]> git.pld-linux.org Git - packages/mysql.git/blame - mysql-innodb_extend_slow.patch
projects / packages / mysql.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
- new perfona url, new patch names (but still have to wait for the patches to be...
[packages/mysql.git] / mysql-innodb_extend_slow.patch
CommitLineData
d4e9320e
AM		 1# name : innodb_extend_slow.patch
2# introduced : 11 or before
3# maintainer : Yasufumi
4#
5#!!! notice !!!
6# Any small change to this file in the main branch
7# should be done or reviewed by the maintainer!
8diff -ruN a/storage/innobase/buf/buf0buf.c b/storage/innobase/buf/buf0buf.c
9--- a/storage/innobase/buf/buf0buf.c 2010-12-03 15:49:59.175955882 +0900
10+++ b/storage/innobase/buf/buf0buf.c 2010-12-03 17:42:42.074307123 +0900
11@@ -51,6 +51,40 @@
12 #include "dict0dict.h"
13 #include "log0recv.h"
14 #include "page0zip.h"
15+#include "trx0trx.h"
16+
17+/* prototypes for new functions added to ha_innodb.cc */
18+trx_t* innobase_get_trx();
19+
20+inline void _increment_page_get_statistics(buf_block_t* block, trx_t* trx)
21+{
22+ ulint block_hash;
23+ ulint block_hash_byte;
24+ byte block_hash_offset;
25+
26+ ut_ad(block);
27+
28+ if (!innobase_get_slow_log() || !trx || !trx->take_stats)
29+ return;
30+
31+ if (!trx->distinct_page_access_hash) {
32+ trx->distinct_page_access_hash = mem_alloc(DPAH_SIZE);
33+ memset(trx->distinct_page_access_hash, 0, DPAH_SIZE);
34+ }
35+
36+ block_hash = ut_hash_ulint((block->page.space << 20) + block->page.space +
37+ block->page.offset, DPAH_SIZE << 3);
38+ block_hash_byte = block_hash >> 3;
39+ block_hash_offset = (byte) block_hash & 0x07;
40+ if (block_hash_byte < 0 || block_hash_byte >= DPAH_SIZE)
41+ fprintf(stderr, "!!! block_hash_byte = %lu block_hash_offset = %d !!!\n", block_hash_byte, block_hash_offset);
42+ if (block_hash_offset < 0 || block_hash_offset > 7)
43+ fprintf(stderr, "!!! block_hash_byte = %lu block_hash_offset = %d !!!\n", block_hash_byte, block_hash_offset);
44+ if ((trx->distinct_page_access_hash[block_hash_byte] & ((byte) 0x01 << block_hash_offset)) == 0)
45+ trx->distinct_page_access++;
46+ trx->distinct_page_access_hash[block_hash_byte] |= (byte) 0x01 << block_hash_offset;
47+ return;
48+}
49
50 /*
51 IMPLEMENTATION OF THE BUFFER POOL
52@@ -2399,11 +2433,19 @@
53 mutex_t* block_mutex;
54 ibool must_read;
55 unsigned access_time;
56+ trx_t* trx = NULL;
57+ ulint sec;
58+ ulint ms;
59+ ib_uint64_t start_time;
60+ ib_uint64_t finish_time;
61 buf_pool_t* buf_pool = buf_pool_get(space, offset);
62
63 #ifndef UNIV_LOG_DEBUG
64 ut_ad(!ibuf_inside());
65 #endif
66+ if (innobase_get_slow_log()) {
67+ trx = innobase_get_trx();
68+ }
69 buf_pool->stat.n_page_gets++;
70
71 for (;;) {
72@@ -2421,7 +2463,7 @@
73 //buf_pool_mutex_exit(buf_pool);
74 rw_lock_s_unlock(&buf_pool->page_hash_latch);
75
76- buf_read_page(space, zip_size, offset);
77+ buf_read_page(space, zip_size, offset, trx);
78
79 #if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
80 ut_a(++buf_dbg_counter % 37 || buf_validate());
81@@ -2499,6 +2541,13 @@
82 /* Let us wait until the read operation
83 completes */
84
85+ if (innobase_get_slow_log() && trx && trx->take_stats)
86+ {
87+ ut_usectime(&sec, &ms);
88+ start_time = (ib_uint64_t)sec * 1000000 + ms;
89+ } else {
90+ start_time = 0;
91+ }
92 for (;;) {
93 enum buf_io_fix io_fix;
94
95@@ -2513,6 +2562,12 @@
96 break;
97 }
98 }
99+ if (innobase_get_slow_log() && trx && trx->take_stats && start_time)
100+ {
101+ ut_usectime(&sec, &ms);
102+ finish_time = (ib_uint64_t)sec * 1000000 + ms;
103+ trx->io_reads_wait_timer += (ulint)(finish_time - start_time);
104+ }
105 }
106
107 #ifdef UNIV_IBUF_COUNT_DEBUG
108@@ -2825,6 +2880,11 @@
109 ibool must_read;
110 ulint retries = 0;
111 mutex_t* block_mutex = NULL;
112+ trx_t* trx = NULL;
113+ ulint sec;
114+ ulint ms;
115+ ib_uint64_t start_time;
116+ ib_uint64_t finish_time;
117 buf_pool_t* buf_pool = buf_pool_get(space, offset);
118
119 ut_ad(mtr);
120@@ -2842,6 +2902,9 @@
121 #ifndef UNIV_LOG_DEBUG
122 ut_ad(!ibuf_inside() || ibuf_page(space, zip_size, offset, NULL));
123 #endif
124+ if (innobase_get_slow_log()) {
125+ trx = innobase_get_trx();
126+ }
127 buf_pool->stat.n_page_gets++;
128 fold = buf_page_address_fold(space, offset);
129 loop:
130@@ -2915,7 +2978,7 @@
131 return(NULL);
132 }
133
134- if (buf_read_page(space, zip_size, offset)) {
135+ if (buf_read_page(space, zip_size, offset, trx)) {
136 retries = 0;
137 } else if (retries < BUF_PAGE_READ_MAX_RETRIES) {
138 ++retries;
139@@ -3178,6 +3241,13 @@
140 /* Let us wait until the read operation
141 completes */
142
143+ if (innobase_get_slow_log() && trx && trx->take_stats)
144+ {
145+ ut_usectime(&sec, &ms);
146+ start_time = (ib_uint64_t)sec * 1000000 + ms;
147+ } else {
148+ start_time = 0;
149+ }
150 for (;;) {
151 enum buf_io_fix io_fix;
152
153@@ -3192,6 +3262,12 @@
154 break;
155 }
156 }
157+ if (innobase_get_slow_log() && trx && trx->take_stats && start_time)
158+ {
159+ ut_usectime(&sec, &ms);
160+ finish_time = (ib_uint64_t)sec * 1000000 + ms;
161+ trx->io_reads_wait_timer += (ulint)(finish_time - start_time);
162+ }
163 }
164
165 fix_type = MTR_MEMO_BUF_FIX;
166@@ -3217,13 +3293,17 @@
167 /* In the case of a first access, try to apply linear
168 read-ahead */
169
170- buf_read_ahead_linear(space, zip_size, offset);
171+ buf_read_ahead_linear(space, zip_size, offset, trx);
172 }
173
174 #ifdef UNIV_IBUF_COUNT_DEBUG
175 ut_a(ibuf_count_get(buf_block_get_space(block),
176 buf_block_get_page_no(block)) == 0);
177 #endif
178+ if (innobase_get_slow_log()) {
179+ _increment_page_get_statistics(block, trx);
180+ }
181+
182 return(block);
183 }
184
185@@ -3247,6 +3327,7 @@
186 unsigned access_time;
187 ibool success;
188 ulint fix_type;
189+ trx_t* trx = NULL;
190
191 ut_ad(block);
192 ut_ad(mtr);
193@@ -3324,13 +3405,17 @@
194 #ifdef UNIV_DEBUG_FILE_ACCESSES
195 ut_a(block->page.file_page_was_freed == FALSE);
196 #endif
197+ if (innobase_get_slow_log()) {
198+ trx = innobase_get_trx();
199+ }
200+
201 if (UNIV_UNLIKELY(!access_time)) {
202 /* In the case of a first access, try to apply linear
203 read-ahead */
204
205 buf_read_ahead_linear(buf_block_get_space(block),
206 buf_block_get_zip_size(block),
207- buf_block_get_page_no(block));
208+ buf_block_get_page_no(block), trx);
209 }
210
211 #ifdef UNIV_IBUF_COUNT_DEBUG
212@@ -3340,6 +3425,9 @@
213 buf_pool = buf_pool_from_block(block);
214 buf_pool->stat.n_page_gets++;
215
216+ if (innobase_get_slow_log()) {
217+ _increment_page_get_statistics(block, trx);
218+ }
219 return(TRUE);
220 }
221
222@@ -3362,6 +3450,7 @@
223 buf_pool_t* buf_pool;
224 ibool success;
225 ulint fix_type;
226+ trx_t* trx = NULL;
227
228 ut_ad(mtr);
229 ut_ad(mtr->state == MTR_ACTIVE);
230@@ -3448,6 +3537,11 @@
231 #endif
232 buf_pool->stat.n_page_gets++;
233
234+ if (innobase_get_slow_log()) {
235+ trx = innobase_get_trx();
236+ _increment_page_get_statistics(block, trx);
237+ }
238+
239 return(TRUE);
240 }
241
242diff -ruN a/storage/innobase/buf/buf0rea.c b/storage/innobase/buf/buf0rea.c
243--- a/storage/innobase/buf/buf0rea.c 2010-12-03 17:32:15.617037263 +0900
244+++ b/storage/innobase/buf/buf0rea.c 2010-12-03 17:42:42.075297193 +0900
245@@ -77,7 +77,8 @@
246 treat the tablespace as dropped; this is a timestamp we
247 use to stop dangling page reads from a tablespace
248 which we have DISCARDed + IMPORTed back */
249- ulint offset) /*!< in: page number */
250+ ulint offset, /*!< in: page number */
251+ trx_t* trx)
252 {
253 buf_page_t* bpage;
254 ulint wake_later;
255@@ -179,15 +180,15 @@
256
257 thd_wait_begin(NULL, THD_WAIT_DISKIO);
258 if (zip_size) {
259- *err = fil_io(OS_FILE_READ | wake_later,
260+ *err = _fil_io(OS_FILE_READ | wake_later,
261 sync, space, zip_size, offset, 0, zip_size,
262- bpage->zip.data, bpage);
263+ bpage->zip.data, bpage, trx);
264 } else {
265 ut_a(buf_page_get_state(bpage) == BUF_BLOCK_FILE_PAGE);
266
267- *err = fil_io(OS_FILE_READ | wake_later,
268+ *err = _fil_io(OS_FILE_READ | wake_later,
269 sync, space, 0, offset, 0, UNIV_PAGE_SIZE,
270- ((buf_block_t*) bpage)->frame, bpage);
271+ ((buf_block_t*) bpage)->frame, bpage, trx);
272 }
273 thd_wait_end(NULL);
274 ut_a(*err == DB_SUCCESS);
275@@ -213,7 +214,8 @@
276 /*==========*/
277 ulint space, /*!< in: space id */
278 ulint zip_size,/*!< in: compressed page size in bytes, or 0 */
279- ulint offset) /*!< in: page number */
280+ ulint offset, /*!< in: page number */
281+ trx_t* trx)
282 {
283 buf_pool_t* buf_pool = buf_pool_get(space, offset);
284 ib_int64_t tablespace_version;
285@@ -227,7 +229,7 @@
286
287 count = buf_read_page_low(&err, TRUE, BUF_READ_ANY_PAGE, space,
288 zip_size, FALSE,
289- tablespace_version, offset);
290+ tablespace_version, offset, trx);
291 srv_buf_pool_reads += count;
292 if (err == DB_TABLESPACE_DELETED) {
293 ut_print_timestamp(stderr);
294@@ -278,8 +280,9 @@
295 /*==================*/
296 ulint space, /*!< in: space id */
297 ulint zip_size,/*!< in: compressed page size in bytes, or 0 */
298- ulint offset) /*!< in: page number of a page; NOTE: the current thread
299+ ulint offset, /*!< in: page number of a page; NOTE: the current thread
300 must want access to this page (see NOTE 3 above) */
301+ trx_t* trx)
302 {
303 buf_pool_t* buf_pool = buf_pool_get(space, offset);
304 ib_int64_t tablespace_version;
305@@ -500,7 +503,7 @@
306 count += buf_read_page_low(
307 &err, FALSE,
308 ibuf_mode | OS_AIO_SIMULATED_WAKE_LATER,
309- space, zip_size, FALSE, tablespace_version, i);
310+ space, zip_size, FALSE, tablespace_version, i, trx);
311 if (err == DB_TABLESPACE_DELETED) {
312 ut_print_timestamp(stderr);
313 fprintf(stderr,
314@@ -594,7 +597,7 @@
315 buf_read_page_low(&err, sync && (i + 1 == n_stored),
316 BUF_READ_ANY_PAGE, space_ids[i],
317 zip_size, TRUE, space_versions[i],
318- page_nos[i]);
319+ page_nos[i], NULL);
320
321 if (UNIV_UNLIKELY(err == DB_TABLESPACE_DELETED)) {
322 tablespace_deleted:
323@@ -736,12 +739,12 @@
324 if ((i + 1 == n_stored) && sync) {
325 buf_read_page_low(&err, TRUE, BUF_READ_ANY_PAGE, space,
326 zip_size, TRUE, tablespace_version,
327- page_nos[i]);
328+ page_nos[i], NULL);
329 } else {
330 buf_read_page_low(&err, FALSE, BUF_READ_ANY_PAGE
331 | OS_AIO_SIMULATED_WAKE_LATER,
332 space, zip_size, TRUE,
333- tablespace_version, page_nos[i]);
334+ tablespace_version, page_nos[i], NULL);
335 }
336 }
337
338diff -ruN a/storage/innobase/fil/fil0fil.c b/storage/innobase/fil/fil0fil.c
339--- a/storage/innobase/fil/fil0fil.c 2010-12-03 15:53:54.610037199 +0900
340+++ b/storage/innobase/fil/fil0fil.c 2010-12-03 17:42:42.079064198 +0900
341@@ -4349,7 +4349,7 @@
342 node->name, node->handle, buf,
343 offset_low, offset_high,
344 page_size * n_pages,
345- NULL, NULL);
346+ NULL, NULL, NULL);
347 #endif
348 if (success) {
349 node->size += n_pages;
350@@ -4676,7 +4676,7 @@
351 i/o on a tablespace which does not exist */
352 UNIV_INTERN
353 ulint
354-fil_io(
355+_fil_io(
356 /*===*/
357 ulint type, /*!< in: OS_FILE_READ or OS_FILE_WRITE,
358 ORed to OS_FILE_LOG, if a log i/o
359@@ -4701,8 +4701,9 @@
360 void* buf, /*!< in/out: buffer where to store read data
361 or from where to write; in aio this must be
362 appropriately aligned */
363- void* message) /*!< in: message for aio handler if non-sync
364+ void* message, /*!< in: message for aio handler if non-sync
365 aio used, else ignored */
366+ trx_t* trx)
367 {
368 ulint mode;
369 fil_space_t* space;
370@@ -4872,7 +4873,7 @@
371 #else
372 /* Queue the aio request */
373 ret = os_aio(type, mode | wake_later, node->name, node->handle, buf,
374- offset_low, offset_high, len, node, message);
375+ offset_low, offset_high, len, node, message, trx);
376 #endif
377 ut_a(ret);
378
379diff -ruN a/storage/innobase/handler/ha_innodb.cc b/storage/innobase/handler/ha_innodb.cc
380--- a/storage/innobase/handler/ha_innodb.cc 2010-12-03 17:36:44.293955189 +0900
381+++ b/storage/innobase/handler/ha_innodb.cc 2010-12-03 17:42:42.090024586 +0900
382@@ -1528,6 +1528,16 @@
383 trx->check_unique_secondary = !thd_test_options(
384 thd, OPTION_RELAXED_UNIQUE_CHECKS);
385
386+#ifdef EXTENDED_SLOWLOG
387+ if (thd_log_slow_verbosity(thd) & SLOG_V_INNODB) {
388+ trx->take_stats = TRUE;
389+ } else {
390+ trx->take_stats = FALSE;
391+ }
392+#else
393+ trx->take_stats = FALSE;
394+#endif
395+
396 DBUG_VOID_RETURN;
397 }
398
399@@ -1583,6 +1593,32 @@
400 }
401
402
403+/*************************************************************************
404+Gets current trx. */
405+extern "C"
406+trx_t*
407+innobase_get_trx()
408+{
409+ THD *thd=current_thd;
410+ if (likely(thd != 0)) {
411+ trx_t*& trx = thd_to_trx(thd);
412+ return(trx);
413+ } else {
414+ return(NULL);
415+ }
416+}
417+
418+extern "C"
419+ibool
420+innobase_get_slow_log()
421+{
422+#ifdef EXTENDED_SLOWLOG
423+ return((ibool) thd_opt_slow_log());
424+#else
425+ return(FALSE);
426+#endif
427+}
428+
429 /*********************************************************************//**
430 Construct ha_innobase handler. */
431 UNIV_INTERN
432@@ -9179,6 +9215,25 @@
433 statement has ended */
434
435 if (trx->n_mysql_tables_in_use == 0) {
436+#ifdef EXTENDED_SLOWLOG
437+ increment_thd_innodb_stats(thd,
438+ (unsigned long long) trx->id,
439+ trx->io_reads,
440+ trx->io_read,
441+ trx->io_reads_wait_timer,
442+ trx->lock_que_wait_timer,
443+ trx->innodb_que_wait_timer,
444+ trx->distinct_page_access);
445+
446+ trx->io_reads = 0;
447+ trx->io_read = 0;
448+ trx->io_reads_wait_timer = 0;
449+ trx->lock_que_wait_timer = 0;
450+ trx->innodb_que_wait_timer = 0;
451+ trx->distinct_page_access = 0;
452+ if (trx->distinct_page_access_hash)
453+ memset(trx->distinct_page_access_hash, 0, DPAH_SIZE);
454+#endif
455
456 trx->mysql_n_tables_locked = 0;
457 prebuilt->used_in_HANDLER = FALSE;
458diff -ruN a/storage/innobase/handler/innodb_patch_info.h b/storage/innobase/handler/innodb_patch_info.h
459--- a/storage/innobase/handler/innodb_patch_info.h 2010-12-03 17:36:44.293955189 +0900
460+++ b/storage/innobase/handler/innodb_patch_info.h 2010-12-03 17:42:42.094955866 +0900
461@@ -38,5 +38,6 @@
462 {"innodb_recovery_patches","Bugfixes and adjustments about recovery process","","http://www.percona.com/docs/wiki/percona-xtradb"},
463 {"innodb_admin_command_base","XtraDB specific command interface through i_s","","http://www.percona.com/docs/wiki/percona-xtradb"},
464 {"innodb_show_lock_name","Show mutex/lock name instead of crated file/line","","http://www.percona.com/docs/wiki/percona-xtradb"},
465+{"innodb_extend_slow","Extended statistics in slow.log","It is InnoDB-part only. It needs to patch also to mysqld.","http://www.percona.com/docs/wiki/percona-xtradb"},
466 {NULL, NULL, NULL, NULL}
467 };
468diff -ruN a/storage/innobase/include/buf0rea.h b/storage/innobase/include/buf0rea.h
469--- a/storage/innobase/include/buf0rea.h 2010-12-03 15:18:48.891024406 +0900
470+++ b/storage/innobase/include/buf0rea.h 2010-12-03 17:42:42.096026873 +0900
471@@ -27,6 +27,7 @@
472 #define buf0rea_h
473
474 #include "univ.i"
475+#include "trx0types.h"
476 #include "buf0types.h"
477
478 /********************************************************************//**
479@@ -41,7 +42,8 @@
480 /*==========*/
481 ulint space, /*!< in: space id */
482 ulint zip_size,/*!< in: compressed page size in bytes, or 0 */
483- ulint offset);/*!< in: page number */
484+ ulint offset, /*!< in: page number */
485+ trx_t* trx);
486 /********************************************************************//**
487 Applies linear read-ahead if in the buf_pool the page is a border page of
488 a linear read-ahead area and all the pages in the area have been accessed.
489@@ -72,8 +74,9 @@
490 /*==================*/
491 ulint space, /*!< in: space id */
492 ulint zip_size,/*!< in: compressed page size in bytes, or 0 */
493- ulint offset);/*!< in: page number of a page; NOTE: the current thread
494+ ulint offset, /*!< in: page number of a page; NOTE: the current thread
495 must want access to this page (see NOTE 3 above) */
496+ trx_t* trx);
497 /********************************************************************//**
498 Issues read requests for pages which the ibuf module wants to read in, in
499 order to contract the insert buffer tree. Technically, this function is like
500diff -ruN a/storage/innobase/include/fil0fil.h b/storage/innobase/include/fil0fil.h
501--- a/storage/innobase/include/fil0fil.h 2010-12-03 15:09:51.290958543 +0900
502+++ b/storage/innobase/include/fil0fil.h 2010-12-03 17:42:42.097027548 +0900
503@@ -611,9 +611,12 @@
504 Reads or writes data. This operation is asynchronous (aio).
505 @return DB_SUCCESS, or DB_TABLESPACE_DELETED if we are trying to do
506 i/o on a tablespace which does not exist */
507+#define fil_io(type, sync, space_id, zip_size, block_offset, byte_offset, len, buf, message) \
508+ _fil_io(type, sync, space_id, zip_size, block_offset, byte_offset, len, buf, message, NULL)
509+
510 UNIV_INTERN
511 ulint
512-fil_io(
513+_fil_io(
514 /*===*/
515 ulint type, /*!< in: OS_FILE_READ or OS_FILE_WRITE,
516 ORed to OS_FILE_LOG, if a log i/o
517@@ -638,8 +641,9 @@
518 void* buf, /*!< in/out: buffer where to store read data
519 or from where to write; in aio this must be
520 appropriately aligned */
521- void* message); /*!< in: message for aio handler if non-sync
522+ void* message, /*!< in: message for aio handler if non-sync
523 aio used, else ignored */
524+ trx_t* trx);
525 /**********************************************************************//**
526 Waits for an aio operation to complete. This function is used to write the
527 handler for completed requests. The aio array of pending requests is divided
528diff -ruN a/storage/innobase/include/os0file.h b/storage/innobase/include/os0file.h
529--- a/storage/innobase/include/os0file.h 2010-11-03 07:01:13.000000000 +0900
530+++ b/storage/innobase/include/os0file.h 2010-12-03 17:42:42.100023783 +0900
531@@ -36,6 +36,7 @@
532 #define os0file_h
533
534 #include "univ.i"
535+#include "trx0types.h"
536
537 #ifndef __WIN__
538 #include <dirent.h>
539@@ -277,13 +278,17 @@
540 pfs_os_file_close_func(file, __FILE__, __LINE__)
541
542 # define os_aio(type, mode, name, file, buf, offset, offset_high, \
543- n, message1, message2) \
544+ n, message1, message2, trx) \
545 pfs_os_aio_func(type, mode, name, file, buf, offset, \
546- offset_high, n, message1, message2, \
547+ offset_high, n, message1, message2, trx, \
548 __FILE__, __LINE__)
549
550 # define os_file_read(file, buf, offset, offset_high, n) \
551- pfs_os_file_read_func(file, buf, offset, offset_high, n, \
552+ pfs_os_file_read_func(file, buf, offset, offset_high, n, NULL, \
553+ __FILE__, __LINE__)
554+
555+# define os_file_read_trx(file, buf, offset, offset_high, n, trx) \
556+ pfs_os_file_read_func(file, buf, offset, offset_high, n, trx, \
557 __FILE__, __LINE__)
558
559 # define os_file_read_no_error_handling(file, buf, offset, \
560@@ -319,12 +324,15 @@
561 # define os_file_close(file) os_file_close_func(file)
562
563 # define os_aio(type, mode, name, file, buf, offset, offset_high, \
564- n, message1, message2) \
565+ n, message1, message2, trx) \
566 os_aio_func(type, mode, name, file, buf, offset, offset_high, n,\
567- message1, message2)
568+ message1, message2, trx)
569
570 # define os_file_read(file, buf, offset, offset_high, n) \
571- os_file_read_func(file, buf, offset, offset_high, n)
572+ os_file_read_func(file, buf, offset, offset_high, n, NULL)
573+
574+# define os_file_read_trx(file, buf, offset, offset_high, n, trx) \
575+ os_file_read_func(file, buf, offset, offset_high, n, trx)
576
577 # define os_file_read_no_error_handling(file, buf, offset, \
578 offset_high, n) \
579@@ -690,6 +698,7 @@
580 ulint offset_high,/*!< in: most significant 32 bits of
581 offset */
582 ulint n, /*!< in: number of bytes to read */
583+ trx_t* trx,
584 const char* src_file,/*!< in: file name where func invoked */
585 ulint src_line);/*!< in: line where the func invoked */
586
587@@ -744,6 +753,7 @@
588 (can be used to identify a completed
589 aio operation); ignored if mode is
590 OS_AIO_SYNC */
591+ trx_t* trx,
592 const char* src_file,/*!< in: file name where func invoked */
593 ulint src_line);/*!< in: line where the func invoked */
594 /*******************************************************************//**
595@@ -885,7 +895,8 @@
596 offset where to read */
597 ulint offset_high,/*!< in: most significant 32 bits of
598 offset */
599- ulint n); /*!< in: number of bytes to read */
600+ ulint n, /*!< in: number of bytes to read */
601+ trx_t* trx);
602 /*******************************************************************//**
603 Rewind file to its start, read at most size - 1 bytes from it to str, and
604 NUL-terminate str. All errors are silently ignored. This function is
605@@ -1044,10 +1055,11 @@
606 (can be used to identify a completed
607 aio operation); ignored if mode is
608 OS_AIO_SYNC */
609- void* message2);/*!< in: message for the aio handler
610+ void* message2,/*!< in: message for the aio handler
611 (can be used to identify a completed
612 aio operation); ignored if mode is
613 OS_AIO_SYNC */
614+ trx_t* trx);
615 /************************************************************************//**
616 Wakes up all async i/o threads so that they know to exit themselves in
617 shutdown. */
618diff -ruN a/storage/innobase/include/os0file.ic b/storage/innobase/include/os0file.ic
619--- a/storage/innobase/include/os0file.ic 2010-11-03 07:01:13.000000000 +0900
620+++ b/storage/innobase/include/os0file.ic 2010-12-03 17:42:42.102024458 +0900
621@@ -229,6 +229,7 @@
622 (can be used to identify a completed
623 aio operation); ignored if mode is
624 OS_AIO_SYNC */
625+ trx_t* trx,
626 const char* src_file,/*!< in: file name where func invoked */
627 ulint src_line)/*!< in: line where the func invoked */
628 {
629@@ -244,7 +245,7 @@
630 src_file, src_line);
631
632 result = os_aio_func(type, mode, name, file, buf, offset, offset_high,
633- n, message1, message2);
634+ n, message1, message2, trx);
635
636 register_pfs_file_io_end(locker, n);
637
638@@ -268,6 +269,7 @@
639 ulint offset_high,/*!< in: most significant 32 bits of
640 offset */
641 ulint n, /*!< in: number of bytes to read */
642+ trx_t* trx,
643 const char* src_file,/*!< in: file name where func invoked */
644 ulint src_line)/*!< in: line where the func invoked */
645 {
646@@ -278,7 +280,7 @@
647 register_pfs_file_io_begin(&state, locker, file, n, PSI_FILE_READ,
648 src_file, src_line);
649
650- result = os_file_read_func(file, buf, offset, offset_high, n);
651+ result = os_file_read_func(file, buf, offset, offset_high, n, trx);
652
653 register_pfs_file_io_end(locker, n);
654
655diff -ruN a/storage/innobase/include/srv0srv.h b/storage/innobase/include/srv0srv.h
656--- a/storage/innobase/include/srv0srv.h 2010-12-03 17:32:15.634987408 +0900
657+++ b/storage/innobase/include/srv0srv.h 2010-12-03 17:42:42.104028644 +0900
658@@ -62,6 +62,9 @@
659 #define SRV_AUTO_EXTEND_INCREMENT \
660 (srv_auto_extend_increment * ((1024 * 1024) / UNIV_PAGE_SIZE))
661
662+/* prototypes for new functions added to ha_innodb.cc */
663+ibool innobase_get_slow_log();
664+
665 /* This is set to TRUE if the MySQL user has set it in MySQL */
666 extern ibool srv_lower_case_table_names;
667
668diff -ruN a/storage/innobase/include/trx0trx.h b/storage/innobase/include/trx0trx.h
669--- a/storage/innobase/include/trx0trx.h 2010-12-03 15:41:52.049372966 +0900
670+++ b/storage/innobase/include/trx0trx.h 2010-12-03 17:42:42.107024532 +0900
671@@ -728,6 +728,17 @@
672 /*------------------------------*/
673 char detailed_error[256]; /*!< detailed error message for last
674 error, or empty. */
675+ /*------------------------------*/
676+ ulint io_reads;
677+ ib_uint64_t io_read;
678+ ulint io_reads_wait_timer;
679+ ib_uint64_t lock_que_wait_ustarted;
680+ ulint lock_que_wait_timer;
681+ ulint innodb_que_wait_timer;
682+ ulint distinct_page_access;
683+#define DPAH_SIZE 8192
684+ byte* distinct_page_access_hash;
685+ ibool take_stats;
686 };
687
688 #define TRX_MAX_N_THREADS 32 /* maximum number of
689diff -ruN a/storage/innobase/lock/lock0lock.c b/storage/innobase/lock/lock0lock.c
690--- a/storage/innobase/lock/lock0lock.c 2010-12-03 15:09:51.297986437 +0900
691+++ b/storage/innobase/lock/lock0lock.c 2010-12-03 17:42:42.111024587 +0900
692@@ -1755,6 +1755,8 @@
693 {
694 lock_t* lock;
695 trx_t* trx;
696+ ulint sec;
697+ ulint ms;
698
699 ut_ad(mutex_own(&kernel_mutex));
700
701@@ -1813,6 +1815,10 @@
702 trx->que_state = TRX_QUE_LOCK_WAIT;
703 trx->was_chosen_as_deadlock_victim = FALSE;
704 trx->wait_started = time(NULL);
705+ if (innobase_get_slow_log() && trx->take_stats) {
706+ ut_usectime(&sec, &ms);
707+ trx->lock_que_wait_ustarted = (ib_uint64_t)sec * 1000000 + ms;
708+ }
709
710 ut_a(que_thr_stop(thr));
711
712@@ -3692,6 +3698,8 @@
713 {
714 lock_t* lock;
715 trx_t* trx;
716+ ulint sec;
717+ ulint ms;
718
719 ut_ad(mutex_own(&kernel_mutex));
720
721@@ -3747,6 +3755,10 @@
722 return(DB_SUCCESS);
723 }
724
725+ if (innobase_get_slow_log() && trx->take_stats) {
726+ ut_usectime(&sec, &ms);
727+ trx->lock_que_wait_ustarted = (ib_uint64_t)sec * 1000000 + ms;
728+ }
729 trx->que_state = TRX_QUE_LOCK_WAIT;
730 trx->was_chosen_as_deadlock_victim = FALSE;
731 trx->wait_started = time(NULL);
732diff -ruN a/storage/innobase/os/os0file.c b/storage/innobase/os/os0file.c
733--- a/storage/innobase/os/os0file.c 2010-12-03 17:32:15.644024974 +0900
734+++ b/storage/innobase/os/os0file.c 2010-12-03 17:42:42.117023467 +0900
735@@ -43,6 +43,8 @@
736 #include "srv0start.h"
737 #include "fil0fil.h"
738 #include "buf0buf.h"
739+#include "trx0sys.h"
740+#include "trx0trx.h"
741 #include "log0recv.h"
742 #ifndef UNIV_HOTBACKUP
743 # include "os0sync.h"
744@@ -2175,13 +2177,18 @@
745 ulint n, /*!< in: number of bytes to read */
746 ulint offset, /*!< in: least significant 32 bits of file
747 offset from where to read */
748- ulint offset_high) /*!< in: most significant 32 bits of
749+ ulint offset_high, /*!< in: most significant 32 bits of
750 offset */
751+ trx_t* trx)
752 {
753 off_t offs;
754 #if defined(HAVE_PREAD) && !defined(HAVE_BROKEN_PREAD)
755 ssize_t n_bytes;
756 #endif /* HAVE_PREAD && !HAVE_BROKEN_PREAD */
757+ ulint sec;
758+ ulint ms;
759+ ib_uint64_t start_time;
760+ ib_uint64_t finish_time;
761
762 ut_a((offset & 0xFFFFFFFFUL) == offset);
763
764@@ -2202,6 +2209,15 @@
765
766 os_n_file_reads++;
767
768+ if (innobase_get_slow_log() && trx && trx->take_stats)
769+ {
770+ trx->io_reads++;
771+ trx->io_read += n;
772+ ut_usectime(&sec, &ms);
773+ start_time = (ib_uint64_t)sec * 1000000 + ms;
774+ } else {
775+ start_time = 0;
776+ }
777 #if defined(HAVE_PREAD) && !defined(HAVE_BROKEN_PREAD)
778 os_mutex_enter(os_file_count_mutex);
779 os_file_n_pending_preads++;
780@@ -2215,6 +2231,13 @@
781 os_n_pending_reads--;
782 os_mutex_exit(os_file_count_mutex);
783
784+ if (innobase_get_slow_log() && trx && trx->take_stats && start_time)
785+ {
786+ ut_usectime(&sec, &ms);
787+ finish_time = (ib_uint64_t)sec * 1000000 + ms;
788+ trx->io_reads_wait_timer += (ulint)(finish_time - start_time);
789+ }
790+
791 return(n_bytes);
792 #else
793 {
794@@ -2251,6 +2274,13 @@
795 os_n_pending_reads--;
796 os_mutex_exit(os_file_count_mutex);
797
798+ if (innobase_get_slow_log() && trx && trx->take_stats && start_time)
799+ {
800+ ut_usectime(&sec, &ms);
801+ finish_time = (ib_uint64_t)sec * 1000000 + ms;
802+ trx->io_reads_wait_timer += (ulint)(finish_time - start_time);
803+ }
804+
805 return(ret);
806 }
807 #endif
808@@ -2391,7 +2421,8 @@
809 offset where to read */
810 ulint offset_high, /*!< in: most significant 32 bits of
811 offset */
812- ulint n) /*!< in: number of bytes to read */
813+ ulint n, /*!< in: number of bytes to read */
814+ trx_t* trx)
815 {
816 #ifdef __WIN__
817 BOOL ret;
818@@ -2463,7 +2494,7 @@
819 os_bytes_read_since_printout += n;
820
821 try_again:
822- ret = os_file_pread(file, buf, n, offset, offset_high);
823+ ret = os_file_pread(file, buf, n, offset, offset_high, trx);
824
825 if ((ulint)ret == n) {
826
827@@ -2589,7 +2620,7 @@
828 os_bytes_read_since_printout += n;
829
830 try_again:
831- ret = os_file_pread(file, buf, n, offset, offset_high);
832+ ret = os_file_pread(file, buf, n, offset, offset_high, NULL);
833
834 if ((ulint)ret == n) {
835
836@@ -3608,7 +3639,8 @@
837 offset */
838 ulint offset_high, /*!< in: most significant 32 bits of
839 offset */
840- ulint len) /*!< in: length of the block to read or write */
841+ ulint len, /*!< in: length of the block to read or write */
842+ trx_t* trx)
843 {
844 os_aio_slot_t* slot = NULL;
845 #ifdef WIN_ASYNC_IO
846@@ -3976,10 +4008,11 @@
847 (can be used to identify a completed
848 aio operation); ignored if mode is
849 OS_AIO_SYNC */
850- void* message2)/*!< in: message for the aio handler
851+ void* message2,/*!< in: message for the aio handler
852 (can be used to identify a completed
853 aio operation); ignored if mode is
854 OS_AIO_SYNC */
855+ trx_t* trx)
856 {
857 os_aio_array_t* array;
858 os_aio_slot_t* slot;
859@@ -4017,8 +4050,8 @@
860 wait in the Windows case. */
861
862 if (type == OS_FILE_READ) {
863- return(os_file_read(file, buf, offset,
864- offset_high, n));
865+ return(os_file_read_trx(file, buf, offset,
866+ offset_high, n, trx));
867 }
868
869 ut_a(type == OS_FILE_WRITE);
870@@ -4056,8 +4089,13 @@
871 ut_error;
872 }
873
874+ if (trx && type == OS_FILE_READ)
875+ {
876+ trx->io_reads++;
877+ trx->io_read += n;
878+ }
879 slot = os_aio_array_reserve_slot(type, array, message1, message2, file,
880- name, buf, offset, offset_high, n);
881+ name, buf, offset, offset_high, n, trx);
882 if (type == OS_FILE_READ) {
883 if (srv_use_native_aio) {
884 os_n_file_reads++;
885diff -ruN a/storage/innobase/srv/srv0srv.c b/storage/innobase/srv/srv0srv.c
886--- a/storage/innobase/srv/srv0srv.c 2010-12-03 17:32:15.648024399 +0900
887+++ b/storage/innobase/srv/srv0srv.c 2010-12-03 17:45:05.067023254 +0900
888@@ -87,6 +87,9 @@
889 #include "mysql/plugin.h"
890 #include "mysql/service_thd_wait.h"
891
892+/* prototypes for new functions added to ha_innodb.cc */
893+ibool innobase_get_slow_log();
894+
895 /* This is set to TRUE if the MySQL user has set it in MySQL; currently
896 affects only FOREIGN KEY definition parsing */
897 UNIV_INTERN ibool srv_lower_case_table_names = FALSE;
898@@ -1204,6 +1207,10 @@
899 ibool has_slept = FALSE;
900 srv_conc_slot_t* slot = NULL;
901 ulint i;
902+ ib_uint64_t start_time = 0L;
903+ ib_uint64_t finish_time = 0L;
904+ ulint sec;
905+ ulint ms;
906
907 if (trx->mysql_thd != NULL
908 && thd_is_replication_slave_thread(trx->mysql_thd)) {
909@@ -1280,6 +1287,7 @@
910 switches. */
911 if (SRV_THREAD_SLEEP_DELAY > 0) {
912 os_thread_sleep(SRV_THREAD_SLEEP_DELAY);
913+ trx->innodb_que_wait_timer += SRV_THREAD_SLEEP_DELAY;
914 }
915
916 trx->op_info = "";
917@@ -1335,6 +1343,13 @@
918 /* Go to wait for the event; when a thread leaves InnoDB it will
919 release this thread */
920
921+ if (innobase_get_slow_log() && trx->take_stats) {
922+ ut_usectime(&sec, &ms);
923+ start_time = (ib_uint64_t)sec * 1000000 + ms;
924+ } else {
925+ start_time = 0;
926+ }
927+
928 trx->op_info = "waiting in InnoDB queue";
929
930 thd_wait_begin(trx->mysql_thd, THD_WAIT_ROW_TABLE_LOCK);
931@@ -1343,6 +1358,12 @@
932
933 trx->op_info = "";
934
935+ if (innobase_get_slow_log() && trx->take_stats && start_time) {
936+ ut_usectime(&sec, &ms);
937+ finish_time = (ib_uint64_t)sec * 1000000 + ms;
938+ trx->innodb_que_wait_timer += (ulint)(finish_time - start_time);
939+ }
940+
941 os_fast_mutex_lock(&srv_conc_mutex);
942
943 srv_conc_n_waiting_threads--;
944diff -ruN a/storage/innobase/trx/trx0trx.c b/storage/innobase/trx/trx0trx.c
945--- a/storage/innobase/trx/trx0trx.c 2010-12-03 15:41:52.053955669 +0900
946+++ b/storage/innobase/trx/trx0trx.c 2010-12-03 17:42:42.127023410 +0900
947@@ -184,6 +184,15 @@
948 trx->global_read_view = NULL;
949 trx->read_view = NULL;
950
951+ trx->io_reads = 0;
952+ trx->io_read = 0;
953+ trx->io_reads_wait_timer = 0;
954+ trx->lock_que_wait_timer = 0;
955+ trx->innodb_que_wait_timer = 0;
956+ trx->distinct_page_access = 0;
957+ trx->distinct_page_access_hash = NULL;
958+ trx->take_stats = FALSE;
959+
960 /* Set X/Open XA transaction identification to NULL */
961 memset(&trx->xid, 0, sizeof(trx->xid));
962 trx->xid.formatID = -1;
963@@ -221,6 +230,11 @@
964
965 trx->mysql_process_no = os_proc_get_number();
966
967+ if (innobase_get_slow_log() && trx->take_stats) {
968+ trx->distinct_page_access_hash = mem_alloc(DPAH_SIZE);
969+ memset(trx->distinct_page_access_hash, 0, DPAH_SIZE);
970+ }
971+
972 return(trx);
973 }
974
975@@ -352,6 +366,12 @@
976 /*===============*/
977 trx_t* trx) /*!< in, own: trx object */
978 {
979+ if (trx->distinct_page_access_hash)
980+ {
981+ mem_free(trx->distinct_page_access_hash);
982+ trx->distinct_page_access_hash= NULL;
983+ }
984+
985 mutex_enter(&kernel_mutex);
986
987 UT_LIST_REMOVE(mysql_trx_list, trx_sys->mysql_trx_list, trx);
988@@ -373,6 +393,12 @@
989 /*====================*/
990 trx_t* trx) /*!< in, own: trx object */
991 {
992+ if (trx->distinct_page_access_hash)
993+ {
994+ mem_free(trx->distinct_page_access_hash);
995+ trx->distinct_page_access_hash= NULL;
996+ }
997+
998 mutex_enter(&kernel_mutex);
999
1000 trx_free(trx);
1001@@ -1094,6 +1120,9 @@
1002 trx_t* trx) /*!< in: transaction */
1003 {
1004 que_thr_t* thr;
1005+ ulint sec;
1006+ ulint ms;
1007+ ib_uint64_t now;
1008
1009 ut_ad(mutex_own(&kernel_mutex));
1010 ut_ad(trx->que_state == TRX_QUE_LOCK_WAIT);
1011@@ -1108,6 +1137,11 @@
1012 thr = UT_LIST_GET_FIRST(trx->wait_thrs);
1013 }
1014
1015+ if (innobase_get_slow_log() && trx->take_stats) {
1016+ ut_usectime(&sec, &ms);
1017+ now = (ib_uint64_t)sec * 1000000 + ms;
1018+ trx->lock_que_wait_timer += (ulint)(now - trx->lock_que_wait_ustarted);
1019+ }
1020 trx->que_state = TRX_QUE_RUNNING;
1021 }
1022
1023@@ -1121,6 +1155,9 @@
1024 trx_t* trx) /*!< in: transaction in the TRX_QUE_LOCK_WAIT state */
1025 {
1026 que_thr_t* thr;
1027+ ulint sec;
1028+ ulint ms;
1029+ ib_uint64_t now;
1030
1031 ut_ad(mutex_own(&kernel_mutex));
1032 ut_ad(trx->que_state == TRX_QUE_LOCK_WAIT);
1033@@ -1135,6 +1172,11 @@
1034 thr = UT_LIST_GET_FIRST(trx->wait_thrs);
1035 }
1036
1037+ if (innobase_get_slow_log() && trx->take_stats) {
1038+ ut_usectime(&sec, &ms);
1039+ now = (ib_uint64_t)sec * 1000000 + ms;
1040+ trx->lock_que_wait_timer += (ulint)(now - trx->lock_que_wait_ustarted);
1041+ }
1042 trx->que_state = TRX_QUE_RUNNING;
1043 }
1044
MySQL: a very fast and reliable SQL database engine
This page took 0.249167 seconds and 4 git commands to generate.