innodb_extend_slow.patch

   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!
   8 diff -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 @@ -2397,11 +2431,19 @@
  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 @@ -2419,7 +2461,7 @@
  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 @@ -2497,6 +2539,13 @@
  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 @@ -2511,6 +2560,12 @@
  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 @@ -2823,6 +2878,11 @@
 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 @@ -2840,6 +2900,9 @@
 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 @@ -2913,7 +2976,7 @@
 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 @@ -3216,6 +3279,13 @@
 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 @@ -3230,6 +3300,12 @@
 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 @@ -3255,13 +3331,17 @@
 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 @@ -3285,6 +3365,7 @@
 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 @@ -3362,13 +3443,17 @@
 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 @@ -3378,6 +3463,9 @@
 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 @@ -3400,6 +3488,7 @@
 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 @@ -3486,6 +3575,11 @@
 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
 242 diff -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
 338 diff -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 @@ -4395,7 +4395,7 @@
 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 @@ -4722,7 +4722,7 @@
 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 @@ -4747,8 +4747,9 @@
 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 @@ -4918,7 +4919,7 @@
 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
 379 diff -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 @@ -1526,6 +1526,16 @@
 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 @@ -1580,6 +1590,32 @@
 400         return(trx);
 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  Note that a transaction has been registered with MySQL.
 431  @return true if transaction is registered with MySQL 2PC coordinator */
 432 @@ -9200,6 +9236,25 @@
 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;
 458 diff -ruN a/storage/innobase/include/buf0rea.h b/storage/innobase/include/buf0rea.h
 459 --- a/storage/innobase/include/buf0rea.h        2010-12-03 15:18:48.891024406 +0900
 460 +++ b/storage/innobase/include/buf0rea.h        2010-12-03 17:42:42.096026873 +0900
 461 @@ -27,6 +27,7 @@
 462  #define buf0rea_h
 463
 464  #include "univ.i"
 465 +#include "trx0types.h"
 466  #include "buf0types.h"
 467
 468  /********************************************************************//**
 469 @@ -41,7 +42,8 @@
 470  /*==========*/
 471         ulint   space,  /*!< in: space id */
 472         ulint   zip_size,/*!< in: compressed page size in bytes, or 0 */
 473 -       ulint   offset);/*!< in: page number */
 474 +       ulint   offset, /*!< in: page number */
 475 +       trx_t*  trx);
 476  /********************************************************************//**
 477  Applies linear read-ahead if in the buf_pool the page is a border page of
 478  a linear read-ahead area and all the pages in the area have been accessed.
 479 @@ -72,8 +74,9 @@
 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 of a page; NOTE: the current thread
 484 +       ulint   offset, /*!< in: page number of a page; NOTE: the current thread
 485                         must want access to this page (see NOTE 3 above) */
 486 +       trx_t*  trx);
 487  /********************************************************************//**
 488  Issues read requests for pages which the ibuf module wants to read in, in
 489  order to contract the insert buffer tree. Technically, this function is like
 490 diff -ruN a/storage/innobase/include/fil0fil.h b/storage/innobase/include/fil0fil.h
 491 --- a/storage/innobase/include/fil0fil.h        2010-12-03 15:09:51.290958543 +0900
 492 +++ b/storage/innobase/include/fil0fil.h        2010-12-03 17:42:42.097027548 +0900
 493 @@ -611,9 +611,12 @@
 494  Reads or writes data. This operation is asynchronous (aio).
 495  @return DB_SUCCESS, or DB_TABLESPACE_DELETED if we are trying to do
 496  i/o on a tablespace which does not exist */
 497 +#define fil_io(type, sync, space_id, zip_size, block_offset, byte_offset, len, buf, message) \
 498 +       _fil_io(type, sync, space_id, zip_size, block_offset, byte_offset, len, buf, message, NULL)
 499 +
 500  UNIV_INTERN
 501  ulint
 502 -fil_io(
 503 +_fil_io(
 504  /*===*/
 505         ulint   type,           /*!< in: OS_FILE_READ or OS_FILE_WRITE,
 506                                 ORed to OS_FILE_LOG, if a log i/o
 507 @@ -638,8 +641,9 @@
 508         void*   buf,            /*!< in/out: buffer where to store read data
 509                                 or from where to write; in aio this must be
 510                                 appropriately aligned */
 511 -       void*   message);       /*!< in: message for aio handler if non-sync
 512 +       void*   message,        /*!< in: message for aio handler if non-sync
 513                                 aio used, else ignored */
 514 +       trx_t*  trx);
 515  /**********************************************************************//**
 516  Waits for an aio operation to complete. This function is used to write the
 517  handler for completed requests. The aio array of pending requests is divided
 518 diff -ruN a/storage/innobase/include/os0file.h b/storage/innobase/include/os0file.h
 519 --- a/storage/innobase/include/os0file.h        2010-11-03 07:01:13.000000000 +0900
 520 +++ b/storage/innobase/include/os0file.h        2010-12-03 17:42:42.100023783 +0900
 521 @@ -36,6 +36,7 @@
 522  #define os0file_h
 523
 524  #include "univ.i"
 525 +#include "trx0types.h"
 526
 527  #ifndef __WIN__
 528  #include <dirent.h>
 529 @@ -277,13 +278,17 @@
 530         pfs_os_file_close_func(file, __FILE__, __LINE__)
 531
 532  # define os_aio(type, mode, name, file, buf, offset, offset_high,      \
 533 -               n, message1, message2)                                  \
 534 +               n, message1, message2, trx)                             \
 535         pfs_os_aio_func(type, mode, name, file, buf, offset,            \
 536 -                       offset_high, n, message1, message2,             \
 537 +                       offset_high, n, message1, message2, trx,        \
 538                         __FILE__, __LINE__)
 539
 540  # define os_file_read(file, buf, offset, offset_high, n)               \
 541 -       pfs_os_file_read_func(file, buf, offset, offset_high, n,        \
 542 +       pfs_os_file_read_func(file, buf, offset, offset_high, n, NULL,  \
 543 +                             __FILE__, __LINE__)
 544 +
 545 +# define os_file_read_trx(file, buf, offset, offset_high, n, trx)      \
 546 +       pfs_os_file_read_func(file, buf, offset, offset_high, n, trx,   \
 547                               __FILE__, __LINE__)
 548
 549  # define os_file_read_no_error_handling(file, buf, offset,             \
 550 @@ -319,12 +324,15 @@
 551  # define os_file_close(file)   os_file_close_func(file)
 552
 553  # define os_aio(type, mode, name, file, buf, offset, offset_high,      \
 554 -              n, message1, message2)                                   \
 555 +              n, message1, message2, trx)                              \
 556         os_aio_func(type, mode, name, file, buf, offset, offset_high, n,\
 557 -                   message1, message2)
 558 +                   message1, message2, trx)
 559
 560  # define os_file_read(file, buf, offset, offset_high, n)               \
 561 -       os_file_read_func(file, buf, offset, offset_high, n)
 562 +       os_file_read_func(file, buf, offset, offset_high, n, NULL)
 563 +
 564 +# define os_file_read_trx(file, buf, offset, offset_high, n, trx)      \
 565 +       os_file_read_func(file, buf, offset, offset_high, n, trx)
 566
 567  # define os_file_read_no_error_handling(file, buf, offset,             \
 568                                        offset_high, n)                  \
 569 @@ -690,6 +698,7 @@
 570         ulint           offset_high,/*!< in: most significant 32 bits of
 571                                 offset */
 572         ulint           n,      /*!< in: number of bytes to read */
 573 +       trx_t*          trx,
 574         const char*     src_file,/*!< in: file name where func invoked */
 575         ulint           src_line);/*!< in: line where the func invoked */
 576
 577 @@ -744,6 +753,7 @@
 578                                 (can be used to identify a completed
 579                                 aio operation); ignored if mode is
 580                                  OS_AIO_SYNC */
 581 +       trx_t*          trx,
 582         const char*     src_file,/*!< in: file name where func invoked */
 583         ulint           src_line);/*!< in: line where the func invoked */
 584  /*******************************************************************//**
 585 @@ -885,7 +895,8 @@
 586                                 offset where to read */
 587         ulint           offset_high,/*!< in: most significant 32 bits of
 588                                 offset */
 589 -       ulint           n);     /*!< in: number of bytes to read */
 590 +       ulint           n,      /*!< in: number of bytes to read */
 591 +       trx_t*          trx);
 592  /*******************************************************************//**
 593  Rewind file to its start, read at most size - 1 bytes from it to str, and
 594  NUL-terminate str. All errors are silently ignored. This function is
 595 @@ -1044,10 +1055,11 @@
 596                                 (can be used to identify a completed
 597                                 aio operation); ignored if mode is
 598                                 OS_AIO_SYNC */
 599 -       void*           message2);/*!< in: message for the aio handler
 600 +       void*           message2,/*!< in: message for the aio handler
 601                                 (can be used to identify a completed
 602                                 aio operation); ignored if mode is
 603                                 OS_AIO_SYNC */
 604 +       trx_t*          trx);
 605  /************************************************************************//**
 606  Wakes up all async i/o threads so that they know to exit themselves in
 607  shutdown. */
 608 diff -ruN a/storage/innobase/include/os0file.ic b/storage/innobase/include/os0file.ic
 609 --- a/storage/innobase/include/os0file.ic       2010-11-03 07:01:13.000000000 +0900
 610 +++ b/storage/innobase/include/os0file.ic       2010-12-03 17:42:42.102024458 +0900
 611 @@ -229,6 +229,7 @@
 612                                 (can be used to identify a completed
 613                                 aio operation); ignored if mode is
 614                                  OS_AIO_SYNC */
 615 +       trx_t*          trx,
 616         const char*     src_file,/*!< in: file name where func invoked */
 617         ulint           src_line)/*!< in: line where the func invoked */
 618  {
 619 @@ -244,7 +245,7 @@
 620                                    src_file, src_line);
 621
 622         result = os_aio_func(type, mode, name, file, buf, offset, offset_high,
 623 -                            n, message1, message2);
 624 +                            n, message1, message2, trx);
 625
 626         register_pfs_file_io_end(locker, n);
 627
 628 @@ -268,6 +269,7 @@
 629         ulint           offset_high,/*!< in: most significant 32 bits of
 630                                 offset */
 631         ulint           n,      /*!< in: number of bytes to read */
 632 +       trx_t*          trx,
 633         const char*     src_file,/*!< in: file name where func invoked */
 634         ulint           src_line)/*!< in: line where the func invoked */
 635  {
 636 @@ -278,7 +280,7 @@
 637         register_pfs_file_io_begin(&state, locker, file, n, PSI_FILE_READ,
 638                                    src_file, src_line);
 639
 640 -       result = os_file_read_func(file, buf, offset, offset_high, n);
 641 +       result = os_file_read_func(file, buf, offset, offset_high, n, trx);
 642
 643         register_pfs_file_io_end(locker, n);
 644
 645 diff -ruN a/storage/innobase/include/srv0srv.h b/storage/innobase/include/srv0srv.h
 646 --- a/storage/innobase/include/srv0srv.h        2010-12-03 17:32:15.634987408 +0900
 647 +++ b/storage/innobase/include/srv0srv.h        2010-12-03 17:42:42.104028644 +0900
 648 @@ -71,6 +71,9 @@
 649  #define SRV_AUTO_EXTEND_INCREMENT      \
 650         (srv_auto_extend_increment * ((1024 * 1024) / UNIV_PAGE_SIZE))
 651
 652 +/* prototypes for new functions added to ha_innodb.cc */
 653 +ibool  innobase_get_slow_log();
 654 +
 655  /* This is set to TRUE if the MySQL user has set it in MySQL */
 656  extern ibool   srv_lower_case_table_names;
 657
 658 diff -ruN a/storage/innobase/include/trx0trx.h b/storage/innobase/include/trx0trx.h
 659 --- a/storage/innobase/include/trx0trx.h        2010-12-03 15:41:52.049372966 +0900
 660 +++ b/storage/innobase/include/trx0trx.h        2010-12-03 17:42:42.107024532 +0900
 661 @@ -738,6 +738,17 @@
 662         /*------------------------------*/
 663         char detailed_error[256];       /*!< detailed error message for last
 664                                         error, or empty. */
 665 +       /*------------------------------*/
 666 +       ulint           io_reads;
 667 +       ib_uint64_t     io_read;
 668 +       ulint           io_reads_wait_timer;
 669 +       ib_uint64_t     lock_que_wait_ustarted;
 670 +       ulint           lock_que_wait_timer;
 671 +       ulint           innodb_que_wait_timer;
 672 +       ulint           distinct_page_access;
 673 +#define        DPAH_SIZE       8192
 674 +       byte*           distinct_page_access_hash;
 675 +       ibool           take_stats;
 676  };
 677
 678  #define TRX_MAX_N_THREADS      32      /* maximum number of
 679 diff -ruN a/storage/innobase/lock/lock0lock.c b/storage/innobase/lock/lock0lock.c
 680 --- a/storage/innobase/lock/lock0lock.c 2010-12-03 15:09:51.297986437 +0900
 681 +++ b/storage/innobase/lock/lock0lock.c 2010-12-03 17:42:42.111024587 +0900
 682 @@ -1755,6 +1755,8 @@
 683  {
 684         lock_t* lock;
 685         trx_t*  trx;
 686 +       ulint   sec;
 687 +       ulint   ms;
 688
 689         ut_ad(mutex_own(&kernel_mutex));
 690
 691 @@ -1813,6 +1815,10 @@
 692         trx->que_state = TRX_QUE_LOCK_WAIT;
 693         trx->was_chosen_as_deadlock_victim = FALSE;
 694         trx->wait_started = time(NULL);
 695 +       if (innobase_get_slow_log() && trx->take_stats) {
 696 +               ut_usectime(&sec, &ms);
 697 +               trx->lock_que_wait_ustarted = (ib_uint64_t)sec * 1000000 + ms;
 698 +       }
 699
 700         ut_a(que_thr_stop(thr));
 701
 702 @@ -3692,6 +3698,8 @@
 703  {
 704         lock_t* lock;
 705         trx_t*  trx;
 706 +       ulint   sec;
 707 +       ulint   ms;
 708
 709         ut_ad(mutex_own(&kernel_mutex));
 710
 711 @@ -3747,6 +3755,10 @@
 712                 return(DB_SUCCESS);
 713         }
 714
 715 +       if (innobase_get_slow_log() && trx->take_stats) {
 716 +               ut_usectime(&sec, &ms);
 717 +               trx->lock_que_wait_ustarted = (ib_uint64_t)sec * 1000000 + ms;
 718 +       }
 719         trx->que_state = TRX_QUE_LOCK_WAIT;
 720         trx->was_chosen_as_deadlock_victim = FALSE;
 721         trx->wait_started = time(NULL);
 722 diff -ruN a/storage/innobase/os/os0file.c b/storage/innobase/os/os0file.c
 723 --- a/storage/innobase/os/os0file.c     2010-12-03 17:32:15.644024974 +0900
 724 +++ b/storage/innobase/os/os0file.c     2010-12-03 17:42:42.117023467 +0900
 725 @@ -43,6 +43,8 @@
 726  #include "srv0start.h"
 727  #include "fil0fil.h"
 728  #include "buf0buf.h"
 729 +#include "trx0sys.h"
 730 +#include "trx0trx.h"
 731  #include "log0recv.h"
 732  #ifndef UNIV_HOTBACKUP
 733  # include "os0sync.h"
 734 @@ -2177,13 +2179,18 @@
 735         ulint           n,      /*!< in: number of bytes to read */
 736         ulint           offset, /*!< in: least significant 32 bits of file
 737                                 offset from where to read */
 738 -       ulint           offset_high) /*!< in: most significant 32 bits of
 739 +       ulint           offset_high, /*!< in: most significant 32 bits of
 740                                 offset */
 741 +       trx_t*          trx)
 742  {
 743         off_t   offs;
 744  #if defined(HAVE_PREAD) && !defined(HAVE_BROKEN_PREAD)
 745         ssize_t n_bytes;
 746  #endif /* HAVE_PREAD && !HAVE_BROKEN_PREAD */
 747 +       ulint           sec;
 748 +       ulint           ms;
 749 +       ib_uint64_t     start_time;
 750 +       ib_uint64_t     finish_time;
 751
 752         ut_a((offset & 0xFFFFFFFFUL) == offset);
 753
 754 @@ -2204,6 +2211,15 @@
 755
 756         os_n_file_reads++;
 757
 758 +       if (innobase_get_slow_log() && trx && trx->take_stats)
 759 +       {
 760 +               trx->io_reads++;
 761 +               trx->io_read += n;
 762 +               ut_usectime(&sec, &ms);
 763 +               start_time = (ib_uint64_t)sec * 1000000 + ms;
 764 +       } else {
 765 +               start_time = 0;
 766 +       }
 767  #if defined(HAVE_PREAD) && !defined(HAVE_BROKEN_PREAD)
 768         os_mutex_enter(os_file_count_mutex);
 769         os_file_n_pending_preads++;
 770 @@ -2217,6 +2233,13 @@
 771         os_n_pending_reads--;
 772         os_mutex_exit(os_file_count_mutex);
 773
 774 +       if (innobase_get_slow_log() && trx && trx->take_stats && start_time)
 775 +       {
 776 +               ut_usectime(&sec, &ms);
 777 +               finish_time = (ib_uint64_t)sec * 1000000 + ms;
 778 +               trx->io_reads_wait_timer += (ulint)(finish_time - start_time);
 779 +       }
 780 +
 781         return(n_bytes);
 782  #else
 783         {
 784 @@ -2253,6 +2276,13 @@
 785                 os_n_pending_reads--;
 786                 os_mutex_exit(os_file_count_mutex);
 787
 788 +               if (innobase_get_slow_log() && trx && trx->take_stats && start_time)
 789 +               {
 790 +                       ut_usectime(&sec, &ms);
 791 +                       finish_time = (ib_uint64_t)sec * 1000000 + ms;
 792 +                       trx->io_reads_wait_timer += (ulint)(finish_time - start_time);
 793 +               }
 794 +
 795                 return(ret);
 796         }
 797  #endif
 798 @@ -2393,7 +2423,8 @@
 799                                 offset where to read */
 800         ulint           offset_high, /*!< in: most significant 32 bits of
 801                                 offset */
 802 -       ulint           n)      /*!< in: number of bytes to read */
 803 +       ulint           n,      /*!< in: number of bytes to read */
 804 +       trx_t*          trx)
 805  {
 806  #ifdef __WIN__
 807         BOOL            ret;
 808 @@ -2468,7 +2499,7 @@
 809         os_bytes_read_since_printout += n;
 810
 811  try_again:
 812 -       ret = os_file_pread(file, buf, n, offset, offset_high);
 813 +       ret = os_file_pread(file, buf, n, offset, offset_high, trx);
 814
 815         if ((ulint)ret == n) {
 816
 817 @@ -2597,7 +2628,7 @@
 818         os_bytes_read_since_printout += n;
 819
 820  try_again:
 821 -       ret = os_file_pread(file, buf, n, offset, offset_high);
 822 +       ret = os_file_pread(file, buf, n, offset, offset_high, NULL);
 823
 824         if ((ulint)ret == n) {
 825
 826 @@ -3619,7 +3650,8 @@
 827                                 offset */
 828         ulint           offset_high, /*!< in: most significant 32 bits of
 829                                 offset */
 830 -       ulint           len)    /*!< in: length of the block to read or write */
 831 +       ulint           len,    /*!< in: length of the block to read or write */
 832 +       trx_t*          trx)
 833  {
 834         os_aio_slot_t*  slot = NULL;
 835  #ifdef WIN_ASYNC_IO
 836 @@ -3991,10 +4023,11 @@
 837                                 (can be used to identify a completed
 838                                 aio operation); ignored if mode is
 839                                 OS_AIO_SYNC */
 840 -       void*           message2)/*!< in: message for the aio handler
 841 +       void*           message2,/*!< in: message for the aio handler
 842                                 (can be used to identify a completed
 843                                 aio operation); ignored if mode is
 844                                 OS_AIO_SYNC */
 845 +       trx_t*          trx)
 846  {
 847         os_aio_array_t* array;
 848         os_aio_slot_t*  slot;
 849 @@ -4035,8 +4068,8 @@
 850                 wait in the Windows case. */
 851
 852                 if (type == OS_FILE_READ) {
 853 -                       return(os_file_read(file, buf, offset,
 854 -                                           offset_high, n));
 855 +                       return(os_file_read_trx(file, buf, offset,
 856 +                                           offset_high, n, trx));
 857                 }
 858
 859                 ut_a(type == OS_FILE_WRITE);
 860 @@ -4074,8 +4107,13 @@
 861                 ut_error;
 862         }
 863
 864 +       if (trx && type == OS_FILE_READ)
 865 +       {
 866 +               trx->io_reads++;
 867 +               trx->io_read += n;
 868 +       }
 869         slot = os_aio_array_reserve_slot(type, array, message1, message2, file,
 870 -                                        name, buf, offset, offset_high, n);
 871 +                                        name, buf, offset, offset_high, n, trx);
 872         if (type == OS_FILE_READ) {
 873                 if (srv_use_native_aio) {
 874                         os_n_file_reads++;
 875 diff -ruN a/storage/innobase/srv/srv0srv.c b/storage/innobase/srv/srv0srv.c
 876 --- a/storage/innobase/srv/srv0srv.c    2010-12-03 17:32:15.648024399 +0900
 877 +++ b/storage/innobase/srv/srv0srv.c    2010-12-03 17:45:05.067023254 +0900
 878 @@ -88,6 +88,9 @@
 879  #include "mysql/plugin.h"
 880  #include "mysql/service_thd_wait.h"
 881
 882 +/* prototypes for new functions added to ha_innodb.cc */
 883 +ibool  innobase_get_slow_log();
 884 +
 885  /* This is set to TRUE if the MySQL user has set it in MySQL; currently
 886  affects only FOREIGN KEY definition parsing */
 887  UNIV_INTERN ibool      srv_lower_case_table_names      = FALSE;
 888 @@ -1219,6 +1222,10 @@
 889         ibool                   has_slept = FALSE;
 890         srv_conc_slot_t*        slot      = NULL;
 891         ulint                   i;
 892 +       ib_uint64_t             start_time = 0L;
 893 +       ib_uint64_t             finish_time = 0L;
 894 +       ulint                   sec;
 895 +       ulint                   ms;
 896
 897         if (trx->mysql_thd != NULL
 898             && thd_is_replication_slave_thread(trx->mysql_thd)) {
 899 @@ -1295,6 +1302,7 @@
 900                 switches. */
 901                 if (SRV_THREAD_SLEEP_DELAY > 0) {
 902                         os_thread_sleep(SRV_THREAD_SLEEP_DELAY);
 903 +                       trx->innodb_que_wait_timer += SRV_THREAD_SLEEP_DELAY;
 904                 }
 905
 906                 trx->op_info = "";
 907 @@ -1350,6 +1358,13 @@
 908         /* Go to wait for the event; when a thread leaves InnoDB it will
 909         release this thread */
 910
 911 +       if (innobase_get_slow_log() && trx->take_stats) {
 912 +               ut_usectime(&sec, &ms);
 913 +               start_time = (ib_uint64_t)sec * 1000000 + ms;
 914 +       } else {
 915 +               start_time = 0;
 916 +       }
 917 +
 918         trx->op_info = "waiting in InnoDB queue";
 919
 920         thd_wait_begin(trx->mysql_thd, THD_WAIT_ROW_TABLE_LOCK);
 921 @@ -1358,6 +1373,12 @@
 922
 923         trx->op_info = "";
 924
 925 +       if (innobase_get_slow_log() && trx->take_stats && start_time) {
 926 +               ut_usectime(&sec, &ms);
 927 +               finish_time = (ib_uint64_t)sec * 1000000 + ms;
 928 +               trx->innodb_que_wait_timer += (ulint)(finish_time - start_time);
 929 +       }
 930 +
 931         os_fast_mutex_lock(&srv_conc_mutex);
 932
 933         srv_conc_n_waiting_threads--;
 934 diff -ruN a/storage/innobase/trx/trx0trx.c b/storage/innobase/trx/trx0trx.c
 935 --- a/storage/innobase/trx/trx0trx.c    2010-12-03 15:41:52.053955669 +0900
 936 +++ b/storage/innobase/trx/trx0trx.c    2010-12-03 17:42:42.127023410 +0900
 937 @@ -185,6 +185,15 @@
 938         trx->global_read_view = NULL;
 939         trx->read_view = NULL;
 940
 941 +       trx->io_reads = 0;
 942 +       trx->io_read = 0;
 943 +       trx->io_reads_wait_timer = 0;
 944 +       trx->lock_que_wait_timer = 0;
 945 +       trx->innodb_que_wait_timer = 0;
 946 +       trx->distinct_page_access = 0;
 947 +       trx->distinct_page_access_hash = NULL;
 948 +       trx->take_stats = FALSE;
 949 +
 950         /* Set X/Open XA transaction identification to NULL */
 951         memset(&trx->xid, 0, sizeof(trx->xid));
 952         trx->xid.formatID = -1;
 953 @@ -222,6 +231,11 @@
 954
 955         trx->mysql_process_no = os_proc_get_number();
 956
 957 +       if (innobase_get_slow_log() && trx->take_stats) {
 958 +               trx->distinct_page_access_hash = mem_alloc(DPAH_SIZE);
 959 +               memset(trx->distinct_page_access_hash, 0, DPAH_SIZE);
 960 +       }
 961 +
 962         return(trx);
 963  }
 964
 965 @@ -353,6 +367,12 @@
 966  /*===============*/
 967         trx_t*  trx)    /*!< in, own: trx object */
 968  {
 969 +       if (trx->distinct_page_access_hash)
 970 +       {
 971 +               mem_free(trx->distinct_page_access_hash);
 972 +               trx->distinct_page_access_hash= NULL;
 973 +       }
 974 +
 975         mutex_enter(&kernel_mutex);
 976
 977         UT_LIST_REMOVE(mysql_trx_list, trx_sys->mysql_trx_list, trx);
 978 @@ -374,6 +394,12 @@
 979  /*====================*/
 980         trx_t*  trx)    /*!< in, own: trx object */
 981  {
 982 +       if (trx->distinct_page_access_hash)
 983 +       {
 984 +               mem_free(trx->distinct_page_access_hash);
 985 +               trx->distinct_page_access_hash= NULL;
 986 +       }
 987 +
 988         mutex_enter(&kernel_mutex);
 989
 990         trx_free(trx);
 991 @@ -1091,6 +1117,9 @@
 992         trx_t*  trx)    /*!< in: transaction */
 993  {
 994         que_thr_t*      thr;
 995 +       ulint           sec;
 996 +       ulint           ms;
 997 +       ib_uint64_t     now;
 998
 999         ut_ad(mutex_own(&kernel_mutex));
1000         ut_ad(trx->que_state == TRX_QUE_LOCK_WAIT);
1001 @@ -1105,6 +1134,11 @@
1002                 thr = UT_LIST_GET_FIRST(trx->wait_thrs);
1003         }
1004
1005 +       if (innobase_get_slow_log() && trx->take_stats) {
1006 +               ut_usectime(&sec, &ms);
1007 +               now = (ib_uint64_t)sec * 1000000 + ms;
1008 +               trx->lock_que_wait_timer += (ulint)(now - trx->lock_que_wait_ustarted);
1009 +       }
1010         trx->que_state = TRX_QUE_RUNNING;
1011  }
1012
1013 @@ -1118,6 +1152,9 @@
1014         trx_t*  trx)    /*!< in: transaction in the TRX_QUE_LOCK_WAIT state */
1015  {
1016         que_thr_t*      thr;
1017 +       ulint           sec;
1018 +       ulint           ms;
1019 +       ib_uint64_t     now;
1020
1021         ut_ad(mutex_own(&kernel_mutex));
1022         ut_ad(trx->que_state == TRX_QUE_LOCK_WAIT);
1023 @@ -1132,6 +1169,11 @@
1024                 thr = UT_LIST_GET_FIRST(trx->wait_thrs);
1025         }
1026
1027 +       if (innobase_get_slow_log() && trx->take_stats) {
1028 +               ut_usectime(&sec, &ms);
1029 +               now = (ib_uint64_t)sec * 1000000 + ms;
1030 +               trx->lock_que_wait_timer += (ulint)(now - trx->lock_que_wait_ustarted);
1031 +       }
1032         trx->que_state = TRX_QUE_RUNNING;
1033  }
1034