slave.cc 149 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"));
514 515 516 517 518 519 520 521 522 523 524 525 526 527

    pthread_mutex_lock(&thd->LOCK_delete);
#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);
    pthread_mutex_unlock(&thd->LOCK_delete);

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
}

773

unknown's avatar
unknown committed
774
/*
775 776
  skip_load_data_infile()

unknown's avatar
unknown committed
777 778 779
  NOTES
    This is used to tell a 3.23 master to break send_file()
*/
780 781 782

void skip_load_data_infile(NET *net)
{
783 784
  DBUG_ENTER("skip_load_data_infile");

785
  (void)net_request_file(net, "/dev/null");
unknown's avatar
unknown committed
786
  (void)my_net_read(net);                               // discard response
787
  (void)net_write_command(net, 0, (uchar*) "", 0, (uchar*) "", 0); // ok
788
  DBUG_VOID_RETURN;
789 790
}

791

792
bool net_request_file(NET* net, const char* fname)
793
{
794
  DBUG_ENTER("net_request_file");
795 796
  DBUG_RETURN(net_write_command(net, 251, (uchar*) fname, strlen(fname),
                                (uchar*) "", 0));
797 798
}

799 800
/*
  From other comments and tests in code, it looks like
801
  sometimes Query_log_event and Load_log_event can have db == 0
802 803 804
  (see rewrite_db() above for example)
  (cases where this happens are unclear; it may be when the master is 3.23).
*/
805 806

const char *print_slave_db_safe(const char* db)
807
{
808 809 810
  DBUG_ENTER("*print_slave_db_safe");

  DBUG_RETURN((db ? db : ""));
811
}
812

813
int init_strvar_from_file(char *var, int max_size, IO_CACHE *f,
unknown's avatar
unknown committed
814
                                 const char *default_val)
unknown's avatar
unknown committed
815
{
unknown's avatar
unknown committed
816
  uint length;
817 818
  DBUG_ENTER("init_strvar_from_file");

unknown's avatar
unknown committed
819 820 821 822 823 824
  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
825
    {
unknown's avatar
unknown committed
826
      /*
unknown's avatar
unknown committed
827 828
        If we truncated a line or stopped on last char, remove all chars
        up to and including newline.
unknown's avatar
unknown committed
829
      */
unknown's avatar
unknown committed
830
      int c;
831
      while (((c=my_b_get(f)) != '\n' && c != my_b_EOF)) ;
unknown's avatar
unknown committed
832
    }
833
    DBUG_RETURN(0);
unknown's avatar
unknown committed
834 835 836
  }
  else if (default_val)
  {
unknown's avatar
unknown committed
837
    strmake(var,  default_val, max_size-1);
838
    DBUG_RETURN(0);
unknown's avatar
unknown committed
839
  }
840
  DBUG_RETURN(1);
unknown's avatar
unknown committed
841 842
}

843

844
int init_intvar_from_file(int* var, IO_CACHE* f, int default_val)
unknown's avatar
unknown committed
845 846
{
  char buf[32];
847 848
  DBUG_ENTER("init_intvar_from_file");

unknown's avatar
unknown committed
849 850

  if (my_b_gets(f, buf, sizeof(buf)))
unknown's avatar
unknown committed
851 852
  {
    *var = atoi(buf);
853
    DBUG_RETURN(0);
unknown's avatar
unknown committed
854
  }
unknown's avatar
unknown committed
855
  else if (default_val)
unknown's avatar
unknown committed
856 857
  {
    *var = default_val;
858
    DBUG_RETURN(0);
unknown's avatar
unknown committed
859
  }
860
  DBUG_RETURN(1);
unknown's avatar
unknown committed
861 862
}

863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885

/*
  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;   
}


886 887 888 889 890 891 892 893
/*
  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
894

895 896 897
  RETURNS
  0       ok
  1       error
898
  2       transient network problem, the caller should try to reconnect
899
*/
900

901
static int get_master_version_and_clock(MYSQL* mysql, Master_info* mi)
902
{
903
  char err_buff[MAX_SLAVE_ERRMSG];
904
  const char* errmsg= 0;
905 906 907
  int err_code= 0;
  MYSQL_RES *master_res= 0;
  MYSQL_ROW master_row;
908
  DBUG_ENTER("get_master_version_and_clock");
909 910 911 912 913 914 915

  /*
    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
916

917
  if (!my_isdigit(&my_charset_bin,*mysql->server_version))
918
  {
unknown's avatar
unknown committed
919
    errmsg = "Master reported unrecognized MySQL version";
920 921 922
    err_code= ER_SLAVE_FATAL_ERROR;
    sprintf(err_buff, ER(err_code), errmsg);
  }
923 924 925 926 927
  else
  {
    /*
      Note the following switch will bug when we have MySQL branch 30 ;)
    */
unknown's avatar
unknown committed
928
    switch (*mysql->server_version)
929 930 931 932 933
    {
    case '0':
    case '1':
    case '2':
      errmsg = "Master reported unrecognized MySQL version";
934 935
      err_code= ER_SLAVE_FATAL_ERROR;
      sprintf(err_buff, ER(err_code), errmsg);
936 937 938
      break;
    case '3':
      mi->rli.relay_log.description_event_for_queue= new
unknown's avatar
unknown committed
939
        Format_description_log_event(1, mysql->server_version);
940 941 942
      break;
    case '4':
      mi->rli.relay_log.description_event_for_queue= new
unknown's avatar
unknown committed
943
        Format_description_log_event(3, mysql->server_version);
944
      break;
unknown's avatar
unknown committed
945
    default:
946 947 948 949 950 951 952 953 954
      /*
        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
955
        Format_description_log_event(4, mysql->server_version);
956 957
      break;
    }
958
  }
unknown's avatar
unknown committed
959 960

  /*
961 962 963 964 965
     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
966

967
  if (errmsg)
968
    goto err;
969 970 971 972

  /* as we are here, we tried to allocate the event */
  if (!mi->rli.relay_log.description_event_for_queue)
  {
973 974 975 976
    errmsg= "default Format_description_log_event";
    err_code= ER_SLAVE_CREATE_EVENT_FAILURE;
    sprintf(err_buff, ER(err_code), errmsg);
    goto err;
977 978
  }

979 980 981 982
  /*
    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
983

984 985
  DBUG_SYNC_POINT("debug_lock.before_get_UNIX_TIMESTAMP", 10);
  master_res= NULL;
986
  if (!mysql_real_query(mysql, STRING_WITH_LEN("SELECT UNIX_TIMESTAMP()")) &&
987 988
      (master_res= mysql_store_result(mysql)) &&
      (master_row= mysql_fetch_row(master_res)))
unknown's avatar
unknown committed
989
  {
unknown's avatar
unknown committed
990
    mi->clock_diff_with_master=
991
      (long) (time((time_t*) 0) - strtoul(master_row[0], 0, 10));
unknown's avatar
unknown committed
992
  }
993 994 995 996 997 998 999
  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
1000
  {
1001
    mi->clock_diff_with_master= 0; /* The "most sensible" value */
1002 1003 1004 1005
    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));
1006 1007
  }
  if (master_res)
1008
  {
unknown's avatar
unknown committed
1009
    mysql_free_result(master_res);
1010 1011
    master_res= NULL;
  }
unknown's avatar
unknown committed
1012

1013 1014 1015 1016 1017 1018 1019 1020 1021 1022
  /*
    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.
  */
1023 1024 1025
  DBUG_SYNC_POINT("debug_lock.before_get_SERVER_ID", 10);
  master_res= NULL;
  master_row= NULL;
1026 1027
  if (!mysql_real_query(mysql,
                        STRING_WITH_LEN("SHOW VARIABLES LIKE 'SERVER_ID'")) &&
1028 1029
      (master_res= mysql_store_result(mysql)) &&
      (master_row= mysql_fetch_row(master_res)))
1030
  {
1031
    if ((::server_id == strtoul(master_row[1], 0, 10)) &&
1032
        !mi->rli.replicate_same_server_id)
1033
    {
1034 1035 1036 1037
      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).";
1038 1039
      err_code= ER_SLAVE_FATAL_ERROR;
      sprintf(err_buff, ER(err_code), errmsg);
1040
      goto err;
1041
    }
1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065
  }
  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)
  {
1066
    mysql_free_result(master_res);
1067
    master_res= NULL;
unknown's avatar
unknown committed
1068 1069
  }

1070 1071 1072
  /*
    Check that the master's global character_set_server and ours are the same.
    Not fatal if query fails (old master?).
1073 1074 1075 1076 1077 1078
    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.
1079 1080 1081 1082 1083 1084
    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.
1085
  */
1086 1087 1088

  /* redundant with rest of code but safer against later additions */
  if (*mysql->server_version == '3')
1089
    goto err;
1090

1091
  if (*mysql->server_version == '4')
unknown's avatar
unknown committed
1092
  {
1093 1094 1095 1096 1097
    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)))
1098
    {
1099 1100 1101
      if (strcmp(master_row[0], global_system_variables.collation_server->name))
      {
        errmsg= "The slave I/O thread stops because master and slave have \
1102
different values for the COLLATION_SERVER global variable. The values must \
1103 1104 1105 1106 1107
be equal for the Statement-format replication to work";
        err_code= ER_SLAVE_FATAL_ERROR;
        sprintf(err_buff, ER(err_code), errmsg);
        goto err;
      }
1108
    }
1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121
    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));
1122
      goto err;
1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134
    }
    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
1135
  }
1136

1137 1138 1139 1140 1141 1142 1143 1144
  /*
    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
1145 1146 1147
    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
1148
    'SYSTEM' time zone).
1149 1150
    This check is only necessary for 4.x masters (and < 5.0.4 masters but
    those were alpha).
1151
  */
1152
  if (*mysql->server_version == '4')
unknown's avatar
unknown committed
1153
  {
1154 1155 1156 1157
    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)))
1158
    {
1159 1160 1161 1162
      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 \
1163
different values for the TIME_ZONE global variable. The values must \
1164 1165 1166 1167 1168
be equal for the Statement-format replication to work";
        err_code= ER_SLAVE_FATAL_ERROR;
        sprintf(err_buff, ER(err_code), errmsg);
        goto err;
      }
1169
    }
1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182
    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));
1183
      goto err;
1184 1185 1186 1187 1188 1189
    }
    if (master_res)
    {
      mysql_free_result(master_res);
      master_res= NULL;
    }
unknown's avatar
unknown committed
1190 1191
  }

1192
err:
1193 1194
  if (errmsg)
  {
1195 1196
    if (master_res)
      mysql_free_result(master_res);
1197 1198
    DBUG_ASSERT(err_code != 0);
    mi->report(ERROR_LEVEL, err_code, err_buff);
1199
    DBUG_RETURN(1);
1200
  }
1201

1202
  DBUG_RETURN(0);
1203 1204 1205 1206 1207

network_err:
  if (master_res)
    mysql_free_result(master_res);
  DBUG_RETURN(2);
1208 1209
}

1210 1211 1212 1213
/*
  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.
1214
  db must be non-zero (guarded by assertion).
1215 1216 1217 1218 1219

  RETURN VALUES
    0           success
    1           error
*/
1220

1221
static int create_table_from_dump(THD* thd, MYSQL *mysql, const char* db,
unknown's avatar
unknown committed
1222
                                  const char* table_name, bool overwrite)
unknown's avatar
unknown committed
1223
{
1224
  ulong packet_len;
1225 1226
  char *query, *save_db;
  uint32 save_db_length;
1227 1228
  Vio* save_vio;
  HA_CHECK_OPT check_opt;
unknown's avatar
unknown committed
1229
  TABLE_LIST tables;
1230 1231
  int error= 1;
  handler *file;
1232
  ulonglong save_options;
1233
  NET *net= &mysql->net;
1234
  const char *found_semicolon= NULL;
1235
  DBUG_ENTER("create_table_from_dump");
unknown's avatar
unknown committed
1236

1237
  packet_len= my_net_read(net); // read create table statement
1238 1239
  if (packet_len == packet_error)
  {
unknown's avatar
unknown committed
1240
    my_message(ER_MASTER_NET_READ, ER(ER_MASTER_NET_READ), MYF(0));
unknown's avatar
unknown committed
1241
    DBUG_RETURN(1);
1242 1243 1244
  }
  if (net->read_pos[0] == 255) // error from master
  {
unknown's avatar
unknown committed
1245
    char *err_msg;
1246
    err_msg= (char*) net->read_pos + ((mysql->server_capabilities &
unknown's avatar
unknown committed
1247 1248
                                       CLIENT_PROTOCOL_41) ?
                                      3+SQLSTATE_LENGTH+1 : 3);
1249
    my_error(ER_MASTER, MYF(0), err_msg);
unknown's avatar
unknown committed
1250
    DBUG_RETURN(1);
1251
  }
unknown's avatar
unknown committed
1252
  thd->command = COM_TABLE_DUMP;
1253
  thd->query_length= packet_len;
1254
  /* Note that we should not set thd->query until the area is initalized */
1255
  if (!(query = thd->strmake((char*) net->read_pos, packet_len)))
1256 1257
  {
    sql_print_error("create_table_from_dump: out of memory");
unknown's avatar
unknown committed
1258
    my_message(ER_GET_ERRNO, "Out of memory", MYF(0));
unknown's avatar
unknown committed
1259
    DBUG_RETURN(1);
1260
  }
1261
  thd->query= query;
1262
  thd->is_slave_error = 0;
1263

unknown's avatar
unknown committed
1264 1265
  bzero((char*) &tables,sizeof(tables));
  tables.db = (char*)db;
1266
  tables.alias= tables.table_name= (char*)table_name;
unknown's avatar
unknown committed
1267

unknown's avatar
unknown committed
1268
  /* Drop the table if 'overwrite' is true */
1269
  if (overwrite)
unknown's avatar
unknown committed
1270
  {
1271 1272 1273 1274 1275 1276 1277 1278 1279 1280
    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
1281 1282
  }

1283
  /* Create the table. We do not want to log the "create table" statement */
1284
  save_options = thd->options;
1285 1286
  thd->options &= ~ (OPTION_BIN_LOG);
  thd_proc_info(thd, "Creating table from master dump");
unknown's avatar
unknown committed
1287
  // save old db in case we are creating in a different database
1288
  save_db = thd->db;
1289
  save_db_length= thd->db_length;
1290
  thd->db = (char*)db;
1291
  DBUG_ASSERT(thd->db != 0);
1292
  thd->db_length= strlen(thd->db);
1293
  mysql_parse(thd, thd->query, packet_len, &found_semicolon); // run create table
unknown's avatar
unknown committed
1294
  thd->db = save_db;            // leave things the way the were before
1295
  thd->db_length= save_db_length;
1296
  thd->options = save_options;
unknown's avatar
unknown committed
1297

1298
  if (thd->is_slave_error)
unknown's avatar
unknown committed
1299
    goto err;                   // mysql_parse took care of the error send
unknown's avatar
unknown committed
1300

1301
  thd_proc_info(thd, "Opening master dump table");
1302
  thd->main_da.reset_diagnostics_area(); /* cleanup from CREATE_TABLE */
1303 1304 1305 1306 1307 1308
  /*
    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))
  */
1309
  tables.lock_type = TL_WRITE;
1310
  if (!open_ltable(thd, &tables, TL_WRITE, 0))
unknown's avatar
unknown committed
1311 1312
  {
    sql_print_error("create_table_from_dump: could not open created table");
1313
    goto err;
unknown's avatar
unknown committed
1314
  }
unknown's avatar
unknown committed
1315

1316
  file = tables.table->file;
1317
  thd_proc_info(thd, "Reading master dump table data");
1318
  /* Copy the data file */
unknown's avatar
unknown committed
1319 1320
  if (file->net_read_dump(net))
  {
unknown's avatar
unknown committed
1321
    my_message(ER_MASTER_NET_READ, ER(ER_MASTER_NET_READ), MYF(0));
1322
    sql_print_error("create_table_from_dump: failed in\
unknown's avatar
unknown committed
1323
 handler::net_read_dump()");
1324
    goto err;
unknown's avatar
unknown committed
1325
  }
unknown's avatar
unknown committed
1326 1327

  check_opt.init();
unknown's avatar
unknown committed
1328
  check_opt.flags|= T_VERY_SILENT | T_CALC_CHECKSUM | T_QUICK;
1329
  thd_proc_info(thd, "Rebuilding the index on master dump table");
unknown's avatar
unknown committed
1330 1331 1332 1333 1334
  /*
    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.
  */
1335
  save_vio = thd->net.vio;
unknown's avatar
unknown committed
1336
  thd->net.vio = 0;
1337
  /* Rebuild the index file from the copied data file (with REPAIR) */
unknown's avatar
unknown committed
1338
  error=file->ha_repair(thd,&check_opt) != 0;
unknown's avatar
unknown committed
1339
  thd->net.vio = save_vio;
1340
  if (error)
unknown's avatar
unknown committed
1341
    my_error(ER_INDEX_REBUILD, MYF(0), tables.table->s->table_name.str);
1342 1343

err:
unknown's avatar
unknown committed
1344
  close_thread_tables(thd);
unknown's avatar
unknown committed
1345
  DBUG_RETURN(error);
unknown's avatar
unknown committed
1346 1347
}

1348

1349
int fetch_master_table(THD *thd, const char *db_name, const char *table_name,
1350
                       Master_info *mi, MYSQL *mysql, bool overwrite)
unknown's avatar
unknown committed
1351
{
1352 1353 1354 1355 1356
  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
1357
                       db_name,table_name));
unknown's avatar
unknown committed
1358

unknown's avatar
merge  
unknown committed
1359
  if (!called_connected)
unknown's avatar
unknown committed
1360
  {
unknown's avatar
SCRUM  
unknown committed
1361
    if (!(mysql = mysql_init(NULL)))
1362 1363 1364
    {
      DBUG_RETURN(1);
    }
unknown's avatar
merge  
unknown committed
1365
    if (connect_to_master(thd, mysql, mi))
1366
    {
1367
      my_error(ER_CONNECT_TO_MASTER, MYF(0), mysql_error(mysql));
1368 1369 1370 1371 1372 1373 1374 1375 1376
      /*
        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
1377
      mysql_close(mysql);
1378
      DBUG_RETURN(1);
1379
    }
1380 1381
    if (thd->killed)
      goto err;
1382
  }
unknown's avatar
unknown committed
1383

unknown's avatar
unknown committed
1384
  if (request_table_dump(mysql, db_name, table_name))
1385
  {
1386 1387
    error= ER_UNKNOWN_ERROR;
    errmsg= "Failed on table dump request";
1388 1389
    goto err;
  }
unknown's avatar
unknown committed
1390
  if (create_table_from_dump(thd, mysql, db_name,
unknown's avatar
unknown committed
1391
                             table_name, overwrite))
unknown's avatar
unknown committed
1392
    goto err;    // create_table_from_dump have sent the error already
unknown's avatar
unknown committed
1393
  error = 0;
1394

unknown's avatar
unknown committed
1395
 err:
1396
  if (!called_connected)
unknown's avatar
SCRUM  
unknown committed
1397
    mysql_close(mysql);
1398
  if (errmsg && thd->vio_ok())
unknown's avatar
unknown committed
1399
    my_message(error, errmsg, MYF(0));
unknown's avatar
unknown committed
1400
  DBUG_RETURN(test(error));                     // Return 1 on error
unknown's avatar
unknown committed
1401
}
unknown's avatar
unknown committed
1402

1403

1404
static bool wait_for_relay_log_space(Relay_log_info* rli)
unknown's avatar
unknown committed
1405
{
1406
  bool slave_killed=0;
1407
  Master_info* mi = rli->mi;
unknown's avatar
unknown committed
1408
  const char *save_proc_info;
unknown's avatar
unknown committed
1409 1410
  THD* thd = mi->io_thd;
  DBUG_ENTER("wait_for_relay_log_space");
1411

unknown's avatar
unknown committed
1412
  pthread_mutex_lock(&rli->log_space_lock);
unknown's avatar
unknown committed
1413
  save_proc_info= thd->enter_cond(&rli->log_space_cond,
unknown's avatar
unknown committed
1414 1415
                                  &rli->log_space_lock,
                                  "\
1416
Waiting for the slave SQL thread to free enough relay log space");
unknown's avatar
unknown committed
1417
  while (rli->log_space_limit < rli->log_space_total &&
unknown's avatar
unknown committed
1418
         !(slave_killed=io_slave_killed(thd,mi)) &&
1419
         !rli->ignore_log_space_limit)
unknown's avatar
unknown committed
1420
    pthread_cond_wait(&rli->log_space_cond, &rli->log_space_lock);
unknown's avatar
unknown committed
1421
  thd->exit_cond(save_proc_info);
unknown's avatar
unknown committed
1422 1423
  DBUG_RETURN(slave_killed);
}
1424

1425

1426 1427
/*
  Builds a Rotate from the ignored events' info and writes it to relay log.
1428

1429 1430 1431 1432 1433 1434 1435 1436 1437 1438
  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).
 */
1439
static void write_ignored_events_info_to_relay_log(THD *thd, Master_info *mi)
1440
{
1441
  Relay_log_info *rli= &mi->rli;
1442
  pthread_mutex_t *log_lock= rli->relay_log.get_log_lock();
1443 1444
  DBUG_ENTER("write_ignored_events_info_to_relay_log");

1445 1446 1447
  DBUG_ASSERT(thd == mi->io_thd);
  pthread_mutex_lock(log_lock);
  if (rli->ign_master_log_name_end[0])
unknown's avatar
unknown committed
1448
  {
1449
    DBUG_PRINT("info",("writing a Rotate event to track down ignored events"));
1450
    Rotate_log_event *ev= new Rotate_log_event(rli->ign_master_log_name_end,
1451 1452 1453 1454 1455 1456 1457 1458 1459
                                               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)))
1460 1461 1462 1463 1464
        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");
1465
      rli->relay_log.harvest_bytes_written(&rli->log_space_total);
1466 1467
      if (flush_master_info(mi, 1))
        sql_print_error("Failed to flush master info file");
1468 1469 1470
      delete ev;
    }
    else
1471 1472 1473 1474
      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
1475
  }
1476 1477
  else
    pthread_mutex_unlock(log_lock);
1478
  DBUG_VOID_RETURN;
unknown's avatar
unknown committed
1479 1480
}

1481

1482
int register_slave_on_master(MYSQL* mysql, Master_info *mi,
1483
                             bool *suppress_warnings)
1484
{
1485
  uchar buf[1024], *pos= buf;
1486
  uint report_host_len, report_user_len=0, report_password_len=0;
1487
  DBUG_ENTER("register_slave_on_master");
1488

1489
  *suppress_warnings= FALSE;
1490
  if (!report_host)
1491
    DBUG_RETURN(0);
1492
  report_host_len= strlen(report_host);
1493
  if (report_user)
1494
    report_user_len= strlen(report_user);
unknown's avatar
unknown committed
1495
  if (report_password)
1496 1497 1498 1499
    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
1500
    DBUG_RETURN(0);                                     // safety
1501 1502

  int4store(pos, server_id); pos+= 4;
1503 1504 1505
  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);
1506
  int2store(pos, (uint16) report_port); pos+= 2;
unknown's avatar
unknown committed
1507
  int4store(pos, rpl_recovery_rank);    pos+= 4;
1508
  /* The master will fill in master_id */
unknown's avatar
unknown committed
1509
  int4store(pos, 0);                    pos+= 4;
1510

1511
  if (simple_command(mysql, COM_REGISTER_SLAVE, buf, (size_t) (pos- buf), 0))
1512
  {
1513 1514 1515 1516
    if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
    {
      *suppress_warnings= TRUE;                 // Suppress reconnect warning
    }
1517
    else if (!check_io_slave_killed(mi->io_thd, mi, NULL))
1518 1519 1520 1521 1522 1523 1524
    {
      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);
    }
1525
    DBUG_RETURN(1);
1526
  }
1527
  DBUG_RETURN(0);
1528 1529
}

1530

1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541
/**
  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
*/
1542
bool show_master_info(THD* thd, Master_info* mi)
unknown's avatar
unknown committed
1543
{
1544
  // TODO: fix this for multi-master
unknown's avatar
unknown committed
1545
  List<Item> field_list;
1546 1547 1548
  Protocol *protocol= thd->protocol;
  DBUG_ENTER("show_master_info");

unknown's avatar
unknown committed
1549
  field_list.push_back(new Item_empty_string("Slave_IO_State",
unknown's avatar
unknown committed
1550
                                                     14));
unknown's avatar
unknown committed
1551
  field_list.push_back(new Item_empty_string("Master_Host",
unknown's avatar
unknown committed
1552
                                                     sizeof(mi->host)));
unknown's avatar
unknown committed
1553
  field_list.push_back(new Item_empty_string("Master_User",
unknown's avatar
unknown committed
1554
                                                     sizeof(mi->user)));
1555
  field_list.push_back(new Item_return_int("Master_Port", 7,
unknown's avatar
unknown committed
1556
                                           MYSQL_TYPE_LONG));
1557
  field_list.push_back(new Item_return_int("Connect_Retry", 10,
unknown's avatar
unknown committed
1558
                                           MYSQL_TYPE_LONG));
1559
  field_list.push_back(new Item_empty_string("Master_Log_File",
unknown's avatar
unknown committed
1560
                                             FN_REFLEN));
1561
  field_list.push_back(new Item_return_int("Read_Master_Log_Pos", 10,
unknown's avatar
unknown committed
1562
                                           MYSQL_TYPE_LONGLONG));
1563
  field_list.push_back(new Item_empty_string("Relay_Log_File",
unknown's avatar
unknown committed
1564
                                             FN_REFLEN));
1565
  field_list.push_back(new Item_return_int("Relay_Log_Pos", 10,
unknown's avatar
unknown committed
1566
                                           MYSQL_TYPE_LONGLONG));
1567
  field_list.push_back(new Item_empty_string("Relay_Master_Log_File",
unknown's avatar
unknown committed
1568
                                             FN_REFLEN));
1569 1570
  field_list.push_back(new Item_empty_string("Slave_IO_Running", 3));
  field_list.push_back(new Item_empty_string("Slave_SQL_Running", 3));
1571 1572 1573 1574 1575 1576
  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
1577
                                             28));
1578 1579 1580
  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
1581
                                           MYSQL_TYPE_LONG));
1582
  field_list.push_back(new Item_return_int("Exec_Master_Log_Pos", 10,
unknown's avatar
unknown committed
1583
                                           MYSQL_TYPE_LONGLONG));
1584
  field_list.push_back(new Item_return_int("Relay_Log_Space", 10,
unknown's avatar
unknown committed
1585
                                           MYSQL_TYPE_LONGLONG));
1586
  field_list.push_back(new Item_empty_string("Until_Condition", 6));
1587
  field_list.push_back(new Item_empty_string("Until_Log_File", FN_REFLEN));
unknown's avatar
unknown committed
1588
  field_list.push_back(new Item_return_int("Until_Log_Pos", 10,
1589
                                           MYSQL_TYPE_LONGLONG));
unknown's avatar
unknown committed
1590 1591 1592
  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
1593
  field_list.push_back(new Item_empty_string("Master_SSL_CA_Path",
unknown's avatar
unknown committed
1594
                                             sizeof(mi->ssl_capath)));
unknown's avatar
unknown committed
1595
  field_list.push_back(new Item_empty_string("Master_SSL_Cert",
unknown's avatar
unknown committed
1596
                                             sizeof(mi->ssl_cert)));
unknown's avatar
unknown committed
1597
  field_list.push_back(new Item_empty_string("Master_SSL_Cipher",
unknown's avatar
unknown committed
1598
                                             sizeof(mi->ssl_cipher)));
unknown's avatar
unknown committed
1599
  field_list.push_back(new Item_empty_string("Master_SSL_Key",
unknown's avatar
unknown committed
1600
                                             sizeof(mi->ssl_key)));
1601
  field_list.push_back(new Item_return_int("Seconds_Behind_Master", 10,
unknown's avatar
unknown committed
1602
                                           MYSQL_TYPE_LONGLONG));
1603 1604
  field_list.push_back(new Item_empty_string("Master_SSL_Verify_Server_Cert",
                                             3));
1605 1606 1607 1608
  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
1609

1610 1611
  if (protocol->send_fields(&field_list,
                            Protocol::SEND_NUM_ROWS | Protocol::SEND_EOF))
unknown's avatar
unknown committed
1612
    DBUG_RETURN(TRUE);
unknown's avatar
unknown committed
1613

1614 1615
  if (mi->host[0])
  {
1616
    DBUG_PRINT("info",("host is set: '%s'", mi->host));
1617
    String *packet= &thd->packet;
1618
    protocol->prepare_for_resend();
unknown's avatar
unknown committed
1619

1620
    /*
1621 1622
      slave_running can be accessed without run_lock but not other
      non-volotile members like mi->io_thd, which is guarded by the mutex.
1623
    */
1624 1625 1626 1627
    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);

1628 1629
    pthread_mutex_lock(&mi->data_lock);
    pthread_mutex_lock(&mi->rli.data_lock);
1630 1631
    pthread_mutex_lock(&mi->err_lock);
    pthread_mutex_lock(&mi->rli.err_lock);
1632 1633
    protocol->store(mi->host, &my_charset_bin);
    protocol->store(mi->user, &my_charset_bin);
1634 1635
    protocol->store((uint32) mi->port);
    protocol->store((uint32) mi->connect_retry);
1636
    protocol->store(mi->master_log_name, &my_charset_bin);
1637
    protocol->store((ulonglong) mi->master_log_pos);
1638
    protocol->store(mi->rli.group_relay_log_name +
unknown's avatar
unknown committed
1639 1640
                    dirname_length(mi->rli.group_relay_log_name),
                    &my_charset_bin);
1641 1642
    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
1643 1644
    protocol->store(mi->slave_running == MYSQL_SLAVE_RUN_CONNECT ?
                    "Yes" : "No", &my_charset_bin);
1645
    protocol->store(mi->rli.slave_running ? "Yes":"No", &my_charset_bin);
1646 1647 1648
    protocol->store(rpl_filter->get_do_db());
    protocol->store(rpl_filter->get_ignore_db());

1649 1650
    char buf[256];
    String tmp(buf, sizeof(buf), &my_charset_bin);
1651
    rpl_filter->get_do_table(&tmp);
1652
    protocol->store(&tmp);
1653
    rpl_filter->get_ignore_table(&tmp);
1654
    protocol->store(&tmp);
1655
    rpl_filter->get_wild_do_table(&tmp);
1656
    protocol->store(&tmp);
1657
    rpl_filter->get_wild_ignore_table(&tmp);
1658 1659
    protocol->store(&tmp);

1660 1661
    protocol->store(mi->rli.last_error().number);
    protocol->store(mi->rli.last_error().message, &my_charset_bin);
1662
    protocol->store((uint32) mi->rli.slave_skip_counter);
1663
    protocol->store((ulonglong) mi->rli.group_master_log_pos);
1664
    protocol->store((ulonglong) mi->rli.log_space_total);
1665 1666

    protocol->store(
1667 1668
      mi->rli.until_condition==Relay_log_info::UNTIL_NONE ? "None":
        ( mi->rli.until_condition==Relay_log_info::UNTIL_MASTER_POS? "Master":
1669 1670 1671
          "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
1672 1673

#ifdef HAVE_OPENSSL
unknown's avatar
unknown committed
1674 1675 1676 1677 1678 1679 1680 1681 1682
    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
1683

1684 1685 1686 1687 1688 1689
    /*
      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)
1690
    {
1691
      long time_diff= ((long)(time(0) - mi->rli.last_master_timestamp)
unknown's avatar
unknown committed
1692
                       - mi->clock_diff_with_master);
1693
      /*
unknown's avatar
unknown committed
1694 1695
        Apparently on some systems time_diff can be <0. Here are possible
        reasons related to MySQL:
1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707
        - 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.
1708 1709 1710 1711
        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".
1712
      */
unknown's avatar
unknown committed
1713 1714
      protocol->store((longlong)(mi->rli.last_master_timestamp ?
                                 max(0, time_diff) : 0));
1715
    }
unknown's avatar
unknown committed
1716
    else
1717
    {
unknown's avatar
unknown committed
1718
      protocol->store_null();
1719 1720
    }
    protocol->store(mi->ssl_verify_server_cert? "Yes":"No", &my_charset_bin);
unknown's avatar
unknown committed
1721

1722
    // Last_IO_Errno
1723
    protocol->store(mi->last_error().number);
1724
    // Last_IO_Error
1725
    protocol->store(mi->last_error().message, &my_charset_bin);
1726
    // Last_SQL_Errno
1727
    protocol->store(mi->rli.last_error().number);
1728
    // Last_SQL_Error
1729
    protocol->store(mi->rli.last_error().message, &my_charset_bin);
1730

1731 1732
    pthread_mutex_unlock(&mi->rli.err_lock);
    pthread_mutex_unlock(&mi->err_lock);
1733 1734
    pthread_mutex_unlock(&mi->rli.data_lock);
    pthread_mutex_unlock(&mi->data_lock);
1735

1736
    if (my_net_write(&thd->net, (uchar*) thd->packet.ptr(), packet->length()))
unknown's avatar
unknown committed
1737
      DBUG_RETURN(TRUE);
1738
  }
1739
  my_eof(thd);
unknown's avatar
unknown committed
1740
  DBUG_RETURN(FALSE);
unknown's avatar
unknown committed
1741 1742
}

unknown's avatar
unknown committed
1743

1744 1745
void set_slave_thread_options(THD* thd)
{
1746
  DBUG_ENTER("set_slave_thread_options");
1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761
  /*
     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;
1762
  thd->variables.completion_type= 0;
1763
  DBUG_VOID_RETURN;
1764
}
1765

1766
void set_slave_thread_default_charset(THD* thd, Relay_log_info const *rli)
1767
{
1768 1769
  DBUG_ENTER("set_slave_thread_default_charset");

1770 1771 1772 1773 1774 1775 1776
  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();
1777 1778 1779 1780 1781 1782 1783

  /*
    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.
   */
1784
  const_cast<Relay_log_info*>(rli)->cached_charset_invalidate();
1785
  DBUG_VOID_RETURN;
1786 1787
}

unknown's avatar
unknown committed
1788
/*
1789
  init_slave_thread()
unknown's avatar
unknown committed
1790
*/
1791

1792
static int init_slave_thread(THD* thd, SLAVE_THD_TYPE thd_type)
unknown's avatar
unknown committed
1793 1794
{
  DBUG_ENTER("init_slave_thread");
1795 1796 1797
#if !defined(DBUG_OFF)
  int simulate_error= 0;
#endif
1798
  thd->system_thread = (thd_type == SLAVE_THD_SQL) ?
unknown's avatar
unknown committed
1799
    SYSTEM_THREAD_SLAVE_SQL : SYSTEM_THREAD_SLAVE_IO;
1800
  thd->security_ctx->skip_grants();
unknown's avatar
unknown committed
1801
  my_net_init(&thd->net, 0);
1802 1803 1804 1805 1806
/*
  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.
*/
1807
  thd->variables.max_allowed_packet= global_system_variables.max_allowed_packet
1808
    + MAX_LOG_EVENT_HEADER;  /* note, incr over the global not session var */
1809
  thd->slave_thread = 1;
1810
  thd->enable_slow_log= opt_log_slow_slave_statements;
1811
  set_slave_thread_options(thd);
unknown's avatar
unknown committed
1812 1813
  thd->client_capabilities = CLIENT_LOCAL_FILES;
  pthread_mutex_lock(&LOCK_thread_count);
unknown's avatar
unknown committed
1814
  thd->thread_id= thd->variables.pseudo_thread_id= thread_id++;
unknown's avatar
unknown committed
1815 1816
  pthread_mutex_unlock(&LOCK_thread_count);

1817 1818 1819 1820 1821 1822 1823
  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
1824
  if (init_thr_lock() || thd->store_globals())
1825
#endif
unknown's avatar
unknown committed
1826
  {
1827
    thd->cleanup();
unknown's avatar
unknown committed
1828 1829
    DBUG_RETURN(-1);
  }
1830
  lex_start(thd);
unknown's avatar
unknown committed
1831

1832
  if (thd_type == SLAVE_THD_SQL)
1833
    thd_proc_info(thd, "Waiting for the next event in relay log");
1834
  else
1835
    thd_proc_info(thd, "Waiting for master update");
unknown's avatar
unknown committed
1836 1837 1838 1839 1840
  thd->version=refresh_version;
  thd->set_time();
  DBUG_RETURN(0);
}

1841

1842
static int safe_sleep(THD* thd, int sec, CHECK_KILLED_FUNC thread_killed,
unknown's avatar
unknown committed
1843
                      void* thread_killed_arg)
unknown's avatar
unknown committed
1844
{
1845
  int nap_time;
unknown's avatar
unknown committed
1846
  thr_alarm_t alarmed;
1847 1848
  DBUG_ENTER("safe_sleep");

unknown's avatar
unknown committed
1849
  thr_alarm_init(&alarmed);
1850
  time_t start_time= my_time(0);
unknown's avatar
unknown committed
1851 1852
  time_t end_time= start_time+sec;

1853
  while ((nap_time= (int) (end_time - start_time)) > 0)
unknown's avatar
unknown committed
1854
  {
1855
    ALARM alarm_buff;
unknown's avatar
unknown committed
1856
    /*
1857
      The only reason we are asking for alarm is so that
unknown's avatar
unknown committed
1858 1859 1860
      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
    */
1861
    thr_alarm(&alarmed, 2 * nap_time, &alarm_buff);
unknown's avatar
unknown committed
1862
    sleep(nap_time);
1863
    thr_end_alarm(&alarmed);
unknown's avatar
unknown committed
1864

1865
    if ((*thread_killed)(thd,thread_killed_arg))
1866
      DBUG_RETURN(1);
1867
    start_time= my_time(0);
unknown's avatar
unknown committed
1868
  }
1869
  DBUG_RETURN(0);
unknown's avatar
unknown committed
1870 1871
}

1872

1873
static int request_dump(MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
1874
                        bool *suppress_warnings)
unknown's avatar
unknown committed
1875
{
1876
  uchar buf[FN_REFLEN + 10];
unknown's avatar
unknown committed
1877 1878
  int len;
  int binlog_flags = 0; // for now
1879
  char* logname = mi->master_log_name;
1880
  DBUG_ENTER("request_dump");
1881 1882
  
  *suppress_warnings= FALSE;
1883

unknown's avatar
unknown committed
1884
  // TODO if big log files: Change next to int8store()
unknown's avatar
unknown committed
1885
  int4store(buf, (ulong) mi->master_log_pos);
unknown's avatar
unknown committed
1886
  int2store(buf + 4, binlog_flags);
1887
  int4store(buf + 6, server_id);
unknown's avatar
unknown committed
1888
  len = (uint) strlen(logname);
1889
  memcpy(buf + 10, logname,len);
unknown's avatar
SCRUM  
unknown committed
1890
  if (simple_command(mysql, COM_BINLOG_DUMP, buf, len + 10, 1))
unknown's avatar
unknown committed
1891
  {
unknown's avatar
unknown committed
1892 1893 1894 1895 1896
    /*
      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
1897
    if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
1898
      *suppress_warnings= TRUE;                 // Suppress reconnect warning
unknown's avatar
unknown committed
1899
    else
1900
      sql_print_error("Error on COM_BINLOG_DUMP: %d  %s, will retry in %d secs",
unknown's avatar
unknown committed
1901 1902
                      mysql_errno(mysql), mysql_error(mysql),
                      master_connect_retry);
1903
    DBUG_RETURN(1);
unknown's avatar
unknown committed
1904
  }
unknown's avatar
unknown committed
1905

1906
  DBUG_RETURN(0);
unknown's avatar
unknown committed
1907 1908
}

1909

1910
static int request_table_dump(MYSQL* mysql, const char* db, const char* table)
unknown's avatar
unknown committed
1911
{
1912
  uchar buf[1024], *p = buf;
1913 1914
  DBUG_ENTER("request_table_dump");

unknown's avatar
unknown committed
1915 1916
  uint table_len = (uint) strlen(table);
  uint db_len = (uint) strlen(db);
unknown's avatar
unknown committed
1917
  if (table_len + db_len > sizeof(buf) - 2)
unknown's avatar
unknown committed
1918 1919
  {
    sql_print_error("request_table_dump: Buffer overrun");
1920
    DBUG_RETURN(1);
unknown's avatar
unknown committed
1921 1922
  }

unknown's avatar
unknown committed
1923 1924 1925 1926 1927
  *p++ = db_len;
  memcpy(p, db, db_len);
  p += db_len;
  *p++ = table_len;
  memcpy(p, table, table_len);
unknown's avatar
unknown committed
1928

unknown's avatar
SCRUM  
unknown committed
1929
  if (simple_command(mysql, COM_TABLE_DUMP, buf, p - buf + table_len, 1))
unknown's avatar
unknown committed
1930 1931
  {
    sql_print_error("request_table_dump: Error sending the table dump \
unknown's avatar
unknown committed
1932
command");
1933
    DBUG_RETURN(1);
unknown's avatar
unknown committed
1934
  }
unknown's avatar
unknown committed
1935

1936
  DBUG_RETURN(0);
unknown's avatar
unknown committed
1937 1938
}

1939

unknown's avatar
unknown committed
1940
/*
1941
  Read one event from the master
unknown's avatar
unknown committed
1942

unknown's avatar
unknown committed
1943 1944
  SYNOPSIS
    read_event()
unknown's avatar
unknown committed
1945 1946 1947 1948 1949 1950
    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.
1951

unknown's avatar
unknown committed
1952
    RETURN VALUES
unknown's avatar
unknown committed
1953 1954
    'packet_error'      Error
    number              Length of packet
unknown's avatar
unknown committed
1955 1956
*/

1957
static ulong read_event(MYSQL* mysql, Master_info *mi, bool* suppress_warnings)
unknown's avatar
unknown committed
1958
{
1959
  ulong len;
1960
  DBUG_ENTER("read_event");
unknown's avatar
unknown committed
1961

1962
  *suppress_warnings= FALSE;
unknown's avatar
unknown committed
1963 1964 1965 1966
  /*
    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
1967
#ifndef DBUG_OFF
1968
  if (disconnect_slave_event_count && !(mi->events_till_disconnect--))
unknown's avatar
unknown committed
1969
    DBUG_RETURN(packet_error);
unknown's avatar
unknown committed
1970
#endif
unknown's avatar
unknown committed
1971

1972
  len = cli_safe_read(mysql);
1973
  if (len == packet_error || (long) len < 1)
unknown's avatar
unknown committed
1974
  {
unknown's avatar
SCRUM  
unknown committed
1975
    if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
unknown's avatar
unknown committed
1976 1977
    {
      /*
unknown's avatar
unknown committed
1978 1979 1980
        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
1981 1982 1983 1984
      */
      *suppress_warnings= TRUE;
    }
    else
1985
      sql_print_error("Error reading packet from server: %s ( server_errno=%d)",
unknown's avatar
unknown committed
1986
                      mysql_error(mysql), mysql_errno(mysql));
1987
    DBUG_RETURN(packet_error);
unknown's avatar
unknown committed
1988 1989
  }

1990 1991
  /* Check if eof packet */
  if (len < 8 && mysql->net.read_pos[0] == 254)
unknown's avatar
unknown committed
1992
  {
1993 1994
    sql_print_information("Slave: received end packet from server, apparent "
                          "master shutdown: %s",
unknown's avatar
unknown committed
1995
                     mysql_error(mysql));
1996
     DBUG_RETURN(packet_error);
unknown's avatar
unknown committed
1997
  }
unknown's avatar
unknown committed
1998

unknown's avatar
unknown committed
1999
  DBUG_PRINT("exit", ("len: %lu  net->read_pos[4]: %d",
unknown's avatar
unknown committed
2000 2001
                      len, mysql->net.read_pos[4]));
  DBUG_RETURN(len - 1);
unknown's avatar
unknown committed
2002 2003
}

unknown's avatar
unknown committed
2004 2005 2006 2007 2008 2009 2010 2011 2012
/*
  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)
{
2013 2014
  DBUG_ENTER("has_temporary_error");

unknown's avatar
unknown committed
2015
  if (thd->is_fatal_error)
2016
    DBUG_RETURN(0);
2017 2018

  DBUG_EXECUTE_IF("all_errors_are_temporary_errors",
2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031
                  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
2032 2033 2034 2035 2036 2037

  /*
    Temporary error codes:
    currently, InnoDB deadlock detected by InnoDB or lock
    wait timeout (innodb_lock_wait_timeout exceeded
  */
2038 2039
  if (thd->main_da.sql_errno() == ER_LOCK_DEADLOCK ||
      thd->main_da.sql_errno() == ER_LOCK_WAIT_TIMEOUT)
2040
    DBUG_RETURN(1);
unknown's avatar
unknown committed
2041 2042 2043 2044 2045 2046 2047 2048 2049

#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
2050
    DBUG_PRINT("info", ("has warning %d %s", err->code, err->msg));
unknown's avatar
unknown committed
2051 2052 2053
    switch (err->code)
    {
    case ER_GET_TEMPORARY_ERRMSG:
2054
      DBUG_RETURN(1);
unknown's avatar
unknown committed
2055 2056 2057 2058 2059
    default:
      break;
    }
  }
#endif
2060
  DBUG_RETURN(0);
unknown's avatar
unknown committed
2061
}
2062

2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171

/**
  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);
2172 2173 2174 2175
#ifdef HAVE_purify
    if (!rli->is_fake)
#endif
    {
2176
#ifndef DBUG_OFF
2177
      char buf[22];
2178
#endif
2179 2180 2181 2182 2183 2184 2185 2186
      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));
    }
2187 2188 2189 2190 2191 2192 2193 2194 2195
    /*
      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)
    {
2196
      char buf[22];
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 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231
      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
2232
     if a failure occurs when applying the event.)
2233 2234 2235 2236 2237 2238 2239

   - An error occurred when updating the binlog position.

  @retval 0 The event was applied.

  @retval 1 The event was not applied.
*/
2240
static int exec_relay_log_event(THD* thd, Relay_log_info* rli)
unknown's avatar
unknown committed
2241
{
2242 2243
  DBUG_ENTER("exec_relay_log_event");

2244 2245
  /*
     We acquire this mutex since we need it for all operations except
2246
     event execution. But we will release it in places where we will
2247 2248 2249
     wait for something for example inside of next_event().
   */
  pthread_mutex_lock(&rli->data_lock);
2250

2251
  Log_event * ev = next_event(rli);
2252

2253
  DBUG_ASSERT(rli->sql_thd==thd);
2254

2255
  if (sql_slave_killed(thd,rli))
2256
  {
2257
    pthread_mutex_unlock(&rli->data_lock);
unknown's avatar
unknown committed
2258
    delete ev;
2259
    DBUG_RETURN(1);
2260
  }
2261 2262
  if (ev)
  {
2263
    int exec_res;
2264

2265 2266 2267 2268
    /*
      This tests if the position of the beginning of the current event
      hits the UNTIL barrier.
    */
2269
    if (rli->until_condition != Relay_log_info::UNTIL_NONE &&
2270
        rli->is_until_satisfied((rli->is_in_group() || !ev->log_pos) ?
2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283
                                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;
2284
      DBUG_RETURN(1);
2285
    }
2286
    exec_res= apply_event_and_update_pos(ev, thd, rli, TRUE);
2287

2288
    /*
2289 2290 2291
      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.
2292
    */
2293
    if (ev->get_type_code() != FORMAT_DESCRIPTION_EVENT)
2294
    {
2295 2296
      DBUG_PRINT("info", ("Deleting the event after it has been executed"));
      delete ev;
2297 2298
    }

2299
    /*
2300 2301
      update_log_pos failed: this should not happen, so we don't
      retry.
2302
    */
2303 2304 2305
    if (exec_res == 2)
      DBUG_RETURN(1);

2306 2307
    if (slave_trans_retries)
    {
2308 2309
      int temp_err;
      if (exec_res && (temp_err= has_temporary_error(thd)))
2310 2311 2312 2313
      {
        const char *errmsg;
        /*
          We were in a transaction which has been rolled back because of a
unknown's avatar
unknown committed
2314 2315
          temporary error;
          let's seek back to BEGIN log event and retry it all again.
2316
	  Note, if lock wait timeout (innodb_lock_wait_timeout exceeded)
2317
	  there is no rollback since 5.0.13 (ref: manual).
2318 2319 2320 2321 2322 2323 2324 2325
          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.
        */
2326
        if (rli->trans_retries < slave_trans_retries)
2327 2328 2329 2330 2331 2332
        {
          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,
2333
                                      1, &errmsg, 1))
2334 2335 2336 2337 2338
            sql_print_error("Error initializing relay log position: %s",
                            errmsg);
          else
          {
            exec_res= 0;
2339
            end_trans(thd, ROLLBACK);
unknown's avatar
unknown committed
2340
            /* chance for concurrent connection to get more locks */
2341
            safe_sleep(thd, min(rli->trans_retries, MAX_SLAVE_RETRY_PAUSE),
unknown's avatar
unknown committed
2342
                       (CHECK_KILLED_FUNC)sql_slave_killed, (void*)rli);
2343
            pthread_mutex_lock(&rli->data_lock); // because of SHOW STATUS
unknown's avatar
unknown committed
2344
            rli->trans_retries++;
2345 2346 2347 2348
            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
2349
          }
2350 2351 2352 2353 2354 2355 2356
        }
        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);
      }
2357
      else if ((exec_res && !temp_err) ||
2358 2359
               (opt_using_transactions &&
                rli->group_relay_log_pos == rli->event_relay_log_pos))
2360 2361
      {
        /*
2362 2363 2364
          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
2365 2366 2367
          non-transient error, the slave will stop with an error.
         */
        rli->trans_retries= 0; // restart from fresh
unknown's avatar
unknown committed
2368
        DBUG_PRINT("info", ("Resetting retry counter, rli->trans_retries: %lu",
2369
                            rli->trans_retries));
2370
      }
2371 2372
    }
    DBUG_RETURN(exec_res);
2373
  }
2374
  pthread_mutex_unlock(&rli->data_lock);
2375 2376
  rli->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_READ_FAILURE,
              ER(ER_SLAVE_RELAY_LOG_READ_FAILURE), "\
2377 2378 2379 2380 2381 2382 2383
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
2384
");
2385
  DBUG_RETURN(1);
unknown's avatar
unknown committed
2386 2387
}

2388

2389
static bool check_io_slave_killed(THD *thd, Master_info *mi, const char *info)
2390 2391 2392
{
  if (io_slave_killed(thd, mi))
  {
2393
    if (info && global_system_variables.log_warnings)
2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426
      sql_print_information(info);
    return TRUE;
  }
  return FALSE;
}


/**
  @brief Try to reconnect slave IO thread.

  @details Terminates current connection to master, sleeps for
  @c mi->connect_retry msecs and initiates new connection with
  @c safe_reconnect(). Variable pointed by @c retry_count is increased -
  if it exceeds @c master_retry_count then connection is not re-established
  and function signals error.
  Unless @c suppres_warnings is TRUE, a warning is put in the server error log
  when reconnecting. The warning message and messages used to report errors
  are taken from @c messages array. In case @c master_retry_count is exceeded,
  no messages are added to the log.

  @param[in]     thd                 Thread context.
  @param[in]     mysql               MySQL connection.
  @param[in]     mi                  Master connection information.
  @param[in,out] retry_count         Number of attempts to reconnect.
  @param[in]     suppress_warnings   TRUE when a normal net read timeout 
                                     has caused to reconnecting.
  @param[in]     messages            Messages to print/log, see 
                                     reconnect_messages[] array.

  @retval        0                   OK.
  @retval        1                   There was an error.
*/

2427
static int try_to_reconnect(THD *thd, MYSQL *mysql, Master_info *mi,
2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476
                            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;
}


2477 2478 2479 2480 2481
/**
  Slave IO thread entry point.

  @param arg Pointer to Master_info struct that holds information for
  the IO thread.
2482

2483 2484
  @return Always 0.
*/
2485
pthread_handler_t handle_slave_io(void *arg)
unknown's avatar
unknown committed
2486
{
unknown's avatar
unknown committed
2487 2488
  THD *thd; // needs to be first for thread_stack
  MYSQL *mysql;
2489
  Master_info *mi = (Master_info*)arg;
2490
  Relay_log_info *rli= &mi->rli;
unknown's avatar
unknown committed
2491 2492
  char llbuff[22];
  uint retry_count;
2493
  bool suppress_warnings;
2494
  int ret;
2495 2496 2497
#ifndef DBUG_OFF
  uint retry_count_reg= 0, retry_count_dump= 0, retry_count_event= 0;
#endif
unknown's avatar
unknown committed
2498 2499
  // needs to call my_thread_init(), otherwise we get a coredump in DBUG_ stuff
  my_thread_init();
2500
  DBUG_ENTER("handle_slave_io");
unknown's avatar
unknown committed
2501

2502
  DBUG_ASSERT(mi->inited);
unknown's avatar
unknown committed
2503 2504 2505
  mysql= NULL ;
  retry_count= 0;

2506
  pthread_mutex_lock(&mi->run_lock);
unknown's avatar
unknown committed
2507 2508 2509
  /* Inform waiting threads that slave has started */
  mi->slave_run_id++;

2510
#ifndef DBUG_OFF
2511
  mi->events_till_disconnect = disconnect_slave_event_count;
2512 2513
#endif

2514
  thd= new THD; // note that contructor of THD uses DBUG_ !
2515
  THD_CHECK_SENTRY(thd);
2516
  mi->io_thd = thd;
unknown's avatar
unknown committed
2517 2518

  pthread_detach_this_thread();
2519
  thd->thread_stack= (char*) &thd; // remember where our stack is
2520
  mi->clear_error();
2521
  if (init_slave_thread(thd, SLAVE_THD_IO))
unknown's avatar
unknown committed
2522 2523 2524 2525 2526 2527
  {
    pthread_cond_broadcast(&mi->start_cond);
    pthread_mutex_unlock(&mi->run_lock);
    sql_print_error("Failed during slave I/O thread initialization");
    goto err;
  }
2528
  pthread_mutex_lock(&LOCK_thread_count);
unknown's avatar
unknown committed
2529
  threads.append(thd);
2530
  pthread_mutex_unlock(&LOCK_thread_count);
2531 2532 2533
  mi->slave_running = 1;
  mi->abort_slave = 0;
  pthread_mutex_unlock(&mi->run_lock);
2534
  pthread_cond_broadcast(&mi->start_cond);
2535

2536
  DBUG_PRINT("master_info",("log_file_name: '%s'  position: %s",
unknown's avatar
unknown committed
2537 2538
                            mi->master_log_name,
                            llstr(mi->master_log_pos,llbuff)));
2539

unknown's avatar
SCRUM  
unknown committed
2540
  if (!(mi->mysql = mysql = mysql_init(NULL)))
unknown's avatar
unknown committed
2541
  {
2542 2543
    mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
               ER(ER_SLAVE_FATAL_ERROR), "error in mysql_init()");
unknown's avatar
unknown committed
2544 2545
    goto err;
  }
2546

2547
  thd_proc_info(thd, "Connecting to master");
2548
  // we can get killed during safe_connect
2549
  if (!safe_connect(thd, mysql, mi))
2550
  {
unknown's avatar
unknown committed
2551 2552 2553
    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,
2554 2555
			  IO_RPL_LOG_NAME,
			  llstr(mi->master_log_pos,llbuff));
2556 2557 2558 2559 2560
  /*
    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.
  */
2561 2562
    mysql->net.max_packet_size= thd->net.max_packet_size+= MAX_LOG_EVENT_HEADER;
  }
2563
  else
unknown's avatar
unknown committed
2564
  {
2565
    sql_print_information("Slave I/O thread killed while connecting to master");
unknown's avatar
unknown committed
2566 2567
    goto err;
  }
2568

2569
connected:
2570

2571 2572
  // TODO: the assignment below should be under mutex (5.0)
  mi->slave_running= MYSQL_SLAVE_RUN_CONNECT;
2573
  thd->slave_net = &mysql->net;
2574
  thd_proc_info(thd, "Checking master version");
2575 2576 2577
  ret= get_master_version_and_clock(mysql, mi);
  if (ret == 1)
    /* Fatal error */
2578
    goto err;
2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591
  
  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;
  } 
2592 2593

  if (mi->rli.relay_log.description_event_for_queue->binlog_version > 1)
2594
  {
unknown's avatar
unknown committed
2595 2596 2597
    /*
      Register ourselves with the master.
    */
2598
    thd_proc_info(thd, "Registering slave on master");
2599
    if (register_slave_on_master(mysql, mi, &suppress_warnings))
2600
    {
2601 2602 2603 2604 2605 2606 2607 2608 2609
      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
2610 2611
        goto err;
      goto connected;
2612
    }
2613 2614 2615 2616 2617 2618 2619 2620 2621 2622
    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;
      });
2623
  }
2624

2625
  DBUG_PRINT("info",("Starting reading binary log from master"));
2626
  while (!io_slave_killed(thd,mi))
unknown's avatar
unknown committed
2627
  {
2628
    thd_proc_info(thd, "Requesting binlog dump");
unknown's avatar
unknown committed
2629
    if (request_dump(mysql, mi, &suppress_warnings))
unknown's avatar
unknown committed
2630 2631
    {
      sql_print_error("Failed on request_dump()");
2632 2633 2634 2635
      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
2636
        goto err;
unknown's avatar
unknown committed
2637 2638
      goto connected;
    }
2639 2640 2641 2642 2643 2644 2645 2646 2647 2648
    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
2649

2650
    DBUG_ASSERT(mi->last_error().number == 0);
2651
    while (!io_slave_killed(thd,mi))
unknown's avatar
unknown committed
2652
    {
2653
      ulong event_len;
2654
      /*
2655
         We say "waiting" because read_event() will wait if there's nothing to
2656 2657 2658
         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.
2659
      */
2660
      thd_proc_info(thd, "Waiting for master to send event");
2661
      event_len= read_event(mysql, mi, &suppress_warnings);
2662 2663
      if (check_io_slave_killed(thd, mi, "Slave I/O thread killed while \
reading event"))
unknown's avatar
unknown committed
2664
        goto err;
2665 2666 2667 2668 2669 2670 2671 2672 2673 2674
      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;
        });
2675

unknown's avatar
unknown committed
2676 2677
      if (event_len == packet_error)
      {
unknown's avatar
unknown committed
2678
        uint mysql_error_number= mysql_errno(mysql);
2679 2680
        switch (mysql_error_number) {
        case CR_NET_PACKET_TOO_LARGE:
unknown's avatar
unknown committed
2681
          sql_print_error("\
2682 2683 2684
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
2685 2686
                          thd->variables.max_allowed_packet);
          goto err;
2687
        case ER_MASTER_FATAL_ERROR_READING_BINLOG:
unknown's avatar
unknown committed
2688 2689 2690
          sql_print_error(ER(mysql_error_number), mysql_error_number,
                          mysql_error(mysql));
          goto err;
2691 2692 2693 2694
        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
2695 2696
          goto err;
        }
2697 2698
        if (try_to_reconnect(thd, mysql, mi, &retry_count, suppress_warnings,
                             reconnect_messages[SLAVE_RECON_ACT_EVENT]))
unknown's avatar
unknown committed
2699 2700
          goto err;
        goto connected;
unknown's avatar
unknown committed
2701
      } // if (event_len == packet_error)
2702

unknown's avatar
unknown committed
2703
      retry_count=0;                    // ok event, reset retry counter
2704
      thd_proc_info(thd, "Queueing master event to the relay log");
unknown's avatar
unknown committed
2705
      if (queue_event(mi,(const char*)mysql->net.read_pos + 1,
unknown's avatar
unknown committed
2706
                      event_len))
unknown's avatar
unknown committed
2707
      {
2708 2709 2710
        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
2711
        goto err;
unknown's avatar
unknown committed
2712
      }
2713 2714 2715 2716 2717
      if (flush_master_info(mi, 1))
      {
        sql_print_error("Failed to flush master info file");
        goto err;
      }
2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729
      /*
        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.
      */
2730 2731 2732 2733 2734
#ifndef DBUG_OFF
      {
        char llbuf1[22], llbuf2[22];
        DBUG_PRINT("info", ("log_space_limit=%s log_space_total=%s \
ignore_log_space_limit=%d",
2735 2736
                            llstr(rli->log_space_limit,llbuf1),
                            llstr(rli->log_space_total,llbuf2),
unknown's avatar
unknown committed
2737
                            (int) rli->ignore_log_space_limit));
2738 2739 2740
      }
#endif

2741
      if (rli->log_space_limit && rli->log_space_limit <
unknown's avatar
unknown committed
2742
          rli->log_space_total &&
2743
          !rli->ignore_log_space_limit)
unknown's avatar
unknown committed
2744 2745 2746
        if (wait_for_relay_log_space(rli))
        {
          sql_print_error("Slave I/O thread aborted while waiting for relay \
unknown's avatar
unknown committed
2747
log space");
unknown's avatar
unknown committed
2748 2749 2750
          goto err;
        }
    }
2751
  }
unknown's avatar
unknown committed
2752

unknown's avatar
unknown committed
2753
  // error = 0;
unknown's avatar
unknown committed
2754
err:
2755
  // print the current replication position
2756
  sql_print_information("Slave I/O thread exiting, read up to log '%s', position %s",
unknown's avatar
unknown committed
2757
                  IO_RPL_LOG_NAME, llstr(mi->master_log_pos,llbuff));
2758
  VOID(pthread_mutex_lock(&LOCK_thread_count));
unknown's avatar
unknown committed
2759
  thd->query = thd->db = 0; // extra safety
2760
  thd->query_length= thd->db_length= 0;
2761
  VOID(pthread_mutex_unlock(&LOCK_thread_count));
unknown's avatar
unknown committed
2762 2763
  if (mysql)
  {
2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774
    /*
      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
2775
    mysql_close(mysql);
unknown's avatar
unknown committed
2776 2777
    mi->mysql=0;
  }
2778
  write_ignored_events_info_to_relay_log(thd, mi);
2779
  thd_proc_info(thd, "Waiting for slave mutex on exit");
2780
  pthread_mutex_lock(&mi->run_lock);
2781

2782 2783 2784
  /* Forget the relay log's format */
  delete mi->rli.relay_log.description_event_for_queue;
  mi->rli.relay_log.description_event_for_queue= 0;
2785
  // TODO: make rpl_status part of Master_info
2786
  change_rpl_status(RPL_ACTIVE_SLAVE,RPL_IDLE_SLAVE);
2787
  DBUG_ASSERT(thd->net.buff != 0);
unknown's avatar
unknown committed
2788
  net_end(&thd->net); // destructor will not free it, because net.vio is 0
2789
  close_thread_tables(thd);
2790
  pthread_mutex_lock(&LOCK_thread_count);
2791
  THD_CHECK_SENTRY(thd);
unknown's avatar
unknown committed
2792
  delete thd;
2793
  pthread_mutex_unlock(&LOCK_thread_count);
unknown's avatar
unknown committed
2794 2795 2796
  mi->abort_slave= 0;
  mi->slave_running= 0;
  mi->io_thd= 0;
2797 2798 2799 2800 2801
  /*
    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)
   */ 
2802
  pthread_cond_broadcast(&mi->stop_cond);       // tell the world we are done
2803
  DBUG_EXECUTE_IF("simulate_slave_delay_at_terminate_bug38694", sleep(5););
2804
  pthread_mutex_unlock(&mi->run_lock);
2805
  my_thread_end();
unknown's avatar
unknown committed
2806
  pthread_exit(0);
unknown's avatar
unknown committed
2807
  DBUG_RETURN(0);                               // Can't return anything here
unknown's avatar
unknown committed
2808 2809
}

2810 2811 2812 2813 2814
/*
  Check the temporary directory used by commands like
  LOAD DATA INFILE.
 */
static 
2815
int check_temp_dir(char* tmp_file)
2816 2817 2818
{
  int fd;
  MY_DIR *dirp;
2819 2820
  char tmp_dir[FN_REFLEN];
  size_t tmp_dir_size;
2821 2822 2823

  DBUG_ENTER("check_temp_dir");

2824 2825 2826 2827 2828
  /*
    Get the directory from the temporary file.
  */
  dirname_part(tmp_dir, tmp_file, &tmp_dir_size);

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

2853 2854 2855 2856 2857
/**
  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
2858

2859 2860
  @return Always 0.
*/
2861
pthread_handler_t handle_slave_sql(void *arg)
2862
{
unknown's avatar
unknown committed
2863
  THD *thd;                     /* needs to be first for thread_stack */
2864
  char llbuff[22],llbuff1[22];
2865

2866
  Relay_log_info* rli = &((Master_info*)arg)->rli;
unknown's avatar
unknown committed
2867 2868 2869 2870
  const char *errmsg;

  // needs to call my_thread_init(), otherwise we get a coredump in DBUG_ stuff
  my_thread_init();
2871
  DBUG_ENTER("handle_slave_sql");
unknown's avatar
unknown committed
2872

2873 2874 2875
  DBUG_ASSERT(rli->inited);
  pthread_mutex_lock(&rli->run_lock);
  DBUG_ASSERT(!rli->slave_running);
unknown's avatar
unknown committed
2876
  errmsg= 0;
unknown's avatar
unknown committed
2877
#ifndef DBUG_OFF
2878
  rli->events_till_abort = abort_slave_event_count;
unknown's avatar
unknown committed
2879
#endif
2880

unknown's avatar
unknown committed
2881
  thd = new THD; // note that contructor of THD uses DBUG_ !
2882
  thd->thread_stack = (char*)&thd; // remember where our stack is
2883
  rli->sql_thd= thd;
2884
  
unknown's avatar
unknown committed
2885 2886
  /* Inform waiting threads that slave has started */
  rli->slave_run_id++;
2887
  rli->slave_running = 1;
unknown's avatar
unknown committed
2888

2889 2890
  pthread_detach_this_thread();
  if (init_slave_thread(thd, SLAVE_THD_SQL))
unknown's avatar
unknown committed
2891 2892 2893 2894 2895 2896 2897
  {
    /*
      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);
2898 2899
    rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, 
                "Failed during slave thread initialization");
unknown's avatar
unknown committed
2900 2901
    goto err;
  }
2902
  thd->init_for_queries();
2903
  thd->temporary_tables = rli->save_temporary_tables; // restore temp tables
2904
  set_thd_in_use_temporary_tables(rli);   // (re)set sql_thd in use for saved temp tables
2905
  pthread_mutex_lock(&LOCK_thread_count);
2906
  threads.append(thd);
2907
  pthread_mutex_unlock(&LOCK_thread_count);
2908 2909 2910 2911 2912 2913 2914 2915
  /*
    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.
  */
2916 2917
  rli->abort_slave = 0;
  pthread_mutex_unlock(&rli->run_lock);
2918
  pthread_cond_broadcast(&rli->start_cond);
2919

unknown's avatar
unknown committed
2920 2921 2922
  /*
    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
2923
    though there's no problem anymore). Do not reset the master timestamp
2924 2925 2926 2927
    (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
2928
    But the master timestamp is reset by RESET SLAVE & CHANGE MASTER.
unknown's avatar
unknown committed
2929
  */
2930
  rli->clear_error();
2931 2932

  //tell the I/O thread to take relay_log_space_limit into account from now on
2933
  pthread_mutex_lock(&rli->log_space_lock);
2934
  rli->ignore_log_space_limit= 0;
2935
  pthread_mutex_unlock(&rli->log_space_lock);
2936
  rli->trans_retries= 0; // start from "no error"
unknown's avatar
unknown committed
2937
  DBUG_PRINT("info", ("rli->trans_retries: %lu", rli->trans_retries));
2938

2939
  if (init_relay_log_pos(rli,
unknown's avatar
unknown committed
2940 2941 2942
                         rli->group_relay_log_name,
                         rli->group_relay_log_pos,
                         1 /*need data lock*/, &errmsg,
2943
                         1 /*look for a description_event*/))
2944 2945 2946
  { 
    rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, 
                "Error initializing relay log position: %s", errmsg);
2947 2948
    goto err;
  }
2949
  THD_CHECK_SENTRY(thd);
2950 2951 2952 2953
#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
2954
                        llstr(my_b_tell(rli->cur_log),llbuf1),
2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973
                        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
2974
  DBUG_ASSERT(rli->sql_thd == thd);
2975 2976

  DBUG_PRINT("master_info",("log_file_name: %s  position: %s",
unknown's avatar
unknown committed
2977 2978
                            rli->group_master_log_name,
                            llstr(rli->group_master_log_pos,llbuff)));
2979
  if (global_system_variables.log_warnings)
2980
    sql_print_information("Slave SQL thread initialized, starting replication in \
unknown's avatar
unknown committed
2981
log '%s' at position %s, relay log '%s' position: %s", RPL_LOG_NAME,
unknown's avatar
unknown committed
2982 2983
                    llstr(rli->group_master_log_pos,llbuff),rli->group_relay_log_name,
                    llstr(rli->group_relay_log_pos,llbuff1));
2984

2985
  if (check_temp_dir(rli->slave_patternload_file))
2986 2987 2988 2989 2990 2991 2992
  {
    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
2993
  /* execute init_slave variable */
unknown's avatar
unknown committed
2994
  if (sys_init_slave.value_length)
unknown's avatar
unknown committed
2995
  {
2996
    execute_init_command(thd, &sys_init_slave, &LOCK_sys_init_slave);
2997
    if (thd->is_slave_error)
unknown's avatar
unknown committed
2998
    {
2999 3000
      rli->report(ERROR_LEVEL, thd->main_da.sql_errno(), 
                  "Slave SQL thread aborted. Can't execute init_slave query");
unknown's avatar
unknown committed
3001 3002 3003 3004
      goto err;
    }
  }

3005 3006 3007 3008 3009
  /*
    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);
3010
  if (rli->until_condition != Relay_log_info::UNTIL_NONE &&
3011 3012 3013 3014 3015 3016 3017 3018 3019 3020
      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);

3021 3022
  /* Read queries from the IO/THREAD until this thread is killed */

3023
  while (!sql_slave_killed(thd,rli))
3024
  {
3025
    thd_proc_info(thd, "Reading event from the relay log");
3026
    DBUG_ASSERT(rli->sql_thd == thd);
3027
    THD_CHECK_SENTRY(thd);
3028 3029
    if (exec_relay_log_event(thd,rli))
    {
3030
      DBUG_PRINT("info", ("exec_relay_log_event() failed"));
3031
      // do not scare the user if SQL thread was simply killed or stopped
3032
      if (!sql_slave_killed(thd,rli))
unknown's avatar
unknown committed
3033 3034
      {
        /*
3035 3036 3037
          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
3038
        */
3039
        uint32 const last_errno= rli->last_error().number;
3040

3041
        if (thd->is_error())
unknown's avatar
unknown committed
3042
        {
3043 3044 3045 3046 3047
          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));
3048
          if (last_errno == 0)
unknown's avatar
unknown committed
3049
          {
3050 3051 3052 3053
            /*
 	      This function is reporting an error which was not reported
 	      while executing exec_relay_log_event().
 	    */ 
3054
            rli->report(ERROR_LEVEL, thd->main_da.sql_errno(), errmsg);
unknown's avatar
unknown committed
3055
          }
3056
          else if (last_errno != thd->main_da.sql_errno())
unknown's avatar
unknown committed
3057
          {
3058 3059 3060 3061 3062 3063
            /*
             * 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
3064
            sql_print_error("Slave (additional info): %s Error_code: %d",
3065
                            errmsg, thd->main_da.sql_errno());
unknown's avatar
unknown committed
3066 3067 3068 3069 3070 3071
          }
        }

        /* Print any warnings issued */
        List_iterator_fast<MYSQL_ERROR> it(thd->warn_list);
        MYSQL_ERROR *err;
3072 3073 3074 3075 3076
        /*
          Added controlled slave thread cancel for replication
          of user-defined variables.
        */
        bool udf_error = false;
unknown's avatar
unknown committed
3077
        while ((err= it++))
3078 3079 3080
        {
          if (err->code == ER_CANT_OPEN_LIBRARY)
            udf_error = true;
unknown's avatar
unknown committed
3081
          sql_print_warning("Slave: %s Error_code: %d",err->msg, err->code);
3082 3083 3084 3085 3086 3087 3088 3089 3090
        }
        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("\
3091
Error running query, slave SQL thread aborted. Fix the problem, and restart \
unknown's avatar
unknown committed
3092
the slave SQL thread with \"SLAVE START\". We stopped at log \
3093
'%s' position %s", RPL_LOG_NAME, llstr(rli->group_master_log_pos, llbuff));
unknown's avatar
unknown committed
3094
      }
3095 3096
      goto err;
    }
3097
  }
3098

3099
  /* Thread stopped. Print the current replication position to the log */
unknown's avatar
unknown committed
3100
  sql_print_information("Slave SQL thread exiting, replication stopped in log "
unknown's avatar
unknown committed
3101 3102
                        "'%s' at position %s",
                        RPL_LOG_NAME, llstr(rli->group_master_log_pos,llbuff));
3103 3104

 err:
3105 3106 3107 3108 3109 3110 3111 3112

  /*
    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);
3113
  VOID(pthread_mutex_lock(&LOCK_thread_count));
3114 3115 3116 3117 3118
  /*
    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)).
  */
unknown's avatar
unknown committed
3119
  thd->query= thd->db= thd->catalog= 0;
3120
  thd->query_length= thd->db_length= 0;
3121
  VOID(pthread_mutex_unlock(&LOCK_thread_count));
3122
  thd_proc_info(thd, "Waiting for slave mutex on exit");
3123
  pthread_mutex_lock(&rli->run_lock);
3124 3125
  /* We need data_lock, at least to wake up any waiting master_pos_wait() */
  pthread_mutex_lock(&rli->data_lock);
3126
  DBUG_ASSERT(rli->slave_running == 1); // tracking buffer overrun
3127
  /* When master_pos_wait() wakes up it will check this and terminate */
unknown's avatar
unknown committed
3128
  rli->slave_running= 0;
3129 3130 3131
  /* Forget the relay log's format */
  delete rli->relay_log.description_event_for_exec;
  rli->relay_log.description_event_for_exec= 0;
3132 3133 3134 3135
  /* 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
3136
  rli->ignore_log_space_limit= 0; /* don't need any lock */
3137 3138
  /* we die so won't remember charset - re-update them on next thread start */
  rli->cached_charset_invalidate();
3139
  rli->save_temporary_tables = thd->temporary_tables;
unknown's avatar
unknown committed
3140 3141 3142 3143 3144

  /*
    TODO: see if we can do this conditionally in next_event() instead
    to avoid unneeded position re-init
  */
3145 3146 3147 3148
  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);
3149
  THD_CHECK_SENTRY(thd);
3150
  rli->sql_thd= 0;
3151
  set_thd_in_use_temporary_tables(rli);  // (re)set sql_thd in use for saved temp tables
3152
  pthread_mutex_lock(&LOCK_thread_count);
3153
  THD_CHECK_SENTRY(thd);
3154 3155
  delete thd;
  pthread_mutex_unlock(&LOCK_thread_count);
3156 3157 3158 3159 3160
 /*
  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)
 */ 
3161
  pthread_cond_broadcast(&rli->stop_cond);
3162
  DBUG_EXECUTE_IF("simulate_slave_delay_at_terminate_bug38694", sleep(5););
3163 3164
  pthread_mutex_unlock(&rli->run_lock);  // tell the world we are done
  
unknown's avatar
unknown committed
3165
  my_thread_end();
3166
  pthread_exit(0);
unknown's avatar
unknown committed
3167
  DBUG_RETURN(0);                               // Can't return anything here
3168
}
unknown's avatar
unknown committed
3169

3170

unknown's avatar
unknown committed
3171
/*
3172
  process_io_create_file()
unknown's avatar
unknown committed
3173
*/
3174

3175
static int process_io_create_file(Master_info* mi, Create_file_log_event* cev)
unknown's avatar
unknown committed
3176 3177 3178 3179
{
  int error = 1;
  ulong num_bytes;
  bool cev_not_written;
3180 3181
  THD *thd = mi->io_thd;
  NET *net = &mi->mysql->net;
unknown's avatar
unknown committed
3182
  DBUG_ENTER("process_io_create_file");
unknown's avatar
unknown committed
3183 3184

  if (unlikely(!cev->is_valid()))
unknown's avatar
unknown committed
3185
    DBUG_RETURN(1);
3186 3187

  if (!rpl_filter->db_ok(cev->db))
unknown's avatar
unknown committed
3188 3189
  {
    skip_load_data_infile(net);
unknown's avatar
unknown committed
3190
    DBUG_RETURN(0);
unknown's avatar
unknown committed
3191 3192 3193
  }
  DBUG_ASSERT(cev->inited_from_old);
  thd->file_id = cev->file_id = mi->file_id++;
3194
  thd->server_id = cev->server_id;
unknown's avatar
unknown committed
3195
  cev_not_written = 1;
unknown's avatar
unknown committed
3196

unknown's avatar
unknown committed
3197 3198 3199
  if (unlikely(net_request_file(net,cev->fname)))
  {
    sql_print_error("Slave I/O: failed requesting download of '%s'",
unknown's avatar
unknown committed
3200
                    cev->fname);
unknown's avatar
unknown committed
3201 3202 3203
    goto err;
  }

unknown's avatar
unknown committed
3204 3205 3206 3207
  /*
    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
3208 3209
  */
  {
3210
    Append_block_log_event aev(thd,0,0,0,0);
unknown's avatar
unknown committed
3211

unknown's avatar
unknown committed
3212 3213 3214 3215
    for (;;)
    {
      if (unlikely((num_bytes=my_net_read(net)) == packet_error))
      {
unknown's avatar
unknown committed
3216 3217 3218
        sql_print_error("Network read error downloading '%s' from master",
                        cev->fname);
        goto err;
unknown's avatar
unknown committed
3219 3220 3221
      }
      if (unlikely(!num_bytes)) /* eof */
      {
3222 3223
	/* 3.23 master wants it */
        net_write_command(net, 0, (uchar*) "", 0, (uchar*) "", 0);
unknown's avatar
unknown committed
3224 3225 3226 3227 3228 3229 3230
        /*
          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
3231 3232 3233 3234 3235
          break;
        Execute_load_log_event xev(thd,0,0);
        xev.log_pos = cev->log_pos;
        if (unlikely(mi->rli.relay_log.append(&xev)))
        {
3236 3237 3238
          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
3239 3240 3241 3242
          goto err;
        }
        mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total);
        break;
unknown's avatar
unknown committed
3243 3244 3245
      }
      if (unlikely(cev_not_written))
      {
unknown's avatar
unknown committed
3246
        cev->block = net->read_pos;
unknown's avatar
unknown committed
3247 3248 3249
        cev->block_len = num_bytes;
        if (unlikely(mi->rli.relay_log.append(cev)))
        {
3250 3251 3252
          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
3253 3254 3255 3256
          goto err;
        }
        cev_not_written=0;
        mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total);
unknown's avatar
unknown committed
3257 3258 3259
      }
      else
      {
unknown's avatar
unknown committed
3260
        aev.block = net->read_pos;
unknown's avatar
unknown committed
3261 3262 3263 3264
        aev.block_len = num_bytes;
        aev.log_pos = cev->log_pos;
        if (unlikely(mi->rli.relay_log.append(&aev)))
        {
3265 3266 3267
          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
3268 3269 3270
          goto err;
        }
        mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total) ;
unknown's avatar
unknown committed
3271 3272 3273 3274 3275
      }
    }
  }
  error=0;
err:
unknown's avatar
unknown committed
3276
  DBUG_RETURN(error);
unknown's avatar
unknown committed
3277
}
unknown's avatar
unknown committed
3278

3279

unknown's avatar
unknown committed
3280
/*
unknown's avatar
unknown committed
3281 3282 3283 3284
  Start using a new binary log on the master

  SYNOPSIS
    process_io_rotate()
unknown's avatar
unknown committed
3285 3286
    mi                  master_info for the slave
    rev                 The rotate log event read from the binary log
unknown's avatar
unknown committed
3287 3288

  DESCRIPTION
3289
    Updates the master info with the place in the next binary
unknown's avatar
unknown committed
3290
    log where we should start reading.
3291
    Rotate the relay log to avoid mixed-format relay logs.
unknown's avatar
unknown committed
3292 3293 3294 3295 3296

  NOTES
    We assume we already locked mi->data_lock

  RETURN VALUES
unknown's avatar
unknown committed
3297 3298
    0           ok
    1           Log event is illegal
unknown's avatar
unknown committed
3299 3300 3301

*/

3302
static int process_io_rotate(Master_info *mi, Rotate_log_event *rev)
3303
{
3304
  DBUG_ENTER("process_io_rotate");
unknown's avatar
unknown committed
3305
  safe_mutex_assert_owner(&mi->data_lock);
3306

unknown's avatar
unknown committed
3307
  if (unlikely(!rev->is_valid()))
3308
    DBUG_RETURN(1);
unknown's avatar
unknown committed
3309

3310
  /* Safe copy as 'rev' has been "sanitized" in Rotate_log_event's ctor */
unknown's avatar
unknown committed
3311 3312
  memcpy(mi->master_log_name, rev->new_log_ident, rev->ident_len+1);
  mi->master_log_pos= rev->pos;
unknown's avatar
unknown committed
3313
  DBUG_PRINT("info", ("master_log_pos: '%s' %lu",
unknown's avatar
unknown committed
3314
                      mi->master_log_name, (ulong) mi->master_log_pos));
3315
#ifndef DBUG_OFF
unknown's avatar
unknown committed
3316 3317 3318 3319 3320
  /*
    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)
3321
    mi->events_till_disconnect++;
3322
#endif
3323

3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337
  /*
    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);
  }
3338 3339 3340 3341
  /*
    Rotate the relay log makes binlog format detection easier (at next slave
    start or mysqlbinlog)
  */
3342
  rotate_relay_log(mi); /* will take the right mutexes */
3343
  DBUG_RETURN(0);
3344 3345
}

unknown's avatar
unknown committed
3346
/*
3347 3348
  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
3349
*/
3350
static int queue_binlog_ver_1_event(Master_info *mi, const char *buf,
unknown's avatar
unknown committed
3351
                           ulong event_len)
3352
{
unknown's avatar
unknown committed
3353
  const char *errmsg = 0;
unknown's avatar
unknown committed
3354 3355 3356
  ulong inc_pos;
  bool ignore_event= 0;
  char *tmp_buf = 0;
3357
  Relay_log_info *rli= &mi->rli;
3358
  DBUG_ENTER("queue_binlog_ver_1_event");
unknown's avatar
unknown committed
3359

unknown's avatar
unknown committed
3360 3361 3362
  /*
    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
3363 3364 3365 3366 3367
  */
  if (buf[EVENT_TYPE_OFFSET] == LOAD_EVENT)
  {
    if (unlikely(!(tmp_buf=(char*)my_malloc(event_len+1,MYF(MY_WME)))))
    {
3368 3369
      mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
                 ER(ER_SLAVE_FATAL_ERROR), "Memory allocation failed");
unknown's avatar
unknown committed
3370
      DBUG_RETURN(1);
unknown's avatar
unknown committed
3371 3372
    }
    memcpy(tmp_buf,buf,event_len);
3373 3374 3375 3376 3377 3378 3379 3380
    /*
      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
3381
    int4store(tmp_buf+EVENT_LEN_OFFSET, event_len);
unknown's avatar
unknown committed
3382 3383
    buf = (const char*)tmp_buf;
  }
3384 3385 3386 3387 3388 3389
  /*
    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
3390
  Log_event *ev = Log_event::read_log_event(buf,event_len, &errmsg,
3391
                                            mi->rli.relay_log.description_event_for_queue);
3392
  if (unlikely(!ev))
3393 3394
  {
    sql_print_error("Read invalid event from master: '%s',\
unknown's avatar
unknown committed
3395
 master could be corrupt but a more likely cause of this is a bug",
unknown's avatar
unknown committed
3396
                    errmsg);
unknown's avatar
unknown committed
3397 3398
    my_free((char*) tmp_buf, MYF(MY_ALLOW_ZERO_PTR));
    DBUG_RETURN(1);
3399
  }
3400

3401
  pthread_mutex_lock(&mi->data_lock);
3402
  ev->log_pos= mi->master_log_pos; /* 3.23 events don't contain log_pos */
unknown's avatar
unknown committed
3403
  switch (ev->get_type_code()) {
unknown's avatar
unknown committed
3404
  case STOP_EVENT:
3405
    ignore_event= 1;
unknown's avatar
unknown committed
3406 3407
    inc_pos= event_len;
    break;
3408
  case ROTATE_EVENT:
3409
    if (unlikely(process_io_rotate(mi,(Rotate_log_event*)ev)))
3410 3411
    {
      delete ev;
3412
      pthread_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
3413
      DBUG_RETURN(1);
3414
    }
unknown's avatar
unknown committed
3415
    inc_pos= 0;
3416
    break;
unknown's avatar
unknown committed
3417
  case CREATE_FILE_EVENT:
3418 3419 3420 3421 3422 3423
    /*
      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
3424
  {
unknown's avatar
unknown committed
3425
    /* We come here when and only when tmp_buf != 0 */
3426
    DBUG_ASSERT(tmp_buf != 0);
3427 3428
    inc_pos=event_len;
    ev->log_pos+= inc_pos;
unknown's avatar
unknown committed
3429
    int error = process_io_create_file(mi,(Create_file_log_event*)ev);
3430
    delete ev;
3431
    mi->master_log_pos += inc_pos;
unknown's avatar
unknown committed
3432
    DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
3433
    pthread_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
3434
    my_free((char*)tmp_buf, MYF(0));
unknown's avatar
unknown committed
3435
    DBUG_RETURN(error);
unknown's avatar
unknown committed
3436
  }
3437
  default:
unknown's avatar
unknown committed
3438
    inc_pos= event_len;
3439 3440
    break;
  }
unknown's avatar
unknown committed
3441
  if (likely(!ignore_event))
3442
  {
unknown's avatar
unknown committed
3443 3444
    if (ev->log_pos)
      /*
3445 3446 3447 3448
         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
3449
    if (unlikely(rli->relay_log.append(ev)))
3450 3451 3452
    {
      delete ev;
      pthread_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
3453
      DBUG_RETURN(1);
3454
    }
unknown's avatar
unknown committed
3455
    rli->relay_log.harvest_bytes_written(&rli->log_space_total);
3456 3457
  }
  delete ev;
unknown's avatar
unknown committed
3458
  mi->master_log_pos+= inc_pos;
unknown's avatar
unknown committed
3459
  DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
3460
  pthread_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
3461
  DBUG_RETURN(0);
3462 3463
}

3464 3465 3466 3467
/*
  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.
*/
3468
static int queue_binlog_ver_3_event(Master_info *mi, const char *buf,
unknown's avatar
unknown committed
3469
                           ulong event_len)
3470 3471 3472 3473
{
  const char *errmsg = 0;
  ulong inc_pos;
  char *tmp_buf = 0;
3474
  Relay_log_info *rli= &mi->rli;
3475 3476 3477 3478 3479 3480 3481 3482 3483
  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
3484
                    errmsg);
3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514
    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
3515
  DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526
  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
3527
  TODO:
3528 3529 3530 3531
    Test this code before release - it has to be tested on a separate
    setup with 3.23 master or 4.0 master
*/

3532
static int queue_old_event(Master_info *mi, const char *buf,
unknown's avatar
unknown committed
3533
                           ulong event_len)
3534
{
3535 3536
  DBUG_ENTER("queue_old_event");

3537 3538 3539
  switch (mi->rli.relay_log.description_event_for_queue->binlog_version)
  {
  case 1:
3540
      DBUG_RETURN(queue_binlog_ver_1_event(mi,buf,event_len));
3541
  case 3:
3542
      DBUG_RETURN(queue_binlog_ver_3_event(mi,buf,event_len));
3543 3544
  default: /* unsupported format; eg version 2 */
    DBUG_PRINT("info",("unsupported binlog format %d in queue_old_event()",
unknown's avatar
unknown committed
3545
                       mi->rli.relay_log.description_event_for_queue->binlog_version));
3546
    DBUG_RETURN(1);
3547 3548
  }
}
3549

unknown's avatar
unknown committed
3550
/*
3551 3552
  queue_event()

3553 3554 3555 3556 3557
  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
3558 3559
*/

3560
static int queue_event(Master_info* mi,const char* buf, ulong event_len)
3561
{
unknown's avatar
unknown committed
3562 3563
  int error= 0;
  ulong inc_pos;
3564
  Relay_log_info *rli= &mi->rli;
3565
  pthread_mutex_t *log_lock= rli->relay_log.get_log_lock();
unknown's avatar
unknown committed
3566 3567
  DBUG_ENTER("queue_event");

unknown's avatar
unknown committed
3568 3569
  LINT_INIT(inc_pos);

3570 3571
  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
3572
    DBUG_RETURN(queue_old_event(mi,buf,event_len));
3573

3574
  LINT_INIT(inc_pos);
3575
  pthread_mutex_lock(&mi->data_lock);
unknown's avatar
unknown committed
3576

unknown's avatar
unknown committed
3577
  switch (buf[EVENT_TYPE_OFFSET]) {
3578
  case STOP_EVENT:
3579 3580
    /*
      We needn't write this event to the relay log. Indeed, it just indicates a
unknown's avatar
unknown committed
3581 3582 3583
      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
3584
      prepared statements' deletion are TODO only when we binlog prep stmts).
unknown's avatar
unknown committed
3585

unknown's avatar
unknown committed
3586 3587 3588 3589
      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).
3590 3591
    */
    goto err;
3592 3593
  case ROTATE_EVENT:
  {
unknown's avatar
unknown committed
3594
    Rotate_log_event rev(buf,event_len,mi->rli.relay_log.description_event_for_queue);
3595
    if (unlikely(process_io_rotate(mi,&rev)))
unknown's avatar
unknown committed
3596
    {
3597 3598
      error= 1;
      goto err;
unknown's avatar
unknown committed
3599
    }
3600 3601 3602 3603
    /*
      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
3604
    inc_pos= 0;
3605 3606
    break;
  }
3607 3608 3609 3610 3611 3612 3613
  case FORMAT_DESCRIPTION_EVENT:
  {
    /*
      Create an event, and save it (when we rotate the relay log, we will have
      to write this event again).
    */
    /*
3614 3615 3616
      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).
3617
    */
3618
    Format_description_log_event* tmp;
3619
    const char* errmsg;
3620
    if (!(tmp= (Format_description_log_event*)
3621
          Log_event::read_log_event(buf, event_len, &errmsg,
3622
                                    mi->rli.relay_log.description_event_for_queue)))
3623 3624 3625 3626
    {
      error= 2;
      goto err;
    }
3627 3628
    delete mi->rli.relay_log.description_event_for_queue;
    mi->rli.relay_log.description_event_for_queue= tmp;
unknown's avatar
unknown committed
3629
    /*
3630
       Though this does some conversion to the slave's format, this will
unknown's avatar
unknown committed
3631
       preserve the master's binlog format version, and number of event types.
3632
    */
unknown's avatar
unknown committed
3633
    /*
3634
       If the event was not requested by the slave (the slave did not ask for
unknown's avatar
unknown committed
3635
       it), i.e. has end_log_pos=0, we do not increment mi->master_log_pos
3636 3637 3638
    */
    inc_pos= uint4korr(buf+LOG_POS_OFFSET) ? event_len : 0;
    DBUG_PRINT("info",("binlog format is now %d",
unknown's avatar
unknown committed
3639
                       mi->rli.relay_log.description_event_for_queue->binlog_version));
3640 3641 3642

  }
  break;
3643
  default:
unknown's avatar
unknown committed
3644
    inc_pos= event_len;
3645 3646
    break;
  }
3647

unknown's avatar
unknown committed
3648 3649
  /*
     If this event is originating from this server, don't queue it.
3650
     We don't check this for 3.23 events because it's simpler like this; 3.23
unknown's avatar
unknown committed
3651 3652
     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
3653 3654 3655 3656 3657 3658 3659 3660
     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!).
  */

3661 3662
  pthread_mutex_lock(log_lock);

unknown's avatar
unknown committed
3663
  if ((uint4korr(buf + SERVER_ID_OFFSET) == ::server_id) &&
3664
      !mi->rli.replicate_same_server_id)
3665
  {
3666 3667
    /*
      Do not write it to the relay log.
3668 3669 3670 3671 3672 3673 3674 3675
      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.
3676 3677 3678
      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.
3679
    */
3680 3681 3682
    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
3683
    {
3684
      mi->master_log_pos+= inc_pos;
unknown's avatar
unknown committed
3685 3686 3687 3688
      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;
    }
3689
    rli->relay_log.signal_update(); // the slave SQL thread needs to re-check
unknown's avatar
unknown committed
3690
    DBUG_PRINT("info", ("master_log_pos: %lu, event originating from the same server, ignored",
unknown's avatar
unknown committed
3691
                        (ulong) mi->master_log_pos));
unknown's avatar
unknown committed
3692
  }
unknown's avatar
unknown committed
3693 3694 3695
  else
  {
    /* write the event to the relay log */
3696
    if (likely(!(rli->relay_log.appendv(buf,event_len,0))))
3697 3698
    {
      mi->master_log_pos+= inc_pos;
unknown's avatar
unknown committed
3699
      DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
3700 3701
      rli->relay_log.harvest_bytes_written(&rli->log_space_total);
    }
3702
    else
3703
      error= 3;
3704
    rli->ign_master_log_name_end[0]= 0; // last event is not ignored
unknown's avatar
unknown committed
3705
  }
3706
  pthread_mutex_unlock(log_lock);
3707 3708 3709


err:
3710
  pthread_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
3711
  DBUG_PRINT("info", ("error: %d", error));
unknown's avatar
unknown committed
3712
  DBUG_RETURN(error);
3713 3714
}

3715

3716
void end_relay_log_info(Relay_log_info* rli)
3717
{
3718 3719
  DBUG_ENTER("end_relay_log_info");

3720
  if (!rli->inited)
3721
    DBUG_VOID_RETURN;
3722
  if (rli->info_fd >= 0)
unknown's avatar
unknown committed
3723 3724
  {
    end_io_cache(&rli->info_file);
3725
    (void) my_close(rli->info_fd, MYF(MY_WME));
unknown's avatar
unknown committed
3726 3727
    rli->info_fd = -1;
  }
3728
  if (rli->cur_log_fd >= 0)
unknown's avatar
unknown committed
3729 3730 3731 3732 3733
  {
    end_io_cache(&rli->cache_buf);
    (void)my_close(rli->cur_log_fd, MYF(MY_WME));
    rli->cur_log_fd = -1;
  }
3734
  rli->inited = 0;
3735
  rli->relay_log.close(LOG_CLOSE_INDEX | LOG_CLOSE_STOP_EVENT);
3736
  rli->relay_log.harvest_bytes_written(&rli->log_space_total);
unknown's avatar
unknown committed
3737 3738 3739 3740 3741 3742
  /*
    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();
3743
  DBUG_VOID_RETURN;
3744 3745
}

unknown's avatar
unknown committed
3746 3747
/*
  Try to connect until successful or slave killed
3748

unknown's avatar
unknown committed
3749 3750
  SYNPOSIS
    safe_connect()
unknown's avatar
unknown committed
3751 3752 3753
    thd                 Thread handler for slave
    mysql               MySQL connection handle
    mi                  Replication handle
3754

unknown's avatar
unknown committed
3755
  RETURN
unknown's avatar
unknown committed
3756 3757
    0   ok
    #   Error
unknown's avatar
unknown committed
3758
*/
3759

3760
static int safe_connect(THD* thd, MYSQL* mysql, Master_info* mi)
unknown's avatar
unknown committed
3761
{
3762 3763 3764
  DBUG_ENTER("safe_connect");

  DBUG_RETURN(connect_to_master(thd, mysql, mi, 0, 0));
unknown's avatar
unknown committed
3765 3766
}

3767

3768
/*
unknown's avatar
unknown committed
3769 3770
  SYNPOSIS
    connect_to_master()
unknown's avatar
unknown committed
3771

unknown's avatar
unknown committed
3772 3773 3774
  IMPLEMENTATION
    Try to connect until successful or slave killed or we have retried
    master_retry_count times
3775
*/
unknown's avatar
unknown committed
3776

3777
static int connect_to_master(THD* thd, MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
3778
                             bool reconnect, bool suppress_warnings)
unknown's avatar
unknown committed
3779
{
3780
  int slave_was_killed;
unknown's avatar
unknown committed
3781
  int last_errno= -2;                           // impossible error
3782
  ulong err_count=0;
unknown's avatar
unknown committed
3783
  char llbuff[22];
3784
  DBUG_ENTER("connect_to_master");
unknown's avatar
unknown committed
3785

unknown's avatar
unknown committed
3786
#ifndef DBUG_OFF
3787
  mi->events_till_disconnect = disconnect_slave_event_count;
unknown's avatar
unknown committed
3788
#endif
3789
  ulong client_flag= CLIENT_REMEMBER_OPTIONS;
3790
  if (opt_slave_compressed_protocol)
unknown's avatar
unknown committed
3791
    client_flag=CLIENT_COMPRESS;                /* We will use compression */
3792

3793 3794
  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
3795

unknown's avatar
unknown committed
3796 3797
#ifdef HAVE_OPENSSL
  if (mi->ssl)
3798
  {
unknown's avatar
unknown committed
3799
    mysql_ssl_set(mysql,
unknown's avatar
unknown committed
3800
                  mi->ssl_key[0]?mi->ssl_key:0,
unknown's avatar
unknown committed
3801
                  mi->ssl_cert[0]?mi->ssl_cert:0,
unknown's avatar
unknown committed
3802 3803 3804
                  mi->ssl_ca[0]?mi->ssl_ca:0,
                  mi->ssl_capath[0]?mi->ssl_capath:0,
                  mi->ssl_cipher[0]?mi->ssl_cipher:0);
3805 3806 3807
    mysql_options(mysql, MYSQL_OPT_SSL_VERIFY_SERVER_CERT,
                  &mi->ssl_verify_server_cert);
  }
unknown's avatar
unknown committed
3808 3809
#endif

3810 3811 3812 3813
  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);

3814
  while (!(slave_was_killed = io_slave_killed(thd,mi)) &&
unknown's avatar
unknown committed
3815 3816 3817
         (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
3818
  {
3819
    /* Don't repeat last error */
unknown's avatar
SCRUM  
unknown committed
3820
    if ((int)mysql_errno(mysql) != last_errno)
3821
    {
unknown's avatar
SCRUM  
unknown committed
3822
      last_errno=mysql_errno(mysql);
unknown's avatar
unknown committed
3823
      suppress_warnings= 0;
3824 3825 3826 3827 3828 3829
      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);
3830
    }
unknown's avatar
unknown committed
3831 3832 3833
    /*
      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
3834
      do not want to have election triggered on the first failure to
unknown's avatar
unknown committed
3835
      connect
3836
    */
3837
    if (++err_count == master_retry_count)
3838 3839
    {
      slave_was_killed=1;
unknown's avatar
unknown committed
3840 3841
      if (reconnect)
        change_rpl_status(RPL_ACTIVE_SLAVE,RPL_LOST_SOLDIER);
3842 3843
      break;
    }
3844
    safe_sleep(thd,mi->connect_retry,(CHECK_KILLED_FUNC)io_slave_killed,
unknown's avatar
unknown committed
3845
               (void*)mi);
unknown's avatar
unknown committed
3846
  }
3847

3848 3849
  if (!slave_was_killed)
  {
3850
    mi->clear_error(); // clear possible left over reconnect error
unknown's avatar
unknown committed
3851
    if (reconnect)
unknown's avatar
unknown committed
3852
    {
3853
      if (!suppress_warnings && global_system_variables.log_warnings)
unknown's avatar
unknown committed
3854
        sql_print_information("Slave: connected to master '%s@%s:%d',\
3855
replication resumed in log '%s' at position %s", mi->user,
unknown's avatar
unknown committed
3856 3857 3858
                        mi->host, mi->port,
                        IO_RPL_LOG_NAME,
                        llstr(mi->master_log_pos,llbuff));
unknown's avatar
unknown committed
3859
    }
unknown's avatar
unknown committed
3860 3861 3862
    else
    {
      change_rpl_status(RPL_IDLE_SLAVE,RPL_ACTIVE_SLAVE);
3863 3864
      general_log_print(thd, COM_CONNECT_OUT, "%s@%s:%d",
                        mi->user, mi->host, mi->port);
unknown's avatar
unknown committed
3865
    }
3866
#ifdef SIGNAL_WITH_VIO_CLOSE
3867
    thd->set_active_vio(mysql->net.vio);
unknown's avatar
unknown committed
3868
#endif
3869
  }
3870
  mysql->reconnect= 1;
3871 3872
  DBUG_PRINT("exit",("slave_was_killed: %d", slave_was_killed));
  DBUG_RETURN(slave_was_killed);
unknown's avatar
unknown committed
3873 3874
}

3875

unknown's avatar
unknown committed
3876
/*
3877
  safe_reconnect()
unknown's avatar
unknown committed
3878

unknown's avatar
unknown committed
3879 3880 3881
  IMPLEMENTATION
    Try to connect until successful or slave killed or we have retried
    master_retry_count times
unknown's avatar
unknown committed
3882 3883
*/

3884
static int safe_reconnect(THD* thd, MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
3885
                          bool suppress_warnings)
unknown's avatar
unknown committed
3886
{
unknown's avatar
unknown committed
3887 3888
  DBUG_ENTER("safe_reconnect");
  DBUG_RETURN(connect_to_master(thd, mysql, mi, 1, suppress_warnings));
unknown's avatar
unknown committed
3889 3890
}

unknown's avatar
unknown committed
3891

3892 3893 3894 3895 3896 3897
/*
  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
3898
    rli                 Relay log information
3899 3900 3901 3902 3903 3904 3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915 3916

  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
3917 3918
    0   ok
    1   write error
3919 3920
*/

3921
bool flush_relay_log_info(Relay_log_info* rli)
3922
{
3923
  bool error=0;
3924
  DBUG_ENTER("flush_relay_log_info");
3925 3926

  if (unlikely(rli->no_storage))
3927
    DBUG_RETURN(0);
3928

3929 3930 3931
  IO_CACHE *file = &rli->info_file;
  char buff[FN_REFLEN*2+22*2+4], *pos;

3932
  my_b_seek(file, 0L);
3933
  pos=strmov(buff, rli->group_relay_log_name);
3934
  *pos++='\n';
3935
  pos=longlong2str(rli->group_relay_log_pos, pos, 10);
3936
  *pos++='\n';
3937
  pos=strmov(pos, rli->group_master_log_name);
3938
  *pos++='\n';
3939
  pos=longlong2str(rli->group_master_log_pos, pos, 10);
3940
  *pos='\n';
3941
  if (my_b_write(file, (uchar*) buff, (size_t) (pos-buff)+1))
3942 3943 3944
    error=1;
  if (flush_io_cache(file))
    error=1;
3945

3946
  /* Flushing the relay log is done by the slave I/O thread */
3947
  DBUG_RETURN(error);
3948 3949
}

3950

unknown's avatar
unknown committed
3951
/*
3952
  Called when we notice that the current "hot" log got rotated under our feet.
unknown's avatar
unknown committed
3953 3954
*/

3955
static IO_CACHE *reopen_relay_log(Relay_log_info *rli, const char **errmsg)
3956
{
3957
  DBUG_ENTER("reopen_relay_log");
3958 3959
  DBUG_ASSERT(rli->cur_log != &rli->cache_buf);
  DBUG_ASSERT(rli->cur_log_fd == -1);
unknown's avatar
unknown committed
3960 3961

  IO_CACHE *cur_log = rli->cur_log=&rli->cache_buf;
3962
  if ((rli->cur_log_fd=open_binlog(cur_log,rli->event_relay_log_name,
unknown's avatar
unknown committed
3963
                                   errmsg)) <0)
unknown's avatar
unknown committed
3964
    DBUG_RETURN(0);
3965 3966
  /*
    We want to start exactly where we was before:
unknown's avatar
unknown committed
3967 3968
    relay_log_pos       Current log pos
    pending             Number of bytes already processed from the event
3969
  */
3970
  rli->event_relay_log_pos= max(rli->event_relay_log_pos, BIN_LOG_HEADER_SIZE);
3971
  my_b_seek(cur_log,rli->event_relay_log_pos);
unknown's avatar
unknown committed
3972
  DBUG_RETURN(cur_log);
3973 3974
}

unknown's avatar
unknown committed
3975

3976 3977 3978 3979 3980 3981 3982 3983 3984 3985
/**
  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.
*/
3986
static Log_event* next_event(Relay_log_info* rli)
3987 3988 3989
{
  Log_event* ev;
  IO_CACHE* cur_log = rli->cur_log;
3990
  pthread_mutex_t *log_lock = rli->relay_log.get_log_lock();
3991 3992
  const char* errmsg=0;
  THD* thd = rli->sql_thd;
unknown's avatar
unknown committed
3993
  DBUG_ENTER("next_event");
3994

3995 3996
  DBUG_ASSERT(thd != 0);

3997 3998 3999 4000 4001
#ifndef DBUG_OFF
  if (abort_slave_event_count && !rli->events_till_abort--)
    DBUG_RETURN(0);
#endif

unknown's avatar
unknown committed
4002 4003
  /*
    For most operations we need to protect rli members with data_lock,
4004 4005 4006 4007
    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
4008
  */
4009
  safe_mutex_assert_owner(&rli->data_lock);
unknown's avatar
unknown committed
4010

4011
  while (!sql_slave_killed(thd,rli))
unknown's avatar
unknown committed
4012 4013 4014
  {
    /*
      We can have two kinds of log reading:
unknown's avatar
unknown committed
4015 4016 4017 4018 4019 4020 4021 4022
      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
4023
    */
4024 4025 4026 4027 4028
    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
4029 4030

      /*
unknown's avatar
unknown committed
4031 4032
        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
4033
      */
unknown's avatar
unknown committed
4034
      if (rli->relay_log.get_open_count() != rli->cur_log_old_open_count)
4035
      {
unknown's avatar
unknown committed
4036 4037 4038 4039 4040 4041
        // 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
4042 4043
      }
    }
4044 4045 4046 4047 4048 4049 4050
    /* 
      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;
4051 4052
#ifndef DBUG_OFF
    {
4053
      /* This is an assertion which sometimes fails, let's try to track it */
4054
      char llbuf1[22], llbuf2[22];
4055
      DBUG_PRINT("info", ("my_b_tell(cur_log)=%s rli->event_relay_log_pos=%s",
4056
                          llstr(my_b_tell(cur_log),llbuf1),
4057 4058 4059
                          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);
4060 4061
    }
#endif
unknown's avatar
unknown committed
4062 4063
    /*
      Relay log is always in new format - if the master is 3.23, the
4064
      I/O thread will convert the format for us.
4065 4066
      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
4067 4068
      logs, which may even have been deleted. So we need to write this
      description event at the beginning of the relay log.
4069 4070
      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.
4071
      But if the relay log is created by new_file(): then the solution is:
4072
      MYSQL_BIN_LOG::open() will write the buffered description event.
unknown's avatar
unknown committed
4073
    */
4074 4075
    if ((ev=Log_event::read_log_event(cur_log,0,
                                      rli->relay_log.description_event_for_exec)))
4076

4077 4078
    {
      DBUG_ASSERT(thd==rli->sql_thd);
4079 4080 4081 4082 4083
      /*
        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);
4084
      if (hot_log)
unknown's avatar
unknown committed
4085
        pthread_mutex_unlock(log_lock);
unknown's avatar
unknown committed
4086
      DBUG_RETURN(ev);
4087 4088
    }
    DBUG_ASSERT(thd==rli->sql_thd);
unknown's avatar
unknown committed
4089
    if (opt_reckless_slave)                     // For mysql-test
unknown's avatar
unknown committed
4090
      cur_log->error = 0;
unknown's avatar
unknown committed
4091
    if (cur_log->error < 0)
unknown's avatar
unknown committed
4092 4093
    {
      errmsg = "slave SQL thread aborted because of I/O error";
unknown's avatar
unknown committed
4094
      if (hot_log)
unknown's avatar
unknown committed
4095
        pthread_mutex_unlock(log_lock);
unknown's avatar
unknown committed
4096 4097
      goto err;
    }
4098 4099
    if (!cur_log->error) /* EOF */
    {
unknown's avatar
unknown committed
4100
      /*
unknown's avatar
unknown committed
4101 4102 4103
        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
4104
      */
4105 4106
      if (hot_log)
      {
4107 4108 4109 4110 4111 4112 4113 4114 4115 4116
        /*
          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").
4117 4118 4119 4120 4121 4122 4123 4124 4125 4126

          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.
4127
        */
4128
        time_t save_timestamp= rli->last_master_timestamp;
4129 4130
        rli->last_master_timestamp= 0;

unknown's avatar
unknown committed
4131
        DBUG_ASSERT(rli->relay_log.get_open_count() ==
unknown's avatar
unknown committed
4132
                    rli->cur_log_old_open_count);
4133 4134 4135 4136 4137

        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"));
4138
          ev= new Rotate_log_event(rli->ign_master_log_name_end,
4139
                                   0, rli->ign_master_log_pos_end,
unknown's avatar
unknown committed
4140
                                   Rotate_log_event::DUP_NAME);
4141
          rli->ign_master_log_name_end[0]= 0;
unknown's avatar
unknown committed
4142
          pthread_mutex_unlock(log_lock);
4143 4144 4145 4146 4147 4148 4149 4150 4151 4152
          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
4153 4154 4155 4156 4157
        /*
          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);
4158 4159

        /*
unknown's avatar
unknown committed
4160
          Possible deadlock :
4161 4162 4163 4164 4165 4166 4167 4168 4169 4170 4171
          - 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
4172
          master does not send anything). Then the I/O thread stops waiting
4173 4174
          and reads more events.
          The SQL thread decides when the I/O thread should take log_space_limit
unknown's avatar
unknown committed
4175
          into account again : ignore_log_space_limit is reset to 0
4176 4177 4178 4179
          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
4180 4181
          be stopped, and the SQL thread sets ignore_log_space_limit to 0 when
          it stops.
4182 4183 4184
        */
        pthread_mutex_lock(&rli->log_space_lock);
        // prevent the I/O thread from blocking next times
unknown's avatar
unknown committed
4185
        rli->ignore_log_space_limit= 1;
4186
        /*
4187 4188 4189 4190
          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.
4191
        */
4192
        pthread_mutex_unlock(&rli->log_space_lock);
4193
        pthread_cond_broadcast(&rli->log_space_cond);
4194
        // Note that wait_for_update unlocks lock_log !
4195
        rli->relay_log.wait_for_update(rli->sql_thd, 1);
4196 4197
        // re-acquire data lock since we released it earlier
        pthread_mutex_lock(&rli->data_lock);
4198
        rli->last_master_timestamp= save_timestamp;
unknown's avatar
unknown committed
4199
        continue;
4200
      }
unknown's avatar
unknown committed
4201
      /*
unknown's avatar
unknown committed
4202 4203 4204
        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
4205 4206 4207 4208 4209
      */
      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
4210

4211
      if (relay_log_purge)
unknown's avatar
unknown committed
4212
      {
unknown's avatar
unknown committed
4213
        /*
4214 4215 4216 4217 4218 4219 4220 4221 4222 4223
          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
4224
        if (rli->relay_log.purge_first_log
4225 4226 4227
            (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
4228 4229 4230 4231
        {
          errmsg = "Error purging processed logs";
          goto err;
        }
unknown's avatar
unknown committed
4232
      }
4233 4234
      else
      {
unknown's avatar
unknown committed
4235 4236 4237 4238 4239 4240 4241 4242 4243 4244 4245 4246 4247 4248 4249 4250
        /*
          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
4251
      }
4252 4253 4254 4255 4256 4257 4258 4259 4260 4261 4262 4263 4264 4265

      /*
        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
4266 4267
      if (rli->relay_log.is_active(rli->linfo.log_file_name))
      {
4268
#ifdef EXTRA_DEBUG
unknown's avatar
unknown committed
4269 4270
        if (global_system_variables.log_warnings)
          sql_print_information("next log '%s' is currently active",
4271
                                rli->linfo.log_file_name);
unknown's avatar
unknown committed
4272 4273 4274 4275 4276 4277 4278 4279
#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.
4280 4281 4282
          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
4283 4284
        */
        if (check_binlog_magic(cur_log,&errmsg))
4285 4286
        {
          if (!hot_log) pthread_mutex_unlock(log_lock);
unknown's avatar
unknown committed
4287
          goto err;
4288 4289
        }
        if (!hot_log) pthread_mutex_unlock(log_lock);
unknown's avatar
unknown committed
4290
        continue;
4291
      }
4292
      if (!hot_log) pthread_mutex_unlock(log_lock);
unknown's avatar
unknown committed
4293
      /*
unknown's avatar
unknown committed
4294 4295 4296
        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
4297 4298
      */
#ifdef EXTRA_DEBUG
unknown's avatar
unknown committed
4299
      if (global_system_variables.log_warnings)
unknown's avatar
unknown committed
4300
        sql_print_information("next log '%s' is not active",
4301
                              rli->linfo.log_file_name);
unknown's avatar
unknown committed
4302
#endif
unknown's avatar
unknown committed
4303 4304
      // 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
4305 4306
                                       &errmsg)) <0)
        goto err;
4307
    }
unknown's avatar
unknown committed
4308
    else
4309
    {
unknown's avatar
unknown committed
4310
      /*
unknown's avatar
unknown committed
4311 4312
        Read failed with a non-EOF error.
        TODO: come up with something better to handle this error
unknown's avatar
unknown committed
4313 4314
      */
      if (hot_log)
unknown's avatar
unknown committed
4315
        pthread_mutex_unlock(log_lock);
4316
      sql_print_error("Slave SQL thread: I/O error reading \
unknown's avatar
unknown committed
4317
event(errno: %d  cur_log->error: %d)",
unknown's avatar
unknown committed
4318
                      my_errno,cur_log->error);
unknown's avatar
unknown committed
4319
      // set read position to the beginning of the event
4320
      my_b_seek(cur_log,rli->event_relay_log_pos);
unknown's avatar
unknown committed
4321 4322
      /* otherwise, we have had a partial read */
      errmsg = "Aborting slave SQL thread because of partial event read";
unknown's avatar
unknown committed
4323
      break;                                    // To end of function
4324 4325
    }
  }
4326
  if (!errmsg && global_system_variables.log_warnings)
4327
  {
unknown's avatar
unknown committed
4328
    sql_print_information("Error reading relay log event: %s",
4329 4330 4331
                          "slave SQL thread was killed");
    DBUG_RETURN(0);
  }
unknown's avatar
unknown committed
4332

4333
err:
4334 4335
  if (errmsg)
    sql_print_error("Error reading relay log event: %s", errmsg);
unknown's avatar
unknown committed
4336
  DBUG_RETURN(0);
4337 4338
}

4339 4340 4341
/*
  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
4342
  locks; here we don't, so this function is mainly taking locks).
4343 4344
  Returns nothing as we cannot catch any error (MYSQL_BIN_LOG::new_file()
  is void).
4345 4346
*/

4347
void rotate_relay_log(Master_info* mi)
4348 4349
{
  DBUG_ENTER("rotate_relay_log");
4350
  Relay_log_info* rli= &mi->rli;
unknown's avatar
unknown committed
4351

4352 4353 4354
  /* We don't lock rli->run_lock. This would lead to deadlocks. */
  pthread_mutex_lock(&mi->run_lock);

unknown's avatar
unknown committed
4355
  /*
unknown's avatar
unknown committed
4356 4357 4358
     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).
  */
4359 4360
  if (!rli->inited)
  {
unknown's avatar
unknown committed
4361
    DBUG_PRINT("info", ("rli->inited == 0"));
unknown's avatar
unknown committed
4362
    goto end;
4363
  }
unknown's avatar
unknown committed
4364

4365
  /* If the relay log is closed, new_file() will do nothing. */
4366
  rli->relay_log.new_file();
unknown's avatar
unknown committed
4367

4368 4369 4370 4371
  /*
    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
4372 4373 4374 4375 4376 4377
    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.
4378 4379 4380 4381
    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
4382
end:
4383
  pthread_mutex_unlock(&mi->run_lock);
4384 4385 4386
  DBUG_VOID_RETURN;
}

unknown's avatar
unknown committed
4387

unknown's avatar
unknown committed
4388 4389 4390
/**
   Detects, based on master's version (as found in the relay log), if master
   has a certain bug.
4391
   @param rli Relay_log_info which tells the master's version
unknown's avatar
unknown committed
4392
   @param bug_id Number of the bug as found in bugs.mysql.com
4393
   @param report bool report error message, default TRUE
4394 4395 4396 4397 4398 4399 4400

   @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
4401 4402
   @return TRUE if master has the bug, FALSE if it does not.
*/
4403 4404
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
4405 4406 4407 4408 4409 4410 4411 4412
{
  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
4413
    {24432, { 5, 0, 24 }, { 5, 0, 38 } },
4414 4415 4416
    {24432, { 5, 1, 12 }, { 5, 1, 17 } },
    {33029, { 5, 0,  0 }, { 5, 0, 58 } },
    {33029, { 5, 1,  0 }, { 5, 1, 12 } },
4417
    {37426, { 5, 1,  0 }, { 5, 1, 26 } },
unknown's avatar
unknown committed
4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430
  };
  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) &&
4431 4432
        (memcmp(fixed_in,      master_ver, 3) >  0) &&
        (pred == NULL || (*pred)(param)))
unknown's avatar
unknown committed
4433
    {
4434 4435
      if (!report)
	return TRUE;
unknown's avatar
unknown committed
4436 4437 4438 4439 4440
      // 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);
4441
      // a verbose message for the error log
4442 4443 4444
      rli->report(ERROR_LEVEL, ER_UNKNOWN_ERROR,
                  "According to the master's version ('%s'),"
                  " it is probable that master suffers from this bug:"
4445 4446 4447 4448 4449 4450 4451 4452 4453 4454 4455 4456 4457 4458 4459 4460
                      " 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
4461 4462 4463 4464 4465 4466
      return TRUE;
    }
  }
  return FALSE;
}

4467 4468 4469 4470 4471 4472 4473 4474 4475 4476 4477 4478 4479 4480 4481
/**
   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;
4482
    DBUG_EXECUTE_IF("simulate_bug33029", return TRUE;);
4483
    return rpl_master_has_bug(rli, 33029, FALSE, NULL, NULL);
4484 4485 4486 4487
  }
  return FALSE;
}

4488
#ifdef HAVE_EXPLICIT_TEMPLATE_INSTANTIATION
unknown's avatar
unknown committed
4489
template class I_List_iterator<i_string>;
unknown's avatar
unknown committed
4490
template class I_List_iterator<i_string_pair>;
unknown's avatar
unknown committed
4491
#endif
4492

4493 4494 4495 4496
/**
  @} (end of group Replication)
*/

unknown's avatar
SCRUM  
unknown committed
4497
#endif /* HAVE_REPLICATION */