1
/* Copyright (C) 2000-2003 MySQL AB
3
This program is free software; you can redistribute it and/or modify
4
it under the terms of the GNU General Public License as published by
5
the Free Software Foundation; version 2 of the License.
7
This program is distributed in the hope that it will be useful,
8
but WITHOUT ANY WARRANTY; without even the implied warranty of
9
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
10
GNU General Public License for more details.
12
You should have received a copy of the GNU General Public License
13
along with this program; if not, write to the Free Software
14
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA */
24
Abort logging when we get an error in reading or writing log files
27
#include <drizzled/server_includes.h>
29
#include "rpl_filter.h"
32
#include <mysys/my_dir.h>
35
#include <drizzled/plugin.h>
36
#include <drizzled/drizzled_error_messages.h>
37
#include <libdrizzle/gettext.h>
39
/* max size of the log message */
40
#define MAX_LOG_BUFFER_SIZE 1024
41
#define MAX_USER_HOST_SIZE 512
42
#define MAX_TIME_SIZE 32
43
#define MY_OFF_T_UNDEF (~(my_off_t)0UL)
45
#define FLAGSTR(V,F) ((V)&(F)?#F" ":"")
49
DRIZZLE_BIN_LOG mysql_bin_log;
50
ulong sync_binlog_counter= 0;
52
static bool test_if_number(const char *str,
53
long *res, bool allow_wildcards);
54
static int binlog_init(void *p);
55
static int binlog_close_connection(handlerton *hton, THD *thd);
56
static int binlog_savepoint_set(handlerton *hton, THD *thd, void *sv);
57
static int binlog_savepoint_rollback(handlerton *hton, THD *thd, void *sv);
58
static int binlog_commit(handlerton *hton, THD *thd, bool all);
59
static int binlog_rollback(handlerton *hton, THD *thd, bool all);
60
static int binlog_prepare(handlerton *hton, THD *thd, bool all);
63
sql_print_message_func sql_print_message_handlers[3] =
65
sql_print_information,
71
char *make_default_log_name(char *buff,const char* log_ext)
73
strmake(buff, pidfile_name, FN_REFLEN-5);
74
return fn_format(buff, buff, mysql_data_home, log_ext,
75
MYF(MY_UNPACK_FILENAME|MY_REPLACE_EXT));
79
Helper class to hold a mutex for the duration of the
82
Eliminates the need for explicit unlocking of mutexes on, e.g.,
83
error returns. On passing a null pointer, the sentry will not do
89
Mutex_sentry(pthread_mutex_t *mutex)
93
pthread_mutex_lock(mutex);
99
pthread_mutex_unlock(m_mutex);
104
pthread_mutex_t *m_mutex;
106
// It's not allowed to copy this object in any way
107
Mutex_sentry(Mutex_sentry const&);
108
void operator=(Mutex_sentry const&);
112
Helper class to store binary log transaction data.
114
class binlog_trx_data {
117
: at_least_one_stmt(0), m_pending(0), before_stmt_pos(MY_OFF_T_UNDEF)
119
trans_log.end_of_file= max_binlog_cache_size;
124
assert(pending() == NULL);
125
close_cached_file(&trans_log);
128
my_off_t position() const {
129
return my_b_tell(&trans_log);
134
return pending() == NULL && my_b_tell(&trans_log) == 0;
138
Truncate the transaction cache to a certain position. This
139
includes deleting the pending event.
141
void truncate(my_off_t pos)
145
reinit_io_cache(&trans_log, WRITE_CACHE, pos, 0, 0);
146
if (pos < before_stmt_pos)
147
before_stmt_pos= MY_OFF_T_UNDEF;
150
The only valid positions that can be truncated to are at the
151
beginning of a statement. We are relying on this fact to be able
152
to set the at_least_one_stmt flag correctly. In other word, if
153
we are truncating to the beginning of the transaction cache,
154
there will be no statements in the cache, otherwhise, we will
155
have at least one statement in the transaction cache.
157
at_least_one_stmt= (pos > 0);
161
Reset the entire contents of the transaction cache, emptying it
167
before_stmt_pos= MY_OFF_T_UNDEF;
168
trans_log.end_of_file= max_binlog_cache_size;
171
Rows_log_event *pending() const
176
void set_pending(Rows_log_event *const pending)
181
IO_CACHE trans_log; // The transaction cache
184
Boolean that is true if there is at least one statement in the
187
bool at_least_one_stmt;
191
Pending binrows event. This event is the event where the rows are
194
Rows_log_event *m_pending;
198
Binlog position before the start of the current statement.
200
my_off_t before_stmt_pos;
203
handlerton *binlog_hton;
206
/* Check if a given table is opened log table */
207
int check_if_log_table(uint32_t db_len __attribute__((unused)),
208
const char *db __attribute__((unused)),
209
uint32_t table_name_len __attribute__((unused)),
210
const char *table_name __attribute__((unused)),
211
uint32_t check_if_opened __attribute__((unused)))
217
Log error with all enabled log event handlers
222
level The level of the error significance: NOTE,
224
format format string for the error message
225
args list of arguments for the format string
232
bool LOGGER::error_log_print(enum loglevel level, const char *format,
236
Log_event_handler **current_handler;
238
/* currently we don't need locking here as there is no error_log table */
239
for (current_handler= error_log_handler_list ; *current_handler ;)
240
error= (*current_handler++)->log_error(level, format, args) || error;
246
void LOGGER::cleanup_base()
249
rwlock_destroy(&LOCK_logger);
253
void LOGGER::cleanup_end()
260
Perform basic log initialization: create file-based log handler and
263
void LOGGER::init_base()
268
/* by default we use traditional error log */
269
init_error_log(LOG_FILE);
271
my_rwlock_init(&LOCK_logger, NULL);
275
bool LOGGER::flush_logs(THD *thd __attribute__((unused)))
280
Now we lock logger, as nobody should be able to use logging routines while
281
log tables are closed
283
logger.lock_exclusive();
285
/* end of log flush */
290
void LOGGER::init_error_log(uint32_t error_log_printer)
292
if (error_log_printer & LOG_NONE)
294
error_log_handler_list[0]= 0;
300
int LOGGER::set_handlers(uint32_t error_log_printer)
302
/* error log table is not supported yet */
305
init_error_log(error_log_printer);
313
Save position of binary log transaction cache.
316
binlog_trans_log_savepos()
318
thd The thread to take the binlog data from
319
pos Pointer to variable where the position will be stored
323
Save the current position in the binary log transaction cache into
324
the variable pointed to by 'pos'
328
binlog_trans_log_savepos(THD *thd, my_off_t *pos)
331
if (thd_get_ha_data(thd, binlog_hton) == NULL)
332
thd->binlog_setup_trx_data();
333
binlog_trx_data *const trx_data=
334
(binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
335
assert(mysql_bin_log.is_open());
336
*pos= trx_data->position();
342
Truncate the binary log transaction cache.
345
binlog_trans_log_truncate()
347
thd The thread to take the binlog data from
348
pos Position to truncate to
352
Truncate the binary log to the given position. Will not change
357
binlog_trans_log_truncate(THD *thd, my_off_t pos)
359
assert(thd_get_ha_data(thd, binlog_hton) != NULL);
360
/* Only true if binlog_trans_log_savepos() wasn't called before */
361
assert(pos != ~(my_off_t) 0);
363
binlog_trx_data *const trx_data=
364
(binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
365
trx_data->truncate(pos);
371
this function is mostly a placeholder.
372
conceptually, binlog initialization (now mostly done in DRIZZLE_BIN_LOG::open)
373
should be moved here.
376
int binlog_init(void *p)
378
binlog_hton= (handlerton *)p;
379
binlog_hton->state=opt_bin_log ? SHOW_OPTION_YES : SHOW_OPTION_NO;
380
binlog_hton->savepoint_offset= sizeof(my_off_t);
381
binlog_hton->close_connection= binlog_close_connection;
382
binlog_hton->savepoint_set= binlog_savepoint_set;
383
binlog_hton->savepoint_rollback= binlog_savepoint_rollback;
384
binlog_hton->commit= binlog_commit;
385
binlog_hton->rollback= binlog_rollback;
386
binlog_hton->prepare= binlog_prepare;
387
binlog_hton->flags= HTON_NOT_USER_SELECTABLE | HTON_HIDDEN;
392
static int binlog_close_connection(handlerton *hton __attribute__((unused)),
395
binlog_trx_data *const trx_data=
396
(binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
397
assert(trx_data->empty());
398
thd_set_ha_data(thd, binlog_hton, NULL);
399
trx_data->~binlog_trx_data();
400
free((unsigned char*)trx_data);
410
thd The thread whose transaction should be ended
411
trx_data Pointer to the transaction data to use
412
end_ev The end event to use, or NULL
413
all True if the entire transaction should be ended, false if
414
only the statement transaction should be ended.
418
End the currently open transaction. The transaction can be either
419
a real transaction (if 'all' is true) or a statement transaction
422
If 'end_ev' is NULL, the transaction is a rollback of only
423
transactional tables, so the transaction cache will be truncated
424
to either just before the last opened statement transaction (if
425
'all' is false), or reset completely (if 'all' is true).
428
binlog_end_trans(THD *thd, binlog_trx_data *trx_data,
429
Log_event *end_ev, bool all)
432
IO_CACHE *trans_log= &trx_data->trans_log;
435
NULL denotes ROLLBACK with nothing to replicate: i.e., rollback of
436
only transactional tables. If the transaction contain changes to
437
any non-transactiona tables, we need write the transaction and log
443
Doing a commit or a rollback including non-transactional tables,
444
i.e., ending a transaction where we might write the transaction
445
cache to the binary log.
447
We can always end the statement when ending a transaction since
448
transactions are not allowed inside stored functions. If they
449
were, we would have to ensure that we're not ending a statement
450
inside a stored function.
452
thd->binlog_flush_pending_rows_event(true);
454
error= mysql_bin_log.write(thd, &trx_data->trans_log, end_ev);
458
We need to step the table map version after writing the
459
transaction cache to disk.
461
mysql_bin_log.update_table_map_version();
462
statistic_increment(binlog_cache_use, &LOCK_status);
463
if (trans_log->disk_writes != 0)
465
statistic_increment(binlog_cache_disk_use, &LOCK_status);
466
trans_log->disk_writes= 0;
472
If rolling back an entire transaction or a single statement not
473
inside a transaction, we reset the transaction cache.
475
If rolling back a statement in a transaction, we truncate the
476
transaction cache to remove the statement.
478
if (all || !(thd->options & (OPTION_BEGIN | OPTION_NOT_AUTOCOMMIT)))
482
assert(!thd->binlog_get_pending_rows_event());
483
thd->clear_binlog_table_maps();
486
trx_data->truncate(trx_data->before_stmt_pos);
489
We need to step the table map version on a rollback to ensure
490
that a new table map event is generated instead of the one that
491
was written to the thrown-away transaction cache.
493
mysql_bin_log.update_table_map_version();
499
static int binlog_prepare(handlerton *hton __attribute__((unused)),
500
THD *thd __attribute__((unused)),
501
bool all __attribute__((unused)))
505
just pretend we can do 2pc, so that MySQL won't
507
real work will be done in DRIZZLE_BIN_LOG::log_xid()
512
#define YESNO(X) ((X) ? "yes" : "no")
515
This function is called once after each statement.
517
It has the responsibility to flush the transaction cache to the
518
binlog file on commits.
520
@param hton The binlog handlerton.
521
@param thd The client thread that executes the transaction.
522
@param all This is @c true if this is a real transaction commit, and
525
@see handlerton::commit
527
static int binlog_commit(handlerton *hton __attribute__((unused)),
530
binlog_trx_data *const trx_data=
531
(binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
533
if (trx_data->empty())
535
// we're here because trans_log was flushed in DRIZZLE_BIN_LOG::log_xid()
541
Decision table for committing a transaction. The top part, the
542
*conditions* represent different cases that can occur, and hte
543
bottom part, the *actions*, represent what should be done in that
546
Real transaction 'all' was true
548
Statement in cache There were at least one statement in the
551
In transaction We are inside a transaction
553
Stmt modified non-trans The statement being committed modified a
554
non-transactional table
556
All modified non-trans Some statement before this one in the
557
transaction modified a non-transactional
561
============================= = = = = = = = = = = = = = = = =
562
Real transaction N N N N N N N N N N N N N N N N
563
Statement in cache N N N N N N N N Y Y Y Y Y Y Y Y
564
In transaction N N N N Y Y Y Y N N N N Y Y Y Y
565
Stmt modified non-trans N N Y Y N N Y Y N N Y Y N N Y Y
566
All modified non-trans N Y N Y N Y N Y N Y N Y N Y N Y
568
Action: (C)ommit/(A)ccumulate C C - C A C - C - - - - A A - A
569
============================= = = = = = = = = = = = = = = = =
572
============================= = = = = = = = = = = = = = = = =
573
Real transaction Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y
574
Statement in cache N N N N N N N N Y Y Y Y Y Y Y Y
575
In transaction N N N N Y Y Y Y N N N N Y Y Y Y
576
Stmt modified non-trans N N Y Y N N Y Y N N Y Y N N Y Y
577
All modified non-trans N Y N Y N Y N Y N Y N Y N Y N Y
579
(C)ommit/(A)ccumulate/(-) - - - - C C - C - - - - C C - C
580
============================= = = = = = = = = = = = = = = = =
582
In other words, we commit the transaction if and only if both of
583
the following are true:
584
- We are not in a transaction and committing a statement
586
- We are in a transaction and one (or more) of the following are
589
- A full transaction is committed
593
- A non-transactional statement is committed and there is
596
Otherwise, we accumulate the statement
598
uint64_t const in_transaction=
599
thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN);
600
if ((in_transaction && (all || (!trx_data->at_least_one_stmt && thd->transaction.stmt.modified_non_trans_table))) || (!in_transaction && !all))
602
Query_log_event qev(thd, STRING_WITH_LEN("COMMIT"), true, false);
603
qev.error_code= 0; // see comment in DRIZZLE_LOG::write(THD, IO_CACHE)
604
int error= binlog_end_trans(thd, trx_data, &qev, all);
611
This function is called when a transaction involving a transactional
612
table is rolled back.
614
It has the responsibility to flush the transaction cache to the
615
binlog file. However, if the transaction does not involve
616
non-transactional tables, nothing needs to be logged.
618
@param hton The binlog handlerton.
619
@param thd The client thread that executes the transaction.
620
@param all This is @c true if this is a real transaction rollback, and
623
@see handlerton::rollback
625
static int binlog_rollback(handlerton *hton __attribute__((unused)),
629
binlog_trx_data *const trx_data=
630
(binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
632
if (trx_data->empty()) {
637
if ((all && thd->transaction.all.modified_non_trans_table) ||
638
(!all && thd->transaction.stmt.modified_non_trans_table) ||
639
(thd->options & OPTION_KEEP_LOG))
642
We write the transaction cache with a rollback last if we have
643
modified any non-transactional table. We do this even if we are
644
committing a single statement that has modified a
645
non-transactional table since it can have modified a
646
transactional table in that statement as well, which needs to be
647
rolled back on the slave.
649
Query_log_event qev(thd, STRING_WITH_LEN("ROLLBACK"), true, false);
650
qev.error_code= 0; // see comment in DRIZZLE_LOG::write(THD, IO_CACHE)
651
error= binlog_end_trans(thd, trx_data, &qev, all);
653
else if ((all && !thd->transaction.all.modified_non_trans_table) ||
654
(!all && !thd->transaction.stmt.modified_non_trans_table))
657
If we have modified only transactional tables, we can truncate
658
the transaction cache without writing anything to the binary
661
error= binlog_end_trans(thd, trx_data, 0, all);
668
How do we handle this (unlikely but legal) case:
670
[transaction] + [update to non-trans table] + [rollback to savepoint] ?
672
The problem occurs when a savepoint is before the update to the
673
non-transactional table. Then when there's a rollback to the savepoint, if we
674
simply truncate the binlog cache, we lose the part of the binlog cache where
675
the update is. If we want to not lose it, we need to write the SAVEPOINT
676
command and the ROLLBACK TO SAVEPOINT command to the binlog cache. The latter
677
is easy: it's just write at the end of the binlog cache, but the former
678
should be *inserted* to the place where the user called SAVEPOINT. The
679
solution is that when the user calls SAVEPOINT, we write it to the binlog
680
cache (so no need to later insert it). As transactions are never intermixed
681
in the binary log (i.e. they are serialized), we won't have conflicts with
682
savepoint names when using mysqlbinlog or in the slave SQL thread.
683
Then when ROLLBACK TO SAVEPOINT is called, if we updated some
684
non-transactional table, we don't truncate the binlog cache but instead write
685
ROLLBACK TO SAVEPOINT to it; otherwise we truncate the binlog cache (which
686
will chop the SAVEPOINT command from the binlog cache, which is good as in
687
that case there is no need to have it in the binlog).
690
static int binlog_savepoint_set(handlerton *hton __attribute__((unused)),
693
binlog_trans_log_savepos(thd, (my_off_t*) sv);
694
/* Write it to the binary log */
697
thd->binlog_query(THD::STMT_QUERY_TYPE,
698
thd->query, thd->query_length, true, false);
702
static int binlog_savepoint_rollback(handlerton *hton __attribute__((unused)),
706
Write ROLLBACK TO SAVEPOINT to the binlog cache if we have updated some
707
non-transactional table. Otherwise, truncate the binlog cache starting
708
from the SAVEPOINT command.
710
if (unlikely(thd->transaction.all.modified_non_trans_table ||
711
(thd->options & OPTION_KEEP_LOG)))
714
thd->binlog_query(THD::STMT_QUERY_TYPE,
715
thd->query, thd->query_length, true, false);
718
binlog_trans_log_truncate(thd, *(my_off_t*)sv);
723
int check_binlog_magic(IO_CACHE* log, const char** errmsg)
726
assert(my_b_tell(log) == 0);
728
if (my_b_read(log, (unsigned char*) magic, sizeof(magic)))
730
*errmsg = _("I/O error reading the header from the binary log");
731
sql_print_error("%s, errno=%d, io cache code=%d", *errmsg, my_errno,
735
if (memcmp(magic, BINLOG_MAGIC, sizeof(magic)))
737
*errmsg = _("Binlog has bad magic number; It's not a binary log file "
738
"that can be used by this version of Drizzle");
745
File open_binlog(IO_CACHE *log, const char *log_file_name, const char **errmsg)
749
if ((file = my_open(log_file_name, O_RDONLY,
752
sql_print_error(_("Failed to open log (file '%s', errno %d)"),
753
log_file_name, my_errno);
754
*errmsg = _("Could not open log file");
757
if (init_io_cache(log, file, IO_SIZE*2, READ_CACHE, 0, 0,
758
MYF(MY_WME|MY_DONT_CHECK_FILESIZE)))
760
sql_print_error(_("Failed to create a cache on log (file '%s')"),
762
*errmsg = _("Could not open log file");
765
if (check_binlog_magic(log,errmsg))
772
my_close(file,MYF(0));
780
Find a unique filename for 'filename.#'.
782
Set '#' to a number as low as possible.
785
nonzero if not possible to get unique filename
788
static int find_uniq_filename(char *name)
792
char buff[FN_REFLEN];
793
struct st_my_dir *dir_info;
794
register struct fileinfo *file_info;
796
size_t buf_length, length;
799
length= dirname_part(buff, name, &buf_length);
800
start= name + length;
801
end= strchr(start, '\0');
804
length= (size_t) (end-start+1);
806
if (!(dir_info = my_dir(buff,MYF(MY_DONT_SORT))))
807
{ // This shouldn't happen
808
my_stpcpy(end,".1"); // use name+1
811
file_info= dir_info->dir_entry;
812
for (i=dir_info->number_off_files ; i-- ; file_info++)
814
if (memcmp(file_info->name, start, length) == 0 &&
815
test_if_number(file_info->name+length, &number,0))
817
set_if_bigger(max_found,(ulong) number);
823
sprintf(end,"%06ld",max_found+1);
828
void DRIZZLE_LOG::init(enum_log_type log_type_arg,
829
enum cache_type io_cache_type_arg)
831
log_type= log_type_arg;
832
io_cache_type= io_cache_type_arg;
838
Open a (new) log file.
843
log_name The name of the log to open
844
log_type_arg The type of the log. E.g. LOG_NORMAL
845
new_name The new name for the logfile. This is only needed
846
when the method is used to open the binlog file.
847
io_cache_type_arg The type of the IO_CACHE to use for this log file
850
Open the logfile, init IO_CACHE and write startup messages
851
(in case of general and slow query logs).
858
bool DRIZZLE_LOG::open(const char *log_name, enum_log_type log_type_arg,
859
const char *new_name, enum cache_type io_cache_type_arg)
861
char buff[FN_REFLEN];
863
int open_flags= O_CREAT;
867
init(log_type_arg, io_cache_type_arg);
869
if (!(name= my_strdup(log_name, MYF(MY_WME))))
871
name= (char *)log_name; // for the error message
876
my_stpcpy(log_file_name, new_name);
877
else if (generate_new_name(log_file_name, name))
880
if (io_cache_type == SEQ_READ_APPEND)
881
open_flags |= O_RDWR | O_APPEND;
883
open_flags |= O_WRONLY | (log_type == LOG_BIN ? 0 : O_APPEND);
887
if ((file= my_open(log_file_name, open_flags,
888
MYF(MY_WME | ME_WAITTANG))) < 0 ||
889
init_io_cache(&log_file, file, IO_SIZE, io_cache_type,
890
my_tell(file, MYF(MY_WME)), 0,
891
MYF(MY_WME | MY_NABP |
892
((log_type == LOG_BIN) ? MY_WAIT_IF_FULL : 0))))
895
if (log_type == LOG_NORMAL)
898
int len=snprintf(buff, sizeof(buff), "%s, Version: %s (%s). "
899
"started with:\nTCP Port: %d, Named Pipe: %s\n",
900
my_progname, server_version, DRIZZLE_COMPILATION_COMMENT,
903
end= my_stpncpy(buff + len, "Time Id Command Argument\n",
905
if (my_b_write(&log_file, (unsigned char*) buff, (uint) (end-buff)) ||
906
flush_io_cache(&log_file))
910
log_state= LOG_OPENED;
914
sql_print_error(_("Could not use %s for logging (error %d). "
915
"Turning logging off for the whole duration of the "
916
"Drizzle server process. "
917
"To turn it on again: fix the cause, "
918
"shutdown the Drizzle server and restart it."),
921
my_close(file, MYF(0));
922
end_io_cache(&log_file);
928
log_state= LOG_CLOSED;
932
DRIZZLE_LOG::DRIZZLE_LOG()
933
: name(0), write_error(false), inited(false), log_type(LOG_UNKNOWN),
934
log_state(LOG_CLOSED)
937
We don't want to initialize LOCK_Log here as such initialization depends on
938
safe_mutex (when using safe_mutex) which depends on MY_INIT(), which is
939
called only in main(). Doing initialization here would make it happen
942
memset(&log_file, 0, sizeof(log_file));
945
void DRIZZLE_LOG::init_pthread_objects()
949
(void) pthread_mutex_init(&LOCK_log, MY_MUTEX_INIT_SLOW);
957
exiting Bitmask. For the slow and general logs the only used bit is
958
LOG_CLOSE_TO_BE_OPENED. This is used if we intend to call
959
open at once after close.
962
One can do an open on the object at once after doing a close.
963
The internal structures are not freed until cleanup() is called
966
void DRIZZLE_LOG::close(uint32_t exiting)
967
{ // One can't set log_type here!
968
if (log_state == LOG_OPENED)
970
end_io_cache(&log_file);
972
if (my_sync(log_file.file, MYF(MY_WME)) && ! write_error)
975
sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
978
if (my_close(log_file.file, MYF(MY_WME)) && ! write_error)
981
sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
985
log_state= (exiting & LOG_CLOSE_TO_BE_OPENED) ? LOG_TO_BE_OPENED : LOG_CLOSED;
994
/** This is called only once. */
996
void DRIZZLE_LOG::cleanup()
1001
(void) pthread_mutex_destroy(&LOCK_log);
1008
int DRIZZLE_LOG::generate_new_name(char *new_name, const char *log_name)
1010
fn_format(new_name, log_name, mysql_data_home, "", 4);
1011
if (log_type == LOG_BIN)
1013
if (!fn_ext(log_name)[0])
1015
if (find_uniq_filename(new_name))
1017
sql_print_error(ER(ER_NO_UNIQUE_LOGFILE), log_name);
1028
The following should be using fn_format(); We just need to
1029
first change fn_format() to cut the file name if it's too long.
1031
const char *DRIZZLE_LOG::generate_name(const char *log_name,
1033
bool strip_ext, char *buff)
1035
if (!log_name || !log_name[0])
1037
strmake(buff, pidfile_name, FN_REFLEN - strlen(suffix) - 1);
1038
return (const char *)
1039
fn_format(buff, buff, "", suffix, MYF(MY_REPLACE_EXT|MY_REPLACE_DIR));
1041
// get rid of extension if the log is binary to avoid problems
1044
char *p= fn_ext(log_name);
1045
uint32_t length= (uint) (p - log_name);
1046
strmake(buff, log_name, cmin(length, (uint)FN_REFLEN));
1047
return (const char*)buff;
1054
DRIZZLE_BIN_LOG::DRIZZLE_BIN_LOG()
1055
:bytes_written(0), prepared_xids(0), file_id(1), open_count(1),
1056
need_start_event(true), m_table_map_version(0),
1057
description_event_for_exec(0), description_event_for_queue(0)
1060
We don't want to initialize locks here as such initialization depends on
1061
safe_mutex (when using safe_mutex) which depends on MY_INIT(), which is
1062
called only in main(). Doing initialization here would make it happen
1065
index_file_name[0] = 0;
1066
memset(&index_file, 0, sizeof(index_file));
1069
/* this is called only once */
1071
void DRIZZLE_BIN_LOG::cleanup()
1076
close(LOG_CLOSE_INDEX|LOG_CLOSE_STOP_EVENT);
1077
delete description_event_for_queue;
1078
delete description_event_for_exec;
1079
(void) pthread_mutex_destroy(&LOCK_log);
1080
(void) pthread_mutex_destroy(&LOCK_index);
1081
(void) pthread_cond_destroy(&update_cond);
1087
/* Init binlog-specific vars */
1088
void DRIZZLE_BIN_LOG::init(bool no_auto_events_arg, ulong max_size_arg)
1090
no_auto_events= no_auto_events_arg;
1091
max_size= max_size_arg;
1096
void DRIZZLE_BIN_LOG::init_pthread_objects()
1098
assert(inited == 0);
1100
(void) pthread_mutex_init(&LOCK_log, MY_MUTEX_INIT_SLOW);
1101
(void) pthread_mutex_init(&LOCK_index, MY_MUTEX_INIT_SLOW);
1102
(void) pthread_cond_init(&update_cond, 0);
1106
bool DRIZZLE_BIN_LOG::open_index_file(const char *index_file_name_arg,
1107
const char *log_name)
1109
File index_file_nr= -1;
1110
assert(!my_b_inited(&index_file));
1113
First open of this class instance
1114
Create an index file that will hold all file names uses for logging.
1115
Add new entries to the end of it.
1117
myf opt= MY_UNPACK_FILENAME;
1118
if (!index_file_name_arg)
1120
index_file_name_arg= log_name; // Use same basename for index file
1121
opt= MY_UNPACK_FILENAME | MY_REPLACE_EXT;
1123
fn_format(index_file_name, index_file_name_arg, mysql_data_home,
1125
if ((index_file_nr= my_open(index_file_name,
1127
MYF(MY_WME))) < 0 ||
1128
my_sync(index_file_nr, MYF(MY_WME)) ||
1129
init_io_cache(&index_file, index_file_nr,
1130
IO_SIZE, WRITE_CACHE,
1131
my_seek(index_file_nr,0L,MY_SEEK_END,MYF(0)),
1132
0, MYF(MY_WME | MY_WAIT_IF_FULL)))
1135
TODO: all operations creating/deleting the index file or a log, should
1136
call my_sync_dir() or my_sync_dir_by_file() to be durable.
1137
TODO: file creation should be done with my_create() not my_open().
1139
if (index_file_nr >= 0)
1140
my_close(index_file_nr,MYF(0));
1148
Open a (new) binlog file.
1150
- Open the log file and the index file. Register the new
1152
- When calling this when the file is in use, you must have a locks
1153
on LOCK_log and LOCK_index.
1161
bool DRIZZLE_BIN_LOG::open(const char *log_name,
1162
enum_log_type log_type_arg,
1163
const char *new_name,
1164
enum cache_type io_cache_type_arg,
1165
bool no_auto_events_arg,
1167
bool null_created_arg)
1173
/* open the main log file */
1174
if (DRIZZLE_LOG::open(log_name, log_type_arg, new_name, io_cache_type_arg))
1175
return(1); /* all warnings issued */
1177
init(no_auto_events_arg, max_size_arg);
1181
assert(log_type == LOG_BIN);
1184
bool write_file_name_to_index_file=0;
1186
if (!my_b_filelength(&log_file))
1189
The binary log file was empty (probably newly created)
1190
This is the normal case and happens when the user doesn't specify
1191
an extension for the binary log files.
1192
In this case we write a standard header to it.
1194
if (my_b_safe_write(&log_file, (unsigned char*) BINLOG_MAGIC,
1195
BIN_LOG_HEADER_SIZE))
1197
bytes_written+= BIN_LOG_HEADER_SIZE;
1198
write_file_name_to_index_file= 1;
1201
assert(my_b_inited(&index_file) != 0);
1202
reinit_io_cache(&index_file, WRITE_CACHE,
1203
my_b_filelength(&index_file), 0, 0);
1204
if (need_start_event && !no_auto_events)
1207
In 4.x we set need_start_event=0 here, but in 5.0 we want a Start event
1208
even if this is not the very first binlog.
1210
Format_description_log_event s(BINLOG_VERSION);
1212
don't set LOG_EVENT_BINLOG_IN_USE_F for SEQ_READ_APPEND io_cache
1213
as we won't be able to reset it later
1215
if (io_cache_type == WRITE_CACHE)
1216
s.flags|= LOG_EVENT_BINLOG_IN_USE_F;
1219
s.dont_set_created= null_created_arg;
1220
if (s.write(&log_file))
1222
bytes_written+= s.data_written;
1224
if (description_event_for_queue &&
1225
description_event_for_queue->binlog_version>=4)
1228
This is a relay log written to by the I/O slave thread.
1229
Write the event so that others can later know the format of this relay
1231
Note that this event is very close to the original event from the
1232
master (it has binlog version of the master, event types of the
1233
master), so this is suitable to parse the next relay log's event. It
1234
has been produced by
1235
Format_description_log_event::Format_description_log_event(char* buf,).
1236
Why don't we want to write the description_event_for_queue if this
1237
event is for format<4 (3.23 or 4.x): this is because in that case, the
1238
description_event_for_queue describes the data received from the
1239
master, but not the data written to the relay log (*conversion*),
1240
which is in format 4 (slave's).
1243
Set 'created' to 0, so that in next relay logs this event does not
1244
trigger cleaning actions on the slave in
1245
Format_description_log_event::apply_event_impl().
1247
description_event_for_queue->created= 0;
1248
/* Don't set log_pos in event header */
1249
description_event_for_queue->artificial_event=1;
1251
if (description_event_for_queue->write(&log_file))
1253
bytes_written+= description_event_for_queue->data_written;
1255
if (flush_io_cache(&log_file) ||
1256
my_sync(log_file.file, MYF(MY_WME)))
1259
if (write_file_name_to_index_file)
1262
As this is a new log file, we write the file name to the index
1263
file. As every time we write to the index file, we sync it.
1265
if (my_b_write(&index_file, (unsigned char*) log_file_name,
1266
strlen(log_file_name)) ||
1267
my_b_write(&index_file, (unsigned char*) "\n", 1) ||
1268
flush_io_cache(&index_file) ||
1269
my_sync(index_file.file, MYF(MY_WME)))
1273
log_state= LOG_OPENED;
1278
sql_print_error(_("Could not use %s for logging (error %d). "
1279
"Turning logging off for the whole duration of the "
1280
"Drizzle server process. "
1281
"To turn it on again: fix the cause, "
1282
"shutdown the Drizzle server and restart it."),
1285
my_close(file,MYF(0));
1286
end_io_cache(&log_file);
1287
end_io_cache(&index_file);
1293
log_state= LOG_CLOSED;
1298
int DRIZZLE_BIN_LOG::get_current_log(LOG_INFO* linfo)
1300
pthread_mutex_lock(&LOCK_log);
1301
int ret = raw_get_current_log(linfo);
1302
pthread_mutex_unlock(&LOCK_log);
1306
int DRIZZLE_BIN_LOG::raw_get_current_log(LOG_INFO* linfo)
1308
strmake(linfo->log_file_name, log_file_name, sizeof(linfo->log_file_name)-1);
1309
linfo->pos = my_b_tell(&log_file);
1314
Move all data up in a file in an filename index file.
1316
We do the copy outside of the IO_CACHE as the cache buffers would just
1317
make things slower and more complicated.
1318
In most cases the copy loop should only do one read.
1320
@param index_file File to move
1321
@param offset Move everything from here to beginning
1324
File will be truncated to be 'offset' shorter or filled up with newlines
1330
static bool copy_up_file_and_fill(IO_CACHE *index_file, my_off_t offset)
1333
my_off_t init_offset= offset;
1334
File file= index_file->file;
1335
unsigned char io_buf[IO_SIZE*2];
1337
for (;; offset+= bytes_read)
1339
(void) my_seek(file, offset, MY_SEEK_SET, MYF(0));
1340
if ((bytes_read= (int) my_read(file, io_buf, sizeof(io_buf), MYF(MY_WME)))
1344
break; // end of file
1345
(void) my_seek(file, offset-init_offset, MY_SEEK_SET, MYF(0));
1346
if (my_write(file, io_buf, bytes_read, MYF(MY_WME | MY_NABP)))
1349
/* The following will either truncate the file or fill the end with \n' */
1350
if (ftruncate(file, offset - init_offset) || my_sync(file, MYF(MY_WME)))
1353
/* Reset data in old index cache */
1354
reinit_io_cache(index_file, READ_CACHE, (my_off_t) 0, 0, 1);
1362
Find the position in the log-index-file for the given log name.
1364
@param linfo Store here the found log file name and position to
1365
the NEXT log file name in the index file.
1366
@param log_name Filename to find in the index file.
1367
Is a null pointer if we want to read the first entry
1368
@param need_lock Set this to 1 if the parent doesn't already have a
1372
On systems without the truncate function the file will end with one or
1373
more empty lines. These will be ignored when reading the file.
1378
LOG_INFO_EOF End of log-index-file found
1380
LOG_INFO_IO Got IO error while reading file
1383
int DRIZZLE_BIN_LOG::find_log_pos(LOG_INFO *linfo, const char *log_name,
1387
char *fname= linfo->log_file_name;
1388
uint32_t log_name_len= log_name ? (uint) strlen(log_name) : 0;
1391
Mutex needed because we need to make sure the file pointer does not
1392
move from under our feet
1395
pthread_mutex_lock(&LOCK_index);
1396
safe_mutex_assert_owner(&LOCK_index);
1398
/* As the file is flushed, we can't get an error here */
1399
(void) reinit_io_cache(&index_file, READ_CACHE, (my_off_t) 0, 0, 0);
1404
my_off_t offset= my_b_tell(&index_file);
1405
/* If we get 0 or 1 characters, this is the end of the file */
1407
if ((length= my_b_gets(&index_file, fname, FN_REFLEN)) <= 1)
1409
/* Did not find the given entry; Return not found or error */
1410
error= !index_file.error ? LOG_INFO_EOF : LOG_INFO_IO;
1414
// if the log entry matches, null string matching anything
1416
(log_name_len == length-1 && fname[log_name_len] == '\n' &&
1417
!memcmp(fname, log_name, log_name_len)))
1419
fname[length-1]=0; // remove last \n
1420
linfo->index_file_start_offset= offset;
1421
linfo->index_file_offset = my_b_tell(&index_file);
1427
pthread_mutex_unlock(&LOCK_index);
1433
Find the position in the log-index-file for the given log name.
1436
linfo Store here the next log file name and position to
1437
the file name after that.
1439
need_lock Set this to 1 if the parent doesn't already have a
1443
- Before calling this function, one has to call find_log_pos()
1445
- Mutex needed because we need to make sure the file pointer does not move
1451
LOG_INFO_EOF End of log-index-file found
1453
LOG_INFO_IO Got IO error while reading file
1456
int DRIZZLE_BIN_LOG::find_next_log(LOG_INFO* linfo, bool need_lock)
1460
char *fname= linfo->log_file_name;
1463
pthread_mutex_lock(&LOCK_index);
1464
safe_mutex_assert_owner(&LOCK_index);
1466
/* As the file is flushed, we can't get an error here */
1467
(void) reinit_io_cache(&index_file, READ_CACHE, linfo->index_file_offset, 0,
1470
linfo->index_file_start_offset= linfo->index_file_offset;
1471
if ((length=my_b_gets(&index_file, fname, FN_REFLEN)) <= 1)
1473
error = !index_file.error ? LOG_INFO_EOF : LOG_INFO_IO;
1476
fname[length-1]=0; // kill \n
1477
linfo->index_file_offset = my_b_tell(&index_file);
1481
pthread_mutex_unlock(&LOCK_index);
1487
Delete all logs refered to in the index file.
1488
Start writing to a new log file.
1490
The new index file will only contain this file.
1495
If not called from slave thread, write start event to new log
1503
bool DRIZZLE_BIN_LOG::reset_logs(THD* thd)
1507
const char* save_name;
1510
We need to get both locks to be sure that no one is trying to
1511
write to the index log file.
1513
pthread_mutex_lock(&LOCK_log);
1514
pthread_mutex_lock(&LOCK_index);
1517
The following mutex is needed to ensure that no threads call
1518
'delete thd' as we would then risk missing a 'rollback' from this
1519
thread. If the transaction involved MyISAM tables, it should go
1520
into binlog even on rollback.
1522
pthread_mutex_lock(&LOCK_thread_count);
1524
/* Save variables so that we can reopen the log */
1526
name=0; // Protect against free
1527
close(LOG_CLOSE_TO_BE_OPENED);
1529
/* First delete all old log files */
1531
if (find_log_pos(&linfo, NULL, 0))
1539
if ((error= my_delete_allow_opened(linfo.log_file_name, MYF(0))) != 0)
1541
if (my_errno == ENOENT)
1543
push_warning_printf(current_thd, DRIZZLE_ERROR::WARN_LEVEL_WARN,
1544
ER_LOG_PURGE_NO_FILE, ER(ER_LOG_PURGE_NO_FILE),
1545
linfo.log_file_name);
1546
sql_print_information(_("Failed to delete file '%s'"),
1547
linfo.log_file_name);
1553
push_warning_printf(current_thd, DRIZZLE_ERROR::WARN_LEVEL_ERROR,
1554
ER_BINLOG_PURGE_FATAL_ERR,
1555
_("a problem with deleting %s; "
1556
"consider examining correspondence "
1557
"of your binlog index file "
1558
"to the actual binlog files"),
1559
linfo.log_file_name);
1564
if (find_next_log(&linfo, 0))
1568
/* Start logging with a new file */
1569
close(LOG_CLOSE_INDEX);
1570
if ((error= my_delete_allow_opened(index_file_name, MYF(0)))) // Reset (open will update)
1572
if (my_errno == ENOENT)
1574
push_warning_printf(current_thd, DRIZZLE_ERROR::WARN_LEVEL_WARN,
1575
ER_LOG_PURGE_NO_FILE, ER(ER_LOG_PURGE_NO_FILE),
1577
sql_print_information(_("Failed to delete file '%s'"),
1584
push_warning_printf(current_thd, DRIZZLE_ERROR::WARN_LEVEL_ERROR,
1585
ER_BINLOG_PURGE_FATAL_ERR,
1586
"a problem with deleting %s; "
1587
"consider examining correspondence "
1588
"of your binlog index file "
1589
"to the actual binlog files",
1595
if (!thd->slave_thread)
1597
if (!open_index_file(index_file_name, 0))
1598
open(save_name, log_type, 0, io_cache_type, no_auto_events, max_size, 0);
1599
free((unsigned char*) save_name);
1602
pthread_mutex_unlock(&LOCK_thread_count);
1603
pthread_mutex_unlock(&LOCK_index);
1604
pthread_mutex_unlock(&LOCK_log);
1610
Delete relay log files prior to rli->group_relay_log_name
1611
(i.e. all logs which are not involved in a non-finished group
1612
(transaction)), remove them from the index file and start on next
1616
- Protects index file with LOCK_index
1617
- Delete relevant relay log files
1618
- Copy all file names after these ones to the front of the index file
1619
- If the OS has truncate, truncate the file, else fill it with \n'
1620
- Read the next file name from the index file and store in rli->linfo
1622
@param rli Relay log information
1623
@param included If false, all relay logs that are strictly before
1624
rli->group_relay_log_name are deleted ; if true, the
1625
latter is deleted too (i.e. all relay logs
1626
read by the SQL slave thread are deleted).
1629
- This is only called from the slave-execute thread when it has read
1630
all commands from a relay log and want to switch to a new relay log.
1631
- When this happens, we can be in an active transaction as
1632
a transaction can span over two relay logs
1633
(although it is always written as a single block to the master's binary
1634
log, hence cannot span over two master's binary logs).
1639
LOG_INFO_EOF End of log-index-file found
1641
LOG_INFO_SEEK Could not allocate IO cache
1643
LOG_INFO_IO Got IO error while reading file
1647
int DRIZZLE_BIN_LOG::purge_first_log(Relay_log_info* rli, bool included)
1652
assert(rli->slave_running == 1);
1653
assert(!strcmp(rli->linfo.log_file_name,rli->event_relay_log_name.c_str()));
1655
pthread_mutex_lock(&LOCK_index);
1656
pthread_mutex_lock(&rli->log_space_lock);
1657
rli->relay_log.purge_logs(rli->group_relay_log_name.c_str(), included,
1658
0, 0, &rli->log_space_total);
1659
// Tell the I/O thread to take the relay_log_space_limit into account
1660
rli->ignore_log_space_limit= 0;
1661
pthread_mutex_unlock(&rli->log_space_lock);
1664
Ok to broadcast after the critical region as there is no risk of
1665
the mutex being destroyed by this thread later - this helps save
1668
pthread_cond_broadcast(&rli->log_space_cond);
1671
Read the next log file name from the index file and pass it back to
1673
If included is true, we want the first relay log;
1674
otherwise we want the one after event_relay_log_name.
1676
if ((included && (error=find_log_pos(&rli->linfo, NULL, 0))) ||
1678
((error=find_log_pos(&rli->linfo, rli->event_relay_log_name.c_str(), 0)) ||
1679
(error=find_next_log(&rli->linfo, 0)))))
1682
sql_print_error(_("next log error: %d offset: %s log: %s included: %d"),
1684
llstr(rli->linfo.index_file_offset,buff),
1685
rli->group_relay_log_name.c_str(),
1691
Reset rli's coordinates to the current log.
1693
rli->event_relay_log_pos= BIN_LOG_HEADER_SIZE;
1694
rli->event_relay_log_name.assign(rli->linfo.log_file_name);
1697
If we removed the rli->group_relay_log_name file,
1698
we must update the rli->group* coordinates, otherwise do not touch it as the
1699
group's execution is not finished (e.g. COMMIT not executed)
1703
rli->group_relay_log_pos = BIN_LOG_HEADER_SIZE;
1704
rli->group_relay_log_name.assign(rli->linfo.log_file_name);
1705
rli->notify_group_relay_log_name_update();
1708
/* Store where we are in the new file for the execution thread */
1709
flush_relay_log_info(rli);
1712
pthread_mutex_unlock(&LOCK_index);
1717
Update log index_file.
1720
int DRIZZLE_BIN_LOG::update_log_index(LOG_INFO* log_info, bool need_update_threads)
1722
if (copy_up_file_and_fill(&index_file, log_info->index_file_start_offset))
1725
// now update offsets in index file for running threads
1726
if (need_update_threads)
1727
adjust_linfo_offsets(log_info->index_file_start_offset);
1732
Remove all logs before the given log from disk and from the index file.
1734
@param to_log Delete all log file name before this file.
1735
@param included If true, to_log is deleted too.
1737
@param need_update_threads If we want to update the log coordinates of
1738
all threads. False for relay logs, true otherwise.
1739
@param freed_log_space If not null, decrement this variable of
1740
the amount of log space freed
1743
If any of the logs before the deleted one is in use,
1744
only purge logs up to this one.
1749
LOG_INFO_EOF to_log not found
1750
LOG_INFO_EMFILE too many files opened
1751
LOG_INFO_FATAL if any other than ENOENT error from
1752
stat() or my_delete()
1755
int DRIZZLE_BIN_LOG::purge_logs(const char *to_log,
1758
bool need_update_threads,
1759
uint64_t *decrease_log_space)
1767
pthread_mutex_lock(&LOCK_index);
1768
if ((error=find_log_pos(&log_info, to_log, 0 /*no mutex*/)))
1772
File name exists in index file; delete until we find this file
1773
or a file that is used.
1775
if ((error=find_log_pos(&log_info, NULL, 0 /*no mutex*/)))
1777
while ((strcmp(to_log,log_info.log_file_name) || (exit_loop=included)) &&
1778
!log_in_use(log_info.log_file_name))
1781
if (stat(log_info.log_file_name, &s))
1783
if (errno == ENOENT)
1786
It's not fatal if we can't stat a log file that does not exist;
1787
If we could not stat, we won't delete.
1789
push_warning_printf(current_thd, DRIZZLE_ERROR::WARN_LEVEL_WARN,
1790
ER_LOG_PURGE_NO_FILE, ER(ER_LOG_PURGE_NO_FILE),
1791
log_info.log_file_name);
1792
sql_print_information(_("Failed to execute stat() on file '%s'"),
1793
log_info.log_file_name);
1799
Other than ENOENT are fatal
1801
push_warning_printf(current_thd, DRIZZLE_ERROR::WARN_LEVEL_ERROR,
1802
ER_BINLOG_PURGE_FATAL_ERR,
1803
_("a problem with getting info on being purged %s; "
1804
"consider examining correspondence "
1805
"of your binlog index file "
1806
"to the actual binlog files"),
1807
log_info.log_file_name);
1808
error= LOG_INFO_FATAL;
1814
if (!my_delete(log_info.log_file_name, MYF(0)))
1816
if (decrease_log_space)
1817
*decrease_log_space-= s.st_size;
1821
if (my_errno == ENOENT)
1823
push_warning_printf(current_thd, DRIZZLE_ERROR::WARN_LEVEL_WARN,
1824
ER_LOG_PURGE_NO_FILE, ER(ER_LOG_PURGE_NO_FILE),
1825
log_info.log_file_name);
1826
sql_print_information(_("Failed to delete file '%s'"),
1827
log_info.log_file_name);
1832
push_warning_printf(current_thd, DRIZZLE_ERROR::WARN_LEVEL_ERROR,
1833
ER_BINLOG_PURGE_FATAL_ERR,
1834
_("a problem with deleting %s; "
1835
"consider examining correspondence "
1836
"of your binlog index file "
1837
"to the actual binlog files"),
1838
log_info.log_file_name);
1839
if (my_errno == EMFILE)
1841
error= LOG_INFO_EMFILE;
1843
error= LOG_INFO_FATAL;
1849
if (find_next_log(&log_info, 0) || exit_loop)
1854
If we get killed -9 here, the sysadmin would have to edit
1855
the log index file after restart - otherwise, this should be safe
1857
error= update_log_index(&log_info, need_update_threads);
1864
pthread_mutex_unlock(&LOCK_index);
1869
Remove all logs before the given file date from disk and from the
1872
@param thd Thread pointer
1873
@param before_date Delete all log files before given date.
1876
If any of the logs before the deleted one is in use,
1877
only purge logs up to this one.
1882
LOG_INFO_PURGE_NO_ROTATE Binary file that can't be rotated
1883
LOG_INFO_FATAL if any other than ENOENT error from
1884
stat() or my_delete()
1887
int DRIZZLE_BIN_LOG::purge_logs_before_date(time_t purge_time)
1891
struct stat stat_area;
1893
pthread_mutex_lock(&LOCK_index);
1896
Delete until we find curren file
1897
or a file that is used or a file
1898
that is older than purge_time.
1900
if ((error=find_log_pos(&log_info, NULL, 0 /*no mutex*/)))
1903
while (strcmp(log_file_name, log_info.log_file_name) &&
1904
!log_in_use(log_info.log_file_name))
1906
if (stat(log_info.log_file_name, &stat_area))
1908
if (errno == ENOENT)
1911
It's not fatal if we can't stat a log file that does not exist.
1913
push_warning_printf(current_thd, DRIZZLE_ERROR::WARN_LEVEL_WARN,
1914
ER_LOG_PURGE_NO_FILE, ER(ER_LOG_PURGE_NO_FILE),
1915
log_info.log_file_name);
1916
sql_print_information(_("Failed to execute stat() on file '%s'"),
1917
log_info.log_file_name);
1923
Other than ENOENT are fatal
1925
push_warning_printf(current_thd, DRIZZLE_ERROR::WARN_LEVEL_ERROR,
1926
ER_BINLOG_PURGE_FATAL_ERR,
1927
_("a problem with getting info on being purged %s; "
1928
"consider examining correspondence "
1929
"of your binlog index file "
1930
"to the actual binlog files"),
1931
log_info.log_file_name);
1932
error= LOG_INFO_FATAL;
1938
if (stat_area.st_mtime >= purge_time)
1940
if (my_delete(log_info.log_file_name, MYF(0)))
1942
if (my_errno == ENOENT)
1944
/* It's not fatal even if we can't delete a log file */
1945
push_warning_printf(current_thd, DRIZZLE_ERROR::WARN_LEVEL_WARN,
1946
ER_LOG_PURGE_NO_FILE, ER(ER_LOG_PURGE_NO_FILE),
1947
log_info.log_file_name);
1948
sql_print_information(_("Failed to delete file '%s'"),
1949
log_info.log_file_name);
1954
push_warning_printf(current_thd, DRIZZLE_ERROR::WARN_LEVEL_ERROR,
1955
ER_BINLOG_PURGE_FATAL_ERR,
1956
_("a problem with deleting %s; "
1957
"consider examining correspondence "
1958
"of your binlog index file "
1959
"to the actual binlog files"),
1960
log_info.log_file_name);
1961
error= LOG_INFO_FATAL;
1966
if (find_next_log(&log_info, 0))
1971
If we get killed -9 here, the sysadmin would have to edit
1972
the log index file after restart - otherwise, this should be safe
1974
error= update_log_index(&log_info, 1);
1977
pthread_mutex_unlock(&LOCK_index);
1983
Create a new log file name.
1985
@param buf buf of at least FN_REFLEN where new name is stored
1988
If file name will be longer then FN_REFLEN it will be truncated
1991
void DRIZZLE_BIN_LOG::make_log_name(char* buf, const char* log_ident)
1993
uint32_t dir_len = dirname_length(log_file_name);
1994
if (dir_len >= FN_REFLEN)
1995
dir_len=FN_REFLEN-1;
1996
my_stpncpy(buf, log_file_name, dir_len);
1997
strmake(buf+dir_len, log_ident, FN_REFLEN - dir_len -1);
2002
Check if we are writing/reading to the given log file.
2005
bool DRIZZLE_BIN_LOG::is_active(const char *log_file_name_arg)
2007
return !strcmp(log_file_name, log_file_name_arg);
2012
Wrappers around new_file_impl to avoid using argument
2013
to control locking. The argument 1) less readable 2) breaks
2014
incapsulation 3) allows external access to the class without
2015
a lock (which is not possible with private new_file_without_locking
2019
void DRIZZLE_BIN_LOG::new_file()
2025
void DRIZZLE_BIN_LOG::new_file_without_locking()
2032
Start writing to a new log file or reopen the old file.
2034
@param need_lock Set to 1 if caller has not locked LOCK_log
2037
The new file name is stored last in the index file
2040
void DRIZZLE_BIN_LOG::new_file_impl(bool need_lock)
2042
char new_name[FN_REFLEN], *new_name_ptr, *old_name;
2050
pthread_mutex_lock(&LOCK_log);
2051
pthread_mutex_lock(&LOCK_index);
2053
safe_mutex_assert_owner(&LOCK_log);
2054
safe_mutex_assert_owner(&LOCK_index);
2057
if binlog is used as tc log, be sure all xids are "unlogged",
2058
so that on recover we only need to scan one - latest - binlog file
2059
for prepared xids. As this is expected to be a rare event,
2060
simple wait strategy is enough. We're locking LOCK_log to be sure no
2061
new Xid_log_event's are added to the log (and prepared_xids is not
2062
increased), and waiting on COND_prep_xids for late threads to
2067
tc_log_page_waits++;
2068
pthread_mutex_lock(&LOCK_prep_xids);
2069
while (prepared_xids) {
2070
pthread_cond_wait(&COND_prep_xids, &LOCK_prep_xids);
2072
pthread_mutex_unlock(&LOCK_prep_xids);
2075
/* Reuse old name if not binlog and not update log */
2079
If user hasn't specified an extension, generate a new log name
2080
We have to do this here and not in open as we want to store the
2081
new file name in the current binary log file.
2083
if (generate_new_name(new_name, name))
2085
new_name_ptr=new_name;
2087
if (log_type == LOG_BIN)
2089
if (!no_auto_events)
2092
We log the whole file name for log file as the user may decide
2093
to change base names at some point.
2095
Rotate_log_event r(new_name+dirname_length(new_name),
2096
0, LOG_EVENT_OFFSET, 0);
2098
bytes_written += r.data_written;
2101
Update needs to be signalled even if there is no rotate event
2102
log rotation should give the waiting thread a signal to
2103
discover EOF and move on to the next log.
2108
name=0; // Don't free name
2109
close(LOG_CLOSE_TO_BE_OPENED);
2112
Note that at this point, log_state != LOG_CLOSED (important for is_open()).
2116
new_file() is only used for rotation (in FLUSH LOGS or because size >
2117
max_binlog_size or max_relay_log_size).
2118
If this is a binary log, the Format_description_log_event at the beginning of
2119
the new file should have created=0 (to distinguish with the
2120
Format_description_log_event written at server startup, which should
2121
trigger temp tables deletion on slaves.
2124
open(old_name, log_type, new_name_ptr,
2125
io_cache_type, no_auto_events, max_size, 1);
2130
pthread_mutex_unlock(&LOCK_log);
2131
pthread_mutex_unlock(&LOCK_index);
2137
bool DRIZZLE_BIN_LOG::append(Log_event* ev)
2140
pthread_mutex_lock(&LOCK_log);
2142
assert(log_file.type == SEQ_READ_APPEND);
2144
Log_event::write() is smart enough to use my_b_write() or
2145
my_b_append() depending on the kind of cache we have.
2147
if (ev->write(&log_file))
2152
bytes_written+= ev->data_written;
2153
if ((uint) my_b_append_tell(&log_file) > max_size)
2154
new_file_without_locking();
2157
pthread_mutex_unlock(&LOCK_log);
2158
signal_update(); // Safe as we don't call close
2163
bool DRIZZLE_BIN_LOG::appendv(const char* buf, uint32_t len,...)
2169
assert(log_file.type == SEQ_READ_APPEND);
2171
safe_mutex_assert_owner(&LOCK_log);
2174
if (my_b_append(&log_file,(unsigned char*) buf,len))
2179
bytes_written += len;
2180
} while ((buf=va_arg(args,const char*)) && (len=va_arg(args,uint)));
2181
if ((uint) my_b_append_tell(&log_file) > max_size)
2182
new_file_without_locking();
2191
bool DRIZZLE_BIN_LOG::flush_and_sync()
2193
int err=0, fd=log_file.file;
2194
safe_mutex_assert_owner(&LOCK_log);
2195
if (flush_io_cache(&log_file))
2197
if (++sync_binlog_counter >= sync_binlog_period && sync_binlog_period)
2199
sync_binlog_counter= 0;
2200
err=my_sync(fd, MYF(MY_WME));
2205
void DRIZZLE_BIN_LOG::start_union_events(THD *thd, query_id_t query_id_param)
2207
assert(!thd->binlog_evt_union.do_union);
2208
thd->binlog_evt_union.do_union= true;
2209
thd->binlog_evt_union.unioned_events= false;
2210
thd->binlog_evt_union.unioned_events_trans= false;
2211
thd->binlog_evt_union.first_query_id= query_id_param;
2214
void DRIZZLE_BIN_LOG::stop_union_events(THD *thd)
2216
assert(thd->binlog_evt_union.do_union);
2217
thd->binlog_evt_union.do_union= false;
2220
bool DRIZZLE_BIN_LOG::is_query_in_union(THD *thd, query_id_t query_id_param)
2222
return (thd->binlog_evt_union.do_union &&
2223
query_id_param >= thd->binlog_evt_union.first_query_id);
2228
These functions are placed in this file since they need access to
2229
binlog_hton, which has internal linkage.
2232
int THD::binlog_setup_trx_data()
2234
binlog_trx_data *trx_data=
2235
(binlog_trx_data*) thd_get_ha_data(this, binlog_hton);
2238
return(0); // Already set up
2240
trx_data= (binlog_trx_data*) my_malloc(sizeof(binlog_trx_data), MYF(MY_ZEROFILL));
2242
open_cached_file(&trx_data->trans_log, mysql_tmpdir,
2243
LOG_PREFIX, binlog_cache_size, MYF(MY_WME)))
2245
free((unsigned char*)trx_data);
2246
return(1); // Didn't manage to set it up
2248
thd_set_ha_data(this, binlog_hton, trx_data);
2250
trx_data= new (thd_get_ha_data(this, binlog_hton)) binlog_trx_data;
2256
Function to start a statement and optionally a transaction for the
2260
binlog_start_trans_and_stmt()
2264
This function does three things:
2265
- Start a transaction if not in autocommit mode or if a BEGIN
2266
statement has been seen.
2268
- Start a statement transaction to allow us to truncate the binary
2271
- Save the currrent binlog position so that we can roll back the
2272
statement by truncating the transaction log.
2274
We only update the saved position if the old one was undefined,
2275
the reason is that there are some cases (e.g., for CREATE-SELECT)
2276
where the position is saved twice (e.g., both in
2277
select_create::prepare() and THD::binlog_write_table_map()) , but
2278
we should use the first. This means that calls to this function
2279
can be used to start the statement before the first table map
2280
event, to include some extra events.
2284
THD::binlog_start_trans_and_stmt()
2286
binlog_trx_data *trx_data= (binlog_trx_data*) thd_get_ha_data(this, binlog_hton);
2288
if (trx_data == NULL ||
2289
trx_data->before_stmt_pos == MY_OFF_T_UNDEF)
2291
this->binlog_set_stmt_begin();
2292
if (options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN))
2293
trans_register_ha(this, true, binlog_hton);
2294
trans_register_ha(this, false, binlog_hton);
2296
Mark statement transaction as read/write. We never start
2297
a binary log transaction and keep it read-only,
2298
therefore it's best to mark the transaction read/write just
2299
at the same time we start it.
2300
Not necessary to mark the normal transaction read/write
2301
since the statement-level flag will be propagated automatically
2302
inside ha_commit_trans.
2304
ha_data[binlog_hton->slot].ha_info[0].set_trx_read_write();
2309
void THD::binlog_set_stmt_begin() {
2310
binlog_trx_data *trx_data=
2311
(binlog_trx_data*) thd_get_ha_data(this, binlog_hton);
2314
The call to binlog_trans_log_savepos() might create the trx_data
2315
structure, if it didn't exist before, so we save the position
2316
into an auto variable and then write it into the transaction
2317
data for the binary log (i.e., trx_data).
2320
binlog_trans_log_savepos(this, &pos);
2321
trx_data= (binlog_trx_data*) thd_get_ha_data(this, binlog_hton);
2322
trx_data->before_stmt_pos= pos;
2327
Write a table map to the binary log.
2330
int THD::binlog_write_table_map(Table *table, bool is_trans)
2334
/* Pre-conditions */
2335
assert(current_stmt_binlog_row_based && mysql_bin_log.is_open());
2336
assert(table->s->table_map_id != UINT32_MAX);
2338
Table_map_log_event::flag_set const
2339
flags= Table_map_log_event::TM_NO_FLAGS;
2342
the_event(this, table, table->s->table_map_id, is_trans, flags);
2344
if (is_trans && binlog_table_maps == 0)
2345
binlog_start_trans_and_stmt();
2347
if ((error= mysql_bin_log.write(&the_event)))
2350
binlog_table_maps++;
2351
table->s->table_map_version= mysql_bin_log.table_map_version();
2356
THD::binlog_get_pending_rows_event() const
2358
binlog_trx_data *const trx_data=
2359
(binlog_trx_data*) thd_get_ha_data(this, binlog_hton);
2361
This is less than ideal, but here's the story: If there is no
2362
trx_data, prepare_pending_rows_event() has never been called
2363
(since the trx_data is set up there). In that case, we just return
2366
return trx_data ? trx_data->pending() : NULL;
2370
THD::binlog_set_pending_rows_event(Rows_log_event* ev)
2372
if (thd_get_ha_data(this, binlog_hton) == NULL)
2373
binlog_setup_trx_data();
2375
binlog_trx_data *const trx_data=
2376
(binlog_trx_data*) thd_get_ha_data(this, binlog_hton);
2379
trx_data->set_pending(ev);
2384
Moves the last bunch of rows from the pending Rows event to the binlog
2385
(either cached binlog if transaction, or disk binlog). Sets a new pending
2389
DRIZZLE_BIN_LOG::flush_and_set_pending_rows_event(THD *thd,
2390
Rows_log_event* event)
2392
assert(mysql_bin_log.is_open());
2396
binlog_trx_data *const trx_data=
2397
(binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
2401
if (Rows_log_event* pending= trx_data->pending())
2403
IO_CACHE *file= &log_file;
2406
Decide if we should write to the log file directly or to the
2409
if (pending->get_cache_stmt() || my_b_tell(&trx_data->trans_log))
2410
file= &trx_data->trans_log;
2413
If we are writing to the log file directly, we could avoid
2414
locking the log. This does not work since we need to step the
2415
m_table_map_version below, and that change has to be protected
2416
by the LOCK_log mutex.
2418
pthread_mutex_lock(&LOCK_log);
2421
Write pending event to log file or transaction cache
2423
if (pending->write(file))
2425
pthread_mutex_unlock(&LOCK_log);
2430
We step the table map version if we are writing an event
2431
representing the end of a statement. We do this regardless of
2432
wheather we write to the transaction cache or to directly to the
2435
In an ideal world, we could avoid stepping the table map version
2436
if we were writing to a transaction cache, since we could then
2437
reuse the table map that was written earlier in the transaction
2438
cache. This does not work since STMT_END_F implies closing all
2439
table mappings on the slave side.
2441
TODO: Find a solution so that table maps does not have to be
2442
written several times within a transaction.
2444
if (pending->get_flags(Rows_log_event::STMT_END_F))
2445
++m_table_map_version;
2449
if (file == &log_file)
2451
error= flush_and_sync();
2455
rotate_and_purge(RP_LOCK_LOG_IS_ALREADY_LOCKED);
2459
pthread_mutex_unlock(&LOCK_log);
2462
thd->binlog_set_pending_rows_event(event);
2468
Write an event to the binary log.
2471
bool DRIZZLE_BIN_LOG::write(Log_event *event_info)
2473
THD *thd= event_info->thd;
2476
if (thd->binlog_evt_union.do_union)
2479
In Stored function; Remember that function call caused an update.
2480
We will log the function call to the binary log on function exit
2482
thd->binlog_evt_union.unioned_events= true;
2483
thd->binlog_evt_union.unioned_events_trans |= event_info->cache_stmt;
2488
Flush the pending rows event to the transaction cache or to the
2489
log file. Since this function potentially aquire the LOCK_log
2490
mutex, we do this before aquiring the LOCK_log mutex in this
2493
We only end the statement if we are in a top-level statement. If
2494
we are inside a stored function, we do not end the statement since
2495
this will close all tables on the slave.
2497
bool const end_stmt= false;
2498
thd->binlog_flush_pending_rows_event(end_stmt);
2500
pthread_mutex_lock(&LOCK_log);
2503
In most cases this is only called if 'is_open()' is true; in fact this is
2504
mostly called if is_open() *was* true a few instructions before, but it
2505
could have changed since.
2507
if (likely(is_open()))
2509
IO_CACHE *file= &log_file;
2511
In the future we need to add to the following if tests like
2512
"do the involved tables match (to be implemented)
2513
binlog_[wild_]{do|ignore}_table?" (WL#1049)"
2515
const char *local_db= event_info->get_db();
2516
if ((thd && !(thd->options & OPTION_BIN_LOG)) ||
2517
(!binlog_filter->db_ok(local_db)))
2519
pthread_mutex_unlock(&LOCK_log);
2524
Should we write to the binlog cache or to the binlog on disk?
2525
Write to the binlog cache if:
2526
- it is already not empty (meaning we're in a transaction; note that the
2527
present event could be about a non-transactional table, but still we need
2528
to write to the binlog cache in that case to handle updates to mixed
2529
trans/non-trans table types the best possible in binlogging)
2530
- or if the event asks for it (cache_stmt == TRUE).
2532
if (opt_using_transactions && thd)
2534
if (thd->binlog_setup_trx_data())
2537
binlog_trx_data *const trx_data=
2538
(binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
2539
IO_CACHE *trans_log= &trx_data->trans_log;
2540
my_off_t trans_log_pos= my_b_tell(trans_log);
2541
if (event_info->get_cache_stmt() || trans_log_pos != 0)
2543
if (trans_log_pos == 0)
2544
thd->binlog_start_trans_and_stmt();
2548
TODO as Mats suggested, for all the cases above where we write to
2549
trans_log, it sounds unnecessary to lock LOCK_log. We should rather
2550
test first if we want to write to trans_log, and if not, lock
2556
No check for auto events flag here - this write method should
2557
never be called if auto-events are enabled
2561
1. Write first log events which describe the 'run environment'
2566
If row-based binlogging, Insert_id, Rand and other kind of "setting
2567
context" events are not needed.
2571
if (!thd->current_stmt_binlog_row_based)
2573
if (thd->stmt_depends_on_first_successful_insert_id_in_prev_stmt)
2575
Intvar_log_event e(thd,(unsigned char) LAST_INSERT_ID_EVENT,
2576
thd->first_successful_insert_id_in_prev_stmt_for_binlog);
2580
if (thd->auto_inc_intervals_in_cur_stmt_for_binlog.nb_elements() > 0)
2583
If the auto_increment was second in a table's index (possible with
2584
MyISAM or BDB) (table->next_number_keypart != 0), such event is
2585
in fact not necessary. We could avoid logging it.
2587
Intvar_log_event e(thd, (unsigned char) INSERT_ID_EVENT,
2588
thd->auto_inc_intervals_in_cur_stmt_for_binlog.
2595
Rand_log_event e(thd,thd->rand_saved_seed1,thd->rand_saved_seed2);
2599
if (thd->user_var_events.elements)
2601
for (uint32_t i= 0; i < thd->user_var_events.elements; i++)
2603
BINLOG_USER_VAR_EVENT *user_var_event;
2604
get_dynamic(&thd->user_var_events,(unsigned char*) &user_var_event, i);
2605
User_var_log_event e(thd, user_var_event->user_var_event->name.str,
2606
user_var_event->user_var_event->name.length,
2607
user_var_event->value,
2608
user_var_event->length,
2609
user_var_event->type,
2610
user_var_event->charset_number);
2619
Write the SQL command
2622
if (event_info->write(file))
2625
if (file == &log_file) // we are writing to the real log (disk)
2627
if (flush_and_sync())
2630
rotate_and_purge(RP_LOCK_LOG_IS_ALREADY_LOCKED);
2637
if (my_errno == EFBIG)
2638
my_message(ER_TRANS_CACHE_FULL, ER(ER_TRANS_CACHE_FULL), MYF(0));
2640
my_error(ER_ERROR_ON_WRITE, MYF(0), name, errno);
2645
if (event_info->flags & LOG_EVENT_UPDATE_TABLE_MAP_VERSION_F)
2646
++m_table_map_version;
2648
pthread_mutex_unlock(&LOCK_log);
2653
int error_log_print(enum loglevel level, const char *format,
2656
return logger.error_log_print(level, format, args);
2659
void DRIZZLE_BIN_LOG::rotate_and_purge(uint32_t flags)
2661
if (!(flags & RP_LOCK_LOG_IS_ALREADY_LOCKED))
2662
pthread_mutex_lock(&LOCK_log);
2663
if ((flags & RP_FORCE_ROTATE) ||
2664
(my_b_tell(&log_file) >= (my_off_t) max_size))
2666
new_file_without_locking();
2667
if (expire_logs_days)
2669
time_t purge_time= my_time(0) - expire_logs_days*24*60*60;
2670
if (purge_time >= 0)
2671
purge_logs_before_date(purge_time);
2674
if (!(flags & RP_LOCK_LOG_IS_ALREADY_LOCKED))
2675
pthread_mutex_unlock(&LOCK_log);
2678
uint32_t DRIZZLE_BIN_LOG::next_file_id()
2681
pthread_mutex_lock(&LOCK_log);
2683
pthread_mutex_unlock(&LOCK_log);
2689
Write the contents of a cache to the binary log.
2693
cache Cache to write to the binary log
2694
lock_log True if the LOCK_log mutex should be aquired, false otherwise
2695
sync_log True if the log should be flushed and sync:ed
2698
Write the contents of the cache to the binary log. The cache will
2699
be reset as a READ_CACHE to be able to read the contents from it.
2702
int DRIZZLE_BIN_LOG::write_cache(IO_CACHE *cache, bool lock_log, bool sync_log)
2704
Mutex_sentry sentry(lock_log ? &LOCK_log : NULL);
2706
if (reinit_io_cache(cache, READ_CACHE, 0, 0, 0))
2707
return ER_ERROR_ON_WRITE;
2708
uint32_t length= my_b_bytes_in_cache(cache), group, carry, hdr_offs;
2710
unsigned char header[LOG_EVENT_HEADER_LEN];
2713
The events in the buffer have incorrect end_log_pos data
2714
(relative to beginning of group rather than absolute),
2715
so we'll recalculate them in situ so the binlog is always
2716
correct, even in the middle of a group. This is possible
2717
because we now know the start position of the group (the
2718
offset of this cache in the log, if you will); all we need
2719
to do is to find all event-headers, and add the position of
2720
the group to the end_log_pos of each event. This is pretty
2721
straight forward, except that we read the cache in segments,
2722
so an event-header might end up on the cache-border and get
2726
group= (uint)my_b_tell(&log_file);
2733
if we only got a partial header in the last iteration,
2734
get the other half now and process a full header.
2736
if (unlikely(carry > 0))
2738
assert(carry < LOG_EVENT_HEADER_LEN);
2740
/* assemble both halves */
2741
memcpy(&header[carry], cache->read_pos, LOG_EVENT_HEADER_LEN - carry);
2743
/* fix end_log_pos */
2744
val= uint4korr(&header[LOG_POS_OFFSET]) + group;
2745
int4store(&header[LOG_POS_OFFSET], val);
2747
/* write the first half of the split header */
2748
if (my_b_write(&log_file, header, carry))
2749
return ER_ERROR_ON_WRITE;
2752
copy fixed second half of header to cache so the correct
2753
version will be written later.
2755
memcpy(cache->read_pos, &header[carry], LOG_EVENT_HEADER_LEN - carry);
2757
/* next event header at ... */
2758
hdr_offs = uint4korr(&header[EVENT_LEN_OFFSET]) - carry;
2763
/* if there is anything to write, process it. */
2765
if (likely(length > 0))
2768
process all event-headers in this (partial) cache.
2769
if next header is beyond current read-buffer,
2770
we'll get it later (though not necessarily in the
2771
very next iteration, just "eventually").
2774
while (hdr_offs < length)
2777
partial header only? save what we can get, process once
2781
if (hdr_offs + LOG_EVENT_HEADER_LEN > length)
2783
carry= length - hdr_offs;
2784
memcpy(header, cache->read_pos + hdr_offs, carry);
2789
/* we've got a full event-header, and it came in one piece */
2791
unsigned char *log_pos= (unsigned char *)cache->read_pos + hdr_offs + LOG_POS_OFFSET;
2793
/* fix end_log_pos */
2794
val= uint4korr(log_pos) + group;
2795
int4store(log_pos, val);
2797
/* next event header at ... */
2798
log_pos= (unsigned char *)cache->read_pos + hdr_offs + EVENT_LEN_OFFSET;
2799
hdr_offs += uint4korr(log_pos);
2805
Adjust hdr_offs. Note that it may still point beyond the segment
2806
read in the next iteration; if the current event is very long,
2807
it may take a couple of read-iterations (and subsequent adjustments
2808
of hdr_offs) for it to point into the then-current segment.
2809
If we have a split header (!carry), hdr_offs will be set at the
2810
beginning of the next iteration, overwriting the value we set here:
2815
/* Write data to the binary log file */
2816
if (my_b_write(&log_file, cache->read_pos, length))
2817
return ER_ERROR_ON_WRITE;
2818
cache->read_pos=cache->read_end; // Mark buffer used up
2819
} while ((length= my_b_fill(cache)));
2830
Write a cached log entry to the binary log.
2831
- To support transaction over replication, we wrap the transaction
2832
with BEGIN/COMMIT or BEGIN/ROLLBACK in the binary log.
2833
We want to write a BEGIN/ROLLBACK block when a non-transactional table
2834
was updated in a transaction which was rolled back. This is to ensure
2835
that the same updates are run on the slave.
2838
@param cache The cache to copy to the binlog
2839
@param commit_event The commit event to print after writing the
2840
contents of the cache.
2843
We only come here if there is something in the cache.
2845
The thing in the cache is always a complete transaction.
2847
'cache' needs to be reinitialized after this functions returns.
2850
bool DRIZZLE_BIN_LOG::write(THD *thd, IO_CACHE *cache, Log_event *commit_event)
2852
pthread_mutex_lock(&LOCK_log);
2854
/* NULL would represent nothing to replicate after ROLLBACK */
2855
assert(commit_event != NULL);
2858
if (likely(is_open())) // Should always be true
2861
We only bother to write to the binary log if there is anything
2864
if (my_b_tell(cache) > 0)
2867
Log "BEGIN" at the beginning of every transaction. Here, a
2868
transaction is either a BEGIN..COMMIT block or a single
2869
statement in autocommit mode.
2871
Query_log_event qinfo(thd, STRING_WITH_LEN("BEGIN"), true, false);
2873
Imagine this is rollback due to net timeout, after all
2874
statements of the transaction succeeded. Then we want a
2875
zero-error code in BEGIN. In other words, if there was a
2876
really serious error code it's already in the statement's
2877
events, there is no need to put it also in this internally
2878
generated event, and as this event is generated late it would
2879
lead to false alarms.
2881
This is safer than thd->clear_error() against kills at shutdown.
2883
qinfo.error_code= 0;
2885
Now this Query_log_event has artificial log_pos 0. It must be
2886
adjusted to reflect the real position in the log. Not doing it
2887
would confuse the slave: it would prevent this one from
2888
knowing where he is in the master's binlog, which would result
2889
in wrong positions being shown to the user, MASTER_POS_WAIT
2892
if (qinfo.write(&log_file))
2895
if ((write_error= write_cache(cache, false, false)))
2898
if (commit_event && commit_event->write(&log_file))
2900
if (flush_and_sync())
2902
if (cache->error) // Error on read
2904
sql_print_error(ER(ER_ERROR_ON_READ), cache->file_name, errno);
2905
write_error=1; // Don't give more errors
2912
if commit_event is Xid_log_event, increase the number of
2913
prepared_xids (it's decreasd in ::unlog()). Binlog cannot be rotated
2914
if there're prepared xids in it - see the comment in new_file() for
2916
If the commit_event is not Xid_log_event (then it's a Query_log_event)
2917
rotate binlog, if necessary.
2919
if (commit_event && commit_event->get_type_code() == XID_EVENT)
2921
pthread_mutex_lock(&LOCK_prep_xids);
2923
pthread_mutex_unlock(&LOCK_prep_xids);
2926
rotate_and_purge(RP_LOCK_LOG_IS_ALREADY_LOCKED);
2928
pthread_mutex_unlock(&LOCK_log);
2936
sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
2938
pthread_mutex_unlock(&LOCK_log);
2944
Wait until we get a signal that the relay log has been updated
2946
@param[in] thd a THD struct
2948
LOCK_log must be taken before calling this function.
2949
It will be released at the end of the function.
2952
void DRIZZLE_BIN_LOG::wait_for_update_relay_log(THD* thd)
2954
const char *old_msg;
2955
old_msg= thd->enter_cond(&update_cond, &LOCK_log,
2956
"Slave has read all relay log; "
2957
"waiting for the slave I/O "
2958
"thread to update it" );
2959
pthread_cond_wait(&update_cond, &LOCK_log);
2960
thd->exit_cond(old_msg);
2966
Wait until we get a signal that the binary log has been updated.
2967
Applies to master only.
2970
@param[in] thd a THD struct
2971
@param[in] timeout a pointer to a timespec;
2972
NULL means to wait w/o timeout.
2973
@retval 0 if got signalled on update
2974
@retval non-0 if wait timeout elapsed
2976
LOCK_log must be taken before calling this function.
2977
LOCK_log is being released while the thread is waiting.
2978
LOCK_log is released by the caller.
2981
int DRIZZLE_BIN_LOG::wait_for_update_bin_log(THD* thd,
2982
const struct timespec *timeout)
2985
const char* old_msg = thd->get_proc_info();
2986
old_msg= thd->enter_cond(&update_cond, &LOCK_log,
2987
"Master has sent all binlog to slave; "
2988
"waiting for binlog to be updated");
2990
pthread_cond_wait(&update_cond, &LOCK_log);
2992
ret= pthread_cond_timedwait(&update_cond, &LOCK_log,
2993
const_cast<struct timespec *>(timeout));
3001
@param exiting Bitmask for one or more of the following bits:
3002
- LOG_CLOSE_INDEX : if we should close the index file
3003
- LOG_CLOSE_TO_BE_OPENED : if we intend to call open
3004
at once after close.
3005
- LOG_CLOSE_STOP_EVENT : write a 'stop' event to the log
3008
One can do an open on the object at once after doing a close.
3009
The internal structures are not freed until cleanup() is called
3012
void DRIZZLE_BIN_LOG::close(uint32_t exiting)
3013
{ // One can't set log_type here!
3014
if (log_state == LOG_OPENED)
3016
if (log_type == LOG_BIN && !no_auto_events &&
3017
(exiting & LOG_CLOSE_STOP_EVENT))
3021
bytes_written+= s.data_written;
3025
/* don't pwrite in a file opened with O_APPEND - it doesn't work */
3026
if (log_file.type == WRITE_CACHE && log_type == LOG_BIN)
3028
my_off_t offset= BIN_LOG_HEADER_SIZE + FLAGS_OFFSET;
3029
unsigned char flags= 0; // clearing LOG_EVENT_BINLOG_IN_USE_F
3030
assert(pwrite(log_file.file, &flags, 1, offset)==1);
3033
/* this will cleanup IO_CACHE, sync and close the file */
3034
DRIZZLE_LOG::close(exiting);
3038
The following test is needed even if is_open() is not set, as we may have
3039
called a not complete close earlier and the index file is still open.
3042
if ((exiting & LOG_CLOSE_INDEX) && my_b_inited(&index_file))
3044
end_io_cache(&index_file);
3045
if (my_close(index_file.file, MYF(0)) < 0 && ! write_error)
3048
sql_print_error(ER(ER_ERROR_ON_WRITE), index_file_name, errno);
3051
log_state= (exiting & LOG_CLOSE_TO_BE_OPENED) ? LOG_TO_BE_OPENED : LOG_CLOSED;
3061
void DRIZZLE_BIN_LOG::set_max_size(ulong max_size_arg)
3064
We need to take locks, otherwise this may happen:
3065
new_file() is called, calls open(old_max_size), then before open() starts,
3066
set_max_size() sets max_size to max_size_arg, then open() starts and
3067
uses the old_max_size argument, so max_size_arg has been overwritten and
3068
it's like if the SET command was never run.
3070
pthread_mutex_lock(&LOCK_log);
3072
max_size= max_size_arg;
3073
pthread_mutex_unlock(&LOCK_log);
3079
Check if a string is a valid number.
3081
@param str String to test
3082
@param res Store value here
3083
@param allow_wildcards Set to 1 if we should ignore '%' and '_'
3086
For the moment the allow_wildcards argument is not used
3087
Should be move to some other file.
3090
1 String is a number
3095
static bool test_if_number(register const char *str,
3096
long *res, bool allow_wildcards)
3103
while (*str++ == ' ') ;
3104
if (*--str == '-' || *str == '+')
3106
while (my_isdigit(files_charset_info,*str) ||
3107
(allow_wildcards && (*str == wild_many || *str == wild_one)))
3115
my_isdigit(files_charset_info,*str) ||
3116
(allow_wildcards && (*str == wild_many || *str == wild_one)) ;
3119
if (*str != 0 || flag == 0)
3123
return(1); /* Number ok */
3124
} /* test_if_number */
3127
void sql_perror(const char *message)
3129
sql_print_error("%s: %s",message, strerror(errno));
3133
bool flush_error_log()
3138
char err_renamed[FN_REFLEN], *end;
3139
end= strmake(err_renamed,log_error_file,FN_REFLEN-4);
3140
my_stpcpy(end, "-old");
3141
pthread_mutex_lock(&LOCK_error_log);
3142
char err_temp[FN_REFLEN+4];
3144
On Windows is necessary a temporary file for to rename
3145
the current error file.
3147
strxmov(err_temp, err_renamed,"-tmp",NULL);
3148
(void) my_delete(err_temp, MYF(0));
3149
if (freopen(err_temp,"a+",stdout))
3153
unsigned char buf[IO_SIZE];
3155
if(freopen(err_temp,"a+",stderr)==NULL)
3157
(void) my_delete(err_renamed, MYF(0));
3158
my_rename(log_error_file,err_renamed,MYF(0));
3159
if (freopen(log_error_file,"a+",stdout)==NULL)
3162
if(freopen(log_error_file,"a+",stderr)==NULL)
3165
if ((fd = my_open(err_temp, O_RDONLY, MYF(0))) >= 0)
3167
while ((bytes= my_read(fd, buf, IO_SIZE, MYF(0))) &&
3168
bytes != MY_FILE_ERROR)
3169
my_fwrite(stderr, buf, bytes, MYF(0));
3170
my_close(fd, MYF(0));
3172
(void) my_delete(err_temp, MYF(0));
3176
pthread_mutex_unlock(&LOCK_error_log);
3181
void DRIZZLE_BIN_LOG::signal_update()
3183
pthread_cond_broadcast(&update_cond);
3188
Prints a printf style message to the error log and, under NT, to the
3191
This function prints the message into a buffer and then sends that buffer
3192
to other functions to write that message to other logging sources.
3194
@param event_type Type of event to write (Error, Warning, or Info)
3195
@param format Printf style format of message
3196
@param args va_list list of arguments for the message
3199
The function always returns 0. The return value is present in the
3200
signature to be compatible with other logging routines, which could
3201
return an error (e.g. logging to the log tables)
3203
static void print_buffer_to_file(enum loglevel level,
3204
int error_code __attribute__((unused)),
3206
size_t buffer_length __attribute__((unused)))
3212
pthread_mutex_lock(&LOCK_error_log);
3215
localtime_r(&skr, &tm_tmp);
3218
fprintf(stderr, "%02d%02d%02d %2d:%02d:%02d [%s] %s\n",
3219
start->tm_year % 100,
3225
(level == ERROR_LEVEL ? "ERROR" : level == WARNING_LEVEL ?
3226
"Warning" : "Note"),
3231
pthread_mutex_unlock(&LOCK_error_log);
3236
int vprint_msg_to_log(enum loglevel level, const char *format, va_list args)
3240
int error_code= errno;
3242
length= vsnprintf(buff, sizeof(buff), format, args);
3244
print_buffer_to_file(level, error_code, buff, length);
3250
void sql_print_error(const char *format, ...)
3254
va_start(args, format);
3255
error_log_print(ERROR_LEVEL, format, args);
3262
void sql_print_warning(const char *format, ...)
3266
va_start(args, format);
3267
error_log_print(WARNING_LEVEL, format, args);
3274
void sql_print_information(const char *format, ...)
3278
va_start(args, format);
3279
error_log_print(INFORMATION_LEVEL, format, args);
3286
/********* transaction coordinator log for 2pc - mmap() based solution *******/
3289
the log consists of a file, mmapped to a memory.
3290
file is divided on pages of tc_log_page_size size.
3291
(usable size of the first page is smaller because of log header)
3292
there's PAGE control structure for each page
3293
each page (or rather PAGE control structure) can be in one of three
3294
states - active, syncing, pool.
3295
there could be only one page in active or syncing states,
3296
but many in pool - pool is fifo queue.
3297
usual lifecycle of a page is pool->active->syncing->pool
3298
"active" page - is a page where new xid's are logged.
3299
the page stays active as long as syncing slot is taken.
3300
"syncing" page is being synced to disk. no new xid can be added to it.
3301
when the sync is done the page is moved to a pool and an active page
3304
the result of such an architecture is a natural "commit grouping" -
3305
If commits are coming faster than the system can sync, they do not
3306
stall. Instead, all commit that came since the last sync are
3307
logged to the same page, and they all are synced with the next -
3308
one - sync. Thus, thought individual commits are delayed, throughput
3311
when a xid is added to an active page, the thread of this xid waits
3312
for a page's condition until the page is synced. when syncing slot
3313
becomes vacant one of these waiters is awaken to take care of syncing.
3314
it syncs the page and signals all waiters that the page is synced.
3315
PAGE::waiters is used to count these waiters, and a page may never
3316
become active again until waiters==0 (that is all waiters from the
3317
previous sync have noticed the sync was completed)
3319
note, that the page becomes "dirty" and has to be synced only when a
3320
new xid is added into it. Removing a xid from a page does not make it
3321
dirty - we don't sync removals to disk.
3324
ulong tc_log_page_waits= 0;
3328
#define TC_LOG_HEADER_SIZE (sizeof(tc_log_magic)+1)
3330
static const char tc_log_magic[]={(char) 254, 0x23, 0x05, 0x74};
3332
ulong opt_tc_log_size= TC_LOG_MIN_SIZE;
3333
ulong tc_log_max_pages_used=0, tc_log_page_size=0, tc_log_cur_pages_used=0;
3335
int TC_LOG_MMAP::open(const char *opt_name)
3338
bool crashed= false;
3341
assert(total_ha_2pc > 1);
3342
assert(opt_name && opt_name[0]);
3344
tc_log_page_size= getpagesize();
3345
assert(TC_LOG_PAGE_SIZE % tc_log_page_size == 0);
3347
fn_format(logname,opt_name,mysql_data_home,"",MY_UNPACK_FILENAME);
3348
if ((fd= my_open(logname, O_RDWR, MYF(0))) < 0)
3350
if (my_errno != ENOENT)
3352
if (using_heuristic_recover())
3354
if ((fd= my_create(logname, CREATE_MODE, O_RDWR, MYF(MY_WME))) < 0)
3357
file_length= opt_tc_log_size;
3358
if (ftruncate(fd, file_length))
3365
sql_print_information(_("Recovering after a crash using %s"), opt_name);
3366
if (tc_heuristic_recover)
3368
sql_print_error(_("Cannot perform automatic crash recovery when "
3369
"--tc-heuristic-recover is used"));
3372
file_length= my_seek(fd, 0L, MY_SEEK_END, MYF(MY_WME+MY_FAE));
3373
if (file_length == MY_FILEPOS_ERROR || file_length % tc_log_page_size)
3377
data= (unsigned char *)my_mmap(0, (size_t)file_length, PROT_READ|PROT_WRITE,
3378
MAP_NOSYNC|MAP_SHARED, fd, 0);
3379
if (data == MAP_FAILED)
3386
npages=(uint)file_length/tc_log_page_size;
3387
assert(npages >= 3); // to guarantee non-empty pool
3388
if (!(pages=(PAGE *)my_malloc(npages*sizeof(PAGE), MYF(MY_WME|MY_ZEROFILL))))
3391
for (pg=pages, i=0; i < npages; i++, pg++)
3396
pthread_mutex_init(&pg->lock, MY_MUTEX_INIT_FAST);
3397
pthread_cond_init (&pg->cond, 0);
3398
pg->start=(my_xid *)(data + i*tc_log_page_size);
3400
pg->end=(my_xid *)(pg->start + tc_log_page_size);
3401
pg->size=pg->free=tc_log_page_size/sizeof(my_xid);
3403
pages[0].size=pages[0].free=
3404
(tc_log_page_size-TC_LOG_HEADER_SIZE)/sizeof(my_xid);
3405
pages[0].start=pages[0].end-pages[0].size;
3406
pages[npages-1].next=0;
3409
if (crashed && recover())
3412
memcpy(data, tc_log_magic, sizeof(tc_log_magic));
3413
data[sizeof(tc_log_magic)]= (unsigned char)total_ha_2pc;
3414
msync(data, tc_log_page_size, MS_SYNC);
3415
my_sync(fd, MYF(0));
3418
pthread_mutex_init(&LOCK_sync, MY_MUTEX_INIT_FAST);
3419
pthread_mutex_init(&LOCK_active, MY_MUTEX_INIT_FAST);
3420
pthread_mutex_init(&LOCK_pool, MY_MUTEX_INIT_FAST);
3421
pthread_cond_init(&COND_active, 0);
3422
pthread_cond_init(&COND_pool, 0);
3429
pool_last=pages+npages-1;
3439
there is no active page, let's got one from the pool.
3441
Two strategies here:
3442
-# take the first from the pool
3443
-# if there're waiters - take the one with the most free space.
3446
TODO page merging. try to allocate adjacent page first,
3447
so that they can be flushed both in one sync
3450
void TC_LOG_MMAP::get_active_from_pool()
3452
PAGE **p, **best_p=0;
3456
pthread_mutex_lock(&LOCK_pool);
3461
if ((*p)->waiters == 0) // can the first page be used ?
3462
break; // yes - take it.
3464
best_free=0; // no - trying second strategy
3465
for (p=&(*p)->next; *p; p=&(*p)->next)
3467
if ((*p)->waiters == 0 && (*p)->free > best_free)
3469
best_free=(*p)->free;
3474
while ((*best_p == 0 || best_free == 0) && overflow());
3477
if (active->free == active->size) // we've chosen an empty page
3479
tc_log_cur_pages_used++;
3480
set_if_bigger(tc_log_max_pages_used, tc_log_cur_pages_used);
3483
if ((*best_p)->next) // unlink the page from the pool
3484
*best_p=(*best_p)->next;
3489
pthread_mutex_unlock(&LOCK_pool);
3494
perhaps, increase log size ?
3496
int TC_LOG_MMAP::overflow()
3499
simple overflow handling - just wait
3500
TODO perhaps, increase log size ?
3501
let's check the behaviour of tc_log_page_waits first
3503
tc_log_page_waits++;
3504
pthread_cond_wait(&COND_pool, &LOCK_pool);
3505
return 1; // always return 1
3509
Record that transaction XID is committed on the persistent storage.
3511
This function is called in the middle of two-phase commit:
3512
First all resources prepare the transaction, then tc_log->log() is called,
3513
then all resources commit the transaction, then tc_log->unlog() is called.
3515
All access to active page is serialized but it's not a problem, as
3516
we're assuming that fsync() will be a main bottleneck.
3517
That is, parallelizing writes to log pages we'll decrease number of
3518
threads waiting for a page, but then all these threads will be waiting
3519
for a fsync() anyway
3521
If tc_log == DRIZZLE_LOG then tc_log writes transaction to binlog and
3522
records XID in a special Xid_log_event.
3523
If tc_log = TC_LOG_MMAP then xid is written in a special memory-mapped
3529
\# - otherwise, "cookie", a number that will be passed as an argument
3530
to unlog() call. tc_log can define it any way it wants,
3531
and use for whatever purposes. TC_LOG_MMAP sets it
3532
to the position in memory where xid was logged to.
3535
int TC_LOG_MMAP::log_xid(THD *thd __attribute__((unused)), my_xid xid)
3541
pthread_mutex_lock(&LOCK_active);
3544
if active page is full - just wait...
3545
frankly speaking, active->free here accessed outside of mutex
3546
protection, but it's safe, because it only means we may miss an
3547
unlog() for the active page, and we're not waiting for it here -
3548
unlog() does not signal COND_active.
3550
while (unlikely(active && active->free == 0))
3551
pthread_cond_wait(&COND_active, &LOCK_active);
3553
/* no active page ? take one from the pool */
3555
get_active_from_pool();
3558
pthread_mutex_lock(&p->lock);
3560
/* searching for an empty slot */
3564
assert(p->ptr < p->end); // because p->free > 0
3567
/* found! store xid there and mark the page dirty */
3568
cookie= (ulong)((unsigned char *)p->ptr - data); // can never be zero
3573
/* to sync or not to sync - this is the question */
3574
pthread_mutex_unlock(&LOCK_active);
3575
pthread_mutex_lock(&LOCK_sync);
3576
pthread_mutex_unlock(&p->lock);
3579
{ // somebody's syncing. let's wait
3582
note - it must be while (), not do ... while () here
3583
as p->state may be not DIRTY when we come here
3585
while (p->state == DIRTY && syncing)
3586
pthread_cond_wait(&p->cond, &LOCK_sync);
3588
err= p->state == ERROR;
3589
if (p->state != DIRTY) // page was synced
3591
if (p->waiters == 0)
3592
pthread_cond_signal(&COND_pool); // in case somebody's waiting
3593
pthread_mutex_unlock(&LOCK_sync);
3594
goto done; // we're done
3596
} // page was not synced! do it now
3597
assert(active == p && syncing == 0);
3598
pthread_mutex_lock(&LOCK_active);
3599
syncing=p; // place is vacant - take it
3600
active=0; // page is not active anymore
3601
pthread_cond_broadcast(&COND_active); // in case somebody's waiting
3602
pthread_mutex_unlock(&LOCK_active);
3603
pthread_mutex_unlock(&LOCK_sync);
3607
return err ? 0 : cookie;
3610
int TC_LOG_MMAP::sync()
3614
assert(syncing != active);
3617
sit down and relax - this can take a while...
3618
note - no locks are held at this point
3620
err= msync(syncing->start, 1, MS_SYNC);
3622
err= my_sync(fd, MYF(0));
3624
/* page is synced. let's move it to the pool */
3625
pthread_mutex_lock(&LOCK_pool);
3626
pool_last->next=syncing;
3629
syncing->state= err ? ERROR : POOL;
3630
pthread_cond_broadcast(&syncing->cond); // signal "sync done"
3631
pthread_cond_signal(&COND_pool); // in case somebody's waiting
3632
pthread_mutex_unlock(&LOCK_pool);
3634
/* marking 'syncing' slot free */
3635
pthread_mutex_lock(&LOCK_sync);
3637
pthread_cond_signal(&active->cond); // wake up a new syncer
3638
pthread_mutex_unlock(&LOCK_sync);
3643
erase xid from the page, update page free space counters/pointers.
3644
cookie points directly to the memory where xid was logged.
3647
void TC_LOG_MMAP::unlog(ulong cookie, my_xid xid __attribute__((unused)))
3649
PAGE *p=pages+(cookie/tc_log_page_size);
3650
my_xid *x=(my_xid *)(data+cookie);
3653
assert(x >= p->start && x < p->end);
3656
pthread_mutex_lock(&p->lock);
3658
assert(p->free <= p->size);
3659
set_if_smaller(p->ptr, x);
3660
if (p->free == p->size) // the page is completely empty
3661
statistic_decrement(tc_log_cur_pages_used, &LOCK_status);
3662
if (p->waiters == 0) // the page is in pool and ready to rock
3663
pthread_cond_signal(&COND_pool); // ping ... for overflow()
3664
pthread_mutex_unlock(&p->lock);
3667
void TC_LOG_MMAP::close()
3672
pthread_mutex_destroy(&LOCK_sync);
3673
pthread_mutex_destroy(&LOCK_active);
3674
pthread_mutex_destroy(&LOCK_pool);
3675
pthread_cond_destroy(&COND_pool);
3677
data[0]='A'; // garble the first (signature) byte, in case my_delete fails
3679
for (i=0; i < npages; i++)
3681
if (pages[i].ptr == 0)
3683
pthread_mutex_destroy(&pages[i].lock);
3684
pthread_cond_destroy(&pages[i].cond);
3687
free((unsigned char*)pages);
3689
my_munmap((char*)data, (size_t)file_length);
3691
my_close(fd, MYF(0));
3693
if (inited>=5) // cannot do in the switch because of Windows
3694
my_delete(logname, MYF(MY_WME));
3698
int TC_LOG_MMAP::recover()
3701
PAGE *p=pages, *end_p=pages+npages;
3703
if (memcmp(data, tc_log_magic, sizeof(tc_log_magic)))
3705
sql_print_error(_("Bad magic header in tc log"));
3710
the first byte after magic signature is set to current
3711
number of storage engines on startup
3713
if (data[sizeof(tc_log_magic)] != total_ha_2pc)
3715
sql_print_error(_("Recovery failed! You must enable "
3716
"exactly %d storage engines that support "
3717
"two-phase commit protocol"),
3718
data[sizeof(tc_log_magic)]);
3722
if (hash_init(&xids, &my_charset_bin, tc_log_page_size/3, 0,
3723
sizeof(my_xid), 0, 0, MYF(0)))
3726
for ( ; p < end_p ; p++)
3728
for (my_xid *x=p->start; x < p->end; x++)
3729
if (*x && my_hash_insert(&xids, (unsigned char *)x))
3733
if (ha_recover(&xids))
3737
memset(data, 0, (size_t)file_length);
3743
sql_print_error(_("Crash recovery failed. Either correct the problem "
3744
"(if it's, for example, out of memory error) and restart, "
3745
"or delete tc log and start drizzled with "
3746
"--tc-heuristic-recover={commit|rollback}"));
3752
TC_LOG_DUMMY tc_log_dummy;
3753
TC_LOG_MMAP tc_log_mmap;
3756
Perform heuristic recovery, if --tc-heuristic-recover was used.
3759
no matter whether heuristic recovery was successful or not
3760
mysqld must exit. So, return value is the same in both cases.
3763
0 no heuristic recovery was requested
3765
1 heuristic recovery was performed
3768
int TC_LOG::using_heuristic_recover()
3770
if (!tc_heuristic_recover)
3773
sql_print_information(_("Heuristic crash recovery mode"));
3775
sql_print_error(_("Heuristic crash recovery failed"));
3776
sql_print_information(_("Please restart mysqld without --tc-heuristic-recover"));
3780
/****** transaction coordinator log for 2pc - binlog() based solution ******/
3781
#define TC_LOG_BINLOG DRIZZLE_BIN_LOG
3785
keep in-memory list of prepared transactions
3786
(add to list in log(), remove on unlog())
3787
and copy it to the new binlog if rotated
3788
but let's check the behaviour of tc_log_page_waits first!
3791
int TC_LOG_BINLOG::open(const char *opt_name)
3796
assert(total_ha_2pc > 1);
3797
assert(opt_name && opt_name[0]);
3799
pthread_mutex_init(&LOCK_prep_xids, MY_MUTEX_INIT_FAST);
3800
pthread_cond_init (&COND_prep_xids, 0);
3802
if (!my_b_inited(&index_file))
3804
/* There was a failure to open the index file, can't open the binlog */
3809
if (using_heuristic_recover())
3811
/* generate a new binlog to mask a corrupted one */
3812
open(opt_name, LOG_BIN, 0, WRITE_CACHE, 0, max_binlog_size, 0);
3817
if ((error= find_log_pos(&log_info, NULL, 1)))
3819
if (error != LOG_INFO_EOF)
3820
sql_print_error(_("find_log_pos() failed (error: %d)"), error);
3831
Format_description_log_event fdle(BINLOG_VERSION);
3832
char log_name[FN_REFLEN];
3834
if (! fdle.is_valid())
3839
strmake(log_name, log_info.log_file_name, sizeof(log_name)-1);
3840
} while (!(error= find_next_log(&log_info, 1)));
3842
if (error != LOG_INFO_EOF)
3844
sql_print_error(_("find_log_pos() failed (error: %d)"), error);
3848
if ((file= open_binlog(&log, log_name, &errmsg)) < 0)
3850
sql_print_error("%s", errmsg);
3854
if ((ev= Log_event::read_log_event(&log, 0, &fdle)) &&
3855
ev->get_type_code() == FORMAT_DESCRIPTION_EVENT &&
3856
ev->flags & LOG_EVENT_BINLOG_IN_USE_F)
3858
sql_print_information(_("Recovering after a crash using %s"), opt_name);
3859
error= recover(&log, (Format_description_log_event *)ev);
3866
my_close(file, MYF(MY_WME));
3876
/** This is called on shutdown, after ha_panic. */
3877
void TC_LOG_BINLOG::close()
3879
assert(prepared_xids==0);
3880
pthread_mutex_destroy(&LOCK_prep_xids);
3881
pthread_cond_destroy (&COND_prep_xids);
3893
int TC_LOG_BINLOG::log_xid(THD *thd, my_xid xid)
3895
Xid_log_event xle(thd, xid);
3896
binlog_trx_data *trx_data=
3897
(binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
3899
We always commit the entire transaction when writing an XID. Also
3900
note that the return value is inverted.
3902
return(!binlog_end_trans(thd, trx_data, &xle, true));
3905
void TC_LOG_BINLOG::unlog(ulong cookie __attribute__((unused)),
3906
my_xid xid __attribute__((unused)))
3908
pthread_mutex_lock(&LOCK_prep_xids);
3909
assert(prepared_xids > 0);
3910
if (--prepared_xids == 0) {
3911
pthread_cond_signal(&COND_prep_xids);
3913
pthread_mutex_unlock(&LOCK_prep_xids);
3914
rotate_and_purge(0); // as ::write() did not rotate
3917
int TC_LOG_BINLOG::recover(IO_CACHE *log, Format_description_log_event *fdle)
3923
if (! fdle->is_valid() ||
3924
hash_init(&xids, &my_charset_bin, TC_LOG_PAGE_SIZE/3, 0,
3925
sizeof(my_xid), 0, 0, MYF(0)))
3928
init_alloc_root(&mem_root, TC_LOG_PAGE_SIZE, TC_LOG_PAGE_SIZE);
3930
fdle->flags&= ~LOG_EVENT_BINLOG_IN_USE_F; // abort on the first error
3932
while ((ev= Log_event::read_log_event(log,0,fdle)) && ev->is_valid())
3934
if (ev->get_type_code() == XID_EVENT)
3936
Xid_log_event *xev=(Xid_log_event *)ev;
3937
unsigned char *x= (unsigned char *) memdup_root(&mem_root, (unsigned char*) &xev->xid,
3941
my_hash_insert(&xids, x);
3946
if (ha_recover(&xids))
3949
free_root(&mem_root, MYF(0));
3954
free_root(&mem_root, MYF(0));
3957
sql_print_error(_("Crash recovery failed. Either correct the problem "
3958
"(if it's, for example, out of memory error) and restart, "
3959
"or delete (or rename) binary log and start mysqld with "
3960
"--tc-heuristic-recover={commit|rollback}"));
3965
#ifdef INNODB_COMPATIBILITY_HOOKS
3967
Get the file name of the MySQL binlog.
3968
@return the name of the binlog file
3971
const char* mysql_bin_log_file_name(void)
3973
return mysql_bin_log.get_log_fname();
3976
Get the current position of the MySQL binlog.
3977
@return byte offset from the beginning of the binlog
3980
uint64_t mysql_bin_log_file_pos(void)
3982
return (uint64_t) mysql_bin_log.get_log_file()->pos_in_file;
3984
#endif /* INNODB_COMPATIBILITY_HOOKS */
3987
mysql_declare_plugin(binlog)
3989
DRIZZLE_STORAGE_ENGINE_PLUGIN,
3993
"This is a pseudo storage engine to represent the binlog in a transaction",
3995
binlog_init, /* Plugin Init */
3996
NULL, /* Plugin Deinit */
3997
NULL, /* status variables */
3998
NULL, /* system variables */
3999
NULL /* config options */
4001
mysql_declare_plugin_end;