1
1
/*****************************************************************************
3
Copyright (c) 1996, 2010, Innobase Oy. All Rights Reserved.
3
Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
4
4
Copyright (c) 2008, Google Inc.
5
Copyright (c) 2009, Percona Inc.
7
6
Portions of this file contain modifications contributed and copyrighted by
8
7
Google, Inc. Those modifications are gratefully acknowledged and are described
10
9
incorporated with their permission, and subject to the conditions contained in
11
10
the file COPYING.Google.
13
Portions of this file contain modifications contributed and copyrighted
14
by Percona Inc.. Those modifications are
15
gratefully acknowledged and are described briefly in the InnoDB
16
documentation. The contributions by Percona Inc. are incorporated with
17
their permission, and subject to the conditions contained in the file
20
12
This program is free software; you can redistribute it and/or modify it under
21
13
the terms of the GNU General Public License as published by the Free Software
22
14
Foundation; version 2 of the License.
30
22
St, Fifth Floor, Boston, MA 02110-1301 USA
32
24
*****************************************************************************/
25
/***********************************************************************
27
Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
28
Copyright (c) 2009, Percona Inc.
30
Portions of this file contain modifications contributed and copyrighted
31
by Percona Inc.. Those modifications are
32
gratefully acknowledged and are described briefly in the InnoDB
33
documentation. The contributions by Percona Inc. are incorporated with
34
their permission, and subject to the conditions contained in the file
37
This program is free software; you can redistribute it and/or modify it
38
under the terms of the GNU General Public License as published by the
39
Free Software Foundation; version 2 of the License.
41
This program is distributed in the hope that it will be useful, but
42
WITHOUT ANY WARRANTY; without even the implied warranty of
43
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General
44
Public License for more details.
46
You should have received a copy of the GNU General Public License along
47
with this program; if not, write to the Free Software Foundation, Inc.,
48
51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
50
***********************************************************************/
34
52
/********************************************************************//**
35
53
@file srv/srv0start.c
129
146
static ulint ios;
131
148
/** io_handler_thread parameters for thread identification */
132
static ulint n[SRV_MAX_N_IO_THREADS + 6];
149
static ulint n[SRV_MAX_N_IO_THREADS + 5];
133
150
/** io_handler_thread identifiers */
134
static os_thread_id_t thread_ids[SRV_MAX_N_IO_THREADS + 6];
151
static os_thread_id_t thread_ids[SRV_MAX_N_IO_THREADS + 5];
136
153
/** We use this mutex to test the return value of pthread_mutex_trylock
137
154
on successful locking. HP-UX does NOT return 0, though Linux et al do. */
145
162
#define SRV_N_PENDING_IOS_PER_THREAD OS_AIO_N_PENDING_IOS_PER_THREAD
146
163
#define SRV_MAX_N_PENDING_SYNC_IOS 100
148
#ifdef UNIV_PFS_THREAD
149
/* Keys to register InnoDB threads with performance schema */
150
UNIV_INTERN mysql_pfs_key_t io_handler_thread_key;
151
UNIV_INTERN mysql_pfs_key_t srv_lock_timeout_thread_key;
152
UNIV_INTERN mysql_pfs_key_t srv_error_monitor_thread_key;
153
UNIV_INTERN mysql_pfs_key_t srv_monitor_thread_key;
154
UNIV_INTERN mysql_pfs_key_t srv_master_thread_key;
155
#endif /* UNIV_PFS_THREAD */
157
#ifdef UNIV_PFS_MUTEX
158
/* Key to register ios_mutex_key with performance schema */
159
UNIV_INTERN mysql_pfs_key_t ios_mutex_key;
160
#endif /* UNIV_PFS_MUTEX */
162
166
/*********************************************************************//**
163
167
Convert a numeric string that optionally ends in G or M, to a number
486
491
fprintf(stderr, "Io handler thread %lu starts, id %lu\n", segment,
487
492
os_thread_pf(os_thread_get_curr_id()));
490
#ifdef UNIV_PFS_THREAD
491
pfs_register_thread(io_handler_thread_key);
492
#endif /* UNIV_PFS_THREAD */
494
while (srv_shutdown_state != SRV_SHUTDOWN_EXIT_THREADS) {
495
495
fil_aio_wait(segment);
497
497
mutex_enter(&ios_mutex);
602
604
sprintf(name + dirnamelen, "%s%lu", "ib_logfile", (ulong) i);
604
files[i] = os_file_create(innodb_file_log_key, name,
605
OS_FILE_CREATE, OS_FILE_NORMAL,
606
files[i] = os_file_create(name, OS_FILE_CREATE, OS_FILE_NORMAL,
606
607
OS_LOG_FILE, &ret);
607
608
if (ret == FALSE) {
608
609
if (os_file_get_last_error(FALSE) != OS_FILE_ALREADY_EXISTS
620
621
return(DB_ERROR);
623
files[i] = os_file_create(innodb_file_log_key, name,
624
OS_FILE_OPEN, OS_FILE_AIO,
624
files[i] = os_file_create(name, OS_FILE_OPEN, OS_FILE_AIO,
625
625
OS_LOG_FILE, &ret);
786
786
/* First we try to create the file: if it already
787
787
exists, ret will get value FALSE */
789
files[i] = os_file_create(innodb_file_data_key,
790
name, OS_FILE_CREATE,
789
files[i] = os_file_create(name, OS_FILE_CREATE,
792
791
OS_DATA_FILE, &ret);
814
813
srv_start_raw_disk_in_use = TRUE;
815
814
srv_created_new_raw = TRUE;
817
files[i] = os_file_create(innodb_file_data_key,
818
name, OS_FILE_OPEN_RAW,
816
files[i] = os_file_create(name, OS_FILE_OPEN_RAW,
820
818
OS_DATA_FILE, &ret);
849
847
if (srv_data_file_is_raw_partition[i] == SRV_OLD_RAW) {
850
848
files[i] = os_file_create(
851
innodb_file_data_key,
852
849
name, OS_FILE_OPEN_RAW,
853
850
OS_FILE_NORMAL, OS_DATA_FILE, &ret);
854
851
} else if (i == 0) {
855
852
files[i] = os_file_create(
856
innodb_file_data_key,
857
853
name, OS_FILE_OPEN_RETRY,
858
854
OS_FILE_NORMAL, OS_DATA_FILE, &ret);
860
856
files[i] = os_file_create(
861
innodb_file_data_key,
862
857
name, OS_FILE_OPEN, OS_FILE_NORMAL,
863
858
OS_DATA_FILE, &ret);
1080
1076
#ifdef UNIV_IBUF_DEBUG
1081
1077
fprintf(stderr,
1082
1078
"InnoDB: !!!!!!!! UNIV_IBUF_DEBUG switched on !!!!!!!!!\n"
1083
# ifdef UNIV_IBUF_COUNT_DEBUG
1084
"InnoDB: !!!!!!!! UNIV_IBUF_COUNT_DEBUG switched on !!!!!!!!!\n"
1085
"InnoDB: Crash recovery will fail with UNIV_IBUF_COUNT_DEBUG\n"
1079
"InnoDB: Crash recovery will fail with UNIV_IBUF_DEBUG\n");
1090
1082
#ifdef UNIV_SYNC_DEBUG
1111
1103
"InnoDB: The InnoDB memory heap is disabled\n");
1114
fputs("InnoDB: " IB_ATOMICS_STARTUP_MSG
1115
"\nInnoDB: Compressed tables use zlib " ZLIB_VERSION
1116
#ifdef UNIV_ZIP_DEBUG
1118
#endif /* UNIV_ZIP_DEBUG */
1119
#ifdef UNIV_ZIP_COPY
1120
" and extra copying"
1121
#endif /* UNIV_ZIP_COPY */
1106
fprintf(stderr, "InnoDB: %s\n", IB_ATOMICS_STARTUP_MSG);
1125
1108
/* Since InnoDB does not currently clean up all its internal data
1126
1109
structures in MySQL Embedded Server Library server_end(), we
1157
1141
but when run in conjunction with InnoDB Hot Backup, it seemed
1158
1142
to corrupt the data files. */
1160
srv_use_native_aio = FALSE;
1165
/* On 2000 and XP, async IO is available. */
1166
srv_use_native_aio = TRUE;
1144
os_aio_use_native_aio = FALSE;
1170
/* Vista and later have both async IO and condition variables */
1171
srv_use_native_aio = TRUE;
1172
srv_use_native_conditions = TRUE;
1147
/* On Win 2000 and XP use async i/o */
1148
os_aio_use_native_aio = TRUE;
1176
#elif defined(LINUX_NATIVE_AIO)
1178
if (srv_use_native_aio) {
1179
ut_print_timestamp(stderr);
1181
" InnoDB: Using Linux native AIO\n");
1184
/* Currently native AIO is supported only on windows and linux
1185
and that also when the support is compiled in. In all other
1186
cases, we ignore the setting of innodb_use_native_aio. */
1187
srv_use_native_aio = FALSE;
1191
1152
if (srv_file_flush_method_str == NULL) {
1192
1153
/* These are the default options */
1213
1174
} else if (0 == ut_strcmp(srv_file_flush_method_str, "normal")) {
1214
1175
srv_win_file_flush_method = SRV_WIN_IO_NORMAL;
1215
srv_use_native_aio = FALSE;
1176
os_aio_use_native_aio = FALSE;
1217
1178
} else if (0 == ut_strcmp(srv_file_flush_method_str, "unbuffered")) {
1218
1179
srv_win_file_flush_method = SRV_WIN_IO_UNBUFFERED;
1219
srv_use_native_aio = FALSE;
1180
os_aio_use_native_aio = FALSE;
1221
1182
} else if (0 == ut_strcmp(srv_file_flush_method_str,
1222
1183
"async_unbuffered")) {
1238
1199
maximum number of threads that can wait in the 'srv_conc array' for
1239
1200
their time to enter InnoDB. */
1202
#if defined(__NETWARE__)
1204
/* Create less event semaphores because Win 98/ME had
1205
difficulty creating 40000 event semaphores. Comment from
1206
Novell, Inc.: also, these just take a lot of memory on
1208
srv_max_n_threads = 1000;
1241
1210
if (srv_buf_pool_size >= 1000 * 1024 * 1024) {
1242
1211
/* If buffer pool is less than 1000 MB,
1243
assume fewer threads. Also use only one
1244
buffer pool instance */
1212
assume fewer threads. */
1245
1213
srv_max_n_threads = 50000;
1247
1215
} else if (srv_buf_pool_size >= 8 * 1024 * 1024) {
1249
srv_buf_pool_instances = 1;
1250
1217
srv_max_n_threads = 10000;
1252
srv_buf_pool_instances = 1;
1253
1219
srv_max_n_threads = 1000; /* saves several MB of memory,
1254
1220
especially in 64-bit
1258
1224
err = srv_boot();
1260
1226
if (err != DB_SUCCESS) {
1262
1228
return((int) err);
1265
mutex_create(srv_monitor_file_mutex_key,
1266
&srv_monitor_file_mutex, SYNC_NO_ORDER_CHECK);
1231
mutex_create(&srv_monitor_file_mutex, SYNC_NO_ORDER_CHECK);
1268
1233
if (srv_innodb_status) {
1269
1234
srv_monitor_file_name = mem_alloc(
1288
mutex_create(srv_dict_tmpfile_mutex_key,
1289
&srv_dict_tmpfile_mutex, SYNC_DICT_OPERATION);
1253
mutex_create(&srv_dict_tmpfile_mutex, SYNC_DICT_OPERATION);
1291
1255
srv_dict_tmpfile = os_file_create_tmpfile();
1292
1256
if (!srv_dict_tmpfile) {
1293
1257
return(DB_ERROR);
1296
mutex_create(srv_misc_tmpfile_mutex_key,
1297
&srv_misc_tmpfile_mutex, SYNC_ANY_LATCH);
1260
mutex_create(&srv_misc_tmpfile_mutex, SYNC_ANY_LATCH);
1299
1262
srv_misc_tmpfile = os_file_create_tmpfile();
1300
1263
if (!srv_misc_tmpfile) {
1314
1277
/* TODO: Investigate if SRV_N_PENDING_IOS_PER_THREAD (32) limit
1315
1278
still applies to windows. */
1316
if (!srv_use_native_aio) {
1279
if (!os_aio_use_native_aio) {
1317
1280
io_limit = 8 * SRV_N_PENDING_IOS_PER_THREAD;
1319
1282
io_limit = SRV_N_PENDING_IOS_PER_THREAD;
1327
1290
fil_init(srv_file_per_table ? 50000 : 5000,
1328
1291
srv_max_n_open_files);
1330
/* Print time to initialize the buffer pool */
1331
ut_print_timestamp(stderr);
1333
" InnoDB: Initializing buffer pool, size =");
1335
if (srv_buf_pool_size >= 1024 * 1024 * 1024) {
1338
((double) srv_buf_pool_size) / (1024 * 1024 * 1024));
1342
((double) srv_buf_pool_size) / (1024 * 1024));
1345
err = buf_pool_init(srv_buf_pool_size, srv_buf_pool_instances);
1347
ut_print_timestamp(stderr);
1349
" InnoDB: Completed initialization of buffer pool\n");
1351
if (err != DB_SUCCESS) {
1293
ret = buf_pool_init();
1352
1296
fprintf(stderr,
1353
1297
"InnoDB: Fatal error: cannot allocate the memory"
1354
1298
" for the buffer pool\n");
1546
1490
if (create_new_db) {
1547
1491
mtr_start(&mtr);
1549
1492
fsp_header_init(0, sum_of_new_sizes, &mtr);
1551
1494
mtr_commit(&mtr);
1553
/* To maintain backward compatibility we create only
1554
the first rollback segment before the double write buffer.
1555
All the remaining rollback segments will be created later,
1556
after the double write buffer has been created. */
1557
1496
trx_sys_create();
1561
1498
srv_startup_is_before_trx_rollback_phase = FALSE;
1563
1500
#ifdef UNIV_LOG_ARCHIVE
1637
1572
trx_sys_init_at_db_start();
1639
/* Initialize the fsp free limit global variable in the log
1641
fsp_header_get_free_limit();
1643
/* recv_recovery_from_checkpoint_finish needs trx lists which
1644
are initialized in trx_sys_init_at_db_start(). */
1646
recv_recovery_from_checkpoint_finish();
1647
1574
if (srv_force_recovery < SRV_FORCE_NO_IBUF_MERGE) {
1648
1575
/* The following call is necessary for the insert
1649
1576
buffer to work with multiple tablespaces. We must
1659
1586
every table in the InnoDB data dictionary that has
1662
We also determine the maximum tablespace id used. */
1589
We also determine the maximum tablespace id used.
1591
TODO: We may have incomplete transactions in the
1592
data dictionary tables. Does that harm the scanning of
1593
the data dictionary below? */
1664
1595
dict_check_tablespaces_and_store_max_id(
1665
1596
recv_needed_recovery);
1668
1599
srv_startup_is_before_trx_rollback_phase = FALSE;
1669
recv_recovery_rollback_active();
1601
/* Initialize the fsp free limit global variable in the log
1603
fsp_header_get_free_limit();
1605
/* recv_recovery_from_checkpoint_finish needs trx lists which
1606
are initialized in trx_sys_init_at_db_start(). */
1608
recv_recovery_from_checkpoint_finish();
1671
1610
/* It is possible that file_format tag has never
1672
1611
been set. In this case we initialize it to minimum
1715
1654
/* fprintf(stderr, "Max allowed record size %lu\n",
1716
1655
page_get_free_space_of_empty() / 2); */
1718
if (trx_doublewrite == NULL) {
1719
/* Create the doublewrite buffer to a new tablespace */
1721
trx_sys_create_doublewrite_buf();
1724
/* Here the double write buffer has already been created and so
1725
any new rollback segments will be allocated after the double
1726
write buffer. The default segment should already exist.
1727
We create the new segments only if it's a new database or
1728
the database was shutdown cleanly. */
1730
/* Note: When creating the extra rollback segments during an upgrade
1731
we violate the latching order, even if the change buffer is empty.
1732
We make an exception in sync0sync.c and check srv_is_being_started
1733
for that violation. It cannot create a deadlock because we are still
1734
running in single threaded mode essentially. Only the IO threads
1735
should be running at this stage. */
1737
trx_sys_create_rsegs(TRX_SYS_N_RSEGS - 1);
1739
/* Create the thread which watches the timeouts for lock waits */
1740
os_thread_create(&srv_lock_timeout_thread, NULL,
1657
/* Create the thread which watches the timeouts for lock waits
1658
and prints InnoDB monitor info */
1660
os_thread_create(&srv_lock_timeout_and_monitor_thread, NULL,
1741
1661
thread_ids + 2 + SRV_MAX_N_IO_THREADS);
1743
1663
/* Create the thread which warns of long semaphore waits */
1744
1664
os_thread_create(&srv_error_monitor_thread, NULL,
1745
1665
thread_ids + 3 + SRV_MAX_N_IO_THREADS);
1747
/* Create the thread which prints InnoDB monitor info */
1748
os_thread_create(&srv_monitor_thread, NULL,
1749
thread_ids + 4 + SRV_MAX_N_IO_THREADS);
1751
1666
srv_is_being_started = FALSE;
1668
if (trx_doublewrite == NULL) {
1669
/* Create the doublewrite buffer to a new tablespace */
1671
trx_sys_create_doublewrite_buf();
1753
1674
err = dict_create_or_check_foreign_constraint_tables();
1755
1676
if (err != DB_SUCCESS) {
1762
1683
os_thread_create(&srv_master_thread, NULL, thread_ids
1763
1684
+ (1 + SRV_MAX_N_IO_THREADS));
1765
/* Currently we allow only a single purge thread. */
1766
ut_a(srv_n_purge_threads == 0 || srv_n_purge_threads == 1);
1768
/* If the user has requested a separate purge thread then
1769
start the purge thread. */
1770
if (srv_n_purge_threads == 1) {
1771
os_thread_create(&srv_purge_thread, NULL, NULL);
1774
1685
#ifdef UNIV_DEBUG
1775
1686
/* buf_debug_prints = TRUE; */
1776
1687
#endif /* UNIV_DEBUG */
1864
1775
if (srv_print_verbose_log) {
1865
1776
ut_print_timestamp(stderr);
1866
1777
fprintf(stderr,
1867
" InnoDB %s started; "
1778
" InnoDB Plugin %s started; "
1868
1779
"log sequence number %"PRIu64"\n",
1869
1780
INNODB_VERSION_STR, srv_start_lsn);
2018
1935
/* c. We wake the master thread so that it exits */
2019
1936
srv_wake_master_thread();
2021
/* d. We wake the purge thread so that it exits */
2022
srv_wake_purge_thread();
2024
/* e. Exit the i/o threads */
1938
/* d. Exit the i/o threads */
2026
1940
os_aio_wake_all_threads_at_shutdown();
2106
2020
pars_lexer_close();
2107
2021
log_mem_free();
2108
buf_pool_free(srv_buf_pool_instances);
2111
/* ut_free_all_mem() frees all allocated memory not freed yet
2112
in shutdown, and it will also free the ut_list_mutex, so it
2113
should be the last one for all operation */
2116
2026
if (os_thread_count != 0
2117
2027
|| os_event_count != 0
2118
2028
|| os_mutex_count != 0