= 0;

    if (opt_bin_log)
    {
      int err= mysql_bin_log.get_most_recent_gtid_list(&binlog_gtid_list,
                                                       &num_binlog_gtids);
      if (err)
      {
        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;
      }
    }
1840 1841 1842

    connect_state.append(STRING_WITH_LEN("SET @slave_connect_state='"),
                         system_charset_info);
unknown's avatar
unknown committed
1843 1844 1845 1846
    rpl_global_gtid_slave_state.tostring(&connect_state, binlog_gtid_list,
                                         num_binlog_gtids);
    if (binlog_gtid_list)
      my_free(binlog_gtid_list);
1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868
    connect_state.append(STRING_WITH_LEN("'"), system_charset_info);
    rc= mysql_real_query(mysql, connect_state.ptr(), connect_state.length());
    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
1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893
  else
  {
    /*
      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
1894 1895
      rpl_global_gtid_slave_state.load(mi->io_thd, master_row[0],
                                       strlen(master_row[0]), false);
unknown's avatar
unknown committed
1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923
    }
    else if (check_io_slave_killed(mi->io_thd, mi, NULL))
      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;
    }
  }
1924

1925
err:
1926 1927
  if (errmsg)
  {
1928 1929
    if (master_res)
      mysql_free_result(master_res);
1930
    DBUG_ASSERT(err_code != 0);
1931
    mi->report(ERROR_LEVEL, err_code, "%s", err_buff);
1932
    DBUG_RETURN(1);
1933
  }
1934

1935
  DBUG_RETURN(0);
1936 1937 1938 1939 1940

network_err:
  if (master_res)
    mysql_free_result(master_res);
  DBUG_RETURN(2);
unknown's avatar
unknown committed
1941

1942 1943 1944 1945
slave_killed_err:
  if (master_res)
    mysql_free_result(master_res);
  DBUG_RETURN(2);
unknown's avatar
unknown committed
1946
}
unknown's avatar
unknown committed
1947

1948

1949
static bool wait_for_relay_log_space(Relay_log_info* rli)
unknown's avatar
unknown committed
1950
{
1951
  bool slave_killed=0;
1952
  Master_info* mi = rli->mi;
unknown's avatar
unknown committed
1953
  const char *save_proc_info;
unknown's avatar
unknown committed
1954 1955
  THD* thd = mi->io_thd;
  DBUG_ENTER("wait_for_relay_log_space");
1956

Marc Alff's avatar
Marc Alff committed
1957
  mysql_mutex_lock(&rli->log_space_lock);
unknown's avatar
unknown committed
1958
  save_proc_info= thd->enter_cond(&rli->log_space_cond,
unknown's avatar
unknown committed
1959 1960
                                  &rli->log_space_lock,
                                  "\
1961
Waiting for the slave SQL thread to free enough relay log space");
unknown's avatar
unknown committed
1962
  while (rli->log_space_limit < rli->log_space_total &&
unknown's avatar
unknown committed
1963
         !(slave_killed=io_slave_killed(thd,mi)) &&
1964
         !rli->ignore_log_space_limit)
Marc Alff's avatar
Marc Alff committed
1965
    mysql_cond_wait(&rli->log_space_cond, &rli->log_space_lock);
1966 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 2001 2002 2003 2004 2005 2006

  /* 
    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)
    {
2007
      mysql_mutex_lock(&mi->data_lock);
2008
      rotate_relay_log(rli->mi);
2009
      mysql_mutex_unlock(&mi->data_lock);
2010 2011 2012 2013 2014 2015
      rli->sql_force_rotate_relay= false;
    }

    rli->ignore_log_space_limit= false;
  }

unknown's avatar
unknown committed
2016
  thd->exit_cond(save_proc_info);
unknown's avatar
unknown committed
2017 2018
  DBUG_RETURN(slave_killed);
}
2019

2020

2021 2022
/*
  Builds a Rotate from the ignored events' info and writes it to relay log.
2023

2024 2025 2026 2027 2028 2029 2030 2031 2032 2033
  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).
 */
2034
static void write_ignored_events_info_to_relay_log(THD *thd, Master_info *mi)
2035
{
2036
  Relay_log_info *rli= &mi->rli;
Marc Alff's avatar
Marc Alff committed
2037
  mysql_mutex_t *log_lock= rli->relay_log.get_log_lock();
2038 2039
  DBUG_ENTER("write_ignored_events_info_to_relay_log");

2040
  DBUG_ASSERT(thd == mi->io_thd);
Marc Alff's avatar
Marc Alff committed
2041
  mysql_mutex_lock(log_lock);
2042
  if (rli->ign_master_log_name_end[0])
unknown's avatar
unknown committed
2043
  {
2044
    DBUG_PRINT("info",("writing a Rotate event to track down ignored events"));
2045
    Rotate_log_event *ev= new Rotate_log_event(rli->ign_master_log_name_end,
2046 2047 2048 2049
                                               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
2050
    mysql_mutex_unlock(log_lock);
2051 2052 2053 2054
    if (likely((bool)ev))
    {
      ev->server_id= 0; // don't be ignored by slave SQL thread
      if (unlikely(rli->relay_log.append(ev)))
2055 2056 2057 2058 2059
        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");
2060
      rli->relay_log.harvest_bytes_written(&rli->log_space_total);
Luis Soares's avatar
Luis Soares committed
2061
      if (flush_master_info(mi, TRUE, TRUE))
2062
        sql_print_error("Failed to flush master info file");
2063 2064 2065
      delete ev;
    }
    else
2066 2067 2068 2069
      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
2070
  }
2071
  else
Marc Alff's avatar
Marc Alff committed
2072
    mysql_mutex_unlock(log_lock);
2073
  DBUG_VOID_RETURN;
unknown's avatar
unknown committed
2074 2075
}

2076

2077
int register_slave_on_master(MYSQL* mysql, Master_info *mi,
2078
                             bool *suppress_warnings)
2079
{
2080
  uchar buf[1024], *pos= buf;
2081
  uint report_host_len=0, report_user_len=0, report_password_len=0;
2082
  DBUG_ENTER("register_slave_on_master");
2083

2084
  *suppress_warnings= FALSE;
2085 2086 2087 2088 2089 2090 2091 2092
  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);
2093
    DBUG_RETURN(0);
2094 2095
  }

2096
  if (report_user)
2097
    report_user_len= strlen(report_user);
2098 2099 2100 2101 2102 2103 2104 2105 2106
  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
2107
  if (report_password)
2108
    report_password_len= strlen(report_password);
2109 2110 2111 2112 2113 2114 2115 2116
  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);
  }
2117

2118
  int4store(pos, global_system_variables.server_id); pos+= 4;
2119 2120 2121
  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);
2122
  int2store(pos, (uint16) report_port); pos+= 2;
2123 2124 2125 2126 2127 2128
  /* 
    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;
2129
  /* The master will fill in master_id */
unknown's avatar
unknown committed
2130
  int4store(pos, 0);                    pos+= 4;
2131

2132
  if (simple_command(mysql, COM_REGISTER_SLAVE, buf, (size_t) (pos- buf), 0))
2133
  {
2134 2135 2136 2137
    if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
    {
      *suppress_warnings= TRUE;                 // Suppress reconnect warning
    }
2138
    else if (!check_io_slave_killed(mi->io_thd, mi, NULL))
2139 2140 2141 2142 2143 2144 2145
    {
      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);
    }
2146
    DBUG_RETURN(1);
2147
  }
2148
  DBUG_RETURN(0);
2149 2150
}

2151

2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162
/**
  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
*/
2163 2164 2165 2166

bool show_master_info(THD *thd, Master_info *mi, bool full)
{
  DBUG_ENTER("show_master_info");
unknown's avatar
unknown committed
2167
  String gtid_pos;
2168

unknown's avatar
unknown committed
2169
  if (full && rpl_global_gtid_slave_state.tostring(&gtid_pos, NULL, 0))
2170
    DBUG_RETURN(TRUE);
unknown's avatar
unknown committed
2171 2172 2173
  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))
2174 2175 2176 2177 2178
    DBUG_RETURN(TRUE);
  my_eof(thd);
  DBUG_RETURN(FALSE);
}

unknown's avatar
unknown committed
2179 2180
static bool send_show_master_info_header(THD *thd, bool full,
                                         size_t gtid_pos_length)
unknown's avatar
unknown committed
2181 2182
{
  List<Item> field_list;
2183
  Protocol *protocol= thd->protocol;
2184
  Master_info *mi;
2185 2186 2187
  DBUG_ENTER("show_master_info_header");

  if (full)
2188
  {
2189 2190
    field_list.push_back(new Item_empty_string("Connection_name",
                                               MAX_CONNECTION_NAME));
2191 2192 2193
    field_list.push_back(new Item_empty_string("Slave_SQL_State",
                                               30));
  }
2194

unknown's avatar
unknown committed
2195
  field_list.push_back(new Item_empty_string("Slave_IO_State",
2196
                                                     30));
unknown's avatar
unknown committed
2197
  field_list.push_back(new Item_empty_string("Master_Host",
unknown's avatar
unknown committed
2198
                                                     sizeof(mi->host)));
unknown's avatar
unknown committed
2199
  field_list.push_back(new Item_empty_string("Master_User",
unknown's avatar
unknown committed
2200
                                                     sizeof(mi->user)));
2201
  field_list.push_back(new Item_return_int("Master_Port", 7,
unknown's avatar
unknown committed
2202
                                           MYSQL_TYPE_LONG));
2203
  field_list.push_back(new Item_return_int("Connect_Retry", 10,
unknown's avatar
unknown committed
2204
                                           MYSQL_TYPE_LONG));
2205
  field_list.push_back(new Item_empty_string("Master_Log_File",
unknown's avatar
unknown committed
2206
                                             FN_REFLEN));
2207
  field_list.push_back(new Item_return_int("Read_Master_Log_Pos", 10,
unknown's avatar
unknown committed
2208
                                           MYSQL_TYPE_LONGLONG));
2209
  field_list.push_back(new Item_empty_string("Relay_Log_File",
unknown's avatar
unknown committed
2210
                                             FN_REFLEN));
2211
  field_list.push_back(new Item_return_int("Relay_Log_Pos", 10,
unknown's avatar
unknown committed
2212
                                           MYSQL_TYPE_LONGLONG));
2213
  field_list.push_back(new Item_empty_string("Relay_Master_Log_File",
unknown's avatar
unknown committed
2214
                                             FN_REFLEN));
2215 2216
  field_list.push_back(new Item_empty_string("Slave_IO_Running", 3));
  field_list.push_back(new Item_empty_string("Slave_SQL_Running", 3));
2217 2218 2219 2220 2221 2222
  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
2223
                                             28));
2224 2225 2226
  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
2227
                                           MYSQL_TYPE_LONG));
2228
  field_list.push_back(new Item_return_int("Exec_Master_Log_Pos", 10,
unknown's avatar
unknown committed
2229
                                           MYSQL_TYPE_LONGLONG));
2230
  field_list.push_back(new Item_return_int("Relay_Log_Space", 10,
unknown's avatar
unknown committed
2231
                                           MYSQL_TYPE_LONGLONG));
2232
  field_list.push_back(new Item_empty_string("Until_Condition", 6));
2233
  field_list.push_back(new Item_empty_string("Until_Log_File", FN_REFLEN));
unknown's avatar
unknown committed
2234
  field_list.push_back(new Item_return_int("Until_Log_Pos", 10,
2235
                                           MYSQL_TYPE_LONGLONG));
unknown's avatar
unknown committed
2236 2237 2238
  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
2239
  field_list.push_back(new Item_empty_string("Master_SSL_CA_Path",
unknown's avatar
unknown committed
2240
                                             sizeof(mi->ssl_capath)));
unknown's avatar
unknown committed
2241
  field_list.push_back(new Item_empty_string("Master_SSL_Cert",
unknown's avatar
unknown committed
2242
                                             sizeof(mi->ssl_cert)));
unknown's avatar
unknown committed
2243
  field_list.push_back(new Item_empty_string("Master_SSL_Cipher",
unknown's avatar
unknown committed
2244
                                             sizeof(mi->ssl_cipher)));
unknown's avatar
unknown committed
2245
  field_list.push_back(new Item_empty_string("Master_SSL_Key",
unknown's avatar
unknown committed
2246
                                             sizeof(mi->ssl_key)));
2247
  field_list.push_back(new Item_return_int("Seconds_Behind_Master", 10,
unknown's avatar
unknown committed
2248
                                           MYSQL_TYPE_LONGLONG));
2249 2250
  field_list.push_back(new Item_empty_string("Master_SSL_Verify_Server_Cert",
                                             3));
2251 2252 2253 2254
  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
2255 2256 2257 2258
  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
2259
  field_list.push_back(new Item_return_int("Using_Gtid", sizeof(ulong),
unknown's avatar
unknown committed
2260
                                           MYSQL_TYPE_LONG));
2261 2262 2263 2264 2265 2266 2267 2268
  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));
2269 2270 2271 2272
    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
2273
    field_list.push_back(new Item_empty_string("Gtid_Pos", gtid_pos_length));
2274
  }
2275
  if (protocol->send_result_set_metadata(&field_list,
2276
                            Protocol::SEND_NUM_ROWS | Protocol::SEND_EOF))
unknown's avatar
unknown committed
2277
    DBUG_RETURN(TRUE);
2278 2279 2280 2281
  DBUG_RETURN(FALSE);
}


unknown's avatar
unknown committed
2282 2283
static bool send_show_master_info_data(THD *thd, Master_info *mi, bool full,
                                       String *gtid_pos)
2284 2285
{
  DBUG_ENTER("send_show_master_info_data");
unknown's avatar
unknown committed
2286

2287 2288
  if (mi->host[0])
  {
2289
    DBUG_PRINT("info",("host is set: '%s'", mi->host));
2290
    String *packet= &thd->packet;
2291
    Protocol *protocol= thd->protocol;
2292 2293
    char buf[256];
    String tmp(buf, sizeof(buf), &my_charset_bin);
2294

2295
    protocol->prepare_for_resend();
unknown's avatar
unknown committed
2296

2297
    /*
2298 2299
      slave_running can be accessed without run_lock but not other
      non-volotile members like mi->io_thd, which is guarded by the mutex.
2300
    */
2301 2302 2303
    if (full)
      protocol->store(mi->connection_name.str, mi->connection_name.length,
                      &my_charset_bin);
Marc Alff's avatar
Marc Alff committed
2304
    mysql_mutex_lock(&mi->run_lock);
2305 2306 2307
    if (full)
      protocol->store(mi->rli.sql_thd ? mi->rli.sql_thd->proc_info : "",
                      &my_charset_bin);
2308
    protocol->store(mi->io_thd ? mi->io_thd->proc_info : "", &my_charset_bin);
Marc Alff's avatar
Marc Alff committed
2309
    mysql_mutex_unlock(&mi->run_lock);
2310

Marc Alff's avatar
Marc Alff committed
2311 2312 2313 2314
    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);
2315 2316
    protocol->store(mi->host, &my_charset_bin);
    protocol->store(mi->user, &my_charset_bin);
2317 2318
    protocol->store((uint32) mi->port);
    protocol->store((uint32) mi->connect_retry);
2319
    protocol->store(mi->master_log_name, &my_charset_bin);
2320
    protocol->store((ulonglong) mi->master_log_pos);
2321
    protocol->store(mi->rli.group_relay_log_name +
unknown's avatar
unknown committed
2322 2323
                    dirname_length(mi->rli.group_relay_log_name),
                    &my_charset_bin);
2324 2325
    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
2326
    protocol->store(mi->slave_running == MYSQL_SLAVE_RUN_CONNECT ?
2327 2328
                    "Yes" : (mi->slave_running == MYSQL_SLAVE_RUN_NOT_CONNECT ?
                             "Connecting" : "No"), &my_charset_bin);
2329
    protocol->store(mi->rli.slave_running ? "Yes":"No", &my_charset_bin);
2330 2331 2332 2333
    protocol->store(rpl_filter->get_do_db());
    protocol->store(rpl_filter->get_ignore_db());

    rpl_filter->get_do_table(&tmp);
2334
    protocol->store(&tmp);
2335
    rpl_filter->get_ignore_table(&tmp);
2336
    protocol->store(&tmp);
2337
    rpl_filter->get_wild_do_table(&tmp);
2338
    protocol->store(&tmp);
2339
    rpl_filter->get_wild_ignore_table(&tmp);
2340 2341
    protocol->store(&tmp);

2342 2343
    protocol->store(mi->rli.last_error().number);
    protocol->store(mi->rli.last_error().message, &my_charset_bin);
2344
    protocol->store((uint32) mi->rli.slave_skip_counter);
2345
    protocol->store((ulonglong) mi->rli.group_master_log_pos);
2346
    protocol->store((ulonglong) mi->rli.log_space_total);
2347 2348

    protocol->store(
2349 2350
      mi->rli.until_condition==Relay_log_info::UNTIL_NONE ? "None":
        ( mi->rli.until_condition==Relay_log_info::UNTIL_MASTER_POS? "Master":
2351 2352 2353
          "Relay"), &my_charset_bin);
    protocol->store(mi->rli.until_log_name, &my_charset_bin);
    protocol->store((ulonglong) mi->rli.until_log_pos);
unknown's avatar
unknown committed
2354 2355

#ifdef HAVE_OPENSSL
unknown's avatar
unknown committed
2356 2357 2358 2359 2360 2361 2362 2363 2364
    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
2365

2366 2367 2368 2369 2370 2371
    /*
      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)
2372
    {
2373
      long time_diff= ((long)(time(0) - mi->rli.last_master_timestamp)
unknown's avatar
unknown committed
2374
                       - mi->clock_diff_with_master);
2375
      /*
unknown's avatar
unknown committed
2376 2377
        Apparently on some systems time_diff can be <0. Here are possible
        reasons related to MySQL:
2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389
        - 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.
2390 2391 2392 2393
        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".
2394
      */
unknown's avatar
unknown committed
2395 2396
      protocol->store((longlong)(mi->rli.last_master_timestamp ?
                                 max(0, time_diff) : 0));
2397
    }
unknown's avatar
unknown committed
2398
    else
2399
    {
unknown's avatar
unknown committed
2400
      protocol->store_null();
2401 2402
    }
    protocol->store(mi->ssl_verify_server_cert? "Yes":"No", &my_charset_bin);
unknown's avatar
unknown committed
2403

2404
    // Last_IO_Errno
2405
    protocol->store(mi->last_error().number);
2406
    // Last_IO_Error
2407
    protocol->store(mi->last_error().message, &my_charset_bin);
2408
    // Last_SQL_Errno
2409
    protocol->store(mi->rli.last_error().number);
2410
    // Last_SQL_Error
2411
    protocol->store(mi->rli.last_error().message, &my_charset_bin);
Andrei Elkin's avatar
Andrei Elkin committed
2412 2413 2414 2415 2416 2417 2418 2419 2420 2421
    // 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);
2422
        slen= sprintf(sbuff, (i==0? "%lu" : ", %lu"), s_id);
Andrei Elkin's avatar
Andrei Elkin committed
2423 2424 2425 2426 2427 2428
        if (cur_len + slen + 4 > FN_REFLEN)
        {
          /*
            break the loop whenever remained space could not fit
            ellipses on the next cycle
          */
2429
          sprintf(buff + cur_len, "...");
Andrei Elkin's avatar
Andrei Elkin committed
2430 2431
          break;
        }
2432
        cur_len += sprintf(buff + cur_len, "%s", sbuff);
Andrei Elkin's avatar
Andrei Elkin committed
2433 2434 2435 2436 2437
      }
      protocol->store(buff, &my_charset_bin);
    }
    // Master_Server_id
    protocol->store((uint32) mi->master_id);
unknown's avatar
unknown committed
2438
    protocol->store((uint32) (mi->gtid_pos_auto != 0));
2439 2440 2441 2442 2443
    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);
2444 2445
      protocol->store((uint32)    mi->received_heartbeats);
      protocol->store((double)    mi->heartbeat_period, 3, &tmp);
unknown's avatar
unknown committed
2446
      protocol->store(gtid_pos->ptr(), gtid_pos->length(), &my_charset_bin);
2447
    }
2448

Marc Alff's avatar
Marc Alff committed
2449 2450 2451 2452
    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);
2453

2454
    if (my_net_write(&thd->net, (uchar*) thd->packet.ptr(), packet->length()))
unknown's avatar
unknown committed
2455
      DBUG_RETURN(TRUE);
2456
  }
2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480
  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
2481 2482 2483

  @note
  master_info_index is protected by LOCK_active_mi.
2484 2485 2486 2487 2488
*/

bool show_all_master_info(THD* thd)
{
  uint i, elements;
unknown's avatar
unknown committed
2489
  String gtid_pos;
2490 2491
  Master_info **tmp;
  DBUG_ENTER("show_master_info");
2492
  mysql_mutex_assert_owner(&LOCK_active_mi);
2493

unknown's avatar
unknown committed
2494 2495 2496
  rpl_global_gtid_slave_state.tostring(&gtid_pos, NULL, 0);

  if (send_show_master_info_header(thd, 1, gtid_pos.length()))
2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517
    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
2518
    if (send_show_master_info_data(thd, tmp[i], 1, &gtid_pos))
2519 2520 2521 2522
      DBUG_RETURN(TRUE);
  }

end:
2523
  my_eof(thd);
unknown's avatar
unknown committed
2524
  DBUG_RETURN(FALSE);
unknown's avatar
unknown committed
2525 2526
}

unknown's avatar
unknown committed
2527

2528 2529
void set_slave_thread_options(THD* thd)
{
2530
  DBUG_ENTER("set_slave_thread_options");
2531 2532 2533 2534 2535 2536 2537 2538 2539
  /*
     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.
  */
2540
  ulonglong options= thd->variables.option_bits | OPTION_BIG_SELECTS;
2541 2542 2543 2544
  if (opt_log_slave_updates)
    options|= OPTION_BIN_LOG;
  else
    options&= ~OPTION_BIN_LOG;
2545
  thd->variables.option_bits= options;
2546
  thd->variables.completion_type= 0;
2547
  DBUG_VOID_RETURN;
2548
}
2549

2550
void set_slave_thread_default_charset(THD* thd, Relay_log_info const *rli)
2551
{
2552 2553
  DBUG_ENTER("set_slave_thread_default_charset");

2554 2555 2556 2557 2558 2559 2560
  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();
2561 2562 2563 2564 2565 2566 2567

  /*
    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.
   */
2568
  const_cast<Relay_log_info*>(rli)->cached_charset_invalidate();
2569
  DBUG_VOID_RETURN;
2570 2571
}

unknown's avatar
unknown committed
2572
/*
2573
  init_slave_thread()
unknown's avatar
unknown committed
2574
*/
2575

2576 2577
static int init_slave_thread(THD* thd, Master_info *mi,
                             SLAVE_THD_TYPE thd_type)
unknown's avatar
unknown committed
2578 2579
{
  DBUG_ENTER("init_slave_thread");
Sergei Golubchik's avatar
Sergei Golubchik committed
2580
  int simulate_error __attribute__((unused))= 0;
2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593
  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);
  }

2594
  thd->system_thread = (thd_type == SLAVE_THD_SQL) ?
unknown's avatar
unknown committed
2595
    SYSTEM_THREAD_SLAVE_SQL : SYSTEM_THREAD_SLAVE_IO;
2596
  thd->security_ctx->skip_grants();
2597 2598 2599 2600 2601
/*
  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.
*/
2602
  thd->variables.max_allowed_packet= slave_max_allowed_packet;
2603 2604
  thd->slave_thread= 1;
  thd->connection_name= mi->connection_name;
2605
  thd->enable_slow_log= opt_log_slow_slave_statements;
2606
  thd->variables.log_slow_filter= global_system_variables.log_slow_filter;
2607
  set_slave_thread_options(thd);
unknown's avatar
unknown committed
2608
  thd->client_capabilities = CLIENT_LOCAL_FILES;
Marc Alff's avatar
Marc Alff committed
2609
  mysql_mutex_lock(&LOCK_thread_count);
unknown's avatar
unknown committed
2610
  thd->thread_id= thd->variables.pseudo_thread_id= thread_id++;
Marc Alff's avatar
Marc Alff committed
2611
  mysql_mutex_unlock(&LOCK_thread_count);
unknown's avatar
unknown committed
2612

2613
  if (thd_type == SLAVE_THD_SQL)
2614
    thd_proc_info(thd, "Waiting for the next event in relay log");
2615
  else
2616
    thd_proc_info(thd, "Waiting for master update");
unknown's avatar
unknown committed
2617
  thd->set_time();
2618 2619
  /* Do not use user-supplied timeout value for system threads. */
  thd->variables.lock_wait_timeout= LONG_TIMEOUT;
unknown's avatar
unknown committed
2620 2621 2622
  DBUG_RETURN(0);
}

2623 2624 2625 2626 2627 2628 2629
/*
  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.
2630

2631 2632 2633 2634 2635
  @retval True if the thread has been killed, false otherwise.
*/
template <typename killed_func, typename rpl_info>
static inline bool slave_sleep(THD *thd, time_t seconds,
                               killed_func func, rpl_info info)
unknown's avatar
unknown committed
2636
{
2637

2638 2639 2640
  bool ret;
  struct timespec abstime;
  const char *old_proc_info;
unknown's avatar
unknown committed
2641

2642 2643
  mysql_mutex_t *lock= &info->sleep_lock;
  mysql_cond_t *cond= &info->sleep_cond;
unknown's avatar
unknown committed
2644

2645 2646 2647 2648 2649 2650 2651 2652 2653 2654
  /* 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);

  while (! (ret= func(thd, info)))
  {
    int error= mysql_cond_timedwait(cond, lock, &abstime);
    if (error == ETIMEDOUT || error == ETIME)
      break;
unknown's avatar
unknown committed
2655
  }
2656 2657 2658
  /* Implicitly unlocks the mutex. */
  thd->exit_cond(old_proc_info);
  return ret;
unknown's avatar
unknown committed
2659 2660
}

2661

He Zhenxing's avatar
He Zhenxing committed
2662 2663
static int request_dump(THD *thd, MYSQL* mysql, Master_info* mi,
			bool *suppress_warnings)
unknown's avatar
unknown committed
2664
{
2665
  uchar buf[FN_REFLEN + 10];
unknown's avatar
unknown committed
2666
  int len;
He Zhenxing's avatar
He Zhenxing committed
2667
  ushort binlog_flags = 0; // for now
2668
  char* logname = mi->master_log_name;
2669
  DBUG_ENTER("request_dump");
2670 2671
  
  *suppress_warnings= FALSE;
2672

2673
  if (opt_log_slave_updates && opt_replicate_annotate_row_events)
2674 2675
    binlog_flags|= BINLOG_SEND_ANNOTATE_ROWS_EVENT;

He Zhenxing's avatar
He Zhenxing committed
2676 2677 2678 2679 2680
  if (RUN_HOOK(binlog_relay_io,
               before_request_transmit,
               (thd, mi, binlog_flags)))
    DBUG_RETURN(1);
  
unknown's avatar
unknown committed
2681
  // TODO if big log files: Change next to int8store()
unknown's avatar
unknown committed
2682
  int4store(buf, (ulong) mi->master_log_pos);
unknown's avatar
unknown committed
2683
  int2store(buf + 4, binlog_flags);
2684
  int4store(buf + 6, global_system_variables.server_id);
unknown's avatar
unknown committed
2685
  len = (uint) strlen(logname);
2686
  memcpy(buf + 10, logname,len);
unknown's avatar
SCRUM  
unknown committed
2687
  if (simple_command(mysql, COM_BINLOG_DUMP, buf, len + 10, 1))
unknown's avatar
unknown committed
2688
  {
unknown's avatar
unknown committed
2689 2690 2691 2692 2693
    /*
      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
2694
    if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
2695
      *suppress_warnings= TRUE;                 // Suppress reconnect warning
unknown's avatar
unknown committed
2696
    else
2697
      sql_print_error("Error on COM_BINLOG_DUMP: %d  %s, will retry in %d secs",
unknown's avatar
unknown committed
2698
                      mysql_errno(mysql), mysql_error(mysql),
2699
                      mi->connect_retry);
2700
    DBUG_RETURN(1);
unknown's avatar
unknown committed
2701
  }
unknown's avatar
unknown committed
2702

2703
  DBUG_RETURN(0);
unknown's avatar
unknown committed
2704 2705
}

2706

unknown's avatar
unknown committed
2707
/*
2708
  Read one event from the master
unknown's avatar
unknown committed
2709

unknown's avatar
unknown committed
2710 2711
  SYNOPSIS
    read_event()
unknown's avatar
unknown committed
2712 2713 2714 2715 2716 2717
    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.
2718

unknown's avatar
unknown committed
2719
    RETURN VALUES
unknown's avatar
unknown committed
2720 2721
    'packet_error'      Error
    number              Length of packet
unknown's avatar
unknown committed
2722 2723
*/

2724
static ulong read_event(MYSQL* mysql, Master_info *mi, bool* suppress_warnings)
unknown's avatar
unknown committed
2725
{
2726
  ulong len;
2727
  DBUG_ENTER("read_event");
unknown's avatar
unknown committed
2728

2729
  *suppress_warnings= FALSE;
unknown's avatar
unknown committed
2730 2731 2732 2733
  /*
    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
2734
#ifndef DBUG_OFF
2735
  if (disconnect_slave_event_count && !(mi->events_till_disconnect--))
unknown's avatar
unknown committed
2736
    DBUG_RETURN(packet_error);
unknown's avatar
unknown committed
2737
#endif
unknown's avatar
unknown committed
2738

2739
  len = cli_safe_read(mysql);
2740
  if (len == packet_error || (long) len < 1)
unknown's avatar
unknown committed
2741
  {
unknown's avatar
SCRUM  
unknown committed
2742
    if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
unknown's avatar
unknown committed
2743 2744
    {
      /*
unknown's avatar
unknown committed
2745 2746 2747
        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
2748 2749 2750 2751
      */
      *suppress_warnings= TRUE;
    }
    else
2752
      sql_print_error("Error reading packet from server: %s ( server_errno=%d)",
unknown's avatar
unknown committed
2753
                      mysql_error(mysql), mysql_errno(mysql));
2754
    DBUG_RETURN(packet_error);
unknown's avatar
unknown committed
2755 2756
  }

2757 2758
  /* Check if eof packet */
  if (len < 8 && mysql->net.read_pos[0] == 254)
unknown's avatar
unknown committed
2759
  {
2760 2761
    sql_print_information("Slave: received end packet from server, apparent "
                          "master shutdown: %s",
unknown's avatar
unknown committed
2762
                     mysql_error(mysql));
2763
     DBUG_RETURN(packet_error);
unknown's avatar
unknown committed
2764
  }
unknown's avatar
unknown committed
2765

unknown's avatar
unknown committed
2766
  DBUG_PRINT("exit", ("len: %lu  net->read_pos[4]: %d",
unknown's avatar
unknown committed
2767 2768
                      len, mysql->net.read_pos[4]));
  DBUG_RETURN(len - 1);
unknown's avatar
unknown committed
2769 2770
}

unknown's avatar
unknown committed
2771 2772 2773 2774 2775 2776 2777 2778 2779
/*
  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)
{
2780 2781
  DBUG_ENTER("has_temporary_error");

2782
  DBUG_EXECUTE_IF("all_errors_are_temporary_errors",
Marc Alff's avatar
Marc Alff committed
2783
                  if (thd->stmt_da->is_error())
2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795
                  {
                    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
2796 2797 2798 2799 2800 2801

  /*
    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
2802 2803
  if (thd->stmt_da->sql_errno() == ER_LOCK_DEADLOCK ||
      thd->stmt_da->sql_errno() == ER_LOCK_WAIT_TIMEOUT)
2804
    DBUG_RETURN(1);
unknown's avatar
unknown committed
2805 2806 2807 2808 2809

#ifdef HAVE_NDB_BINLOG
  /*
    currently temporary error set in ndbcluster
  */
Marc Alff's avatar
Marc Alff committed
2810
  List_iterator_fast<MYSQL_ERROR> it(thd->warning_info->warn_list());
unknown's avatar
unknown committed
2811 2812 2813
  MYSQL_ERROR *err;
  while ((err= it++))
  {
Marc Alff's avatar
Marc Alff committed
2814 2815 2816
    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
2817 2818
    {
    case ER_GET_TEMPORARY_ERRMSG:
2819
      DBUG_RETURN(1);
unknown's avatar
unknown committed
2820 2821 2822 2823 2824
    default:
      break;
    }
  }
#endif
2825
  DBUG_RETURN(0);
unknown's avatar
unknown committed
2826
}
2827

2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854

/**
  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().
*/
2855
int apply_event_and_update_pos(Log_event* ev, THD* thd, Relay_log_info* rli)
2856 2857 2858 2859 2860 2861 2862 2863 2864
{
  int exec_res= 0;

  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));
  DBUG_PRINT("info", ("thd->options: %s%s; rli->last_event_start_time: %lu",
2865 2866
                      FLAGSTR(thd->variables.option_bits, OPTION_NOT_AUTOCOMMIT),
                      FLAGSTR(thd->variables.option_bits, OPTION_BEGIN),
2867
                      (ulong) rli->last_event_start_time));
2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892

  /*
    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).
  */

2893 2894
  /* Use the original server id for logging. */
  thd->variables.server_id = ev->server_id;
2895 2896 2897
  thd->set_time();                            // time the query
  thd->lex->current_select= 0;
  if (!ev->when)
2898 2899
  {
    my_hrtime_t hrtime= my_hrtime();
2900
    ev->when= hrtime_to_my_time(hrtime);
2901 2902
    ev->when_sec_part= hrtime_sec_part(hrtime);
  }
2903 2904
  thd->variables.option_bits=
    (thd->variables.option_bits & ~OPTION_SKIP_REPLICATION) |
2905
    (ev->flags & LOG_EVENT_SKIP_REPLICATION_F ? OPTION_SKIP_REPLICATION : 0);
2906 2907
  ev->thd = thd; // because up to this point, ev->thd == 0

2908 2909
  int reason= ev->shall_skip(rli);
  if (reason == Log_event::EVENT_SKIP_COUNT)
2910 2911 2912 2913
  {
    DBUG_ASSERT(rli->slave_skip_counter > 0);
    rli->slave_skip_counter--;
  }
Marc Alff's avatar
Marc Alff committed
2914
  mysql_mutex_unlock(&rli->data_lock);
2915 2916 2917
  if (reason == Log_event::EVENT_SKIP_NOT)
    exec_res= ev->apply_event(rli);

2918
#ifndef DBUG_OFF
2919 2920
  /*
    This only prints information to the debug trace.
2921

2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932
    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"
  };
  DBUG_PRINT("info", ("OPTION_BEGIN: %d; IN_STMT: %d",
2933
                      test(thd->variables.option_bits & OPTION_BEGIN),
2934 2935 2936
                      rli->get_flag(Relay_log_info::IN_STMT)));
  DBUG_PRINT("skip_event", ("%s event was %s",
                            ev->get_type_str(), explain[reason]));
2937 2938 2939 2940 2941 2942
#endif

  DBUG_PRINT("info", ("apply_event error = %d", exec_res));
  if (exec_res == 0)
  {
    int error= ev->update_pos(rli);
2943
#ifdef HAVE_valgrind
2944 2945 2946
    if (!rli->is_fake)
#endif
    {
2947
#ifndef DBUG_OFF
2948
      char buf[22];
2949
#endif
2950 2951 2952 2953 2954 2955 2956 2957
      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));
    }
2958 2959 2960 2961 2962 2963 2964 2965 2966
    /*
      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)
    {
2967
      char buf[22];
2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002
      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);
}


/**
  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
3003
     if a failure occurs when applying the event.)
3004 3005 3006 3007 3008 3009 3010

   - An error occurred when updating the binlog position.

  @retval 0 The event was applied.

  @retval 1 The event was not applied.
*/
3011
static int exec_relay_log_event(THD* thd, Relay_log_info* rli)
unknown's avatar
unknown committed
3012
{
3013 3014
  DBUG_ENTER("exec_relay_log_event");

3015 3016
  /*
     We acquire this mutex since we need it for all operations except
3017
     event execution. But we will release it in places where we will
3018 3019
     wait for something for example inside of next_event().
   */
Marc Alff's avatar
Marc Alff committed
3020
  mysql_mutex_lock(&rli->data_lock);
3021

3022
  Log_event * ev = next_event(rli);
3023

3024
  DBUG_ASSERT(rli->sql_thd==thd);
3025

3026
  if (sql_slave_killed(thd,rli))
3027
  {
Marc Alff's avatar
Marc Alff committed
3028
    mysql_mutex_unlock(&rli->data_lock);
unknown's avatar
unknown committed
3029
    delete ev;
3030
    DBUG_RETURN(1);
3031
  }
3032 3033
  if (ev)
  {
3034
    int exec_res;
3035

3036 3037 3038 3039
    /*
      This tests if the position of the beginning of the current event
      hits the UNTIL barrier.
    */
3040
    if (rli->until_condition != Relay_log_info::UNTIL_NONE &&
3041
        rli->is_until_satisfied(thd, ev))
3042 3043 3044 3045 3046 3047 3048 3049 3050
    {
      char buf[22];
      sql_print_information("Slave SQL thread stopped because it reached its"
                            " UNTIL position %s", llstr(rli->until_pos(), buf));
      /*
        Setting abort_slave flag because we do not want additional message about
        error in query execution to be printed.
      */
      rli->abort_slave= 1;
Marc Alff's avatar
Marc Alff committed
3051
      mysql_mutex_unlock(&rli->data_lock);
3052
      delete ev;
3053
      DBUG_RETURN(1);
3054
    }
3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068

    { /**
         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",
                      if ((ev->get_type_code() == XID_EVENT) ||
                          ((ev->get_type_code() == QUERY_EVENT) &&
                           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
3069
                        mysql_mutex_unlock(&rli->data_lock);
3070 3071 3072 3073 3074 3075
                        delete ev;
                        rli->inc_event_relay_log_pos();
                        DBUG_RETURN(0);
                      };);
    }

3076
    exec_res= apply_event_and_update_pos(ev, thd, rli);
3077

3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107
    switch (ev->get_type_code()) {
      case FORMAT_DESCRIPTION_EVENT:
        /*
          Format_description_log_event should not be deleted because it
          will be used to read info about the relay log's format;
          it will be deleted when the SQL thread does not need it,
          i.e. when this thread terminates.
        */
        break;
      case ANNOTATE_ROWS_EVENT:
        /*
          Annotate_rows event should not be deleted because after it has
          been applied, thd->query points to the string inside this event.
          The thd->query will be used to generate new Annotate_rows event
          during applying the subsequent Rows events.
        */
        rli->set_annotate_event((Annotate_rows_log_event*) ev);
        break;
      case DELETE_ROWS_EVENT:
      case UPDATE_ROWS_EVENT:
      case WRITE_ROWS_EVENT:
        /*
          After the last Rows event has been applied, the saved Annotate_rows
          event (if any) is not needed anymore and can be deleted.
        */
        if (((Rows_log_event*)ev)->get_flags(Rows_log_event::STMT_END_F))
          rli->free_annotate_event();
        /* fall through */
      default:
        DBUG_PRINT("info", ("Deleting the event after it has been executed"));
Sergei Golubchik's avatar
Sergei Golubchik committed
3108 3109
        if (!rli->is_deferred_event(ev))
          delete ev;
3110
        break;
3111 3112
    }

3113

3114
    /*
3115 3116
      update_log_pos failed: this should not happen, so we don't
      retry.
3117
    */
3118 3119 3120
    if (exec_res == 2)
      DBUG_RETURN(1);

3121 3122
    if (slave_trans_retries)
    {
3123
      int temp_err;
Michael Widenius's avatar
Michael Widenius committed
3124
      LINT_INIT(temp_err);
3125
      if (exec_res && (temp_err= has_temporary_error(thd)))
3126 3127 3128 3129
      {
        const char *errmsg;
        /*
          We were in a transaction which has been rolled back because of a
unknown's avatar
unknown committed
3130 3131
          temporary error;
          let's seek back to BEGIN log event and retry it all again.
3132
	  Note, if lock wait timeout (innodb_lock_wait_timeout exceeded)
3133
	  there is no rollback since 5.0.13 (ref: manual).
3134 3135 3136 3137 3138 3139 3140 3141
          We have to not only seek but also
          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
          init_master_info()).
          b) init_relay_log_pos(), because the BEGIN may be an older relay log.
        */
3142
        if (rli->trans_retries < slave_trans_retries)
3143 3144 3145 3146 3147 3148
        {
          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,
3149
                                      1, &errmsg, 1))
3150 3151 3152 3153 3154
            sql_print_error("Error initializing relay log position: %s",
                            errmsg);
          else
          {
            exec_res= 0;
3155
            rli->cleanup_context(thd, 1);
unknown's avatar
unknown committed
3156
            /* chance for concurrent connection to get more locks */
3157 3158
            slave_sleep(thd, min(rli->trans_retries, MAX_SLAVE_RETRY_PAUSE),
                       sql_slave_killed, rli);
Marc Alff's avatar
Marc Alff committed
3159
            mysql_mutex_lock(&rli->data_lock); // because of SHOW STATUS
unknown's avatar
unknown committed
3160
            rli->trans_retries++;
3161
            rli->retried_trans++;
3162
            statistic_increment(slave_retried_transactions, LOCK_status);
Marc Alff's avatar
Marc Alff committed
3163
            mysql_mutex_unlock(&rli->data_lock);
3164 3165
            DBUG_PRINT("info", ("Slave retries transaction "
                                "rli->trans_retries: %lu", rli->trans_retries));
unknown's avatar
unknown committed
3166
          }
3167 3168 3169 3170 3171 3172 3173
        }
        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);
      }
3174
      else if ((exec_res && !temp_err) ||
3175 3176
               (opt_using_transactions &&
                rli->group_relay_log_pos == rli->event_relay_log_pos))
3177 3178
      {
        /*
3179 3180 3181
          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
3182 3183 3184
          non-transient error, the slave will stop with an error.
         */
        rli->trans_retries= 0; // restart from fresh
unknown's avatar
unknown committed
3185
        DBUG_PRINT("info", ("Resetting retry counter, rli->trans_retries: %lu",
3186
                            rli->trans_retries));
3187
      }
3188 3189
    }
    DBUG_RETURN(exec_res);
3190
  }
Marc Alff's avatar
Marc Alff committed
3191
  mysql_mutex_unlock(&rli->data_lock);
3192 3193
  rli->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_READ_FAILURE,
              ER(ER_SLAVE_RELAY_LOG_READ_FAILURE), "\
3194 3195 3196 3197 3198 3199 3200
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
3201
");
3202
  DBUG_RETURN(1);
unknown's avatar
unknown committed
3203 3204
}

3205

3206
static bool check_io_slave_killed(THD *thd, Master_info *mi, const char *info)
3207 3208 3209
{
  if (io_slave_killed(thd, mi))
  {
3210
    if (info && global_system_variables.log_warnings)
3211
      sql_print_information("%s", info);
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
    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.
*/

3243
static int try_to_reconnect(THD *thd, MYSQL *mysql, Master_info *mi,
3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256
                            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
3257
    slave_sleep(thd, mi->connect_retry, io_slave_killed, mi);
3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278
  }
  if (check_io_slave_killed(thd, mi, messages[SLAVE_RECON_MSG_KILLED_WAITING]))
    return 1;
  thd->proc_info = messages[SLAVE_RECON_MSG_AFTER];
  if (!suppress_warnings) 
  {
    char buf[256], llbuff[22];
    my_snprintf(buf, sizeof(buf), messages[SLAVE_RECON_MSG_FAILED], 
                IO_RPL_LOG_NAME, llstr(mi->master_log_pos, llbuff));
    /* 
      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
    {
3279
      sql_print_information("%s", buf);
3280 3281 3282 3283 3284
    }
  }
  if (safe_reconnect(thd, mysql, mi, 1) || io_slave_killed(thd, mi))
  {
    if (global_system_variables.log_warnings)
3285
      sql_print_information("%s", messages[SLAVE_RECON_MSG_KILLED_AFTER]);
3286 3287 3288 3289 3290 3291
    return 1;
  }
  return 0;
}


3292 3293 3294 3295 3296
/**
  Slave IO thread entry point.

  @param arg Pointer to Master_info struct that holds information for
  the IO thread.
3297

3298 3299
  @return Always 0.
*/
3300
pthread_handler_t handle_slave_io(void *arg)
unknown's avatar
unknown committed
3301
{
unknown's avatar
unknown committed
3302 3303
  THD *thd; // needs to be first for thread_stack
  MYSQL *mysql;
3304
  Master_info *mi = (Master_info*)arg;
3305
  Relay_log_info *rli= &mi->rli;
unknown's avatar
unknown committed
3306 3307
  char llbuff[22];
  uint retry_count;
3308
  bool suppress_warnings;
3309
  int ret;
3310 3311 3312
#ifndef DBUG_OFF
  uint retry_count_reg= 0, retry_count_dump= 0, retry_count_event= 0;
#endif
unknown's avatar
unknown committed
3313 3314
  // needs to call my_thread_init(), otherwise we get a coredump in DBUG_ stuff
  my_thread_init();
3315
  DBUG_ENTER("handle_slave_io");
unknown's avatar
unknown committed
3316

3317
  DBUG_ASSERT(mi->inited);
unknown's avatar
unknown committed
3318 3319 3320
  mysql= NULL ;
  retry_count= 0;

3321 3322
  thd= new THD; // note that contructor of THD uses DBUG_ !

Marc Alff's avatar
Marc Alff committed
3323
  mysql_mutex_lock(&mi->run_lock);
unknown's avatar
unknown committed
3324 3325 3326
  /* Inform waiting threads that slave has started */
  mi->slave_run_id++;

3327
#ifndef DBUG_OFF
3328
  mi->events_till_disconnect = disconnect_slave_event_count;
3329 3330
#endif

3331
  THD_CHECK_SENTRY(thd);
3332
  mi->io_thd = thd;
unknown's avatar
unknown committed
3333 3334

  pthread_detach_this_thread();
3335
  thd->thread_stack= (char*) &thd; // remember where our stack is
3336
  mi->clear_error();
3337
  if (init_slave_thread(thd, mi, SLAVE_THD_IO))
unknown's avatar
unknown committed
3338
  {
Marc Alff's avatar
Marc Alff committed
3339
    mysql_cond_broadcast(&mi->start_cond);
unknown's avatar
unknown committed
3340
    sql_print_error("Failed during slave I/O thread initialization");
3341
    goto err_during_init;
unknown's avatar
unknown committed
3342
  }
Marc Alff's avatar
Marc Alff committed
3343
  mysql_mutex_lock(&LOCK_thread_count);
unknown's avatar
unknown committed
3344
  threads.append(thd);
Marc Alff's avatar
Marc Alff committed
3345
  mysql_mutex_unlock(&LOCK_thread_count);
3346
  mi->slave_running = MYSQL_SLAVE_RUN_NOT_CONNECT;
3347
  mi->abort_slave = 0;
Marc Alff's avatar
Marc Alff committed
3348 3349
  mysql_mutex_unlock(&mi->run_lock);
  mysql_cond_broadcast(&mi->start_cond);
3350

3351
  DBUG_PRINT("master_info",("log_file_name: '%s'  position: %s",
unknown's avatar
unknown committed
3352 3353
                            mi->master_log_name,
                            llstr(mi->master_log_pos,llbuff)));
3354

He Zhenxing's avatar
He Zhenxing committed
3355 3356 3357 3358 3359 3360 3361 3362 3363 3364
  /* This must be called before run any binlog_relay_io hooks */
  my_pthread_setspecific_ptr(RPL_MASTER_INFO, mi);

  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
3365
  if (!(mi->mysql = mysql = mysql_init(NULL)))
unknown's avatar
unknown committed
3366
  {
3367 3368
    mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
               ER(ER_SLAVE_FATAL_ERROR), "error in mysql_init()");
unknown's avatar
unknown committed
3369 3370
    goto err;
  }
3371

3372
  thd_proc_info(thd, "Connecting to master");
3373
  // we can get killed during safe_connect
3374
  if (!safe_connect(thd, mysql, mi))
3375
  {
unknown's avatar
unknown committed
3376 3377 3378
    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,
3379 3380
			  IO_RPL_LOG_NAME,
			  llstr(mi->master_log_pos,llbuff));
3381 3382 3383 3384 3385
  /*
    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.
  */
3386
    thd->net.max_packet_size= slave_max_allowed_packet;
3387 3388
    mysql->net.max_packet_size= thd->net.max_packet_size+= MAX_LOG_EVENT_HEADER;
  }
3389
  else
unknown's avatar
unknown committed
3390
  {
3391
    sql_print_information("Slave I/O thread killed while connecting to master");
unknown's avatar
unknown committed
3392 3393
    goto err;
  }
3394

3395
connected:
3396

3397 3398 3399 3400 3401
    DBUG_EXECUTE_IF("dbug.before_get_running_status_yes",
                    {
                      const char act[]=
                        "now "
                        "wait_for signal.io_thread_let_running";
3402
                      DBUG_ASSERT(debug_sync_service);
3403 3404 3405 3406
                      DBUG_ASSERT(!debug_sync_set_action(thd, 
                                                         STRING_WITH_LEN(act)));
                    };);

3407 3408
  // TODO: the assignment below should be under mutex (5.0)
  mi->slave_running= MYSQL_SLAVE_RUN_CONNECT;
3409
  thd->slave_net = &mysql->net;
3410
  thd_proc_info(thd, "Checking master version");
3411 3412 3413
  ret= get_master_version_and_clock(mysql, mi);
  if (ret == 1)
    /* Fatal error */
3414
    goto err;
unknown's avatar
unknown committed
3415 3416
  if (mi->gtid_pos_auto)
    reset_coordinates_for_gtid(mi, rli);
3417

3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429
  if (ret == 2) 
  { 
    if (check_io_slave_killed(mi->io_thd, mi, "Slave I/O thread killed"
                              "while calling get_master_version_and_clock(...)"))
      goto err;
    suppress_warnings= FALSE;
    /* Try to reconnect because the error was caused by a transient network problem */
    if (try_to_reconnect(thd, mysql, mi, &retry_count, suppress_warnings,
                             reconnect_messages[SLAVE_RECON_ACT_REG]))
      goto err;
    goto connected;
  } 
3430 3431

  if (mi->rli.relay_log.description_event_for_queue->binlog_version > 1)
3432
  {
unknown's avatar
unknown committed
3433 3434 3435
    /*
      Register ourselves with the master.
    */
3436
    thd_proc_info(thd, "Registering slave on master");
3437
    if (register_slave_on_master(mysql, mi, &suppress_warnings))
3438
    {
3439 3440 3441 3442 3443 3444 3445 3446 3447
      if (!check_io_slave_killed(thd, mi, "Slave I/O thread killed "
                                "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
3448 3449
        goto err;
      goto connected;
3450
    }
3451 3452 3453 3454 3455 3456 3457 3458 3459 3460
    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;
      });
3461
  }
3462

3463
  DBUG_PRINT("info",("Starting reading binary log from master"));
3464
  while (!io_slave_killed(thd,mi))
unknown's avatar
unknown committed
3465
  {
3466
    thd_proc_info(thd, "Requesting binlog dump");
He Zhenxing's avatar
He Zhenxing committed
3467
    if (request_dump(thd, mysql, mi, &suppress_warnings))
unknown's avatar
unknown committed
3468 3469
    {
      sql_print_error("Failed on request_dump()");
3470 3471 3472 3473
      if (check_io_slave_killed(thd, mi, "Slave I/O thread killed while \
requesting master dump") ||
          try_to_reconnect(thd, mysql, mi, &retry_count, suppress_warnings,
                           reconnect_messages[SLAVE_RECON_ACT_DUMP]))
unknown's avatar
unknown committed
3474
        goto err;
unknown's avatar
unknown committed
3475 3476
      goto connected;
    }
3477 3478 3479 3480 3481 3482 3483 3484 3485 3486
    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
3487
    const char *event_buf;
unknown's avatar
unknown committed
3488

3489
    DBUG_ASSERT(mi->last_error().number == 0);
3490
    while (!io_slave_killed(thd,mi))
unknown's avatar
unknown committed
3491
    {
3492
      ulong event_len;
3493
      /*
3494
         We say "waiting" because read_event() will wait if there's nothing to
3495 3496 3497
         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.
3498
      */
3499
      thd_proc_info(thd, "Waiting for master to send event");
3500
      event_len= read_event(mysql, mi, &suppress_warnings);
3501 3502
      if (check_io_slave_killed(thd, mi, "Slave I/O thread killed while \
reading event"))
unknown's avatar
unknown committed
3503
        goto err;
3504 3505 3506 3507 3508 3509 3510 3511 3512 3513
      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;
        });
3514

unknown's avatar
unknown committed
3515 3516
      if (event_len == packet_error)
      {
unknown's avatar
unknown committed
3517
        uint mysql_error_number= mysql_errno(mysql);
3518 3519
        switch (mysql_error_number) {
        case CR_NET_PACKET_TOO_LARGE:
unknown's avatar
unknown committed
3520
          sql_print_error("\
3521
Log entry on master is longer than slave_max_allowed_packet (%lu) on \
3522
slave. If the entry is correct, restart the server with a higher value of \
3523 3524
slave_max_allowed_packet",
                         slave_max_allowed_packet);
3525
          mi->report(ERROR_LEVEL, ER_NET_PACKET_TOO_LARGE,
3526
                     "%s", "Got a packet bigger than 'slave_max_allowed_packet' bytes");
unknown's avatar
unknown committed
3527
          goto err;
3528
        case ER_MASTER_FATAL_ERROR_READING_BINLOG:
3529 3530 3531
          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
3532
          goto err;
3533
        case ER_OUT_OF_RESOURCES:
3534 3535
          sql_print_error("\
Stopping slave I/O thread due to out-of-memory error from master");
3536
          mi->report(ERROR_LEVEL, ER_OUT_OF_RESOURCES,
3537
                     "%s", ER(ER_OUT_OF_RESOURCES));
unknown's avatar
unknown committed
3538 3539
          goto err;
        }
3540 3541
        if (try_to_reconnect(thd, mysql, mi, &retry_count, suppress_warnings,
                             reconnect_messages[SLAVE_RECON_ACT_EVENT]))
unknown's avatar
unknown committed
3542 3543
          goto err;
        goto connected;
unknown's avatar
unknown committed
3544
      } // if (event_len == packet_error)
3545

unknown's avatar
unknown committed
3546
      retry_count=0;                    // ok event, reset retry counter
3547
      thd_proc_info(thd, "Queueing master event to the relay log");
He Zhenxing's avatar
He Zhenxing committed
3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562
      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
3563
      {
3564 3565 3566
        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
3567
        goto err;
unknown's avatar
unknown committed
3568
      }
He Zhenxing's avatar
He Zhenxing committed
3569 3570 3571 3572 3573 3574 3575 3576 3577 3578

      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;
      }

3579
      if (flush_master_info(mi, TRUE, TRUE))
3580 3581 3582 3583
      {
        sql_print_error("Failed to flush master info file");
        goto err;
      }
3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595
      /*
        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.
        - 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.
      */
3596 3597 3598 3599 3600
#ifndef DBUG_OFF
      {
        char llbuf1[22], llbuf2[22];
        DBUG_PRINT("info", ("log_space_limit=%s log_space_total=%s \
ignore_log_space_limit=%d",
3601 3602
                            llstr(rli->log_space_limit,llbuf1),
                            llstr(rli->log_space_total,llbuf2),
unknown's avatar
unknown committed
3603
                            (int) rli->ignore_log_space_limit));
3604 3605 3606
      }
#endif

3607
      if (rli->log_space_limit && rli->log_space_limit <
unknown's avatar
unknown committed
3608
          rli->log_space_total &&
3609
          !rli->ignore_log_space_limit)
unknown's avatar
unknown committed
3610 3611 3612
        if (wait_for_relay_log_space(rli))
        {
          sql_print_error("Slave I/O thread aborted while waiting for relay \
unknown's avatar
unknown committed
3613
log space");
unknown's avatar
unknown committed
3614 3615 3616
          goto err;
        }
    }
3617
  }
unknown's avatar
unknown committed
3618

unknown's avatar
unknown committed
3619
  // error = 0;
unknown's avatar
unknown committed
3620
err:
3621
  // print the current replication position
3622
  sql_print_information("Slave I/O thread exiting, read up to log '%s', position %s",
unknown's avatar
unknown committed
3623
                  IO_RPL_LOG_NAME, llstr(mi->master_log_pos,llbuff));
He Zhenxing's avatar
He Zhenxing committed
3624
  RUN_HOOK(binlog_relay_io, thread_stop, (thd, mi));
3625
  thd->reset_query();
unknown's avatar
unknown committed
3626
  thd->reset_db(NULL, 0);
unknown's avatar
unknown committed
3627 3628
  if (mysql)
  {
3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639
    /*
      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
3640
    mysql_close(mysql);
unknown's avatar
unknown committed
3641 3642
    mi->mysql=0;
  }
3643
  write_ignored_events_info_to_relay_log(thd, mi);
3644
  thd_proc_info(thd, "Waiting for slave mutex on exit");
Marc Alff's avatar
Marc Alff committed
3645
  mysql_mutex_lock(&mi->run_lock);
3646

3647
err_during_init:
3648 3649 3650
  /* Forget the relay log's format */
  delete mi->rli.relay_log.description_event_for_queue;
  mi->rli.relay_log.description_event_for_queue= 0;
3651
  // TODO: make rpl_status part of Master_info
3652
  change_rpl_status(RPL_ACTIVE_SLAVE,RPL_IDLE_SLAVE);
Marc Alff's avatar
Marc Alff committed
3653
  mysql_mutex_lock(&LOCK_thread_count);
3654
  THD_CHECK_SENTRY(thd);
unknown's avatar
unknown committed
3655
  delete thd;
Marc Alff's avatar
Marc Alff committed
3656
  mysql_mutex_unlock(&LOCK_thread_count);
unknown's avatar
unknown committed
3657
  mi->abort_slave= 0;
3658
  mi->slave_running= MYSQL_SLAVE_NOT_RUN;
unknown's avatar
unknown committed
3659
  mi->io_thd= 0;
3660 3661 3662 3663 3664
  /*
    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
3665
  mysql_cond_broadcast(&mi->stop_cond);       // tell the world we are done
3666
  DBUG_EXECUTE_IF("simulate_slave_delay_at_terminate_bug38694", sleep(5););
Marc Alff's avatar
Marc Alff committed
3667
  mysql_mutex_unlock(&mi->run_lock);
3668 3669

  DBUG_LEAVE;                                   // Must match DBUG_ENTER()
3670
  my_thread_end();
unknown's avatar
unknown committed
3671
  pthread_exit(0);
3672
  return 0;                                     // Avoid compiler warnings
unknown's avatar
unknown committed
3673 3674
}

3675 3676 3677 3678 3679
/*
  Check the temporary directory used by commands like
  LOAD DATA INFILE.
 */
static 
3680
int check_temp_dir(char* tmp_file)
3681 3682 3683
{
  int fd;
  MY_DIR *dirp;
3684 3685
  char tmp_dir[FN_REFLEN];
  size_t tmp_dir_size;
3686 3687 3688

  DBUG_ENTER("check_temp_dir");

3689 3690 3691 3692 3693
  /*
    Get the directory from the temporary file.
  */
  dirname_part(tmp_dir, tmp_file, &tmp_dir_size);

3694 3695 3696 3697 3698 3699 3700 3701 3702 3703
  /*
    Check if the directory exists.
   */
  if (!(dirp=my_dir(tmp_dir,MYF(MY_WME))))
    DBUG_RETURN(1);
  my_dirend(dirp);

  /*
    Check permissions to create a file.
   */
Marc Alff's avatar
Marc Alff committed
3704 3705 3706 3707
  if ((fd= mysql_file_create(key_file_misc,
                             tmp_file, CREATE_MODE,
                             O_WRONLY | O_BINARY | O_EXCL | O_NOFOLLOW,
                             MYF(MY_WME))) < 0)
3708 3709 3710 3711 3712
  DBUG_RETURN(1);

  /*
    Clean up.
   */
Marc Alff's avatar
Marc Alff committed
3713 3714
  mysql_file_close(fd, MYF(0));
  mysql_file_delete(key_file_misc, tmp_file, MYF(0));
3715 3716 3717

  DBUG_RETURN(0);
}
unknown's avatar
unknown committed
3718

3719 3720 3721 3722 3723
/**
  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
3724

3725 3726
  @return Always 0.
*/
3727
pthread_handler_t handle_slave_sql(void *arg)
3728
{
unknown's avatar
unknown committed
3729
  THD *thd;                     /* needs to be first for thread_stack */
3730
  char llbuff[22],llbuff1[22];
3731 3732
  char saved_log_name[FN_REFLEN];
  char saved_master_log_name[FN_REFLEN];
3733 3734
  my_off_t UNINIT_VAR(saved_log_pos);
  my_off_t UNINIT_VAR(saved_master_log_pos);
3735
  my_off_t saved_skip= 0;
3736 3737
  Master_info *mi= ((Master_info*)arg);
  Relay_log_info* rli = &mi->rli;
unknown's avatar
unknown committed
3738 3739 3740 3741
  const char *errmsg;

  // needs to call my_thread_init(), otherwise we get a coredump in DBUG_ stuff
  my_thread_init();
3742
  DBUG_ENTER("handle_slave_sql");
unknown's avatar
unknown committed
3743

Michael Widenius's avatar
Michael Widenius committed
3744 3745
  LINT_INIT(saved_master_log_pos);
  LINT_INIT(saved_log_pos);
3746 3747 3748 3749

  thd = new THD; // note that contructor of THD uses DBUG_ !
  thd->thread_stack = (char*)&thd; // remember where our stack is

3750
  DBUG_ASSERT(rli->inited);
3751
  DBUG_ASSERT(rli->mi == mi);
Marc Alff's avatar
Marc Alff committed
3752
  mysql_mutex_lock(&rli->run_lock);
3753
  DBUG_ASSERT(!rli->slave_running);
unknown's avatar
unknown committed
3754
  errmsg= 0;
unknown's avatar
unknown committed
3755
#ifndef DBUG_OFF
3756
  rli->events_till_abort = abort_slave_event_count;
unknown's avatar
unknown committed
3757
#endif
3758

3759
  rli->sql_thd= thd;
3760
  
unknown's avatar
unknown committed
3761 3762
  /* Inform waiting threads that slave has started */
  rli->slave_run_id++;
3763
  rli->slave_running= MYSQL_SLAVE_RUN_NOT_CONNECT;
unknown's avatar
unknown committed
3764

3765
  pthread_detach_this_thread();
3766
  if (init_slave_thread(thd, mi, SLAVE_THD_SQL))
unknown's avatar
unknown committed
3767 3768 3769 3770 3771
  {
    /*
      TODO: this is currently broken - slave start and change master
      will be stuck if we fail here
    */
Marc Alff's avatar
Marc Alff committed
3772
    mysql_cond_broadcast(&rli->start_cond);
3773 3774
    rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, 
                "Failed during slave thread initialization");
3775
    goto err_during_init;
unknown's avatar
unknown committed
3776
  }
3777
  thd->init_for_queries();
3778 3779 3780 3781 3782 3783
  thd->rli_slave= rli;
  if ((rli->deferred_events_collecting= rpl_filter->is_on()))
  {
    rli->deferred_events= new Deferred_log_events(rli);
  }

3784
  thd->temporary_tables = rli->save_temporary_tables; // restore temp tables
3785
  set_thd_in_use_temporary_tables(rli);   // (re)set sql_thd in use for saved temp tables
3786
  /*
3787
    binlog_annotate_row_events must be TRUE only after an Annotate_rows event
3788 3789 3790
    has been recieved and only till the last corresponding rbr event has been
    applied. In all other cases it must be FALSE.
  */
3791
  thd->variables.binlog_annotate_row_events= 0;
Marc Alff's avatar
Marc Alff committed
3792
  mysql_mutex_lock(&LOCK_thread_count);
3793
  threads.append(thd);
Marc Alff's avatar
Marc Alff committed
3794
  mysql_mutex_unlock(&LOCK_thread_count);
3795 3796 3797 3798 3799 3800 3801 3802
  /*
    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.
  */
3803
  rli->abort_slave = 0;
Marc Alff's avatar
Marc Alff committed
3804 3805
  mysql_mutex_unlock(&rli->run_lock);
  mysql_cond_broadcast(&rli->start_cond);
3806

unknown's avatar
unknown committed
3807 3808 3809
  /*
    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
3810
    though there's no problem anymore). Do not reset the master timestamp
3811 3812 3813 3814
    (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
3815
    But the master timestamp is reset by RESET SLAVE & CHANGE MASTER.
unknown's avatar
unknown committed
3816
  */
3817
  rli->clear_error();
3818 3819

  //tell the I/O thread to take relay_log_space_limit into account from now on
Marc Alff's avatar
Marc Alff committed
3820
  mysql_mutex_lock(&rli->log_space_lock);
3821
  rli->ignore_log_space_limit= 0;
Marc Alff's avatar
Marc Alff committed
3822
  mysql_mutex_unlock(&rli->log_space_lock);
3823
  rli->trans_retries= 0; // start from "no error"
unknown's avatar
unknown committed
3824
  DBUG_PRINT("info", ("rli->trans_retries: %lu", rli->trans_retries));
3825

3826
  if (init_relay_log_pos(rli,
unknown's avatar
unknown committed
3827 3828 3829
                         rli->group_relay_log_name,
                         rli->group_relay_log_pos,
                         1 /*need data lock*/, &errmsg,
3830
                         1 /*look for a description_event*/))
3831 3832 3833
  { 
    rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, 
                "Error initializing relay log position: %s", errmsg);
3834 3835
    goto err;
  }
3836
  THD_CHECK_SENTRY(thd);
3837 3838 3839 3840
#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
3841
                        llstr(my_b_tell(rli->cur_log),llbuf1),
3842 3843 3844 3845 3846 3847 3848 3849 3850 3851 3852 3853 3854 3855 3856 3857 3858 3859 3860
                        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
3861
  DBUG_ASSERT(rli->sql_thd == thd);
3862 3863

  DBUG_PRINT("master_info",("log_file_name: %s  position: %s",
unknown's avatar
unknown committed
3864 3865
                            rli->group_master_log_name,
                            llstr(rli->group_master_log_pos,llbuff)));
3866
  if (global_system_variables.log_warnings)
3867
    sql_print_information("Slave SQL thread initialized, starting replication in \
unknown's avatar
unknown committed
3868
log '%s' at position %s, relay log '%s' position: %s", RPL_LOG_NAME,
unknown's avatar
unknown committed
3869 3870
                    llstr(rli->group_master_log_pos,llbuff),rli->group_relay_log_name,
                    llstr(rli->group_relay_log_pos,llbuff1));
3871

3872
  if (check_temp_dir(rli->slave_patternload_file))
3873
  {
Marc Alff's avatar
Marc Alff committed
3874
    rli->report(ERROR_LEVEL, thd->stmt_da->sql_errno(), 
3875
                "Unable to use slave's temporary directory %s - %s", 
Marc Alff's avatar
Marc Alff committed
3876
                slave_load_tmpdir, thd->stmt_da->message());
3877 3878 3879
    goto err;
  }

3880 3881 3882 3883 3884 3885 3886 3887 3888
  /* 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());
    goto err;
  }

unknown's avatar
unknown committed
3889
  /* execute init_slave variable */
3890
  if (opt_init_slave.length)
unknown's avatar
unknown committed
3891
  {
3892
    execute_init_command(thd, &opt_init_slave, &LOCK_sys_init_slave);
3893
    if (thd->is_slave_error)
unknown's avatar
unknown committed
3894
    {
Marc Alff's avatar
Marc Alff committed
3895
      rli->report(ERROR_LEVEL, thd->stmt_da->sql_errno(),
3896
                  "Slave SQL thread aborted. Can't execute init_slave query");
unknown's avatar
unknown committed
3897 3898 3899 3900
      goto err;
    }
  }

3901 3902 3903 3904
  /*
    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
3905
  mysql_mutex_lock(&rli->data_lock);
3906 3907
  if (rli->slave_skip_counter)
  {
3908 3909
    strmake(saved_log_name, rli->group_relay_log_name, FN_REFLEN - 1);
    strmake(saved_master_log_name, rli->group_master_log_name, FN_REFLEN - 1);
3910 3911 3912 3913
    saved_log_pos= rli->group_relay_log_pos;
    saved_master_log_pos= rli->group_master_log_pos;
    saved_skip= rli->slave_skip_counter;
  }
3914
  if (rli->until_condition != Relay_log_info::UNTIL_NONE &&
3915
      rli->is_until_satisfied(thd, NULL))
3916 3917 3918 3919
  {
    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
3920
    mysql_mutex_unlock(&rli->data_lock);
3921 3922
    goto err;
  }
Marc Alff's avatar
Marc Alff committed
3923
  mysql_mutex_unlock(&rli->data_lock);
3924

3925 3926
  /* Read queries from the IO/THREAD until this thread is killed */

3927
  while (!sql_slave_killed(thd,rli))
3928
  {
3929
    thd_proc_info(thd, "Reading event from the relay log");
3930
    DBUG_ASSERT(rli->sql_thd == thd);
3931
    THD_CHECK_SENTRY(thd);
3932 3933 3934 3935 3936 3937 3938 3939 3940 3941 3942 3943 3944 3945 3946

    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;
    }
    
3947 3948
    if (exec_relay_log_event(thd,rli))
    {
3949
      DBUG_PRINT("info", ("exec_relay_log_event() failed"));
3950
      // do not scare the user if SQL thread was simply killed or stopped
3951
      if (!sql_slave_killed(thd,rli))
unknown's avatar
unknown committed
3952 3953
      {
        /*
3954 3955 3956
          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
3957
        */
3958
        uint32 const last_errno= rli->last_error().number;
3959

3960
        if (thd->is_error())
unknown's avatar
unknown committed
3961
        {
Marc Alff's avatar
Marc Alff committed
3962
          char const *const errmsg= thd->stmt_da->message();
3963 3964

          DBUG_PRINT("info",
Marc Alff's avatar
Marc Alff committed
3965 3966
                     ("thd->stmt_da->sql_errno()=%d; rli->last_error.number=%d",
                      thd->stmt_da->sql_errno(), last_errno));
3967
          if (last_errno == 0)
unknown's avatar
unknown committed
3968
          {
3969 3970 3971 3972
            /*
 	      This function is reporting an error which was not reported
 	      while executing exec_relay_log_event().
 	    */ 
3973
            rli->report(ERROR_LEVEL, thd->stmt_da->sql_errno(), "%s", errmsg);
unknown's avatar
unknown committed
3974
          }
Marc Alff's avatar
Marc Alff committed
3975
          else if (last_errno != thd->stmt_da->sql_errno())
unknown's avatar
unknown committed
3976
          {
3977 3978 3979 3980 3981 3982
            /*
             * 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
3983
            sql_print_error("Slave (additional info): %s Error_code: %d",
Marc Alff's avatar
Marc Alff committed
3984
                            errmsg, thd->stmt_da->sql_errno());
unknown's avatar
unknown committed
3985 3986 3987 3988
          }
        }

        /* Print any warnings issued */
Marc Alff's avatar
Marc Alff committed
3989
        List_iterator_fast<MYSQL_ERROR> it(thd->warning_info->warn_list());
unknown's avatar
unknown committed
3990
        MYSQL_ERROR *err;
3991 3992 3993 3994 3995
        /*
          Added controlled slave thread cancel for replication
          of user-defined variables.
        */
        bool udf_error = false;
unknown's avatar
unknown committed
3996
        while ((err= it++))
3997
        {
Marc Alff's avatar
Marc Alff committed
3998
          if (err->get_sql_errno() == ER_CANT_OPEN_LIBRARY)
3999
            udf_error = true;
Marc Alff's avatar
Marc Alff committed
4000
          sql_print_warning("Slave: %s Error_code: %d", err->get_message_text(), err->get_sql_errno());
4001 4002 4003 4004 4005 4006 4007 4008 4009
        }
        if (udf_error)
          sql_print_error("Error loading user-defined library, slave SQL "
            "thread aborted. Install the missing library, and restart the "
            "slave SQL thread with \"SLAVE START\". We stopped at log '%s' "
            "position %s", RPL_LOG_NAME, llstr(rli->group_master_log_pos, 
            llbuff));
        else
          sql_print_error("\
4010
Error running query, slave SQL thread aborted. Fix the problem, and restart \
unknown's avatar
unknown committed
4011
the slave SQL thread with \"SLAVE START\". We stopped at log \
4012
'%s' position %s", RPL_LOG_NAME, llstr(rli->group_master_log_pos, llbuff));
unknown's avatar
unknown committed
4013
      }
4014 4015
      goto err;
    }
4016
    rli->executed_entries++;
4017
  }
4018

4019
  /* Thread stopped. Print the current replication position to the log */
unknown's avatar
unknown committed
4020
  sql_print_information("Slave SQL thread exiting, replication stopped in log "
unknown's avatar
unknown committed
4021 4022
                        "'%s' at position %s",
                        RPL_LOG_NAME, llstr(rli->group_master_log_pos,llbuff));
4023 4024

 err:
4025 4026 4027 4028 4029 4030 4031

  /*
    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:
  */
4032
  thd->clear_error();
4033
  rli->cleanup_context(thd, 1);
4034 4035 4036 4037 4038
  /*
    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)).
  */
4039
  thd->catalog= 0;
4040
  thd->reset_query();
unknown's avatar
unknown committed
4041
  thd->reset_db(NULL, 0);
4042
  thd_proc_info(thd, "Waiting for slave mutex on exit");
Marc Alff's avatar
Marc Alff committed
4043
  mysql_mutex_lock(&rli->run_lock);
4044
err_during_init:
4045
  /* We need data_lock, at least to wake up any waiting master_pos_wait() */
Marc Alff's avatar
Marc Alff committed
4046
  mysql_mutex_lock(&rli->data_lock);
4047
  DBUG_ASSERT(rli->slave_running == MYSQL_SLAVE_RUN_NOT_CONNECT); // tracking buffer overrun
4048
  /* When master_pos_wait() wakes up it will check this and terminate */
4049
  rli->slave_running= MYSQL_SLAVE_NOT_RUN;
4050 4051 4052
  /* Forget the relay log's format */
  delete rli->relay_log.description_event_for_exec;
  rli->relay_log.description_event_for_exec= 0;
4053
  /* Wake up master_pos_wait() */
Marc Alff's avatar
Marc Alff committed
4054
  mysql_mutex_unlock(&rli->data_lock);
4055
  DBUG_PRINT("info",("Signaling possibly waiting master_pos_wait() functions"));
Marc Alff's avatar
Marc Alff committed
4056
  mysql_cond_broadcast(&rli->data_cond);
unknown's avatar
unknown committed
4057
  rli->ignore_log_space_limit= 0; /* don't need any lock */
4058 4059
  /* we die so won't remember charset - re-update them on next thread start */
  rli->cached_charset_invalidate();
4060
  rli->save_temporary_tables = thd->temporary_tables;
unknown's avatar
unknown committed
4061 4062 4063 4064 4065

  /*
    TODO: see if we can do this conditionally in next_event() instead
    to avoid unneeded position re-init
  */
4066 4067
  thd->temporary_tables = 0; // remove tempation from destructor to close them
  DBUG_ASSERT(rli->sql_thd == thd);
4068
  THD_CHECK_SENTRY(thd);
4069
  rli->sql_thd= 0;
4070
  set_thd_in_use_temporary_tables(rli);  // (re)set sql_thd in use for saved temp tables
Marc Alff's avatar
Marc Alff committed
4071
  mysql_mutex_lock(&LOCK_thread_count);
4072
  THD_CHECK_SENTRY(thd);
4073
  delete thd;
Marc Alff's avatar
Marc Alff committed
4074
  mysql_mutex_unlock(&LOCK_thread_count);
4075 4076 4077 4078 4079
 /*
  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
4080
  mysql_cond_broadcast(&rli->stop_cond);
4081
  DBUG_EXECUTE_IF("simulate_slave_delay_at_terminate_bug38694", sleep(5););
Marc Alff's avatar
Marc Alff committed
4082
  mysql_mutex_unlock(&rli->run_lock);  // tell the world we are done
4083 4084

  DBUG_LEAVE;                                   // Must match DBUG_ENTER()
unknown's avatar
unknown committed
4085
  my_thread_end();
4086
  pthread_exit(0);
4087
  return 0;                                     // Avoid compiler warnings
4088
}
unknown's avatar
unknown committed
4089

4090

unknown's avatar
unknown committed
4091
/*
4092
  process_io_create_file()
unknown's avatar
unknown committed
4093
*/
4094

4095
static int process_io_create_file(Master_info* mi, Create_file_log_event* cev)
unknown's avatar
unknown committed
4096 4097 4098 4099
{
  int error = 1;
  ulong num_bytes;
  bool cev_not_written;
4100 4101
  THD *thd = mi->io_thd;
  NET *net = &mi->mysql->net;
unknown's avatar
unknown committed
4102
  DBUG_ENTER("process_io_create_file");
unknown's avatar
unknown committed
4103 4104

  if (unlikely(!cev->is_valid()))
unknown's avatar
unknown committed
4105
    DBUG_RETURN(1);
4106 4107

  if (!rpl_filter->db_ok(cev->db))
unknown's avatar
unknown committed
4108 4109
  {
    skip_load_data_infile(net);
unknown's avatar
unknown committed
4110
    DBUG_RETURN(0);
unknown's avatar
unknown committed
4111 4112 4113
  }
  DBUG_ASSERT(cev->inited_from_old);
  thd->file_id = cev->file_id = mi->file_id++;
4114
  thd->variables.server_id = cev->server_id;
unknown's avatar
unknown committed
4115
  cev_not_written = 1;
unknown's avatar
unknown committed
4116

unknown's avatar
unknown committed
4117 4118 4119
  if (unlikely(net_request_file(net,cev->fname)))
  {
    sql_print_error("Slave I/O: failed requesting download of '%s'",
unknown's avatar
unknown committed
4120
                    cev->fname);
unknown's avatar
unknown committed
4121 4122 4123
    goto err;
  }

unknown's avatar
unknown committed
4124 4125 4126 4127
  /*
    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
4128 4129
  */
  {
4130
    Append_block_log_event aev(thd,0,0,0,0);
unknown's avatar
unknown committed
4131

unknown's avatar
unknown committed
4132 4133 4134 4135
    for (;;)
    {
      if (unlikely((num_bytes=my_net_read(net)) == packet_error))
      {
unknown's avatar
unknown committed
4136 4137 4138
        sql_print_error("Network read error downloading '%s' from master",
                        cev->fname);
        goto err;
unknown's avatar
unknown committed
4139 4140 4141
      }
      if (unlikely(!num_bytes)) /* eof */
      {
4142 4143
	/* 3.23 master wants it */
        net_write_command(net, 0, (uchar*) "", 0, (uchar*) "", 0);
unknown's avatar
unknown committed
4144 4145 4146 4147 4148 4149 4150
        /*
          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
4151 4152 4153 4154 4155
          break;
        Execute_load_log_event xev(thd,0,0);
        xev.log_pos = cev->log_pos;
        if (unlikely(mi->rli.relay_log.append(&xev)))
        {
4156 4157 4158
          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
4159 4160 4161 4162
          goto err;
        }
        mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total);
        break;
unknown's avatar
unknown committed
4163 4164 4165
      }
      if (unlikely(cev_not_written))
      {
unknown's avatar
unknown committed
4166
        cev->block = net->read_pos;
unknown's avatar
unknown committed
4167 4168 4169
        cev->block_len = num_bytes;
        if (unlikely(mi->rli.relay_log.append(cev)))
        {
4170 4171 4172
          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
4173 4174 4175 4176
          goto err;
        }
        cev_not_written=0;
        mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total);
unknown's avatar
unknown committed
4177 4178 4179
      }
      else
      {
unknown's avatar
unknown committed
4180
        aev.block = net->read_pos;
unknown's avatar
unknown committed
4181 4182 4183 4184
        aev.block_len = num_bytes;
        aev.log_pos = cev->log_pos;
        if (unlikely(mi->rli.relay_log.append(&aev)))
        {
4185 4186 4187
          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
4188 4189 4190
          goto err;
        }
        mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total) ;
unknown's avatar
unknown committed
4191 4192 4193 4194 4195
      }
    }
  }
  error=0;
err:
unknown's avatar
unknown committed
4196
  DBUG_RETURN(error);
unknown's avatar
unknown committed
4197
}
unknown's avatar
unknown committed
4198

4199

unknown's avatar
unknown committed
4200
/*
unknown's avatar
unknown committed
4201 4202 4203 4204
  Start using a new binary log on the master

  SYNOPSIS
    process_io_rotate()
unknown's avatar
unknown committed
4205 4206
    mi                  master_info for the slave
    rev                 The rotate log event read from the binary log
unknown's avatar
unknown committed
4207 4208

  DESCRIPTION
4209
    Updates the master info with the place in the next binary
unknown's avatar
unknown committed
4210
    log where we should start reading.
4211
    Rotate the relay log to avoid mixed-format relay logs.
unknown's avatar
unknown committed
4212 4213 4214 4215 4216

  NOTES
    We assume we already locked mi->data_lock

  RETURN VALUES
unknown's avatar
unknown committed
4217 4218
    0           ok
    1           Log event is illegal
unknown's avatar
unknown committed
4219 4220 4221

*/

4222
static int process_io_rotate(Master_info *mi, Rotate_log_event *rev)
4223
{
4224
  DBUG_ENTER("process_io_rotate");
Marc Alff's avatar
Marc Alff committed
4225
  mysql_mutex_assert_owner(&mi->data_lock);
4226

unknown's avatar
unknown committed
4227
  if (unlikely(!rev->is_valid()))
4228
    DBUG_RETURN(1);
unknown's avatar
unknown committed
4229

4230
  /* Safe copy as 'rev' has been "sanitized" in Rotate_log_event's ctor */
unknown's avatar
unknown committed
4231 4232
  memcpy(mi->master_log_name, rev->new_log_ident, rev->ident_len+1);
  mi->master_log_pos= rev->pos;
unknown's avatar
unknown committed
4233
  DBUG_PRINT("info", ("master_log_pos: '%s' %lu",
unknown's avatar
unknown committed
4234
                      mi->master_log_name, (ulong) mi->master_log_pos));
4235
#ifndef DBUG_OFF
unknown's avatar
unknown committed
4236 4237 4238 4239 4240
  /*
    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)
4241
    mi->events_till_disconnect++;
4242
#endif
4243

4244 4245 4246 4247 4248 4249 4250 4251 4252
  /*
    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)
  {
4253 4254 4255
    DBUG_ASSERT(mi->rli.relay_log.description_event_for_queue->checksum_alg ==
                mi->rli.relay_log.relay_log_checksum_alg);
    
4256 4257 4258 4259
    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);
4260 4261
    mi->rli.relay_log.description_event_for_queue->checksum_alg=
      mi->rli.relay_log.relay_log_checksum_alg;    
4262
  }
4263 4264 4265 4266
  /*
    Rotate the relay log makes binlog format detection easier (at next slave
    start or mysqlbinlog)
  */
4267
  DBUG_RETURN(rotate_relay_log(mi) /* will take the right mutexes */);
4268 4269
}

unknown's avatar
unknown committed
4270
/*
4271 4272
  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
4273
*/
4274
static int queue_binlog_ver_1_event(Master_info *mi, const char *buf,
unknown's avatar
unknown committed
4275
                           ulong event_len)
4276
{
unknown's avatar
unknown committed
4277
  const char *errmsg = 0;
unknown's avatar
unknown committed
4278 4279 4280
  ulong inc_pos;
  bool ignore_event= 0;
  char *tmp_buf = 0;
4281
  Relay_log_info *rli= &mi->rli;
4282
  DBUG_ENTER("queue_binlog_ver_1_event");
unknown's avatar
unknown committed
4283

unknown's avatar
unknown committed
4284 4285 4286
  /*
    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
4287
  */
4288
  if ((uchar)buf[EVENT_TYPE_OFFSET] == LOAD_EVENT)
unknown's avatar
unknown committed
4289 4290 4291
  {
    if (unlikely(!(tmp_buf=(char*)my_malloc(event_len+1,MYF(MY_WME)))))
    {
4292 4293
      mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
                 ER(ER_SLAVE_FATAL_ERROR), "Memory allocation failed");
unknown's avatar
unknown committed
4294
      DBUG_RETURN(1);
unknown's avatar
unknown committed
4295 4296
    }
    memcpy(tmp_buf,buf,event_len);
4297 4298 4299 4300 4301 4302 4303 4304
    /*
      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
4305
    int4store(tmp_buf+EVENT_LEN_OFFSET, event_len);
unknown's avatar
unknown committed
4306 4307
    buf = (const char*)tmp_buf;
  }
4308 4309 4310 4311 4312 4313
  /*
    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).
  */
4314 4315 4316
  Log_event *ev=
    Log_event::read_log_event(buf, event_len, &errmsg,
                              mi->rli.relay_log.description_event_for_queue, 0);
4317
  if (unlikely(!ev))
4318 4319
  {
    sql_print_error("Read invalid event from master: '%s',\
unknown's avatar
unknown committed
4320
 master could be corrupt but a more likely cause of this is a bug",
unknown's avatar
unknown committed
4321
                    errmsg);
4322
    my_free(tmp_buf);
unknown's avatar
unknown committed
4323
    DBUG_RETURN(1);
4324
  }
4325

Marc Alff's avatar
Marc Alff committed
4326
  mysql_mutex_lock(&mi->data_lock);
4327
  ev->log_pos= mi->master_log_pos; /* 3.23 events don't contain log_pos */
unknown's avatar
unknown committed
4328
  switch (ev->get_type_code()) {
unknown's avatar
unknown committed
4329
  case STOP_EVENT:
4330
    ignore_event= 1;
unknown's avatar
unknown committed
4331 4332
    inc_pos= event_len;
    break;
4333
  case ROTATE_EVENT:
4334
    if (unlikely(process_io_rotate(mi,(Rotate_log_event*)ev)))
4335 4336
    {
      delete ev;
Marc Alff's avatar
Marc Alff committed
4337
      mysql_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
4338
      DBUG_RETURN(1);
4339
    }
unknown's avatar
unknown committed
4340
    inc_pos= 0;
4341
    break;
unknown's avatar
unknown committed
4342
  case CREATE_FILE_EVENT:
4343 4344 4345 4346 4347 4348
    /*
      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
4349
  {
unknown's avatar
unknown committed
4350
    /* We come here when and only when tmp_buf != 0 */
4351
    DBUG_ASSERT(tmp_buf != 0);
4352 4353
    inc_pos=event_len;
    ev->log_pos+= inc_pos;
unknown's avatar
unknown committed
4354
    int error = process_io_create_file(mi,(Create_file_log_event*)ev);
4355
    delete ev;
4356
    mi->master_log_pos += inc_pos;
unknown's avatar
unknown committed
4357
    DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
Marc Alff's avatar
Marc Alff committed
4358
    mysql_mutex_unlock(&mi->data_lock);
4359
    my_free(tmp_buf);
unknown's avatar
unknown committed
4360
    DBUG_RETURN(error);
unknown's avatar
unknown committed
4361
  }
4362
  default:
unknown's avatar
unknown committed
4363
    inc_pos= event_len;
4364 4365
    break;
  }
unknown's avatar
unknown committed
4366
  if (likely(!ignore_event))
4367
  {
unknown's avatar
unknown committed
4368 4369
    if (ev->log_pos)
      /*
4370 4371 4372 4373
         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
4374
    if (unlikely(rli->relay_log.append(ev)))
4375 4376
    {
      delete ev;
Marc Alff's avatar
Marc Alff committed
4377
      mysql_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
4378
      DBUG_RETURN(1);
4379
    }
unknown's avatar
unknown committed
4380
    rli->relay_log.harvest_bytes_written(&rli->log_space_total);
4381 4382
  }
  delete ev;
unknown's avatar
unknown committed
4383
  mi->master_log_pos+= inc_pos;
unknown's avatar
unknown committed
4384
  DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
Marc Alff's avatar
Marc Alff committed
4385
  mysql_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
4386
  DBUG_RETURN(0);
4387 4388
}

4389 4390 4391 4392
/*
  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.
*/
4393
static int queue_binlog_ver_3_event(Master_info *mi, const char *buf,
unknown's avatar
unknown committed
4394
                           ulong event_len)
4395 4396 4397 4398
{
  const char *errmsg = 0;
  ulong inc_pos;
  char *tmp_buf = 0;
4399
  Relay_log_info *rli= &mi->rli;
4400 4401 4402
  DBUG_ENTER("queue_binlog_ver_3_event");

  /* read_log_event() will adjust log_pos to be end_log_pos */
4403 4404 4405
  Log_event *ev=
    Log_event::read_log_event(buf,event_len, &errmsg,
                              mi->rli.relay_log.description_event_for_queue, 0);
4406 4407 4408 4409
  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
4410
                    errmsg);
4411
    my_free(tmp_buf);
4412 4413
    DBUG_RETURN(1);
  }
Marc Alff's avatar
Marc Alff committed
4414
  mysql_mutex_lock(&mi->data_lock);
4415 4416 4417 4418 4419 4420 4421
  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
4422
      mysql_mutex_unlock(&mi->data_lock);
4423 4424 4425 4426 4427 4428 4429 4430
      DBUG_RETURN(1);
    }
    inc_pos= 0;
    break;
  default:
    inc_pos= event_len;
    break;
  }
4431

4432 4433 4434
  if (unlikely(rli->relay_log.append(ev)))
  {
    delete ev;
Marc Alff's avatar
Marc Alff committed
4435
    mysql_mutex_unlock(&mi->data_lock);
4436 4437 4438 4439 4440 4441
    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
4442
  DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
Marc Alff's avatar
Marc Alff committed
4443
  mysql_mutex_unlock(&mi->data_lock);
4444 4445 4446 4447 4448 4449 4450 4451 4452 4453
  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
4454
  TODO:
4455 4456 4457 4458
    Test this code before release - it has to be tested on a separate
    setup with 3.23 master or 4.0 master
*/

4459
static int queue_old_event(Master_info *mi, const char *buf,
unknown's avatar
unknown committed
4460
                           ulong event_len)
4461
{
4462 4463
  DBUG_ENTER("queue_old_event");

4464 4465 4466
  switch (mi->rli.relay_log.description_event_for_queue->binlog_version)
  {
  case 1:
4467
      DBUG_RETURN(queue_binlog_ver_1_event(mi,buf,event_len));
4468
  case 3:
4469
      DBUG_RETURN(queue_binlog_ver_3_event(mi,buf,event_len));
4470 4471
  default: /* unsupported format; eg version 2 */
    DBUG_PRINT("info",("unsupported binlog format %d in queue_old_event()",
unknown's avatar
unknown committed
4472
                       mi->rli.relay_log.description_event_for_queue->binlog_version));
4473
    DBUG_RETURN(1);
4474 4475
  }
}
4476

unknown's avatar
unknown committed
4477
/*
4478 4479
  queue_event()

4480 4481 4482 4483 4484
  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
4485 4486
*/

4487
static int queue_event(Master_info* mi,const char* buf, ulong event_len)
4488
{
unknown's avatar
unknown committed
4489
  int error= 0;
Andrei Elkin's avatar
Andrei Elkin committed
4490
  String error_msg;
4491 4492
  ulonglong inc_pos;
  ulonglong event_pos;
4493
  Relay_log_info *rli= &mi->rli;
Marc Alff's avatar
Marc Alff committed
4494
  mysql_mutex_t *log_lock= rli->relay_log.get_log_lock();
Andrei Elkin's avatar
Andrei Elkin committed
4495
  ulong s_id;
4496 4497 4498 4499 4500 4501 4502 4503 4504 4505 4506 4507 4508 4509 4510 4511 4512 4513
  bool unlock_data_lock= TRUE;
  /*
    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
4514
  DBUG_ENTER("queue_event");
4515 4516 4517 4518 4519 4520 4521 4522 4523 4524 4525 4526 4527 4528 4529 4530 4531 4532 4533 4534 4535 4536 4537 4538 4539 4540 4541 4542 4543 4544 4545 4546 4547 4548 4549 4550 4551 4552 4553 4554 4555 4556 4557
  /*
    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
4558

unknown's avatar
unknown committed
4559 4560
  LINT_INIT(inc_pos);

4561
  if (mi->rli.relay_log.description_event_for_queue->binlog_version<4 &&
4562
      (uchar)buf[EVENT_TYPE_OFFSET] != FORMAT_DESCRIPTION_EVENT /* a way to escape */)
unknown's avatar
unknown committed
4563
    DBUG_RETURN(queue_old_event(mi,buf,event_len));
4564

Marc Alff's avatar
Marc Alff committed
4565
  mysql_mutex_lock(&mi->data_lock);
unknown's avatar
unknown committed
4566

4567
  switch ((uchar)buf[EVENT_TYPE_OFFSET]) {
4568
  case STOP_EVENT:
4569 4570
    /*
      We needn't write this event to the relay log. Indeed, it just indicates a
unknown's avatar
unknown committed
4571 4572 4573
      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
4574
      prepared statements' deletion are TODO only when we binlog prep stmts).
unknown's avatar
unknown committed
4575

unknown's avatar
unknown committed
4576 4577 4578 4579
      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).
4580 4581
    */
    goto err;
4582 4583
  case ROTATE_EVENT:
  {
4584 4585 4586 4587 4588
    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
4589
    {
Andrei Elkin's avatar
Andrei Elkin committed
4590
      error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
4591
      goto err;
unknown's avatar
unknown committed
4592
    }
4593 4594 4595 4596 4597 4598 4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613 4614 4615 4616 4617 4618 4619 4620 4621 4622 4623 4624 4625 4626 4627 4628 4629 4630 4631 4632 4633 4634 4635 4636 4637 4638 4639 4640 4641 4642 4643 4644
    /* 
       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;
      }
4645 4646 4647 4648
    /*
      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
4649
    inc_pos= 0;
4650 4651
    break;
  }
4652 4653 4654 4655 4656 4657 4658
  case FORMAT_DESCRIPTION_EVENT:
  {
    /*
      Create an event, and save it (when we rotate the relay log, we will have
      to write this event again).
    */
    /*
4659 4660 4661
      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).
4662
    */
4663
    Format_description_log_event* tmp;
4664
    const char* errmsg;
4665 4666
    // mark it as undefined that is irrelevant anymore
    mi->checksum_alg_before_fd= BINLOG_CHECKSUM_ALG_UNDEF;
4667
    if (!(tmp= (Format_description_log_event*)
4668
          Log_event::read_log_event(buf, event_len, &errmsg,
4669 4670
                                    mi->rli.relay_log.description_event_for_queue,
                                    1)))
4671
    {
Andrei Elkin's avatar
Andrei Elkin committed
4672
      error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
4673 4674
      goto err;
    }
4675 4676
    delete mi->rli.relay_log.description_event_for_queue;
    mi->rli.relay_log.description_event_for_queue= tmp;
4677 4678 4679 4680 4681 4682
    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
4683
    /*
4684
       Though this does some conversion to the slave's format, this will
unknown's avatar
unknown committed
4685
       preserve the master's binlog format version, and number of event types.
4686
    */
unknown's avatar
unknown committed
4687
    /*
4688
       If the event was not requested by the slave (the slave did not ask for
unknown's avatar
unknown committed
4689
       it), i.e. has end_log_pos=0, we do not increment mi->master_log_pos
4690 4691 4692
    */
    inc_pos= uint4korr(buf+LOG_POS_OFFSET) ? event_len : 0;
    DBUG_PRINT("info",("binlog format is now %d",
unknown's avatar
unknown committed
4693
                       mi->rli.relay_log.description_event_for_queue->binlog_version));
4694 4695 4696

  }
  break;
Andrei Elkin's avatar
Andrei Elkin committed
4697 4698 4699 4700 4701 4702 4703

  case HEARTBEAT_LOG_EVENT:
  {
    /*
      HB (heartbeat) cannot come before RL (Relay)
    */
    char  llbuf[22];
unknown's avatar
unknown committed
4704 4705 4706 4707 4708
    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
4709 4710 4711 4712 4713 4714 4715 4716 4717 4718 4719 4720 4721 4722 4723 4724 4725
    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
4726
       Slave can not have a higher coordinate except in the only
Andrei Elkin's avatar
Andrei Elkin committed
4727 4728 4729 4730
       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
4731 4732
       Slave can have lower coordinates, if some event from master was omitted.

Andrei Elkin's avatar
Andrei Elkin committed
4733 4734 4735 4736
       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
4737
        || mi->master_log_pos > hb.log_pos)
Andrei Elkin's avatar
Andrei Elkin committed
4738 4739 4740 4741 4742 4743 4744 4745 4746 4747 4748 4749 4750 4751
    {
      /* 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;
4752

4753
  default:
unknown's avatar
unknown committed
4754
    inc_pos= event_len;
4755 4756
    break;
  }
4757

4758
  /*
4759
    If we filter events master-side (eg. @@skip_replication), we will see holes
4760 4761 4762 4763 4764 4765 4766 4767 4768
    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
4769
    DBUG_PRINT("info", ("Adjust master_log_pos %llu->%llu to account for "
4770
                        "master-side filtering",
Sergei Golubchik's avatar
Sergei Golubchik committed
4771
                        mi->master_log_pos + inc_pos, event_pos));
4772 4773
  }

unknown's avatar
unknown committed
4774 4775
  /*
     If this event is originating from this server, don't queue it.
4776
     We don't check this for 3.23 events because it's simpler like this; 3.23
unknown's avatar
unknown committed
4777 4778
     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
4779 4780 4781 4782 4783 4784 4785 4786
     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
4787
  mysql_mutex_lock(log_lock);
Andrei Elkin's avatar
Andrei Elkin committed
4788
  s_id= uint4korr(buf + SERVER_ID_OFFSET);
4789 4790
  if ((s_id == global_system_variables.server_id &&
       !mi->rli.replicate_same_server_id) ||
Andrei Elkin's avatar
Andrei Elkin committed
4791 4792 4793 4794 4795 4796 4797 4798 4799 4800
      /*
        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 */
4801 4802
        (buf[EVENT_TYPE_OFFSET] != FORMAT_DESCRIPTION_EVENT &&
         buf[EVENT_TYPE_OFFSET] != ROTATE_EVENT))))
4803
  {
4804 4805
    /*
      Do not write it to the relay log.
4806 4807 4808 4809 4810 4811 4812 4813
      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.
4814 4815 4816
      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
4817 4818 4819
      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.
4820
    */
4821 4822
    if (!(s_id == global_system_variables.server_id &&
          !mi->rli.replicate_same_server_id) ||
4823 4824 4825
        (buf[EVENT_TYPE_OFFSET] != FORMAT_DESCRIPTION_EVENT &&
         buf[EVENT_TYPE_OFFSET] != ROTATE_EVENT &&
         buf[EVENT_TYPE_OFFSET] != STOP_EVENT))
unknown's avatar
unknown committed
4826
    {
4827
      mi->master_log_pos+= inc_pos;
unknown's avatar
unknown committed
4828 4829 4830 4831
      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;
    }
4832
    rli->relay_log.signal_update(); // the slave SQL thread needs to re-check
Andrei Elkin's avatar
Andrei Elkin committed
4833 4834
    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
4835
  }
unknown's avatar
unknown committed
4836 4837 4838
  else
  {
    /* write the event to the relay log */
4839
    if (likely(!(rli->relay_log.appendv(buf,event_len,0))))
4840 4841
    {
      mi->master_log_pos+= inc_pos;
unknown's avatar
unknown committed
4842
      DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
4843 4844
      rli->relay_log.harvest_bytes_written(&rli->log_space_total);
    }
4845
    else
Andrei Elkin's avatar
Andrei Elkin committed
4846 4847 4848
    {
      error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
    }
4849
    rli->ign_master_log_name_end[0]= 0; // last event is not ignored
4850 4851
    if (save_buf != NULL)
      buf= save_buf;
unknown's avatar
unknown committed
4852
  }
Marc Alff's avatar
Marc Alff committed
4853
  mysql_mutex_unlock(log_lock);
4854

Andrei Elkin's avatar
Andrei Elkin committed
4855 4856
skip_relay_logging:
  
4857
err:
4858
  if (unlock_data_lock)
Sergei Golubchik's avatar
Sergei Golubchik committed
4859
    mysql_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
4860
  DBUG_PRINT("info", ("error: %d", error));
Andrei Elkin's avatar
Andrei Elkin committed
4861 4862 4863 4864 4865
  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
4866
  DBUG_RETURN(error);
4867 4868
}

4869

4870
void end_relay_log_info(Relay_log_info* rli)
4871
{
4872 4873
  DBUG_ENTER("end_relay_log_info");

4874
  if (!rli->inited)
4875
    DBUG_VOID_RETURN;
4876
  if (rli->info_fd >= 0)
unknown's avatar
unknown committed
4877 4878
  {
    end_io_cache(&rli->info_file);
Marc Alff's avatar
Marc Alff committed
4879
    mysql_file_close(rli->info_fd, MYF(MY_WME));
unknown's avatar
unknown committed
4880 4881
    rli->info_fd = -1;
  }
4882
  if (rli->cur_log_fd >= 0)
unknown's avatar
unknown committed
4883 4884
  {
    end_io_cache(&rli->cache_buf);
Marc Alff's avatar
Marc Alff committed
4885
    mysql_file_close(rli->cur_log_fd, MYF(MY_WME));
unknown's avatar
unknown committed
4886 4887
    rli->cur_log_fd = -1;
  }
4888
  rli->inited = 0;
4889
  rli->relay_log.close(LOG_CLOSE_INDEX | LOG_CLOSE_STOP_EVENT);
4890
  rli->relay_log.harvest_bytes_written(&rli->log_space_total);
unknown's avatar
unknown committed
4891 4892 4893 4894 4895 4896
  /*
    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();
4897
  DBUG_VOID_RETURN;
4898 4899
}

4900 4901 4902 4903 4904 4905 4906 4907 4908 4909 4910 4911 4912 4913 4914 4915 4916 4917 4918

/**
  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;
4919
  if (thd && thd->slave_thread)
4920 4921 4922 4923 4924
    thd->clear_active_vio();
#endif
}


unknown's avatar
unknown committed
4925 4926
/*
  Try to connect until successful or slave killed
4927

unknown's avatar
unknown committed
4928 4929
  SYNPOSIS
    safe_connect()
unknown's avatar
unknown committed
4930 4931 4932
    thd                 Thread handler for slave
    mysql               MySQL connection handle
    mi                  Replication handle
4933

unknown's avatar
unknown committed
4934
  RETURN
unknown's avatar
unknown committed
4935 4936
    0   ok
    #   Error
unknown's avatar
unknown committed
4937
*/
4938

4939
static int safe_connect(THD* thd, MYSQL* mysql, Master_info* mi)
unknown's avatar
unknown committed
4940
{
4941 4942 4943
  DBUG_ENTER("safe_connect");

  DBUG_RETURN(connect_to_master(thd, mysql, mi, 0, 0));
unknown's avatar
unknown committed
4944 4945
}

4946

4947
/*
unknown's avatar
unknown committed
4948 4949
  SYNPOSIS
    connect_to_master()
unknown's avatar
unknown committed
4950

unknown's avatar
unknown committed
4951 4952 4953
  IMPLEMENTATION
    Try to connect until successful or slave killed or we have retried
    master_retry_count times
4954
*/
unknown's avatar
unknown committed
4955

4956
static int connect_to_master(THD* thd, MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
4957
                             bool reconnect, bool suppress_warnings)
unknown's avatar
unknown committed
4958
{
4959
  int slave_was_killed;
unknown's avatar
unknown committed
4960
  int last_errno= -2;                           // impossible error
4961
  ulong err_count=0;
unknown's avatar
unknown committed
4962
  char llbuff[22];
4963
  DBUG_ENTER("connect_to_master");
unknown's avatar
unknown committed
4964

unknown's avatar
unknown committed
4965
#ifndef DBUG_OFF
4966
  mi->events_till_disconnect = disconnect_slave_event_count;
unknown's avatar
unknown committed
4967
#endif
4968
  ulong client_flag= CLIENT_REMEMBER_OPTIONS;
4969
  if (opt_slave_compressed_protocol)
unknown's avatar
unknown committed
4970
    client_flag=CLIENT_COMPRESS;                /* We will use compression */
4971

4972 4973
  mysql_options(mysql, MYSQL_OPT_CONNECT_TIMEOUT, (char *) &slave_net_timeout);
  mysql_options(mysql, MYSQL_OPT_READ_TIMEOUT, (char *) &slave_net_timeout);
unknown's avatar
unknown committed
4974

unknown's avatar
unknown committed
4975 4976
#ifdef HAVE_OPENSSL
  if (mi->ssl)
4977
  {
unknown's avatar
unknown committed
4978
    mysql_ssl_set(mysql,
unknown's avatar
unknown committed
4979
                  mi->ssl_key[0]?mi->ssl_key:0,
unknown's avatar
unknown committed
4980
                  mi->ssl_cert[0]?mi->ssl_cert:0,
unknown's avatar
unknown committed
4981 4982 4983
                  mi->ssl_ca[0]?mi->ssl_ca:0,
                  mi->ssl_capath[0]?mi->ssl_capath:0,
                  mi->ssl_cipher[0]?mi->ssl_cipher:0);
4984 4985 4986
    mysql_options(mysql, MYSQL_OPT_SSL_VERIFY_SERVER_CERT,
                  &mi->ssl_verify_server_cert);
  }
unknown's avatar
unknown committed
4987 4988
#endif

4989 4990 4991 4992
  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);

4993 4994 4995 4996
  /* 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);

4997 4998 4999 5000 5001 5002 5003 5004 5005 5006
  /* 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);
  }
5007
  while (!(slave_was_killed = io_slave_killed(thd,mi)) &&
unknown's avatar
unknown committed
5008 5009 5010
         (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
5011
  {
5012
    /* Don't repeat last error */
unknown's avatar
SCRUM  
unknown committed
5013
    if ((int)mysql_errno(mysql) != last_errno)
5014
    {
unknown's avatar
SCRUM  
unknown committed
5015
      last_errno=mysql_errno(mysql);
unknown's avatar
unknown committed
5016
      suppress_warnings= 0;
5017 5018
      mi->report(ERROR_LEVEL, last_errno,
                 "error %s to master '%s@%s:%d'"
5019
                 " - retry-time: %d  retries: %lu  message: %s",
5020 5021
                 (reconnect ? "reconnecting" : "connecting"),
                 mi->user, mi->host, mi->port,
5022 5023
                 mi->connect_retry, master_retry_count,
                 mysql_error(mysql));
5024
    }
unknown's avatar
unknown committed
5025 5026 5027
    /*
      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
5028
      do not want to have election triggered on the first failure to
unknown's avatar
unknown committed
5029
      connect
5030
    */
5031
    if (++err_count == master_retry_count)
5032 5033
    {
      slave_was_killed=1;
unknown's avatar
unknown committed
5034 5035
      if (reconnect)
        change_rpl_status(RPL_ACTIVE_SLAVE,RPL_LOST_SOLDIER);
5036 5037
      break;
    }
5038
    slave_sleep(thd,mi->connect_retry,io_slave_killed, mi);
unknown's avatar
unknown committed
5039
  }
5040

5041 5042
  if (!slave_was_killed)
  {
5043
    mi->clear_error(); // clear possible left over reconnect error
unknown's avatar
unknown committed
5044
    if (reconnect)
unknown's avatar
unknown committed
5045
    {
5046
      if (!suppress_warnings && global_system_variables.log_warnings)
unknown's avatar
unknown committed
5047
        sql_print_information("Slave: connected to master '%s@%s:%d',\
5048
replication resumed in log '%s' at position %s", mi->user,
unknown's avatar
unknown committed
5049 5050 5051
                        mi->host, mi->port,
                        IO_RPL_LOG_NAME,
                        llstr(mi->master_log_pos,llbuff));
unknown's avatar
unknown committed
5052
    }
unknown's avatar
unknown committed
5053 5054 5055
    else
    {
      change_rpl_status(RPL_IDLE_SLAVE,RPL_ACTIVE_SLAVE);
5056 5057
      general_log_print(thd, COM_CONNECT_OUT, "%s@%s:%d",
                        mi->user, mi->host, mi->port);
unknown's avatar
unknown committed
5058
    }
5059
#ifdef SIGNAL_WITH_VIO_CLOSE
5060
    thd->set_active_vio(mysql->net.vio);
unknown's avatar
unknown committed
5061
#endif
5062
  }
5063
  mysql->reconnect= 1;
5064 5065
  DBUG_PRINT("exit",("slave_was_killed: %d", slave_was_killed));
  DBUG_RETURN(slave_was_killed);
unknown's avatar
unknown committed
5066 5067
}

5068

unknown's avatar
unknown committed
5069
/*
5070
  safe_reconnect()
unknown's avatar
unknown committed
5071

unknown's avatar
unknown committed
5072 5073 5074
  IMPLEMENTATION
    Try to connect until successful or slave killed or we have retried
    master_retry_count times
unknown's avatar
unknown committed
5075 5076
*/

5077
static int safe_reconnect(THD* thd, MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
5078
                          bool suppress_warnings)
unknown's avatar
unknown committed
5079
{
unknown's avatar
unknown committed
5080 5081
  DBUG_ENTER("safe_reconnect");
  DBUG_RETURN(connect_to_master(thd, mysql, mi, 1, suppress_warnings));
unknown's avatar
unknown committed
5082 5083
}

unknown's avatar
unknown committed
5084

He Zhenxing's avatar
He Zhenxing committed
5085 5086 5087 5088 5089 5090 5091 5092 5093 5094 5095 5096 5097 5098 5099 5100 5101 5102 5103 5104 5105 5106 5107 5108 5109 5110 5111 5112 5113 5114 5115 5116 5117 5118 5119 5120 5121 5122 5123 5124 5125 5126 5127 5128 5129 5130 5131 5132 5133 5134
MYSQL *rpl_connect_master(MYSQL *mysql)
{
  THD *thd= current_thd;
  Master_info *mi= my_pthread_getspecific_ptr(Master_info*, RPL_MASTER_INFO);
  if (!mi)
  {
    sql_print_error("'rpl_connect_master' must be called in slave I/O thread context.");
    return NULL;
  }

  bool allocated= false;
  
  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);

#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);

5135 5136 5137
  if (mi->user == NULL
      || mi->user[0] == 0
      || io_slave_killed(thd, mi)
He Zhenxing's avatar
He Zhenxing committed
5138 5139 5140 5141 5142 5143 5144 5145 5146 5147 5148 5149 5150 5151
      || !mysql_real_connect(mysql, mi->host, mi->user, mi->password, 0,
                             mi->port, 0, 0))
  {
    if (!io_slave_killed(thd, mi))
      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;
}

5152 5153 5154 5155 5156 5157
/*
  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
5158
    rli                 Relay log information
5159 5160

  NOTES
5161 5162 5163
    - 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.
5164 5165 5166 5167 5168 5169 5170 5171 5172 5173 5174 5175 5176 5177
    - 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
5178 5179
    0   ok
    1   write error
5180 5181
*/

5182
bool flush_relay_log_info(Relay_log_info* rli)
5183
{
5184
  bool error=0;
5185
  DBUG_ENTER("flush_relay_log_info");
5186 5187

  if (unlikely(rli->no_storage))
5188
    DBUG_RETURN(0);
5189

5190 5191 5192
  IO_CACHE *file = &rli->info_file;
  char buff[FN_REFLEN*2+22*2+4], *pos;

5193
  my_b_seek(file, 0L);
5194
  pos=strmov(buff, rli->group_relay_log_name);
5195
  *pos++='\n';
5196
  pos= longlong10_to_str(rli->group_relay_log_pos, pos, 10);
5197
  *pos++='\n';
5198
  pos=strmov(pos, rli->group_master_log_name);
5199
  *pos++='\n';
5200
  pos=longlong10_to_str(rli->group_master_log_pos, pos, 10);
5201
  *pos='\n';
5202
  if (my_b_write(file, (uchar*) buff, (size_t) (pos-buff)+1))
5203 5204 5205
    error=1;
  if (flush_io_cache(file))
    error=1;
5206 5207 5208 5209 5210 5211 5212 5213
  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;
  }
5214 5215 5216 5217
  /* 
    Flushing the relay log is done by the slave I/O thread 
    or by the user on STOP SLAVE. 
   */
5218
  DBUG_RETURN(error);
5219 5220
}

5221

unknown's avatar
unknown committed
5222
/*
5223
  Called when we notice that the current "hot" log got rotated under our feet.
unknown's avatar
unknown committed
5224 5225
*/

5226
static IO_CACHE *reopen_relay_log(Relay_log_info *rli, const char **errmsg)
5227
{
5228
  DBUG_ENTER("reopen_relay_log");
5229 5230
  DBUG_ASSERT(rli->cur_log != &rli->cache_buf);
  DBUG_ASSERT(rli->cur_log_fd == -1);
unknown's avatar
unknown committed
5231 5232

  IO_CACHE *cur_log = rli->cur_log=&rli->cache_buf;
5233
  if ((rli->cur_log_fd=open_binlog(cur_log,rli->event_relay_log_name,
unknown's avatar
unknown committed
5234
                                   errmsg)) <0)
unknown's avatar
unknown committed
5235
    DBUG_RETURN(0);
5236 5237
  /*
    We want to start exactly where we was before:
unknown's avatar
unknown committed
5238 5239
    relay_log_pos       Current log pos
    pending             Number of bytes already processed from the event
5240
  */
5241
  rli->event_relay_log_pos= max(rli->event_relay_log_pos, BIN_LOG_HEADER_SIZE);
5242
  my_b_seek(cur_log,rli->event_relay_log_pos);
unknown's avatar
unknown committed
5243
  DBUG_RETURN(cur_log);
5244 5245
}

unknown's avatar
unknown committed
5246

5247 5248 5249 5250 5251 5252 5253 5254 5255 5256
/**
  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.
*/
5257
static Log_event* next_event(Relay_log_info* rli)
5258 5259 5260
{
  Log_event* ev;
  IO_CACHE* cur_log = rli->cur_log;
Marc Alff's avatar
Marc Alff committed
5261
  mysql_mutex_t *log_lock = rli->relay_log.get_log_lock();
5262 5263
  const char* errmsg=0;
  THD* thd = rli->sql_thd;
unknown's avatar
unknown committed
5264
  DBUG_ENTER("next_event");
5265

5266 5267
  DBUG_ASSERT(thd != 0);

5268 5269 5270 5271 5272
#ifndef DBUG_OFF
  if (abort_slave_event_count && !rli->events_till_abort--)
    DBUG_RETURN(0);
#endif

unknown's avatar
unknown committed
5273 5274
  /*
    For most operations we need to protect rli members with data_lock,
5275 5276 5277
    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
5278
    mysql_cond_wait() with the non-data_lock mutex
unknown's avatar
unknown committed
5279
  */
Marc Alff's avatar
Marc Alff committed
5280
  mysql_mutex_assert_owner(&rli->data_lock);
unknown's avatar
unknown committed
5281

5282
  while (!sql_slave_killed(thd,rli))
unknown's avatar
unknown committed
5283 5284 5285
  {
    /*
      We can have two kinds of log reading:
unknown's avatar
unknown committed
5286 5287 5288 5289 5290 5291 5292 5293
      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
5294
    */
5295 5296 5297 5298
    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
5299
      mysql_mutex_lock(log_lock);
unknown's avatar
unknown committed
5300 5301

      /*
unknown's avatar
unknown committed
5302 5303
        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
5304
      */
unknown's avatar
unknown committed
5305
      if (rli->relay_log.get_open_count() != rli->cur_log_old_open_count)
5306
      {
unknown's avatar
unknown committed
5307 5308
        // 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
5309
        mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
5310 5311 5312
        if (!cur_log)                           // No more log files
          goto err;
        hot_log=0;                              // Using old binary log
5313 5314
      }
    }
5315 5316 5317 5318 5319 5320 5321
    /* 
      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;
5322 5323
#ifndef DBUG_OFF
    {
5324
      /* This is an assertion which sometimes fails, let's try to track it */
5325
      char llbuf1[22], llbuf2[22];
5326
      DBUG_PRINT("info", ("my_b_tell(cur_log)=%s rli->event_relay_log_pos=%s",
5327
                          llstr(my_b_tell(cur_log),llbuf1),
5328 5329 5330
                          llstr(rli->event_relay_log_pos,llbuf2)));
      DBUG_ASSERT(my_b_tell(cur_log) >= BIN_LOG_HEADER_SIZE);
      DBUG_ASSERT(my_b_tell(cur_log) == rli->event_relay_log_pos);
5331 5332
    }
#endif
unknown's avatar
unknown committed
5333 5334
    /*
      Relay log is always in new format - if the master is 3.23, the
5335
      I/O thread will convert the format for us.
5336 5337
      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
5338 5339
      logs, which may even have been deleted. So we need to write this
      description event at the beginning of the relay log.
5340 5341
      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.
5342
      But if the relay log is created by new_file(): then the solution is:
5343
      MYSQL_BIN_LOG::open() will write the buffered description event.
unknown's avatar
unknown committed
5344
    */
5345 5346 5347
    if ((ev= Log_event::read_log_event(cur_log,0,
                                       rli->relay_log.description_event_for_exec,
                                       opt_slave_sql_verify_checksum)))
5348

5349 5350
    {
      DBUG_ASSERT(thd==rli->sql_thd);
5351 5352 5353 5354 5355
      /*
        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);
5356 5357 5358 5359 5360 5361 5362 5363 5364 5365 5366 5367 5368 5369 5370 5371 5372 5373 5374 5375 5376
      /*
        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 (ev->get_type_code() == GTID_EVENT)
      {
        Gtid_log_event *gev= static_cast<Gtid_log_event *>(ev);
        uint64 sub_id= rpl_global_gtid_slave_state.next_subid(gev->domain_id);
        if (!sub_id)
        {
          errmsg = "slave SQL thread aborted because of out-of-memory error";
          if (hot_log)
            mysql_mutex_unlock(log_lock);
          goto err;
        }
        rli->gtid_sub_id= sub_id;
        rli->current_gtid.server_id= gev->server_id;
        rli->current_gtid.domain_id= gev->domain_id;
        rli->current_gtid.seq_no= gev->seq_no;
      }

5377
      if (hot_log)
Marc Alff's avatar
Marc Alff committed
5378
        mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
5379
      DBUG_RETURN(ev);
5380 5381
    }
    DBUG_ASSERT(thd==rli->sql_thd);
unknown's avatar
unknown committed
5382
    if (opt_reckless_slave)                     // For mysql-test
unknown's avatar
unknown committed
5383
      cur_log->error = 0;
unknown's avatar
unknown committed
5384
    if (cur_log->error < 0)
unknown's avatar
unknown committed
5385 5386
    {
      errmsg = "slave SQL thread aborted because of I/O error";
unknown's avatar
unknown committed
5387
      if (hot_log)
Marc Alff's avatar
Marc Alff committed
5388
        mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
5389 5390
      goto err;
    }
5391 5392
    if (!cur_log->error) /* EOF */
    {
unknown's avatar
unknown committed
5393
      /*
unknown's avatar
unknown committed
5394 5395 5396
        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
5397
      */
5398 5399
      if (hot_log)
      {
5400 5401 5402 5403 5404 5405 5406 5407 5408 5409
        /*
          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").
5410 5411 5412 5413 5414 5415 5416 5417 5418 5419

          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.
5420
        */
5421
        time_t save_timestamp= rli->last_master_timestamp;
5422 5423
        rli->last_master_timestamp= 0;

unknown's avatar
unknown committed
5424
        DBUG_ASSERT(rli->relay_log.get_open_count() ==
unknown's avatar
unknown committed
5425
                    rli->cur_log_old_open_count);
5426 5427 5428 5429 5430

        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"));
5431
          ev= new Rotate_log_event(rli->ign_master_log_name_end,
5432
                                   0, rli->ign_master_log_pos_end,
unknown's avatar
unknown committed
5433
                                   Rotate_log_event::DUP_NAME);
5434
          rli->ign_master_log_name_end[0]= 0;
Marc Alff's avatar
Marc Alff committed
5435
          mysql_mutex_unlock(log_lock);
5436 5437 5438 5439 5440 5441 5442 5443 5444 5445
          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
5446 5447 5448 5449
        /*
          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
5450
        mysql_mutex_unlock(&rli->data_lock);
5451 5452

        /*
unknown's avatar
unknown committed
5453
          Possible deadlock :
5454 5455 5456 5457 5458 5459 5460 5461 5462 5463 5464
          - 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
5465
          master does not send anything). Then the I/O thread stops waiting
5466 5467 5468 5469 5470 5471 5472 5473 5474 5475
          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
          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,
          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 
          RESET SLAVE, but in fact, no need as RESET SLAVE requires that the slave
unknown's avatar
unknown committed
5476 5477
          be stopped, and the SQL thread sets ignore_log_space_limit to 0 when
          it stops.
5478
        */
Marc Alff's avatar
Marc Alff committed
5479
        mysql_mutex_lock(&rli->log_space_lock);
5480 5481 5482 5483 5484 5485 5486 5487 5488 5489 5490 5491 5492 5493 5494 5495 5496 5497 5498 5499 5500 5501 5502 5503 5504

        /* 
          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;
        }

5505
        /*
5506 5507 5508 5509
          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.
5510
        */
Marc Alff's avatar
Marc Alff committed
5511 5512
        mysql_mutex_unlock(&rli->log_space_lock);
        mysql_cond_broadcast(&rli->log_space_cond);
Andrei Elkin's avatar
Andrei Elkin committed
5513 5514
        // Note that wait_for_update_relay_log unlocks lock_log !
        rli->relay_log.wait_for_update_relay_log(rli->sql_thd);
5515
        // re-acquire data lock since we released it earlier
Marc Alff's avatar
Marc Alff committed
5516
        mysql_mutex_lock(&rli->data_lock);
5517
        rli->last_master_timestamp= save_timestamp;
unknown's avatar
unknown committed
5518
        continue;
5519
      }
unknown's avatar
unknown committed
5520
      /*
unknown's avatar
unknown committed
5521 5522 5523
        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
5524 5525 5526
      */
      end_io_cache(cur_log);
      DBUG_ASSERT(rli->cur_log_fd >= 0);
Marc Alff's avatar
Marc Alff committed
5527
      mysql_file_close(rli->cur_log_fd, MYF(MY_WME));
unknown's avatar
unknown committed
5528
      rli->cur_log_fd = -1;
unknown's avatar
unknown committed
5529

5530
      if (relay_log_purge)
unknown's avatar
unknown committed
5531
      {
unknown's avatar
unknown committed
5532
        /*
5533 5534 5535 5536 5537 5538 5539 5540 5541 5542
          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
5543
        if (rli->relay_log.purge_first_log
5544 5545 5546
            (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
5547 5548 5549 5550
        {
          errmsg = "Error purging processed logs";
          goto err;
        }
unknown's avatar
unknown committed
5551
      }
5552 5553
      else
      {
unknown's avatar
unknown committed
5554 5555 5556 5557 5558 5559 5560 5561 5562 5563 5564 5565 5566 5567 5568 5569
        /*
          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;
        strmake(rli->event_relay_log_name,rli->linfo.log_file_name,
                sizeof(rli->event_relay_log_name)-1);
        flush_relay_log_info(rli);
unknown's avatar
unknown committed
5570
      }
5571 5572 5573 5574 5575 5576 5577 5578 5579 5580 5581 5582 5583

      /*
        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
5584
        mysql_mutex_lock(log_lock);
unknown's avatar
unknown committed
5585 5586
      if (rli->relay_log.is_active(rli->linfo.log_file_name))
      {
unknown's avatar
unknown committed
5587 5588 5589 5590 5591
        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);

        /*
5592 5593 5594 5595 5596 5597 5598 5599 5600 5601 5602 5603 5604 5605 5606 5607 5608 5609 5610 5611 5612 5613 5614 5615 5616 5617 5618 5619 5620 5621 5622 5623 5624 5625 5626 5627 5628 5629 5630 5631 5632 5633 5634 5635 5636 5637 5638 5639 5640 5641 5642 5643 5644 5645 5646 5647 5648
           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
5649
        */
5650 5651

        my_b_seek(cur_log, (my_off_t) 0);
unknown's avatar
unknown committed
5652
        if (check_binlog_magic(cur_log,&errmsg))
5653
        {
Marc Alff's avatar
Marc Alff committed
5654 5655
          if (!hot_log)
            mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
5656
          goto err;
5657
        }
Marc Alff's avatar
Marc Alff committed
5658 5659
        if (!hot_log)
          mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
5660
        continue;
5661
      }
Marc Alff's avatar
Marc Alff committed
5662 5663
      if (!hot_log)
        mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
5664
      /*
unknown's avatar
unknown committed
5665 5666 5667
        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
5668 5669 5670
      */
      // 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
5671 5672
                                       &errmsg)) <0)
        goto err;
5673
    }
unknown's avatar
unknown committed
5674
    else
5675
    {
unknown's avatar
unknown committed
5676
      /*
unknown's avatar
unknown committed
5677 5678
        Read failed with a non-EOF error.
        TODO: come up with something better to handle this error
unknown's avatar
unknown committed
5679 5680
      */
      if (hot_log)
Marc Alff's avatar
Marc Alff committed
5681
        mysql_mutex_unlock(log_lock);
5682
      sql_print_error("Slave SQL thread: I/O error reading \
unknown's avatar
unknown committed
5683
event(errno: %d  cur_log->error: %d)",
unknown's avatar
unknown committed
5684
                      my_errno,cur_log->error);
unknown's avatar
unknown committed
5685
      // set read position to the beginning of the event
5686
      my_b_seek(cur_log,rli->event_relay_log_pos);
unknown's avatar
unknown committed
5687 5688
      /* otherwise, we have had a partial read */
      errmsg = "Aborting slave SQL thread because of partial event read";
unknown's avatar
unknown committed
5689
      break;                                    // To end of function
5690 5691
    }
  }
5692
  if (!errmsg && global_system_variables.log_warnings)
5693
  {
unknown's avatar
unknown committed
5694
    sql_print_information("Error reading relay log event: %s",
5695 5696 5697
                          "slave SQL thread was killed");
    DBUG_RETURN(0);
  }
unknown's avatar
unknown committed
5698

5699
err:
5700 5701
  if (errmsg)
    sql_print_error("Error reading relay log event: %s", errmsg);
unknown's avatar
unknown committed
5702
  DBUG_RETURN(0);
5703 5704
}

5705 5706 5707
/*
  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
5708
  locks; here we don't, so this function is mainly taking locks).
5709 5710
  Returns nothing as we cannot catch any error (MYSQL_BIN_LOG::new_file()
  is void).
5711 5712
*/

5713
int rotate_relay_log(Master_info* mi)
5714 5715
{
  DBUG_ENTER("rotate_relay_log");
5716
  Relay_log_info* rli= &mi->rli;
5717
  int error= 0;
unknown's avatar
unknown committed
5718

Luis Soares's avatar
Luis Soares committed
5719
  DBUG_EXECUTE_IF("crash_before_rotate_relaylog", DBUG_SUICIDE(););
5720

unknown's avatar
unknown committed
5721
  /*
unknown's avatar
unknown committed
5722 5723 5724
     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).
  */
5725 5726
  if (!rli->inited)
  {
unknown's avatar
unknown committed
5727
    DBUG_PRINT("info", ("rli->inited == 0"));
unknown's avatar
unknown committed
5728
    goto end;
5729
  }
unknown's avatar
unknown committed
5730

5731
  /* If the relay log is closed, new_file() will do nothing. */
5732 5733
  if ((error= rli->relay_log.new_file()))
    goto end;
unknown's avatar
unknown committed
5734

5735 5736 5737 5738
  /*
    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
5739 5740 5741 5742 5743 5744
    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.
5745 5746
    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
5747 5748

    Note that it needs to be protected by mi->data_lock.
5749
  */
Sergei Golubchik's avatar
Sergei Golubchik committed
5750
  mysql_mutex_assert_owner(&mi->data_lock);
5751
  rli->relay_log.harvest_bytes_written(&rli->log_space_total);
unknown's avatar
unknown committed
5752
end:
5753
  DBUG_RETURN(error);
5754 5755
}

unknown's avatar
unknown committed
5756

unknown's avatar
unknown committed
5757 5758 5759
/**
   Detects, based on master's version (as found in the relay log), if master
   has a certain bug.
5760
   @param rli Relay_log_info which tells the master's version
unknown's avatar
unknown committed
5761
   @param bug_id Number of the bug as found in bugs.mysql.com
5762
   @param report bool report error message, default TRUE
5763 5764 5765 5766 5767 5768 5769

   @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
5770 5771
   @return TRUE if master has the bug, FALSE if it does not.
*/
5772 5773
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
5774 5775 5776 5777 5778 5779 5780 5781
{
  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
5782
    {24432, { 5, 0, 24 }, { 5, 0, 38 } },
5783 5784 5785
    {24432, { 5, 1, 12 }, { 5, 1, 17 } },
    {33029, { 5, 0,  0 }, { 5, 0, 58 } },
    {33029, { 5, 1,  0 }, { 5, 1, 12 } },
5786
    {37426, { 5, 1,  0 }, { 5, 1, 26 } },
unknown's avatar
unknown committed
5787 5788
  };
  const uchar *master_ver=
5789
    rli->relay_log.description_event_for_exec->server_version_split.ver;
unknown's avatar
unknown committed
5790

5791
  DBUG_ASSERT(sizeof(rli->relay_log.description_event_for_exec->server_version_split.ver) == 3);
unknown's avatar
unknown committed
5792 5793 5794 5795 5796 5797 5798 5799

  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) &&
5800 5801
        (memcmp(fixed_in,      master_ver, 3) >  0) &&
        (pred == NULL || (*pred)(param)))
unknown's avatar
unknown committed
5802
    {
5803 5804
      if (!report)
	return TRUE;
unknown's avatar
unknown committed
5805 5806 5807 5808 5809
      // 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);
5810
      // a verbose message for the error log
5811 5812 5813
      rli->report(ERROR_LEVEL, ER_UNKNOWN_ERROR,
                  "According to the master's version ('%s'),"
                  " it is probable that master suffers from this bug:"
5814 5815 5816 5817 5818 5819 5820 5821 5822 5823 5824 5825 5826 5827 5828 5829
                      " 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
5830 5831 5832 5833 5834 5835
      return TRUE;
    }
  }
  return FALSE;
}

5836 5837 5838 5839 5840 5841 5842 5843 5844 5845 5846 5847
/**
   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)
{
5848
  if (thd->rli_slave)
5849
  {
5850
    DBUG_EXECUTE_IF("simulate_bug33029", return TRUE;);
5851
    return rpl_master_has_bug(thd->rli_slave, 33029, FALSE, NULL, NULL);
5852 5853 5854 5855
  }
  return FALSE;
}

5856 5857 5858 5859
/**
  @} (end of group Replication)
*/

unknown's avatar
SCRUM  
unknown committed
5860
#endif /* HAVE_REPLICATION */