slave.cc 152 KB
Newer Older
1
/* Copyright (C) 2000-2003 MySQL AB
2

unknown's avatar
unknown committed
3 4
   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
5
   the Free Software Foundation; version 2 of the License.
6

unknown's avatar
unknown committed
7 8 9 10
   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.
11

unknown's avatar
unknown committed
12 13 14 15
   You should have received a copy of the GNU General Public License
   along with this program; if not, write to the Free Software
   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA */

16 17 18 19 20 21 22 23 24 25 26

/**
  @addtogroup Replication
  @{

  @file

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

unknown's avatar
unknown committed
27
#include "mysql_priv.h"
unknown's avatar
SCRUM  
unknown committed
28

unknown's avatar
unknown committed
29
#include <mysql.h>
30
#include <myisam.h>
31
#include "slave.h"
32 33
#include "rpl_mi.h"
#include "rpl_rli.h"
34
#include "sql_repl.h"
unknown's avatar
unknown committed
35
#include "rpl_filter.h"
36
#include "repl_failsafe.h"
unknown's avatar
unknown committed
37
#include <thr_alarm.h>
unknown's avatar
unknown committed
38
#include <my_dir.h>
unknown's avatar
unknown committed
39
#include <sql_common.h>
40
#include <errmsg.h>
41
#include <mysqld_error.h>
42
#include <mysys_err.h>
unknown's avatar
SCRUM  
unknown committed
43

44 45 46
#ifdef HAVE_REPLICATION

#include "rpl_tblmap.h"
47
#include "debug_sync.h"
48

49
#define FLAGSTR(V,F) ((V)&(F)?#F" ":"")
50

51
#define MAX_SLAVE_RETRY_PAUSE 5
52 53
bool use_slave_mask = 0;
MY_BITMAP slave_error_mask;
54
char slave_skip_error_names[SHOW_VAR_FUNC_BUFF_SIZE];
55

56 57
typedef bool (*CHECK_KILLED_FUNC)(THD*,void*);

58
char* slave_load_tmpdir = 0;
59
Master_info *active_mi= 0;
60
my_bool replicate_same_server_id;
61
ulonglong relay_log_space_limit = 0;
unknown's avatar
unknown committed
62 63 64 65 66

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

67
  TODO: move the vars below under Master_info
unknown's avatar
unknown committed
68
*/
69

70
int disconnect_slave_event_count = 0, abort_slave_event_count = 0;
71
int events_till_abort = -1;
unknown's avatar
unknown committed
72

73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124
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, \
log '%s' at postion %s",
    "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",
    "failed dump request, reconnecting to try again, log '%s' at postion %s",
    "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, \
log '%s' at postion %s",
    "",
    "Slave I/O thread killed during or after a reconnect done to recover from \
failed read"
  }
};
 

125
typedef enum { SLAVE_THD_IO, SLAVE_THD_SQL} SLAVE_THD_TYPE;
unknown's avatar
unknown committed
126

127 128
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);
129
static bool wait_for_relay_log_space(Relay_log_info* rli);
130
static inline bool io_slave_killed(THD* thd,Master_info* mi);
131
static inline bool sql_slave_killed(THD* thd,Relay_log_info* rli);
132
static int init_slave_thread(THD* thd, SLAVE_THD_TYPE thd_type);
133
static void print_slave_skip_errors(void);
134 135
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
136
                          bool suppress_warnings);
137
static int connect_to_master(THD* thd, MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
138
                             bool reconnect, bool suppress_warnings);
139
static int safe_sleep(THD* thd, int sec, CHECK_KILLED_FUNC thread_killed,
unknown's avatar
unknown committed
140
                      void* thread_killed_arg);
141
static int request_table_dump(MYSQL* mysql, const char* db, const char* table);
142
static int create_table_from_dump(THD* thd, MYSQL *mysql, const char* db,
unknown's avatar
unknown committed
143
                                  const char* table_name, bool overwrite);
144
static int get_master_version_and_clock(MYSQL* mysql, Master_info* mi);
145
static Log_event* next_event(Relay_log_info* rli);
146
static int queue_event(Master_info* mi,const char* buf,ulong event_len);
147
static int terminate_slave_thread(THD *thd,
148 149
                                  pthread_mutex_t *term_lock,
                                  pthread_cond_t *term_cond,
150 151
                                  volatile uint *slave_running,
                                  bool skip_lock);
152
static bool check_io_slave_killed(THD *thd, Master_info *mi, const char *info);
153 154

/*
unknown's avatar
unknown committed
155
  Find out which replications threads are running
156

unknown's avatar
unknown committed
157 158
  SYNOPSIS
    init_thread_mask()
unknown's avatar
unknown committed
159 160 161
    mask                Return value here
    mi                  master_info for slave
    inverse             If set, returns which threads are not running
unknown's avatar
unknown committed
162 163 164 165 166 167

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

  RETURN
unknown's avatar
unknown committed
168 169
    mask        If inverse == 0, running threads
                If inverse == 1, stopped threads
170 171
*/

172
void init_thread_mask(int* mask,Master_info* mi,bool inverse)
173 174 175
{
  bool set_io = mi->slave_running, set_sql = mi->rli.slave_running;
  register int tmp_mask=0;
176 177
  DBUG_ENTER("init_thread_mask");

178 179 180 181
  if (set_io)
    tmp_mask |= SLAVE_IO;
  if (set_sql)
    tmp_mask |= SLAVE_SQL;
182 183
  if (inverse)
    tmp_mask^= (SLAVE_IO | SLAVE_SQL);
184
  *mask = tmp_mask;
185
  DBUG_VOID_RETURN;
186 187
}

188

unknown's avatar
unknown committed
189
/*
190
  lock_slave_threads()
unknown's avatar
unknown committed
191
*/
192

193
void lock_slave_threads(Master_info* mi)
194
{
195 196
  DBUG_ENTER("lock_slave_threads");

197 198 199
  //TODO: see if we can do this without dual mutex
  pthread_mutex_lock(&mi->run_lock);
  pthread_mutex_lock(&mi->rli.run_lock);
200
  DBUG_VOID_RETURN;
201 202
}

203

unknown's avatar
unknown committed
204
/*
205
  unlock_slave_threads()
unknown's avatar
unknown committed
206
*/
207

208
void unlock_slave_threads(Master_info* mi)
209
{
210 211
  DBUG_ENTER("unlock_slave_threads");

212 213 214
  //TODO: see if we can do this without dual mutex
  pthread_mutex_unlock(&mi->rli.run_lock);
  pthread_mutex_unlock(&mi->run_lock);
215
  DBUG_VOID_RETURN;
216 217
}

218

unknown's avatar
unknown committed
219
/* Initialize slave structures */
220

221 222
int init_slave()
{
223
  DBUG_ENTER("init_slave");
224

225 226 227 228 229 230
  /*
    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.
  */
  pthread_mutex_lock(&LOCK_active_mi);
231 232 233 234
  /*
    TODO: re-write this to interate through the list of files
    for multi-master
  */
235
  active_mi= new Master_info;
236

237 238 239 240 241 242 243 244 245
  /*
    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();
  }

246
  /*
247 248 249
    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.
250
  */
251 252 253 254 255
  if (!active_mi)
  {
    sql_print_error("Failed to allocate memory for the master info structure");
    goto err;
  }
256

unknown's avatar
unknown committed
257
  if (init_master_info(active_mi,master_info_file,relay_log_info_file,
unknown's avatar
unknown committed
258
                       !master_host, (SLAVE_IO | SLAVE_SQL)))
259
  {
260
    sql_print_error("Failed to initialize the master info structure");
unknown's avatar
unknown committed
261
    goto err;
262
  }
263 264 265 266

  if (server_id && !master_host && active_mi->host[0])
    master_host= active_mi->host;

267 268
  /* If server id is not set, start_slave_thread() will say it */

269
  if (master_host && !opt_skip_slave_start)
270
  {
271
    if (start_slave_threads(1 /* need mutex */,
unknown's avatar
unknown committed
272 273 274 275 276
                            0 /* no wait for start*/,
                            active_mi,
                            master_info_file,
                            relay_log_info_file,
                            SLAVE_IO | SLAVE_SQL))
unknown's avatar
unknown committed
277
    {
278
      sql_print_error("Failed to create slave threads");
unknown's avatar
unknown committed
279 280
      goto err;
    }
281
  }
282
  pthread_mutex_unlock(&LOCK_active_mi);
283
  DBUG_RETURN(0);
284

unknown's avatar
unknown committed
285
err:
286
  pthread_mutex_unlock(&LOCK_active_mi);
unknown's avatar
unknown committed
287
  DBUG_RETURN(1);
288
}
289

290

291 292 293 294 295 296 297 298 299 300 301 302
/**
  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.
  */
303
  const size_t MIN_ROOM= 10;
304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325
  DBUG_ENTER("print_slave_skip_errors");
  DBUG_ASSERT(sizeof(slave_skip_error_names) > MIN_ROOM);
  DBUG_ASSERT(MAX_SLAVE_ERROR <= 999999); // 6 digits

  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;

326
    for (errnum= 0; errnum < MAX_SLAVE_ERROR; errnum++)
327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348
    {
      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
349
/*
350
  Init function to set up array for errors that should be skipped for slave
unknown's avatar
unknown committed
351

unknown's avatar
unknown committed
352 353
  SYNOPSIS
    init_slave_skip_errors()
unknown's avatar
unknown committed
354
    arg         List of errors numbers to skip, separated with ','
unknown's avatar
unknown committed
355 356 357 358

  NOTES
    Called from get_options() in mysqld.cc on start-up
*/
359

unknown's avatar
unknown committed
360
void init_slave_skip_errors(const char* arg)
361
{
unknown's avatar
unknown committed
362
  const char *p;
363 364
  DBUG_ENTER("init_slave_skip_errors");

365
  if (bitmap_init(&slave_error_mask,0,MAX_SLAVE_ERROR,0))
366 367 368 369 370
  {
    fprintf(stderr, "Badly out of memory, please check your system status\n");
    exit(1);
  }
  use_slave_mask = 1;
371
  for (;my_isspace(system_charset_info,*arg);++arg)
372
    /* empty */;
373
  if (!my_strnncoll(system_charset_info,(uchar*)arg,4,(const uchar*)"all",4))
374 375
  {
    bitmap_set_all(&slave_error_mask);
376
    print_slave_skip_errors();
377
    DBUG_VOID_RETURN;
378 379 380 381 382 383 384 385
  }
  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);
386
    while (!my_isdigit(system_charset_info,*p) && *p)
387 388
      p++;
  }
389 390
  /* Convert slave skip errors bitmap into a printable string. */
  print_slave_skip_errors();
391
  DBUG_VOID_RETURN;
392 393
}

394
static void set_thd_in_use_temporary_tables(Relay_log_info *rli)
395 396 397 398 399 400
{
  TABLE *table;

  for (table= rli->save_temporary_tables ; table ; table= table->next)
    table->in_use= rli->sql_thd;
}
401

402
int terminate_slave_threads(Master_info* mi,int thread_mask,bool skip_lock)
403
{
404 405
  DBUG_ENTER("terminate_slave_threads");

406
  if (!mi->inited)
407
    DBUG_RETURN(0); /* successfully do nothing */
408 409
  int error,force_all = (thread_mask & SLAVE_FORCE_ALL);
  pthread_mutex_t *sql_lock = &mi->rli.run_lock, *io_lock = &mi->run_lock;
unknown's avatar
unknown committed
410

411
  if (thread_mask & (SLAVE_IO|SLAVE_FORCE_ALL))
412
  {
unknown's avatar
unknown committed
413
    DBUG_PRINT("info",("Terminating IO thread"));
414
    mi->abort_slave=1;
415
    if ((error=terminate_slave_thread(mi->io_thd, io_lock,
unknown's avatar
unknown committed
416
                                      &mi->stop_cond,
417 418
                                      &mi->slave_running,
                                      skip_lock)) &&
unknown's avatar
unknown committed
419
        !force_all)
420
      DBUG_RETURN(error);
421
  }
422
  if (thread_mask & (SLAVE_SQL|SLAVE_FORCE_ALL))
423
  {
unknown's avatar
unknown committed
424
    DBUG_PRINT("info",("Terminating SQL thread"));
425
    mi->rli.abort_slave=1;
426
    if ((error=terminate_slave_thread(mi->rli.sql_thd, sql_lock,
unknown's avatar
unknown committed
427
                                      &mi->rli.stop_cond,
428 429
                                      &mi->rli.slave_running,
                                      skip_lock)) &&
unknown's avatar
unknown committed
430
        !force_all)
431
      DBUG_RETURN(error);
432
  }
433
  DBUG_RETURN(0);
434 435
}

436

437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463
/**
   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
464
   @retval 0 All OK ER_SLAVE_NOT_RUNNING otherwise.
465

Andrei Elkin's avatar
Andrei Elkin committed
466 467 468
   @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.
469

470 471 472
 */
static int
terminate_slave_thread(THD *thd,
Andrei Elkin's avatar
Andrei Elkin committed
473 474
                       pthread_mutex_t *term_lock,
                       pthread_cond_t *term_cond,
475 476
                       volatile uint *slave_running,
                       bool skip_lock)
477
{
478
  DBUG_ENTER("terminate_slave_thread");
479
  if (!skip_lock)
480 481
  {
    pthread_mutex_lock(term_lock);
482 483 484 485 486
  }
  else
  {
    safe_mutex_assert_owner(term_lock);
  }
487 488 489
  if (!*slave_running)
  {
    if (!skip_lock)
490
    {
491 492 493 494
      /*
        if run_lock (term_lock) is acquired locally then either
        slave_running status is fine
      */
495
      pthread_mutex_unlock(term_lock);
496 497 498 499
      DBUG_RETURN(0);
    }
    else
    {
500
      DBUG_RETURN(ER_SLAVE_NOT_RUNNING);
501 502 503
    }
  }
  DBUG_ASSERT(thd != 0);
504
  THD_CHECK_SENTRY(thd);
505

unknown's avatar
unknown committed
506
  /*
507
    Is is critical to test if the slave is running. Otherwise, we might
unknown's avatar
unknown committed
508
    be referening freed memory trying to kick it
509
  */
unknown's avatar
unknown committed
510

unknown's avatar
unknown committed
511
  while (*slave_running)                        // Should always be true
512
  {
513
    int error;
514
    DBUG_PRINT("loop", ("killing slave thread"));
515

516
    pthread_mutex_lock(&thd->LOCK_thd_data);
517 518 519 520 521 522 523 524 525 526
#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)
    */
    IF_DBUG(int err= ) pthread_kill(thd->real_id, thr_client_alarm);
    DBUG_ASSERT(err != EINVAL);
#endif
    thd->awake(THD::NOT_KILLED);
527
    pthread_mutex_unlock(&thd->LOCK_thd_data);
528

unknown's avatar
unknown committed
529 530 531
    /*
      There is a small chance that slave thread might miss the first
      alarm. To protect againts it, resend the signal until it reacts
532 533
    */
    struct timespec abstime;
unknown's avatar
unknown committed
534
    set_timespec(abstime,2);
535 536
    error= pthread_cond_timedwait(term_cond, term_lock, &abstime);
    DBUG_ASSERT(error == ETIMEDOUT || error == 0);
537
  }
538 539 540 541

  DBUG_ASSERT(*slave_running == 0);

  if (!skip_lock)
542
    pthread_mutex_unlock(term_lock);
543
  DBUG_RETURN(0);
544 545
}

546

unknown's avatar
unknown committed
547
int start_slave_thread(pthread_handler h_func, pthread_mutex_t *start_lock,
unknown's avatar
unknown committed
548 549 550 551
                       pthread_mutex_t *cond_lock,
                       pthread_cond_t *start_cond,
                       volatile uint *slave_running,
                       volatile ulong *slave_run_id,
552
                       Master_info* mi,
553
                       bool high_priority)
554 555
{
  pthread_t th;
unknown's avatar
unknown committed
556 557 558
  ulong start_id;
  DBUG_ENTER("start_slave_thread");

559 560
  DBUG_ASSERT(mi->inited);

561 562 563 564 565 566 567 568 569
  if (start_lock)
    pthread_mutex_lock(start_lock);
  if (!server_id)
  {
    if (start_cond)
      pthread_cond_broadcast(start_cond);
    if (start_lock)
      pthread_mutex_unlock(start_lock);
    sql_print_error("Server id not set, will not start slave");
unknown's avatar
unknown committed
570
    DBUG_RETURN(ER_BAD_SLAVE);
571
  }
unknown's avatar
unknown committed
572

573
  if (*slave_running)
574 575 576 577 578
  {
    if (start_cond)
      pthread_cond_broadcast(start_cond);
    if (start_lock)
      pthread_mutex_unlock(start_lock);
unknown's avatar
unknown committed
579
    DBUG_RETURN(ER_SLAVE_MUST_STOP);
580
  }
unknown's avatar
unknown committed
581 582
  start_id= *slave_run_id;
  DBUG_PRINT("info",("Creating new slave thread"));
583 584
  if (high_priority)
    my_pthread_attr_setprio(&connection_attrib,CONNECT_PRIOR);
585 586 587 588
  if (pthread_create(&th, &connection_attrib, h_func, (void*)mi))
  {
    if (start_lock)
      pthread_mutex_unlock(start_lock);
unknown's avatar
unknown committed
589
    DBUG_RETURN(ER_SLAVE_THREAD);
590
  }
unknown's avatar
unknown committed
591
  if (start_cond && cond_lock) // caller has cond_lock
592 593
  {
    THD* thd = current_thd;
unknown's avatar
unknown committed
594
    while (start_id == *slave_run_id)
595
    {
unknown's avatar
unknown committed
596
      DBUG_PRINT("sleep",("Waiting for slave thread to start"));
597
      const char* old_msg = thd->enter_cond(start_cond,cond_lock,
unknown's avatar
unknown committed
598
                                            "Waiting for slave thread to start");
599 600
      pthread_cond_wait(start_cond,cond_lock);
      thd->exit_cond(old_msg);
unknown's avatar
unknown committed
601
      pthread_mutex_lock(cond_lock); // re-acquire it as exit_cond() released
602
      if (thd->killed)
unknown's avatar
unknown committed
603
        DBUG_RETURN(thd->killed_errno());
604 605 606 607
    }
  }
  if (start_lock)
    pthread_mutex_unlock(start_lock);
unknown's avatar
unknown committed
608
  DBUG_RETURN(0);
609
}
unknown's avatar
unknown committed
610

611

unknown's avatar
unknown committed
612
/*
613
  start_slave_threads()
unknown's avatar
unknown committed
614

unknown's avatar
unknown committed
615 616 617 618
  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
619
*/
unknown's avatar
unknown committed
620

621
int start_slave_threads(bool need_slave_mutex, bool wait_for_start,
622
                        Master_info* mi, const char* master_info_fname,
unknown's avatar
unknown committed
623
                        const char* slave_info_fname, int thread_mask)
624 625 626 627
{
  pthread_mutex_t *lock_io=0,*lock_sql=0,*lock_cond_io=0,*lock_cond_sql=0;
  pthread_cond_t* cond_io=0,*cond_sql=0;
  int error=0;
628
  DBUG_ENTER("start_slave_threads");
unknown's avatar
unknown committed
629

630 631 632 633 634 635 636 637 638 639 640 641
  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;
  }
642 643 644

  if (thread_mask & SLAVE_IO)
    error=start_slave_thread(handle_slave_io,lock_io,lock_cond_io,
unknown's avatar
unknown committed
645 646 647
                             cond_io,
                             &mi->slave_running, &mi->slave_run_id,
                             mi, 1); //high priority, to read the most possible
648
  if (!error && (thread_mask & SLAVE_SQL))
649
  {
650
    error=start_slave_thread(handle_slave_sql,lock_sql,lock_cond_sql,
unknown's avatar
unknown committed
651 652 653
                             cond_sql,
                             &mi->rli.slave_running, &mi->rli.slave_run_id,
                             mi, 0);
654
    if (error)
655
      terminate_slave_threads(mi, thread_mask & SLAVE_IO, !need_slave_mutex);
656
  }
657
  DBUG_RETURN(error);
658
}
659

660

661
#ifdef NOT_USED_YET
662
static int end_slave_on_walk(Master_info* mi, uchar* /*unused*/)
663
{
664 665
  DBUG_ENTER("end_slave_on_walk");

666
  end_master_info(mi);
667
  DBUG_RETURN(0);
668
}
669
#endif
670

671

unknown's avatar
unknown committed
672
/*
673
  Release slave threads at time of executing shutdown.
unknown's avatar
unknown committed
674 675 676 677

  SYNOPSIS
    end_slave()
*/
678

679 680
void end_slave()
{
681 682
  DBUG_ENTER("end_slave");

683 684 685 686 687 688 689 690
  /*
    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.
  */
  pthread_mutex_lock(&LOCK_active_mi);
unknown's avatar
unknown committed
691 692 693 694 695 696 697 698
  if (active_mi)
  {
    /*
      TODO: replace the line below with
      list_walk(&master_list, (list_walk_action)end_slave_on_walk,0);
      once multi-master code is ready.
    */
    terminate_slave_threads(active_mi,SLAVE_FORCE_ALL);
699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717
  }
  pthread_mutex_unlock(&LOCK_active_mi);
  DBUG_VOID_RETURN;
}

/**
   Free all resources used by slave threads at time of executing shutdown.
   The routine must be called after all possible users of @c active_mi
   have left.

   SYNOPSIS
     close_active_mi()

*/
void close_active_mi()
{
  pthread_mutex_lock(&LOCK_active_mi);
  if (active_mi)
  {
unknown's avatar
unknown committed
718 719 720 721
    end_master_info(active_mi);
    delete active_mi;
    active_mi= 0;
  }
722
  pthread_mutex_unlock(&LOCK_active_mi);
723
}
unknown's avatar
unknown committed
724

725
static bool io_slave_killed(THD* thd, Master_info* mi)
unknown's avatar
unknown committed
726
{
727 728
  DBUG_ENTER("io_slave_killed");

729
  DBUG_ASSERT(mi->io_thd == thd);
730
  DBUG_ASSERT(mi->slave_running); // tracking buffer overrun
731
  DBUG_RETURN(mi->abort_slave || abort_loop || thd->killed);
unknown's avatar
unknown committed
732 733
}

734

735
static bool sql_slave_killed(THD* thd, Relay_log_info* rli)
736
{
737 738
  DBUG_ENTER("sql_slave_killed");

739 740
  DBUG_ASSERT(rli->sql_thd == thd);
  DBUG_ASSERT(rli->slave_running == 1);// tracking buffer overrun
741 742
  if (abort_loop || thd->killed || rli->abort_slave)
  {
743 744 745
    if (rli->abort_slave && rli->is_in_group() &&
        thd->transaction.all.modified_non_trans_table)
      DBUG_RETURN(0);
746 747 748 749 750 751 752 753 754
    /*
      If we are in an unsafe situation (stopping could corrupt replication),
      we give one minute to the slave SQL thread of grace before really
      terminating, in the hope that it will be able to read more events and
      the unsafe situation will soon be left. Note that this one minute starts
      from the last time anything happened in the slave SQL thread. So it's
      really one minute of idleness, we don't timeout if the slave SQL thread
      is actively working.
    */
755
    if (rli->last_event_start_time == 0)
756
      DBUG_RETURN(1);
757 758
    DBUG_PRINT("info", ("Slave SQL thread is in an unsafe situation, giving "
                        "it some grace period"));
759
    if (difftime(time(0), rli->last_event_start_time) > 60)
760
    {
761 762 763 764 765 766 767
      rli->report(ERROR_LEVEL, 0,
                  "SQL thread had to stop in an unsafe situation, in "
                  "the middle of applying updates to a "
                  "non-transactional table without any primary key. "
                  "There is a risk of duplicate updates when the slave "
                  "SQL thread is restarted. Please check your tables' "
                  "contents after restart.");
768
      DBUG_RETURN(1);
769 770
    }
  }
771
  DBUG_RETURN(0);
772 773
}

774

unknown's avatar
unknown committed
775
/*
776 777
  skip_load_data_infile()

unknown's avatar
unknown committed
778 779 780
  NOTES
    This is used to tell a 3.23 master to break send_file()
*/
781 782 783

void skip_load_data_infile(NET *net)
{
784 785
  DBUG_ENTER("skip_load_data_infile");

786
  (void)net_request_file(net, "/dev/null");
unknown's avatar
unknown committed
787
  (void)my_net_read(net);                               // discard response
788
  (void)net_write_command(net, 0, (uchar*) "", 0, (uchar*) "", 0); // ok
789
  DBUG_VOID_RETURN;
790 791
}

792

793
bool net_request_file(NET* net, const char* fname)
794
{
795
  DBUG_ENTER("net_request_file");
796 797
  DBUG_RETURN(net_write_command(net, 251, (uchar*) fname, strlen(fname),
                                (uchar*) "", 0));
798 799
}

800 801
/*
  From other comments and tests in code, it looks like
802
  sometimes Query_log_event and Load_log_event can have db == 0
803 804 805
  (see rewrite_db() above for example)
  (cases where this happens are unclear; it may be when the master is 3.23).
*/
806 807

const char *print_slave_db_safe(const char* db)
808
{
809 810 811
  DBUG_ENTER("*print_slave_db_safe");

  DBUG_RETURN((db ? db : ""));
812
}
813

814
int init_strvar_from_file(char *var, int max_size, IO_CACHE *f,
unknown's avatar
unknown committed
815
                                 const char *default_val)
unknown's avatar
unknown committed
816
{
unknown's avatar
unknown committed
817
  uint length;
818 819
  DBUG_ENTER("init_strvar_from_file");

unknown's avatar
unknown committed
820 821 822 823 824 825
  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
826
    {
unknown's avatar
unknown committed
827
      /*
unknown's avatar
unknown committed
828 829
        If we truncated a line or stopped on last char, remove all chars
        up to and including newline.
unknown's avatar
unknown committed
830
      */
unknown's avatar
unknown committed
831
      int c;
832
      while (((c=my_b_get(f)) != '\n' && c != my_b_EOF)) ;
unknown's avatar
unknown committed
833
    }
834
    DBUG_RETURN(0);
unknown's avatar
unknown committed
835 836 837
  }
  else if (default_val)
  {
unknown's avatar
unknown committed
838
    strmake(var,  default_val, max_size-1);
839
    DBUG_RETURN(0);
unknown's avatar
unknown committed
840
  }
841
  DBUG_RETURN(1);
unknown's avatar
unknown committed
842 843
}

844

845
int init_intvar_from_file(int* var, IO_CACHE* f, int default_val)
unknown's avatar
unknown committed
846 847
{
  char buf[32];
848 849
  DBUG_ENTER("init_intvar_from_file");

unknown's avatar
unknown committed
850 851

  if (my_b_gets(f, buf, sizeof(buf)))
unknown's avatar
unknown committed
852 853
  {
    *var = atoi(buf);
854
    DBUG_RETURN(0);
unknown's avatar
unknown committed
855
  }
unknown's avatar
unknown committed
856
  else if (default_val)
unknown's avatar
unknown committed
857 858
  {
    *var = default_val;
859
    DBUG_RETURN(0);
unknown's avatar
unknown committed
860
  }
861
  DBUG_RETURN(1);
unknown's avatar
unknown committed
862 863
}

864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886

/*
  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 ||
      errorno == ER_SERVER_SHUTDOWN)
    return TRUE;

  return FALSE;   
}


887 888 889 890 891 892 893 894
/*
  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
895

896 897 898
  RETURNS
  0       ok
  1       error
899
  2       transient network problem, the caller should try to reconnect
900
*/
901

902
static int get_master_version_and_clock(MYSQL* mysql, Master_info* mi)
903
{
904
  char err_buff[MAX_SLAVE_ERRMSG];
905
  const char* errmsg= 0;
906 907 908
  int err_code= 0;
  MYSQL_RES *master_res= 0;
  MYSQL_ROW master_row;
909
  DBUG_ENTER("get_master_version_and_clock");
910 911 912 913 914 915 916

  /*
    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
917

918
  if (!my_isdigit(&my_charset_bin,*mysql->server_version))
919
  {
unknown's avatar
unknown committed
920
    errmsg = "Master reported unrecognized MySQL version";
921 922 923
    err_code= ER_SLAVE_FATAL_ERROR;
    sprintf(err_buff, ER(err_code), errmsg);
  }
924 925 926 927 928
  else
  {
    /*
      Note the following switch will bug when we have MySQL branch 30 ;)
    */
unknown's avatar
unknown committed
929
    switch (*mysql->server_version)
930 931 932 933 934
    {
    case '0':
    case '1':
    case '2':
      errmsg = "Master reported unrecognized MySQL version";
935 936
      err_code= ER_SLAVE_FATAL_ERROR;
      sprintf(err_buff, ER(err_code), errmsg);
937 938 939
      break;
    case '3':
      mi->rli.relay_log.description_event_for_queue= new
unknown's avatar
unknown committed
940
        Format_description_log_event(1, mysql->server_version);
941 942 943
      break;
    case '4':
      mi->rli.relay_log.description_event_for_queue= new
unknown's avatar
unknown committed
944
        Format_description_log_event(3, mysql->server_version);
945
      break;
unknown's avatar
unknown committed
946
    default:
947 948 949 950 951 952 953 954 955
      /*
        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
956
        Format_description_log_event(4, mysql->server_version);
957 958
      break;
    }
959
  }
unknown's avatar
unknown committed
960 961

  /*
962 963 964 965 966
     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
967

968
  if (errmsg)
969
    goto err;
970 971 972 973

  /* as we are here, we tried to allocate the event */
  if (!mi->rli.relay_log.description_event_for_queue)
  {
974 975 976 977
    errmsg= "default Format_description_log_event";
    err_code= ER_SLAVE_CREATE_EVENT_FAILURE;
    sprintf(err_buff, ER(err_code), errmsg);
    goto err;
978 979
  }

980 981 982 983
  /*
    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
984

985 986 987 988 989 990 991 992 993 994
  DBUG_EXECUTE_IF("dbug.before_get_UNIX_TIMESTAMP",
                  {
                    const char act[]=
                      "now "
                      "wait_for signal.get_unix_timestamp";
                    DBUG_ASSERT(opt_debug_sync_timeout > 0);
                    DBUG_ASSERT(!debug_sync_set_action(current_thd,
                                                       STRING_WITH_LEN(act)));
                  };);

995
  master_res= NULL;
996
  if (!mysql_real_query(mysql, STRING_WITH_LEN("SELECT UNIX_TIMESTAMP()")) &&
997 998
      (master_res= mysql_store_result(mysql)) &&
      (master_row= mysql_fetch_row(master_res)))
unknown's avatar
unknown committed
999
  {
unknown's avatar
unknown committed
1000
    mi->clock_diff_with_master=
1001
      (long) (time((time_t*) 0) - strtoul(master_row[0], 0, 10));
unknown's avatar
unknown committed
1002
  }
1003 1004 1005 1006 1007 1008 1009
  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
1010
  {
1011
    mi->clock_diff_with_master= 0; /* The "most sensible" value */
1012 1013 1014 1015
    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));
1016 1017
  }
  if (master_res)
1018
  {
unknown's avatar
unknown committed
1019
    mysql_free_result(master_res);
1020 1021
    master_res= NULL;
  }
unknown's avatar
unknown committed
1022

1023 1024 1025 1026 1027 1028 1029 1030 1031 1032
  /*
    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.
  */
1033 1034 1035 1036 1037 1038 1039 1040 1041
  DBUG_EXECUTE_IF("dbug.before_get_SERVER_ID",
                  {
                    const char act[]=
                      "now "
                      "wait_for signal.get_server_id";
                    DBUG_ASSERT(opt_debug_sync_timeout > 0);
                    DBUG_ASSERT(!debug_sync_set_action(current_thd, 
                                                       STRING_WITH_LEN(act)));
                  };);
1042 1043
  master_res= NULL;
  master_row= NULL;
1044 1045
  if (!mysql_real_query(mysql,
                        STRING_WITH_LEN("SHOW VARIABLES LIKE 'SERVER_ID'")) &&
1046 1047
      (master_res= mysql_store_result(mysql)) &&
      (master_row= mysql_fetch_row(master_res)))
1048
  {
1049
    if ((::server_id == strtoul(master_row[1], 0, 10)) &&
1050
        !mi->rli.replicate_same_server_id)
1051
    {
1052 1053 1054 1055
      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).";
1056 1057
      err_code= ER_SLAVE_FATAL_ERROR;
      sprintf(err_buff, ER(err_code), errmsg);
1058
      goto err;
1059
    }
1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083
  }
  else if (mysql_errno(mysql))
  {
    if (is_network_error(mysql_errno(mysql)))
    {
      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)
  {
1084
    mysql_free_result(master_res);
1085
    master_res= NULL;
unknown's avatar
unknown committed
1086 1087
  }

1088 1089 1090
  /*
    Check that the master's global character_set_server and ours are the same.
    Not fatal if query fails (old master?).
1091 1092 1093 1094 1095 1096
    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.
1097 1098 1099 1100 1101 1102
    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.
1103
  */
1104 1105 1106

  /* redundant with rest of code but safer against later additions */
  if (*mysql->server_version == '3')
1107
    goto err;
1108

1109
  if (*mysql->server_version == '4')
unknown's avatar
unknown committed
1110
  {
1111 1112 1113 1114 1115
    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)))
1116
    {
1117 1118 1119
      if (strcmp(master_row[0], global_system_variables.collation_server->name))
      {
        errmsg= "The slave I/O thread stops because master and slave have \
1120
different values for the COLLATION_SERVER global variable. The values must \
1121 1122 1123 1124 1125
be equal for the Statement-format replication to work";
        err_code= ER_SLAVE_FATAL_ERROR;
        sprintf(err_buff, ER(err_code), errmsg);
        goto err;
      }
1126
    }
1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139
    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));
1140
      goto err;
1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152
    }
    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
1153
  }
1154

1155 1156 1157 1158 1159 1160 1161 1162
  /*
    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
1163 1164 1165
    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
1166
    'SYSTEM' time zone).
1167 1168
    This check is only necessary for 4.x masters (and < 5.0.4 masters but
    those were alpha).
1169
  */
1170
  if (*mysql->server_version == '4')
unknown's avatar
unknown committed
1171
  {
1172 1173 1174 1175
    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)))
1176
    {
1177 1178 1179 1180
      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 \
1181
different values for the TIME_ZONE global variable. The values must \
1182 1183 1184 1185 1186
be equal for the Statement-format replication to work";
        err_code= ER_SLAVE_FATAL_ERROR;
        sprintf(err_buff, ER(err_code), errmsg);
        goto err;
      }
1187
    }
1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200
    else if (is_network_error(mysql_errno(mysql)))
    {
      mi->report(WARNING_LEVEL, mysql_errno(mysql),
                 "Get master TIME_ZONE 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 try to get the value of TIME_ZONE global variable from master.";
      err_code= mysql_errno(mysql);
      sprintf(err_buff, "%s Error: %s", errmsg, mysql_error(mysql));
1201
      goto err;
1202 1203 1204 1205 1206 1207
    }
    if (master_res)
    {
      mysql_free_result(master_res);
      master_res= NULL;
    }
unknown's avatar
unknown committed
1208 1209
  }

1210
err:
1211 1212
  if (errmsg)
  {
1213 1214
    if (master_res)
      mysql_free_result(master_res);
1215
    DBUG_ASSERT(err_code != 0);
1216
    mi->report(ERROR_LEVEL, err_code, "%s", err_buff);
1217
    DBUG_RETURN(1);
1218
  }
1219

1220
  DBUG_RETURN(0);
1221 1222 1223 1224 1225

network_err:
  if (master_res)
    mysql_free_result(master_res);
  DBUG_RETURN(2);
1226 1227
}

1228 1229 1230 1231
/*
  Used by fetch_master_table (used by LOAD TABLE tblname FROM MASTER and LOAD
  DATA FROM MASTER). Drops the table (if 'overwrite' is true) and recreates it
  from the dump. Honours replication inclusion/exclusion rules.
1232
  db must be non-zero (guarded by assertion).
1233 1234 1235 1236 1237

  RETURN VALUES
    0           success
    1           error
*/
1238

1239
static int create_table_from_dump(THD* thd, MYSQL *mysql, const char* db,
unknown's avatar
unknown committed
1240
                                  const char* table_name, bool overwrite)
unknown's avatar
unknown committed
1241
{
1242
  ulong packet_len;
1243 1244
  char *query, *save_db;
  uint32 save_db_length;
1245 1246
  Vio* save_vio;
  HA_CHECK_OPT check_opt;
unknown's avatar
unknown committed
1247
  TABLE_LIST tables;
1248 1249
  int error= 1;
  handler *file;
1250
  ulonglong save_options;
1251
  NET *net= &mysql->net;
1252
  const char *found_semicolon= NULL;
1253
  DBUG_ENTER("create_table_from_dump");
unknown's avatar
unknown committed
1254

1255
  packet_len= my_net_read(net); // read create table statement
1256 1257
  if (packet_len == packet_error)
  {
unknown's avatar
unknown committed
1258
    my_message(ER_MASTER_NET_READ, ER(ER_MASTER_NET_READ), MYF(0));
unknown's avatar
unknown committed
1259
    DBUG_RETURN(1);
1260 1261 1262
  }
  if (net->read_pos[0] == 255) // error from master
  {
unknown's avatar
unknown committed
1263
    char *err_msg;
1264
    err_msg= (char*) net->read_pos + ((mysql->server_capabilities &
unknown's avatar
unknown committed
1265 1266
                                       CLIENT_PROTOCOL_41) ?
                                      3+SQLSTATE_LENGTH+1 : 3);
1267
    my_error(ER_MASTER, MYF(0), err_msg);
unknown's avatar
unknown committed
1268
    DBUG_RETURN(1);
1269
  }
unknown's avatar
unknown committed
1270
  thd->command = COM_TABLE_DUMP;
1271
  if (!(query = thd->strmake((char*) net->read_pos, packet_len)))
1272 1273
  {
    sql_print_error("create_table_from_dump: out of memory");
unknown's avatar
unknown committed
1274
    my_message(ER_GET_ERRNO, "Out of memory", MYF(0));
unknown's avatar
unknown committed
1275
    DBUG_RETURN(1);
1276
  }
1277
  thd->set_query(query, packet_len);
1278
  thd->is_slave_error = 0;
1279

unknown's avatar
unknown committed
1280 1281
  bzero((char*) &tables,sizeof(tables));
  tables.db = (char*)db;
1282
  tables.alias= tables.table_name= (char*)table_name;
unknown's avatar
unknown committed
1283

unknown's avatar
unknown committed
1284
  /* Drop the table if 'overwrite' is true */
1285
  if (overwrite)
unknown's avatar
unknown committed
1286
  {
1287 1288 1289 1290 1291 1292 1293 1294 1295 1296
    if (mysql_rm_table(thd,&tables,1,0)) /* drop if exists */
    {
      sql_print_error("create_table_from_dump: failed to drop the table");
      goto err;
    }
    else
    {
      /* Clear the OK result of mysql_rm_table(). */
      thd->main_da.reset_diagnostics_area();
    }
unknown's avatar
unknown committed
1297 1298
  }

1299
  /* Create the table. We do not want to log the "create table" statement */
1300
  save_options = thd->options;
1301 1302
  thd->options &= ~ (OPTION_BIN_LOG);
  thd_proc_info(thd, "Creating table from master dump");
unknown's avatar
unknown committed
1303
  // save old db in case we are creating in a different database
1304
  save_db = thd->db;
1305
  save_db_length= thd->db_length;
1306
  thd->db = (char*)db;
1307
  DBUG_ASSERT(thd->db != 0);
1308
  thd->db_length= strlen(thd->db);
1309 1310
  /* run create table */
  mysql_parse(thd, thd->query(), packet_len, &found_semicolon);
unknown's avatar
unknown committed
1311
  thd->db = save_db;            // leave things the way the were before
1312
  thd->db_length= save_db_length;
1313
  thd->options = save_options;
unknown's avatar
unknown committed
1314

1315
  if (thd->is_slave_error)
unknown's avatar
unknown committed
1316
    goto err;                   // mysql_parse took care of the error send
unknown's avatar
unknown committed
1317

1318
  thd_proc_info(thd, "Opening master dump table");
1319
  thd->main_da.reset_diagnostics_area(); /* cleanup from CREATE_TABLE */
1320 1321 1322 1323 1324 1325
  /*
    Note: If this function starts to fail for MERGE tables,
    change the next two lines to these:
    tables.table= NULL; // was set by mysql_rm_table()
    if (!open_n_lock_single_table(thd, &tables, TL_WRITE))
  */
1326
  tables.lock_type = TL_WRITE;
1327
  if (!open_ltable(thd, &tables, TL_WRITE, 0))
unknown's avatar
unknown committed
1328 1329
  {
    sql_print_error("create_table_from_dump: could not open created table");
1330
    goto err;
unknown's avatar
unknown committed
1331
  }
unknown's avatar
unknown committed
1332

1333
  file = tables.table->file;
1334
  thd_proc_info(thd, "Reading master dump table data");
1335
  /* Copy the data file */
unknown's avatar
unknown committed
1336 1337
  if (file->net_read_dump(net))
  {
unknown's avatar
unknown committed
1338
    my_message(ER_MASTER_NET_READ, ER(ER_MASTER_NET_READ), MYF(0));
1339
    sql_print_error("create_table_from_dump: failed in\
unknown's avatar
unknown committed
1340
 handler::net_read_dump()");
1341
    goto err;
unknown's avatar
unknown committed
1342
  }
unknown's avatar
unknown committed
1343 1344

  check_opt.init();
unknown's avatar
unknown committed
1345
  check_opt.flags|= T_VERY_SILENT | T_CALC_CHECKSUM | T_QUICK;
1346
  thd_proc_info(thd, "Rebuilding the index on master dump table");
unknown's avatar
unknown committed
1347 1348 1349 1350 1351
  /*
    We do not want repair() to spam us with messages
    just send them to the error log, and report the failure in case of
    problems.
  */
1352
  save_vio = thd->net.vio;
unknown's avatar
unknown committed
1353
  thd->net.vio = 0;
1354
  /* Rebuild the index file from the copied data file (with REPAIR) */
unknown's avatar
unknown committed
1355
  error=file->ha_repair(thd,&check_opt) != 0;
unknown's avatar
unknown committed
1356
  thd->net.vio = save_vio;
1357
  if (error)
unknown's avatar
unknown committed
1358
    my_error(ER_INDEX_REBUILD, MYF(0), tables.table->s->table_name.str);
1359 1360

err:
unknown's avatar
unknown committed
1361
  close_thread_tables(thd);
unknown's avatar
unknown committed
1362
  DBUG_RETURN(error);
unknown's avatar
unknown committed
1363 1364
}

1365

1366
int fetch_master_table(THD *thd, const char *db_name, const char *table_name,
1367
                       Master_info *mi, MYSQL *mysql, bool overwrite)
unknown's avatar
unknown committed
1368
{
1369 1370 1371 1372 1373
  int error= 1;
  const char *errmsg=0;
  bool called_connected= (mysql != NULL);
  DBUG_ENTER("fetch_master_table");
  DBUG_PRINT("enter", ("db_name: '%s'  table_name: '%s'",
unknown's avatar
unknown committed
1374
                       db_name,table_name));
unknown's avatar
unknown committed
1375

unknown's avatar
merge  
unknown committed
1376
  if (!called_connected)
unknown's avatar
unknown committed
1377
  {
unknown's avatar
SCRUM  
unknown committed
1378
    if (!(mysql = mysql_init(NULL)))
1379 1380 1381
    {
      DBUG_RETURN(1);
    }
unknown's avatar
merge  
unknown committed
1382
    if (connect_to_master(thd, mysql, mi))
1383
    {
1384
      my_error(ER_CONNECT_TO_MASTER, MYF(0), mysql_error(mysql));
1385 1386 1387 1388 1389 1390 1391 1392 1393
      /*
        We need to clear the active VIO since, theoretically, somebody
        might issue an awake() on this thread.  If we are then in the
        middle of closing and destroying the VIO inside the
        mysql_close(), we will have a problem.
       */
#ifdef SIGNAL_WITH_VIO_CLOSE
      thd->clear_active_vio();
#endif
unknown's avatar
SCRUM  
unknown committed
1394
      mysql_close(mysql);
1395
      DBUG_RETURN(1);
1396
    }
1397 1398
    if (thd->killed)
      goto err;
1399
  }
unknown's avatar
unknown committed
1400

unknown's avatar
unknown committed
1401
  if (request_table_dump(mysql, db_name, table_name))
1402
  {
1403 1404
    error= ER_UNKNOWN_ERROR;
    errmsg= "Failed on table dump request";
1405 1406
    goto err;
  }
unknown's avatar
unknown committed
1407
  if (create_table_from_dump(thd, mysql, db_name,
unknown's avatar
unknown committed
1408
                             table_name, overwrite))
unknown's avatar
unknown committed
1409
    goto err;    // create_table_from_dump have sent the error already
unknown's avatar
unknown committed
1410
  error = 0;
1411

unknown's avatar
unknown committed
1412
 err:
1413
  if (!called_connected)
unknown's avatar
SCRUM  
unknown committed
1414
    mysql_close(mysql);
1415
  if (errmsg && thd->vio_ok())
unknown's avatar
unknown committed
1416
    my_message(error, errmsg, MYF(0));
unknown's avatar
unknown committed
1417
  DBUG_RETURN(test(error));                     // Return 1 on error
unknown's avatar
unknown committed
1418
}
unknown's avatar
unknown committed
1419

1420

1421
static bool wait_for_relay_log_space(Relay_log_info* rli)
unknown's avatar
unknown committed
1422
{
1423
  bool slave_killed=0;
1424
  Master_info* mi = rli->mi;
unknown's avatar
unknown committed
1425
  const char *save_proc_info;
unknown's avatar
unknown committed
1426 1427
  THD* thd = mi->io_thd;
  DBUG_ENTER("wait_for_relay_log_space");
1428

unknown's avatar
unknown committed
1429
  pthread_mutex_lock(&rli->log_space_lock);
unknown's avatar
unknown committed
1430
  save_proc_info= thd->enter_cond(&rli->log_space_cond,
unknown's avatar
unknown committed
1431 1432
                                  &rli->log_space_lock,
                                  "\
1433
Waiting for the slave SQL thread to free enough relay log space");
unknown's avatar
unknown committed
1434
  while (rli->log_space_limit < rli->log_space_total &&
unknown's avatar
unknown committed
1435
         !(slave_killed=io_slave_killed(thd,mi)) &&
1436
         !rli->ignore_log_space_limit)
unknown's avatar
unknown committed
1437
    pthread_cond_wait(&rli->log_space_cond, &rli->log_space_lock);
unknown's avatar
unknown committed
1438
  thd->exit_cond(save_proc_info);
unknown's avatar
unknown committed
1439 1440
  DBUG_RETURN(slave_killed);
}
1441

1442

1443 1444
/*
  Builds a Rotate from the ignored events' info and writes it to relay log.
1445

1446 1447 1448 1449 1450 1451 1452 1453 1454 1455
  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).
 */
1456
static void write_ignored_events_info_to_relay_log(THD *thd, Master_info *mi)
1457
{
1458
  Relay_log_info *rli= &mi->rli;
1459
  pthread_mutex_t *log_lock= rli->relay_log.get_log_lock();
1460 1461
  DBUG_ENTER("write_ignored_events_info_to_relay_log");

1462 1463 1464
  DBUG_ASSERT(thd == mi->io_thd);
  pthread_mutex_lock(log_lock);
  if (rli->ign_master_log_name_end[0])
unknown's avatar
unknown committed
1465
  {
1466
    DBUG_PRINT("info",("writing a Rotate event to track down ignored events"));
1467
    Rotate_log_event *ev= new Rotate_log_event(rli->ign_master_log_name_end,
1468 1469 1470 1471 1472 1473 1474 1475 1476
                                               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 */
    pthread_mutex_unlock(log_lock);
    if (likely((bool)ev))
    {
      ev->server_id= 0; // don't be ignored by slave SQL thread
      if (unlikely(rli->relay_log.append(ev)))
1477 1478 1479 1480 1481
        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");
1482
      rli->relay_log.harvest_bytes_written(&rli->log_space_total);
1483
      if (flush_master_info(mi, TRUE, TRUE))
1484
        sql_print_error("Failed to flush master info file");
1485 1486 1487
      delete ev;
    }
    else
1488 1489 1490 1491
      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
1492
  }
1493 1494
  else
    pthread_mutex_unlock(log_lock);
1495
  DBUG_VOID_RETURN;
unknown's avatar
unknown committed
1496 1497
}

1498

1499
int register_slave_on_master(MYSQL* mysql, Master_info *mi,
1500
                             bool *suppress_warnings)
1501
{
1502
  uchar buf[1024], *pos= buf;
1503
  uint report_host_len, report_user_len=0, report_password_len=0;
1504
  DBUG_ENTER("register_slave_on_master");
1505

1506
  *suppress_warnings= FALSE;
1507
  if (!report_host)
1508
    DBUG_RETURN(0);
1509
  report_host_len= strlen(report_host);
1510
  if (report_user)
1511
    report_user_len= strlen(report_user);
unknown's avatar
unknown committed
1512
  if (report_password)
1513 1514 1515 1516
    report_password_len= strlen(report_password);
  /* 30 is a good safety margin */
  if (report_host_len + report_user_len + report_password_len + 30 >
      sizeof(buf))
unknown's avatar
unknown committed
1517
    DBUG_RETURN(0);                                     // safety
1518 1519

  int4store(pos, server_id); pos+= 4;
1520 1521 1522
  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);
1523
  int2store(pos, (uint16) report_port); pos+= 2;
unknown's avatar
unknown committed
1524
  int4store(pos, rpl_recovery_rank);    pos+= 4;
1525
  /* The master will fill in master_id */
unknown's avatar
unknown committed
1526
  int4store(pos, 0);                    pos+= 4;
1527

1528
  if (simple_command(mysql, COM_REGISTER_SLAVE, buf, (size_t) (pos- buf), 0))
1529
  {
1530 1531 1532 1533
    if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
    {
      *suppress_warnings= TRUE;                 // Suppress reconnect warning
    }
1534
    else if (!check_io_slave_killed(mi->io_thd, mi, NULL))
1535 1536 1537 1538 1539 1540 1541
    {
      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);
    }
1542
    DBUG_RETURN(1);
1543
  }
1544
  DBUG_RETURN(0);
1545 1546
}

1547

1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558
/**
  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
*/
1559
bool show_master_info(THD* thd, Master_info* mi)
unknown's avatar
unknown committed
1560
{
1561
  // TODO: fix this for multi-master
unknown's avatar
unknown committed
1562
  List<Item> field_list;
1563 1564 1565
  Protocol *protocol= thd->protocol;
  DBUG_ENTER("show_master_info");

unknown's avatar
unknown committed
1566
  field_list.push_back(new Item_empty_string("Slave_IO_State",
unknown's avatar
unknown committed
1567
                                                     14));
unknown's avatar
unknown committed
1568
  field_list.push_back(new Item_empty_string("Master_Host",
unknown's avatar
unknown committed
1569
                                                     sizeof(mi->host)));
unknown's avatar
unknown committed
1570
  field_list.push_back(new Item_empty_string("Master_User",
unknown's avatar
unknown committed
1571
                                                     sizeof(mi->user)));
1572
  field_list.push_back(new Item_return_int("Master_Port", 7,
unknown's avatar
unknown committed
1573
                                           MYSQL_TYPE_LONG));
1574
  field_list.push_back(new Item_return_int("Connect_Retry", 10,
unknown's avatar
unknown committed
1575
                                           MYSQL_TYPE_LONG));
1576
  field_list.push_back(new Item_empty_string("Master_Log_File",
unknown's avatar
unknown committed
1577
                                             FN_REFLEN));
1578
  field_list.push_back(new Item_return_int("Read_Master_Log_Pos", 10,
unknown's avatar
unknown committed
1579
                                           MYSQL_TYPE_LONGLONG));
1580
  field_list.push_back(new Item_empty_string("Relay_Log_File",
unknown's avatar
unknown committed
1581
                                             FN_REFLEN));
1582
  field_list.push_back(new Item_return_int("Relay_Log_Pos", 10,
unknown's avatar
unknown committed
1583
                                           MYSQL_TYPE_LONGLONG));
1584
  field_list.push_back(new Item_empty_string("Relay_Master_Log_File",
unknown's avatar
unknown committed
1585
                                             FN_REFLEN));
1586 1587
  field_list.push_back(new Item_empty_string("Slave_IO_Running", 3));
  field_list.push_back(new Item_empty_string("Slave_SQL_Running", 3));
1588 1589 1590 1591 1592 1593
  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
1594
                                             28));
1595 1596 1597
  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
1598
                                           MYSQL_TYPE_LONG));
1599
  field_list.push_back(new Item_return_int("Exec_Master_Log_Pos", 10,
unknown's avatar
unknown committed
1600
                                           MYSQL_TYPE_LONGLONG));
1601
  field_list.push_back(new Item_return_int("Relay_Log_Space", 10,
unknown's avatar
unknown committed
1602
                                           MYSQL_TYPE_LONGLONG));
1603
  field_list.push_back(new Item_empty_string("Until_Condition", 6));
1604
  field_list.push_back(new Item_empty_string("Until_Log_File", FN_REFLEN));
unknown's avatar
unknown committed
1605
  field_list.push_back(new Item_return_int("Until_Log_Pos", 10,
1606
                                           MYSQL_TYPE_LONGLONG));
unknown's avatar
unknown committed
1607 1608 1609
  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
1610
  field_list.push_back(new Item_empty_string("Master_SSL_CA_Path",
unknown's avatar
unknown committed
1611
                                             sizeof(mi->ssl_capath)));
unknown's avatar
unknown committed
1612
  field_list.push_back(new Item_empty_string("Master_SSL_Cert",
unknown's avatar
unknown committed
1613
                                             sizeof(mi->ssl_cert)));
unknown's avatar
unknown committed
1614
  field_list.push_back(new Item_empty_string("Master_SSL_Cipher",
unknown's avatar
unknown committed
1615
                                             sizeof(mi->ssl_cipher)));
unknown's avatar
unknown committed
1616
  field_list.push_back(new Item_empty_string("Master_SSL_Key",
unknown's avatar
unknown committed
1617
                                             sizeof(mi->ssl_key)));
1618
  field_list.push_back(new Item_return_int("Seconds_Behind_Master", 10,
unknown's avatar
unknown committed
1619
                                           MYSQL_TYPE_LONGLONG));
1620 1621
  field_list.push_back(new Item_empty_string("Master_SSL_Verify_Server_Cert",
                                             3));
1622 1623 1624 1625
  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));
unknown's avatar
unknown committed
1626

1627 1628
  if (protocol->send_fields(&field_list,
                            Protocol::SEND_NUM_ROWS | Protocol::SEND_EOF))
unknown's avatar
unknown committed
1629
    DBUG_RETURN(TRUE);
unknown's avatar
unknown committed
1630

1631 1632
  if (mi->host[0])
  {
1633
    DBUG_PRINT("info",("host is set: '%s'", mi->host));
1634
    String *packet= &thd->packet;
1635
    protocol->prepare_for_resend();
unknown's avatar
unknown committed
1636

1637
    /*
1638 1639
      slave_running can be accessed without run_lock but not other
      non-volotile members like mi->io_thd, which is guarded by the mutex.
1640
    */
1641 1642 1643 1644
    pthread_mutex_lock(&mi->run_lock);
    protocol->store(mi->io_thd ? mi->io_thd->proc_info : "", &my_charset_bin);
    pthread_mutex_unlock(&mi->run_lock);

1645 1646
    pthread_mutex_lock(&mi->data_lock);
    pthread_mutex_lock(&mi->rli.data_lock);
1647 1648
    pthread_mutex_lock(&mi->err_lock);
    pthread_mutex_lock(&mi->rli.err_lock);
1649 1650
    protocol->store(mi->host, &my_charset_bin);
    protocol->store(mi->user, &my_charset_bin);
1651 1652
    protocol->store((uint32) mi->port);
    protocol->store((uint32) mi->connect_retry);
1653
    protocol->store(mi->master_log_name, &my_charset_bin);
1654
    protocol->store((ulonglong) mi->master_log_pos);
1655
    protocol->store(mi->rli.group_relay_log_name +
unknown's avatar
unknown committed
1656 1657
                    dirname_length(mi->rli.group_relay_log_name),
                    &my_charset_bin);
1658 1659
    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
1660 1661
    protocol->store(mi->slave_running == MYSQL_SLAVE_RUN_CONNECT ?
                    "Yes" : "No", &my_charset_bin);
1662
    protocol->store(mi->rli.slave_running ? "Yes":"No", &my_charset_bin);
1663 1664 1665
    protocol->store(rpl_filter->get_do_db());
    protocol->store(rpl_filter->get_ignore_db());

1666 1667
    char buf[256];
    String tmp(buf, sizeof(buf), &my_charset_bin);
1668
    rpl_filter->get_do_table(&tmp);
1669
    protocol->store(&tmp);
1670
    rpl_filter->get_ignore_table(&tmp);
1671
    protocol->store(&tmp);
1672
    rpl_filter->get_wild_do_table(&tmp);
1673
    protocol->store(&tmp);
1674
    rpl_filter->get_wild_ignore_table(&tmp);
1675 1676
    protocol->store(&tmp);

1677 1678
    protocol->store(mi->rli.last_error().number);
    protocol->store(mi->rli.last_error().message, &my_charset_bin);
1679
    protocol->store((uint32) mi->rli.slave_skip_counter);
1680
    protocol->store((ulonglong) mi->rli.group_master_log_pos);
1681
    protocol->store((ulonglong) mi->rli.log_space_total);
1682 1683

    protocol->store(
1684 1685
      mi->rli.until_condition==Relay_log_info::UNTIL_NONE ? "None":
        ( mi->rli.until_condition==Relay_log_info::UNTIL_MASTER_POS? "Master":
1686 1687 1688
          "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
1689 1690

#ifdef HAVE_OPENSSL
unknown's avatar
unknown committed
1691 1692 1693 1694 1695 1696 1697 1698 1699
    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
1700

1701 1702 1703 1704 1705 1706
    /*
      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)
1707
    {
1708
      long time_diff= ((long)(time(0) - mi->rli.last_master_timestamp)
unknown's avatar
unknown committed
1709
                       - mi->clock_diff_with_master);
1710
      /*
unknown's avatar
unknown committed
1711 1712
        Apparently on some systems time_diff can be <0. Here are possible
        reasons related to MySQL:
1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724
        - 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.
1725 1726 1727 1728
        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".
1729
      */
unknown's avatar
unknown committed
1730 1731
      protocol->store((longlong)(mi->rli.last_master_timestamp ?
                                 max(0, time_diff) : 0));
1732
    }
unknown's avatar
unknown committed
1733
    else
1734
    {
unknown's avatar
unknown committed
1735
      protocol->store_null();
1736 1737
    }
    protocol->store(mi->ssl_verify_server_cert? "Yes":"No", &my_charset_bin);
unknown's avatar
unknown committed
1738

1739
    // Last_IO_Errno
1740
    protocol->store(mi->last_error().number);
1741
    // Last_IO_Error
1742
    protocol->store(mi->last_error().message, &my_charset_bin);
1743
    // Last_SQL_Errno
1744
    protocol->store(mi->rli.last_error().number);
1745
    // Last_SQL_Error
1746
    protocol->store(mi->rli.last_error().message, &my_charset_bin);
1747

1748 1749
    pthread_mutex_unlock(&mi->rli.err_lock);
    pthread_mutex_unlock(&mi->err_lock);
1750 1751
    pthread_mutex_unlock(&mi->rli.data_lock);
    pthread_mutex_unlock(&mi->data_lock);
1752

1753
    if (my_net_write(&thd->net, (uchar*) thd->packet.ptr(), packet->length()))
unknown's avatar
unknown committed
1754
      DBUG_RETURN(TRUE);
1755
  }
1756
  my_eof(thd);
unknown's avatar
unknown committed
1757
  DBUG_RETURN(FALSE);
unknown's avatar
unknown committed
1758 1759
}

unknown's avatar
unknown committed
1760

1761 1762
void set_slave_thread_options(THD* thd)
{
1763
  DBUG_ENTER("set_slave_thread_options");
1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778
  /*
     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.
  */
  ulonglong options= thd->options | OPTION_BIG_SELECTS;
  if (opt_log_slave_updates)
    options|= OPTION_BIN_LOG;
  else
    options&= ~OPTION_BIN_LOG;
  thd->options= options;
1779
  thd->variables.completion_type= 0;
1780
  DBUG_VOID_RETURN;
1781
}
1782

1783
void set_slave_thread_default_charset(THD* thd, Relay_log_info const *rli)
1784
{
1785 1786
  DBUG_ENTER("set_slave_thread_default_charset");

1787 1788 1789 1790 1791 1792 1793
  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();
1794 1795 1796 1797 1798 1799 1800

  /*
    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.
   */
1801
  const_cast<Relay_log_info*>(rli)->cached_charset_invalidate();
1802
  DBUG_VOID_RETURN;
1803 1804
}

unknown's avatar
unknown committed
1805
/*
1806
  init_slave_thread()
unknown's avatar
unknown committed
1807
*/
1808

1809
static int init_slave_thread(THD* thd, SLAVE_THD_TYPE thd_type)
unknown's avatar
unknown committed
1810 1811
{
  DBUG_ENTER("init_slave_thread");
1812 1813 1814
#if !defined(DBUG_OFF)
  int simulate_error= 0;
#endif
1815
  thd->system_thread = (thd_type == SLAVE_THD_SQL) ?
unknown's avatar
unknown committed
1816
    SYSTEM_THREAD_SLAVE_SQL : SYSTEM_THREAD_SLAVE_IO;
1817
  thd->security_ctx->skip_grants();
unknown's avatar
unknown committed
1818
  my_net_init(&thd->net, 0);
1819 1820 1821 1822 1823
/*
  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.
*/
1824
  thd->variables.max_allowed_packet= global_system_variables.max_allowed_packet
1825
    + MAX_LOG_EVENT_HEADER;  /* note, incr over the global not session var */
1826
  thd->slave_thread = 1;
1827
  thd->enable_slow_log= opt_log_slow_slave_statements;
1828
  set_slave_thread_options(thd);
unknown's avatar
unknown committed
1829 1830
  thd->client_capabilities = CLIENT_LOCAL_FILES;
  pthread_mutex_lock(&LOCK_thread_count);
unknown's avatar
unknown committed
1831
  thd->thread_id= thd->variables.pseudo_thread_id= thread_id++;
unknown's avatar
unknown committed
1832 1833
  pthread_mutex_unlock(&LOCK_thread_count);

1834 1835 1836 1837 1838 1839 1840
  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););
#if !defined(DBUG_OFF)
  if (init_thr_lock() || thd->store_globals() || simulate_error & (1<< thd_type))
#else
1841
  if (init_thr_lock() || thd->store_globals())
1842
#endif
unknown's avatar
unknown committed
1843
  {
1844
    thd->cleanup();
unknown's avatar
unknown committed
1845 1846
    DBUG_RETURN(-1);
  }
1847
  lex_start(thd);
unknown's avatar
unknown committed
1848

1849
  if (thd_type == SLAVE_THD_SQL)
1850
    thd_proc_info(thd, "Waiting for the next event in relay log");
1851
  else
1852
    thd_proc_info(thd, "Waiting for master update");
unknown's avatar
unknown committed
1853 1854 1855 1856 1857
  thd->version=refresh_version;
  thd->set_time();
  DBUG_RETURN(0);
}

1858

1859
static int safe_sleep(THD* thd, int sec, CHECK_KILLED_FUNC thread_killed,
unknown's avatar
unknown committed
1860
                      void* thread_killed_arg)
unknown's avatar
unknown committed
1861
{
1862
  int nap_time;
unknown's avatar
unknown committed
1863
  thr_alarm_t alarmed;
1864 1865
  DBUG_ENTER("safe_sleep");

unknown's avatar
unknown committed
1866
  thr_alarm_init(&alarmed);
1867
  time_t start_time= my_time(0);
unknown's avatar
unknown committed
1868 1869
  time_t end_time= start_time+sec;

1870
  while ((nap_time= (int) (end_time - start_time)) > 0)
unknown's avatar
unknown committed
1871
  {
1872
    ALARM alarm_buff;
unknown's avatar
unknown committed
1873
    /*
1874
      The only reason we are asking for alarm is so that
unknown's avatar
unknown committed
1875 1876 1877
      we will be woken up in case of murder, so if we do not get killed,
      set the alarm so it goes off after we wake up naturally
    */
1878
    thr_alarm(&alarmed, 2 * nap_time, &alarm_buff);
unknown's avatar
unknown committed
1879
    sleep(nap_time);
1880
    thr_end_alarm(&alarmed);
unknown's avatar
unknown committed
1881

1882
    if ((*thread_killed)(thd,thread_killed_arg))
1883
      DBUG_RETURN(1);
1884
    start_time= my_time(0);
unknown's avatar
unknown committed
1885
  }
1886
  DBUG_RETURN(0);
unknown's avatar
unknown committed
1887 1888
}

1889

1890
static int request_dump(MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
1891
                        bool *suppress_warnings)
unknown's avatar
unknown committed
1892
{
1893
  uchar buf[FN_REFLEN + 10];
unknown's avatar
unknown committed
1894 1895
  int len;
  int binlog_flags = 0; // for now
1896
  char* logname = mi->master_log_name;
1897
  DBUG_ENTER("request_dump");
1898 1899
  
  *suppress_warnings= FALSE;
1900

unknown's avatar
unknown committed
1901
  // TODO if big log files: Change next to int8store()
unknown's avatar
unknown committed
1902
  int4store(buf, (ulong) mi->master_log_pos);
unknown's avatar
unknown committed
1903
  int2store(buf + 4, binlog_flags);
1904
  int4store(buf + 6, server_id);
unknown's avatar
unknown committed
1905
  len = (uint) strlen(logname);
1906
  memcpy(buf + 10, logname,len);
unknown's avatar
SCRUM  
unknown committed
1907
  if (simple_command(mysql, COM_BINLOG_DUMP, buf, len + 10, 1))
unknown's avatar
unknown committed
1908
  {
unknown's avatar
unknown committed
1909 1910 1911 1912 1913
    /*
      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
1914
    if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
1915
      *suppress_warnings= TRUE;                 // Suppress reconnect warning
unknown's avatar
unknown committed
1916
    else
1917
      sql_print_error("Error on COM_BINLOG_DUMP: %d  %s, will retry in %d secs",
unknown's avatar
unknown committed
1918 1919
                      mysql_errno(mysql), mysql_error(mysql),
                      master_connect_retry);
1920
    DBUG_RETURN(1);
unknown's avatar
unknown committed
1921
  }
unknown's avatar
unknown committed
1922

1923
  DBUG_RETURN(0);
unknown's avatar
unknown committed
1924 1925
}

1926

1927
static int request_table_dump(MYSQL* mysql, const char* db, const char* table)
unknown's avatar
unknown committed
1928
{
1929
  uchar buf[1024], *p = buf;
1930 1931
  DBUG_ENTER("request_table_dump");

unknown's avatar
unknown committed
1932 1933
  uint table_len = (uint) strlen(table);
  uint db_len = (uint) strlen(db);
unknown's avatar
unknown committed
1934
  if (table_len + db_len > sizeof(buf) - 2)
unknown's avatar
unknown committed
1935 1936
  {
    sql_print_error("request_table_dump: Buffer overrun");
1937
    DBUG_RETURN(1);
unknown's avatar
unknown committed
1938 1939
  }

unknown's avatar
unknown committed
1940 1941 1942 1943 1944
  *p++ = db_len;
  memcpy(p, db, db_len);
  p += db_len;
  *p++ = table_len;
  memcpy(p, table, table_len);
unknown's avatar
unknown committed
1945

unknown's avatar
SCRUM  
unknown committed
1946
  if (simple_command(mysql, COM_TABLE_DUMP, buf, p - buf + table_len, 1))
unknown's avatar
unknown committed
1947 1948
  {
    sql_print_error("request_table_dump: Error sending the table dump \
unknown's avatar
unknown committed
1949
command");
1950
    DBUG_RETURN(1);
unknown's avatar
unknown committed
1951
  }
unknown's avatar
unknown committed
1952

1953
  DBUG_RETURN(0);
unknown's avatar
unknown committed
1954 1955
}

1956

unknown's avatar
unknown committed
1957
/*
1958
  Read one event from the master
unknown's avatar
unknown committed
1959

unknown's avatar
unknown committed
1960 1961
  SYNOPSIS
    read_event()
unknown's avatar
unknown committed
1962 1963 1964 1965 1966 1967
    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.
1968

unknown's avatar
unknown committed
1969
    RETURN VALUES
unknown's avatar
unknown committed
1970 1971
    'packet_error'      Error
    number              Length of packet
unknown's avatar
unknown committed
1972 1973
*/

1974
static ulong read_event(MYSQL* mysql, Master_info *mi, bool* suppress_warnings)
unknown's avatar
unknown committed
1975
{
1976
  ulong len;
1977
  DBUG_ENTER("read_event");
unknown's avatar
unknown committed
1978

1979
  *suppress_warnings= FALSE;
unknown's avatar
unknown committed
1980 1981 1982 1983
  /*
    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
1984
#ifndef DBUG_OFF
1985
  if (disconnect_slave_event_count && !(mi->events_till_disconnect--))
unknown's avatar
unknown committed
1986
    DBUG_RETURN(packet_error);
unknown's avatar
unknown committed
1987
#endif
unknown's avatar
unknown committed
1988

1989
  len = cli_safe_read(mysql);
1990
  if (len == packet_error || (long) len < 1)
unknown's avatar
unknown committed
1991
  {
unknown's avatar
SCRUM  
unknown committed
1992
    if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
unknown's avatar
unknown committed
1993 1994
    {
      /*
unknown's avatar
unknown committed
1995 1996 1997
        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
1998 1999 2000 2001
      */
      *suppress_warnings= TRUE;
    }
    else
2002
      sql_print_error("Error reading packet from server: %s ( server_errno=%d)",
unknown's avatar
unknown committed
2003
                      mysql_error(mysql), mysql_errno(mysql));
2004
    DBUG_RETURN(packet_error);
unknown's avatar
unknown committed
2005 2006
  }

2007 2008
  /* Check if eof packet */
  if (len < 8 && mysql->net.read_pos[0] == 254)
unknown's avatar
unknown committed
2009
  {
2010 2011
    sql_print_information("Slave: received end packet from server, apparent "
                          "master shutdown: %s",
unknown's avatar
unknown committed
2012
                     mysql_error(mysql));
2013
     DBUG_RETURN(packet_error);
unknown's avatar
unknown committed
2014
  }
unknown's avatar
unknown committed
2015

unknown's avatar
unknown committed
2016
  DBUG_PRINT("exit", ("len: %lu  net->read_pos[4]: %d",
unknown's avatar
unknown committed
2017 2018
                      len, mysql->net.read_pos[4]));
  DBUG_RETURN(len - 1);
unknown's avatar
unknown committed
2019 2020
}

unknown's avatar
unknown committed
2021 2022 2023 2024 2025 2026 2027 2028 2029
/*
  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)
{
2030 2031
  DBUG_ENTER("has_temporary_error");

2032
  DBUG_EXECUTE_IF("all_errors_are_temporary_errors",
2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045
                  if (thd->main_da.is_error())
                  {
                    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
2046 2047 2048 2049 2050 2051

  /*
    Temporary error codes:
    currently, InnoDB deadlock detected by InnoDB or lock
    wait timeout (innodb_lock_wait_timeout exceeded
  */
2052 2053
  if (thd->main_da.sql_errno() == ER_LOCK_DEADLOCK ||
      thd->main_da.sql_errno() == ER_LOCK_WAIT_TIMEOUT)
2054
    DBUG_RETURN(1);
unknown's avatar
unknown committed
2055 2056 2057 2058 2059 2060 2061 2062 2063

#ifdef HAVE_NDB_BINLOG
  /*
    currently temporary error set in ndbcluster
  */
  List_iterator_fast<MYSQL_ERROR> it(thd->warn_list);
  MYSQL_ERROR *err;
  while ((err= it++))
  {
unknown's avatar
unknown committed
2064
    DBUG_PRINT("info", ("has warning %d %s", err->code, err->msg));
unknown's avatar
unknown committed
2065 2066 2067
    switch (err->code)
    {
    case ER_GET_TEMPORARY_ERRMSG:
2068
      DBUG_RETURN(1);
unknown's avatar
unknown committed
2069 2070 2071 2072 2073
    default:
      break;
    }
  }
#endif
2074
  DBUG_RETURN(0);
unknown's avatar
unknown committed
2075
}
2076

2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103

/**
  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().
*/
2104
int apply_event_and_update_pos(Log_event* ev, THD* thd, Relay_log_info* rli)
2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148
{
  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",
                      FLAGSTR(thd->options, OPTION_NOT_AUTOCOMMIT),
                      FLAGSTR(thd->options, OPTION_BEGIN),
                      rli->last_event_start_time));

  /*
    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).
  */

  thd->server_id = ev->server_id; // use the original server id for logging
  thd->set_time();                            // time the query
  thd->lex->current_select= 0;
  if (!ev->when)
    ev->when= my_time(0);
  ev->thd = thd; // because up to this point, ev->thd == 0

2149 2150 2151 2152 2153 2154 2155
  int reason= ev->shall_skip(rli);
  if (reason == Log_event::EVENT_SKIP_COUNT)
    --rli->slave_skip_counter;
  pthread_mutex_unlock(&rli->data_lock);
  if (reason == Log_event::EVENT_SKIP_NOT)
    exec_res= ev->apply_event(rli);

2156
#ifndef DBUG_OFF
2157 2158
  /*
    This only prints information to the debug trace.
2159

2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174
    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",
                      thd->options & OPTION_BEGIN ? 1 : 0,
                      rli->get_flag(Relay_log_info::IN_STMT)));
  DBUG_PRINT("skip_event", ("%s event was %s",
                            ev->get_type_str(), explain[reason]));
2175 2176 2177 2178 2179 2180
#endif

  DBUG_PRINT("info", ("apply_event error = %d", exec_res));
  if (exec_res == 0)
  {
    int error= ev->update_pos(rli);
2181 2182 2183 2184
#ifdef HAVE_purify
    if (!rli->is_fake)
#endif
    {
2185
#ifndef DBUG_OFF
2186
      char buf[22];
2187
#endif
2188 2189 2190 2191 2192 2193 2194 2195
      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));
    }
2196 2197 2198 2199 2200 2201 2202 2203 2204
    /*
      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)
    {
2205
      char buf[22];
2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240
      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
2241
     if a failure occurs when applying the event.)
2242 2243 2244 2245 2246 2247 2248

   - An error occurred when updating the binlog position.

  @retval 0 The event was applied.

  @retval 1 The event was not applied.
*/
2249
static int exec_relay_log_event(THD* thd, Relay_log_info* rli)
unknown's avatar
unknown committed
2250
{
2251 2252
  DBUG_ENTER("exec_relay_log_event");

2253 2254
  /*
     We acquire this mutex since we need it for all operations except
2255
     event execution. But we will release it in places where we will
2256 2257 2258
     wait for something for example inside of next_event().
   */
  pthread_mutex_lock(&rli->data_lock);
2259

2260
  Log_event * ev = next_event(rli);
2261

2262
  DBUG_ASSERT(rli->sql_thd==thd);
2263

2264
  if (sql_slave_killed(thd,rli))
2265
  {
2266
    pthread_mutex_unlock(&rli->data_lock);
unknown's avatar
unknown committed
2267
    delete ev;
2268
    DBUG_RETURN(1);
2269
  }
2270 2271
  if (ev)
  {
2272
    int exec_res;
2273

2274 2275 2276 2277
    /*
      This tests if the position of the beginning of the current event
      hits the UNTIL barrier.
    */
2278
    if (rli->until_condition != Relay_log_info::UNTIL_NONE &&
2279
        rli->is_until_satisfied(thd, ev))
2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290
    {
      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;
      pthread_mutex_unlock(&rli->data_lock);
      delete ev;
2291
      DBUG_RETURN(1);
2292
    }
2293
    exec_res= apply_event_and_update_pos(ev, thd, rli);
2294

2295
    /*
2296 2297 2298
      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.
2299
    */
2300
    if (ev->get_type_code() != FORMAT_DESCRIPTION_EVENT)
2301
    {
2302 2303
      DBUG_PRINT("info", ("Deleting the event after it has been executed"));
      delete ev;
2304 2305
    }

2306
    /*
2307 2308
      update_log_pos failed: this should not happen, so we don't
      retry.
2309
    */
2310 2311 2312
    if (exec_res == 2)
      DBUG_RETURN(1);

2313 2314
    if (slave_trans_retries)
    {
2315 2316
      int temp_err;
      if (exec_res && (temp_err= has_temporary_error(thd)))
2317 2318 2319 2320
      {
        const char *errmsg;
        /*
          We were in a transaction which has been rolled back because of a
unknown's avatar
unknown committed
2321 2322
          temporary error;
          let's seek back to BEGIN log event and retry it all again.
2323
	  Note, if lock wait timeout (innodb_lock_wait_timeout exceeded)
2324
	  there is no rollback since 5.0.13 (ref: manual).
2325 2326 2327 2328 2329 2330 2331 2332
          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.
        */
2333
        if (rli->trans_retries < slave_trans_retries)
2334 2335 2336 2337 2338 2339
        {
          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,
2340
                                      1, &errmsg, 1))
2341 2342 2343 2344 2345
            sql_print_error("Error initializing relay log position: %s",
                            errmsg);
          else
          {
            exec_res= 0;
2346
            end_trans(thd, ROLLBACK);
unknown's avatar
unknown committed
2347
            /* chance for concurrent connection to get more locks */
2348
            safe_sleep(thd, min(rli->trans_retries, MAX_SLAVE_RETRY_PAUSE),
unknown's avatar
unknown committed
2349
                       (CHECK_KILLED_FUNC)sql_slave_killed, (void*)rli);
2350
            pthread_mutex_lock(&rli->data_lock); // because of SHOW STATUS
unknown's avatar
unknown committed
2351
            rli->trans_retries++;
2352 2353 2354 2355
            rli->retried_trans++;
            pthread_mutex_unlock(&rli->data_lock);
            DBUG_PRINT("info", ("Slave retries transaction "
                                "rli->trans_retries: %lu", rli->trans_retries));
unknown's avatar
unknown committed
2356
          }
2357 2358 2359 2360 2361 2362 2363
        }
        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);
      }
2364
      else if ((exec_res && !temp_err) ||
2365 2366
               (opt_using_transactions &&
                rli->group_relay_log_pos == rli->event_relay_log_pos))
2367 2368
      {
        /*
2369 2370 2371
          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
2372 2373 2374
          non-transient error, the slave will stop with an error.
         */
        rli->trans_retries= 0; // restart from fresh
unknown's avatar
unknown committed
2375
        DBUG_PRINT("info", ("Resetting retry counter, rli->trans_retries: %lu",
2376
                            rli->trans_retries));
2377
      }
2378 2379
    }
    DBUG_RETURN(exec_res);
2380
  }
2381
  pthread_mutex_unlock(&rli->data_lock);
2382 2383
  rli->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_READ_FAILURE,
              ER(ER_SLAVE_RELAY_LOG_READ_FAILURE), "\
2384 2385 2386 2387 2388 2389 2390
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
2391
");
2392
  DBUG_RETURN(1);
unknown's avatar
unknown committed
2393 2394
}

2395

2396
static bool check_io_slave_killed(THD *thd, Master_info *mi, const char *info)
2397 2398 2399
{
  if (io_slave_killed(thd, mi))
  {
2400
    if (info && global_system_variables.log_warnings)
2401
      sql_print_information("%s", info);
2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433
    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.
*/

2434
static int try_to_reconnect(THD *thd, MYSQL *mysql, Master_info *mi,
2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470
                            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
    safe_sleep(thd, mi->connect_retry, (CHECK_KILLED_FUNC) io_slave_killed,
               (void *) mi);
  }
  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
    {
2471
      sql_print_information("%s", buf);
2472 2473 2474 2475 2476
    }
  }
  if (safe_reconnect(thd, mysql, mi, 1) || io_slave_killed(thd, mi))
  {
    if (global_system_variables.log_warnings)
2477
      sql_print_information("%s", messages[SLAVE_RECON_MSG_KILLED_AFTER]);
2478 2479 2480 2481 2482 2483
    return 1;
  }
  return 0;
}


2484 2485 2486 2487 2488
/**
  Slave IO thread entry point.

  @param arg Pointer to Master_info struct that holds information for
  the IO thread.
2489

2490 2491
  @return Always 0.
*/
2492
pthread_handler_t handle_slave_io(void *arg)
unknown's avatar
unknown committed
2493
{
unknown's avatar
unknown committed
2494 2495
  THD *thd; // needs to be first for thread_stack
  MYSQL *mysql;
2496
  Master_info *mi = (Master_info*)arg;
2497
  Relay_log_info *rli= &mi->rli;
unknown's avatar
unknown committed
2498 2499
  char llbuff[22];
  uint retry_count;
2500
  bool suppress_warnings;
2501
  int ret;
2502 2503 2504
#ifndef DBUG_OFF
  uint retry_count_reg= 0, retry_count_dump= 0, retry_count_event= 0;
#endif
unknown's avatar
unknown committed
2505 2506
  // needs to call my_thread_init(), otherwise we get a coredump in DBUG_ stuff
  my_thread_init();
2507
  DBUG_ENTER("handle_slave_io");
unknown's avatar
unknown committed
2508

2509
  DBUG_ASSERT(mi->inited);
unknown's avatar
unknown committed
2510 2511 2512
  mysql= NULL ;
  retry_count= 0;

2513
  pthread_mutex_lock(&mi->run_lock);
unknown's avatar
unknown committed
2514 2515 2516
  /* Inform waiting threads that slave has started */
  mi->slave_run_id++;

2517
#ifndef DBUG_OFF
2518
  mi->events_till_disconnect = disconnect_slave_event_count;
2519 2520
#endif

2521
  thd= new THD; // note that contructor of THD uses DBUG_ !
2522
  THD_CHECK_SENTRY(thd);
2523
  mi->io_thd = thd;
unknown's avatar
unknown committed
2524 2525

  pthread_detach_this_thread();
2526
  thd->thread_stack= (char*) &thd; // remember where our stack is
2527
  mi->clear_error();
2528
  if (init_slave_thread(thd, SLAVE_THD_IO))
unknown's avatar
unknown committed
2529 2530 2531 2532 2533 2534
  {
    pthread_cond_broadcast(&mi->start_cond);
    pthread_mutex_unlock(&mi->run_lock);
    sql_print_error("Failed during slave I/O thread initialization");
    goto err;
  }
2535
  pthread_mutex_lock(&LOCK_thread_count);
unknown's avatar
unknown committed
2536
  threads.append(thd);
2537
  pthread_mutex_unlock(&LOCK_thread_count);
2538 2539 2540
  mi->slave_running = 1;
  mi->abort_slave = 0;
  pthread_mutex_unlock(&mi->run_lock);
2541
  pthread_cond_broadcast(&mi->start_cond);
2542

2543
  DBUG_PRINT("master_info",("log_file_name: '%s'  position: %s",
unknown's avatar
unknown committed
2544 2545
                            mi->master_log_name,
                            llstr(mi->master_log_pos,llbuff)));
2546

unknown's avatar
SCRUM  
unknown committed
2547
  if (!(mi->mysql = mysql = mysql_init(NULL)))
unknown's avatar
unknown committed
2548
  {
2549 2550
    mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
               ER(ER_SLAVE_FATAL_ERROR), "error in mysql_init()");
unknown's avatar
unknown committed
2551 2552
    goto err;
  }
2553

2554
  thd_proc_info(thd, "Connecting to master");
2555
  // we can get killed during safe_connect
2556
  if (!safe_connect(thd, mysql, mi))
2557
  {
unknown's avatar
unknown committed
2558 2559 2560
    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,
2561 2562
			  IO_RPL_LOG_NAME,
			  llstr(mi->master_log_pos,llbuff));
2563 2564 2565 2566 2567
  /*
    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.
  */
2568 2569
    mysql->net.max_packet_size= thd->net.max_packet_size+= MAX_LOG_EVENT_HEADER;
  }
2570
  else
unknown's avatar
unknown committed
2571
  {
2572
    sql_print_information("Slave I/O thread killed while connecting to master");
unknown's avatar
unknown committed
2573 2574
    goto err;
  }
2575

2576
connected:
2577

2578 2579 2580 2581 2582 2583 2584 2585 2586 2587
    DBUG_EXECUTE_IF("dbug.before_get_running_status_yes",
                    {
                      const char act[]=
                        "now "
                        "wait_for signal.io_thread_let_running";
                      DBUG_ASSERT(opt_debug_sync_timeout > 0);
                      DBUG_ASSERT(!debug_sync_set_action(thd, 
                                                         STRING_WITH_LEN(act)));
                    };);

2588 2589
  // TODO: the assignment below should be under mutex (5.0)
  mi->slave_running= MYSQL_SLAVE_RUN_CONNECT;
2590
  thd->slave_net = &mysql->net;
2591
  thd_proc_info(thd, "Checking master version");
2592 2593 2594
  ret= get_master_version_and_clock(mysql, mi);
  if (ret == 1)
    /* Fatal error */
2595
    goto err;
2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608
  
  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;
  } 
2609 2610

  if (mi->rli.relay_log.description_event_for_queue->binlog_version > 1)
2611
  {
unknown's avatar
unknown committed
2612 2613 2614
    /*
      Register ourselves with the master.
    */
2615
    thd_proc_info(thd, "Registering slave on master");
2616
    if (register_slave_on_master(mysql, mi, &suppress_warnings))
2617
    {
2618 2619 2620 2621 2622 2623 2624 2625 2626
      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
2627 2628
        goto err;
      goto connected;
2629
    }
2630 2631 2632 2633 2634 2635 2636 2637 2638 2639
    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;
      });
2640
  }
2641

2642
  DBUG_PRINT("info",("Starting reading binary log from master"));
2643
  while (!io_slave_killed(thd,mi))
unknown's avatar
unknown committed
2644
  {
2645
    thd_proc_info(thd, "Requesting binlog dump");
unknown's avatar
unknown committed
2646
    if (request_dump(mysql, mi, &suppress_warnings))
unknown's avatar
unknown committed
2647 2648
    {
      sql_print_error("Failed on request_dump()");
2649 2650 2651 2652
      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
2653
        goto err;
unknown's avatar
unknown committed
2654 2655
      goto connected;
    }
2656 2657 2658 2659 2660 2661 2662 2663 2664 2665
    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;
      });
unknown's avatar
unknown committed
2666

2667
    DBUG_ASSERT(mi->last_error().number == 0);
2668
    while (!io_slave_killed(thd,mi))
unknown's avatar
unknown committed
2669
    {
2670
      ulong event_len;
2671
      /*
2672
         We say "waiting" because read_event() will wait if there's nothing to
2673 2674 2675
         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.
2676
      */
2677
      thd_proc_info(thd, "Waiting for master to send event");
2678
      event_len= read_event(mysql, mi, &suppress_warnings);
2679 2680
      if (check_io_slave_killed(thd, mi, "Slave I/O thread killed while \
reading event"))
unknown's avatar
unknown committed
2681
        goto err;
2682 2683 2684 2685 2686 2687 2688 2689 2690 2691
      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;
        });
2692

unknown's avatar
unknown committed
2693 2694
      if (event_len == packet_error)
      {
unknown's avatar
unknown committed
2695
        uint mysql_error_number= mysql_errno(mysql);
2696 2697
        switch (mysql_error_number) {
        case CR_NET_PACKET_TOO_LARGE:
unknown's avatar
unknown committed
2698
          sql_print_error("\
2699 2700 2701
Log entry on master is longer than max_allowed_packet (%ld) on \
slave. If the entry is correct, restart the server with a higher value of \
max_allowed_packet",
unknown's avatar
unknown committed
2702
                          thd->variables.max_allowed_packet);
2703
          mi->report(ERROR_LEVEL, ER_NET_PACKET_TOO_LARGE,
2704
                     "%s", ER(ER_NET_PACKET_TOO_LARGE));
unknown's avatar
unknown committed
2705
          goto err;
2706
        case ER_MASTER_FATAL_ERROR_READING_BINLOG:
2707 2708 2709
          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
2710
          goto err;
2711
        case ER_OUT_OF_RESOURCES:
2712 2713
          sql_print_error("\
Stopping slave I/O thread due to out-of-memory error from master");
2714
          mi->report(ERROR_LEVEL, ER_OUT_OF_RESOURCES,
2715
                     "%s", ER(ER_OUT_OF_RESOURCES));
unknown's avatar
unknown committed
2716 2717
          goto err;
        }
2718 2719
        if (try_to_reconnect(thd, mysql, mi, &retry_count, suppress_warnings,
                             reconnect_messages[SLAVE_RECON_ACT_EVENT]))
unknown's avatar
unknown committed
2720 2721
          goto err;
        goto connected;
unknown's avatar
unknown committed
2722
      } // if (event_len == packet_error)
2723

unknown's avatar
unknown committed
2724
      retry_count=0;                    // ok event, reset retry counter
2725
      thd_proc_info(thd, "Queueing master event to the relay log");
unknown's avatar
unknown committed
2726
      if (queue_event(mi,(const char*)mysql->net.read_pos + 1,
unknown's avatar
unknown committed
2727
                      event_len))
unknown's avatar
unknown committed
2728
      {
2729 2730 2731
        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
2732
        goto err;
unknown's avatar
unknown committed
2733
      }
2734
      if (flush_master_info(mi, TRUE, TRUE))
2735 2736 2737 2738
      {
        sql_print_error("Failed to flush master info file");
        goto err;
      }
2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750
      /*
        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.
      */
2751 2752 2753 2754 2755
#ifndef DBUG_OFF
      {
        char llbuf1[22], llbuf2[22];
        DBUG_PRINT("info", ("log_space_limit=%s log_space_total=%s \
ignore_log_space_limit=%d",
2756 2757
                            llstr(rli->log_space_limit,llbuf1),
                            llstr(rli->log_space_total,llbuf2),
unknown's avatar
unknown committed
2758
                            (int) rli->ignore_log_space_limit));
2759 2760 2761
      }
#endif

2762
      if (rli->log_space_limit && rli->log_space_limit <
unknown's avatar
unknown committed
2763
          rli->log_space_total &&
2764
          !rli->ignore_log_space_limit)
unknown's avatar
unknown committed
2765 2766 2767
        if (wait_for_relay_log_space(rli))
        {
          sql_print_error("Slave I/O thread aborted while waiting for relay \
unknown's avatar
unknown committed
2768
log space");
unknown's avatar
unknown committed
2769 2770 2771
          goto err;
        }
    }
2772
  }
unknown's avatar
unknown committed
2773

unknown's avatar
unknown committed
2774
  // error = 0;
unknown's avatar
unknown committed
2775
err:
2776
  // print the current replication position
2777
  sql_print_information("Slave I/O thread exiting, read up to log '%s', position %s",
unknown's avatar
unknown committed
2778
                  IO_RPL_LOG_NAME, llstr(mi->master_log_pos,llbuff));
2779
  thd->set_query(NULL, 0);
unknown's avatar
unknown committed
2780
  thd->reset_db(NULL, 0);
unknown's avatar
unknown committed
2781 2782
  if (mysql)
  {
2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793
    /*
      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
2794
    mysql_close(mysql);
unknown's avatar
unknown committed
2795 2796
    mi->mysql=0;
  }
2797
  write_ignored_events_info_to_relay_log(thd, mi);
2798
  thd_proc_info(thd, "Waiting for slave mutex on exit");
2799
  pthread_mutex_lock(&mi->run_lock);
2800

2801 2802 2803
  /* Forget the relay log's format */
  delete mi->rli.relay_log.description_event_for_queue;
  mi->rli.relay_log.description_event_for_queue= 0;
2804
  // TODO: make rpl_status part of Master_info
2805
  change_rpl_status(RPL_ACTIVE_SLAVE,RPL_IDLE_SLAVE);
2806
  DBUG_ASSERT(thd->net.buff != 0);
unknown's avatar
unknown committed
2807
  net_end(&thd->net); // destructor will not free it, because net.vio is 0
2808
  close_thread_tables(thd);
2809
  pthread_mutex_lock(&LOCK_thread_count);
2810
  THD_CHECK_SENTRY(thd);
unknown's avatar
unknown committed
2811
  delete thd;
2812
  pthread_mutex_unlock(&LOCK_thread_count);
unknown's avatar
unknown committed
2813 2814 2815
  mi->abort_slave= 0;
  mi->slave_running= 0;
  mi->io_thd= 0;
2816 2817 2818 2819 2820
  /*
    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)
   */ 
2821
  pthread_cond_broadcast(&mi->stop_cond);       // tell the world we are done
2822
  DBUG_EXECUTE_IF("simulate_slave_delay_at_terminate_bug38694", sleep(5););
2823
  pthread_mutex_unlock(&mi->run_lock);
2824 2825

  DBUG_LEAVE;                                   // Must match DBUG_ENTER()
2826
  my_thread_end();
unknown's avatar
unknown committed
2827
  pthread_exit(0);
2828
  return 0;                                     // Avoid compiler warnings
unknown's avatar
unknown committed
2829 2830
}

2831 2832 2833 2834 2835
/*
  Check the temporary directory used by commands like
  LOAD DATA INFILE.
 */
static 
2836
int check_temp_dir(char* tmp_file)
2837 2838 2839
{
  int fd;
  MY_DIR *dirp;
2840 2841
  char tmp_dir[FN_REFLEN];
  size_t tmp_dir_size;
2842 2843 2844

  DBUG_ENTER("check_temp_dir");

2845 2846 2847 2848 2849
  /*
    Get the directory from the temporary file.
  */
  dirname_part(tmp_dir, tmp_file, &tmp_dir_size);

2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872
  /*
    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.
   */
  if ((fd= my_create(tmp_file, CREATE_MODE,
                     O_WRONLY | O_BINARY | O_EXCL | O_NOFOLLOW,
                     MYF(MY_WME))) < 0)
  DBUG_RETURN(1);

  /*
    Clean up.
   */
  my_close(fd, MYF(0));
  my_delete(tmp_file, MYF(0));

  DBUG_RETURN(0);
}
unknown's avatar
unknown committed
2873

2874 2875 2876 2877 2878
/**
  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
2879

2880 2881
  @return Always 0.
*/
2882
pthread_handler_t handle_slave_sql(void *arg)
2883
{
unknown's avatar
unknown committed
2884
  THD *thd;                     /* needs to be first for thread_stack */
2885
  char llbuff[22],llbuff1[22];
2886 2887 2888 2889 2890
  char saved_log_name[FN_REFLEN];
  char saved_master_log_name[FN_REFLEN];
  my_off_t saved_log_pos;
  my_off_t saved_master_log_pos;
  my_off_t saved_skip= 0;
2891

2892
  Relay_log_info* rli = &((Master_info*)arg)->rli;
unknown's avatar
unknown committed
2893 2894 2895 2896
  const char *errmsg;

  // needs to call my_thread_init(), otherwise we get a coredump in DBUG_ stuff
  my_thread_init();
2897
  DBUG_ENTER("handle_slave_sql");
unknown's avatar
unknown committed
2898

2899 2900 2901
  DBUG_ASSERT(rli->inited);
  pthread_mutex_lock(&rli->run_lock);
  DBUG_ASSERT(!rli->slave_running);
unknown's avatar
unknown committed
2902
  errmsg= 0;
unknown's avatar
unknown committed
2903
#ifndef DBUG_OFF
2904
  rli->events_till_abort = abort_slave_event_count;
unknown's avatar
unknown committed
2905
#endif
2906

unknown's avatar
unknown committed
2907
  thd = new THD; // note that contructor of THD uses DBUG_ !
2908
  thd->thread_stack = (char*)&thd; // remember where our stack is
2909
  rli->sql_thd= thd;
2910
  
unknown's avatar
unknown committed
2911 2912
  /* Inform waiting threads that slave has started */
  rli->slave_run_id++;
2913
  rli->slave_running = 1;
unknown's avatar
unknown committed
2914

2915 2916
  pthread_detach_this_thread();
  if (init_slave_thread(thd, SLAVE_THD_SQL))
unknown's avatar
unknown committed
2917 2918 2919 2920 2921 2922 2923
  {
    /*
      TODO: this is currently broken - slave start and change master
      will be stuck if we fail here
    */
    pthread_cond_broadcast(&rli->start_cond);
    pthread_mutex_unlock(&rli->run_lock);
2924 2925
    rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, 
                "Failed during slave thread initialization");
unknown's avatar
unknown committed
2926 2927
    goto err;
  }
2928
  thd->init_for_queries();
2929
  thd->temporary_tables = rli->save_temporary_tables; // restore temp tables
2930
  set_thd_in_use_temporary_tables(rli);   // (re)set sql_thd in use for saved temp tables
2931
  pthread_mutex_lock(&LOCK_thread_count);
2932
  threads.append(thd);
2933
  pthread_mutex_unlock(&LOCK_thread_count);
2934 2935 2936 2937 2938 2939 2940 2941
  /*
    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.
  */
2942 2943
  rli->abort_slave = 0;
  pthread_mutex_unlock(&rli->run_lock);
2944
  pthread_cond_broadcast(&rli->start_cond);
2945

unknown's avatar
unknown committed
2946 2947 2948
  /*
    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
2949
    though there's no problem anymore). Do not reset the master timestamp
2950 2951 2952 2953
    (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
2954
    But the master timestamp is reset by RESET SLAVE & CHANGE MASTER.
unknown's avatar
unknown committed
2955
  */
2956
  rli->clear_error();
2957 2958

  //tell the I/O thread to take relay_log_space_limit into account from now on
2959
  pthread_mutex_lock(&rli->log_space_lock);
2960
  rli->ignore_log_space_limit= 0;
2961
  pthread_mutex_unlock(&rli->log_space_lock);
2962
  rli->trans_retries= 0; // start from "no error"
unknown's avatar
unknown committed
2963
  DBUG_PRINT("info", ("rli->trans_retries: %lu", rli->trans_retries));
2964

2965
  if (init_relay_log_pos(rli,
unknown's avatar
unknown committed
2966 2967 2968
                         rli->group_relay_log_name,
                         rli->group_relay_log_pos,
                         1 /*need data lock*/, &errmsg,
2969
                         1 /*look for a description_event*/))
2970 2971 2972
  { 
    rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, 
                "Error initializing relay log position: %s", errmsg);
2973 2974
    goto err;
  }
2975
  THD_CHECK_SENTRY(thd);
2976 2977 2978 2979
#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
2980
                        llstr(my_b_tell(rli->cur_log),llbuf1),
2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999
                        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
3000
  DBUG_ASSERT(rli->sql_thd == thd);
3001 3002

  DBUG_PRINT("master_info",("log_file_name: %s  position: %s",
unknown's avatar
unknown committed
3003 3004
                            rli->group_master_log_name,
                            llstr(rli->group_master_log_pos,llbuff)));
3005
  if (global_system_variables.log_warnings)
3006
    sql_print_information("Slave SQL thread initialized, starting replication in \
unknown's avatar
unknown committed
3007
log '%s' at position %s, relay log '%s' position: %s", RPL_LOG_NAME,
unknown's avatar
unknown committed
3008 3009
                    llstr(rli->group_master_log_pos,llbuff),rli->group_relay_log_name,
                    llstr(rli->group_relay_log_pos,llbuff1));
3010

3011
  if (check_temp_dir(rli->slave_patternload_file))
3012 3013 3014 3015 3016 3017 3018
  {
    rli->report(ERROR_LEVEL, thd->main_da.sql_errno(), 
                "Unable to use slave's temporary directory %s - %s", 
                slave_load_tmpdir, thd->main_da.message());
    goto err;
  }

unknown's avatar
unknown committed
3019
  /* execute init_slave variable */
unknown's avatar
unknown committed
3020
  if (sys_init_slave.value_length)
unknown's avatar
unknown committed
3021
  {
3022
    execute_init_command(thd, &sys_init_slave, &LOCK_sys_init_slave);
3023
    if (thd->is_slave_error)
unknown's avatar
unknown committed
3024
    {
3025 3026
      rli->report(ERROR_LEVEL, thd->main_da.sql_errno(), 
                  "Slave SQL thread aborted. Can't execute init_slave query");
unknown's avatar
unknown committed
3027 3028 3029 3030
      goto err;
    }
  }

3031 3032 3033 3034 3035
  /*
    First check until condition - probably there is nothing to execute. We
    do not want to wait for next event in this case.
  */
  pthread_mutex_lock(&rli->data_lock);
3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046
  if (rli->slave_skip_counter)
  {
    char *pos;
    pos= strmake(saved_log_name, rli->group_relay_log_name, FN_REFLEN - 1);
    pos= '\0';
    pos= strmake(saved_master_log_name, rli->group_master_log_name, FN_REFLEN - 1);
    pos= '\0';
    saved_log_pos= rli->group_relay_log_pos;
    saved_master_log_pos= rli->group_master_log_pos;
    saved_skip= rli->slave_skip_counter;
  }
3047
  if (rli->until_condition != Relay_log_info::UNTIL_NONE &&
3048
      rli->is_until_satisfied(thd, NULL))
3049 3050 3051 3052 3053 3054 3055 3056 3057
  {
    char buf[22];
    sql_print_information("Slave SQL thread stopped because it reached its"
                          " UNTIL position %s", llstr(rli->until_pos(), buf));
    pthread_mutex_unlock(&rli->data_lock);
    goto err;
  }
  pthread_mutex_unlock(&rli->data_lock);

3058 3059
  /* Read queries from the IO/THREAD until this thread is killed */

3060
  while (!sql_slave_killed(thd,rli))
3061
  {
3062
    thd_proc_info(thd, "Reading event from the relay log");
3063
    DBUG_ASSERT(rli->sql_thd == thd);
3064
    THD_CHECK_SENTRY(thd);
3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079

    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;
    }
    
3080 3081
    if (exec_relay_log_event(thd,rli))
    {
3082
      DBUG_PRINT("info", ("exec_relay_log_event() failed"));
3083
      // do not scare the user if SQL thread was simply killed or stopped
3084
      if (!sql_slave_killed(thd,rli))
unknown's avatar
unknown committed
3085 3086
      {
        /*
3087 3088 3089
          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
3090
        */
3091
        uint32 const last_errno= rli->last_error().number;
3092

3093
        if (thd->is_error())
unknown's avatar
unknown committed
3094
        {
3095 3096 3097 3098 3099
          char const *const errmsg= thd->main_da.message();

          DBUG_PRINT("info",
                     ("thd->main_da.sql_errno()=%d; rli->last_error.number=%d",
                      thd->main_da.sql_errno(), last_errno));
3100
          if (last_errno == 0)
unknown's avatar
unknown committed
3101
          {
3102 3103 3104 3105
            /*
 	      This function is reporting an error which was not reported
 	      while executing exec_relay_log_event().
 	    */ 
3106
            rli->report(ERROR_LEVEL, thd->main_da.sql_errno(), "%s", errmsg);
unknown's avatar
unknown committed
3107
          }
3108
          else if (last_errno != thd->main_da.sql_errno())
unknown's avatar
unknown committed
3109
          {
3110 3111 3112 3113 3114 3115
            /*
             * 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
3116
            sql_print_error("Slave (additional info): %s Error_code: %d",
3117
                            errmsg, thd->main_da.sql_errno());
unknown's avatar
unknown committed
3118 3119 3120 3121 3122 3123
          }
        }

        /* Print any warnings issued */
        List_iterator_fast<MYSQL_ERROR> it(thd->warn_list);
        MYSQL_ERROR *err;
3124 3125 3126 3127 3128
        /*
          Added controlled slave thread cancel for replication
          of user-defined variables.
        */
        bool udf_error = false;
unknown's avatar
unknown committed
3129
        while ((err= it++))
3130 3131 3132
        {
          if (err->code == ER_CANT_OPEN_LIBRARY)
            udf_error = true;
unknown's avatar
unknown committed
3133
          sql_print_warning("Slave: %s Error_code: %d",err->msg, err->code);
3134 3135 3136 3137 3138 3139 3140 3141 3142
        }
        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("\
3143
Error running query, slave SQL thread aborted. Fix the problem, and restart \
unknown's avatar
unknown committed
3144
the slave SQL thread with \"SLAVE START\". We stopped at log \
3145
'%s' position %s", RPL_LOG_NAME, llstr(rli->group_master_log_pos, llbuff));
unknown's avatar
unknown committed
3146
      }
3147 3148
      goto err;
    }
3149
  }
3150

3151
  /* Thread stopped. Print the current replication position to the log */
unknown's avatar
unknown committed
3152
  sql_print_information("Slave SQL thread exiting, replication stopped in log "
unknown's avatar
unknown committed
3153 3154
                        "'%s' at position %s",
                        RPL_LOG_NAME, llstr(rli->group_master_log_pos,llbuff));
3155 3156

 err:
3157 3158 3159 3160 3161 3162 3163 3164

  /*
    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:
  */
  rli->cleanup_context(thd, 1);
3165 3166 3167 3168 3169
  /*
    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)).
  */
3170 3171
  thd->catalog= 0;
  thd->set_query(NULL, 0);
unknown's avatar
unknown committed
3172
  thd->reset_db(NULL, 0);
3173
  thd_proc_info(thd, "Waiting for slave mutex on exit");
3174
  pthread_mutex_lock(&rli->run_lock);
3175 3176
  /* We need data_lock, at least to wake up any waiting master_pos_wait() */
  pthread_mutex_lock(&rli->data_lock);
3177
  DBUG_ASSERT(rli->slave_running == 1); // tracking buffer overrun
3178
  /* When master_pos_wait() wakes up it will check this and terminate */
unknown's avatar
unknown committed
3179
  rli->slave_running= 0;
3180 3181 3182
  /* Forget the relay log's format */
  delete rli->relay_log.description_event_for_exec;
  rli->relay_log.description_event_for_exec= 0;
3183 3184 3185 3186
  /* Wake up master_pos_wait() */
  pthread_mutex_unlock(&rli->data_lock);
  DBUG_PRINT("info",("Signaling possibly waiting master_pos_wait() functions"));
  pthread_cond_broadcast(&rli->data_cond);
unknown's avatar
unknown committed
3187
  rli->ignore_log_space_limit= 0; /* don't need any lock */
3188 3189
  /* we die so won't remember charset - re-update them on next thread start */
  rli->cached_charset_invalidate();
3190
  rli->save_temporary_tables = thd->temporary_tables;
unknown's avatar
unknown committed
3191 3192 3193 3194 3195

  /*
    TODO: see if we can do this conditionally in next_event() instead
    to avoid unneeded position re-init
  */
3196 3197 3198 3199
  thd->temporary_tables = 0; // remove tempation from destructor to close them
  DBUG_ASSERT(thd->net.buff != 0);
  net_end(&thd->net); // destructor will not free it, because we are weird
  DBUG_ASSERT(rli->sql_thd == thd);
3200
  THD_CHECK_SENTRY(thd);
3201
  rli->sql_thd= 0;
3202
  set_thd_in_use_temporary_tables(rli);  // (re)set sql_thd in use for saved temp tables
3203
  pthread_mutex_lock(&LOCK_thread_count);
3204
  THD_CHECK_SENTRY(thd);
3205 3206
  delete thd;
  pthread_mutex_unlock(&LOCK_thread_count);
3207 3208 3209 3210 3211
 /*
  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)
 */ 
3212
  pthread_cond_broadcast(&rli->stop_cond);
3213
  DBUG_EXECUTE_IF("simulate_slave_delay_at_terminate_bug38694", sleep(5););
3214
  pthread_mutex_unlock(&rli->run_lock);  // tell the world we are done
3215 3216

  DBUG_LEAVE;                                   // Must match DBUG_ENTER()
unknown's avatar
unknown committed
3217
  my_thread_end();
3218
  pthread_exit(0);
3219
  return 0;                                     // Avoid compiler warnings
3220
}
unknown's avatar
unknown committed
3221

3222

unknown's avatar
unknown committed
3223
/*
3224
  process_io_create_file()
unknown's avatar
unknown committed
3225
*/
3226

3227
static int process_io_create_file(Master_info* mi, Create_file_log_event* cev)
unknown's avatar
unknown committed
3228 3229 3230 3231
{
  int error = 1;
  ulong num_bytes;
  bool cev_not_written;
3232 3233
  THD *thd = mi->io_thd;
  NET *net = &mi->mysql->net;
unknown's avatar
unknown committed
3234
  DBUG_ENTER("process_io_create_file");
unknown's avatar
unknown committed
3235 3236

  if (unlikely(!cev->is_valid()))
unknown's avatar
unknown committed
3237
    DBUG_RETURN(1);
3238 3239

  if (!rpl_filter->db_ok(cev->db))
unknown's avatar
unknown committed
3240 3241
  {
    skip_load_data_infile(net);
unknown's avatar
unknown committed
3242
    DBUG_RETURN(0);
unknown's avatar
unknown committed
3243 3244 3245
  }
  DBUG_ASSERT(cev->inited_from_old);
  thd->file_id = cev->file_id = mi->file_id++;
3246
  thd->server_id = cev->server_id;
unknown's avatar
unknown committed
3247
  cev_not_written = 1;
unknown's avatar
unknown committed
3248

unknown's avatar
unknown committed
3249 3250 3251
  if (unlikely(net_request_file(net,cev->fname)))
  {
    sql_print_error("Slave I/O: failed requesting download of '%s'",
unknown's avatar
unknown committed
3252
                    cev->fname);
unknown's avatar
unknown committed
3253 3254 3255
    goto err;
  }

unknown's avatar
unknown committed
3256 3257 3258 3259
  /*
    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
3260 3261
  */
  {
3262
    Append_block_log_event aev(thd,0,0,0,0);
unknown's avatar
unknown committed
3263

unknown's avatar
unknown committed
3264 3265 3266 3267
    for (;;)
    {
      if (unlikely((num_bytes=my_net_read(net)) == packet_error))
      {
unknown's avatar
unknown committed
3268 3269 3270
        sql_print_error("Network read error downloading '%s' from master",
                        cev->fname);
        goto err;
unknown's avatar
unknown committed
3271 3272 3273
      }
      if (unlikely(!num_bytes)) /* eof */
      {
3274 3275
	/* 3.23 master wants it */
        net_write_command(net, 0, (uchar*) "", 0, (uchar*) "", 0);
unknown's avatar
unknown committed
3276 3277 3278 3279 3280 3281 3282
        /*
          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
3283 3284 3285 3286 3287
          break;
        Execute_load_log_event xev(thd,0,0);
        xev.log_pos = cev->log_pos;
        if (unlikely(mi->rli.relay_log.append(&xev)))
        {
3288 3289 3290
          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
3291 3292 3293 3294
          goto err;
        }
        mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total);
        break;
unknown's avatar
unknown committed
3295 3296 3297
      }
      if (unlikely(cev_not_written))
      {
unknown's avatar
unknown committed
3298
        cev->block = net->read_pos;
unknown's avatar
unknown committed
3299 3300 3301
        cev->block_len = num_bytes;
        if (unlikely(mi->rli.relay_log.append(cev)))
        {
3302 3303 3304
          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
3305 3306 3307 3308
          goto err;
        }
        cev_not_written=0;
        mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total);
unknown's avatar
unknown committed
3309 3310 3311
      }
      else
      {
unknown's avatar
unknown committed
3312
        aev.block = net->read_pos;
unknown's avatar
unknown committed
3313 3314 3315 3316
        aev.block_len = num_bytes;
        aev.log_pos = cev->log_pos;
        if (unlikely(mi->rli.relay_log.append(&aev)))
        {
3317 3318 3319
          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
3320 3321 3322
          goto err;
        }
        mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total) ;
unknown's avatar
unknown committed
3323 3324 3325 3326 3327
      }
    }
  }
  error=0;
err:
unknown's avatar
unknown committed
3328
  DBUG_RETURN(error);
unknown's avatar
unknown committed
3329
}
unknown's avatar
unknown committed
3330

3331

unknown's avatar
unknown committed
3332
/*
unknown's avatar
unknown committed
3333 3334 3335 3336
  Start using a new binary log on the master

  SYNOPSIS
    process_io_rotate()
unknown's avatar
unknown committed
3337 3338
    mi                  master_info for the slave
    rev                 The rotate log event read from the binary log
unknown's avatar
unknown committed
3339 3340

  DESCRIPTION
3341
    Updates the master info with the place in the next binary
unknown's avatar
unknown committed
3342
    log where we should start reading.
3343
    Rotate the relay log to avoid mixed-format relay logs.
unknown's avatar
unknown committed
3344 3345 3346 3347 3348

  NOTES
    We assume we already locked mi->data_lock

  RETURN VALUES
unknown's avatar
unknown committed
3349 3350
    0           ok
    1           Log event is illegal
unknown's avatar
unknown committed
3351 3352 3353

*/

3354
static int process_io_rotate(Master_info *mi, Rotate_log_event *rev)
3355
{
3356
  DBUG_ENTER("process_io_rotate");
unknown's avatar
unknown committed
3357
  safe_mutex_assert_owner(&mi->data_lock);
3358

unknown's avatar
unknown committed
3359
  if (unlikely(!rev->is_valid()))
3360
    DBUG_RETURN(1);
unknown's avatar
unknown committed
3361

3362
  /* Safe copy as 'rev' has been "sanitized" in Rotate_log_event's ctor */
unknown's avatar
unknown committed
3363 3364
  memcpy(mi->master_log_name, rev->new_log_ident, rev->ident_len+1);
  mi->master_log_pos= rev->pos;
unknown's avatar
unknown committed
3365
  DBUG_PRINT("info", ("master_log_pos: '%s' %lu",
unknown's avatar
unknown committed
3366
                      mi->master_log_name, (ulong) mi->master_log_pos));
3367
#ifndef DBUG_OFF
unknown's avatar
unknown committed
3368 3369 3370 3371 3372
  /*
    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)
3373
    mi->events_till_disconnect++;
3374
#endif
3375

3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389
  /*
    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)
  {
    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);
  }
3390 3391 3392 3393
  /*
    Rotate the relay log makes binlog format detection easier (at next slave
    start or mysqlbinlog)
  */
3394
  rotate_relay_log(mi); /* will take the right mutexes */
3395
  DBUG_RETURN(0);
3396 3397
}

unknown's avatar
unknown committed
3398
/*
3399 3400
  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
3401
*/
3402
static int queue_binlog_ver_1_event(Master_info *mi, const char *buf,
unknown's avatar
unknown committed
3403
                           ulong event_len)
3404
{
unknown's avatar
unknown committed
3405
  const char *errmsg = 0;
unknown's avatar
unknown committed
3406 3407 3408
  ulong inc_pos;
  bool ignore_event= 0;
  char *tmp_buf = 0;
3409
  Relay_log_info *rli= &mi->rli;
3410
  DBUG_ENTER("queue_binlog_ver_1_event");
unknown's avatar
unknown committed
3411

unknown's avatar
unknown committed
3412 3413 3414
  /*
    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
3415 3416 3417 3418 3419
  */
  if (buf[EVENT_TYPE_OFFSET] == LOAD_EVENT)
  {
    if (unlikely(!(tmp_buf=(char*)my_malloc(event_len+1,MYF(MY_WME)))))
    {
3420 3421
      mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
                 ER(ER_SLAVE_FATAL_ERROR), "Memory allocation failed");
unknown's avatar
unknown committed
3422
      DBUG_RETURN(1);
unknown's avatar
unknown committed
3423 3424
    }
    memcpy(tmp_buf,buf,event_len);
3425 3426 3427 3428 3429 3430 3431 3432
    /*
      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
3433
    int4store(tmp_buf+EVENT_LEN_OFFSET, event_len);
unknown's avatar
unknown committed
3434 3435
    buf = (const char*)tmp_buf;
  }
3436 3437 3438 3439 3440 3441
  /*
    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).
  */
unknown's avatar
unknown committed
3442
  Log_event *ev = Log_event::read_log_event(buf,event_len, &errmsg,
3443
                                            mi->rli.relay_log.description_event_for_queue);
3444
  if (unlikely(!ev))
3445 3446
  {
    sql_print_error("Read invalid event from master: '%s',\
unknown's avatar
unknown committed
3447
 master could be corrupt but a more likely cause of this is a bug",
unknown's avatar
unknown committed
3448
                    errmsg);
unknown's avatar
unknown committed
3449 3450
    my_free((char*) tmp_buf, MYF(MY_ALLOW_ZERO_PTR));
    DBUG_RETURN(1);
3451
  }
3452

3453
  pthread_mutex_lock(&mi->data_lock);
3454
  ev->log_pos= mi->master_log_pos; /* 3.23 events don't contain log_pos */
unknown's avatar
unknown committed
3455
  switch (ev->get_type_code()) {
unknown's avatar
unknown committed
3456
  case STOP_EVENT:
3457
    ignore_event= 1;
unknown's avatar
unknown committed
3458 3459
    inc_pos= event_len;
    break;
3460
  case ROTATE_EVENT:
3461
    if (unlikely(process_io_rotate(mi,(Rotate_log_event*)ev)))
3462 3463
    {
      delete ev;
3464
      pthread_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
3465
      DBUG_RETURN(1);
3466
    }
unknown's avatar
unknown committed
3467
    inc_pos= 0;
3468
    break;
unknown's avatar
unknown committed
3469
  case CREATE_FILE_EVENT:
3470 3471 3472 3473 3474 3475
    /*
      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
3476
  {
unknown's avatar
unknown committed
3477
    /* We come here when and only when tmp_buf != 0 */
3478
    DBUG_ASSERT(tmp_buf != 0);
3479 3480
    inc_pos=event_len;
    ev->log_pos+= inc_pos;
unknown's avatar
unknown committed
3481
    int error = process_io_create_file(mi,(Create_file_log_event*)ev);
3482
    delete ev;
3483
    mi->master_log_pos += inc_pos;
unknown's avatar
unknown committed
3484
    DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
3485
    pthread_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
3486
    my_free((char*)tmp_buf, MYF(0));
unknown's avatar
unknown committed
3487
    DBUG_RETURN(error);
unknown's avatar
unknown committed
3488
  }
3489
  default:
unknown's avatar
unknown committed
3490
    inc_pos= event_len;
3491 3492
    break;
  }
unknown's avatar
unknown committed
3493
  if (likely(!ignore_event))
3494
  {
unknown's avatar
unknown committed
3495 3496
    if (ev->log_pos)
      /*
3497 3498 3499 3500
         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
3501
    if (unlikely(rli->relay_log.append(ev)))
3502 3503 3504
    {
      delete ev;
      pthread_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
3505
      DBUG_RETURN(1);
3506
    }
unknown's avatar
unknown committed
3507
    rli->relay_log.harvest_bytes_written(&rli->log_space_total);
3508 3509
  }
  delete ev;
unknown's avatar
unknown committed
3510
  mi->master_log_pos+= inc_pos;
unknown's avatar
unknown committed
3511
  DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
3512
  pthread_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
3513
  DBUG_RETURN(0);
3514 3515
}

3516 3517 3518 3519
/*
  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.
*/
3520
static int queue_binlog_ver_3_event(Master_info *mi, const char *buf,
unknown's avatar
unknown committed
3521
                           ulong event_len)
3522 3523 3524 3525
{
  const char *errmsg = 0;
  ulong inc_pos;
  char *tmp_buf = 0;
3526
  Relay_log_info *rli= &mi->rli;
3527 3528 3529 3530 3531 3532 3533 3534 3535
  DBUG_ENTER("queue_binlog_ver_3_event");

  /* read_log_event() will adjust log_pos to be end_log_pos */
  Log_event *ev = Log_event::read_log_event(buf,event_len, &errmsg,
                                            mi->rli.relay_log.description_event_for_queue);
  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
3536
                    errmsg);
3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566
    my_free((char*) tmp_buf, MYF(MY_ALLOW_ZERO_PTR));
    DBUG_RETURN(1);
  }
  pthread_mutex_lock(&mi->data_lock);
  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;
      pthread_mutex_unlock(&mi->data_lock);
      DBUG_RETURN(1);
    }
    inc_pos= 0;
    break;
  default:
    inc_pos= event_len;
    break;
  }
  if (unlikely(rli->relay_log.append(ev)))
  {
    delete ev;
    pthread_mutex_unlock(&mi->data_lock);
    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
3567
  DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578
  pthread_mutex_unlock(&mi->data_lock);
  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
3579
  TODO:
3580 3581 3582 3583
    Test this code before release - it has to be tested on a separate
    setup with 3.23 master or 4.0 master
*/

3584
static int queue_old_event(Master_info *mi, const char *buf,
unknown's avatar
unknown committed
3585
                           ulong event_len)
3586
{
3587 3588
  DBUG_ENTER("queue_old_event");

3589 3590 3591
  switch (mi->rli.relay_log.description_event_for_queue->binlog_version)
  {
  case 1:
3592
      DBUG_RETURN(queue_binlog_ver_1_event(mi,buf,event_len));
3593
  case 3:
3594
      DBUG_RETURN(queue_binlog_ver_3_event(mi,buf,event_len));
3595 3596
  default: /* unsupported format; eg version 2 */
    DBUG_PRINT("info",("unsupported binlog format %d in queue_old_event()",
unknown's avatar
unknown committed
3597
                       mi->rli.relay_log.description_event_for_queue->binlog_version));
3598
    DBUG_RETURN(1);
3599 3600
  }
}
3601

unknown's avatar
unknown committed
3602
/*
3603 3604
  queue_event()

3605 3606 3607 3608 3609
  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
3610 3611
*/

3612
static int queue_event(Master_info* mi,const char* buf, ulong event_len)
3613
{
unknown's avatar
unknown committed
3614 3615
  int error= 0;
  ulong inc_pos;
3616
  Relay_log_info *rli= &mi->rli;
3617
  pthread_mutex_t *log_lock= rli->relay_log.get_log_lock();
unknown's avatar
unknown committed
3618 3619
  DBUG_ENTER("queue_event");

unknown's avatar
unknown committed
3620 3621
  LINT_INIT(inc_pos);

3622 3623
  if (mi->rli.relay_log.description_event_for_queue->binlog_version<4 &&
      buf[EVENT_TYPE_OFFSET] != FORMAT_DESCRIPTION_EVENT /* a way to escape */)
unknown's avatar
unknown committed
3624
    DBUG_RETURN(queue_old_event(mi,buf,event_len));
3625

3626
  LINT_INIT(inc_pos);
3627
  pthread_mutex_lock(&mi->data_lock);
unknown's avatar
unknown committed
3628

unknown's avatar
unknown committed
3629
  switch (buf[EVENT_TYPE_OFFSET]) {
3630
  case STOP_EVENT:
3631 3632
    /*
      We needn't write this event to the relay log. Indeed, it just indicates a
unknown's avatar
unknown committed
3633 3634 3635
      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
3636
      prepared statements' deletion are TODO only when we binlog prep stmts).
unknown's avatar
unknown committed
3637

unknown's avatar
unknown committed
3638 3639 3640 3641
      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).
3642 3643
    */
    goto err;
3644 3645
  case ROTATE_EVENT:
  {
unknown's avatar
unknown committed
3646
    Rotate_log_event rev(buf,event_len,mi->rli.relay_log.description_event_for_queue);
3647
    if (unlikely(process_io_rotate(mi,&rev)))
unknown's avatar
unknown committed
3648
    {
3649 3650
      error= 1;
      goto err;
unknown's avatar
unknown committed
3651
    }
3652 3653 3654 3655
    /*
      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
3656
    inc_pos= 0;
3657 3658
    break;
  }
3659 3660 3661 3662 3663 3664 3665
  case FORMAT_DESCRIPTION_EVENT:
  {
    /*
      Create an event, and save it (when we rotate the relay log, we will have
      to write this event again).
    */
    /*
3666 3667 3668
      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).
3669
    */
3670
    Format_description_log_event* tmp;
3671
    const char* errmsg;
3672
    if (!(tmp= (Format_description_log_event*)
3673
          Log_event::read_log_event(buf, event_len, &errmsg,
3674
                                    mi->rli.relay_log.description_event_for_queue)))
3675 3676 3677 3678
    {
      error= 2;
      goto err;
    }
3679 3680
    delete mi->rli.relay_log.description_event_for_queue;
    mi->rli.relay_log.description_event_for_queue= tmp;
unknown's avatar
unknown committed
3681
    /*
3682
       Though this does some conversion to the slave's format, this will
unknown's avatar
unknown committed
3683
       preserve the master's binlog format version, and number of event types.
3684
    */
unknown's avatar
unknown committed
3685
    /*
3686
       If the event was not requested by the slave (the slave did not ask for
unknown's avatar
unknown committed
3687
       it), i.e. has end_log_pos=0, we do not increment mi->master_log_pos
3688 3689 3690
    */
    inc_pos= uint4korr(buf+LOG_POS_OFFSET) ? event_len : 0;
    DBUG_PRINT("info",("binlog format is now %d",
unknown's avatar
unknown committed
3691
                       mi->rli.relay_log.description_event_for_queue->binlog_version));
3692 3693 3694

  }
  break;
3695
  default:
unknown's avatar
unknown committed
3696
    inc_pos= event_len;
3697 3698
    break;
  }
3699

unknown's avatar
unknown committed
3700 3701
  /*
     If this event is originating from this server, don't queue it.
3702
     We don't check this for 3.23 events because it's simpler like this; 3.23
unknown's avatar
unknown committed
3703 3704
     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
3705 3706 3707 3708 3709 3710 3711 3712
     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!).
  */

3713 3714
  pthread_mutex_lock(log_lock);

unknown's avatar
unknown committed
3715
  if ((uint4korr(buf + SERVER_ID_OFFSET) == ::server_id) &&
3716
      !mi->rli.replicate_same_server_id)
3717
  {
3718 3719
    /*
      Do not write it to the relay log.
3720 3721 3722 3723 3724 3725 3726 3727
      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.
3728 3729 3730
      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.
3731
    */
3732 3733 3734
    if (buf[EVENT_TYPE_OFFSET]!=FORMAT_DESCRIPTION_EVENT &&
        buf[EVENT_TYPE_OFFSET]!=ROTATE_EVENT &&
        buf[EVENT_TYPE_OFFSET]!=STOP_EVENT)
unknown's avatar
unknown committed
3735
    {
3736
      mi->master_log_pos+= inc_pos;
unknown's avatar
unknown committed
3737 3738 3739 3740
      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;
    }
3741
    rli->relay_log.signal_update(); // the slave SQL thread needs to re-check
unknown's avatar
unknown committed
3742
    DBUG_PRINT("info", ("master_log_pos: %lu, event originating from the same server, ignored",
unknown's avatar
unknown committed
3743
                        (ulong) mi->master_log_pos));
unknown's avatar
unknown committed
3744
  }
unknown's avatar
unknown committed
3745 3746 3747
  else
  {
    /* write the event to the relay log */
3748
    if (likely(!(rli->relay_log.appendv(buf,event_len,0))))
3749 3750
    {
      mi->master_log_pos+= inc_pos;
unknown's avatar
unknown committed
3751
      DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
3752 3753
      rli->relay_log.harvest_bytes_written(&rli->log_space_total);
    }
3754
    else
3755
      error= 3;
3756
    rli->ign_master_log_name_end[0]= 0; // last event is not ignored
unknown's avatar
unknown committed
3757
  }
3758
  pthread_mutex_unlock(log_lock);
3759 3760 3761


err:
3762
  pthread_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
3763
  DBUG_PRINT("info", ("error: %d", error));
unknown's avatar
unknown committed
3764
  DBUG_RETURN(error);
3765 3766
}

3767

3768
void end_relay_log_info(Relay_log_info* rli)
3769
{
3770 3771
  DBUG_ENTER("end_relay_log_info");

3772
  if (!rli->inited)
3773
    DBUG_VOID_RETURN;
3774
  if (rli->info_fd >= 0)
unknown's avatar
unknown committed
3775 3776
  {
    end_io_cache(&rli->info_file);
3777
    (void) my_close(rli->info_fd, MYF(MY_WME));
unknown's avatar
unknown committed
3778 3779
    rli->info_fd = -1;
  }
3780
  if (rli->cur_log_fd >= 0)
unknown's avatar
unknown committed
3781 3782 3783 3784 3785
  {
    end_io_cache(&rli->cache_buf);
    (void)my_close(rli->cur_log_fd, MYF(MY_WME));
    rli->cur_log_fd = -1;
  }
3786
  rli->inited = 0;
3787
  rli->relay_log.close(LOG_CLOSE_INDEX | LOG_CLOSE_STOP_EVENT);
3788
  rli->relay_log.harvest_bytes_written(&rli->log_space_total);
unknown's avatar
unknown committed
3789 3790 3791 3792 3793 3794
  /*
    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();
3795
  DBUG_VOID_RETURN;
3796 3797
}

3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816

/**
  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;
3817
  if (thd && thd->slave_thread)
3818 3819 3820 3821 3822
    thd->clear_active_vio();
#endif
}


unknown's avatar
unknown committed
3823 3824
/*
  Try to connect until successful or slave killed
3825

unknown's avatar
unknown committed
3826 3827
  SYNPOSIS
    safe_connect()
unknown's avatar
unknown committed
3828 3829 3830
    thd                 Thread handler for slave
    mysql               MySQL connection handle
    mi                  Replication handle
3831

unknown's avatar
unknown committed
3832
  RETURN
unknown's avatar
unknown committed
3833 3834
    0   ok
    #   Error
unknown's avatar
unknown committed
3835
*/
3836

3837
static int safe_connect(THD* thd, MYSQL* mysql, Master_info* mi)
unknown's avatar
unknown committed
3838
{
3839 3840 3841
  DBUG_ENTER("safe_connect");

  DBUG_RETURN(connect_to_master(thd, mysql, mi, 0, 0));
unknown's avatar
unknown committed
3842 3843
}

3844

3845
/*
unknown's avatar
unknown committed
3846 3847
  SYNPOSIS
    connect_to_master()
unknown's avatar
unknown committed
3848

unknown's avatar
unknown committed
3849 3850 3851
  IMPLEMENTATION
    Try to connect until successful or slave killed or we have retried
    master_retry_count times
3852
*/
unknown's avatar
unknown committed
3853

3854
static int connect_to_master(THD* thd, MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
3855
                             bool reconnect, bool suppress_warnings)
unknown's avatar
unknown committed
3856
{
3857
  int slave_was_killed;
unknown's avatar
unknown committed
3858
  int last_errno= -2;                           // impossible error
3859
  ulong err_count=0;
unknown's avatar
unknown committed
3860
  char llbuff[22];
3861
  DBUG_ENTER("connect_to_master");
unknown's avatar
unknown committed
3862

unknown's avatar
unknown committed
3863
#ifndef DBUG_OFF
3864
  mi->events_till_disconnect = disconnect_slave_event_count;
unknown's avatar
unknown committed
3865
#endif
3866
  ulong client_flag= CLIENT_REMEMBER_OPTIONS;
3867
  if (opt_slave_compressed_protocol)
unknown's avatar
unknown committed
3868
    client_flag=CLIENT_COMPRESS;                /* We will use compression */
3869

3870 3871
  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
3872

unknown's avatar
unknown committed
3873 3874
#ifdef HAVE_OPENSSL
  if (mi->ssl)
3875
  {
unknown's avatar
unknown committed
3876
    mysql_ssl_set(mysql,
unknown's avatar
unknown committed
3877
                  mi->ssl_key[0]?mi->ssl_key:0,
unknown's avatar
unknown committed
3878
                  mi->ssl_cert[0]?mi->ssl_cert:0,
unknown's avatar
unknown committed
3879 3880 3881
                  mi->ssl_ca[0]?mi->ssl_ca:0,
                  mi->ssl_capath[0]?mi->ssl_capath:0,
                  mi->ssl_cipher[0]?mi->ssl_cipher:0);
3882 3883 3884
    mysql_options(mysql, MYSQL_OPT_SSL_VERIFY_SERVER_CERT,
                  &mi->ssl_verify_server_cert);
  }
unknown's avatar
unknown committed
3885 3886
#endif

3887 3888 3889 3890
  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);

3891
  while (!(slave_was_killed = io_slave_killed(thd,mi)) &&
unknown's avatar
unknown committed
3892 3893 3894
         (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
3895
  {
3896
    /* Don't repeat last error */
unknown's avatar
SCRUM  
unknown committed
3897
    if ((int)mysql_errno(mysql) != last_errno)
3898
    {
unknown's avatar
SCRUM  
unknown committed
3899
      last_errno=mysql_errno(mysql);
unknown's avatar
unknown committed
3900
      suppress_warnings= 0;
3901 3902 3903 3904 3905 3906
      mi->report(ERROR_LEVEL, last_errno,
                 "error %s to master '%s@%s:%d'"
                 " - retry-time: %d  retries: %lu",
                 (reconnect ? "reconnecting" : "connecting"),
                 mi->user, mi->host, mi->port,
                 mi->connect_retry, master_retry_count);
3907
    }
unknown's avatar
unknown committed
3908 3909 3910
    /*
      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
3911
      do not want to have election triggered on the first failure to
unknown's avatar
unknown committed
3912
      connect
3913
    */
3914
    if (++err_count == master_retry_count)
3915 3916
    {
      slave_was_killed=1;
unknown's avatar
unknown committed
3917 3918
      if (reconnect)
        change_rpl_status(RPL_ACTIVE_SLAVE,RPL_LOST_SOLDIER);
3919 3920
      break;
    }
3921
    safe_sleep(thd,mi->connect_retry,(CHECK_KILLED_FUNC)io_slave_killed,
unknown's avatar
unknown committed
3922
               (void*)mi);
unknown's avatar
unknown committed
3923
  }
3924

3925 3926
  if (!slave_was_killed)
  {
3927
    mi->clear_error(); // clear possible left over reconnect error
unknown's avatar
unknown committed
3928
    if (reconnect)
unknown's avatar
unknown committed
3929
    {
3930
      if (!suppress_warnings && global_system_variables.log_warnings)
unknown's avatar
unknown committed
3931
        sql_print_information("Slave: connected to master '%s@%s:%d',\
3932
replication resumed in log '%s' at position %s", mi->user,
unknown's avatar
unknown committed
3933 3934 3935
                        mi->host, mi->port,
                        IO_RPL_LOG_NAME,
                        llstr(mi->master_log_pos,llbuff));
unknown's avatar
unknown committed
3936
    }
unknown's avatar
unknown committed
3937 3938 3939
    else
    {
      change_rpl_status(RPL_IDLE_SLAVE,RPL_ACTIVE_SLAVE);
3940 3941
      general_log_print(thd, COM_CONNECT_OUT, "%s@%s:%d",
                        mi->user, mi->host, mi->port);
unknown's avatar
unknown committed
3942
    }
3943
#ifdef SIGNAL_WITH_VIO_CLOSE
3944
    thd->set_active_vio(mysql->net.vio);
unknown's avatar
unknown committed
3945
#endif
3946
  }
3947
  mysql->reconnect= 1;
3948 3949
  DBUG_PRINT("exit",("slave_was_killed: %d", slave_was_killed));
  DBUG_RETURN(slave_was_killed);
unknown's avatar
unknown committed
3950 3951
}

3952

unknown's avatar
unknown committed
3953
/*
3954
  safe_reconnect()
unknown's avatar
unknown committed
3955

unknown's avatar
unknown committed
3956 3957 3958
  IMPLEMENTATION
    Try to connect until successful or slave killed or we have retried
    master_retry_count times
unknown's avatar
unknown committed
3959 3960
*/

3961
static int safe_reconnect(THD* thd, MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
3962
                          bool suppress_warnings)
unknown's avatar
unknown committed
3963
{
unknown's avatar
unknown committed
3964 3965
  DBUG_ENTER("safe_reconnect");
  DBUG_RETURN(connect_to_master(thd, mysql, mi, 1, suppress_warnings));
unknown's avatar
unknown committed
3966 3967
}

unknown's avatar
unknown committed
3968

3969 3970 3971 3972 3973 3974
/*
  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
3975
    rli                 Relay log information
3976 3977 3978 3979 3980 3981 3982 3983 3984 3985 3986 3987 3988 3989 3990 3991 3992 3993

  NOTES
    - As this is only called by the slave thread, we don't need to
      have a lock on this.
    - 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
3994 3995
    0   ok
    1   write error
3996 3997
*/

3998
bool flush_relay_log_info(Relay_log_info* rli)
3999
{
4000
  bool error=0;
4001
  DBUG_ENTER("flush_relay_log_info");
4002 4003

  if (unlikely(rli->no_storage))
4004
    DBUG_RETURN(0);
4005

4006 4007 4008
  IO_CACHE *file = &rli->info_file;
  char buff[FN_REFLEN*2+22*2+4], *pos;

4009
  my_b_seek(file, 0L);
4010
  pos=strmov(buff, rli->group_relay_log_name);
4011
  *pos++='\n';
4012
  pos=longlong2str(rli->group_relay_log_pos, pos, 10);
4013
  *pos++='\n';
4014
  pos=strmov(pos, rli->group_master_log_name);
4015
  *pos++='\n';
4016
  pos=longlong2str(rli->group_master_log_pos, pos, 10);
4017
  *pos='\n';
4018
  if (my_b_write(file, (uchar*) buff, (size_t) (pos-buff)+1))
4019 4020 4021
    error=1;
  if (flush_io_cache(file))
    error=1;
4022

4023
  /* Flushing the relay log is done by the slave I/O thread */
4024
  DBUG_RETURN(error);
4025 4026
}

4027

unknown's avatar
unknown committed
4028
/*
4029
  Called when we notice that the current "hot" log got rotated under our feet.
unknown's avatar
unknown committed
4030 4031
*/

4032
static IO_CACHE *reopen_relay_log(Relay_log_info *rli, const char **errmsg)
4033
{
4034
  DBUG_ENTER("reopen_relay_log");
4035 4036
  DBUG_ASSERT(rli->cur_log != &rli->cache_buf);
  DBUG_ASSERT(rli->cur_log_fd == -1);
unknown's avatar
unknown committed
4037 4038

  IO_CACHE *cur_log = rli->cur_log=&rli->cache_buf;
4039
  if ((rli->cur_log_fd=open_binlog(cur_log,rli->event_relay_log_name,
unknown's avatar
unknown committed
4040
                                   errmsg)) <0)
unknown's avatar
unknown committed
4041
    DBUG_RETURN(0);
4042 4043
  /*
    We want to start exactly where we was before:
unknown's avatar
unknown committed
4044 4045
    relay_log_pos       Current log pos
    pending             Number of bytes already processed from the event
4046
  */
4047
  rli->event_relay_log_pos= max(rli->event_relay_log_pos, BIN_LOG_HEADER_SIZE);
4048
  my_b_seek(cur_log,rli->event_relay_log_pos);
unknown's avatar
unknown committed
4049
  DBUG_RETURN(cur_log);
4050 4051
}

unknown's avatar
unknown committed
4052

4053 4054 4055 4056 4057 4058 4059 4060 4061 4062
/**
  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.
*/
4063
static Log_event* next_event(Relay_log_info* rli)
4064 4065 4066
{
  Log_event* ev;
  IO_CACHE* cur_log = rli->cur_log;
4067
  pthread_mutex_t *log_lock = rli->relay_log.get_log_lock();
4068 4069
  const char* errmsg=0;
  THD* thd = rli->sql_thd;
unknown's avatar
unknown committed
4070
  DBUG_ENTER("next_event");
4071

4072 4073
  DBUG_ASSERT(thd != 0);

4074 4075 4076 4077 4078
#ifndef DBUG_OFF
  if (abort_slave_event_count && !rli->events_till_abort--)
    DBUG_RETURN(0);
#endif

unknown's avatar
unknown committed
4079 4080
  /*
    For most operations we need to protect rli members with data_lock,
4081 4082 4083 4084
    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
    pthread_cond_wait() with the non-data_lock mutex
unknown's avatar
unknown committed
4085
  */
4086
  safe_mutex_assert_owner(&rli->data_lock);
unknown's avatar
unknown committed
4087

4088
  while (!sql_slave_killed(thd,rli))
unknown's avatar
unknown committed
4089 4090 4091
  {
    /*
      We can have two kinds of log reading:
unknown's avatar
unknown committed
4092 4093 4094 4095 4096 4097 4098 4099
      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
4100
    */
4101 4102 4103 4104 4105
    bool hot_log;
    if ((hot_log = (cur_log != &rli->cache_buf)))
    {
      DBUG_ASSERT(rli->cur_log_fd == -1); // foreign descriptor
      pthread_mutex_lock(log_lock);
unknown's avatar
unknown committed
4106 4107

      /*
unknown's avatar
unknown committed
4108 4109
        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
4110
      */
unknown's avatar
unknown committed
4111
      if (rli->relay_log.get_open_count() != rli->cur_log_old_open_count)
4112
      {
unknown's avatar
unknown committed
4113 4114 4115 4116 4117 4118
        // The master has switched to a new log file; Reopen the old log file
        cur_log=reopen_relay_log(rli, &errmsg);
        pthread_mutex_unlock(log_lock);
        if (!cur_log)                           // No more log files
          goto err;
        hot_log=0;                              // Using old binary log
4119 4120
      }
    }
4121 4122 4123 4124 4125 4126 4127
    /* 
      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;
4128 4129
#ifndef DBUG_OFF
    {
4130
      /* This is an assertion which sometimes fails, let's try to track it */
4131
      char llbuf1[22], llbuf2[22];
4132
      DBUG_PRINT("info", ("my_b_tell(cur_log)=%s rli->event_relay_log_pos=%s",
4133
                          llstr(my_b_tell(cur_log),llbuf1),
4134 4135 4136
                          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);
4137 4138
    }
#endif
unknown's avatar
unknown committed
4139 4140
    /*
      Relay log is always in new format - if the master is 3.23, the
4141
      I/O thread will convert the format for us.
4142 4143
      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
4144 4145
      logs, which may even have been deleted. So we need to write this
      description event at the beginning of the relay log.
4146 4147
      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.
4148
      But if the relay log is created by new_file(): then the solution is:
4149
      MYSQL_BIN_LOG::open() will write the buffered description event.
unknown's avatar
unknown committed
4150
    */
4151 4152
    if ((ev=Log_event::read_log_event(cur_log,0,
                                      rli->relay_log.description_event_for_exec)))
4153

4154 4155
    {
      DBUG_ASSERT(thd==rli->sql_thd);
4156 4157 4158 4159 4160
      /*
        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);
4161
      if (hot_log)
unknown's avatar
unknown committed
4162
        pthread_mutex_unlock(log_lock);
unknown's avatar
unknown committed
4163
      DBUG_RETURN(ev);
4164 4165
    }
    DBUG_ASSERT(thd==rli->sql_thd);
unknown's avatar
unknown committed
4166
    if (opt_reckless_slave)                     // For mysql-test
unknown's avatar
unknown committed
4167
      cur_log->error = 0;
unknown's avatar
unknown committed
4168
    if (cur_log->error < 0)
unknown's avatar
unknown committed
4169 4170
    {
      errmsg = "slave SQL thread aborted because of I/O error";
unknown's avatar
unknown committed
4171
      if (hot_log)
unknown's avatar
unknown committed
4172
        pthread_mutex_unlock(log_lock);
unknown's avatar
unknown committed
4173 4174
      goto err;
    }
4175 4176
    if (!cur_log->error) /* EOF */
    {
unknown's avatar
unknown committed
4177
      /*
unknown's avatar
unknown committed
4178 4179 4180
        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
4181
      */
4182 4183
      if (hot_log)
      {
4184 4185 4186 4187 4188 4189 4190 4191 4192 4193
        /*
          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").
4194 4195 4196 4197 4198 4199 4200 4201 4202 4203

          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.
4204
        */
4205
        time_t save_timestamp= rli->last_master_timestamp;
4206 4207
        rli->last_master_timestamp= 0;

unknown's avatar
unknown committed
4208
        DBUG_ASSERT(rli->relay_log.get_open_count() ==
unknown's avatar
unknown committed
4209
                    rli->cur_log_old_open_count);
4210 4211 4212 4213 4214

        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"));
4215
          ev= new Rotate_log_event(rli->ign_master_log_name_end,
4216
                                   0, rli->ign_master_log_pos_end,
unknown's avatar
unknown committed
4217
                                   Rotate_log_event::DUP_NAME);
4218
          rli->ign_master_log_name_end[0]= 0;
unknown's avatar
unknown committed
4219
          pthread_mutex_unlock(log_lock);
4220 4221 4222 4223 4224 4225 4226 4227 4228 4229
          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
4230 4231 4232 4233 4234
        /*
          We can, and should release data_lock while we are waiting for
          update. If we do not, show slave status will block
        */
        pthread_mutex_unlock(&rli->data_lock);
4235 4236

        /*
unknown's avatar
unknown committed
4237
          Possible deadlock :
4238 4239 4240 4241 4242 4243 4244 4245 4246 4247 4248
          - 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
4249
          master does not send anything). Then the I/O thread stops waiting
4250 4251
          and reads more events.
          The SQL thread decides when the I/O thread should take log_space_limit
unknown's avatar
unknown committed
4252
          into account again : ignore_log_space_limit is reset to 0
4253 4254 4255 4256
          in purge_first_log (when the SQL thread purges the just-read relay
          log), and also when the SQL thread starts. 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
4257 4258
          be stopped, and the SQL thread sets ignore_log_space_limit to 0 when
          it stops.
4259 4260 4261
        */
        pthread_mutex_lock(&rli->log_space_lock);
        // prevent the I/O thread from blocking next times
unknown's avatar
unknown committed
4262
        rli->ignore_log_space_limit= 1;
4263
        /*
4264 4265 4266 4267
          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.
4268
        */
4269
        pthread_mutex_unlock(&rli->log_space_lock);
4270
        pthread_cond_broadcast(&rli->log_space_cond);
4271
        // Note that wait_for_update unlocks lock_log !
4272
        rli->relay_log.wait_for_update(rli->sql_thd, 1);
4273 4274
        // re-acquire data lock since we released it earlier
        pthread_mutex_lock(&rli->data_lock);
4275
        rli->last_master_timestamp= save_timestamp;
unknown's avatar
unknown committed
4276
        continue;
4277
      }
unknown's avatar
unknown committed
4278
      /*
unknown's avatar
unknown committed
4279 4280 4281
        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
4282 4283 4284 4285 4286
      */
      end_io_cache(cur_log);
      DBUG_ASSERT(rli->cur_log_fd >= 0);
      my_close(rli->cur_log_fd, MYF(MY_WME));
      rli->cur_log_fd = -1;
unknown's avatar
unknown committed
4287

4288
      if (relay_log_purge)
unknown's avatar
unknown committed
4289
      {
unknown's avatar
unknown committed
4290
        /*
4291 4292 4293 4294 4295 4296 4297 4298 4299 4300
          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
4301
        if (rli->relay_log.purge_first_log
4302 4303 4304
            (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
4305 4306 4307 4308
        {
          errmsg = "Error purging processed logs";
          goto err;
        }
unknown's avatar
unknown committed
4309
      }
4310 4311
      else
      {
unknown's avatar
unknown committed
4312 4313 4314 4315 4316 4317 4318 4319 4320 4321 4322 4323 4324 4325 4326 4327
        /*
          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
4328
      }
4329 4330 4331 4332 4333 4334 4335 4336 4337 4338 4339 4340 4341 4342

      /*
        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 */
        pthread_mutex_lock(log_lock);
unknown's avatar
unknown committed
4343 4344
      if (rli->relay_log.is_active(rli->linfo.log_file_name))
      {
4345
#ifdef EXTRA_DEBUG
unknown's avatar
unknown committed
4346 4347
        if (global_system_variables.log_warnings)
          sql_print_information("next log '%s' is currently active",
4348
                                rli->linfo.log_file_name);
unknown's avatar
unknown committed
4349 4350 4351 4352 4353 4354 4355 4356
#endif
        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);

        /*
          Read pointer has to be at the start since we are the only
          reader.
4357 4358 4359
          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).
unknown's avatar
unknown committed
4360 4361
        */
        if (check_binlog_magic(cur_log,&errmsg))
4362 4363
        {
          if (!hot_log) pthread_mutex_unlock(log_lock);
unknown's avatar
unknown committed
4364
          goto err;
4365 4366
        }
        if (!hot_log) pthread_mutex_unlock(log_lock);
unknown's avatar
unknown committed
4367
        continue;
4368
      }
4369
      if (!hot_log) pthread_mutex_unlock(log_lock);
unknown's avatar
unknown committed
4370
      /*
unknown's avatar
unknown committed
4371 4372 4373
        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
4374 4375
      */
#ifdef EXTRA_DEBUG
unknown's avatar
unknown committed
4376
      if (global_system_variables.log_warnings)
unknown's avatar
unknown committed
4377
        sql_print_information("next log '%s' is not active",
4378
                              rli->linfo.log_file_name);
unknown's avatar
unknown committed
4379
#endif
unknown's avatar
unknown committed
4380 4381
      // 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
4382 4383
                                       &errmsg)) <0)
        goto err;
4384
    }
unknown's avatar
unknown committed
4385
    else
4386
    {
unknown's avatar
unknown committed
4387
      /*
unknown's avatar
unknown committed
4388 4389
        Read failed with a non-EOF error.
        TODO: come up with something better to handle this error
unknown's avatar
unknown committed
4390 4391
      */
      if (hot_log)
unknown's avatar
unknown committed
4392
        pthread_mutex_unlock(log_lock);
4393
      sql_print_error("Slave SQL thread: I/O error reading \
unknown's avatar
unknown committed
4394
event(errno: %d  cur_log->error: %d)",
unknown's avatar
unknown committed
4395
                      my_errno,cur_log->error);
unknown's avatar
unknown committed
4396
      // set read position to the beginning of the event
4397
      my_b_seek(cur_log,rli->event_relay_log_pos);
unknown's avatar
unknown committed
4398 4399
      /* otherwise, we have had a partial read */
      errmsg = "Aborting slave SQL thread because of partial event read";
unknown's avatar
unknown committed
4400
      break;                                    // To end of function
4401 4402
    }
  }
4403
  if (!errmsg && global_system_variables.log_warnings)
4404
  {
unknown's avatar
unknown committed
4405
    sql_print_information("Error reading relay log event: %s",
4406 4407 4408
                          "slave SQL thread was killed");
    DBUG_RETURN(0);
  }
unknown's avatar
unknown committed
4409

4410
err:
4411 4412
  if (errmsg)
    sql_print_error("Error reading relay log event: %s", errmsg);
unknown's avatar
unknown committed
4413
  DBUG_RETURN(0);
4414 4415
}

4416 4417 4418
/*
  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
4419
  locks; here we don't, so this function is mainly taking locks).
4420 4421
  Returns nothing as we cannot catch any error (MYSQL_BIN_LOG::new_file()
  is void).
4422 4423
*/

4424
void rotate_relay_log(Master_info* mi)
4425 4426
{
  DBUG_ENTER("rotate_relay_log");
4427
  Relay_log_info* rli= &mi->rli;
unknown's avatar
unknown committed
4428

4429 4430 4431
  /* We don't lock rli->run_lock. This would lead to deadlocks. */
  pthread_mutex_lock(&mi->run_lock);

unknown's avatar
unknown committed
4432
  /*
unknown's avatar
unknown committed
4433 4434 4435
     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).
  */
4436 4437
  if (!rli->inited)
  {
unknown's avatar
unknown committed
4438
    DBUG_PRINT("info", ("rli->inited == 0"));
unknown's avatar
unknown committed
4439
    goto end;
4440
  }
unknown's avatar
unknown committed
4441

4442
  /* If the relay log is closed, new_file() will do nothing. */
4443
  rli->relay_log.new_file();
unknown's avatar
unknown committed
4444

4445 4446 4447 4448
  /*
    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
4449 4450 4451 4452 4453 4454
    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.
4455 4456 4457 4458
    If the log is closed, then this will just harvest the last writes, probably
    0 as they probably have been harvested.
  */
  rli->relay_log.harvest_bytes_written(&rli->log_space_total);
unknown's avatar
unknown committed
4459
end:
4460
  pthread_mutex_unlock(&mi->run_lock);
4461 4462 4463
  DBUG_VOID_RETURN;
}

unknown's avatar
unknown committed
4464

unknown's avatar
unknown committed
4465 4466 4467
/**
   Detects, based on master's version (as found in the relay log), if master
   has a certain bug.
4468
   @param rli Relay_log_info which tells the master's version
unknown's avatar
unknown committed
4469
   @param bug_id Number of the bug as found in bugs.mysql.com
4470
   @param report bool report error message, default TRUE
4471 4472 4473 4474 4475 4476 4477

   @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
4478 4479
   @return TRUE if master has the bug, FALSE if it does not.
*/
4480 4481
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
4482 4483 4484 4485 4486 4487 4488 4489
{
  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
4490
    {24432, { 5, 0, 24 }, { 5, 0, 38 } },
4491 4492 4493
    {24432, { 5, 1, 12 }, { 5, 1, 17 } },
    {33029, { 5, 0,  0 }, { 5, 0, 58 } },
    {33029, { 5, 1,  0 }, { 5, 1, 12 } },
4494
    {37426, { 5, 1,  0 }, { 5, 1, 26 } },
unknown's avatar
unknown committed
4495 4496 4497 4498 4499 4500 4501 4502 4503 4504 4505 4506 4507
  };
  const uchar *master_ver=
    rli->relay_log.description_event_for_exec->server_version_split;

  DBUG_ASSERT(sizeof(rli->relay_log.description_event_for_exec->server_version_split) == 3);

  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) &&
4508 4509
        (memcmp(fixed_in,      master_ver, 3) >  0) &&
        (pred == NULL || (*pred)(param)))
unknown's avatar
unknown committed
4510
    {
4511 4512
      if (!report)
	return TRUE;
unknown's avatar
unknown committed
4513 4514 4515 4516 4517
      // 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);
4518
      // a verbose message for the error log
4519 4520 4521
      rli->report(ERROR_LEVEL, ER_UNKNOWN_ERROR,
                  "According to the master's version ('%s'),"
                  " it is probable that master suffers from this bug:"
4522 4523 4524 4525 4526 4527 4528 4529 4530 4531 4532 4533 4534 4535 4536 4537
                      " 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
4538 4539 4540 4541 4542 4543
      return TRUE;
    }
  }
  return FALSE;
}

4544 4545 4546 4547 4548 4549 4550 4551 4552 4553 4554 4555 4556 4557 4558
/**
   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)
{
  if (active_mi && active_mi->rli.sql_thd == thd)
  {
    Relay_log_info *rli= &active_mi->rli;
4559
    DBUG_EXECUTE_IF("simulate_bug33029", return TRUE;);
4560
    return rpl_master_has_bug(rli, 33029, FALSE, NULL, NULL);
4561 4562 4563 4564
  }
  return FALSE;
}

4565
#ifdef HAVE_EXPLICIT_TEMPLATE_INSTANTIATION
unknown's avatar
unknown committed
4566
template class I_List_iterator<i_string>;
unknown's avatar
unknown committed
4567
template class I_List_iterator<i_string_pair>;
unknown's avatar
unknown committed
4568
#endif
4569

4570 4571 4572 4573
/**
  @} (end of group Replication)
*/

unknown's avatar
SCRUM  
unknown committed
4574
#endif /* HAVE_REPLICATION */