slave.cc 169 KB
Newer Older
Marc Alff's avatar
Marc Alff committed
1
/* Copyright (C) 2000-2003 MySQL AB, 2008-2009 Sun Microsystems, Inc
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.
*/

27 28
#include "sql_priv.h"
#include "my_global.h"
29
#include "slave.h"
30 31
#include "sql_parse.h"                         // execute_init_command
#include "sql_table.h"                         // mysql_rm_table
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"
Konstantin Osipov's avatar
Konstantin Osipov committed
37
#include "transaction.h"
unknown's avatar
unknown committed
38
#include <thr_alarm.h>
unknown's avatar
unknown committed
39
#include <my_dir.h>
unknown's avatar
unknown committed
40
#include <sql_common.h>
41
#include <errmsg.h>
42
#include <mysqld_error.h>
43
#include <mysys_err.h>
He Zhenxing's avatar
He Zhenxing committed
44
#include "rpl_handler.h"
45 46 47 48 49 50 51 52 53
#include <signal.h>
#include <mysql.h>
#include <myisam.h>

#include "sql_base.h"                           // close_thread_tables
#include "tztime.h"                             // struct Time_zone
#include "log_event.h"                          // Rotate_log_event,
                                                // Create_file_log_event,
                                                // Format_description_log_event
unknown's avatar
SCRUM  
unknown committed
54

55 56 57
#ifdef HAVE_REPLICATION

#include "rpl_tblmap.h"
58
#include "debug_sync.h"
59

60
#define FLAGSTR(V,F) ((V)&(F)?#F" ":"")
61

62
#define MAX_SLAVE_RETRY_PAUSE 5
63 64 65 66
/*
  a parameter of sql_slave_killed() to defer the killed status
*/
#define SLAVE_WAIT_GROUP_DONE 60
67 68
bool use_slave_mask = 0;
MY_BITMAP slave_error_mask;
69
char slave_skip_error_names[SHOW_VAR_FUNC_BUFF_SIZE];
70

71 72
typedef bool (*CHECK_KILLED_FUNC)(THD*,void*);

73
char* slave_load_tmpdir = 0;
74
Master_info *active_mi= 0;
75
my_bool replicate_same_server_id;
76
ulonglong relay_log_space_limit = 0;
unknown's avatar
unknown committed
77 78 79 80 81

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

82
  TODO: move the vars below under Master_info
unknown's avatar
unknown committed
83
*/
84

85
int disconnect_slave_event_count = 0, abort_slave_event_count = 0;
unknown's avatar
unknown committed
86

He Zhenxing's avatar
He Zhenxing committed
87 88
static pthread_key(Master_info*, RPL_MASTER_INFO);

89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140
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"
  }
};
 

141
typedef enum { SLAVE_THD_IO, SLAVE_THD_SQL} SLAVE_THD_TYPE;
unknown's avatar
unknown committed
142

143 144
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);
145
static bool wait_for_relay_log_space(Relay_log_info* rli);
146
static inline bool io_slave_killed(THD* thd,Master_info* mi);
147
static inline bool sql_slave_killed(THD* thd,Relay_log_info* rli);
148
static int init_slave_thread(THD* thd, SLAVE_THD_TYPE thd_type);
149
static void print_slave_skip_errors(void);
150 151
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
152
                          bool suppress_warnings);
153
static int connect_to_master(THD* thd, MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
154
                             bool reconnect, bool suppress_warnings);
155
static int safe_sleep(THD* thd, int sec, CHECK_KILLED_FUNC thread_killed,
unknown's avatar
unknown committed
156
                      void* thread_killed_arg);
157
static int get_master_version_and_clock(MYSQL* mysql, Master_info* mi);
158
static Log_event* next_event(Relay_log_info* rli);
159
static int queue_event(Master_info* mi,const char* buf,ulong event_len);
160
static int terminate_slave_thread(THD *thd,
Marc Alff's avatar
Marc Alff committed
161 162
                                  mysql_mutex_t *term_lock,
                                  mysql_cond_t *term_cond,
163 164
                                  volatile uint *slave_running,
                                  bool skip_lock);
165
static bool check_io_slave_killed(THD *thd, Master_info *mi, const char *info);
166 167

/*
unknown's avatar
unknown committed
168
  Find out which replications threads are running
169

unknown's avatar
unknown committed
170 171
  SYNOPSIS
    init_thread_mask()
unknown's avatar
unknown committed
172 173 174
    mask                Return value here
    mi                  master_info for slave
    inverse             If set, returns which threads are not running
unknown's avatar
unknown committed
175 176 177 178 179 180

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

  RETURN
unknown's avatar
unknown committed
181 182
    mask        If inverse == 0, running threads
                If inverse == 1, stopped threads
183 184
*/

185
void init_thread_mask(int* mask,Master_info* mi,bool inverse)
186 187 188
{
  bool set_io = mi->slave_running, set_sql = mi->rli.slave_running;
  register int tmp_mask=0;
189 190
  DBUG_ENTER("init_thread_mask");

191 192 193 194
  if (set_io)
    tmp_mask |= SLAVE_IO;
  if (set_sql)
    tmp_mask |= SLAVE_SQL;
195 196
  if (inverse)
    tmp_mask^= (SLAVE_IO | SLAVE_SQL);
197
  *mask = tmp_mask;
198
  DBUG_VOID_RETURN;
199 200
}

201

unknown's avatar
unknown committed
202
/*
203
  lock_slave_threads()
unknown's avatar
unknown committed
204
*/
205

206
void lock_slave_threads(Master_info* mi)
207
{
208 209
  DBUG_ENTER("lock_slave_threads");

210
  //TODO: see if we can do this without dual mutex
Marc Alff's avatar
Marc Alff committed
211 212
  mysql_mutex_lock(&mi->run_lock);
  mysql_mutex_lock(&mi->rli.run_lock);
213
  DBUG_VOID_RETURN;
214 215
}

216

unknown's avatar
unknown committed
217
/*
218
  unlock_slave_threads()
unknown's avatar
unknown committed
219
*/
220

221
void unlock_slave_threads(Master_info* mi)
222
{
223 224
  DBUG_ENTER("unlock_slave_threads");

225
  //TODO: see if we can do this without dual mutex
Marc Alff's avatar
Marc Alff committed
226 227
  mysql_mutex_unlock(&mi->rli.run_lock);
  mysql_mutex_unlock(&mi->run_lock);
228
  DBUG_VOID_RETURN;
229 230
}

Marc Alff's avatar
Marc Alff committed
231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251
#ifdef HAVE_PSI_INTERFACE
static PSI_thread_key key_thread_slave_io, key_thread_slave_sql;

static PSI_thread_info all_slave_threads[]=
{
  { &key_thread_slave_io, "slave_io", PSI_FLAG_GLOBAL},
  { &key_thread_slave_sql, "slave_sql", PSI_FLAG_GLOBAL}
};

static void init_slave_psi_keys(void)
{
  const char* category= "sql";
  int count;

  if (PSI_server == NULL)
    return;

  count= array_elements(all_slave_threads);
  PSI_server->register_thread(category, all_slave_threads, count);
}
#endif /* HAVE_PSI_INTERFACE */
252

unknown's avatar
unknown committed
253
/* Initialize slave structures */
254

255 256
int init_slave()
{
257
  DBUG_ENTER("init_slave");
258
  int error= 0;
259

Marc Alff's avatar
Marc Alff committed
260 261 262 263
#ifdef HAVE_PSI_INTERFACE
  init_slave_psi_keys();
#endif

264 265 266 267 268
  /*
    This is called when mysqld starts. Before client connections are
    accepted. However bootstrap may conflict with us if it does START SLAVE.
    So it's safer to take the lock.
  */
Marc Alff's avatar
Marc Alff committed
269
  mysql_mutex_lock(&LOCK_active_mi);
270 271 272 273
  /*
    TODO: re-write this to interate through the list of files
    for multi-master
  */
274
  active_mi= new Master_info(relay_log_recovery);
275

He Zhenxing's avatar
He Zhenxing committed
276 277
  if (pthread_key_create(&RPL_MASTER_INFO, NULL))
    goto err;
278

279 280 281 282 283 284 285 286 287
  /*
    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();
  }

288
  /*
289 290 291
    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.
292
  */
293 294 295
  if (!active_mi)
  {
    sql_print_error("Failed to allocate memory for the master info structure");
296
    error= 1;
297 298
    goto err;
  }
299

unknown's avatar
unknown committed
300
  if (init_master_info(active_mi,master_info_file,relay_log_info_file,
301
                       1, (SLAVE_IO | SLAVE_SQL)))
302
  {
303
    sql_print_error("Failed to initialize the master info structure");
304
    error= 1;
unknown's avatar
unknown committed
305
    goto err;
306
  }
307

308 309
  /* If server id is not set, start_slave_thread() will say it */

310
  if (active_mi->host[0] && !opt_skip_slave_start)
311
  {
312
    if (start_slave_threads(1 /* need mutex */,
unknown's avatar
unknown committed
313 314 315 316 317
                            0 /* no wait for start*/,
                            active_mi,
                            master_info_file,
                            relay_log_info_file,
                            SLAVE_IO | SLAVE_SQL))
unknown's avatar
unknown committed
318
    {
319
      sql_print_error("Failed to create slave threads");
320
      error= 1;
unknown's avatar
unknown committed
321 322
      goto err;
    }
323 324
  }

unknown's avatar
unknown committed
325
err:
Marc Alff's avatar
Marc Alff committed
326
  mysql_mutex_unlock(&LOCK_active_mi);
327
  DBUG_RETURN(error);
328
}
329

330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356
/*
  Updates the master info based on the information stored in the
  relay info and ignores relay logs previously retrieved by the IO 
  thread, which thus starts fetching again based on to the  
  group_master_log_pos and group_master_log_name. Eventually, the old
  relay logs will be purged by the normal purge mechanism.

  In the feature, we should improve this routine in order to avoid throwing
  away logs that are safely stored in the disk. Note also that this recovery 
  routine relies on the correctness of the relay-log.info and only tolerates 
  coordinate problems in master.info.
  
  In this function, there is no need for a mutex as the caller 
  (i.e. init_slave) already has one acquired.
  
  Specifically, the following structures are updated:
 
  1 - mi->master_log_pos  <-- rli->group_master_log_pos
  2 - mi->master_log_name <-- rli->group_master_log_name
  3 - It moves the relay log to the new relay log file, by
      rli->group_relay_log_pos  <-- BIN_LOG_HEADER_SIZE;
      rli->event_relay_log_pos  <-- BIN_LOG_HEADER_SIZE;
      rli->group_relay_log_name <-- rli->relay_log.get_log_fname();
      rli->event_relay_log_name <-- rli->relay_log.get_log_fname();
  
   If there is an error, it returns (1), otherwise returns (0).
 */
357
int init_recovery(Master_info* mi, const char** errmsg)
358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378
{
  DBUG_ENTER("init_recovery");
 
  Relay_log_info *rli= &mi->rli;
  if (rli->group_master_log_name[0])
  {
    mi->master_log_pos= max(BIN_LOG_HEADER_SIZE,
                             rli->group_master_log_pos);
    strmake(mi->master_log_name, rli->group_master_log_name,
            sizeof(mi->master_log_name)-1);
 
    sql_print_warning("Recovery from master pos %ld and file %s.",
                      (ulong) mi->master_log_pos, mi->master_log_name);
 
    strmake(rli->group_relay_log_name, rli->relay_log.get_log_fname(),
            sizeof(rli->group_relay_log_name)-1);
    strmake(rli->event_relay_log_name, rli->relay_log.get_log_fname(),
            sizeof(mi->rli.event_relay_log_name)-1);
 
    rli->group_relay_log_pos= rli->event_relay_log_pos= BIN_LOG_HEADER_SIZE;
  }
379

380 381 382
  DBUG_RETURN(0);
}
 
383 384 385 386 387 388 389 390 391 392 393 394
/**
  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.
  */
395
  const size_t MIN_ROOM= 10;
396 397 398 399
  DBUG_ENTER("print_slave_skip_errors");
  DBUG_ASSERT(sizeof(slave_skip_error_names) > MIN_ROOM);
  DBUG_ASSERT(MAX_SLAVE_ERROR <= 999999); // 6 digits

400 401 402
  /* Make @@slave_skip_errors show the nice human-readable value.  */
  opt_slave_skip_errors= slave_skip_error_names;

403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420
  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;

421
    for (errnum= 0; errnum < MAX_SLAVE_ERROR; errnum++)
422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443
    {
      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
444
/*
445
  Init function to set up array for errors that should be skipped for slave
unknown's avatar
unknown committed
446

unknown's avatar
unknown committed
447 448
  SYNOPSIS
    init_slave_skip_errors()
unknown's avatar
unknown committed
449
    arg         List of errors numbers to skip, separated with ','
unknown's avatar
unknown committed
450 451 452 453

  NOTES
    Called from get_options() in mysqld.cc on start-up
*/
454

unknown's avatar
unknown committed
455
void init_slave_skip_errors(const char* arg)
456
{
unknown's avatar
unknown committed
457
  const char *p;
458 459
  DBUG_ENTER("init_slave_skip_errors");

460
  if (bitmap_init(&slave_error_mask,0,MAX_SLAVE_ERROR,0))
461 462 463 464 465
  {
    fprintf(stderr, "Badly out of memory, please check your system status\n");
    exit(1);
  }
  use_slave_mask = 1;
466
  for (;my_isspace(system_charset_info,*arg);++arg)
467
    /* empty */;
468
  if (!my_strnncoll(system_charset_info,(uchar*)arg,4,(const uchar*)"all",4))
469 470
  {
    bitmap_set_all(&slave_error_mask);
471
    print_slave_skip_errors();
472
    DBUG_VOID_RETURN;
473 474 475 476 477 478 479 480
  }
  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);
481
    while (!my_isdigit(system_charset_info,*p) && *p)
482 483
      p++;
  }
484 485
  /* Convert slave skip errors bitmap into a printable string. */
  print_slave_skip_errors();
486
  DBUG_VOID_RETURN;
487 488
}

489
static void set_thd_in_use_temporary_tables(Relay_log_info *rli)
490 491 492 493 494 495
{
  TABLE *table;

  for (table= rli->save_temporary_tables ; table ; table= table->next)
    table->in_use= rli->sql_thd;
}
496

497
int terminate_slave_threads(Master_info* mi,int thread_mask,bool skip_lock)
498
{
499 500
  DBUG_ENTER("terminate_slave_threads");

501
  if (!mi->inited)
502
    DBUG_RETURN(0); /* successfully do nothing */
503
  int error,force_all = (thread_mask & SLAVE_FORCE_ALL);
Marc Alff's avatar
Marc Alff committed
504
  mysql_mutex_t *sql_lock = &mi->rli.run_lock, *io_lock = &mi->run_lock;
505
  mysql_mutex_t *log_lock= mi->rli.relay_log.get_log_lock();
unknown's avatar
unknown committed
506

507
  if (thread_mask & (SLAVE_IO|SLAVE_FORCE_ALL))
508
  {
unknown's avatar
unknown committed
509
    DBUG_PRINT("info",("Terminating IO thread"));
510
    mi->abort_slave=1;
511
    if ((error=terminate_slave_thread(mi->io_thd, io_lock,
unknown's avatar
unknown committed
512
                                      &mi->stop_cond,
513 514
                                      &mi->slave_running,
                                      skip_lock)) &&
unknown's avatar
unknown committed
515
        !force_all)
516
      DBUG_RETURN(error);
517

518
    mysql_mutex_lock(log_lock);
519 520 521 522

    DBUG_PRINT("info",("Flushing relay log and master info file."));
    if (current_thd)
      thd_proc_info(current_thd, "Flushing relay log and master info files.");
523
    if (flush_master_info(mi, TRUE, FALSE))
524 525 526 527 528 529 530 531
      DBUG_RETURN(ER_ERROR_DURING_FLUSH_LOGS);

    if (my_sync(mi->rli.relay_log.get_log_file()->file, MYF(MY_WME)))
      DBUG_RETURN(ER_ERROR_DURING_FLUSH_LOGS);

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

532
    mysql_mutex_unlock(log_lock);
533
  }
534
  if (thread_mask & (SLAVE_SQL|SLAVE_FORCE_ALL))
535
  {
unknown's avatar
unknown committed
536
    DBUG_PRINT("info",("Terminating SQL thread"));
537
    mi->rli.abort_slave=1;
538
    if ((error=terminate_slave_thread(mi->rli.sql_thd, sql_lock,
unknown's avatar
unknown committed
539
                                      &mi->rli.stop_cond,
540 541
                                      &mi->rli.slave_running,
                                      skip_lock)) &&
unknown's avatar
unknown committed
542
        !force_all)
543
      DBUG_RETURN(error);
544

545
    mysql_mutex_lock(log_lock);
546 547 548 549 550 551 552 553 554 555

    DBUG_PRINT("info",("Flushing relay-log info file."));
    if (current_thd)
      thd_proc_info(current_thd, "Flushing relay-log info file.");
    if (flush_relay_log_info(&mi->rli))
      DBUG_RETURN(ER_ERROR_DURING_FLUSH_LOGS);
    
    if (my_sync(mi->rli.info_fd, MYF(MY_WME)))
      DBUG_RETURN(ER_ERROR_DURING_FLUSH_LOGS);

556
    mysql_mutex_unlock(log_lock);
557
  }
558
  DBUG_RETURN(0); 
559 560
}

561

562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588
/**
   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
589
   @retval 0 All OK ER_SLAVE_NOT_RUNNING otherwise.
590

Andrei Elkin's avatar
Andrei Elkin committed
591 592 593
   @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.
594

595 596 597
 */
static int
terminate_slave_thread(THD *thd,
Marc Alff's avatar
Marc Alff committed
598 599
                       mysql_mutex_t *term_lock,
                       mysql_cond_t *term_cond,
600 601
                       volatile uint *slave_running,
                       bool skip_lock)
602
{
603
  DBUG_ENTER("terminate_slave_thread");
604
  if (!skip_lock)
605
  {
Marc Alff's avatar
Marc Alff committed
606
    mysql_mutex_lock(term_lock);
607 608 609
  }
  else
  {
Marc Alff's avatar
Marc Alff committed
610
    mysql_mutex_assert_owner(term_lock);
611
  }
612 613 614
  if (!*slave_running)
  {
    if (!skip_lock)
615
    {
616 617 618 619
      /*
        if run_lock (term_lock) is acquired locally then either
        slave_running status is fine
      */
Marc Alff's avatar
Marc Alff committed
620
      mysql_mutex_unlock(term_lock);
621 622 623 624
      DBUG_RETURN(0);
    }
    else
    {
625
      DBUG_RETURN(ER_SLAVE_NOT_RUNNING);
626 627 628
    }
  }
  DBUG_ASSERT(thd != 0);
629
  THD_CHECK_SENTRY(thd);
630

unknown's avatar
unknown committed
631
  /*
632
    Is is critical to test if the slave is running. Otherwise, we might
unknown's avatar
unknown committed
633
    be referening freed memory trying to kick it
634
  */
unknown's avatar
unknown committed
635

unknown's avatar
unknown committed
636
  while (*slave_running)                        // Should always be true
637
  {
638
    int error;
639
    DBUG_PRINT("loop", ("killing slave thread"));
640

Marc Alff's avatar
Marc Alff committed
641
    mysql_mutex_lock(&thd->LOCK_thd_data);
642 643 644 645 646 647
#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)
    */
648
    int err __attribute__((unused))= pthread_kill(thd->real_id, thr_client_alarm);
649 650 651
    DBUG_ASSERT(err != EINVAL);
#endif
    thd->awake(THD::NOT_KILLED);
Marc Alff's avatar
Marc Alff committed
652
    mysql_mutex_unlock(&thd->LOCK_thd_data);
653

unknown's avatar
unknown committed
654 655 656
    /*
      There is a small chance that slave thread might miss the first
      alarm. To protect againts it, resend the signal until it reacts
657 658
    */
    struct timespec abstime;
unknown's avatar
unknown committed
659
    set_timespec(abstime,2);
Marc Alff's avatar
Marc Alff committed
660
    error= mysql_cond_timedwait(term_cond, term_lock, &abstime);
661
    DBUG_ASSERT(error == ETIMEDOUT || error == 0);
662
  }
663 664 665 666

  DBUG_ASSERT(*slave_running == 0);

  if (!skip_lock)
Marc Alff's avatar
Marc Alff committed
667
    mysql_mutex_unlock(term_lock);
668
  DBUG_RETURN(0);
669 670
}

671

Marc Alff's avatar
Marc Alff committed
672 673 674 675 676 677 678
int start_slave_thread(
#ifdef HAVE_PSI_INTERFACE
                       PSI_thread_key thread_key,
#endif
                       pthread_handler h_func, mysql_mutex_t *start_lock,
                       mysql_mutex_t *cond_lock,
                       mysql_cond_t *start_cond,
unknown's avatar
unknown committed
679 680
                       volatile uint *slave_running,
                       volatile ulong *slave_run_id,
Konstantin Osipov's avatar
Konstantin Osipov committed
681
                       Master_info* mi)
682 683
{
  pthread_t th;
unknown's avatar
unknown committed
684 685 686
  ulong start_id;
  DBUG_ENTER("start_slave_thread");

687 688
  DBUG_ASSERT(mi->inited);

689
  if (start_lock)
Marc Alff's avatar
Marc Alff committed
690
    mysql_mutex_lock(start_lock);
691 692 693
  if (!server_id)
  {
    if (start_cond)
Marc Alff's avatar
Marc Alff committed
694
      mysql_cond_broadcast(start_cond);
695
    if (start_lock)
Marc Alff's avatar
Marc Alff committed
696
      mysql_mutex_unlock(start_lock);
697
    sql_print_error("Server id not set, will not start slave");
unknown's avatar
unknown committed
698
    DBUG_RETURN(ER_BAD_SLAVE);
699
  }
unknown's avatar
unknown committed
700

701
  if (*slave_running)
702 703
  {
    if (start_cond)
Marc Alff's avatar
Marc Alff committed
704
      mysql_cond_broadcast(start_cond);
705
    if (start_lock)
Marc Alff's avatar
Marc Alff committed
706
      mysql_mutex_unlock(start_lock);
unknown's avatar
unknown committed
707
    DBUG_RETURN(ER_SLAVE_MUST_STOP);
708
  }
unknown's avatar
unknown committed
709 710
  start_id= *slave_run_id;
  DBUG_PRINT("info",("Creating new slave thread"));
Marc Alff's avatar
Marc Alff committed
711 712
  if (mysql_thread_create(thread_key,
                          &th, &connection_attrib, h_func, (void*)mi))
713 714
  {
    if (start_lock)
Marc Alff's avatar
Marc Alff committed
715
      mysql_mutex_unlock(start_lock);
unknown's avatar
unknown committed
716
    DBUG_RETURN(ER_SLAVE_THREAD);
717
  }
unknown's avatar
unknown committed
718
  if (start_cond && cond_lock) // caller has cond_lock
719 720
  {
    THD* thd = current_thd;
unknown's avatar
unknown committed
721
    while (start_id == *slave_run_id)
722
    {
unknown's avatar
unknown committed
723
      DBUG_PRINT("sleep",("Waiting for slave thread to start"));
724 725 726 727 728 729 730 731 732 733 734
      const char *old_msg= thd->enter_cond(start_cond, cond_lock,
                                           "Waiting for slave thread to start");
      /*
        It is not sufficient to test this at loop bottom. We must test
        it after registering the mutex in enter_cond(). If the kill
        happens after testing of thd->killed and before the mutex is
        registered, we could otherwise go waiting though thd->killed is
        set.
      */
      if (!thd->killed)
        mysql_cond_wait(start_cond, cond_lock);
735
      thd->exit_cond(old_msg);
Marc Alff's avatar
Marc Alff committed
736
      mysql_mutex_lock(cond_lock); // re-acquire it as exit_cond() released
737
      if (thd->killed)
738 739
      {
        if (start_lock)
Marc Alff's avatar
Marc Alff committed
740
          mysql_mutex_unlock(start_lock);
unknown's avatar
unknown committed
741
        DBUG_RETURN(thd->killed_errno());
742
      }
743 744 745
    }
  }
  if (start_lock)
Marc Alff's avatar
Marc Alff committed
746
    mysql_mutex_unlock(start_lock);
unknown's avatar
unknown committed
747
  DBUG_RETURN(0);
748
}
unknown's avatar
unknown committed
749

750

unknown's avatar
unknown committed
751
/*
752
  start_slave_threads()
unknown's avatar
unknown committed
753

unknown's avatar
unknown committed
754 755 756 757
  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
758
*/
unknown's avatar
unknown committed
759

760
int start_slave_threads(bool need_slave_mutex, bool wait_for_start,
761
                        Master_info* mi, const char* master_info_fname,
unknown's avatar
unknown committed
762
                        const char* slave_info_fname, int thread_mask)
763
{
Marc Alff's avatar
Marc Alff committed
764 765
  mysql_mutex_t *lock_io=0, *lock_sql=0, *lock_cond_io=0, *lock_cond_sql=0;
  mysql_cond_t* cond_io=0, *cond_sql=0;
766
  int error=0;
767
  DBUG_ENTER("start_slave_threads");
unknown's avatar
unknown committed
768

769 770 771 772 773 774 775 776 777 778 779 780
  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;
  }
781 782

  if (thread_mask & SLAVE_IO)
Marc Alff's avatar
Marc Alff committed
783 784 785 786 787 788 789 790
    error= start_slave_thread(
#ifdef HAVE_PSI_INTERFACE
                              key_thread_slave_io,
#endif
                              handle_slave_io, lock_io, lock_cond_io,
                              cond_io,
                              &mi->slave_running, &mi->slave_run_id,
                              mi);
791
  if (!error && (thread_mask & SLAVE_SQL))
792
  {
Marc Alff's avatar
Marc Alff committed
793 794 795 796 797 798 799 800
    error= start_slave_thread(
#ifdef HAVE_PSI_INTERFACE
                              key_thread_slave_sql,
#endif
                              handle_slave_sql, lock_sql, lock_cond_sql,
                              cond_sql,
                              &mi->rli.slave_running, &mi->rli.slave_run_id,
                              mi);
801
    if (error)
802
      terminate_slave_threads(mi, thread_mask & SLAVE_IO, !need_slave_mutex);
803
  }
804
  DBUG_RETURN(error);
805
}
806

807

unknown's avatar
unknown committed
808
/*
809
  Release slave threads at time of executing shutdown.
unknown's avatar
unknown committed
810 811 812 813

  SYNOPSIS
    end_slave()
*/
814

815 816
void end_slave()
{
817 818
  DBUG_ENTER("end_slave");

819 820 821 822 823 824 825
  /*
    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.
  */
Marc Alff's avatar
Marc Alff committed
826
  mysql_mutex_lock(&LOCK_active_mi);
unknown's avatar
unknown committed
827 828 829 830 831 832 833 834
  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);
835
  }
Marc Alff's avatar
Marc Alff committed
836
  mysql_mutex_unlock(&LOCK_active_mi);
837 838 839 840 841 842 843 844 845 846 847 848 849 850
  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()
{
Marc Alff's avatar
Marc Alff committed
851
  mysql_mutex_lock(&LOCK_active_mi);
852 853
  if (active_mi)
  {
unknown's avatar
unknown committed
854 855 856 857
    end_master_info(active_mi);
    delete active_mi;
    active_mi= 0;
  }
Marc Alff's avatar
Marc Alff committed
858
  mysql_mutex_unlock(&LOCK_active_mi);
859
}
unknown's avatar
unknown committed
860

861
static bool io_slave_killed(THD* thd, Master_info* mi)
unknown's avatar
unknown committed
862
{
863 864
  DBUG_ENTER("io_slave_killed");

865
  DBUG_ASSERT(mi->io_thd == thd);
866
  DBUG_ASSERT(mi->slave_running); // tracking buffer overrun
867
  DBUG_RETURN(mi->abort_slave || abort_loop || thd->killed);
unknown's avatar
unknown committed
868 869
}

870 871 872 873 874 875
/**
   The function analyzes a possible killed status and makes
   a decision whether to accept it or not.
   Normally upon accepting the sql thread goes to shutdown.
   In the event of deffering decision @rli->last_event_start_time waiting
   timer is set to force the killed status be accepted upon its expiration.
876

877 878
   @param thd   pointer to a THD instance
   @param rli   pointer to Relay_log_info instance
879

880 881 882
   @return TRUE the killed status is recognized, FALSE a possible killed
           status is deferred.
*/
883
static bool sql_slave_killed(THD* thd, Relay_log_info* rli)
884
{
885
  bool ret= FALSE;
886 887
  DBUG_ENTER("sql_slave_killed");

888 889
  DBUG_ASSERT(rli->sql_thd == thd);
  DBUG_ASSERT(rli->slave_running == 1);// tracking buffer overrun
890 891
  if (abort_loop || thd->killed || rli->abort_slave)
  {
892 893 894 895 896 897 898 899
    /*
      The transaction should always be binlogged if OPTION_KEEP_LOG is set
      (it implies that something can not be rolled back). And such case
      should be regarded similarly as modifing a non-transactional table
      because retrying of the transaction will lead to an error or inconsistency
      as well.
      Example: OPTION_KEEP_LOG is set if a temporary table is created or dropped.
    */
unknown's avatar
unknown committed
900 901 902
    if ((thd->transaction.all.modified_non_trans_table ||
         (thd->variables.option_bits & OPTION_KEEP_LOG))
        && rli->is_in_group())
903
    {
904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959
      char msg_stopped[]=
        "... The slave SQL is stopped, leaving the current group "
        "of events unfinished with a non-transaction table changed. "
        "If the group consists solely of Row-based events, you can try "
        "restarting the slave with --slave-exec-mode=IDEMPOTENT, which "
        "ignores duplicate key, key not found, and similar errors (see "
        "documentation for details).";

      if (rli->abort_slave)
      {
        DBUG_PRINT("info", ("Slave SQL thread is being stopped in the middle of"
                            " a group having updated a non-trans table, giving"
                            " it some grace period"));

        /*
          Slave sql thread shutdown in face of unfinished group modified 
          Non-trans table is handled via a timer. The slave may eventually
          give out to complete the current group and in that case there
          might be issues at consequent slave restart, see the error message.
          WL#2975 offers a robust solution requiring to store the last exectuted
          event's coordinates along with the group's coordianates
          instead of waiting with @c last_event_start_time the timer.
        */

        if (rli->last_event_start_time == 0)
          rli->last_event_start_time= my_time(0);
        ret= difftime(my_time(0), rli->last_event_start_time) <=
          SLAVE_WAIT_GROUP_DONE ? FALSE : TRUE;

        DBUG_EXECUTE_IF("stop_slave_middle_group", 
                        DBUG_EXECUTE_IF("incomplete_group_in_relay_log",
                                        ret= TRUE;);); // time is over

        if (ret == 0)
        {
          rli->report(WARNING_LEVEL, 0,
                      "slave SQL thread is being stopped in the middle "
                      "of applying of a group having updated a non-transaction "
                      "table; waiting for the group completion ... ");
        }
        else
        {
          rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
                      ER(ER_SLAVE_FATAL_ERROR), msg_stopped);
        }
      }
      else
      {
        ret= TRUE;
        rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, ER(ER_SLAVE_FATAL_ERROR),
                    msg_stopped);
      }
    }
    else
    {
      ret= TRUE;
960 961
    }
  }
962 963 964 965
  if (ret)
    rli->last_event_start_time= 0;
  
  DBUG_RETURN(ret);
966 967
}

968

unknown's avatar
unknown committed
969
/*
970 971
  skip_load_data_infile()

unknown's avatar
unknown committed
972 973 974
  NOTES
    This is used to tell a 3.23 master to break send_file()
*/
975 976 977

void skip_load_data_infile(NET *net)
{
978 979
  DBUG_ENTER("skip_load_data_infile");

980
  (void)net_request_file(net, "/dev/null");
unknown's avatar
unknown committed
981
  (void)my_net_read(net);                               // discard response
982
  (void)net_write_command(net, 0, (uchar*) "", 0, (uchar*) "", 0); // ok
983
  DBUG_VOID_RETURN;
984 985
}

986

987
bool net_request_file(NET* net, const char* fname)
988
{
989
  DBUG_ENTER("net_request_file");
990 991
  DBUG_RETURN(net_write_command(net, 251, (uchar*) fname, strlen(fname),
                                (uchar*) "", 0));
992 993
}

994 995
/*
  From other comments and tests in code, it looks like
996
  sometimes Query_log_event and Load_log_event can have db == 0
997 998 999
  (see rewrite_db() above for example)
  (cases where this happens are unclear; it may be when the master is 3.23).
*/
1000 1001

const char *print_slave_db_safe(const char* db)
1002
{
1003 1004 1005
  DBUG_ENTER("*print_slave_db_safe");

  DBUG_RETURN((db ? db : ""));
1006
}
1007

1008
int init_strvar_from_file(char *var, int max_size, IO_CACHE *f,
unknown's avatar
unknown committed
1009
                                 const char *default_val)
unknown's avatar
unknown committed
1010
{
unknown's avatar
unknown committed
1011
  uint length;
1012 1013
  DBUG_ENTER("init_strvar_from_file");

unknown's avatar
unknown committed
1014 1015 1016 1017 1018 1019
  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
1020
    {
unknown's avatar
unknown committed
1021
      /*
unknown's avatar
unknown committed
1022 1023
        If we truncated a line or stopped on last char, remove all chars
        up to and including newline.
unknown's avatar
unknown committed
1024
      */
unknown's avatar
unknown committed
1025
      int c;
1026
      while (((c=my_b_get(f)) != '\n' && c != my_b_EOF)) ;
unknown's avatar
unknown committed
1027
    }
1028
    DBUG_RETURN(0);
unknown's avatar
unknown committed
1029 1030 1031
  }
  else if (default_val)
  {
unknown's avatar
unknown committed
1032
    strmake(var,  default_val, max_size-1);
1033
    DBUG_RETURN(0);
unknown's avatar
unknown committed
1034
  }
1035
  DBUG_RETURN(1);
unknown's avatar
unknown committed
1036 1037
}

1038

1039
int init_intvar_from_file(int* var, IO_CACHE* f, int default_val)
unknown's avatar
unknown committed
1040 1041
{
  char buf[32];
1042 1043
  DBUG_ENTER("init_intvar_from_file");

unknown's avatar
unknown committed
1044 1045

  if (my_b_gets(f, buf, sizeof(buf)))
unknown's avatar
unknown committed
1046 1047
  {
    *var = atoi(buf);
1048
    DBUG_RETURN(0);
unknown's avatar
unknown committed
1049
  }
unknown's avatar
unknown committed
1050
  else if (default_val)
unknown's avatar
unknown committed
1051 1052
  {
    *var = default_val;
1053
    DBUG_RETURN(0);
unknown's avatar
unknown committed
1054
  }
1055
  DBUG_RETURN(1);
unknown's avatar
unknown committed
1056 1057
}

Andrei Elkin's avatar
Andrei Elkin committed
1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078
int init_floatvar_from_file(float* var, IO_CACHE* f, float default_val)
{
  char buf[16];
  DBUG_ENTER("init_floatvar_from_file");


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

Andrei Elkin's avatar
Andrei Elkin committed
1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130

/**
   A master info read method

   This function is called from @c init_master_info() along with
   relatives to restore some of @c active_mi members.
   Particularly, this function is responsible for restoring
   IGNORE_SERVER_IDS list of servers whose events the slave is
   going to ignore (to not log them in the relay log).
   Items being read are supposed to be decimal output of values of a
   type shorter or equal of @c long and separated by the single space.

   @param arr         @c DYNAMIC_ARRAY pointer to storage for servers id
   @param f           @c IO_CACHE pointer to the source file

   @retval 0         All OK
   @retval non-zero  An error
*/

int init_dynarray_intvar_from_file(DYNAMIC_ARRAY* arr, IO_CACHE* f)
{
  int ret= 0;
  char buf[16 * (sizeof(long)*4 + 1)]; // static buffer to use most of times
  char *buf_act= buf; // actual buffer can be dynamic if static is short
  char *token, *last;
  uint num_items;     // number of items of `arr'
  size_t read_size;
  DBUG_ENTER("init_dynarray_intvar_from_file");

  if ((read_size= my_b_gets(f, buf_act, sizeof(buf))) == 0)
  {
    return 0; // no line in master.info
  }
  if (read_size + 1 == sizeof(buf) && buf[sizeof(buf) - 2] != '\n')
  {
    /*
      short read happend; allocate sufficient memory and make the 2nd read
    */
    char buf_work[(sizeof(long)*3 + 1)*16];
    memcpy(buf_work, buf, sizeof(buf_work));
    num_items= atoi(strtok_r(buf_work, " ", &last));
    size_t snd_size;
    /*
      max size lower bound approximate estimation bases on the formula:
      (the items number + items themselves) * 
          (decimal size + space) - 1 + `\n' + '\0'
    */
    size_t max_size= (1 + num_items) * (sizeof(long)*3 + 1) + 1;
    buf_act= (char*) my_malloc(max_size, MYF(MY_WME));
    memcpy(buf_act, buf, read_size);
    snd_size= my_b_gets(f, buf_act + read_size, max_size - read_size);
    if (snd_size == 0 ||
1131
        ((snd_size + 1 == max_size - read_size) &&  buf[max_size - 2] != '\n'))
Andrei Elkin's avatar
Andrei Elkin committed
1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162
    {
      /*
        failure to make the 2nd read or short read again
      */
      ret= 1;
      goto err;
    }
  }
  token= strtok_r(buf_act, " ", &last);
  if (token == NULL)
  {
    ret= 1;
    goto err;
  }
  num_items= atoi(token);
  for (uint i=0; i < num_items; i++)
  {
    token= strtok_r(NULL, " ", &last);
    if (token == NULL)
    {
      ret= 1;
      goto err;
    }
    else
    {
      ulong val= atol(token);
      insert_dynamic(arr, (uchar *) &val);
    }
  }
err:
  if (buf_act != buf)
1163
    my_free(buf_act);
Andrei Elkin's avatar
Andrei Elkin committed
1164 1165 1166 1167
  DBUG_RETURN(ret);
}


1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189
/*
  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;   
}


1190 1191 1192 1193 1194 1195 1196 1197
/*
  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
1198

1199 1200 1201
  RETURNS
  0       ok
  1       error
1202
  2       transient network problem, the caller should try to reconnect
1203
*/
1204

1205
static int get_master_version_and_clock(MYSQL* mysql, Master_info* mi)
1206
{
1207
  char err_buff[MAX_SLAVE_ERRMSG];
1208
  const char* errmsg= 0;
1209 1210 1211
  int err_code= 0;
  MYSQL_RES *master_res= 0;
  MYSQL_ROW master_row;
1212
  DBUG_ENTER("get_master_version_and_clock");
1213 1214 1215 1216 1217 1218 1219

  /*
    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
1220

1221
  if (!my_isdigit(&my_charset_bin,*mysql->server_version))
1222
  {
unknown's avatar
unknown committed
1223
    errmsg = "Master reported unrecognized MySQL version";
1224 1225 1226
    err_code= ER_SLAVE_FATAL_ERROR;
    sprintf(err_buff, ER(err_code), errmsg);
  }
1227 1228 1229 1230 1231
  else
  {
    /*
      Note the following switch will bug when we have MySQL branch 30 ;)
    */
unknown's avatar
unknown committed
1232
    switch (*mysql->server_version)
1233 1234 1235 1236 1237
    {
    case '0':
    case '1':
    case '2':
      errmsg = "Master reported unrecognized MySQL version";
1238 1239
      err_code= ER_SLAVE_FATAL_ERROR;
      sprintf(err_buff, ER(err_code), errmsg);
1240 1241 1242
      break;
    case '3':
      mi->rli.relay_log.description_event_for_queue= new
unknown's avatar
unknown committed
1243
        Format_description_log_event(1, mysql->server_version);
1244 1245 1246
      break;
    case '4':
      mi->rli.relay_log.description_event_for_queue= new
unknown's avatar
unknown committed
1247
        Format_description_log_event(3, mysql->server_version);
1248
      break;
unknown's avatar
unknown committed
1249
    default:
1250 1251 1252 1253 1254 1255 1256 1257 1258
      /*
        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
1259
        Format_description_log_event(4, mysql->server_version);
1260 1261
      break;
    }
1262
  }
unknown's avatar
unknown committed
1263 1264

  /*
1265 1266 1267 1268 1269
     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
1270

1271
  if (errmsg)
1272
    goto err;
1273 1274 1275 1276

  /* as we are here, we tried to allocate the event */
  if (!mi->rli.relay_log.description_event_for_queue)
  {
1277 1278 1279 1280
    errmsg= "default Format_description_log_event";
    err_code= ER_SLAVE_CREATE_EVENT_FAILURE;
    sprintf(err_buff, ER(err_code), errmsg);
    goto err;
1281 1282
  }

1283 1284 1285 1286
  /*
    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
1287

1288 1289 1290 1291 1292 1293 1294 1295 1296 1297
  DBUG_EXECUTE_IF("dbug.before_get_UNIX_TIMESTAMP",
                  {
                    const char act[]=
                      "now "
                      "wait_for signal.get_unix_timestamp";
                    DBUG_ASSERT(opt_debug_sync_timeout > 0);
                    DBUG_ASSERT(!debug_sync_set_action(current_thd,
                                                       STRING_WITH_LEN(act)));
                  };);

1298
  master_res= NULL;
1299
  if (!mysql_real_query(mysql, STRING_WITH_LEN("SELECT UNIX_TIMESTAMP()")) &&
1300 1301
      (master_res= mysql_store_result(mysql)) &&
      (master_row= mysql_fetch_row(master_res)))
unknown's avatar
unknown committed
1302
  {
unknown's avatar
unknown committed
1303
    mi->clock_diff_with_master=
1304
      (long) (time((time_t*) 0) - strtoul(master_row[0], 0, 10));
unknown's avatar
unknown committed
1305
  }
1306 1307
  else if (check_io_slave_killed(mi->io_thd, mi, NULL))
    goto slave_killed_err;
1308 1309 1310 1311 1312 1313 1314
  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
1315
  {
1316
    mi->clock_diff_with_master= 0; /* The "most sensible" value */
1317 1318 1319 1320
    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));
1321 1322
  }
  if (master_res)
1323
  {
unknown's avatar
unknown committed
1324
    mysql_free_result(master_res);
1325 1326
    master_res= NULL;
  }
unknown's avatar
unknown committed
1327

1328 1329 1330 1331 1332 1333 1334 1335 1336 1337
  /*
    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.
  */
1338 1339 1340 1341 1342 1343 1344 1345 1346
  DBUG_EXECUTE_IF("dbug.before_get_SERVER_ID",
                  {
                    const char act[]=
                      "now "
                      "wait_for signal.get_server_id";
                    DBUG_ASSERT(opt_debug_sync_timeout > 0);
                    DBUG_ASSERT(!debug_sync_set_action(current_thd, 
                                                       STRING_WITH_LEN(act)));
                  };);
1347 1348
  master_res= NULL;
  master_row= NULL;
1349 1350
  if (!mysql_real_query(mysql,
                        STRING_WITH_LEN("SHOW VARIABLES LIKE 'SERVER_ID'")) &&
1351 1352
      (master_res= mysql_store_result(mysql)) &&
      (master_row= mysql_fetch_row(master_res)))
1353
  {
Andrei Elkin's avatar
Andrei Elkin committed
1354
    if ((::server_id == (mi->master_id= strtoul(master_row[1], 0, 10))) &&
1355
        !mi->rli.replicate_same_server_id)
1356
    {
1357 1358 1359 1360
      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).";
1361 1362
      err_code= ER_SLAVE_FATAL_ERROR;
      sprintf(err_buff, ER(err_code), errmsg);
1363
      goto err;
1364
    }
1365 1366 1367
  }
  else if (mysql_errno(mysql))
  {
1368 1369 1370
    if (check_io_slave_killed(mi->io_thd, mi, NULL))
      goto slave_killed_err;
    else if (is_network_error(mysql_errno(mysql)))
1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390
    {
      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)
  {
1391
    mysql_free_result(master_res);
1392
    master_res= NULL;
unknown's avatar
unknown committed
1393
  }
Andrei Elkin's avatar
Andrei Elkin committed
1394 1395 1396 1397 1398 1399 1400
  if (mi->master_id == 0 && mi->ignore_server_ids.elements > 0)
  {
    errmsg= "Slave configured with server id filtering could not detect the master server id.";
    err_code= ER_SLAVE_FATAL_ERROR;
    sprintf(err_buff, ER(err_code), errmsg);
    goto err;
  }
unknown's avatar
unknown committed
1401

1402 1403 1404
  /*
    Check that the master's global character_set_server and ours are the same.
    Not fatal if query fails (old master?).
1405 1406 1407 1408 1409 1410
    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.
1411 1412 1413 1414 1415 1416
    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.
1417
  */
1418 1419 1420

  /* redundant with rest of code but safer against later additions */
  if (*mysql->server_version == '3')
1421
    goto err;
1422

1423
  if (*mysql->server_version == '4')
unknown's avatar
unknown committed
1424
  {
1425 1426 1427 1428 1429
    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)))
1430
    {
1431 1432 1433
      if (strcmp(master_row[0], global_system_variables.collation_server->name))
      {
        errmsg= "The slave I/O thread stops because master and slave have \
1434
different values for the COLLATION_SERVER global variable. The values must \
1435 1436 1437 1438 1439
be equal for the Statement-format replication to work";
        err_code= ER_SLAVE_FATAL_ERROR;
        sprintf(err_buff, ER(err_code), errmsg);
        goto err;
      }
1440
    }
1441 1442
    else if (check_io_slave_killed(mi->io_thd, mi, NULL))
      goto slave_killed_err;
1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455
    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));
1456
      goto err;
1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468
    }
    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
1469
  }
1470

1471 1472 1473 1474 1475 1476 1477 1478
  /*
    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
1479 1480 1481
    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
1482
    'SYSTEM' time zone).
1483 1484
    This check is only necessary for 4.x masters (and < 5.0.4 masters but
    those were alpha).
1485
  */
1486
  if (*mysql->server_version == '4')
unknown's avatar
unknown committed
1487
  {
1488 1489 1490 1491
    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)))
1492
    {
1493 1494 1495 1496
      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 \
1497
different values for the TIME_ZONE global variable. The values must \
1498 1499 1500 1501 1502
be equal for the Statement-format replication to work";
        err_code= ER_SLAVE_FATAL_ERROR;
        sprintf(err_buff, ER(err_code), errmsg);
        goto err;
      }
1503
    }
1504 1505
    else if (check_io_slave_killed(mi->io_thd, mi, NULL))
      goto slave_killed_err;
1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518
    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));
1519
      goto err;
1520 1521 1522 1523 1524 1525
    }
    if (master_res)
    {
      mysql_free_result(master_res);
      master_res= NULL;
    }
unknown's avatar
unknown committed
1526 1527
  }

Andrei Elkin's avatar
Andrei Elkin committed
1528 1529 1530 1531 1532 1533 1534 1535 1536
  if (mi->heartbeat_period != 0.0)
  {
    char llbuf[22];
    const char query_format[]= "SET @master_heartbeat_period= %s";
    char query[sizeof(query_format) - 2 + sizeof(llbuf)];
    /* 
       the period is an ulonglong of nano-secs. 
    */
    llstr((ulonglong) (mi->heartbeat_period*1000000000UL), llbuf);
1537
    sprintf(query, query_format, llbuf);
Andrei Elkin's avatar
Andrei Elkin committed
1538 1539 1540 1541

    if (mysql_real_query(mysql, query, strlen(query))
        && !check_io_slave_killed(mi->io_thd, mi, NULL))
    {
Andrei Elkin's avatar
Andrei Elkin committed
1542 1543
      errmsg= "The slave I/O thread stops because SET @master_heartbeat_period "
        "on master failed.";
Andrei Elkin's avatar
Andrei Elkin committed
1544
      err_code= ER_SLAVE_FATAL_ERROR;
Andrei Elkin's avatar
Andrei Elkin committed
1545
      sprintf(err_buff, "%s Error: %s", errmsg, mysql_error(mysql));
Andrei Elkin's avatar
Andrei Elkin committed
1546 1547 1548 1549 1550 1551 1552
      mysql_free_result(mysql_store_result(mysql));
      goto err;
    }
    mysql_free_result(mysql_store_result(mysql));
  }
 

1553
err:
1554 1555
  if (errmsg)
  {
1556 1557
    if (master_res)
      mysql_free_result(master_res);
1558
    DBUG_ASSERT(err_code != 0);
1559
    mi->report(ERROR_LEVEL, err_code, "%s", err_buff);
1560
    DBUG_RETURN(1);
1561
  }
1562

1563
  DBUG_RETURN(0);
1564 1565 1566 1567 1568

network_err:
  if (master_res)
    mysql_free_result(master_res);
  DBUG_RETURN(2);
unknown's avatar
unknown committed
1569

1570 1571 1572 1573
slave_killed_err:
  if (master_res)
    mysql_free_result(master_res);
  DBUG_RETURN(2);
unknown's avatar
unknown committed
1574
}
unknown's avatar
unknown committed
1575

1576
static bool wait_for_relay_log_space(Relay_log_info* rli)
unknown's avatar
unknown committed
1577
{
1578
  bool slave_killed=0;
1579
  Master_info* mi = rli->mi;
unknown's avatar
unknown committed
1580
  const char *save_proc_info;
unknown's avatar
unknown committed
1581 1582
  THD* thd = mi->io_thd;
  DBUG_ENTER("wait_for_relay_log_space");
1583

Marc Alff's avatar
Marc Alff committed
1584
  mysql_mutex_lock(&rli->log_space_lock);
unknown's avatar
unknown committed
1585
  save_proc_info= thd->enter_cond(&rli->log_space_cond,
unknown's avatar
unknown committed
1586 1587
                                  &rli->log_space_lock,
                                  "\
1588
Waiting for the slave SQL thread to free enough relay log space");
unknown's avatar
unknown committed
1589
  while (rli->log_space_limit < rli->log_space_total &&
unknown's avatar
unknown committed
1590
         !(slave_killed=io_slave_killed(thd,mi)) &&
1591
         !rli->ignore_log_space_limit)
Marc Alff's avatar
Marc Alff committed
1592
    mysql_cond_wait(&rli->log_space_cond, &rli->log_space_lock);
unknown's avatar
unknown committed
1593
  thd->exit_cond(save_proc_info);
unknown's avatar
unknown committed
1594 1595
  DBUG_RETURN(slave_killed);
}
1596

1597

1598 1599
/*
  Builds a Rotate from the ignored events' info and writes it to relay log.
1600

1601 1602 1603 1604 1605 1606 1607 1608 1609 1610
  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).
 */
1611
static void write_ignored_events_info_to_relay_log(THD *thd, Master_info *mi)
1612
{
1613
  Relay_log_info *rli= &mi->rli;
Marc Alff's avatar
Marc Alff committed
1614
  mysql_mutex_t *log_lock= rli->relay_log.get_log_lock();
1615 1616
  DBUG_ENTER("write_ignored_events_info_to_relay_log");

1617
  DBUG_ASSERT(thd == mi->io_thd);
Marc Alff's avatar
Marc Alff committed
1618
  mysql_mutex_lock(log_lock);
1619
  if (rli->ign_master_log_name_end[0])
unknown's avatar
unknown committed
1620
  {
1621
    DBUG_PRINT("info",("writing a Rotate event to track down ignored events"));
1622
    Rotate_log_event *ev= new Rotate_log_event(rli->ign_master_log_name_end,
1623 1624 1625 1626
                                               0, rli->ign_master_log_pos_end,
                                               Rotate_log_event::DUP_NAME);
    rli->ign_master_log_name_end[0]= 0;
    /* can unlock before writing as slave SQL thd will soon see our Rotate */
Marc Alff's avatar
Marc Alff committed
1627
    mysql_mutex_unlock(log_lock);
1628 1629 1630 1631
    if (likely((bool)ev))
    {
      ev->server_id= 0; // don't be ignored by slave SQL thread
      if (unlikely(rli->relay_log.append(ev)))
1632 1633 1634 1635 1636
        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");
1637
      rli->relay_log.harvest_bytes_written(&rli->log_space_total);
Luis Soares's avatar
Luis Soares committed
1638
      if (flush_master_info(mi, TRUE, TRUE))
1639
        sql_print_error("Failed to flush master info file");
1640 1641 1642
      delete ev;
    }
    else
1643 1644 1645 1646
      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
1647
  }
1648
  else
Marc Alff's avatar
Marc Alff committed
1649
    mysql_mutex_unlock(log_lock);
1650
  DBUG_VOID_RETURN;
unknown's avatar
unknown committed
1651 1652
}

1653

1654
int register_slave_on_master(MYSQL* mysql, Master_info *mi,
1655
                             bool *suppress_warnings)
1656
{
1657
  uchar buf[1024], *pos= buf;
1658
  uint report_host_len=0, report_user_len=0, report_password_len=0;
1659
  DBUG_ENTER("register_slave_on_master");
1660

1661
  *suppress_warnings= FALSE;
1662 1663 1664 1665 1666 1667 1668 1669
  if (report_host)
    report_host_len= strlen(report_host);
  if (report_host_len > HOSTNAME_LENGTH)
  {
    sql_print_warning("The length of report_host is %d. "
                      "It is larger than the max length(%d), so this "
                      "slave cannot be registered to the master.",
                      report_host_len, HOSTNAME_LENGTH);
1670
    DBUG_RETURN(0);
1671 1672
  }

1673
  if (report_user)
1674
    report_user_len= strlen(report_user);
1675 1676 1677 1678 1679 1680 1681 1682 1683
  if (report_user_len > USERNAME_LENGTH)
  {
    sql_print_warning("The length of report_user is %d. "
                      "It is larger than the max length(%d), so this "
                      "slave cannot be registered to the master.",
                      report_user_len, USERNAME_LENGTH);
    DBUG_RETURN(0);
  }

unknown's avatar
unknown committed
1684
  if (report_password)
1685
    report_password_len= strlen(report_password);
1686 1687 1688 1689 1690 1691 1692 1693
  if (report_password_len > MAX_PASSWORD_LENGTH)
  {
    sql_print_warning("The length of report_password is %d. "
                      "It is larger than the max length(%d), so this "
                      "slave cannot be registered to the master.",
                      report_password_len, MAX_PASSWORD_LENGTH);
    DBUG_RETURN(0);
  }
1694 1695

  int4store(pos, server_id); pos+= 4;
1696 1697 1698
  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);
1699
  int2store(pos, (uint16) report_port); pos+= 2;
1700 1701 1702 1703 1704 1705
  /* 
    Fake rpl_recovery_rank, which was removed in BUG#13963,
    so that this server can register itself on old servers,
    see BUG#49259.
   */
  int4store(pos, /* rpl_recovery_rank */ 0);    pos+= 4;
1706
  /* The master will fill in master_id */
unknown's avatar
unknown committed
1707
  int4store(pos, 0);                    pos+= 4;
1708

1709
  if (simple_command(mysql, COM_REGISTER_SLAVE, buf, (size_t) (pos- buf), 0))
1710
  {
1711 1712 1713 1714
    if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
    {
      *suppress_warnings= TRUE;                 // Suppress reconnect warning
    }
1715
    else if (!check_io_slave_killed(mi->io_thd, mi, NULL))
1716 1717 1718 1719 1720 1721 1722
    {
      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);
    }
1723
    DBUG_RETURN(1);
1724
  }
1725
  DBUG_RETURN(0);
1726 1727
}

1728

1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739
/**
  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
*/
1740
bool show_master_info(THD* thd, Master_info* mi)
unknown's avatar
unknown committed
1741
{
1742
  // TODO: fix this for multi-master
unknown's avatar
unknown committed
1743
  List<Item> field_list;
1744 1745 1746
  Protocol *protocol= thd->protocol;
  DBUG_ENTER("show_master_info");

unknown's avatar
unknown committed
1747
  field_list.push_back(new Item_empty_string("Slave_IO_State",
unknown's avatar
unknown committed
1748
                                                     14));
unknown's avatar
unknown committed
1749
  field_list.push_back(new Item_empty_string("Master_Host",
unknown's avatar
unknown committed
1750
                                                     sizeof(mi->host)));
unknown's avatar
unknown committed
1751
  field_list.push_back(new Item_empty_string("Master_User",
unknown's avatar
unknown committed
1752
                                                     sizeof(mi->user)));
1753
  field_list.push_back(new Item_return_int("Master_Port", 7,
unknown's avatar
unknown committed
1754
                                           MYSQL_TYPE_LONG));
1755
  field_list.push_back(new Item_return_int("Connect_Retry", 10,
unknown's avatar
unknown committed
1756
                                           MYSQL_TYPE_LONG));
1757
  field_list.push_back(new Item_empty_string("Master_Log_File",
unknown's avatar
unknown committed
1758
                                             FN_REFLEN));
1759
  field_list.push_back(new Item_return_int("Read_Master_Log_Pos", 10,
unknown's avatar
unknown committed
1760
                                           MYSQL_TYPE_LONGLONG));
1761
  field_list.push_back(new Item_empty_string("Relay_Log_File",
unknown's avatar
unknown committed
1762
                                             FN_REFLEN));
1763
  field_list.push_back(new Item_return_int("Relay_Log_Pos", 10,
unknown's avatar
unknown committed
1764
                                           MYSQL_TYPE_LONGLONG));
1765
  field_list.push_back(new Item_empty_string("Relay_Master_Log_File",
unknown's avatar
unknown committed
1766
                                             FN_REFLEN));
1767 1768
  field_list.push_back(new Item_empty_string("Slave_IO_Running", 3));
  field_list.push_back(new Item_empty_string("Slave_SQL_Running", 3));
1769 1770 1771 1772 1773 1774
  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
1775
                                             28));
1776 1777 1778
  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
1779
                                           MYSQL_TYPE_LONG));
1780
  field_list.push_back(new Item_return_int("Exec_Master_Log_Pos", 10,
unknown's avatar
unknown committed
1781
                                           MYSQL_TYPE_LONGLONG));
1782
  field_list.push_back(new Item_return_int("Relay_Log_Space", 10,
unknown's avatar
unknown committed
1783
                                           MYSQL_TYPE_LONGLONG));
1784
  field_list.push_back(new Item_empty_string("Until_Condition", 6));
1785
  field_list.push_back(new Item_empty_string("Until_Log_File", FN_REFLEN));
unknown's avatar
unknown committed
1786
  field_list.push_back(new Item_return_int("Until_Log_Pos", 10,
1787
                                           MYSQL_TYPE_LONGLONG));
unknown's avatar
unknown committed
1788 1789 1790
  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
1791
  field_list.push_back(new Item_empty_string("Master_SSL_CA_Path",
unknown's avatar
unknown committed
1792
                                             sizeof(mi->ssl_capath)));
unknown's avatar
unknown committed
1793
  field_list.push_back(new Item_empty_string("Master_SSL_Cert",
unknown's avatar
unknown committed
1794
                                             sizeof(mi->ssl_cert)));
unknown's avatar
unknown committed
1795
  field_list.push_back(new Item_empty_string("Master_SSL_Cipher",
unknown's avatar
unknown committed
1796
                                             sizeof(mi->ssl_cipher)));
unknown's avatar
unknown committed
1797
  field_list.push_back(new Item_empty_string("Master_SSL_Key",
unknown's avatar
unknown committed
1798
                                             sizeof(mi->ssl_key)));
1799
  field_list.push_back(new Item_return_int("Seconds_Behind_Master", 10,
unknown's avatar
unknown committed
1800
                                           MYSQL_TYPE_LONGLONG));
1801 1802
  field_list.push_back(new Item_empty_string("Master_SSL_Verify_Server_Cert",
                                             3));
1803 1804 1805 1806
  field_list.push_back(new Item_return_int("Last_IO_Errno", 4, MYSQL_TYPE_LONG));
  field_list.push_back(new Item_empty_string("Last_IO_Error", 20));
  field_list.push_back(new Item_return_int("Last_SQL_Errno", 4, MYSQL_TYPE_LONG));
  field_list.push_back(new Item_empty_string("Last_SQL_Error", 20));
Andrei Elkin's avatar
Andrei Elkin committed
1807 1808 1809 1810
  field_list.push_back(new Item_empty_string("Replicate_Ignore_Server_Ids",
                                             FN_REFLEN));
  field_list.push_back(new Item_return_int("Master_Server_Id", sizeof(ulong),
                                           MYSQL_TYPE_LONG));
unknown's avatar
unknown committed
1811

1812
  if (protocol->send_result_set_metadata(&field_list,
1813
                            Protocol::SEND_NUM_ROWS | Protocol::SEND_EOF))
unknown's avatar
unknown committed
1814
    DBUG_RETURN(TRUE);
unknown's avatar
unknown committed
1815

1816 1817
  if (mi->host[0])
  {
1818
    DBUG_PRINT("info",("host is set: '%s'", mi->host));
1819
    String *packet= &thd->packet;
1820
    protocol->prepare_for_resend();
unknown's avatar
unknown committed
1821

1822
    /*
1823 1824
      slave_running can be accessed without run_lock but not other
      non-volotile members like mi->io_thd, which is guarded by the mutex.
1825
    */
Marc Alff's avatar
Marc Alff committed
1826
    mysql_mutex_lock(&mi->run_lock);
1827
    protocol->store(mi->io_thd ? mi->io_thd->proc_info : "", &my_charset_bin);
Marc Alff's avatar
Marc Alff committed
1828
    mysql_mutex_unlock(&mi->run_lock);
1829

Marc Alff's avatar
Marc Alff committed
1830 1831 1832 1833
    mysql_mutex_lock(&mi->data_lock);
    mysql_mutex_lock(&mi->rli.data_lock);
    mysql_mutex_lock(&mi->err_lock);
    mysql_mutex_lock(&mi->rli.err_lock);
1834 1835
    protocol->store(mi->host, &my_charset_bin);
    protocol->store(mi->user, &my_charset_bin);
1836 1837
    protocol->store((uint32) mi->port);
    protocol->store((uint32) mi->connect_retry);
1838
    protocol->store(mi->master_log_name, &my_charset_bin);
1839
    protocol->store((ulonglong) mi->master_log_pos);
1840
    protocol->store(mi->rli.group_relay_log_name +
unknown's avatar
unknown committed
1841 1842
                    dirname_length(mi->rli.group_relay_log_name),
                    &my_charset_bin);
1843 1844
    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
1845
    protocol->store(mi->slave_running == MYSQL_SLAVE_RUN_CONNECT ?
1846 1847
                    "Yes" : (mi->slave_running == MYSQL_SLAVE_RUN_NOT_CONNECT ?
                             "Connecting" : "No"), &my_charset_bin);
1848
    protocol->store(mi->rli.slave_running ? "Yes":"No", &my_charset_bin);
1849 1850 1851
    protocol->store(rpl_filter->get_do_db());
    protocol->store(rpl_filter->get_ignore_db());

1852 1853
    char buf[256];
    String tmp(buf, sizeof(buf), &my_charset_bin);
1854
    rpl_filter->get_do_table(&tmp);
1855
    protocol->store(&tmp);
1856
    rpl_filter->get_ignore_table(&tmp);
1857
    protocol->store(&tmp);
1858
    rpl_filter->get_wild_do_table(&tmp);
1859
    protocol->store(&tmp);
1860
    rpl_filter->get_wild_ignore_table(&tmp);
1861 1862
    protocol->store(&tmp);

1863 1864
    protocol->store(mi->rli.last_error().number);
    protocol->store(mi->rli.last_error().message, &my_charset_bin);
1865
    protocol->store((uint32) mi->rli.slave_skip_counter);
1866
    protocol->store((ulonglong) mi->rli.group_master_log_pos);
1867
    protocol->store((ulonglong) mi->rli.log_space_total);
1868 1869

    protocol->store(
1870 1871
      mi->rli.until_condition==Relay_log_info::UNTIL_NONE ? "None":
        ( mi->rli.until_condition==Relay_log_info::UNTIL_MASTER_POS? "Master":
1872 1873 1874
          "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
1875 1876

#ifdef HAVE_OPENSSL
unknown's avatar
unknown committed
1877 1878 1879 1880 1881 1882 1883 1884 1885
    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
1886

1887 1888 1889 1890 1891 1892
    /*
      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)
1893
    {
1894
      long time_diff= ((long)(time(0) - mi->rli.last_master_timestamp)
unknown's avatar
unknown committed
1895
                       - mi->clock_diff_with_master);
1896
      /*
unknown's avatar
unknown committed
1897 1898
        Apparently on some systems time_diff can be <0. Here are possible
        reasons related to MySQL:
1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910
        - 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.
1911 1912 1913 1914
        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".
1915
      */
unknown's avatar
unknown committed
1916 1917
      protocol->store((longlong)(mi->rli.last_master_timestamp ?
                                 max(0, time_diff) : 0));
1918
    }
unknown's avatar
unknown committed
1919
    else
1920
    {
unknown's avatar
unknown committed
1921
      protocol->store_null();
1922 1923
    }
    protocol->store(mi->ssl_verify_server_cert? "Yes":"No", &my_charset_bin);
unknown's avatar
unknown committed
1924

1925
    // Last_IO_Errno
1926
    protocol->store(mi->last_error().number);
1927
    // Last_IO_Error
1928
    protocol->store(mi->last_error().message, &my_charset_bin);
1929
    // Last_SQL_Errno
1930
    protocol->store(mi->rli.last_error().number);
1931
    // Last_SQL_Error
1932
    protocol->store(mi->rli.last_error().message, &my_charset_bin);
Andrei Elkin's avatar
Andrei Elkin committed
1933 1934 1935 1936 1937 1938 1939 1940 1941 1942
    // Replicate_Ignore_Server_Ids
    {
      char buff[FN_REFLEN];
      ulong i, cur_len;
      for (i= 0, buff[0]= 0, cur_len= 0;
           i < mi->ignore_server_ids.elements; i++)
      {
        ulong s_id, slen;
        char sbuff[FN_REFLEN];
        get_dynamic(&mi->ignore_server_ids, (uchar*) &s_id, i);
1943
        slen= sprintf(sbuff, (i==0? "%lu" : ", %lu"), s_id);
Andrei Elkin's avatar
Andrei Elkin committed
1944 1945 1946 1947 1948 1949
        if (cur_len + slen + 4 > FN_REFLEN)
        {
          /*
            break the loop whenever remained space could not fit
            ellipses on the next cycle
          */
1950
          sprintf(buff + cur_len, "...");
Andrei Elkin's avatar
Andrei Elkin committed
1951 1952
          break;
        }
1953
        cur_len += sprintf(buff + cur_len, "%s", sbuff);
Andrei Elkin's avatar
Andrei Elkin committed
1954 1955 1956 1957 1958
      }
      protocol->store(buff, &my_charset_bin);
    }
    // Master_Server_id
    protocol->store((uint32) mi->master_id);
1959

Marc Alff's avatar
Marc Alff committed
1960 1961 1962 1963
    mysql_mutex_unlock(&mi->rli.err_lock);
    mysql_mutex_unlock(&mi->err_lock);
    mysql_mutex_unlock(&mi->rli.data_lock);
    mysql_mutex_unlock(&mi->data_lock);
1964

1965
    if (my_net_write(&thd->net, (uchar*) thd->packet.ptr(), packet->length()))
unknown's avatar
unknown committed
1966
      DBUG_RETURN(TRUE);
1967
  }
1968
  my_eof(thd);
unknown's avatar
unknown committed
1969
  DBUG_RETURN(FALSE);
unknown's avatar
unknown committed
1970 1971
}

unknown's avatar
unknown committed
1972

1973 1974
void set_slave_thread_options(THD* thd)
{
1975
  DBUG_ENTER("set_slave_thread_options");
1976 1977 1978 1979 1980 1981 1982 1983 1984
  /*
     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.
  */
1985
  ulonglong options= thd->variables.option_bits | OPTION_BIG_SELECTS;
1986 1987 1988 1989
  if (opt_log_slave_updates)
    options|= OPTION_BIN_LOG;
  else
    options&= ~OPTION_BIN_LOG;
1990
  thd->variables.option_bits= options;
1991
  thd->variables.completion_type= 0;
1992
  DBUG_VOID_RETURN;
1993
}
1994

1995
void set_slave_thread_default_charset(THD* thd, Relay_log_info const *rli)
1996
{
1997 1998
  DBUG_ENTER("set_slave_thread_default_charset");

1999 2000 2001 2002 2003 2004 2005
  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();
2006 2007 2008 2009 2010 2011 2012

  /*
    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.
   */
2013
  const_cast<Relay_log_info*>(rli)->cached_charset_invalidate();
2014
  DBUG_VOID_RETURN;
2015 2016
}

unknown's avatar
unknown committed
2017
/*
2018
  init_slave_thread()
unknown's avatar
unknown committed
2019
*/
2020

2021
static int init_slave_thread(THD* thd, SLAVE_THD_TYPE thd_type)
unknown's avatar
unknown committed
2022 2023
{
  DBUG_ENTER("init_slave_thread");
2024 2025 2026
#if !defined(DBUG_OFF)
  int simulate_error= 0;
#endif
2027
  thd->system_thread = (thd_type == SLAVE_THD_SQL) ?
unknown's avatar
unknown committed
2028
    SYSTEM_THREAD_SLAVE_SQL : SYSTEM_THREAD_SLAVE_IO;
2029
  thd->security_ctx->skip_grants();
unknown's avatar
unknown committed
2030
  my_net_init(&thd->net, 0);
2031 2032 2033 2034 2035
/*
  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.
*/
2036
  thd->variables.max_allowed_packet= global_system_variables.max_allowed_packet
2037
    + MAX_LOG_EVENT_HEADER;  /* note, incr over the global not session var */
2038
  thd->slave_thread = 1;
2039
  thd->enable_slow_log= opt_log_slow_slave_statements;
2040
  set_slave_thread_options(thd);
unknown's avatar
unknown committed
2041
  thd->client_capabilities = CLIENT_LOCAL_FILES;
Marc Alff's avatar
Marc Alff committed
2042
  mysql_mutex_lock(&LOCK_thread_count);
unknown's avatar
unknown committed
2043
  thd->thread_id= thd->variables.pseudo_thread_id= thread_id++;
Marc Alff's avatar
Marc Alff committed
2044
  mysql_mutex_unlock(&LOCK_thread_count);
unknown's avatar
unknown committed
2045

2046 2047 2048 2049 2050 2051 2052
  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
2053
  if (init_thr_lock() || thd->store_globals())
2054
#endif
unknown's avatar
unknown committed
2055
  {
2056
    thd->cleanup();
unknown's avatar
unknown committed
2057 2058 2059
    DBUG_RETURN(-1);
  }

2060
  if (thd_type == SLAVE_THD_SQL)
2061
    thd_proc_info(thd, "Waiting for the next event in relay log");
2062
  else
2063
    thd_proc_info(thd, "Waiting for master update");
unknown's avatar
unknown committed
2064
  thd->set_time();
2065 2066
  /* Do not use user-supplied timeout value for system threads. */
  thd->variables.lock_wait_timeout= LONG_TIMEOUT;
unknown's avatar
unknown committed
2067 2068 2069
  DBUG_RETURN(0);
}

2070

2071
static int safe_sleep(THD* thd, int sec, CHECK_KILLED_FUNC thread_killed,
unknown's avatar
unknown committed
2072
                      void* thread_killed_arg)
unknown's avatar
unknown committed
2073
{
2074
  int nap_time;
unknown's avatar
unknown committed
2075
  thr_alarm_t alarmed;
2076 2077
  DBUG_ENTER("safe_sleep");

unknown's avatar
unknown committed
2078
  thr_alarm_init(&alarmed);
2079
  time_t start_time= my_time(0);
unknown's avatar
unknown committed
2080 2081
  time_t end_time= start_time+sec;

2082
  while ((nap_time= (int) (end_time - start_time)) > 0)
unknown's avatar
unknown committed
2083
  {
2084
    ALARM alarm_buff;
unknown's avatar
unknown committed
2085
    /*
2086
      The only reason we are asking for alarm is so that
unknown's avatar
unknown committed
2087 2088 2089
      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
    */
2090
    thr_alarm(&alarmed, 2 * nap_time, &alarm_buff);
unknown's avatar
unknown committed
2091
    sleep(nap_time);
2092
    thr_end_alarm(&alarmed);
unknown's avatar
unknown committed
2093

2094
    if ((*thread_killed)(thd,thread_killed_arg))
2095
      DBUG_RETURN(1);
2096
    start_time= my_time(0);
unknown's avatar
unknown committed
2097
  }
2098
  DBUG_RETURN(0);
unknown's avatar
unknown committed
2099 2100
}

2101

He Zhenxing's avatar
He Zhenxing committed
2102 2103
static int request_dump(THD *thd, MYSQL* mysql, Master_info* mi,
			bool *suppress_warnings)
unknown's avatar
unknown committed
2104
{
2105
  uchar buf[FN_REFLEN + 10];
unknown's avatar
unknown committed
2106
  int len;
He Zhenxing's avatar
He Zhenxing committed
2107
  ushort binlog_flags = 0; // for now
2108
  char* logname = mi->master_log_name;
2109
  DBUG_ENTER("request_dump");
2110 2111
  
  *suppress_warnings= FALSE;
2112

He Zhenxing's avatar
He Zhenxing committed
2113 2114 2115 2116 2117
  if (RUN_HOOK(binlog_relay_io,
               before_request_transmit,
               (thd, mi, binlog_flags)))
    DBUG_RETURN(1);
  
unknown's avatar
unknown committed
2118
  // TODO if big log files: Change next to int8store()
unknown's avatar
unknown committed
2119
  int4store(buf, (ulong) mi->master_log_pos);
unknown's avatar
unknown committed
2120
  int2store(buf + 4, binlog_flags);
2121
  int4store(buf + 6, server_id);
unknown's avatar
unknown committed
2122
  len = (uint) strlen(logname);
2123
  memcpy(buf + 10, logname,len);
unknown's avatar
SCRUM  
unknown committed
2124
  if (simple_command(mysql, COM_BINLOG_DUMP, buf, len + 10, 1))
unknown's avatar
unknown committed
2125
  {
unknown's avatar
unknown committed
2126 2127 2128 2129 2130
    /*
      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
2131
    if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
2132
      *suppress_warnings= TRUE;                 // Suppress reconnect warning
unknown's avatar
unknown committed
2133
    else
2134
      sql_print_error("Error on COM_BINLOG_DUMP: %d  %s, will retry in %d secs",
unknown's avatar
unknown committed
2135
                      mysql_errno(mysql), mysql_error(mysql),
2136
                      mi->connect_retry);
2137
    DBUG_RETURN(1);
unknown's avatar
unknown committed
2138
  }
unknown's avatar
unknown committed
2139

2140
  DBUG_RETURN(0);
unknown's avatar
unknown committed
2141 2142
}

2143

unknown's avatar
unknown committed
2144
/*
2145
  Read one event from the master
unknown's avatar
unknown committed
2146

unknown's avatar
unknown committed
2147 2148
  SYNOPSIS
    read_event()
unknown's avatar
unknown committed
2149 2150 2151 2152 2153 2154
    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.
2155

unknown's avatar
unknown committed
2156
    RETURN VALUES
unknown's avatar
unknown committed
2157 2158
    'packet_error'      Error
    number              Length of packet
unknown's avatar
unknown committed
2159 2160
*/

2161
static ulong read_event(MYSQL* mysql, Master_info *mi, bool* suppress_warnings)
unknown's avatar
unknown committed
2162
{
2163
  ulong len;
2164
  DBUG_ENTER("read_event");
unknown's avatar
unknown committed
2165

2166
  *suppress_warnings= FALSE;
unknown's avatar
unknown committed
2167 2168 2169 2170
  /*
    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
2171
#ifndef DBUG_OFF
2172
  if (disconnect_slave_event_count && !(mi->events_till_disconnect--))
unknown's avatar
unknown committed
2173
    DBUG_RETURN(packet_error);
unknown's avatar
unknown committed
2174
#endif
unknown's avatar
unknown committed
2175

2176
  len = cli_safe_read(mysql);
2177
  if (len == packet_error || (long) len < 1)
unknown's avatar
unknown committed
2178
  {
unknown's avatar
SCRUM  
unknown committed
2179
    if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED)
unknown's avatar
unknown committed
2180 2181
    {
      /*
unknown's avatar
unknown committed
2182 2183 2184
        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
2185 2186 2187 2188
      */
      *suppress_warnings= TRUE;
    }
    else
2189
      sql_print_error("Error reading packet from server: %s ( server_errno=%d)",
unknown's avatar
unknown committed
2190
                      mysql_error(mysql), mysql_errno(mysql));
2191
    DBUG_RETURN(packet_error);
unknown's avatar
unknown committed
2192 2193
  }

2194 2195
  /* Check if eof packet */
  if (len < 8 && mysql->net.read_pos[0] == 254)
unknown's avatar
unknown committed
2196
  {
2197 2198
    sql_print_information("Slave: received end packet from server, apparent "
                          "master shutdown: %s",
unknown's avatar
unknown committed
2199
                     mysql_error(mysql));
2200
     DBUG_RETURN(packet_error);
unknown's avatar
unknown committed
2201
  }
unknown's avatar
unknown committed
2202

unknown's avatar
unknown committed
2203
  DBUG_PRINT("exit", ("len: %lu  net->read_pos[4]: %d",
unknown's avatar
unknown committed
2204 2205
                      len, mysql->net.read_pos[4]));
  DBUG_RETURN(len - 1);
unknown's avatar
unknown committed
2206 2207
}

unknown's avatar
unknown committed
2208 2209 2210 2211 2212 2213 2214 2215 2216
/*
  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)
{
2217 2218
  DBUG_ENTER("has_temporary_error");

2219
  DBUG_EXECUTE_IF("all_errors_are_temporary_errors",
Marc Alff's avatar
Marc Alff committed
2220
                  if (thd->stmt_da->is_error())
2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232
                  {
                    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
2233 2234 2235 2236 2237 2238

  /*
    Temporary error codes:
    currently, InnoDB deadlock detected by InnoDB or lock
    wait timeout (innodb_lock_wait_timeout exceeded
  */
Marc Alff's avatar
Marc Alff committed
2239 2240
  if (thd->stmt_da->sql_errno() == ER_LOCK_DEADLOCK ||
      thd->stmt_da->sql_errno() == ER_LOCK_WAIT_TIMEOUT)
2241
    DBUG_RETURN(1);
unknown's avatar
unknown committed
2242 2243 2244 2245 2246

#ifdef HAVE_NDB_BINLOG
  /*
    currently temporary error set in ndbcluster
  */
Marc Alff's avatar
Marc Alff committed
2247
  List_iterator_fast<MYSQL_ERROR> it(thd->warning_info->warn_list());
unknown's avatar
unknown committed
2248 2249 2250
  MYSQL_ERROR *err;
  while ((err= it++))
  {
Marc Alff's avatar
Marc Alff committed
2251 2252 2253
    DBUG_PRINT("info", ("has condition %d %s", err->get_sql_errno(),
                        err->get_message_text()));
    switch (err->get_sql_errno())
unknown's avatar
unknown committed
2254 2255
    {
    case ER_GET_TEMPORARY_ERRMSG:
2256
      DBUG_RETURN(1);
unknown's avatar
unknown committed
2257 2258 2259 2260 2261
    default:
      break;
    }
  }
#endif
2262
  DBUG_RETURN(0);
unknown's avatar
unknown committed
2263
}
2264

2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291

/**
  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().
*/
2292
int apply_event_and_update_pos(Log_event* ev, THD* thd, Relay_log_info* rli)
2293 2294 2295 2296 2297 2298 2299 2300 2301
{
  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",
2302 2303
                      FLAGSTR(thd->variables.option_bits, OPTION_NOT_AUTOCOMMIT),
                      FLAGSTR(thd->variables.option_bits, OPTION_BEGIN),
2304
                      (ulong) rli->last_event_start_time));
2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336

  /*
    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

2337 2338
  int reason= ev->shall_skip(rli);
  if (reason == Log_event::EVENT_SKIP_COUNT)
2339
    sql_slave_skip_counter= --rli->slave_skip_counter;
Marc Alff's avatar
Marc Alff committed
2340
  mysql_mutex_unlock(&rli->data_lock);
2341 2342 2343
  if (reason == Log_event::EVENT_SKIP_NOT)
    exec_res= ev->apply_event(rli);

2344
#ifndef DBUG_OFF
2345 2346
  /*
    This only prints information to the debug trace.
2347

2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358
    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",
2359
                      test(thd->variables.option_bits & OPTION_BEGIN),
2360 2361 2362
                      rli->get_flag(Relay_log_info::IN_STMT)));
  DBUG_PRINT("skip_event", ("%s event was %s",
                            ev->get_type_str(), explain[reason]));
2363 2364 2365 2366 2367 2368
#endif

  DBUG_PRINT("info", ("apply_event error = %d", exec_res));
  if (exec_res == 0)
  {
    int error= ev->update_pos(rli);
2369 2370 2371 2372
#ifdef HAVE_purify
    if (!rli->is_fake)
#endif
    {
2373
#ifndef DBUG_OFF
2374
      char buf[22];
2375
#endif
2376 2377 2378 2379 2380 2381 2382 2383
      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));
    }
2384 2385 2386 2387 2388 2389 2390 2391 2392
    /*
      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)
    {
2393
      char buf[22];
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 2427 2428
      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
2429
     if a failure occurs when applying the event.)
2430 2431 2432 2433 2434 2435 2436

   - An error occurred when updating the binlog position.

  @retval 0 The event was applied.

  @retval 1 The event was not applied.
*/
2437
static int exec_relay_log_event(THD* thd, Relay_log_info* rli)
unknown's avatar
unknown committed
2438
{
2439 2440
  DBUG_ENTER("exec_relay_log_event");

2441 2442
  /*
     We acquire this mutex since we need it for all operations except
2443
     event execution. But we will release it in places where we will
2444 2445
     wait for something for example inside of next_event().
   */
Marc Alff's avatar
Marc Alff committed
2446
  mysql_mutex_lock(&rli->data_lock);
2447

2448
  Log_event * ev = next_event(rli);
2449

2450
  DBUG_ASSERT(rli->sql_thd==thd);
2451

2452
  if (sql_slave_killed(thd,rli))
2453
  {
Marc Alff's avatar
Marc Alff committed
2454
    mysql_mutex_unlock(&rli->data_lock);
unknown's avatar
unknown committed
2455
    delete ev;
2456
    DBUG_RETURN(1);
2457
  }
2458 2459
  if (ev)
  {
2460
    int exec_res;
2461

2462 2463 2464 2465
    /*
      This tests if the position of the beginning of the current event
      hits the UNTIL barrier.
    */
2466
    if (rli->until_condition != Relay_log_info::UNTIL_NONE &&
2467
        rli->is_until_satisfied(thd, ev))
2468 2469 2470 2471 2472 2473 2474 2475 2476
    {
      char buf[22];
      sql_print_information("Slave SQL thread stopped because it reached its"
                            " UNTIL position %s", llstr(rli->until_pos(), buf));
      /*
        Setting abort_slave flag because we do not want additional message about
        error in query execution to be printed.
      */
      rli->abort_slave= 1;
Marc Alff's avatar
Marc Alff committed
2477
      mysql_mutex_unlock(&rli->data_lock);
2478
      delete ev;
2479
      DBUG_RETURN(1);
2480
    }
2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494

    { /**
         The following failure injecion works in cooperation with tests 
         setting @@global.debug= 'd,incomplete_group_in_relay_log'.
         Xid or Commit events are not executed to force the slave sql
         read hanging if the realy log does not have any more events.
      */
      DBUG_EXECUTE_IF("incomplete_group_in_relay_log",
                      if ((ev->get_type_code() == XID_EVENT) ||
                          ((ev->get_type_code() == QUERY_EVENT) &&
                           strcmp("COMMIT", ((Query_log_event *) ev)->query) == 0))
                      {
                        DBUG_ASSERT(thd->transaction.all.modified_non_trans_table);
                        rli->abort_slave= 1;
Marc Alff's avatar
Marc Alff committed
2495
                        mysql_mutex_unlock(&rli->data_lock);
2496 2497 2498 2499 2500 2501
                        delete ev;
                        rli->inc_event_relay_log_pos();
                        DBUG_RETURN(0);
                      };);
    }

2502
    exec_res= apply_event_and_update_pos(ev, thd, rli);
2503

2504
    /*
2505 2506 2507
      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.
2508
    */
2509
    if (ev->get_type_code() != FORMAT_DESCRIPTION_EVENT)
2510
    {
2511 2512
      DBUG_PRINT("info", ("Deleting the event after it has been executed"));
      delete ev;
2513 2514
    }

2515
    /*
2516 2517
      update_log_pos failed: this should not happen, so we don't
      retry.
2518
    */
2519 2520 2521
    if (exec_res == 2)
      DBUG_RETURN(1);

2522 2523
    if (slave_trans_retries)
    {
2524
      int UNINIT_VAR(temp_err);
2525
      if (exec_res && (temp_err= has_temporary_error(thd)))
2526 2527 2528 2529
      {
        const char *errmsg;
        /*
          We were in a transaction which has been rolled back because of a
unknown's avatar
unknown committed
2530 2531
          temporary error;
          let's seek back to BEGIN log event and retry it all again.
2532
	  Note, if lock wait timeout (innodb_lock_wait_timeout exceeded)
2533
	  there is no rollback since 5.0.13 (ref: manual).
2534 2535 2536 2537 2538 2539 2540 2541
          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.
        */
2542
        if (rli->trans_retries < slave_trans_retries)
2543 2544 2545 2546 2547 2548
        {
          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,
2549
                                      1, &errmsg, 1))
2550 2551 2552 2553 2554
            sql_print_error("Error initializing relay log position: %s",
                            errmsg);
          else
          {
            exec_res= 0;
2555
            rli->cleanup_context(thd, 1);
unknown's avatar
unknown committed
2556
            /* chance for concurrent connection to get more locks */
2557
            safe_sleep(thd, min(rli->trans_retries, MAX_SLAVE_RETRY_PAUSE),
unknown's avatar
unknown committed
2558
                       (CHECK_KILLED_FUNC)sql_slave_killed, (void*)rli);
Marc Alff's avatar
Marc Alff committed
2559
            mysql_mutex_lock(&rli->data_lock); // because of SHOW STATUS
unknown's avatar
unknown committed
2560
            rli->trans_retries++;
2561
            rli->retried_trans++;
Marc Alff's avatar
Marc Alff committed
2562
            mysql_mutex_unlock(&rli->data_lock);
2563 2564
            DBUG_PRINT("info", ("Slave retries transaction "
                                "rli->trans_retries: %lu", rli->trans_retries));
unknown's avatar
unknown committed
2565
          }
2566 2567 2568 2569 2570 2571 2572
        }
        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);
      }
2573
      else if ((exec_res && !temp_err) ||
2574 2575
               (opt_using_transactions &&
                rli->group_relay_log_pos == rli->event_relay_log_pos))
2576 2577
      {
        /*
2578 2579 2580
          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
2581 2582 2583
          non-transient error, the slave will stop with an error.
         */
        rli->trans_retries= 0; // restart from fresh
unknown's avatar
unknown committed
2584
        DBUG_PRINT("info", ("Resetting retry counter, rli->trans_retries: %lu",
2585
                            rli->trans_retries));
2586
      }
2587 2588
    }
    DBUG_RETURN(exec_res);
2589
  }
Marc Alff's avatar
Marc Alff committed
2590
  mysql_mutex_unlock(&rli->data_lock);
2591 2592
  rli->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_READ_FAILURE,
              ER(ER_SLAVE_RELAY_LOG_READ_FAILURE), "\
2593 2594 2595 2596 2597 2598 2599
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
2600
");
2601
  DBUG_RETURN(1);
unknown's avatar
unknown committed
2602 2603
}

2604

2605
static bool check_io_slave_killed(THD *thd, Master_info *mi, const char *info)
2606 2607 2608
{
  if (io_slave_killed(thd, mi))
  {
2609
    if (info && global_system_variables.log_warnings)
2610
      sql_print_information("%s", info);
2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641
    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.
*/

2642
static int try_to_reconnect(THD *thd, MYSQL *mysql, Master_info *mi,
2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678
                            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
    {
2679
      sql_print_information("%s", buf);
2680 2681 2682 2683 2684
    }
  }
  if (safe_reconnect(thd, mysql, mi, 1) || io_slave_killed(thd, mi))
  {
    if (global_system_variables.log_warnings)
2685
      sql_print_information("%s", messages[SLAVE_RECON_MSG_KILLED_AFTER]);
2686 2687 2688 2689 2690 2691
    return 1;
  }
  return 0;
}


2692 2693 2694 2695 2696
/**
  Slave IO thread entry point.

  @param arg Pointer to Master_info struct that holds information for
  the IO thread.
2697

2698 2699
  @return Always 0.
*/
2700
pthread_handler_t handle_slave_io(void *arg)
unknown's avatar
unknown committed
2701
{
unknown's avatar
unknown committed
2702 2703
  THD *thd; // needs to be first for thread_stack
  MYSQL *mysql;
2704
  Master_info *mi = (Master_info*)arg;
2705
  Relay_log_info *rli= &mi->rli;
unknown's avatar
unknown committed
2706 2707
  char llbuff[22];
  uint retry_count;
2708
  bool suppress_warnings;
2709
  int ret;
2710 2711 2712
#ifndef DBUG_OFF
  uint retry_count_reg= 0, retry_count_dump= 0, retry_count_event= 0;
#endif
unknown's avatar
unknown committed
2713 2714
  // needs to call my_thread_init(), otherwise we get a coredump in DBUG_ stuff
  my_thread_init();
2715
  DBUG_ENTER("handle_slave_io");
unknown's avatar
unknown committed
2716

2717
  DBUG_ASSERT(mi->inited);
unknown's avatar
unknown committed
2718 2719 2720
  mysql= NULL ;
  retry_count= 0;

Marc Alff's avatar
Marc Alff committed
2721
  mysql_mutex_lock(&mi->run_lock);
unknown's avatar
unknown committed
2722 2723 2724
  /* Inform waiting threads that slave has started */
  mi->slave_run_id++;

2725
#ifndef DBUG_OFF
2726
  mi->events_till_disconnect = disconnect_slave_event_count;
2727 2728
#endif

2729
  thd= new THD; // note that contructor of THD uses DBUG_ !
2730
  THD_CHECK_SENTRY(thd);
2731
  mi->io_thd = thd;
unknown's avatar
unknown committed
2732 2733

  pthread_detach_this_thread();
2734
  thd->thread_stack= (char*) &thd; // remember where our stack is
2735
  mi->clear_error();
2736
  if (init_slave_thread(thd, SLAVE_THD_IO))
unknown's avatar
unknown committed
2737
  {
Marc Alff's avatar
Marc Alff committed
2738 2739
    mysql_cond_broadcast(&mi->start_cond);
    mysql_mutex_unlock(&mi->run_lock);
unknown's avatar
unknown committed
2740 2741 2742
    sql_print_error("Failed during slave I/O thread initialization");
    goto err;
  }
Marc Alff's avatar
Marc Alff committed
2743
  mysql_mutex_lock(&LOCK_thread_count);
unknown's avatar
unknown committed
2744
  threads.append(thd);
Marc Alff's avatar
Marc Alff committed
2745
  mysql_mutex_unlock(&LOCK_thread_count);
2746 2747
  mi->slave_running = 1;
  mi->abort_slave = 0;
Marc Alff's avatar
Marc Alff committed
2748 2749
  mysql_mutex_unlock(&mi->run_lock);
  mysql_cond_broadcast(&mi->start_cond);
2750

2751
  DBUG_PRINT("master_info",("log_file_name: '%s'  position: %s",
unknown's avatar
unknown committed
2752 2753
                            mi->master_log_name,
                            llstr(mi->master_log_pos,llbuff)));
2754

He Zhenxing's avatar
He Zhenxing committed
2755 2756 2757 2758 2759 2760 2761 2762 2763 2764
  /* This must be called before run any binlog_relay_io hooks */
  my_pthread_setspecific_ptr(RPL_MASTER_INFO, mi);

  if (RUN_HOOK(binlog_relay_io, thread_start, (thd, mi)))
  {
    mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
               ER(ER_SLAVE_FATAL_ERROR), "Failed to run 'thread_start' hook");
    goto err;
  }

unknown's avatar
SCRUM  
unknown committed
2765
  if (!(mi->mysql = mysql = mysql_init(NULL)))
unknown's avatar
unknown committed
2766
  {
2767 2768
    mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
               ER(ER_SLAVE_FATAL_ERROR), "error in mysql_init()");
unknown's avatar
unknown committed
2769 2770
    goto err;
  }
2771

2772
  thd_proc_info(thd, "Connecting to master");
2773
  // we can get killed during safe_connect
2774
  if (!safe_connect(thd, mysql, mi))
2775
  {
unknown's avatar
unknown committed
2776 2777 2778
    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,
2779 2780
			  IO_RPL_LOG_NAME,
			  llstr(mi->master_log_pos,llbuff));
2781 2782 2783 2784 2785
  /*
    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.
  */
2786 2787
    mysql->net.max_packet_size= thd->net.max_packet_size+= MAX_LOG_EVENT_HEADER;
  }
2788
  else
unknown's avatar
unknown committed
2789
  {
2790
    sql_print_information("Slave I/O thread killed while connecting to master");
unknown's avatar
unknown committed
2791 2792
    goto err;
  }
2793

2794
connected:
2795

2796 2797 2798 2799 2800 2801 2802 2803 2804 2805
    DBUG_EXECUTE_IF("dbug.before_get_running_status_yes",
                    {
                      const char act[]=
                        "now "
                        "wait_for signal.io_thread_let_running";
                      DBUG_ASSERT(opt_debug_sync_timeout > 0);
                      DBUG_ASSERT(!debug_sync_set_action(thd, 
                                                         STRING_WITH_LEN(act)));
                    };);

2806 2807
  // TODO: the assignment below should be under mutex (5.0)
  mi->slave_running= MYSQL_SLAVE_RUN_CONNECT;
2808
  thd->slave_net = &mysql->net;
2809
  thd_proc_info(thd, "Checking master version");
2810 2811 2812
  ret= get_master_version_and_clock(mysql, mi);
  if (ret == 1)
    /* Fatal error */
2813
    goto err;
2814

2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826
  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;
  } 
2827 2828

  if (mi->rli.relay_log.description_event_for_queue->binlog_version > 1)
2829
  {
unknown's avatar
unknown committed
2830 2831 2832
    /*
      Register ourselves with the master.
    */
2833
    thd_proc_info(thd, "Registering slave on master");
2834
    if (register_slave_on_master(mysql, mi, &suppress_warnings))
2835
    {
2836 2837 2838 2839 2840 2841 2842 2843 2844
      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
2845 2846
        goto err;
      goto connected;
2847
    }
2848 2849 2850 2851 2852 2853 2854 2855 2856 2857
    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;
      });
2858
  }
2859

2860
  DBUG_PRINT("info",("Starting reading binary log from master"));
2861
  while (!io_slave_killed(thd,mi))
unknown's avatar
unknown committed
2862
  {
2863
    thd_proc_info(thd, "Requesting binlog dump");
He Zhenxing's avatar
He Zhenxing committed
2864
    if (request_dump(thd, mysql, mi, &suppress_warnings))
unknown's avatar
unknown committed
2865 2866
    {
      sql_print_error("Failed on request_dump()");
2867 2868 2869 2870
      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
2871
        goto err;
unknown's avatar
unknown committed
2872 2873
      goto connected;
    }
2874 2875 2876 2877 2878 2879 2880 2881 2882 2883
    DBUG_EXECUTE_IF("FORCE_SLAVE_TO_RECONNECT_DUMP", 
      if (!retry_count_dump)
      {
        retry_count_dump++;
        sql_print_information("Forcing to reconnect slave I/O thread");
        if (try_to_reconnect(thd, mysql, mi, &retry_count, suppress_warnings,
                             reconnect_messages[SLAVE_RECON_ACT_DUMP]))
          goto err;
        goto connected;
      });
He Zhenxing's avatar
He Zhenxing committed
2884
    const char *event_buf;
unknown's avatar
unknown committed
2885

2886
    DBUG_ASSERT(mi->last_error().number == 0);
2887
    while (!io_slave_killed(thd,mi))
unknown's avatar
unknown committed
2888
    {
2889
      ulong event_len;
2890
      /*
2891
         We say "waiting" because read_event() will wait if there's nothing to
2892 2893 2894
         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.
2895
      */
2896
      thd_proc_info(thd, "Waiting for master to send event");
2897
      event_len= read_event(mysql, mi, &suppress_warnings);
2898 2899
      if (check_io_slave_killed(thd, mi, "Slave I/O thread killed while \
reading event"))
unknown's avatar
unknown committed
2900
        goto err;
2901 2902 2903 2904 2905 2906 2907 2908 2909 2910
      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;
        });
2911

unknown's avatar
unknown committed
2912 2913
      if (event_len == packet_error)
      {
unknown's avatar
unknown committed
2914
        uint mysql_error_number= mysql_errno(mysql);
2915 2916
        switch (mysql_error_number) {
        case CR_NET_PACKET_TOO_LARGE:
unknown's avatar
unknown committed
2917
          sql_print_error("\
2918 2919 2920
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
2921
                          thd->variables.max_allowed_packet);
2922
          mi->report(ERROR_LEVEL, ER_NET_PACKET_TOO_LARGE,
2923
                     "%s", ER(ER_NET_PACKET_TOO_LARGE));
unknown's avatar
unknown committed
2924
          goto err;
2925
        case ER_MASTER_FATAL_ERROR_READING_BINLOG:
2926 2927 2928
          mi->report(ERROR_LEVEL, ER_MASTER_FATAL_ERROR_READING_BINLOG,
                     ER(ER_MASTER_FATAL_ERROR_READING_BINLOG),
                     mysql_error_number, mysql_error(mysql));
unknown's avatar
unknown committed
2929
          goto err;
2930
        case ER_OUT_OF_RESOURCES:
2931 2932
          sql_print_error("\
Stopping slave I/O thread due to out-of-memory error from master");
2933
          mi->report(ERROR_LEVEL, ER_OUT_OF_RESOURCES,
2934
                     "%s", ER(ER_OUT_OF_RESOURCES));
unknown's avatar
unknown committed
2935 2936
          goto err;
        }
2937 2938
        if (try_to_reconnect(thd, mysql, mi, &retry_count, suppress_warnings,
                             reconnect_messages[SLAVE_RECON_ACT_EVENT]))
unknown's avatar
unknown committed
2939 2940
          goto err;
        goto connected;
unknown's avatar
unknown committed
2941
      } // if (event_len == packet_error)
2942

unknown's avatar
unknown committed
2943
      retry_count=0;                    // ok event, reset retry counter
2944
      thd_proc_info(thd, "Queueing master event to the relay log");
He Zhenxing's avatar
He Zhenxing committed
2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959
      event_buf= (const char*)mysql->net.read_pos + 1;
      if (RUN_HOOK(binlog_relay_io, after_read_event,
                   (thd, mi,(const char*)mysql->net.read_pos + 1,
                    event_len, &event_buf, &event_len)))
      {
        mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
                   ER(ER_SLAVE_FATAL_ERROR),
                   "Failed to run 'after_read_event' hook");
        goto err;
      }

      /* XXX: 'synced' should be updated by queue_event to indicate
         whether event has been synced to disk */
      bool synced= 0;
      if (queue_event(mi, event_buf, event_len))
unknown's avatar
unknown committed
2960
      {
2961 2962 2963
        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
2964
        goto err;
unknown's avatar
unknown committed
2965
      }
He Zhenxing's avatar
He Zhenxing committed
2966 2967 2968 2969 2970 2971 2972 2973 2974 2975

      if (RUN_HOOK(binlog_relay_io, after_queue_event,
                   (thd, mi, event_buf, event_len, synced)))
      {
        mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
                   ER(ER_SLAVE_FATAL_ERROR),
                   "Failed to run 'after_queue_event' hook");
        goto err;
      }

2976
      if (flush_master_info(mi, TRUE, TRUE))
2977 2978 2979 2980
      {
        sql_print_error("Failed to flush master info file");
        goto err;
      }
2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992
      /*
        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.
      */
2993 2994 2995 2996 2997
#ifndef DBUG_OFF
      {
        char llbuf1[22], llbuf2[22];
        DBUG_PRINT("info", ("log_space_limit=%s log_space_total=%s \
ignore_log_space_limit=%d",
2998 2999
                            llstr(rli->log_space_limit,llbuf1),
                            llstr(rli->log_space_total,llbuf2),
unknown's avatar
unknown committed
3000
                            (int) rli->ignore_log_space_limit));
3001 3002 3003
      }
#endif

3004
      if (rli->log_space_limit && rli->log_space_limit <
unknown's avatar
unknown committed
3005
          rli->log_space_total &&
3006
          !rli->ignore_log_space_limit)
unknown's avatar
unknown committed
3007 3008 3009
        if (wait_for_relay_log_space(rli))
        {
          sql_print_error("Slave I/O thread aborted while waiting for relay \
unknown's avatar
unknown committed
3010
log space");
unknown's avatar
unknown committed
3011 3012 3013
          goto err;
        }
    }
3014
  }
unknown's avatar
unknown committed
3015

unknown's avatar
unknown committed
3016
  // error = 0;
unknown's avatar
unknown committed
3017
err:
3018
  // print the current replication position
3019
  sql_print_information("Slave I/O thread exiting, read up to log '%s', position %s",
unknown's avatar
unknown committed
3020
                  IO_RPL_LOG_NAME, llstr(mi->master_log_pos,llbuff));
He Zhenxing's avatar
He Zhenxing committed
3021
  RUN_HOOK(binlog_relay_io, thread_stop, (thd, mi));
3022
  thd->reset_query();
unknown's avatar
unknown committed
3023
  thd->reset_db(NULL, 0);
unknown's avatar
unknown committed
3024 3025
  if (mysql)
  {
3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036
    /*
      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
3037
    mysql_close(mysql);
unknown's avatar
unknown committed
3038 3039
    mi->mysql=0;
  }
3040
  write_ignored_events_info_to_relay_log(thd, mi);
3041
  thd_proc_info(thd, "Waiting for slave mutex on exit");
Marc Alff's avatar
Marc Alff committed
3042
  mysql_mutex_lock(&mi->run_lock);
3043

3044 3045 3046
  /* Forget the relay log's format */
  delete mi->rli.relay_log.description_event_for_queue;
  mi->rli.relay_log.description_event_for_queue= 0;
3047
  // TODO: make rpl_status part of Master_info
3048
  change_rpl_status(RPL_ACTIVE_SLAVE,RPL_IDLE_SLAVE);
3049
  DBUG_ASSERT(thd->net.buff != 0);
unknown's avatar
unknown committed
3050
  net_end(&thd->net); // destructor will not free it, because net.vio is 0
Marc Alff's avatar
Marc Alff committed
3051
  mysql_mutex_lock(&LOCK_thread_count);
3052
  THD_CHECK_SENTRY(thd);
unknown's avatar
unknown committed
3053
  delete thd;
Marc Alff's avatar
Marc Alff committed
3054
  mysql_mutex_unlock(&LOCK_thread_count);
unknown's avatar
unknown committed
3055 3056 3057
  mi->abort_slave= 0;
  mi->slave_running= 0;
  mi->io_thd= 0;
3058 3059 3060 3061 3062
  /*
    Note: the order of the two following calls (first broadcast, then unlock)
    is important. Otherwise a killer_thread can execute between the calls and
    delete the mi structure leading to a crash! (see BUG#25306 for details)
   */ 
Marc Alff's avatar
Marc Alff committed
3063
  mysql_cond_broadcast(&mi->stop_cond);       // tell the world we are done
3064
  DBUG_EXECUTE_IF("simulate_slave_delay_at_terminate_bug38694", sleep(5););
Marc Alff's avatar
Marc Alff committed
3065
  mysql_mutex_unlock(&mi->run_lock);
3066 3067

  DBUG_LEAVE;                                   // Must match DBUG_ENTER()
3068
  my_thread_end();
unknown's avatar
unknown committed
3069
  pthread_exit(0);
3070
  return 0;                                     // Avoid compiler warnings
unknown's avatar
unknown committed
3071 3072
}

3073 3074 3075 3076 3077
/*
  Check the temporary directory used by commands like
  LOAD DATA INFILE.
 */
static 
3078
int check_temp_dir(char* tmp_file)
3079 3080 3081
{
  int fd;
  MY_DIR *dirp;
3082 3083
  char tmp_dir[FN_REFLEN];
  size_t tmp_dir_size;
3084 3085 3086

  DBUG_ENTER("check_temp_dir");

3087 3088 3089 3090 3091
  /*
    Get the directory from the temporary file.
  */
  dirname_part(tmp_dir, tmp_file, &tmp_dir_size);

3092 3093 3094 3095 3096 3097 3098 3099 3100 3101
  /*
    Check if the directory exists.
   */
  if (!(dirp=my_dir(tmp_dir,MYF(MY_WME))))
    DBUG_RETURN(1);
  my_dirend(dirp);

  /*
    Check permissions to create a file.
   */
Marc Alff's avatar
Marc Alff committed
3102 3103 3104 3105
  if ((fd= mysql_file_create(key_file_misc,
                             tmp_file, CREATE_MODE,
                             O_WRONLY | O_BINARY | O_EXCL | O_NOFOLLOW,
                             MYF(MY_WME))) < 0)
3106 3107 3108 3109 3110
  DBUG_RETURN(1);

  /*
    Clean up.
   */
Marc Alff's avatar
Marc Alff committed
3111 3112
  mysql_file_close(fd, MYF(0));
  mysql_file_delete(key_file_misc, tmp_file, MYF(0));
3113 3114 3115

  DBUG_RETURN(0);
}
unknown's avatar
unknown committed
3116

3117 3118 3119 3120 3121
/**
  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
3122

3123 3124
  @return Always 0.
*/
3125
pthread_handler_t handle_slave_sql(void *arg)
3126
{
unknown's avatar
unknown committed
3127
  THD *thd;                     /* needs to be first for thread_stack */
3128
  char llbuff[22],llbuff1[22];
3129 3130
  char saved_log_name[FN_REFLEN];
  char saved_master_log_name[FN_REFLEN];
3131 3132
  my_off_t UNINIT_VAR(saved_log_pos);
  my_off_t UNINIT_VAR(saved_master_log_pos);
3133
  my_off_t saved_skip= 0;
3134

3135
  Relay_log_info* rli = &((Master_info*)arg)->rli;
unknown's avatar
unknown committed
3136 3137 3138 3139
  const char *errmsg;

  // needs to call my_thread_init(), otherwise we get a coredump in DBUG_ stuff
  my_thread_init();
3140
  DBUG_ENTER("handle_slave_sql");
unknown's avatar
unknown committed
3141

3142
  DBUG_ASSERT(rli->inited);
Marc Alff's avatar
Marc Alff committed
3143
  mysql_mutex_lock(&rli->run_lock);
3144
  DBUG_ASSERT(!rli->slave_running);
unknown's avatar
unknown committed
3145
  errmsg= 0;
unknown's avatar
unknown committed
3146
#ifndef DBUG_OFF
3147
  rli->events_till_abort = abort_slave_event_count;
unknown's avatar
unknown committed
3148
#endif
3149

unknown's avatar
unknown committed
3150
  thd = new THD; // note that contructor of THD uses DBUG_ !
3151
  thd->thread_stack = (char*)&thd; // remember where our stack is
3152
  rli->sql_thd= thd;
3153
  
unknown's avatar
unknown committed
3154 3155
  /* Inform waiting threads that slave has started */
  rli->slave_run_id++;
3156
  rli->slave_running = 1;
unknown's avatar
unknown committed
3157

3158 3159
  pthread_detach_this_thread();
  if (init_slave_thread(thd, SLAVE_THD_SQL))
unknown's avatar
unknown committed
3160 3161 3162 3163 3164
  {
    /*
      TODO: this is currently broken - slave start and change master
      will be stuck if we fail here
    */
Marc Alff's avatar
Marc Alff committed
3165 3166
    mysql_cond_broadcast(&rli->start_cond);
    mysql_mutex_unlock(&rli->run_lock);
3167 3168
    rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, 
                "Failed during slave thread initialization");
unknown's avatar
unknown committed
3169 3170
    goto err;
  }
3171
  thd->init_for_queries();
3172
  thd->temporary_tables = rli->save_temporary_tables; // restore temp tables
3173
  set_thd_in_use_temporary_tables(rli);   // (re)set sql_thd in use for saved temp tables
Marc Alff's avatar
Marc Alff committed
3174
  mysql_mutex_lock(&LOCK_thread_count);
3175
  threads.append(thd);
Marc Alff's avatar
Marc Alff committed
3176
  mysql_mutex_unlock(&LOCK_thread_count);
3177 3178 3179 3180 3181 3182 3183 3184
  /*
    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.
  */
3185
  rli->abort_slave = 0;
Marc Alff's avatar
Marc Alff committed
3186 3187
  mysql_mutex_unlock(&rli->run_lock);
  mysql_cond_broadcast(&rli->start_cond);
3188

unknown's avatar
unknown committed
3189 3190 3191
  /*
    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
3192
    though there's no problem anymore). Do not reset the master timestamp
3193 3194 3195 3196
    (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
3197
    But the master timestamp is reset by RESET SLAVE & CHANGE MASTER.
unknown's avatar
unknown committed
3198
  */
3199
  rli->clear_error();
3200 3201

  //tell the I/O thread to take relay_log_space_limit into account from now on
Marc Alff's avatar
Marc Alff committed
3202
  mysql_mutex_lock(&rli->log_space_lock);
3203
  rli->ignore_log_space_limit= 0;
Marc Alff's avatar
Marc Alff committed
3204
  mysql_mutex_unlock(&rli->log_space_lock);
3205
  rli->trans_retries= 0; // start from "no error"
unknown's avatar
unknown committed
3206
  DBUG_PRINT("info", ("rli->trans_retries: %lu", rli->trans_retries));
3207

3208
  if (init_relay_log_pos(rli,
unknown's avatar
unknown committed
3209 3210 3211
                         rli->group_relay_log_name,
                         rli->group_relay_log_pos,
                         1 /*need data lock*/, &errmsg,
3212
                         1 /*look for a description_event*/))
3213 3214 3215
  { 
    rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, 
                "Error initializing relay log position: %s", errmsg);
3216 3217
    goto err;
  }
3218
  THD_CHECK_SENTRY(thd);
3219 3220 3221 3222
#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
3223
                        llstr(my_b_tell(rli->cur_log),llbuf1),
3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242
                        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
3243
  DBUG_ASSERT(rli->sql_thd == thd);
3244 3245

  DBUG_PRINT("master_info",("log_file_name: %s  position: %s",
unknown's avatar
unknown committed
3246 3247
                            rli->group_master_log_name,
                            llstr(rli->group_master_log_pos,llbuff)));
3248
  if (global_system_variables.log_warnings)
3249
    sql_print_information("Slave SQL thread initialized, starting replication in \
unknown's avatar
unknown committed
3250
log '%s' at position %s, relay log '%s' position: %s", RPL_LOG_NAME,
unknown's avatar
unknown committed
3251 3252
                    llstr(rli->group_master_log_pos,llbuff),rli->group_relay_log_name,
                    llstr(rli->group_relay_log_pos,llbuff1));
3253

3254
  if (check_temp_dir(rli->slave_patternload_file))
3255
  {
Marc Alff's avatar
Marc Alff committed
3256
    rli->report(ERROR_LEVEL, thd->stmt_da->sql_errno(), 
3257
                "Unable to use slave's temporary directory %s - %s", 
Marc Alff's avatar
Marc Alff committed
3258
                slave_load_tmpdir, thd->stmt_da->message());
3259 3260 3261
    goto err;
  }

unknown's avatar
unknown committed
3262
  /* execute init_slave variable */
3263
  if (opt_init_slave.length)
unknown's avatar
unknown committed
3264
  {
3265
    execute_init_command(thd, &opt_init_slave, &LOCK_sys_init_slave);
3266
    if (thd->is_slave_error)
unknown's avatar
unknown committed
3267
    {
Marc Alff's avatar
Marc Alff committed
3268
      rli->report(ERROR_LEVEL, thd->stmt_da->sql_errno(),
3269
                  "Slave SQL thread aborted. Can't execute init_slave query");
unknown's avatar
unknown committed
3270 3271 3272 3273
      goto err;
    }
  }

3274 3275 3276 3277
  /*
    First check until condition - probably there is nothing to execute. We
    do not want to wait for next event in this case.
  */
Marc Alff's avatar
Marc Alff committed
3278
  mysql_mutex_lock(&rli->data_lock);
3279 3280
  if (rli->slave_skip_counter)
  {
3281 3282
    strmake(saved_log_name, rli->group_relay_log_name, FN_REFLEN - 1);
    strmake(saved_master_log_name, rli->group_master_log_name, FN_REFLEN - 1);
3283 3284 3285 3286
    saved_log_pos= rli->group_relay_log_pos;
    saved_master_log_pos= rli->group_master_log_pos;
    saved_skip= rli->slave_skip_counter;
  }
3287
  if (rli->until_condition != Relay_log_info::UNTIL_NONE &&
3288
      rli->is_until_satisfied(thd, NULL))
3289 3290 3291 3292
  {
    char buf[22];
    sql_print_information("Slave SQL thread stopped because it reached its"
                          " UNTIL position %s", llstr(rli->until_pos(), buf));
Marc Alff's avatar
Marc Alff committed
3293
    mysql_mutex_unlock(&rli->data_lock);
3294 3295
    goto err;
  }
Marc Alff's avatar
Marc Alff committed
3296
  mysql_mutex_unlock(&rli->data_lock);
3297

3298 3299
  /* Read queries from the IO/THREAD until this thread is killed */

3300
  while (!sql_slave_killed(thd,rli))
3301
  {
3302
    thd_proc_info(thd, "Reading event from the relay log");
3303
    DBUG_ASSERT(rli->sql_thd == thd);
3304
    THD_CHECK_SENTRY(thd);
3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318 3319

    if (saved_skip && rli->slave_skip_counter == 0)
    {
      sql_print_information("'SQL_SLAVE_SKIP_COUNTER=%ld' executed at "
        "relay_log_file='%s', relay_log_pos='%ld', master_log_name='%s', "
        "master_log_pos='%ld' and new position at "
        "relay_log_file='%s', relay_log_pos='%ld', master_log_name='%s', "
        "master_log_pos='%ld' ",
        (ulong) saved_skip, saved_log_name, (ulong) saved_log_pos,
        saved_master_log_name, (ulong) saved_master_log_pos,
        rli->group_relay_log_name, (ulong) rli->group_relay_log_pos,
        rli->group_master_log_name, (ulong) rli->group_master_log_pos);
      saved_skip= 0;
    }
    
3320 3321
    if (exec_relay_log_event(thd,rli))
    {
3322
      DBUG_PRINT("info", ("exec_relay_log_event() failed"));
3323
      // do not scare the user if SQL thread was simply killed or stopped
3324
      if (!sql_slave_killed(thd,rli))
unknown's avatar
unknown committed
3325 3326
      {
        /*
3327 3328 3329
          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
3330
        */
3331
        uint32 const last_errno= rli->last_error().number;
3332

3333
        if (thd->is_error())
unknown's avatar
unknown committed
3334
        {
Marc Alff's avatar
Marc Alff committed
3335
          char const *const errmsg= thd->stmt_da->message();
3336 3337

          DBUG_PRINT("info",
Marc Alff's avatar
Marc Alff committed
3338 3339
                     ("thd->stmt_da->sql_errno()=%d; rli->last_error.number=%d",
                      thd->stmt_da->sql_errno(), last_errno));
3340
          if (last_errno == 0)
unknown's avatar
unknown committed
3341
          {
3342 3343 3344 3345
            /*
 	      This function is reporting an error which was not reported
 	      while executing exec_relay_log_event().
 	    */ 
3346
            rli->report(ERROR_LEVEL, thd->stmt_da->sql_errno(), "%s", errmsg);
unknown's avatar
unknown committed
3347
          }
Marc Alff's avatar
Marc Alff committed
3348
          else if (last_errno != thd->stmt_da->sql_errno())
unknown's avatar
unknown committed
3349
          {
3350 3351 3352 3353 3354 3355
            /*
             * 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
3356
            sql_print_error("Slave (additional info): %s Error_code: %d",
Marc Alff's avatar
Marc Alff committed
3357
                            errmsg, thd->stmt_da->sql_errno());
unknown's avatar
unknown committed
3358 3359 3360 3361
          }
        }

        /* Print any warnings issued */
Marc Alff's avatar
Marc Alff committed
3362
        List_iterator_fast<MYSQL_ERROR> it(thd->warning_info->warn_list());
unknown's avatar
unknown committed
3363
        MYSQL_ERROR *err;
3364 3365 3366 3367 3368
        /*
          Added controlled slave thread cancel for replication
          of user-defined variables.
        */
        bool udf_error = false;
unknown's avatar
unknown committed
3369
        while ((err= it++))
3370
        {
Marc Alff's avatar
Marc Alff committed
3371
          if (err->get_sql_errno() == ER_CANT_OPEN_LIBRARY)
3372
            udf_error = true;
Marc Alff's avatar
Marc Alff committed
3373
          sql_print_warning("Slave: %s Error_code: %d", err->get_message_text(), err->get_sql_errno());
3374 3375 3376 3377 3378 3379 3380 3381 3382
        }
        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("\
3383
Error running query, slave SQL thread aborted. Fix the problem, and restart \
unknown's avatar
unknown committed
3384
the slave SQL thread with \"SLAVE START\". We stopped at log \
3385
'%s' position %s", RPL_LOG_NAME, llstr(rli->group_master_log_pos, llbuff));
unknown's avatar
unknown committed
3386
      }
3387 3388
      goto err;
    }
3389
  }
3390

3391
  /* Thread stopped. Print the current replication position to the log */
unknown's avatar
unknown committed
3392
  sql_print_information("Slave SQL thread exiting, replication stopped in log "
unknown's avatar
unknown committed
3393 3394
                        "'%s' at position %s",
                        RPL_LOG_NAME, llstr(rli->group_master_log_pos,llbuff));
3395 3396

 err:
3397 3398 3399 3400 3401 3402 3403

  /*
    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:
  */
3404
  thd->clear_error();
3405
  rli->cleanup_context(thd, 1);
3406 3407 3408 3409 3410
  /*
    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)).
  */
3411
  thd->catalog= 0;
3412
  thd->reset_query();
unknown's avatar
unknown committed
3413
  thd->reset_db(NULL, 0);
3414
  thd_proc_info(thd, "Waiting for slave mutex on exit");
Marc Alff's avatar
Marc Alff committed
3415
  mysql_mutex_lock(&rli->run_lock);
3416
  /* We need data_lock, at least to wake up any waiting master_pos_wait() */
Marc Alff's avatar
Marc Alff committed
3417
  mysql_mutex_lock(&rli->data_lock);
3418
  DBUG_ASSERT(rli->slave_running == 1); // tracking buffer overrun
3419
  /* When master_pos_wait() wakes up it will check this and terminate */
unknown's avatar
unknown committed
3420
  rli->slave_running= 0;
3421 3422 3423
  /* Forget the relay log's format */
  delete rli->relay_log.description_event_for_exec;
  rli->relay_log.description_event_for_exec= 0;
3424
  /* Wake up master_pos_wait() */
Marc Alff's avatar
Marc Alff committed
3425
  mysql_mutex_unlock(&rli->data_lock);
3426
  DBUG_PRINT("info",("Signaling possibly waiting master_pos_wait() functions"));
Marc Alff's avatar
Marc Alff committed
3427
  mysql_cond_broadcast(&rli->data_cond);
unknown's avatar
unknown committed
3428
  rli->ignore_log_space_limit= 0; /* don't need any lock */
3429 3430
  /* we die so won't remember charset - re-update them on next thread start */
  rli->cached_charset_invalidate();
3431
  rli->save_temporary_tables = thd->temporary_tables;
unknown's avatar
unknown committed
3432 3433 3434 3435 3436

  /*
    TODO: see if we can do this conditionally in next_event() instead
    to avoid unneeded position re-init
  */
3437 3438 3439 3440
  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);
3441
  THD_CHECK_SENTRY(thd);
3442
  rli->sql_thd= 0;
3443
  set_thd_in_use_temporary_tables(rli);  // (re)set sql_thd in use for saved temp tables
Marc Alff's avatar
Marc Alff committed
3444
  mysql_mutex_lock(&LOCK_thread_count);
3445
  THD_CHECK_SENTRY(thd);
3446
  delete thd;
Marc Alff's avatar
Marc Alff committed
3447
  mysql_mutex_unlock(&LOCK_thread_count);
3448 3449 3450 3451 3452
 /*
  Note: the order of the broadcast and unlock calls below (first broadcast, then unlock)
  is important. Otherwise a killer_thread can execute between the calls and
  delete the mi structure leading to a crash! (see BUG#25306 for details)
 */ 
Marc Alff's avatar
Marc Alff committed
3453
  mysql_cond_broadcast(&rli->stop_cond);
3454
  DBUG_EXECUTE_IF("simulate_slave_delay_at_terminate_bug38694", sleep(5););
Marc Alff's avatar
Marc Alff committed
3455
  mysql_mutex_unlock(&rli->run_lock);  // tell the world we are done
3456 3457

  DBUG_LEAVE;                                   // Must match DBUG_ENTER()
unknown's avatar
unknown committed
3458
  my_thread_end();
3459
  pthread_exit(0);
3460
  return 0;                                     // Avoid compiler warnings
3461
}
unknown's avatar
unknown committed
3462

3463

unknown's avatar
unknown committed
3464
/*
3465
  process_io_create_file()
unknown's avatar
unknown committed
3466
*/
3467

3468
static int process_io_create_file(Master_info* mi, Create_file_log_event* cev)
unknown's avatar
unknown committed
3469 3470 3471 3472
{
  int error = 1;
  ulong num_bytes;
  bool cev_not_written;
3473 3474
  THD *thd = mi->io_thd;
  NET *net = &mi->mysql->net;
unknown's avatar
unknown committed
3475
  DBUG_ENTER("process_io_create_file");
unknown's avatar
unknown committed
3476 3477

  if (unlikely(!cev->is_valid()))
unknown's avatar
unknown committed
3478
    DBUG_RETURN(1);
3479 3480

  if (!rpl_filter->db_ok(cev->db))
unknown's avatar
unknown committed
3481 3482
  {
    skip_load_data_infile(net);
unknown's avatar
unknown committed
3483
    DBUG_RETURN(0);
unknown's avatar
unknown committed
3484 3485 3486
  }
  DBUG_ASSERT(cev->inited_from_old);
  thd->file_id = cev->file_id = mi->file_id++;
3487
  thd->server_id = cev->server_id;
unknown's avatar
unknown committed
3488
  cev_not_written = 1;
unknown's avatar
unknown committed
3489

unknown's avatar
unknown committed
3490 3491 3492
  if (unlikely(net_request_file(net,cev->fname)))
  {
    sql_print_error("Slave I/O: failed requesting download of '%s'",
unknown's avatar
unknown committed
3493
                    cev->fname);
unknown's avatar
unknown committed
3494 3495 3496
    goto err;
  }

unknown's avatar
unknown committed
3497 3498 3499 3500
  /*
    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
3501 3502
  */
  {
3503
    Append_block_log_event aev(thd,0,0,0,0);
unknown's avatar
unknown committed
3504

unknown's avatar
unknown committed
3505 3506 3507 3508
    for (;;)
    {
      if (unlikely((num_bytes=my_net_read(net)) == packet_error))
      {
unknown's avatar
unknown committed
3509 3510 3511
        sql_print_error("Network read error downloading '%s' from master",
                        cev->fname);
        goto err;
unknown's avatar
unknown committed
3512 3513 3514
      }
      if (unlikely(!num_bytes)) /* eof */
      {
3515 3516
	/* 3.23 master wants it */
        net_write_command(net, 0, (uchar*) "", 0, (uchar*) "", 0);
unknown's avatar
unknown committed
3517 3518 3519 3520 3521 3522 3523
        /*
          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
3524 3525 3526 3527 3528
          break;
        Execute_load_log_event xev(thd,0,0);
        xev.log_pos = cev->log_pos;
        if (unlikely(mi->rli.relay_log.append(&xev)))
        {
3529 3530 3531
          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
3532 3533 3534 3535
          goto err;
        }
        mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total);
        break;
unknown's avatar
unknown committed
3536 3537 3538
      }
      if (unlikely(cev_not_written))
      {
unknown's avatar
unknown committed
3539
        cev->block = net->read_pos;
unknown's avatar
unknown committed
3540 3541 3542
        cev->block_len = num_bytes;
        if (unlikely(mi->rli.relay_log.append(cev)))
        {
3543 3544 3545
          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
3546 3547 3548 3549
          goto err;
        }
        cev_not_written=0;
        mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total);
unknown's avatar
unknown committed
3550 3551 3552
      }
      else
      {
unknown's avatar
unknown committed
3553
        aev.block = net->read_pos;
unknown's avatar
unknown committed
3554 3555 3556 3557
        aev.block_len = num_bytes;
        aev.log_pos = cev->log_pos;
        if (unlikely(mi->rli.relay_log.append(&aev)))
        {
3558 3559 3560
          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
3561 3562 3563
          goto err;
        }
        mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total) ;
unknown's avatar
unknown committed
3564 3565 3566 3567 3568
      }
    }
  }
  error=0;
err:
unknown's avatar
unknown committed
3569
  DBUG_RETURN(error);
unknown's avatar
unknown committed
3570
}
unknown's avatar
unknown committed
3571

3572

unknown's avatar
unknown committed
3573
/*
unknown's avatar
unknown committed
3574 3575 3576 3577
  Start using a new binary log on the master

  SYNOPSIS
    process_io_rotate()
unknown's avatar
unknown committed
3578 3579
    mi                  master_info for the slave
    rev                 The rotate log event read from the binary log
unknown's avatar
unknown committed
3580 3581

  DESCRIPTION
3582
    Updates the master info with the place in the next binary
unknown's avatar
unknown committed
3583
    log where we should start reading.
3584
    Rotate the relay log to avoid mixed-format relay logs.
unknown's avatar
unknown committed
3585 3586 3587 3588 3589

  NOTES
    We assume we already locked mi->data_lock

  RETURN VALUES
unknown's avatar
unknown committed
3590 3591
    0           ok
    1           Log event is illegal
unknown's avatar
unknown committed
3592 3593 3594

*/

3595
static int process_io_rotate(Master_info *mi, Rotate_log_event *rev)
3596
{
3597
  DBUG_ENTER("process_io_rotate");
Marc Alff's avatar
Marc Alff committed
3598
  mysql_mutex_assert_owner(&mi->data_lock);
3599

unknown's avatar
unknown committed
3600
  if (unlikely(!rev->is_valid()))
3601
    DBUG_RETURN(1);
unknown's avatar
unknown committed
3602

3603
  /* Safe copy as 'rev' has been "sanitized" in Rotate_log_event's ctor */
unknown's avatar
unknown committed
3604 3605
  memcpy(mi->master_log_name, rev->new_log_ident, rev->ident_len+1);
  mi->master_log_pos= rev->pos;
unknown's avatar
unknown committed
3606
  DBUG_PRINT("info", ("master_log_pos: '%s' %lu",
unknown's avatar
unknown committed
3607
                      mi->master_log_name, (ulong) mi->master_log_pos));
3608
#ifndef DBUG_OFF
unknown's avatar
unknown committed
3609 3610 3611 3612 3613
  /*
    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)
3614
    mi->events_till_disconnect++;
3615
#endif
3616

3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630
  /*
    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);
  }
3631 3632 3633 3634
  /*
    Rotate the relay log makes binlog format detection easier (at next slave
    start or mysqlbinlog)
  */
3635
  rotate_relay_log(mi); /* will take the right mutexes */
3636
  DBUG_RETURN(0);
3637 3638
}

unknown's avatar
unknown committed
3639
/*
3640 3641
  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
3642
*/
3643
static int queue_binlog_ver_1_event(Master_info *mi, const char *buf,
unknown's avatar
unknown committed
3644
                           ulong event_len)
3645
{
unknown's avatar
unknown committed
3646
  const char *errmsg = 0;
unknown's avatar
unknown committed
3647 3648 3649
  ulong inc_pos;
  bool ignore_event= 0;
  char *tmp_buf = 0;
3650
  Relay_log_info *rli= &mi->rli;
3651
  DBUG_ENTER("queue_binlog_ver_1_event");
unknown's avatar
unknown committed
3652

unknown's avatar
unknown committed
3653 3654 3655
  /*
    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
3656 3657 3658 3659 3660
  */
  if (buf[EVENT_TYPE_OFFSET] == LOAD_EVENT)
  {
    if (unlikely(!(tmp_buf=(char*)my_malloc(event_len+1,MYF(MY_WME)))))
    {
3661 3662
      mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR,
                 ER(ER_SLAVE_FATAL_ERROR), "Memory allocation failed");
unknown's avatar
unknown committed
3663
      DBUG_RETURN(1);
unknown's avatar
unknown committed
3664 3665
    }
    memcpy(tmp_buf,buf,event_len);
3666 3667 3668 3669 3670 3671 3672 3673
    /*
      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
3674
    int4store(tmp_buf+EVENT_LEN_OFFSET, event_len);
unknown's avatar
unknown committed
3675 3676
    buf = (const char*)tmp_buf;
  }
3677 3678 3679 3680 3681 3682
  /*
    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
3683
  Log_event *ev = Log_event::read_log_event(buf,event_len, &errmsg,
3684
                                            mi->rli.relay_log.description_event_for_queue);
3685
  if (unlikely(!ev))
3686 3687
  {
    sql_print_error("Read invalid event from master: '%s',\
unknown's avatar
unknown committed
3688
 master could be corrupt but a more likely cause of this is a bug",
unknown's avatar
unknown committed
3689
                    errmsg);
3690
    my_free(tmp_buf);
unknown's avatar
unknown committed
3691
    DBUG_RETURN(1);
3692
  }
3693

Marc Alff's avatar
Marc Alff committed
3694
  mysql_mutex_lock(&mi->data_lock);
3695
  ev->log_pos= mi->master_log_pos; /* 3.23 events don't contain log_pos */
unknown's avatar
unknown committed
3696
  switch (ev->get_type_code()) {
unknown's avatar
unknown committed
3697
  case STOP_EVENT:
3698
    ignore_event= 1;
unknown's avatar
unknown committed
3699 3700
    inc_pos= event_len;
    break;
3701
  case ROTATE_EVENT:
3702
    if (unlikely(process_io_rotate(mi,(Rotate_log_event*)ev)))
3703 3704
    {
      delete ev;
Marc Alff's avatar
Marc Alff committed
3705
      mysql_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
3706
      DBUG_RETURN(1);
3707
    }
unknown's avatar
unknown committed
3708
    inc_pos= 0;
3709
    break;
unknown's avatar
unknown committed
3710
  case CREATE_FILE_EVENT:
3711 3712 3713 3714 3715 3716
    /*
      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
3717
  {
unknown's avatar
unknown committed
3718
    /* We come here when and only when tmp_buf != 0 */
3719
    DBUG_ASSERT(tmp_buf != 0);
3720 3721
    inc_pos=event_len;
    ev->log_pos+= inc_pos;
unknown's avatar
unknown committed
3722
    int error = process_io_create_file(mi,(Create_file_log_event*)ev);
3723
    delete ev;
3724
    mi->master_log_pos += inc_pos;
unknown's avatar
unknown committed
3725
    DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
Marc Alff's avatar
Marc Alff committed
3726
    mysql_mutex_unlock(&mi->data_lock);
3727
    my_free(tmp_buf);
unknown's avatar
unknown committed
3728
    DBUG_RETURN(error);
unknown's avatar
unknown committed
3729
  }
3730
  default:
unknown's avatar
unknown committed
3731
    inc_pos= event_len;
3732 3733
    break;
  }
unknown's avatar
unknown committed
3734
  if (likely(!ignore_event))
3735
  {
unknown's avatar
unknown committed
3736 3737
    if (ev->log_pos)
      /*
3738 3739 3740 3741
         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
3742
    if (unlikely(rli->relay_log.append(ev)))
3743 3744
    {
      delete ev;
Marc Alff's avatar
Marc Alff committed
3745
      mysql_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
3746
      DBUG_RETURN(1);
3747
    }
unknown's avatar
unknown committed
3748
    rli->relay_log.harvest_bytes_written(&rli->log_space_total);
3749 3750
  }
  delete ev;
unknown's avatar
unknown committed
3751
  mi->master_log_pos+= inc_pos;
unknown's avatar
unknown committed
3752
  DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
Marc Alff's avatar
Marc Alff committed
3753
  mysql_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
3754
  DBUG_RETURN(0);
3755 3756
}

3757 3758 3759 3760
/*
  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.
*/
3761
static int queue_binlog_ver_3_event(Master_info *mi, const char *buf,
unknown's avatar
unknown committed
3762
                           ulong event_len)
3763 3764 3765 3766
{
  const char *errmsg = 0;
  ulong inc_pos;
  char *tmp_buf = 0;
3767
  Relay_log_info *rli= &mi->rli;
3768 3769 3770 3771 3772 3773 3774 3775 3776
  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
3777
                    errmsg);
3778
    my_free(tmp_buf);
3779 3780
    DBUG_RETURN(1);
  }
Marc Alff's avatar
Marc Alff committed
3781
  mysql_mutex_lock(&mi->data_lock);
3782 3783 3784 3785 3786 3787 3788
  switch (ev->get_type_code()) {
  case STOP_EVENT:
    goto err;
  case ROTATE_EVENT:
    if (unlikely(process_io_rotate(mi,(Rotate_log_event*)ev)))
    {
      delete ev;
Marc Alff's avatar
Marc Alff committed
3789
      mysql_mutex_unlock(&mi->data_lock);
3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800
      DBUG_RETURN(1);
    }
    inc_pos= 0;
    break;
  default:
    inc_pos= event_len;
    break;
  }
  if (unlikely(rli->relay_log.append(ev)))
  {
    delete ev;
Marc Alff's avatar
Marc Alff committed
3801
    mysql_mutex_unlock(&mi->data_lock);
3802 3803 3804 3805 3806 3807
    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
3808
  DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
Marc Alff's avatar
Marc Alff committed
3809
  mysql_mutex_unlock(&mi->data_lock);
3810 3811 3812 3813 3814 3815 3816 3817 3818 3819
  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
3820
  TODO:
3821 3822 3823 3824
    Test this code before release - it has to be tested on a separate
    setup with 3.23 master or 4.0 master
*/

3825
static int queue_old_event(Master_info *mi, const char *buf,
unknown's avatar
unknown committed
3826
                           ulong event_len)
3827
{
3828 3829
  DBUG_ENTER("queue_old_event");

3830 3831 3832
  switch (mi->rli.relay_log.description_event_for_queue->binlog_version)
  {
  case 1:
3833
      DBUG_RETURN(queue_binlog_ver_1_event(mi,buf,event_len));
3834
  case 3:
3835
      DBUG_RETURN(queue_binlog_ver_3_event(mi,buf,event_len));
3836 3837
  default: /* unsupported format; eg version 2 */
    DBUG_PRINT("info",("unsupported binlog format %d in queue_old_event()",
unknown's avatar
unknown committed
3838
                       mi->rli.relay_log.description_event_for_queue->binlog_version));
3839
    DBUG_RETURN(1);
3840 3841
  }
}
3842

unknown's avatar
unknown committed
3843
/*
3844 3845
  queue_event()

3846 3847 3848 3849 3850
  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
3851 3852
*/

3853
static int queue_event(Master_info* mi,const char* buf, ulong event_len)
3854
{
unknown's avatar
unknown committed
3855
  int error= 0;
Andrei Elkin's avatar
Andrei Elkin committed
3856
  String error_msg;
unknown's avatar
unknown committed
3857
  ulong inc_pos;
3858
  Relay_log_info *rli= &mi->rli;
Marc Alff's avatar
Marc Alff committed
3859
  mysql_mutex_t *log_lock= rli->relay_log.get_log_lock();
Andrei Elkin's avatar
Andrei Elkin committed
3860
  ulong s_id;
unknown's avatar
unknown committed
3861 3862
  DBUG_ENTER("queue_event");

unknown's avatar
unknown committed
3863 3864
  LINT_INIT(inc_pos);

3865 3866
  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
3867
    DBUG_RETURN(queue_old_event(mi,buf,event_len));
3868

3869
  LINT_INIT(inc_pos);
Marc Alff's avatar
Marc Alff committed
3870
  mysql_mutex_lock(&mi->data_lock);
unknown's avatar
unknown committed
3871

unknown's avatar
unknown committed
3872
  switch (buf[EVENT_TYPE_OFFSET]) {
3873
  case STOP_EVENT:
3874 3875
    /*
      We needn't write this event to the relay log. Indeed, it just indicates a
unknown's avatar
unknown committed
3876 3877 3878
      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
3879
      prepared statements' deletion are TODO only when we binlog prep stmts).
unknown's avatar
unknown committed
3880

unknown's avatar
unknown committed
3881 3882 3883 3884
      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).
3885 3886
    */
    goto err;
3887 3888
  case ROTATE_EVENT:
  {
unknown's avatar
unknown committed
3889
    Rotate_log_event rev(buf,event_len,mi->rli.relay_log.description_event_for_queue);
3890
    if (unlikely(process_io_rotate(mi,&rev)))
unknown's avatar
unknown committed
3891
    {
Andrei Elkin's avatar
Andrei Elkin committed
3892
      error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
3893
      goto err;
unknown's avatar
unknown committed
3894
    }
3895 3896 3897 3898
    /*
      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
3899
    inc_pos= 0;
3900 3901
    break;
  }
3902 3903 3904 3905 3906 3907 3908
  case FORMAT_DESCRIPTION_EVENT:
  {
    /*
      Create an event, and save it (when we rotate the relay log, we will have
      to write this event again).
    */
    /*
3909 3910 3911
      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).
3912
    */
3913
    Format_description_log_event* tmp;
3914
    const char* errmsg;
3915
    if (!(tmp= (Format_description_log_event*)
3916
          Log_event::read_log_event(buf, event_len, &errmsg,
3917
                                    mi->rli.relay_log.description_event_for_queue)))
3918
    {
Andrei Elkin's avatar
Andrei Elkin committed
3919
      error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
3920 3921
      goto err;
    }
3922 3923
    delete mi->rli.relay_log.description_event_for_queue;
    mi->rli.relay_log.description_event_for_queue= tmp;
unknown's avatar
unknown committed
3924
    /*
3925
       Though this does some conversion to the slave's format, this will
unknown's avatar
unknown committed
3926
       preserve the master's binlog format version, and number of event types.
3927
    */
unknown's avatar
unknown committed
3928
    /*
3929
       If the event was not requested by the slave (the slave did not ask for
unknown's avatar
unknown committed
3930
       it), i.e. has end_log_pos=0, we do not increment mi->master_log_pos
3931 3932 3933
    */
    inc_pos= uint4korr(buf+LOG_POS_OFFSET) ? event_len : 0;
    DBUG_PRINT("info",("binlog format is now %d",
unknown's avatar
unknown committed
3934
                       mi->rli.relay_log.description_event_for_queue->binlog_version));
3935 3936 3937

  }
  break;
Andrei Elkin's avatar
Andrei Elkin committed
3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967 3968 3969 3970 3971 3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982 3983 3984 3985 3986

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

       TODO: handling `when' for SHOW SLAVE STATUS' snds behind
    */
    if ((memcmp(mi->master_log_name, hb.get_log_ident(), hb.get_ident_len())
         && mi->master_log_name != NULL)
        || mi->master_log_pos != hb.log_pos)
    {
      /* missed events of heartbeat from the past */
      error= ER_SLAVE_HEARTBEAT_FAILURE;
      error_msg.append(STRING_WITH_LEN("heartbeat is not compatible with local info;"));
      error_msg.append(STRING_WITH_LEN("the event's data: log_file_name "));
      error_msg.append(hb.get_log_ident(), (uint) strlen(hb.get_log_ident()));
      error_msg.append(STRING_WITH_LEN(" log_pos "));
      llstr(hb.log_pos, llbuf);
      error_msg.append(llbuf, strlen(llbuf));
      goto err;
    }
    goto skip_relay_logging;
  }
  break;
3987

3988
  default:
unknown's avatar
unknown committed
3989
    inc_pos= event_len;
3990 3991
    break;
  }
3992

unknown's avatar
unknown committed
3993 3994
  /*
     If this event is originating from this server, don't queue it.
3995
     We don't check this for 3.23 events because it's simpler like this; 3.23
unknown's avatar
unknown committed
3996 3997
     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
3998 3999 4000 4001 4002 4003 4004 4005
     upgrades a 4.0 slave which has a not-filtered relay log).

     ANY event coming from ourselves can be ignored: it is obvious for queries;
     for STOP_EVENT/ROTATE_EVENT/START_EVENT: these cannot come from ourselves
     (--log-slave-updates would not log that) unless this slave is also its
     direct master (an unsupported, useless setup!).
  */

Marc Alff's avatar
Marc Alff committed
4006
  mysql_mutex_lock(log_lock);
Andrei Elkin's avatar
Andrei Elkin committed
4007 4008 4009 4010 4011 4012 4013 4014 4015 4016 4017 4018
  s_id= uint4korr(buf + SERVER_ID_OFFSET);
  if ((s_id == ::server_id && !mi->rli.replicate_same_server_id) ||
      /*
        the following conjunction deals with IGNORE_SERVER_IDS, if set
        If the master is on the ignore list, execution of
        format description log events and rotate events is necessary.
      */
      (mi->ignore_server_ids.elements > 0 &&
       mi->shall_ignore_server_id(s_id) &&
       /* everything is filtered out from non-master */
       (s_id != mi->master_id ||
        /* for the master meta information is necessary */
4019 4020
        (buf[EVENT_TYPE_OFFSET] != FORMAT_DESCRIPTION_EVENT &&
         buf[EVENT_TYPE_OFFSET] != ROTATE_EVENT))))
4021
  {
4022 4023
    /*
      Do not write it to the relay log.
4024 4025 4026 4027 4028 4029 4030 4031
      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.
4032 4033 4034
      But events which were generated by this slave and which do not exist in
      the master's binlog (i.e. Format_desc, Rotate & Stop) should not increment
      mi->master_log_pos.
Andrei Elkin's avatar
Andrei Elkin committed
4035 4036 4037
      If the event is originated remotely and is being filtered out by
      IGNORE_SERVER_IDS it increments mi->master_log_pos
      as well as rli->group_relay_log_pos.
4038
    */
Andrei Elkin's avatar
Andrei Elkin committed
4039
    if (!(s_id == ::server_id && !mi->rli.replicate_same_server_id) ||
4040 4041 4042
        (buf[EVENT_TYPE_OFFSET] != FORMAT_DESCRIPTION_EVENT &&
         buf[EVENT_TYPE_OFFSET] != ROTATE_EVENT &&
         buf[EVENT_TYPE_OFFSET] != STOP_EVENT))
unknown's avatar
unknown committed
4043
    {
4044
      mi->master_log_pos+= inc_pos;
unknown's avatar
unknown committed
4045 4046 4047 4048
      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;
    }
4049
    rli->relay_log.signal_update(); // the slave SQL thread needs to re-check
Andrei Elkin's avatar
Andrei Elkin committed
4050 4051
    DBUG_PRINT("info", ("master_log_pos: %lu, event originating from %u server, ignored",
                        (ulong) mi->master_log_pos, uint4korr(buf + SERVER_ID_OFFSET)));
unknown's avatar
unknown committed
4052
  }
unknown's avatar
unknown committed
4053 4054 4055
  else
  {
    /* write the event to the relay log */
4056
    if (likely(!(rli->relay_log.appendv(buf,event_len,0))))
4057 4058
    {
      mi->master_log_pos+= inc_pos;
unknown's avatar
unknown committed
4059
      DBUG_PRINT("info", ("master_log_pos: %lu", (ulong) mi->master_log_pos));
4060 4061
      rli->relay_log.harvest_bytes_written(&rli->log_space_total);
    }
4062
    else
Andrei Elkin's avatar
Andrei Elkin committed
4063 4064 4065
    {
      error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE;
    }
4066
    rli->ign_master_log_name_end[0]= 0; // last event is not ignored
unknown's avatar
unknown committed
4067
  }
Marc Alff's avatar
Marc Alff committed
4068
  mysql_mutex_unlock(log_lock);
4069

Andrei Elkin's avatar
Andrei Elkin committed
4070 4071
skip_relay_logging:
  
4072
err:
Marc Alff's avatar
Marc Alff committed
4073
  mysql_mutex_unlock(&mi->data_lock);
unknown's avatar
unknown committed
4074
  DBUG_PRINT("info", ("error: %d", error));
Andrei Elkin's avatar
Andrei Elkin committed
4075 4076 4077 4078 4079
  if (error)
    mi->report(ERROR_LEVEL, error, ER(error), 
               (error == ER_SLAVE_RELAY_LOG_WRITE_FAILURE)?
               "could not queue event from master" :
               error_msg.ptr());
unknown's avatar
unknown committed
4080
  DBUG_RETURN(error);
4081 4082
}

4083

4084
void end_relay_log_info(Relay_log_info* rli)
4085
{
4086 4087
  DBUG_ENTER("end_relay_log_info");

4088
  if (!rli->inited)
4089
    DBUG_VOID_RETURN;
4090
  if (rli->info_fd >= 0)
unknown's avatar
unknown committed
4091 4092
  {
    end_io_cache(&rli->info_file);
Marc Alff's avatar
Marc Alff committed
4093
    mysql_file_close(rli->info_fd, MYF(MY_WME));
unknown's avatar
unknown committed
4094 4095
    rli->info_fd = -1;
  }
4096
  if (rli->cur_log_fd >= 0)
unknown's avatar
unknown committed
4097 4098
  {
    end_io_cache(&rli->cache_buf);
Marc Alff's avatar
Marc Alff committed
4099
    mysql_file_close(rli->cur_log_fd, MYF(MY_WME));
unknown's avatar
unknown committed
4100 4101
    rli->cur_log_fd = -1;
  }
4102
  rli->inited = 0;
4103
  rli->relay_log.close(LOG_CLOSE_INDEX | LOG_CLOSE_STOP_EVENT);
4104
  rli->relay_log.harvest_bytes_written(&rli->log_space_total);
unknown's avatar
unknown committed
4105 4106 4107 4108 4109 4110
  /*
    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();
4111
  DBUG_VOID_RETURN;
4112 4113
}

4114 4115 4116 4117 4118 4119 4120 4121 4122 4123 4124 4125 4126 4127 4128 4129 4130 4131 4132

/**
  Hook to detach the active VIO before closing a connection handle.

  The client API might close the connection (and associated data)
  in case it encounters a unrecoverable (network) error. This hook
  is called from the client code before the VIO handle is deleted
  allows the thread to detach the active vio so it does not point
  to freed memory.

  Other calls to THD::clear_active_vio throughout this module are
  redundant due to the hook but are left in place for illustrative
  purposes.
*/

extern "C" void slave_io_thread_detach_vio()
{
#ifdef SIGNAL_WITH_VIO_CLOSE
  THD *thd= current_thd;
4133
  if (thd && thd->slave_thread)
4134 4135 4136 4137 4138
    thd->clear_active_vio();
#endif
}


unknown's avatar
unknown committed
4139 4140
/*
  Try to connect until successful or slave killed
4141

unknown's avatar
unknown committed
4142 4143
  SYNPOSIS
    safe_connect()
unknown's avatar
unknown committed
4144 4145 4146
    thd                 Thread handler for slave
    mysql               MySQL connection handle
    mi                  Replication handle
4147

unknown's avatar
unknown committed
4148
  RETURN
unknown's avatar
unknown committed
4149 4150
    0   ok
    #   Error
unknown's avatar
unknown committed
4151
*/
4152

4153
static int safe_connect(THD* thd, MYSQL* mysql, Master_info* mi)
unknown's avatar
unknown committed
4154
{
4155 4156 4157
  DBUG_ENTER("safe_connect");

  DBUG_RETURN(connect_to_master(thd, mysql, mi, 0, 0));
unknown's avatar
unknown committed
4158 4159
}

4160

4161
/*
unknown's avatar
unknown committed
4162 4163
  SYNPOSIS
    connect_to_master()
unknown's avatar
unknown committed
4164

unknown's avatar
unknown committed
4165 4166 4167
  IMPLEMENTATION
    Try to connect until successful or slave killed or we have retried
    master_retry_count times
4168
*/
unknown's avatar
unknown committed
4169

4170
static int connect_to_master(THD* thd, MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
4171
                             bool reconnect, bool suppress_warnings)
unknown's avatar
unknown committed
4172
{
4173
  int slave_was_killed;
unknown's avatar
unknown committed
4174
  int last_errno= -2;                           // impossible error
4175
  ulong err_count=0;
unknown's avatar
unknown committed
4176
  char llbuff[22];
4177
  DBUG_ENTER("connect_to_master");
unknown's avatar
unknown committed
4178

unknown's avatar
unknown committed
4179
#ifndef DBUG_OFF
4180
  mi->events_till_disconnect = disconnect_slave_event_count;
unknown's avatar
unknown committed
4181
#endif
4182
  ulong client_flag= CLIENT_REMEMBER_OPTIONS;
4183
  if (opt_slave_compressed_protocol)
unknown's avatar
unknown committed
4184
    client_flag=CLIENT_COMPRESS;                /* We will use compression */
4185

4186 4187
  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
4188

unknown's avatar
unknown committed
4189 4190
#ifdef HAVE_OPENSSL
  if (mi->ssl)
4191
  {
unknown's avatar
unknown committed
4192
    mysql_ssl_set(mysql,
unknown's avatar
unknown committed
4193
                  mi->ssl_key[0]?mi->ssl_key:0,
unknown's avatar
unknown committed
4194
                  mi->ssl_cert[0]?mi->ssl_cert:0,
unknown's avatar
unknown committed
4195 4196 4197
                  mi->ssl_ca[0]?mi->ssl_ca:0,
                  mi->ssl_capath[0]?mi->ssl_capath:0,
                  mi->ssl_cipher[0]?mi->ssl_cipher:0);
4198 4199 4200
    mysql_options(mysql, MYSQL_OPT_SSL_VERIFY_SERVER_CERT,
                  &mi->ssl_verify_server_cert);
  }
unknown's avatar
unknown committed
4201 4202
#endif

4203 4204 4205 4206
  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);

4207
  while (!(slave_was_killed = io_slave_killed(thd,mi)) &&
unknown's avatar
unknown committed
4208 4209 4210
         (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
4211
  {
4212
    /* Don't repeat last error */
unknown's avatar
SCRUM  
unknown committed
4213
    if ((int)mysql_errno(mysql) != last_errno)
4214
    {
unknown's avatar
SCRUM  
unknown committed
4215
      last_errno=mysql_errno(mysql);
unknown's avatar
unknown committed
4216
      suppress_warnings= 0;
4217 4218 4219 4220 4221 4222
      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);
4223
    }
unknown's avatar
unknown committed
4224 4225 4226
    /*
      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
4227
      do not want to have election triggered on the first failure to
unknown's avatar
unknown committed
4228
      connect
4229
    */
4230
    if (++err_count == master_retry_count)
4231 4232
    {
      slave_was_killed=1;
unknown's avatar
unknown committed
4233 4234
      if (reconnect)
        change_rpl_status(RPL_ACTIVE_SLAVE,RPL_LOST_SOLDIER);
4235 4236
      break;
    }
4237
    safe_sleep(thd,mi->connect_retry,(CHECK_KILLED_FUNC)io_slave_killed,
unknown's avatar
unknown committed
4238
               (void*)mi);
unknown's avatar
unknown committed
4239
  }
4240

4241 4242
  if (!slave_was_killed)
  {
4243
    mi->clear_error(); // clear possible left over reconnect error
unknown's avatar
unknown committed
4244
    if (reconnect)
unknown's avatar
unknown committed
4245
    {
4246
      if (!suppress_warnings && global_system_variables.log_warnings)
unknown's avatar
unknown committed
4247
        sql_print_information("Slave: connected to master '%s@%s:%d',\
4248
replication resumed in log '%s' at position %s", mi->user,
unknown's avatar
unknown committed
4249 4250 4251
                        mi->host, mi->port,
                        IO_RPL_LOG_NAME,
                        llstr(mi->master_log_pos,llbuff));
unknown's avatar
unknown committed
4252
    }
unknown's avatar
unknown committed
4253 4254 4255
    else
    {
      change_rpl_status(RPL_IDLE_SLAVE,RPL_ACTIVE_SLAVE);
4256 4257
      general_log_print(thd, COM_CONNECT_OUT, "%s@%s:%d",
                        mi->user, mi->host, mi->port);
unknown's avatar
unknown committed
4258
    }
4259
#ifdef SIGNAL_WITH_VIO_CLOSE
4260
    thd->set_active_vio(mysql->net.vio);
unknown's avatar
unknown committed
4261
#endif
4262
  }
4263
  mysql->reconnect= 1;
4264 4265
  DBUG_PRINT("exit",("slave_was_killed: %d", slave_was_killed));
  DBUG_RETURN(slave_was_killed);
unknown's avatar
unknown committed
4266 4267
}

4268

unknown's avatar
unknown committed
4269
/*
4270
  safe_reconnect()
unknown's avatar
unknown committed
4271

unknown's avatar
unknown committed
4272 4273 4274
  IMPLEMENTATION
    Try to connect until successful or slave killed or we have retried
    master_retry_count times
unknown's avatar
unknown committed
4275 4276
*/

4277
static int safe_reconnect(THD* thd, MYSQL* mysql, Master_info* mi,
unknown's avatar
unknown committed
4278
                          bool suppress_warnings)
unknown's avatar
unknown committed
4279
{
unknown's avatar
unknown committed
4280 4281
  DBUG_ENTER("safe_reconnect");
  DBUG_RETURN(connect_to_master(thd, mysql, mi, 1, suppress_warnings));
unknown's avatar
unknown committed
4282 4283
}

unknown's avatar
unknown committed
4284

He Zhenxing's avatar
He Zhenxing committed
4285 4286 4287 4288 4289 4290 4291 4292 4293 4294 4295 4296 4297 4298 4299 4300 4301 4302 4303 4304 4305 4306 4307 4308 4309 4310 4311 4312 4313 4314 4315 4316 4317 4318 4319 4320 4321 4322 4323 4324 4325 4326 4327 4328 4329 4330 4331 4332 4333 4334 4335 4336 4337 4338 4339 4340 4341 4342 4343 4344 4345 4346 4347 4348 4349
MYSQL *rpl_connect_master(MYSQL *mysql)
{
  THD *thd= current_thd;
  Master_info *mi= my_pthread_getspecific_ptr(Master_info*, RPL_MASTER_INFO);
  if (!mi)
  {
    sql_print_error("'rpl_connect_master' must be called in slave I/O thread context.");
    return NULL;
  }

  bool allocated= false;
  
  if (!mysql)
  {
    if(!(mysql= mysql_init(NULL)))
    {
      sql_print_error("rpl_connect_master: failed in mysql_init()");
      return NULL;
    }
    allocated= true;
  }

  /*
    XXX: copied from connect_to_master, this function should not
    change the slave status, so we cannot use connect_to_master
    directly
    
    TODO: make this part a seperate function to eliminate duplication
  */
  mysql_options(mysql, MYSQL_OPT_CONNECT_TIMEOUT, (char *) &slave_net_timeout);
  mysql_options(mysql, MYSQL_OPT_READ_TIMEOUT, (char *) &slave_net_timeout);

#ifdef HAVE_OPENSSL
  if (mi->ssl)
  {
    mysql_ssl_set(mysql,
                  mi->ssl_key[0]?mi->ssl_key:0,
                  mi->ssl_cert[0]?mi->ssl_cert:0,
                  mi->ssl_ca[0]?mi->ssl_ca:0,
                  mi->ssl_capath[0]?mi->ssl_capath:0,
                  mi->ssl_cipher[0]?mi->ssl_cipher:0);
    mysql_options(mysql, MYSQL_OPT_SSL_VERIFY_SERVER_CERT,
                  &mi->ssl_verify_server_cert);
  }
#endif

  mysql_options(mysql, MYSQL_SET_CHARSET_NAME, default_charset_info->csname);
  /* This one is not strictly needed but we have it here for completeness */
  mysql_options(mysql, MYSQL_SET_CHARSET_DIR, (char *) charsets_dir);

  if (io_slave_killed(thd, mi)
      || !mysql_real_connect(mysql, mi->host, mi->user, mi->password, 0,
                             mi->port, 0, 0))
  {
    if (!io_slave_killed(thd, mi))
      sql_print_error("rpl_connect_master: error connecting to master: %s (server_error: %d)",
                      mysql_error(mysql), mysql_errno(mysql));
    
    if (allocated)
      mysql_close(mysql);                       // this will free the object
    return NULL;
  }
  return mysql;
}

4350 4351 4352 4353 4354 4355
/*
  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
4356
    rli                 Relay log information
4357 4358

  NOTES
4359 4360 4361
    - As this is only called by the slave thread or on STOP SLAVE, with the
      log_lock grabbed and the slave thread stopped, we don't need to have 
      a lock here.
4362 4363 4364 4365 4366 4367 4368 4369 4370 4371 4372 4373 4374 4375
    - 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
4376 4377
    0   ok
    1   write error
4378 4379
*/

4380
bool flush_relay_log_info(Relay_log_info* rli)
4381
{
4382
  bool error=0;
4383
  DBUG_ENTER("flush_relay_log_info");
4384 4385

  if (unlikely(rli->no_storage))
4386
    DBUG_RETURN(0);
4387

4388 4389 4390
  IO_CACHE *file = &rli->info_file;
  char buff[FN_REFLEN*2+22*2+4], *pos;

4391
  my_b_seek(file, 0L);
4392
  pos=strmov(buff, rli->group_relay_log_name);
4393
  *pos++='\n';
4394
  pos=longlong2str(rli->group_relay_log_pos, pos, 10);
4395
  *pos++='\n';
4396
  pos=strmov(pos, rli->group_master_log_name);
4397
  *pos++='\n';
4398
  pos=longlong2str(rli->group_master_log_pos, pos, 10);
4399
  *pos='\n';
4400
  if (my_b_write(file, (uchar*) buff, (size_t) (pos-buff)+1))
4401 4402 4403
    error=1;
  if (flush_io_cache(file))
    error=1;
4404 4405 4406 4407 4408 4409 4410 4411
  if (sync_relayloginfo_period &&
      !error &&
      ++(rli->sync_counter) >= sync_relayloginfo_period)
  {
    if (my_sync(rli->info_fd, MYF(MY_WME)))
      error=1;
    rli->sync_counter= 0;
  }
4412 4413 4414 4415
  /* 
    Flushing the relay log is done by the slave I/O thread 
    or by the user on STOP SLAVE. 
   */
4416
  DBUG_RETURN(error);
4417 4418
}

4419

unknown's avatar
unknown committed
4420
/*
4421
  Called when we notice that the current "hot" log got rotated under our feet.
unknown's avatar
unknown committed
4422 4423
*/

4424
static IO_CACHE *reopen_relay_log(Relay_log_info *rli, const char **errmsg)
4425
{
4426
  DBUG_ENTER("reopen_relay_log");
4427 4428
  DBUG_ASSERT(rli->cur_log != &rli->cache_buf);
  DBUG_ASSERT(rli->cur_log_fd == -1);
unknown's avatar
unknown committed
4429 4430

  IO_CACHE *cur_log = rli->cur_log=&rli->cache_buf;
4431
  if ((rli->cur_log_fd=open_binlog(cur_log,rli->event_relay_log_name,
unknown's avatar
unknown committed
4432
                                   errmsg)) <0)
unknown's avatar
unknown committed
4433
    DBUG_RETURN(0);
4434 4435
  /*
    We want to start exactly where we was before:
unknown's avatar
unknown committed
4436 4437
    relay_log_pos       Current log pos
    pending             Number of bytes already processed from the event
4438
  */
4439
  rli->event_relay_log_pos= max(rli->event_relay_log_pos, BIN_LOG_HEADER_SIZE);
4440
  my_b_seek(cur_log,rli->event_relay_log_pos);
unknown's avatar
unknown committed
4441
  DBUG_RETURN(cur_log);
4442 4443
}

unknown's avatar
unknown committed
4444

4445 4446 4447 4448 4449 4450 4451 4452 4453 4454
/**
  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.
*/
4455
static Log_event* next_event(Relay_log_info* rli)
4456 4457 4458
{
  Log_event* ev;
  IO_CACHE* cur_log = rli->cur_log;
Marc Alff's avatar
Marc Alff committed
4459
  mysql_mutex_t *log_lock = rli->relay_log.get_log_lock();
4460 4461
  const char* errmsg=0;
  THD* thd = rli->sql_thd;
unknown's avatar
unknown committed
4462
  DBUG_ENTER("next_event");
4463

4464 4465
  DBUG_ASSERT(thd != 0);

4466 4467 4468 4469 4470
#ifndef DBUG_OFF
  if (abort_slave_event_count && !rli->events_till_abort--)
    DBUG_RETURN(0);
#endif

unknown's avatar
unknown committed
4471 4472
  /*
    For most operations we need to protect rli members with data_lock,
4473 4474 4475
    so we assume calling function acquired this mutex for us and we will
    hold it for the most of the loop below However, we will release it
    whenever it is worth the hassle,  and in the cases when we go into a
Marc Alff's avatar
Marc Alff committed
4476
    mysql_cond_wait() with the non-data_lock mutex
unknown's avatar
unknown committed
4477
  */
Marc Alff's avatar
Marc Alff committed
4478
  mysql_mutex_assert_owner(&rli->data_lock);
unknown's avatar
unknown committed
4479

4480
  while (!sql_slave_killed(thd,rli))
unknown's avatar
unknown committed
4481 4482 4483
  {
    /*
      We can have two kinds of log reading:
unknown's avatar
unknown committed
4484 4485 4486 4487 4488 4489 4490 4491
      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
4492
    */
4493 4494 4495 4496
    bool hot_log;
    if ((hot_log = (cur_log != &rli->cache_buf)))
    {
      DBUG_ASSERT(rli->cur_log_fd == -1); // foreign descriptor
Marc Alff's avatar
Marc Alff committed
4497
      mysql_mutex_lock(log_lock);
unknown's avatar
unknown committed
4498 4499

      /*
unknown's avatar
unknown committed
4500 4501
        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
4502
      */
unknown's avatar
unknown committed
4503
      if (rli->relay_log.get_open_count() != rli->cur_log_old_open_count)
4504
      {
unknown's avatar
unknown committed
4505 4506
        // The master has switched to a new log file; Reopen the old log file
        cur_log=reopen_relay_log(rli, &errmsg);
Marc Alff's avatar
Marc Alff committed
4507
        mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
4508 4509 4510
        if (!cur_log)                           // No more log files
          goto err;
        hot_log=0;                              // Using old binary log
4511 4512
      }
    }
4513 4514 4515 4516 4517 4518 4519
    /* 
      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;
4520 4521
#ifndef DBUG_OFF
    {
4522
      /* This is an assertion which sometimes fails, let's try to track it */
4523
      char llbuf1[22], llbuf2[22];
4524
      DBUG_PRINT("info", ("my_b_tell(cur_log)=%s rli->event_relay_log_pos=%s",
4525
                          llstr(my_b_tell(cur_log),llbuf1),
4526 4527 4528
                          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);
4529 4530
    }
#endif
unknown's avatar
unknown committed
4531 4532
    /*
      Relay log is always in new format - if the master is 3.23, the
4533
      I/O thread will convert the format for us.
4534 4535
      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
4536 4537
      logs, which may even have been deleted. So we need to write this
      description event at the beginning of the relay log.
4538 4539
      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.
4540
      But if the relay log is created by new_file(): then the solution is:
4541
      MYSQL_BIN_LOG::open() will write the buffered description event.
unknown's avatar
unknown committed
4542
    */
4543 4544
    if ((ev=Log_event::read_log_event(cur_log,0,
                                      rli->relay_log.description_event_for_exec)))
4545

4546 4547
    {
      DBUG_ASSERT(thd==rli->sql_thd);
4548 4549 4550 4551 4552
      /*
        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);
4553
      if (hot_log)
Marc Alff's avatar
Marc Alff committed
4554
        mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
4555
      DBUG_RETURN(ev);
4556 4557
    }
    DBUG_ASSERT(thd==rli->sql_thd);
unknown's avatar
unknown committed
4558
    if (opt_reckless_slave)                     // For mysql-test
unknown's avatar
unknown committed
4559
      cur_log->error = 0;
unknown's avatar
unknown committed
4560
    if (cur_log->error < 0)
unknown's avatar
unknown committed
4561 4562
    {
      errmsg = "slave SQL thread aborted because of I/O error";
unknown's avatar
unknown committed
4563
      if (hot_log)
Marc Alff's avatar
Marc Alff committed
4564
        mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
4565 4566
      goto err;
    }
4567 4568
    if (!cur_log->error) /* EOF */
    {
unknown's avatar
unknown committed
4569
      /*
unknown's avatar
unknown committed
4570 4571 4572
        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
4573
      */
4574 4575
      if (hot_log)
      {
4576 4577 4578 4579 4580 4581 4582 4583 4584 4585
        /*
          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").
4586 4587 4588 4589 4590 4591 4592 4593 4594 4595

          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.
4596
        */
4597
        time_t save_timestamp= rli->last_master_timestamp;
4598 4599
        rli->last_master_timestamp= 0;

unknown's avatar
unknown committed
4600
        DBUG_ASSERT(rli->relay_log.get_open_count() ==
unknown's avatar
unknown committed
4601
                    rli->cur_log_old_open_count);
4602 4603 4604 4605 4606

        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"));
4607
          ev= new Rotate_log_event(rli->ign_master_log_name_end,
4608
                                   0, rli->ign_master_log_pos_end,
unknown's avatar
unknown committed
4609
                                   Rotate_log_event::DUP_NAME);
4610
          rli->ign_master_log_name_end[0]= 0;
Marc Alff's avatar
Marc Alff committed
4611
          mysql_mutex_unlock(log_lock);
4612 4613 4614 4615 4616 4617 4618 4619 4620 4621
          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
4622 4623 4624 4625
        /*
          We can, and should release data_lock while we are waiting for
          update. If we do not, show slave status will block
        */
Marc Alff's avatar
Marc Alff committed
4626
        mysql_mutex_unlock(&rli->data_lock);
4627 4628

        /*
unknown's avatar
unknown committed
4629
          Possible deadlock :
4630 4631 4632 4633 4634 4635 4636 4637 4638 4639 4640
          - 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
4641
          master does not send anything). Then the I/O thread stops waiting
4642 4643
          and reads more events.
          The SQL thread decides when the I/O thread should take log_space_limit
unknown's avatar
unknown committed
4644
          into account again : ignore_log_space_limit is reset to 0
4645 4646 4647 4648
          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
4649 4650
          be stopped, and the SQL thread sets ignore_log_space_limit to 0 when
          it stops.
4651
        */
Marc Alff's avatar
Marc Alff committed
4652
        mysql_mutex_lock(&rli->log_space_lock);
4653
        // prevent the I/O thread from blocking next times
unknown's avatar
unknown committed
4654
        rli->ignore_log_space_limit= 1;
4655
        /*
4656 4657 4658 4659
          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.
4660
        */
Marc Alff's avatar
Marc Alff committed
4661 4662
        mysql_mutex_unlock(&rli->log_space_lock);
        mysql_cond_broadcast(&rli->log_space_cond);
Andrei Elkin's avatar
Andrei Elkin committed
4663 4664
        // Note that wait_for_update_relay_log unlocks lock_log !
        rli->relay_log.wait_for_update_relay_log(rli->sql_thd);
4665
        // re-acquire data lock since we released it earlier
Marc Alff's avatar
Marc Alff committed
4666
        mysql_mutex_lock(&rli->data_lock);
4667
        rli->last_master_timestamp= save_timestamp;
unknown's avatar
unknown committed
4668
        continue;
4669
      }
unknown's avatar
unknown committed
4670
      /*
unknown's avatar
unknown committed
4671 4672 4673
        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
4674 4675 4676
      */
      end_io_cache(cur_log);
      DBUG_ASSERT(rli->cur_log_fd >= 0);
Marc Alff's avatar
Marc Alff committed
4677
      mysql_file_close(rli->cur_log_fd, MYF(MY_WME));
unknown's avatar
unknown committed
4678
      rli->cur_log_fd = -1;
unknown's avatar
unknown committed
4679

4680
      if (relay_log_purge)
unknown's avatar
unknown committed
4681
      {
unknown's avatar
unknown committed
4682
        /*
4683 4684 4685 4686 4687 4688 4689 4690 4691 4692
          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
4693
        if (rli->relay_log.purge_first_log
4694 4695 4696
            (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
4697 4698 4699 4700
        {
          errmsg = "Error purging processed logs";
          goto err;
        }
unknown's avatar
unknown committed
4701
      }
4702 4703
      else
      {
unknown's avatar
unknown committed
4704 4705 4706 4707 4708 4709 4710 4711 4712 4713 4714 4715 4716 4717 4718 4719
        /*
          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
4720
      }
4721 4722 4723 4724 4725 4726 4727 4728 4729 4730 4731 4732 4733

      /*
        Now we want to open this next log. To know if it's a hot log (the one
        being written by the I/O thread now) or a cold log, we can use
        is_active(); if it is hot, we use the I/O cache; if it's cold we open
        the file normally. But if is_active() reports that the log is hot, this
        may change between the test and the consequence of the test. So we may
        open the I/O cache whereas the log is now cold, which is nonsense.
        To guard against this, we need to have LOCK_log.
      */

      DBUG_PRINT("info",("hot_log: %d",hot_log));
      if (!hot_log) /* if hot_log, we already have this mutex */
Marc Alff's avatar
Marc Alff committed
4734
        mysql_mutex_lock(log_lock);
unknown's avatar
unknown committed
4735 4736
      if (rli->relay_log.is_active(rli->linfo.log_file_name))
      {
4737
#ifdef EXTRA_DEBUG
unknown's avatar
unknown committed
4738 4739
        if (global_system_variables.log_warnings)
          sql_print_information("next log '%s' is currently active",
4740
                                rli->linfo.log_file_name);
unknown's avatar
unknown committed
4741 4742 4743 4744 4745 4746
#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);

        /*
4747 4748 4749 4750 4751 4752 4753 4754 4755 4756 4757 4758 4759 4760 4761 4762 4763 4764 4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775 4776 4777 4778 4779 4780 4781 4782 4783 4784 4785 4786 4787 4788 4789 4790 4791 4792 4793 4794 4795 4796 4797 4798 4799 4800 4801 4802 4803
           When the SQL thread is [stopped and] (re)started the
           following may happen:

           1. Log was hot at stop time and remains hot at restart

              SQL thread reads again from hot_log (SQL thread was
              reading from the active log when it was stopped and the
              very same log is still active on SQL thread restart).

              In this case, my_b_seek is performed on cur_log, while
              cur_log points to relay_log.get_log_file();

           2. Log was hot at stop time but got cold before restart

              The log was hot when SQL thread stopped, but it is not
              anymore when the SQL thread restarts.

              In this case, the SQL thread reopens the log, using
              cache_buf, ie, cur_log points to &cache_buf, and thence
              its coordinates are reset.

           3. Log was already cold at stop time

              The log was not hot when the SQL thread stopped, and, of
              course, it will not be hot when it restarts.

              In this case, the SQL thread opens the cold log again,
              using cache_buf, ie, cur_log points to &cache_buf, and
              thence its coordinates are reset.

           4. Log was hot at stop time, DBA changes to previous cold
              log and restarts SQL thread

              The log was hot when the SQL thread was stopped, but the
              user changed the coordinates of the SQL thread to
              restart from a previous cold log.

              In this case, at start time, cur_log points to a cold
              log, opened using &cache_buf as cache, and coordinates
              are reset. However, as it moves on to the next logs, it
              will eventually reach the hot log. If the hot log is the
              same at the time the SQL thread was stopped, then
              coordinates were not reset - the cur_log will point to
              relay_log.get_log_file(), and not a freshly opened
              IO_CACHE through cache_buf. For this reason we need to
              deploy a my_b_seek before calling check_binlog_magic at
              this point of the code (see: BUG#55263 for more
              details).
          
          NOTES: 
            - We must keep the LOCK_log to read the 4 first bytes, as
              this is a hot log (same as when we call read_log_event()
              above: for a hot log we take the mutex).

            - Because of scenario #4 above, we need to have a
              my_b_seek here. Otherwise, we might hit the assertion
              inside check_binlog_magic.
unknown's avatar
unknown committed
4804
        */
4805 4806

        my_b_seek(cur_log, (my_off_t) 0);
unknown's avatar
unknown committed
4807
        if (check_binlog_magic(cur_log,&errmsg))
4808
        {
Marc Alff's avatar
Marc Alff committed
4809 4810
          if (!hot_log)
            mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
4811
          goto err;
4812
        }
Marc Alff's avatar
Marc Alff committed
4813 4814
        if (!hot_log)
          mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
4815
        continue;
4816
      }
Marc Alff's avatar
Marc Alff committed
4817 4818
      if (!hot_log)
        mysql_mutex_unlock(log_lock);
unknown's avatar
unknown committed
4819
      /*
unknown's avatar
unknown committed
4820 4821 4822
        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
4823 4824
      */
#ifdef EXTRA_DEBUG
unknown's avatar
unknown committed
4825
      if (global_system_variables.log_warnings)
unknown's avatar
unknown committed
4826
        sql_print_information("next log '%s' is not active",
4827
                              rli->linfo.log_file_name);
unknown's avatar
unknown committed
4828
#endif
unknown's avatar
unknown committed
4829 4830
      // 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
4831 4832
                                       &errmsg)) <0)
        goto err;
4833
    }
unknown's avatar
unknown committed
4834
    else
4835
    {
unknown's avatar
unknown committed
4836
      /*
unknown's avatar
unknown committed
4837 4838
        Read failed with a non-EOF error.
        TODO: come up with something better to handle this error
unknown's avatar
unknown committed
4839 4840
      */
      if (hot_log)
Marc Alff's avatar
Marc Alff committed
4841
        mysql_mutex_unlock(log_lock);
4842
      sql_print_error("Slave SQL thread: I/O error reading \
unknown's avatar
unknown committed
4843
event(errno: %d  cur_log->error: %d)",
unknown's avatar
unknown committed
4844
                      my_errno,cur_log->error);
unknown's avatar
unknown committed
4845
      // set read position to the beginning of the event
4846
      my_b_seek(cur_log,rli->event_relay_log_pos);
unknown's avatar
unknown committed
4847 4848
      /* otherwise, we have had a partial read */
      errmsg = "Aborting slave SQL thread because of partial event read";
unknown's avatar
unknown committed
4849
      break;                                    // To end of function
4850 4851
    }
  }
4852
  if (!errmsg && global_system_variables.log_warnings)
4853
  {
unknown's avatar
unknown committed
4854
    sql_print_information("Error reading relay log event: %s",
4855 4856 4857
                          "slave SQL thread was killed");
    DBUG_RETURN(0);
  }
unknown's avatar
unknown committed
4858

4859
err:
4860 4861
  if (errmsg)
    sql_print_error("Error reading relay log event: %s", errmsg);
unknown's avatar
unknown committed
4862
  DBUG_RETURN(0);
4863 4864
}

4865 4866 4867
/*
  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
4868
  locks; here we don't, so this function is mainly taking locks).
4869 4870
  Returns nothing as we cannot catch any error (MYSQL_BIN_LOG::new_file()
  is void).
4871 4872
*/

4873
void rotate_relay_log(Master_info* mi)
4874 4875
{
  DBUG_ENTER("rotate_relay_log");
4876
  Relay_log_info* rli= &mi->rli;
unknown's avatar
unknown committed
4877

4878 4879
  DBUG_EXECUTE_IF("crash_before_rotate_relaylog", abort(););

unknown's avatar
unknown committed
4880
  /*
unknown's avatar
unknown committed
4881 4882 4883
     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).
  */
4884 4885
  if (!rli->inited)
  {
unknown's avatar
unknown committed
4886
    DBUG_PRINT("info", ("rli->inited == 0"));
unknown's avatar
unknown committed
4887
    goto end;
4888
  }
unknown's avatar
unknown committed
4889

4890
  /* If the relay log is closed, new_file() will do nothing. */
4891
  rli->relay_log.new_file();
unknown's avatar
unknown committed
4892

4893 4894 4895 4896
  /*
    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
4897 4898 4899 4900 4901 4902
    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.
4903 4904 4905 4906
    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
4907
end:
4908 4909 4910
  DBUG_VOID_RETURN;
}

unknown's avatar
unknown committed
4911

unknown's avatar
unknown committed
4912 4913 4914
/**
   Detects, based on master's version (as found in the relay log), if master
   has a certain bug.
4915
   @param rli Relay_log_info which tells the master's version
unknown's avatar
unknown committed
4916
   @param bug_id Number of the bug as found in bugs.mysql.com
4917
   @param report bool report error message, default TRUE
4918 4919 4920 4921 4922 4923 4924

   @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
4925 4926
   @return TRUE if master has the bug, FALSE if it does not.
*/
4927 4928
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
4929 4930 4931 4932 4933 4934 4935 4936
{
  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
4937
    {24432, { 5, 0, 24 }, { 5, 0, 38 } },
4938 4939 4940
    {24432, { 5, 1, 12 }, { 5, 1, 17 } },
    {33029, { 5, 0,  0 }, { 5, 0, 58 } },
    {33029, { 5, 1,  0 }, { 5, 1, 12 } },
4941
    {37426, { 5, 1,  0 }, { 5, 1, 26 } },
unknown's avatar
unknown committed
4942 4943 4944 4945 4946 4947 4948 4949 4950 4951 4952 4953 4954
  };
  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) &&
4955 4956
        (memcmp(fixed_in,      master_ver, 3) >  0) &&
        (pred == NULL || (*pred)(param)))
unknown's avatar
unknown committed
4957
    {
4958 4959
      if (!report)
	return TRUE;
unknown's avatar
unknown committed
4960 4961 4962 4963 4964
      // 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);
4965
      // a verbose message for the error log
4966 4967 4968
      rli->report(ERROR_LEVEL, ER_UNKNOWN_ERROR,
                  "According to the master's version ('%s'),"
                  " it is probable that master suffers from this bug:"
4969 4970 4971 4972 4973 4974 4975 4976 4977 4978 4979 4980 4981 4982 4983 4984
                      " 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
4985 4986 4987 4988 4989 4990
      return TRUE;
    }
  }
  return FALSE;
}

4991 4992 4993 4994 4995 4996 4997 4998 4999 5000 5001 5002 5003 5004 5005
/**
   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;
5006
    DBUG_EXECUTE_IF("simulate_bug33029", return TRUE;);
5007
    return rpl_master_has_bug(rli, 33029, FALSE, NULL, NULL);
5008 5009 5010 5011
  }
  return FALSE;
}

5012
#ifdef HAVE_EXPLICIT_TEMPLATE_INSTANTIATION
unknown's avatar
unknown committed
5013
template class I_List_iterator<i_string>;
unknown's avatar
unknown committed
5014
template class I_List_iterator<i_string_pair>;
unknown's avatar
unknown committed
5015
#endif
5016

5017 5018 5019 5020
/**
  @} (end of group Replication)
*/

unknown's avatar
SCRUM  
unknown committed
5021
#endif /* HAVE_REPLICATION */