1 | /* Copyright (c) 2000, 2017, Oracle and/or its affiliates. |
2 | Copyright (c) 2009, 2017, MariaDB Corporation |
3 | |
4 | This program is free software; you can redistribute it and/or modify |
5 | it under the terms of the GNU General Public License as published by |
6 | the Free Software Foundation; version 2 of the License. |
7 | |
8 | This program is distributed in the hope that it will be useful, |
9 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
10 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
11 | GNU General Public License for more details. |
12 | |
13 | You should have received a copy of the GNU General Public License |
14 | along with this program; if not, write to the Free Software |
15 | Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA */ |
16 | |
17 | |
18 | /** |
19 | @addtogroup Replication |
20 | @{ |
21 | |
22 | @file |
23 | |
24 | @brief Code to run the io thread and the sql thread on the |
25 | replication slave. |
26 | */ |
27 | |
28 | #include "mariadb.h" |
29 | #include "sql_priv.h" |
30 | #include "slave.h" |
31 | #include "sql_parse.h" // execute_init_command |
32 | #include "sql_table.h" // mysql_rm_table |
33 | #include "rpl_mi.h" |
34 | #include "rpl_rli.h" |
35 | #include "sql_repl.h" |
36 | #include "rpl_filter.h" |
37 | #include "repl_failsafe.h" |
38 | #include "transaction.h" |
39 | #include <thr_alarm.h> |
40 | #include <my_dir.h> |
41 | #include <sql_common.h> |
42 | #include <errmsg.h> |
43 | #include <ssl_compat.h> |
44 | #include "unireg.h" |
45 | #include <mysys_err.h> |
46 | #include <signal.h> |
47 | #include <mysql.h> |
48 | #include <myisam.h> |
49 | |
50 | #include "sql_base.h" // close_thread_tables |
51 | #include "tztime.h" // struct Time_zone |
52 | #include "log_event.h" // Rotate_log_event, |
53 | // Create_file_log_event, |
54 | // Format_description_log_event |
55 | #include "wsrep_mysqld.h" |
56 | |
57 | #ifdef HAVE_REPLICATION |
58 | |
59 | #include "rpl_tblmap.h" |
60 | #include "debug_sync.h" |
61 | #include "rpl_parallel.h" |
62 | #include "sql_show.h" |
63 | #include "semisync_slave.h" |
64 | |
65 | #define FLAGSTR(V,F) ((V)&(F)?#F" ":"") |
66 | |
67 | #define MAX_SLAVE_RETRY_PAUSE 5 |
68 | /* |
69 | a parameter of sql_slave_killed() to defer the killed status |
70 | */ |
71 | #define SLAVE_WAIT_GROUP_DONE 60 |
72 | bool use_slave_mask = 0; |
73 | MY_BITMAP slave_error_mask; |
74 | char slave_skip_error_names[SHOW_VAR_FUNC_BUFF_SIZE]; |
75 | uint *slave_transaction_retry_errors; |
76 | uint slave_transaction_retry_error_length= 0; |
77 | char slave_transaction_retry_error_names[SHOW_VAR_FUNC_BUFF_SIZE]; |
78 | |
79 | char* slave_load_tmpdir = 0; |
80 | Master_info *active_mi= 0; |
81 | Master_info_index *master_info_index; |
82 | my_bool replicate_same_server_id; |
83 | ulonglong relay_log_space_limit = 0; |
84 | ulonglong opt_read_binlog_speed_limit = 0; |
85 | |
86 | const char *relay_log_index= 0; |
87 | const char *relay_log_basename= 0; |
88 | |
89 | LEX_CSTRING default_master_connection_name= { (char*) "" , 0 }; |
90 | |
91 | /* |
92 | When slave thread exits, we need to remember the temporary tables so we |
93 | can re-use them on slave start. |
94 | |
95 | TODO: move the vars below under Master_info |
96 | */ |
97 | |
98 | int disconnect_slave_event_count = 0, abort_slave_event_count = 0; |
99 | |
100 | static pthread_key(Master_info*, RPL_MASTER_INFO); |
101 | |
102 | enum enum_slave_reconnect_actions |
103 | { |
104 | SLAVE_RECON_ACT_REG= 0, |
105 | SLAVE_RECON_ACT_DUMP= 1, |
106 | SLAVE_RECON_ACT_EVENT= 2, |
107 | SLAVE_RECON_ACT_MAX |
108 | }; |
109 | |
110 | enum enum_slave_reconnect_messages |
111 | { |
112 | SLAVE_RECON_MSG_WAIT= 0, |
113 | SLAVE_RECON_MSG_KILLED_WAITING= 1, |
114 | SLAVE_RECON_MSG_AFTER= 2, |
115 | SLAVE_RECON_MSG_FAILED= 3, |
116 | SLAVE_RECON_MSG_COMMAND= 4, |
117 | SLAVE_RECON_MSG_KILLED_AFTER= 5, |
118 | SLAVE_RECON_MSG_MAX |
119 | }; |
120 | |
121 | static const char *reconnect_messages[SLAVE_RECON_ACT_MAX][SLAVE_RECON_MSG_MAX]= |
122 | { |
123 | { |
124 | "Waiting to reconnect after a failed registration on master" , |
125 | "Slave I/O thread killed while waiting to reconnect after a failed \ |
126 | registration on master" , |
127 | "Reconnecting after a failed registration on master" , |
128 | "failed registering on master, reconnecting to try again, \ |
129 | log '%s' at position %llu%s" , |
130 | "COM_REGISTER_SLAVE" , |
131 | "Slave I/O thread killed during or after reconnect" |
132 | }, |
133 | { |
134 | "Waiting to reconnect after a failed binlog dump request" , |
135 | "Slave I/O thread killed while retrying master dump" , |
136 | "Reconnecting after a failed binlog dump request" , |
137 | "failed dump request, reconnecting to try again, log '%s' at position %llu%s" , |
138 | "COM_BINLOG_DUMP" , |
139 | "Slave I/O thread killed during or after reconnect" |
140 | }, |
141 | { |
142 | "Waiting to reconnect after a failed master event read" , |
143 | "Slave I/O thread killed while waiting to reconnect after a failed read" , |
144 | "Reconnecting after a failed master event read" , |
145 | "Slave I/O thread: Failed reading log event, reconnecting to retry, \ |
146 | log '%s' at position %llu%s" , |
147 | "" , |
148 | "Slave I/O thread killed during or after a reconnect done to recover from \ |
149 | failed read" |
150 | } |
151 | }; |
152 | |
153 | |
154 | typedef enum { SLAVE_THD_IO, SLAVE_THD_SQL} SLAVE_THD_TYPE; |
155 | |
156 | static int process_io_rotate(Master_info* mi, Rotate_log_event* rev); |
157 | static int process_io_create_file(Master_info* mi, Create_file_log_event* cev); |
158 | static bool wait_for_relay_log_space(Relay_log_info* rli); |
159 | static bool io_slave_killed(Master_info* mi); |
160 | static bool sql_slave_killed(rpl_group_info *rgi); |
161 | static int init_slave_thread(THD*, Master_info *, SLAVE_THD_TYPE); |
162 | static void make_slave_skip_errors_printable(void); |
163 | static void make_slave_transaction_retry_errors_printable(void); |
164 | static int safe_connect(THD* thd, MYSQL* mysql, Master_info* mi); |
165 | static int safe_reconnect(THD*, MYSQL*, Master_info*, bool); |
166 | static int connect_to_master(THD*, MYSQL*, Master_info*, bool, bool); |
167 | static Log_event* next_event(rpl_group_info* rgi, ulonglong *event_size); |
168 | static int queue_event(Master_info* mi,const char* buf,ulong event_len); |
169 | static int terminate_slave_thread(THD *, mysql_mutex_t *, mysql_cond_t *, |
170 | volatile uint *, bool); |
171 | static bool check_io_slave_killed(Master_info *mi, const char *info); |
172 | static bool send_show_master_info_data(THD *, Master_info *, bool, String *); |
173 | /* |
174 | Function to set the slave's max_allowed_packet based on the value |
175 | of slave_max_allowed_packet. |
176 | |
177 | @in_param thd Thread handler for slave |
178 | @in_param mysql MySQL connection handle |
179 | */ |
180 | |
181 | static void set_slave_max_allowed_packet(THD *thd, MYSQL *mysql) |
182 | { |
183 | DBUG_ENTER("set_slave_max_allowed_packet" ); |
184 | // thd and mysql must be valid |
185 | DBUG_ASSERT(thd && mysql); |
186 | |
187 | thd->variables.max_allowed_packet= slave_max_allowed_packet; |
188 | thd->net.max_packet_size= slave_max_allowed_packet; |
189 | /* |
190 | Adding MAX_LOG_EVENT_HEADER_LEN to the max_packet_size on the I/O |
191 | thread and the mysql->option max_allowed_packet, since a |
192 | replication event can become this much larger than |
193 | the corresponding packet (query) sent from client to master. |
194 | */ |
195 | thd->net.max_packet_size+= MAX_LOG_EVENT_HEADER; |
196 | /* |
197 | Skipping the setting of mysql->net.max_packet size to slave |
198 | max_allowed_packet since this is done during mysql_real_connect. |
199 | */ |
200 | mysql->options.max_allowed_packet= |
201 | slave_max_allowed_packet+MAX_LOG_EVENT_HEADER; |
202 | DBUG_VOID_RETURN; |
203 | } |
204 | |
205 | /* |
206 | Find out which replications threads are running |
207 | |
208 | SYNOPSIS |
209 | init_thread_mask() |
210 | mask Return value here |
211 | mi master_info for slave |
212 | inverse If set, returns which threads are not running |
213 | |
214 | IMPLEMENTATION |
215 | Get a bit mask for which threads are running so that we can later restart |
216 | these threads. |
217 | |
218 | RETURN |
219 | mask If inverse == 0, running threads |
220 | If inverse == 1, stopped threads |
221 | */ |
222 | |
223 | void init_thread_mask(int* mask,Master_info* mi,bool inverse) |
224 | { |
225 | bool set_io = mi->slave_running, set_sql = mi->rli.slave_running; |
226 | int tmp_mask=0; |
227 | DBUG_ENTER("init_thread_mask" ); |
228 | |
229 | if (set_io) |
230 | tmp_mask |= SLAVE_IO; |
231 | if (set_sql) |
232 | tmp_mask |= SLAVE_SQL; |
233 | if (inverse) |
234 | tmp_mask^= (SLAVE_IO | SLAVE_SQL); |
235 | *mask = tmp_mask; |
236 | DBUG_VOID_RETURN; |
237 | } |
238 | |
239 | |
240 | /* |
241 | lock_slave_threads() against other threads doing STOP, START or RESET SLAVE |
242 | |
243 | */ |
244 | |
245 | void Master_info::lock_slave_threads() |
246 | { |
247 | DBUG_ENTER("lock_slave_threads" ); |
248 | mysql_mutex_lock(&start_stop_lock); |
249 | DBUG_VOID_RETURN; |
250 | } |
251 | |
252 | |
253 | /* |
254 | unlock_slave_threads() |
255 | */ |
256 | |
257 | void Master_info::unlock_slave_threads() |
258 | { |
259 | DBUG_ENTER("unlock_slave_threads" ); |
260 | mysql_mutex_unlock(&start_stop_lock); |
261 | DBUG_VOID_RETURN; |
262 | } |
263 | |
264 | #ifdef HAVE_PSI_INTERFACE |
265 | static PSI_thread_key key_thread_slave_io, key_thread_slave_sql; |
266 | |
267 | static PSI_thread_info all_slave_threads[]= |
268 | { |
269 | { &key_thread_slave_io, "slave_io" , PSI_FLAG_GLOBAL}, |
270 | { &key_thread_slave_sql, "slave_sql" , PSI_FLAG_GLOBAL} |
271 | }; |
272 | |
273 | static void init_slave_psi_keys(void) |
274 | { |
275 | const char* category= "sql" ; |
276 | int count; |
277 | |
278 | if (PSI_server == NULL) |
279 | return; |
280 | |
281 | count= array_elements(all_slave_threads); |
282 | PSI_server->register_thread(category, all_slave_threads, count); |
283 | } |
284 | #endif /* HAVE_PSI_INTERFACE */ |
285 | |
286 | |
287 | /* |
288 | Note: This definition needs to be kept in sync with the one in |
289 | mysql_system_tables.sql which is used by mysql_create_db. |
290 | */ |
291 | static const char gtid_pos_table_definition1[]= |
292 | "CREATE TABLE " ; |
293 | static const char gtid_pos_table_definition2[]= |
294 | " (domain_id INT UNSIGNED NOT NULL, " |
295 | "sub_id BIGINT UNSIGNED NOT NULL, " |
296 | "server_id INT UNSIGNED NOT NULL, " |
297 | "seq_no BIGINT UNSIGNED NOT NULL, " |
298 | "PRIMARY KEY (domain_id, sub_id)) CHARSET=latin1 " |
299 | "COMMENT='Replication slave GTID position' " |
300 | "ENGINE=" ; |
301 | |
302 | /* |
303 | Build a query string |
304 | CREATE TABLE mysql.gtid_slave_pos_<engine> ... ENGINE=<engine> |
305 | */ |
306 | static bool |
307 | build_gtid_pos_create_query(THD *thd, String *query, |
308 | LEX_CSTRING *table_name, |
309 | LEX_CSTRING *engine_name) |
310 | { |
311 | bool err= false; |
312 | err|= query->append(gtid_pos_table_definition1); |
313 | err|= append_identifier(thd, query, table_name); |
314 | err|= query->append(gtid_pos_table_definition2); |
315 | err|= append_identifier(thd, query, engine_name); |
316 | return err; |
317 | } |
318 | |
319 | |
320 | static int |
321 | gtid_pos_table_creation(THD *thd, plugin_ref engine, LEX_CSTRING *table_name) |
322 | { |
323 | int err; |
324 | StringBuffer<sizeof(gtid_pos_table_definition1) + |
325 | sizeof(gtid_pos_table_definition1) + |
326 | 2*FN_REFLEN> query; |
327 | |
328 | if (build_gtid_pos_create_query(thd, &query, table_name, plugin_name(engine))) |
329 | { |
330 | my_error(ER_OUT_OF_RESOURCES, MYF(0)); |
331 | return 1; |
332 | } |
333 | |
334 | thd->set_db(&MYSQL_SCHEMA_NAME); |
335 | thd->clear_error(); |
336 | ulonglong thd_saved_option= thd->variables.option_bits; |
337 | /* This query shuold not be binlogged. */ |
338 | thd->variables.option_bits&= ~(ulonglong)OPTION_BIN_LOG; |
339 | thd->set_query_and_id(query.c_ptr(), query.length(), thd->charset(), |
340 | next_query_id()); |
341 | Parser_state parser_state; |
342 | err= parser_state.init(thd, thd->query(), thd->query_length()); |
343 | if (err) |
344 | goto end; |
345 | mysql_parse(thd, thd->query(), thd->query_length(), &parser_state, |
346 | FALSE, FALSE); |
347 | if (unlikely(thd->is_error())) |
348 | err= 1; |
349 | /* The warning is relevant to 10.3 and earlier. */ |
350 | sql_print_warning("The automatically created table '%s' name may not be " |
351 | "entirely in lowercase. The table name will be converted " |
352 | "to lowercase to any future upgrade to 10.4.0 and later " |
353 | "version where it will be auto-created at once " |
354 | "in lowercase." , |
355 | table_name->str); |
356 | end: |
357 | thd->variables.option_bits= thd_saved_option; |
358 | thd->reset_query(); |
359 | return err; |
360 | } |
361 | |
362 | |
363 | static void |
364 | handle_gtid_pos_auto_create_request(THD *thd, void *hton) |
365 | { |
366 | int UNINIT_VAR(err); |
367 | plugin_ref engine= NULL, *auto_engines; |
368 | rpl_slave_state::gtid_pos_table *entry; |
369 | StringBuffer<FN_REFLEN> loc_table_name; |
370 | LEX_CSTRING table_name; |
371 | |
372 | /* |
373 | Check that the plugin is still in @@gtid_pos_auto_engines, and lock |
374 | it. |
375 | */ |
376 | mysql_mutex_lock(&LOCK_global_system_variables); |
377 | engine= NULL; |
378 | for (auto_engines= opt_gtid_pos_auto_plugins; |
379 | auto_engines && *auto_engines; |
380 | ++auto_engines) |
381 | { |
382 | if (plugin_hton(*auto_engines) == hton) |
383 | { |
384 | engine= my_plugin_lock(NULL, *auto_engines); |
385 | break; |
386 | } |
387 | } |
388 | mysql_mutex_unlock(&LOCK_global_system_variables); |
389 | if (!engine) |
390 | { |
391 | /* The engine is gone from @@gtid_pos_auto_engines, so no action. */ |
392 | goto end; |
393 | } |
394 | |
395 | /* Find the entry for the table to auto-create. */ |
396 | mysql_mutex_lock(&rpl_global_gtid_slave_state->LOCK_slave_state); |
397 | entry= (rpl_slave_state::gtid_pos_table *) |
398 | rpl_global_gtid_slave_state->gtid_pos_tables; |
399 | while (entry) |
400 | { |
401 | if (entry->table_hton == hton && |
402 | entry->state == rpl_slave_state::GTID_POS_CREATE_REQUESTED) |
403 | break; |
404 | entry= entry->next; |
405 | } |
406 | if (entry) |
407 | { |
408 | entry->state = rpl_slave_state::GTID_POS_CREATE_IN_PROGRESS; |
409 | err= loc_table_name.append(entry->table_name.str, entry->table_name.length); |
410 | } |
411 | mysql_mutex_unlock(&rpl_global_gtid_slave_state->LOCK_slave_state); |
412 | if (!entry) |
413 | goto end; |
414 | if (err) |
415 | { |
416 | sql_print_error("Out of memory while trying to auto-create GTID position table" ); |
417 | goto end; |
418 | } |
419 | table_name.str= loc_table_name.c_ptr_safe(); |
420 | table_name.length= loc_table_name.length(); |
421 | |
422 | err= gtid_pos_table_creation(thd, engine, &table_name); |
423 | if (err) |
424 | { |
425 | sql_print_error("Error auto-creating GTID position table `mysql.%s`: %s Error_code: %d" , |
426 | table_name.str, thd->get_stmt_da()->message(), |
427 | thd->get_stmt_da()->sql_errno()); |
428 | thd->clear_error(); |
429 | goto end; |
430 | } |
431 | |
432 | /* Now enable the entry for the auto-created table. */ |
433 | mysql_mutex_lock(&rpl_global_gtid_slave_state->LOCK_slave_state); |
434 | entry= (rpl_slave_state::gtid_pos_table *) |
435 | rpl_global_gtid_slave_state->gtid_pos_tables; |
436 | while (entry) |
437 | { |
438 | if (entry->table_hton == hton && |
439 | entry->state == rpl_slave_state::GTID_POS_CREATE_IN_PROGRESS) |
440 | { |
441 | entry->state= rpl_slave_state::GTID_POS_AVAILABLE; |
442 | break; |
443 | } |
444 | entry= entry->next; |
445 | } |
446 | mysql_mutex_unlock(&rpl_global_gtid_slave_state->LOCK_slave_state); |
447 | |
448 | end: |
449 | if (engine) |
450 | plugin_unlock(NULL, engine); |
451 | } |
452 | |
453 | |
454 | static bool slave_background_thread_running; |
455 | static bool slave_background_thread_stop; |
456 | static bool slave_background_thread_gtid_loaded; |
457 | |
458 | static struct slave_background_kill_t { |
459 | slave_background_kill_t *next; |
460 | THD *to_kill; |
461 | } *slave_background_kill_list; |
462 | |
463 | static struct slave_background_gtid_pos_create_t { |
464 | slave_background_gtid_pos_create_t *next; |
465 | void *hton; |
466 | } *slave_background_gtid_pos_create_list; |
467 | |
468 | |
469 | pthread_handler_t |
470 | handle_slave_background(void *arg __attribute__((unused))) |
471 | { |
472 | THD *thd; |
473 | PSI_stage_info old_stage; |
474 | bool stop; |
475 | |
476 | my_thread_init(); |
477 | thd= new THD(next_thread_id()); |
478 | thd->thread_stack= (char*) &thd; /* Set approximate stack start */ |
479 | thd->system_thread = SYSTEM_THREAD_SLAVE_BACKGROUND; |
480 | thread_safe_increment32(&service_thread_count); |
481 | thd->store_globals(); |
482 | thd->security_ctx->skip_grants(); |
483 | thd->set_command(COM_DAEMON); |
484 | |
485 | thd_proc_info(thd, "Loading slave GTID position from table" ); |
486 | if (rpl_load_gtid_slave_state(thd)) |
487 | sql_print_warning("Failed to load slave replication state from table " |
488 | "%s.%s: %u: %s" , "mysql" , |
489 | rpl_gtid_slave_state_table_name.str, |
490 | thd->get_stmt_da()->sql_errno(), |
491 | thd->get_stmt_da()->message()); |
492 | |
493 | mysql_mutex_lock(&LOCK_slave_background); |
494 | slave_background_thread_gtid_loaded= true; |
495 | mysql_cond_broadcast(&COND_slave_background); |
496 | |
497 | THD_STAGE_INFO(thd, stage_slave_background_process_request); |
498 | do |
499 | { |
500 | slave_background_kill_t *kill_list; |
501 | slave_background_gtid_pos_create_t *create_list; |
502 | |
503 | thd->ENTER_COND(&COND_slave_background, &LOCK_slave_background, |
504 | &stage_slave_background_wait_request, |
505 | &old_stage); |
506 | for (;;) |
507 | { |
508 | stop= abort_loop || thd->killed || slave_background_thread_stop; |
509 | kill_list= slave_background_kill_list; |
510 | create_list= slave_background_gtid_pos_create_list; |
511 | if (stop || kill_list || create_list) |
512 | break; |
513 | mysql_cond_wait(&COND_slave_background, &LOCK_slave_background); |
514 | } |
515 | |
516 | slave_background_kill_list= NULL; |
517 | slave_background_gtid_pos_create_list= NULL; |
518 | thd->EXIT_COND(&old_stage); |
519 | |
520 | while (kill_list) |
521 | { |
522 | slave_background_kill_t *p = kill_list; |
523 | THD *to_kill= p->to_kill; |
524 | kill_list= p->next; |
525 | |
526 | to_kill->awake(KILL_CONNECTION); |
527 | mysql_mutex_lock(&to_kill->LOCK_wakeup_ready); |
528 | to_kill->rgi_slave->killed_for_retry= |
529 | rpl_group_info::RETRY_KILL_KILLED; |
530 | mysql_cond_broadcast(&to_kill->COND_wakeup_ready); |
531 | mysql_mutex_unlock(&to_kill->LOCK_wakeup_ready); |
532 | my_free(p); |
533 | } |
534 | |
535 | while (create_list) |
536 | { |
537 | slave_background_gtid_pos_create_t *next= create_list->next; |
538 | void *hton= create_list->hton; |
539 | handle_gtid_pos_auto_create_request(thd, hton); |
540 | my_free(create_list); |
541 | create_list= next; |
542 | } |
543 | |
544 | mysql_mutex_lock(&LOCK_slave_background); |
545 | } while (!stop); |
546 | |
547 | slave_background_thread_running= false; |
548 | mysql_cond_broadcast(&COND_slave_background); |
549 | mysql_mutex_unlock(&LOCK_slave_background); |
550 | |
551 | delete thd; |
552 | thread_safe_decrement32(&service_thread_count); |
553 | signal_thd_deleted(); |
554 | |
555 | my_thread_end(); |
556 | return 0; |
557 | } |
558 | |
559 | |
560 | |
561 | void |
562 | slave_background_kill_request(THD *to_kill) |
563 | { |
564 | if (to_kill->rgi_slave->killed_for_retry) |
565 | return; // Already deadlock killed. |
566 | slave_background_kill_t *p= |
567 | (slave_background_kill_t *)my_malloc(sizeof(*p), MYF(MY_WME)); |
568 | if (p) |
569 | { |
570 | p->to_kill= to_kill; |
571 | to_kill->rgi_slave->killed_for_retry= |
572 | rpl_group_info::RETRY_KILL_PENDING; |
573 | mysql_mutex_lock(&LOCK_slave_background); |
574 | p->next= slave_background_kill_list; |
575 | slave_background_kill_list= p; |
576 | mysql_cond_signal(&COND_slave_background); |
577 | mysql_mutex_unlock(&LOCK_slave_background); |
578 | } |
579 | } |
580 | |
581 | |
582 | /* |
583 | This function must only be called from a slave SQL thread (or worker thread), |
584 | to ensure that the table_entry will not go away before we can lock the |
585 | LOCK_slave_state. |
586 | */ |
587 | void |
588 | slave_background_gtid_pos_create_request( |
589 | rpl_slave_state::gtid_pos_table *table_entry) |
590 | { |
591 | slave_background_gtid_pos_create_t *p; |
592 | |
593 | if (table_entry->state != rpl_slave_state::GTID_POS_AUTO_CREATE) |
594 | return; |
595 | p= (slave_background_gtid_pos_create_t *)my_malloc(sizeof(*p), MYF(MY_WME)); |
596 | if (!p) |
597 | return; |
598 | mysql_mutex_lock(&rpl_global_gtid_slave_state->LOCK_slave_state); |
599 | if (table_entry->state != rpl_slave_state::GTID_POS_AUTO_CREATE) |
600 | { |
601 | my_free(p); |
602 | mysql_mutex_unlock(&rpl_global_gtid_slave_state->LOCK_slave_state); |
603 | return; |
604 | } |
605 | table_entry->state= rpl_slave_state::GTID_POS_CREATE_REQUESTED; |
606 | mysql_mutex_unlock(&rpl_global_gtid_slave_state->LOCK_slave_state); |
607 | |
608 | p->hton= table_entry->table_hton; |
609 | mysql_mutex_lock(&LOCK_slave_background); |
610 | p->next= slave_background_gtid_pos_create_list; |
611 | slave_background_gtid_pos_create_list= p; |
612 | mysql_cond_signal(&COND_slave_background); |
613 | mysql_mutex_unlock(&LOCK_slave_background); |
614 | } |
615 | |
616 | |
617 | /* |
618 | Start the slave background thread. |
619 | |
620 | This thread is currently used for two purposes: |
621 | |
622 | 1. To load the GTID state from mysql.gtid_slave_pos at server start; reading |
623 | from table requires valid THD, which is otherwise not available during |
624 | server init. |
625 | |
626 | 2. To kill worker thread transactions during parallel replication, when a |
627 | storage engine attempts to take an errorneous conflicting lock that would |
628 | cause a deadlock. Killing is done asynchroneously, as the kill may not |
629 | be safe within the context of a callback from inside storage engine |
630 | locking code. |
631 | */ |
632 | static int |
633 | start_slave_background_thread() |
634 | { |
635 | pthread_t th; |
636 | |
637 | slave_background_thread_running= true; |
638 | slave_background_thread_stop= false; |
639 | slave_background_thread_gtid_loaded= false; |
640 | if (mysql_thread_create(key_thread_slave_background, |
641 | &th, &connection_attrib, handle_slave_background, |
642 | NULL)) |
643 | { |
644 | sql_print_error("Failed to create thread while initialising slave" ); |
645 | return 1; |
646 | } |
647 | mysql_mutex_lock(&LOCK_slave_background); |
648 | while (!slave_background_thread_gtid_loaded) |
649 | mysql_cond_wait(&COND_slave_background, &LOCK_slave_background); |
650 | mysql_mutex_unlock(&LOCK_slave_background); |
651 | |
652 | return 0; |
653 | } |
654 | |
655 | |
656 | static void |
657 | stop_slave_background_thread() |
658 | { |
659 | mysql_mutex_lock(&LOCK_slave_background); |
660 | slave_background_thread_stop= true; |
661 | mysql_cond_broadcast(&COND_slave_background); |
662 | while (slave_background_thread_running) |
663 | mysql_cond_wait(&COND_slave_background, &LOCK_slave_background); |
664 | mysql_mutex_unlock(&LOCK_slave_background); |
665 | } |
666 | |
667 | |
668 | /* Initialize slave structures */ |
669 | |
670 | int init_slave() |
671 | { |
672 | DBUG_ENTER("init_slave" ); |
673 | int error= 0; |
674 | |
675 | #ifdef HAVE_PSI_INTERFACE |
676 | init_slave_psi_keys(); |
677 | #endif |
678 | |
679 | if (start_slave_background_thread()) |
680 | return 1; |
681 | |
682 | if (global_rpl_thread_pool.init(opt_slave_parallel_threads)) |
683 | return 1; |
684 | |
685 | /* |
686 | This is called when mysqld starts. Before client connections are |
687 | accepted. However bootstrap may conflict with us if it does START SLAVE. |
688 | So it's safer to take the lock. |
689 | */ |
690 | |
691 | if (pthread_key_create(&RPL_MASTER_INFO, NULL)) |
692 | goto err; |
693 | |
694 | master_info_index= new Master_info_index; |
695 | if (!master_info_index || master_info_index->init_all_master_info()) |
696 | { |
697 | sql_print_error("Failed to initialize multi master structures" ); |
698 | DBUG_RETURN(1); |
699 | } |
700 | if (!(active_mi= new Master_info(&default_master_connection_name, |
701 | relay_log_recovery)) || |
702 | active_mi->error()) |
703 | { |
704 | delete active_mi; |
705 | active_mi= 0; |
706 | sql_print_error("Failed to allocate memory for the Master Info structure" ); |
707 | goto err; |
708 | } |
709 | |
710 | if (master_info_index->add_master_info(active_mi, FALSE)) |
711 | { |
712 | delete active_mi; |
713 | active_mi= 0; |
714 | goto err; |
715 | } |
716 | |
717 | /* |
718 | If master_host is not specified, try to read it from the master_info file. |
719 | If master_host is specified, create the master_info file if it doesn't |
720 | exists. |
721 | */ |
722 | |
723 | if (init_master_info(active_mi,master_info_file,relay_log_info_file, |
724 | 1, (SLAVE_IO | SLAVE_SQL))) |
725 | { |
726 | sql_print_error("Failed to initialize the master info structure" ); |
727 | goto err; |
728 | } |
729 | |
730 | /* If server id is not set, start_slave_thread() will say it */ |
731 | |
732 | if (active_mi->host[0] && !opt_skip_slave_start) |
733 | { |
734 | int error; |
735 | THD *thd= new THD(next_thread_id()); |
736 | thd->thread_stack= (char*) &thd; |
737 | thd->store_globals(); |
738 | |
739 | error= start_slave_threads(0, /* No active thd */ |
740 | 1 /* need mutex */, |
741 | 1 /* wait for start*/, |
742 | active_mi, |
743 | master_info_file, |
744 | relay_log_info_file, |
745 | SLAVE_IO | SLAVE_SQL); |
746 | |
747 | thd->reset_globals(); |
748 | delete thd; |
749 | if (unlikely(error)) |
750 | { |
751 | sql_print_error("Failed to create slave threads" ); |
752 | goto err; |
753 | } |
754 | } |
755 | |
756 | end: |
757 | DBUG_RETURN(error); |
758 | |
759 | err: |
760 | error= 1; |
761 | goto end; |
762 | } |
763 | |
764 | /* |
765 | Updates the master info based on the information stored in the |
766 | relay info and ignores relay logs previously retrieved by the IO |
767 | thread, which thus starts fetching again based on to the |
768 | group_master_log_pos and group_master_log_name. Eventually, the old |
769 | relay logs will be purged by the normal purge mechanism. |
770 | |
771 | In the feature, we should improve this routine in order to avoid throwing |
772 | away logs that are safely stored in the disk. Note also that this recovery |
773 | routine relies on the correctness of the relay-log.info and only tolerates |
774 | coordinate problems in master.info. |
775 | |
776 | In this function, there is no need for a mutex as the caller |
777 | (i.e. init_slave) already has one acquired. |
778 | |
779 | Specifically, the following structures are updated: |
780 | |
781 | 1 - mi->master_log_pos <-- rli->group_master_log_pos |
782 | 2 - mi->master_log_name <-- rli->group_master_log_name |
783 | 3 - It moves the relay log to the new relay log file, by |
784 | rli->group_relay_log_pos <-- BIN_LOG_HEADER_SIZE; |
785 | rli->event_relay_log_pos <-- BIN_LOG_HEADER_SIZE; |
786 | rli->group_relay_log_name <-- rli->relay_log.get_log_fname(); |
787 | rli->event_relay_log_name <-- rli->relay_log.get_log_fname(); |
788 | |
789 | If there is an error, it returns (1), otherwise returns (0). |
790 | */ |
791 | int init_recovery(Master_info* mi, const char** errmsg) |
792 | { |
793 | DBUG_ENTER("init_recovery" ); |
794 | |
795 | Relay_log_info *rli= &mi->rli; |
796 | if (rli->group_master_log_name[0]) |
797 | { |
798 | mi->master_log_pos= MY_MAX(BIN_LOG_HEADER_SIZE, |
799 | rli->group_master_log_pos); |
800 | strmake_buf(mi->master_log_name, rli->group_master_log_name); |
801 | |
802 | sql_print_warning("Recovery from master pos %ld and file %s." , |
803 | (ulong) mi->master_log_pos, mi->master_log_name); |
804 | |
805 | strmake_buf(rli->group_relay_log_name, rli->relay_log.get_log_fname()); |
806 | strmake_buf(rli->event_relay_log_name, rli->relay_log.get_log_fname()); |
807 | |
808 | rli->group_relay_log_pos= rli->event_relay_log_pos= BIN_LOG_HEADER_SIZE; |
809 | } |
810 | |
811 | DBUG_RETURN(0); |
812 | } |
813 | |
814 | |
815 | /** |
816 | Convert slave skip errors bitmap into a printable string. |
817 | */ |
818 | |
819 | static void make_slave_skip_errors_printable(void) |
820 | { |
821 | /* |
822 | To be safe, we want 10 characters of room in the buffer for a number |
823 | plus terminators. Also, we need some space for constant strings. |
824 | 10 characters must be sufficient for a number plus {',' | '...'} |
825 | plus a NUL terminator. That is a max 6 digit number. |
826 | */ |
827 | const size_t MIN_ROOM= 10; |
828 | DBUG_ENTER("make_slave_skip_errors_printable" ); |
829 | DBUG_ASSERT(sizeof(slave_skip_error_names) > MIN_ROOM); |
830 | DBUG_ASSERT(MAX_SLAVE_ERROR <= 999999); // 6 digits |
831 | |
832 | /* Make @@slave_skip_errors show the nice human-readable value. */ |
833 | opt_slave_skip_errors= slave_skip_error_names; |
834 | |
835 | if (!use_slave_mask || bitmap_is_clear_all(&slave_error_mask)) |
836 | { |
837 | /* purecov: begin tested */ |
838 | memcpy(slave_skip_error_names, STRING_WITH_LEN("OFF" )); |
839 | /* purecov: end */ |
840 | } |
841 | else if (bitmap_is_set_all(&slave_error_mask)) |
842 | { |
843 | /* purecov: begin tested */ |
844 | memcpy(slave_skip_error_names, STRING_WITH_LEN("ALL" )); |
845 | /* purecov: end */ |
846 | } |
847 | else |
848 | { |
849 | char *buff= slave_skip_error_names; |
850 | char *bend= buff + sizeof(slave_skip_error_names) - MIN_ROOM; |
851 | int errnum; |
852 | |
853 | for (errnum= 0; errnum < MAX_SLAVE_ERROR; errnum++) |
854 | { |
855 | if (bitmap_is_set(&slave_error_mask, errnum)) |
856 | { |
857 | if (buff >= bend) |
858 | break; /* purecov: tested */ |
859 | buff= int10_to_str(errnum, buff, 10); |
860 | *buff++= ','; |
861 | } |
862 | } |
863 | if (buff != slave_skip_error_names) |
864 | buff--; // Remove last ',' |
865 | if (errnum < MAX_SLAVE_ERROR) |
866 | { |
867 | /* Couldn't show all errors */ |
868 | buff= strmov(buff, "..." ); /* purecov: tested */ |
869 | } |
870 | *buff=0; |
871 | } |
872 | DBUG_PRINT("init" , ("error_names: '%s'" , slave_skip_error_names)); |
873 | DBUG_VOID_RETURN; |
874 | } |
875 | |
876 | /* |
877 | Init function to set up array for errors that should be skipped for slave |
878 | |
879 | SYNOPSIS |
880 | init_slave_skip_errors() |
881 | arg List of errors numbers to skip, separated with ',' |
882 | |
883 | NOTES |
884 | Called from get_options() in mysqld.cc on start-up |
885 | */ |
886 | |
887 | bool init_slave_skip_errors(const char* arg) |
888 | { |
889 | const char *p; |
890 | DBUG_ENTER("init_slave_skip_errors" ); |
891 | |
892 | if (!arg || !*arg) // No errors defined |
893 | goto end; |
894 | |
895 | if (unlikely(my_bitmap_init(&slave_error_mask,0,MAX_SLAVE_ERROR,0))) |
896 | DBUG_RETURN(1); |
897 | |
898 | use_slave_mask= 1; |
899 | for (;my_isspace(system_charset_info,*arg);++arg) |
900 | /* empty */; |
901 | if (!my_strnncoll(system_charset_info,(uchar*)arg,4,(const uchar*)"all" ,4)) |
902 | { |
903 | bitmap_set_all(&slave_error_mask); |
904 | goto end; |
905 | } |
906 | for (p= arg ; *p; ) |
907 | { |
908 | long err_code; |
909 | if (!(p= str2int(p, 10, 0, LONG_MAX, &err_code))) |
910 | break; |
911 | if (err_code < MAX_SLAVE_ERROR) |
912 | bitmap_set_bit(&slave_error_mask,(uint)err_code); |
913 | while (!my_isdigit(system_charset_info,*p) && *p) |
914 | p++; |
915 | } |
916 | |
917 | end: |
918 | make_slave_skip_errors_printable(); |
919 | DBUG_RETURN(0); |
920 | } |
921 | |
922 | /** |
923 | Make printable version if slave_transaction_retry_errors |
924 | This is never empty as at least ER_LOCK_DEADLOCK and ER_LOCK_WAIT_TIMEOUT |
925 | will be there |
926 | */ |
927 | |
928 | static void make_slave_transaction_retry_errors_printable(void) |
929 | { |
930 | /* |
931 | To be safe, we want 10 characters of room in the buffer for a number |
932 | plus terminators. Also, we need some space for constant strings. |
933 | 10 characters must be sufficient for a number plus {',' | '...'} |
934 | plus a NUL terminator. That is a max 6 digit number. |
935 | */ |
936 | const size_t MIN_ROOM= 10; |
937 | char *buff= slave_transaction_retry_error_names; |
938 | char *bend= buff + sizeof(slave_transaction_retry_error_names) - MIN_ROOM; |
939 | uint i; |
940 | DBUG_ENTER("make_slave_transaction_retry_errors_printable" ); |
941 | DBUG_ASSERT(sizeof(slave_transaction_retry_error_names) > MIN_ROOM); |
942 | |
943 | /* Make @@slave_transaction_retry_errors show a human-readable value */ |
944 | opt_slave_transaction_retry_errors= slave_transaction_retry_error_names; |
945 | |
946 | for (i= 0; i < slave_transaction_retry_error_length && buff < bend; i++) |
947 | { |
948 | buff= int10_to_str(slave_transaction_retry_errors[i], buff, 10); |
949 | *buff++= ','; |
950 | } |
951 | if (buff != slave_transaction_retry_error_names) |
952 | buff--; // Remove last ',' |
953 | if (i < slave_transaction_retry_error_length) |
954 | { |
955 | /* Couldn't show all errors */ |
956 | buff= strmov(buff, "..." ); /* purecov: tested */ |
957 | } |
958 | *buff=0; |
959 | DBUG_PRINT("exit" , ("error_names: '%s'" , |
960 | slave_transaction_retry_error_names)); |
961 | DBUG_VOID_RETURN; |
962 | } |
963 | |
964 | |
965 | bool init_slave_transaction_retry_errors(const char* arg) |
966 | { |
967 | const char *p; |
968 | long err_code; |
969 | uint i; |
970 | DBUG_ENTER("init_slave_transaction_retry_errors" ); |
971 | |
972 | /* Handle empty strings */ |
973 | if (!arg) |
974 | arg= "" ; |
975 | |
976 | slave_transaction_retry_error_length= 2; |
977 | for (;my_isspace(system_charset_info,*arg);++arg) |
978 | /* empty */; |
979 | for (p= arg; *p; ) |
980 | { |
981 | if (!(p= str2int(p, 10, 0, LONG_MAX, &err_code))) |
982 | break; |
983 | slave_transaction_retry_error_length++; |
984 | while (!my_isdigit(system_charset_info,*p) && *p) |
985 | p++; |
986 | } |
987 | |
988 | if (unlikely(!(slave_transaction_retry_errors= |
989 | (uint *) my_once_alloc(sizeof(int) * |
990 | slave_transaction_retry_error_length, |
991 | MYF(MY_WME))))) |
992 | DBUG_RETURN(1); |
993 | |
994 | /* |
995 | Temporary error codes: |
996 | currently, InnoDB deadlock detected by InnoDB or lock |
997 | wait timeout (innodb_lock_wait_timeout exceeded |
998 | */ |
999 | slave_transaction_retry_errors[0]= ER_LOCK_DEADLOCK; |
1000 | slave_transaction_retry_errors[1]= ER_LOCK_WAIT_TIMEOUT; |
1001 | |
1002 | /* Add user codes after this */ |
1003 | for (p= arg, i= 2; *p; ) |
1004 | { |
1005 | if (!(p= str2int(p, 10, 0, LONG_MAX, &err_code))) |
1006 | break; |
1007 | if (err_code > 0 && err_code < ER_ERROR_LAST) |
1008 | slave_transaction_retry_errors[i++]= (uint) err_code; |
1009 | while (!my_isdigit(system_charset_info,*p) && *p) |
1010 | p++; |
1011 | } |
1012 | slave_transaction_retry_error_length= i; |
1013 | |
1014 | make_slave_transaction_retry_errors_printable(); |
1015 | DBUG_RETURN(0); |
1016 | } |
1017 | |
1018 | |
1019 | int terminate_slave_threads(Master_info* mi,int thread_mask,bool skip_lock) |
1020 | { |
1021 | DBUG_ENTER("terminate_slave_threads" ); |
1022 | |
1023 | if (!mi->inited) |
1024 | DBUG_RETURN(0); /* successfully do nothing */ |
1025 | int error,force_all = (thread_mask & SLAVE_FORCE_ALL); |
1026 | int retval= 0; |
1027 | mysql_mutex_t *sql_lock = &mi->rli.run_lock, *io_lock = &mi->run_lock; |
1028 | mysql_mutex_t *log_lock= mi->rli.relay_log.get_log_lock(); |
1029 | |
1030 | if (thread_mask & (SLAVE_SQL|SLAVE_FORCE_ALL)) |
1031 | { |
1032 | DBUG_PRINT("info" ,("Terminating SQL thread" )); |
1033 | if (mi->using_parallel() && mi->rli.abort_slave && mi->rli.stop_for_until) |
1034 | { |
1035 | mi->rli.stop_for_until= false; |
1036 | mi->rli.parallel.stop_during_until(); |
1037 | } |
1038 | else |
1039 | mi->rli.abort_slave=1; |
1040 | if (unlikely((error= terminate_slave_thread(mi->rli.sql_driver_thd, |
1041 | sql_lock, |
1042 | &mi->rli.stop_cond, |
1043 | &mi->rli.slave_running, |
1044 | skip_lock))) && |
1045 | !force_all) |
1046 | DBUG_RETURN(error); |
1047 | retval= error; |
1048 | |
1049 | mysql_mutex_lock(log_lock); |
1050 | |
1051 | DBUG_PRINT("info" ,("Flushing relay-log info file." )); |
1052 | if (current_thd) |
1053 | THD_STAGE_INFO(current_thd, stage_flushing_relay_log_info_file); |
1054 | if (mi->rli.flush() || my_sync(mi->rli.info_fd, MYF(MY_WME))) |
1055 | retval= ER_ERROR_DURING_FLUSH_LOGS; |
1056 | |
1057 | mysql_mutex_unlock(log_lock); |
1058 | } |
1059 | if (thread_mask & (SLAVE_IO|SLAVE_FORCE_ALL)) |
1060 | { |
1061 | DBUG_PRINT("info" ,("Terminating IO thread" )); |
1062 | mi->abort_slave=1; |
1063 | if (unlikely((error= terminate_slave_thread(mi->io_thd, io_lock, |
1064 | &mi->stop_cond, |
1065 | &mi->slave_running, |
1066 | skip_lock))) && |
1067 | !force_all) |
1068 | DBUG_RETURN(error); |
1069 | if (!retval) |
1070 | retval= error; |
1071 | |
1072 | mysql_mutex_lock(log_lock); |
1073 | |
1074 | DBUG_PRINT("info" ,("Flushing relay log and master info file." )); |
1075 | if (current_thd) |
1076 | THD_STAGE_INFO(current_thd, stage_flushing_relay_log_and_master_info_repository); |
1077 | if (likely(mi->fd >= 0)) |
1078 | { |
1079 | if (flush_master_info(mi, TRUE, FALSE) || my_sync(mi->fd, MYF(MY_WME))) |
1080 | retval= ER_ERROR_DURING_FLUSH_LOGS; |
1081 | } |
1082 | if (mi->rli.relay_log.is_open() && |
1083 | my_sync(mi->rli.relay_log.get_log_file()->file, MYF(MY_WME))) |
1084 | retval= ER_ERROR_DURING_FLUSH_LOGS; |
1085 | |
1086 | mysql_mutex_unlock(log_lock); |
1087 | } |
1088 | DBUG_RETURN(retval); |
1089 | } |
1090 | |
1091 | |
1092 | /** |
1093 | Wait for a slave thread to terminate. |
1094 | |
1095 | This function is called after requesting the thread to terminate |
1096 | (by setting @c abort_slave member of @c Relay_log_info or @c |
1097 | Master_info structure to 1). Termination of the thread is |
1098 | controlled with the the predicate <code>*slave_running</code>. |
1099 | |
1100 | Function will acquire @c term_lock before waiting on the condition |
1101 | unless @c skip_lock is true in which case the mutex should be owned |
1102 | by the caller of this function and will remain acquired after |
1103 | return from the function. |
1104 | |
1105 | @param term_lock |
1106 | Associated lock to use when waiting for @c term_cond |
1107 | |
1108 | @param term_cond |
1109 | Condition that is signalled when the thread has terminated |
1110 | |
1111 | @param slave_running |
1112 | Pointer to predicate to check for slave thread termination |
1113 | |
1114 | @param skip_lock |
1115 | If @c true the lock will not be acquired before waiting on |
1116 | the condition. In this case, it is assumed that the calling |
1117 | function acquires the lock before calling this function. |
1118 | |
1119 | @retval 0 All OK ER_SLAVE_NOT_RUNNING otherwise. |
1120 | |
1121 | @note If the executing thread has to acquire term_lock (skip_lock |
1122 | is false), the negative running status does not represent |
1123 | any issue therefore no error is reported. |
1124 | |
1125 | */ |
1126 | static int |
1127 | terminate_slave_thread(THD *thd, |
1128 | mysql_mutex_t *term_lock, |
1129 | mysql_cond_t *term_cond, |
1130 | volatile uint *slave_running, |
1131 | bool skip_lock) |
1132 | { |
1133 | DBUG_ENTER("terminate_slave_thread" ); |
1134 | if (!skip_lock) |
1135 | { |
1136 | mysql_mutex_lock(term_lock); |
1137 | } |
1138 | else |
1139 | { |
1140 | mysql_mutex_assert_owner(term_lock); |
1141 | } |
1142 | if (!*slave_running) |
1143 | { |
1144 | if (!skip_lock) |
1145 | { |
1146 | /* |
1147 | if run_lock (term_lock) is acquired locally then either |
1148 | slave_running status is fine |
1149 | */ |
1150 | mysql_mutex_unlock(term_lock); |
1151 | DBUG_RETURN(0); |
1152 | } |
1153 | else |
1154 | { |
1155 | DBUG_RETURN(ER_SLAVE_NOT_RUNNING); |
1156 | } |
1157 | } |
1158 | DBUG_ASSERT(thd != 0); |
1159 | THD_CHECK_SENTRY(thd); |
1160 | |
1161 | /* |
1162 | Is is critical to test if the slave is running. Otherwise, we might |
1163 | be referening freed memory trying to kick it |
1164 | */ |
1165 | |
1166 | while (*slave_running) // Should always be true |
1167 | { |
1168 | int error __attribute__((unused)); |
1169 | DBUG_PRINT("loop" , ("killing slave thread" )); |
1170 | |
1171 | mysql_mutex_lock(&thd->LOCK_thd_kill); |
1172 | #ifndef DONT_USE_THR_ALARM |
1173 | /* |
1174 | Error codes from pthread_kill are: |
1175 | EINVAL: invalid signal number (can't happen) |
1176 | ESRCH: thread already killed (can happen, should be ignored) |
1177 | */ |
1178 | int err __attribute__((unused))= pthread_kill(thd->real_id, thr_client_alarm); |
1179 | DBUG_ASSERT(err != EINVAL); |
1180 | #endif |
1181 | thd->awake_no_mutex(NOT_KILLED); |
1182 | |
1183 | mysql_mutex_unlock(&thd->LOCK_thd_kill); |
1184 | |
1185 | /* |
1186 | There is a small chance that slave thread might miss the first |
1187 | alarm. To protect againts it, resend the signal until it reacts |
1188 | */ |
1189 | struct timespec abstime; |
1190 | set_timespec(abstime,2); |
1191 | error= mysql_cond_timedwait(term_cond, term_lock, &abstime); |
1192 | DBUG_ASSERT(error == ETIMEDOUT || error == 0); |
1193 | } |
1194 | |
1195 | DBUG_ASSERT(*slave_running == 0); |
1196 | |
1197 | if (!skip_lock) |
1198 | mysql_mutex_unlock(term_lock); |
1199 | DBUG_RETURN(0); |
1200 | } |
1201 | |
1202 | |
1203 | int start_slave_thread( |
1204 | #ifdef HAVE_PSI_INTERFACE |
1205 | PSI_thread_key thread_key, |
1206 | #endif |
1207 | pthread_handler h_func, mysql_mutex_t *start_lock, |
1208 | mysql_mutex_t *cond_lock, |
1209 | mysql_cond_t *start_cond, |
1210 | volatile uint *slave_running, |
1211 | volatile ulong *slave_run_id, |
1212 | Master_info* mi) |
1213 | { |
1214 | pthread_t th; |
1215 | ulong start_id; |
1216 | int error; |
1217 | DBUG_ENTER("start_slave_thread" ); |
1218 | |
1219 | DBUG_ASSERT(mi->inited); |
1220 | |
1221 | if (start_lock) |
1222 | mysql_mutex_lock(start_lock); |
1223 | if (!global_system_variables.server_id) |
1224 | { |
1225 | if (start_cond) |
1226 | mysql_cond_broadcast(start_cond); |
1227 | if (start_lock) |
1228 | mysql_mutex_unlock(start_lock); |
1229 | sql_print_error("Server id not set, will not start slave" ); |
1230 | DBUG_RETURN(ER_BAD_SLAVE); |
1231 | } |
1232 | |
1233 | if (*slave_running) |
1234 | { |
1235 | if (start_cond) |
1236 | mysql_cond_broadcast(start_cond); |
1237 | if (start_lock) |
1238 | mysql_mutex_unlock(start_lock); |
1239 | DBUG_RETURN(ER_SLAVE_MUST_STOP); |
1240 | } |
1241 | start_id= *slave_run_id; |
1242 | DBUG_PRINT("info" ,("Creating new slave thread" )); |
1243 | if (unlikely((error= mysql_thread_create(thread_key, |
1244 | &th, &connection_attrib, h_func, |
1245 | (void*)mi)))) |
1246 | { |
1247 | sql_print_error("Can't create slave thread (errno= %d)." , error); |
1248 | if (start_lock) |
1249 | mysql_mutex_unlock(start_lock); |
1250 | DBUG_RETURN(ER_SLAVE_THREAD); |
1251 | } |
1252 | |
1253 | /* |
1254 | In the following loop we can't check for thd->killed as we have to |
1255 | wait until THD structures for the slave thread are created |
1256 | before we can return. |
1257 | This should be ok as there is no major work done in the slave |
1258 | threads before they signal that we can stop waiting. |
1259 | */ |
1260 | |
1261 | if (start_cond && cond_lock) // caller has cond_lock |
1262 | { |
1263 | THD* thd = current_thd; |
1264 | while (start_id == *slave_run_id) |
1265 | { |
1266 | DBUG_PRINT("sleep" ,("Waiting for slave thread to start" )); |
1267 | PSI_stage_info saved_stage= {0, "" , 0}; |
1268 | thd->ENTER_COND(start_cond, cond_lock, |
1269 | & stage_waiting_for_slave_thread_to_start, |
1270 | & saved_stage); |
1271 | /* |
1272 | It is not sufficient to test this at loop bottom. We must test |
1273 | it after registering the mutex in enter_cond(). If the kill |
1274 | happens after testing of thd->killed and before the mutex is |
1275 | registered, we could otherwise go waiting though thd->killed is |
1276 | set. |
1277 | */ |
1278 | mysql_cond_wait(start_cond, cond_lock); |
1279 | thd->EXIT_COND(& saved_stage); |
1280 | mysql_mutex_lock(cond_lock); // re-acquire it as exit_cond() released |
1281 | } |
1282 | } |
1283 | if (start_lock) |
1284 | mysql_mutex_unlock(start_lock); |
1285 | DBUG_RETURN(0); |
1286 | } |
1287 | |
1288 | |
1289 | /* |
1290 | start_slave_threads() |
1291 | |
1292 | NOTES |
1293 | SLAVE_FORCE_ALL is not implemented here on purpose since it does not make |
1294 | sense to do that for starting a slave--we always care if it actually |
1295 | started the threads that were not previously running |
1296 | */ |
1297 | |
1298 | int start_slave_threads(THD *thd, |
1299 | bool need_slave_mutex, bool wait_for_start, |
1300 | Master_info* mi, const char* master_info_fname, |
1301 | const char* slave_info_fname, int thread_mask) |
1302 | { |
1303 | mysql_mutex_t *lock_io=0, *lock_sql=0, *lock_cond_io=0, *lock_cond_sql=0; |
1304 | mysql_cond_t* cond_io=0, *cond_sql=0; |
1305 | int error=0; |
1306 | const char *errmsg; |
1307 | DBUG_ENTER("start_slave_threads" ); |
1308 | |
1309 | if (need_slave_mutex) |
1310 | { |
1311 | lock_io = &mi->run_lock; |
1312 | lock_sql = &mi->rli.run_lock; |
1313 | } |
1314 | if (wait_for_start) |
1315 | { |
1316 | cond_io = &mi->start_cond; |
1317 | cond_sql = &mi->rli.start_cond; |
1318 | lock_cond_io = &mi->run_lock; |
1319 | lock_cond_sql = &mi->rli.run_lock; |
1320 | } |
1321 | |
1322 | /* |
1323 | If we are using GTID and both SQL and IO threads are stopped, then get |
1324 | rid of all relay logs. |
1325 | |
1326 | Relay logs are not very useful when using GTID, except as a buffer |
1327 | between the fetch in the IO thread and the apply in SQL thread. However |
1328 | while one of the threads is running, they are in use and cannot be |
1329 | removed. |
1330 | */ |
1331 | if (mi->using_gtid != Master_info::USE_GTID_NO && |
1332 | !mi->slave_running && !mi->rli.slave_running) |
1333 | { |
1334 | /* |
1335 | purge_relay_logs() clears the mi->rli.group_master_log_pos. |
1336 | So save and restore them, like we do in CHANGE MASTER. |
1337 | (We are not going to use them for GTID, but it might be worth to |
1338 | keep them in case connection with GTID fails and user wants to go |
1339 | back and continue with previous old-style replication coordinates). |
1340 | */ |
1341 | mi->master_log_pos = MY_MAX(BIN_LOG_HEADER_SIZE, |
1342 | mi->rli.group_master_log_pos); |
1343 | strmake(mi->master_log_name, mi->rli.group_master_log_name, |
1344 | sizeof(mi->master_log_name)-1); |
1345 | purge_relay_logs(&mi->rli, thd, 0, &errmsg); |
1346 | mi->rli.group_master_log_pos= mi->master_log_pos; |
1347 | strmake(mi->rli.group_master_log_name, mi->master_log_name, |
1348 | sizeof(mi->rli.group_master_log_name)-1); |
1349 | |
1350 | error= rpl_load_gtid_state(&mi->gtid_current_pos, mi->using_gtid == |
1351 | Master_info::USE_GTID_CURRENT_POS); |
1352 | mi->events_queued_since_last_gtid= 0; |
1353 | mi->gtid_reconnect_event_skip_count= 0; |
1354 | |
1355 | mi->rli.restart_gtid_pos.reset(); |
1356 | } |
1357 | |
1358 | if (likely(!error) && likely((thread_mask & SLAVE_IO))) |
1359 | error= start_slave_thread( |
1360 | #ifdef HAVE_PSI_INTERFACE |
1361 | key_thread_slave_io, |
1362 | #endif |
1363 | handle_slave_io, lock_io, lock_cond_io, |
1364 | cond_io, |
1365 | &mi->slave_running, &mi->slave_run_id, |
1366 | mi); |
1367 | if (likely(!error) && likely(thread_mask & SLAVE_SQL)) |
1368 | { |
1369 | error= start_slave_thread( |
1370 | #ifdef HAVE_PSI_INTERFACE |
1371 | key_thread_slave_sql, |
1372 | #endif |
1373 | handle_slave_sql, lock_sql, lock_cond_sql, |
1374 | cond_sql, |
1375 | &mi->rli.slave_running, &mi->rli.slave_run_id, |
1376 | mi); |
1377 | if (unlikely(error)) |
1378 | terminate_slave_threads(mi, thread_mask & SLAVE_IO, !need_slave_mutex); |
1379 | } |
1380 | DBUG_RETURN(error); |
1381 | } |
1382 | |
1383 | |
1384 | /* |
1385 | Kill slaves preparing for shutdown |
1386 | */ |
1387 | |
1388 | void slave_prepare_for_shutdown() |
1389 | { |
1390 | mysql_mutex_lock(&LOCK_active_mi); |
1391 | master_info_index->free_connections(); |
1392 | mysql_mutex_unlock(&LOCK_active_mi); |
1393 | stop_slave_background_thread(); |
1394 | } |
1395 | |
1396 | /* |
1397 | Release slave threads at time of executing shutdown. |
1398 | */ |
1399 | |
1400 | void end_slave() |
1401 | { |
1402 | DBUG_ENTER("end_slave" ); |
1403 | |
1404 | /* |
1405 | This is called when the server terminates, in close_connections(). |
1406 | It terminates slave threads. However, some CHANGE MASTER etc may still be |
1407 | running presently. If a START SLAVE was in progress, the mutex lock below |
1408 | will make us wait until slave threads have started, and START SLAVE |
1409 | returns, then we terminate them here. |
1410 | |
1411 | We can also be called by cleanup(), which only happens if some |
1412 | startup parameter to the server was wrong. |
1413 | */ |
1414 | mysql_mutex_lock(&LOCK_active_mi); |
1415 | /* |
1416 | master_info_index should not have any threads anymore as they where |
1417 | killed as part of slave_prepare_for_shutdown() |
1418 | */ |
1419 | delete master_info_index; |
1420 | master_info_index= 0; |
1421 | active_mi= 0; |
1422 | mysql_mutex_unlock(&LOCK_active_mi); |
1423 | |
1424 | stop_slave_background_thread(); |
1425 | |
1426 | global_rpl_thread_pool.destroy(); |
1427 | free_all_rpl_filters(); |
1428 | DBUG_VOID_RETURN; |
1429 | } |
1430 | |
1431 | static bool io_slave_killed(Master_info* mi) |
1432 | { |
1433 | DBUG_ENTER("io_slave_killed" ); |
1434 | |
1435 | DBUG_ASSERT(mi->slave_running); // tracking buffer overrun |
1436 | DBUG_RETURN(mi->abort_slave || abort_loop || mi->io_thd->killed); |
1437 | } |
1438 | |
1439 | /** |
1440 | The function analyzes a possible killed status and makes |
1441 | a decision whether to accept it or not. |
1442 | Normally upon accepting the sql thread goes to shutdown. |
1443 | In the event of deffering decision @rli->last_event_start_time waiting |
1444 | timer is set to force the killed status be accepted upon its expiration. |
1445 | |
1446 | @param thd pointer to a THD instance |
1447 | @param rli pointer to Relay_log_info instance |
1448 | |
1449 | @return TRUE the killed status is recognized, FALSE a possible killed |
1450 | status is deferred. |
1451 | */ |
1452 | static bool sql_slave_killed(rpl_group_info *rgi) |
1453 | { |
1454 | bool ret= FALSE; |
1455 | Relay_log_info *rli= rgi->rli; |
1456 | THD *thd= rgi->thd; |
1457 | DBUG_ENTER("sql_slave_killed" ); |
1458 | |
1459 | DBUG_ASSERT(rli->sql_driver_thd == thd); |
1460 | DBUG_ASSERT(rli->slave_running == 1);// tracking buffer overrun |
1461 | if (abort_loop || rli->sql_driver_thd->killed || rli->abort_slave) |
1462 | { |
1463 | /* |
1464 | The transaction should always be binlogged if OPTION_KEEP_LOG is |
1465 | set (it implies that something can not be rolled back). And such |
1466 | case should be regarded similarly as modifing a |
1467 | non-transactional table because retrying of the transaction will |
1468 | lead to an error or inconsistency as well. |
1469 | |
1470 | Example: OPTION_KEEP_LOG is set if a temporary table is created |
1471 | or dropped. |
1472 | |
1473 | Note that transaction.all.modified_non_trans_table may be 1 |
1474 | if last statement was a single row transaction without begin/end. |
1475 | Testing this flag must always be done in connection with |
1476 | rli->is_in_group(). |
1477 | */ |
1478 | |
1479 | if ((thd->transaction.all.modified_non_trans_table || |
1480 | (thd->variables.option_bits & OPTION_KEEP_LOG)) && |
1481 | rli->is_in_group()) |
1482 | { |
1483 | char msg_stopped[]= |
1484 | "... Slave SQL Thread stopped with incomplete event group " |
1485 | "having non-transactional changes. " |
1486 | "If the group consists solely of row-based events, you can try " |
1487 | "to restart the slave with --slave-exec-mode=IDEMPOTENT, which " |
1488 | "ignores duplicate key, key not found, and similar errors (see " |
1489 | "documentation for details)." ; |
1490 | |
1491 | DBUG_PRINT("info" , ("modified_non_trans_table: %d OPTION_BEGIN: %d " |
1492 | "OPTION_KEEP_LOG: %d is_in_group: %d" , |
1493 | thd->transaction.all.modified_non_trans_table, |
1494 | MY_TEST(thd->variables.option_bits & OPTION_BEGIN), |
1495 | MY_TEST(thd->variables.option_bits & OPTION_KEEP_LOG), |
1496 | rli->is_in_group())); |
1497 | |
1498 | if (rli->abort_slave) |
1499 | { |
1500 | DBUG_PRINT("info" , |
1501 | ("Request to stop slave SQL Thread received while " |
1502 | "applying a group that has non-transactional " |
1503 | "changes; waiting for completion of the group ... " )); |
1504 | |
1505 | /* |
1506 | Slave sql thread shutdown in face of unfinished group |
1507 | modified Non-trans table is handled via a timer. The slave |
1508 | may eventually give out to complete the current group and in |
1509 | that case there might be issues at consequent slave restart, |
1510 | see the error message. WL#2975 offers a robust solution |
1511 | requiring to store the last exectuted event's coordinates |
1512 | along with the group's coordianates instead of waiting with |
1513 | @c last_event_start_time the timer. |
1514 | */ |
1515 | |
1516 | if (rgi->last_event_start_time == 0) |
1517 | rgi->last_event_start_time= my_time(0); |
1518 | ret= difftime(my_time(0), rgi->last_event_start_time) <= |
1519 | SLAVE_WAIT_GROUP_DONE ? FALSE : TRUE; |
1520 | |
1521 | DBUG_EXECUTE_IF("stop_slave_middle_group" , |
1522 | DBUG_EXECUTE_IF("incomplete_group_in_relay_log" , |
1523 | ret= TRUE;);); // time is over |
1524 | |
1525 | if (ret == 0) |
1526 | { |
1527 | rli->report(WARNING_LEVEL, 0, rgi->gtid_info(), |
1528 | "Request to stop slave SQL Thread received while " |
1529 | "applying a group that has non-transactional " |
1530 | "changes; waiting for completion of the group ... " ); |
1531 | } |
1532 | else |
1533 | { |
1534 | rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, rgi->gtid_info(), |
1535 | ER_THD(thd, ER_SLAVE_FATAL_ERROR), msg_stopped); |
1536 | } |
1537 | } |
1538 | else |
1539 | { |
1540 | ret= TRUE; |
1541 | rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, rgi->gtid_info(), |
1542 | ER_THD(thd, ER_SLAVE_FATAL_ERROR), |
1543 | msg_stopped); |
1544 | } |
1545 | } |
1546 | else |
1547 | { |
1548 | ret= TRUE; |
1549 | } |
1550 | } |
1551 | if (ret) |
1552 | rgi->last_event_start_time= 0; |
1553 | |
1554 | DBUG_RETURN(ret); |
1555 | } |
1556 | |
1557 | |
1558 | /* |
1559 | skip_load_data_infile() |
1560 | |
1561 | NOTES |
1562 | This is used to tell a 3.23 master to break send_file() |
1563 | */ |
1564 | |
1565 | void skip_load_data_infile(NET *net) |
1566 | { |
1567 | DBUG_ENTER("skip_load_data_infile" ); |
1568 | |
1569 | (void)net_request_file(net, "/dev/null" ); |
1570 | (void)my_net_read(net); // discard response |
1571 | (void)net_write_command(net, 0, (uchar*) "" , 0, (uchar*) "" , 0); // ok |
1572 | DBUG_VOID_RETURN; |
1573 | } |
1574 | |
1575 | |
1576 | bool net_request_file(NET* net, const char* fname) |
1577 | { |
1578 | DBUG_ENTER("net_request_file" ); |
1579 | DBUG_RETURN(net_write_command(net, 251, (uchar*) fname, strlen(fname), |
1580 | (uchar*) "" , 0)); |
1581 | } |
1582 | |
1583 | /* |
1584 | From other comments and tests in code, it looks like |
1585 | sometimes Query_log_event and Load_log_event can have db == 0 |
1586 | (see rewrite_db() above for example) |
1587 | (cases where this happens are unclear; it may be when the master is 3.23). |
1588 | */ |
1589 | |
1590 | const char *print_slave_db_safe(const char* db) |
1591 | { |
1592 | DBUG_ENTER("*print_slave_db_safe" ); |
1593 | |
1594 | DBUG_RETURN((db ? db : "" )); |
1595 | } |
1596 | |
1597 | #endif /* HAVE_REPLICATION */ |
1598 | |
1599 | int init_strvar_from_file(char *var, int max_size, IO_CACHE *f, |
1600 | const char *default_val) |
1601 | { |
1602 | size_t length; |
1603 | DBUG_ENTER("init_strvar_from_file" ); |
1604 | |
1605 | if ((length=my_b_gets(f,var, max_size))) |
1606 | { |
1607 | char* last_p = var + length -1; |
1608 | if (*last_p == '\n') |
1609 | *last_p = 0; // if we stopped on newline, kill it |
1610 | else |
1611 | { |
1612 | /* |
1613 | If we truncated a line or stopped on last char, remove all chars |
1614 | up to and including newline. |
1615 | */ |
1616 | int c; |
1617 | while (((c=my_b_get(f)) != '\n' && c != my_b_EOF)) ; |
1618 | } |
1619 | DBUG_RETURN(0); |
1620 | } |
1621 | else if (default_val) |
1622 | { |
1623 | strmake(var, default_val, max_size-1); |
1624 | DBUG_RETURN(0); |
1625 | } |
1626 | DBUG_RETURN(1); |
1627 | } |
1628 | |
1629 | /* |
1630 | when moving these functions to mysys, don't forget to |
1631 | remove slave.cc from libmysqld/CMakeLists.txt |
1632 | */ |
1633 | int init_intvar_from_file(int* var, IO_CACHE* f, int default_val) |
1634 | { |
1635 | char buf[32]; |
1636 | DBUG_ENTER("init_intvar_from_file" ); |
1637 | |
1638 | |
1639 | if (my_b_gets(f, buf, sizeof(buf))) |
1640 | { |
1641 | *var = atoi(buf); |
1642 | DBUG_RETURN(0); |
1643 | } |
1644 | else if (default_val) |
1645 | { |
1646 | *var = default_val; |
1647 | DBUG_RETURN(0); |
1648 | } |
1649 | DBUG_RETURN(1); |
1650 | } |
1651 | |
1652 | int init_floatvar_from_file(float* var, IO_CACHE* f, float default_val) |
1653 | { |
1654 | char buf[16]; |
1655 | DBUG_ENTER("init_floatvar_from_file" ); |
1656 | |
1657 | |
1658 | if (my_b_gets(f, buf, sizeof(buf))) |
1659 | { |
1660 | if (sscanf(buf, "%f" , var) != 1) |
1661 | DBUG_RETURN(1); |
1662 | else |
1663 | DBUG_RETURN(0); |
1664 | } |
1665 | else if (default_val != 0.0) |
1666 | { |
1667 | *var = default_val; |
1668 | DBUG_RETURN(0); |
1669 | } |
1670 | DBUG_RETURN(1); |
1671 | } |
1672 | |
1673 | |
1674 | /** |
1675 | A master info read method |
1676 | |
1677 | This function is called from @c init_master_info() along with |
1678 | relatives to restore some of @c active_mi members. |
1679 | Particularly, this function is responsible for restoring |
1680 | IGNORE_SERVER_IDS list of servers whose events the slave is |
1681 | going to ignore (to not log them in the relay log). |
1682 | Items being read are supposed to be decimal output of values of a |
1683 | type shorter or equal of @c long and separated by the single space. |
1684 | It also used to restore DO_DOMAIN_IDS & IGNORE_DOMAIN_IDS lists. |
1685 | |
1686 | @param arr @c DYNAMIC_ARRAY pointer to storage for servers id |
1687 | @param f @c IO_CACHE pointer to the source file |
1688 | |
1689 | @retval 0 All OK |
1690 | @retval non-zero An error |
1691 | */ |
1692 | |
1693 | int init_dynarray_intvar_from_file(DYNAMIC_ARRAY* arr, IO_CACHE* f) |
1694 | { |
1695 | int ret= 0; |
1696 | char buf[16 * (sizeof(long)*4 + 1)]; // static buffer to use most of times |
1697 | char *buf_act= buf; // actual buffer can be dynamic if static is short |
1698 | char *token, *last; |
1699 | uint num_items; // number of items of `arr' |
1700 | size_t read_size; |
1701 | DBUG_ENTER("init_dynarray_intvar_from_file" ); |
1702 | |
1703 | if ((read_size= my_b_gets(f, buf_act, sizeof(buf))) == 0) |
1704 | { |
1705 | DBUG_RETURN(0); // no line in master.info |
1706 | } |
1707 | if (read_size + 1 == sizeof(buf) && buf[sizeof(buf) - 2] != '\n') |
1708 | { |
1709 | /* |
1710 | short read happend; allocate sufficient memory and make the 2nd read |
1711 | */ |
1712 | char buf_work[(sizeof(long)*3 + 1)*16]; |
1713 | memcpy(buf_work, buf, sizeof(buf_work)); |
1714 | num_items= atoi(strtok_r(buf_work, " " , &last)); |
1715 | size_t snd_size; |
1716 | /* |
1717 | max size lower bound approximate estimation bases on the formula: |
1718 | (the items number + items themselves) * |
1719 | (decimal size + space) - 1 + `\n' + '\0' |
1720 | */ |
1721 | size_t max_size= (1 + num_items) * (sizeof(long)*3 + 1) + 1; |
1722 | buf_act= (char*) my_malloc(max_size, MYF(MY_WME)); |
1723 | memcpy(buf_act, buf, read_size); |
1724 | snd_size= my_b_gets(f, buf_act + read_size, max_size - read_size); |
1725 | if (snd_size == 0 || |
1726 | ((snd_size + 1 == max_size - read_size) && buf_act[max_size - 2] != '\n')) |
1727 | { |
1728 | /* |
1729 | failure to make the 2nd read or short read again |
1730 | */ |
1731 | ret= 1; |
1732 | goto err; |
1733 | } |
1734 | } |
1735 | token= strtok_r(buf_act, " " , &last); |
1736 | if (token == NULL) |
1737 | { |
1738 | ret= 1; |
1739 | goto err; |
1740 | } |
1741 | num_items= atoi(token); |
1742 | for (uint i=0; i < num_items; i++) |
1743 | { |
1744 | token= strtok_r(NULL, " " , &last); |
1745 | if (token == NULL) |
1746 | { |
1747 | ret= 1; |
1748 | goto err; |
1749 | } |
1750 | else |
1751 | { |
1752 | ulong val= atol(token); |
1753 | insert_dynamic(arr, (uchar *) &val); |
1754 | } |
1755 | } |
1756 | err: |
1757 | if (buf_act != buf) |
1758 | my_free(buf_act); |
1759 | DBUG_RETURN(ret); |
1760 | } |
1761 | |
1762 | #ifdef HAVE_REPLICATION |
1763 | |
1764 | /* |
1765 | Check if the error is caused by network. |
1766 | @param[in] errorno Number of the error. |
1767 | RETURNS: |
1768 | TRUE network error |
1769 | FALSE not network error |
1770 | */ |
1771 | |
1772 | bool is_network_error(uint errorno) |
1773 | { |
1774 | if (errorno == CR_CONNECTION_ERROR || |
1775 | errorno == CR_CONN_HOST_ERROR || |
1776 | errorno == CR_SERVER_GONE_ERROR || |
1777 | errorno == CR_SERVER_LOST || |
1778 | errorno == ER_CON_COUNT_ERROR || |
1779 | errorno == ER_CONNECTION_KILLED || |
1780 | errorno == ER_NEW_ABORTING_CONNECTION || |
1781 | errorno == ER_NET_READ_INTERRUPTED || |
1782 | errorno == ER_SERVER_SHUTDOWN) |
1783 | return TRUE; |
1784 | #ifdef WITH_WSREP |
1785 | if (errorno == ER_UNKNOWN_COM_ERROR) |
1786 | return TRUE; |
1787 | #endif |
1788 | |
1789 | return FALSE; |
1790 | } |
1791 | |
1792 | |
1793 | /* |
1794 | Note that we rely on the master's version (3.23, 4.0.14 etc) instead of |
1795 | relying on the binlog's version. This is not perfect: imagine an upgrade |
1796 | of the master without waiting that all slaves are in sync with the master; |
1797 | then a slave could be fooled about the binlog's format. This is what happens |
1798 | when people upgrade a 3.23 master to 4.0 without doing RESET MASTER: 4.0 |
1799 | slaves are fooled. So we do this only to distinguish between 3.23 and more |
1800 | recent masters (it's too late to change things for 3.23). |
1801 | |
1802 | RETURNS |
1803 | 0 ok |
1804 | 1 error |
1805 | 2 transient network problem, the caller should try to reconnect |
1806 | */ |
1807 | |
1808 | static int get_master_version_and_clock(MYSQL* mysql, Master_info* mi) |
1809 | { |
1810 | char err_buff[MAX_SLAVE_ERRMSG], err_buff2[MAX_SLAVE_ERRMSG]; |
1811 | const char* errmsg= 0; |
1812 | int err_code= 0; |
1813 | MYSQL_RES *master_res= 0; |
1814 | MYSQL_ROW master_row; |
1815 | uint version= mysql_get_server_version(mysql) / 10000; |
1816 | DBUG_ENTER("get_master_version_and_clock" ); |
1817 | |
1818 | /* |
1819 | Free old description_event_for_queue (that is needed if we are in |
1820 | a reconnection). |
1821 | */ |
1822 | delete mi->rli.relay_log.description_event_for_queue; |
1823 | mi->rli.relay_log.description_event_for_queue= 0; |
1824 | |
1825 | if (!my_isdigit(&my_charset_bin,*mysql->server_version)) |
1826 | { |
1827 | errmsg= err_buff2; |
1828 | snprintf(err_buff2, sizeof(err_buff2), |
1829 | "Master reported unrecognized MySQL version: %s" , |
1830 | mysql->server_version); |
1831 | err_code= ER_SLAVE_FATAL_ERROR; |
1832 | sprintf(err_buff, ER_DEFAULT(err_code), err_buff2); |
1833 | } |
1834 | else |
1835 | { |
1836 | /* |
1837 | Note the following switch will bug when we have MySQL branch 30 ;) |
1838 | */ |
1839 | switch (version) { |
1840 | case 0: |
1841 | case 1: |
1842 | case 2: |
1843 | errmsg= err_buff2; |
1844 | snprintf(err_buff2, sizeof(err_buff2), |
1845 | "Master reported unrecognized MySQL version: %s" , |
1846 | mysql->server_version); |
1847 | err_code= ER_SLAVE_FATAL_ERROR; |
1848 | sprintf(err_buff, ER_DEFAULT(err_code), err_buff2); |
1849 | break; |
1850 | case 3: |
1851 | mi->rli.relay_log.description_event_for_queue= new |
1852 | Format_description_log_event(1, mysql->server_version); |
1853 | break; |
1854 | case 4: |
1855 | mi->rli.relay_log.description_event_for_queue= new |
1856 | Format_description_log_event(3, mysql->server_version); |
1857 | break; |
1858 | default: |
1859 | /* |
1860 | Master is MySQL >=5.0. Give a default Format_desc event, so that we can |
1861 | take the early steps (like tests for "is this a 3.23 master") which we |
1862 | have to take before we receive the real master's Format_desc which will |
1863 | override this one. Note that the Format_desc we create below is garbage |
1864 | (it has the format of the *slave*); it's only good to help know if the |
1865 | master is 3.23, 4.0, etc. |
1866 | */ |
1867 | mi->rli.relay_log.description_event_for_queue= new |
1868 | Format_description_log_event(4, mysql->server_version); |
1869 | break; |
1870 | } |
1871 | } |
1872 | |
1873 | /* |
1874 | This does not mean that a 5.0 slave will be able to read a 6.0 master; but |
1875 | as we don't know yet, we don't want to forbid this for now. If a 5.0 slave |
1876 | can't read a 6.0 master, this will show up when the slave can't read some |
1877 | events sent by the master, and there will be error messages. |
1878 | */ |
1879 | |
1880 | if (errmsg) |
1881 | goto err; |
1882 | |
1883 | /* as we are here, we tried to allocate the event */ |
1884 | if (!mi->rli.relay_log.description_event_for_queue) |
1885 | { |
1886 | errmsg= "default Format_description_log_event" ; |
1887 | err_code= ER_SLAVE_CREATE_EVENT_FAILURE; |
1888 | sprintf(err_buff, ER_DEFAULT(err_code), errmsg); |
1889 | goto err; |
1890 | } |
1891 | |
1892 | /* |
1893 | FD_q's (A) is set initially from RL's (A): FD_q.(A) := RL.(A). |
1894 | It's necessary to adjust FD_q.(A) at this point because in the following |
1895 | course FD_q is going to be dumped to RL. |
1896 | Generally FD_q is derived from a received FD_m (roughly FD_q := FD_m) |
1897 | in queue_event and the master's (A) is installed. |
1898 | At one step with the assignment the Relay-Log's checksum alg is set to |
1899 | a new value: RL.(A) := FD_q.(A). If the slave service is stopped |
1900 | the last time assigned RL.(A) will be passed over to the restarting |
1901 | service (to the current execution point). |
1902 | RL.A is a "codec" to verify checksum in queue_event() almost all the time |
1903 | the first fake Rotate event. |
1904 | Starting from this point IO thread will executes the following checksum |
1905 | warmup sequence of actions: |
1906 | |
1907 | FD_q.A := RL.A, |
1908 | A_m^0 := master.@@global.binlog_checksum, |
1909 | {queue_event(R_f): verifies(R_f, A_m^0)}, |
1910 | {queue_event(FD_m): verifies(FD_m, FD_m.A), dump(FD_q), rotate(RL), |
1911 | FD_q := FD_m, RL.A := FD_q.A)} |
1912 | |
1913 | See legends definition on MYSQL_BIN_LOG::relay_log_checksum_alg |
1914 | docs lines (binlog.h). |
1915 | In above A_m^0 - the value of master's |
1916 | @@binlog_checksum determined in the upcoming handshake (stored in |
1917 | mi->checksum_alg_before_fd). |
1918 | |
1919 | |
1920 | After the warm-up sequence IO gets to "normal" checksum verification mode |
1921 | to use RL.A in |
1922 | |
1923 | {queue_event(E_m): verifies(E_m, RL.A)} |
1924 | |
1925 | until it has received a new FD_m. |
1926 | */ |
1927 | mi->rli.relay_log.description_event_for_queue->checksum_alg= |
1928 | mi->rli.relay_log.relay_log_checksum_alg; |
1929 | |
1930 | DBUG_ASSERT(mi->rli.relay_log.description_event_for_queue->checksum_alg != |
1931 | BINLOG_CHECKSUM_ALG_UNDEF); |
1932 | DBUG_ASSERT(mi->rli.relay_log.relay_log_checksum_alg != |
1933 | BINLOG_CHECKSUM_ALG_UNDEF); |
1934 | /* |
1935 | Compare the master and slave's clock. Do not die if master's clock is |
1936 | unavailable (very old master not supporting UNIX_TIMESTAMP()?). |
1937 | */ |
1938 | |
1939 | #ifdef ENABLED_DEBUG_SYNC |
1940 | DBUG_EXECUTE_IF("dbug.before_get_UNIX_TIMESTAMP" , |
1941 | { |
1942 | const char act[]= |
1943 | "now " |
1944 | "wait_for signal.get_unix_timestamp" ; |
1945 | DBUG_ASSERT(debug_sync_service); |
1946 | DBUG_ASSERT(!debug_sync_set_action(current_thd, |
1947 | STRING_WITH_LEN(act))); |
1948 | };); |
1949 | #endif |
1950 | |
1951 | master_res= NULL; |
1952 | if (!mysql_real_query(mysql, STRING_WITH_LEN("SELECT UNIX_TIMESTAMP()" )) && |
1953 | (master_res= mysql_store_result(mysql)) && |
1954 | (master_row= mysql_fetch_row(master_res))) |
1955 | { |
1956 | mysql_mutex_lock(&mi->data_lock); |
1957 | mi->clock_diff_with_master= |
1958 | (long) (time((time_t*) 0) - strtoul(master_row[0], 0, 10)); |
1959 | mysql_mutex_unlock(&mi->data_lock); |
1960 | } |
1961 | else if (check_io_slave_killed(mi, NULL)) |
1962 | goto slave_killed_err; |
1963 | else if (is_network_error(mysql_errno(mysql))) |
1964 | { |
1965 | mi->report(WARNING_LEVEL, mysql_errno(mysql), NULL, |
1966 | "Get master clock failed with error: %s" , mysql_error(mysql)); |
1967 | goto network_err; |
1968 | } |
1969 | else |
1970 | { |
1971 | mysql_mutex_lock(&mi->data_lock); |
1972 | mi->clock_diff_with_master= 0; /* The "most sensible" value */ |
1973 | mysql_mutex_unlock(&mi->data_lock); |
1974 | sql_print_warning("\"SELECT UNIX_TIMESTAMP()\" failed on master, " |
1975 | "do not trust column Seconds_Behind_Master of SHOW " |
1976 | "SLAVE STATUS. Error: %s (%d)" , |
1977 | mysql_error(mysql), mysql_errno(mysql)); |
1978 | } |
1979 | if (master_res) |
1980 | { |
1981 | mysql_free_result(master_res); |
1982 | master_res= NULL; |
1983 | } |
1984 | |
1985 | /* |
1986 | Check that the master's server id and ours are different. Because if they |
1987 | are equal (which can result from a simple copy of master's datadir to slave, |
1988 | thus copying some my.cnf), replication will work but all events will be |
1989 | skipped. |
1990 | Do not die if SHOW VARIABLES LIKE 'SERVER_ID' fails on master (very old |
1991 | master?). |
1992 | Note: we could have put a @@SERVER_ID in the previous SELECT |
1993 | UNIX_TIMESTAMP() instead, but this would not have worked on 3.23 masters. |
1994 | */ |
1995 | #ifdef ENABLED_DEBUG_SYNC |
1996 | DBUG_EXECUTE_IF("dbug.before_get_SERVER_ID" , |
1997 | { |
1998 | const char act[]= |
1999 | "now " |
2000 | "wait_for signal.get_server_id" ; |
2001 | DBUG_ASSERT(debug_sync_service); |
2002 | DBUG_ASSERT(!debug_sync_set_action(current_thd, |
2003 | STRING_WITH_LEN(act))); |
2004 | };); |
2005 | #endif |
2006 | master_res= NULL; |
2007 | master_row= NULL; |
2008 | if (!mysql_real_query(mysql, |
2009 | STRING_WITH_LEN("SHOW VARIABLES LIKE 'SERVER_ID'" )) && |
2010 | (master_res= mysql_store_result(mysql)) && |
2011 | (master_row= mysql_fetch_row(master_res))) |
2012 | { |
2013 | if ((global_system_variables.server_id == |
2014 | (mi->master_id= strtoul(master_row[1], 0, 10))) && |
2015 | !mi->rli.replicate_same_server_id) |
2016 | { |
2017 | errmsg= "The slave I/O thread stops because master and slave have equal \ |
2018 | MySQL server ids; these ids must be different for replication to work (or \ |
2019 | the --replicate-same-server-id option must be used on slave but this does \ |
2020 | not always make sense; please check the manual before using it)." ; |
2021 | err_code= ER_SLAVE_FATAL_ERROR; |
2022 | sprintf(err_buff, ER_DEFAULT(err_code), errmsg); |
2023 | goto err; |
2024 | } |
2025 | } |
2026 | else if (mysql_errno(mysql)) |
2027 | { |
2028 | if (check_io_slave_killed(mi, NULL)) |
2029 | goto slave_killed_err; |
2030 | else if (is_network_error(mysql_errno(mysql))) |
2031 | { |
2032 | mi->report(WARNING_LEVEL, mysql_errno(mysql), NULL, |
2033 | "Get master SERVER_ID failed with error: %s" , mysql_error(mysql)); |
2034 | goto network_err; |
2035 | } |
2036 | /* Fatal error */ |
2037 | errmsg= "The slave I/O thread stops because a fatal error is encountered \ |
2038 | when it try to get the value of SERVER_ID variable from master." ; |
2039 | err_code= mysql_errno(mysql); |
2040 | sprintf(err_buff, "%s Error: %s" , errmsg, mysql_error(mysql)); |
2041 | goto err; |
2042 | } |
2043 | else if (!master_row && master_res) |
2044 | { |
2045 | mi->report(WARNING_LEVEL, ER_UNKNOWN_SYSTEM_VARIABLE, NULL, |
2046 | "Unknown system variable 'SERVER_ID' on master, \ |
2047 | maybe it is a *VERY OLD MASTER*." ); |
2048 | } |
2049 | if (master_res) |
2050 | { |
2051 | mysql_free_result(master_res); |
2052 | master_res= NULL; |
2053 | } |
2054 | if (mi->master_id == 0 && mi->ignore_server_ids.elements > 0) |
2055 | { |
2056 | errmsg= "Slave configured with server id filtering could not detect the master server id." ; |
2057 | err_code= ER_SLAVE_FATAL_ERROR; |
2058 | sprintf(err_buff, ER_DEFAULT(err_code), errmsg); |
2059 | goto err; |
2060 | } |
2061 | |
2062 | /* |
2063 | Check that the master's global character_set_server and ours are the same. |
2064 | Not fatal if query fails (old master?). |
2065 | Note that we don't check for equality of global character_set_client and |
2066 | collation_connection (neither do we prevent their setting in |
2067 | set_var.cc). That's because from what I (Guilhem) have tested, the global |
2068 | values of these 2 are never used (new connections don't use them). |
2069 | We don't test equality of global collation_database either as it's is |
2070 | going to be deprecated (made read-only) in 4.1 very soon. |
2071 | The test is only relevant if master < 5.0.3 (we'll test only if it's older |
2072 | than the 5 branch; < 5.0.3 was alpha...), as >= 5.0.3 master stores |
2073 | charset info in each binlog event. |
2074 | We don't do it for 3.23 because masters <3.23.50 hang on |
2075 | SELECT @@unknown_var (BUG#7965 - see changelog of 3.23.50). So finally we |
2076 | test only if master is 4.x. |
2077 | */ |
2078 | |
2079 | /* redundant with rest of code but safer against later additions */ |
2080 | if (version == 3) |
2081 | goto err; |
2082 | |
2083 | if (version == 4) |
2084 | { |
2085 | master_res= NULL; |
2086 | if (!mysql_real_query(mysql, |
2087 | STRING_WITH_LEN("SELECT @@GLOBAL.COLLATION_SERVER" )) && |
2088 | (master_res= mysql_store_result(mysql)) && |
2089 | (master_row= mysql_fetch_row(master_res))) |
2090 | { |
2091 | if (strcmp(master_row[0], global_system_variables.collation_server->name)) |
2092 | { |
2093 | errmsg= "The slave I/O thread stops because master and slave have \ |
2094 | different values for the COLLATION_SERVER global variable. The values must \ |
2095 | be equal for the Statement-format replication to work" ; |
2096 | err_code= ER_SLAVE_FATAL_ERROR; |
2097 | sprintf(err_buff, ER_DEFAULT(err_code), errmsg); |
2098 | goto err; |
2099 | } |
2100 | } |
2101 | else if (check_io_slave_killed(mi, NULL)) |
2102 | goto slave_killed_err; |
2103 | else if (is_network_error(mysql_errno(mysql))) |
2104 | { |
2105 | mi->report(WARNING_LEVEL, mysql_errno(mysql), NULL, |
2106 | "Get master COLLATION_SERVER failed with error: %s" , mysql_error(mysql)); |
2107 | goto network_err; |
2108 | } |
2109 | else if (mysql_errno(mysql) != ER_UNKNOWN_SYSTEM_VARIABLE) |
2110 | { |
2111 | /* Fatal error */ |
2112 | errmsg= "The slave I/O thread stops because a fatal error is encountered \ |
2113 | when it try to get the value of COLLATION_SERVER global variable from master." ; |
2114 | err_code= mysql_errno(mysql); |
2115 | sprintf(err_buff, "%s Error: %s" , errmsg, mysql_error(mysql)); |
2116 | goto err; |
2117 | } |
2118 | else |
2119 | mi->report(WARNING_LEVEL, ER_UNKNOWN_SYSTEM_VARIABLE, NULL, |
2120 | "Unknown system variable 'COLLATION_SERVER' on master, \ |
2121 | maybe it is a *VERY OLD MASTER*. *NOTE*: slave may experience \ |
2122 | inconsistency if replicated data deals with collation." ); |
2123 | |
2124 | if (master_res) |
2125 | { |
2126 | mysql_free_result(master_res); |
2127 | master_res= NULL; |
2128 | } |
2129 | } |
2130 | |
2131 | /* |
2132 | Perform analogous check for time zone. Theoretically we also should |
2133 | perform check here to verify that SYSTEM time zones are the same on |
2134 | slave and master, but we can't rely on value of @@system_time_zone |
2135 | variable (it is time zone abbreviation) since it determined at start |
2136 | time and so could differ for slave and master even if they are really |
2137 | in the same system time zone. So we are omiting this check and just |
2138 | relying on documentation. Also according to Monty there are many users |
2139 | who are using replication between servers in various time zones. Hence |
2140 | such check will broke everything for them. (And now everything will |
2141 | work for them because by default both their master and slave will have |
2142 | 'SYSTEM' time zone). |
2143 | This check is only necessary for 4.x masters (and < 5.0.4 masters but |
2144 | those were alpha). |
2145 | */ |
2146 | if (version == 4) |
2147 | { |
2148 | master_res= NULL; |
2149 | if (!mysql_real_query(mysql, STRING_WITH_LEN("SELECT @@GLOBAL.TIME_ZONE" )) && |
2150 | (master_res= mysql_store_result(mysql)) && |
2151 | (master_row= mysql_fetch_row(master_res))) |
2152 | { |
2153 | if (strcmp(master_row[0], |
2154 | global_system_variables.time_zone->get_name()->ptr())) |
2155 | { |
2156 | errmsg= "The slave I/O thread stops because master and slave have \ |
2157 | different values for the TIME_ZONE global variable. The values must \ |
2158 | be equal for the Statement-format replication to work" ; |
2159 | err_code= ER_SLAVE_FATAL_ERROR; |
2160 | sprintf(err_buff, ER_DEFAULT(err_code), errmsg); |
2161 | goto err; |
2162 | } |
2163 | } |
2164 | else if (check_io_slave_killed(mi, NULL)) |
2165 | goto slave_killed_err; |
2166 | else if (is_network_error(err_code= mysql_errno(mysql))) |
2167 | { |
2168 | mi->report(ERROR_LEVEL, err_code, NULL, |
2169 | "Get master TIME_ZONE failed with error: %s" , |
2170 | mysql_error(mysql)); |
2171 | goto network_err; |
2172 | } |
2173 | else if (err_code == ER_UNKNOWN_SYSTEM_VARIABLE) |
2174 | { |
2175 | /* We use ERROR_LEVEL to get the error logged to file */ |
2176 | mi->report(ERROR_LEVEL, err_code, NULL, |
2177 | |
2178 | "MySQL master doesn't have a TIME_ZONE variable. Note that" |
2179 | "if your timezone is not same between master and slave, your " |
2180 | "slave may get wrong data into timestamp columns" ); |
2181 | } |
2182 | else |
2183 | { |
2184 | /* Fatal error */ |
2185 | errmsg= "The slave I/O thread stops because a fatal error is encountered \ |
2186 | when it try to get the value of TIME_ZONE global variable from master." ; |
2187 | sprintf(err_buff, "%s Error: %s" , errmsg, mysql_error(mysql)); |
2188 | goto err; |
2189 | } |
2190 | if (master_res) |
2191 | { |
2192 | mysql_free_result(master_res); |
2193 | master_res= NULL; |
2194 | } |
2195 | } |
2196 | |
2197 | if (mi->heartbeat_period != 0.0) |
2198 | { |
2199 | const char query_format[]= "SET @master_heartbeat_period= %llu" ; |
2200 | char query[sizeof(query_format) + 32]; |
2201 | /* |
2202 | the period is an ulonglong of nano-secs. |
2203 | */ |
2204 | my_snprintf(query, sizeof(query), query_format, |
2205 | (ulonglong) (mi->heartbeat_period*1000000000UL)); |
2206 | |
2207 | DBUG_EXECUTE_IF("simulate_slave_heartbeat_network_error" , |
2208 | { static ulong dbug_count= 0; |
2209 | if (++dbug_count < 3) |
2210 | goto heartbeat_network_error; |
2211 | }); |
2212 | if (mysql_real_query(mysql, query, (ulong)strlen(query))) |
2213 | { |
2214 | if (check_io_slave_killed(mi, NULL)) |
2215 | goto slave_killed_err; |
2216 | |
2217 | if (is_network_error(mysql_errno(mysql))) |
2218 | { |
2219 | IF_DBUG(heartbeat_network_error: , ) |
2220 | mi->report(WARNING_LEVEL, mysql_errno(mysql), NULL, |
2221 | "SET @master_heartbeat_period to master failed with error: %s" , |
2222 | mysql_error(mysql)); |
2223 | mysql_free_result(mysql_store_result(mysql)); |
2224 | goto network_err; |
2225 | } |
2226 | else |
2227 | { |
2228 | /* Fatal error */ |
2229 | errmsg= "The slave I/O thread stops because a fatal error is encountered " |
2230 | "when it tries to SET @master_heartbeat_period on master." ; |
2231 | err_code= ER_SLAVE_FATAL_ERROR; |
2232 | sprintf(err_buff, "%s Error: %s" , errmsg, mysql_error(mysql)); |
2233 | mysql_free_result(mysql_store_result(mysql)); |
2234 | goto err; |
2235 | } |
2236 | } |
2237 | mysql_free_result(mysql_store_result(mysql)); |
2238 | } |
2239 | |
2240 | /* |
2241 | Querying if master is capable to checksum and notifying it about own |
2242 | CRC-awareness. The master's side instant value of @@global.binlog_checksum |
2243 | is stored in the dump thread's uservar area as well as cached locally |
2244 | to become known in consensus by master and slave. |
2245 | */ |
2246 | DBUG_EXECUTE_IF("simulate_slave_unaware_checksum" , |
2247 | mi->checksum_alg_before_fd= BINLOG_CHECKSUM_ALG_OFF; |
2248 | goto past_checksum;); |
2249 | { |
2250 | int rc; |
2251 | const char query[]= "SET @master_binlog_checksum= @@global.binlog_checksum" ; |
2252 | master_res= NULL; |
2253 | mi->checksum_alg_before_fd= BINLOG_CHECKSUM_ALG_UNDEF; //initially undefined |
2254 | /* |
2255 | @c checksum_alg_before_fd is queried from master in this block. |
2256 | If master is old checksum-unaware the value stays undefined. |
2257 | Once the first FD will be received its alg descriptor will replace |
2258 | the being queried one. |
2259 | */ |
2260 | rc= mysql_real_query(mysql, query,(ulong)strlen(query)); |
2261 | if (rc != 0) |
2262 | { |
2263 | if (check_io_slave_killed(mi, NULL)) |
2264 | goto slave_killed_err; |
2265 | |
2266 | if (mysql_errno(mysql) == ER_UNKNOWN_SYSTEM_VARIABLE) |
2267 | { |
2268 | /* Ignore this expected error if not a high error level */ |
2269 | if (global_system_variables.log_warnings > 1) |
2270 | { |
2271 | // this is tolerable as OM -> NS is supported |
2272 | mi->report(WARNING_LEVEL, mysql_errno(mysql), NULL, |
2273 | "Notifying master by %s failed with " |
2274 | "error: %s" , query, mysql_error(mysql)); |
2275 | } |
2276 | } |
2277 | else |
2278 | { |
2279 | if (is_network_error(mysql_errno(mysql))) |
2280 | { |
2281 | mi->report(WARNING_LEVEL, mysql_errno(mysql), NULL, |
2282 | "Notifying master by %s failed with " |
2283 | "error: %s" , query, mysql_error(mysql)); |
2284 | mysql_free_result(mysql_store_result(mysql)); |
2285 | goto network_err; |
2286 | } |
2287 | else |
2288 | { |
2289 | errmsg= "The slave I/O thread stops because a fatal error is encountered " |
2290 | "when it tried to SET @master_binlog_checksum on master." ; |
2291 | err_code= ER_SLAVE_FATAL_ERROR; |
2292 | sprintf(err_buff, "%s Error: %s" , errmsg, mysql_error(mysql)); |
2293 | mysql_free_result(mysql_store_result(mysql)); |
2294 | goto err; |
2295 | } |
2296 | } |
2297 | } |
2298 | else |
2299 | { |
2300 | mysql_free_result(mysql_store_result(mysql)); |
2301 | if (!mysql_real_query(mysql, |
2302 | STRING_WITH_LEN("SELECT @master_binlog_checksum" )) && |
2303 | (master_res= mysql_store_result(mysql)) && |
2304 | (master_row= mysql_fetch_row(master_res)) && |
2305 | (master_row[0] != NULL)) |
2306 | { |
2307 | mi->checksum_alg_before_fd= (enum_binlog_checksum_alg) |
2308 | (find_type(master_row[0], &binlog_checksum_typelib, 1) - 1); |
2309 | // valid outcome is either of |
2310 | DBUG_ASSERT(mi->checksum_alg_before_fd == BINLOG_CHECKSUM_ALG_OFF || |
2311 | mi->checksum_alg_before_fd == BINLOG_CHECKSUM_ALG_CRC32); |
2312 | } |
2313 | else if (check_io_slave_killed(mi, NULL)) |
2314 | goto slave_killed_err; |
2315 | else if (is_network_error(mysql_errno(mysql))) |
2316 | { |
2317 | mi->report(WARNING_LEVEL, mysql_errno(mysql), NULL, |
2318 | "Get master BINLOG_CHECKSUM failed with error: %s" , mysql_error(mysql)); |
2319 | goto network_err; |
2320 | } |
2321 | else |
2322 | { |
2323 | errmsg= "The slave I/O thread stops because a fatal error is encountered " |
2324 | "when it tried to SELECT @master_binlog_checksum." ; |
2325 | err_code= ER_SLAVE_FATAL_ERROR; |
2326 | sprintf(err_buff, "%s Error: %s" , errmsg, mysql_error(mysql)); |
2327 | mysql_free_result(mysql_store_result(mysql)); |
2328 | goto err; |
2329 | } |
2330 | } |
2331 | if (master_res) |
2332 | { |
2333 | mysql_free_result(master_res); |
2334 | master_res= NULL; |
2335 | } |
2336 | } |
2337 | |
2338 | #ifndef DBUG_OFF |
2339 | past_checksum: |
2340 | #endif |
2341 | |
2342 | /* |
2343 | Request the master to filter away events with the @@skip_replication flag |
2344 | set, if we are running with |
2345 | --replicate-events-marked-for-skip=FILTER_ON_MASTER. |
2346 | */ |
2347 | if (opt_replicate_events_marked_for_skip == RPL_SKIP_FILTER_ON_MASTER) |
2348 | { |
2349 | if (unlikely(mysql_real_query(mysql, |
2350 | STRING_WITH_LEN("SET skip_replication=1" )))) |
2351 | { |
2352 | err_code= mysql_errno(mysql); |
2353 | if (is_network_error(err_code)) |
2354 | { |
2355 | mi->report(ERROR_LEVEL, err_code, NULL, |
2356 | "Setting master-side filtering of @@skip_replication failed " |
2357 | "with error: %s" , mysql_error(mysql)); |
2358 | goto network_err; |
2359 | } |
2360 | else if (err_code == ER_UNKNOWN_SYSTEM_VARIABLE) |
2361 | { |
2362 | /* |
2363 | The master is older than the slave and does not support the |
2364 | @@skip_replication feature. |
2365 | This is not a problem, as such master will not generate events with |
2366 | the @@skip_replication flag set in the first place. We will still |
2367 | do slave-side filtering of such events though, to handle the (rare) |
2368 | case of downgrading a master and receiving old events generated from |
2369 | before the downgrade with the @@skip_replication flag set. |
2370 | */ |
2371 | DBUG_PRINT("info" , ("Old master does not support master-side filtering " |
2372 | "of @@skip_replication events." )); |
2373 | } |
2374 | else |
2375 | { |
2376 | /* Fatal error */ |
2377 | errmsg= "The slave I/O thread stops because a fatal error is " |
2378 | "encountered when it tries to request filtering of events marked " |
2379 | "with the @@skip_replication flag." ; |
2380 | sprintf(err_buff, "%s Error: %s" , errmsg, mysql_error(mysql)); |
2381 | goto err; |
2382 | } |
2383 | } |
2384 | } |
2385 | |
2386 | /* Announce MariaDB slave capabilities. */ |
2387 | DBUG_EXECUTE_IF("simulate_slave_capability_none" , goto after_set_capability;); |
2388 | { |
2389 | int rc= DBUG_EVALUATE_IF("simulate_slave_capability_old_53" , |
2390 | mysql_real_query(mysql, STRING_WITH_LEN("SET @mariadb_slave_capability=" |
2391 | STRINGIFY_ARG(MARIA_SLAVE_CAPABILITY_ANNOTATE))), |
2392 | mysql_real_query(mysql, STRING_WITH_LEN("SET @mariadb_slave_capability=" |
2393 | STRINGIFY_ARG(MARIA_SLAVE_CAPABILITY_MINE)))); |
2394 | if (unlikely(rc)) |
2395 | { |
2396 | err_code= mysql_errno(mysql); |
2397 | if (is_network_error(err_code)) |
2398 | { |
2399 | mi->report(ERROR_LEVEL, err_code, NULL, |
2400 | "Setting @mariadb_slave_capability failed with error: %s" , |
2401 | mysql_error(mysql)); |
2402 | goto network_err; |
2403 | } |
2404 | else |
2405 | { |
2406 | /* Fatal error */ |
2407 | errmsg= "The slave I/O thread stops because a fatal error is " |
2408 | "encountered when it tries to set @mariadb_slave_capability." ; |
2409 | sprintf(err_buff, "%s Error: %s" , errmsg, mysql_error(mysql)); |
2410 | goto err; |
2411 | } |
2412 | } |
2413 | } |
2414 | #ifndef DBUG_OFF |
2415 | after_set_capability: |
2416 | #endif |
2417 | |
2418 | if (mi->using_gtid != Master_info::USE_GTID_NO) |
2419 | { |
2420 | /* Request dump to start from slave replication GTID state. */ |
2421 | int rc; |
2422 | char str_buf[256]; |
2423 | String query_str(str_buf, sizeof(str_buf), system_charset_info); |
2424 | query_str.length(0); |
2425 | |
2426 | /* |
2427 | Read the master @@GLOBAL.gtid_domain_id variable. |
2428 | This is mostly to check that master is GTID aware, but we could later |
2429 | perhaps use it to check that different multi-source masters are correctly |
2430 | configured with distinct domain_id. |
2431 | */ |
2432 | if (mysql_real_query(mysql, |
2433 | STRING_WITH_LEN("SELECT @@GLOBAL.gtid_domain_id" )) || |
2434 | !(master_res= mysql_store_result(mysql)) || |
2435 | !(master_row= mysql_fetch_row(master_res))) |
2436 | { |
2437 | err_code= mysql_errno(mysql); |
2438 | if (is_network_error(err_code)) |
2439 | { |
2440 | mi->report(ERROR_LEVEL, err_code, NULL, |
2441 | "Get master @@GLOBAL.gtid_domain_id failed with error: %s" , |
2442 | mysql_error(mysql)); |
2443 | goto network_err; |
2444 | } |
2445 | else |
2446 | { |
2447 | errmsg= "The slave I/O thread stops because master does not support " |
2448 | "MariaDB global transaction id. A fatal error is encountered when " |
2449 | "it tries to SELECT @@GLOBAL.gtid_domain_id." ; |
2450 | sprintf(err_buff, "%s Error: %s" , errmsg, mysql_error(mysql)); |
2451 | goto err; |
2452 | } |
2453 | } |
2454 | mysql_free_result(master_res); |
2455 | master_res= NULL; |
2456 | |
2457 | query_str.append(STRING_WITH_LEN("SET @slave_connect_state='" ), |
2458 | system_charset_info); |
2459 | if (mi->gtid_current_pos.append_to_string(&query_str)) |
2460 | { |
2461 | err_code= ER_OUTOFMEMORY; |
2462 | errmsg= "The slave I/O thread stops because a fatal out-of-memory " |
2463 | "error is encountered when it tries to compute @slave_connect_state." ; |
2464 | sprintf(err_buff, "%s Error: Out of memory" , errmsg); |
2465 | goto err; |
2466 | } |
2467 | query_str.append(STRING_WITH_LEN("'" ), system_charset_info); |
2468 | |
2469 | rc= mysql_real_query(mysql, query_str.ptr(), query_str.length()); |
2470 | if (unlikely(rc)) |
2471 | { |
2472 | err_code= mysql_errno(mysql); |
2473 | if (is_network_error(err_code)) |
2474 | { |
2475 | mi->report(ERROR_LEVEL, err_code, NULL, |
2476 | "Setting @slave_connect_state failed with error: %s" , |
2477 | mysql_error(mysql)); |
2478 | goto network_err; |
2479 | } |
2480 | else |
2481 | { |
2482 | /* Fatal error */ |
2483 | errmsg= "The slave I/O thread stops because a fatal error is " |
2484 | "encountered when it tries to set @slave_connect_state." ; |
2485 | sprintf(err_buff, "%s Error: %s" , errmsg, mysql_error(mysql)); |
2486 | goto err; |
2487 | } |
2488 | } |
2489 | |
2490 | query_str.length(0); |
2491 | if (query_str.append(STRING_WITH_LEN("SET @slave_gtid_strict_mode=" ), |
2492 | system_charset_info) || |
2493 | query_str.append_ulonglong(opt_gtid_strict_mode != false)) |
2494 | { |
2495 | err_code= ER_OUTOFMEMORY; |
2496 | errmsg= "The slave I/O thread stops because a fatal out-of-memory " |
2497 | "error is encountered when it tries to set @slave_gtid_strict_mode." ; |
2498 | sprintf(err_buff, "%s Error: Out of memory" , errmsg); |
2499 | goto err; |
2500 | } |
2501 | |
2502 | rc= mysql_real_query(mysql, query_str.ptr(), query_str.length()); |
2503 | if (unlikely(rc)) |
2504 | { |
2505 | err_code= mysql_errno(mysql); |
2506 | if (is_network_error(err_code)) |
2507 | { |
2508 | mi->report(ERROR_LEVEL, err_code, NULL, |
2509 | "Setting @slave_gtid_strict_mode failed with error: %s" , |
2510 | mysql_error(mysql)); |
2511 | goto network_err; |
2512 | } |
2513 | else |
2514 | { |
2515 | /* Fatal error */ |
2516 | errmsg= "The slave I/O thread stops because a fatal error is " |
2517 | "encountered when it tries to set @slave_gtid_strict_mode." ; |
2518 | sprintf(err_buff, "%s Error: %s" , errmsg, mysql_error(mysql)); |
2519 | goto err; |
2520 | } |
2521 | } |
2522 | |
2523 | query_str.length(0); |
2524 | if (query_str.append(STRING_WITH_LEN("SET @slave_gtid_ignore_duplicates=" ), |
2525 | system_charset_info) || |
2526 | query_str.append_ulonglong(opt_gtid_ignore_duplicates != false)) |
2527 | { |
2528 | err_code= ER_OUTOFMEMORY; |
2529 | errmsg= "The slave I/O thread stops because a fatal out-of-memory error " |
2530 | "is encountered when it tries to set @slave_gtid_ignore_duplicates." ; |
2531 | sprintf(err_buff, "%s Error: Out of memory" , errmsg); |
2532 | goto err; |
2533 | } |
2534 | |
2535 | rc= mysql_real_query(mysql, query_str.ptr(), query_str.length()); |
2536 | if (unlikely(rc)) |
2537 | { |
2538 | err_code= mysql_errno(mysql); |
2539 | if (is_network_error(err_code)) |
2540 | { |
2541 | mi->report(ERROR_LEVEL, err_code, NULL, |
2542 | "Setting @slave_gtid_ignore_duplicates failed with " |
2543 | "error: %s" , mysql_error(mysql)); |
2544 | goto network_err; |
2545 | } |
2546 | else |
2547 | { |
2548 | /* Fatal error */ |
2549 | errmsg= "The slave I/O thread stops because a fatal error is " |
2550 | "encountered when it tries to set @slave_gtid_ignore_duplicates." ; |
2551 | sprintf(err_buff, "%s Error: %s" , errmsg, mysql_error(mysql)); |
2552 | goto err; |
2553 | } |
2554 | } |
2555 | |
2556 | if (mi->rli.until_condition == Relay_log_info::UNTIL_GTID) |
2557 | { |
2558 | query_str.length(0); |
2559 | query_str.append(STRING_WITH_LEN("SET @slave_until_gtid='" ), |
2560 | system_charset_info); |
2561 | if (mi->rli.until_gtid_pos.append_to_string(&query_str)) |
2562 | { |
2563 | err_code= ER_OUTOFMEMORY; |
2564 | errmsg= "The slave I/O thread stops because a fatal out-of-memory " |
2565 | "error is encountered when it tries to compute @slave_until_gtid." ; |
2566 | sprintf(err_buff, "%s Error: Out of memory" , errmsg); |
2567 | goto err; |
2568 | } |
2569 | query_str.append(STRING_WITH_LEN("'" ), system_charset_info); |
2570 | |
2571 | rc= mysql_real_query(mysql, query_str.ptr(), query_str.length()); |
2572 | if (unlikely(rc)) |
2573 | { |
2574 | err_code= mysql_errno(mysql); |
2575 | if (is_network_error(err_code)) |
2576 | { |
2577 | mi->report(ERROR_LEVEL, err_code, NULL, |
2578 | "Setting @slave_until_gtid failed with error: %s" , |
2579 | mysql_error(mysql)); |
2580 | goto network_err; |
2581 | } |
2582 | else |
2583 | { |
2584 | /* Fatal error */ |
2585 | errmsg= "The slave I/O thread stops because a fatal error is " |
2586 | "encountered when it tries to set @slave_until_gtid." ; |
2587 | sprintf(err_buff, "%s Error: %s" , errmsg, mysql_error(mysql)); |
2588 | goto err; |
2589 | } |
2590 | } |
2591 | } |
2592 | } |
2593 | else |
2594 | { |
2595 | /* |
2596 | If we are not using GTID to connect this time, then instead request |
2597 | the corresponding GTID position from the master, so that the user |
2598 | can reconnect the next time using MASTER_GTID_POS=AUTO. |
2599 | */ |
2600 | char quote_buf[2*sizeof(mi->master_log_name)+1]; |
2601 | char str_buf[28+2*sizeof(mi->master_log_name)+10]; |
2602 | String query(str_buf, sizeof(str_buf), system_charset_info); |
2603 | query.length(0); |
2604 | |
2605 | query.append("SELECT binlog_gtid_pos('" ); |
2606 | escape_quotes_for_mysql(&my_charset_bin, quote_buf, sizeof(quote_buf), |
2607 | mi->master_log_name, strlen(mi->master_log_name)); |
2608 | query.append(quote_buf); |
2609 | query.append("'," ); |
2610 | query.append_ulonglong(mi->master_log_pos); |
2611 | query.append(")" ); |
2612 | |
2613 | if (!mysql_real_query(mysql, query.c_ptr_safe(), query.length()) && |
2614 | (master_res= mysql_store_result(mysql)) && |
2615 | (master_row= mysql_fetch_row(master_res)) && |
2616 | (master_row[0] != NULL)) |
2617 | { |
2618 | rpl_global_gtid_slave_state->load(mi->io_thd, master_row[0], |
2619 | strlen(master_row[0]), false, false); |
2620 | } |
2621 | else if (check_io_slave_killed(mi, NULL)) |
2622 | goto slave_killed_err; |
2623 | else if (is_network_error(mysql_errno(mysql))) |
2624 | { |
2625 | mi->report(WARNING_LEVEL, mysql_errno(mysql), NULL, |
2626 | "Get master GTID position failed with error: %s" , mysql_error(mysql)); |
2627 | goto network_err; |
2628 | } |
2629 | else |
2630 | { |
2631 | /* |
2632 | ToDo: If the master does not have the binlog_gtid_pos() function, it |
2633 | just means that it is an old master with no GTID support, so we should |
2634 | do nothing. |
2635 | |
2636 | However, if binlog_gtid_pos() exists, but fails or returns NULL, then |
2637 | it means that the requested position is not valid. We could use this |
2638 | to catch attempts to replicate from within the middle of an event, |
2639 | avoiding strange failures or possible corruption. |
2640 | */ |
2641 | } |
2642 | if (master_res) |
2643 | { |
2644 | mysql_free_result(master_res); |
2645 | master_res= NULL; |
2646 | } |
2647 | } |
2648 | |
2649 | err: |
2650 | if (errmsg) |
2651 | { |
2652 | if (master_res) |
2653 | mysql_free_result(master_res); |
2654 | DBUG_ASSERT(err_code != 0); |
2655 | mi->report(ERROR_LEVEL, err_code, NULL, "%s" , err_buff); |
2656 | DBUG_RETURN(1); |
2657 | } |
2658 | |
2659 | DBUG_RETURN(0); |
2660 | |
2661 | network_err: |
2662 | if (master_res) |
2663 | mysql_free_result(master_res); |
2664 | DBUG_RETURN(2); |
2665 | |
2666 | slave_killed_err: |
2667 | if (master_res) |
2668 | mysql_free_result(master_res); |
2669 | DBUG_RETURN(2); |
2670 | } |
2671 | |
2672 | |
2673 | static bool wait_for_relay_log_space(Relay_log_info* rli) |
2674 | { |
2675 | bool slave_killed=0; |
2676 | bool ignore_log_space_limit; |
2677 | Master_info* mi = rli->mi; |
2678 | PSI_stage_info old_stage; |
2679 | THD* thd = mi->io_thd; |
2680 | DBUG_ENTER("wait_for_relay_log_space" ); |
2681 | |
2682 | mysql_mutex_lock(&rli->log_space_lock); |
2683 | thd->ENTER_COND(&rli->log_space_cond, |
2684 | &rli->log_space_lock, |
2685 | &stage_waiting_for_relay_log_space, |
2686 | &old_stage); |
2687 | while (rli->log_space_limit < rli->log_space_total && |
2688 | !(slave_killed=io_slave_killed(mi)) && |
2689 | !rli->ignore_log_space_limit) |
2690 | mysql_cond_wait(&rli->log_space_cond, &rli->log_space_lock); |
2691 | |
2692 | ignore_log_space_limit= rli->ignore_log_space_limit; |
2693 | rli->ignore_log_space_limit= 0; |
2694 | |
2695 | thd->EXIT_COND(&old_stage); |
2696 | |
2697 | /* |
2698 | Makes the IO thread read only one event at a time |
2699 | until the SQL thread is able to purge the relay |
2700 | logs, freeing some space. |
2701 | |
2702 | Therefore, once the SQL thread processes this next |
2703 | event, it goes to sleep (no more events in the queue), |
2704 | sets ignore_log_space_limit=true and wakes the IO thread. |
2705 | However, this event may have been enough already for |
2706 | the SQL thread to purge some log files, freeing |
2707 | rli->log_space_total . |
2708 | |
2709 | This guarantees that the SQL and IO thread move |
2710 | forward only one event at a time (to avoid deadlocks), |
2711 | when the relay space limit is reached. It also |
2712 | guarantees that when the SQL thread is prepared to |
2713 | rotate (to be able to purge some logs), the IO thread |
2714 | will know about it and will rotate. |
2715 | |
2716 | NOTE: The ignore_log_space_limit is only set when the SQL |
2717 | thread sleeps waiting for events. |
2718 | |
2719 | */ |
2720 | |
2721 | if (ignore_log_space_limit) |
2722 | { |
2723 | #ifndef DBUG_OFF |
2724 | { |
2725 | DBUG_PRINT("info" , ("log_space_limit=%llu log_space_total=%llu " |
2726 | "ignore_log_space_limit=%d " |
2727 | "sql_force_rotate_relay=%d" , |
2728 | rli->log_space_limit, rli->log_space_total, |
2729 | (int) rli->ignore_log_space_limit, |
2730 | (int) rli->sql_force_rotate_relay)); |
2731 | } |
2732 | #endif |
2733 | if (rli->sql_force_rotate_relay) |
2734 | { |
2735 | mysql_mutex_lock(&mi->data_lock); |
2736 | rotate_relay_log(rli->mi); |
2737 | mysql_mutex_unlock(&mi->data_lock); |
2738 | rli->sql_force_rotate_relay= false; |
2739 | } |
2740 | } |
2741 | |
2742 | DBUG_RETURN(slave_killed); |
2743 | } |
2744 | |
2745 | |
2746 | /* |
2747 | Builds a Rotate from the ignored events' info and writes it to relay log. |
2748 | |
2749 | SYNOPSIS |
2750 | write_ignored_events_info_to_relay_log() |
2751 | thd pointer to I/O thread's thd |
2752 | mi |
2753 | |
2754 | DESCRIPTION |
2755 | Slave I/O thread, going to die, must leave a durable trace of the |
2756 | ignored events' end position for the use of the slave SQL thread, by |
2757 | calling this function. Only that thread can call it (see assertion). |
2758 | */ |
2759 | static void write_ignored_events_info_to_relay_log(THD *thd, Master_info *mi) |
2760 | { |
2761 | Relay_log_info *rli= &mi->rli; |
2762 | mysql_mutex_t *log_lock= rli->relay_log.get_log_lock(); |
2763 | DBUG_ENTER("write_ignored_events_info_to_relay_log" ); |
2764 | |
2765 | DBUG_ASSERT(thd == mi->io_thd); |
2766 | mysql_mutex_lock(log_lock); |
2767 | if (rli->ign_master_log_name_end[0] || rli->ign_gtids.count()) |
2768 | { |
2769 | Rotate_log_event *rev= NULL; |
2770 | Gtid_list_log_event *glev= NULL; |
2771 | if (rli->ign_master_log_name_end[0]) |
2772 | { |
2773 | rev= new Rotate_log_event(rli->ign_master_log_name_end, |
2774 | 0, rli->ign_master_log_pos_end, |
2775 | Rotate_log_event::DUP_NAME); |
2776 | rli->ign_master_log_name_end[0]= 0; |
2777 | if (unlikely(!(bool)rev)) |
2778 | mi->report(ERROR_LEVEL, ER_SLAVE_CREATE_EVENT_FAILURE, NULL, |
2779 | ER_THD(thd, ER_SLAVE_CREATE_EVENT_FAILURE), |
2780 | "Rotate_event (out of memory?)," |
2781 | " SHOW SLAVE STATUS may be inaccurate" ); |
2782 | } |
2783 | if (rli->ign_gtids.count()) |
2784 | { |
2785 | DBUG_ASSERT(!rli->is_in_group()); // Ensure no active transaction |
2786 | glev= new Gtid_list_log_event(&rli->ign_gtids, |
2787 | Gtid_list_log_event::FLAG_IGN_GTIDS); |
2788 | rli->ign_gtids.reset(); |
2789 | if (unlikely(!(bool)glev)) |
2790 | mi->report(ERROR_LEVEL, ER_SLAVE_CREATE_EVENT_FAILURE, NULL, |
2791 | ER_THD(thd, ER_SLAVE_CREATE_EVENT_FAILURE), |
2792 | "Gtid_list_event (out of memory?)," |
2793 | " gtid_slave_pos may be inaccurate" ); |
2794 | } |
2795 | |
2796 | /* Can unlock before writing as slave SQL thd will soon see our event. */ |
2797 | mysql_mutex_unlock(log_lock); |
2798 | if (rev) |
2799 | { |
2800 | DBUG_PRINT("info" ,("writing a Rotate event to track down ignored events" )); |
2801 | rev->server_id= 0; // don't be ignored by slave SQL thread |
2802 | if (unlikely(rli->relay_log.append(rev))) |
2803 | mi->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_WRITE_FAILURE, NULL, |
2804 | ER_THD(thd, ER_SLAVE_RELAY_LOG_WRITE_FAILURE), |
2805 | "failed to write a Rotate event" |
2806 | " to the relay log, SHOW SLAVE STATUS may be" |
2807 | " inaccurate" ); |
2808 | delete rev; |
2809 | } |
2810 | if (glev) |
2811 | { |
2812 | DBUG_PRINT("info" ,("writing a Gtid_list event to track down ignored events" )); |
2813 | glev->server_id= 0; // don't be ignored by slave SQL thread |
2814 | glev->set_artificial_event(); // Don't mess up Exec_Master_Log_Pos |
2815 | if (unlikely(rli->relay_log.append(glev))) |
2816 | mi->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_WRITE_FAILURE, NULL, |
2817 | ER_THD(thd, ER_SLAVE_RELAY_LOG_WRITE_FAILURE), |
2818 | "failed to write a Gtid_list event to the relay log, " |
2819 | "gtid_slave_pos may be inaccurate" ); |
2820 | delete glev; |
2821 | } |
2822 | if (likely (rev || glev)) |
2823 | { |
2824 | rli->relay_log.harvest_bytes_written(&rli->log_space_total); |
2825 | if (flush_master_info(mi, TRUE, TRUE)) |
2826 | sql_print_error("Failed to flush master info file" ); |
2827 | } |
2828 | } |
2829 | else |
2830 | mysql_mutex_unlock(log_lock); |
2831 | DBUG_VOID_RETURN; |
2832 | } |
2833 | |
2834 | |
2835 | int register_slave_on_master(MYSQL* mysql, Master_info *mi, |
2836 | bool *suppress_warnings) |
2837 | { |
2838 | uchar buf[1024], *pos= buf; |
2839 | size_t report_host_len=0, report_user_len=0, report_password_len=0; |
2840 | DBUG_ENTER("register_slave_on_master" ); |
2841 | |
2842 | *suppress_warnings= FALSE; |
2843 | if (report_host) |
2844 | report_host_len= strlen(report_host); |
2845 | if (report_host_len > HOSTNAME_LENGTH) |
2846 | { |
2847 | sql_print_warning("The length of report_host is %zu. " |
2848 | "It is larger than the max length(%d), so this " |
2849 | "slave cannot be registered to the master." , |
2850 | report_host_len, HOSTNAME_LENGTH); |
2851 | DBUG_RETURN(0); |
2852 | } |
2853 | |
2854 | if (report_user) |
2855 | report_user_len= strlen(report_user); |
2856 | if (report_user_len > USERNAME_LENGTH) |
2857 | { |
2858 | sql_print_warning("The length of report_user is %zu. " |
2859 | "It is larger than the max length(%d), so this " |
2860 | "slave cannot be registered to the master." , |
2861 | report_user_len, USERNAME_LENGTH); |
2862 | DBUG_RETURN(0); |
2863 | } |
2864 | |
2865 | if (report_password) |
2866 | report_password_len= strlen(report_password); |
2867 | if (report_password_len > MAX_PASSWORD_LENGTH) |
2868 | { |
2869 | sql_print_warning("The length of report_password is %zu. " |
2870 | "It is larger than the max length(%d), so this " |
2871 | "slave cannot be registered to the master." , |
2872 | report_password_len, MAX_PASSWORD_LENGTH); |
2873 | DBUG_RETURN(0); |
2874 | } |
2875 | |
2876 | int4store(pos, global_system_variables.server_id); pos+= 4; |
2877 | pos= net_store_data(pos, (uchar*) report_host, report_host_len); |
2878 | pos= net_store_data(pos, (uchar*) report_user, report_user_len); |
2879 | pos= net_store_data(pos, (uchar*) report_password, report_password_len); |
2880 | int2store(pos, (uint16) report_port); pos+= 2; |
2881 | /* |
2882 | Fake rpl_recovery_rank, which was removed in BUG#13963, |
2883 | so that this server can register itself on old servers, |
2884 | see BUG#49259. |
2885 | */ |
2886 | int4store(pos, /* rpl_recovery_rank */ 0); pos+= 4; |
2887 | /* The master will fill in master_id */ |
2888 | int4store(pos, 0); pos+= 4; |
2889 | |
2890 | if (simple_command(mysql, COM_REGISTER_SLAVE, buf, (ulong) (pos- buf), 0)) |
2891 | { |
2892 | if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED) |
2893 | { |
2894 | *suppress_warnings= TRUE; // Suppress reconnect warning |
2895 | } |
2896 | else if (!check_io_slave_killed(mi, NULL)) |
2897 | { |
2898 | char buf[256]; |
2899 | my_snprintf(buf, sizeof(buf), "%s (Errno: %d)" , mysql_error(mysql), |
2900 | mysql_errno(mysql)); |
2901 | mi->report(ERROR_LEVEL, ER_SLAVE_MASTER_COM_FAILURE, NULL, |
2902 | ER(ER_SLAVE_MASTER_COM_FAILURE), "COM_REGISTER_SLAVE" , buf); |
2903 | } |
2904 | DBUG_RETURN(1); |
2905 | } |
2906 | DBUG_RETURN(0); |
2907 | } |
2908 | |
2909 | |
2910 | /** |
2911 | Execute a SHOW SLAVE STATUS statement. |
2912 | |
2913 | @param thd Pointer to THD object for the client thread executing the |
2914 | statement. |
2915 | |
2916 | @param mi Pointer to Master_info object for the IO thread. |
2917 | |
2918 | @retval FALSE success |
2919 | @retval TRUE failure |
2920 | */ |
2921 | |
2922 | bool show_master_info(THD *thd, Master_info *mi, bool full) |
2923 | { |
2924 | DBUG_ENTER("show_master_info" ); |
2925 | String gtid_pos; |
2926 | List<Item> field_list; |
2927 | |
2928 | if (full && rpl_global_gtid_slave_state->tostring(>id_pos, NULL, 0)) |
2929 | DBUG_RETURN(TRUE); |
2930 | show_master_info_get_fields(thd, &field_list, full, gtid_pos.length()); |
2931 | if (thd->protocol->send_result_set_metadata(&field_list, |
2932 | Protocol::SEND_NUM_ROWS | Protocol::SEND_EOF)) |
2933 | DBUG_RETURN(TRUE); |
2934 | if (send_show_master_info_data(thd, mi, full, >id_pos)) |
2935 | DBUG_RETURN(TRUE); |
2936 | my_eof(thd); |
2937 | DBUG_RETURN(FALSE); |
2938 | } |
2939 | |
2940 | void show_master_info_get_fields(THD *thd, List<Item> *field_list, |
2941 | bool full, size_t gtid_pos_length) |
2942 | { |
2943 | Master_info *mi; |
2944 | MEM_ROOT *mem_root= thd->mem_root; |
2945 | DBUG_ENTER("show_master_info_get_fields" ); |
2946 | |
2947 | if (full) |
2948 | { |
2949 | field_list->push_back(new (mem_root) |
2950 | Item_empty_string(thd, "Connection_name" , |
2951 | MAX_CONNECTION_NAME), |
2952 | mem_root); |
2953 | field_list->push_back(new (mem_root) |
2954 | Item_empty_string(thd, "Slave_SQL_State" , 30), |
2955 | mem_root); |
2956 | } |
2957 | |
2958 | field_list->push_back(new (mem_root) |
2959 | Item_empty_string(thd, "Slave_IO_State" , 30), |
2960 | mem_root); |
2961 | field_list->push_back(new (mem_root) |
2962 | Item_empty_string(thd, "Master_Host" , sizeof(mi->host)), |
2963 | mem_root); |
2964 | field_list->push_back(new (mem_root) |
2965 | Item_empty_string(thd, "Master_User" , sizeof(mi->user)), |
2966 | mem_root); |
2967 | field_list->push_back(new (mem_root) |
2968 | Item_return_int(thd, "Master_Port" , 7, MYSQL_TYPE_LONG), |
2969 | mem_root); |
2970 | field_list->push_back(new (mem_root) |
2971 | Item_return_int(thd, "Connect_Retry" , 10, |
2972 | MYSQL_TYPE_LONG), |
2973 | mem_root); |
2974 | field_list->push_back(new (mem_root) |
2975 | Item_empty_string(thd, "Master_Log_File" , FN_REFLEN), |
2976 | mem_root); |
2977 | field_list->push_back(new (mem_root) |
2978 | Item_return_int(thd, "Read_Master_Log_Pos" , 10, |
2979 | MYSQL_TYPE_LONGLONG), |
2980 | mem_root); |
2981 | field_list->push_back(new (mem_root) |
2982 | Item_empty_string(thd, "Relay_Log_File" , FN_REFLEN), |
2983 | mem_root); |
2984 | field_list->push_back(new (mem_root) |
2985 | Item_return_int(thd, "Relay_Log_Pos" , 10, |
2986 | MYSQL_TYPE_LONGLONG), |
2987 | mem_root); |
2988 | field_list->push_back(new (mem_root) |
2989 | Item_empty_string(thd, "Relay_Master_Log_File" , |
2990 | FN_REFLEN), |
2991 | mem_root); |
2992 | field_list->push_back(new (mem_root) |
2993 | Item_empty_string(thd, "Slave_IO_Running" , 3), |
2994 | mem_root); |
2995 | field_list->push_back(new (mem_root) |
2996 | Item_empty_string(thd, "Slave_SQL_Running" , 3), |
2997 | mem_root); |
2998 | field_list->push_back(new (mem_root) |
2999 | Item_empty_string(thd, "Replicate_Do_DB" , 20), |
3000 | mem_root); |
3001 | field_list->push_back(new (mem_root) |
3002 | Item_empty_string(thd, "Replicate_Ignore_DB" , 20), |
3003 | mem_root); |
3004 | field_list->push_back(new (mem_root) |
3005 | Item_empty_string(thd, "Replicate_Do_Table" , 20), |
3006 | mem_root); |
3007 | field_list->push_back(new (mem_root) |
3008 | Item_empty_string(thd, "Replicate_Ignore_Table" , 23), |
3009 | mem_root); |
3010 | field_list->push_back(new (mem_root) |
3011 | Item_empty_string(thd, "Replicate_Wild_Do_Table" , 24), |
3012 | mem_root); |
3013 | field_list->push_back(new (mem_root) |
3014 | Item_empty_string(thd, "Replicate_Wild_Ignore_Table" , |
3015 | 28), |
3016 | mem_root); |
3017 | field_list->push_back(new (mem_root) |
3018 | Item_return_int(thd, "Last_Errno" , 4, MYSQL_TYPE_LONG), |
3019 | mem_root); |
3020 | field_list->push_back(new (mem_root) |
3021 | Item_empty_string(thd, "Last_Error" , 20), |
3022 | mem_root); |
3023 | field_list->push_back(new (mem_root) |
3024 | Item_return_int(thd, "Skip_Counter" , 10, |
3025 | MYSQL_TYPE_LONG), |
3026 | mem_root); |
3027 | field_list->push_back(new (mem_root) |
3028 | Item_return_int(thd, "Exec_Master_Log_Pos" , 10, |
3029 | MYSQL_TYPE_LONGLONG), |
3030 | mem_root); |
3031 | field_list->push_back(new (mem_root) |
3032 | Item_return_int(thd, "Relay_Log_Space" , 10, |
3033 | MYSQL_TYPE_LONGLONG), |
3034 | mem_root); |
3035 | field_list->push_back(new (mem_root) |
3036 | Item_empty_string(thd, "Until_Condition" , 6), |
3037 | mem_root); |
3038 | field_list->push_back(new (mem_root) |
3039 | Item_empty_string(thd, "Until_Log_File" , FN_REFLEN), |
3040 | mem_root); |
3041 | field_list->push_back(new (mem_root) |
3042 | Item_return_int(thd, "Until_Log_Pos" , 10, |
3043 | MYSQL_TYPE_LONGLONG), |
3044 | mem_root); |
3045 | field_list->push_back(new (mem_root) |
3046 | Item_empty_string(thd, "Master_SSL_Allowed" , 7), |
3047 | mem_root); |
3048 | field_list->push_back(new (mem_root) |
3049 | Item_empty_string(thd, "Master_SSL_CA_File" , |
3050 | sizeof(mi->ssl_ca)), |
3051 | mem_root); |
3052 | field_list->push_back(new (mem_root) |
3053 | Item_empty_string(thd, "Master_SSL_CA_Path" , |
3054 | sizeof(mi->ssl_capath)), |
3055 | mem_root); |
3056 | field_list->push_back(new (mem_root) |
3057 | Item_empty_string(thd, "Master_SSL_Cert" , |
3058 | sizeof(mi->ssl_cert)), |
3059 | mem_root); |
3060 | field_list->push_back(new (mem_root) |
3061 | Item_empty_string(thd, "Master_SSL_Cipher" , |
3062 | sizeof(mi->ssl_cipher)), |
3063 | mem_root); |
3064 | field_list->push_back(new (mem_root) |
3065 | Item_empty_string(thd, "Master_SSL_Key" , |
3066 | sizeof(mi->ssl_key)), |
3067 | mem_root); |
3068 | field_list->push_back(new (mem_root) |
3069 | Item_return_int(thd, "Seconds_Behind_Master" , 10, |
3070 | MYSQL_TYPE_LONGLONG), |
3071 | mem_root); |
3072 | field_list->push_back(new (mem_root) |
3073 | Item_empty_string(thd, "Master_SSL_Verify_Server_Cert" , |
3074 | 3), |
3075 | mem_root); |
3076 | field_list->push_back(new (mem_root) |
3077 | Item_return_int(thd, "Last_IO_Errno" , 4, |
3078 | MYSQL_TYPE_LONG), |
3079 | mem_root); |
3080 | field_list->push_back(new (mem_root) |
3081 | Item_empty_string(thd, "Last_IO_Error" , 20), |
3082 | mem_root); |
3083 | field_list->push_back(new (mem_root) |
3084 | Item_return_int(thd, "Last_SQL_Errno" , 4, |
3085 | MYSQL_TYPE_LONG), |
3086 | mem_root); |
3087 | field_list->push_back(new (mem_root) |
3088 | Item_empty_string(thd, "Last_SQL_Error" , 20), |
3089 | mem_root); |
3090 | field_list->push_back(new (mem_root) |
3091 | Item_empty_string(thd, "Replicate_Ignore_Server_Ids" , |
3092 | FN_REFLEN), |
3093 | mem_root); |
3094 | field_list->push_back(new (mem_root) |
3095 | Item_return_int(thd, "Master_Server_Id" , sizeof(ulong), |
3096 | MYSQL_TYPE_LONG), |
3097 | mem_root); |
3098 | field_list->push_back(new (mem_root) |
3099 | Item_empty_string(thd, "Master_SSL_Crl" , |
3100 | sizeof(mi->ssl_crl)), |
3101 | mem_root); |
3102 | field_list->push_back(new (mem_root) |
3103 | Item_empty_string(thd, "Master_SSL_Crlpath" , |
3104 | sizeof(mi->ssl_crlpath)), |
3105 | mem_root); |
3106 | field_list->push_back(new (mem_root) |
3107 | Item_empty_string(thd, "Using_Gtid" , |
3108 | sizeof("Current_Pos" )-1), |
3109 | mem_root); |
3110 | field_list->push_back(new (mem_root) |
3111 | Item_empty_string(thd, "Gtid_IO_Pos" , 30), |
3112 | mem_root); |
3113 | field_list->push_back(new (mem_root) |
3114 | Item_empty_string(thd, "Replicate_Do_Domain_Ids" , |
3115 | FN_REFLEN), |
3116 | mem_root); |
3117 | field_list->push_back(new (mem_root) |
3118 | Item_empty_string(thd, "Replicate_Ignore_Domain_Ids" , |
3119 | FN_REFLEN), |
3120 | mem_root); |
3121 | field_list->push_back(new (mem_root) |
3122 | Item_empty_string(thd, "Parallel_Mode" , |
3123 | sizeof("conservative" )-1), |
3124 | mem_root); |
3125 | field_list->push_back(new (mem_root) |
3126 | Item_return_int(thd, "SQL_Delay" , 10, |
3127 | MYSQL_TYPE_LONG)); |
3128 | field_list->push_back(new (mem_root) |
3129 | Item_return_int(thd, "SQL_Remaining_Delay" , 8, |
3130 | MYSQL_TYPE_LONG)); |
3131 | field_list->push_back(new (mem_root) |
3132 | Item_empty_string(thd, "Slave_SQL_Running_State" , |
3133 | 20)); |
3134 | field_list->push_back(new (mem_root) |
3135 | Item_return_int(thd, "Slave_DDL_Groups" , 20, |
3136 | MYSQL_TYPE_LONGLONG), |
3137 | mem_root); |
3138 | field_list->push_back(new (mem_root) |
3139 | Item_return_int(thd, "Slave_Non_Transactional_Groups" , 20, |
3140 | MYSQL_TYPE_LONGLONG), |
3141 | mem_root); |
3142 | field_list->push_back(new (mem_root) |
3143 | Item_return_int(thd, "Slave_Transactional_Groups" , 20, |
3144 | MYSQL_TYPE_LONGLONG), |
3145 | mem_root); |
3146 | |
3147 | if (full) |
3148 | { |
3149 | field_list->push_back(new (mem_root) |
3150 | Item_return_int(thd, "Retried_transactions" , 10, |
3151 | MYSQL_TYPE_LONG), |
3152 | mem_root); |
3153 | field_list->push_back(new (mem_root) |
3154 | Item_return_int(thd, "Max_relay_log_size" , 10, |
3155 | MYSQL_TYPE_LONGLONG), |
3156 | mem_root); |
3157 | field_list->push_back(new (mem_root) |
3158 | Item_return_int(thd, "Executed_log_entries" , 10, |
3159 | MYSQL_TYPE_LONG), |
3160 | mem_root); |
3161 | field_list->push_back(new (mem_root) |
3162 | Item_return_int(thd, "Slave_received_heartbeats" , 10, |
3163 | MYSQL_TYPE_LONG), |
3164 | mem_root); |
3165 | field_list->push_back(new (mem_root) |
3166 | Item_float(thd, "Slave_heartbeat_period" , 0.0, 3, 10), |
3167 | mem_root); |
3168 | field_list->push_back(new (mem_root) |
3169 | Item_empty_string(thd, "Gtid_Slave_Pos" , |
3170 | (uint)gtid_pos_length), |
3171 | mem_root); |
3172 | } |
3173 | DBUG_VOID_RETURN; |
3174 | } |
3175 | |
3176 | /* Text for Slave_IO_Running */ |
3177 | static const char *slave_running[]= { "No" , "Connecting" , "Preparing" , "Yes" }; |
3178 | |
3179 | static bool send_show_master_info_data(THD *thd, Master_info *mi, bool full, |
3180 | String *gtid_pos) |
3181 | { |
3182 | DBUG_ENTER("send_show_master_info_data" ); |
3183 | |
3184 | if (mi->host[0]) |
3185 | { |
3186 | DBUG_PRINT("info" ,("host is set: '%s'" , mi->host)); |
3187 | String *packet= &thd->packet; |
3188 | Protocol *protocol= thd->protocol; |
3189 | Rpl_filter *rpl_filter= mi->rpl_filter; |
3190 | StringBuffer<256> tmp; |
3191 | |
3192 | protocol->prepare_for_resend(); |
3193 | |
3194 | /* |
3195 | slave_running can be accessed without run_lock but not other |
3196 | non-volotile members like mi->io_thd, which is guarded by the mutex. |
3197 | */ |
3198 | if (full) |
3199 | protocol->store(mi->connection_name.str, mi->connection_name.length, |
3200 | &my_charset_bin); |
3201 | mysql_mutex_lock(&mi->run_lock); |
3202 | if (full) |
3203 | { |
3204 | /* |
3205 | Show what the sql driver replication thread is doing |
3206 | This is only meaningful if there is only one slave thread. |
3207 | */ |
3208 | protocol->store(mi->rli.sql_driver_thd ? |
3209 | mi->rli.sql_driver_thd->get_proc_info() : "" , |
3210 | &my_charset_bin); |
3211 | } |
3212 | protocol->store(mi->io_thd ? mi->io_thd->get_proc_info() : "" , &my_charset_bin); |
3213 | mysql_mutex_unlock(&mi->run_lock); |
3214 | |
3215 | mysql_mutex_lock(&mi->data_lock); |
3216 | mysql_mutex_lock(&mi->rli.data_lock); |
3217 | /* err_lock is to protect mi->last_error() */ |
3218 | mysql_mutex_lock(&mi->err_lock); |
3219 | /* err_lock is to protect mi->rli.last_error() */ |
3220 | mysql_mutex_lock(&mi->rli.err_lock); |
3221 | protocol->store(mi->host, &my_charset_bin); |
3222 | protocol->store(mi->user, &my_charset_bin); |
3223 | protocol->store((uint32) mi->port); |
3224 | protocol->store((uint32) mi->connect_retry); |
3225 | protocol->store(mi->master_log_name, &my_charset_bin); |
3226 | protocol->store((ulonglong) mi->master_log_pos); |
3227 | protocol->store(mi->rli.group_relay_log_name + |
3228 | dirname_length(mi->rli.group_relay_log_name), |
3229 | &my_charset_bin); |
3230 | protocol->store((ulonglong) mi->rli.group_relay_log_pos); |
3231 | protocol->store(mi->rli.group_master_log_name, &my_charset_bin); |
3232 | protocol->store(slave_running[mi->slave_running], &my_charset_bin); |
3233 | protocol->store(mi->rli.slave_running ? "Yes" :"No" , &my_charset_bin); |
3234 | protocol->store(rpl_filter->get_do_db()); |
3235 | protocol->store(rpl_filter->get_ignore_db()); |
3236 | |
3237 | rpl_filter->get_do_table(&tmp); |
3238 | protocol->store(&tmp); |
3239 | rpl_filter->get_ignore_table(&tmp); |
3240 | protocol->store(&tmp); |
3241 | rpl_filter->get_wild_do_table(&tmp); |
3242 | protocol->store(&tmp); |
3243 | rpl_filter->get_wild_ignore_table(&tmp); |
3244 | protocol->store(&tmp); |
3245 | |
3246 | protocol->store(mi->rli.last_error().number); |
3247 | protocol->store(mi->rli.last_error().message, &my_charset_bin); |
3248 | protocol->store((uint32) mi->rli.slave_skip_counter); |
3249 | protocol->store((ulonglong) mi->rli.group_master_log_pos); |
3250 | protocol->store((ulonglong) mi->rli.log_space_total); |
3251 | |
3252 | protocol->store( |
3253 | mi->rli.until_condition==Relay_log_info::UNTIL_NONE ? "None" : |
3254 | ( mi->rli.until_condition==Relay_log_info::UNTIL_MASTER_POS? "Master" : |
3255 | ( mi->rli.until_condition==Relay_log_info::UNTIL_RELAY_POS? "Relay" : |
3256 | "Gtid" )), &my_charset_bin); |
3257 | protocol->store(mi->rli.until_log_name, &my_charset_bin); |
3258 | protocol->store((ulonglong) mi->rli.until_log_pos); |
3259 | |
3260 | #ifdef HAVE_OPENSSL |
3261 | protocol->store(mi->ssl? "Yes" :"No" , &my_charset_bin); |
3262 | #else |
3263 | protocol->store(mi->ssl? "Ignored" :"No" , &my_charset_bin); |
3264 | #endif |
3265 | protocol->store(mi->ssl_ca, &my_charset_bin); |
3266 | protocol->store(mi->ssl_capath, &my_charset_bin); |
3267 | protocol->store(mi->ssl_cert, &my_charset_bin); |
3268 | protocol->store(mi->ssl_cipher, &my_charset_bin); |
3269 | protocol->store(mi->ssl_key, &my_charset_bin); |
3270 | |
3271 | /* |
3272 | Seconds_Behind_Master: if SQL thread is running and I/O thread is |
3273 | connected, we can compute it otherwise show NULL (i.e. unknown). |
3274 | */ |
3275 | if ((mi->slave_running == MYSQL_SLAVE_RUN_READING) && |
3276 | mi->rli.slave_running) |
3277 | { |
3278 | long time_diff; |
3279 | bool idle; |
3280 | time_t stamp= mi->rli.last_master_timestamp; |
3281 | |
3282 | if (!stamp) |
3283 | idle= true; |
3284 | else |
3285 | { |
3286 | idle= mi->rli.sql_thread_caught_up; |
3287 | if (mi->using_parallel() && idle && !mi->rli.parallel.workers_idle()) |
3288 | idle= false; |
3289 | } |
3290 | if (idle) |
3291 | time_diff= 0; |
3292 | else |
3293 | { |
3294 | time_diff= ((long)(time(0) - stamp) - mi->clock_diff_with_master); |
3295 | /* |
3296 | Apparently on some systems time_diff can be <0. Here are possible |
3297 | reasons related to MySQL: |
3298 | - the master is itself a slave of another master whose time is ahead. |
3299 | - somebody used an explicit SET TIMESTAMP on the master. |
3300 | Possible reason related to granularity-to-second of time functions |
3301 | (nothing to do with MySQL), which can explain a value of -1: |
3302 | assume the master's and slave's time are perfectly synchronized, and |
3303 | that at slave's connection time, when the master's timestamp is read, |
3304 | it is at the very end of second 1, and (a very short time later) when |
3305 | the slave's timestamp is read it is at the very beginning of second |
3306 | 2. Then the recorded value for master is 1 and the recorded value for |
3307 | slave is 2. At SHOW SLAVE STATUS time, assume that the difference |
3308 | between timestamp of slave and rli->last_master_timestamp is 0 |
3309 | (i.e. they are in the same second), then we get 0-(2-1)=-1 as a result. |
3310 | This confuses users, so we don't go below 0. |
3311 | |
3312 | last_master_timestamp == 0 (an "impossible" timestamp 1970) is a |
3313 | special marker to say "consider we have caught up". |
3314 | */ |
3315 | if (time_diff < 0) |
3316 | time_diff= 0; |
3317 | } |
3318 | protocol->store((longlong)time_diff); |
3319 | } |
3320 | else |
3321 | { |
3322 | protocol->store_null(); |
3323 | } |
3324 | protocol->store(mi->ssl_verify_server_cert? "Yes" :"No" , &my_charset_bin); |
3325 | |
3326 | // Last_IO_Errno |
3327 | protocol->store(mi->last_error().number); |
3328 | // Last_IO_Error |
3329 | protocol->store(mi->last_error().message, &my_charset_bin); |
3330 | // Last_SQL_Errno |
3331 | protocol->store(mi->rli.last_error().number); |
3332 | // Last_SQL_Error |
3333 | protocol->store(mi->rli.last_error().message, &my_charset_bin); |
3334 | // Replicate_Ignore_Server_Ids |
3335 | prot_store_ids(thd, &mi->ignore_server_ids); |
3336 | // Master_Server_id |
3337 | protocol->store((uint32) mi->master_id); |
3338 | // SQL_Delay |
3339 | // Master_Ssl_Crl |
3340 | protocol->store(mi->ssl_ca, &my_charset_bin); |
3341 | // Master_Ssl_Crlpath |
3342 | protocol->store(mi->ssl_capath, &my_charset_bin); |
3343 | // Using_Gtid |
3344 | protocol->store(mi->using_gtid_astext(mi->using_gtid), &my_charset_bin); |
3345 | // Gtid_IO_Pos |
3346 | { |
3347 | mi->gtid_current_pos.to_string(&tmp); |
3348 | protocol->store(tmp.ptr(), tmp.length(), &my_charset_bin); |
3349 | } |
3350 | |
3351 | // Replicate_Do_Domain_Ids & Replicate_Ignore_Domain_Ids |
3352 | mi->domain_id_filter.store_ids(thd); |
3353 | |
3354 | // Parallel_Mode |
3355 | { |
3356 | const char *mode_name= get_type(&slave_parallel_mode_typelib, |
3357 | mi->parallel_mode); |
3358 | protocol->store(mode_name, strlen(mode_name), &my_charset_bin); |
3359 | } |
3360 | |
3361 | protocol->store((uint32) mi->rli.get_sql_delay()); |
3362 | // SQL_Remaining_Delay |
3363 | // THD::proc_info is not protected by any lock, so we read it once |
3364 | // to ensure that we use the same value throughout this function. |
3365 | const char *slave_sql_running_state= |
3366 | mi->rli.sql_driver_thd ? mi->rli.sql_driver_thd->proc_info : "" ; |
3367 | if (slave_sql_running_state == Relay_log_info::state_delaying_string) |
3368 | { |
3369 | time_t t= my_time(0), sql_delay_end= mi->rli.get_sql_delay_end(); |
3370 | protocol->store((uint32)(t < sql_delay_end ? sql_delay_end - t : 0)); |
3371 | } |
3372 | else |
3373 | protocol->store_null(); |
3374 | // Slave_SQL_Running_State |
3375 | protocol->store(slave_sql_running_state, &my_charset_bin); |
3376 | |
3377 | uint64 events; |
3378 | events= (uint64)my_atomic_load64_explicit((volatile int64 *) |
3379 | &mi->total_ddl_groups, MY_MEMORY_ORDER_RELAXED); |
3380 | protocol->store(events); |
3381 | events= (uint64)my_atomic_load64_explicit((volatile int64 *) |
3382 | &mi->total_non_trans_groups, MY_MEMORY_ORDER_RELAXED); |
3383 | protocol->store(events); |
3384 | events= (uint64)my_atomic_load64_explicit((volatile int64 *) |
3385 | &mi->total_trans_groups, MY_MEMORY_ORDER_RELAXED); |
3386 | protocol->store(events); |
3387 | |
3388 | if (full) |
3389 | { |
3390 | protocol->store((uint32) mi->rli.retried_trans); |
3391 | protocol->store((ulonglong) mi->rli.max_relay_log_size); |
3392 | protocol->store((uint32) mi->rli.executed_entries); |
3393 | protocol->store((uint32) mi->received_heartbeats); |
3394 | protocol->store((double) mi->heartbeat_period, 3, &tmp); |
3395 | protocol->store(gtid_pos->ptr(), gtid_pos->length(), &my_charset_bin); |
3396 | } |
3397 | |
3398 | mysql_mutex_unlock(&mi->rli.err_lock); |
3399 | mysql_mutex_unlock(&mi->err_lock); |
3400 | mysql_mutex_unlock(&mi->rli.data_lock); |
3401 | mysql_mutex_unlock(&mi->data_lock); |
3402 | |
3403 | if (my_net_write(&thd->net, (uchar*) thd->packet.ptr(), packet->length())) |
3404 | DBUG_RETURN(TRUE); |
3405 | } |
3406 | DBUG_RETURN(FALSE); |
3407 | } |
3408 | |
3409 | |
3410 | /* Used to sort connections by name */ |
3411 | |
3412 | static int cmp_mi_by_name(const Master_info **arg1, |
3413 | const Master_info **arg2) |
3414 | { |
3415 | return my_strcasecmp(system_charset_info, (*arg1)->connection_name.str, |
3416 | (*arg2)->connection_name.str); |
3417 | } |
3418 | |
3419 | |
3420 | /** |
3421 | Execute a SHOW FULL SLAVE STATUS statement. |
3422 | |
3423 | @param thd Pointer to THD object for the client thread executing the |
3424 | statement. |
3425 | |
3426 | Elements are sorted according to the original connection_name. |
3427 | |
3428 | @retval FALSE success |
3429 | @retval TRUE failure |
3430 | |
3431 | @note |
3432 | master_info_index is protected by LOCK_active_mi. |
3433 | */ |
3434 | |
3435 | bool show_all_master_info(THD* thd) |
3436 | { |
3437 | uint i, elements; |
3438 | String gtid_pos; |
3439 | Master_info **tmp; |
3440 | List<Item> field_list; |
3441 | DBUG_ENTER("show_master_info" ); |
3442 | mysql_mutex_assert_owner(&LOCK_active_mi); |
3443 | |
3444 | gtid_pos.length(0); |
3445 | if (rpl_append_gtid_state(>id_pos, true)) |
3446 | { |
3447 | my_error(ER_OUT_OF_RESOURCES, MYF(0)); |
3448 | DBUG_RETURN(TRUE); |
3449 | } |
3450 | |
3451 | show_master_info_get_fields(thd, &field_list, 1, gtid_pos.length()); |
3452 | if (thd->protocol->send_result_set_metadata(&field_list, |
3453 | Protocol::SEND_NUM_ROWS | Protocol::SEND_EOF)) |
3454 | DBUG_RETURN(TRUE); |
3455 | |
3456 | if (!master_info_index || |
3457 | !(elements= master_info_index->master_info_hash.records)) |
3458 | goto end; |
3459 | |
3460 | /* |
3461 | Sort lines to get them into a predicted order |
3462 | (needed for test cases and to not confuse users) |
3463 | */ |
3464 | if (!(tmp= (Master_info**) thd->alloc(sizeof(Master_info*) * elements))) |
3465 | DBUG_RETURN(TRUE); |
3466 | |
3467 | for (i= 0; i < elements; i++) |
3468 | { |
3469 | tmp[i]= (Master_info *) my_hash_element(&master_info_index-> |
3470 | master_info_hash, i); |
3471 | } |
3472 | my_qsort(tmp, elements, sizeof(Master_info*), (qsort_cmp) cmp_mi_by_name); |
3473 | |
3474 | for (i= 0; i < elements; i++) |
3475 | { |
3476 | if (send_show_master_info_data(thd, tmp[i], 1, >id_pos)) |
3477 | DBUG_RETURN(TRUE); |
3478 | } |
3479 | |
3480 | end: |
3481 | my_eof(thd); |
3482 | DBUG_RETURN(FALSE); |
3483 | } |
3484 | |
3485 | |
3486 | void set_slave_thread_options(THD* thd) |
3487 | { |
3488 | DBUG_ENTER("set_slave_thread_options" ); |
3489 | /* |
3490 | It's nonsense to constrain the slave threads with max_join_size; if a |
3491 | query succeeded on master, we HAVE to execute it. So set |
3492 | OPTION_BIG_SELECTS. Setting max_join_size to HA_POS_ERROR is not enough |
3493 | (and it's not needed if we have OPTION_BIG_SELECTS) because an INSERT |
3494 | SELECT examining more than 4 billion rows would still fail (yes, because |
3495 | when max_join_size is 4G, OPTION_BIG_SELECTS is automatically set, but |
3496 | only for client threads. |
3497 | */ |
3498 | ulonglong options= thd->variables.option_bits | OPTION_BIG_SELECTS; |
3499 | if (opt_log_slave_updates) |
3500 | options|= OPTION_BIN_LOG; |
3501 | else |
3502 | options&= ~OPTION_BIN_LOG; |
3503 | thd->variables.option_bits= options; |
3504 | thd->variables.completion_type= 0; |
3505 | |
3506 | /* For easier test in LOGGER::log_command */ |
3507 | if (thd->variables.log_disabled_statements & LOG_DISABLE_SLAVE) |
3508 | thd->variables.option_bits|= OPTION_LOG_OFF; |
3509 | |
3510 | thd->variables.sql_log_slow= !MY_TEST(thd->variables.log_slow_disabled_statements & |
3511 | LOG_SLOW_DISABLE_SLAVE); |
3512 | DBUG_VOID_RETURN; |
3513 | } |
3514 | |
3515 | void set_slave_thread_default_charset(THD* thd, rpl_group_info *rgi) |
3516 | { |
3517 | DBUG_ENTER("set_slave_thread_default_charset" ); |
3518 | |
3519 | thd->variables.collation_server= |
3520 | global_system_variables.collation_server; |
3521 | thd->update_charset(global_system_variables.character_set_client, |
3522 | global_system_variables.collation_connection); |
3523 | |
3524 | thd->system_thread_info.rpl_sql_info->cached_charset_invalidate(); |
3525 | DBUG_VOID_RETURN; |
3526 | } |
3527 | |
3528 | /* |
3529 | init_slave_thread() |
3530 | */ |
3531 | |
3532 | static int init_slave_thread(THD* thd, Master_info *mi, |
3533 | SLAVE_THD_TYPE thd_type) |
3534 | { |
3535 | DBUG_ENTER("init_slave_thread" ); |
3536 | int simulate_error __attribute__((unused))= 0; |
3537 | DBUG_EXECUTE_IF("simulate_io_slave_error_on_init" , |
3538 | simulate_error|= (1 << SLAVE_THD_IO);); |
3539 | DBUG_EXECUTE_IF("simulate_sql_slave_error_on_init" , |
3540 | simulate_error|= (1 << SLAVE_THD_SQL);); |
3541 | |
3542 | thd->system_thread = (thd_type == SLAVE_THD_SQL) ? |
3543 | SYSTEM_THREAD_SLAVE_SQL : SYSTEM_THREAD_SLAVE_IO; |
3544 | thread_safe_increment32(&service_thread_count); |
3545 | |
3546 | /* We must call store_globals() before doing my_net_init() */ |
3547 | if (init_thr_lock() || thd->store_globals() || |
3548 | my_net_init(&thd->net, 0, thd, MYF(MY_THREAD_SPECIFIC)) || |
3549 | IF_DBUG(simulate_error & (1<< thd_type), 0)) |
3550 | { |
3551 | thd->cleanup(); |
3552 | DBUG_RETURN(-1); |
3553 | } |
3554 | |
3555 | thd->security_ctx->skip_grants(); |
3556 | thd->slave_thread= 1; |
3557 | thd->connection_name= mi->connection_name; |
3558 | thd->variables.sql_log_slow= !MY_TEST(thd->variables.log_slow_disabled_statements & LOG_SLOW_DISABLE_SLAVE); |
3559 | set_slave_thread_options(thd); |
3560 | |
3561 | if (thd_type == SLAVE_THD_SQL) |
3562 | THD_STAGE_INFO(thd, stage_waiting_for_the_next_event_in_relay_log); |
3563 | else |
3564 | THD_STAGE_INFO(thd, stage_waiting_for_master_update); |
3565 | thd->set_time(); |
3566 | /* Do not use user-supplied timeout value for system threads. */ |
3567 | thd->variables.lock_wait_timeout= LONG_TIMEOUT; |
3568 | DBUG_RETURN(0); |
3569 | } |
3570 | |
3571 | /* |
3572 | Sleep for a given amount of time or until killed. |
3573 | |
3574 | @param thd Thread context of the current thread. |
3575 | @param seconds The number of seconds to sleep. |
3576 | @param func Function object to check if the thread has been killed. |
3577 | @param info The Rpl_info object associated with this sleep. |
3578 | |
3579 | @retval True if the thread has been killed, false otherwise. |
3580 | */ |
3581 | template <typename killed_func, typename rpl_info> |
3582 | static bool slave_sleep(THD *thd, time_t seconds, |
3583 | killed_func func, rpl_info info) |
3584 | { |
3585 | |
3586 | bool ret; |
3587 | struct timespec abstime; |
3588 | |
3589 | mysql_mutex_t *lock= &info->sleep_lock; |
3590 | mysql_cond_t *cond= &info->sleep_cond; |
3591 | |
3592 | /* Absolute system time at which the sleep time expires. */ |
3593 | set_timespec(abstime, seconds); |
3594 | mysql_mutex_lock(lock); |
3595 | thd->ENTER_COND(cond, lock, NULL, NULL); |
3596 | |
3597 | while (! (ret= func(info))) |
3598 | { |
3599 | int error= mysql_cond_timedwait(cond, lock, &abstime); |
3600 | if (error == ETIMEDOUT || error == ETIME) |
3601 | break; |
3602 | } |
3603 | /* Implicitly unlocks the mutex. */ |
3604 | thd->EXIT_COND(NULL); |
3605 | return ret; |
3606 | } |
3607 | |
3608 | |
3609 | static int request_dump(THD *thd, MYSQL* mysql, Master_info* mi, |
3610 | bool *suppress_warnings) |
3611 | { |
3612 | uchar buf[FN_REFLEN + 10]; |
3613 | int len; |
3614 | ushort binlog_flags = 0; // for now |
3615 | char* logname = mi->master_log_name; |
3616 | DBUG_ENTER("request_dump" ); |
3617 | |
3618 | *suppress_warnings= FALSE; |
3619 | |
3620 | if (opt_log_slave_updates && opt_replicate_annotate_row_events) |
3621 | binlog_flags|= BINLOG_SEND_ANNOTATE_ROWS_EVENT; |
3622 | |
3623 | if (repl_semisync_slave.request_transmit(mi)) |
3624 | DBUG_RETURN(1); |
3625 | |
3626 | // TODO if big log files: Change next to int8store() |
3627 | int4store(buf, (ulong) mi->master_log_pos); |
3628 | int2store(buf + 4, binlog_flags); |
3629 | int4store(buf + 6, global_system_variables.server_id); |
3630 | len = (uint) strlen(logname); |
3631 | memcpy(buf + 10, logname,len); |
3632 | if (simple_command(mysql, COM_BINLOG_DUMP, buf, len + 10, 1)) |
3633 | { |
3634 | /* |
3635 | Something went wrong, so we will just reconnect and retry later |
3636 | in the future, we should do a better error analysis, but for |
3637 | now we just fill up the error log :-) |
3638 | */ |
3639 | if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED) |
3640 | *suppress_warnings= TRUE; // Suppress reconnect warning |
3641 | else |
3642 | sql_print_error("Error on COM_BINLOG_DUMP: %d %s, will retry in %d secs" , |
3643 | mysql_errno(mysql), mysql_error(mysql), |
3644 | mi->connect_retry); |
3645 | DBUG_RETURN(1); |
3646 | } |
3647 | |
3648 | DBUG_RETURN(0); |
3649 | } |
3650 | |
3651 | |
3652 | /* |
3653 | Read one event from the master |
3654 | |
3655 | SYNOPSIS |
3656 | read_event() |
3657 | mysql MySQL connection |
3658 | mi Master connection information |
3659 | suppress_warnings TRUE when a normal net read timeout has caused us to |
3660 | try a reconnect. We do not want to print anything to |
3661 | the error log in this case because this a anormal |
3662 | event in an idle server. |
3663 | network_read_len get the real network read length in VIO, especially using compressed protocol |
3664 | |
3665 | RETURN VALUES |
3666 | 'packet_error' Error |
3667 | number Length of packet |
3668 | */ |
3669 | |
3670 | static ulong read_event(MYSQL* mysql, Master_info *mi, bool* suppress_warnings, |
3671 | ulong* network_read_len) |
3672 | { |
3673 | ulong len; |
3674 | DBUG_ENTER("read_event" ); |
3675 | |
3676 | *suppress_warnings= FALSE; |
3677 | /* |
3678 | my_real_read() will time us out |
3679 | We check if we were told to die, and if not, try reading again |
3680 | */ |
3681 | #ifndef DBUG_OFF |
3682 | if (disconnect_slave_event_count && !(mi->events_till_disconnect--)) |
3683 | DBUG_RETURN(packet_error); |
3684 | #endif |
3685 | |
3686 | len = cli_safe_read_reallen(mysql, network_read_len); |
3687 | if (unlikely(len == packet_error || (long) len < 1)) |
3688 | { |
3689 | if (mysql_errno(mysql) == ER_NET_READ_INTERRUPTED) |
3690 | { |
3691 | /* |
3692 | We are trying a normal reconnect after a read timeout; |
3693 | we suppress prints to .err file as long as the reconnect |
3694 | happens without problems |
3695 | */ |
3696 | *suppress_warnings= |
3697 | global_system_variables.log_warnings < 2 ? TRUE : FALSE; |
3698 | } |
3699 | else |
3700 | { |
3701 | if (!mi->rli.abort_slave) |
3702 | { |
3703 | sql_print_error("Error reading packet from server: %s (server_errno=%d)" , |
3704 | mysql_error(mysql), mysql_errno(mysql)); |
3705 | } |
3706 | } |
3707 | DBUG_RETURN(packet_error); |
3708 | } |
3709 | |
3710 | /* Check if eof packet */ |
3711 | if (len < 8 && mysql->net.read_pos[0] == 254) |
3712 | { |
3713 | sql_print_information("Slave: received end packet from server, apparent " |
3714 | "master shutdown: %s" , |
3715 | mysql_error(mysql)); |
3716 | DBUG_RETURN(packet_error); |
3717 | } |
3718 | |
3719 | DBUG_PRINT("exit" , ("len: %lu net->read_pos[4]: %d" , |
3720 | len, mysql->net.read_pos[4])); |
3721 | DBUG_RETURN(len - 1); |
3722 | } |
3723 | |
3724 | |
3725 | /** |
3726 | Check if the current error is of temporary nature of not. |
3727 | Some errors are temporary in nature, such as |
3728 | ER_LOCK_DEADLOCK and ER_LOCK_WAIT_TIMEOUT. |
3729 | |
3730 | @retval 0 if fatal error |
3731 | @retval 1 temporary error, do retry |
3732 | */ |
3733 | |
3734 | int |
3735 | has_temporary_error(THD *thd) |
3736 | { |
3737 | uint current_errno; |
3738 | DBUG_ENTER("has_temporary_error" ); |
3739 | |
3740 | DBUG_EXECUTE_IF("all_errors_are_temporary_errors" , |
3741 | if (thd->get_stmt_da()->is_error()) |
3742 | { |
3743 | thd->clear_error(); |
3744 | my_error(ER_LOCK_DEADLOCK, MYF(0)); |
3745 | }); |
3746 | |
3747 | /* |
3748 | If there is no message in THD, we can't say if it's a temporary |
3749 | error or not. This is currently the case for Incident_log_event, |
3750 | which sets no message. Return FALSE. |
3751 | */ |
3752 | if (!likely(thd->is_error())) |
3753 | DBUG_RETURN(0); |
3754 | |
3755 | current_errno= thd->get_stmt_da()->sql_errno(); |
3756 | for (uint i= 0; i < slave_transaction_retry_error_length; i++) |
3757 | { |
3758 | if (current_errno == slave_transaction_retry_errors[i]) |
3759 | DBUG_RETURN(1); |
3760 | } |
3761 | |
3762 | DBUG_RETURN(0); |
3763 | } |
3764 | |
3765 | |
3766 | /** |
3767 | If this is a lagging slave (specified with CHANGE MASTER TO MASTER_DELAY = X), delays accordingly. Also unlocks rli->data_lock. |
3768 | |
3769 | Design note: this is the place to unlock rli->data_lock. The lock |
3770 | must be held when reading delay info from rli, but it should not be |
3771 | held while sleeping. |
3772 | |
3773 | @param ev Event that is about to be executed. |
3774 | |
3775 | @param thd The sql thread's THD object. |
3776 | |
3777 | @param rli The sql thread's Relay_log_info structure. |
3778 | |
3779 | @retval 0 If the delay timed out and the event shall be executed. |
3780 | |
3781 | @retval nonzero If the delay was interrupted and the event shall be skipped. |
3782 | */ |
3783 | int |
3784 | sql_delay_event(Log_event *ev, THD *thd, rpl_group_info *rgi) |
3785 | { |
3786 | Relay_log_info* rli= rgi->rli; |
3787 | long sql_delay= rli->get_sql_delay(); |
3788 | |
3789 | DBUG_ENTER("sql_delay_event" ); |
3790 | mysql_mutex_assert_owner(&rli->data_lock); |
3791 | DBUG_ASSERT(!rli->belongs_to_client()); |
3792 | |
3793 | int type= ev->get_type_code(); |
3794 | if (sql_delay && type != ROTATE_EVENT && |
3795 | type != FORMAT_DESCRIPTION_EVENT && type != START_EVENT_V3) |
3796 | { |
3797 | // The time when we should execute the event. |
3798 | time_t sql_delay_end= |
3799 | ev->when + rli->mi->clock_diff_with_master + sql_delay; |
3800 | // The current time. |
3801 | time_t now= my_time(0); |
3802 | // The time we will have to sleep before executing the event. |
3803 | unsigned long nap_time= 0; |
3804 | if (sql_delay_end > now) |
3805 | nap_time= (ulong)(sql_delay_end - now); |
3806 | |
3807 | DBUG_PRINT("info" , ("sql_delay= %lu " |
3808 | "ev->when= %lu " |
3809 | "rli->mi->clock_diff_with_master= %lu " |
3810 | "now= %ld " |
3811 | "sql_delay_end= %llu " |
3812 | "nap_time= %ld" , |
3813 | sql_delay, (long)ev->when, |
3814 | rli->mi->clock_diff_with_master, |
3815 | (long)now, (ulonglong)sql_delay_end, (long)nap_time)); |
3816 | |
3817 | if (sql_delay_end > now) |
3818 | { |
3819 | DBUG_PRINT("info" , ("delaying replication event %lu secs" , |
3820 | nap_time)); |
3821 | rli->start_sql_delay(sql_delay_end); |
3822 | mysql_mutex_unlock(&rli->data_lock); |
3823 | DBUG_RETURN(slave_sleep(thd, nap_time, sql_slave_killed, rgi)); |
3824 | } |
3825 | } |
3826 | |
3827 | mysql_mutex_unlock(&rli->data_lock); |
3828 | |
3829 | DBUG_RETURN(0); |
3830 | } |
3831 | |
3832 | |
3833 | /* |
3834 | First half of apply_event_and_update_pos(), see below. |
3835 | Setup some THD variables for applying the event. |
3836 | |
3837 | Split out so that it can run with rli->data_lock held in non-parallel |
3838 | replication, but without the mutex held in the parallel case. |
3839 | */ |
3840 | static int |
3841 | apply_event_and_update_pos_setup(Log_event* ev, THD* thd, rpl_group_info *rgi) |
3842 | { |
3843 | DBUG_ENTER("apply_event_and_update_pos_setup" ); |
3844 | |
3845 | DBUG_PRINT("exec_event" ,("%s(type_code: %d; server_id: %d)" , |
3846 | ev->get_type_str(), ev->get_type_code(), |
3847 | ev->server_id)); |
3848 | DBUG_PRINT("info" , ("thd->options: '%s%s%s' rgi->last_event_start_time: %lu" , |
3849 | FLAGSTR(thd->variables.option_bits, OPTION_NOT_AUTOCOMMIT), |
3850 | FLAGSTR(thd->variables.option_bits, OPTION_BEGIN), |
3851 | FLAGSTR(thd->variables.option_bits, OPTION_GTID_BEGIN), |
3852 | (ulong) rgi->last_event_start_time)); |
3853 | |
3854 | /* |
3855 | Execute the event to change the database and update the binary |
3856 | log coordinates, but first we set some data that is needed for |
3857 | the thread. |
3858 | |
3859 | The event will be executed unless it is supposed to be skipped. |
3860 | |
3861 | Queries originating from this server must be skipped. Low-level |
3862 | events (Format_description_log_event, Rotate_log_event, |
3863 | Stop_log_event) from this server must also be skipped. But for |
3864 | those we don't want to modify 'group_master_log_pos', because |
3865 | these events did not exist on the master. |
3866 | Format_description_log_event is not completely skipped. |
3867 | |
3868 | Skip queries specified by the user in 'slave_skip_counter'. We |
3869 | can't however skip events that has something to do with the log |
3870 | files themselves. |
3871 | |
3872 | Filtering on own server id is extremely important, to ignore |
3873 | execution of events created by the creation/rotation of the relay |
3874 | log (remember that now the relay log starts with its Format_desc, |
3875 | has a Rotate etc). |
3876 | */ |
3877 | |
3878 | /* Use the original server id for logging. */ |
3879 | thd->variables.server_id = ev->server_id; |
3880 | thd->set_time(); // time the query |
3881 | thd->lex->current_select= 0; |
3882 | if (!ev->when) |
3883 | { |
3884 | my_hrtime_t hrtime= my_hrtime(); |
3885 | ev->when= hrtime_to_my_time(hrtime); |
3886 | ev->when_sec_part= hrtime_sec_part(hrtime); |
3887 | } |
3888 | thd->variables.option_bits= |
3889 | (thd->variables.option_bits & ~OPTION_SKIP_REPLICATION) | |
3890 | (ev->flags & LOG_EVENT_SKIP_REPLICATION_F ? OPTION_SKIP_REPLICATION : 0); |
3891 | ev->thd = thd; // because up to this point, ev->thd == 0 |
3892 | |
3893 | DBUG_RETURN(ev->shall_skip(rgi)); |
3894 | } |
3895 | |
3896 | |
3897 | /* |
3898 | Second half of apply_event_and_update_pos(), see below. |
3899 | |
3900 | Do the actual event apply (or skip), and position update. |
3901 | */ |
3902 | static int |
3903 | apply_event_and_update_pos_apply(Log_event* ev, THD* thd, rpl_group_info *rgi, |
3904 | int reason) |
3905 | { |
3906 | int exec_res= 0; |
3907 | Relay_log_info* rli= rgi->rli; |
3908 | |
3909 | DBUG_ENTER("apply_event_and_update_pos_apply" ); |
3910 | DBUG_EXECUTE_IF("inject_slave_sql_before_apply_event" , |
3911 | { |
3912 | DBUG_ASSERT(!debug_sync_set_action |
3913 | (thd, STRING_WITH_LEN("now WAIT_FOR continue" ))); |
3914 | DBUG_SET_INITIAL("-d,inject_slave_sql_before_apply_event" ); |
3915 | };); |
3916 | if (reason == Log_event::EVENT_SKIP_NOT) |
3917 | exec_res= ev->apply_event(rgi); |
3918 | |
3919 | #ifdef WITH_WSREP |
3920 | if (exec_res && thd->wsrep_conflict_state != NO_CONFLICT) |
3921 | { |
3922 | WSREP_DEBUG("SQL apply failed, res %d conflict state: %d" , |
3923 | exec_res, thd->wsrep_conflict_state); |
3924 | rli->abort_slave= 1; |
3925 | rli->report(ERROR_LEVEL, ER_UNKNOWN_COM_ERROR, rgi->gtid_info(), |
3926 | "Node has dropped from cluster" ); |
3927 | } |
3928 | #endif |
3929 | |
3930 | #ifndef DBUG_OFF |
3931 | /* |
3932 | This only prints information to the debug trace. |
3933 | |
3934 | TODO: Print an informational message to the error log? |
3935 | */ |
3936 | static const char *const explain[] = { |
3937 | // EVENT_SKIP_NOT, |
3938 | "not skipped" , |
3939 | // EVENT_SKIP_IGNORE, |
3940 | "skipped because event should be ignored" , |
3941 | // EVENT_SKIP_COUNT |
3942 | "skipped because event skip counter was non-zero" |
3943 | }; |
3944 | DBUG_PRINT("info" , ("OPTION_BEGIN: %d IN_STMT: %d IN_TRANSACTION: %d" , |
3945 | MY_TEST(thd->variables.option_bits & OPTION_BEGIN), |
3946 | rli->get_flag(Relay_log_info::IN_STMT), |
3947 | rli->get_flag(Relay_log_info::IN_TRANSACTION))); |
3948 | DBUG_PRINT("skip_event" , ("%s event was %s" , |
3949 | ev->get_type_str(), explain[reason])); |
3950 | #endif |
3951 | |
3952 | DBUG_PRINT("info" , ("apply_event error = %d" , exec_res)); |
3953 | if (exec_res == 0) |
3954 | { |
3955 | int error= ev->update_pos(rgi); |
3956 | #ifndef DBUG_OFF |
3957 | DBUG_PRINT("info" , ("update_pos error = %d" , error)); |
3958 | if (!rli->belongs_to_client()) |
3959 | { |
3960 | DBUG_PRINT("info" , ("group %llu %s" , rli->group_relay_log_pos, |
3961 | rli->group_relay_log_name)); |
3962 | DBUG_PRINT("info" , ("event %llu %s" , rli->event_relay_log_pos, |
3963 | rli->event_relay_log_name)); |
3964 | } |
3965 | #endif |
3966 | /* |
3967 | The update should not fail, so print an error message and |
3968 | return an error code. |
3969 | |
3970 | TODO: Replace this with a decent error message when merged |
3971 | with BUG#24954 (which adds several new error message). |
3972 | */ |
3973 | if (unlikely(error)) |
3974 | { |
3975 | rli->report(ERROR_LEVEL, ER_UNKNOWN_ERROR, rgi->gtid_info(), |
3976 | "It was not possible to update the positions" |
3977 | " of the relay log information: the slave may" |
3978 | " be in an inconsistent state." |
3979 | " Stopped in %s position %llu" , |
3980 | rli->group_relay_log_name, rli->group_relay_log_pos); |
3981 | DBUG_RETURN(2); |
3982 | } |
3983 | } |
3984 | else |
3985 | { |
3986 | /* |
3987 | Make sure we do not erroneously update gtid_slave_pos with a lingering |
3988 | GTID from this failed event group (MDEV-4906). |
3989 | */ |
3990 | rgi->gtid_pending= false; |
3991 | } |
3992 | |
3993 | DBUG_RETURN(exec_res ? 1 : 0); |
3994 | } |
3995 | |
3996 | |
3997 | /** |
3998 | Applies the given event and advances the relay log position. |
3999 | |
4000 | This is needed by the sql thread to execute events from the binlog, |
4001 | and by clients executing BINLOG statements. Conceptually, this |
4002 | function does: |
4003 | |
4004 | @code |
4005 | ev->apply_event(rli); |
4006 | ev->update_pos(rli); |
4007 | @endcode |
4008 | |
4009 | It also does the following maintainance: |
4010 | |
4011 | - Initializes the thread's server_id and time; and the event's |
4012 | thread. |
4013 | |
4014 | - If !rli->belongs_to_client() (i.e., if it belongs to the slave |
4015 | sql thread instead of being used for executing BINLOG |
4016 | statements), it does the following things: (1) skips events if it |
4017 | is needed according to the server id or slave_skip_counter; (2) |
4018 | unlocks rli->data_lock; (3) sleeps if required by 'CHANGE MASTER |
4019 | TO MASTER_DELAY=X'; (4) maintains the running state of the sql |
4020 | thread (rli->thread_state). |
4021 | |
4022 | - Reports errors as needed. |
4023 | |
4024 | @param ev The event to apply. |
4025 | |
4026 | @param thd The client thread that executes the event (i.e., the |
4027 | slave sql thread if called from a replication slave, or the client |
4028 | thread if called to execute a BINLOG statement). |
4029 | |
4030 | @param rli The relay log info (i.e., the slave's rli if called from |
4031 | a replication slave, or the client's thd->rli_fake if called to |
4032 | execute a BINLOG statement). |
4033 | |
4034 | @retval 0 OK. |
4035 | |
4036 | @retval 1 Error calling ev->apply_event(). |
4037 | |
4038 | @retval 2 No error calling ev->apply_event(), but error calling |
4039 | ev->update_pos(). |
4040 | |
4041 | This function is only used in non-parallel replication, where it is called |
4042 | with rli->data_lock held; this lock is released during this function. |
4043 | */ |
4044 | int |
4045 | apply_event_and_update_pos(Log_event* ev, THD* thd, rpl_group_info *rgi) |
4046 | { |
4047 | Relay_log_info* rli= rgi->rli; |
4048 | mysql_mutex_assert_owner(&rli->data_lock); |
4049 | int reason= apply_event_and_update_pos_setup(ev, thd, rgi); |
4050 | if (reason == Log_event::EVENT_SKIP_COUNT) |
4051 | { |
4052 | DBUG_ASSERT(rli->slave_skip_counter > 0); |
4053 | rli->slave_skip_counter--; |
4054 | } |
4055 | |
4056 | if (reason == Log_event::EVENT_SKIP_NOT) |
4057 | { |
4058 | // Sleeps if needed, and unlocks rli->data_lock. |
4059 | if (sql_delay_event(ev, thd, rgi)) |
4060 | return 0; |
4061 | } |
4062 | else |
4063 | mysql_mutex_unlock(&rli->data_lock); |
4064 | |
4065 | return apply_event_and_update_pos_apply(ev, thd, rgi, reason); |
4066 | } |
4067 | |
4068 | |
4069 | /* |
4070 | The version of above apply_event_and_update_pos() used in parallel |
4071 | replication. Unlike the non-parallel case, this function is called without |
4072 | rli->data_lock held. |
4073 | */ |
4074 | int |
4075 | apply_event_and_update_pos_for_parallel(Log_event* ev, THD* thd, |
4076 | rpl_group_info *rgi) |
4077 | { |
4078 | mysql_mutex_assert_not_owner(&rgi->rli->data_lock); |
4079 | int reason= apply_event_and_update_pos_setup(ev, thd, rgi); |
4080 | /* |
4081 | In parallel replication, sql_slave_skip_counter is handled in the SQL |
4082 | driver thread, so 23 should never see EVENT_SKIP_COUNT here. |
4083 | */ |
4084 | DBUG_ASSERT(reason != Log_event::EVENT_SKIP_COUNT); |
4085 | /* |
4086 | Calling sql_delay_event() was handled in the SQL driver thread when |
4087 | doing parallel replication. |
4088 | */ |
4089 | return apply_event_and_update_pos_apply(ev, thd, rgi, reason); |
4090 | } |
4091 | |
4092 | |
4093 | /** |
4094 | Keep the relay log transaction state up to date. |
4095 | |
4096 | The state reflects how things are after the given event, that has just been |
4097 | read from the relay log, is executed. |
4098 | |
4099 | This is only needed to ensure we: |
4100 | - Don't abort the sql driver thread in the middle of an event group. |
4101 | - Don't rotate the io thread in the middle of a statement or transaction. |
4102 | The mechanism is that the io thread, when it needs to rotate the relay |
4103 | log, will wait until the sql driver has read all the cached events |
4104 | and then continue reading events one by one from the master until |
4105 | the sql threads signals that log doesn't have an active group anymore. |
4106 | |
4107 | There are two possible cases. We keep them as 2 separate flags mainly |
4108 | to make debugging easier. |
4109 | |
4110 | - IN_STMT is set when we have read an event that should be used |
4111 | together with the next event. This is for example setting a |
4112 | variable that is used when executing the next statement. |
4113 | - IN_TRANSACTION is set when we are inside a BEGIN...COMMIT group |
4114 | |
4115 | To test the state one should use the is_in_group() function. |
4116 | */ |
4117 | |
4118 | inline void update_state_of_relay_log(Relay_log_info *rli, Log_event *ev) |
4119 | { |
4120 | Log_event_type typ= ev->get_type_code(); |
4121 | |
4122 | /* check if we are in a multi part event */ |
4123 | if (ev->is_part_of_group()) |
4124 | rli->set_flag(Relay_log_info::IN_STMT); |
4125 | else if (Log_event::is_group_event(typ)) |
4126 | { |
4127 | /* |
4128 | If it was not a is_part_of_group() and not a group event (like |
4129 | rotate) then we can reset the IN_STMT flag. We have the above |
4130 | if only to allow us to have a rotate element anywhere. |
4131 | */ |
4132 | rli->clear_flag(Relay_log_info::IN_STMT); |
4133 | } |
4134 | |
4135 | /* Check for an event that starts or stops a transaction */ |
4136 | if (LOG_EVENT_IS_QUERY(typ)) |
4137 | { |
4138 | Query_log_event *qev= (Query_log_event*) ev; |
4139 | /* |
4140 | Trivial optimization to avoid the following somewhat expensive |
4141 | checks. |
4142 | */ |
4143 | if (qev->q_len <= sizeof("ROLLBACK" )) |
4144 | { |
4145 | if (qev->is_begin()) |
4146 | rli->set_flag(Relay_log_info::IN_TRANSACTION); |
4147 | if (qev->is_commit() || qev->is_rollback()) |
4148 | rli->clear_flag(Relay_log_info::IN_TRANSACTION); |
4149 | } |
4150 | } |
4151 | if (typ == XID_EVENT) |
4152 | rli->clear_flag(Relay_log_info::IN_TRANSACTION); |
4153 | if (typ == GTID_EVENT && |
4154 | !(((Gtid_log_event*) ev)->flags2 & Gtid_log_event::FL_STANDALONE)) |
4155 | { |
4156 | /* This GTID_EVENT will generate a BEGIN event */ |
4157 | rli->set_flag(Relay_log_info::IN_TRANSACTION); |
4158 | } |
4159 | |
4160 | DBUG_PRINT("info" , ("event: %u IN_STMT: %d IN_TRANSACTION: %d" , |
4161 | (uint) typ, |
4162 | rli->get_flag(Relay_log_info::IN_STMT), |
4163 | rli->get_flag(Relay_log_info::IN_TRANSACTION))); |
4164 | } |
4165 | |
4166 | |
4167 | /** |
4168 | Top-level function for executing the next event in the relay log. |
4169 | This is called from the SQL thread. |
4170 | |
4171 | This function reads the event from the relay log, executes it, and |
4172 | advances the relay log position. It also handles errors, etc. |
4173 | |
4174 | This function may fail to apply the event for the following reasons: |
4175 | |
4176 | - The position specfied by the UNTIL condition of the START SLAVE |
4177 | command is reached. |
4178 | |
4179 | - It was not possible to read the event from the log. |
4180 | |
4181 | - The slave is killed. |
4182 | |
4183 | - An error occurred when applying the event, and the event has been |
4184 | tried slave_trans_retries times. If the event has been retried |
4185 | fewer times, 0 is returned. |
4186 | |
4187 | - init_master_info or init_relay_log_pos failed. (These are called |
4188 | if a failure occurs when applying the event.) |
4189 | |
4190 | - An error occurred when updating the binlog position. |
4191 | |
4192 | @retval 0 The event was applied. |
4193 | |
4194 | @retval 1 The event was not applied. |
4195 | */ |
4196 | |
4197 | static int exec_relay_log_event(THD* thd, Relay_log_info* rli, |
4198 | rpl_group_info *serial_rgi) |
4199 | { |
4200 | ulonglong event_size; |
4201 | DBUG_ENTER("exec_relay_log_event" ); |
4202 | |
4203 | /* |
4204 | We acquire this mutex since we need it for all operations except |
4205 | event execution. But we will release it in places where we will |
4206 | wait for something for example inside of next_event(). |
4207 | */ |
4208 | mysql_mutex_lock(&rli->data_lock); |
4209 | |
4210 | Log_event *ev= next_event(serial_rgi, &event_size); |
4211 | |
4212 | if (sql_slave_killed(serial_rgi)) |
4213 | { |
4214 | mysql_mutex_unlock(&rli->data_lock); |
4215 | delete ev; |
4216 | DBUG_RETURN(1); |
4217 | } |
4218 | if (ev) |
4219 | { |
4220 | int exec_res; |
4221 | Log_event_type typ= ev->get_type_code(); |
4222 | |
4223 | /* |
4224 | Even if we don't execute this event, we keep the master timestamp, |
4225 | so that seconds behind master shows correct delta (there are events |
4226 | that are not replayed, so we keep falling behind). |
4227 | |
4228 | If it is an artificial event, or a relay log event (IO thread generated |
4229 | event) or ev->when is set to 0, we don't update the |
4230 | last_master_timestamp. |
4231 | |
4232 | In parallel replication, we might queue a large number of events, and |
4233 | the user might be surprised to see a claim that the slave is up to date |
4234 | long before those queued events are actually executed. |
4235 | */ |
4236 | if (!rli->mi->using_parallel() && |
4237 | !(ev->is_artificial_event() || ev->is_relay_log_event() || (ev->when == 0))) |
4238 | { |
4239 | rli->last_master_timestamp= ev->when + (time_t) ev->exec_time; |
4240 | DBUG_ASSERT(rli->last_master_timestamp >= 0); |
4241 | } |
4242 | |
4243 | /* |
4244 | This tests if the position of the beginning of the current event |
4245 | hits the UNTIL barrier. |
4246 | */ |
4247 | if ((rli->until_condition == Relay_log_info::UNTIL_MASTER_POS || |
4248 | rli->until_condition == Relay_log_info::UNTIL_RELAY_POS) && |
4249 | (ev->server_id != global_system_variables.server_id || |
4250 | rli->replicate_same_server_id) && |
4251 | rli->is_until_satisfied((rli->get_flag(Relay_log_info::IN_TRANSACTION) || !ev->log_pos) |
4252 | ? rli->group_master_log_pos |
4253 | : ev->log_pos - ev->data_written)) |
4254 | { |
4255 | sql_print_information("Slave SQL thread stopped because it reached its" |
4256 | " UNTIL position %llu" , rli->until_pos()); |
4257 | /* |
4258 | Setting abort_slave flag because we do not want additional |
4259 | message about error in query execution to be printed. |
4260 | */ |
4261 | rli->abort_slave= 1; |
4262 | rli->stop_for_until= true; |
4263 | mysql_mutex_unlock(&rli->data_lock); |
4264 | delete ev; |
4265 | DBUG_RETURN(1); |
4266 | } |
4267 | |
4268 | { /** |
4269 | The following failure injecion works in cooperation with tests |
4270 | setting @@global.debug= 'd,incomplete_group_in_relay_log'. |
4271 | Xid or Commit events are not executed to force the slave sql |
4272 | read hanging if the realy log does not have any more events. |
4273 | */ |
4274 | DBUG_EXECUTE_IF("incomplete_group_in_relay_log" , |
4275 | if ((typ == XID_EVENT) || |
4276 | (LOG_EVENT_IS_QUERY(typ) && |
4277 | strcmp("COMMIT" , ((Query_log_event *) ev)->query) == 0)) |
4278 | { |
4279 | DBUG_ASSERT(thd->transaction.all.modified_non_trans_table); |
4280 | rli->abort_slave= 1; |
4281 | mysql_mutex_unlock(&rli->data_lock); |
4282 | delete ev; |
4283 | serial_rgi->inc_event_relay_log_pos(); |
4284 | DBUG_RETURN(0); |
4285 | };); |
4286 | } |
4287 | |
4288 | update_state_of_relay_log(rli, ev); |
4289 | |
4290 | if (rli->mi->using_parallel()) |
4291 | { |
4292 | int res= rli->parallel.do_event(serial_rgi, ev, event_size); |
4293 | /* |
4294 | In parallel replication, we need to update the relay log position |
4295 | immediately so that it will be the correct position from which to |
4296 | read the next event. |
4297 | */ |
4298 | if (res == 0) |
4299 | rli->event_relay_log_pos= rli->future_event_relay_log_pos; |
4300 | if (res >= 0) |
4301 | DBUG_RETURN(res); |
4302 | /* |
4303 | Else we proceed to execute the event non-parallel. |
4304 | This is the case for pre-10.0 events without GTID, and for handling |
4305 | slave_skip_counter. |
4306 | */ |
4307 | } |
4308 | |
4309 | if (typ == GTID_EVENT) |
4310 | { |
4311 | Gtid_log_event *gev= static_cast<Gtid_log_event *>(ev); |
4312 | |
4313 | /* |
4314 | For GTID, allocate a new sub_id for the given domain_id. |
4315 | The sub_id must be allocated in increasing order of binlog order. |
4316 | */ |
4317 | if (event_group_new_gtid(serial_rgi, gev)) |
4318 | { |
4319 | sql_print_error("Error reading relay log event: %s" , "slave SQL thread " |
4320 | "aborted because of out-of-memory error" ); |
4321 | mysql_mutex_unlock(&rli->data_lock); |
4322 | delete ev; |
4323 | DBUG_RETURN(1); |
4324 | } |
4325 | |
4326 | if (opt_gtid_ignore_duplicates && |
4327 | rli->mi->using_gtid != Master_info::USE_GTID_NO) |
4328 | { |
4329 | int res= rpl_global_gtid_slave_state->check_duplicate_gtid |
4330 | (&serial_rgi->current_gtid, serial_rgi); |
4331 | if (res < 0) |
4332 | { |
4333 | sql_print_error("Error processing GTID event: %s" , "slave SQL " |
4334 | "thread aborted because of out-of-memory error" ); |
4335 | mysql_mutex_unlock(&rli->data_lock); |
4336 | delete ev; |
4337 | DBUG_RETURN(1); |
4338 | } |
4339 | /* |
4340 | If we need to skip this event group (because the GTID was already |
4341 | applied), then do it using the code for slave_skip_counter, which |
4342 | is able to handle skipping until the end of the event group. |
4343 | */ |
4344 | if (!res) |
4345 | rli->slave_skip_counter= 1; |
4346 | } |
4347 | } |
4348 | |
4349 | serial_rgi->future_event_relay_log_pos= rli->future_event_relay_log_pos; |
4350 | serial_rgi->event_relay_log_name= rli->event_relay_log_name; |
4351 | serial_rgi->event_relay_log_pos= rli->event_relay_log_pos; |
4352 | exec_res= apply_event_and_update_pos(ev, thd, serial_rgi); |
4353 | |
4354 | #ifdef WITH_WSREP |
4355 | WSREP_DEBUG("apply_event_and_update_pos() result: %d" , exec_res); |
4356 | #endif /* WITH_WSREP */ |
4357 | |
4358 | delete_or_keep_event_post_apply(serial_rgi, typ, ev); |
4359 | |
4360 | /* |
4361 | update_log_pos failed: this should not happen, so we don't |
4362 | retry. |
4363 | */ |
4364 | if (unlikely(exec_res == 2)) |
4365 | DBUG_RETURN(1); |
4366 | |
4367 | #ifdef WITH_WSREP |
4368 | mysql_mutex_lock(&thd->LOCK_thd_data); |
4369 | if (thd->wsrep_conflict_state == NO_CONFLICT) |
4370 | { |
4371 | mysql_mutex_unlock(&thd->LOCK_thd_data); |
4372 | #endif /* WITH_WSREP */ |
4373 | if (slave_trans_retries) |
4374 | { |
4375 | int UNINIT_VAR(temp_err); |
4376 | if (unlikely(exec_res) && (temp_err= has_temporary_error(thd))) |
4377 | { |
4378 | const char *errmsg; |
4379 | rli->clear_error(); |
4380 | /* |
4381 | We were in a transaction which has been rolled back because of a |
4382 | temporary error; |
4383 | let's seek back to BEGIN log event and retry it all again. |
4384 | Note, if lock wait timeout (innodb_lock_wait_timeout exceeded) |
4385 | there is no rollback since 5.0.13 (ref: manual). |
4386 | We have to not only seek but also |
4387 | |
4388 | a) init_master_info(), to seek back to hot relay log's start |
4389 | for later (for when we will come back to this hot log after |
4390 | re-processing the possibly existing old logs where BEGIN is: |
4391 | check_binlog_magic() will then need the cache to be at |
4392 | position 0 (see comments at beginning of |
4393 | init_master_info()). |
4394 | b) init_relay_log_pos(), because the BEGIN may be an older relay log. |
4395 | */ |
4396 | if (serial_rgi->trans_retries < slave_trans_retries) |
4397 | { |
4398 | if (init_master_info(rli->mi, 0, 0, 0, SLAVE_SQL)) |
4399 | sql_print_error("Failed to initialize the master info structure" ); |
4400 | else if (init_relay_log_pos(rli, |
4401 | rli->group_relay_log_name, |
4402 | rli->group_relay_log_pos, |
4403 | 1, &errmsg, 1)) |
4404 | sql_print_error("Error initializing relay log position: %s" , |
4405 | errmsg); |
4406 | else |
4407 | { |
4408 | exec_res= 0; |
4409 | serial_rgi->cleanup_context(thd, 1); |
4410 | /* chance for concurrent connection to get more locks */ |
4411 | slave_sleep(thd, MY_MAX(MY_MIN(serial_rgi->trans_retries, |
4412 | MAX_SLAVE_RETRY_PAUSE), |
4413 | slave_trans_retry_interval), |
4414 | sql_slave_killed, serial_rgi); |
4415 | serial_rgi->trans_retries++; |
4416 | mysql_mutex_lock(&rli->data_lock); // because of SHOW STATUS |
4417 | rli->retried_trans++; |
4418 | statistic_increment(slave_retried_transactions, LOCK_status); |
4419 | mysql_mutex_unlock(&rli->data_lock); |
4420 | DBUG_PRINT("info" , ("Slave retries transaction " |
4421 | "rgi->trans_retries: %lu" , |
4422 | serial_rgi->trans_retries)); |
4423 | } |
4424 | } |
4425 | else |
4426 | sql_print_error("Slave SQL thread retried transaction %lu time(s) " |
4427 | "in vain, giving up. Consider raising the value of " |
4428 | "the slave_transaction_retries variable." , |
4429 | slave_trans_retries); |
4430 | } |
4431 | else if ((exec_res && !temp_err) || |
4432 | (opt_using_transactions && |
4433 | rli->group_relay_log_pos == rli->event_relay_log_pos)) |
4434 | { |
4435 | /* |
4436 | Only reset the retry counter if the entire group succeeded |
4437 | or failed with a non-transient error. On a successful |
4438 | event, the execution will proceed as usual; in the case of a |
4439 | non-transient error, the slave will stop with an error. |
4440 | */ |
4441 | serial_rgi->trans_retries= 0; // restart from fresh |
4442 | DBUG_PRINT("info" , ("Resetting retry counter, rgi->trans_retries: %lu" , |
4443 | serial_rgi->trans_retries)); |
4444 | } |
4445 | } |
4446 | #ifdef WITH_WSREP |
4447 | } |
4448 | else |
4449 | mysql_mutex_unlock(&thd->LOCK_thd_data); |
4450 | #endif /* WITH_WSREP */ |
4451 | |
4452 | thread_safe_increment64(&rli->executed_entries); |
4453 | DBUG_RETURN(exec_res); |
4454 | } |
4455 | mysql_mutex_unlock(&rli->data_lock); |
4456 | rli->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_READ_FAILURE, NULL, |
4457 | ER_THD(thd, ER_SLAVE_RELAY_LOG_READ_FAILURE), "\ |
4458 | Could not parse relay log event entry. The possible reasons are: the master's \ |
4459 | binary log is corrupted (you can check this by running 'mysqlbinlog' on the \ |
4460 | binary log), the slave's relay log is corrupted (you can check this by running \ |
4461 | 'mysqlbinlog' on the relay log), a network problem, or a bug in the master's \ |
4462 | or slave's MySQL code. If you want to check the master's binary log or slave's \ |
4463 | relay log, you will be able to know their names by issuing 'SHOW SLAVE STATUS' \ |
4464 | on this slave.\ |
4465 | " ); |
4466 | DBUG_RETURN(1); |
4467 | } |
4468 | |
4469 | |
4470 | static bool check_io_slave_killed(Master_info *mi, const char *info) |
4471 | { |
4472 | if (io_slave_killed(mi)) |
4473 | { |
4474 | if (info && global_system_variables.log_warnings) |
4475 | sql_print_information("%s" , info); |
4476 | return TRUE; |
4477 | } |
4478 | return FALSE; |
4479 | } |
4480 | |
4481 | /** |
4482 | @brief Try to reconnect slave IO thread. |
4483 | |
4484 | @details Terminates current connection to master, sleeps for |
4485 | @c mi->connect_retry msecs and initiates new connection with |
4486 | @c safe_reconnect(). Variable pointed by @c retry_count is increased - |
4487 | if it exceeds @c master_retry_count then connection is not re-established |
4488 | and function signals error. |
4489 | Unless @c suppres_warnings is TRUE, a warning is put in the server error log |
4490 | when reconnecting. The warning message and messages used to report errors |
4491 | are taken from @c messages array. In case @c master_retry_count is exceeded, |
4492 | no messages are added to the log. |
4493 | |
4494 | @param[in] thd Thread context. |
4495 | @param[in] mysql MySQL connection. |
4496 | @param[in] mi Master connection information. |
4497 | @param[in,out] retry_count Number of attempts to reconnect. |
4498 | @param[in] suppress_warnings TRUE when a normal net read timeout |
4499 | has caused to reconnecting. |
4500 | @param[in] messages Messages to print/log, see |
4501 | reconnect_messages[] array. |
4502 | |
4503 | @retval 0 OK. |
4504 | @retval 1 There was an error. |
4505 | */ |
4506 | |
4507 | static int try_to_reconnect(THD *thd, MYSQL *mysql, Master_info *mi, |
4508 | uint *retry_count, bool suppress_warnings, |
4509 | const char *messages[SLAVE_RECON_MSG_MAX]) |
4510 | { |
4511 | mi->slave_running= MYSQL_SLAVE_RUN_NOT_CONNECT; |
4512 | thd->proc_info= messages[SLAVE_RECON_MSG_WAIT]; |
4513 | #ifdef SIGNAL_WITH_VIO_CLOSE |
4514 | thd->clear_active_vio(); |
4515 | #endif |
4516 | end_server(mysql); |
4517 | if ((*retry_count)++) |
4518 | { |
4519 | if (*retry_count > master_retry_count) |
4520 | return 1; // Don't retry forever |
4521 | slave_sleep(thd, mi->connect_retry, io_slave_killed, mi); |
4522 | } |
4523 | if (check_io_slave_killed(mi, messages[SLAVE_RECON_MSG_KILLED_WAITING])) |
4524 | return 1; |
4525 | thd->proc_info = messages[SLAVE_RECON_MSG_AFTER]; |
4526 | if (!suppress_warnings) |
4527 | { |
4528 | char buf[256]; |
4529 | StringBuffer<100> tmp; |
4530 | if (mi->using_gtid != Master_info::USE_GTID_NO) |
4531 | { |
4532 | tmp.append(STRING_WITH_LEN("; GTID position '" )); |
4533 | mi->gtid_current_pos.append_to_string(&tmp); |
4534 | if (mi->events_queued_since_last_gtid == 0) |
4535 | tmp.append(STRING_WITH_LEN("'" )); |
4536 | else |
4537 | { |
4538 | tmp.append(STRING_WITH_LEN("', GTID event skip " )); |
4539 | tmp.append_ulonglong((ulonglong)mi->events_queued_since_last_gtid); |
4540 | } |
4541 | } |
4542 | my_snprintf(buf, sizeof(buf), messages[SLAVE_RECON_MSG_FAILED], |
4543 | IO_RPL_LOG_NAME, mi->master_log_pos, |
4544 | tmp.c_ptr_safe()); |
4545 | /* |
4546 | Raise a warining during registering on master/requesting dump. |
4547 | Log a message reading event. |
4548 | */ |
4549 | if (messages[SLAVE_RECON_MSG_COMMAND][0]) |
4550 | { |
4551 | mi->report(WARNING_LEVEL, ER_SLAVE_MASTER_COM_FAILURE, NULL, |
4552 | ER_THD(thd, ER_SLAVE_MASTER_COM_FAILURE), |
4553 | messages[SLAVE_RECON_MSG_COMMAND], buf); |
4554 | } |
4555 | else |
4556 | { |
4557 | sql_print_information("%s" , buf); |
4558 | } |
4559 | } |
4560 | if (safe_reconnect(thd, mysql, mi, 1) || io_slave_killed(mi)) |
4561 | { |
4562 | if (global_system_variables.log_warnings) |
4563 | sql_print_information("%s" , messages[SLAVE_RECON_MSG_KILLED_AFTER]); |
4564 | return 1; |
4565 | } |
4566 | return 0; |
4567 | } |
4568 | |
4569 | |
4570 | /** |
4571 | Slave IO thread entry point. |
4572 | |
4573 | @param arg Pointer to Master_info struct that holds information for |
4574 | the IO thread. |
4575 | |
4576 | @return Always 0. |
4577 | */ |
4578 | pthread_handler_t handle_slave_io(void *arg) |
4579 | { |
4580 | THD *thd; // needs to be first for thread_stack |
4581 | MYSQL *mysql; |
4582 | Master_info *mi = (Master_info*)arg; |
4583 | Relay_log_info *rli= &mi->rli; |
4584 | uint retry_count; |
4585 | bool suppress_warnings; |
4586 | int ret; |
4587 | rpl_io_thread_info io_info; |
4588 | #ifndef DBUG_OFF |
4589 | mi->dbug_do_disconnect= false; |
4590 | #endif |
4591 | // needs to call my_thread_init(), otherwise we get a coredump in DBUG_ stuff |
4592 | my_thread_init(); |
4593 | DBUG_ENTER("handle_slave_io" ); |
4594 | |
4595 | DBUG_ASSERT(mi->inited); |
4596 | mysql= NULL ; |
4597 | retry_count= 0; |
4598 | |
4599 | thd= new THD(next_thread_id()); // note that contructor of THD uses DBUG_ ! |
4600 | |
4601 | mysql_mutex_lock(&mi->run_lock); |
4602 | /* Inform waiting threads that slave has started */ |
4603 | mi->slave_run_id++; |
4604 | |
4605 | #ifndef DBUG_OFF |
4606 | mi->events_till_disconnect = disconnect_slave_event_count; |
4607 | #endif |
4608 | |
4609 | THD_CHECK_SENTRY(thd); |
4610 | mi->io_thd = thd; |
4611 | |
4612 | pthread_detach_this_thread(); |
4613 | thd->thread_stack= (char*) &thd; // remember where our stack is |
4614 | mi->clear_error(); |
4615 | if (init_slave_thread(thd, mi, SLAVE_THD_IO)) |
4616 | { |
4617 | mysql_cond_broadcast(&mi->start_cond); |
4618 | sql_print_error("Failed during slave I/O thread initialization" ); |
4619 | goto err_during_init; |
4620 | } |
4621 | thd->system_thread_info.rpl_io_info= &io_info; |
4622 | add_to_active_threads(thd); |
4623 | mi->slave_running = MYSQL_SLAVE_RUN_NOT_CONNECT; |
4624 | mi->abort_slave = 0; |
4625 | mysql_mutex_unlock(&mi->run_lock); |
4626 | mysql_cond_broadcast(&mi->start_cond); |
4627 | mi->rows_event_tracker.reset(); |
4628 | |
4629 | DBUG_PRINT("master_info" ,("log_file_name: '%s' position: %llu" , |
4630 | mi->master_log_name, mi->master_log_pos)); |
4631 | |
4632 | /* This must be called before run any binlog_relay_io hooks */ |
4633 | my_pthread_setspecific_ptr(RPL_MASTER_INFO, mi); |
4634 | |
4635 | /* Load the set of seen GTIDs, if we did not already. */ |
4636 | if (rpl_load_gtid_slave_state(thd)) |
4637 | { |
4638 | mi->report(ERROR_LEVEL, thd->get_stmt_da()->sql_errno(), NULL, |
4639 | "Unable to load replication GTID slave state from mysql.%s: %s" , |
4640 | rpl_gtid_slave_state_table_name.str, |
4641 | thd->get_stmt_da()->message()); |
4642 | /* |
4643 | If we are using old-style replication, we can continue, even though we |
4644 | then will not be able to record the GTIDs we receive. But if using GTID, |
4645 | we must give up. |
4646 | */ |
4647 | if (mi->using_gtid != Master_info::USE_GTID_NO || opt_gtid_strict_mode) |
4648 | goto err; |
4649 | } |
4650 | |
4651 | |
4652 | if (DBUG_EVALUATE_IF("failed_slave_start" , 1, 0) |
4653 | || repl_semisync_slave.slave_start(mi)) |
4654 | { |
4655 | mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, NULL, |
4656 | ER_THD(thd, ER_SLAVE_FATAL_ERROR), |
4657 | "Failed to run 'thread_start' hook" ); |
4658 | goto err; |
4659 | } |
4660 | |
4661 | if (!(mi->mysql = mysql = mysql_init(NULL))) |
4662 | { |
4663 | mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, NULL, |
4664 | ER_THD(thd, ER_SLAVE_FATAL_ERROR), "error in mysql_init()" ); |
4665 | goto err; |
4666 | } |
4667 | |
4668 | THD_STAGE_INFO(thd, stage_connecting_to_master); |
4669 | // we can get killed during safe_connect |
4670 | if (!safe_connect(thd, mysql, mi)) |
4671 | { |
4672 | if (mi->using_gtid == Master_info::USE_GTID_NO) |
4673 | sql_print_information("Slave I/O thread: connected to master '%s@%s:%d'," |
4674 | "replication started in log '%s' at position %llu" , |
4675 | mi->user, mi->host, mi->port, |
4676 | IO_RPL_LOG_NAME, mi->master_log_pos); |
4677 | else |
4678 | { |
4679 | StringBuffer<100> tmp; |
4680 | mi->gtid_current_pos.to_string(&tmp); |
4681 | sql_print_information("Slave I/O thread: connected to master '%s@%s:%d'," |
4682 | "replication starts at GTID position '%s'" , |
4683 | mi->user, mi->host, mi->port, tmp.c_ptr_safe()); |
4684 | } |
4685 | } |
4686 | else |
4687 | { |
4688 | sql_print_information("Slave I/O thread killed while connecting to master" ); |
4689 | goto err; |
4690 | } |
4691 | |
4692 | connected: |
4693 | |
4694 | if (mi->using_gtid != Master_info::USE_GTID_NO) |
4695 | { |
4696 | /* |
4697 | When the IO thread (re)connects to the master using GTID, it will |
4698 | connect at the start of an event group. But the IO thread may have |
4699 | previously logged part of the following event group to the relay |
4700 | log. |
4701 | |
4702 | When the IO and SQL thread are started together, we erase any previous |
4703 | relay logs, but this is not possible/desirable while the SQL thread is |
4704 | running. To avoid duplicating partial event groups in the relay logs in |
4705 | this case, we remember the count of events in any partially logged event |
4706 | group before the reconnect, and then here at connect we set up a counter |
4707 | to skip the already-logged part of the group. |
4708 | */ |
4709 | mi->gtid_reconnect_event_skip_count= mi->events_queued_since_last_gtid; |
4710 | mi->gtid_event_seen= false; |
4711 | /* |
4712 | Reset stale state of the rows-event group tracker at reconnect. |
4713 | */ |
4714 | mi->rows_event_tracker.reset(); |
4715 | } |
4716 | |
4717 | #ifdef ENABLED_DEBUG_SYNC |
4718 | DBUG_EXECUTE_IF("dbug.before_get_running_status_yes" , |
4719 | { |
4720 | const char act[]= |
4721 | "now " |
4722 | "wait_for signal.io_thread_let_running" ; |
4723 | DBUG_ASSERT(debug_sync_service); |
4724 | DBUG_ASSERT(!debug_sync_set_action(thd, |
4725 | STRING_WITH_LEN(act))); |
4726 | };); |
4727 | #endif |
4728 | |
4729 | mysql_mutex_lock(&mi->run_lock); |
4730 | mi->slave_running= MYSQL_SLAVE_RUN_CONNECT; |
4731 | mysql_mutex_unlock(&mi->run_lock); |
4732 | |
4733 | thd->slave_net = &mysql->net; |
4734 | THD_STAGE_INFO(thd, stage_checking_master_version); |
4735 | ret= get_master_version_and_clock(mysql, mi); |
4736 | if (ret == 1) |
4737 | /* Fatal error */ |
4738 | goto err; |
4739 | |
4740 | if (ret == 2) |
4741 | { |
4742 | if (check_io_slave_killed(mi, "Slave I/O thread killed " |
4743 | "while calling get_master_version_and_clock(...)" )) |
4744 | goto err; |
4745 | suppress_warnings= FALSE; |
4746 | /* |
4747 | Try to reconnect because the error was caused by a transient network |
4748 | problem |
4749 | */ |
4750 | if (try_to_reconnect(thd, mysql, mi, &retry_count, suppress_warnings, |
4751 | reconnect_messages[SLAVE_RECON_ACT_REG])) |
4752 | goto err; |
4753 | goto connected; |
4754 | } |
4755 | |
4756 | if (mi->rli.relay_log.description_event_for_queue->binlog_version > 1) |
4757 | { |
4758 | /* |
4759 | Register ourselves with the master. |
4760 | */ |
4761 | THD_STAGE_INFO(thd, stage_registering_slave_on_master); |
4762 | if (register_slave_on_master(mysql, mi, &suppress_warnings)) |
4763 | { |
4764 | if (!check_io_slave_killed(mi, "Slave I/O thread killed " |
4765 | "while registering slave on master" )) |
4766 | { |
4767 | sql_print_error("Slave I/O thread couldn't register on master" ); |
4768 | if (try_to_reconnect(thd, mysql, mi, &retry_count, suppress_warnings, |
4769 | reconnect_messages[SLAVE_RECON_ACT_REG])) |
4770 | goto err; |
4771 | } |
4772 | else |
4773 | goto err; |
4774 | goto connected; |
4775 | } |
4776 | } |
4777 | |
4778 | DBUG_PRINT("info" ,("Starting reading binary log from master" )); |
4779 | thd->set_command(COM_SLAVE_IO); |
4780 | while (!io_slave_killed(mi)) |
4781 | { |
4782 | THD_STAGE_INFO(thd, stage_requesting_binlog_dump); |
4783 | if (request_dump(thd, mysql, mi, &suppress_warnings)) |
4784 | { |
4785 | sql_print_error("Failed on request_dump()" ); |
4786 | if (check_io_slave_killed(mi, NullS) || |
4787 | try_to_reconnect(thd, mysql, mi, &retry_count, suppress_warnings, |
4788 | reconnect_messages[SLAVE_RECON_ACT_DUMP])) |
4789 | goto err; |
4790 | goto connected; |
4791 | } |
4792 | |
4793 | const char *event_buf; |
4794 | |
4795 | mi->slave_running= MYSQL_SLAVE_RUN_READING; |
4796 | DBUG_ASSERT(mi->last_error().number == 0); |
4797 | ulonglong lastchecktime = my_hrtime().val; |
4798 | ulonglong tokenamount = opt_read_binlog_speed_limit*1024; |
4799 | while (!io_slave_killed(mi)) |
4800 | { |
4801 | ulong event_len, network_read_len = 0; |
4802 | /* |
4803 | We say "waiting" because read_event() will wait if there's nothing to |
4804 | read. But if there's something to read, it will not wait. The |
4805 | important thing is to not confuse users by saying "reading" whereas |
4806 | we're in fact receiving nothing. |
4807 | */ |
4808 | THD_STAGE_INFO(thd, stage_waiting_for_master_to_send_event); |
4809 | event_len= read_event(mysql, mi, &suppress_warnings, &network_read_len); |
4810 | if (check_io_slave_killed(mi, NullS)) |
4811 | goto err; |
4812 | |
4813 | if (unlikely(event_len == packet_error)) |
4814 | { |
4815 | uint mysql_error_number= mysql_errno(mysql); |
4816 | switch (mysql_error_number) { |
4817 | case CR_NET_PACKET_TOO_LARGE: |
4818 | sql_print_error("\ |
4819 | Log entry on master is longer than slave_max_allowed_packet (%lu) on \ |
4820 | slave. If the entry is correct, restart the server with a higher value of \ |
4821 | slave_max_allowed_packet" , |
4822 | slave_max_allowed_packet); |
4823 | mi->report(ERROR_LEVEL, ER_NET_PACKET_TOO_LARGE, NULL, |
4824 | "%s" , "Got a packet bigger than 'slave_max_allowed_packet' bytes" ); |
4825 | goto err; |
4826 | case ER_MASTER_FATAL_ERROR_READING_BINLOG: |
4827 | mi->report(ERROR_LEVEL, ER_MASTER_FATAL_ERROR_READING_BINLOG, NULL, |
4828 | ER_THD(thd, ER_MASTER_FATAL_ERROR_READING_BINLOG), |
4829 | mysql_error_number, mysql_error(mysql)); |
4830 | goto err; |
4831 | case ER_OUT_OF_RESOURCES: |
4832 | sql_print_error("\ |
4833 | Stopping slave I/O thread due to out-of-memory error from master" ); |
4834 | mi->report(ERROR_LEVEL, ER_OUT_OF_RESOURCES, NULL, |
4835 | "%s" , ER_THD(thd, ER_OUT_OF_RESOURCES)); |
4836 | goto err; |
4837 | } |
4838 | if (try_to_reconnect(thd, mysql, mi, &retry_count, suppress_warnings, |
4839 | reconnect_messages[SLAVE_RECON_ACT_EVENT])) |
4840 | goto err; |
4841 | goto connected; |
4842 | } // if (event_len == packet_error) |
4843 | |
4844 | retry_count=0; // ok event, reset retry counter |
4845 | THD_STAGE_INFO(thd, stage_queueing_master_event_to_the_relay_log); |
4846 | event_buf= (const char*)mysql->net.read_pos + 1; |
4847 | mi->semi_ack= 0; |
4848 | if (repl_semisync_slave. |
4849 | slave_read_sync_header((const char*)mysql->net.read_pos + 1, event_len, |
4850 | &(mi->semi_ack), &event_buf, &event_len)) |
4851 | { |
4852 | mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, NULL, |
4853 | ER_THD(thd, ER_SLAVE_FATAL_ERROR), |
4854 | "Failed to run 'after_read_event' hook" ); |
4855 | goto err; |
4856 | } |
4857 | |
4858 | /* Control the binlog read speed of master |
4859 | when read_binlog_speed_limit is non-zero |
4860 | */ |
4861 | ulonglong speed_limit_in_bytes = opt_read_binlog_speed_limit * 1024; |
4862 | if (speed_limit_in_bytes) |
4863 | { |
4864 | /* Prevent the tokenamount become a large value, |
4865 | for example, the IO thread doesn't work for a long time |
4866 | */ |
4867 | if (tokenamount > speed_limit_in_bytes * 2) |
4868 | { |
4869 | lastchecktime = my_hrtime().val; |
4870 | tokenamount = speed_limit_in_bytes * 2; |
4871 | } |
4872 | |
4873 | do |
4874 | { |
4875 | ulonglong currenttime = my_hrtime().val; |
4876 | tokenamount += (currenttime - lastchecktime) * speed_limit_in_bytes / (1000*1000); |
4877 | lastchecktime = currenttime; |
4878 | if(tokenamount < network_read_len) |
4879 | { |
4880 | ulonglong duration =1000ULL*1000 * (network_read_len - tokenamount) / speed_limit_in_bytes; |
4881 | time_t second_time = (time_t)(duration / (1000 * 1000)); |
4882 | uint micro_time = duration % (1000 * 1000); |
4883 | |
4884 | // at least sleep 1000 micro second |
4885 | my_sleep(MY_MAX(micro_time,1000)); |
4886 | |
4887 | /* |
4888 | If it sleep more than one second, |
4889 | it should use slave_sleep() to avoid the STOP SLAVE hang. |
4890 | */ |
4891 | if (second_time) |
4892 | slave_sleep(thd, second_time, io_slave_killed, mi); |
4893 | |
4894 | } |
4895 | }while(tokenamount < network_read_len); |
4896 | tokenamount -= network_read_len; |
4897 | } |
4898 | |
4899 | if (queue_event(mi, event_buf, event_len)) |
4900 | { |
4901 | mi->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_WRITE_FAILURE, NULL, |
4902 | ER_THD(thd, ER_SLAVE_RELAY_LOG_WRITE_FAILURE), |
4903 | "could not queue event from master" ); |
4904 | goto err; |
4905 | } |
4906 | |
4907 | if (rpl_semi_sync_slave_status && (mi->semi_ack & SEMI_SYNC_NEED_ACK) && |
4908 | repl_semisync_slave.slave_reply(mi)) |
4909 | { |
4910 | mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, NULL, |
4911 | ER_THD(thd, ER_SLAVE_FATAL_ERROR), |
4912 | "Failed to run 'after_queue_event' hook" ); |
4913 | goto err; |
4914 | } |
4915 | |
4916 | if (mi->using_gtid == Master_info::USE_GTID_NO && |
4917 | /* |
4918 | If rpl_semi_sync_slave_delay_master is enabled, we will flush |
4919 | master info only when ack is needed. This may lead to at least one |
4920 | group transaction delay but affords better performance improvement. |
4921 | */ |
4922 | (!repl_semisync_slave.get_slave_enabled() || |
4923 | (!(mi->semi_ack & SEMI_SYNC_SLAVE_DELAY_SYNC) || |
4924 | (mi->semi_ack & (SEMI_SYNC_NEED_ACK)))) && |
4925 | (DBUG_EVALUATE_IF("failed_flush_master_info" , 1, 0) || |
4926 | flush_master_info(mi, TRUE, TRUE))) |
4927 | { |
4928 | sql_print_error("Failed to flush master info file" ); |
4929 | goto err; |
4930 | } |
4931 | /* |
4932 | See if the relay logs take too much space. |
4933 | We don't lock mi->rli.log_space_lock here; this dirty read saves time |
4934 | and does not introduce any problem: |
4935 | - if mi->rli.ignore_log_space_limit is 1 but becomes 0 just after (so |
4936 | the clean value is 0), then we are reading only one more event as we |
4937 | should, and we'll block only at the next event. No big deal. |
4938 | - if mi->rli.ignore_log_space_limit is 0 but becomes 1 just |
4939 | after (so the clean value is 1), then we are going into |
4940 | wait_for_relay_log_space() for no reason, but this function |
4941 | will do a clean read, notice the clean value and exit |
4942 | immediately. |
4943 | */ |
4944 | #ifndef DBUG_OFF |
4945 | { |
4946 | DBUG_PRINT("info" , ("log_space_limit=%llu log_space_total=%llu " |
4947 | "ignore_log_space_limit=%d" , |
4948 | rli->log_space_limit, rli->log_space_total, |
4949 | (int) rli->ignore_log_space_limit)); |
4950 | } |
4951 | #endif |
4952 | |
4953 | if (rli->log_space_limit && rli->log_space_limit < |
4954 | rli->log_space_total && |
4955 | !rli->ignore_log_space_limit) |
4956 | if (wait_for_relay_log_space(rli)) |
4957 | { |
4958 | sql_print_error("Slave I/O thread aborted while waiting for relay \ |
4959 | log space" ); |
4960 | goto err; |
4961 | } |
4962 | } |
4963 | } |
4964 | |
4965 | // error = 0; |
4966 | err: |
4967 | // print the current replication position |
4968 | if (mi->using_gtid == Master_info::USE_GTID_NO) |
4969 | sql_print_information("Slave I/O thread exiting, read up to log '%s', " |
4970 | "position %llu" , IO_RPL_LOG_NAME, mi->master_log_pos); |
4971 | else |
4972 | { |
4973 | StringBuffer<100> tmp; |
4974 | mi->gtid_current_pos.to_string(&tmp); |
4975 | sql_print_information("Slave I/O thread exiting, read up to log '%s', " |
4976 | "position %llu; GTID position %s" , |
4977 | IO_RPL_LOG_NAME, mi->master_log_pos, |
4978 | tmp.c_ptr_safe()); |
4979 | } |
4980 | repl_semisync_slave.slave_stop(mi); |
4981 | thd->reset_query(); |
4982 | thd->reset_db(&null_clex_str); |
4983 | if (mysql) |
4984 | { |
4985 | /* |
4986 | Here we need to clear the active VIO before closing the |
4987 | connection with the master. The reason is that THD::awake() |
4988 | might be called from terminate_slave_thread() because somebody |
4989 | issued a STOP SLAVE. If that happends, the close_active_vio() |
4990 | can be called in the middle of closing the VIO associated with |
4991 | the 'mysql' object, causing a crash. |
4992 | */ |
4993 | #ifdef SIGNAL_WITH_VIO_CLOSE |
4994 | thd->clear_active_vio(); |
4995 | #endif |
4996 | mysql_close(mysql); |
4997 | mi->mysql=0; |
4998 | } |
4999 | write_ignored_events_info_to_relay_log(thd, mi); |
5000 | if (mi->using_gtid != Master_info::USE_GTID_NO) |
5001 | flush_master_info(mi, TRUE, TRUE); |
5002 | THD_STAGE_INFO(thd, stage_waiting_for_slave_mutex_on_exit); |
5003 | thd->add_status_to_global(); |
5004 | unlink_not_visible_thd(thd); |
5005 | mysql_mutex_lock(&mi->run_lock); |
5006 | |
5007 | err_during_init: |
5008 | /* Forget the relay log's format */ |
5009 | delete mi->rli.relay_log.description_event_for_queue; |
5010 | mi->rli.relay_log.description_event_for_queue= 0; |
5011 | // TODO: make rpl_status part of Master_info |
5012 | change_rpl_status(RPL_ACTIVE_SLAVE,RPL_IDLE_SLAVE); |
5013 | |
5014 | thd->assert_not_linked(); |
5015 | delete thd; |
5016 | thread_safe_decrement32(&service_thread_count); |
5017 | signal_thd_deleted(); |
5018 | |
5019 | mi->abort_slave= 0; |
5020 | mi->slave_running= MYSQL_SLAVE_NOT_RUN; |
5021 | mi->io_thd= 0; |
5022 | /* |
5023 | Note: the order of the two following calls (first broadcast, then unlock) |
5024 | is important. Otherwise a killer_thread can execute between the calls and |
5025 | delete the mi structure leading to a crash! (see BUG#25306 for details) |
5026 | */ |
5027 | mysql_cond_broadcast(&mi->stop_cond); // tell the world we are done |
5028 | DBUG_EXECUTE_IF("simulate_slave_delay_at_terminate_bug38694" , sleep(5);); |
5029 | mysql_mutex_unlock(&mi->run_lock); |
5030 | |
5031 | DBUG_LEAVE; // Must match DBUG_ENTER() |
5032 | my_thread_end(); |
5033 | ERR_remove_state(0); |
5034 | pthread_exit(0); |
5035 | return 0; // Avoid compiler warnings |
5036 | } |
5037 | |
5038 | /* |
5039 | Check the temporary directory used by commands like |
5040 | LOAD DATA INFILE. |
5041 | |
5042 | As the directory never changes during a mysqld run, we only |
5043 | test this once and cache the result. This also resolve a race condition |
5044 | when this can be run by multiple threads at the same time. |
5045 | */ |
5046 | |
5047 | static bool check_temp_dir_run= 0; |
5048 | static int check_temp_dir_result= 0; |
5049 | |
5050 | static |
5051 | int check_temp_dir(char* tmp_file) |
5052 | { |
5053 | File fd; |
5054 | int result= 1; // Assume failure |
5055 | MY_DIR *dirp; |
5056 | char tmp_dir[FN_REFLEN]; |
5057 | size_t tmp_dir_size; |
5058 | DBUG_ENTER("check_temp_dir" ); |
5059 | |
5060 | /* This look is safe to use as this function is only called once */ |
5061 | mysql_mutex_lock(&LOCK_start_thread); |
5062 | if (check_temp_dir_run) |
5063 | { |
5064 | if ((result= check_temp_dir_result)) |
5065 | my_message(result, tmp_file, MYF(0)); |
5066 | goto end; |
5067 | } |
5068 | check_temp_dir_run= 1; |
5069 | |
5070 | /* |
5071 | Get the directory from the temporary file. |
5072 | */ |
5073 | dirname_part(tmp_dir, tmp_file, &tmp_dir_size); |
5074 | |
5075 | /* |
5076 | Check if the directory exists. |
5077 | */ |
5078 | if (!(dirp=my_dir(tmp_dir,MYF(MY_WME)))) |
5079 | goto end; |
5080 | my_dirend(dirp); |
5081 | |
5082 | /* |
5083 | Check permissions to create a file. We use O_TRUNC to ensure that |
5084 | things works even if we happen to have and old file laying around. |
5085 | */ |
5086 | if ((fd= mysql_file_create(key_file_misc, |
5087 | tmp_file, CREATE_MODE, |
5088 | O_WRONLY | O_BINARY | O_TRUNC | O_NOFOLLOW, |
5089 | MYF(MY_WME))) < 0) |
5090 | goto end; |
5091 | |
5092 | result= 0; // Directory name ok |
5093 | /* |
5094 | Clean up. |
5095 | */ |
5096 | mysql_file_close(fd, MYF(0)); |
5097 | mysql_file_delete(key_file_misc, tmp_file, MYF(0)); |
5098 | |
5099 | end: |
5100 | mysql_mutex_unlock(&LOCK_start_thread); |
5101 | DBUG_RETURN(result); |
5102 | } |
5103 | |
5104 | |
5105 | void |
5106 | slave_output_error_info(rpl_group_info *rgi, THD *thd) |
5107 | { |
5108 | /* |
5109 | retrieve as much info as possible from the thd and, error |
5110 | codes and warnings and print this to the error log as to |
5111 | allow the user to locate the error |
5112 | */ |
5113 | Relay_log_info *rli= rgi->rli; |
5114 | uint32 const last_errno= rli->last_error().number; |
5115 | |
5116 | if (unlikely(thd->is_error())) |
5117 | { |
5118 | char const *const errmsg= thd->get_stmt_da()->message(); |
5119 | |
5120 | DBUG_PRINT("info" , |
5121 | ("thd->get_stmt_da()->sql_errno()=%d; rli->last_error.number=%d" , |
5122 | thd->get_stmt_da()->sql_errno(), last_errno)); |
5123 | if (last_errno == 0) |
5124 | { |
5125 | /* |
5126 | This function is reporting an error which was not reported |
5127 | while executing exec_relay_log_event(). |
5128 | */ |
5129 | rli->report(ERROR_LEVEL, thd->get_stmt_da()->sql_errno(), |
5130 | rgi->gtid_info(), "%s" , errmsg); |
5131 | } |
5132 | else if (last_errno != thd->get_stmt_da()->sql_errno()) |
5133 | { |
5134 | /* |
5135 | * An error was reported while executing exec_relay_log_event() |
5136 | * however the error code differs from what is in the thread. |
5137 | * This function prints out more information to help finding |
5138 | * what caused the problem. |
5139 | */ |
5140 | sql_print_error("Slave (additional info): %s Error_code: %d" , |
5141 | errmsg, thd->get_stmt_da()->sql_errno()); |
5142 | } |
5143 | } |
5144 | |
5145 | /* Print any warnings issued */ |
5146 | Diagnostics_area::Sql_condition_iterator it= |
5147 | thd->get_stmt_da()->sql_conditions(); |
5148 | const Sql_condition *err; |
5149 | /* |
5150 | Added controlled slave thread cancel for replication |
5151 | of user-defined variables. |
5152 | */ |
5153 | bool udf_error = false; |
5154 | while ((err= it++)) |
5155 | { |
5156 | if (err->get_sql_errno() == ER_CANT_OPEN_LIBRARY) |
5157 | udf_error = true; |
5158 | sql_print_warning("Slave: %s Error_code: %d" , err->get_message_text(), err->get_sql_errno()); |
5159 | } |
5160 | if (unlikely(udf_error)) |
5161 | { |
5162 | StringBuffer<100> tmp; |
5163 | if (rli->mi->using_gtid != Master_info::USE_GTID_NO) |
5164 | { |
5165 | tmp.append(STRING_WITH_LEN("; GTID position '" )); |
5166 | rpl_append_gtid_state(&tmp, false); |
5167 | tmp.append(STRING_WITH_LEN("'" )); |
5168 | } |
5169 | sql_print_error("Error loading user-defined library, slave SQL " |
5170 | "thread aborted. Install the missing library, and restart the " |
5171 | "slave SQL thread with \"SLAVE START\". We stopped at log '%s' " |
5172 | "position %llu%s" , RPL_LOG_NAME, rli->group_master_log_pos, |
5173 | tmp.c_ptr_safe()); |
5174 | } |
5175 | else |
5176 | { |
5177 | StringBuffer<100> tmp; |
5178 | if (rli->mi->using_gtid != Master_info::USE_GTID_NO) |
5179 | { |
5180 | tmp.append(STRING_WITH_LEN("; GTID position '" )); |
5181 | rpl_append_gtid_state(&tmp, false); |
5182 | tmp.append(STRING_WITH_LEN("'" )); |
5183 | } |
5184 | sql_print_error("Error running query, slave SQL thread aborted. " |
5185 | "Fix the problem, and restart the slave SQL thread " |
5186 | "with \"SLAVE START\". We stopped at log '%s' position " |
5187 | "%llu%s" , RPL_LOG_NAME, rli->group_master_log_pos, |
5188 | tmp.c_ptr_safe()); |
5189 | } |
5190 | } |
5191 | |
5192 | |
5193 | /** |
5194 | Slave SQL thread entry point. |
5195 | |
5196 | @param arg Pointer to Relay_log_info object that holds information |
5197 | for the SQL thread. |
5198 | |
5199 | @return Always 0. |
5200 | */ |
5201 | pthread_handler_t handle_slave_sql(void *arg) |
5202 | { |
5203 | THD *thd; /* needs to be first for thread_stack */ |
5204 | char saved_log_name[FN_REFLEN]; |
5205 | char saved_master_log_name[FN_REFLEN]; |
5206 | my_off_t UNINIT_VAR(saved_log_pos); |
5207 | my_off_t UNINIT_VAR(saved_master_log_pos); |
5208 | String saved_skip_gtid_pos; |
5209 | my_off_t saved_skip= 0; |
5210 | Master_info *mi= ((Master_info*)arg); |
5211 | Relay_log_info* rli = &mi->rli; |
5212 | my_bool wsrep_node_dropped __attribute__((unused)) = FALSE; |
5213 | const char *errmsg; |
5214 | rpl_group_info *serial_rgi; |
5215 | rpl_sql_thread_info sql_info(mi->rpl_filter); |
5216 | |
5217 | // needs to call my_thread_init(), otherwise we get a coredump in DBUG_ stuff |
5218 | my_thread_init(); |
5219 | DBUG_ENTER("handle_slave_sql" ); |
5220 | |
5221 | #ifdef WITH_WSREP |
5222 | wsrep_restart_point: |
5223 | #endif |
5224 | |
5225 | serial_rgi= new rpl_group_info(rli); |
5226 | thd = new THD(next_thread_id()); // note that contructor of THD uses DBUG_ ! |
5227 | thd->thread_stack = (char*)&thd; // remember where our stack is |
5228 | thd->system_thread_info.rpl_sql_info= &sql_info; |
5229 | |
5230 | DBUG_ASSERT(rli->inited); |
5231 | DBUG_ASSERT(rli->mi == mi); |
5232 | mysql_mutex_lock(&rli->run_lock); |
5233 | DBUG_ASSERT(!rli->slave_running); |
5234 | errmsg= 0; |
5235 | #ifndef DBUG_OFF |
5236 | rli->events_till_abort = abort_slave_event_count; |
5237 | #endif |
5238 | |
5239 | /* |
5240 | THD for the sql driver thd. In parallel replication this is the thread |
5241 | that reads things from the relay log and calls rpl_parallel::do_event() |
5242 | to execute queries. |
5243 | |
5244 | In single thread replication this is the THD for the thread that is |
5245 | executing SQL queries too. |
5246 | */ |
5247 | serial_rgi->thd= rli->sql_driver_thd= thd; |
5248 | |
5249 | /* Inform waiting threads that slave has started */ |
5250 | rli->slave_run_id++; |
5251 | rli->slave_running= MYSQL_SLAVE_RUN_NOT_CONNECT; |
5252 | |
5253 | pthread_detach_this_thread(); |
5254 | |
5255 | if (opt_slave_parallel_threads > 0 && |
5256 | rpl_parallel_activate_pool(&global_rpl_thread_pool)) |
5257 | { |
5258 | mysql_cond_broadcast(&rli->start_cond); |
5259 | rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, NULL, |
5260 | "Failed during parallel slave pool activation" ); |
5261 | goto err_during_init; |
5262 | } |
5263 | |
5264 | if (init_slave_thread(thd, mi, SLAVE_THD_SQL)) |
5265 | { |
5266 | /* |
5267 | TODO: this is currently broken - slave start and change master |
5268 | will be stuck if we fail here |
5269 | */ |
5270 | mysql_cond_broadcast(&rli->start_cond); |
5271 | rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, NULL, |
5272 | "Failed during slave thread initialization" ); |
5273 | goto err_during_init; |
5274 | } |
5275 | thd->init_for_queries(); |
5276 | thd->rgi_slave= serial_rgi; |
5277 | if ((serial_rgi->deferred_events_collecting= mi->rpl_filter->is_on())) |
5278 | { |
5279 | serial_rgi->deferred_events= new Deferred_log_events(rli); |
5280 | } |
5281 | |
5282 | /* |
5283 | binlog_annotate_row_events must be TRUE only after an Annotate_rows event |
5284 | has been received and only till the last corresponding rbr event has been |
5285 | applied. In all other cases it must be FALSE. |
5286 | */ |
5287 | thd->variables.binlog_annotate_row_events= 0; |
5288 | |
5289 | /* Ensure that slave can exeute any alter table it gets from master */ |
5290 | thd->variables.alter_algorithm= (ulong) Alter_info::ALTER_TABLE_ALGORITHM_DEFAULT; |
5291 | |
5292 | add_to_active_threads(thd); |
5293 | /* |
5294 | We are going to set slave_running to 1. Assuming slave I/O thread is |
5295 | alive and connected, this is going to make Seconds_Behind_Master be 0 |
5296 | i.e. "caught up". Even if we're just at start of thread. Well it's ok, at |
5297 | the moment we start we can think we are caught up, and the next second we |
5298 | start receiving data so we realize we are not caught up and |
5299 | Seconds_Behind_Master grows. No big deal. |
5300 | */ |
5301 | rli->abort_slave = 0; |
5302 | rli->stop_for_until= false; |
5303 | mysql_mutex_unlock(&rli->run_lock); |
5304 | mysql_cond_broadcast(&rli->start_cond); |
5305 | |
5306 | /* |
5307 | Reset errors for a clean start (otherwise, if the master is idle, the SQL |
5308 | thread may execute no Query_log_event, so the error will remain even |
5309 | though there's no problem anymore). Do not reset the master timestamp |
5310 | (imagine the slave has caught everything, the STOP SLAVE and START SLAVE: |
5311 | as we are not sure that we are going to receive a query, we want to |
5312 | remember the last master timestamp (to say how many seconds behind we are |
5313 | now. |
5314 | But the master timestamp is reset by RESET SLAVE & CHANGE MASTER. |
5315 | */ |
5316 | rli->clear_error(); |
5317 | rli->parallel.reset(); |
5318 | |
5319 | //tell the I/O thread to take relay_log_space_limit into account from now on |
5320 | rli->ignore_log_space_limit= 0; |
5321 | |
5322 | serial_rgi->gtid_sub_id= 0; |
5323 | serial_rgi->gtid_pending= false; |
5324 | if (mi->using_gtid != Master_info::USE_GTID_NO && mi->using_parallel() && |
5325 | rli->restart_gtid_pos.count() > 0) |
5326 | { |
5327 | /* |
5328 | With parallel replication in GTID mode, if we have a multi-domain GTID |
5329 | position, we need to start some way back in the relay log and skip any |
5330 | GTID that was already applied before. Since event groups can be split |
5331 | across multiple relay logs, this earlier starting point may be in the |
5332 | middle of an already applied event group, so we also need to skip any |
5333 | remaining part of such group. |
5334 | */ |
5335 | rli->gtid_skip_flag = GTID_SKIP_TRANSACTION; |
5336 | } |
5337 | else |
5338 | rli->gtid_skip_flag = GTID_SKIP_NOT; |
5339 | if (init_relay_log_pos(rli, |
5340 | rli->group_relay_log_name, |
5341 | rli->group_relay_log_pos, |
5342 | 1 /*need data lock*/, &errmsg, |
5343 | 1 /*look for a description_event*/)) |
5344 | { |
5345 | rli->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, NULL, |
5346 | "Error initializing relay log position: %s" , errmsg); |
5347 | goto err_before_start; |
5348 | } |
5349 | rli->reset_inuse_relaylog(); |
5350 | if (rli->alloc_inuse_relaylog(rli->group_relay_log_name)) |
5351 | goto err_before_start; |
5352 | |
5353 | strcpy(rli->future_event_master_log_name, rli->group_master_log_name); |
5354 | THD_CHECK_SENTRY(thd); |
5355 | #ifndef DBUG_OFF |
5356 | { |
5357 | DBUG_PRINT("info" , ("my_b_tell(rli->cur_log)=%llu " |
5358 | "rli->event_relay_log_pos=%llu" , |
5359 | my_b_tell(rli->cur_log), rli->event_relay_log_pos)); |
5360 | DBUG_ASSERT(rli->event_relay_log_pos >= BIN_LOG_HEADER_SIZE); |
5361 | /* |
5362 | Wonder if this is correct. I (Guilhem) wonder if my_b_tell() returns the |
5363 | correct position when it's called just after my_b_seek() (the questionable |
5364 | stuff is those "seek is done on next read" comments in the my_b_seek() |
5365 | source code). |
5366 | The crude reality is that this assertion randomly fails whereas |
5367 | replication seems to work fine. And there is no easy explanation why it |
5368 | fails (as we my_b_seek(rli->event_relay_log_pos) at the very end of |
5369 | init_relay_log_pos() called above). Maybe the assertion would be |
5370 | meaningful if we held rli->data_lock between the my_b_seek() and the |
5371 | DBUG_ASSERT(). |
5372 | */ |
5373 | #ifdef SHOULD_BE_CHECKED |
5374 | DBUG_ASSERT(my_b_tell(rli->cur_log) == rli->event_relay_log_pos); |
5375 | #endif |
5376 | } |
5377 | #endif |
5378 | |
5379 | #ifdef WITH_WSREP |
5380 | thd->wsrep_exec_mode= LOCAL_STATE; |
5381 | /* synchronize with wsrep replication */ |
5382 | if (WSREP_ON) |
5383 | wsrep_ready_wait(); |
5384 | #endif |
5385 | DBUG_PRINT("master_info" ,("log_file_name: %s position: %llu" , |
5386 | rli->group_master_log_name, |
5387 | rli->group_master_log_pos)); |
5388 | if (global_system_variables.log_warnings) |
5389 | { |
5390 | StringBuffer<100> tmp; |
5391 | if (mi->using_gtid != Master_info::USE_GTID_NO) |
5392 | { |
5393 | tmp.append(STRING_WITH_LEN("; GTID position '" )); |
5394 | rpl_append_gtid_state(&tmp, |
5395 | mi->using_gtid==Master_info::USE_GTID_CURRENT_POS); |
5396 | tmp.append(STRING_WITH_LEN("'" )); |
5397 | } |
5398 | sql_print_information("Slave SQL thread initialized, starting replication " |
5399 | "in log '%s' at position %llu, relay log '%s' " |
5400 | "position: %llu%s" , RPL_LOG_NAME, |
5401 | rli->group_master_log_pos, rli->group_relay_log_name, |
5402 | rli->group_relay_log_pos, tmp.c_ptr_safe()); |
5403 | } |
5404 | |
5405 | if (check_temp_dir(rli->slave_patternload_file)) |
5406 | { |
5407 | check_temp_dir_result= thd->get_stmt_da()->sql_errno(); |
5408 | rli->report(ERROR_LEVEL, thd->get_stmt_da()->sql_errno(), NULL, |
5409 | "Unable to use slave's temporary directory %s - %s" , |
5410 | slave_load_tmpdir, thd->get_stmt_da()->message()); |
5411 | goto err; |
5412 | } |
5413 | else |
5414 | check_temp_dir_result= 0; |
5415 | |
5416 | /* Load the set of seen GTIDs, if we did not already. */ |
5417 | if (rpl_load_gtid_slave_state(thd)) |
5418 | { |
5419 | rli->report(ERROR_LEVEL, thd->get_stmt_da()->sql_errno(), NULL, |
5420 | "Unable to load replication GTID slave state from mysql.%s: %s" , |
5421 | rpl_gtid_slave_state_table_name.str, |
5422 | thd->get_stmt_da()->message()); |
5423 | /* |
5424 | If we are using old-style replication, we can continue, even though we |
5425 | then will not be able to record the GTIDs we receive. But if using GTID, |
5426 | we must give up. |
5427 | */ |
5428 | if (mi->using_gtid != Master_info::USE_GTID_NO || opt_gtid_strict_mode) |
5429 | goto err; |
5430 | } |
5431 | /* Re-load the set of mysql.gtid_slave_posXXX tables available. */ |
5432 | if (find_gtid_slave_pos_tables(thd)) |
5433 | { |
5434 | rli->report(ERROR_LEVEL, thd->get_stmt_da()->sql_errno(), NULL, |
5435 | "Error processing replication GTID position tables: %s" , |
5436 | thd->get_stmt_da()->message()); |
5437 | goto err; |
5438 | } |
5439 | |
5440 | /* execute init_slave variable */ |
5441 | if (opt_init_slave.length) |
5442 | { |
5443 | execute_init_command(thd, &opt_init_slave, &LOCK_sys_init_slave); |
5444 | if (unlikely(thd->is_slave_error)) |
5445 | { |
5446 | rli->report(ERROR_LEVEL, thd->get_stmt_da()->sql_errno(), NULL, |
5447 | "Slave SQL thread aborted. Can't execute init_slave query" ); |
5448 | goto err; |
5449 | } |
5450 | } |
5451 | |
5452 | /* |
5453 | First check until condition - probably there is nothing to execute. We |
5454 | do not want to wait for next event in this case. |
5455 | */ |
5456 | mysql_mutex_lock(&rli->data_lock); |
5457 | if (rli->slave_skip_counter) |
5458 | { |
5459 | strmake_buf(saved_log_name, rli->group_relay_log_name); |
5460 | strmake_buf(saved_master_log_name, rli->group_master_log_name); |
5461 | saved_log_pos= rli->group_relay_log_pos; |
5462 | saved_master_log_pos= rli->group_master_log_pos; |
5463 | if (mi->using_gtid != Master_info::USE_GTID_NO) |
5464 | { |
5465 | saved_skip_gtid_pos.append(STRING_WITH_LEN(", GTID '" )); |
5466 | rpl_append_gtid_state(&saved_skip_gtid_pos, false); |
5467 | saved_skip_gtid_pos.append(STRING_WITH_LEN("'; " )); |
5468 | } |
5469 | saved_skip= rli->slave_skip_counter; |
5470 | } |
5471 | if ((rli->until_condition == Relay_log_info::UNTIL_MASTER_POS || |
5472 | rli->until_condition == Relay_log_info::UNTIL_RELAY_POS) && |
5473 | rli->is_until_satisfied(rli->group_master_log_pos)) |
5474 | { |
5475 | sql_print_information("Slave SQL thread stopped because it reached its" |
5476 | " UNTIL position %llu" , rli->until_pos()); |
5477 | mysql_mutex_unlock(&rli->data_lock); |
5478 | goto err; |
5479 | } |
5480 | mysql_mutex_unlock(&rli->data_lock); |
5481 | |
5482 | /* Read queries from the IO/THREAD until this thread is killed */ |
5483 | |
5484 | thd->set_command(COM_SLAVE_SQL); |
5485 | while (!sql_slave_killed(serial_rgi)) |
5486 | { |
5487 | THD_STAGE_INFO(thd, stage_reading_event_from_the_relay_log); |
5488 | THD_CHECK_SENTRY(thd); |
5489 | |
5490 | if (saved_skip && rli->slave_skip_counter == 0) |
5491 | { |
5492 | StringBuffer<100> tmp; |
5493 | if (mi->using_gtid != Master_info::USE_GTID_NO) |
5494 | { |
5495 | tmp.append(STRING_WITH_LEN(", GTID '" )); |
5496 | rpl_append_gtid_state(&tmp, false); |
5497 | tmp.append(STRING_WITH_LEN("'; " )); |
5498 | } |
5499 | |
5500 | sql_print_information("'SQL_SLAVE_SKIP_COUNTER=%ld' executed at " |
5501 | "relay_log_file='%s', relay_log_pos='%ld', master_log_name='%s', " |
5502 | "master_log_pos='%ld'%s and new position at " |
5503 | "relay_log_file='%s', relay_log_pos='%ld', master_log_name='%s', " |
5504 | "master_log_pos='%ld'%s " , |
5505 | (ulong) saved_skip, saved_log_name, (ulong) saved_log_pos, |
5506 | saved_master_log_name, (ulong) saved_master_log_pos, |
5507 | saved_skip_gtid_pos.c_ptr_safe(), |
5508 | rli->group_relay_log_name, (ulong) rli->group_relay_log_pos, |
5509 | rli->group_master_log_name, (ulong) rli->group_master_log_pos, |
5510 | tmp.c_ptr_safe()); |
5511 | saved_skip= 0; |
5512 | saved_skip_gtid_pos.free(); |
5513 | } |
5514 | |
5515 | if (exec_relay_log_event(thd, rli, serial_rgi)) |
5516 | { |
5517 | #ifdef WITH_WSREP |
5518 | if (thd->wsrep_conflict_state != NO_CONFLICT) |
5519 | { |
5520 | wsrep_node_dropped= TRUE; |
5521 | rli->abort_slave= TRUE; |
5522 | } |
5523 | #endif /* WITH_WSREP */ |
5524 | |
5525 | DBUG_PRINT("info" , ("exec_relay_log_event() failed" )); |
5526 | // do not scare the user if SQL thread was simply killed or stopped |
5527 | if (!sql_slave_killed(serial_rgi)) |
5528 | { |
5529 | slave_output_error_info(serial_rgi, thd); |
5530 | if (WSREP_ON && rli->last_error().number == ER_UNKNOWN_COM_ERROR) |
5531 | wsrep_node_dropped= TRUE; |
5532 | } |
5533 | goto err; |
5534 | } |
5535 | } |
5536 | |
5537 | err: |
5538 | if (mi->using_parallel()) |
5539 | rli->parallel.wait_for_done(thd, rli); |
5540 | |
5541 | /* Thread stopped. Print the current replication position to the log */ |
5542 | { |
5543 | StringBuffer<100> tmp; |
5544 | if (mi->using_gtid != Master_info::USE_GTID_NO) |
5545 | { |
5546 | tmp.append(STRING_WITH_LEN("; GTID position '" )); |
5547 | rpl_append_gtid_state(&tmp, false); |
5548 | tmp.append(STRING_WITH_LEN("'" )); |
5549 | } |
5550 | sql_print_information("Slave SQL thread exiting, replication stopped in " |
5551 | "log '%s' at position %llu%s" , RPL_LOG_NAME, |
5552 | rli->group_master_log_pos, tmp.c_ptr_safe()); |
5553 | } |
5554 | |
5555 | err_before_start: |
5556 | |
5557 | /* |
5558 | Some events set some playgrounds, which won't be cleared because thread |
5559 | stops. Stopping of this thread may not be known to these events ("stop" |
5560 | request is detected only by the present function, not by events), so we |
5561 | must "proactively" clear playgrounds: |
5562 | */ |
5563 | thd->clear_error(); |
5564 | serial_rgi->cleanup_context(thd, 1); |
5565 | /* |
5566 | Some extra safety, which should not been needed (normally, event deletion |
5567 | should already have done these assignments (each event which sets these |
5568 | variables is supposed to set them to 0 before terminating)). |
5569 | */ |
5570 | thd->catalog= 0; |
5571 | thd->reset_query(); |
5572 | thd->reset_db(&null_clex_str); |
5573 | if (rli->mi->using_gtid != Master_info::USE_GTID_NO) |
5574 | { |
5575 | ulong domain_count; |
5576 | my_bool save_log_all_errors= thd->log_all_errors; |
5577 | |
5578 | /* |
5579 | We don't need to check return value for rli->flush() |
5580 | as any errors should be logged to stderr |
5581 | */ |
5582 | thd->log_all_errors= 1; |
5583 | rli->flush(); |
5584 | thd->log_all_errors= save_log_all_errors; |
5585 | if (mi->using_parallel()) |
5586 | { |
5587 | /* |
5588 | In parallel replication GTID mode, we may stop with different domains |
5589 | at different positions in the relay log. |
5590 | |
5591 | To handle this when we restart the SQL thread, mark the current |
5592 | per-domain position in the Relay_log_info. |
5593 | */ |
5594 | mysql_mutex_lock(&rpl_global_gtid_slave_state->LOCK_slave_state); |
5595 | domain_count= rpl_global_gtid_slave_state->count(); |
5596 | mysql_mutex_unlock(&rpl_global_gtid_slave_state->LOCK_slave_state); |
5597 | if (domain_count > 1) |
5598 | { |
5599 | inuse_relaylog *ir; |
5600 | |
5601 | /* |
5602 | Load the starting GTID position, so that we can skip already applied |
5603 | GTIDs when we restart the SQL thread. And set the start position in |
5604 | the relay log back to a known safe place to start (prior to any not |
5605 | yet applied transaction in any domain). |
5606 | */ |
5607 | rli->restart_gtid_pos.load(rpl_global_gtid_slave_state, NULL, 0); |
5608 | if ((ir= rli->inuse_relaylog_list)) |
5609 | { |
5610 | rpl_gtid *gtid= ir->relay_log_state; |
5611 | uint32 count= ir->relay_log_state_count; |
5612 | while (count > 0) |
5613 | { |
5614 | process_gtid_for_restart_pos(rli, gtid); |
5615 | ++gtid; |
5616 | --count; |
5617 | } |
5618 | strmake_buf(rli->group_relay_log_name, ir->name); |
5619 | rli->group_relay_log_pos= BIN_LOG_HEADER_SIZE; |
5620 | rli->relay_log_state.load(ir->relay_log_state, ir->relay_log_state_count); |
5621 | } |
5622 | } |
5623 | } |
5624 | } |
5625 | THD_STAGE_INFO(thd, stage_waiting_for_slave_mutex_on_exit); |
5626 | thd->add_status_to_global(); |
5627 | unlink_not_visible_thd(thd); |
5628 | mysql_mutex_lock(&rli->run_lock); |
5629 | |
5630 | err_during_init: |
5631 | /* We need data_lock, at least to wake up any waiting master_pos_wait() */ |
5632 | mysql_mutex_lock(&rli->data_lock); |
5633 | DBUG_ASSERT(rli->slave_running == MYSQL_SLAVE_RUN_NOT_CONNECT); // tracking buffer overrun |
5634 | /* When master_pos_wait() wakes up it will check this and terminate */ |
5635 | rli->slave_running= MYSQL_SLAVE_NOT_RUN; |
5636 | /* Forget the relay log's format */ |
5637 | delete rli->relay_log.description_event_for_exec; |
5638 | rli->relay_log.description_event_for_exec= 0; |
5639 | rli->reset_inuse_relaylog(); |
5640 | /* Wake up master_pos_wait() */ |
5641 | mysql_mutex_unlock(&rli->data_lock); |
5642 | DBUG_PRINT("info" ,("Signaling possibly waiting master_pos_wait() functions" )); |
5643 | mysql_cond_broadcast(&rli->data_cond); |
5644 | rli->ignore_log_space_limit= 0; /* don't need any lock */ |
5645 | /* we die so won't remember charset - re-update them on next thread start */ |
5646 | thd->system_thread_info.rpl_sql_info->cached_charset_invalidate(); |
5647 | |
5648 | /* |
5649 | TODO: see if we can do this conditionally in next_event() instead |
5650 | to avoid unneeded position re-init |
5651 | |
5652 | We only reset THD::temporary_tables to 0 here and not free it, as this |
5653 | could be used by slave through Relay_log_info::save_temporary_tables. |
5654 | */ |
5655 | thd->temporary_tables= 0; |
5656 | rli->sql_driver_thd= 0; |
5657 | thd->rgi_fake= thd->rgi_slave= NULL; |
5658 | |
5659 | #ifdef WITH_WSREP |
5660 | /* |
5661 | If slave stopped due to node going non primary, we set global flag to |
5662 | trigger automatic restart of slave when node joins back to cluster. |
5663 | */ |
5664 | if (WSREP_ON && wsrep_node_dropped && wsrep_restart_slave) |
5665 | { |
5666 | if (wsrep_ready) |
5667 | { |
5668 | WSREP_INFO("Slave error due to node temporarily non-primary" |
5669 | "SQL slave will continue" ); |
5670 | wsrep_node_dropped= FALSE; |
5671 | mysql_mutex_unlock(&rli->run_lock); |
5672 | WSREP_DEBUG("wsrep_conflict_state now: %d" , thd->wsrep_conflict_state); |
5673 | WSREP_INFO("slave restart: %d" , thd->wsrep_conflict_state); |
5674 | thd->wsrep_conflict_state= NO_CONFLICT; |
5675 | goto wsrep_restart_point; |
5676 | } else { |
5677 | WSREP_INFO("Slave error due to node going non-primary" ); |
5678 | WSREP_INFO("wsrep_restart_slave was set and therefore slave will be " |
5679 | "automatically restarted when node joins back to cluster." ); |
5680 | wsrep_restart_slave_activated= TRUE; |
5681 | } |
5682 | } |
5683 | #endif /* WITH_WSREP */ |
5684 | |
5685 | /* |
5686 | Note: the order of the broadcast and unlock calls below (first |
5687 | broadcast, then unlock) is important. Otherwise a killer_thread can |
5688 | execute between the calls and delete the mi structure leading to a |
5689 | crash! (see BUG#25306 for details) |
5690 | */ |
5691 | mysql_cond_broadcast(&rli->stop_cond); |
5692 | DBUG_EXECUTE_IF("simulate_slave_delay_at_terminate_bug38694" , sleep(5);); |
5693 | mysql_mutex_unlock(&rli->run_lock); // tell the world we are done |
5694 | |
5695 | rpl_parallel_resize_pool_if_no_slaves(); |
5696 | |
5697 | delete serial_rgi; |
5698 | delete thd; |
5699 | thread_safe_decrement32(&service_thread_count); |
5700 | signal_thd_deleted(); |
5701 | |
5702 | DBUG_LEAVE; // Must match DBUG_ENTER() |
5703 | my_thread_end(); |
5704 | ERR_remove_state(0); |
5705 | pthread_exit(0); |
5706 | return 0; // Avoid compiler warnings |
5707 | } |
5708 | |
5709 | |
5710 | /* |
5711 | process_io_create_file() |
5712 | */ |
5713 | |
5714 | static int process_io_create_file(Master_info* mi, Create_file_log_event* cev) |
5715 | { |
5716 | int error = 1; |
5717 | ulong num_bytes; |
5718 | bool cev_not_written; |
5719 | THD *thd = mi->io_thd; |
5720 | NET *net = &mi->mysql->net; |
5721 | DBUG_ENTER("process_io_create_file" ); |
5722 | |
5723 | if (unlikely(!cev->is_valid())) |
5724 | DBUG_RETURN(1); |
5725 | |
5726 | if (!mi->rpl_filter->db_ok(cev->db)) |
5727 | { |
5728 | skip_load_data_infile(net); |
5729 | DBUG_RETURN(0); |
5730 | } |
5731 | DBUG_ASSERT(cev->inited_from_old); |
5732 | thd->file_id = cev->file_id = mi->file_id++; |
5733 | thd->variables.server_id = cev->server_id; |
5734 | cev_not_written = 1; |
5735 | |
5736 | if (unlikely(net_request_file(net,cev->fname))) |
5737 | { |
5738 | sql_print_error("Slave I/O: failed requesting download of '%s'" , |
5739 | cev->fname); |
5740 | goto err; |
5741 | } |
5742 | |
5743 | /* |
5744 | This dummy block is so we could instantiate Append_block_log_event |
5745 | once and then modify it slightly instead of doing it multiple times |
5746 | in the loop |
5747 | */ |
5748 | { |
5749 | Append_block_log_event aev(thd,0,0,0,0); |
5750 | |
5751 | for (;;) |
5752 | { |
5753 | if (unlikely((num_bytes=my_net_read(net)) == packet_error)) |
5754 | { |
5755 | sql_print_error("Network read error downloading '%s' from master" , |
5756 | cev->fname); |
5757 | goto err; |
5758 | } |
5759 | if (unlikely(!num_bytes)) /* eof */ |
5760 | { |
5761 | /* 3.23 master wants it */ |
5762 | net_write_command(net, 0, (uchar*) "" , 0, (uchar*) "" , 0); |
5763 | /* |
5764 | If we wrote Create_file_log_event, then we need to write |
5765 | Execute_load_log_event. If we did not write Create_file_log_event, |
5766 | then this is an empty file and we can just do as if the LOAD DATA |
5767 | INFILE had not existed, i.e. write nothing. |
5768 | */ |
5769 | if (unlikely(cev_not_written)) |
5770 | break; |
5771 | Execute_load_log_event xev(thd,0,0); |
5772 | xev.log_pos = cev->log_pos; |
5773 | if (unlikely(mi->rli.relay_log.append(&xev))) |
5774 | { |
5775 | mi->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_WRITE_FAILURE, NULL, |
5776 | ER_THD(thd, ER_SLAVE_RELAY_LOG_WRITE_FAILURE), |
5777 | "error writing Exec_load event to relay log" ); |
5778 | goto err; |
5779 | } |
5780 | mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total); |
5781 | break; |
5782 | } |
5783 | if (unlikely(cev_not_written)) |
5784 | { |
5785 | cev->block = net->read_pos; |
5786 | cev->block_len = num_bytes; |
5787 | if (unlikely(mi->rli.relay_log.append(cev))) |
5788 | { |
5789 | mi->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_WRITE_FAILURE, NULL, |
5790 | ER_THD(thd, ER_SLAVE_RELAY_LOG_WRITE_FAILURE), |
5791 | "error writing Create_file event to relay log" ); |
5792 | goto err; |
5793 | } |
5794 | cev_not_written=0; |
5795 | mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total); |
5796 | } |
5797 | else |
5798 | { |
5799 | aev.block = net->read_pos; |
5800 | aev.block_len = num_bytes; |
5801 | aev.log_pos = cev->log_pos; |
5802 | if (unlikely(mi->rli.relay_log.append(&aev))) |
5803 | { |
5804 | mi->report(ERROR_LEVEL, ER_SLAVE_RELAY_LOG_WRITE_FAILURE, NULL, |
5805 | ER_THD(thd, ER_SLAVE_RELAY_LOG_WRITE_FAILURE), |
5806 | "error writing Append_block event to relay log" ); |
5807 | goto err; |
5808 | } |
5809 | mi->rli.relay_log.harvest_bytes_written(&mi->rli.log_space_total) ; |
5810 | } |
5811 | } |
5812 | } |
5813 | error=0; |
5814 | err: |
5815 | DBUG_RETURN(error); |
5816 | } |
5817 | |
5818 | |
5819 | /* |
5820 | Start using a new binary log on the master |
5821 | |
5822 | SYNOPSIS |
5823 | process_io_rotate() |
5824 | mi master_info for the slave |
5825 | rev The rotate log event read from the binary log |
5826 | |
5827 | DESCRIPTION |
5828 | Updates the master info with the place in the next binary |
5829 | log where we should start reading. |
5830 | Rotate the relay log to avoid mixed-format relay logs. |
5831 | |
5832 | NOTES |
5833 | We assume we already locked mi->data_lock |
5834 | |
5835 | RETURN VALUES |
5836 | 0 ok |
5837 | 1 Log event is illegal |
5838 | |
5839 | */ |
5840 | |
5841 | static int process_io_rotate(Master_info *mi, Rotate_log_event *rev) |
5842 | { |
5843 | DBUG_ENTER("process_io_rotate" ); |
5844 | mysql_mutex_assert_owner(&mi->data_lock); |
5845 | |
5846 | if (unlikely(!rev->is_valid())) |
5847 | DBUG_RETURN(1); |
5848 | |
5849 | /* Safe copy as 'rev' has been "sanitized" in Rotate_log_event's ctor */ |
5850 | memcpy(mi->master_log_name, rev->new_log_ident, rev->ident_len+1); |
5851 | mi->master_log_pos= rev->pos; |
5852 | DBUG_PRINT("info" , ("master_log_pos: '%s' %lu" , |
5853 | mi->master_log_name, (ulong) mi->master_log_pos)); |
5854 | #ifndef DBUG_OFF |
5855 | /* |
5856 | If we do not do this, we will be getting the first |
5857 | rotate event forever, so we need to not disconnect after one. |
5858 | */ |
5859 | if (disconnect_slave_event_count) |
5860 | mi->events_till_disconnect++; |
5861 | #endif |
5862 | |
5863 | /* |
5864 | If description_event_for_queue is format <4, there is conversion in the |
5865 | relay log to the slave's format (4). And Rotate can mean upgrade or |
5866 | nothing. If upgrade, it's to 5.0 or newer, so we will get a Format_desc, so |
5867 | no need to reset description_event_for_queue now. And if it's nothing (same |
5868 | master version as before), no need (still using the slave's format). |
5869 | */ |
5870 | if (mi->rli.relay_log.description_event_for_queue->binlog_version >= 4) |
5871 | { |
5872 | DBUG_ASSERT(mi->rli.relay_log.description_event_for_queue->checksum_alg == |
5873 | mi->rli.relay_log.relay_log_checksum_alg); |
5874 | |
5875 | delete mi->rli.relay_log.description_event_for_queue; |
5876 | /* start from format 3 (MySQL 4.0) again */ |
5877 | mi->rli.relay_log.description_event_for_queue= new |
5878 | Format_description_log_event(3); |
5879 | mi->rli.relay_log.description_event_for_queue->checksum_alg= |
5880 | mi->rli.relay_log.relay_log_checksum_alg; |
5881 | } |
5882 | /* |
5883 | Rotate the relay log makes binlog format detection easier (at next slave |
5884 | start or mysqlbinlog) |
5885 | */ |
5886 | DBUG_RETURN(rotate_relay_log(mi) /* will take the right mutexes */); |
5887 | } |
5888 | |
5889 | /* |
5890 | Reads a 3.23 event and converts it to the slave's format. This code was |
5891 | copied from MySQL 4.0. |
5892 | */ |
5893 | static int queue_binlog_ver_1_event(Master_info *mi, const char *buf, |
5894 | ulong event_len) |
5895 | { |
5896 | const char *errmsg = 0; |
5897 | ulong inc_pos; |
5898 | bool ignore_event= 0; |
5899 | char *tmp_buf = 0; |
5900 | Relay_log_info *rli= &mi->rli; |
5901 | DBUG_ENTER("queue_binlog_ver_1_event" ); |
5902 | |
5903 | /* |
5904 | If we get Load event, we need to pass a non-reusable buffer |
5905 | to read_log_event, so we do a trick |
5906 | */ |
5907 | if ((uchar)buf[EVENT_TYPE_OFFSET] == LOAD_EVENT) |
5908 | { |
5909 | if (unlikely(!(tmp_buf=(char*)my_malloc(event_len+1,MYF(MY_WME))))) |
5910 | { |
5911 | mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, NULL, |
5912 | ER(ER_SLAVE_FATAL_ERROR), "Memory allocation failed" ); |
5913 | DBUG_RETURN(1); |
5914 | } |
5915 | memcpy(tmp_buf,buf,event_len); |
5916 | /* |
5917 | Create_file constructor wants a 0 as last char of buffer, this 0 will |
5918 | serve as the string-termination char for the file's name (which is at the |
5919 | end of the buffer) |
5920 | We must increment event_len, otherwise the event constructor will not see |
5921 | this end 0, which leads to segfault. |
5922 | */ |
5923 | tmp_buf[event_len++]=0; |
5924 | int4store(tmp_buf+EVENT_LEN_OFFSET, event_len); |
5925 | buf = (const char*)tmp_buf; |
5926 | } |
5927 | /* |
5928 | This will transform LOAD_EVENT into CREATE_FILE_EVENT, ask the master to |
5929 | send the loaded file, and write it to the relay log in the form of |
5930 | Append_block/Exec_load (the SQL thread needs the data, as that thread is not |
5931 | connected to the master). |
5932 | */ |
5933 | Log_event *ev= |
5934 | Log_event::read_log_event(buf, event_len, &errmsg, |
5935 | mi->rli.relay_log.description_event_for_queue, 0); |
5936 | if (unlikely(!ev)) |
5937 | { |
5938 | sql_print_error("Read invalid event from master: '%s',\ |
5939 | master could be corrupt but a more likely cause of this is a bug" , |
5940 | errmsg); |
5941 | my_free(tmp_buf); |
5942 | DBUG_RETURN(1); |
5943 | } |
5944 | |
5945 | mysql_mutex_lock(&mi->data_lock); |
5946 | ev->log_pos= mi->master_log_pos; /* 3.23 events don't contain log_pos */ |
5947 | switch (ev->get_type_code()) { |
5948 | case STOP_EVENT: |
5949 | ignore_event= 1; |
5950 | inc_pos= event_len; |
5951 | break; |
5952 | case ROTATE_EVENT: |
5953 | if (unlikely(process_io_rotate(mi,(Rotate_log_event*)ev))) |
5954 | { |
5955 | delete ev; |
5956 | mysql_mutex_unlock(&mi->data_lock); |
5957 | DBUG_RETURN(1); |
5958 | } |
5959 | inc_pos= 0; |
5960 | break; |
5961 | case CREATE_FILE_EVENT: |
5962 | /* |
5963 | Yes it's possible to have CREATE_FILE_EVENT here, even if we're in |
5964 | queue_old_event() which is for 3.23 events which don't comprise |
5965 | CREATE_FILE_EVENT. This is because read_log_event() above has just |
5966 | transformed LOAD_EVENT into CREATE_FILE_EVENT. |
5967 | */ |
5968 | { |
5969 | /* We come here when and only when tmp_buf != 0 */ |
5970 | DBUG_ASSERT(tmp_buf != 0); |
5971 | inc_pos=event_len; |
5972 | ev->log_pos+= inc_pos; |
5973 | int error = process_io_create_file(mi,(Create_file_log_event*)ev); |
5974 | delete ev; |
5975 | mi->master_log_pos += inc_pos; |
5976 | DBUG_PRINT("info" , ("master_log_pos: %lu" , (ulong) mi->master_log_pos)); |
5977 | mysql_mutex_unlock(&mi->data_lock); |
5978 | my_free(tmp_buf); |
5979 | DBUG_RETURN(error); |
5980 | } |
5981 | default: |
5982 | inc_pos= event_len; |
5983 | break; |
5984 | } |
5985 | if (likely(!ignore_event)) |
5986 | { |
5987 | if (ev->log_pos) |
5988 | /* |
5989 | Don't do it for fake Rotate events (see comment in |
5990 | Log_event::Log_event(const char* buf...) in log_event.cc). |
5991 | */ |
5992 | ev->log_pos+= event_len; /* make log_pos be the pos of the end of the event */ |
5993 | if (unlikely(rli->relay_log.append(ev))) |
5994 | { |
5995 | delete ev; |
5996 | mysql_mutex_unlock(&mi->data_lock); |
5997 | DBUG_RETURN(1); |
5998 | } |
5999 | rli->relay_log.harvest_bytes_written(&rli->log_space_total); |
6000 | } |
6001 | delete ev; |
6002 | mi->master_log_pos+= inc_pos; |
6003 | DBUG_PRINT("info" , ("master_log_pos: %lu" , (ulong) mi->master_log_pos)); |
6004 | mysql_mutex_unlock(&mi->data_lock); |
6005 | DBUG_RETURN(0); |
6006 | } |
6007 | |
6008 | /* |
6009 | Reads a 4.0 event and converts it to the slave's format. This code was copied |
6010 | from queue_binlog_ver_1_event(), with some affordable simplifications. |
6011 | */ |
6012 | static int queue_binlog_ver_3_event(Master_info *mi, const char *buf, |
6013 | ulong event_len) |
6014 | { |
6015 | const char *errmsg = 0; |
6016 | ulong inc_pos; |
6017 | char *tmp_buf = 0; |
6018 | Relay_log_info *rli= &mi->rli; |
6019 | DBUG_ENTER("queue_binlog_ver_3_event" ); |
6020 | |
6021 | /* read_log_event() will adjust log_pos to be end_log_pos */ |
6022 | Log_event *ev= |
6023 | Log_event::read_log_event(buf,event_len, &errmsg, |
6024 | mi->rli.relay_log.description_event_for_queue, 0); |
6025 | if (unlikely(!ev)) |
6026 | { |
6027 | sql_print_error("Read invalid event from master: '%s',\ |
6028 | master could be corrupt but a more likely cause of this is a bug" , |
6029 | errmsg); |
6030 | my_free(tmp_buf); |
6031 | DBUG_RETURN(1); |
6032 | } |
6033 | mysql_mutex_lock(&mi->data_lock); |
6034 | switch (ev->get_type_code()) { |
6035 | case STOP_EVENT: |
6036 | goto err; |
6037 | case ROTATE_EVENT: |
6038 | if (unlikely(process_io_rotate(mi,(Rotate_log_event*)ev))) |
6039 | { |
6040 | delete ev; |
6041 | mysql_mutex_unlock(&mi->data_lock); |
6042 | DBUG_RETURN(1); |
6043 | } |
6044 | inc_pos= 0; |
6045 | break; |
6046 | default: |
6047 | inc_pos= event_len; |
6048 | break; |
6049 | } |
6050 | |
6051 | if (unlikely(rli->relay_log.append(ev))) |
6052 | { |
6053 | delete ev; |
6054 | mysql_mutex_unlock(&mi->data_lock); |
6055 | DBUG_RETURN(1); |
6056 | } |
6057 | rli->relay_log.harvest_bytes_written(&rli->log_space_total); |
6058 | delete ev; |
6059 | mi->master_log_pos+= inc_pos; |
6060 | err: |
6061 | DBUG_PRINT("info" , ("master_log_pos: %lu" , (ulong) mi->master_log_pos)); |
6062 | mysql_mutex_unlock(&mi->data_lock); |
6063 | DBUG_RETURN(0); |
6064 | } |
6065 | |
6066 | /* |
6067 | queue_old_event() |
6068 | |
6069 | Writes a 3.23 or 4.0 event to the relay log, after converting it to the 5.0 |
6070 | (exactly, slave's) format. To do the conversion, we create a 5.0 event from |
6071 | the 3.23/4.0 bytes, then write this event to the relay log. |
6072 | |
6073 | TODO: |
6074 | Test this code before release - it has to be tested on a separate |
6075 | setup with 3.23 master or 4.0 master |
6076 | */ |
6077 | |
6078 | static int queue_old_event(Master_info *mi, const char *buf, |
6079 | ulong event_len) |
6080 | { |
6081 | DBUG_ENTER("queue_old_event" ); |
6082 | |
6083 | switch (mi->rli.relay_log.description_event_for_queue->binlog_version) |
6084 | { |
6085 | case 1: |
6086 | DBUG_RETURN(queue_binlog_ver_1_event(mi,buf,event_len)); |
6087 | case 3: |
6088 | DBUG_RETURN(queue_binlog_ver_3_event(mi,buf,event_len)); |
6089 | default: /* unsupported format; eg version 2 */ |
6090 | DBUG_PRINT("info" ,("unsupported binlog format %d in queue_old_event()" , |
6091 | mi->rli.relay_log.description_event_for_queue->binlog_version)); |
6092 | DBUG_RETURN(1); |
6093 | } |
6094 | } |
6095 | |
6096 | /* |
6097 | queue_event() |
6098 | |
6099 | If the event is 3.23/4.0, passes it to queue_old_event() which will convert |
6100 | it. Otherwise, writes a 5.0 (or newer) event to the relay log. Then there is |
6101 | no format conversion, it's pure read/write of bytes. |
6102 | So a 5.0.0 slave's relay log can contain events in the slave's format or in |
6103 | any >=5.0.0 format. |
6104 | */ |
6105 | |
6106 | static int queue_event(Master_info* mi,const char* buf, ulong event_len) |
6107 | { |
6108 | int error= 0; |
6109 | StringBuffer<1024> error_msg; |
6110 | ulonglong inc_pos= 0; |
6111 | ulonglong event_pos; |
6112 | Relay_log_info *rli= &mi->rli; |
6113 | mysql_mutex_t *log_lock= rli->relay_log.get_log_lock(); |
6114 | ulong s_id; |
6115 | bool unlock_data_lock= TRUE; |
6116 | bool gtid_skip_enqueue= false; |
6117 | bool got_gtid_event= false; |
6118 | rpl_gtid event_gtid; |
6119 | static uint dbug_rows_event_count __attribute__((unused))= 0; |
6120 | bool is_compress_event = false; |
6121 | char* new_buf = NULL; |
6122 | char new_buf_arr[4096]; |
6123 | bool is_malloc = false; |
6124 | bool is_rows_event= false; |
6125 | /* |
6126 | FD_q must have been prepared for the first R_a event |
6127 | inside get_master_version_and_clock() |
6128 | Show-up of FD:s affects checksum_alg at once because |
6129 | that changes FD_queue. |
6130 | */ |
6131 | enum enum_binlog_checksum_alg checksum_alg= |
6132 | mi->checksum_alg_before_fd != BINLOG_CHECKSUM_ALG_UNDEF ? |
6133 | mi->checksum_alg_before_fd : mi->rli.relay_log.relay_log_checksum_alg; |
6134 | |
6135 | char *save_buf= NULL; // needed for checksumming the fake Rotate event |
6136 | char rot_buf[LOG_EVENT_HEADER_LEN + ROTATE_HEADER_LEN + FN_REFLEN]; |
6137 | |
6138 | DBUG_ASSERT(checksum_alg == BINLOG_CHECKSUM_ALG_OFF || |
6139 | checksum_alg == BINLOG_CHECKSUM_ALG_UNDEF || |
6140 | checksum_alg == BINLOG_CHECKSUM_ALG_CRC32); |
6141 | |
6142 | DBUG_ENTER("queue_event" ); |
6143 | /* |
6144 | FD_queue checksum alg description does not apply in a case of |
6145 | FD itself. The one carries both parts of the checksum data. |
6146 | */ |
6147 | if (buf[EVENT_TYPE_OFFSET] == FORMAT_DESCRIPTION_EVENT) |
6148 | { |
6149 | checksum_alg= get_checksum_alg(buf, event_len); |
6150 | } |
6151 | else if (buf[EVENT_TYPE_OFFSET] == START_EVENT_V3) |
6152 | { |
6153 | // checksum behaviour is similar to the pre-checksum FD handling |
6154 | mi->checksum_alg_before_fd= BINLOG_CHECKSUM_ALG_UNDEF; |
6155 | mi->rli.relay_log.description_event_for_queue->checksum_alg= |
6156 | mi->rli.relay_log.relay_log_checksum_alg= checksum_alg= |
6157 | BINLOG_CHECKSUM_ALG_OFF; |
6158 | } |
6159 | |
6160 | // does not hold always because of old binlog can work with NM |
6161 | // DBUG_ASSERT(checksum_alg != BINLOG_CHECKSUM_ALG_UNDEF); |
6162 | |
6163 | // should hold unless manipulations with RL. Tests that do that |
6164 | // will have to refine the clause. |
6165 | DBUG_ASSERT(mi->rli.relay_log.relay_log_checksum_alg != |
6166 | BINLOG_CHECKSUM_ALG_UNDEF); |
6167 | |
6168 | // Emulate the network corruption |
6169 | DBUG_EXECUTE_IF("corrupt_queue_event" , |
6170 | if ((uchar)buf[EVENT_TYPE_OFFSET] != FORMAT_DESCRIPTION_EVENT) |
6171 | { |
6172 | char *debug_event_buf_c = (char*) buf; |
6173 | int debug_cor_pos = rand() % (event_len - BINLOG_CHECKSUM_LEN); |
6174 | debug_event_buf_c[debug_cor_pos] =~ debug_event_buf_c[debug_cor_pos]; |
6175 | DBUG_PRINT("info" , ("Corrupt the event at queue_event: byte on position %d" , debug_cor_pos)); |
6176 | DBUG_SET("-d,corrupt_queue_event" ); |
6177 | } |
6178 | ); |
6179 | |
6180 | if (event_checksum_test((uchar *) buf, event_len, checksum_alg)) |
6181 | { |
6182 | error= ER_NETWORK_READ_EVENT_CHECKSUM_FAILURE; |
6183 | unlock_data_lock= FALSE; |
6184 | goto err; |
6185 | } |
6186 | |
6187 | if (mi->rli.relay_log.description_event_for_queue->binlog_version<4 && |
6188 | (uchar)buf[EVENT_TYPE_OFFSET] != FORMAT_DESCRIPTION_EVENT /* a way to escape */) |
6189 | DBUG_RETURN(queue_old_event(mi,buf,event_len)); |
6190 | |
6191 | #ifdef ENABLED_DEBUG_SYNC |
6192 | /* |
6193 | A (+d,dbug.rows_events_to_delay_relay_logging)-test is supposed to |
6194 | create a few Write_log_events and after receiving the 1st of them |
6195 | the IO thread signals to launch the SQL thread, and sets itself to |
6196 | wait for a release signal. |
6197 | */ |
6198 | DBUG_EXECUTE_IF("dbug.rows_events_to_delay_relay_logging" , |
6199 | if ((buf[EVENT_TYPE_OFFSET] == WRITE_ROWS_EVENT_V1 || |
6200 | buf[EVENT_TYPE_OFFSET] == WRITE_ROWS_EVENT) && |
6201 | ++dbug_rows_event_count == 2) |
6202 | { |
6203 | const char act[]= |
6204 | "now SIGNAL start_sql_thread " |
6205 | "WAIT_FOR go_on_relay_logging" ; |
6206 | DBUG_ASSERT(debug_sync_service); |
6207 | DBUG_ASSERT(!debug_sync_set_action(current_thd, |
6208 | STRING_WITH_LEN(act))); |
6209 | };); |
6210 | #endif |
6211 | mysql_mutex_lock(&mi->data_lock); |
6212 | |
6213 | switch ((uchar)buf[EVENT_TYPE_OFFSET]) { |
6214 | case STOP_EVENT: |
6215 | /* |
6216 | We needn't write this event to the relay log. Indeed, it just indicates a |
6217 | master server shutdown. The only thing this does is cleaning. But |
6218 | cleaning is already done on a per-master-thread basis (as the master |
6219 | server is shutting down cleanly, it has written all DROP TEMPORARY TABLE |
6220 | prepared statements' deletion are TODO only when we binlog prep stmts). |
6221 | |
6222 | We don't even increment mi->master_log_pos, because we may be just after |
6223 | a Rotate event. Btw, in a few milliseconds we are going to have a Start |
6224 | event from the next binlog (unless the master is presently running |
6225 | without --log-bin). |
6226 | */ |
6227 | goto err; |
6228 | case ROTATE_EVENT: |
6229 | { |
6230 | Rotate_log_event rev(buf, checksum_alg != BINLOG_CHECKSUM_ALG_OFF ? |
6231 | event_len - BINLOG_CHECKSUM_LEN : event_len, |
6232 | mi->rli.relay_log.description_event_for_queue); |
6233 | |
6234 | if (unlikely(mi->gtid_reconnect_event_skip_count) && |
6235 | unlikely(!mi->gtid_event_seen) && |
6236 | rev.is_artificial_event() && |
6237 | (mi->prev_master_id != mi->master_id || |
6238 | strcmp(rev.new_log_ident, mi->master_log_name) != 0)) |
6239 | { |
6240 | /* |
6241 | Artificial Rotate_log_event is the first event we receive at the start |
6242 | of each master binlog file. It gives the name of the new binlog file. |
6243 | |
6244 | Normally, we already have this name from the real rotate event at the |
6245 | end of the previous binlog file (unless we are making a new connection |
6246 | using GTID). But if the master server restarted/crashed, there is no |
6247 | rotate event at the end of the prior binlog file, so the name is new. |
6248 | |
6249 | We use this fact to handle a special case of master crashing. If the |
6250 | master crashed while writing the binlog, it might end with a partial |
6251 | event group lacking the COMMIT/XID event, which must be rolled |
6252 | back. If the slave IO thread happens to get a disconnect in the middle |
6253 | of exactly this event group, it will try to reconnect at the same GTID |
6254 | and skip already fetched events. However, that GTID did not commit on |
6255 | the master before the crash, so it does not really exist, and the |
6256 | master will connect the slave at the next following GTID starting in |
6257 | the next binlog. This could confuse the slave and make it mix the |
6258 | start of one event group with the end of another. |
6259 | |
6260 | But we detect this case here, by noticing the change of binlog name |
6261 | which detects the missing rotate event at the end of the previous |
6262 | binlog file. In this case, we reset the counters to make us not skip |
6263 | the next event group, and queue an artificial Format Description |
6264 | event. The previously fetched incomplete event group will then be |
6265 | rolled back when the Format Description event is executed by the SQL |
6266 | thread. |
6267 | |
6268 | A similar case is if the reconnect somehow connects to a different |
6269 | master server (like due to a network proxy or IP address takeover). |
6270 | We detect this case by noticing a change of server_id and in this |
6271 | case likewise rollback the partially received event group. |
6272 | */ |
6273 | Format_description_log_event fdle(4); |
6274 | |
6275 | if (mi->prev_master_id != mi->master_id) |
6276 | sql_print_warning("The server_id of master server changed in the " |
6277 | "middle of GTID %u-%u-%llu. Assuming a change of " |
6278 | "master server, so rolling back the previously " |
6279 | "received partial transaction. Expected: %lu, " |
6280 | "received: %lu" , mi->last_queued_gtid.domain_id, |
6281 | mi->last_queued_gtid.server_id, |
6282 | mi->last_queued_gtid.seq_no, |
6283 | mi->prev_master_id, mi->master_id); |
6284 | else if (strcmp(rev.new_log_ident, mi->master_log_name) != 0) |
6285 | sql_print_warning("Unexpected change of master binlog file name in the " |
6286 | "middle of GTID %u-%u-%llu, assuming that master has " |
6287 | "crashed and rolling back the transaction. Expected: " |
6288 | "'%s', received: '%s'" , |
6289 | mi->last_queued_gtid.domain_id, |
6290 | mi->last_queued_gtid.server_id, |
6291 | mi->last_queued_gtid.seq_no, |
6292 | mi->master_log_name, rev.new_log_ident); |
6293 | |
6294 | mysql_mutex_lock(log_lock); |
6295 | if (likely(!rli->relay_log.write_event(&fdle) && |
6296 | !rli->relay_log.flush_and_sync(NULL))) |
6297 | { |
6298 | rli->relay_log.harvest_bytes_written(&rli->log_space_total); |
6299 | } |
6300 | else |
6301 | { |
6302 | error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE; |
6303 | mysql_mutex_unlock(log_lock); |
6304 | goto err; |
6305 | } |
6306 | rli->relay_log.signal_relay_log_update(); |
6307 | mysql_mutex_unlock(log_lock); |
6308 | |
6309 | mi->gtid_reconnect_event_skip_count= 0; |
6310 | mi->events_queued_since_last_gtid= 0; |
6311 | } |
6312 | mi->prev_master_id= mi->master_id; |
6313 | |
6314 | if (unlikely(process_io_rotate(mi, &rev))) |
6315 | { |
6316 | error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE; |
6317 | goto err; |
6318 | } |
6319 | /* |
6320 | Checksum special cases for the fake Rotate (R_f) event caused by the protocol |
6321 | of events generation and serialization in RL where Rotate of master is |
6322 | queued right next to FD of slave. |
6323 | Since it's only FD that carries the alg desc of FD_s has to apply to R_m. |
6324 | Two special rules apply only to the first R_f which comes in before any FD_m. |
6325 | The 2nd R_f should be compatible with the FD_s that must have taken over |
6326 | the last seen FD_m's (A). |
6327 | |
6328 | RSC_1: If OM \and fake Rotate \and slave is configured to |
6329 | to compute checksum for its first FD event for RL |
6330 | the fake Rotate gets checksummed here. |
6331 | */ |
6332 | if (uint4korr(&buf[0]) == 0 && checksum_alg == BINLOG_CHECKSUM_ALG_OFF && |
6333 | mi->rli.relay_log.relay_log_checksum_alg != BINLOG_CHECKSUM_ALG_OFF) |
6334 | { |
6335 | ha_checksum rot_crc= 0; |
6336 | event_len += BINLOG_CHECKSUM_LEN; |
6337 | memcpy(rot_buf, buf, event_len - BINLOG_CHECKSUM_LEN); |
6338 | int4store(&rot_buf[EVENT_LEN_OFFSET], |
6339 | uint4korr(&rot_buf[EVENT_LEN_OFFSET]) + BINLOG_CHECKSUM_LEN); |
6340 | rot_crc= my_checksum(rot_crc, (const uchar *) rot_buf, |
6341 | event_len - BINLOG_CHECKSUM_LEN); |
6342 | int4store(&rot_buf[event_len - BINLOG_CHECKSUM_LEN], rot_crc); |
6343 | DBUG_ASSERT(event_len == uint4korr(&rot_buf[EVENT_LEN_OFFSET])); |
6344 | DBUG_ASSERT(mi->rli.relay_log.description_event_for_queue->checksum_alg == |
6345 | mi->rli.relay_log.relay_log_checksum_alg); |
6346 | /* the first one */ |
6347 | DBUG_ASSERT(mi->checksum_alg_before_fd != BINLOG_CHECKSUM_ALG_UNDEF); |
6348 | save_buf= (char *) buf; |
6349 | buf= rot_buf; |
6350 | } |
6351 | else |
6352 | /* |
6353 | RSC_2: If NM \and fake Rotate \and slave does not compute checksum |
6354 | the fake Rotate's checksum is stripped off before relay-logging. |
6355 | */ |
6356 | if (uint4korr(&buf[0]) == 0 && checksum_alg != BINLOG_CHECKSUM_ALG_OFF && |
6357 | mi->rli.relay_log.relay_log_checksum_alg == BINLOG_CHECKSUM_ALG_OFF) |
6358 | { |
6359 | event_len -= BINLOG_CHECKSUM_LEN; |
6360 | memcpy(rot_buf, buf, event_len); |
6361 | int4store(&rot_buf[EVENT_LEN_OFFSET], |
6362 | uint4korr(&rot_buf[EVENT_LEN_OFFSET]) - BINLOG_CHECKSUM_LEN); |
6363 | DBUG_ASSERT(event_len == uint4korr(&rot_buf[EVENT_LEN_OFFSET])); |
6364 | DBUG_ASSERT(mi->rli.relay_log.description_event_for_queue->checksum_alg == |
6365 | mi->rli.relay_log.relay_log_checksum_alg); |
6366 | /* the first one */ |
6367 | DBUG_ASSERT(mi->checksum_alg_before_fd != BINLOG_CHECKSUM_ALG_UNDEF); |
6368 | save_buf= (char *) buf; |
6369 | buf= rot_buf; |
6370 | } |
6371 | /* |
6372 | Now the I/O thread has just changed its mi->master_log_name, so |
6373 | incrementing mi->master_log_pos is nonsense. |
6374 | */ |
6375 | inc_pos= 0; |
6376 | break; |
6377 | } |
6378 | case FORMAT_DESCRIPTION_EVENT: |
6379 | { |
6380 | /* |
6381 | Create an event, and save it (when we rotate the relay log, we will have |
6382 | to write this event again). |
6383 | */ |
6384 | /* |
6385 | We are the only thread which reads/writes description_event_for_queue. |
6386 | The relay_log struct does not move (though some members of it can |
6387 | change), so we needn't any lock (no rli->data_lock, no log lock). |
6388 | */ |
6389 | Format_description_log_event* tmp; |
6390 | const char* errmsg; |
6391 | // mark it as undefined that is irrelevant anymore |
6392 | mi->checksum_alg_before_fd= BINLOG_CHECKSUM_ALG_UNDEF; |
6393 | if (!(tmp= (Format_description_log_event*) |
6394 | Log_event::read_log_event(buf, event_len, &errmsg, |
6395 | mi->rli.relay_log.description_event_for_queue, |
6396 | 1))) |
6397 | { |
6398 | error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE; |
6399 | goto err; |
6400 | } |
6401 | tmp->copy_crypto_data(mi->rli.relay_log.description_event_for_queue); |
6402 | delete mi->rli.relay_log.description_event_for_queue; |
6403 | mi->rli.relay_log.description_event_for_queue= tmp; |
6404 | if (tmp->checksum_alg == BINLOG_CHECKSUM_ALG_UNDEF) |
6405 | tmp->checksum_alg= BINLOG_CHECKSUM_ALG_OFF; |
6406 | |
6407 | /* installing new value of checksum Alg for relay log */ |
6408 | mi->rli.relay_log.relay_log_checksum_alg= tmp->checksum_alg; |
6409 | |
6410 | /* |
6411 | Do not queue any format description event that we receive after a |
6412 | reconnect where we are skipping over a partial event group received |
6413 | before the reconnect. |
6414 | |
6415 | (If we queued such an event, and it was the first format_description |
6416 | event after master restart, the slave SQL thread would think that |
6417 | the partial event group before it in the relay log was from a |
6418 | previous master crash and should be rolled back). |
6419 | */ |
6420 | if (unlikely(mi->gtid_reconnect_event_skip_count && !mi->gtid_event_seen)) |
6421 | gtid_skip_enqueue= true; |
6422 | |
6423 | /* |
6424 | Though this does some conversion to the slave's format, this will |
6425 | preserve the master's binlog format version, and number of event types. |
6426 | */ |
6427 | /* |
6428 | If the event was not requested by the slave (the slave did not ask for |
6429 | it), i.e. has end_log_pos=0, we do not increment mi->master_log_pos |
6430 | */ |
6431 | inc_pos= uint4korr(buf+LOG_POS_OFFSET) ? event_len : 0; |
6432 | DBUG_PRINT("info" ,("binlog format is now %d" , |
6433 | mi->rli.relay_log.description_event_for_queue->binlog_version)); |
6434 | |
6435 | } |
6436 | break; |
6437 | |
6438 | case HEARTBEAT_LOG_EVENT: |
6439 | { |
6440 | /* |
6441 | HB (heartbeat) cannot come before RL (Relay) |
6442 | */ |
6443 | Heartbeat_log_event hb(buf, |
6444 | mi->rli.relay_log.relay_log_checksum_alg |
6445 | != BINLOG_CHECKSUM_ALG_OFF ? |
6446 | event_len - BINLOG_CHECKSUM_LEN : event_len, |
6447 | mi->rli.relay_log.description_event_for_queue); |
6448 | if (!hb.is_valid()) |
6449 | { |
6450 | error= ER_SLAVE_HEARTBEAT_FAILURE; |
6451 | error_msg.append(STRING_WITH_LEN("inconsistent heartbeat event content;" )); |
6452 | error_msg.append(STRING_WITH_LEN("the event's data: log_file_name " )); |
6453 | error_msg.append(hb.get_log_ident(), (uint) strlen(hb.get_log_ident())); |
6454 | error_msg.append(STRING_WITH_LEN(" log_pos " )); |
6455 | error_msg.append_ulonglong(hb.log_pos); |
6456 | goto err; |
6457 | } |
6458 | mi->received_heartbeats++; |
6459 | /* |
6460 | compare local and event's versions of log_file, log_pos. |
6461 | |
6462 | Heartbeat is sent only after an event corresponding to the corrdinates |
6463 | the heartbeat carries. |
6464 | Slave can not have a higher coordinate except in the only |
6465 | special case when mi->master_log_name, master_log_pos have never |
6466 | been updated by Rotate event i.e when slave does not have any history |
6467 | with the master (and thereafter mi->master_log_pos is NULL). |
6468 | |
6469 | Slave can have lower coordinates, if some event from master was omitted. |
6470 | |
6471 | TODO: handling `when' for SHOW SLAVE STATUS' snds behind |
6472 | */ |
6473 | if (memcmp(mi->master_log_name, hb.get_log_ident(), hb.get_ident_len()) || |
6474 | mi->master_log_pos > hb.log_pos) { |
6475 | /* missed events of heartbeat from the past */ |
6476 | error= ER_SLAVE_HEARTBEAT_FAILURE; |
6477 | error_msg.append(STRING_WITH_LEN("heartbeat is not compatible with local info;" )); |
6478 | error_msg.append(STRING_WITH_LEN("the event's data: log_file_name " )); |
6479 | error_msg.append(hb.get_log_ident(), (uint) strlen(hb.get_log_ident())); |
6480 | error_msg.append(STRING_WITH_LEN(" log_pos " )); |
6481 | error_msg.append_ulonglong(hb.log_pos); |
6482 | goto err; |
6483 | } |
6484 | |
6485 | /* |
6486 | Heartbeat events doesn't count in the binlog size, so we don't have to |
6487 | increment mi->master_log_pos |
6488 | */ |
6489 | goto skip_relay_logging; |
6490 | } |
6491 | break; |
6492 | |
6493 | case GTID_LIST_EVENT: |
6494 | { |
6495 | const char *errmsg; |
6496 | Gtid_list_log_event *glev; |
6497 | Log_event *tmp; |
6498 | uint32 flags; |
6499 | |
6500 | if (!(tmp= Log_event::read_log_event(buf, event_len, &errmsg, |
6501 | mi->rli.relay_log.description_event_for_queue, |
6502 | opt_slave_sql_verify_checksum))) |
6503 | { |
6504 | error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE; |
6505 | goto err; |
6506 | } |
6507 | glev= static_cast<Gtid_list_log_event *>(tmp); |
6508 | event_pos= glev->log_pos; |
6509 | flags= glev->gl_flags; |
6510 | delete glev; |
6511 | |
6512 | /* |
6513 | We use fake Gtid_list events to update the old-style position (among |
6514 | other things). |
6515 | |
6516 | Early code created fake Gtid_list events with zero log_pos, those should |
6517 | not modify old-style position. |
6518 | */ |
6519 | if (event_pos == 0 || event_pos <= mi->master_log_pos) |
6520 | inc_pos= 0; |
6521 | else |
6522 | inc_pos= event_pos - mi->master_log_pos; |
6523 | |
6524 | if (mi->rli.until_condition == Relay_log_info::UNTIL_GTID && |
6525 | flags & Gtid_list_log_event::FLAG_UNTIL_REACHED) |
6526 | { |
6527 | char str_buf[128]; |
6528 | String str(str_buf, sizeof(str_buf), system_charset_info); |
6529 | mi->rli.until_gtid_pos.to_string(&str); |
6530 | sql_print_information("Slave I/O thread stops because it reached its" |
6531 | " UNTIL master_gtid_pos %s" , str.c_ptr_safe()); |
6532 | mi->abort_slave= true; |
6533 | } |
6534 | } |
6535 | break; |
6536 | |
6537 | case GTID_EVENT: |
6538 | { |
6539 | DBUG_EXECUTE_IF("kill_slave_io_after_2_events" , |
6540 | { |
6541 | mi->dbug_do_disconnect= true; |
6542 | mi->dbug_event_counter= 2; |
6543 | };); |
6544 | |
6545 | uchar gtid_flag; |
6546 | |
6547 | if (Gtid_log_event::peek(buf, event_len, checksum_alg, |
6548 | &event_gtid.domain_id, &event_gtid.server_id, |
6549 | &event_gtid.seq_no, >id_flag, |
6550 | rli->relay_log.description_event_for_queue)) |
6551 | { |
6552 | error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE; |
6553 | goto err; |
6554 | } |
6555 | got_gtid_event= true; |
6556 | if (mi->using_gtid == Master_info::USE_GTID_NO) |
6557 | goto default_action; |
6558 | if (unlikely(mi->gtid_reconnect_event_skip_count)) |
6559 | { |
6560 | if (likely(!mi->gtid_event_seen)) |
6561 | { |
6562 | mi->gtid_event_seen= true; |
6563 | /* |
6564 | If we are reconnecting, and we need to skip a partial event group |
6565 | already queued to the relay log before the reconnect, then we check |
6566 | that we actually get the same event group (same GTID) as before, so |
6567 | we do not end up with half of one group and half another. |
6568 | |
6569 | The only way we should be able to receive a different GTID than what |
6570 | we expect is if the binlog on the master (or more likely the whole |
6571 | master server) was replaced with a different one, on the same IP |
6572 | address, _and_ the new master happens to have domains in a different |
6573 | order so we get the GTID from a different domain first. Still, it is |
6574 | best to protect against this case. |
6575 | */ |
6576 | if (event_gtid.domain_id != mi->last_queued_gtid.domain_id || |
6577 | event_gtid.server_id != mi->last_queued_gtid.server_id || |
6578 | event_gtid.seq_no != mi->last_queued_gtid.seq_no) |
6579 | { |
6580 | bool first; |
6581 | error= ER_SLAVE_UNEXPECTED_MASTER_SWITCH; |
6582 | error_msg.append(STRING_WITH_LEN("Expected: " )); |
6583 | first= true; |
6584 | rpl_slave_state_tostring_helper(&error_msg, &mi->last_queued_gtid, |
6585 | &first); |
6586 | error_msg.append(STRING_WITH_LEN(", received: " )); |
6587 | first= true; |
6588 | rpl_slave_state_tostring_helper(&error_msg, &event_gtid, &first); |
6589 | goto err; |
6590 | } |
6591 | if (global_system_variables.log_warnings > 1) |
6592 | { |
6593 | bool first= true; |
6594 | StringBuffer<1024> gtid_text; |
6595 | rpl_slave_state_tostring_helper(>id_text, &mi->last_queued_gtid, |
6596 | &first); |
6597 | sql_print_information("Slave IO thread is reconnected to " |
6598 | "receive Gtid_log_event %s. It is to skip %llu " |
6599 | "already received events including the gtid one" , |
6600 | gtid_text.ptr(), |
6601 | mi->events_queued_since_last_gtid); |
6602 | } |
6603 | goto default_action; |
6604 | } |
6605 | else |
6606 | { |
6607 | bool first; |
6608 | StringBuffer<1024> gtid_text; |
6609 | |
6610 | gtid_text.append(STRING_WITH_LEN("Last received gtid: " )); |
6611 | first= true; |
6612 | rpl_slave_state_tostring_helper(>id_text, &mi->last_queued_gtid, |
6613 | &first); |
6614 | gtid_text.append(STRING_WITH_LEN(", currently received: " )); |
6615 | first= true; |
6616 | rpl_slave_state_tostring_helper(>id_text, &event_gtid, &first); |
6617 | |
6618 | error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE; |
6619 | sql_print_error("Slave IO thread has received a new Gtid_log_event " |
6620 | "while skipping already logged events " |
6621 | "after reconnect. %s. %llu remains to be skipped. " |
6622 | "The number of originally read events was %llu" , |
6623 | gtid_text.ptr(), |
6624 | mi->gtid_reconnect_event_skip_count, |
6625 | mi->events_queued_since_last_gtid); |
6626 | goto err; |
6627 | } |
6628 | } |
6629 | mi->gtid_event_seen= true; |
6630 | |
6631 | /* |
6632 | We have successfully queued to relay log everything before this GTID, so |
6633 | in case of reconnect we can start from after any previous GTID. |
6634 | (Normally we would have updated gtid_current_pos earlier at the end of |
6635 | the previous event group, but better leave an extra check here for |
6636 | safety). |
6637 | */ |
6638 | if (mi->events_queued_since_last_gtid) |
6639 | { |
6640 | mi->gtid_current_pos.update(&mi->last_queued_gtid); |
6641 | mi->events_queued_since_last_gtid= 0; |
6642 | } |
6643 | mi->last_queued_gtid= event_gtid; |
6644 | mi->last_queued_gtid_standalone= |
6645 | (gtid_flag & Gtid_log_event::FL_STANDALONE) != 0; |
6646 | |
6647 | /* Should filter all the subsequent events in the current GTID group? */ |
6648 | mi->domain_id_filter.do_filter(event_gtid.domain_id); |
6649 | |
6650 | ++mi->events_queued_since_last_gtid; |
6651 | inc_pos= event_len; |
6652 | } |
6653 | break; |
6654 | /* |
6655 | Binlog compressed event should uncompress in IO thread |
6656 | */ |
6657 | case QUERY_COMPRESSED_EVENT: |
6658 | inc_pos= event_len; |
6659 | if (query_event_uncompress(rli->relay_log.description_event_for_queue, |
6660 | checksum_alg == BINLOG_CHECKSUM_ALG_CRC32, |
6661 | buf, event_len, new_buf_arr, sizeof(new_buf_arr), |
6662 | &is_malloc, (char **)&new_buf, &event_len)) |
6663 | { |
6664 | char llbuf[22]; |
6665 | error = ER_BINLOG_UNCOMPRESS_ERROR; |
6666 | error_msg.append(STRING_WITH_LEN("binlog uncompress error, master log_pos: " )); |
6667 | llstr(mi->master_log_pos, llbuf); |
6668 | error_msg.append(llbuf, strlen(llbuf)); |
6669 | goto err; |
6670 | } |
6671 | buf = new_buf; |
6672 | is_compress_event = true; |
6673 | goto default_action; |
6674 | |
6675 | case WRITE_ROWS_COMPRESSED_EVENT: |
6676 | case UPDATE_ROWS_COMPRESSED_EVENT: |
6677 | case DELETE_ROWS_COMPRESSED_EVENT: |
6678 | case WRITE_ROWS_COMPRESSED_EVENT_V1: |
6679 | case UPDATE_ROWS_COMPRESSED_EVENT_V1: |
6680 | case DELETE_ROWS_COMPRESSED_EVENT_V1: |
6681 | inc_pos = event_len; |
6682 | { |
6683 | if (row_log_event_uncompress(rli->relay_log.description_event_for_queue, |
6684 | checksum_alg == BINLOG_CHECKSUM_ALG_CRC32, |
6685 | buf, event_len, new_buf_arr, sizeof(new_buf_arr), |
6686 | &is_malloc, (char **)&new_buf, &event_len)) |
6687 | { |
6688 | char llbuf[22]; |
6689 | error = ER_BINLOG_UNCOMPRESS_ERROR; |
6690 | error_msg.append(STRING_WITH_LEN("binlog uncompress error, master log_pos: " )); |
6691 | llstr(mi->master_log_pos, llbuf); |
6692 | error_msg.append(llbuf, strlen(llbuf)); |
6693 | goto err; |
6694 | } |
6695 | } |
6696 | is_compress_event = true; |
6697 | buf = new_buf; |
6698 | /* |
6699 | As we are uncertain about compressed V2 rows events, we don't track |
6700 | them |
6701 | */ |
6702 | if (LOG_EVENT_IS_ROW_V2((Log_event_type) buf[EVENT_TYPE_OFFSET])) |
6703 | goto default_action; |
6704 | /* fall through */ |
6705 | case WRITE_ROWS_EVENT_V1: |
6706 | case UPDATE_ROWS_EVENT_V1: |
6707 | case DELETE_ROWS_EVENT_V1: |
6708 | case WRITE_ROWS_EVENT: |
6709 | case UPDATE_ROWS_EVENT: |
6710 | case DELETE_ROWS_EVENT: |
6711 | { |
6712 | is_rows_event= true; |
6713 | mi->rows_event_tracker.update(mi->master_log_name, |
6714 | mi->master_log_pos, |
6715 | buf, |
6716 | mi->rli.relay_log. |
6717 | description_event_for_queue); |
6718 | |
6719 | DBUG_EXECUTE_IF("simulate_stmt_end_rows_event_loss" , |
6720 | { |
6721 | mi->rows_event_tracker.stmt_end_seen= false; |
6722 | }); |
6723 | } |
6724 | goto default_action; |
6725 | |
6726 | #ifndef DBUG_OFF |
6727 | case XID_EVENT: |
6728 | DBUG_EXECUTE_IF("slave_discard_xid_for_gtid_0_x_1000" , |
6729 | { |
6730 | /* Inject an event group that is missing its XID commit event. */ |
6731 | if (mi->last_queued_gtid.domain_id == 0 && |
6732 | mi->last_queued_gtid.seq_no == 1000) |
6733 | goto skip_relay_logging; |
6734 | }); |
6735 | #endif |
6736 | /* fall through */ |
6737 | default: |
6738 | default_action: |
6739 | DBUG_EXECUTE_IF("kill_slave_io_after_2_events" , |
6740 | { |
6741 | if (mi->dbug_do_disconnect && |
6742 | (LOG_EVENT_IS_QUERY((Log_event_type)(uchar)buf[EVENT_TYPE_OFFSET]) || |
6743 | ((uchar)buf[EVENT_TYPE_OFFSET] == TABLE_MAP_EVENT)) |
6744 | && (--mi->dbug_event_counter == 0)) |
6745 | { |
6746 | error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE; |
6747 | mi->dbug_do_disconnect= false; /* Safety */ |
6748 | goto err; |
6749 | } |
6750 | };); |
6751 | |
6752 | DBUG_EXECUTE_IF("kill_slave_io_before_commit" , |
6753 | { |
6754 | if ((uchar)buf[EVENT_TYPE_OFFSET] == XID_EVENT || |
6755 | ((uchar)buf[EVENT_TYPE_OFFSET] == QUERY_EVENT && /* QUERY_COMPRESSED_EVENT would never be commmit or rollback */ |
6756 | Query_log_event::peek_is_commit_rollback(buf, event_len, |
6757 | checksum_alg))) |
6758 | { |
6759 | error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE; |
6760 | goto err; |
6761 | } |
6762 | };); |
6763 | |
6764 | if (mi->using_gtid != Master_info::USE_GTID_NO && mi->gtid_event_seen) |
6765 | { |
6766 | if (unlikely(mi->gtid_reconnect_event_skip_count)) |
6767 | { |
6768 | --mi->gtid_reconnect_event_skip_count; |
6769 | gtid_skip_enqueue= true; |
6770 | } |
6771 | else if (mi->events_queued_since_last_gtid) |
6772 | ++mi->events_queued_since_last_gtid; |
6773 | } |
6774 | |
6775 | if (!is_compress_event) |
6776 | inc_pos= event_len; |
6777 | |
6778 | break; |
6779 | } |
6780 | |
6781 | /* |
6782 | Integrity of Rows- event group check. |
6783 | A sequence of Rows- events must end with STMT_END_F flagged one. |
6784 | Even when Heartbeat event interrupts Rows- events flow this must indicate a |
6785 | malfunction e.g logging on the master. |
6786 | */ |
6787 | if (((uchar) buf[EVENT_TYPE_OFFSET] != HEARTBEAT_LOG_EVENT) && |
6788 | !is_rows_event && |
6789 | mi->rows_event_tracker.check_and_report(mi->master_log_name, |
6790 | mi->master_log_pos)) |
6791 | { |
6792 | error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE; |
6793 | goto err; |
6794 | } |
6795 | |
6796 | /* |
6797 | If we filter events master-side (eg. @@skip_replication), we will see holes |
6798 | in the event positions from the master. If we see such a hole, adjust |
6799 | mi->master_log_pos accordingly so we maintain the correct position (for |
6800 | reconnect, MASTER_POS_WAIT(), etc.) |
6801 | */ |
6802 | if (inc_pos > 0 && |
6803 | event_len >= LOG_POS_OFFSET+4 && |
6804 | (event_pos= uint4korr(buf+LOG_POS_OFFSET)) > mi->master_log_pos + inc_pos) |
6805 | { |
6806 | inc_pos= event_pos - mi->master_log_pos; |
6807 | DBUG_PRINT("info" , ("Adjust master_log_pos %llu->%llu to account for " |
6808 | "master-side filtering" , |
6809 | mi->master_log_pos + inc_pos, event_pos)); |
6810 | } |
6811 | |
6812 | /* |
6813 | If this event is originating from this server, don't queue it. |
6814 | We don't check this for 3.23 events because it's simpler like this; 3.23 |
6815 | will be filtered anyway by the SQL slave thread which also tests the |
6816 | server id (we must also keep this test in the SQL thread, in case somebody |
6817 | upgrades a 4.0 slave which has a not-filtered relay log). |
6818 | |
6819 | ANY event coming from ourselves can be ignored: it is obvious for queries; |
6820 | for STOP_EVENT/ROTATE_EVENT/START_EVENT: these cannot come from ourselves |
6821 | (--log-slave-updates would not log that) unless this slave is also its |
6822 | direct master (an unsupported, useless setup!). |
6823 | */ |
6824 | |
6825 | mysql_mutex_lock(log_lock); |
6826 | s_id= uint4korr(buf + SERVER_ID_OFFSET); |
6827 | /* |
6828 | Write the event to the relay log, unless we reconnected in the middle |
6829 | of an event group and now need to skip the initial part of the group that |
6830 | we already wrote before reconnecting. |
6831 | */ |
6832 | if (unlikely(gtid_skip_enqueue)) |
6833 | { |
6834 | mi->master_log_pos+= inc_pos; |
6835 | if ((uchar)buf[EVENT_TYPE_OFFSET] == FORMAT_DESCRIPTION_EVENT && |
6836 | s_id == mi->master_id) |
6837 | { |
6838 | /* |
6839 | If we write this master's description event in the middle of an event |
6840 | group due to GTID reconnect, SQL thread will think that master crashed |
6841 | in the middle of the group and roll back the first half, so we must not. |
6842 | |
6843 | But we still have to write an artificial copy of the masters description |
6844 | event, to override the initial slave-version description event so that |
6845 | SQL thread has the right information for parsing the events it reads. |
6846 | */ |
6847 | rli->relay_log.description_event_for_queue->created= 0; |
6848 | rli->relay_log.description_event_for_queue->set_artificial_event(); |
6849 | if (rli->relay_log.append_no_lock |
6850 | (rli->relay_log.description_event_for_queue)) |
6851 | error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE; |
6852 | else |
6853 | rli->relay_log.harvest_bytes_written(&rli->log_space_total); |
6854 | } |
6855 | else if (mi->gtid_reconnect_event_skip_count == 0) |
6856 | { |
6857 | /* |
6858 | Add a fake rotate event so that SQL thread can see the old-style |
6859 | position where we re-connected in the middle of a GTID event group. |
6860 | */ |
6861 | Rotate_log_event fake_rev(mi->master_log_name, 0, mi->master_log_pos, 0); |
6862 | fake_rev.server_id= mi->master_id; |
6863 | if (rli->relay_log.append_no_lock(&fake_rev)) |
6864 | error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE; |
6865 | else |
6866 | rli->relay_log.harvest_bytes_written(&rli->log_space_total); |
6867 | } |
6868 | } |
6869 | else |
6870 | if ((s_id == global_system_variables.server_id && |
6871 | !mi->rli.replicate_same_server_id) || |
6872 | event_that_should_be_ignored(buf) || |
6873 | /* |
6874 | the following conjunction deals with IGNORE_SERVER_IDS, if set |
6875 | If the master is on the ignore list, execution of |
6876 | format description log events and rotate events is necessary. |
6877 | */ |
6878 | (mi->ignore_server_ids.elements > 0 && |
6879 | mi->shall_ignore_server_id(s_id) && |
6880 | /* everything is filtered out from non-master */ |
6881 | (s_id != mi->master_id || |
6882 | /* for the master meta information is necessary */ |
6883 | ((uchar)buf[EVENT_TYPE_OFFSET] != FORMAT_DESCRIPTION_EVENT && |
6884 | (uchar)buf[EVENT_TYPE_OFFSET] != ROTATE_EVENT))) || |
6885 | |
6886 | /* |
6887 | Check whether it needs to be filtered based on domain_id |
6888 | (DO_DOMAIN_IDS/IGNORE_DOMAIN_IDS). |
6889 | */ |
6890 | (mi->domain_id_filter.is_group_filtered() && |
6891 | Log_event::is_group_event((Log_event_type)(uchar) |
6892 | buf[EVENT_TYPE_OFFSET]))) |
6893 | { |
6894 | /* |
6895 | Do not write it to the relay log. |
6896 | a) We still want to increment mi->master_log_pos, so that we won't |
6897 | re-read this event from the master if the slave IO thread is now |
6898 | stopped/restarted (more efficient if the events we are ignoring are big |
6899 | LOAD DATA INFILE). |
6900 | b) We want to record that we are skipping events, for the information of |
6901 | the slave SQL thread, otherwise that thread may let |
6902 | rli->group_relay_log_pos stay too small if the last binlog's event is |
6903 | ignored. |
6904 | But events which were generated by this slave and which do not exist in |
6905 | the master's binlog (i.e. Format_desc, Rotate & Stop) should not increment |
6906 | mi->master_log_pos. |
6907 | If the event is originated remotely and is being filtered out by |
6908 | IGNORE_SERVER_IDS it increments mi->master_log_pos |
6909 | as well as rli->group_relay_log_pos. |
6910 | */ |
6911 | if (!(s_id == global_system_variables.server_id && |
6912 | !mi->rli.replicate_same_server_id) || |
6913 | ((uchar)buf[EVENT_TYPE_OFFSET] != FORMAT_DESCRIPTION_EVENT && |
6914 | (uchar)buf[EVENT_TYPE_OFFSET] != ROTATE_EVENT && |
6915 | (uchar)buf[EVENT_TYPE_OFFSET] != STOP_EVENT)) |
6916 | { |
6917 | mi->master_log_pos+= inc_pos; |
6918 | memcpy(rli->ign_master_log_name_end, mi->master_log_name, FN_REFLEN); |
6919 | DBUG_ASSERT(rli->ign_master_log_name_end[0]); |
6920 | rli->ign_master_log_pos_end= mi->master_log_pos; |
6921 | if (got_gtid_event) |
6922 | rli->ign_gtids.update(&event_gtid); |
6923 | } |
6924 | // the slave SQL thread needs to re-check |
6925 | rli->relay_log.signal_relay_log_update(); |
6926 | DBUG_PRINT("info" , ("master_log_pos: %lu, event originating from %u server, ignored" , |
6927 | (ulong) mi->master_log_pos, uint4korr(buf + SERVER_ID_OFFSET))); |
6928 | } |
6929 | else |
6930 | { |
6931 | if (likely(!rli->relay_log.write_event_buffer((uchar*)buf, event_len))) |
6932 | { |
6933 | mi->master_log_pos+= inc_pos; |
6934 | DBUG_PRINT("info" , ("master_log_pos: %lu" , (ulong) mi->master_log_pos)); |
6935 | rli->relay_log.harvest_bytes_written(&rli->log_space_total); |
6936 | } |
6937 | else |
6938 | { |
6939 | error= ER_SLAVE_RELAY_LOG_WRITE_FAILURE; |
6940 | } |
6941 | rli->ign_master_log_name_end[0]= 0; // last event is not ignored |
6942 | if (got_gtid_event) |
6943 | rli->ign_gtids.remove_if_present(&event_gtid); |
6944 | if (save_buf != NULL) |
6945 | buf= save_buf; |
6946 | } |
6947 | mysql_mutex_unlock(log_lock); |
6948 | |
6949 | if (likely(!error) && |
6950 | mi->using_gtid != Master_info::USE_GTID_NO && |
6951 | mi->events_queued_since_last_gtid > 0 && |
6952 | ( (mi->last_queued_gtid_standalone && |
6953 | !Log_event::is_part_of_group((Log_event_type)(uchar) |
6954 | buf[EVENT_TYPE_OFFSET])) || |
6955 | (!mi->last_queued_gtid_standalone && |
6956 | ((uchar)buf[EVENT_TYPE_OFFSET] == XID_EVENT || |
6957 | ((uchar)buf[EVENT_TYPE_OFFSET] == QUERY_EVENT && /* QUERY_COMPRESSED_EVENT would never be commmit or rollback */ |
6958 | Query_log_event::peek_is_commit_rollback(buf, event_len, |
6959 | checksum_alg)))))) |
6960 | { |
6961 | /* |
6962 | The whole of the current event group is queued. So in case of |
6963 | reconnect we can start from after the current GTID. |
6964 | */ |
6965 | if (mi->gtid_reconnect_event_skip_count) |
6966 | { |
6967 | bool first= true; |
6968 | StringBuffer<1024> gtid_text; |
6969 | |
6970 | rpl_slave_state_tostring_helper(>id_text, &mi->last_queued_gtid, |
6971 | &first); |
6972 | sql_print_error("Slave IO thread received a terminal event from " |
6973 | "group %s whose retrieval was interrupted " |
6974 | "with reconnect. We still had %llu events to read. " |
6975 | "The number of originally read events was %llu" , |
6976 | gtid_text.ptr(), |
6977 | mi->gtid_reconnect_event_skip_count, |
6978 | mi->events_queued_since_last_gtid); |
6979 | } |
6980 | mi->gtid_current_pos.update(&mi->last_queued_gtid); |
6981 | mi->events_queued_since_last_gtid= 0; |
6982 | |
6983 | /* Reset the domain_id_filter flag. */ |
6984 | mi->domain_id_filter.reset_filter(); |
6985 | } |
6986 | |
6987 | skip_relay_logging: |
6988 | |
6989 | err: |
6990 | if (unlock_data_lock) |
6991 | mysql_mutex_unlock(&mi->data_lock); |
6992 | DBUG_PRINT("info" , ("error: %d" , error)); |
6993 | |
6994 | /* |
6995 | Do not print ER_SLAVE_RELAY_LOG_WRITE_FAILURE error here, as the caller |
6996 | handle_slave_io() prints it on return. |
6997 | */ |
6998 | if (unlikely(error) && error != ER_SLAVE_RELAY_LOG_WRITE_FAILURE) |
6999 | mi->report(ERROR_LEVEL, error, NULL, ER_DEFAULT(error), |
7000 | error_msg.ptr()); |
7001 | |
7002 | if (unlikely(is_malloc)) |
7003 | my_free((void *)new_buf); |
7004 | |
7005 | DBUG_RETURN(error); |
7006 | } |
7007 | |
7008 | |
7009 | void end_relay_log_info(Relay_log_info* rli) |
7010 | { |
7011 | mysql_mutex_t *log_lock; |
7012 | DBUG_ENTER("end_relay_log_info" ); |
7013 | |
7014 | rli->error_on_rli_init_info= false; |
7015 | if (!rli->inited) |
7016 | DBUG_VOID_RETURN; |
7017 | if (rli->info_fd >= 0) |
7018 | { |
7019 | end_io_cache(&rli->info_file); |
7020 | mysql_file_close(rli->info_fd, MYF(MY_WME)); |
7021 | rli->info_fd = -1; |
7022 | } |
7023 | if (rli->cur_log_fd >= 0) |
7024 | { |
7025 | end_io_cache(&rli->cache_buf); |
7026 | mysql_file_close(rli->cur_log_fd, MYF(MY_WME)); |
7027 | rli->cur_log_fd = -1; |
7028 | } |
7029 | rli->inited = 0; |
7030 | log_lock= rli->relay_log.get_log_lock(); |
7031 | mysql_mutex_lock(log_lock); |
7032 | rli->relay_log.close(LOG_CLOSE_INDEX | LOG_CLOSE_STOP_EVENT); |
7033 | rli->relay_log.harvest_bytes_written(&rli->log_space_total); |
7034 | mysql_mutex_unlock(log_lock); |
7035 | /* |
7036 | Delete the slave's temporary tables from memory. |
7037 | In the future there will be other actions than this, to ensure persistance |
7038 | of slave's temp tables after shutdown. |
7039 | */ |
7040 | rli->close_temporary_tables(); |
7041 | DBUG_VOID_RETURN; |
7042 | } |
7043 | |
7044 | |
7045 | /** |
7046 | Hook to detach the active VIO before closing a connection handle. |
7047 | |
7048 | The client API might close the connection (and associated data) |
7049 | in case it encounters a unrecoverable (network) error. This hook |
7050 | is called from the client code before the VIO handle is deleted |
7051 | allows the thread to detach the active vio so it does not point |
7052 | to freed memory. |
7053 | |
7054 | Other calls to THD::clear_active_vio throughout this module are |
7055 | redundant due to the hook but are left in place for illustrative |
7056 | purposes. |
7057 | */ |
7058 | |
7059 | extern "C" void slave_io_thread_detach_vio() |
7060 | { |
7061 | #ifdef SIGNAL_WITH_VIO_CLOSE |
7062 | THD *thd= current_thd; |
7063 | if (thd && thd->slave_thread) |
7064 | thd->clear_active_vio(); |
7065 | #endif |
7066 | } |
7067 | |
7068 | |
7069 | /* |
7070 | Try to connect until successful or slave killed |
7071 | |
7072 | SYNPOSIS |
7073 | safe_connect() |
7074 | thd Thread handler for slave |
7075 | mysql MySQL connection handle |
7076 | mi Replication handle |
7077 | |
7078 | RETURN |
7079 | 0 ok |
7080 | # Error |
7081 | */ |
7082 | |
7083 | static int safe_connect(THD* thd, MYSQL* mysql, Master_info* mi) |
7084 | { |
7085 | DBUG_ENTER("safe_connect" ); |
7086 | |
7087 | DBUG_RETURN(connect_to_master(thd, mysql, mi, 0, 0)); |
7088 | } |
7089 | |
7090 | |
7091 | /* |
7092 | SYNPOSIS |
7093 | connect_to_master() |
7094 | |
7095 | IMPLEMENTATION |
7096 | Try to connect until successful or slave killed or we have retried |
7097 | master_retry_count times |
7098 | */ |
7099 | |
7100 | static int connect_to_master(THD* thd, MYSQL* mysql, Master_info* mi, |
7101 | bool reconnect, bool suppress_warnings) |
7102 | { |
7103 | int slave_was_killed; |
7104 | int last_errno= -2; // impossible error |
7105 | ulong err_count=0; |
7106 | my_bool my_true= 1; |
7107 | DBUG_ENTER("connect_to_master" ); |
7108 | set_slave_max_allowed_packet(thd, mysql); |
7109 | #ifndef DBUG_OFF |
7110 | mi->events_till_disconnect = disconnect_slave_event_count; |
7111 | #endif |
7112 | ulong client_flag= CLIENT_REMEMBER_OPTIONS; |
7113 | if (opt_slave_compressed_protocol) |
7114 | client_flag|= CLIENT_COMPRESS; /* We will use compression */ |
7115 | |
7116 | mysql_options(mysql, MYSQL_OPT_CONNECT_TIMEOUT, (char *) &slave_net_timeout); |
7117 | mysql_options(mysql, MYSQL_OPT_READ_TIMEOUT, (char *) &slave_net_timeout); |
7118 | mysql_options(mysql, MYSQL_OPT_USE_THREAD_SPECIFIC_MEMORY, |
7119 | (char*) &my_true); |
7120 | |
7121 | #ifdef HAVE_OPENSSL |
7122 | if (mi->ssl) |
7123 | { |
7124 | mysql_ssl_set(mysql, |
7125 | mi->ssl_key[0]?mi->ssl_key:0, |
7126 | mi->ssl_cert[0]?mi->ssl_cert:0, |
7127 | mi->ssl_ca[0]?mi->ssl_ca:0, |
7128 | mi->ssl_capath[0]?mi->ssl_capath:0, |
7129 | mi->ssl_cipher[0]?mi->ssl_cipher:0); |
7130 | mysql_options(mysql, MYSQL_OPT_SSL_VERIFY_SERVER_CERT, |
7131 | &mi->ssl_verify_server_cert); |
7132 | mysql_options(mysql, MYSQL_OPT_SSL_CRLPATH, |
7133 | mi->ssl_crlpath[0] ? mi->ssl_crlpath : 0); |
7134 | mysql_options(mysql, MYSQL_OPT_SSL_VERIFY_SERVER_CERT, |
7135 | &mi->ssl_verify_server_cert); |
7136 | } |
7137 | #endif |
7138 | |
7139 | /* |
7140 | If server's default charset is not supported (like utf16, utf32) as client |
7141 | charset, then set client charset to 'latin1' (default client charset). |
7142 | */ |
7143 | if (is_supported_parser_charset(default_charset_info)) |
7144 | mysql_options(mysql, MYSQL_SET_CHARSET_NAME, default_charset_info->csname); |
7145 | else |
7146 | { |
7147 | sql_print_information("'%s' can not be used as client character set. " |
7148 | "'%s' will be used as default client character set " |
7149 | "while connecting to master." , |
7150 | default_charset_info->csname, |
7151 | default_client_charset_info->csname); |
7152 | mysql_options(mysql, MYSQL_SET_CHARSET_NAME, |
7153 | default_client_charset_info->csname); |
7154 | } |
7155 | |
7156 | /* This one is not strictly needed but we have it here for completeness */ |
7157 | mysql_options(mysql, MYSQL_SET_CHARSET_DIR, (char *) charsets_dir); |
7158 | |
7159 | /* Set MYSQL_PLUGIN_DIR in case master asks for an external authentication plugin */ |
7160 | if (opt_plugin_dir_ptr && *opt_plugin_dir_ptr) |
7161 | mysql_options(mysql, MYSQL_PLUGIN_DIR, opt_plugin_dir_ptr); |
7162 | |
7163 | /* we disallow empty users */ |
7164 | if (mi->user[0] == 0) |
7165 | { |
7166 | mi->report(ERROR_LEVEL, ER_SLAVE_FATAL_ERROR, NULL, |
7167 | ER_THD(thd, ER_SLAVE_FATAL_ERROR), |
7168 | "Invalid (empty) username when attempting to " |
7169 | "connect to the master server. Connection attempt " |
7170 | "terminated." ); |
7171 | DBUG_RETURN(1); |
7172 | } |
7173 | while (!(slave_was_killed = io_slave_killed(mi)) && |
7174 | (reconnect ? mysql_reconnect(mysql) != 0 : |
7175 | mysql_real_connect(mysql, mi->host, mi->user, mi->password, 0, |
7176 | mi->port, 0, client_flag) == 0)) |
7177 | { |
7178 | /* Don't repeat last error */ |
7179 | if ((int)mysql_errno(mysql) != last_errno) |
7180 | { |
7181 | last_errno=mysql_errno(mysql); |
7182 | suppress_warnings= 0; |
7183 | mi->report(ERROR_LEVEL, last_errno, NULL, |
7184 | "error %s to master '%s@%s:%d'" |
7185 | " - retry-time: %d maximum-retries: %lu message: %s" , |
7186 | (reconnect ? "reconnecting" : "connecting" ), |
7187 | mi->user, mi->host, mi->port, |
7188 | mi->connect_retry, master_retry_count, |
7189 | mysql_error(mysql)); |
7190 | } |
7191 | /* |
7192 | By default we try forever. The reason is that failure will trigger |
7193 | master election, so if the user did not set master_retry_count we |
7194 | do not want to have election triggered on the first failure to |
7195 | connect |
7196 | */ |
7197 | if (++err_count == master_retry_count) |
7198 | { |
7199 | slave_was_killed=1; |
7200 | if (reconnect) |
7201 | change_rpl_status(RPL_ACTIVE_SLAVE,RPL_LOST_SOLDIER); |
7202 | break; |
7203 | } |
7204 | slave_sleep(thd,mi->connect_retry,io_slave_killed, mi); |
7205 | } |
7206 | |
7207 | if (!slave_was_killed) |
7208 | { |
7209 | mi->clear_error(); // clear possible left over reconnect error |
7210 | if (reconnect) |
7211 | { |
7212 | if (!suppress_warnings && global_system_variables.log_warnings) |
7213 | sql_print_information("Slave: connected to master '%s@%s:%d'," |
7214 | "replication resumed in log '%s' at " |
7215 | "position %llu" , mi->user, mi->host, mi->port, |
7216 | IO_RPL_LOG_NAME, mi->master_log_pos); |
7217 | } |
7218 | else |
7219 | { |
7220 | change_rpl_status(RPL_IDLE_SLAVE,RPL_ACTIVE_SLAVE); |
7221 | general_log_print(thd, COM_CONNECT_OUT, "%s@%s:%d" , |
7222 | mi->user, mi->host, mi->port); |
7223 | } |
7224 | #ifdef SIGNAL_WITH_VIO_CLOSE |
7225 | thd->set_active_vio(mysql->net.vio); |
7226 | #endif |
7227 | } |
7228 | mysql->reconnect= 1; |
7229 | DBUG_PRINT("exit" ,("slave_was_killed: %d" , slave_was_killed)); |
7230 | DBUG_RETURN(slave_was_killed); |
7231 | } |
7232 | |
7233 | |
7234 | /* |
7235 | safe_reconnect() |
7236 | |
7237 | IMPLEMENTATION |
7238 | Try to connect until successful or slave killed or we have retried |
7239 | master_retry_count times |
7240 | */ |
7241 | |
7242 | static int safe_reconnect(THD* thd, MYSQL* mysql, Master_info* mi, |
7243 | bool suppress_warnings) |
7244 | { |
7245 | DBUG_ENTER("safe_reconnect" ); |
7246 | DBUG_RETURN(connect_to_master(thd, mysql, mi, 1, suppress_warnings)); |
7247 | } |
7248 | |
7249 | |
7250 | #ifdef NOT_USED |
7251 | MYSQL *rpl_connect_master(MYSQL *mysql) |
7252 | { |
7253 | Master_info *mi= my_pthread_getspecific_ptr(Master_info*, RPL_MASTER_INFO); |
7254 | bool allocated= false; |
7255 | my_bool my_true= 1; |
7256 | THD *thd; |
7257 | |
7258 | if (!mi) |
7259 | { |
7260 | sql_print_error("'rpl_connect_master' must be called in slave I/O thread context." ); |
7261 | return NULL; |
7262 | } |
7263 | thd= mi->io_thd; |
7264 | if (!mysql) |
7265 | { |
7266 | if(!(mysql= mysql_init(NULL))) |
7267 | { |
7268 | sql_print_error("rpl_connect_master: failed in mysql_init()" ); |
7269 | return NULL; |
7270 | } |
7271 | allocated= true; |
7272 | } |
7273 | |
7274 | /* |
7275 | XXX: copied from connect_to_master, this function should not |
7276 | change the slave status, so we cannot use connect_to_master |
7277 | directly |
7278 | |
7279 | TODO: make this part a seperate function to eliminate duplication |
7280 | */ |
7281 | mysql_options(mysql, MYSQL_OPT_CONNECT_TIMEOUT, (char *) &slave_net_timeout); |
7282 | mysql_options(mysql, MYSQL_OPT_READ_TIMEOUT, (char *) &slave_net_timeout); |
7283 | mysql_options(mysql, MYSQL_OPT_USE_THREAD_SPECIFIC_MEMORY, |
7284 | (char*) &my_true); |
7285 | |
7286 | #ifdef HAVE_OPENSSL |
7287 | if (mi->ssl) |
7288 | { |
7289 | mysql_ssl_set(mysql, |
7290 | mi->ssl_key[0]?mi->ssl_key:0, |
7291 | mi->ssl_cert[0]?mi->ssl_cert:0, |
7292 | mi->ssl_ca[0]?mi->ssl_ca:0, |
7293 | mi->ssl_capath[0]?mi->ssl_capath:0, |
7294 | mi->ssl_cipher[0]?mi->ssl_cipher:0); |
7295 | mysql_options(mysql, MYSQL_OPT_SSL_VERIFY_SERVER_CERT, |
7296 | &mi->ssl_verify_server_cert); |
7297 | } |
7298 | #endif |
7299 | |
7300 | mysql_options(mysql, MYSQL_SET_CHARSET_NAME, default_charset_info->csname); |
7301 | /* This one is not strictly needed but we have it here for completeness */ |
7302 | mysql_options(mysql, MYSQL_SET_CHARSET_DIR, (char *) charsets_dir); |
7303 | |
7304 | if (mi->user == NULL |
7305 | || mi->user[0] == 0 |
7306 | || io_slave_killed( mi) |
7307 | || !mysql_real_connect(mysql, mi->host, mi->user, mi->password, 0, |
7308 | mi->port, 0, 0)) |
7309 | { |
7310 | if (!io_slave_killed( mi)) |
7311 | sql_print_error("rpl_connect_master: error connecting to master: %s (server_error: %d)" , |
7312 | mysql_error(mysql), mysql_errno(mysql)); |
7313 | |
7314 | if (allocated) |
7315 | mysql_close(mysql); // this will free the object |
7316 | return NULL; |
7317 | } |
7318 | return mysql; |
7319 | } |
7320 | #endif |
7321 | |
7322 | |
7323 | /* |
7324 | Called when we notice that the current "hot" log got rotated under our feet. |
7325 | */ |
7326 | |
7327 | static IO_CACHE *reopen_relay_log(Relay_log_info *rli, const char **errmsg) |
7328 | { |
7329 | DBUG_ENTER("reopen_relay_log" ); |
7330 | DBUG_ASSERT(rli->cur_log != &rli->cache_buf); |
7331 | DBUG_ASSERT(rli->cur_log_fd == -1); |
7332 | |
7333 | IO_CACHE *cur_log = rli->cur_log=&rli->cache_buf; |
7334 | if ((rli->cur_log_fd=open_binlog(cur_log,rli->event_relay_log_name, |
7335 | errmsg)) <0) |
7336 | DBUG_RETURN(0); |
7337 | /* |
7338 | We want to start exactly where we was before: |
7339 | relay_log_pos Current log pos |
7340 | pending Number of bytes already processed from the event |
7341 | */ |
7342 | rli->event_relay_log_pos= MY_MAX(rli->event_relay_log_pos, BIN_LOG_HEADER_SIZE); |
7343 | my_b_seek(cur_log,rli->event_relay_log_pos); |
7344 | DBUG_RETURN(cur_log); |
7345 | } |
7346 | |
7347 | |
7348 | /** |
7349 | Reads next event from the relay log. Should be called from the |
7350 | slave IO thread. |
7351 | |
7352 | @param rli Relay_log_info structure for the slave IO thread. |
7353 | |
7354 | @return The event read, or NULL on error. If an error occurs, the |
7355 | error is reported through the sql_print_information() or |
7356 | sql_print_error() functions. |
7357 | |
7358 | The size of the read event (in bytes) is returned in *event_size. |
7359 | */ |
7360 | static Log_event* next_event(rpl_group_info *rgi, ulonglong *event_size) |
7361 | { |
7362 | Log_event* ev; |
7363 | Relay_log_info *rli= rgi->rli; |
7364 | IO_CACHE* cur_log = rli->cur_log; |
7365 | mysql_mutex_t *log_lock = rli->relay_log.get_log_lock(); |
7366 | const char* errmsg=0; |
7367 | DBUG_ENTER("next_event" ); |
7368 | |
7369 | DBUG_ASSERT(rgi->thd != 0 && rgi->thd == rli->sql_driver_thd); |
7370 | *event_size= 0; |
7371 | |
7372 | #ifndef DBUG_OFF |
7373 | if (abort_slave_event_count && !rli->events_till_abort--) |
7374 | DBUG_RETURN(0); |
7375 | #endif |
7376 | |
7377 | /* |
7378 | For most operations we need to protect rli members with data_lock, |
7379 | so we assume calling function acquired this mutex for us and we will |
7380 | hold it for the most of the loop below However, we will release it |
7381 | whenever it is worth the hassle, and in the cases when we go into a |
7382 | mysql_cond_wait() with the non-data_lock mutex |
7383 | */ |
7384 | mysql_mutex_assert_owner(&rli->data_lock); |
7385 | |
7386 | while (!sql_slave_killed(rgi)) |
7387 | { |
7388 | /* |
7389 | We can have two kinds of log reading: |
7390 | hot_log: |
7391 | rli->cur_log points at the IO_CACHE of relay_log, which |
7392 | is actively being updated by the I/O thread. We need to be careful |
7393 | in this case and make sure that we are not looking at a stale log that |
7394 | has already been rotated. If it has been, we reopen the log. |
7395 | |
7396 | The other case is much simpler: |
7397 | We just have a read only log that nobody else will be updating. |
7398 | */ |
7399 | ulonglong old_pos; |
7400 | bool hot_log; |
7401 | if ((hot_log = (cur_log != &rli->cache_buf))) |
7402 | { |
7403 | DBUG_ASSERT(rli->cur_log_fd == -1); // foreign descriptor |
7404 | mysql_mutex_lock(log_lock); |
7405 | |
7406 | /* |
7407 | Reading xxx_file_id is safe because the log will only |
7408 | be rotated when we hold relay_log.LOCK_log |
7409 | */ |
7410 | if (rli->relay_log.get_open_count() != rli->cur_log_old_open_count) |
7411 | { |
7412 | // The master has switched to a new log file; Reopen the old log file |
7413 | cur_log=reopen_relay_log(rli, &errmsg); |
7414 | mysql_mutex_unlock(log_lock); |
7415 | if (!cur_log) // No more log files |
7416 | goto err; |
7417 | hot_log=0; // Using old binary log |
7418 | } |
7419 | } |
7420 | /* |
7421 | As there is no guarantee that the relay is open (for example, an I/O |
7422 | error during a write by the slave I/O thread may have closed it), we |
7423 | have to test it. |
7424 | */ |
7425 | if (!my_b_inited(cur_log)) |
7426 | goto err; |
7427 | #ifndef DBUG_OFF |
7428 | { |
7429 | /* This is an assertion which sometimes fails, let's try to track it */ |
7430 | DBUG_PRINT("info" , ("my_b_tell(cur_log)=%llu rli->event_relay_log_pos=%llu" , |
7431 | my_b_tell(cur_log), rli->event_relay_log_pos)); |
7432 | DBUG_ASSERT(my_b_tell(cur_log) >= BIN_LOG_HEADER_SIZE); |
7433 | DBUG_ASSERT(rli->mi->using_parallel() || |
7434 | my_b_tell(cur_log) == rli->event_relay_log_pos); |
7435 | } |
7436 | #endif |
7437 | /* |
7438 | Relay log is always in new format - if the master is 3.23, the |
7439 | I/O thread will convert the format for us. |
7440 | A problem: the description event may be in a previous relay log. So if |
7441 | the slave has been shutdown meanwhile, we would have to look in old relay |
7442 | logs, which may even have been deleted. So we need to write this |
7443 | description event at the beginning of the relay log. |
7444 | When the relay log is created when the I/O thread starts, easy: the |
7445 | master will send the description event and we will queue it. |
7446 | But if the relay log is created by new_file(): then the solution is: |
7447 | MYSQL_BIN_LOG::open() will write the buffered description event. |
7448 | */ |
7449 | old_pos= rli->event_relay_log_pos; |
7450 | if ((ev= Log_event::read_log_event(cur_log, |
7451 | rli->relay_log.description_event_for_exec, |
7452 | opt_slave_sql_verify_checksum))) |
7453 | |
7454 | { |
7455 | /* |
7456 | read it while we have a lock, to avoid a mutex lock in |
7457 | inc_event_relay_log_pos() |
7458 | */ |
7459 | rli->future_event_relay_log_pos= my_b_tell(cur_log); |
7460 | *event_size= rli->future_event_relay_log_pos - old_pos; |
7461 | |
7462 | if (hot_log) |
7463 | mysql_mutex_unlock(log_lock); |
7464 | rli->sql_thread_caught_up= false; |
7465 | DBUG_RETURN(ev); |
7466 | } |
7467 | if (opt_reckless_slave) // For mysql-test |
7468 | cur_log->error = 0; |
7469 | if (unlikely(cur_log->error < 0)) |
7470 | { |
7471 | errmsg = "slave SQL thread aborted because of I/O error" ; |
7472 | if (hot_log) |
7473 | mysql_mutex_unlock(log_lock); |
7474 | goto err; |
7475 | } |
7476 | if (!cur_log->error) /* EOF */ |
7477 | { |
7478 | /* |
7479 | On a hot log, EOF means that there are no more updates to |
7480 | process and we must block until I/O thread adds some and |
7481 | signals us to continue |
7482 | */ |
7483 | if (hot_log) |
7484 | { |
7485 | /* |
7486 | We say in Seconds_Behind_Master that we have "caught up". Note that |
7487 | for example if network link is broken but I/O slave thread hasn't |
7488 | noticed it (slave_net_timeout not elapsed), then we'll say "caught |
7489 | up" whereas we're not really caught up. Fixing that would require |
7490 | internally cutting timeout in smaller pieces in network read, no |
7491 | thanks. Another example: SQL has caught up on I/O, now I/O has read |
7492 | a new event and is queuing it; the false "0" will exist until SQL |
7493 | finishes executing the new event; it will be look abnormal only if |
7494 | the events have old timestamps (then you get "many", 0, "many"). |
7495 | |
7496 | Transient phases like this can be fixed with implemeting |
7497 | Heartbeat event which provides the slave the status of the |
7498 | master at time the master does not have any new update to send. |
7499 | Seconds_Behind_Master would be zero only when master has no |
7500 | more updates in binlog for slave. The heartbeat can be sent |
7501 | in a (small) fraction of slave_net_timeout. Until it's done |
7502 | rli->sql_thread_caught_up is temporarely (for time of waiting for |
7503 | the following event) set whenever EOF is reached. |
7504 | */ |
7505 | rli->sql_thread_caught_up= true; |
7506 | |
7507 | DBUG_ASSERT(rli->relay_log.get_open_count() == |
7508 | rli->cur_log_old_open_count); |
7509 | |
7510 | if (rli->ign_master_log_name_end[0]) |
7511 | { |
7512 | /* We generate and return a Rotate, to make our positions advance */ |
7513 | DBUG_PRINT("info" ,("seeing an ignored end segment" )); |
7514 | ev= new Rotate_log_event(rli->ign_master_log_name_end, |
7515 | 0, rli->ign_master_log_pos_end, |
7516 | Rotate_log_event::DUP_NAME); |
7517 | rli->ign_master_log_name_end[0]= 0; |
7518 | mysql_mutex_unlock(log_lock); |
7519 | if (unlikely(!ev)) |
7520 | { |
7521 | errmsg= "Slave SQL thread failed to create a Rotate event " |
7522 | "(out of memory?), SHOW SLAVE STATUS may be inaccurate" ; |
7523 | goto err; |
7524 | } |
7525 | ev->server_id= 0; // don't be ignored by slave SQL thread |
7526 | DBUG_RETURN(ev); |
7527 | } |
7528 | |
7529 | if (rli->ign_gtids.count() && !rli->is_in_group()) |
7530 | { |
7531 | /* |
7532 | We generate and return a Gtid_list, to update gtid_slave_pos, |
7533 | unless being in the middle of a group. |
7534 | */ |
7535 | DBUG_PRINT("info" ,("seeing ignored end gtids" )); |
7536 | ev= new Gtid_list_log_event(&rli->ign_gtids, |
7537 | Gtid_list_log_event::FLAG_IGN_GTIDS); |
7538 | rli->ign_gtids.reset(); |
7539 | mysql_mutex_unlock(log_lock); |
7540 | if (unlikely(!ev)) |
7541 | { |
7542 | errmsg= "Slave SQL thread failed to create a Gtid_list event " |
7543 | "(out of memory?), gtid_slave_pos may be inaccurate" ; |
7544 | goto err; |
7545 | } |
7546 | ev->server_id= 0; // don't be ignored by slave SQL thread |
7547 | ev->set_artificial_event(); // Don't mess up Exec_Master_Log_Pos |
7548 | DBUG_RETURN(ev); |
7549 | } |
7550 | |
7551 | /* |
7552 | We have to check sql_slave_killed() here an extra time. |
7553 | Otherwise we may miss a wakeup, since last check was done |
7554 | without holding LOCK_log. |
7555 | */ |
7556 | if (sql_slave_killed(rgi)) |
7557 | { |
7558 | mysql_mutex_unlock(log_lock); |
7559 | break; |
7560 | } |
7561 | |
7562 | /* |
7563 | We can, and should release data_lock while we are waiting for |
7564 | update. If we do not, show slave status will block |
7565 | */ |
7566 | mysql_mutex_unlock(&rli->data_lock); |
7567 | |
7568 | /* |
7569 | Possible deadlock : |
7570 | - the I/O thread has reached log_space_limit |
7571 | - the SQL thread has read all relay logs, but cannot purge for some |
7572 | reason: |
7573 | * it has already purged all logs except the current one |
7574 | * there are other logs than the current one but they're involved in |
7575 | a transaction that finishes in the current one (or is not finished) |
7576 | Solution : |
7577 | Wake up the possibly waiting I/O thread, and set a boolean asking |
7578 | the I/O thread to temporarily ignore the log_space_limit |
7579 | constraint, because we do not want the I/O thread to block because of |
7580 | space (it's ok if it blocks for any other reason (e.g. because the |
7581 | master does not send anything). Then the I/O thread stops waiting |
7582 | and reads one more event and starts honoring log_space_limit again. |
7583 | |
7584 | If the SQL thread needs more events to be able to rotate the log (it |
7585 | might need to finish the current group first), then it can ask for |
7586 | one more at a time. Thus we don't outgrow the relay log indefinitely, |
7587 | but rather in a controlled manner, until the next rotate. |
7588 | |
7589 | When the SQL thread starts it sets ignore_log_space_limit to false. |
7590 | We should also reset ignore_log_space_limit to 0 when the user does |
7591 | RESET SLAVE, but in fact, no need as RESET SLAVE requires that the |
7592 | slave be stopped, and the SQL thread sets ignore_log_space_limit |
7593 | to 0 when |
7594 | it stops. |
7595 | */ |
7596 | mysql_mutex_lock(&rli->log_space_lock); |
7597 | |
7598 | /* |
7599 | If we have reached the limit of the relay space and we |
7600 | are going to sleep, waiting for more events: |
7601 | |
7602 | 1. If outside a group, SQL thread asks the IO thread |
7603 | to force a rotation so that the SQL thread purges |
7604 | logs next time it processes an event (thus space is |
7605 | freed). |
7606 | |
7607 | 2. If in a group, SQL thread asks the IO thread to |
7608 | ignore the limit and queues yet one more event |
7609 | so that the SQL thread finishes the group and |
7610 | is are able to rotate and purge sometime soon. |
7611 | */ |
7612 | if (rli->log_space_limit && |
7613 | rli->log_space_limit < rli->log_space_total) |
7614 | { |
7615 | /* force rotation if not in an unfinished group */ |
7616 | rli->sql_force_rotate_relay= !rli->is_in_group(); |
7617 | |
7618 | /* ask for one more event */ |
7619 | rli->ignore_log_space_limit= true; |
7620 | } |
7621 | |
7622 | mysql_cond_broadcast(&rli->log_space_cond); |
7623 | mysql_mutex_unlock(&rli->log_space_lock); |
7624 | // Note that wait_for_update_relay_log unlocks lock_log ! |
7625 | rli->relay_log.wait_for_update_relay_log(rli->sql_driver_thd); |
7626 | // re-acquire data lock since we released it earlier |
7627 | mysql_mutex_lock(&rli->data_lock); |
7628 | rli->sql_thread_caught_up= false; |
7629 | continue; |
7630 | } |
7631 | /* |
7632 | If the log was not hot, we need to move to the next log in |
7633 | sequence. The next log could be hot or cold, we deal with both |
7634 | cases separately after doing some common initialization |
7635 | */ |
7636 | end_io_cache(cur_log); |
7637 | DBUG_ASSERT(rli->cur_log_fd >= 0); |
7638 | mysql_file_close(rli->cur_log_fd, MYF(MY_WME)); |
7639 | rli->cur_log_fd = -1; |
7640 | rli->last_inuse_relaylog->completed= true; |
7641 | rli->relay_log.description_event_for_exec->reset_crypto(); |
7642 | |
7643 | if (relay_log_purge) |
7644 | { |
7645 | /* |
7646 | purge_first_log will properly set up relay log coordinates in rli. |
7647 | If the group's coordinates are equal to the event's coordinates |
7648 | (i.e. the relay log was not rotated in the middle of a group), |
7649 | we can purge this relay log too. |
7650 | We do ulonglong and string comparisons, this may be slow but |
7651 | - purging the last relay log is nice (it can save 1GB of disk), so we |
7652 | like to detect the case where we can do it, and given this, |
7653 | - I see no better detection method |
7654 | - purge_first_log is not called that often |
7655 | */ |
7656 | if (rli->relay_log.purge_first_log |
7657 | (rli, |
7658 | rli->group_relay_log_pos == rli->event_relay_log_pos |
7659 | && !strcmp(rli->group_relay_log_name,rli->event_relay_log_name))) |
7660 | { |
7661 | errmsg = "Error purging processed logs" ; |
7662 | goto err; |
7663 | } |
7664 | } |
7665 | else |
7666 | { |
7667 | /* |
7668 | If hot_log is set, then we already have a lock on |
7669 | LOCK_log. If not, we have to get the lock. |
7670 | |
7671 | According to Sasha, the only time this code will ever be executed |
7672 | is if we are recovering from a bug. |
7673 | */ |
7674 | if (rli->relay_log.find_next_log(&rli->linfo, !hot_log)) |
7675 | { |
7676 | errmsg = "error switching to the next log" ; |
7677 | goto err; |
7678 | } |
7679 | rli->event_relay_log_pos = BIN_LOG_HEADER_SIZE; |
7680 | strmake_buf(rli->event_relay_log_name,rli->linfo.log_file_name); |
7681 | if (rli->flush()) |
7682 | { |
7683 | errmsg= "error flushing relay log" ; |
7684 | goto err; |
7685 | } |
7686 | } |
7687 | /* |
7688 | Now we want to open this next log. To know if it's a hot log (the one |
7689 | being written by the I/O thread now) or a cold log, we can use |
7690 | is_active(); if it is hot, we use the I/O cache; if it's cold we open |
7691 | the file normally. But if is_active() reports that the log is hot, this |
7692 | may change between the test and the consequence of the test. So we may |
7693 | open the I/O cache whereas the log is now cold, which is nonsense. |
7694 | To guard against this, we need to have LOCK_log. |
7695 | */ |
7696 | |
7697 | DBUG_PRINT("info" ,("hot_log: %d" ,hot_log)); |
7698 | if (!hot_log) /* if hot_log, we already have this mutex */ |
7699 | mysql_mutex_lock(log_lock); |
7700 | if (rli->relay_log.is_active(rli->linfo.log_file_name)) |
7701 | { |
7702 | rli->cur_log= cur_log= rli->relay_log.get_log_file(); |
7703 | rli->cur_log_old_open_count= rli->relay_log.get_open_count(); |
7704 | DBUG_ASSERT(rli->cur_log_fd == -1); |
7705 | |
7706 | /* |
7707 | When the SQL thread is [stopped and] (re)started the |
7708 | following may happen: |
7709 | |
7710 | 1. Log was hot at stop time and remains hot at restart |
7711 | |
7712 | SQL thread reads again from hot_log (SQL thread was |
7713 | reading from the active log when it was stopped and the |
7714 | very same log is still active on SQL thread restart). |
7715 | |
7716 | In this case, my_b_seek is performed on cur_log, while |
7717 | cur_log points to relay_log.get_log_file(); |
7718 | |
7719 | 2. Log was hot at stop time but got cold before restart |
7720 | |
7721 | The log was hot when SQL thread stopped, but it is not |
7722 | anymore when the SQL thread restarts. |
7723 | |
7724 | In this case, the SQL thread reopens the log, using |
7725 | cache_buf, ie, cur_log points to &cache_buf, and thence |
7726 | its coordinates are reset. |
7727 | |
7728 | 3. Log was already cold at stop time |
7729 | |
7730 | The log was not hot when the SQL thread stopped, and, of |
7731 | course, it will not be hot when it restarts. |
7732 | |
7733 | In this case, the SQL thread opens the cold log again, |
7734 | using cache_buf, ie, cur_log points to &cache_buf, and |
7735 | thence its coordinates are reset. |
7736 | |
7737 | 4. Log was hot at stop time, DBA changes to previous cold |
7738 | log and restarts SQL thread |
7739 | |
7740 | The log was hot when the SQL thread was stopped, but the |
7741 | user changed the coordinates of the SQL thread to |
7742 | restart from a previous cold log. |
7743 | |
7744 | In this case, at start time, cur_log points to a cold |
7745 | log, opened using &cache_buf as cache, and coordinates |
7746 | are reset. However, as it moves on to the next logs, it |
7747 | will eventually reach the hot log. If the hot log is the |
7748 | same at the time the SQL thread was stopped, then |
7749 | coordinates were not reset - the cur_log will point to |
7750 | relay_log.get_log_file(), and not a freshly opened |
7751 | IO_CACHE through cache_buf. For this reason we need to |
7752 | deploy a my_b_seek before calling check_binlog_magic at |
7753 | this point of the code (see: BUG#55263 for more |
7754 | details). |
7755 | |
7756 | NOTES: |
7757 | - We must keep the LOCK_log to read the 4 first bytes, as |
7758 | this is a hot log (same as when we call read_log_event() |
7759 | above: for a hot log we take the mutex). |
7760 | |
7761 | - Because of scenario #4 above, we need to have a |
7762 | my_b_seek here. Otherwise, we might hit the assertion |
7763 | inside check_binlog_magic. |
7764 | */ |
7765 | |
7766 | my_b_seek(cur_log, (my_off_t) 0); |
7767 | if (check_binlog_magic(cur_log,&errmsg)) |
7768 | { |
7769 | if (!hot_log) |
7770 | mysql_mutex_unlock(log_lock); |
7771 | goto err; |
7772 | } |
7773 | if (rli->alloc_inuse_relaylog(rli->linfo.log_file_name)) |
7774 | { |
7775 | if (!hot_log) |
7776 | mysql_mutex_unlock(log_lock); |
7777 | goto err; |
7778 | } |
7779 | if (!hot_log) |
7780 | mysql_mutex_unlock(log_lock); |
7781 | continue; |
7782 | } |
7783 | if (!hot_log) |
7784 | mysql_mutex_unlock(log_lock); |
7785 | /* |
7786 | if we get here, the log was not hot, so we will have to open it |
7787 | ourselves. We are sure that the log is still not hot now (a log can get |
7788 | from hot to cold, but not from cold to hot). No need for LOCK_log. |
7789 | */ |
7790 | // open_binlog() will check the magic header |
7791 | if ((rli->cur_log_fd=open_binlog(cur_log,rli->linfo.log_file_name, |
7792 | &errmsg)) <0) |
7793 | goto err; |
7794 | if (rli->alloc_inuse_relaylog(rli->linfo.log_file_name)) |
7795 | goto err; |
7796 | } |
7797 | else |
7798 | { |
7799 | /* |
7800 | Read failed with a non-EOF error. |
7801 | TODO: come up with something better to handle this error |
7802 | */ |
7803 | if (hot_log) |
7804 | mysql_mutex_unlock(log_lock); |
7805 | sql_print_error("Slave SQL thread: I/O error reading \ |
7806 | event(errno: %d cur_log->error: %d)" , |
7807 | my_errno,cur_log->error); |
7808 | // set read position to the beginning of the event |
7809 | my_b_seek(cur_log,rli->event_relay_log_pos); |
7810 | /* otherwise, we have had a partial read */ |
7811 | errmsg = "Aborting slave SQL thread because of partial event read" ; |
7812 | break; // To end of function |
7813 | } |
7814 | } |
7815 | if (!errmsg && global_system_variables.log_warnings) |
7816 | { |
7817 | sql_print_information("Error reading relay log event: %s" , |
7818 | "slave SQL thread was killed" ); |
7819 | DBUG_RETURN(0); |
7820 | } |
7821 | |
7822 | err: |
7823 | if (errmsg) |
7824 | sql_print_error("Error reading relay log event: %s" , errmsg); |
7825 | DBUG_RETURN(0); |
7826 | } |
7827 | |
7828 | /* |
7829 | Rotate a relay log (this is used only by FLUSH LOGS; the automatic rotation |
7830 | because of size is simpler because when we do it we already have all relevant |
7831 | locks; here we don't, so this function is mainly taking locks). |
7832 | Returns nothing as we cannot catch any error (MYSQL_BIN_LOG::new_file() |
7833 | is void). |
7834 | */ |
7835 | |
7836 | int rotate_relay_log(Master_info* mi) |
7837 | { |
7838 | DBUG_ENTER("rotate_relay_log" ); |
7839 | Relay_log_info* rli= &mi->rli; |
7840 | int error= 0; |
7841 | |
7842 | DBUG_EXECUTE_IF("crash_before_rotate_relaylog" , DBUG_SUICIDE();); |
7843 | |
7844 | /* |
7845 | We need to test inited because otherwise, new_file() will attempt to lock |
7846 | LOCK_log, which may not be inited (if we're not a slave). |
7847 | */ |
7848 | if (!rli->inited) |
7849 | { |
7850 | DBUG_PRINT("info" , ("rli->inited == 0" )); |
7851 | goto end; |
7852 | } |
7853 | |
7854 | /* If the relay log is closed, new_file() will do nothing. */ |
7855 | if ((error= rli->relay_log.new_file())) |
7856 | goto end; |
7857 | |
7858 | /* |
7859 | We harvest now, because otherwise BIN_LOG_HEADER_SIZE will not immediately |
7860 | be counted, so imagine a succession of FLUSH LOGS and assume the slave |
7861 | threads are started: |
7862 | relay_log_space decreases by the size of the deleted relay log, but does |
7863 | not increase, so flush-after-flush we may become negative, which is wrong. |
7864 | Even if this will be corrected as soon as a query is replicated on the |
7865 | slave (because the I/O thread will then call harvest_bytes_written() which |
7866 | will harvest all these BIN_LOG_HEADER_SIZE we forgot), it may give strange |
7867 | output in SHOW SLAVE STATUS meanwhile. So we harvest now. |
7868 | If the log is closed, then this will just harvest the last writes, probably |
7869 | 0 as they probably have been harvested. |
7870 | |
7871 | Note that it needs to be protected by mi->data_lock. |
7872 | */ |
7873 | mysql_mutex_assert_owner(&mi->data_lock); |
7874 | rli->relay_log.harvest_bytes_written(&rli->log_space_total); |
7875 | end: |
7876 | DBUG_RETURN(error); |
7877 | } |
7878 | |
7879 | |
7880 | /** |
7881 | Detects, based on master's version (as found in the relay log), if master |
7882 | has a certain bug. |
7883 | @param rli Relay_log_info which tells the master's version |
7884 | @param bug_id Number of the bug as found in bugs.mysql.com |
7885 | @param report bool report error message, default TRUE |
7886 | |
7887 | @param pred Predicate function that will be called with @c param to |
7888 | check for the bug. If the function return @c true, the bug is present, |
7889 | otherwise, it is not. |
7890 | |
7891 | @param param State passed to @c pred function. |
7892 | |
7893 | @return TRUE if master has the bug, FALSE if it does not. |
7894 | */ |
7895 | bool rpl_master_has_bug(const Relay_log_info *rli, uint bug_id, bool report, |
7896 | bool (*pred)(const void *), const void *param) |
7897 | { |
7898 | struct st_version_range_for_one_bug { |
7899 | uint bug_id; |
7900 | const uchar introduced_in[3]; // first version with bug |
7901 | const uchar fixed_in[3]; // first version with fix |
7902 | }; |
7903 | static struct st_version_range_for_one_bug versions_for_all_bugs[]= |
7904 | { |
7905 | {24432, { 5, 0, 24 }, { 5, 0, 38 } }, |
7906 | {24432, { 5, 1, 12 }, { 5, 1, 17 } }, |
7907 | {33029, { 5, 0, 0 }, { 5, 0, 58 } }, |
7908 | {33029, { 5, 1, 0 }, { 5, 1, 12 } }, |
7909 | {37426, { 5, 1, 0 }, { 5, 1, 26 } }, |
7910 | }; |
7911 | const uchar *master_ver= |
7912 | rli->relay_log.description_event_for_exec->server_version_split.ver; |
7913 | |
7914 | DBUG_ASSERT(sizeof(rli->relay_log.description_event_for_exec->server_version_split.ver) == 3); |
7915 | |
7916 | for (uint i= 0; |
7917 | i < sizeof(versions_for_all_bugs)/sizeof(*versions_for_all_bugs);i++) |
7918 | { |
7919 | const uchar *introduced_in= versions_for_all_bugs[i].introduced_in, |
7920 | *fixed_in= versions_for_all_bugs[i].fixed_in; |
7921 | if ((versions_for_all_bugs[i].bug_id == bug_id) && |
7922 | (memcmp(introduced_in, master_ver, 3) <= 0) && |
7923 | (memcmp(fixed_in, master_ver, 3) > 0) && |
7924 | (pred == NULL || (*pred)(param))) |
7925 | { |
7926 | if (!report) |
7927 | return TRUE; |
7928 | // a short message for SHOW SLAVE STATUS (message length constraints) |
7929 | my_printf_error(ER_UNKNOWN_ERROR, "master may suffer from" |
7930 | " http://bugs.mysql.com/bug.php?id=%u" |
7931 | " so slave stops; check error log on slave" |
7932 | " for more info" , MYF(0), bug_id); |
7933 | // a verbose message for the error log |
7934 | rli->report(ERROR_LEVEL, ER_UNKNOWN_ERROR, NULL, |
7935 | "According to the master's version ('%s')," |
7936 | " it is probable that master suffers from this bug:" |
7937 | " http://bugs.mysql.com/bug.php?id=%u" |
7938 | " and thus replicating the current binary log event" |
7939 | " may make the slave's data become different from the" |
7940 | " master's data." |
7941 | " To take no risk, slave refuses to replicate" |
7942 | " this event and stops." |
7943 | " We recommend that all updates be stopped on the" |
7944 | " master and slave, that the data of both be" |
7945 | " manually synchronized," |
7946 | " that master's binary logs be deleted," |
7947 | " that master be upgraded to a version at least" |
7948 | " equal to '%d.%d.%d'. Then replication can be" |
7949 | " restarted." , |
7950 | rli->relay_log.description_event_for_exec->server_version, |
7951 | bug_id, |
7952 | fixed_in[0], fixed_in[1], fixed_in[2]); |
7953 | return TRUE; |
7954 | } |
7955 | } |
7956 | return FALSE; |
7957 | } |
7958 | |
7959 | /** |
7960 | BUG#33029, For all 5.0 up to 5.0.58 exclusive, and 5.1 up to 5.1.12 |
7961 | exclusive, if one statement in a SP generated AUTO_INCREMENT value |
7962 | by the top statement, all statements after it would be considered |
7963 | generated AUTO_INCREMENT value by the top statement, and a |
7964 | erroneous INSERT_ID value might be associated with these statement, |
7965 | which could cause duplicate entry error and stop the slave. |
7966 | |
7967 | Detect buggy master to work around. |
7968 | */ |
7969 | bool rpl_master_erroneous_autoinc(THD *thd) |
7970 | { |
7971 | if (thd->rgi_slave) |
7972 | { |
7973 | DBUG_EXECUTE_IF("simulate_bug33029" , return TRUE;); |
7974 | return rpl_master_has_bug(thd->rgi_slave->rli, 33029, FALSE, NULL, NULL); |
7975 | } |
7976 | return FALSE; |
7977 | } |
7978 | |
7979 | |
7980 | static bool get_row_event_stmt_end(const char* buf, |
7981 | const Format_description_log_event *fdle) |
7982 | { |
7983 | uint8 const = fdle->common_header_len; |
7984 | Log_event_type event_type= (Log_event_type)(uchar)buf[EVENT_TYPE_OFFSET]; |
7985 | |
7986 | uint8 const = fdle->post_header_len[event_type-1]; |
7987 | const char *flag_start= buf + common_header_len; |
7988 | /* |
7989 | The term 4 below signifies that master is of 'an intermediate source', see |
7990 | Rows_log_event::Rows_log_event. |
7991 | */ |
7992 | flag_start += RW_MAPID_OFFSET + ((post_header_len == 6) ? 4 : RW_FLAGS_OFFSET); |
7993 | |
7994 | return (uint2korr(flag_start) & Rows_log_event::STMT_END_F) != 0; |
7995 | } |
7996 | |
7997 | |
7998 | /* |
7999 | Reset log event tracking data. |
8000 | */ |
8001 | |
8002 | void Rows_event_tracker::reset() |
8003 | { |
8004 | binlog_file_name[0]= 0; |
8005 | first_seen= last_seen= 0; |
8006 | stmt_end_seen= false; |
8007 | } |
8008 | |
8009 | |
8010 | /* |
8011 | Update log event tracking data. |
8012 | |
8013 | The first- and last- seen event binlog position get memorized, as |
8014 | well as the end-of-statement status of the last one. |
8015 | */ |
8016 | |
8017 | void Rows_event_tracker::update(const char* file_name, my_off_t pos, |
8018 | const char* buf, |
8019 | const Format_description_log_event *fdle) |
8020 | { |
8021 | if (!first_seen) |
8022 | { |
8023 | first_seen= pos; |
8024 | strmake(binlog_file_name, file_name, sizeof(binlog_file_name) - 1); |
8025 | } |
8026 | last_seen= pos; |
8027 | DBUG_ASSERT(stmt_end_seen == 0); // We can only have one |
8028 | stmt_end_seen= get_row_event_stmt_end(buf, fdle); |
8029 | }; |
8030 | |
8031 | |
8032 | /** |
8033 | The function is called at next event reading |
8034 | after a sequence of Rows- log-events. It checks the end-of-statement status |
8035 | of the past sequence to report on any isssue. |
8036 | In the positive case the tracker gets reset. |
8037 | |
8038 | @return true when the Rows- event group integrity found compromised, |
8039 | false otherwise. |
8040 | */ |
8041 | bool Rows_event_tracker::check_and_report(const char* file_name, |
8042 | my_off_t pos) |
8043 | { |
8044 | if (last_seen) |
8045 | { |
8046 | // there was at least one "block" event previously |
8047 | if (!stmt_end_seen) |
8048 | { |
8049 | sql_print_error("Slave IO thread did not receive an expected " |
8050 | "Rows-log end-of-statement for event starting " |
8051 | "at log '%s' position %llu " |
8052 | "whose last block was seen at log '%s' position %llu. " |
8053 | "The end-of-statement should have been delivered " |
8054 | "before the current one at log '%s' position %llu" , |
8055 | binlog_file_name, first_seen, |
8056 | binlog_file_name, last_seen, file_name, pos); |
8057 | return true; |
8058 | } |
8059 | reset(); |
8060 | } |
8061 | |
8062 | return false; |
8063 | } |
8064 | |
8065 | /** |
8066 | @} (end of group Replication) |
8067 | */ |
8068 | |
8069 | #endif /* HAVE_REPLICATION */ |
8070 | |