slave.cc 153 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 47
#ifdef HAVE_REPLICATION

#include "rpl_tblmap.h"

48
#define FLAGSTR(V,F) ((V)&(F)?#F" ":"")
49

50
#define MAX_SLAVE_RETRY_PAUSE 5
51 52
bool use_slave_mask = 0;
MY_BITMAP slave_error_mask;
53
char slave_skip_error_names[SHOW_VAR_FUNC_BUFF_SIZE];
54

55 56
typedef bool (*CHECK_KILLED_FUNC)(THD*,void*);

57
char* slave_load_tmpdir = 0;
58
Master_info *active_mi= 0;
59
my_bool replicate_same_server_id;
60
ulonglong relay_log_space_limit = 0;
unknown's avatar
unknown committed
61 62 63 64 65

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

66
  TODO: move the vars below under Master_info
unknown's avatar
unknown committed
67
*/
68

69
int disconnect_slave_event_count = 0, abort_slave_event_count = 0;
70
int events_till_abort = -1;
unknown's avatar
unknown committed
71

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
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"
  }
};
 

124
typedef enum { SLAVE_THD_IO, SLAVE_THD_SQL} SLAVE_THD_TYPE;
unknown's avatar
unknown committed
125

126 127
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);
128
static bool wait_for_relay_log_space(Relay_log_info* rli);
129
static inline bool io_slave_killed(THD* thd,Master_info* mi);
130
static inline bool sql_slave_killed(THD* thd,Relay_log_info* rli);
131
static int init_slave_thread(THD* thd, SLAVE_THD_TYPE thd_type);
132
static void print_slave_skip_errors(void);
133 134
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
135
                          bool suppress_warnings);
136
static int connect_to_master(THD* thd, MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
137
                             bool reconnect, bool suppress_warnings);
138
static int safe_sleep(THD* thd, int sec, CHECK_KILLED_FUNC thread_killed,
unknown's avatar
unknown committed
139
                      void* thread_killed_arg);
140
static int request_table_dump(MYSQL* mysql, const char* db, const char* table);
141
static int create_table_from_dump(THD* thd, MYSQL *mysql, const char* db,
unknown's avatar
unknown committed
142
                                  const char* table_name, bool overwrite);
143
static int get_master_version_and_clock(MYSQL* mysql, Master_info* mi);
144
static Log_event* next_event(Relay_log_info* rli);
145
static int queue_event(Master_info* mi,const char* buf,ulong event_len);
146
static int terminate_slave_thread(THD *thd,
147 148
                                  pthread_mutex_t *term_lock,
                                  pthread_cond_t *term_cond,
149 150
                                  volatile uint *slave_running,
                                  bool skip_lock);
151
static bool check_io_slave_killed(THD *thd, Master_info *mi, const char *info);
152 153

/*
unknown's avatar
unknown committed
154
  Find out which replications threads are running
155

unknown's avatar
unknown committed
156 157
  SYNOPSIS
    init_thread_mask()
unknown's avatar
unknown committed
158 159 160
    mask                Return value here
    mi                  master_info for slave
    inverse             If set, returns which threads are not running
unknown's avatar
unknown committed
161 162 163 164 165 166

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

  RETURN
unknown's avatar
unknown committed
167 168
    mask        If inverse == 0, running threads
                If inverse == 1, stopped threads
169 170
*/

171
void init_thread_mask(int* mask,Master_info* mi,bool inverse)
172 173 174
{
  bool set_io = mi->slave_running, set_sql = mi->rli.slave_running;
  register int tmp_mask=0;
175 176
  DBUG_ENTER("init_thread_mask");

177 178 179 180
  if (set_io)
    tmp_mask |= SLAVE_IO;
  if (set_sql)
    tmp_mask |= SLAVE_SQL;
181 182
  if (inverse)
    tmp_mask^= (SLAVE_IO | SLAVE_SQL);
183
  *mask = tmp_mask;
184
  DBUG_VOID_RETURN;
185 186
}

187

unknown's avatar
unknown committed
188
/*
189
  lock_slave_threads()
unknown's avatar
unknown committed
190
*/
191

192
void lock_slave_threads(Master_info* mi)
193
{
194 195
  DBUG_ENTER("lock_slave_threads");

196 197 198
  //TODO: see if we can do this without dual mutex
  pthread_mutex_lock(&mi->run_lock);
  pthread_mutex_lock(&mi->rli.run_lock);
199
  DBUG_VOID_RETURN;
200 201
}

202

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

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

211 212 213
  //TODO: see if we can do this without dual mutex
  pthread_mutex_unlock(&mi->rli.run_lock);
  pthread_mutex_unlock(&mi->run_lock);
214
  DBUG_VOID_RETURN;
215 216
}

217

unknown's avatar
unknown committed
218
/* Initialize slave structures */
219

220 221
int init_slave()
{
222
  DBUG_ENTER("init_slave");
223

224 225 226 227 228 229
  /*
    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);
230 231 232 233
  /*
    TODO: re-write this to interate through the list of files
    for multi-master
  */
234
  active_mi= new Master_info;
235

236 237 238 239 240 241 242 243 244
  /*
    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();
  }

245
  /*
246 247 248
    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.
249
  */
250 251 252 253 254
  if (!active_mi)
  {
    sql_print_error("Failed to allocate memory for the master info structure");
    goto err;
  }
255

unknown's avatar
unknown committed
256
  if (init_master_info(active_mi,master_info_file,relay_log_info_file,
unknown's avatar
unknown committed
257
                       !master_host, (SLAVE_IO | SLAVE_SQL)))
258
  {
259
    sql_print_error("Failed to initialize the master info structure");
unknown's avatar
unknown committed
260
    goto err;
261
  }
262 263 264 265

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

266 267
  /* If server id is not set, start_slave_thread() will say it */

268
  if (master_host && !opt_skip_slave_start)
269
  {
270
    if (start_slave_threads(1 /* need mutex */,
unknown's avatar
unknown committed
271 272 273 274 275
                            0 /* no wait for start*/,
                            active_mi,
                            master_info_file,
                            relay_log_info_file,
                            SLAVE_IO | SLAVE_SQL))
unknown's avatar
unknown committed
276
    {
277
      sql_print_error("Failed to create slave threads");
unknown's avatar
unknown committed
278 279
      goto err;
    }
280
  }
281
  pthread_mutex_unlock(&LOCK_active_mi);
282
  DBUG_RETURN(0);
283

unknown's avatar
unknown committed
284
err:
285
  pthread_mutex_unlock(&LOCK_active_mi);
unknown's avatar
unknown committed
286
  DBUG_RETURN(1);
287
}
288

289

290 291 292 293 294 295 296 297 298 299 300 301
/**
  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.
  */
302
  const size_t MIN_ROOM= 10;
303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324
  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;

325
    for (errnum= 0; errnum < MAX_SLAVE_ERROR; errnum++)
326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347
    {
      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
348
/*
349
  Init function to set up array for errors that should be skipped for slave
unknown's avatar
unknown committed
350

unknown's avatar
unknown committed
351 352
  SYNOPSIS
    init_slave_skip_errors()
unknown's avatar
unknown committed
353
    arg         List of errors numbers to skip, separated with ','
unknown's avatar
unknown committed
354 355 356 357

  NOTES
    Called from get_options() in mysqld.cc on start-up
*/
358

unknown's avatar
unknown committed
359
void init_slave_skip_errors(const char* arg)
360
{
unknown's avatar
unknown committed
361
  const char *p;
362 363
  DBUG_ENTER("init_slave_skip_errors");

364
  if (bitmap_init(&slave_error_mask,0,MAX_SLAVE_ERROR,0))
365 366 367 368 369
  {
    fprintf(stderr, "Badly out of memory, please check your system status\n");
    exit(1);
  }
  use_slave_mask = 1;
370
  for (;my_isspace(system_charset_info,*arg);++arg)
371
    /* empty */;
372
  if (!my_strnncoll(system_charset_info,(uchar*)arg,4,(const uchar*)"all",4))
373 374
  {
    bitmap_set_all(&slave_error_mask);
375
    print_slave_skip_errors();
376
    DBUG_VOID_RETURN;
377 378 379 380 381 382 383 384
  }
  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);
385
    while (!my_isdigit(system_charset_info,*p) && *p)
386 387
      p++;
  }
388 389
  /* Convert slave skip errors bitmap into a printable string. */
  print_slave_skip_errors();
390
  DBUG_VOID_RETURN;
391 392
}

393
static void set_thd_in_use_temporary_tables(Relay_log_info *rli)
394 395 396 397 398 399
{
  TABLE *table;

  for (table= rli->save_temporary_tables ; table ; table= table->next)
    table->in_use= rli->sql_thd;
}
400

401
int terminate_slave_threads(Master_info* mi,int thread_mask,bool skip_lock)
402
{
403 404
  DBUG_ENTER("terminate_slave_threads");

405
  if (!mi->inited)
406
    DBUG_RETURN(0); /* successfully do nothing */
407 408
  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
409

410
  if (thread_mask & (SLAVE_IO|SLAVE_FORCE_ALL))
411
  {
unknown's avatar
unknown committed
412
    DBUG_PRINT("info",("Terminating IO thread"));
413
    mi->abort_slave=1;
414
    if ((error=terminate_slave_thread(mi->io_thd, io_lock,
unknown's avatar
unknown committed
415
                                      &mi->stop_cond,
416 417
                                      &mi->slave_running,
                                      skip_lock)) &&
unknown's avatar
unknown committed
418
        !force_all)
419
      DBUG_RETURN(error);
420
  }
421
  if (thread_mask & (SLAVE_SQL|SLAVE_FORCE_ALL))
422
  {
unknown's avatar
unknown committed
423
    DBUG_PRINT("info",("Terminating SQL thread"));
424
    mi->rli.abort_slave=1;
425
    if ((error=terminate_slave_thread(mi->rli.sql_thd, sql_lock,
unknown's avatar
unknown committed
426
                                      &mi->rli.stop_cond,
427 428
                                      &mi->rli.slave_running,
                                      skip_lock)) &&
unknown's avatar
unknown committed
429
        !force_all)
430
      DBUG_RETURN(error);
431
  }
432
  DBUG_RETURN(0);
433 434
}

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
/**
   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
463
   @retval 0 All OK ER_SLAVE_NOT_RUNNING otherwise.
464

Andrei Elkin's avatar
Andrei Elkin committed
465 466 467
   @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.
468

469 470 471
 */
static int
terminate_slave_thread(THD *thd,
Andrei Elkin's avatar
Andrei Elkin committed
472 473
                       pthread_mutex_t *term_lock,
                       pthread_cond_t *term_cond,
474 475
                       volatile uint *slave_running,
                       bool skip_lock)
476
{
477
  DBUG_ENTER("terminate_slave_thread");
478
  if (!skip_lock)
479 480
  {
    pthread_mutex_lock(term_lock);
481 482 483 484 485
  }
  else
  {
    safe_mutex_assert_owner(term_lock);
  }
486 487 488
  if (!*slave_running)
  {
    if (!skip_lock)
489
    {
490 491 492 493
      /*
        if run_lock (term_lock) is acquired locally then either
        slave_running status is fine
      */
494
      pthread_mutex_unlock(term_lock);
495 496 497 498
      DBUG_RETURN(0);
    }
    else
    {
499
      DBUG_RETURN(ER_SLAVE_NOT_RUNNING);
500 501 502
    }
  }
  DBUG_ASSERT(thd != 0);
503
  THD_CHECK_SENTRY(thd);
504

unknown's avatar
unknown committed
505
  /*
506
    Is is critical to test if the slave is running. Otherwise, we might
unknown's avatar
unknown committed
507
    be referening freed memory trying to kick it
508
  */
unknown's avatar
unknown committed
509

unknown's avatar
unknown committed
510
  while (*slave_running)                        // Should always be true
511
  {
512
    int error;
513
    DBUG_PRINT("loop", ("killing slave thread"));
514

515
    pthread_mutex_lock(&thd->LOCK_thd_data);
516 517 518 519 520 521 522 523 524 525
#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);
526
    pthread_mutex_unlock(&thd->LOCK_thd_data);
527

unknown's avatar
unknown committed
528 529 530
    /*
      There is a small chance that slave thread might miss the first
      alarm. To protect againts it, resend the signal until it reacts
531 532
    */
    struct timespec abstime;
unknown's avatar
unknown committed
533
    set_timespec(abstime,2);
534 535
    error= pthread_cond_timedwait(term_cond, term_lock, &abstime);
    DBUG_ASSERT(error == ETIMEDOUT || error == 0);
536
  }
537 538 539 540

  DBUG_ASSERT(*slave_running == 0);

  if (!skip_lock)
541
    pthread_mutex_unlock(term_lock);
542
  DBUG_RETURN(0);
543 544
}

545

unknown's avatar
unknown committed
546
int start_slave_thread(pthread_handler h_func, pthread_mutex_t *start_lock,
unknown's avatar
unknown committed
547 548 549 550
                       pthread_mutex_t *cond_lock,
                       pthread_cond_t *start_cond,
                       volatile uint *slave_running,
                       volatile ulong *slave_run_id,
551
                       Master_info* mi,
552
                       bool high_priority)
553 554
{
  pthread_t th;
unknown's avatar
unknown committed
555 556 557
  ulong start_id;
  DBUG_ENTER("start_slave_thread");

558 559
  DBUG_ASSERT(mi->inited);

560 561 562 563 564 565 566 567 568
  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
569
    DBUG_RETURN(ER_BAD_SLAVE);
570
  }
unknown's avatar
unknown committed
571

572
  if (*slave_running)
573 574 575 576 577
  {
    if (start_cond)
      pthread_cond_broadcast(start_cond);
    if (start_lock)
      pthread_mutex_unlock(start_lock);
unknown's avatar
unknown committed
578
    DBUG_RETURN(ER_SLAVE_MUST_STOP);
579
  }
unknown's avatar
unknown committed
580 581
  start_id= *slave_run_id;
  DBUG_PRINT("info",("Creating new slave thread"));
582 583
  if (high_priority)
    my_pthread_attr_setprio(&connection_attrib,CONNECT_PRIOR);
584 585 586 587
  if (pthread_create(&th, &connection_attrib, h_func, (void*)mi))
  {
    if (start_lock)
      pthread_mutex_unlock(start_lock);
unknown's avatar
unknown committed
588
    DBUG_RETURN(ER_SLAVE_THREAD);
589
  }
unknown's avatar
unknown committed
590
  if (start_cond && cond_lock) // caller has cond_lock
591 592
  {
    THD* thd = current_thd;
unknown's avatar
unknown committed
593
    while (start_id == *slave_run_id)
594
    {
unknown's avatar
unknown committed
595
      DBUG_PRINT("sleep",("Waiting for slave thread to start"));
596
      const char* old_msg = thd->enter_cond(start_cond,cond_lock,
unknown's avatar
unknown committed
597
                                            "Waiting for slave thread to start");
598 599
      pthread_cond_wait(start_cond,cond_lock);
      thd->exit_cond(old_msg);
unknown's avatar
unknown committed
600
      pthread_mutex_lock(cond_lock); // re-acquire it as exit_cond() released
601
      if (thd->killed)
unknown's avatar
unknown committed
602
        DBUG_RETURN(thd->killed_errno());
603 604 605 606
    }
  }
  if (start_lock)
    pthread_mutex_unlock(start_lock);
unknown's avatar
unknown committed
607
  DBUG_RETURN(0);
608
}
unknown's avatar
unknown committed
609

610

unknown's avatar
unknown committed
611
/*
612
  start_slave_threads()
unknown's avatar
unknown committed
613

unknown's avatar
unknown committed
614 615 616 617
  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
618
*/
unknown's avatar
unknown committed
619

620
int start_slave_threads(bool need_slave_mutex, bool wait_for_start,
621
                        Master_info* mi, const char* master_info_fname,
unknown's avatar
unknown committed
622
                        const char* slave_info_fname, int thread_mask)
623 624 625 626
{
  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;
627
  DBUG_ENTER("start_slave_threads");
unknown's avatar
unknown committed
628

629 630 631 632 633 634 635 636 637 638 639 640
  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;
  }
641 642 643

  if (thread_mask & SLAVE_IO)
    error=start_slave_thread(handle_slave_io,lock_io,lock_cond_io,
unknown's avatar
unknown committed
644 645 646
                             cond_io,
                             &mi->slave_running, &mi->slave_run_id,
                             mi, 1); //high priority, to read the most possible
647
  if (!error && (thread_mask & SLAVE_SQL))
648
  {
649
    error=start_slave_thread(handle_slave_sql,lock_sql,lock_cond_sql,
unknown's avatar
unknown committed
650 651 652
                             cond_sql,
                             &mi->rli.slave_running, &mi->rli.slave_run_id,
                             mi, 0);
653
    if (error)
654
      terminate_slave_threads(mi, thread_mask & SLAVE_IO, !need_slave_mutex);
655
  }
656
  DBUG_RETURN(error);
657
}
658

659

660
#ifdef NOT_USED_YET
661
static int end_slave_on_walk(Master_info* mi, uchar* /*unused*/)
662
{
663 664
  DBUG_ENTER("end_slave_on_walk");

665
  end_master_info(mi);
666
  DBUG_RETURN(0);
667
}
668
#endif
669

670

unknown's avatar
unknown committed
671
/*
672
  Release slave threads at time of executing shutdown.
unknown's avatar
unknown committed
673 674 675 676

  SYNOPSIS
    end_slave()
*/
677

678 679
void end_slave()
{
680 681
  DBUG_ENTER("end_slave");

682 683 684 685 686 687 688 689
  /*
    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
690 691 692 693 694 695 696 697
  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);
698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716
  }
  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
717 718 719 720
    end_master_info(active_mi);
    delete active_mi;
    active_mi= 0;
  }
721
  pthread_mutex_unlock(&LOCK_active_mi);
722
}
unknown's avatar
unknown committed
723

724
static bool io_slave_killed(THD* thd, Master_info* mi)
unknown's avatar
unknown committed
725
{
726 727
  DBUG_ENTER("io_slave_killed");

728
  DBUG_ASSERT(mi->io_thd == thd);
729
  DBUG_ASSERT(mi->slave_running); // tracking buffer overrun
730
  DBUG_RETURN(mi->abort_slave || abort_loop || thd->killed);
unknown's avatar
unknown committed
731 732
}

733

734
static bool sql_slave_killed(THD* thd, Relay_log_info* rli)
735
{
736 737
  DBUG_ENTER("sql_slave_killed");

738 739
  DBUG_ASSERT(rli->sql_thd == thd);
  DBUG_ASSERT(rli->slave_running == 1);// tracking buffer overrun
740 741
  if (abort_loop || thd->killed || rli->abort_slave)
  {
742 743 744
    if (rli->abort_slave && rli->is_in_group() &&
        thd->transaction.all.modified_non_trans_table)
      DBUG_RETURN(0);
745 746 747 748 749 750 751 752 753
    /*
      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.
    */
754
    if (rli->last_event_start_time == 0)
755
      DBUG_RETURN(1);
756 757
    DBUG_PRINT("info", ("Slave SQL thread is in an unsafe situation, giving "
                        "it some grace period"));
758
    if (difftime(time(0), rli->last_event_start_time) > 60)
759
    {
760 761 762 763 764 765 766
      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.");
767
      DBUG_RETURN(1);
768 769
    }
  }
770
  DBUG_RETURN(0);
771 772
}

unknown's avatar
unknown committed
773
/*
774 775
  skip_load_data_infile()

unknown's avatar
unknown committed
776 777 778
  NOTES
    This is used to tell a 3.23 master to break send_file()
*/
779 780 781

void skip_load_data_infile(NET *net)
{
782 783
  DBUG_ENTER("skip_load_data_infile");

784
  (void)net_request_file(net, "/dev/null");
unknown's avatar
unknown committed
785
  (void)my_net_read(net);                               // discard response
786
  (void)net_write_command(net, 0, (uchar*) "", 0, (uchar*) "", 0); // ok
787
  DBUG_VOID_RETURN;
788 789
}

790

791
bool net_request_file(NET* net, const char* fname)
792
{
793
  DBUG_ENTER("net_request_file");
794 795
  DBUG_RETURN(net_write_command(net, 251, (uchar*) fname, strlen(fname),
                                (uchar*) "", 0));
796 797
}

798 799
/*
  From other comments and tests in code, it looks like
800
  sometimes Query_log_event and Load_log_event can have db == 0
801 802 803
  (see rewrite_db() above for example)
  (cases where this happens are unclear; it may be when the master is 3.23).
*/
804 805

const char *print_slave_db_safe(const char* db)
806
{
807 808 809
  DBUG_ENTER("*print_slave_db_safe");

  DBUG_RETURN((db ? db : ""));
810
}
811

812
int init_strvar_from_file(char *var, int max_size, IO_CACHE *f,
unknown's avatar
unknown committed
813
                                 const char *default_val)
unknown's avatar
unknown committed
814
{
unknown's avatar
unknown committed
815
  uint length;
816 817
  DBUG_ENTER("init_strvar_from_file");

unknown's avatar
unknown committed
818 819 820 821 822 823
  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
824
    {
unknown's avatar
unknown committed
825
      /*
unknown's avatar
unknown committed
826 827
        If we truncated a line or stopped on last char, remove all chars
        up to and including newline.
unknown's avatar
unknown committed
828
      */
unknown's avatar
unknown committed
829
      int c;
830
      while (((c=my_b_get(f)) != '\n' && c != my_b_EOF)) ;
unknown's avatar
unknown committed
831
    }
832
    DBUG_RETURN(0);
unknown's avatar
unknown committed
833 834 835
  }
  else if (default_val)
  {
unknown's avatar
unknown committed
836
    strmake(var,  default_val, max_size-1);
837
    DBUG_RETURN(0);
unknown's avatar
unknown committed
838
  }
839
  DBUG_RETURN(1);
unknown's avatar
unknown committed
840 841
}

842

843
int init_intvar_from_file(int* var, IO_CACHE* f, int default_val)
unknown's avatar
unknown committed
844 845
{
  char buf[32];
846 847
  DBUG_ENTER("init_intvar_from_file");

unknown's avatar
unknown committed
848 849

  if (my_b_gets(f, buf, sizeof(buf)))
unknown's avatar
unknown committed
850 851
  {
    *var = atoi(buf);
852
    DBUG_RETURN(0);
unknown's avatar
unknown committed
853
  }
unknown's avatar
unknown committed
854
  else if (default_val)
unknown's avatar
unknown committed
855 856
  {
    *var = default_val;
857
    DBUG_RETURN(0);
unknown's avatar
unknown committed
858
  }
859
  DBUG_RETURN(1);
unknown's avatar
unknown committed
860 861
}

Andrei Elkin's avatar
Andrei Elkin committed
862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892
int init_floatvar_from_file(float* var, IO_CACHE* f, float default_val)
{
  char buf[16];
  DBUG_ENTER("init_floatvar_from_file");


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

static bool check_io_slave_killed(THD *thd, Master_info *mi, const char *info)
{
  if (io_slave_killed(thd, mi))
  {
    if (info && global_system_variables.log_warnings)
      sql_print_information(info);
    return TRUE;
  }
  return FALSE;
}
893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915

/*
  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;   
}


916 917 918 919 920 921 922 923
/*
  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
924

925 926 927
  RETURNS
  0       ok
  1       error
928
  2       transient network problem, the caller should try to reconnect
929
*/
930

931
static int get_master_version_and_clock(MYSQL* mysql, Master_info* mi)
932
{
933
  char err_buff[MAX_SLAVE_ERRMSG];
934
  const char* errmsg= 0;
935 936 937
  int err_code= 0;
  MYSQL_RES *master_res= 0;
  MYSQL_ROW master_row;
938
  DBUG_ENTER("get_master_version_and_clock");
939 940 941 942 943 944 945

  /*
    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
946

947
  if (!my_isdigit(&my_charset_bin,*mysql->server_version))
948
  {
unknown's avatar
unknown committed
949
    errmsg = "Master reported unrecognized MySQL version";
950 951 952
    err_code= ER_SLAVE_FATAL_ERROR;
    sprintf(err_buff, ER(err_code), errmsg);
  }
953 954 955 956 957
  else
  {
    /*
      Note the following switch will bug when we have MySQL branch 30 ;)
    */
unknown's avatar
unknown committed
958
    switch (*mysql->server_version)
959 960 961 962 963
    {
    case '0':
    case '1':
    case '2':
      errmsg = "Master reported unrecognized MySQL version";
964 965
      err_code= ER_SLAVE_FATAL_ERROR;
      sprintf(err_buff, ER(err_code), errmsg);
966 967 968
      break;
    case '3':
      mi->rli.relay_log.description_event_for_queue= new
unknown's avatar
unknown committed
969
        Format_description_log_event(1, mysql->server_version);
970 971 972
      break;
    case '4':
      mi->rli.relay_log.description_event_for_queue= new
unknown's avatar
unknown committed
973
        Format_description_log_event(3, mysql->server_version);
974
      break;
unknown's avatar
unknown committed
975
    default:
976 977 978 979 980 981 982 983 984
      /*
        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
985
        Format_description_log_event(4, mysql->server_version);
986 987
      break;
    }
988
  }
unknown's avatar
unknown committed
989 990

  /*
991 992 993 994 995
     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
996

997
  if (errmsg)
998
    goto err;
999 1000 1001 1002

  /* as we are here, we tried to allocate the event */
  if (!mi->rli.relay_log.description_event_for_queue)
  {
1003 1004 1005 1006
    errmsg= "default Format_description_log_event";
    err_code= ER_SLAVE_CREATE_EVENT_FAILURE;
    sprintf(err_buff, ER(err_code), errmsg);
    goto err;
1007 1008
  }

1009 1010 1011 1012
  /*
    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
1013

1014 1015
  DBUG_SYNC_POINT("debug_lock.before_get_UNIX_TIMESTAMP", 10);
  master_res= NULL;
1016
  if (!mysql_real_query(mysql, STRING_WITH_LEN("SELECT UNIX_TIMESTAMP()")) &&
1017 1018
      (master_res= mysql_store_result(mysql)) &&
      (master_row= mysql_fetch_row(master_res)))
unknown's avatar
unknown committed
1019
  {
unknown's avatar
unknown committed
1020
    mi->clock_diff_with_master=
1021
      (long) (time((time_t*) 0) - strtoul(master_row[0], 0, 10));
unknown's avatar
unknown committed
1022
  }
1023 1024 1025 1026 1027 1028 1029
  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
1030
  {
1031
    mi->clock_diff_with_master= 0; /* The "most sensible" value */
1032 1033 1034 1035
    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));
1036 1037
  }
  if (master_res)
1038
  {
unknown's avatar
unknown committed
1039
    mysql_free_result(master_res);
1040 1041
    master_res= NULL;
  }
unknown's avatar
unknown committed
1042

1043 1044 1045 1046 1047 1048 1049 1050 1051 1052
  /*
    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.
  */
1053 1054 1055
  DBUG_SYNC_POINT("debug_lock.before_get_SERVER_ID", 10);
  master_res= NULL;
  master_row= NULL;
1056 1057
  if (!mysql_real_query(mysql,
                        STRING_WITH_LEN("SHOW VARIABLES LIKE 'SERVER_ID'")) &&
1058 1059
      (master_res= mysql_store_result(mysql)) &&
      (master_row= mysql_fetch_row(master_res)))
1060
  {
1061
    if ((::server_id == strtoul(master_row[1], 0, 10)) &&
1062
        !mi->rli.replicate_same_server_id)
1063
    {
1064 1065 1066 1067
      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).";
1068 1069
      err_code= ER_SLAVE_FATAL_ERROR;
      sprintf(err_buff, ER(err_code), errmsg);
1070
      goto err;
1071
    }
1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095
  }
  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)
  {
1096
    mysql_free_result(master_res);
1097
    master_res= NULL;
unknown's avatar
unknown committed
1098 1099
  }

1100 1101 1102
  /*
    Check that the master's global character_set_server and ours are the same.
    Not fatal if query fails (old master?).
1103 1104 1105 1106 1107 1108
    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.
1109 1110 1111 1112 1113 1114
    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.
1115
  */
1116 1117 1118

  /* redundant with rest of code but safer against later additions */
  if (*mysql->server_version == '3')
1119
    goto err;
1120

1121
  if (*mysql->server_version == '4')
unknown's avatar
unknown committed
1122
  {
1123 1124 1125 1126 1127
    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)))
1128
    {
1129 1130 1131
      if (strcmp(master_row[0], global_system_variables.collation_server->name))
      {
        errmsg= "The slave I/O thread stops because master and slave have \
1132
different values for the COLLATION_SERVER global variable. The values must \
1133 1134 1135 1136 1137
be equal for the Statement-format replication to work";
        err_code= ER_SLAVE_FATAL_ERROR;
        sprintf(err_buff, ER(err_code), errmsg);
        goto err;
      }
1138
    }
1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151
    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));
1152
      goto err;
1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164
    }
    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
1165
  }
1166

1167 1168 1169 1170 1171 1172 1173 1174
  /*
    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
1175 1176 1177
    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
1178
    'SYSTEM' time zone).
1179 1180
    This check is only necessary for 4.x masters (and < 5.0.4 masters but
    those were alpha).
1181
  */
1182
  if (*mysql->server_version == '4')
unknown's avatar
unknown committed
1183
  {
1184 1185 1186 1187
    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)))
1188
    {
1189 1190 1191 1192
      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 \
1193
different values for the TIME_ZONE global variable. The values must \
1194 1195 1196 1197 1198
be equal for the Statement-format replication to work";
        err_code= ER_SLAVE_FATAL_ERROR;
        sprintf(err_buff, ER(err_code), errmsg);
        goto err;
      }
1199
    }
1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212
    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));
1213
      goto err;
1214 1215 1216 1217 1218 1219
    }
    if (master_res)
    {
      mysql_free_result(master_res);
      master_res= NULL;
    }
unknown's avatar
unknown committed
1220 1221
  }

Andrei Elkin's avatar
Andrei Elkin committed
1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235
  if (mi->heartbeat_period != 0.0)
  {
    char llbuf[22];
    const char query_format[]= "SET @master_heartbeat_period= %s";
    char query[sizeof(query_format) - 2 + sizeof(llbuf)];
    /* 
       the period is an ulonglong of nano-secs. 
    */
    llstr((ulonglong) (mi->heartbeat_period*1000000000UL), llbuf);
    my_sprintf(query, (query, query_format, llbuf));

    if (mysql_real_query(mysql, query, strlen(query))
        && !check_io_slave_killed(mi->io_thd, mi, NULL))
    {
Andrei Elkin's avatar
Andrei Elkin committed
1236 1237
      errmsg= "The slave I/O thread stops because SET @master_heartbeat_period "
        "on master failed.";
Andrei Elkin's avatar
Andrei Elkin committed
1238
      err_code= ER_SLAVE_FATAL_ERROR;
Andrei Elkin's avatar
Andrei Elkin committed
1239
      sprintf(err_buff, "%s Error: %s", errmsg, mysql_error(mysql));
Andrei Elkin's avatar
Andrei Elkin committed
1240 1241 1242 1243 1244 1245 1246
      mysql_free_result(mysql_store_result(mysql));
      goto err;
    }
    mysql_free_result(mysql_store_result(mysql));
  }
 

1247
err:
1248 1249
  if (errmsg)
  {
1250 1251
    if (master_res)
      mysql_free_result(master_res);
1252 1253
    DBUG_ASSERT(err_code != 0);
    mi->report(ERROR_LEVEL, err_code, err_buff);
1254
    DBUG_RETURN(1);
1255
  }
1256

1257
  DBUG_RETURN(0);
1258 1259 1260 1261 1262

network_err:
  if (master_res)
    mysql_free_result(master_res);
  DBUG_RETURN(2);
1263 1264
}

1265 1266 1267 1268
/*
  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.
1269
  db must be non-zero (guarded by assertion).
1270 1271 1272 1273 1274

  RETURN VALUES
    0           success
    1           error
*/
1275

1276
static int create_table_from_dump(THD* thd, MYSQL *mysql, const char* db,
unknown's avatar
unknown committed
1277
                                  const char* table_name, bool overwrite)
unknown's avatar
unknown committed
1278
{
1279
  ulong packet_len;
1280 1281
  char *query, *save_db;
  uint32 save_db_length;
1282 1283
  Vio* save_vio;
  HA_CHECK_OPT check_opt;
unknown's avatar
unknown committed
1284
  TABLE_LIST tables;
1285 1286
  int error= 1;
  handler *file;
1287
  ulonglong save_options;
1288
  NET *net= &mysql->net;
1289
  const char *found_semicolon= NULL;
1290
  DBUG_ENTER("create_table_from_dump");
unknown's avatar
unknown committed
1291

1292
  packet_len= my_net_read(net); // read create table statement
1293 1294
  if (packet_len == packet_error)
  {
unknown's avatar
unknown committed
1295
    my_message(ER_MASTER_NET_READ, ER(ER_MASTER_NET_READ), MYF(0));
unknown's avatar
unknown committed
1296
    DBUG_RETURN(1);
1297 1298 1299
  }
  if (net->read_pos[0] == 255) // error from master
  {
unknown's avatar
unknown committed
1300
    char *err_msg;
1301
    err_msg= (char*) net->read_pos + ((mysql->server_capabilities &
unknown's avatar
unknown committed
1302 1303
                                       CLIENT_PROTOCOL_41) ?
                                      3+SQLSTATE_LENGTH+1 : 3);
1304
    my_error(ER_MASTER, MYF(0), err_msg);
unknown's avatar
unknown committed
1305
    DBUG_RETURN(1);
1306
  }
unknown's avatar
unknown committed
1307
  thd->command = COM_TABLE_DUMP;
1308
  if (!(query = thd->strmake((char*) net->read_pos, packet_len)))
1309 1310
  {
    sql_print_error("create_table_from_dump: out of memory");
unknown's avatar
unknown committed
1311
    my_message(ER_GET_ERRNO, "Out of memory", MYF(0));
unknown's avatar
unknown committed
1312
    DBUG_RETURN(1);
1313
  }
1314
  thd->set_query(query, packet_len);
1315
  thd->is_slave_error = 0;
1316

unknown's avatar
unknown committed
1317 1318
  bzero((char*) &tables,sizeof(tables));
  tables.db = (char*)db;
1319
  tables.alias= tables.table_name= (char*)table_name;
unknown's avatar
unknown committed
1320

unknown's avatar
unknown committed
1321
  /* Drop the table if 'overwrite' is true */
1322
  if (overwrite)
unknown's avatar
unknown committed
1323
  {
1324 1325 1326 1327 1328 1329 1330 1331 1332 1333
    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
1334 1335
  }

1336
  /* Create the table. We do not want to log the "create table" statement */
1337
  save_options = thd->options;
1338 1339
  thd->options &= ~ (OPTION_BIN_LOG);
  thd_proc_info(thd, "Creating table from master dump");
unknown's avatar
unknown committed
1340
  // save old db in case we are creating in a different database
1341
  save_db = thd->db;
1342
  save_db_length= thd->db_length;
1343
  thd->db = (char*)db;
1344
  DBUG_ASSERT(thd->db != 0);
1345
  thd->db_length= strlen(thd->db);
1346
  mysql_parse(thd, thd->query, packet_len, &found_semicolon); // run create table
unknown's avatar
unknown committed
1347
  thd->db = save_db;            // leave things the way the were before
1348
  thd->db_length= save_db_length;
1349
  thd->options = save_options;
unknown's avatar
unknown committed
1350

1351
  if (thd->is_slave_error)
unknown's avatar
unknown committed
1352
    goto err;                   // mysql_parse took care of the error send
unknown's avatar
unknown committed
1353

1354
  thd_proc_info(thd, "Opening master dump table");
1355
  thd->main_da.reset_diagnostics_area(); /* cleanup from CREATE_TABLE */
1356 1357 1358 1359 1360 1361
  /*
    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))
  */
1362
  tables.lock_type = TL_WRITE;
1363
  if (!open_ltable(thd, &tables, TL_WRITE, 0))
unknown's avatar
unknown committed
1364 1365
  {
    sql_print_error("create_table_from_dump: could not open created table");
1366
    goto err;
unknown's avatar
unknown committed
1367
  }
unknown's avatar
unknown committed
1368

1369
  file = tables.table->file;
1370
  thd_proc_info(thd, "Reading master dump table data");
1371
  /* Copy the data file */
unknown's avatar
unknown committed
1372 1373
  if (file->net_read_dump(net))
  {
unknown's avatar
unknown committed
1374
    my_message(ER_MASTER_NET_READ, ER(ER_MASTER_NET_READ), MYF(0));
1375
    sql_print_error("create_table_from_dump: failed in\
unknown's avatar
unknown committed
1376
 handler::net_read_dump()");
1377
    goto err;
unknown's avatar
unknown committed
1378
  }
unknown's avatar
unknown committed
1379 1380

  check_opt.init();
unknown's avatar
unknown committed
1381
  check_opt.flags|= T_VERY_SILENT | T_CALC_CHECKSUM | T_QUICK;
1382
  thd_proc_info(thd, "Rebuilding the index on master dump table");
unknown's avatar
unknown committed
1383 1384 1385 1386 1387
  /*
    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.
  */
1388
  save_vio = thd->net.vio;
unknown's avatar
unknown committed
1389
  thd->net.vio = 0;
1390
  /* Rebuild the index file from the copied data file (with REPAIR) */
unknown's avatar
unknown committed
1391
  error=file->ha_repair(thd,&check_opt) != 0;
unknown's avatar
unknown committed
1392
  thd->net.vio = save_vio;
1393
  if (error)
unknown's avatar
unknown committed
1394
    my_error(ER_INDEX_REBUILD, MYF(0), tables.table->s->table_name.str);
1395 1396

err:
unknown's avatar
unknown committed
1397
  close_thread_tables(thd);
unknown's avatar
unknown committed
1398
  DBUG_RETURN(error);
unknown's avatar
unknown committed
1399 1400
}

1401

1402
int fetch_master_table(THD *thd, const char *db_name, const char *table_name,
1403
                       Master_info *mi, MYSQL *mysql, bool overwrite)
unknown's avatar
unknown committed
1404
{
1405 1406 1407 1408 1409
  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
1410
                       db_name,table_name));
unknown's avatar
unknown committed
1411

unknown's avatar
merge  
unknown committed
1412
  if (!called_connected)
unknown's avatar
unknown committed
1413
  {
unknown's avatar
SCRUM  
unknown committed
1414
    if (!(mysql = mysql_init(NULL)))
1415 1416 1417
    {
      DBUG_RETURN(1);
    }
unknown's avatar
merge  
unknown committed
1418
    if (connect_to_master(thd, mysql, mi))
1419
    {
1420
      my_error(ER_CONNECT_TO_MASTER, MYF(0), mysql_error(mysql));
1421 1422 1423 1424 1425 1426 1427 1428 1429
      /*
        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
1430
      mysql_close(mysql);
1431
      DBUG_RETURN(1);
1432
    }
1433 1434
    if (thd->killed)
      goto err;
1435
  }
unknown's avatar
unknown committed
1436

unknown's avatar
unknown committed
1437
  if (request_table_dump(mysql, db_name, table_name))
1438
  {
1439 1440
    error= ER_UNKNOWN_ERROR;
    errmsg= "Failed on table dump request";
1441 1442
    goto err;
  }
unknown's avatar
unknown committed
1443
  if (create_table_from_dump(thd, mysql, db_name,
unknown's avatar
unknown committed
1444
                             table_name, overwrite))
unknown's avatar
unknown committed
1445
    goto err;    // create_table_from_dump have sent the error already
unknown's avatar
unknown committed
1446
  error = 0;
1447

unknown's avatar
unknown committed
1448
 err:
1449
  if (!called_connected)
unknown's avatar
SCRUM  
unknown committed
1450
    mysql_close(mysql);
1451
  if (errmsg && thd->vio_ok())
unknown's avatar
unknown committed
1452
    my_message(error, errmsg, MYF(0));
unknown's avatar
unknown committed
1453
  DBUG_RETURN(test(error));                     // Return 1 on error
unknown's avatar
unknown committed
1454
}
unknown's avatar
unknown committed
1455

1456

1457
static bool wait_for_relay_log_space(Relay_log_info* rli)
unknown's avatar
unknown committed
1458
{
1459
  bool slave_killed=0;
1460
  Master_info* mi = rli->mi;
unknown's avatar
unknown committed
1461
  const char *save_proc_info;
unknown's avatar
unknown committed
1462 1463
  THD* thd = mi->io_thd;
  DBUG_ENTER("wait_for_relay_log_space");
1464

unknown's avatar
unknown committed
1465
  pthread_mutex_lock(&rli->log_space_lock);
unknown's avatar
unknown committed
1466
  save_proc_info= thd->enter_cond(&rli->log_space_cond,
unknown's avatar
unknown committed
1467 1468
                                  &rli->log_space_lock,
                                  "\
1469
Waiting for the slave SQL thread to free enough relay log space");
unknown's avatar
unknown committed
1470
  while (rli->log_space_limit < rli->log_space_total &&
unknown's avatar
unknown committed
1471
         !(slave_killed=io_slave_killed(thd,mi)) &&
1472
         !rli->ignore_log_space_limit)
unknown's avatar
unknown committed
1473
    pthread_cond_wait(&rli->log_space_cond, &rli->log_space_lock);
unknown's avatar
unknown committed
1474
  thd->exit_cond(save_proc_info);
unknown's avatar
unknown committed
1475 1476
  DBUG_RETURN(slave_killed);
}
1477

1478

1479 1480
/*
  Builds a Rotate from the ignored events' info and writes it to relay log.
1481

1482 1483 1484 1485 1486 1487 1488 1489 1490 1491
  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).
 */
1492
static void write_ignored_events_info_to_relay_log(THD *thd, Master_info *mi)
1493
{
1494
  Relay_log_info *rli= &mi->rli;
1495
  pthread_mutex_t *log_lock= rli->relay_log.get_log_lock();
1496 1497
  DBUG_ENTER("write_ignored_events_info_to_relay_log");

1498 1499 1500
  DBUG_ASSERT(thd == mi->io_thd);
  pthread_mutex_lock(log_lock);
  if (rli->ign_master_log_name_end[0])
unknown's avatar
unknown committed
1501
  {
1502
    DBUG_PRINT("info",("writing a Rotate event to track down ignored events"));
1503
    Rotate_log_event *ev= new Rotate_log_event(rli->ign_master_log_name_end,
1504 1505 1506 1507 1508 1509 1510 1511 1512
                                               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)))
1513 1514 1515 1516 1517
        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");
1518
      rli->relay_log.harvest_bytes_written(&rli->log_space_total);
1519 1520
      if (flush_master_info(mi, 1))
        sql_print_error("Failed to flush master info file");
1521 1522 1523
      delete ev;
    }
    else
1524 1525 1526 1527
      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
1528
  }
1529 1530
  else
    pthread_mutex_unlock(log_lock);
1531
  DBUG_VOID_RETURN;
unknown's avatar
unknown committed
1532 1533
}

1534

1535
int register_slave_on_master(MYSQL* mysql, Master_info *mi,
1536
                             bool *suppress_warnings)
1537
{
1538
  uchar buf[1024], *pos= buf;
1539
  uint report_host_len, report_user_len=0, report_password_len=0;
1540
  DBUG_ENTER("register_slave_on_master");
1541

1542
  *suppress_warnings= FALSE;
1543
  if (!report_host)
1544
    DBUG_RETURN(0);
1545
  report_host_len= strlen(report_host);
1546
  if (report_user)
1547
    report_user_len= strlen(report_user);
unknown's avatar
unknown committed
1548
  if (report_password)
1549 1550 1551 1552
    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
1553
    DBUG_RETURN(0);                                     // safety
1554 1555

  int4store(pos, server_id); pos+= 4;
1556 1557 1558
  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);
1559
  int2store(pos, (uint16) report_port); pos+= 2;
unknown's avatar
unknown committed
1560
  int4store(pos, rpl_recovery_rank);    pos+= 4;
1561
  /* The master will fill in master_id */
unknown's avatar
unknown committed
1562
  int4store(pos, 0);                    pos+= 4;
1563

1564
  if (simple_command(mysql, COM_REGISTER_SLAVE, buf, (size_t) (pos- buf), 0))
1565
  {
1566 1567 1568 1569
    if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
    {
      *suppress_warnings= TRUE;                 // Suppress reconnect warning
    }
1570
    else if (!check_io_slave_killed(mi->io_thd, mi, NULL))
1571 1572 1573 1574 1575 1576 1577
    {
      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);
    }
1578
    DBUG_RETURN(1);
1579
  }
1580
  DBUG_RETURN(0);
1581 1582
}

1583

1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594
/**
  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
*/
1595
bool show_master_info(THD* thd, Master_info* mi)
unknown's avatar
unknown committed
1596
{
1597
  // TODO: fix this for multi-master
unknown's avatar
unknown committed
1598
  List<Item> field_list;
1599 1600 1601
  Protocol *protocol= thd->protocol;
  DBUG_ENTER("show_master_info");

unknown's avatar
unknown committed
1602
  field_list.push_back(new Item_empty_string("Slave_IO_State",
unknown's avatar
unknown committed
1603
                                                     14));
unknown's avatar
unknown committed
1604
  field_list.push_back(new Item_empty_string("Master_Host",
unknown's avatar
unknown committed
1605
                                                     sizeof(mi->host)));
unknown's avatar
unknown committed
1606
  field_list.push_back(new Item_empty_string("Master_User",
unknown's avatar
unknown committed
1607
                                                     sizeof(mi->user)));
1608
  field_list.push_back(new Item_return_int("Master_Port", 7,
unknown's avatar
unknown committed
1609
                                           MYSQL_TYPE_LONG));
1610
  field_list.push_back(new Item_return_int("Connect_Retry", 10,
unknown's avatar
unknown committed
1611
                                           MYSQL_TYPE_LONG));
1612
  field_list.push_back(new Item_empty_string("Master_Log_File",
unknown's avatar
unknown committed
1613
                                             FN_REFLEN));
1614
  field_list.push_back(new Item_return_int("Read_Master_Log_Pos", 10,
unknown's avatar
unknown committed
1615
                                           MYSQL_TYPE_LONGLONG));
1616
  field_list.push_back(new Item_empty_string("Relay_Log_File",
unknown's avatar
unknown committed
1617
                                             FN_REFLEN));
1618
  field_list.push_back(new Item_return_int("Relay_Log_Pos", 10,
unknown's avatar
unknown committed
1619
                                           MYSQL_TYPE_LONGLONG));
1620
  field_list.push_back(new Item_empty_string("Relay_Master_Log_File",
unknown's avatar
unknown committed
1621
                                             FN_REFLEN));
1622 1623
  field_list.push_back(new Item_empty_string("Slave_IO_Running", 3));
  field_list.push_back(new Item_empty_string("Slave_SQL_Running", 3));
1624 1625 1626 1627 1628 1629
  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
1630
                                             28));
1631 1632 1633
  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
1634
                                           MYSQL_TYPE_LONG));
1635
  field_list.push_back(new Item_return_int("Exec_Master_Log_Pos", 10,
unknown's avatar
unknown committed
1636
                                           MYSQL_TYPE_LONGLONG));
1637
  field_list.push_back(new Item_return_int("Relay_Log_Space", 10,
unknown's avatar
unknown committed
1638
                                           MYSQL_TYPE_LONGLONG));
1639
  field_list.push_back(new Item_empty_string("Until_Condition", 6));
1640
  field_list.push_back(new Item_empty_string("Until_Log_File", FN_REFLEN));
unknown's avatar
unknown committed
1641
  field_list.push_back(new Item_return_int("Until_Log_Pos", 10,
1642
                                           MYSQL_TYPE_LONGLONG));
unknown's avatar
unknown committed
1643 1644 1645
  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
1646
  field_list.push_back(new Item_empty_string("Master_SSL_CA_Path",
unknown's avatar
unknown committed
1647
                                             sizeof(mi->ssl_capath)));
unknown's avatar
unknown committed
1648
  field_list.push_back(new Item_empty_string("Master_SSL_Cert",
unknown's avatar
unknown committed
1649
                                             sizeof(mi->ssl_cert)));
unknown's avatar
unknown committed
1650
  field_list.push_back(new Item_empty_string("Master_SSL_Cipher",
unknown's avatar
unknown committed
1651
                                             sizeof(mi->ssl_cipher)));
unknown's avatar
unknown committed
1652
  field_list.push_back(new Item_empty_string("Master_SSL_Key",
unknown's avatar
unknown committed
1653
                                             sizeof(mi->ssl_key)));
1654
  field_list.push_back(new Item_return_int("Seconds_Behind_Master", 10,
unknown's avatar
unknown committed
1655
                                           MYSQL_TYPE_LONGLONG));
1656 1657
  field_list.push_back(new Item_empty_string("Master_SSL_Verify_Server_Cert",
                                             3));
1658 1659 1660 1661
  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
1662

1663 1664
  if (protocol->send_fields(&field_list,
                            Protocol::SEND_NUM_ROWS | Protocol::SEND_EOF))
unknown's avatar
unknown committed
1665
    DBUG_RETURN(TRUE);
unknown's avatar
unknown committed
1666

1667 1668
  if (mi->host[0])
  {
1669
    DBUG_PRINT("info",("host is set: '%s'", mi->host));
1670
    String *packet= &thd->packet;
1671
    protocol->prepare_for_resend();
unknown's avatar
unknown committed
1672

1673
    /*
1674 1675
      slave_running can be accessed without run_lock but not other
      non-volotile members like mi->io_thd, which is guarded by the mutex.
1676
    */
1677 1678 1679 1680
    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);

1681 1682
    pthread_mutex_lock(&mi->data_lock);
    pthread_mutex_lock(&mi->rli.data_lock);
1683 1684
    pthread_mutex_lock(&mi->err_lock);
    pthread_mutex_lock(&mi->rli.err_lock);
1685 1686
    protocol->store(mi->host, &my_charset_bin);
    protocol->store(mi->user, &my_charset_bin);
1687 1688
    protocol->store((uint32) mi->port);
    protocol->store((uint32) mi->connect_retry);
1689
    protocol->store(mi->master_log_name, &my_charset_bin);
1690
    protocol->store((ulonglong) mi->master_log_pos);
1691
    protocol->store(mi->rli.group_relay_log_name +
unknown's avatar
unknown committed
1692 1693
                    dirname_length(mi->rli.group_relay_log_name),
                    &my_charset_bin);
1694 1695
    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
1696 1697
    protocol->store(mi->slave_running == MYSQL_SLAVE_RUN_CONNECT ?
                    "Yes" : "No", &my_charset_bin);
1698
    protocol->store(mi->rli.slave_running ? "Yes":"No", &my_charset_bin);
1699 1700 1701
    protocol->store(rpl_filter->get_do_db());
    protocol->store(rpl_filter->get_ignore_db());

1702 1703
    char buf[256];
    String tmp(buf, sizeof(buf), &my_charset_bin);
1704
    rpl_filter->get_do_table(&tmp);
1705
    protocol->store(&tmp);
1706
    rpl_filter->get_ignore_table(&tmp);
1707
    protocol->store(&tmp);
1708
    rpl_filter->get_wild_do_table(&tmp);
1709
    protocol->store(&tmp);
1710
    rpl_filter->get_wild_ignore_table(&tmp);
1711 1712
    protocol->store(&tmp);

1713 1714
    protocol->store(mi->rli.last_error().number);
    protocol->store(mi->rli.last_error().message, &my_charset_bin);
1715
    protocol->store((uint32) mi->rli.slave_skip_counter);
1716
    protocol->store((ulonglong) mi->rli.group_master_log_pos);
1717
    protocol->store((ulonglong) mi->rli.log_space_total);
1718 1719

    protocol->store(
1720 1721
      mi->rli.until_condition==Relay_log_info::UNTIL_NONE ? "None":
        ( mi->rli.until_condition==Relay_log_info::UNTIL_MASTER_POS? "Master":
1722 1723 1724
          "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
1725 1726

#ifdef HAVE_OPENSSL
unknown's avatar
unknown committed
1727 1728 1729 1730 1731 1732 1733 1734 1735
    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
1736

1737 1738 1739 1740 1741 1742
    /*
      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)
1743
    {
1744
      long time_diff= ((long)(time(0) - mi->rli.last_master_timestamp)
unknown's avatar
unknown committed
1745
                       - mi->clock_diff_with_master);
1746
      /*
unknown's avatar
unknown committed
1747 1748
        Apparently on some systems time_diff can be <0. Here are possible
        reasons related to MySQL:
1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760
        - 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.
1761 1762 1763 1764
        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".
1765
      */
unknown's avatar
unknown committed
1766 1767
      protocol->store((longlong)(mi->rli.last_master_timestamp ?
                                 max(0, time_diff) : 0));
1768
    }
unknown's avatar
unknown committed
1769
    else
1770
    {
unknown's avatar
unknown committed
1771
      protocol->store_null();
1772 1773
    }
    protocol->store(mi->ssl_verify_server_cert? "Yes":"No", &my_charset_bin);
unknown's avatar
unknown committed
1774

1775
    // Last_IO_Errno
1776
    protocol->store(mi->last_error().number);
1777
    // Last_IO_Error
1778
    protocol->store(mi->last_error().message, &my_charset_bin);
1779
    // Last_SQL_Errno
1780
    protocol->store(mi->rli.last_error().number);
1781
    // Last_SQL_Error
1782
    protocol->store(mi->rli.last_error().message, &my_charset_bin);
1783

1784 1785
    pthread_mutex_unlock(&mi->rli.err_lock);
    pthread_mutex_unlock(&mi->err_lock);
1786 1787
    pthread_mutex_unlock(&mi->rli.data_lock);
    pthread_mutex_unlock(&mi->data_lock);
1788

1789
    if (my_net_write(&thd->net, (uchar*) thd->packet.ptr(), packet->length()))
unknown's avatar
unknown committed
1790
      DBUG_RETURN(TRUE);
1791
  }
1792
  my_eof(thd);
unknown's avatar
unknown committed
1793
  DBUG_RETURN(FALSE);
unknown's avatar
unknown committed
1794 1795
}

unknown's avatar
unknown committed
1796

1797 1798
void set_slave_thread_options(THD* thd)
{
1799
  DBUG_ENTER("set_slave_thread_options");
1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814
  /*
     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;
1815
  thd->variables.completion_type= 0;
1816
  DBUG_VOID_RETURN;
1817
}
1818

1819
void set_slave_thread_default_charset(THD* thd, Relay_log_info const *rli)
1820
{
1821 1822
  DBUG_ENTER("set_slave_thread_default_charset");

1823 1824 1825 1826 1827 1828 1829
  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();
1830 1831 1832 1833 1834 1835 1836

  /*
    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.
   */
1837
  const_cast<Relay_log_info*>(rli)->cached_charset_invalidate();
1838
  DBUG_VOID_RETURN;
1839 1840
}

unknown's avatar
unknown committed
1841
/*
1842
  init_slave_thread()
unknown's avatar
unknown committed
1843
*/
1844

1845
static int init_slave_thread(THD* thd, SLAVE_THD_TYPE thd_type)
unknown's avatar
unknown committed
1846 1847
{
  DBUG_ENTER("init_slave_thread");
1848 1849 1850
#if !defined(DBUG_OFF)
  int simulate_error= 0;
#endif
1851
  thd->system_thread = (thd_type == SLAVE_THD_SQL) ?
unknown's avatar
unknown committed
1852
    SYSTEM_THREAD_SLAVE_SQL : SYSTEM_THREAD_SLAVE_IO;
1853
  thd->security_ctx->skip_grants();
unknown's avatar
unknown committed
1854
  my_net_init(&thd->net, 0);
1855 1856 1857 1858 1859
/*
  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.
*/
1860
  thd->variables.max_allowed_packet= global_system_variables.max_allowed_packet
1861
    + MAX_LOG_EVENT_HEADER;  /* note, incr over the global not session var */
1862
  thd->slave_thread = 1;
1863
  thd->enable_slow_log= opt_log_slow_slave_statements;
1864
  set_slave_thread_options(thd);
unknown's avatar
unknown committed
1865 1866
  thd->client_capabilities = CLIENT_LOCAL_FILES;
  pthread_mutex_lock(&LOCK_thread_count);
unknown's avatar
unknown committed
1867
  thd->thread_id= thd->variables.pseudo_thread_id= thread_id++;
unknown's avatar
unknown committed
1868 1869
  pthread_mutex_unlock(&LOCK_thread_count);

1870 1871 1872 1873 1874 1875 1876
  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
1877
  if (init_thr_lock() || thd->store_globals())
1878
#endif
unknown's avatar
unknown committed
1879
  {
1880
    thd->cleanup();
unknown's avatar
unknown committed
1881 1882
    DBUG_RETURN(-1);
  }
1883
  lex_start(thd);
unknown's avatar
unknown committed
1884

1885
  if (thd_type == SLAVE_THD_SQL)
1886
    thd_proc_info(thd, "Waiting for the next event in relay log");
1887
  else
1888
    thd_proc_info(thd, "Waiting for master update");
unknown's avatar
unknown committed
1889 1890 1891 1892 1893
  thd->version=refresh_version;
  thd->set_time();
  DBUG_RETURN(0);
}

1894

1895
static int safe_sleep(THD* thd, int sec, CHECK_KILLED_FUNC thread_killed,
unknown's avatar
unknown committed
1896
                      void* thread_killed_arg)
unknown's avatar
unknown committed
1897
{
1898
  int nap_time;
unknown's avatar
unknown committed
1899
  thr_alarm_t alarmed;
1900 1901
  DBUG_ENTER("safe_sleep");

unknown's avatar
unknown committed
1902
  thr_alarm_init(&alarmed);
1903
  time_t start_time= my_time(0);
unknown's avatar
unknown committed
1904 1905
  time_t end_time= start_time+sec;

1906
  while ((nap_time= (int) (end_time - start_time)) > 0)
unknown's avatar
unknown committed
1907
  {
1908
    ALARM alarm_buff;
unknown's avatar
unknown committed
1909
    /*
1910
      The only reason we are asking for alarm is so that
unknown's avatar
unknown committed
1911 1912 1913
      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
    */
1914
    thr_alarm(&alarmed, 2 * nap_time, &alarm_buff);
unknown's avatar
unknown committed
1915
    sleep(nap_time);
1916
    thr_end_alarm(&alarmed);
unknown's avatar
unknown committed
1917

1918
    if ((*thread_killed)(thd,thread_killed_arg))
1919
      DBUG_RETURN(1);
1920
    start_time= my_time(0);
unknown's avatar
unknown committed
1921
  }
1922
  DBUG_RETURN(0);
unknown's avatar
unknown committed
1923 1924
}

1925

1926
static int request_dump(MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
1927
                        bool *suppress_warnings)
unknown's avatar
unknown committed
1928
{
1929
  uchar buf[FN_REFLEN + 10];
unknown's avatar
unknown committed
1930 1931
  int len;
  int binlog_flags = 0; // for now
1932
  char* logname = mi->master_log_name;
1933
  DBUG_ENTER("request_dump");
1934 1935
  
  *suppress_warnings= FALSE;
1936

unknown's avatar
unknown committed
1937
  // TODO if big log files: Change next to int8store()
unknown's avatar
unknown committed
1938
  int4store(buf, (ulong) mi->master_log_pos);
unknown's avatar
unknown committed
1939
  int2store(buf + 4, binlog_flags);
1940
  int4store(buf + 6, server_id);
unknown's avatar
unknown committed
1941
  len = (uint) strlen(logname);
1942
  memcpy(buf + 10, logname,len);
unknown's avatar
SCRUM  
unknown committed
1943
  if (simple_command(mysql, COM_BINLOG_DUMP, buf, len + 10, 1))
unknown's avatar
unknown committed
1944
  {
unknown's avatar
unknown committed
1945 1946 1947 1948 1949
    /*
      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
1950
    if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
1951
      *suppress_warnings= TRUE;                 // Suppress reconnect warning
unknown's avatar
unknown committed
1952
    else
1953
      sql_print_error("Error on COM_BINLOG_DUMP: %d  %s, will retry in %d secs",
unknown's avatar
unknown committed
1954 1955
                      mysql_errno(mysql), mysql_error(mysql),
                      master_connect_retry);
1956
    DBUG_RETURN(1);
unknown's avatar
unknown committed
1957
  }
unknown's avatar
unknown committed
1958

1959
  DBUG_RETURN(0);
unknown's avatar
unknown committed
1960 1961
}

1962

1963
static int request_table_dump(MYSQL* mysql, const char* db, const char* table)
unknown's avatar
unknown committed
1964
{
1965
  uchar buf[1024], *p = buf;
1966 1967
  DBUG_ENTER("request_table_dump");

unknown's avatar
unknown committed
1968 1969
  uint table_len = (uint) strlen(table);
  uint db_len = (uint) strlen(db);
unknown's avatar
unknown committed
1970
  if (table_len + db_len > sizeof(buf) - 2)
unknown's avatar
unknown committed
1971 1972
  {
    sql_print_error("request_table_dump: Buffer overrun");
1973
    DBUG_RETURN(1);
unknown's avatar
unknown committed
1974 1975
  }

unknown's avatar
unknown committed
1976 1977 1978 1979 1980
  *p++ = db_len;
  memcpy(p, db, db_len);
  p += db_len;
  *p++ = table_len;
  memcpy(p, table, table_len);
unknown's avatar
unknown committed
1981

unknown's avatar
SCRUM  
unknown committed
1982
  if (simple_command(mysql, COM_TABLE_DUMP, buf, p - buf + table_len, 1))
unknown's avatar
unknown committed
1983 1984
  {
    sql_print_error("request_table_dump: Error sending the table dump \
unknown's avatar
unknown committed
1985
command");
1986
    DBUG_RETURN(1);
unknown's avatar
unknown committed
1987
  }
unknown's avatar
unknown committed
1988

1989
  DBUG_RETURN(0);
unknown's avatar
unknown committed
1990 1991
}

1992

unknown's avatar
unknown committed
1993
/*
1994
  Read one event from the master
unknown's avatar
unknown committed
1995

unknown's avatar
unknown committed
1996 1997
  SYNOPSIS
    read_event()
unknown's avatar
unknown committed
1998 1999 2000 2001 2002 2003
    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.
2004

unknown's avatar
unknown committed
2005
    RETURN VALUES
unknown's avatar
unknown committed
2006 2007
    'packet_error'      Error
    number              Length of packet
unknown's avatar
unknown committed
2008 2009
*/

2010
static ulong read_event(MYSQL* mysql, Master_info *mi, bool* suppress_warnings)
unknown's avatar
unknown committed
2011
{
2012
  ulong len;
2013
  DBUG_ENTER("read_event");
unknown's avatar
unknown committed
2014

2015
  *suppress_warnings= FALSE;
unknown's avatar
unknown committed
2016 2017 2018 2019
  /*
    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
2020
#ifndef DBUG_OFF
2021
  if (disconnect_slave_event_count && !(mi->events_till_disconnect--))
unknown's avatar
unknown committed
2022
    DBUG_RETURN(packet_error);
unknown's avatar
unknown committed
2023
#endif
unknown's avatar
unknown committed
2024

2025
  len = cli_safe_read(mysql);
2026
  if (len == packet_error || (long) len < 1)
unknown's avatar
unknown committed
2027
  {
unknown's avatar
SCRUM  
unknown committed
2028
    if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
unknown's avatar
unknown committed
2029 2030
    {
      /*
unknown's avatar
unknown committed
2031 2032 2033
        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
2034 2035 2036 2037
      */
      *suppress_warnings= TRUE;
    }
    else
2038
      sql_print_error("Error reading packet from server: %s ( server_errno=%d)",
unknown's avatar
unknown committed
2039
                      mysql_error(mysql), mysql_errno(mysql));
2040
    DBUG_RETURN(packet_error);
unknown's avatar
unknown committed
2041 2042
  }

2043 2044
  /* Check if eof packet */
  if (len < 8 && mysql->net.read_pos[0] == 254)
unknown's avatar
unknown committed
2045
  {
2046 2047
    sql_print_information("Slave: received end packet from server, apparent "
                          "master shutdown: %s",
unknown's avatar
unknown committed
2048
                     mysql_error(mysql));
2049
     DBUG_RETURN(packet_error);
unknown's avatar
unknown committed
2050
  }
unknown's avatar
unknown committed
2051

unknown's avatar
unknown committed
2052
  DBUG_PRINT("exit", ("len: %lu  net->read_pos[4]: %d",
unknown's avatar
unknown committed
2053 2054
                      len, mysql->net.read_pos[4]));
  DBUG_RETURN(len - 1);
unknown's avatar
unknown committed
2055 2056
}

unknown's avatar
unknown committed
2057 2058 2059 2060 2061 2062 2063 2064 2065
/*
  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)
{
2066 2067
  DBUG_ENTER("has_temporary_error");

2068
  DBUG_EXECUTE_IF("all_errors_are_temporary_errors",
2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081
                  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
2082 2083 2084 2085 2086 2087

  /*
    Temporary error codes:
    currently, InnoDB deadlock detected by InnoDB or lock
    wait timeout (innodb_lock_wait_timeout exceeded
  */
2088 2089
  if (thd->main_da.sql_errno() == ER_LOCK_DEADLOCK ||
      thd->main_da.sql_errno() == ER_LOCK_WAIT_TIMEOUT)
2090
    DBUG_RETURN(1);
unknown's avatar
unknown committed
2091 2092 2093 2094 2095 2096 2097 2098 2099

#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
2100
    DBUG_PRINT("info", ("has warning %d %s", err->code, err->msg));
unknown's avatar
unknown committed
2101 2102 2103
    switch (err->code)
    {
    case ER_GET_TEMPORARY_ERRMSG:
2104
      DBUG_RETURN(1);
unknown's avatar
unknown committed
2105 2106 2107 2108 2109
    default:
      break;
    }
  }
#endif
2110
  DBUG_RETURN(0);
unknown's avatar
unknown committed
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 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221

/**
  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().
*/
int apply_event_and_update_pos(Log_event* ev, THD* thd, Relay_log_info* rli,
                               bool skip)
{
  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

  if (skip)
  {
    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);
#ifndef DBUG_OFF
    /*
      This only prints information to the debug trace.

      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]));
#endif
  }
  else
    exec_res= ev->apply_event(rli);

  DBUG_PRINT("info", ("apply_event error = %d", exec_res));
  if (exec_res == 0)
  {
    int error= ev->update_pos(rli);
2222 2223 2224 2225
#ifdef HAVE_purify
    if (!rli->is_fake)
#endif
    {
2226
#ifndef DBUG_OFF
2227
      char buf[22];
2228
#endif
2229 2230 2231 2232 2233 2234 2235 2236
      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));
    }
2237 2238 2239 2240 2241 2242 2243 2244 2245
    /*
      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)
    {
2246
      char buf[22];
2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281
      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
2282
     if a failure occurs when applying the event.)
2283 2284 2285 2286 2287 2288 2289

   - An error occurred when updating the binlog position.

  @retval 0 The event was applied.

  @retval 1 The event was not applied.
*/
2290
static int exec_relay_log_event(THD* thd, Relay_log_info* rli)
unknown's avatar
unknown committed
2291
{
2292 2293
  DBUG_ENTER("exec_relay_log_event");

2294 2295
  /*
     We acquire this mutex since we need it for all operations except
2296
     event execution. But we will release it in places where we will
2297 2298 2299
     wait for something for example inside of next_event().
   */
  pthread_mutex_lock(&rli->data_lock);
2300

2301
  Log_event * ev = next_event(rli);
2302

2303
  DBUG_ASSERT(rli->sql_thd==thd);
2304

2305
  if (sql_slave_killed(thd,rli))
2306
  {
2307
    pthread_mutex_unlock(&rli->data_lock);
unknown's avatar
unknown committed
2308
    delete ev;
2309
    DBUG_RETURN(1);
2310
  }
2311 2312
  if (ev)
  {
2313
    int exec_res;
2314

2315 2316 2317 2318
    /*
      This tests if the position of the beginning of the current event
      hits the UNTIL barrier.
    */
2319
    if (rli->until_condition != Relay_log_info::UNTIL_NONE &&
2320
        rli->is_until_satisfied((rli->is_in_group() || !ev->log_pos) ?
2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333
                                rli->group_master_log_pos :
                                ev->log_pos - ev->data_written))
    {
      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;
2334
      DBUG_RETURN(1);
2335
    }
2336
    exec_res= apply_event_and_update_pos(ev, thd, rli, TRUE);
2337

2338
    /*
2339 2340 2341
      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.
2342
    */
2343
    if (ev->get_type_code() != FORMAT_DESCRIPTION_EVENT)
2344
    {
2345 2346
      DBUG_PRINT("info", ("Deleting the event after it has been executed"));
      delete ev;
2347 2348
    }

2349
    /*
2350 2351
      update_log_pos failed: this should not happen, so we don't
      retry.
2352
    */
2353 2354 2355
    if (exec_res == 2)
      DBUG_RETURN(1);

2356 2357
    if (slave_trans_retries)
    {
2358 2359
      int temp_err;
      if (exec_res && (temp_err= has_temporary_error(thd)))
2360 2361 2362 2363
      {
        const char *errmsg;
        /*
          We were in a transaction which has been rolled back because of a
unknown's avatar
unknown committed
2364 2365
          temporary error;
          let's seek back to BEGIN log event and retry it all again.
2366
	  Note, if lock wait timeout (innodb_lock_wait_timeout exceeded)
2367
	  there is no rollback since 5.0.13 (ref: manual).
2368 2369 2370 2371 2372 2373 2374 2375
          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.
        */
2376
        if (rli->trans_retries < slave_trans_retries)
2377 2378 2379 2380 2381 2382
        {
          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,
2383
                                      1, &errmsg, 1))
2384 2385 2386 2387 2388
            sql_print_error("Error initializing relay log position: %s",
                            errmsg);
          else
          {
            exec_res= 0;
2389
            end_trans(thd, ROLLBACK);
unknown's avatar
unknown committed
2390
            /* chance for concurrent connection to get more locks */
2391
            safe_sleep(thd, min(rli->trans_retries, MAX_SLAVE_RETRY_PAUSE),
unknown's avatar
unknown committed
2392
                       (CHECK_KILLED_FUNC)sql_slave_killed, (void*)rli);
2393
            pthread_mutex_lock(&rli->data_lock); // because of SHOW STATUS
unknown's avatar
unknown committed
2394
            rli->trans_retries++;
2395 2396 2397 2398
            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
2399
          }
2400 2401 2402 2403 2404 2405 2406
        }
        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);
      }
2407
      else if ((exec_res && !temp_err) ||
2408 2409
               (opt_using_transactions &&
                rli->group_relay_log_pos == rli->event_relay_log_pos))
2410 2411
      {
        /*
2412 2413 2414
          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
2415 2416 2417
          non-transient error, the slave will stop with an error.
         */
        rli->trans_retries= 0; // restart from fresh
unknown's avatar
unknown committed
2418
        DBUG_PRINT("info", ("Resetting retry counter, rli->trans_retries: %lu",
2419
                            rli->trans_retries));
2420
      }
2421 2422
    }
    DBUG_RETURN(exec_res);
2423
  }
2424
  pthread_mutex_unlock(&rli->data_lock);
2425 2426
  rli->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_READ_FAILURE,
              ER(ER_SLAVE_RELAY_LOG_READ_FAILURE), "\
2427 2428 2429 2430 2431 2432 2433
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
2434
");
2435
  DBUG_RETURN(1);
unknown's avatar
unknown committed
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
/**
  @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.
*/

2465
static int try_to_reconnect(THD *thd, MYSQL *mysql, Master_info *mi,
2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514
                            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
    {
      sql_print_information(buf);
    }
  }
  if (safe_reconnect(thd, mysql, mi, 1) || io_slave_killed(thd, mi))
  {
    if (global_system_variables.log_warnings)
      sql_print_information(messages[SLAVE_RECON_MSG_KILLED_AFTER]);
    return 1;
  }
  return 0;
}


2515 2516 2517 2518 2519
/**
  Slave IO thread entry point.

  @param arg Pointer to Master_info struct that holds information for
  the IO thread.
2520

2521 2522
  @return Always 0.
*/
2523
pthread_handler_t handle_slave_io(void *arg)
unknown's avatar
unknown committed
2524
{
unknown's avatar
unknown committed
2525 2526
  THD *thd; // needs to be first for thread_stack
  MYSQL *mysql;
2527
  Master_info *mi = (Master_info*)arg;
2528
  Relay_log_info *rli= &mi->rli;
unknown's avatar
unknown committed
2529 2530
  char llbuff[22];
  uint retry_count;
2531
  bool suppress_warnings;
2532
  int ret;
2533 2534 2535
#ifndef DBUG_OFF
  uint retry_count_reg= 0, retry_count_dump= 0, retry_count_event= 0;
#endif
unknown's avatar
unknown committed
2536 2537
  // needs to call my_thread_init(), otherwise we get a coredump in DBUG_ stuff
  my_thread_init();
2538
  DBUG_ENTER("handle_slave_io");
unknown's avatar
unknown committed
2539

2540
  DBUG_ASSERT(mi->inited);
unknown's avatar
unknown committed
2541 2542 2543
  mysql= NULL ;
  retry_count= 0;

2544
  pthread_mutex_lock(&mi->run_lock);
unknown's avatar
unknown committed
2545 2546 2547
  /* Inform waiting threads that slave has started */
  mi->slave_run_id++;

2548
#ifndef DBUG_OFF
2549
  mi->events_till_disconnect = disconnect_slave_event_count;
2550 2551
#endif

2552
  thd= new THD; // note that contructor of THD uses DBUG_ !
2553
  THD_CHECK_SENTRY(thd);
2554
  mi->io_thd = thd;
unknown's avatar
unknown committed
2555 2556

  pthread_detach_this_thread();
2557
  thd->thread_stack= (char*) &thd; // remember where our stack is
2558
  mi->clear_error();
2559
  if (init_slave_thread(thd, SLAVE_THD_IO))
unknown's avatar
unknown committed
2560 2561 2562 2563 2564 2565
  {
    pthread_cond_broadcast(&mi->start_cond);
    pthread_mutex_unlock(&mi->run_lock);
    sql_print_error("Failed during slave I/O thread initialization");
    goto err;
  }
2566
  pthread_mutex_lock(&LOCK_thread_count);
unknown's avatar
unknown committed
2567
  threads.append(thd);
2568
  pthread_mutex_unlock(&LOCK_thread_count);
2569 2570 2571
  mi->slave_running = 1;
  mi->abort_slave = 0;
  pthread_mutex_unlock(&mi->run_lock);
2572
  pthread_cond_broadcast(&mi->start_cond);
2573

2574
  DBUG_PRINT("master_info",("log_file_name: '%s'  position: %s",
unknown's avatar
unknown committed
2575 2576
                            mi->master_log_name,
                            llstr(mi->master_log_pos,llbuff)));
2577

unknown's avatar
SCRUM  
unknown committed
2578
  if (!(mi->mysql = mysql = mysql_init(NULL)))
unknown's avatar
unknown committed
2579
  {
2580 2581
    mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
               ER(ER_SLAVE_FATAL_ERROR), "error in mysql_init()");
unknown's avatar
unknown committed
2582 2583
    goto err;
  }
2584

2585
  thd_proc_info(thd, "Connecting to master");
2586
  // we can get killed during safe_connect
2587
  if (!safe_connect(thd, mysql, mi))
2588
  {
unknown's avatar
unknown committed
2589 2590 2591
    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,
2592 2593
			  IO_RPL_LOG_NAME,
			  llstr(mi->master_log_pos,llbuff));
2594 2595 2596 2597 2598
  /*
    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.
  */
2599 2600
    mysql->net.max_packet_size= thd->net.max_packet_size+= MAX_LOG_EVENT_HEADER;
  }
2601
  else
unknown's avatar
unknown committed
2602
  {
2603
    sql_print_information("Slave I/O thread killed while connecting to master");
unknown's avatar
unknown committed
2604 2605
    goto err;
  }
2606

2607
connected:
2608

2609 2610
  // TODO: the assignment below should be under mutex (5.0)
  mi->slave_running= MYSQL_SLAVE_RUN_CONNECT;
2611
  thd->slave_net = &mysql->net;
2612
  thd_proc_info(thd, "Checking master version");
2613 2614 2615
  ret= get_master_version_and_clock(mysql, mi);
  if (ret == 1)
    /* Fatal error */
2616
    goto err;
2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629
  
  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;
  } 
2630 2631

  if (mi->rli.relay_log.description_event_for_queue->binlog_version > 1)
2632
  {
unknown's avatar
unknown committed
2633 2634 2635
    /*
      Register ourselves with the master.
    */
2636
    thd_proc_info(thd, "Registering slave on master");
2637
    if (register_slave_on_master(mysql, mi, &suppress_warnings))
2638
    {
2639 2640 2641 2642 2643 2644 2645 2646 2647
      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
2648 2649
        goto err;
      goto connected;
2650
    }
2651 2652 2653 2654 2655 2656 2657 2658 2659 2660
    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;
      });
2661
  }
2662

2663
  DBUG_PRINT("info",("Starting reading binary log from master"));
2664
  while (!io_slave_killed(thd,mi))
unknown's avatar
unknown committed
2665
  {
2666
    thd_proc_info(thd, "Requesting binlog dump");
unknown's avatar
unknown committed
2667
    if (request_dump(mysql, mi, &suppress_warnings))
unknown's avatar
unknown committed
2668 2669
    {
      sql_print_error("Failed on request_dump()");
2670 2671 2672 2673
      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
2674
        goto err;
unknown's avatar
unknown committed
2675 2676
      goto connected;
    }
2677 2678 2679 2680 2681 2682 2683 2684 2685 2686
    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
2687

2688
    DBUG_ASSERT(mi->last_error().number == 0);
2689
    while (!io_slave_killed(thd,mi))
unknown's avatar
unknown committed
2690
    {
2691
      ulong event_len;
2692
      /*
2693
         We say "waiting" because read_event() will wait if there's nothing to
2694 2695 2696
         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.
2697
      */
2698
      thd_proc_info(thd, "Waiting for master to send event");
2699
      event_len= read_event(mysql, mi, &suppress_warnings);
2700 2701
      if (check_io_slave_killed(thd, mi, "Slave I/O thread killed while \
reading event"))
unknown's avatar
unknown committed
2702
        goto err;
2703 2704 2705 2706 2707 2708 2709 2710 2711 2712
      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;
        });
2713

unknown's avatar
unknown committed
2714 2715
      if (event_len == packet_error)
      {
unknown's avatar
unknown committed
2716
        uint mysql_error_number= mysql_errno(mysql);
2717 2718
        switch (mysql_error_number) {
        case CR_NET_PACKET_TOO_LARGE:
unknown's avatar
unknown committed
2719
          sql_print_error("\
2720 2721 2722
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
2723 2724
                          thd->variables.max_allowed_packet);
          goto err;
2725
        case ER_MASTER_FATAL_ERROR_READING_BINLOG:
unknown's avatar
unknown committed
2726 2727 2728
          sql_print_error(ER(mysql_error_number), mysql_error_number,
                          mysql_error(mysql));
          goto err;
2729 2730 2731 2732
        case EE_OUTOFMEMORY:
        case ER_OUTOFMEMORY:
          sql_print_error("\
Stopping slave I/O thread due to out-of-memory error from master");
unknown's avatar
unknown committed
2733 2734
          goto err;
        }
2735 2736
        if (try_to_reconnect(thd, mysql, mi, &retry_count, suppress_warnings,
                             reconnect_messages[SLAVE_RECON_ACT_EVENT]))
unknown's avatar
unknown committed
2737 2738
          goto err;
        goto connected;
unknown's avatar
unknown committed
2739
      } // if (event_len == packet_error)
2740

unknown's avatar
unknown committed
2741
      retry_count=0;                    // ok event, reset retry counter
2742
      thd_proc_info(thd, "Queueing master event to the relay log");
unknown's avatar
unknown committed
2743
      if (queue_event(mi,(const char*)mysql->net.read_pos + 1,
unknown's avatar
unknown committed
2744
                      event_len))
unknown's avatar
unknown committed
2745
      {
2746 2747 2748
        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
2749
        goto err;
unknown's avatar
unknown committed
2750
      }
2751 2752 2753 2754 2755
      if (flush_master_info(mi, 1))
      {
        sql_print_error("Failed to flush master info file");
        goto err;
      }
2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767
      /*
        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.
      */
2768 2769 2770 2771 2772
#ifndef DBUG_OFF
      {
        char llbuf1[22], llbuf2[22];
        DBUG_PRINT("info", ("log_space_limit=%s log_space_total=%s \
ignore_log_space_limit=%d",
2773 2774
                            llstr(rli->log_space_limit,llbuf1),
                            llstr(rli->log_space_total,llbuf2),
unknown's avatar
unknown committed
2775
                            (int) rli->ignore_log_space_limit));
2776 2777 2778
      }
#endif

2779
      if (rli->log_space_limit && rli->log_space_limit <
unknown's avatar
unknown committed
2780
          rli->log_space_total &&
2781
          !rli->ignore_log_space_limit)
unknown's avatar
unknown committed
2782 2783 2784
        if (wait_for_relay_log_space(rli))
        {
          sql_print_error("Slave I/O thread aborted while waiting for relay \
unknown's avatar
unknown committed
2785
log space");
unknown's avatar
unknown committed
2786 2787 2788
          goto err;
        }
    }
2789
  }
unknown's avatar
unknown committed
2790

unknown's avatar
unknown committed
2791
  // error = 0;
unknown's avatar
unknown committed
2792
err:
2793
  // print the current replication position
2794
  sql_print_information("Slave I/O thread exiting, read up to log '%s', position %s",
unknown's avatar
unknown committed
2795
                  IO_RPL_LOG_NAME, llstr(mi->master_log_pos,llbuff));
2796
  thd->set_query(NULL, 0);
unknown's avatar
unknown committed
2797
  thd->reset_db(NULL, 0);
unknown's avatar
unknown committed
2798 2799
  if (mysql)
  {
2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810
    /*
      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
2811
    mysql_close(mysql);
unknown's avatar
unknown committed
2812 2813
    mi->mysql=0;
  }
2814
  write_ignored_events_info_to_relay_log(thd, mi);
2815
  thd_proc_info(thd, "Waiting for slave mutex on exit");
2816
  pthread_mutex_lock(&mi->run_lock);
2817

2818 2819 2820
  /* Forget the relay log's format */
  delete mi->rli.relay_log.description_event_for_queue;
  mi->rli.relay_log.description_event_for_queue= 0;
2821
  // TODO: make rpl_status part of Master_info
2822
  change_rpl_status(RPL_ACTIVE_SLAVE,RPL_IDLE_SLAVE);
2823
  DBUG_ASSERT(thd->net.buff != 0);
unknown's avatar
unknown committed
2824
  net_end(&thd->net); // destructor will not free it, because net.vio is 0
2825
  close_thread_tables(thd);
2826
  pthread_mutex_lock(&LOCK_thread_count);
2827
  THD_CHECK_SENTRY(thd);
unknown's avatar
unknown committed
2828
  delete thd;
2829
  pthread_mutex_unlock(&LOCK_thread_count);
unknown's avatar
unknown committed
2830 2831 2832
  mi->abort_slave= 0;
  mi->slave_running= 0;
  mi->io_thd= 0;
2833 2834 2835 2836 2837
  /*
    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)
   */ 
2838
  pthread_cond_broadcast(&mi->stop_cond);       // tell the world we are done
2839
  DBUG_EXECUTE_IF("simulate_slave_delay_at_terminate_bug38694", sleep(5););
2840
  pthread_mutex_unlock(&mi->run_lock);
2841
  my_thread_end();
unknown's avatar
unknown committed
2842
  pthread_exit(0);
unknown's avatar
unknown committed
2843
  DBUG_RETURN(0);                               // Can't return anything here
unknown's avatar
unknown committed
2844 2845
}

2846 2847 2848 2849 2850
/*
  Check the temporary directory used by commands like
  LOAD DATA INFILE.
 */
static 
2851
int check_temp_dir(char* tmp_file)
2852 2853 2854
{
  int fd;
  MY_DIR *dirp;
2855 2856
  char tmp_dir[FN_REFLEN];
  size_t tmp_dir_size;
2857 2858 2859

  DBUG_ENTER("check_temp_dir");

2860 2861 2862 2863 2864
  /*
    Get the directory from the temporary file.
  */
  dirname_part(tmp_dir, tmp_file, &tmp_dir_size);

2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887
  /*
    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
2888

2889 2890 2891 2892 2893
/**
  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
2894

2895 2896
  @return Always 0.
*/
2897
pthread_handler_t handle_slave_sql(void *arg)
2898
{
unknown's avatar
unknown committed
2899
  THD *thd;                     /* needs to be first for thread_stack */
2900
  char llbuff[22],llbuff1[22];
2901

2902
  Relay_log_info* rli = &((Master_info*)arg)->rli;
unknown's avatar
unknown committed
2903 2904 2905 2906
  const char *errmsg;

  // needs to call my_thread_init(), otherwise we get a coredump in DBUG_ stuff
  my_thread_init();
2907
  DBUG_ENTER("handle_slave_sql");
unknown's avatar
unknown committed
2908

2909 2910 2911
  DBUG_ASSERT(rli->inited);
  pthread_mutex_lock(&rli->run_lock);
  DBUG_ASSERT(!rli->slave_running);
unknown's avatar
unknown committed
2912
  errmsg= 0;
unknown's avatar
unknown committed
2913
#ifndef DBUG_OFF
2914
  rli->events_till_abort = abort_slave_event_count;
unknown's avatar
unknown committed
2915
#endif
2916

unknown's avatar
unknown committed
2917
  thd = new THD; // note that contructor of THD uses DBUG_ !
2918
  thd->thread_stack = (char*)&thd; // remember where our stack is
2919
  rli->sql_thd= thd;
2920
  
unknown's avatar
unknown committed
2921 2922
  /* Inform waiting threads that slave has started */
  rli->slave_run_id++;
2923
  rli->slave_running = 1;
unknown's avatar
unknown committed
2924

2925 2926
  pthread_detach_this_thread();
  if (init_slave_thread(thd, SLAVE_THD_SQL))
unknown's avatar
unknown committed
2927 2928 2929 2930 2931 2932 2933
  {
    /*
      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);
2934 2935
    rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, 
                "Failed during slave thread initialization");
unknown's avatar
unknown committed
2936 2937
    goto err;
  }
2938
  thd->init_for_queries();
2939
  thd->temporary_tables = rli->save_temporary_tables; // restore temp tables
2940
  set_thd_in_use_temporary_tables(rli);   // (re)set sql_thd in use for saved temp tables
2941
  pthread_mutex_lock(&LOCK_thread_count);
2942
  threads.append(thd);
2943
  pthread_mutex_unlock(&LOCK_thread_count);
2944 2945 2946 2947 2948 2949 2950 2951
  /*
    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.
  */
2952 2953
  rli->abort_slave = 0;
  pthread_mutex_unlock(&rli->run_lock);
2954
  pthread_cond_broadcast(&rli->start_cond);
2955

unknown's avatar
unknown committed
2956 2957 2958
  /*
    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
2959
    though there's no problem anymore). Do not reset the master timestamp
2960 2961 2962 2963
    (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
2964
    But the master timestamp is reset by RESET SLAVE & CHANGE MASTER.
unknown's avatar
unknown committed
2965
  */
2966
  rli->clear_error();
2967 2968

  //tell the I/O thread to take relay_log_space_limit into account from now on
2969
  pthread_mutex_lock(&rli->log_space_lock);
2970
  rli->ignore_log_space_limit= 0;
2971
  pthread_mutex_unlock(&rli->log_space_lock);
2972
  rli->trans_retries= 0; // start from "no error"
unknown's avatar
unknown committed
2973
  DBUG_PRINT("info", ("rli->trans_retries: %lu", rli->trans_retries));
2974

2975
  if (init_relay_log_pos(rli,
unknown's avatar
unknown committed
2976 2977 2978
                         rli->group_relay_log_name,
                         rli->group_relay_log_pos,
                         1 /*need data lock*/, &errmsg,
2979
                         1 /*look for a description_event*/))
2980 2981 2982
  { 
    rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, 
                "Error initializing relay log position: %s", errmsg);
2983 2984
    goto err;
  }
2985
  THD_CHECK_SENTRY(thd);
2986 2987 2988 2989
#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
2990
                        llstr(my_b_tell(rli->cur_log),llbuf1),
2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009
                        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
3010
  DBUG_ASSERT(rli->sql_thd == thd);
3011 3012

  DBUG_PRINT("master_info",("log_file_name: %s  position: %s",
unknown's avatar
unknown committed
3013 3014
                            rli->group_master_log_name,
                            llstr(rli->group_master_log_pos,llbuff)));
3015
  if (global_system_variables.log_warnings)
3016
    sql_print_information("Slave SQL thread initialized, starting replication in \
unknown's avatar
unknown committed
3017
log '%s' at position %s, relay log '%s' position: %s", RPL_LOG_NAME,
unknown's avatar
unknown committed
3018 3019
                    llstr(rli->group_master_log_pos,llbuff),rli->group_relay_log_name,
                    llstr(rli->group_relay_log_pos,llbuff1));
3020

3021
  if (check_temp_dir(rli->slave_patternload_file))
3022 3023 3024 3025 3026 3027 3028
  {
    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
3029
  /* execute init_slave variable */
unknown's avatar
unknown committed
3030
  if (sys_init_slave.value_length)
unknown's avatar
unknown committed
3031
  {
3032
    execute_init_command(thd, &sys_init_slave, &LOCK_sys_init_slave);
3033
    if (thd->is_slave_error)
unknown's avatar
unknown committed
3034
    {
3035 3036
      rli->report(ERROR_LEVEL, thd->main_da.sql_errno(), 
                  "Slave SQL thread aborted. Can't execute init_slave query");
unknown's avatar
unknown committed
3037 3038 3039 3040
      goto err;
    }
  }

3041 3042 3043 3044 3045
  /*
    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);
3046
  if (rli->until_condition != Relay_log_info::UNTIL_NONE &&
3047 3048 3049 3050 3051 3052 3053 3054 3055 3056
      rli->is_until_satisfied(rli->group_master_log_pos))
  {
    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);

3057 3058
  /* Read queries from the IO/THREAD until this thread is killed */

3059
  while (!sql_slave_killed(thd,rli))
3060
  {
3061
    thd_proc_info(thd, "Reading event from the relay log");
3062
    DBUG_ASSERT(rli->sql_thd == thd);
3063
    THD_CHECK_SENTRY(thd);
3064 3065
    if (exec_relay_log_event(thd,rli))
    {
3066
      DBUG_PRINT("info", ("exec_relay_log_event() failed"));
3067
      // do not scare the user if SQL thread was simply killed or stopped
3068
      if (!sql_slave_killed(thd,rli))
unknown's avatar
unknown committed
3069 3070
      {
        /*
3071 3072 3073
          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
3074
        */
3075
        uint32 const last_errno= rli->last_error().number;
3076

3077
        if (thd->is_error())
unknown's avatar
unknown committed
3078
        {
3079 3080 3081 3082 3083
          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));
3084
          if (last_errno == 0)
unknown's avatar
unknown committed
3085
          {
3086 3087 3088 3089
            /*
 	      This function is reporting an error which was not reported
 	      while executing exec_relay_log_event().
 	    */ 
3090
            rli->report(ERROR_LEVEL, thd->main_da.sql_errno(), errmsg);
unknown's avatar
unknown committed
3091
          }
3092
          else if (last_errno != thd->main_da.sql_errno())
unknown's avatar
unknown committed
3093
          {
3094 3095 3096 3097 3098 3099
            /*
             * 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
3100
            sql_print_error("Slave (additional info): %s Error_code: %d",
3101
                            errmsg, thd->main_da.sql_errno());
unknown's avatar
unknown committed
3102 3103 3104 3105 3106 3107
          }
        }

        /* Print any warnings issued */
        List_iterator_fast<MYSQL_ERROR> it(thd->warn_list);
        MYSQL_ERROR *err;
3108 3109 3110 3111 3112
        /*
          Added controlled slave thread cancel for replication
          of user-defined variables.
        */
        bool udf_error = false;
unknown's avatar
unknown committed
3113
        while ((err= it++))
3114 3115 3116
        {
          if (err->code == ER_CANT_OPEN_LIBRARY)
            udf_error = true;
unknown's avatar
unknown committed
3117
          sql_print_warning("Slave: %s Error_code: %d",err->msg, err->code);
3118 3119 3120 3121 3122 3123 3124 3125 3126
        }
        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("\
3127
Error running query, slave SQL thread aborted. Fix the problem, and restart \
unknown's avatar
unknown committed
3128
the slave SQL thread with \"SLAVE START\". We stopped at log \
3129
'%s' position %s", RPL_LOG_NAME, llstr(rli->group_master_log_pos, llbuff));
unknown's avatar
unknown committed
3130
      }
3131 3132
      goto err;
    }
3133
  }
3134

3135
  /* Thread stopped. Print the current replication position to the log */
unknown's avatar
unknown committed
3136
  sql_print_information("Slave SQL thread exiting, replication stopped in log "
unknown's avatar
unknown committed
3137 3138
                        "'%s' at position %s",
                        RPL_LOG_NAME, llstr(rli->group_master_log_pos,llbuff));
3139 3140

 err:
3141 3142 3143 3144 3145 3146 3147 3148

  /*
    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);
3149 3150 3151 3152 3153
  /*
    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)).
  */
3154 3155
  thd->catalog= 0;
  thd->set_query(NULL, 0);
unknown's avatar
unknown committed
3156
  thd->reset_db(NULL, 0);
3157
  thd_proc_info(thd, "Waiting for slave mutex on exit");
3158
  pthread_mutex_lock(&rli->run_lock);
3159 3160
  /* We need data_lock, at least to wake up any waiting master_pos_wait() */
  pthread_mutex_lock(&rli->data_lock);
3161
  DBUG_ASSERT(rli->slave_running == 1); // tracking buffer overrun
3162
  /* When master_pos_wait() wakes up it will check this and terminate */
unknown's avatar
unknown committed
3163
  rli->slave_running= 0;
3164 3165 3166
  /* Forget the relay log's format */
  delete rli->relay_log.description_event_for_exec;
  rli->relay_log.description_event_for_exec= 0;
3167 3168 3169 3170
  /* 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
3171
  rli->ignore_log_space_limit= 0; /* don't need any lock */
3172 3173
  /* we die so won't remember charset - re-update them on next thread start */
  rli->cached_charset_invalidate();
3174
  rli->save_temporary_tables = thd->temporary_tables;
unknown's avatar
unknown committed
3175 3176 3177 3178 3179

  /*
    TODO: see if we can do this conditionally in next_event() instead
    to avoid unneeded position re-init
  */
3180 3181 3182 3183
  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);
3184
  THD_CHECK_SENTRY(thd);
3185
  rli->sql_thd= 0;
3186
  set_thd_in_use_temporary_tables(rli);  // (re)set sql_thd in use for saved temp tables
3187
  pthread_mutex_lock(&LOCK_thread_count);
3188
  THD_CHECK_SENTRY(thd);
3189 3190
  delete thd;
  pthread_mutex_unlock(&LOCK_thread_count);
3191 3192 3193 3194 3195
 /*
  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)
 */ 
3196
  pthread_cond_broadcast(&rli->stop_cond);
3197
  DBUG_EXECUTE_IF("simulate_slave_delay_at_terminate_bug38694", sleep(5););
3198 3199
  pthread_mutex_unlock(&rli->run_lock);  // tell the world we are done
  
unknown's avatar
unknown committed
3200
  my_thread_end();
3201
  pthread_exit(0);
unknown's avatar
unknown committed
3202
  DBUG_RETURN(0);                               // Can't return anything here
3203
}
unknown's avatar
unknown committed
3204

3205

unknown's avatar
unknown committed
3206
/*
3207
  process_io_create_file()
unknown's avatar
unknown committed
3208
*/
3209

3210
static int process_io_create_file(Master_info* mi, Create_file_log_event* cev)
unknown's avatar
unknown committed
3211 3212 3213 3214
{
  int error = 1;
  ulong num_bytes;
  bool cev_not_written;
3215 3216
  THD *thd = mi->io_thd;
  NET *net = &mi->mysql->net;
unknown's avatar
unknown committed
3217
  DBUG_ENTER("process_io_create_file");
unknown's avatar
unknown committed
3218 3219

  if (unlikely(!cev->is_valid()))
unknown's avatar
unknown committed
3220
    DBUG_RETURN(1);
3221 3222

  if (!rpl_filter->db_ok(cev->db))
unknown's avatar
unknown committed
3223 3224
  {
    skip_load_data_infile(net);
unknown's avatar
unknown committed
3225
    DBUG_RETURN(0);
unknown's avatar
unknown committed
3226 3227 3228
  }
  DBUG_ASSERT(cev->inited_from_old);
  thd->file_id = cev->file_id = mi->file_id++;
3229
  thd->server_id = cev->server_id;
unknown's avatar
unknown committed
3230
  cev_not_written = 1;
unknown's avatar
unknown committed
3231

unknown's avatar
unknown committed
3232 3233 3234
  if (unlikely(net_request_file(net,cev->fname)))
  {
    sql_print_error("Slave I/O: failed requesting download of '%s'",
unknown's avatar
unknown committed
3235
                    cev->fname);
unknown's avatar
unknown committed
3236 3237 3238
    goto err;
  }

unknown's avatar
unknown committed
3239 3240 3241 3242
  /*
    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
3243 3244
  */
  {
3245
    Append_block_log_event aev(thd,0,0,0,0);
unknown's avatar
unknown committed
3246

unknown's avatar
unknown committed
3247 3248 3249 3250
    for (;;)
    {
      if (unlikely((num_bytes=my_net_read(net)) == packet_error))
      {
unknown's avatar
unknown committed
3251 3252 3253
        sql_print_error("Network read error downloading '%s' from master",
                        cev->fname);
        goto err;
unknown's avatar
unknown committed
3254 3255 3256
      }
      if (unlikely(!num_bytes)) /* eof */
      {
3257 3258
	/* 3.23 master wants it */
        net_write_command(net, 0, (uchar*) "", 0, (uchar*) "", 0);
unknown's avatar
unknown committed
3259 3260 3261 3262 3263 3264 3265
        /*
          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
3266 3267 3268 3269 3270
          break;
        Execute_load_log_event xev(thd,0,0);
        xev.log_pos = cev->log_pos;
        if (unlikely(mi->rli.relay_log.append(&xev)))
        {
3271 3272 3273
          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
3274 3275 3276 3277
          goto err;
        }
        mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total);
        break;
unknown's avatar
unknown committed
3278 3279 3280
      }
      if (unlikely(cev_not_written))
      {
unknown's avatar
unknown committed
3281
        cev->block = net->read_pos;
unknown's avatar
unknown committed
3282 3283 3284
        cev->block_len = num_bytes;
        if (unlikely(mi->rli.relay_log.append(cev)))
        {
3285 3286 3287
          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
3288 3289 3290 3291
          goto err;
        }
        cev_not_written=0;
        mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total);
unknown's avatar
unknown committed
3292 3293 3294
      }
      else
      {
unknown's avatar
unknown committed
3295
        aev.block = net->read_pos;
unknown's avatar
unknown committed
3296 3297 3298 3299
        aev.block_len = num_bytes;
        aev.log_pos = cev->log_pos;
        if (unlikely(mi->rli.relay_log.append(&aev)))
        {
3300 3301 3302
          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
3303 3304 3305
          goto err;
        }
        mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total) ;
unknown's avatar
unknown committed
3306 3307 3308 3309 3310
      }
    }
  }
  error=0;
err:
unknown's avatar
unknown committed
3311
  DBUG_RETURN(error);
unknown's avatar
unknown committed
3312
}
unknown's avatar
unknown committed
3313

3314

unknown's avatar
unknown committed
3315
/*
unknown's avatar
unknown committed
3316 3317 3318 3319
  Start using a new binary log on the master

  SYNOPSIS
    process_io_rotate()
unknown's avatar
unknown committed
3320 3321
    mi                  master_info for the slave
    rev                 The rotate log event read from the binary log
unknown's avatar
unknown committed
3322 3323

  DESCRIPTION
3324
    Updates the master info with the place in the next binary
unknown's avatar
unknown committed
3325
    log where we should start reading.
3326
    Rotate the relay log to avoid mixed-format relay logs.
unknown's avatar
unknown committed
3327 3328 3329 3330 3331

  NOTES
    We assume we already locked mi->data_lock

  RETURN VALUES
unknown's avatar
unknown committed
3332 3333
    0           ok
    1           Log event is illegal
unknown's avatar
unknown committed
3334 3335 3336

*/

3337
static int process_io_rotate(Master_info *mi, Rotate_log_event *rev)
3338
{
3339
  DBUG_ENTER("process_io_rotate");
unknown's avatar
unknown committed
3340
  safe_mutex_assert_owner(&mi->data_lock);
3341

unknown's avatar
unknown committed
3342
  if (unlikely(!rev->is_valid()))
3343
    DBUG_RETURN(1);
unknown's avatar
unknown committed
3344

3345
  /* Safe copy as 'rev' has been "sanitized" in Rotate_log_event's ctor */
unknown's avatar
unknown committed
3346 3347
  memcpy(mi->master_log_name, rev->new_log_ident, rev->ident_len+1);
  mi->master_log_pos= rev->pos;
unknown's avatar
unknown committed
3348
  DBUG_PRINT("info", ("master_log_pos: '%s' %lu",
unknown's avatar
unknown committed
3349
                      mi->master_log_name, (ulong) mi->master_log_pos));
3350
#ifndef DBUG_OFF
unknown's avatar
unknown committed
3351 3352 3353 3354 3355
  /*
    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)
3356
    mi->events_till_disconnect++;
3357
#endif
3358

3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372
  /*
    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);
  }
3373 3374 3375 3376
  /*
    Rotate the relay log makes binlog format detection easier (at next slave
    start or mysqlbinlog)
  */
3377
  rotate_relay_log(mi); /* will take the right mutexes */
3378
  DBUG_RETURN(0);
3379 3380
}

unknown's avatar
unknown committed
3381
/*
3382 3383
  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
3384
*/
3385
static int queue_binlog_ver_1_event(Master_info *mi, const char *buf,
unknown's avatar
unknown committed
3386
                           ulong event_len)
3387
{
unknown's avatar
unknown committed
3388
  const char *errmsg = 0;
unknown's avatar
unknown committed
3389 3390 3391
  ulong inc_pos;
  bool ignore_event= 0;
  char *tmp_buf = 0;
3392
  Relay_log_info *rli= &mi->rli;
3393
  DBUG_ENTER("queue_binlog_ver_1_event");
unknown's avatar
unknown committed
3394

unknown's avatar
unknown committed
3395 3396 3397
  /*
    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
3398 3399 3400 3401 3402
  */
  if (buf[EVENT_TYPE_OFFSET] == LOAD_EVENT)
  {
    if (unlikely(!(tmp_buf=(char*)my_malloc(event_len+1,MYF(MY_WME)))))
    {
3403 3404
      mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
                 ER(ER_SLAVE_FATAL_ERROR), "Memory allocation failed");
unknown's avatar
unknown committed
3405
      DBUG_RETURN(1);
unknown's avatar
unknown committed
3406 3407
    }
    memcpy(tmp_buf,buf,event_len);
3408 3409 3410 3411 3412 3413 3414 3415
    /*
      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
3416
    int4store(tmp_buf+EVENT_LEN_OFFSET, event_len);
unknown's avatar
unknown committed
3417 3418
    buf = (const char*)tmp_buf;
  }
3419 3420 3421 3422 3423 3424
  /*
    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
3425
  Log_event *ev = Log_event::read_log_event(buf,event_len, &errmsg,
3426
                                            mi->rli.relay_log.description_event_for_queue);
3427
  if (unlikely(!ev))
3428 3429
  {
    sql_print_error("Read invalid event from master: '%s',\
unknown's avatar
unknown committed
3430
 master could be corrupt but a more likely cause of this is a bug",
unknown's avatar
unknown committed
3431
                    errmsg);
unknown's avatar
unknown committed
3432 3433
    my_free((char*) tmp_buf, MYF(MY_ALLOW_ZERO_PTR));
    DBUG_RETURN(1);
3434
  }
3435

3436
  pthread_mutex_lock(&mi->data_lock);
3437
  ev->log_pos= mi->master_log_pos; /* 3.23 events don't contain log_pos */
unknown's avatar
unknown committed
3438
  switch (ev->get_type_code()) {
unknown's avatar
unknown committed
3439
  case STOP_EVENT:
3440
    ignore_event= 1;
unknown's avatar
unknown committed
3441 3442
    inc_pos= event_len;
    break;
3443
  case ROTATE_EVENT:
3444
    if (unlikely(process_io_rotate(mi,(Rotate_log_event*)ev)))
3445 3446
    {
      delete ev;
3447
      pthread_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
3448
      DBUG_RETURN(1);
3449
    }
unknown's avatar
unknown committed
3450
    inc_pos= 0;
3451
    break;
unknown's avatar
unknown committed
3452
  case CREATE_FILE_EVENT:
3453 3454 3455 3456 3457 3458
    /*
      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
3459
  {
unknown's avatar
unknown committed
3460
    /* We come here when and only when tmp_buf != 0 */
3461
    DBUG_ASSERT(tmp_buf != 0);
3462 3463
    inc_pos=event_len;
    ev->log_pos+= inc_pos;
unknown's avatar
unknown committed
3464
    int error = process_io_create_file(mi,(Create_file_log_event*)ev);
3465
    delete ev;
3466
    mi->master_log_pos += inc_pos;
unknown's avatar
unknown committed
3467
    DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
3468
    pthread_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
3469
    my_free((char*)tmp_buf, MYF(0));
unknown's avatar
unknown committed
3470
    DBUG_RETURN(error);
unknown's avatar
unknown committed
3471
  }
3472
  default:
unknown's avatar
unknown committed
3473
    inc_pos= event_len;
3474 3475
    break;
  }
unknown's avatar
unknown committed
3476
  if (likely(!ignore_event))
3477
  {
unknown's avatar
unknown committed
3478 3479
    if (ev->log_pos)
      /*
3480 3481 3482 3483
         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
3484
    if (unlikely(rli->relay_log.append(ev)))
3485 3486 3487
    {
      delete ev;
      pthread_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
3488
      DBUG_RETURN(1);
3489
    }
unknown's avatar
unknown committed
3490
    rli->relay_log.harvest_bytes_written(&rli->log_space_total);
3491 3492
  }
  delete ev;
unknown's avatar
unknown committed
3493
  mi->master_log_pos+= inc_pos;
unknown's avatar
unknown committed
3494
  DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
3495
  pthread_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
3496
  DBUG_RETURN(0);
3497 3498
}

3499 3500 3501 3502
/*
  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.
*/
3503
static int queue_binlog_ver_3_event(Master_info *mi, const char *buf,
unknown's avatar
unknown committed
3504
                           ulong event_len)
3505 3506 3507 3508
{
  const char *errmsg = 0;
  ulong inc_pos;
  char *tmp_buf = 0;
3509
  Relay_log_info *rli= &mi->rli;
3510 3511 3512 3513 3514 3515 3516 3517 3518
  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
3519
                    errmsg);
3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549
    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
3550
  DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561
  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
3562
  TODO:
3563 3564 3565 3566
    Test this code before release - it has to be tested on a separate
    setup with 3.23 master or 4.0 master
*/

3567
static int queue_old_event(Master_info *mi, const char *buf,
unknown's avatar
unknown committed
3568
                           ulong event_len)
3569
{
3570 3571
  DBUG_ENTER("queue_old_event");

3572 3573 3574
  switch (mi->rli.relay_log.description_event_for_queue->binlog_version)
  {
  case 1:
3575
      DBUG_RETURN(queue_binlog_ver_1_event(mi,buf,event_len));
3576
  case 3:
3577
      DBUG_RETURN(queue_binlog_ver_3_event(mi,buf,event_len));
3578 3579
  default: /* unsupported format; eg version 2 */
    DBUG_PRINT("info",("unsupported binlog format %d in queue_old_event()",
unknown's avatar
unknown committed
3580
                       mi->rli.relay_log.description_event_for_queue->binlog_version));
3581
    DBUG_RETURN(1);
3582 3583
  }
}
3584

unknown's avatar
unknown committed
3585
/*
3586 3587
  queue_event()

3588 3589 3590 3591 3592
  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
3593 3594
*/

3595
static int queue_event(Master_info* mi,const char* buf, ulong event_len)
3596
{
unknown's avatar
unknown committed
3597
  int error= 0;
Andrei Elkin's avatar
Andrei Elkin committed
3598
  String error_msg;
unknown's avatar
unknown committed
3599
  ulong inc_pos;
3600
  Relay_log_info *rli= &mi->rli;
3601
  pthread_mutex_t *log_lock= rli->relay_log.get_log_lock();
unknown's avatar
unknown committed
3602 3603
  DBUG_ENTER("queue_event");

unknown's avatar
unknown committed
3604 3605
  LINT_INIT(inc_pos);

3606 3607
  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
3608
    DBUG_RETURN(queue_old_event(mi,buf,event_len));
3609

3610
  LINT_INIT(inc_pos);
3611
  pthread_mutex_lock(&mi->data_lock);
unknown's avatar
unknown committed
3612

unknown's avatar
unknown committed
3613
  switch (buf[EVENT_TYPE_OFFSET]) {
3614
  case STOP_EVENT:
3615 3616
    /*
      We needn't write this event to the relay log. Indeed, it just indicates a
unknown's avatar
unknown committed
3617 3618 3619
      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
3620
      prepared statements' deletion are TODO only when we binlog prep stmts).
unknown's avatar
unknown committed
3621

unknown's avatar
unknown committed
3622 3623 3624 3625
      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).
3626 3627
    */
    goto err;
3628 3629
  case ROTATE_EVENT:
  {
unknown's avatar
unknown committed
3630
    Rotate_log_event rev(buf,event_len,mi->rli.relay_log.description_event_for_queue);
3631
    if (unlikely(process_io_rotate(mi,&rev)))
unknown's avatar
unknown committed
3632
    {
Andrei Elkin's avatar
Andrei Elkin committed
3633
      error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
3634
      goto err;
unknown's avatar
unknown committed
3635
    }
3636 3637 3638 3639
    /*
      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
3640
    inc_pos= 0;
3641 3642
    break;
  }
3643 3644 3645 3646 3647 3648 3649
  case FORMAT_DESCRIPTION_EVENT:
  {
    /*
      Create an event, and save it (when we rotate the relay log, we will have
      to write this event again).
    */
    /*
3650 3651 3652
      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).
3653
    */
3654
    Format_description_log_event* tmp;
3655
    const char* errmsg;
3656
    if (!(tmp= (Format_description_log_event*)
3657
          Log_event::read_log_event(buf, event_len, &errmsg,
3658
                                    mi->rli.relay_log.description_event_for_queue)))
3659
    {
Andrei Elkin's avatar
Andrei Elkin committed
3660
      error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
3661 3662
      goto err;
    }
3663 3664
    delete mi->rli.relay_log.description_event_for_queue;
    mi->rli.relay_log.description_event_for_queue= tmp;
unknown's avatar
unknown committed
3665
    /*
3666
       Though this does some conversion to the slave's format, this will
unknown's avatar
unknown committed
3667
       preserve the master's binlog format version, and number of event types.
3668
    */
unknown's avatar
unknown committed
3669
    /*
3670
       If the event was not requested by the slave (the slave did not ask for
unknown's avatar
unknown committed
3671
       it), i.e. has end_log_pos=0, we do not increment mi->master_log_pos
3672 3673 3674
    */
    inc_pos= uint4korr(buf+LOG_POS_OFFSET) ? event_len : 0;
    DBUG_PRINT("info",("binlog format is now %d",
unknown's avatar
unknown committed
3675
                       mi->rli.relay_log.description_event_for_queue->binlog_version));
3676 3677 3678

  }
  break;
Andrei Elkin's avatar
Andrei Elkin committed
3679 3680 3681 3682 3683 3684 3685 3686 3687 3688 3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728

  case HEARTBEAT_LOG_EVENT:
  {
    /*
      HB (heartbeat) cannot come before RL (Relay)
    */
    char  llbuf[22];
    Heartbeat_log_event hb(buf, event_len, mi->rli.relay_log.description_event_for_queue);
    if (!hb.is_valid())
    {
      error= ER_SLAVE_HEARTBEAT_FAILURE;
      error_msg.append(STRING_WITH_LEN("inconsistent heartbeat event content;"));
      error_msg.append(STRING_WITH_LEN("the event's data: log_file_name "));
      error_msg.append(hb.get_log_ident(), (uint) strlen(hb.get_log_ident()));
      error_msg.append(STRING_WITH_LEN(" log_pos "));
      llstr(hb.log_pos, llbuf);
      error_msg.append(llbuf, strlen(llbuf));
      goto err;
    }
    mi->received_heartbeats++;
    /* 
       compare local and event's versions of log_file, log_pos.
       
       Heartbeat is sent only after an event corresponding to the corrdinates
       the heartbeat carries.
       Slave can not have a difference in coordinates except in the only
       special case when mi->master_log_name, master_log_pos have never
       been updated by Rotate event i.e when slave does not have any history
       with the master (and thereafter mi->master_log_pos is NULL).

       TODO: handling `when' for SHOW SLAVE STATUS' snds behind
    */
    if ((memcmp(mi->master_log_name, hb.get_log_ident(), hb.get_ident_len())
         && mi->master_log_name != NULL)
        || mi->master_log_pos != hb.log_pos)
    {
      /* missed events of heartbeat from the past */
      error= ER_SLAVE_HEARTBEAT_FAILURE;
      error_msg.append(STRING_WITH_LEN("heartbeat is not compatible with local info;"));
      error_msg.append(STRING_WITH_LEN("the event's data: log_file_name "));
      error_msg.append(hb.get_log_ident(), (uint) strlen(hb.get_log_ident()));
      error_msg.append(STRING_WITH_LEN(" log_pos "));
      llstr(hb.log_pos, llbuf);
      error_msg.append(llbuf, strlen(llbuf));
      goto err;
    }
    goto skip_relay_logging;
  }
  break;
    
3729
  default:
unknown's avatar
unknown committed
3730
    inc_pos= event_len;
3731 3732
    break;
  }
3733

unknown's avatar
unknown committed
3734 3735
  /*
     If this event is originating from this server, don't queue it.
3736
     We don't check this for 3.23 events because it's simpler like this; 3.23
unknown's avatar
unknown committed
3737 3738
     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
3739 3740 3741 3742 3743 3744 3745 3746
     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!).
  */

3747 3748
  pthread_mutex_lock(log_lock);

unknown's avatar
unknown committed
3749
  if ((uint4korr(buf + SERVER_ID_OFFSET) == ::server_id) &&
3750
      !mi->rli.replicate_same_server_id)
3751
  {
3752 3753
    /*
      Do not write it to the relay log.
3754 3755 3756 3757 3758 3759 3760 3761
      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.
3762 3763 3764
      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.
3765
    */
3766 3767 3768
    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
3769
    {
3770
      mi->master_log_pos+= inc_pos;
unknown's avatar
unknown committed
3771 3772 3773 3774
      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;
    }
3775
    rli->relay_log.signal_update(); // the slave SQL thread needs to re-check
unknown's avatar
unknown committed
3776
    DBUG_PRINT("info", ("master_log_pos: %lu, event originating from the same server, ignored",
unknown's avatar
unknown committed
3777
                        (ulong) mi->master_log_pos));
unknown's avatar
unknown committed
3778
  }
unknown's avatar
unknown committed
3779 3780 3781
  else
  {
    /* write the event to the relay log */
3782
    if (likely(!(rli->relay_log.appendv(buf,event_len,0))))
3783 3784
    {
      mi->master_log_pos+= inc_pos;
unknown's avatar
unknown committed
3785
      DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
3786 3787
      rli->relay_log.harvest_bytes_written(&rli->log_space_total);
    }
3788
    else
Andrei Elkin's avatar
Andrei Elkin committed
3789 3790 3791
    {
      error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
    }
3792
    rli->ign_master_log_name_end[0]= 0; // last event is not ignored
unknown's avatar
unknown committed
3793
  }
3794
  pthread_mutex_unlock(log_lock);
3795

Andrei Elkin's avatar
Andrei Elkin committed
3796 3797
skip_relay_logging:
  
3798
err:
3799
  pthread_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
3800
  DBUG_PRINT("info", ("error: %d", error));
Andrei Elkin's avatar
Andrei Elkin committed
3801 3802 3803 3804 3805
  if (error)
    mi->report(ERROR_LEVEL, error, ER(error), 
               (error == ER_SLAVE_RELAY_LOG_WRITE_FAILURE)?
               "could not queue event from master" :
               error_msg.ptr());
unknown's avatar
unknown committed
3806
  DBUG_RETURN(error);
3807 3808
}

3809

3810
void end_relay_log_info(Relay_log_info* rli)
3811
{
3812 3813
  DBUG_ENTER("end_relay_log_info");

3814
  if (!rli->inited)
3815
    DBUG_VOID_RETURN;
3816
  if (rli->info_fd >= 0)
unknown's avatar
unknown committed
3817 3818
  {
    end_io_cache(&rli->info_file);
3819
    (void) my_close(rli->info_fd, MYF(MY_WME));
unknown's avatar
unknown committed
3820 3821
    rli->info_fd = -1;
  }
3822
  if (rli->cur_log_fd >= 0)
unknown's avatar
unknown committed
3823 3824 3825 3826 3827
  {
    end_io_cache(&rli->cache_buf);
    (void)my_close(rli->cur_log_fd, MYF(MY_WME));
    rli->cur_log_fd = -1;
  }
3828
  rli->inited = 0;
3829
  rli->relay_log.close(LOG_CLOSE_INDEX | LOG_CLOSE_STOP_EVENT);
3830
  rli->relay_log.harvest_bytes_written(&rli->log_space_total);
unknown's avatar
unknown committed
3831 3832 3833 3834 3835 3836
  /*
    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();
3837
  DBUG_VOID_RETURN;
3838 3839
}

3840 3841 3842 3843 3844 3845 3846 3847 3848 3849 3850 3851 3852 3853 3854 3855 3856 3857 3858 3859 3860 3861 3862 3863 3864

/**
  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;
  if (thd->slave_thread)
    thd->clear_active_vio();
#endif
}


unknown's avatar
unknown committed
3865 3866
/*
  Try to connect until successful or slave killed
3867

unknown's avatar
unknown committed
3868 3869
  SYNPOSIS
    safe_connect()
unknown's avatar
unknown committed
3870 3871 3872
    thd                 Thread handler for slave
    mysql               MySQL connection handle
    mi                  Replication handle
3873

unknown's avatar
unknown committed
3874
  RETURN
unknown's avatar
unknown committed
3875 3876
    0   ok
    #   Error
unknown's avatar
unknown committed
3877
*/
3878

3879
static int safe_connect(THD* thd, MYSQL* mysql, Master_info* mi)
unknown's avatar
unknown committed
3880
{
3881 3882 3883
  DBUG_ENTER("safe_connect");

  DBUG_RETURN(connect_to_master(thd, mysql, mi, 0, 0));
unknown's avatar
unknown committed
3884 3885
}

3886

3887
/*
unknown's avatar
unknown committed
3888 3889
  SYNPOSIS
    connect_to_master()
unknown's avatar
unknown committed
3890

unknown's avatar
unknown committed
3891 3892 3893
  IMPLEMENTATION
    Try to connect until successful or slave killed or we have retried
    master_retry_count times
3894
*/
unknown's avatar
unknown committed
3895

3896
static int connect_to_master(THD* thd, MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
3897
                             bool reconnect, bool suppress_warnings)
unknown's avatar
unknown committed
3898
{
3899
  int slave_was_killed;
unknown's avatar
unknown committed
3900
  int last_errno= -2;                           // impossible error
3901
  ulong err_count=0;
unknown's avatar
unknown committed
3902
  char llbuff[22];
3903
  DBUG_ENTER("connect_to_master");
unknown's avatar
unknown committed
3904

unknown's avatar
unknown committed
3905
#ifndef DBUG_OFF
3906
  mi->events_till_disconnect = disconnect_slave_event_count;
unknown's avatar
unknown committed
3907
#endif
3908
  ulong client_flag= CLIENT_REMEMBER_OPTIONS;
3909
  if (opt_slave_compressed_protocol)
unknown's avatar
unknown committed
3910
    client_flag=CLIENT_COMPRESS;                /* We will use compression */
3911

3912 3913
  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
3914

unknown's avatar
unknown committed
3915 3916
#ifdef HAVE_OPENSSL
  if (mi->ssl)
3917
  {
unknown's avatar
unknown committed
3918
    mysql_ssl_set(mysql,
unknown's avatar
unknown committed
3919
                  mi->ssl_key[0]?mi->ssl_key:0,
unknown's avatar
unknown committed
3920
                  mi->ssl_cert[0]?mi->ssl_cert:0,
unknown's avatar
unknown committed
3921 3922 3923
                  mi->ssl_ca[0]?mi->ssl_ca:0,
                  mi->ssl_capath[0]?mi->ssl_capath:0,
                  mi->ssl_cipher[0]?mi->ssl_cipher:0);
3924 3925 3926
    mysql_options(mysql, MYSQL_OPT_SSL_VERIFY_SERVER_CERT,
                  &mi->ssl_verify_server_cert);
  }
unknown's avatar
unknown committed
3927 3928
#endif

3929 3930 3931 3932
  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);

3933
  while (!(slave_was_killed = io_slave_killed(thd,mi)) &&
unknown's avatar
unknown committed
3934 3935 3936
         (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
3937
  {
3938
    /* Don't repeat last error */
unknown's avatar
SCRUM  
unknown committed
3939
    if ((int)mysql_errno(mysql) != last_errno)
3940
    {
unknown's avatar
SCRUM  
unknown committed
3941
      last_errno=mysql_errno(mysql);
unknown's avatar
unknown committed
3942
      suppress_warnings= 0;
3943 3944 3945 3946 3947 3948
      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);
3949
    }
unknown's avatar
unknown committed
3950 3951 3952
    /*
      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
3953
      do not want to have election triggered on the first failure to
unknown's avatar
unknown committed
3954
      connect
3955
    */
3956
    if (++err_count == master_retry_count)
3957 3958
    {
      slave_was_killed=1;
unknown's avatar
unknown committed
3959 3960
      if (reconnect)
        change_rpl_status(RPL_ACTIVE_SLAVE,RPL_LOST_SOLDIER);
3961 3962
      break;
    }
3963
    safe_sleep(thd,mi->connect_retry,(CHECK_KILLED_FUNC)io_slave_killed,
unknown's avatar
unknown committed
3964
               (void*)mi);
unknown's avatar
unknown committed
3965
  }
3966

3967 3968
  if (!slave_was_killed)
  {
3969
    mi->clear_error(); // clear possible left over reconnect error
unknown's avatar
unknown committed
3970
    if (reconnect)
unknown's avatar
unknown committed
3971
    {
3972
      if (!suppress_warnings && global_system_variables.log_warnings)
unknown's avatar
unknown committed
3973
        sql_print_information("Slave: connected to master '%s@%s:%d',\
3974
replication resumed in log '%s' at position %s", mi->user,
unknown's avatar
unknown committed
3975 3976 3977
                        mi->host, mi->port,
                        IO_RPL_LOG_NAME,
                        llstr(mi->master_log_pos,llbuff));
unknown's avatar
unknown committed
3978
    }
unknown's avatar
unknown committed
3979 3980 3981
    else
    {
      change_rpl_status(RPL_IDLE_SLAVE,RPL_ACTIVE_SLAVE);
3982 3983
      general_log_print(thd, COM_CONNECT_OUT, "%s@%s:%d",
                        mi->user, mi->host, mi->port);
unknown's avatar
unknown committed
3984
    }
3985
#ifdef SIGNAL_WITH_VIO_CLOSE
3986
    thd->set_active_vio(mysql->net.vio);
unknown's avatar
unknown committed
3987
#endif
3988
  }
3989
  mysql->reconnect= 1;
3990 3991
  DBUG_PRINT("exit",("slave_was_killed: %d", slave_was_killed));
  DBUG_RETURN(slave_was_killed);
unknown's avatar
unknown committed
3992 3993
}

3994

unknown's avatar
unknown committed
3995
/*
3996
  safe_reconnect()
unknown's avatar
unknown committed
3997

unknown's avatar
unknown committed
3998 3999 4000
  IMPLEMENTATION
    Try to connect until successful or slave killed or we have retried
    master_retry_count times
unknown's avatar
unknown committed
4001 4002
*/

4003
static int safe_reconnect(THD* thd, MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
4004
                          bool suppress_warnings)
unknown's avatar
unknown committed
4005
{
unknown's avatar
unknown committed
4006 4007
  DBUG_ENTER("safe_reconnect");
  DBUG_RETURN(connect_to_master(thd, mysql, mi, 1, suppress_warnings));
unknown's avatar
unknown committed
4008 4009
}

unknown's avatar
unknown committed
4010

4011 4012 4013 4014 4015 4016
/*
  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
4017
    rli                 Relay log information
4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035

  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
4036 4037
    0   ok
    1   write error
4038 4039
*/

4040
bool flush_relay_log_info(Relay_log_info* rli)
4041
{
4042
  bool error=0;
4043
  DBUG_ENTER("flush_relay_log_info");
4044 4045

  if (unlikely(rli->no_storage))
4046
    DBUG_RETURN(0);
4047

4048 4049 4050
  IO_CACHE *file = &rli->info_file;
  char buff[FN_REFLEN*2+22*2+4], *pos;

4051
  my_b_seek(file, 0L);
4052
  pos=strmov(buff, rli->group_relay_log_name);
4053
  *pos++='\n';
4054
  pos=longlong2str(rli->group_relay_log_pos, pos, 10);
4055
  *pos++='\n';
4056
  pos=strmov(pos, rli->group_master_log_name);
4057
  *pos++='\n';
4058
  pos=longlong2str(rli->group_master_log_pos, pos, 10);
4059
  *pos='\n';
4060
  if (my_b_write(file, (uchar*) buff, (size_t) (pos-buff)+1))
4061 4062 4063
    error=1;
  if (flush_io_cache(file))
    error=1;
4064

4065
  /* Flushing the relay log is done by the slave I/O thread */
4066
  DBUG_RETURN(error);
4067 4068
}

4069

unknown's avatar
unknown committed
4070
/*
4071
  Called when we notice that the current "hot" log got rotated under our feet.
unknown's avatar
unknown committed
4072 4073
*/

4074
static IO_CACHE *reopen_relay_log(Relay_log_info *rli, const char **errmsg)
4075
{
4076
  DBUG_ENTER("reopen_relay_log");
4077 4078
  DBUG_ASSERT(rli->cur_log != &rli->cache_buf);
  DBUG_ASSERT(rli->cur_log_fd == -1);
unknown's avatar
unknown committed
4079 4080

  IO_CACHE *cur_log = rli->cur_log=&rli->cache_buf;
4081
  if ((rli->cur_log_fd=open_binlog(cur_log,rli->event_relay_log_name,
unknown's avatar
unknown committed
4082
                                   errmsg)) <0)
unknown's avatar
unknown committed
4083
    DBUG_RETURN(0);
4084 4085
  /*
    We want to start exactly where we was before:
unknown's avatar
unknown committed
4086 4087
    relay_log_pos       Current log pos
    pending             Number of bytes already processed from the event
4088
  */
4089
  rli->event_relay_log_pos= max(rli->event_relay_log_pos, BIN_LOG_HEADER_SIZE);
4090
  my_b_seek(cur_log,rli->event_relay_log_pos);
unknown's avatar
unknown committed
4091
  DBUG_RETURN(cur_log);
4092 4093
}

unknown's avatar
unknown committed
4094

4095 4096 4097 4098 4099 4100 4101 4102 4103 4104
/**
  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.
*/
4105
static Log_event* next_event(Relay_log_info* rli)
4106 4107 4108
{
  Log_event* ev;
  IO_CACHE* cur_log = rli->cur_log;
4109
  pthread_mutex_t *log_lock = rli->relay_log.get_log_lock();
4110 4111
  const char* errmsg=0;
  THD* thd = rli->sql_thd;
unknown's avatar
unknown committed
4112
  DBUG_ENTER("next_event");
4113

4114 4115
  DBUG_ASSERT(thd != 0);

4116 4117 4118 4119 4120
#ifndef DBUG_OFF
  if (abort_slave_event_count && !rli->events_till_abort--)
    DBUG_RETURN(0);
#endif

unknown's avatar
unknown committed
4121 4122
  /*
    For most operations we need to protect rli members with data_lock,
4123 4124 4125 4126
    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
4127
  */
4128
  safe_mutex_assert_owner(&rli->data_lock);
unknown's avatar
unknown committed
4129

4130
  while (!sql_slave_killed(thd,rli))
unknown's avatar
unknown committed
4131 4132 4133
  {
    /*
      We can have two kinds of log reading:
unknown's avatar
unknown committed
4134 4135 4136 4137 4138 4139 4140 4141
      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
4142
    */
4143 4144 4145 4146 4147
    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
4148 4149

      /*
unknown's avatar
unknown committed
4150 4151
        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
4152
      */
unknown's avatar
unknown committed
4153
      if (rli->relay_log.get_open_count() != rli->cur_log_old_open_count)
4154
      {
unknown's avatar
unknown committed
4155 4156 4157 4158 4159 4160
        // 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
4161 4162
      }
    }
4163 4164 4165 4166 4167 4168 4169
    /* 
      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;
4170 4171
#ifndef DBUG_OFF
    {
4172
      /* This is an assertion which sometimes fails, let's try to track it */
4173
      char llbuf1[22], llbuf2[22];
4174
      DBUG_PRINT("info", ("my_b_tell(cur_log)=%s rli->event_relay_log_pos=%s",
4175
                          llstr(my_b_tell(cur_log),llbuf1),
4176 4177 4178
                          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);
4179 4180
    }
#endif
unknown's avatar
unknown committed
4181 4182
    /*
      Relay log is always in new format - if the master is 3.23, the
4183
      I/O thread will convert the format for us.
4184 4185
      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
4186 4187
      logs, which may even have been deleted. So we need to write this
      description event at the beginning of the relay log.
4188 4189
      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.
4190
      But if the relay log is created by new_file(): then the solution is:
4191
      MYSQL_BIN_LOG::open() will write the buffered description event.
unknown's avatar
unknown committed
4192
    */
4193 4194
    if ((ev=Log_event::read_log_event(cur_log,0,
                                      rli->relay_log.description_event_for_exec)))
4195

4196 4197
    {
      DBUG_ASSERT(thd==rli->sql_thd);
4198 4199 4200 4201 4202
      /*
        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);
4203
      if (hot_log)
unknown's avatar
unknown committed
4204
        pthread_mutex_unlock(log_lock);
unknown's avatar
unknown committed
4205
      DBUG_RETURN(ev);
4206 4207
    }
    DBUG_ASSERT(thd==rli->sql_thd);
unknown's avatar
unknown committed
4208
    if (opt_reckless_slave)                     // For mysql-test
unknown's avatar
unknown committed
4209
      cur_log->error = 0;
unknown's avatar
unknown committed
4210
    if (cur_log->error < 0)
unknown's avatar
unknown committed
4211 4212
    {
      errmsg = "slave SQL thread aborted because of I/O error";
unknown's avatar
unknown committed
4213
      if (hot_log)
unknown's avatar
unknown committed
4214
        pthread_mutex_unlock(log_lock);
unknown's avatar
unknown committed
4215 4216
      goto err;
    }
4217 4218
    if (!cur_log->error) /* EOF */
    {
unknown's avatar
unknown committed
4219
      /*
unknown's avatar
unknown committed
4220 4221 4222
        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
4223
      */
4224 4225
      if (hot_log)
      {
4226 4227 4228 4229 4230 4231 4232 4233 4234 4235
        /*
          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").
4236 4237 4238 4239 4240 4241 4242 4243 4244 4245

          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.
4246
        */
4247
        time_t save_timestamp= rli->last_master_timestamp;
4248 4249
        rli->last_master_timestamp= 0;

unknown's avatar
unknown committed
4250
        DBUG_ASSERT(rli->relay_log.get_open_count() ==
unknown's avatar
unknown committed
4251
                    rli->cur_log_old_open_count);
4252 4253 4254 4255 4256

        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"));
4257
          ev= new Rotate_log_event(rli->ign_master_log_name_end,
4258
                                   0, rli->ign_master_log_pos_end,
unknown's avatar
unknown committed
4259
                                   Rotate_log_event::DUP_NAME);
4260
          rli->ign_master_log_name_end[0]= 0;
unknown's avatar
unknown committed
4261
          pthread_mutex_unlock(log_lock);
4262 4263 4264 4265 4266 4267 4268 4269 4270 4271
          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
4272 4273 4274 4275 4276
        /*
          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);
4277 4278

        /*
unknown's avatar
unknown committed
4279
          Possible deadlock :
4280 4281 4282 4283 4284 4285 4286 4287 4288 4289 4290
          - 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
4291
          master does not send anything). Then the I/O thread stops waiting
4292 4293
          and reads more events.
          The SQL thread decides when the I/O thread should take log_space_limit
unknown's avatar
unknown committed
4294
          into account again : ignore_log_space_limit is reset to 0
4295 4296 4297 4298
          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
4299 4300
          be stopped, and the SQL thread sets ignore_log_space_limit to 0 when
          it stops.
4301 4302 4303
        */
        pthread_mutex_lock(&rli->log_space_lock);
        // prevent the I/O thread from blocking next times
unknown's avatar
unknown committed
4304
        rli->ignore_log_space_limit= 1;
4305
        /*
4306 4307 4308 4309
          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.
4310
        */
4311
        pthread_mutex_unlock(&rli->log_space_lock);
4312
        pthread_cond_broadcast(&rli->log_space_cond);
Andrei Elkin's avatar
Andrei Elkin committed
4313 4314
        // Note that wait_for_update_relay_log unlocks lock_log !
        rli->relay_log.wait_for_update_relay_log(rli->sql_thd);
4315 4316
        // re-acquire data lock since we released it earlier
        pthread_mutex_lock(&rli->data_lock);
4317
        rli->last_master_timestamp= save_timestamp;
unknown's avatar
unknown committed
4318
        continue;
4319
      }
unknown's avatar
unknown committed
4320
      /*
unknown's avatar
unknown committed
4321 4322 4323
        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
4324 4325 4326 4327 4328
      */
      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
4329

4330
      if (relay_log_purge)
unknown's avatar
unknown committed
4331
      {
unknown's avatar
unknown committed
4332
        /*
4333 4334 4335 4336 4337 4338 4339 4340 4341 4342
          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
4343
        if (rli->relay_log.purge_first_log
4344 4345 4346
            (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
4347 4348 4349 4350
        {
          errmsg = "Error purging processed logs";
          goto err;
        }
unknown's avatar
unknown committed
4351
      }
4352 4353
      else
      {
unknown's avatar
unknown committed
4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365 4366 4367 4368 4369
        /*
          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
4370
      }
4371 4372 4373 4374 4375 4376 4377 4378 4379 4380 4381 4382 4383 4384

      /*
        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
4385 4386
      if (rli->relay_log.is_active(rli->linfo.log_file_name))
      {
4387
#ifdef EXTRA_DEBUG
unknown's avatar
unknown committed
4388 4389
        if (global_system_variables.log_warnings)
          sql_print_information("next log '%s' is currently active",
4390
                                rli->linfo.log_file_name);
unknown's avatar
unknown committed
4391 4392 4393 4394 4395 4396 4397 4398
#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.
4399 4400 4401
          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
4402 4403
        */
        if (check_binlog_magic(cur_log,&errmsg))
4404 4405
        {
          if (!hot_log) pthread_mutex_unlock(log_lock);
unknown's avatar
unknown committed
4406
          goto err;
4407 4408
        }
        if (!hot_log) pthread_mutex_unlock(log_lock);
unknown's avatar
unknown committed
4409
        continue;
4410
      }
4411
      if (!hot_log) pthread_mutex_unlock(log_lock);
unknown's avatar
unknown committed
4412
      /*
unknown's avatar
unknown committed
4413 4414 4415
        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
4416 4417
      */
#ifdef EXTRA_DEBUG
unknown's avatar
unknown committed
4418
      if (global_system_variables.log_warnings)
unknown's avatar
unknown committed
4419
        sql_print_information("next log '%s' is not active",
4420
                              rli->linfo.log_file_name);
unknown's avatar
unknown committed
4421
#endif
unknown's avatar
unknown committed
4422 4423
      // 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
4424 4425
                                       &errmsg)) <0)
        goto err;
4426
    }
unknown's avatar
unknown committed
4427
    else
4428
    {
unknown's avatar
unknown committed
4429
      /*
unknown's avatar
unknown committed
4430 4431
        Read failed with a non-EOF error.
        TODO: come up with something better to handle this error
unknown's avatar
unknown committed
4432 4433
      */
      if (hot_log)
unknown's avatar
unknown committed
4434
        pthread_mutex_unlock(log_lock);
4435
      sql_print_error("Slave SQL thread: I/O error reading \
unknown's avatar
unknown committed
4436
event(errno: %d  cur_log->error: %d)",
unknown's avatar
unknown committed
4437
                      my_errno,cur_log->error);
unknown's avatar
unknown committed
4438
      // set read position to the beginning of the event
4439
      my_b_seek(cur_log,rli->event_relay_log_pos);
unknown's avatar
unknown committed
4440 4441
      /* otherwise, we have had a partial read */
      errmsg = "Aborting slave SQL thread because of partial event read";
unknown's avatar
unknown committed
4442
      break;                                    // To end of function
4443 4444
    }
  }
4445
  if (!errmsg && global_system_variables.log_warnings)
4446
  {
unknown's avatar
unknown committed
4447
    sql_print_information("Error reading relay log event: %s",
4448 4449 4450
                          "slave SQL thread was killed");
    DBUG_RETURN(0);
  }
unknown's avatar
unknown committed
4451

4452
err:
4453 4454
  if (errmsg)
    sql_print_error("Error reading relay log event: %s", errmsg);
unknown's avatar
unknown committed
4455
  DBUG_RETURN(0);
4456 4457
}

4458 4459 4460
/*
  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
4461
  locks; here we don't, so this function is mainly taking locks).
4462 4463
  Returns nothing as we cannot catch any error (MYSQL_BIN_LOG::new_file()
  is void).
4464 4465
*/

4466
void rotate_relay_log(Master_info* mi)
4467 4468
{
  DBUG_ENTER("rotate_relay_log");
4469
  Relay_log_info* rli= &mi->rli;
unknown's avatar
unknown committed
4470

4471 4472 4473
  /* We don't lock rli->run_lock. This would lead to deadlocks. */
  pthread_mutex_lock(&mi->run_lock);

unknown's avatar
unknown committed
4474
  /*
unknown's avatar
unknown committed
4475 4476 4477
     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).
  */
4478 4479
  if (!rli->inited)
  {
unknown's avatar
unknown committed
4480
    DBUG_PRINT("info", ("rli->inited == 0"));
unknown's avatar
unknown committed
4481
    goto end;
4482
  }
unknown's avatar
unknown committed
4483

4484
  /* If the relay log is closed, new_file() will do nothing. */
4485
  rli->relay_log.new_file();
unknown's avatar
unknown committed
4486

4487 4488 4489 4490
  /*
    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
4491 4492 4493 4494 4495 4496
    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.
4497 4498 4499 4500
    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
4501
end:
4502
  pthread_mutex_unlock(&mi->run_lock);
4503 4504 4505
  DBUG_VOID_RETURN;
}

unknown's avatar
unknown committed
4506

unknown's avatar
unknown committed
4507 4508 4509
/**
   Detects, based on master's version (as found in the relay log), if master
   has a certain bug.
4510
   @param rli Relay_log_info which tells the master's version
unknown's avatar
unknown committed
4511
   @param bug_id Number of the bug as found in bugs.mysql.com
4512
   @param report bool report error message, default TRUE
4513 4514 4515 4516 4517 4518 4519

   @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
4520 4521
   @return TRUE if master has the bug, FALSE if it does not.
*/
4522 4523
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
4524 4525 4526 4527 4528 4529 4530 4531
{
  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
4532
    {24432, { 5, 0, 24 }, { 5, 0, 38 } },
4533 4534 4535
    {24432, { 5, 1, 12 }, { 5, 1, 17 } },
    {33029, { 5, 0,  0 }, { 5, 0, 58 } },
    {33029, { 5, 1,  0 }, { 5, 1, 12 } },
4536
    {37426, { 5, 1,  0 }, { 5, 1, 26 } },
unknown's avatar
unknown committed
4537 4538 4539 4540 4541 4542 4543 4544 4545 4546 4547 4548 4549
  };
  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) &&
4550 4551
        (memcmp(fixed_in,      master_ver, 3) >  0) &&
        (pred == NULL || (*pred)(param)))
unknown's avatar
unknown committed
4552
    {
4553 4554
      if (!report)
	return TRUE;
unknown's avatar
unknown committed
4555 4556 4557 4558 4559
      // 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);
4560
      // a verbose message for the error log
4561 4562 4563
      rli->report(ERROR_LEVEL, ER_UNKNOWN_ERROR,
                  "According to the master's version ('%s'),"
                  " it is probable that master suffers from this bug:"
4564 4565 4566 4567 4568 4569 4570 4571 4572 4573 4574 4575 4576 4577 4578 4579
                      " 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
4580 4581 4582 4583 4584 4585
      return TRUE;
    }
  }
  return FALSE;
}

4586 4587 4588 4589 4590 4591 4592 4593 4594 4595 4596 4597 4598 4599 4600
/**
   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;
4601
    DBUG_EXECUTE_IF("simulate_bug33029", return TRUE;);
4602
    return rpl_master_has_bug(rli, 33029, FALSE, NULL, NULL);
4603 4604 4605 4606
  }
  return FALSE;
}

4607
#ifdef HAVE_EXPLICIT_TEMPLATE_INSTANTIATION
unknown's avatar
unknown committed
4608
template class I_List_iterator<i_string>;
unknown's avatar
unknown committed
4609
template class I_List_iterator<i_string_pair>;
unknown's avatar
unknown committed
4610
#endif
4611

4612 4613 4614 4615
/**
  @} (end of group Replication)
*/

unknown's avatar
SCRUM  
unknown committed
4616
#endif /* HAVE_REPLICATION */