slave.cc 194 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);
165
static bool send_show_master_info_header(THD *thd, bool full);
166
static bool send_show_master_info_data(THD *thd, Master_info *mi, bool full);
167 168

/*
unknown's avatar
unknown committed
169
  Find out which replications threads are running
170

unknown's avatar
unknown committed
171 172
  SYNOPSIS
    init_thread_mask()
unknown's avatar
unknown committed
173 174 175
    mask                Return value here
    mi                  master_info for slave
    inverse             If set, returns which threads are not running
unknown's avatar
unknown committed
176 177 178 179 180 181

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

  RETURN
unknown's avatar
unknown committed
182 183
    mask        If inverse == 0, running threads
                If inverse == 1, stopped threads
184 185
*/

186
void init_thread_mask(int* mask,Master_info* mi,bool inverse)
187 188 189
{
  bool set_io = mi->slave_running, set_sql = mi->rli.slave_running;
  register int tmp_mask=0;
190 191
  DBUG_ENTER("init_thread_mask");

192 193 194 195
  if (set_io)
    tmp_mask |= SLAVE_IO;
  if (set_sql)
    tmp_mask |= SLAVE_SQL;
196 197
  if (inverse)
    tmp_mask^= (SLAVE_IO | SLAVE_SQL);
198
  *mask = tmp_mask;
199
  DBUG_VOID_RETURN;
200 201
}

202

unknown's avatar
unknown committed
203
/*
204
  lock_slave_threads()
unknown's avatar
unknown committed
205
*/
206

207
void lock_slave_threads(Master_info* mi)
208
{
209 210
  DBUG_ENTER("lock_slave_threads");

211
  //TODO: see if we can do this without dual mutex
Marc Alff's avatar
Marc Alff committed
212 213
  mysql_mutex_lock(&mi->run_lock);
  mysql_mutex_lock(&mi->rli.run_lock);
214
  DBUG_VOID_RETURN;
215 216
}

217

unknown's avatar
unknown committed
218
/*
219
  unlock_slave_threads()
unknown's avatar
unknown committed
220
*/
221

222
void unlock_slave_threads(Master_info* mi)
223
{
224 225
  DBUG_ENTER("unlock_slave_threads");

226
  //TODO: see if we can do this without dual mutex
Marc Alff's avatar
Marc Alff committed
227 228
  mysql_mutex_unlock(&mi->rli.run_lock);
  mysql_mutex_unlock(&mi->run_lock);
229
  DBUG_VOID_RETURN;
230 231
}

Marc Alff's avatar
Marc Alff committed
232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252
#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 */
253

unknown's avatar
unknown committed
254
/* Initialize slave structures */
255

256 257
int init_slave()
{
258
  DBUG_ENTER("init_slave");
259
  int error= 0;
260

Marc Alff's avatar
Marc Alff committed
261 262 263 264
#ifdef HAVE_PSI_INTERFACE
  init_slave_psi_keys();
#endif

265 266 267 268 269
  /*
    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
270
  mysql_mutex_lock(&LOCK_active_mi);
271

He Zhenxing's avatar
He Zhenxing committed
272 273
  if (pthread_key_create(&RPL_MASTER_INFO, NULL))
    goto err;
274

275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297
  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;
  }

298 299 300 301 302 303 304 305 306
  /*
    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();
  }

307
  /*
308 309 310
    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.
311
  */
312

unknown's avatar
unknown committed
313
  if (init_master_info(active_mi,master_info_file,relay_log_info_file,
314
                       1, (SLAVE_IO | SLAVE_SQL)))
315
  {
316
    sql_print_error("Failed to initialize the master info structure");
unknown's avatar
unknown committed
317
    goto err;
318
  }
319

320 321
  /* If server id is not set, start_slave_thread() will say it */

322
  if (active_mi->host[0] && !opt_skip_slave_start)
323
  {
324
    if (start_slave_threads(1 /* need mutex */,
unknown's avatar
unknown committed
325 326 327 328 329
                            0 /* no wait for start*/,
                            active_mi,
                            master_info_file,
                            relay_log_info_file,
                            SLAVE_IO | SLAVE_SQL))
unknown's avatar
unknown committed
330
    {
331
      sql_print_error("Failed to create slave threads");
unknown's avatar
unknown committed
332 333
      goto err;
    }
334 335
  }

336
end:
Marc Alff's avatar
Marc Alff committed
337
  mysql_mutex_unlock(&LOCK_active_mi);
338
  DBUG_RETURN(error);
339 340 341 342 343

err:
  sql_print_error("Failed to allocate memory for the Master Info structure");
  error= 1;
  goto end;
344
}
345

346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372
/*
  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).
 */
373
int init_recovery(Master_info* mi, const char** errmsg)
374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394
{
  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;
  }
395

396 397 398
  DBUG_RETURN(0);
}
 
399 400 401 402 403 404 405 406 407 408 409 410
/**
  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.
  */
411
  const size_t MIN_ROOM= 10;
412 413 414 415
  DBUG_ENTER("print_slave_skip_errors");
  DBUG_ASSERT(sizeof(slave_skip_error_names) > MIN_ROOM);
  DBUG_ASSERT(MAX_SLAVE_ERROR <= 999999); // 6 digits

416 417 418
  /* Make @@slave_skip_errors show the nice human-readable value.  */
  opt_slave_skip_errors= slave_skip_error_names;

419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436
  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;

437
    for (errnum= 0; errnum < MAX_SLAVE_ERROR; errnum++)
438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459
    {
      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
460
/*
461
  Init function to set up array for errors that should be skipped for slave
unknown's avatar
unknown committed
462

unknown's avatar
unknown committed
463 464
  SYNOPSIS
    init_slave_skip_errors()
unknown's avatar
unknown committed
465
    arg         List of errors numbers to skip, separated with ','
unknown's avatar
unknown committed
466 467 468 469

  NOTES
    Called from get_options() in mysqld.cc on start-up
*/
470

unknown's avatar
unknown committed
471
void init_slave_skip_errors(const char* arg)
472
{
unknown's avatar
unknown committed
473
  const char *p;
474 475
  DBUG_ENTER("init_slave_skip_errors");

476
  if (bitmap_init(&slave_error_mask,0,MAX_SLAVE_ERROR,0))
477 478 479 480 481
  {
    fprintf(stderr, "Badly out of memory, please check your system status\n");
    exit(1);
  }
  use_slave_mask = 1;
482
  for (;my_isspace(system_charset_info,*arg);++arg)
483
    /* empty */;
484
  if (!my_strnncoll(system_charset_info,(uchar*)arg,4,(const uchar*)"all",4))
485 486
  {
    bitmap_set_all(&slave_error_mask);
487
    print_slave_skip_errors();
488
    DBUG_VOID_RETURN;
489 490 491 492 493 494 495 496
  }
  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);
497
    while (!my_isdigit(system_charset_info,*p) && *p)
498 499
      p++;
  }
500 501
  /* Convert slave skip errors bitmap into a printable string. */
  print_slave_skip_errors();
502
  DBUG_VOID_RETURN;
503 504
}

505
static void set_thd_in_use_temporary_tables(Relay_log_info *rli)
506 507 508 509 510 511
{
  TABLE *table;

  for (table= rli->save_temporary_tables ; table ; table= table->next)
    table->in_use= rli->sql_thd;
}
512

513
int terminate_slave_threads(Master_info* mi,int thread_mask,bool skip_lock)
514
{
515 516
  DBUG_ENTER("terminate_slave_threads");

517
  if (!mi->inited)
518
    DBUG_RETURN(0); /* successfully do nothing */
519
  int error,force_all = (thread_mask & SLAVE_FORCE_ALL);
Marc Alff's avatar
Marc Alff committed
520
  mysql_mutex_t *sql_lock = &mi->rli.run_lock, *io_lock = &mi->run_lock;
521
  mysql_mutex_t *log_lock= mi->rli.relay_log.get_log_lock();
unknown's avatar
unknown committed
522

523
  if (thread_mask & (SLAVE_SQL|SLAVE_FORCE_ALL))
524
  {
unknown's avatar
unknown committed
525
    DBUG_PRINT("info",("Terminating SQL thread"));
526
    mi->rli.abort_slave=1;
527
    if ((error=terminate_slave_thread(mi->rli.sql_thd, sql_lock,
unknown's avatar
unknown committed
528
                                      &mi->rli.stop_cond,
529 530
                                      &mi->rli.slave_running,
                                      skip_lock)) &&
unknown's avatar
unknown committed
531
        !force_all)
532
      DBUG_RETURN(error);
533

534
    mysql_mutex_lock(log_lock);
535

Libing Song's avatar
Libing Song committed
536
    DBUG_PRINT("info",("Flushing relay-log info file."));
537
    if (current_thd)
Libing Song's avatar
Libing Song committed
538 539
      thd_proc_info(current_thd, "Flushing relay-log info file.");
    if (flush_relay_log_info(&mi->rli))
540
      DBUG_RETURN(ER_ERROR_DURING_FLUSH_LOGS);
Libing Song's avatar
Libing Song committed
541 542
    
    if (my_sync(mi->rli.info_fd, MYF(MY_WME)))
543 544
      DBUG_RETURN(ER_ERROR_DURING_FLUSH_LOGS);

545
    mysql_mutex_unlock(log_lock);
546
  }
547
  if (thread_mask & (SLAVE_IO|SLAVE_FORCE_ALL))
548
  {
unknown's avatar
unknown committed
549
    DBUG_PRINT("info",("Terminating IO thread"));
550
    mi->abort_slave=1;
551
    if ((error=terminate_slave_thread(mi->io_thd, io_lock,
unknown's avatar
unknown committed
552
                                      &mi->stop_cond,
553 554
                                      &mi->slave_running,
                                      skip_lock)) &&
unknown's avatar
unknown committed
555
        !force_all)
556
      DBUG_RETURN(error);
557

558
    mysql_mutex_lock(log_lock);
559 560 561 562

    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.");
563
    if (flush_master_info(mi, TRUE, FALSE))
564 565
      DBUG_RETURN(ER_ERROR_DURING_FLUSH_LOGS);

Luis Soares's avatar
Luis Soares committed
566 567
    if (mi->rli.relay_log.is_open() &&
        my_sync(mi->rli.relay_log.get_log_file()->file, MYF(MY_WME)))
568 569 570 571 572
      DBUG_RETURN(ER_ERROR_DURING_FLUSH_LOGS);

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

573
    mysql_mutex_unlock(log_lock);
574
  }
575
  DBUG_RETURN(0); 
576 577
}

578

579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605
/**
   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
606
   @retval 0 All OK ER_SLAVE_NOT_RUNNING otherwise.
607

Andrei Elkin's avatar
Andrei Elkin committed
608 609 610
   @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.
611

612 613 614
 */
static int
terminate_slave_thread(THD *thd,
Marc Alff's avatar
Marc Alff committed
615 616
                       mysql_mutex_t *term_lock,
                       mysql_cond_t *term_cond,
617 618
                       volatile uint *slave_running,
                       bool skip_lock)
619
{
620
  DBUG_ENTER("terminate_slave_thread");
621
  if (!skip_lock)
622
  {
Marc Alff's avatar
Marc Alff committed
623
    mysql_mutex_lock(term_lock);
624 625 626
  }
  else
  {
Marc Alff's avatar
Marc Alff committed
627
    mysql_mutex_assert_owner(term_lock);
628
  }
629 630 631
  if (!*slave_running)
  {
    if (!skip_lock)
632
    {
633 634 635 636
      /*
        if run_lock (term_lock) is acquired locally then either
        slave_running status is fine
      */
Marc Alff's avatar
Marc Alff committed
637
      mysql_mutex_unlock(term_lock);
638 639 640 641
      DBUG_RETURN(0);
    }
    else
    {
642
      DBUG_RETURN(ER_SLAVE_NOT_RUNNING);
643 644 645
    }
  }
  DBUG_ASSERT(thd != 0);
646
  THD_CHECK_SENTRY(thd);
647

unknown's avatar
unknown committed
648
  /*
649
    Is is critical to test if the slave is running. Otherwise, we might
unknown's avatar
unknown committed
650
    be referening freed memory trying to kick it
651
  */
unknown's avatar
unknown committed
652

unknown's avatar
unknown committed
653
  while (*slave_running)                        // Should always be true
654
  {
Michael Widenius's avatar
Michael Widenius committed
655
    int error __attribute__((unused));
656
    DBUG_PRINT("loop", ("killing slave thread"));
657

Marc Alff's avatar
Marc Alff committed
658
    mysql_mutex_lock(&thd->LOCK_thd_data);
659 660 661 662 663 664
#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)
    */
665
    int err __attribute__((unused))= pthread_kill(thd->real_id, thr_client_alarm);
666 667
    DBUG_ASSERT(err != EINVAL);
#endif
Sergei Golubchik's avatar
Sergei Golubchik committed
668 669
    thd->awake(NOT_KILLED);

Marc Alff's avatar
Marc Alff committed
670
    mysql_mutex_unlock(&thd->LOCK_thd_data);
671

unknown's avatar
unknown committed
672 673 674
    /*
      There is a small chance that slave thread might miss the first
      alarm. To protect againts it, resend the signal until it reacts
675 676
    */
    struct timespec abstime;
unknown's avatar
unknown committed
677
    set_timespec(abstime,2);
Marc Alff's avatar
Marc Alff committed
678
    error= mysql_cond_timedwait(term_cond, term_lock, &abstime);
679
    DBUG_ASSERT(error == ETIMEDOUT || error == 0);
680
  }
681 682 683 684

  DBUG_ASSERT(*slave_running == 0);

  if (!skip_lock)
Marc Alff's avatar
Marc Alff committed
685
    mysql_mutex_unlock(term_lock);
686
  DBUG_RETURN(0);
687 688
}

689

Marc Alff's avatar
Marc Alff committed
690 691 692 693 694 695 696
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
697 698
                       volatile uint *slave_running,
                       volatile ulong *slave_run_id,
Konstantin Osipov's avatar
Konstantin Osipov committed
699
                       Master_info* mi)
700 701
{
  pthread_t th;
unknown's avatar
unknown committed
702 703 704
  ulong start_id;
  DBUG_ENTER("start_slave_thread");

705 706
  DBUG_ASSERT(mi->inited);

707
  if (start_lock)
Marc Alff's avatar
Marc Alff committed
708
    mysql_mutex_lock(start_lock);
709
  if (!global_system_variables.server_id)
710 711
  {
    if (start_cond)
Marc Alff's avatar
Marc Alff committed
712
      mysql_cond_broadcast(start_cond);
713
    if (start_lock)
Marc Alff's avatar
Marc Alff committed
714
      mysql_mutex_unlock(start_lock);
715
    sql_print_error("Server id not set, will not start slave");
unknown's avatar
unknown committed
716
    DBUG_RETURN(ER_BAD_SLAVE);
717
  }
unknown's avatar
unknown committed
718

719
  if (*slave_running)
720 721
  {
    if (start_cond)
Marc Alff's avatar
Marc Alff committed
722
      mysql_cond_broadcast(start_cond);
723
    if (start_lock)
Marc Alff's avatar
Marc Alff committed
724
      mysql_mutex_unlock(start_lock);
unknown's avatar
unknown committed
725
    DBUG_RETURN(ER_SLAVE_MUST_STOP);
726
  }
unknown's avatar
unknown committed
727 728
  start_id= *slave_run_id;
  DBUG_PRINT("info",("Creating new slave thread"));
Marc Alff's avatar
Marc Alff committed
729 730
  if (mysql_thread_create(thread_key,
                          &th, &connection_attrib, h_func, (void*)mi))
731 732
  {
    if (start_lock)
Marc Alff's avatar
Marc Alff committed
733
      mysql_mutex_unlock(start_lock);
unknown's avatar
unknown committed
734
    DBUG_RETURN(ER_SLAVE_THREAD);
735
  }
unknown's avatar
unknown committed
736
  if (start_cond && cond_lock) // caller has cond_lock
737 738
  {
    THD* thd = current_thd;
unknown's avatar
unknown committed
739
    while (start_id == *slave_run_id)
740
    {
unknown's avatar
unknown committed
741
      DBUG_PRINT("sleep",("Waiting for slave thread to start"));
742 743 744 745 746 747 748 749 750 751 752
      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);
753
      thd->exit_cond(old_msg);
Marc Alff's avatar
Marc Alff committed
754
      mysql_mutex_lock(cond_lock); // re-acquire it as exit_cond() released
755
      if (thd->killed)
756 757
      {
        if (start_lock)
Marc Alff's avatar
Marc Alff committed
758
          mysql_mutex_unlock(start_lock);
unknown's avatar
unknown committed
759
        DBUG_RETURN(thd->killed_errno());
760
      }
761 762 763
    }
  }
  if (start_lock)
Marc Alff's avatar
Marc Alff committed
764
    mysql_mutex_unlock(start_lock);
unknown's avatar
unknown committed
765
  DBUG_RETURN(0);
766
}
unknown's avatar
unknown committed
767

768

unknown's avatar
unknown committed
769
/*
770
  start_slave_threads()
unknown's avatar
unknown committed
771

unknown's avatar
unknown committed
772 773 774 775
  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
776
*/
unknown's avatar
unknown committed
777

778
int start_slave_threads(bool need_slave_mutex, bool wait_for_start,
779
                        Master_info* mi, const char* master_info_fname,
unknown's avatar
unknown committed
780
                        const char* slave_info_fname, int thread_mask)
781
{
Marc Alff's avatar
Marc Alff committed
782 783
  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;
784
  int error=0;
785
  DBUG_ENTER("start_slave_threads");
unknown's avatar
unknown committed
786

787 788 789 790 791 792 793 794 795 796 797 798
  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;
  }
799 800

  if (thread_mask & SLAVE_IO)
Marc Alff's avatar
Marc Alff committed
801 802 803 804 805 806 807 808
    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);
809
  if (!error && (thread_mask & SLAVE_SQL))
810
  {
Marc Alff's avatar
Marc Alff committed
811 812 813 814 815 816 817 818
    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);
819
    if (error)
820
      terminate_slave_threads(mi, thread_mask & SLAVE_IO, !need_slave_mutex);
821
  }
822
  DBUG_RETURN(error);
823
}
824

825

unknown's avatar
unknown committed
826
/*
827
  Release slave threads at time of executing shutdown.
unknown's avatar
unknown committed
828 829 830 831

  SYNOPSIS
    end_slave()
*/
832

833 834
void end_slave()
{
835 836
  DBUG_ENTER("end_slave");

837 838 839 840 841 842
  /*
    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.
843 844 845

    We can also be called by cleanup(), which only happens if some
    startup parameter to the server was wrong.
846
  */
Marc Alff's avatar
Marc Alff committed
847
  mysql_mutex_lock(&LOCK_active_mi);
848 849 850 851
  /* 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
852
  mysql_mutex_unlock(&LOCK_active_mi);
853 854 855
  DBUG_VOID_RETURN;
}

856
static bool io_slave_killed(THD* thd, Master_info* mi)
unknown's avatar
unknown committed
857
{
858 859
  DBUG_ENTER("io_slave_killed");

860
  DBUG_ASSERT(mi->io_thd == thd);
861
  DBUG_ASSERT(mi->slave_running); // tracking buffer overrun
862
  DBUG_RETURN(mi->abort_slave || abort_loop || thd->killed);
unknown's avatar
unknown committed
863 864
}

865 866 867 868 869 870
/**
   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.
871

872 873
   @param thd   pointer to a THD instance
   @param rli   pointer to Relay_log_info instance
874

875 876 877
   @return TRUE the killed status is recognized, FALSE a possible killed
           status is deferred.
*/
878
static bool sql_slave_killed(THD* thd, Relay_log_info* rli)
879
{
880
  bool ret= FALSE;
881 882
  DBUG_ENTER("sql_slave_killed");

883 884
  DBUG_ASSERT(rli->sql_thd == thd);
  DBUG_ASSERT(rli->slave_running == 1);// tracking buffer overrun
885 886
  if (abort_loop || thd->killed || rli->abort_slave)
  {
887 888 889 890 891 892 893 894
    /*
      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
895 896 897
    if ((thd->transaction.all.modified_non_trans_table ||
         (thd->variables.option_bits & OPTION_KEEP_LOG))
        && rli->is_in_group())
898
    {
899
      char msg_stopped[]=
900 901 902 903
        "... 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 "
904 905 906 907 908
        "ignores duplicate key, key not found, and similar errors (see "
        "documentation for details).";

      if (rli->abort_slave)
      {
909 910 911
        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 ... "));
912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934

        /*
          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,
935 936 937
                      "Request to stop slave SQL Thread received while "
                      "applying a group that has non-transactional "
                      "changes; waiting for completion of the group ... ");
938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954
        }
        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;
955 956
    }
  }
957 958 959 960
  if (ret)
    rli->last_event_start_time= 0;
  
  DBUG_RETURN(ret);
961 962
}

963

unknown's avatar
unknown committed
964
/*
965 966
  skip_load_data_infile()

unknown's avatar
unknown committed
967 968 969
  NOTES
    This is used to tell a 3.23 master to break send_file()
*/
970 971 972

void skip_load_data_infile(NET *net)
{
973 974
  DBUG_ENTER("skip_load_data_infile");

975
  (void)net_request_file(net, "/dev/null");
unknown's avatar
unknown committed
976
  (void)my_net_read(net);                               // discard response
977
  (void)net_write_command(net, 0, (uchar*) "", 0, (uchar*) "", 0); // ok
978
  DBUG_VOID_RETURN;
979 980
}

981

982
bool net_request_file(NET* net, const char* fname)
983
{
984
  DBUG_ENTER("net_request_file");
985 986
  DBUG_RETURN(net_write_command(net, 251, (uchar*) fname, strlen(fname),
                                (uchar*) "", 0));
987 988
}

989 990
/*
  From other comments and tests in code, it looks like
991
  sometimes Query_log_event and Load_log_event can have db == 0
992 993 994
  (see rewrite_db() above for example)
  (cases where this happens are unclear; it may be when the master is 3.23).
*/
995 996

const char *print_slave_db_safe(const char* db)
997
{
998 999 1000
  DBUG_ENTER("*print_slave_db_safe");

  DBUG_RETURN((db ? db : ""));
1001
}
1002

Sergei Golubchik's avatar
Sergei Golubchik committed
1003 1004
#endif /* HAVE_REPLICATION */

1005
int init_strvar_from_file(char *var, int max_size, IO_CACHE *f,
unknown's avatar
unknown committed
1006
                                 const char *default_val)
unknown's avatar
unknown committed
1007
{
unknown's avatar
unknown committed
1008
  uint length;
1009 1010
  DBUG_ENTER("init_strvar_from_file");

unknown's avatar
unknown committed
1011 1012 1013 1014 1015 1016
  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
1017
    {
unknown's avatar
unknown committed
1018
      /*
unknown's avatar
unknown committed
1019 1020
        If we truncated a line or stopped on last char, remove all chars
        up to and including newline.
unknown's avatar
unknown committed
1021
      */
unknown's avatar
unknown committed
1022
      int c;
1023
      while (((c=my_b_get(f)) != '\n' && c != my_b_EOF)) ;
unknown's avatar
unknown committed
1024
    }
1025
    DBUG_RETURN(0);
unknown's avatar
unknown committed
1026 1027 1028
  }
  else if (default_val)
  {
unknown's avatar
unknown committed
1029
    strmake(var,  default_val, max_size-1);
1030
    DBUG_RETURN(0);
unknown's avatar
unknown committed
1031
  }
1032
  DBUG_RETURN(1);
unknown's avatar
unknown committed
1033 1034
}

1035

Sergei Golubchik's avatar
Sergei Golubchik committed
1036 1037 1038 1039
/*
  when moving these functions to mysys, don't forget to
  remove slave.cc from libmysqld/CMakeLists.txt
*/
1040
int init_intvar_from_file(int* var, IO_CACHE* f, int default_val)
unknown's avatar
unknown committed
1041 1042
{
  char buf[32];
1043 1044
  DBUG_ENTER("init_intvar_from_file");

unknown's avatar
unknown committed
1045 1046

  if (my_b_gets(f, buf, sizeof(buf)))
unknown's avatar
unknown committed
1047 1048
  {
    *var = atoi(buf);
1049
    DBUG_RETURN(0);
unknown's avatar
unknown committed
1050
  }
unknown's avatar
unknown committed
1051
  else if (default_val)
unknown's avatar
unknown committed
1052 1053
  {
    *var = default_val;
1054
    DBUG_RETURN(0);
unknown's avatar
unknown committed
1055
  }
1056
  DBUG_RETURN(1);
unknown's avatar
unknown committed
1057 1058
}

Andrei Elkin's avatar
Andrei Elkin committed
1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079
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
1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131

/**
   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 ||
1132
        ((snd_size + 1 == max_size - read_size) &&  buf_act[max_size - 2] != '\n'))
Andrei Elkin's avatar
Andrei Elkin committed
1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163
    {
      /*
        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)
1164
    my_free(buf_act);
Andrei Elkin's avatar
Andrei Elkin committed
1165 1166 1167
  DBUG_RETURN(ret);
}

Sergei Golubchik's avatar
Sergei Golubchik committed
1168 1169
#ifdef HAVE_REPLICATION

1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184
/*
  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 ||
1185
      errorno == ER_CONNECTION_KILLED ||
1186
      errorno == ER_NEW_ABORTING_CONNECTION ||
1187 1188 1189 1190 1191 1192 1193
      errorno == ER_SERVER_SHUTDOWN)
    return TRUE;

  return FALSE;   
}


1194 1195 1196 1197 1198 1199 1200 1201
/*
  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
1202

1203 1204 1205
  RETURNS
  0       ok
  1       error
1206
  2       transient network problem, the caller should try to reconnect
1207
*/
1208

1209
static int get_master_version_and_clock(MYSQL* mysql, Master_info* mi)
1210
{
1211
  char err_buff[MAX_SLAVE_ERRMSG], err_buff2[MAX_SLAVE_ERRMSG];
1212
  const char* errmsg= 0;
1213 1214 1215
  int err_code= 0;
  MYSQL_RES *master_res= 0;
  MYSQL_ROW master_row;
1216
  uint version= mysql_get_server_version(mysql) / 10000;
1217
  DBUG_ENTER("get_master_version_and_clock");
1218 1219 1220 1221 1222 1223 1224

  /*
    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
1225

1226
  if (!my_isdigit(&my_charset_bin,*mysql->server_version))
1227
  {
1228 1229 1230 1231
    errmsg= err_buff2;
    snprintf(err_buff2, sizeof(err_buff2),
             "Master reported unrecognized MySQL version: %s",
             mysql->server_version);
1232
    err_code= ER_SLAVE_FATAL_ERROR;
1233
    sprintf(err_buff, ER(err_code), err_buff2);
1234
  }
1235 1236 1237 1238 1239
  else
  {
    /*
      Note the following switch will bug when we have MySQL branch 30 ;)
    */
1240 1241 1242 1243 1244 1245 1246 1247
    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);
1248
      err_code= ER_SLAVE_FATAL_ERROR;
1249
      sprintf(err_buff, ER(err_code), err_buff2);
1250
      break;
1251
    case 3:
1252
      mi->rli.relay_log.description_event_for_queue= new
unknown's avatar
unknown committed
1253
        Format_description_log_event(1, mysql->server_version);
1254
      break;
1255
    case 4:
1256
      mi->rli.relay_log.description_event_for_queue= new
unknown's avatar
unknown committed
1257
        Format_description_log_event(3, mysql->server_version);
1258
      break;
unknown's avatar
unknown committed
1259
    default:
1260 1261 1262 1263 1264 1265 1266 1267 1268
      /*
        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
1269
        Format_description_log_event(4, mysql->server_version);
1270 1271
      break;
    }
1272
  }
unknown's avatar
unknown committed
1273 1274

  /*
1275 1276 1277 1278 1279
     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
1280

1281
  if (errmsg)
1282
    goto err;
1283 1284 1285 1286

  /* as we are here, we tried to allocate the event */
  if (!mi->rli.relay_log.description_event_for_queue)
  {
1287 1288 1289 1290
    errmsg= "default Format_description_log_event";
    err_code= ER_SLAVE_CREATE_EVENT_FAILURE;
    sprintf(err_buff, ER(err_code), errmsg);
    goto err;
1291 1292
  }

1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334
  /*
    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); 
1335 1336 1337 1338
  /*
    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
1339

1340 1341 1342 1343 1344
  DBUG_EXECUTE_IF("dbug.before_get_UNIX_TIMESTAMP",
                  {
                    const char act[]=
                      "now "
                      "wait_for signal.get_unix_timestamp";
1345
                    DBUG_ASSERT(debug_sync_service);
1346 1347 1348 1349
                    DBUG_ASSERT(!debug_sync_set_action(current_thd,
                                                       STRING_WITH_LEN(act)));
                  };);

1350
  master_res= NULL;
1351
  if (!mysql_real_query(mysql, STRING_WITH_LEN("SELECT UNIX_TIMESTAMP()")) &&
1352 1353
      (master_res= mysql_store_result(mysql)) &&
      (master_row= mysql_fetch_row(master_res)))
unknown's avatar
unknown committed
1354
  {
unknown's avatar
unknown committed
1355
    mi->clock_diff_with_master=
1356
      (long) (time((time_t*) 0) - strtoul(master_row[0], 0, 10));
unknown's avatar
unknown committed
1357
  }
1358 1359
  else if (check_io_slave_killed(mi->io_thd, mi, NULL))
    goto slave_killed_err;
1360 1361 1362 1363 1364 1365 1366
  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
1367
  {
1368
    mi->clock_diff_with_master= 0; /* The "most sensible" value */
1369 1370 1371 1372
    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));
1373 1374
  }
  if (master_res)
1375
  {
unknown's avatar
unknown committed
1376
    mysql_free_result(master_res);
1377 1378
    master_res= NULL;
  }
unknown's avatar
unknown committed
1379

1380 1381 1382 1383 1384 1385 1386 1387 1388 1389
  /*
    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.
  */
1390 1391 1392 1393 1394
  DBUG_EXECUTE_IF("dbug.before_get_SERVER_ID",
                  {
                    const char act[]=
                      "now "
                      "wait_for signal.get_server_id";
1395
                    DBUG_ASSERT(debug_sync_service);
1396 1397 1398
                    DBUG_ASSERT(!debug_sync_set_action(current_thd, 
                                                       STRING_WITH_LEN(act)));
                  };);
1399 1400
  master_res= NULL;
  master_row= NULL;
1401 1402
  if (!mysql_real_query(mysql,
                        STRING_WITH_LEN("SHOW VARIABLES LIKE 'SERVER_ID'")) &&
1403 1404
      (master_res= mysql_store_result(mysql)) &&
      (master_row= mysql_fetch_row(master_res)))
1405
  {
1406 1407
    if ((global_system_variables.server_id ==
             (mi->master_id= strtoul(master_row[1], 0, 10))) &&
1408
        !mi->rli.replicate_same_server_id)
1409
    {
1410 1411 1412 1413
      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).";
1414 1415
      err_code= ER_SLAVE_FATAL_ERROR;
      sprintf(err_buff, ER(err_code), errmsg);
1416
      goto err;
1417
    }
1418 1419 1420
  }
  else if (mysql_errno(mysql))
  {
1421 1422 1423
    if (check_io_slave_killed(mi->io_thd, mi, NULL))
      goto slave_killed_err;
    else if (is_network_error(mysql_errno(mysql)))
1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443
    {
      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)
  {
1444
    mysql_free_result(master_res);
1445
    master_res= NULL;
unknown's avatar
unknown committed
1446
  }
Andrei Elkin's avatar
Andrei Elkin committed
1447 1448 1449 1450 1451 1452 1453
  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
1454

1455 1456 1457
  /*
    Check that the master's global character_set_server and ours are the same.
    Not fatal if query fails (old master?).
1458 1459 1460 1461 1462 1463
    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.
1464 1465 1466 1467 1468 1469
    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.
1470
  */
1471 1472

  /* redundant with rest of code but safer against later additions */
1473
  if (version == 3)
1474
    goto err;
1475

1476
  if (version == 4)
unknown's avatar
unknown committed
1477
  {
1478 1479 1480 1481 1482
    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)))
1483
    {
1484 1485 1486
      if (strcmp(master_row[0], global_system_variables.collation_server->name))
      {
        errmsg= "The slave I/O thread stops because master and slave have \
1487
different values for the COLLATION_SERVER global variable. The values must \
1488 1489 1490 1491 1492
be equal for the Statement-format replication to work";
        err_code= ER_SLAVE_FATAL_ERROR;
        sprintf(err_buff, ER(err_code), errmsg);
        goto err;
      }
1493
    }
1494 1495
    else if (check_io_slave_killed(mi->io_thd, mi, NULL))
      goto slave_killed_err;
1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508
    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));
1509
      goto err;
1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521
    }
    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
1522
  }
1523

1524 1525 1526 1527 1528 1529 1530 1531
  /*
    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
1532 1533 1534
    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
1535
    'SYSTEM' time zone).
1536 1537
    This check is only necessary for 4.x masters (and < 5.0.4 masters but
    those were alpha).
1538
  */
1539
  if (version == 4)
unknown's avatar
unknown committed
1540
  {
1541 1542 1543 1544
    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)))
1545
    {
1546 1547 1548 1549
      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 \
1550
different values for the TIME_ZONE global variable. The values must \
1551 1552 1553 1554 1555
be equal for the Statement-format replication to work";
        err_code= ER_SLAVE_FATAL_ERROR;
        sprintf(err_buff, ER(err_code), errmsg);
        goto err;
      }
1556
    }
1557 1558
    else if (check_io_slave_killed(mi->io_thd, mi, NULL))
      goto slave_killed_err;
1559
    else if (is_network_error(err_code= mysql_errno(mysql)))
1560
    {
1561 1562 1563
      mi->report(ERROR_LEVEL, err_code,
                 "Get master TIME_ZONE failed with error: %s",
                 mysql_error(mysql));
1564
      goto network_err;
1565 1566 1567 1568 1569 1570 1571 1572 1573 1574
    }
    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");
    }
1575 1576 1577 1578 1579 1580
    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));
1581
      goto err;
1582 1583 1584 1585 1586 1587
    }
    if (master_res)
    {
      mysql_free_result(master_res);
      master_res= NULL;
    }
unknown's avatar
unknown committed
1588 1589
  }

Andrei Elkin's avatar
Andrei Elkin committed
1590 1591 1592 1593 1594 1595 1596 1597 1598
  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);
1599
    sprintf(query, query_format, llbuf);
Andrei Elkin's avatar
Andrei Elkin committed
1600 1601 1602 1603

    if (mysql_real_query(mysql, query, strlen(query))
        && !check_io_slave_killed(mi->io_thd, mi, NULL))
    {
Andrei Elkin's avatar
Andrei Elkin committed
1604 1605
      errmsg= "The slave I/O thread stops because SET @master_heartbeat_period "
        "on master failed.";
Andrei Elkin's avatar
Andrei Elkin committed
1606
      err_code= ER_SLAVE_FATAL_ERROR;
Andrei Elkin's avatar
Andrei Elkin committed
1607
      sprintf(err_buff, "%s Error: %s", errmsg, mysql_error(mysql));
Andrei Elkin's avatar
Andrei Elkin committed
1608 1609 1610 1611 1612 1613
      mysql_free_result(mysql_store_result(mysql));
      goto err;
    }
    mysql_free_result(mysql_store_result(mysql));
  }
 
1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 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
  /*
    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
1711

1712
  /*
1713
    Request the master to filter away events with the @@skip_replication flag
1714 1715
    set, if we are running with
    --replicate-events-marked-for-skip=FILTER_ON_MASTER.
1716
  */
1717
  if (opt_replicate_events_marked_for_skip == RPL_SKIP_FILTER_ON_MASTER)
1718
  {
1719
    if (mysql_real_query(mysql, STRING_WITH_LEN("SET skip_replication=1")))
1720 1721 1722 1723 1724
    {
      err_code= mysql_errno(mysql);
      if (is_network_error(err_code))
      {
        mi->report(ERROR_LEVEL, err_code,
1725
                   "Setting master-side filtering of @@skip_replication failed "
1726 1727 1728 1729 1730 1731 1732
                   "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
1733
          @@skip_replication feature.
1734
          This is not a problem, as such master will not generate events with
1735
          the @@skip_replication flag set in the first place. We will still
1736 1737
          do slave-side filtering of such events though, to handle the (rare)
          case of downgrading a master and receiving old events generated from
1738
          before the downgrade with the @@skip_replication flag set.
1739 1740
        */
        DBUG_PRINT("info", ("Old master does not support master-side filtering "
1741 1742 1743 1744 1745 1746 1747 1748 1749 1750
                            "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;
1751 1752 1753
      }
    }
  }
1754 1755 1756 1757

  /* Announce MariaDB slave capabilities. */
  DBUG_EXECUTE_IF("simulate_slave_capability_none", goto after_set_capability;);
  {
1758 1759 1760 1761 1762 1763
    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)
1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782
    {
      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;
      }
    }
  }
1783
#ifndef DBUG_OFF
1784
after_set_capability:
1785
#endif
1786

1787
err:
1788 1789
  if (errmsg)
  {
1790 1791
    if (master_res)
      mysql_free_result(master_res);
1792
    DBUG_ASSERT(err_code != 0);
1793
    mi->report(ERROR_LEVEL, err_code, "%s", err_buff);
1794
    DBUG_RETURN(1);
1795
  }
1796

1797
  DBUG_RETURN(0);
1798 1799 1800 1801 1802

network_err:
  if (master_res)
    mysql_free_result(master_res);
  DBUG_RETURN(2);
unknown's avatar
unknown committed
1803

1804 1805 1806 1807
slave_killed_err:
  if (master_res)
    mysql_free_result(master_res);
  DBUG_RETURN(2);
unknown's avatar
unknown committed
1808
}
unknown's avatar
unknown committed
1809

1810

1811
static bool wait_for_relay_log_space(Relay_log_info* rli)
unknown's avatar
unknown committed
1812
{
1813
  bool slave_killed=0;
1814
  Master_info* mi = rli->mi;
unknown's avatar
unknown committed
1815
  const char *save_proc_info;
unknown's avatar
unknown committed
1816 1817
  THD* thd = mi->io_thd;
  DBUG_ENTER("wait_for_relay_log_space");
1818

Marc Alff's avatar
Marc Alff committed
1819
  mysql_mutex_lock(&rli->log_space_lock);
unknown's avatar
unknown committed
1820
  save_proc_info= thd->enter_cond(&rli->log_space_cond,
unknown's avatar
unknown committed
1821 1822
                                  &rli->log_space_lock,
                                  "\
1823
Waiting for the slave SQL thread to free enough relay log space");
unknown's avatar
unknown committed
1824
  while (rli->log_space_limit < rli->log_space_total &&
unknown's avatar
unknown committed
1825
         !(slave_killed=io_slave_killed(thd,mi)) &&
1826
         !rli->ignore_log_space_limit)
Marc Alff's avatar
Marc Alff committed
1827
    mysql_cond_wait(&rli->log_space_cond, &rli->log_space_lock);
1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868

  /* 
    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)
    {
1869
      mysql_mutex_lock(&mi->data_lock);
1870
      rotate_relay_log(rli->mi);
1871
      mysql_mutex_unlock(&mi->data_lock);
1872 1873 1874 1875 1876 1877
      rli->sql_force_rotate_relay= false;
    }

    rli->ignore_log_space_limit= false;
  }

unknown's avatar
unknown committed
1878
  thd->exit_cond(save_proc_info);
unknown's avatar
unknown committed
1879 1880
  DBUG_RETURN(slave_killed);
}
1881

1882

1883 1884
/*
  Builds a Rotate from the ignored events' info and writes it to relay log.
1885

1886 1887 1888 1889 1890 1891 1892 1893 1894 1895
  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).
 */
1896
static void write_ignored_events_info_to_relay_log(THD *thd, Master_info *mi)
1897
{
1898
  Relay_log_info *rli= &mi->rli;
Marc Alff's avatar
Marc Alff committed
1899
  mysql_mutex_t *log_lock= rli->relay_log.get_log_lock();
1900 1901
  DBUG_ENTER("write_ignored_events_info_to_relay_log");

1902
  DBUG_ASSERT(thd == mi->io_thd);
Marc Alff's avatar
Marc Alff committed
1903
  mysql_mutex_lock(log_lock);
1904
  if (rli->ign_master_log_name_end[0])
unknown's avatar
unknown committed
1905
  {
1906
    DBUG_PRINT("info",("writing a Rotate event to track down ignored events"));
1907
    Rotate_log_event *ev= new Rotate_log_event(rli->ign_master_log_name_end,
1908 1909 1910 1911
                                               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
1912
    mysql_mutex_unlock(log_lock);
1913 1914 1915 1916
    if (likely((bool)ev))
    {
      ev->server_id= 0; // don't be ignored by slave SQL thread
      if (unlikely(rli->relay_log.append(ev)))
1917 1918 1919 1920 1921
        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");
1922
      rli->relay_log.harvest_bytes_written(&rli->log_space_total);
Luis Soares's avatar
Luis Soares committed
1923
      if (flush_master_info(mi, TRUE, TRUE))
1924
        sql_print_error("Failed to flush master info file");
1925 1926 1927
      delete ev;
    }
    else
1928 1929 1930 1931
      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
1932
  }
1933
  else
Marc Alff's avatar
Marc Alff committed
1934
    mysql_mutex_unlock(log_lock);
1935
  DBUG_VOID_RETURN;
unknown's avatar
unknown committed
1936 1937
}

1938

1939
int register_slave_on_master(MYSQL* mysql, Master_info *mi,
1940
                             bool *suppress_warnings)
1941
{
1942
  uchar buf[1024], *pos= buf;
1943
  uint report_host_len=0, report_user_len=0, report_password_len=0;
1944
  DBUG_ENTER("register_slave_on_master");
1945

1946
  *suppress_warnings= FALSE;
1947 1948 1949 1950 1951 1952 1953 1954
  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);
1955
    DBUG_RETURN(0);
1956 1957
  }

1958
  if (report_user)
1959
    report_user_len= strlen(report_user);
1960 1961 1962 1963 1964 1965 1966 1967 1968
  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
1969
  if (report_password)
1970
    report_password_len= strlen(report_password);
1971 1972 1973 1974 1975 1976 1977 1978
  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);
  }
1979

1980
  int4store(pos, global_system_variables.server_id); pos+= 4;
1981 1982 1983
  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);
1984
  int2store(pos, (uint16) report_port); pos+= 2;
1985 1986 1987 1988 1989 1990
  /* 
    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;
1991
  /* The master will fill in master_id */
unknown's avatar
unknown committed
1992
  int4store(pos, 0);                    pos+= 4;
1993

1994
  if (simple_command(mysql, COM_REGISTER_SLAVE, buf, (size_t) (pos- buf), 0))
1995
  {
1996 1997 1998 1999
    if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
    {
      *suppress_warnings= TRUE;                 // Suppress reconnect warning
    }
2000
    else if (!check_io_slave_killed(mi->io_thd, mi, NULL))
2001 2002 2003 2004 2005 2006 2007
    {
      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);
    }
2008
    DBUG_RETURN(1);
2009
  }
2010
  DBUG_RETURN(0);
2011 2012
}

2013

2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024
/**
  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
*/
2025 2026 2027 2028 2029

bool show_master_info(THD *thd, Master_info *mi, bool full)
{
  DBUG_ENTER("show_master_info");

2030
  if (send_show_master_info_header(thd, full))
2031 2032 2033 2034 2035 2036 2037
    DBUG_RETURN(TRUE);
  if (send_show_master_info_data(thd, mi, full))
    DBUG_RETURN(TRUE);
  my_eof(thd);
  DBUG_RETURN(FALSE);
}

2038
static bool send_show_master_info_header(THD *thd, bool full)
unknown's avatar
unknown committed
2039 2040
{
  List<Item> field_list;
2041
  Protocol *protocol= thd->protocol;
2042
  Master_info *mi;
2043 2044 2045
  DBUG_ENTER("show_master_info_header");

  if (full)
2046
  {
2047 2048
    field_list.push_back(new Item_empty_string("Connection_name",
                                               MAX_CONNECTION_NAME));
2049 2050 2051
    field_list.push_back(new Item_empty_string("Slave_SQL_State",
                                               30));
  }
2052

unknown's avatar
unknown committed
2053
  field_list.push_back(new Item_empty_string("Slave_IO_State",
2054
                                                     30));
unknown's avatar
unknown committed
2055
  field_list.push_back(new Item_empty_string("Master_Host",
unknown's avatar
unknown committed
2056
                                                     sizeof(mi->host)));
unknown's avatar
unknown committed
2057
  field_list.push_back(new Item_empty_string("Master_User",
unknown's avatar
unknown committed
2058
                                                     sizeof(mi->user)));
2059
  field_list.push_back(new Item_return_int("Master_Port", 7,
unknown's avatar
unknown committed
2060
                                           MYSQL_TYPE_LONG));
2061
  field_list.push_back(new Item_return_int("Connect_Retry", 10,
unknown's avatar
unknown committed
2062
                                           MYSQL_TYPE_LONG));
2063
  field_list.push_back(new Item_empty_string("Master_Log_File",
unknown's avatar
unknown committed
2064
                                             FN_REFLEN));
2065
  field_list.push_back(new Item_return_int("Read_Master_Log_Pos", 10,
unknown's avatar
unknown committed
2066
                                           MYSQL_TYPE_LONGLONG));
2067
  field_list.push_back(new Item_empty_string("Relay_Log_File",
unknown's avatar
unknown committed
2068
                                             FN_REFLEN));
2069
  field_list.push_back(new Item_return_int("Relay_Log_Pos", 10,
unknown's avatar
unknown committed
2070
                                           MYSQL_TYPE_LONGLONG));
2071
  field_list.push_back(new Item_empty_string("Relay_Master_Log_File",
unknown's avatar
unknown committed
2072
                                             FN_REFLEN));
2073 2074
  field_list.push_back(new Item_empty_string("Slave_IO_Running", 3));
  field_list.push_back(new Item_empty_string("Slave_SQL_Running", 3));
2075 2076 2077 2078 2079 2080
  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
2081
                                             28));
2082 2083 2084
  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
2085
                                           MYSQL_TYPE_LONG));
2086
  field_list.push_back(new Item_return_int("Exec_Master_Log_Pos", 10,
unknown's avatar
unknown committed
2087
                                           MYSQL_TYPE_LONGLONG));
2088
  field_list.push_back(new Item_return_int("Relay_Log_Space", 10,
unknown's avatar
unknown committed
2089
                                           MYSQL_TYPE_LONGLONG));
2090
  field_list.push_back(new Item_empty_string("Until_Condition", 6));
2091
  field_list.push_back(new Item_empty_string("Until_Log_File", FN_REFLEN));
unknown's avatar
unknown committed
2092
  field_list.push_back(new Item_return_int("Until_Log_Pos", 10,
2093
                                           MYSQL_TYPE_LONGLONG));
unknown's avatar
unknown committed
2094 2095 2096
  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
2097
  field_list.push_back(new Item_empty_string("Master_SSL_CA_Path",
unknown's avatar
unknown committed
2098
                                             sizeof(mi->ssl_capath)));
unknown's avatar
unknown committed
2099
  field_list.push_back(new Item_empty_string("Master_SSL_Cert",
unknown's avatar
unknown committed
2100
                                             sizeof(mi->ssl_cert)));
unknown's avatar
unknown committed
2101
  field_list.push_back(new Item_empty_string("Master_SSL_Cipher",
unknown's avatar
unknown committed
2102
                                             sizeof(mi->ssl_cipher)));
unknown's avatar
unknown committed
2103
  field_list.push_back(new Item_empty_string("Master_SSL_Key",
unknown's avatar
unknown committed
2104
                                             sizeof(mi->ssl_key)));
2105
  field_list.push_back(new Item_return_int("Seconds_Behind_Master", 10,
unknown's avatar
unknown committed
2106
                                           MYSQL_TYPE_LONGLONG));
2107 2108
  field_list.push_back(new Item_empty_string("Master_SSL_Verify_Server_Cert",
                                             3));
2109 2110 2111 2112
  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
2113 2114 2115 2116
  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));
2117 2118 2119 2120 2121 2122 2123 2124
  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));
2125 2126 2127 2128
    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));
2129
  }
2130
  if (protocol->send_result_set_metadata(&field_list,
2131
                            Protocol::SEND_NUM_ROWS | Protocol::SEND_EOF))
unknown's avatar
unknown committed
2132
    DBUG_RETURN(TRUE);
2133 2134 2135 2136 2137 2138 2139
  DBUG_RETURN(FALSE);
}


static bool send_show_master_info_data(THD *thd, Master_info *mi, bool full)
{
  DBUG_ENTER("send_show_master_info_data");
unknown's avatar
unknown committed
2140

2141 2142
  if (mi->host[0])
  {
2143
    DBUG_PRINT("info",("host is set: '%s'", mi->host));
2144
    String *packet= &thd->packet;
2145
    Protocol *protocol= thd->protocol;
2146 2147
    char buf[256];
    String tmp(buf, sizeof(buf), &my_charset_bin);
2148

2149
    protocol->prepare_for_resend();
unknown's avatar
unknown committed
2150

2151
    /*
2152 2153
      slave_running can be accessed without run_lock but not other
      non-volotile members like mi->io_thd, which is guarded by the mutex.
2154
    */
2155 2156 2157
    if (full)
      protocol->store(mi->connection_name.str, mi->connection_name.length,
                      &my_charset_bin);
Marc Alff's avatar
Marc Alff committed
2158
    mysql_mutex_lock(&mi->run_lock);
2159 2160 2161
    if (full)
      protocol->store(mi->rli.sql_thd ? mi->rli.sql_thd->proc_info : "",
                      &my_charset_bin);
2162
    protocol->store(mi->io_thd ? mi->io_thd->proc_info : "", &my_charset_bin);
Marc Alff's avatar
Marc Alff committed
2163
    mysql_mutex_unlock(&mi->run_lock);
2164

Marc Alff's avatar
Marc Alff committed
2165 2166 2167 2168
    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);
2169 2170
    protocol->store(mi->host, &my_charset_bin);
    protocol->store(mi->user, &my_charset_bin);
2171 2172
    protocol->store((uint32) mi->port);
    protocol->store((uint32) mi->connect_retry);
2173
    protocol->store(mi->master_log_name, &my_charset_bin);
2174
    protocol->store((ulonglong) mi->master_log_pos);
2175
    protocol->store(mi->rli.group_relay_log_name +
unknown's avatar
unknown committed
2176 2177
                    dirname_length(mi->rli.group_relay_log_name),
                    &my_charset_bin);
2178 2179
    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
2180
    protocol->store(mi->slave_running == MYSQL_SLAVE_RUN_CONNECT ?
2181 2182
                    "Yes" : (mi->slave_running == MYSQL_SLAVE_RUN_NOT_CONNECT ?
                             "Connecting" : "No"), &my_charset_bin);
2183
    protocol->store(mi->rli.slave_running ? "Yes":"No", &my_charset_bin);
2184 2185 2186 2187
    protocol->store(rpl_filter->get_do_db());
    protocol->store(rpl_filter->get_ignore_db());

    rpl_filter->get_do_table(&tmp);
2188
    protocol->store(&tmp);
2189
    rpl_filter->get_ignore_table(&tmp);
2190
    protocol->store(&tmp);
2191
    rpl_filter->get_wild_do_table(&tmp);
2192
    protocol->store(&tmp);
2193
    rpl_filter->get_wild_ignore_table(&tmp);
2194 2195
    protocol->store(&tmp);

2196 2197
    protocol->store(mi->rli.last_error().number);
    protocol->store(mi->rli.last_error().message, &my_charset_bin);
2198
    protocol->store((uint32) mi->rli.slave_skip_counter);
2199
    protocol->store((ulonglong) mi->rli.group_master_log_pos);
2200
    protocol->store((ulonglong) mi->rli.log_space_total);
2201 2202

    protocol->store(
2203 2204
      mi->rli.until_condition==Relay_log_info::UNTIL_NONE ? "None":
        ( mi->rli.until_condition==Relay_log_info::UNTIL_MASTER_POS? "Master":
2205 2206 2207
          "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
2208 2209

#ifdef HAVE_OPENSSL
unknown's avatar
unknown committed
2210 2211 2212 2213 2214 2215 2216 2217 2218
    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
2219

2220 2221 2222 2223 2224 2225
    /*
      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)
2226
    {
2227
      long time_diff= ((long)(time(0) - mi->rli.last_master_timestamp)
unknown's avatar
unknown committed
2228
                       - mi->clock_diff_with_master);
2229
      /*
unknown's avatar
unknown committed
2230 2231
        Apparently on some systems time_diff can be <0. Here are possible
        reasons related to MySQL:
2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243
        - 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.
2244 2245 2246 2247
        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".
2248
      */
unknown's avatar
unknown committed
2249 2250
      protocol->store((longlong)(mi->rli.last_master_timestamp ?
                                 max(0, time_diff) : 0));
2251
    }
unknown's avatar
unknown committed
2252
    else
2253
    {
unknown's avatar
unknown committed
2254
      protocol->store_null();
2255 2256
    }
    protocol->store(mi->ssl_verify_server_cert? "Yes":"No", &my_charset_bin);
unknown's avatar
unknown committed
2257

2258
    // Last_IO_Errno
2259
    protocol->store(mi->last_error().number);
2260
    // Last_IO_Error
2261
    protocol->store(mi->last_error().message, &my_charset_bin);
2262
    // Last_SQL_Errno
2263
    protocol->store(mi->rli.last_error().number);
2264
    // Last_SQL_Error
2265
    protocol->store(mi->rli.last_error().message, &my_charset_bin);
Andrei Elkin's avatar
Andrei Elkin committed
2266 2267 2268 2269 2270 2271 2272 2273 2274 2275
    // 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);
2276
        slen= sprintf(sbuff, (i==0? "%lu" : ", %lu"), s_id);
Andrei Elkin's avatar
Andrei Elkin committed
2277 2278 2279 2280 2281 2282
        if (cur_len + slen + 4 > FN_REFLEN)
        {
          /*
            break the loop whenever remained space could not fit
            ellipses on the next cycle
          */
2283
          sprintf(buff + cur_len, "...");
Andrei Elkin's avatar
Andrei Elkin committed
2284 2285
          break;
        }
2286
        cur_len += sprintf(buff + cur_len, "%s", sbuff);
Andrei Elkin's avatar
Andrei Elkin committed
2287 2288 2289 2290 2291
      }
      protocol->store(buff, &my_charset_bin);
    }
    // Master_Server_id
    protocol->store((uint32) mi->master_id);
2292 2293 2294 2295 2296
    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);
2297 2298
      protocol->store((uint32)    mi->received_heartbeats);
      protocol->store((double)    mi->heartbeat_period, 3, &tmp);
2299
    }
2300

Marc Alff's avatar
Marc Alff committed
2301 2302 2303 2304
    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);
2305

2306
    if (my_net_write(&thd->net, (uchar*) thd->packet.ptr(), packet->length()))
unknown's avatar
unknown committed
2307
      DBUG_RETURN(TRUE);
2308
  }
2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332
  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
2333 2334 2335

  @note
  master_info_index is protected by LOCK_active_mi.
2336 2337 2338 2339 2340 2341 2342
*/

bool show_all_master_info(THD* thd)
{
  uint i, elements;
  Master_info **tmp;
  DBUG_ENTER("show_master_info");
2343
  mysql_mutex_assert_owner(&LOCK_active_mi);
2344

2345
  if (send_show_master_info_header(thd, 1))
2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371
    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++)
  {
    if (send_show_master_info_data(thd, tmp[i], 1))
      DBUG_RETURN(TRUE);
  }

end:
2372
  my_eof(thd);
unknown's avatar
unknown committed
2373
  DBUG_RETURN(FALSE);
unknown's avatar
unknown committed
2374 2375
}

unknown's avatar
unknown committed
2376

2377 2378
void set_slave_thread_options(THD* thd)
{
2379
  DBUG_ENTER("set_slave_thread_options");
2380 2381 2382 2383 2384 2385 2386 2387 2388
  /*
     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.
  */
2389
  ulonglong options= thd->variables.option_bits | OPTION_BIG_SELECTS;
2390 2391 2392 2393
  if (opt_log_slave_updates)
    options|= OPTION_BIN_LOG;
  else
    options&= ~OPTION_BIN_LOG;
2394
  thd->variables.option_bits= options;
2395
  thd->variables.completion_type= 0;
2396
  DBUG_VOID_RETURN;
2397
}
2398

2399
void set_slave_thread_default_charset(THD* thd, Relay_log_info const *rli)
2400
{
2401 2402
  DBUG_ENTER("set_slave_thread_default_charset");

2403 2404 2405 2406 2407 2408 2409
  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();
2410 2411 2412 2413 2414 2415 2416

  /*
    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.
   */
2417
  const_cast<Relay_log_info*>(rli)->cached_charset_invalidate();
2418
  DBUG_VOID_RETURN;
2419 2420
}

unknown's avatar
unknown committed
2421
/*
2422
  init_slave_thread()
unknown's avatar
unknown committed
2423
*/
2424

2425 2426
static int init_slave_thread(THD* thd, Master_info *mi,
                             SLAVE_THD_TYPE thd_type)
unknown's avatar
unknown committed
2427 2428
{
  DBUG_ENTER("init_slave_thread");
Sergei Golubchik's avatar
Sergei Golubchik committed
2429
  int simulate_error __attribute__((unused))= 0;
2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442
  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);
  }

2443
  thd->system_thread = (thd_type == SLAVE_THD_SQL) ?
unknown's avatar
unknown committed
2444
    SYSTEM_THREAD_SLAVE_SQL : SYSTEM_THREAD_SLAVE_IO;
2445
  thd->security_ctx->skip_grants();
2446 2447 2448 2449 2450
/*
  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.
*/
2451
  thd->variables.max_allowed_packet= slave_max_allowed_packet;
2452 2453
  thd->slave_thread= 1;
  thd->connection_name= mi->connection_name;
2454
  thd->enable_slow_log= opt_log_slow_slave_statements;
2455
  thd->variables.log_slow_filter= global_system_variables.log_slow_filter;
2456
  set_slave_thread_options(thd);
unknown's avatar
unknown committed
2457
  thd->client_capabilities = CLIENT_LOCAL_FILES;
Marc Alff's avatar
Marc Alff committed
2458
  mysql_mutex_lock(&LOCK_thread_count);
unknown's avatar
unknown committed
2459
  thd->thread_id= thd->variables.pseudo_thread_id= thread_id++;
Marc Alff's avatar
Marc Alff committed
2460
  mysql_mutex_unlock(&LOCK_thread_count);
unknown's avatar
unknown committed
2461

2462
  if (thd_type == SLAVE_THD_SQL)
2463
    thd_proc_info(thd, "Waiting for the next event in relay log");
2464
  else
2465
    thd_proc_info(thd, "Waiting for master update");
unknown's avatar
unknown committed
2466
  thd->set_time();
2467 2468
  /* Do not use user-supplied timeout value for system threads. */
  thd->variables.lock_wait_timeout= LONG_TIMEOUT;
unknown's avatar
unknown committed
2469 2470 2471
  DBUG_RETURN(0);
}

2472 2473 2474 2475 2476 2477 2478
/*
  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.
2479

2480 2481 2482 2483 2484
  @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
2485
{
2486

2487 2488 2489
  bool ret;
  struct timespec abstime;
  const char *old_proc_info;
unknown's avatar
unknown committed
2490

2491 2492
  mysql_mutex_t *lock= &info->sleep_lock;
  mysql_cond_t *cond= &info->sleep_cond;
unknown's avatar
unknown committed
2493

2494 2495 2496 2497 2498 2499 2500 2501 2502 2503
  /* 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
2504
  }
2505 2506 2507
  /* Implicitly unlocks the mutex. */
  thd->exit_cond(old_proc_info);
  return ret;
unknown's avatar
unknown committed
2508 2509
}

2510

He Zhenxing's avatar
He Zhenxing committed
2511 2512
static int request_dump(THD *thd, MYSQL* mysql, Master_info* mi,
			bool *suppress_warnings)
unknown's avatar
unknown committed
2513
{
2514
  uchar buf[FN_REFLEN + 10];
unknown's avatar
unknown committed
2515
  int len;
He Zhenxing's avatar
He Zhenxing committed
2516
  ushort binlog_flags = 0; // for now
2517
  char* logname = mi->master_log_name;
2518
  DBUG_ENTER("request_dump");
2519 2520
  
  *suppress_warnings= FALSE;
2521

2522
  if (opt_log_slave_updates && opt_replicate_annotate_row_events)
2523 2524
    binlog_flags|= BINLOG_SEND_ANNOTATE_ROWS_EVENT;

He Zhenxing's avatar
He Zhenxing committed
2525 2526 2527 2528 2529
  if (RUN_HOOK(binlog_relay_io,
               before_request_transmit,
               (thd, mi, binlog_flags)))
    DBUG_RETURN(1);
  
unknown's avatar
unknown committed
2530
  // TODO if big log files: Change next to int8store()
unknown's avatar
unknown committed
2531
  int4store(buf, (ulong) mi->master_log_pos);
unknown's avatar
unknown committed
2532
  int2store(buf + 4, binlog_flags);
2533
  int4store(buf + 6, global_system_variables.server_id);
unknown's avatar
unknown committed
2534
  len = (uint) strlen(logname);
2535
  memcpy(buf + 10, logname,len);
unknown's avatar
SCRUM  
unknown committed
2536
  if (simple_command(mysql, COM_BINLOG_DUMP, buf, len + 10, 1))
unknown's avatar
unknown committed
2537
  {
unknown's avatar
unknown committed
2538 2539 2540 2541 2542
    /*
      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
2543
    if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
2544
      *suppress_warnings= TRUE;                 // Suppress reconnect warning
unknown's avatar
unknown committed
2545
    else
2546
      sql_print_error("Error on COM_BINLOG_DUMP: %d  %s, will retry in %d secs",
unknown's avatar
unknown committed
2547
                      mysql_errno(mysql), mysql_error(mysql),
2548
                      mi->connect_retry);
2549
    DBUG_RETURN(1);
unknown's avatar
unknown committed
2550
  }
unknown's avatar
unknown committed
2551

2552
  DBUG_RETURN(0);
unknown's avatar
unknown committed
2553 2554
}

2555

unknown's avatar
unknown committed
2556
/*
2557
  Read one event from the master
unknown's avatar
unknown committed
2558

unknown's avatar
unknown committed
2559 2560
  SYNOPSIS
    read_event()
unknown's avatar
unknown committed
2561 2562 2563 2564 2565 2566
    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.
2567

unknown's avatar
unknown committed
2568
    RETURN VALUES
unknown's avatar
unknown committed
2569 2570
    'packet_error'      Error
    number              Length of packet
unknown's avatar
unknown committed
2571 2572
*/

2573
static ulong read_event(MYSQL* mysql, Master_info *mi, bool* suppress_warnings)
unknown's avatar
unknown committed
2574
{
2575
  ulong len;
2576
  DBUG_ENTER("read_event");
unknown's avatar
unknown committed
2577

2578
  *suppress_warnings= FALSE;
unknown's avatar
unknown committed
2579 2580 2581 2582
  /*
    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
2583
#ifndef DBUG_OFF
2584
  if (disconnect_slave_event_count && !(mi->events_till_disconnect--))
unknown's avatar
unknown committed
2585
    DBUG_RETURN(packet_error);
unknown's avatar
unknown committed
2586
#endif
unknown's avatar
unknown committed
2587

2588
  len = cli_safe_read(mysql);
2589
  if (len == packet_error || (long) len < 1)
unknown's avatar
unknown committed
2590
  {
unknown's avatar
SCRUM  
unknown committed
2591
    if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
unknown's avatar
unknown committed
2592 2593
    {
      /*
unknown's avatar
unknown committed
2594 2595 2596
        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
2597 2598 2599 2600
      */
      *suppress_warnings= TRUE;
    }
    else
2601
      sql_print_error("Error reading packet from server: %s ( server_errno=%d)",
unknown's avatar
unknown committed
2602
                      mysql_error(mysql), mysql_errno(mysql));
2603
    DBUG_RETURN(packet_error);
unknown's avatar
unknown committed
2604 2605
  }

2606 2607
  /* Check if eof packet */
  if (len < 8 && mysql->net.read_pos[0] == 254)
unknown's avatar
unknown committed
2608
  {
2609 2610
    sql_print_information("Slave: received end packet from server, apparent "
                          "master shutdown: %s",
unknown's avatar
unknown committed
2611
                     mysql_error(mysql));
2612
     DBUG_RETURN(packet_error);
unknown's avatar
unknown committed
2613
  }
unknown's avatar
unknown committed
2614

unknown's avatar
unknown committed
2615
  DBUG_PRINT("exit", ("len: %lu  net->read_pos[4]: %d",
unknown's avatar
unknown committed
2616 2617
                      len, mysql->net.read_pos[4]));
  DBUG_RETURN(len - 1);
unknown's avatar
unknown committed
2618 2619
}

unknown's avatar
unknown committed
2620 2621 2622 2623 2624 2625 2626 2627 2628
/*
  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)
{
2629 2630
  DBUG_ENTER("has_temporary_error");

2631
  DBUG_EXECUTE_IF("all_errors_are_temporary_errors",
Marc Alff's avatar
Marc Alff committed
2632
                  if (thd->stmt_da->is_error())
2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644
                  {
                    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
2645 2646 2647 2648 2649 2650

  /*
    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
2651 2652
  if (thd->stmt_da->sql_errno() == ER_LOCK_DEADLOCK ||
      thd->stmt_da->sql_errno() == ER_LOCK_WAIT_TIMEOUT)
2653
    DBUG_RETURN(1);
unknown's avatar
unknown committed
2654 2655 2656 2657 2658

#ifdef HAVE_NDB_BINLOG
  /*
    currently temporary error set in ndbcluster
  */
Marc Alff's avatar
Marc Alff committed
2659
  List_iterator_fast<MYSQL_ERROR> it(thd->warning_info->warn_list());
unknown's avatar
unknown committed
2660 2661 2662
  MYSQL_ERROR *err;
  while ((err= it++))
  {
Marc Alff's avatar
Marc Alff committed
2663 2664 2665
    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
2666 2667
    {
    case ER_GET_TEMPORARY_ERRMSG:
2668
      DBUG_RETURN(1);
unknown's avatar
unknown committed
2669 2670 2671 2672 2673
    default:
      break;
    }
  }
#endif
2674
  DBUG_RETURN(0);
unknown's avatar
unknown committed
2675
}
2676

2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703

/**
  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().
*/
2704
int apply_event_and_update_pos(Log_event* ev, THD* thd, Relay_log_info* rli)
2705 2706 2707 2708 2709 2710 2711 2712 2713
{
  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",
2714 2715
                      FLAGSTR(thd->variables.option_bits, OPTION_NOT_AUTOCOMMIT),
                      FLAGSTR(thd->variables.option_bits, OPTION_BEGIN),
2716
                      (ulong) rli->last_event_start_time));
2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741

  /*
    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).
  */

2742 2743
  /* Use the original server id for logging. */
  thd->variables.server_id = ev->server_id;
2744 2745 2746
  thd->set_time();                            // time the query
  thd->lex->current_select= 0;
  if (!ev->when)
2747 2748
  {
    my_hrtime_t hrtime= my_hrtime();
2749
    ev->when= hrtime_to_my_time(hrtime);
2750 2751
    ev->when_sec_part= hrtime_sec_part(hrtime);
  }
2752 2753
  thd->variables.option_bits=
    (thd->variables.option_bits & ~OPTION_SKIP_REPLICATION) |
2754
    (ev->flags & LOG_EVENT_SKIP_REPLICATION_F ? OPTION_SKIP_REPLICATION : 0);
2755 2756
  ev->thd = thd; // because up to this point, ev->thd == 0

2757 2758
  int reason= ev->shall_skip(rli);
  if (reason == Log_event::EVENT_SKIP_COUNT)
2759 2760 2761 2762
  {
    DBUG_ASSERT(rli->slave_skip_counter > 0);
    rli->slave_skip_counter--;
  }
Marc Alff's avatar
Marc Alff committed
2763
  mysql_mutex_unlock(&rli->data_lock);
2764 2765 2766
  if (reason == Log_event::EVENT_SKIP_NOT)
    exec_res= ev->apply_event(rli);

2767
#ifndef DBUG_OFF
2768 2769
  /*
    This only prints information to the debug trace.
2770

2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781
    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",
2782
                      test(thd->variables.option_bits & OPTION_BEGIN),
2783 2784 2785
                      rli->get_flag(Relay_log_info::IN_STMT)));
  DBUG_PRINT("skip_event", ("%s event was %s",
                            ev->get_type_str(), explain[reason]));
2786 2787 2788 2789 2790 2791
#endif

  DBUG_PRINT("info", ("apply_event error = %d", exec_res));
  if (exec_res == 0)
  {
    int error= ev->update_pos(rli);
2792
#ifdef HAVE_valgrind
2793 2794 2795
    if (!rli->is_fake)
#endif
    {
2796
#ifndef DBUG_OFF
2797
      char buf[22];
2798
#endif
2799 2800 2801 2802 2803 2804 2805 2806
      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));
    }
2807 2808 2809 2810 2811 2812 2813 2814 2815
    /*
      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)
    {
2816
      char buf[22];
2817 2818 2819 2820 2821 2822 2823 2824 2825 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
      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
2852
     if a failure occurs when applying the event.)
2853 2854 2855 2856 2857 2858 2859

   - An error occurred when updating the binlog position.

  @retval 0 The event was applied.

  @retval 1 The event was not applied.
*/
2860
static int exec_relay_log_event(THD* thd, Relay_log_info* rli)
unknown's avatar
unknown committed
2861
{
2862 2863
  DBUG_ENTER("exec_relay_log_event");

2864 2865
  /*
     We acquire this mutex since we need it for all operations except
2866
     event execution. But we will release it in places where we will
2867 2868
     wait for something for example inside of next_event().
   */
Marc Alff's avatar
Marc Alff committed
2869
  mysql_mutex_lock(&rli->data_lock);
2870

2871
  Log_event * ev = next_event(rli);
2872

2873
  DBUG_ASSERT(rli->sql_thd==thd);
2874

2875
  if (sql_slave_killed(thd,rli))
2876
  {
Marc Alff's avatar
Marc Alff committed
2877
    mysql_mutex_unlock(&rli->data_lock);
unknown's avatar
unknown committed
2878
    delete ev;
2879
    DBUG_RETURN(1);
2880
  }
2881 2882
  if (ev)
  {
2883
    int exec_res;
2884

2885 2886 2887 2888
    /*
      This tests if the position of the beginning of the current event
      hits the UNTIL barrier.
    */
2889
    if (rli->until_condition != Relay_log_info::UNTIL_NONE &&
2890
        rli->is_until_satisfied(thd, ev))
2891 2892 2893 2894 2895 2896 2897 2898 2899
    {
      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
2900
      mysql_mutex_unlock(&rli->data_lock);
2901
      delete ev;
2902
      DBUG_RETURN(1);
2903
    }
2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917

    { /**
         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
2918
                        mysql_mutex_unlock(&rli->data_lock);
2919 2920 2921 2922 2923 2924
                        delete ev;
                        rli->inc_event_relay_log_pos();
                        DBUG_RETURN(0);
                      };);
    }

2925
    exec_res= apply_event_and_update_pos(ev, thd, rli);
2926

2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956
    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
2957 2958
        if (!rli->is_deferred_event(ev))
          delete ev;
2959
        break;
2960 2961
    }

2962

2963
    /*
2964 2965
      update_log_pos failed: this should not happen, so we don't
      retry.
2966
    */
2967 2968 2969
    if (exec_res == 2)
      DBUG_RETURN(1);

2970 2971
    if (slave_trans_retries)
    {
2972
      int temp_err;
Michael Widenius's avatar
Michael Widenius committed
2973
      LINT_INIT(temp_err);
2974
      if (exec_res && (temp_err= has_temporary_error(thd)))
2975 2976 2977 2978
      {
        const char *errmsg;
        /*
          We were in a transaction which has been rolled back because of a
unknown's avatar
unknown committed
2979 2980
          temporary error;
          let's seek back to BEGIN log event and retry it all again.
2981
	  Note, if lock wait timeout (innodb_lock_wait_timeout exceeded)
2982
	  there is no rollback since 5.0.13 (ref: manual).
2983 2984 2985 2986 2987 2988 2989 2990
          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.
        */
2991
        if (rli->trans_retries < slave_trans_retries)
2992 2993 2994 2995 2996 2997
        {
          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,
2998
                                      1, &errmsg, 1))
2999 3000 3001 3002 3003
            sql_print_error("Error initializing relay log position: %s",
                            errmsg);
          else
          {
            exec_res= 0;
3004
            rli->cleanup_context(thd, 1);
unknown's avatar
unknown committed
3005
            /* chance for concurrent connection to get more locks */
3006 3007
            slave_sleep(thd, min(rli->trans_retries, MAX_SLAVE_RETRY_PAUSE),
                       sql_slave_killed, rli);
Marc Alff's avatar
Marc Alff committed
3008
            mysql_mutex_lock(&rli->data_lock); // because of SHOW STATUS
unknown's avatar
unknown committed
3009
            rli->trans_retries++;
3010
            rli->retried_trans++;
3011
            statistic_increment(slave_retried_transactions, LOCK_status);
Marc Alff's avatar
Marc Alff committed
3012
            mysql_mutex_unlock(&rli->data_lock);
3013 3014
            DBUG_PRINT("info", ("Slave retries transaction "
                                "rli->trans_retries: %lu", rli->trans_retries));
unknown's avatar
unknown committed
3015
          }
3016 3017 3018 3019 3020 3021 3022
        }
        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);
      }
3023
      else if ((exec_res && !temp_err) ||
3024 3025
               (opt_using_transactions &&
                rli->group_relay_log_pos == rli->event_relay_log_pos))
3026 3027
      {
        /*
3028 3029 3030
          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
3031 3032 3033
          non-transient error, the slave will stop with an error.
         */
        rli->trans_retries= 0; // restart from fresh
unknown's avatar
unknown committed
3034
        DBUG_PRINT("info", ("Resetting retry counter, rli->trans_retries: %lu",
3035
                            rli->trans_retries));
3036
      }
3037 3038
    }
    DBUG_RETURN(exec_res);
3039
  }
Marc Alff's avatar
Marc Alff committed
3040
  mysql_mutex_unlock(&rli->data_lock);
3041 3042
  rli->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_READ_FAILURE,
              ER(ER_SLAVE_RELAY_LOG_READ_FAILURE), "\
3043 3044 3045 3046 3047 3048 3049
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
3050
");
3051
  DBUG_RETURN(1);
unknown's avatar
unknown committed
3052 3053
}

3054

3055
static bool check_io_slave_killed(THD *thd, Master_info *mi, const char *info)
3056 3057 3058
{
  if (io_slave_killed(thd, mi))
  {
3059
    if (info && global_system_variables.log_warnings)
3060
      sql_print_information("%s", info);
3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091
    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.
*/

3092
static int try_to_reconnect(THD *thd, MYSQL *mysql, Master_info *mi,
3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105
                            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
3106
    slave_sleep(thd, mi->connect_retry, io_slave_killed, mi);
3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127
  }
  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
    {
3128
      sql_print_information("%s", buf);
3129 3130 3131 3132 3133
    }
  }
  if (safe_reconnect(thd, mysql, mi, 1) || io_slave_killed(thd, mi))
  {
    if (global_system_variables.log_warnings)
3134
      sql_print_information("%s", messages[SLAVE_RECON_MSG_KILLED_AFTER]);
3135 3136 3137 3138 3139 3140
    return 1;
  }
  return 0;
}


3141 3142 3143 3144 3145
/**
  Slave IO thread entry point.

  @param arg Pointer to Master_info struct that holds information for
  the IO thread.
3146

3147 3148
  @return Always 0.
*/
3149
pthread_handler_t handle_slave_io(void *arg)
unknown's avatar
unknown committed
3150
{
unknown's avatar
unknown committed
3151 3152
  THD *thd; // needs to be first for thread_stack
  MYSQL *mysql;
3153
  Master_info *mi = (Master_info*)arg;
3154
  Relay_log_info *rli= &mi->rli;
unknown's avatar
unknown committed
3155 3156
  char llbuff[22];
  uint retry_count;
3157
  bool suppress_warnings;
3158
  int ret;
3159 3160 3161
#ifndef DBUG_OFF
  uint retry_count_reg= 0, retry_count_dump= 0, retry_count_event= 0;
#endif
unknown's avatar
unknown committed
3162 3163
  // needs to call my_thread_init(), otherwise we get a coredump in DBUG_ stuff
  my_thread_init();
3164
  DBUG_ENTER("handle_slave_io");
unknown's avatar
unknown committed
3165

3166
  DBUG_ASSERT(mi->inited);
unknown's avatar
unknown committed
3167 3168 3169
  mysql= NULL ;
  retry_count= 0;

3170 3171
  thd= new THD; // note that contructor of THD uses DBUG_ !

Marc Alff's avatar
Marc Alff committed
3172
  mysql_mutex_lock(&mi->run_lock);
unknown's avatar
unknown committed
3173 3174 3175
  /* Inform waiting threads that slave has started */
  mi->slave_run_id++;

3176
#ifndef DBUG_OFF
3177
  mi->events_till_disconnect = disconnect_slave_event_count;
3178 3179
#endif

3180
  THD_CHECK_SENTRY(thd);
3181
  mi->io_thd = thd;
unknown's avatar
unknown committed
3182 3183

  pthread_detach_this_thread();
3184
  thd->thread_stack= (char*) &thd; // remember where our stack is
3185
  mi->clear_error();
3186
  if (init_slave_thread(thd, mi, SLAVE_THD_IO))
unknown's avatar
unknown committed
3187
  {
Marc Alff's avatar
Marc Alff committed
3188
    mysql_cond_broadcast(&mi->start_cond);
unknown's avatar
unknown committed
3189
    sql_print_error("Failed during slave I/O thread initialization");
3190
    goto err_during_init;
unknown's avatar
unknown committed
3191
  }
Marc Alff's avatar
Marc Alff committed
3192
  mysql_mutex_lock(&LOCK_thread_count);
unknown's avatar
unknown committed
3193
  threads.append(thd);
Marc Alff's avatar
Marc Alff committed
3194
  mysql_mutex_unlock(&LOCK_thread_count);
3195
  mi->slave_running = MYSQL_SLAVE_RUN_NOT_CONNECT;
3196
  mi->abort_slave = 0;
Marc Alff's avatar
Marc Alff committed
3197 3198
  mysql_mutex_unlock(&mi->run_lock);
  mysql_cond_broadcast(&mi->start_cond);
3199

3200
  DBUG_PRINT("master_info",("log_file_name: '%s'  position: %s",
unknown's avatar
unknown committed
3201 3202
                            mi->master_log_name,
                            llstr(mi->master_log_pos,llbuff)));
3203

He Zhenxing's avatar
He Zhenxing committed
3204 3205 3206 3207 3208 3209 3210 3211 3212 3213
  /* 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
3214
  if (!(mi->mysql = mysql = mysql_init(NULL)))
unknown's avatar
unknown committed
3215
  {
3216 3217
    mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
               ER(ER_SLAVE_FATAL_ERROR), "error in mysql_init()");
unknown's avatar
unknown committed
3218 3219
    goto err;
  }
3220

3221
  thd_proc_info(thd, "Connecting to master");
3222
  // we can get killed during safe_connect
3223
  if (!safe_connect(thd, mysql, mi))
3224
  {
unknown's avatar
unknown committed
3225 3226 3227
    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,
3228 3229
			  IO_RPL_LOG_NAME,
			  llstr(mi->master_log_pos,llbuff));
3230 3231 3232 3233 3234
  /*
    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.
  */
3235
    thd->net.max_packet_size= slave_max_allowed_packet;
3236 3237
    mysql->net.max_packet_size= thd->net.max_packet_size+= MAX_LOG_EVENT_HEADER;
  }
3238
  else
unknown's avatar
unknown committed
3239
  {
3240
    sql_print_information("Slave I/O thread killed while connecting to master");
unknown's avatar
unknown committed
3241 3242
    goto err;
  }
3243

3244
connected:
3245

3246 3247 3248 3249 3250
    DBUG_EXECUTE_IF("dbug.before_get_running_status_yes",
                    {
                      const char act[]=
                        "now "
                        "wait_for signal.io_thread_let_running";
3251
                      DBUG_ASSERT(debug_sync_service);
3252 3253 3254 3255
                      DBUG_ASSERT(!debug_sync_set_action(thd, 
                                                         STRING_WITH_LEN(act)));
                    };);

3256 3257
  // TODO: the assignment below should be under mutex (5.0)
  mi->slave_running= MYSQL_SLAVE_RUN_CONNECT;
3258
  thd->slave_net = &mysql->net;
3259
  thd_proc_info(thd, "Checking master version");
3260 3261 3262
  ret= get_master_version_and_clock(mysql, mi);
  if (ret == 1)
    /* Fatal error */
3263
    goto err;
3264

3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276
  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;
  } 
3277 3278

  if (mi->rli.relay_log.description_event_for_queue->binlog_version > 1)
3279
  {
unknown's avatar
unknown committed
3280 3281 3282
    /*
      Register ourselves with the master.
    */
3283
    thd_proc_info(thd, "Registering slave on master");
3284
    if (register_slave_on_master(mysql, mi, &suppress_warnings))
3285
    {
3286 3287 3288 3289 3290 3291 3292 3293 3294
      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
3295 3296
        goto err;
      goto connected;
3297
    }
3298 3299 3300 3301 3302 3303 3304 3305 3306 3307
    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;
      });
3308
  }
3309

3310
  DBUG_PRINT("info",("Starting reading binary log from master"));
3311
  while (!io_slave_killed(thd,mi))
unknown's avatar
unknown committed
3312
  {
3313
    thd_proc_info(thd, "Requesting binlog dump");
He Zhenxing's avatar
He Zhenxing committed
3314
    if (request_dump(thd, mysql, mi, &suppress_warnings))
unknown's avatar
unknown committed
3315 3316
    {
      sql_print_error("Failed on request_dump()");
3317 3318 3319 3320
      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
3321
        goto err;
unknown's avatar
unknown committed
3322 3323
      goto connected;
    }
3324 3325 3326 3327 3328 3329 3330 3331 3332 3333
    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
3334
    const char *event_buf;
unknown's avatar
unknown committed
3335

3336
    DBUG_ASSERT(mi->last_error().number == 0);
3337
    while (!io_slave_killed(thd,mi))
unknown's avatar
unknown committed
3338
    {
3339
      ulong event_len;
3340
      /*
3341
         We say "waiting" because read_event() will wait if there's nothing to
3342 3343 3344
         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.
3345
      */
3346
      thd_proc_info(thd, "Waiting for master to send event");
3347
      event_len= read_event(mysql, mi, &suppress_warnings);
3348 3349
      if (check_io_slave_killed(thd, mi, "Slave I/O thread killed while \
reading event"))
unknown's avatar
unknown committed
3350
        goto err;
3351 3352 3353 3354 3355 3356 3357 3358 3359 3360
      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;
        });
3361

unknown's avatar
unknown committed
3362 3363
      if (event_len == packet_error)
      {
unknown's avatar
unknown committed
3364
        uint mysql_error_number= mysql_errno(mysql);
3365 3366
        switch (mysql_error_number) {
        case CR_NET_PACKET_TOO_LARGE:
unknown's avatar
unknown committed
3367
          sql_print_error("\
3368
Log entry on master is longer than slave_max_allowed_packet (%lu) on \
3369
slave. If the entry is correct, restart the server with a higher value of \
3370 3371
slave_max_allowed_packet",
                         slave_max_allowed_packet);
3372
          mi->report(ERROR_LEVEL, ER_NET_PACKET_TOO_LARGE,
3373
                     "%s", "Got a packet bigger than 'slave_max_allowed_packet' bytes");
unknown's avatar
unknown committed
3374
          goto err;
3375
        case ER_MASTER_FATAL_ERROR_READING_BINLOG:
3376 3377 3378
          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
3379
          goto err;
3380
        case ER_OUT_OF_RESOURCES:
3381 3382
          sql_print_error("\
Stopping slave I/O thread due to out-of-memory error from master");
3383
          mi->report(ERROR_LEVEL, ER_OUT_OF_RESOURCES,
3384
                     "%s", ER(ER_OUT_OF_RESOURCES));
unknown's avatar
unknown committed
3385 3386
          goto err;
        }
3387 3388
        if (try_to_reconnect(thd, mysql, mi, &retry_count, suppress_warnings,
                             reconnect_messages[SLAVE_RECON_ACT_EVENT]))
unknown's avatar
unknown committed
3389 3390
          goto err;
        goto connected;
unknown's avatar
unknown committed
3391
      } // if (event_len == packet_error)
3392

unknown's avatar
unknown committed
3393
      retry_count=0;                    // ok event, reset retry counter
3394
      thd_proc_info(thd, "Queueing master event to the relay log");
He Zhenxing's avatar
He Zhenxing committed
3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409
      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
3410
      {
3411 3412 3413
        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
3414
        goto err;
unknown's avatar
unknown committed
3415
      }
He Zhenxing's avatar
He Zhenxing committed
3416 3417 3418 3419 3420 3421 3422 3423 3424 3425

      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;
      }

3426
      if (flush_master_info(mi, TRUE, TRUE))
3427 3428 3429 3430
      {
        sql_print_error("Failed to flush master info file");
        goto err;
      }
3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442
      /*
        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.
      */
3443 3444 3445 3446 3447
#ifndef DBUG_OFF
      {
        char llbuf1[22], llbuf2[22];
        DBUG_PRINT("info", ("log_space_limit=%s log_space_total=%s \
ignore_log_space_limit=%d",
3448 3449
                            llstr(rli->log_space_limit,llbuf1),
                            llstr(rli->log_space_total,llbuf2),
unknown's avatar
unknown committed
3450
                            (int) rli->ignore_log_space_limit));
3451 3452 3453
      }
#endif

3454
      if (rli->log_space_limit && rli->log_space_limit <
unknown's avatar
unknown committed
3455
          rli->log_space_total &&
3456
          !rli->ignore_log_space_limit)
unknown's avatar
unknown committed
3457 3458 3459
        if (wait_for_relay_log_space(rli))
        {
          sql_print_error("Slave I/O thread aborted while waiting for relay \
unknown's avatar
unknown committed
3460
log space");
unknown's avatar
unknown committed
3461 3462 3463
          goto err;
        }
    }
3464
  }
unknown's avatar
unknown committed
3465

unknown's avatar
unknown committed
3466
  // error = 0;
unknown's avatar
unknown committed
3467
err:
3468
  // print the current replication position
3469
  sql_print_information("Slave I/O thread exiting, read up to log '%s', position %s",
unknown's avatar
unknown committed
3470
                  IO_RPL_LOG_NAME, llstr(mi->master_log_pos,llbuff));
He Zhenxing's avatar
He Zhenxing committed
3471
  RUN_HOOK(binlog_relay_io, thread_stop, (thd, mi));
3472
  thd->reset_query();
unknown's avatar
unknown committed
3473
  thd->reset_db(NULL, 0);
unknown's avatar
unknown committed
3474 3475
  if (mysql)
  {
3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486
    /*
      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
3487
    mysql_close(mysql);
unknown's avatar
unknown committed
3488 3489
    mi->mysql=0;
  }
3490
  write_ignored_events_info_to_relay_log(thd, mi);
3491
  thd_proc_info(thd, "Waiting for slave mutex on exit");
Marc Alff's avatar
Marc Alff committed
3492
  mysql_mutex_lock(&mi->run_lock);
3493

3494
err_during_init:
3495 3496 3497
  /* Forget the relay log's format */
  delete mi->rli.relay_log.description_event_for_queue;
  mi->rli.relay_log.description_event_for_queue= 0;
3498
  // TODO: make rpl_status part of Master_info
3499
  change_rpl_status(RPL_ACTIVE_SLAVE,RPL_IDLE_SLAVE);
Marc Alff's avatar
Marc Alff committed
3500
  mysql_mutex_lock(&LOCK_thread_count);
3501
  THD_CHECK_SENTRY(thd);
unknown's avatar
unknown committed
3502
  delete thd;
Marc Alff's avatar
Marc Alff committed
3503
  mysql_mutex_unlock(&LOCK_thread_count);
unknown's avatar
unknown committed
3504
  mi->abort_slave= 0;
3505
  mi->slave_running= MYSQL_SLAVE_NOT_RUN;
unknown's avatar
unknown committed
3506
  mi->io_thd= 0;
3507 3508 3509 3510 3511
  /*
    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
3512
  mysql_cond_broadcast(&mi->stop_cond);       // tell the world we are done
3513
  DBUG_EXECUTE_IF("simulate_slave_delay_at_terminate_bug38694", sleep(5););
Marc Alff's avatar
Marc Alff committed
3514
  mysql_mutex_unlock(&mi->run_lock);
3515 3516

  DBUG_LEAVE;                                   // Must match DBUG_ENTER()
3517
  my_thread_end();
unknown's avatar
unknown committed
3518
  pthread_exit(0);
3519
  return 0;                                     // Avoid compiler warnings
unknown's avatar
unknown committed
3520 3521
}

3522 3523 3524 3525 3526
/*
  Check the temporary directory used by commands like
  LOAD DATA INFILE.
 */
static 
3527
int check_temp_dir(char* tmp_file)
3528 3529 3530
{
  int fd;
  MY_DIR *dirp;
3531 3532
  char tmp_dir[FN_REFLEN];
  size_t tmp_dir_size;
3533 3534 3535

  DBUG_ENTER("check_temp_dir");

3536 3537 3538 3539 3540
  /*
    Get the directory from the temporary file.
  */
  dirname_part(tmp_dir, tmp_file, &tmp_dir_size);

3541 3542 3543 3544 3545 3546 3547 3548 3549 3550
  /*
    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
3551 3552 3553 3554
  if ((fd= mysql_file_create(key_file_misc,
                             tmp_file, CREATE_MODE,
                             O_WRONLY | O_BINARY | O_EXCL | O_NOFOLLOW,
                             MYF(MY_WME))) < 0)
3555 3556 3557 3558 3559
  DBUG_RETURN(1);

  /*
    Clean up.
   */
Marc Alff's avatar
Marc Alff committed
3560 3561
  mysql_file_close(fd, MYF(0));
  mysql_file_delete(key_file_misc, tmp_file, MYF(0));
3562 3563 3564

  DBUG_RETURN(0);
}
unknown's avatar
unknown committed
3565

3566 3567 3568 3569 3570
/**
  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
3571

3572 3573
  @return Always 0.
*/
3574
pthread_handler_t handle_slave_sql(void *arg)
3575
{
unknown's avatar
unknown committed
3576
  THD *thd;                     /* needs to be first for thread_stack */
3577
  char llbuff[22],llbuff1[22];
3578 3579
  char saved_log_name[FN_REFLEN];
  char saved_master_log_name[FN_REFLEN];
3580 3581
  my_off_t UNINIT_VAR(saved_log_pos);
  my_off_t UNINIT_VAR(saved_master_log_pos);
3582
  my_off_t saved_skip= 0;
3583 3584
  Master_info *mi= ((Master_info*)arg);
  Relay_log_info* rli = &mi->rli;
unknown's avatar
unknown committed
3585 3586 3587 3588
  const char *errmsg;

  // needs to call my_thread_init(), otherwise we get a coredump in DBUG_ stuff
  my_thread_init();
3589
  DBUG_ENTER("handle_slave_sql");
unknown's avatar
unknown committed
3590

Michael Widenius's avatar
Michael Widenius committed
3591 3592
  LINT_INIT(saved_master_log_pos);
  LINT_INIT(saved_log_pos);
3593 3594 3595 3596

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

3597
  DBUG_ASSERT(rli->inited);
3598
  DBUG_ASSERT(rli->mi == mi);
Marc Alff's avatar
Marc Alff committed
3599
  mysql_mutex_lock(&rli->run_lock);
3600
  DBUG_ASSERT(!rli->slave_running);
unknown's avatar
unknown committed
3601
  errmsg= 0;
unknown's avatar
unknown committed
3602
#ifndef DBUG_OFF
3603
  rli->events_till_abort = abort_slave_event_count;
unknown's avatar
unknown committed
3604
#endif
3605

3606
  rli->sql_thd= thd;
3607
  
unknown's avatar
unknown committed
3608 3609
  /* Inform waiting threads that slave has started */
  rli->slave_run_id++;
3610
  rli->slave_running= MYSQL_SLAVE_RUN_NOT_CONNECT;
unknown's avatar
unknown committed
3611

3612
  pthread_detach_this_thread();
3613
  if (init_slave_thread(thd, mi, SLAVE_THD_SQL))
unknown's avatar
unknown committed
3614 3615 3616 3617 3618
  {
    /*
      TODO: this is currently broken - slave start and change master
      will be stuck if we fail here
    */
Marc Alff's avatar
Marc Alff committed
3619
    mysql_cond_broadcast(&rli->start_cond);
3620 3621
    rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, 
                "Failed during slave thread initialization");
3622
    goto err_during_init;
unknown's avatar
unknown committed
3623
  }
3624
  thd->init_for_queries();
3625 3626 3627 3628 3629 3630
  thd->rli_slave= rli;
  if ((rli->deferred_events_collecting= rpl_filter->is_on()))
  {
    rli->deferred_events= new Deferred_log_events(rli);
  }

3631
  thd->temporary_tables = rli->save_temporary_tables; // restore temp tables
3632
  set_thd_in_use_temporary_tables(rli);   // (re)set sql_thd in use for saved temp tables
3633
  /*
3634
    binlog_annotate_row_events must be TRUE only after an Annotate_rows event
3635 3636 3637
    has been recieved and only till the last corresponding rbr event has been
    applied. In all other cases it must be FALSE.
  */
3638
  thd->variables.binlog_annotate_row_events= 0;
Marc Alff's avatar
Marc Alff committed
3639
  mysql_mutex_lock(&LOCK_thread_count);
3640
  threads.append(thd);
Marc Alff's avatar
Marc Alff committed
3641
  mysql_mutex_unlock(&LOCK_thread_count);
3642 3643 3644 3645 3646 3647 3648 3649
  /*
    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.
  */
3650
  rli->abort_slave = 0;
Marc Alff's avatar
Marc Alff committed
3651 3652
  mysql_mutex_unlock(&rli->run_lock);
  mysql_cond_broadcast(&rli->start_cond);
3653

unknown's avatar
unknown committed
3654 3655 3656
  /*
    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
3657
    though there's no problem anymore). Do not reset the master timestamp
3658 3659 3660 3661
    (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
3662
    But the master timestamp is reset by RESET SLAVE & CHANGE MASTER.
unknown's avatar
unknown committed
3663
  */
3664
  rli->clear_error();
3665 3666

  //tell the I/O thread to take relay_log_space_limit into account from now on
Marc Alff's avatar
Marc Alff committed
3667
  mysql_mutex_lock(&rli->log_space_lock);
3668
  rli->ignore_log_space_limit= 0;
Marc Alff's avatar
Marc Alff committed
3669
  mysql_mutex_unlock(&rli->log_space_lock);
3670
  rli->trans_retries= 0; // start from "no error"
unknown's avatar
unknown committed
3671
  DBUG_PRINT("info", ("rli->trans_retries: %lu", rli->trans_retries));
3672

3673
  if (init_relay_log_pos(rli,
unknown's avatar
unknown committed
3674 3675 3676
                         rli->group_relay_log_name,
                         rli->group_relay_log_pos,
                         1 /*need data lock*/, &errmsg,
3677
                         1 /*look for a description_event*/))
3678 3679 3680
  { 
    rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, 
                "Error initializing relay log position: %s", errmsg);
3681 3682
    goto err;
  }
3683
  THD_CHECK_SENTRY(thd);
3684 3685 3686 3687
#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
3688
                        llstr(my_b_tell(rli->cur_log),llbuf1),
3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707
                        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
3708
  DBUG_ASSERT(rli->sql_thd == thd);
3709 3710

  DBUG_PRINT("master_info",("log_file_name: %s  position: %s",
unknown's avatar
unknown committed
3711 3712
                            rli->group_master_log_name,
                            llstr(rli->group_master_log_pos,llbuff)));
3713
  if (global_system_variables.log_warnings)
3714
    sql_print_information("Slave SQL thread initialized, starting replication in \
unknown's avatar
unknown committed
3715
log '%s' at position %s, relay log '%s' position: %s", RPL_LOG_NAME,
unknown's avatar
unknown committed
3716 3717
                    llstr(rli->group_master_log_pos,llbuff),rli->group_relay_log_name,
                    llstr(rli->group_relay_log_pos,llbuff1));
3718

3719
  if (check_temp_dir(rli->slave_patternload_file))
3720
  {
Marc Alff's avatar
Marc Alff committed
3721
    rli->report(ERROR_LEVEL, thd->stmt_da->sql_errno(), 
3722
                "Unable to use slave's temporary directory %s - %s", 
Marc Alff's avatar
Marc Alff committed
3723
                slave_load_tmpdir, thd->stmt_da->message());
3724 3725 3726
    goto err;
  }

unknown's avatar
unknown committed
3727
  /* execute init_slave variable */
3728
  if (opt_init_slave.length)
unknown's avatar
unknown committed
3729
  {
3730
    execute_init_command(thd, &opt_init_slave, &LOCK_sys_init_slave);
3731
    if (thd->is_slave_error)
unknown's avatar
unknown committed
3732
    {
Marc Alff's avatar
Marc Alff committed
3733
      rli->report(ERROR_LEVEL, thd->stmt_da->sql_errno(),
3734
                  "Slave SQL thread aborted. Can't execute init_slave query");
unknown's avatar
unknown committed
3735 3736 3737 3738
      goto err;
    }
  }

3739 3740 3741 3742
  /*
    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
3743
  mysql_mutex_lock(&rli->data_lock);
3744 3745
  if (rli->slave_skip_counter)
  {
3746 3747
    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);
3748 3749 3750 3751
    saved_log_pos= rli->group_relay_log_pos;
    saved_master_log_pos= rli->group_master_log_pos;
    saved_skip= rli->slave_skip_counter;
  }
3752
  if (rli->until_condition != Relay_log_info::UNTIL_NONE &&
3753
      rli->is_until_satisfied(thd, NULL))
3754 3755 3756 3757
  {
    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
3758
    mysql_mutex_unlock(&rli->data_lock);
3759 3760
    goto err;
  }
Marc Alff's avatar
Marc Alff committed
3761
  mysql_mutex_unlock(&rli->data_lock);
3762

3763 3764
  /* Read queries from the IO/THREAD until this thread is killed */

3765
  while (!sql_slave_killed(thd,rli))
3766
  {
3767
    thd_proc_info(thd, "Reading event from the relay log");
3768
    DBUG_ASSERT(rli->sql_thd == thd);
3769
    THD_CHECK_SENTRY(thd);
3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784

    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;
    }
    
3785 3786
    if (exec_relay_log_event(thd,rli))
    {
3787
      DBUG_PRINT("info", ("exec_relay_log_event() failed"));
3788
      // do not scare the user if SQL thread was simply killed or stopped
3789
      if (!sql_slave_killed(thd,rli))
unknown's avatar
unknown committed
3790 3791
      {
        /*
3792 3793 3794
          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
3795
        */
3796
        uint32 const last_errno= rli->last_error().number;
3797

3798
        if (thd->is_error())
unknown's avatar
unknown committed
3799
        {
Marc Alff's avatar
Marc Alff committed
3800
          char const *const errmsg= thd->stmt_da->message();
3801 3802

          DBUG_PRINT("info",
Marc Alff's avatar
Marc Alff committed
3803 3804
                     ("thd->stmt_da->sql_errno()=%d; rli->last_error.number=%d",
                      thd->stmt_da->sql_errno(), last_errno));
3805
          if (last_errno == 0)
unknown's avatar
unknown committed
3806
          {
3807 3808 3809 3810
            /*
 	      This function is reporting an error which was not reported
 	      while executing exec_relay_log_event().
 	    */ 
3811
            rli->report(ERROR_LEVEL, thd->stmt_da->sql_errno(), "%s", errmsg);
unknown's avatar
unknown committed
3812
          }
Marc Alff's avatar
Marc Alff committed
3813
          else if (last_errno != thd->stmt_da->sql_errno())
unknown's avatar
unknown committed
3814
          {
3815 3816 3817 3818 3819 3820
            /*
             * 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
3821
            sql_print_error("Slave (additional info): %s Error_code: %d",
Marc Alff's avatar
Marc Alff committed
3822
                            errmsg, thd->stmt_da->sql_errno());
unknown's avatar
unknown committed
3823 3824 3825 3826
          }
        }

        /* Print any warnings issued */
Marc Alff's avatar
Marc Alff committed
3827
        List_iterator_fast<MYSQL_ERROR> it(thd->warning_info->warn_list());
unknown's avatar
unknown committed
3828
        MYSQL_ERROR *err;
3829 3830 3831 3832 3833
        /*
          Added controlled slave thread cancel for replication
          of user-defined variables.
        */
        bool udf_error = false;
unknown's avatar
unknown committed
3834
        while ((err= it++))
3835
        {
Marc Alff's avatar
Marc Alff committed
3836
          if (err->get_sql_errno() == ER_CANT_OPEN_LIBRARY)
3837
            udf_error = true;
Marc Alff's avatar
Marc Alff committed
3838
          sql_print_warning("Slave: %s Error_code: %d", err->get_message_text(), err->get_sql_errno());
3839 3840 3841 3842 3843 3844 3845 3846 3847
        }
        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("\
3848
Error running query, slave SQL thread aborted. Fix the problem, and restart \
unknown's avatar
unknown committed
3849
the slave SQL thread with \"SLAVE START\". We stopped at log \
3850
'%s' position %s", RPL_LOG_NAME, llstr(rli->group_master_log_pos, llbuff));
unknown's avatar
unknown committed
3851
      }
3852 3853
      goto err;
    }
3854
    rli->executed_entries++;
3855
  }
3856

3857
  /* Thread stopped. Print the current replication position to the log */
unknown's avatar
unknown committed
3858
  sql_print_information("Slave SQL thread exiting, replication stopped in log "
unknown's avatar
unknown committed
3859 3860
                        "'%s' at position %s",
                        RPL_LOG_NAME, llstr(rli->group_master_log_pos,llbuff));
3861 3862

 err:
3863 3864 3865 3866 3867 3868 3869

  /*
    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:
  */
3870
  thd->clear_error();
3871
  rli->cleanup_context(thd, 1);
3872 3873 3874 3875 3876
  /*
    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)).
  */
3877
  thd->catalog= 0;
3878
  thd->reset_query();
unknown's avatar
unknown committed
3879
  thd->reset_db(NULL, 0);
3880
  thd_proc_info(thd, "Waiting for slave mutex on exit");
Marc Alff's avatar
Marc Alff committed
3881
  mysql_mutex_lock(&rli->run_lock);
3882
err_during_init:
3883
  /* We need data_lock, at least to wake up any waiting master_pos_wait() */
Marc Alff's avatar
Marc Alff committed
3884
  mysql_mutex_lock(&rli->data_lock);
3885
  DBUG_ASSERT(rli->slave_running == MYSQL_SLAVE_RUN_NOT_CONNECT); // tracking buffer overrun
3886
  /* When master_pos_wait() wakes up it will check this and terminate */
3887
  rli->slave_running= MYSQL_SLAVE_NOT_RUN;
3888 3889 3890
  /* Forget the relay log's format */
  delete rli->relay_log.description_event_for_exec;
  rli->relay_log.description_event_for_exec= 0;
3891
  /* Wake up master_pos_wait() */
Marc Alff's avatar
Marc Alff committed
3892
  mysql_mutex_unlock(&rli->data_lock);
3893
  DBUG_PRINT("info",("Signaling possibly waiting master_pos_wait() functions"));
Marc Alff's avatar
Marc Alff committed
3894
  mysql_cond_broadcast(&rli->data_cond);
unknown's avatar
unknown committed
3895
  rli->ignore_log_space_limit= 0; /* don't need any lock */
3896 3897
  /* we die so won't remember charset - re-update them on next thread start */
  rli->cached_charset_invalidate();
3898
  rli->save_temporary_tables = thd->temporary_tables;
unknown's avatar
unknown committed
3899 3900 3901 3902 3903

  /*
    TODO: see if we can do this conditionally in next_event() instead
    to avoid unneeded position re-init
  */
3904 3905
  thd->temporary_tables = 0; // remove tempation from destructor to close them
  DBUG_ASSERT(rli->sql_thd == thd);
3906
  THD_CHECK_SENTRY(thd);
3907
  rli->sql_thd= 0;
3908
  set_thd_in_use_temporary_tables(rli);  // (re)set sql_thd in use for saved temp tables
Marc Alff's avatar
Marc Alff committed
3909
  mysql_mutex_lock(&LOCK_thread_count);
3910
  THD_CHECK_SENTRY(thd);
3911
  delete thd;
Marc Alff's avatar
Marc Alff committed
3912
  mysql_mutex_unlock(&LOCK_thread_count);
3913 3914 3915 3916 3917
 /*
  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
3918
  mysql_cond_broadcast(&rli->stop_cond);
3919
  DBUG_EXECUTE_IF("simulate_slave_delay_at_terminate_bug38694", sleep(5););
Marc Alff's avatar
Marc Alff committed
3920
  mysql_mutex_unlock(&rli->run_lock);  // tell the world we are done
3921 3922

  DBUG_LEAVE;                                   // Must match DBUG_ENTER()
unknown's avatar
unknown committed
3923
  my_thread_end();
3924
  pthread_exit(0);
3925
  return 0;                                     // Avoid compiler warnings
3926
}
unknown's avatar
unknown committed
3927

3928

unknown's avatar
unknown committed
3929
/*
3930
  process_io_create_file()
unknown's avatar
unknown committed
3931
*/
3932

3933
static int process_io_create_file(Master_info* mi, Create_file_log_event* cev)
unknown's avatar
unknown committed
3934 3935 3936 3937
{
  int error = 1;
  ulong num_bytes;
  bool cev_not_written;
3938 3939
  THD *thd = mi->io_thd;
  NET *net = &mi->mysql->net;
unknown's avatar
unknown committed
3940
  DBUG_ENTER("process_io_create_file");
unknown's avatar
unknown committed
3941 3942

  if (unlikely(!cev->is_valid()))
unknown's avatar
unknown committed
3943
    DBUG_RETURN(1);
3944 3945

  if (!rpl_filter->db_ok(cev->db))
unknown's avatar
unknown committed
3946 3947
  {
    skip_load_data_infile(net);
unknown's avatar
unknown committed
3948
    DBUG_RETURN(0);
unknown's avatar
unknown committed
3949 3950 3951
  }
  DBUG_ASSERT(cev->inited_from_old);
  thd->file_id = cev->file_id = mi->file_id++;
3952
  thd->variables.server_id = cev->server_id;
unknown's avatar
unknown committed
3953
  cev_not_written = 1;
unknown's avatar
unknown committed
3954

unknown's avatar
unknown committed
3955 3956 3957
  if (unlikely(net_request_file(net,cev->fname)))
  {
    sql_print_error("Slave I/O: failed requesting download of '%s'",
unknown's avatar
unknown committed
3958
                    cev->fname);
unknown's avatar
unknown committed
3959 3960 3961
    goto err;
  }

unknown's avatar
unknown committed
3962 3963 3964 3965
  /*
    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
3966 3967
  */
  {
3968
    Append_block_log_event aev(thd,0,0,0,0);
unknown's avatar
unknown committed
3969

unknown's avatar
unknown committed
3970 3971 3972 3973
    for (;;)
    {
      if (unlikely((num_bytes=my_net_read(net)) == packet_error))
      {
unknown's avatar
unknown committed
3974 3975 3976
        sql_print_error("Network read error downloading '%s' from master",
                        cev->fname);
        goto err;
unknown's avatar
unknown committed
3977 3978 3979
      }
      if (unlikely(!num_bytes)) /* eof */
      {
3980 3981
	/* 3.23 master wants it */
        net_write_command(net, 0, (uchar*) "", 0, (uchar*) "", 0);
unknown's avatar
unknown committed
3982 3983 3984 3985 3986 3987 3988
        /*
          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
3989 3990 3991 3992 3993
          break;
        Execute_load_log_event xev(thd,0,0);
        xev.log_pos = cev->log_pos;
        if (unlikely(mi->rli.relay_log.append(&xev)))
        {
3994 3995 3996
          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
3997 3998 3999 4000
          goto err;
        }
        mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total);
        break;
unknown's avatar
unknown committed
4001 4002 4003
      }
      if (unlikely(cev_not_written))
      {
unknown's avatar
unknown committed
4004
        cev->block = net->read_pos;
unknown's avatar
unknown committed
4005 4006 4007
        cev->block_len = num_bytes;
        if (unlikely(mi->rli.relay_log.append(cev)))
        {
4008 4009 4010
          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
4011 4012 4013 4014
          goto err;
        }
        cev_not_written=0;
        mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total);
unknown's avatar
unknown committed
4015 4016 4017
      }
      else
      {
unknown's avatar
unknown committed
4018
        aev.block = net->read_pos;
unknown's avatar
unknown committed
4019 4020 4021 4022
        aev.block_len = num_bytes;
        aev.log_pos = cev->log_pos;
        if (unlikely(mi->rli.relay_log.append(&aev)))
        {
4023 4024 4025
          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
4026 4027 4028
          goto err;
        }
        mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total) ;
unknown's avatar
unknown committed
4029 4030 4031 4032 4033
      }
    }
  }
  error=0;
err:
unknown's avatar
unknown committed
4034
  DBUG_RETURN(error);
unknown's avatar
unknown committed
4035
}
unknown's avatar
unknown committed
4036

4037

unknown's avatar
unknown committed
4038
/*
unknown's avatar
unknown committed
4039 4040 4041 4042
  Start using a new binary log on the master

  SYNOPSIS
    process_io_rotate()
unknown's avatar
unknown committed
4043 4044
    mi                  master_info for the slave
    rev                 The rotate log event read from the binary log
unknown's avatar
unknown committed
4045 4046

  DESCRIPTION
4047
    Updates the master info with the place in the next binary
unknown's avatar
unknown committed
4048
    log where we should start reading.
4049
    Rotate the relay log to avoid mixed-format relay logs.
unknown's avatar
unknown committed
4050 4051 4052 4053 4054

  NOTES
    We assume we already locked mi->data_lock

  RETURN VALUES
unknown's avatar
unknown committed
4055 4056
    0           ok
    1           Log event is illegal
unknown's avatar
unknown committed
4057 4058 4059

*/

4060
static int process_io_rotate(Master_info *mi, Rotate_log_event *rev)
4061
{
4062
  DBUG_ENTER("process_io_rotate");
Marc Alff's avatar
Marc Alff committed
4063
  mysql_mutex_assert_owner(&mi->data_lock);
4064

unknown's avatar
unknown committed
4065
  if (unlikely(!rev->is_valid()))
4066
    DBUG_RETURN(1);
unknown's avatar
unknown committed
4067

4068
  /* Safe copy as 'rev' has been "sanitized" in Rotate_log_event's ctor */
unknown's avatar
unknown committed
4069 4070
  memcpy(mi->master_log_name, rev->new_log_ident, rev->ident_len+1);
  mi->master_log_pos= rev->pos;
unknown's avatar
unknown committed
4071
  DBUG_PRINT("info", ("master_log_pos: '%s' %lu",
unknown's avatar
unknown committed
4072
                      mi->master_log_name, (ulong) mi->master_log_pos));
4073
#ifndef DBUG_OFF
unknown's avatar
unknown committed
4074 4075 4076 4077 4078
  /*
    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)
4079
    mi->events_till_disconnect++;
4080
#endif
4081

4082 4083 4084 4085 4086 4087 4088 4089 4090
  /*
    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)
  {
4091 4092 4093
    DBUG_ASSERT(mi->rli.relay_log.description_event_for_queue->checksum_alg ==
                mi->rli.relay_log.relay_log_checksum_alg);
    
4094 4095 4096 4097
    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);
4098 4099
    mi->rli.relay_log.description_event_for_queue->checksum_alg=
      mi->rli.relay_log.relay_log_checksum_alg;    
4100
  }
4101 4102 4103 4104
  /*
    Rotate the relay log makes binlog format detection easier (at next slave
    start or mysqlbinlog)
  */
4105
  DBUG_RETURN(rotate_relay_log(mi) /* will take the right mutexes */);
4106 4107
}

unknown's avatar
unknown committed
4108
/*
4109 4110
  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
4111
*/
4112
static int queue_binlog_ver_1_event(Master_info *mi, const char *buf,
unknown's avatar
unknown committed
4113
                           ulong event_len)
4114
{
unknown's avatar
unknown committed
4115
  const char *errmsg = 0;
unknown's avatar
unknown committed
4116 4117 4118
  ulong inc_pos;
  bool ignore_event= 0;
  char *tmp_buf = 0;
4119
  Relay_log_info *rli= &mi->rli;
4120
  DBUG_ENTER("queue_binlog_ver_1_event");
unknown's avatar
unknown committed
4121

unknown's avatar
unknown committed
4122 4123 4124
  /*
    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
4125
  */
4126
  if ((uchar)buf[EVENT_TYPE_OFFSET] == LOAD_EVENT)
unknown's avatar
unknown committed
4127 4128 4129
  {
    if (unlikely(!(tmp_buf=(char*)my_malloc(event_len+1,MYF(MY_WME)))))
    {
4130 4131
      mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
                 ER(ER_SLAVE_FATAL_ERROR), "Memory allocation failed");
unknown's avatar
unknown committed
4132
      DBUG_RETURN(1);
unknown's avatar
unknown committed
4133 4134
    }
    memcpy(tmp_buf,buf,event_len);
4135 4136 4137 4138 4139 4140 4141 4142
    /*
      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
4143
    int4store(tmp_buf+EVENT_LEN_OFFSET, event_len);
unknown's avatar
unknown committed
4144 4145
    buf = (const char*)tmp_buf;
  }
4146 4147 4148 4149 4150 4151
  /*
    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).
  */
4152 4153 4154
  Log_event *ev=
    Log_event::read_log_event(buf, event_len, &errmsg,
                              mi->rli.relay_log.description_event_for_queue, 0);
4155
  if (unlikely(!ev))
4156 4157
  {
    sql_print_error("Read invalid event from master: '%s',\
unknown's avatar
unknown committed
4158
 master could be corrupt but a more likely cause of this is a bug",
unknown's avatar
unknown committed
4159
                    errmsg);
4160
    my_free(tmp_buf);
unknown's avatar
unknown committed
4161
    DBUG_RETURN(1);
4162
  }
4163

Marc Alff's avatar
Marc Alff committed
4164
  mysql_mutex_lock(&mi->data_lock);
4165
  ev->log_pos= mi->master_log_pos; /* 3.23 events don't contain log_pos */
unknown's avatar
unknown committed
4166
  switch (ev->get_type_code()) {
unknown's avatar
unknown committed
4167
  case STOP_EVENT:
4168
    ignore_event= 1;
unknown's avatar
unknown committed
4169 4170
    inc_pos= event_len;
    break;
4171
  case ROTATE_EVENT:
4172
    if (unlikely(process_io_rotate(mi,(Rotate_log_event*)ev)))
4173 4174
    {
      delete ev;
Marc Alff's avatar
Marc Alff committed
4175
      mysql_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
4176
      DBUG_RETURN(1);
4177
    }
unknown's avatar
unknown committed
4178
    inc_pos= 0;
4179
    break;
unknown's avatar
unknown committed
4180
  case CREATE_FILE_EVENT:
4181 4182 4183 4184 4185 4186
    /*
      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
4187
  {
unknown's avatar
unknown committed
4188
    /* We come here when and only when tmp_buf != 0 */
4189
    DBUG_ASSERT(tmp_buf != 0);
4190 4191
    inc_pos=event_len;
    ev->log_pos+= inc_pos;
unknown's avatar
unknown committed
4192
    int error = process_io_create_file(mi,(Create_file_log_event*)ev);
4193
    delete ev;
4194
    mi->master_log_pos += inc_pos;
unknown's avatar
unknown committed
4195
    DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
Marc Alff's avatar
Marc Alff committed
4196
    mysql_mutex_unlock(&mi->data_lock);
4197
    my_free(tmp_buf);
unknown's avatar
unknown committed
4198
    DBUG_RETURN(error);
unknown's avatar
unknown committed
4199
  }
4200
  default:
unknown's avatar
unknown committed
4201
    inc_pos= event_len;
4202 4203
    break;
  }
unknown's avatar
unknown committed
4204
  if (likely(!ignore_event))
4205
  {
unknown's avatar
unknown committed
4206 4207
    if (ev->log_pos)
      /*
4208 4209 4210 4211
         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
4212
    if (unlikely(rli->relay_log.append(ev)))
4213 4214
    {
      delete ev;
Marc Alff's avatar
Marc Alff committed
4215
      mysql_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
4216
      DBUG_RETURN(1);
4217
    }
unknown's avatar
unknown committed
4218
    rli->relay_log.harvest_bytes_written(&rli->log_space_total);
4219 4220
  }
  delete ev;
unknown's avatar
unknown committed
4221
  mi->master_log_pos+= inc_pos;
unknown's avatar
unknown committed
4222
  DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
Marc Alff's avatar
Marc Alff committed
4223
  mysql_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
4224
  DBUG_RETURN(0);
4225 4226
}

4227 4228 4229 4230
/*
  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.
*/
4231
static int queue_binlog_ver_3_event(Master_info *mi, const char *buf,
unknown's avatar
unknown committed
4232
                           ulong event_len)
4233 4234 4235 4236
{
  const char *errmsg = 0;
  ulong inc_pos;
  char *tmp_buf = 0;
4237
  Relay_log_info *rli= &mi->rli;
4238 4239 4240
  DBUG_ENTER("queue_binlog_ver_3_event");

  /* read_log_event() will adjust log_pos to be end_log_pos */
4241 4242 4243
  Log_event *ev=
    Log_event::read_log_event(buf,event_len, &errmsg,
                              mi->rli.relay_log.description_event_for_queue, 0);
4244 4245 4246 4247
  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
4248
                    errmsg);
4249
    my_free(tmp_buf);
4250 4251
    DBUG_RETURN(1);
  }
Marc Alff's avatar
Marc Alff committed
4252
  mysql_mutex_lock(&mi->data_lock);
4253 4254 4255 4256 4257 4258 4259
  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
4260
      mysql_mutex_unlock(&mi->data_lock);
4261 4262 4263 4264 4265 4266 4267 4268
      DBUG_RETURN(1);
    }
    inc_pos= 0;
    break;
  default:
    inc_pos= event_len;
    break;
  }
4269

4270 4271 4272
  if (unlikely(rli->relay_log.append(ev)))
  {
    delete ev;
Marc Alff's avatar
Marc Alff committed
4273
    mysql_mutex_unlock(&mi->data_lock);
4274 4275 4276 4277 4278 4279
    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
4280
  DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
Marc Alff's avatar
Marc Alff committed
4281
  mysql_mutex_unlock(&mi->data_lock);
4282 4283 4284 4285 4286 4287 4288 4289 4290 4291
  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
4292
  TODO:
4293 4294 4295 4296
    Test this code before release - it has to be tested on a separate
    setup with 3.23 master or 4.0 master
*/

4297
static int queue_old_event(Master_info *mi, const char *buf,
unknown's avatar
unknown committed
4298
                           ulong event_len)
4299
{
4300 4301
  DBUG_ENTER("queue_old_event");

4302 4303 4304
  switch (mi->rli.relay_log.description_event_for_queue->binlog_version)
  {
  case 1:
4305
      DBUG_RETURN(queue_binlog_ver_1_event(mi,buf,event_len));
4306
  case 3:
4307
      DBUG_RETURN(queue_binlog_ver_3_event(mi,buf,event_len));
4308 4309
  default: /* unsupported format; eg version 2 */
    DBUG_PRINT("info",("unsupported binlog format %d in queue_old_event()",
unknown's avatar
unknown committed
4310
                       mi->rli.relay_log.description_event_for_queue->binlog_version));
4311
    DBUG_RETURN(1);
4312 4313
  }
}
4314

unknown's avatar
unknown committed
4315
/*
4316 4317
  queue_event()

4318 4319 4320 4321 4322
  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
4323 4324
*/

4325
static int queue_event(Master_info* mi,const char* buf, ulong event_len)
4326
{
unknown's avatar
unknown committed
4327
  int error= 0;
Andrei Elkin's avatar
Andrei Elkin committed
4328
  String error_msg;
4329 4330
  ulonglong inc_pos;
  ulonglong event_pos;
4331
  Relay_log_info *rli= &mi->rli;
Marc Alff's avatar
Marc Alff committed
4332
  mysql_mutex_t *log_lock= rli->relay_log.get_log_lock();
Andrei Elkin's avatar
Andrei Elkin committed
4333
  ulong s_id;
4334 4335 4336 4337 4338 4339 4340 4341 4342 4343 4344 4345 4346 4347 4348 4349 4350 4351
  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
4352
  DBUG_ENTER("queue_event");
4353 4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365 4366 4367 4368 4369 4370 4371 4372 4373 4374 4375 4376 4377 4378 4379 4380 4381 4382 4383 4384 4385 4386 4387 4388 4389 4390 4391 4392 4393 4394 4395
  /*
    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
4396

unknown's avatar
unknown committed
4397 4398
  LINT_INIT(inc_pos);

4399
  if (mi->rli.relay_log.description_event_for_queue->binlog_version<4 &&
4400
      (uchar)buf[EVENT_TYPE_OFFSET] != FORMAT_DESCRIPTION_EVENT /* a way to escape */)
unknown's avatar
unknown committed
4401
    DBUG_RETURN(queue_old_event(mi,buf,event_len));
4402

Marc Alff's avatar
Marc Alff committed
4403
  mysql_mutex_lock(&mi->data_lock);
unknown's avatar
unknown committed
4404

4405
  switch ((uchar)buf[EVENT_TYPE_OFFSET]) {
4406
  case STOP_EVENT:
4407 4408
    /*
      We needn't write this event to the relay log. Indeed, it just indicates a
unknown's avatar
unknown committed
4409 4410 4411
      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
4412
      prepared statements' deletion are TODO only when we binlog prep stmts).
unknown's avatar
unknown committed
4413

unknown's avatar
unknown committed
4414 4415 4416 4417
      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).
4418 4419
    */
    goto err;
4420 4421
  case ROTATE_EVENT:
  {
4422 4423 4424 4425 4426
    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
4427
    {
Andrei Elkin's avatar
Andrei Elkin committed
4428
      error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
4429
      goto err;
unknown's avatar
unknown committed
4430
    }
4431 4432 4433 4434 4435 4436 4437 4438 4439 4440 4441 4442 4443 4444 4445 4446 4447 4448 4449 4450 4451 4452 4453 4454 4455 4456 4457 4458 4459 4460 4461 4462 4463 4464 4465 4466 4467 4468 4469 4470 4471 4472 4473 4474 4475 4476 4477 4478 4479 4480 4481 4482
    /* 
       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;
      }
4483 4484 4485 4486
    /*
      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
4487
    inc_pos= 0;
4488 4489
    break;
  }
4490 4491 4492 4493 4494 4495 4496
  case FORMAT_DESCRIPTION_EVENT:
  {
    /*
      Create an event, and save it (when we rotate the relay log, we will have
      to write this event again).
    */
    /*
4497 4498 4499
      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).
4500
    */
4501
    Format_description_log_event* tmp;
4502
    const char* errmsg;
4503 4504
    // mark it as undefined that is irrelevant anymore
    mi->checksum_alg_before_fd= BINLOG_CHECKSUM_ALG_UNDEF;
4505
    if (!(tmp= (Format_description_log_event*)
4506
          Log_event::read_log_event(buf, event_len, &errmsg,
4507 4508
                                    mi->rli.relay_log.description_event_for_queue,
                                    1)))
4509
    {
Andrei Elkin's avatar
Andrei Elkin committed
4510
      error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
4511 4512
      goto err;
    }
4513 4514
    delete mi->rli.relay_log.description_event_for_queue;
    mi->rli.relay_log.description_event_for_queue= tmp;
4515 4516 4517 4518 4519 4520
    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
4521
    /*
4522
       Though this does some conversion to the slave's format, this will
unknown's avatar
unknown committed
4523
       preserve the master's binlog format version, and number of event types.
4524
    */
unknown's avatar
unknown committed
4525
    /*
4526
       If the event was not requested by the slave (the slave did not ask for
unknown's avatar
unknown committed
4527
       it), i.e. has end_log_pos=0, we do not increment mi->master_log_pos
4528 4529 4530
    */
    inc_pos= uint4korr(buf+LOG_POS_OFFSET) ? event_len : 0;
    DBUG_PRINT("info",("binlog format is now %d",
unknown's avatar
unknown committed
4531
                       mi->rli.relay_log.description_event_for_queue->binlog_version));
4532 4533 4534

  }
  break;
Andrei Elkin's avatar
Andrei Elkin committed
4535 4536 4537 4538 4539 4540 4541

  case HEARTBEAT_LOG_EVENT:
  {
    /*
      HB (heartbeat) cannot come before RL (Relay)
    */
    char  llbuf[22];
unknown's avatar
unknown committed
4542 4543 4544 4545 4546
    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
4547 4548 4549 4550 4551 4552 4553 4554 4555 4556 4557 4558 4559 4560 4561 4562 4563 4564 4565 4566 4567 4568 4569 4570 4571 4572 4573 4574 4575 4576 4577 4578 4579 4580 4581 4582 4583 4584 4585 4586 4587
    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.
       Slave can not have a difference in coordinates except in the only
       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).

       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)
        || mi->master_log_pos != hb.log_pos)
    {
      /* 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;
4588

4589
  default:
unknown's avatar
unknown committed
4590
    inc_pos= event_len;
4591 4592
    break;
  }
4593

4594
  /*
4595
    If we filter events master-side (eg. @@skip_replication), we will see holes
4596 4597 4598 4599 4600 4601 4602 4603 4604
    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
4605
    DBUG_PRINT("info", ("Adjust master_log_pos %llu->%llu to account for "
4606
                        "master-side filtering",
Sergei Golubchik's avatar
Sergei Golubchik committed
4607
                        mi->master_log_pos + inc_pos, event_pos));
4608 4609
  }

unknown's avatar
unknown committed
4610 4611
  /*
     If this event is originating from this server, don't queue it.
4612
     We don't check this for 3.23 events because it's simpler like this; 3.23
unknown's avatar
unknown committed
4613 4614
     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
4615 4616 4617 4618 4619 4620 4621 4622
     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
4623
  mysql_mutex_lock(log_lock);
Andrei Elkin's avatar
Andrei Elkin committed
4624
  s_id= uint4korr(buf + SERVER_ID_OFFSET);
4625 4626
  if ((s_id == global_system_variables.server_id &&
       !mi->rli.replicate_same_server_id) ||
Andrei Elkin's avatar
Andrei Elkin committed
4627 4628 4629 4630 4631 4632 4633 4634 4635 4636
      /*
        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 */
4637 4638
        (buf[EVENT_TYPE_OFFSET] != FORMAT_DESCRIPTION_EVENT &&
         buf[EVENT_TYPE_OFFSET] != ROTATE_EVENT))))
4639
  {
4640 4641
    /*
      Do not write it to the relay log.
4642 4643 4644 4645 4646 4647 4648 4649
      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.
4650 4651 4652
      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
4653 4654 4655
      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.
4656
    */
4657 4658
    if (!(s_id == global_system_variables.server_id &&
          !mi->rli.replicate_same_server_id) ||
4659 4660 4661
        (buf[EVENT_TYPE_OFFSET] != FORMAT_DESCRIPTION_EVENT &&
         buf[EVENT_TYPE_OFFSET] != ROTATE_EVENT &&
         buf[EVENT_TYPE_OFFSET] != STOP_EVENT))
unknown's avatar
unknown committed
4662
    {
4663
      mi->master_log_pos+= inc_pos;
unknown's avatar
unknown committed
4664 4665 4666 4667
      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;
    }
4668
    rli->relay_log.signal_update(); // the slave SQL thread needs to re-check
Andrei Elkin's avatar
Andrei Elkin committed
4669 4670
    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
4671
  }
unknown's avatar
unknown committed
4672 4673 4674
  else
  {
    /* write the event to the relay log */
4675
    if (likely(!(rli->relay_log.appendv(buf,event_len,0))))
4676 4677
    {
      mi->master_log_pos+= inc_pos;
unknown's avatar
unknown committed
4678
      DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
4679 4680
      rli->relay_log.harvest_bytes_written(&rli->log_space_total);
    }
4681
    else
Andrei Elkin's avatar
Andrei Elkin committed
4682 4683 4684
    {
      error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
    }
4685
    rli->ign_master_log_name_end[0]= 0; // last event is not ignored
4686 4687
    if (save_buf != NULL)
      buf= save_buf;
unknown's avatar
unknown committed
4688
  }
Marc Alff's avatar
Marc Alff committed
4689
  mysql_mutex_unlock(log_lock);
4690

Andrei Elkin's avatar
Andrei Elkin committed
4691 4692
skip_relay_logging:
  
4693
err:
4694
  if (unlock_data_lock)
Sergei Golubchik's avatar
Sergei Golubchik committed
4695
    mysql_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
4696
  DBUG_PRINT("info", ("error: %d", error));
Andrei Elkin's avatar
Andrei Elkin committed
4697 4698 4699 4700 4701
  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
4702
  DBUG_RETURN(error);
4703 4704
}

4705

4706
void end_relay_log_info(Relay_log_info* rli)
4707
{
4708 4709
  DBUG_ENTER("end_relay_log_info");

4710
  if (!rli->inited)
4711
    DBUG_VOID_RETURN;
4712
  if (rli->info_fd >= 0)
unknown's avatar
unknown committed
4713 4714
  {
    end_io_cache(&rli->info_file);
Marc Alff's avatar
Marc Alff committed
4715
    mysql_file_close(rli->info_fd, MYF(MY_WME));
unknown's avatar
unknown committed
4716 4717
    rli->info_fd = -1;
  }
4718
  if (rli->cur_log_fd >= 0)
unknown's avatar
unknown committed
4719 4720
  {
    end_io_cache(&rli->cache_buf);
Marc Alff's avatar
Marc Alff committed
4721
    mysql_file_close(rli->cur_log_fd, MYF(MY_WME));
unknown's avatar
unknown committed
4722 4723
    rli->cur_log_fd = -1;
  }
4724
  rli->inited = 0;
4725
  rli->relay_log.close(LOG_CLOSE_INDEX | LOG_CLOSE_STOP_EVENT);
4726
  rli->relay_log.harvest_bytes_written(&rli->log_space_total);
unknown's avatar
unknown committed
4727 4728 4729 4730 4731 4732
  /*
    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();
4733
  DBUG_VOID_RETURN;
4734 4735
}

4736 4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747 4748 4749 4750 4751 4752 4753 4754

/**
  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;
4755
  if (thd && thd->slave_thread)
4756 4757 4758 4759 4760
    thd->clear_active_vio();
#endif
}


unknown's avatar
unknown committed
4761 4762
/*
  Try to connect until successful or slave killed
4763

unknown's avatar
unknown committed
4764 4765
  SYNPOSIS
    safe_connect()
unknown's avatar
unknown committed
4766 4767 4768
    thd                 Thread handler for slave
    mysql               MySQL connection handle
    mi                  Replication handle
4769

unknown's avatar
unknown committed
4770
  RETURN
unknown's avatar
unknown committed
4771 4772
    0   ok
    #   Error
unknown's avatar
unknown committed
4773
*/
4774

4775
static int safe_connect(THD* thd, MYSQL* mysql, Master_info* mi)
unknown's avatar
unknown committed
4776
{
4777 4778 4779
  DBUG_ENTER("safe_connect");

  DBUG_RETURN(connect_to_master(thd, mysql, mi, 0, 0));
unknown's avatar
unknown committed
4780 4781
}

4782

4783
/*
unknown's avatar
unknown committed
4784 4785
  SYNPOSIS
    connect_to_master()
unknown's avatar
unknown committed
4786

unknown's avatar
unknown committed
4787 4788 4789
  IMPLEMENTATION
    Try to connect until successful or slave killed or we have retried
    master_retry_count times
4790
*/
unknown's avatar
unknown committed
4791

4792
static int connect_to_master(THD* thd, MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
4793
                             bool reconnect, bool suppress_warnings)
unknown's avatar
unknown committed
4794
{
4795
  int slave_was_killed;
unknown's avatar
unknown committed
4796
  int last_errno= -2;                           // impossible error
4797
  ulong err_count=0;
unknown's avatar
unknown committed
4798
  char llbuff[22];
4799
  DBUG_ENTER("connect_to_master");
unknown's avatar
unknown committed
4800

unknown's avatar
unknown committed
4801
#ifndef DBUG_OFF
4802
  mi->events_till_disconnect = disconnect_slave_event_count;
unknown's avatar
unknown committed
4803
#endif
4804
  ulong client_flag= CLIENT_REMEMBER_OPTIONS;
4805
  if (opt_slave_compressed_protocol)
unknown's avatar
unknown committed
4806
    client_flag=CLIENT_COMPRESS;                /* We will use compression */
4807

4808 4809
  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
4810

unknown's avatar
unknown committed
4811 4812
#ifdef HAVE_OPENSSL
  if (mi->ssl)
4813
  {
unknown's avatar
unknown committed
4814
    mysql_ssl_set(mysql,
unknown's avatar
unknown committed
4815
                  mi->ssl_key[0]?mi->ssl_key:0,
unknown's avatar
unknown committed
4816
                  mi->ssl_cert[0]?mi->ssl_cert:0,
unknown's avatar
unknown committed
4817 4818 4819
                  mi->ssl_ca[0]?mi->ssl_ca:0,
                  mi->ssl_capath[0]?mi->ssl_capath:0,
                  mi->ssl_cipher[0]?mi->ssl_cipher:0);
4820 4821 4822
    mysql_options(mysql, MYSQL_OPT_SSL_VERIFY_SERVER_CERT,
                  &mi->ssl_verify_server_cert);
  }
unknown's avatar
unknown committed
4823 4824
#endif

4825 4826 4827 4828
  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);

4829 4830 4831 4832
  /* 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);

4833 4834 4835 4836 4837 4838 4839 4840 4841 4842
  /* 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);
  }
4843
  while (!(slave_was_killed = io_slave_killed(thd,mi)) &&
unknown's avatar
unknown committed
4844 4845 4846
         (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
4847
  {
4848
    /* Don't repeat last error */
unknown's avatar
SCRUM  
unknown committed
4849
    if ((int)mysql_errno(mysql) != last_errno)
4850
    {
unknown's avatar
SCRUM  
unknown committed
4851
      last_errno=mysql_errno(mysql);
unknown's avatar
unknown committed
4852
      suppress_warnings= 0;
4853 4854
      mi->report(ERROR_LEVEL, last_errno,
                 "error %s to master '%s@%s:%d'"
4855
                 " - retry-time: %d  retries: %lu  message: %s",
4856 4857
                 (reconnect ? "reconnecting" : "connecting"),
                 mi->user, mi->host, mi->port,
4858 4859
                 mi->connect_retry, master_retry_count,
                 mysql_error(mysql));
4860
    }
unknown's avatar
unknown committed
4861 4862 4863
    /*
      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
4864
      do not want to have election triggered on the first failure to
unknown's avatar
unknown committed
4865
      connect
4866
    */
4867
    if (++err_count == master_retry_count)
4868 4869
    {
      slave_was_killed=1;
unknown's avatar
unknown committed
4870 4871
      if (reconnect)
        change_rpl_status(RPL_ACTIVE_SLAVE,RPL_LOST_SOLDIER);
4872 4873
      break;
    }
4874
    slave_sleep(thd,mi->connect_retry,io_slave_killed, mi);
unknown's avatar
unknown committed
4875
  }
4876

4877 4878
  if (!slave_was_killed)
  {
4879
    mi->clear_error(); // clear possible left over reconnect error
unknown's avatar
unknown committed
4880
    if (reconnect)
unknown's avatar
unknown committed
4881
    {
4882
      if (!suppress_warnings && global_system_variables.log_warnings)
unknown's avatar
unknown committed
4883
        sql_print_information("Slave: connected to master '%s@%s:%d',\
4884
replication resumed in log '%s' at position %s", mi->user,
unknown's avatar
unknown committed
4885 4886 4887
                        mi->host, mi->port,
                        IO_RPL_LOG_NAME,
                        llstr(mi->master_log_pos,llbuff));
unknown's avatar
unknown committed
4888
    }
unknown's avatar
unknown committed
4889 4890 4891
    else
    {
      change_rpl_status(RPL_IDLE_SLAVE,RPL_ACTIVE_SLAVE);
4892 4893
      general_log_print(thd, COM_CONNECT_OUT, "%s@%s:%d",
                        mi->user, mi->host, mi->port);
unknown's avatar
unknown committed
4894
    }
4895
#ifdef SIGNAL_WITH_VIO_CLOSE
4896
    thd->set_active_vio(mysql->net.vio);
unknown's avatar
unknown committed
4897
#endif
4898
  }
4899
  mysql->reconnect= 1;
4900 4901
  DBUG_PRINT("exit",("slave_was_killed: %d", slave_was_killed));
  DBUG_RETURN(slave_was_killed);
unknown's avatar
unknown committed
4902 4903
}

4904

unknown's avatar
unknown committed
4905
/*
4906
  safe_reconnect()
unknown's avatar
unknown committed
4907

unknown's avatar
unknown committed
4908 4909 4910
  IMPLEMENTATION
    Try to connect until successful or slave killed or we have retried
    master_retry_count times
unknown's avatar
unknown committed
4911 4912
*/

4913
static int safe_reconnect(THD* thd, MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
4914
                          bool suppress_warnings)
unknown's avatar
unknown committed
4915
{
unknown's avatar
unknown committed
4916 4917
  DBUG_ENTER("safe_reconnect");
  DBUG_RETURN(connect_to_master(thd, mysql, mi, 1, suppress_warnings));
unknown's avatar
unknown committed
4918 4919
}

unknown's avatar
unknown committed
4920

He Zhenxing's avatar
He Zhenxing committed
4921 4922 4923 4924 4925 4926 4927 4928 4929 4930 4931 4932 4933 4934 4935 4936 4937 4938 4939 4940 4941 4942 4943 4944 4945 4946 4947 4948 4949 4950 4951 4952 4953 4954 4955 4956 4957 4958 4959 4960 4961 4962 4963 4964 4965 4966 4967 4968 4969 4970
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);

4971 4972 4973
  if (mi->user == NULL
      || mi->user[0] == 0
      || io_slave_killed(thd, mi)
He Zhenxing's avatar
He Zhenxing committed
4974 4975 4976 4977 4978 4979 4980 4981 4982 4983 4984 4985 4986 4987
      || !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;
}

4988 4989 4990 4991 4992 4993
/*
  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
4994
    rli                 Relay log information
4995 4996

  NOTES
4997 4998 4999
    - 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.
5000 5001 5002 5003 5004 5005 5006 5007 5008 5009 5010 5011 5012 5013
    - 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
5014 5015
    0   ok
    1   write error
5016 5017
*/

5018
bool flush_relay_log_info(Relay_log_info* rli)
5019
{
5020
  bool error=0;
5021
  DBUG_ENTER("flush_relay_log_info");
5022 5023

  if (unlikely(rli->no_storage))
5024
    DBUG_RETURN(0);
5025

5026 5027 5028
  IO_CACHE *file = &rli->info_file;
  char buff[FN_REFLEN*2+22*2+4], *pos;

5029
  my_b_seek(file, 0L);
5030
  pos=strmov(buff, rli->group_relay_log_name);
5031
  *pos++='\n';
5032
  pos= longlong10_to_str(rli->group_relay_log_pos, pos, 10);
5033
  *pos++='\n';
5034
  pos=strmov(pos, rli->group_master_log_name);
5035
  *pos++='\n';
5036
  pos=longlong10_to_str(rli->group_master_log_pos, pos, 10);
5037
  *pos='\n';
5038
  if (my_b_write(file, (uchar*) buff, (size_t) (pos-buff)+1))
5039 5040 5041
    error=1;
  if (flush_io_cache(file))
    error=1;
5042 5043 5044 5045 5046 5047 5048 5049
  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;
  }
5050 5051 5052 5053
  /* 
    Flushing the relay log is done by the slave I/O thread 
    or by the user on STOP SLAVE. 
   */
5054
  DBUG_RETURN(error);
5055 5056
}

5057

unknown's avatar
unknown committed
5058
/*
5059
  Called when we notice that the current "hot" log got rotated under our feet.
unknown's avatar
unknown committed
5060 5061
*/

5062
static IO_CACHE *reopen_relay_log(Relay_log_info *rli, const char **errmsg)
5063
{
5064
  DBUG_ENTER("reopen_relay_log");
5065 5066
  DBUG_ASSERT(rli->cur_log != &rli->cache_buf);
  DBUG_ASSERT(rli->cur_log_fd == -1);
unknown's avatar
unknown committed
5067 5068

  IO_CACHE *cur_log = rli->cur_log=&rli->cache_buf;
5069
  if ((rli->cur_log_fd=open_binlog(cur_log,rli->event_relay_log_name,
unknown's avatar
unknown committed
5070
                                   errmsg)) <0)
unknown's avatar
unknown committed
5071
    DBUG_RETURN(0);
5072 5073
  /*
    We want to start exactly where we was before:
unknown's avatar
unknown committed
5074 5075
    relay_log_pos       Current log pos
    pending             Number of bytes already processed from the event
5076
  */
5077
  rli->event_relay_log_pos= max(rli->event_relay_log_pos, BIN_LOG_HEADER_SIZE);
5078
  my_b_seek(cur_log,rli->event_relay_log_pos);
unknown's avatar
unknown committed
5079
  DBUG_RETURN(cur_log);
5080 5081
}

unknown's avatar
unknown committed
5082

5083 5084 5085 5086 5087 5088 5089 5090 5091 5092
/**
  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.
*/
5093
static Log_event* next_event(Relay_log_info* rli)
5094 5095 5096
{
  Log_event* ev;
  IO_CACHE* cur_log = rli->cur_log;
Marc Alff's avatar
Marc Alff committed
5097
  mysql_mutex_t *log_lock = rli->relay_log.get_log_lock();
5098 5099
  const char* errmsg=0;
  THD* thd = rli->sql_thd;
unknown's avatar
unknown committed
5100
  DBUG_ENTER("next_event");
5101

5102 5103
  DBUG_ASSERT(thd != 0);

5104 5105 5106 5107 5108
#ifndef DBUG_OFF
  if (abort_slave_event_count && !rli->events_till_abort--)
    DBUG_RETURN(0);
#endif

unknown's avatar
unknown committed
5109 5110
  /*
    For most operations we need to protect rli members with data_lock,
5111 5112 5113
    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
5114
    mysql_cond_wait() with the non-data_lock mutex
unknown's avatar
unknown committed
5115
  */
Marc Alff's avatar
Marc Alff committed
5116
  mysql_mutex_assert_owner(&rli->data_lock);
unknown's avatar
unknown committed
5117

5118
  while (!sql_slave_killed(thd,rli))
unknown's avatar
unknown committed
5119 5120 5121
  {
    /*
      We can have two kinds of log reading:
unknown's avatar
unknown committed
5122 5123 5124 5125 5126 5127 5128 5129
      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
5130
    */
5131 5132 5133 5134
    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
5135
      mysql_mutex_lock(log_lock);
unknown's avatar
unknown committed
5136 5137

      /*
unknown's avatar
unknown committed
5138 5139
        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
5140
      */
unknown's avatar
unknown committed
5141
      if (rli->relay_log.get_open_count() != rli->cur_log_old_open_count)
5142
      {
unknown's avatar
unknown committed
5143 5144
        // 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
5145
        mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
5146 5147 5148
        if (!cur_log)                           // No more log files
          goto err;
        hot_log=0;                              // Using old binary log
5149 5150
      }
    }
5151 5152 5153 5154 5155 5156 5157
    /* 
      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;
5158 5159
#ifndef DBUG_OFF
    {
5160
      /* This is an assertion which sometimes fails, let's try to track it */
5161
      char llbuf1[22], llbuf2[22];
5162
      DBUG_PRINT("info", ("my_b_tell(cur_log)=%s rli->event_relay_log_pos=%s",
5163
                          llstr(my_b_tell(cur_log),llbuf1),
5164 5165 5166
                          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);
5167 5168
    }
#endif
unknown's avatar
unknown committed
5169 5170
    /*
      Relay log is always in new format - if the master is 3.23, the
5171
      I/O thread will convert the format for us.
5172 5173
      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
5174 5175
      logs, which may even have been deleted. So we need to write this
      description event at the beginning of the relay log.
5176 5177
      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.
5178
      But if the relay log is created by new_file(): then the solution is:
5179
      MYSQL_BIN_LOG::open() will write the buffered description event.
unknown's avatar
unknown committed
5180
    */
5181 5182 5183
    if ((ev= Log_event::read_log_event(cur_log,0,
                                       rli->relay_log.description_event_for_exec,
                                       opt_slave_sql_verify_checksum)))
5184

5185 5186
    {
      DBUG_ASSERT(thd==rli->sql_thd);
5187 5188 5189 5190 5191
      /*
        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);
5192
      if (hot_log)
Marc Alff's avatar
Marc Alff committed
5193
        mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
5194
      DBUG_RETURN(ev);
5195 5196
    }
    DBUG_ASSERT(thd==rli->sql_thd);
unknown's avatar
unknown committed
5197
    if (opt_reckless_slave)                     // For mysql-test
unknown's avatar
unknown committed
5198
      cur_log->error = 0;
unknown's avatar
unknown committed
5199
    if (cur_log->error < 0)
unknown's avatar
unknown committed
5200 5201
    {
      errmsg = "slave SQL thread aborted because of I/O error";
unknown's avatar
unknown committed
5202
      if (hot_log)
Marc Alff's avatar
Marc Alff committed
5203
        mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
5204 5205
      goto err;
    }
5206 5207
    if (!cur_log->error) /* EOF */
    {
unknown's avatar
unknown committed
5208
      /*
unknown's avatar
unknown committed
5209 5210 5211
        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
5212
      */
5213 5214
      if (hot_log)
      {
5215 5216 5217 5218 5219 5220 5221 5222 5223 5224
        /*
          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").
5225 5226 5227 5228 5229 5230 5231 5232 5233 5234

          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.
5235
        */
5236
        time_t save_timestamp= rli->last_master_timestamp;
5237 5238
        rli->last_master_timestamp= 0;

unknown's avatar
unknown committed
5239
        DBUG_ASSERT(rli->relay_log.get_open_count() ==
unknown's avatar
unknown committed
5240
                    rli->cur_log_old_open_count);
5241 5242 5243 5244 5245

        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"));
5246
          ev= new Rotate_log_event(rli->ign_master_log_name_end,
5247
                                   0, rli->ign_master_log_pos_end,
unknown's avatar
unknown committed
5248
                                   Rotate_log_event::DUP_NAME);
5249
          rli->ign_master_log_name_end[0]= 0;
Marc Alff's avatar
Marc Alff committed
5250
          mysql_mutex_unlock(log_lock);
5251 5252 5253 5254 5255 5256 5257 5258 5259 5260
          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
5261 5262 5263 5264
        /*
          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
5265
        mysql_mutex_unlock(&rli->data_lock);
5266 5267

        /*
unknown's avatar
unknown committed
5268
          Possible deadlock :
5269 5270 5271 5272 5273 5274 5275 5276 5277 5278 5279
          - 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
5280
          master does not send anything). Then the I/O thread stops waiting
5281 5282 5283 5284 5285 5286 5287 5288 5289 5290
          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
5291 5292
          be stopped, and the SQL thread sets ignore_log_space_limit to 0 when
          it stops.
5293
        */
Marc Alff's avatar
Marc Alff committed
5294
        mysql_mutex_lock(&rli->log_space_lock);
5295 5296 5297 5298 5299 5300 5301 5302 5303 5304 5305 5306 5307 5308 5309 5310 5311 5312 5313 5314 5315 5316 5317 5318 5319

        /* 
          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;
        }

5320
        /*
5321 5322 5323 5324
          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.
5325
        */
Marc Alff's avatar
Marc Alff committed
5326 5327
        mysql_mutex_unlock(&rli->log_space_lock);
        mysql_cond_broadcast(&rli->log_space_cond);
Andrei Elkin's avatar
Andrei Elkin committed
5328 5329
        // Note that wait_for_update_relay_log unlocks lock_log !
        rli->relay_log.wait_for_update_relay_log(rli->sql_thd);
5330
        // re-acquire data lock since we released it earlier
Marc Alff's avatar
Marc Alff committed
5331
        mysql_mutex_lock(&rli->data_lock);
5332
        rli->last_master_timestamp= save_timestamp;
unknown's avatar
unknown committed
5333
        continue;
5334
      }
unknown's avatar
unknown committed
5335
      /*
unknown's avatar
unknown committed
5336 5337 5338
        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
5339 5340 5341
      */
      end_io_cache(cur_log);
      DBUG_ASSERT(rli->cur_log_fd >= 0);
Marc Alff's avatar
Marc Alff committed
5342
      mysql_file_close(rli->cur_log_fd, MYF(MY_WME));
unknown's avatar
unknown committed
5343
      rli->cur_log_fd = -1;
unknown's avatar
unknown committed
5344

5345
      if (relay_log_purge)
unknown's avatar
unknown committed
5346
      {
unknown's avatar
unknown committed
5347
        /*
5348 5349 5350 5351 5352 5353 5354 5355 5356 5357
          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
5358
        if (rli->relay_log.purge_first_log
5359 5360 5361
            (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
5362 5363 5364 5365
        {
          errmsg = "Error purging processed logs";
          goto err;
        }
unknown's avatar
unknown committed
5366
      }
5367 5368
      else
      {
unknown's avatar
unknown committed
5369 5370 5371 5372 5373 5374 5375 5376 5377 5378 5379 5380 5381 5382 5383 5384
        /*
          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
5385
      }
5386 5387 5388 5389 5390 5391 5392 5393 5394 5395 5396 5397 5398

      /*
        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
5399
        mysql_mutex_lock(log_lock);
unknown's avatar
unknown committed
5400 5401
      if (rli->relay_log.is_active(rli->linfo.log_file_name))
      {
unknown's avatar
unknown committed
5402 5403 5404 5405 5406
        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);

        /*
5407 5408 5409 5410 5411 5412 5413 5414 5415 5416 5417 5418 5419 5420 5421 5422 5423 5424 5425 5426 5427 5428 5429 5430 5431 5432 5433 5434 5435 5436 5437 5438 5439 5440 5441 5442 5443 5444 5445 5446 5447 5448 5449 5450 5451 5452 5453 5454 5455 5456 5457 5458 5459 5460 5461 5462 5463
           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
5464
        */
5465 5466

        my_b_seek(cur_log, (my_off_t) 0);
unknown's avatar
unknown committed
5467
        if (check_binlog_magic(cur_log,&errmsg))
5468
        {
Marc Alff's avatar
Marc Alff committed
5469 5470
          if (!hot_log)
            mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
5471
          goto err;
5472
        }
Marc Alff's avatar
Marc Alff committed
5473 5474
        if (!hot_log)
          mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
5475
        continue;
5476
      }
Marc Alff's avatar
Marc Alff committed
5477 5478
      if (!hot_log)
        mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
5479
      /*
unknown's avatar
unknown committed
5480 5481 5482
        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
5483 5484 5485
      */
      // 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
5486 5487
                                       &errmsg)) <0)
        goto err;
5488
    }
unknown's avatar
unknown committed
5489
    else
5490
    {
unknown's avatar
unknown committed
5491
      /*
unknown's avatar
unknown committed
5492 5493
        Read failed with a non-EOF error.
        TODO: come up with something better to handle this error
unknown's avatar
unknown committed
5494 5495
      */
      if (hot_log)
Marc Alff's avatar
Marc Alff committed
5496
        mysql_mutex_unlock(log_lock);
5497
      sql_print_error("Slave SQL thread: I/O error reading \
unknown's avatar
unknown committed
5498
event(errno: %d  cur_log->error: %d)",
unknown's avatar
unknown committed
5499
                      my_errno,cur_log->error);
unknown's avatar
unknown committed
5500
      // set read position to the beginning of the event
5501
      my_b_seek(cur_log,rli->event_relay_log_pos);
unknown's avatar
unknown committed
5502 5503
      /* otherwise, we have had a partial read */
      errmsg = "Aborting slave SQL thread because of partial event read";
unknown's avatar
unknown committed
5504
      break;                                    // To end of function
5505 5506
    }
  }
5507
  if (!errmsg && global_system_variables.log_warnings)
5508
  {
unknown's avatar
unknown committed
5509
    sql_print_information("Error reading relay log event: %s",
5510 5511 5512
                          "slave SQL thread was killed");
    DBUG_RETURN(0);
  }
unknown's avatar
unknown committed
5513

5514
err:
5515 5516
  if (errmsg)
    sql_print_error("Error reading relay log event: %s", errmsg);
unknown's avatar
unknown committed
5517
  DBUG_RETURN(0);
5518 5519
}

5520 5521 5522
/*
  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
5523
  locks; here we don't, so this function is mainly taking locks).
5524 5525
  Returns nothing as we cannot catch any error (MYSQL_BIN_LOG::new_file()
  is void).
5526 5527
*/

5528
int rotate_relay_log(Master_info* mi)
5529 5530
{
  DBUG_ENTER("rotate_relay_log");
5531
  Relay_log_info* rli= &mi->rli;
5532
  int error= 0;
unknown's avatar
unknown committed
5533

Luis Soares's avatar
Luis Soares committed
5534
  DBUG_EXECUTE_IF("crash_before_rotate_relaylog", DBUG_SUICIDE(););
5535

unknown's avatar
unknown committed
5536
  /*
unknown's avatar
unknown committed
5537 5538 5539
     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).
  */
5540 5541
  if (!rli->inited)
  {
unknown's avatar
unknown committed
5542
    DBUG_PRINT("info", ("rli->inited == 0"));
unknown's avatar
unknown committed
5543
    goto end;
5544
  }
unknown's avatar
unknown committed
5545

5546
  /* If the relay log is closed, new_file() will do nothing. */
5547 5548
  if ((error= rli->relay_log.new_file()))
    goto end;
unknown's avatar
unknown committed
5549

5550 5551 5552 5553
  /*
    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
5554 5555 5556 5557 5558 5559
    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.
5560 5561
    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
5562 5563

    Note that it needs to be protected by mi->data_lock.
5564
  */
Sergei Golubchik's avatar
Sergei Golubchik committed
5565
  mysql_mutex_assert_owner(&mi->data_lock);
5566
  rli->relay_log.harvest_bytes_written(&rli->log_space_total);
unknown's avatar
unknown committed
5567
end:
5568
  DBUG_RETURN(error);
5569 5570
}

unknown's avatar
unknown committed
5571

unknown's avatar
unknown committed
5572 5573 5574
/**
   Detects, based on master's version (as found in the relay log), if master
   has a certain bug.
5575
   @param rli Relay_log_info which tells the master's version
unknown's avatar
unknown committed
5576
   @param bug_id Number of the bug as found in bugs.mysql.com
5577
   @param report bool report error message, default TRUE
5578 5579 5580 5581 5582 5583 5584

   @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
5585 5586
   @return TRUE if master has the bug, FALSE if it does not.
*/
5587 5588
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
5589 5590 5591 5592 5593 5594 5595 5596
{
  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
5597
    {24432, { 5, 0, 24 }, { 5, 0, 38 } },
5598 5599 5600
    {24432, { 5, 1, 12 }, { 5, 1, 17 } },
    {33029, { 5, 0,  0 }, { 5, 0, 58 } },
    {33029, { 5, 1,  0 }, { 5, 1, 12 } },
5601
    {37426, { 5, 1,  0 }, { 5, 1, 26 } },
unknown's avatar
unknown committed
5602 5603
  };
  const uchar *master_ver=
5604
    rli->relay_log.description_event_for_exec->server_version_split.ver;
unknown's avatar
unknown committed
5605

5606
  DBUG_ASSERT(sizeof(rli->relay_log.description_event_for_exec->server_version_split.ver) == 3);
unknown's avatar
unknown committed
5607 5608 5609 5610 5611 5612 5613 5614

  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) &&
5615 5616
        (memcmp(fixed_in,      master_ver, 3) >  0) &&
        (pred == NULL || (*pred)(param)))
unknown's avatar
unknown committed
5617
    {
5618 5619
      if (!report)
	return TRUE;
unknown's avatar
unknown committed
5620 5621 5622 5623 5624
      // 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);
5625
      // a verbose message for the error log
5626 5627 5628
      rli->report(ERROR_LEVEL, ER_UNKNOWN_ERROR,
                  "According to the master's version ('%s'),"
                  " it is probable that master suffers from this bug:"
5629 5630 5631 5632 5633 5634 5635 5636 5637 5638 5639 5640 5641 5642 5643 5644
                      " 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
5645 5646 5647 5648 5649 5650
      return TRUE;
    }
  }
  return FALSE;
}

5651 5652 5653 5654 5655 5656 5657 5658 5659 5660 5661 5662
/**
   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)
{
5663
  if (thd->rli_slave)
5664
  {
5665
    DBUG_EXECUTE_IF("simulate_bug33029", return TRUE;);
5666
    return rpl_master_has_bug(thd->rli_slave, 33029, FALSE, NULL, NULL);
5667 5668 5669 5670
  }
  return FALSE;
}

5671 5672 5673 5674
/**
  @} (end of group Replication)
*/

unknown's avatar
SCRUM  
unknown committed
5675
#endif /* HAVE_REPLICATION */