]> git.pld-linux.org Git - packages/mysql.git/blame_incremental - innodb_extend_slow.patch
projects / packages / mysql.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
- make mysql.init a bit more lsb-compatible
[packages/mysql.git] / innodb_extend_slow.patch
... / ...
CommitLineData
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 >= 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 > 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@@ -2403,8 +2437,16 @@
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+ if (innobase_get_slow_log()) {
64+ trx = innobase_get_trx();
65+ }
66 buf_pool->stat.n_page_gets++;
67
68 for (;;) {
69@@ -2422,7 +2464,7 @@
70 //buf_pool_mutex_exit(buf_pool);
71 rw_lock_s_unlock(&buf_pool->page_hash_latch);
72
73- buf_read_page(space, zip_size, offset);
74+ buf_read_page(space, zip_size, offset, trx);
75
76 #if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
77 ut_a(++buf_dbg_counter % 37 || buf_validate());
78@@ -2518,6 +2560,13 @@
79 /* Let us wait until the read operation
80 completes */
81
82+ if (innobase_get_slow_log() && trx && trx->take_stats)
83+ {
84+ ut_usectime(&sec, &ms);
85+ start_time = (ib_uint64_t)sec * 1000000 + ms;
86+ } else {
87+ start_time = 0;
88+ }
89 for (;;) {
90 enum buf_io_fix io_fix;
91
92@@ -2532,6 +2581,12 @@
93 break;
94 }
95 }
96+ if (innobase_get_slow_log() && trx && trx->take_stats && start_time)
97+ {
98+ ut_usectime(&sec, &ms);
99+ finish_time = (ib_uint64_t)sec * 1000000 + ms;
100+ trx->io_reads_wait_timer += (ulint)(finish_time - start_time);
101+ }
102 }
103
104 #ifdef UNIV_IBUF_COUNT_DEBUG
105@@ -2847,6 +2902,11 @@
106 ibool must_read;
107 ulint retries = 0;
108 mutex_t* block_mutex = NULL;
109+ trx_t* trx = NULL;
110+ ulint sec;
111+ ulint ms;
112+ ib_uint64_t start_time;
113+ ib_uint64_t finish_time;
114 buf_pool_t* buf_pool = buf_pool_get(space, offset);
115
116 ut_ad(mtr);
117@@ -2875,6 +2935,9 @@
118 || ibuf_page_low(space, zip_size, offset,
119 FALSE, file, line, NULL));
120 #endif
121+ if (innobase_get_slow_log()) {
122+ trx = innobase_get_trx();
123+ }
124 buf_pool->stat.n_page_gets++;
125 fold = buf_page_address_fold(space, offset);
126 loop:
127@@ -2949,7 +3012,7 @@
128 return(NULL);
129 }
130
131- if (buf_read_page(space, zip_size, offset)) {
132+ if (buf_read_page(space, zip_size, offset, trx)) {
133 retries = 0;
134 } else if (retries < BUF_PAGE_READ_MAX_RETRIES) {
135 ++retries;
136@@ -3258,6 +3321,13 @@
137 /* Let us wait until the read operation
138 completes */
139
140+ if (innobase_get_slow_log() && trx && trx->take_stats)
141+ {
142+ ut_usectime(&sec, &ms);
143+ start_time = (ib_uint64_t)sec * 1000000 + ms;
144+ } else {
145+ start_time = 0;
146+ }
147 for (;;) {
148 enum buf_io_fix io_fix;
149
150@@ -3272,6 +3342,12 @@
151 break;
152 }
153 }
154+ if (innobase_get_slow_log() && trx && trx->take_stats && start_time)
155+ {
156+ ut_usectime(&sec, &ms);
157+ finish_time = (ib_uint64_t)sec * 1000000 + ms;
158+ trx->io_reads_wait_timer += (ulint)(finish_time - start_time);
159+ }
160 }
161
162 fix_type = MTR_MEMO_BUF_FIX;
163@@ -3298,13 +3374,17 @@
164 read-ahead */
165
166 buf_read_ahead_linear(space, zip_size, offset,
167- ibuf_inside(mtr));
168+ ibuf_inside(mtr), trx);
169 }
170
171 #ifdef UNIV_IBUF_COUNT_DEBUG
172 ut_a(ibuf_count_get(buf_block_get_space(block),
173 buf_block_get_page_no(block)) == 0);
174 #endif
175+ if (innobase_get_slow_log()) {
176+ _increment_page_get_statistics(block, trx);
177+ }
178+
179 return(block);
180 }
181
182@@ -3328,6 +3408,7 @@
183 unsigned access_time;
184 ibool success;
185 ulint fix_type;
186+ trx_t* trx = NULL;
187
188 ut_ad(block);
189 ut_ad(mtr);
190@@ -3405,6 +3486,10 @@
191 #if defined UNIV_DEBUG_FILE_ACCESSES || defined UNIV_DEBUG
192 ut_a(block->page.file_page_was_freed == FALSE);
193 #endif
194+ if (innobase_get_slow_log()) {
195+ trx = innobase_get_trx();
196+ }
197+
198 if (UNIV_UNLIKELY(!access_time)) {
199 /* In the case of a first access, try to apply linear
200 read-ahead */
201@@ -3412,7 +3497,7 @@
202 buf_read_ahead_linear(buf_block_get_space(block),
203 buf_block_get_zip_size(block),
204 buf_block_get_page_no(block),
205- ibuf_inside(mtr));
206+ ibuf_inside(mtr), trx);
207 }
208
209 #ifdef UNIV_IBUF_COUNT_DEBUG
210@@ -3422,6 +3507,9 @@
211 buf_pool = buf_pool_from_block(block);
212 buf_pool->stat.n_page_gets++;
213
214+ if (innobase_get_slow_log()) {
215+ _increment_page_get_statistics(block, trx);
216+ }
217 return(TRUE);
218 }
219
220@@ -3444,6 +3532,7 @@
221 buf_pool_t* buf_pool;
222 ibool success;
223 ulint fix_type;
224+ trx_t* trx = NULL;
225
226 ut_ad(mtr);
227 ut_ad(mtr->state == MTR_ACTIVE);
228@@ -3530,6 +3619,11 @@
229 #endif
230 buf_pool->stat.n_page_gets++;
231
232+ if (innobase_get_slow_log()) {
233+ trx = innobase_get_trx();
234+ _increment_page_get_statistics(block, trx);
235+ }
236+
237 return(TRUE);
238 }
239
240diff -ruN a/storage/innobase/buf/buf0rea.c b/storage/innobase/buf/buf0rea.c
241--- a/storage/innobase/buf/buf0rea.c 2010-12-03 17:32:15.617037263 +0900
242+++ b/storage/innobase/buf/buf0rea.c 2010-12-03 17:42:42.075297193 +0900
243@@ -77,7 +77,8 @@
244 treat the tablespace as dropped; this is a timestamp we
245 use to stop dangling page reads from a tablespace
246 which we have DISCARDed + IMPORTed back */
247- ulint offset) /*!< in: page number */
248+ ulint offset, /*!< in: page number */
249+ trx_t* trx)
250 {
251 buf_page_t* bpage;
252 ulint wake_later;
253@@ -179,15 +180,15 @@
254
255 thd_wait_begin(NULL, THD_WAIT_DISKIO);
256 if (zip_size) {
257- *err = fil_io(OS_FILE_READ | wake_later,
258+ *err = _fil_io(OS_FILE_READ | wake_later,
259 sync, space, zip_size, offset, 0, zip_size,
260- bpage->zip.data, bpage);
261+ bpage->zip.data, bpage, trx);
262 } else {
263 ut_a(buf_page_get_state(bpage) == BUF_BLOCK_FILE_PAGE);
264
265- *err = fil_io(OS_FILE_READ | wake_later,
266+ *err = _fil_io(OS_FILE_READ | wake_later,
267 sync, space, 0, offset, 0, UNIV_PAGE_SIZE,
268- ((buf_block_t*) bpage)->frame, bpage);
269+ ((buf_block_t*) bpage)->frame, bpage, trx);
270 }
271 thd_wait_end(NULL);
272 ut_a(*err == DB_SUCCESS);
273@@ -213,7 +214,8 @@
274 /*==========*/
275 ulint space, /*!< in: space id */
276 ulint zip_size,/*!< in: compressed page size in bytes, or 0 */
277- ulint offset) /*!< in: page number */
278+ ulint offset, /*!< in: page number */
279+ trx_t* trx)
280 {
281 buf_pool_t* buf_pool = buf_pool_get(space, offset);
282 ib_int64_t tablespace_version;
283@@ -227,7 +229,7 @@
284
285 count = buf_read_page_low(&err, TRUE, BUF_READ_ANY_PAGE, space,
286 zip_size, FALSE,
287- tablespace_version, offset);
288+ tablespace_version, offset, trx);
289 srv_buf_pool_reads += count;
290 if (err == DB_TABLESPACE_DELETED) {
291 ut_print_timestamp(stderr);
292@@ -279,7 +281,8 @@
293 ulint space, /*!< in: space id */
294 ulint zip_size, /*!< in: compressed page size in bytes, or 0 */
295 ulint offset, /*!< in: page number; see NOTE 3 above */
296- ibool inside_ibuf) /*!< in: TRUE if we are inside ibuf routine */
297+ ibool inside_ibuf, /*!< in: TRUE if we are inside ibuf routine */
298+ trx_t* trx)
299 {
300 buf_pool_t* buf_pool = buf_pool_get(space, offset);
301 ib_int64_t tablespace_version;
302@@ -498,7 +501,7 @@
303 count += buf_read_page_low(
304 &err, FALSE,
305 ibuf_mode,
306- space, zip_size, FALSE, tablespace_version, i);
307+ space, zip_size, FALSE, tablespace_version, i, trx);
308 if (err == DB_TABLESPACE_DELETED) {
309 ut_print_timestamp(stderr);
310 fprintf(stderr,
311@@ -591,7 +594,7 @@
312 buf_read_page_low(&err, sync && (i + 1 == n_stored),
313 BUF_READ_ANY_PAGE, space_ids[i],
314 zip_size, TRUE, space_versions[i],
315- page_nos[i]);
316+ page_nos[i], NULL);
317
318 if (UNIV_UNLIKELY(err == DB_TABLESPACE_DELETED)) {
319 tablespace_deleted:
320@@ -733,12 +736,12 @@
321 if ((i + 1 == n_stored) && sync) {
322 buf_read_page_low(&err, TRUE, BUF_READ_ANY_PAGE, space,
323 zip_size, TRUE, tablespace_version,
324- page_nos[i]);
325+ page_nos[i], NULL);
326 } else {
327 buf_read_page_low(&err, FALSE, BUF_READ_ANY_PAGE
328 | OS_AIO_SIMULATED_WAKE_LATER,
329 space, zip_size, TRUE,
330- tablespace_version, page_nos[i]);
331+ tablespace_version, page_nos[i], NULL);
332 }
333 }
334
335diff -ruN a/storage/innobase/fil/fil0fil.c b/storage/innobase/fil/fil0fil.c
336--- a/storage/innobase/fil/fil0fil.c 2010-12-03 15:53:54.610037199 +0900
337+++ b/storage/innobase/fil/fil0fil.c 2010-12-03 17:42:42.079064198 +0900
338@@ -4747,7 +4747,7 @@
339 node->name, node->handle, buf,
340 offset_low, offset_high,
341 page_size * n_pages,
342- NULL, NULL);
343+ NULL, NULL, NULL);
344 #endif
345 if (success) {
346 node->size += n_pages;
347@@ -5074,7 +5074,7 @@
348 i/o on a tablespace which does not exist */
349 UNIV_INTERN
350 ulint
351-fil_io(
352+_fil_io(
353 /*===*/
354 ulint type, /*!< in: OS_FILE_READ or OS_FILE_WRITE,
355 ORed to OS_FILE_LOG, if a log i/o
356@@ -5099,8 +5099,9 @@
357 void* buf, /*!< in/out: buffer where to store read data
358 or from where to write; in aio this must be
359 appropriately aligned */
360- void* message) /*!< in: message for aio handler if non-sync
361+ void* message, /*!< in: message for aio handler if non-sync
362 aio used, else ignored */
363+ trx_t* trx)
364 {
365 ulint mode;
366 fil_space_t* space;
367@@ -5268,7 +5269,7 @@
368 #else
369 /* Queue the aio request */
370 ret = os_aio(type, mode | wake_later, node->name, node->handle, buf,
371- offset_low, offset_high, len, node, message);
372+ offset_low, offset_high, len, node, message, trx);
373 #endif
374 ut_a(ret);
375
376diff -ruN a/storage/innobase/handler/ha_innodb.cc b/storage/innobase/handler/ha_innodb.cc
377--- a/storage/innobase/handler/ha_innodb.cc 2010-12-03 17:36:44.293955189 +0900
378+++ b/storage/innobase/handler/ha_innodb.cc 2010-12-03 17:42:42.090024586 +0900
379@@ -1573,6 +1573,16 @@
380 trx->check_unique_secondary = !thd_test_options(
381 thd, OPTION_RELAXED_UNIQUE_CHECKS);
382
383+#ifdef EXTENDED_SLOWLOG
384+ if (thd_log_slow_verbosity(thd) & SLOG_V_INNODB) {
385+ trx->take_stats = TRUE;
386+ } else {
387+ trx->take_stats = FALSE;
388+ }
389+#else
390+ trx->take_stats = FALSE;
391+#endif
392+
393 DBUG_VOID_RETURN;
394 }
395
396@@ -1627,6 +1637,32 @@
397 return(trx);
398 }
399
400+/*************************************************************************
401+Gets current trx. */
402+extern "C"
403+trx_t*
404+innobase_get_trx()
405+{
406+ THD *thd=current_thd;
407+ if (likely(thd != 0)) {
408+ trx_t*& trx = thd_to_trx(thd);
409+ return(trx);
410+ } else {
411+ return(NULL);
412+ }
413+}
414+
415+extern "C"
416+ibool
417+innobase_get_slow_log()
418+{
419+#ifdef EXTENDED_SLOWLOG
420+ return((ibool) thd_opt_slow_log());
421+#else
422+ return(FALSE);
423+#endif
424+}
425+
426 /*********************************************************************//**
427 Note that a transaction has been registered with MySQL.
428 @return true if transaction is registered with MySQL 2PC coordinator */
429@@ -9301,6 +9337,25 @@
430 statement has ended */
431
432 if (trx->n_mysql_tables_in_use == 0) {
433+#ifdef EXTENDED_SLOWLOG
434+ increment_thd_innodb_stats(thd,
435+ (unsigned long long) trx->id,
436+ trx->io_reads,
437+ trx->io_read,
438+ trx->io_reads_wait_timer,
439+ trx->lock_que_wait_timer,
440+ trx->innodb_que_wait_timer,
441+ trx->distinct_page_access);
442+
443+ trx->io_reads = 0;
444+ trx->io_read = 0;
445+ trx->io_reads_wait_timer = 0;
446+ trx->lock_que_wait_timer = 0;
447+ trx->innodb_que_wait_timer = 0;
448+ trx->distinct_page_access = 0;
449+ if (trx->distinct_page_access_hash)
450+ memset(trx->distinct_page_access_hash, 0, DPAH_SIZE);
451+#endif
452
453 trx->mysql_n_tables_locked = 0;
454 prebuilt->used_in_HANDLER = FALSE;
455diff -ruN a/storage/innobase/include/buf0rea.h b/storage/innobase/include/buf0rea.h
456--- a/storage/innobase/include/buf0rea.h 2010-12-03 15:18:48.891024406 +0900
457+++ b/storage/innobase/include/buf0rea.h 2010-12-03 17:42:42.096026873 +0900
458@@ -27,6 +27,7 @@
459 #define buf0rea_h
460
461 #include "univ.i"
462+#include "trx0types.h"
463 #include "buf0types.h"
464
465 /********************************************************************//**
466@@ -41,7 +42,8 @@
467 /*==========*/
468 ulint space, /*!< in: space id */
469 ulint zip_size,/*!< in: compressed page size in bytes, or 0 */
470- ulint offset);/*!< in: page number */
471+ ulint offset, /*!< in: page number */
472+ trx_t* trx);
473 /********************************************************************//**
474 Applies linear read-ahead if in the buf_pool the page is a border page of
475 a linear read-ahead area and all the pages in the area have been accessed.
476@@ -73,7 +75,8 @@
477 ulint space, /*!< in: space id */
478 ulint zip_size, /*!< in: compressed page size in bytes, or 0 */
479 ulint offset, /*!< in: page number; see NOTE 3 above */
480- ibool inside_ibuf); /*!< in: TRUE if we are inside ibuf routine */
481+ ibool inside_ibuf, /*!< in: TRUE if we are inside ibuf routine */
482+ trx_t* trx);
483 /********************************************************************//**
484 Issues read requests for pages which the ibuf module wants to read in, in
485 order to contract the insert buffer tree. Technically, this function is like
486diff -ruN a/storage/innobase/include/fil0fil.h b/storage/innobase/include/fil0fil.h
487--- a/storage/innobase/include/fil0fil.h 2010-12-03 15:09:51.290958543 +0900
488+++ b/storage/innobase/include/fil0fil.h 2010-12-03 17:42:42.097027548 +0900
489@@ -611,9 +611,12 @@
490 Reads or writes data. This operation is asynchronous (aio).
491 @return DB_SUCCESS, or DB_TABLESPACE_DELETED if we are trying to do
492 i/o on a tablespace which does not exist */
493+#define fil_io(type, sync, space_id, zip_size, block_offset, byte_offset, len, buf, message) \
494+ _fil_io(type, sync, space_id, zip_size, block_offset, byte_offset, len, buf, message, NULL)
495+
496 UNIV_INTERN
497 ulint
498-fil_io(
499+_fil_io(
500 /*===*/
501 ulint type, /*!< in: OS_FILE_READ or OS_FILE_WRITE,
502 ORed to OS_FILE_LOG, if a log i/o
503@@ -638,8 +641,9 @@
504 void* buf, /*!< in/out: buffer where to store read data
505 or from where to write; in aio this must be
506 appropriately aligned */
507- void* message); /*!< in: message for aio handler if non-sync
508+ void* message, /*!< in: message for aio handler if non-sync
509 aio used, else ignored */
510+ trx_t* trx);
511 /**********************************************************************//**
512 Waits for an aio operation to complete. This function is used to write the
513 handler for completed requests. The aio array of pending requests is divided
514diff -ruN a/storage/innobase/include/os0file.h b/storage/innobase/include/os0file.h
515--- a/storage/innobase/include/os0file.h 2010-11-03 07:01:13.000000000 +0900
516+++ b/storage/innobase/include/os0file.h 2010-12-03 17:42:42.100023783 +0900
517@@ -36,6 +36,7 @@
518 #define os0file_h
519
520 #include "univ.i"
521+#include "trx0types.h"
522
523 #ifndef __WIN__
524 #include <dirent.h>
525@@ -277,13 +278,17 @@
526 pfs_os_file_close_func(file, __FILE__, __LINE__)
527
528 # define os_aio(type, mode, name, file, buf, offset, offset_high, \
529- n, message1, message2) \
530+ n, message1, message2, trx) \
531 pfs_os_aio_func(type, mode, name, file, buf, offset, \
532- offset_high, n, message1, message2, \
533+ offset_high, n, message1, message2, trx, \
534 __FILE__, __LINE__)
535
536 # define os_file_read(file, buf, offset, offset_high, n) \
537- pfs_os_file_read_func(file, buf, offset, offset_high, n, \
538+ pfs_os_file_read_func(file, buf, offset, offset_high, n, NULL, \
539+ __FILE__, __LINE__)
540+
541+# define os_file_read_trx(file, buf, offset, offset_high, n, trx) \
542+ pfs_os_file_read_func(file, buf, offset, offset_high, n, trx, \
543 __FILE__, __LINE__)
544
545 # define os_file_read_no_error_handling(file, buf, offset, \
546@@ -319,12 +324,15 @@
547 # define os_file_close(file) os_file_close_func(file)
548
549 # define os_aio(type, mode, name, file, buf, offset, offset_high, \
550- n, message1, message2) \
551+ n, message1, message2, trx) \
552 os_aio_func(type, mode, name, file, buf, offset, offset_high, n,\
553- message1, message2)
554+ message1, message2, trx)
555
556 # define os_file_read(file, buf, offset, offset_high, n) \
557- os_file_read_func(file, buf, offset, offset_high, n)
558+ os_file_read_func(file, buf, offset, offset_high, n, NULL)
559+
560+# define os_file_read_trx(file, buf, offset, offset_high, n, trx) \
561+ os_file_read_func(file, buf, offset, offset_high, n, trx)
562
563 # define os_file_read_no_error_handling(file, buf, offset, \
564 offset_high, n) \
565@@ -692,6 +700,7 @@
566 ulint offset_high,/*!< in: most significant 32 bits of
567 offset */
568 ulint n, /*!< in: number of bytes to read */
569+ trx_t* trx,
570 const char* src_file,/*!< in: file name where func invoked */
571 ulint src_line);/*!< in: line where the func invoked */
572
573@@ -746,6 +755,7 @@
574 (can be used to identify a completed
575 aio operation); ignored if mode is
576 OS_AIO_SYNC */
577+ trx_t* trx,
578 const char* src_file,/*!< in: file name where func invoked */
579 ulint src_line);/*!< in: line where the func invoked */
580 /*******************************************************************//**
581@@ -889,7 +899,8 @@
582 offset where to read */
583 ulint offset_high,/*!< in: most significant 32 bits of
584 offset */
585- ulint n); /*!< in: number of bytes to read */
586+ ulint n, /*!< in: number of bytes to read */
587+ trx_t* trx);
588 /*******************************************************************//**
589 Rewind file to its start, read at most size - 1 bytes from it to str, and
590 NUL-terminate str. All errors are silently ignored. This function is
591@@ -1048,10 +1059,11 @@
592 (can be used to identify a completed
593 aio operation); ignored if mode is
594 OS_AIO_SYNC */
595- void* message2);/*!< in: message for the aio handler
596+ void* message2,/*!< in: message for the aio handler
597 (can be used to identify a completed
598 aio operation); ignored if mode is
599 OS_AIO_SYNC */
600+ trx_t* trx);
601 /************************************************************************//**
602 Wakes up all async i/o threads so that they know to exit themselves in
603 shutdown. */
604diff -ruN a/storage/innobase/include/os0file.ic b/storage/innobase/include/os0file.ic
605--- a/storage/innobase/include/os0file.ic 2010-11-03 07:01:13.000000000 +0900
606+++ b/storage/innobase/include/os0file.ic 2010-12-03 17:42:42.102024458 +0900
607@@ -229,6 +229,7 @@
608 (can be used to identify a completed
609 aio operation); ignored if mode is
610 OS_AIO_SYNC */
611+ trx_t* trx,
612 const char* src_file,/*!< in: file name where func invoked */
613 ulint src_line)/*!< in: line where the func invoked */
614 {
615@@ -244,7 +245,7 @@
616 src_file, src_line);
617
618 result = os_aio_func(type, mode, name, file, buf, offset, offset_high,
619- n, message1, message2);
620+ n, message1, message2, trx);
621
622 register_pfs_file_io_end(locker, n);
623
624@@ -268,6 +269,7 @@
625 ulint offset_high,/*!< in: most significant 32 bits of
626 offset */
627 ulint n, /*!< in: number of bytes to read */
628+ trx_t* trx,
629 const char* src_file,/*!< in: file name where func invoked */
630 ulint src_line)/*!< in: line where the func invoked */
631 {
632@@ -278,7 +280,7 @@
633 register_pfs_file_io_begin(&state, locker, file, n, PSI_FILE_READ,
634 src_file, src_line);
635
636- result = os_file_read_func(file, buf, offset, offset_high, n);
637+ result = os_file_read_func(file, buf, offset, offset_high, n, trx);
638
639 register_pfs_file_io_end(locker, n);
640
641diff -ruN a/storage/innobase/include/srv0srv.h b/storage/innobase/include/srv0srv.h
642--- a/storage/innobase/include/srv0srv.h 2010-12-03 17:32:15.634987408 +0900
643+++ b/storage/innobase/include/srv0srv.h 2010-12-03 17:42:42.104028644 +0900
644@@ -71,6 +71,9 @@
645 #define SRV_AUTO_EXTEND_INCREMENT \
646 (srv_auto_extend_increment * ((1024 * 1024) / UNIV_PAGE_SIZE))
647
648+/* prototypes for new functions added to ha_innodb.cc */
649+ibool innobase_get_slow_log();
650+
651 /* Mutex for locking srv_monitor_file */
652 extern mutex_t srv_monitor_file_mutex;
653 /* Temporary file for innodb monitor output */
654diff -ruN a/storage/innobase/include/trx0trx.h b/storage/innobase/include/trx0trx.h
655--- a/storage/innobase/include/trx0trx.h 2010-12-03 15:41:52.049372966 +0900
656+++ b/storage/innobase/include/trx0trx.h 2010-12-03 17:42:42.107024532 +0900
657@@ -743,6 +743,17 @@
658 /*------------------------------*/
659 char detailed_error[256]; /*!< detailed error message for last
660 error, or empty. */
661+ /*------------------------------*/
662+ ulint io_reads;
663+ ib_uint64_t io_read;
664+ ulint io_reads_wait_timer;
665+ ib_uint64_t lock_que_wait_ustarted;
666+ ulint lock_que_wait_timer;
667+ ulint innodb_que_wait_timer;
668+ ulint distinct_page_access;
669+#define DPAH_SIZE 8192
670+ byte* distinct_page_access_hash;
671+ ibool take_stats;
672 };
673
674 #define TRX_MAX_N_THREADS 32 /* maximum number of
675diff -ruN a/storage/innobase/lock/lock0lock.c b/storage/innobase/lock/lock0lock.c
676--- a/storage/innobase/lock/lock0lock.c 2010-12-03 15:09:51.297986437 +0900
677+++ b/storage/innobase/lock/lock0lock.c 2010-12-03 17:42:42.111024587 +0900
678@@ -1755,6 +1755,8 @@
679 {
680 lock_t* lock;
681 trx_t* trx;
682+ ulint sec;
683+ ulint ms;
684
685 ut_ad(mutex_own(&kernel_mutex));
686
687@@ -1813,6 +1815,10 @@
688 trx->que_state = TRX_QUE_LOCK_WAIT;
689 trx->was_chosen_as_deadlock_victim = FALSE;
690 trx->wait_started = time(NULL);
691+ if (innobase_get_slow_log() && trx->take_stats) {
692+ ut_usectime(&sec, &ms);
693+ trx->lock_que_wait_ustarted = (ib_uint64_t)sec * 1000000 + ms;
694+ }
695
696 ut_a(que_thr_stop(thr));
697
698@@ -3764,6 +3770,8 @@
699 {
700 lock_t* lock;
701 trx_t* trx;
702+ ulint sec;
703+ ulint ms;
704
705 ut_ad(mutex_own(&kernel_mutex));
706
707@@ -3819,6 +3827,10 @@
708 return(DB_SUCCESS);
709 }
710
711+ if (innobase_get_slow_log() && trx->take_stats) {
712+ ut_usectime(&sec, &ms);
713+ trx->lock_que_wait_ustarted = (ib_uint64_t)sec * 1000000 + ms;
714+ }
715 trx->que_state = TRX_QUE_LOCK_WAIT;
716 trx->was_chosen_as_deadlock_victim = FALSE;
717 trx->wait_started = time(NULL);
718diff -ruN a/storage/innobase/os/os0file.c b/storage/innobase/os/os0file.c
719--- a/storage/innobase/os/os0file.c 2010-12-03 17:32:15.644024974 +0900
720+++ b/storage/innobase/os/os0file.c 2010-12-03 17:42:42.117023467 +0900
721@@ -43,6 +43,8 @@
722 #include "srv0start.h"
723 #include "fil0fil.h"
724 #include "buf0buf.h"
725+#include "trx0sys.h"
726+#include "trx0trx.h"
727 #include "log0recv.h"
728 #ifndef UNIV_HOTBACKUP
729 # include "os0sync.h"
730@@ -2212,13 +2214,18 @@
731 ulint n, /*!< in: number of bytes to read */
732 ulint offset, /*!< in: least significant 32 bits of file
733 offset from where to read */
734- ulint offset_high) /*!< in: most significant 32 bits of
735+ ulint offset_high, /*!< in: most significant 32 bits of
736 offset */
737+ trx_t* trx)
738 {
739 off_t offs;
740 #if defined(HAVE_PREAD) && !defined(HAVE_BROKEN_PREAD)
741 ssize_t n_bytes;
742 #endif /* HAVE_PREAD && !HAVE_BROKEN_PREAD */
743+ ulint sec;
744+ ulint ms;
745+ ib_uint64_t start_time;
746+ ib_uint64_t finish_time;
747
748 ut_a((offset & 0xFFFFFFFFUL) == offset);
749
750@@ -2239,6 +2246,15 @@
751
752 os_n_file_reads++;
753
754+ if (innobase_get_slow_log() && trx && trx->take_stats)
755+ {
756+ trx->io_reads++;
757+ trx->io_read += n;
758+ ut_usectime(&sec, &ms);
759+ start_time = (ib_uint64_t)sec * 1000000 + ms;
760+ } else {
761+ start_time = 0;
762+ }
763 #if defined(HAVE_PREAD) && !defined(HAVE_BROKEN_PREAD)
764 os_mutex_enter(os_file_count_mutex);
765 os_file_n_pending_preads++;
766@@ -2252,6 +2268,13 @@
767 os_n_pending_reads--;
768 os_mutex_exit(os_file_count_mutex);
769
770+ if (innobase_get_slow_log() && trx && trx->take_stats && start_time)
771+ {
772+ ut_usectime(&sec, &ms);
773+ finish_time = (ib_uint64_t)sec * 1000000 + ms;
774+ trx->io_reads_wait_timer += (ulint)(finish_time - start_time);
775+ }
776+
777 return(n_bytes);
778 #else
779 {
780@@ -2288,6 +2311,13 @@
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(ret);
792 }
793 #endif
794@@ -2428,7 +2458,8 @@
795 offset where to read */
796 ulint offset_high, /*!< in: most significant 32 bits of
797 offset */
798- ulint n) /*!< in: number of bytes to read */
799+ ulint n, /*!< in: number of bytes to read */
800+ trx_t* trx)
801 {
802 #ifdef __WIN__
803 BOOL ret;
804@@ -2503,7 +2534,7 @@
805 os_bytes_read_since_printout += n;
806
807 try_again:
808- ret = os_file_pread(file, buf, n, offset, offset_high);
809+ ret = os_file_pread(file, buf, n, offset, offset_high, trx);
810
811 if ((ulint)ret == n) {
812
813@@ -2632,7 +2663,7 @@
814 os_bytes_read_since_printout += n;
815
816 try_again:
817- ret = os_file_pread(file, buf, n, offset, offset_high);
818+ ret = os_file_pread(file, buf, n, offset, offset_high, NULL);
819
820 if ((ulint)ret == n) {
821
822@@ -4026,10 +4057,11 @@
823 (can be used to identify a completed
824 aio operation); ignored if mode is
825 OS_AIO_SYNC */
826- void* message2)/*!< in: message for the aio handler
827+ void* message2,/*!< in: message for the aio handler
828 (can be used to identify a completed
829 aio operation); ignored if mode is
830 OS_AIO_SYNC */
831+ trx_t* trx)
832 {
833 os_aio_array_t* array;
834 os_aio_slot_t* slot;
835@@ -4070,8 +4102,8 @@
836 wait in the Windows case. */
837
838 if (type == OS_FILE_READ) {
839- return(os_file_read(file, buf, offset,
840- offset_high, n));
841+ return(os_file_read_trx(file, buf, offset,
842+ offset_high, n, trx));
843 }
844
845 ut_a(type == OS_FILE_WRITE);
846@@ -4111,6 +4143,11 @@
847 array = NULL; /* Eliminate compiler warning */
848 }
849
850+ if (trx && type == OS_FILE_READ)
851+ {
852+ trx->io_reads++;
853+ trx->io_read += n;
854+ }
855 slot = os_aio_array_reserve_slot(type, array, message1, message2, file,
856 name, buf, offset, offset_high, n);
857 if (type == OS_FILE_READ) {
858diff -ruN a/storage/innobase/srv/srv0srv.c b/storage/innobase/srv/srv0srv.c
859--- a/storage/innobase/srv/srv0srv.c 2010-12-03 17:32:15.648024399 +0900
860+++ b/storage/innobase/srv/srv0srv.c 2010-12-03 17:45:05.067023254 +0900
861@@ -87,6 +87,9 @@
862 #include "mysql/plugin.h"
863 #include "mysql/service_thd_wait.h"
864
865+/* prototypes for new functions added to ha_innodb.cc */
866+ibool innobase_get_slow_log();
867+
868 /* The following counter is incremented whenever there is some user activity
869 in the server */
870 UNIV_INTERN ulint srv_activity_count = 0;
871@@ -1232,6 +1235,10 @@
872 ibool has_slept = FALSE;
873 srv_conc_slot_t* slot = NULL;
874 ulint i;
875+ ib_uint64_t start_time = 0L;
876+ ib_uint64_t finish_time = 0L;
877+ ulint sec;
878+ ulint ms;
879
880 #ifdef UNIV_SYNC_DEBUG
881 ut_ad(!sync_thread_levels_nonempty_trx(trx->has_search_latch));
882@@ -1312,6 +1319,7 @@
883 switches. */
884 if (SRV_THREAD_SLEEP_DELAY > 0) {
885 os_thread_sleep(SRV_THREAD_SLEEP_DELAY);
886+ trx->innodb_que_wait_timer += SRV_THREAD_SLEEP_DELAY;
887 }
888
889 trx->op_info = "";
890@@ -1371,6 +1379,14 @@
891 #ifdef UNIV_SYNC_DEBUG
892 ut_ad(!sync_thread_levels_nonempty_trx(trx->has_search_latch));
893 #endif /* UNIV_SYNC_DEBUG */
894+
895+ if (innobase_get_slow_log() && trx->take_stats) {
896+ ut_usectime(&sec, &ms);
897+ start_time = (ib_uint64_t)sec * 1000000 + ms;
898+ } else {
899+ start_time = 0;
900+ }
901+
902 trx->op_info = "waiting in InnoDB queue";
903
904 thd_wait_begin(trx->mysql_thd, THD_WAIT_ROW_TABLE_LOCK);
905@@ -1379,6 +1395,12 @@
906
907 trx->op_info = "";
908
909+ if (innobase_get_slow_log() && trx->take_stats && start_time) {
910+ ut_usectime(&sec, &ms);
911+ finish_time = (ib_uint64_t)sec * 1000000 + ms;
912+ trx->innodb_que_wait_timer += (ulint)(finish_time - start_time);
913+ }
914+
915 os_fast_mutex_lock(&srv_conc_mutex);
916
917 srv_conc_n_waiting_threads--;
918diff -ruN a/storage/innobase/trx/trx0trx.c b/storage/innobase/trx/trx0trx.c
919--- a/storage/innobase/trx/trx0trx.c 2010-12-03 15:41:52.053955669 +0900
920+++ b/storage/innobase/trx/trx0trx.c 2010-12-03 17:42:42.127023410 +0900
921@@ -188,6 +188,15 @@
922 trx->global_read_view = NULL;
923 trx->read_view = NULL;
924
925+ trx->io_reads = 0;
926+ trx->io_read = 0;
927+ trx->io_reads_wait_timer = 0;
928+ trx->lock_que_wait_timer = 0;
929+ trx->innodb_que_wait_timer = 0;
930+ trx->distinct_page_access = 0;
931+ trx->distinct_page_access_hash = NULL;
932+ trx->take_stats = FALSE;
933+
934 /* Set X/Open XA transaction identification to NULL */
935 memset(&trx->xid, 0, sizeof(trx->xid));
936 trx->xid.formatID = -1;
937@@ -221,6 +230,11 @@
938
939 mutex_exit(&kernel_mutex);
940
941+ if (innobase_get_slow_log() && trx->take_stats) {
942+ trx->distinct_page_access_hash = mem_alloc(DPAH_SIZE);
943+ memset(trx->distinct_page_access_hash, 0, DPAH_SIZE);
944+ }
945+
946 return(trx);
947 }
948
949@@ -406,6 +420,12 @@
950 /*===============*/
951 trx_t* trx) /*!< in, own: trx object */
952 {
953+ if (trx->distinct_page_access_hash)
954+ {
955+ mem_free(trx->distinct_page_access_hash);
956+ trx->distinct_page_access_hash= NULL;
957+ }
958+
959 mutex_enter(&kernel_mutex);
960
961 UT_LIST_REMOVE(mysql_trx_list, trx_sys->mysql_trx_list, trx);
962@@ -427,6 +447,12 @@
963 /*====================*/
964 trx_t* trx) /*!< in, own: trx object */
965 {
966+ if (trx->distinct_page_access_hash)
967+ {
968+ mem_free(trx->distinct_page_access_hash);
969+ trx->distinct_page_access_hash= NULL;
970+ }
971+
972 mutex_enter(&kernel_mutex);
973
974 trx_free(trx);
975@@ -1212,6 +1238,9 @@
976 trx_t* trx) /*!< in: transaction */
977 {
978 que_thr_t* thr;
979+ ulint sec;
980+ ulint ms;
981+ ib_uint64_t now;
982
983 ut_ad(mutex_own(&kernel_mutex));
984 ut_ad(trx->que_state == TRX_QUE_LOCK_WAIT);
985@@ -1226,6 +1255,11 @@
986 thr = UT_LIST_GET_FIRST(trx->wait_thrs);
987 }
988
989+ if (innobase_get_slow_log() && trx->take_stats) {
990+ ut_usectime(&sec, &ms);
991+ now = (ib_uint64_t)sec * 1000000 + ms;
992+ trx->lock_que_wait_timer += (ulint)(now - trx->lock_que_wait_ustarted);
993+ }
994 trx->que_state = TRX_QUE_RUNNING;
995 }
996
997@@ -1239,6 +1273,9 @@
998 trx_t* trx) /*!< in: transaction in the TRX_QUE_LOCK_WAIT state */
999 {
1000 que_thr_t* thr;
1001+ ulint sec;
1002+ ulint ms;
1003+ ib_uint64_t now;
1004
1005 ut_ad(mutex_own(&kernel_mutex));
1006 ut_ad(trx->que_state == TRX_QUE_LOCK_WAIT);
1007@@ -1253,6 +1290,11 @@
1008 thr = UT_LIST_GET_FIRST(trx->wait_thrs);
1009 }
1010
1011+ if (innobase_get_slow_log() && trx->take_stats) {
1012+ ut_usectime(&sec, &ms);
1013+ now = (ib_uint64_t)sec * 1000000 + ms;
1014+ trx->lock_que_wait_timer += (ulint)(now - trx->lock_que_wait_ustarted);
1015+ }
1016 trx->que_state = TRX_QUE_RUNNING;
1017 }
1018
MySQL: a very fast and reliable SQL database engine
This page took 0.074681 seconds and 4 git commands to generate.