diff options
author | Vasil Dimov <vasil.dimov@oracle.com> | 2011-01-30 18:41:58 +0200 |
---|---|---|
committer | Vasil Dimov <vasil.dimov@oracle.com> | 2011-01-30 18:41:58 +0200 |
commit | 6f8c6da4b556480458fcdbc2d90ee5bcf138620e (patch) | |
tree | 6577651ab0bb8a0c98c7690114615d31041f7457 /storage/innodb_plugin | |
parent | 26528be7112fadd5f23994f96174d3c45bd741b5 (diff) | |
parent | 71e8043bae2071ba875b18326504b1058b8deb98 (diff) | |
download | mariadb-git-6f8c6da4b556480458fcdbc2d90ee5bcf138620e.tar.gz |
Merge mysql-5.1-innodb -> mysql-5.1
Diffstat (limited to 'storage/innodb_plugin')
32 files changed, 400 insertions, 153 deletions
diff --git a/storage/innodb_plugin/ChangeLog b/storage/innodb_plugin/ChangeLog index 8eb63fe8c78..18ab48f32a5 100644 --- a/storage/innodb_plugin/ChangeLog +++ b/storage/innodb_plugin/ChangeLog @@ -1,3 +1,54 @@ +2011-01-27 The InnoDB Team + + * btr/btr0cur.c: + Bug#59465 btr_estimate_number_of_different_key_vals use + incorrect offset for external_size + +2011-01-27 The InnoDB Team + + * include/trx0trx.h, trx/trx0trx.c: + Bug#59440 Race condition in XA ROLLBACK and XA COMMIT + after server restart + +2011-01-25 The InnoDB Team + + * row/row0upd.c: + Bug#59585 Fix 58912 introduces compiler warning + due to potentially uninitialized variable + +2011-01-25 The InnoDB Team + + * mtr/mtr0log.c: + Bug#59486 Incorrect usage of UNIV_UNLIKELY() in mlog_parse_string() + +2011-01-25 The InnoDB Team + + * row/row0vers.c: + Fix Bug#59464 Race condition in row_vers_build_for_semi_consistent_read + +2011-01-25 The InnoDB Team + + * btr/btr0btr.c, btr/btr0cur.c, btr/btr0sea.c, + buf/buf0buddy.c, buf/buf0buf.c, buf/buf0lru.c, + include/buf0buf.h, include/buf0buf.ic, include/buf0lru.h, + mem/mem0mem.c, page/page0zip.c: + Fix Bug#59707 Unused compression-related parameters + in buffer pool functions + +2011-01-18 The InnoDB Team + + * include/sync0rw.h, sync/sync0arr.c, sync/sync0rw.c: + Fix Bug#59579 rw_lock_debug_print outputs to stderr, not to + SHOW ENGINE INNODB STATUS + +2011-01-14 The InnoDB Team + * btr/btr0cur.c, dict/dict0dict.c, handler/ha_innodb.cc, + include/btr0cur.h, include/dict0mem.h, include/rem0cmp.h, + include/rem0cmp.ic, include/srv0srv.h, rem/rem0cmp.c, + srv/srv0srv.c, innodb_bug30423.test: + Fix Bug#30423 InnoDBs treatment of NULL in index stats causes + bad "rows examined" estimates + 2011-01-06 The InnoDB Team * handler/i_s.cc, include/trx0i_s.h, trx/trx0i_s.c: Fix Bug#55397 cannot select from innodb_trx when trx_query contains diff --git a/storage/innodb_plugin/btr/btr0btr.c b/storage/innodb_plugin/btr/btr0btr.c index 32e2caecdb8..3d8d6048603 100644 --- a/storage/innodb_plugin/btr/btr0btr.c +++ b/storage/innodb_plugin/btr/btr0btr.c @@ -979,7 +979,7 @@ btr_page_reorganize_low( log_mode = mtr_set_log_mode(mtr, MTR_LOG_NONE); #ifndef UNIV_HOTBACKUP - temp_block = buf_block_alloc(0); + temp_block = buf_block_alloc(); #else /* !UNIV_HOTBACKUP */ ut_ad(block == back_block1); temp_block = back_block2; diff --git a/storage/innodb_plugin/btr/btr0cur.c b/storage/innodb_plugin/btr/btr0cur.c index c57255a25ae..874db3066b5 100644 --- a/storage/innodb_plugin/btr/btr0cur.c +++ b/storage/innodb_plugin/btr/btr0cur.c @@ -100,6 +100,18 @@ can be released by page reorganize, then it is reorganized */ /*--------------------------------------*/ #define BTR_BLOB_HDR_SIZE 8 /*!< Size of a BLOB part header, in bytes */ + +/** Estimated table level stats from sampled value. +@param value sampled stats +@param index index being sampled +@param sample number of sampled rows +@param ext_size external stored data size +@param not_empty table not empty +@return estimated table wide stats from sampled value */ +#define BTR_TABLE_STATS_FROM_SAMPLE(value, index, sample, ext_size, not_empty)\ + (((value) * (ib_int64_t) index->stat_n_leaf_pages \ + + (sample) - 1 + (ext_size) + (not_empty)) / ((sample) + (ext_size))) + /* @} */ #endif /* !UNIV_HOTBACKUP */ @@ -3201,9 +3213,54 @@ btr_estimate_n_rows_in_range( } /*******************************************************************//** +Record the number of non_null key values in a given index for +each n-column prefix of the index where n < dict_index_get_n_unique(index). +The estimates are eventually stored in the array: +index->stat_n_non_null_key_vals. */ +static +void +btr_record_not_null_field_in_rec( +/*=============================*/ + rec_t* rec, /*!< in: physical record */ + ulint n_unique, /*!< in: dict_index_get_n_unique(index), + number of columns uniquely determine + an index entry */ + const ulint* offsets, /*!< in: rec_get_offsets(rec, index), + its size could be for all fields or + that of "n_unique" */ + ib_int64_t* n_not_null) /*!< in/out: array to record number of + not null rows for n-column prefix */ +{ + ulint i; + + ut_ad(rec_offs_n_fields(offsets) >= n_unique); + + if (n_not_null == NULL) { + return; + } + + for (i = 0; i < n_unique; i++) { + ulint rec_len; + byte* field; + + field = rec_get_nth_field(rec, offsets, i, &rec_len); + + if (rec_len != UNIV_SQL_NULL) { + n_not_null[i]++; + } else { + /* Break if we hit the first NULL value */ + break; + } + } +} + +/*******************************************************************//** Estimates the number of different key values in a given index, for each n-column prefix of the index where n <= dict_index_get_n_unique(index). -The estimates are stored in the array index->stat_n_diff_key_vals. */ +The estimates are stored in the array index->stat_n_diff_key_vals. +If innodb_stats_method is "nulls_ignored", we also record the number of +non-null values for each prefix and store the estimates in +array index->stat_n_non_null_key_vals. */ UNIV_INTERN void btr_estimate_number_of_different_key_vals( @@ -3217,6 +3274,8 @@ btr_estimate_number_of_different_key_vals( ulint matched_fields; ulint matched_bytes; ib_int64_t* n_diff; + ib_int64_t* n_not_null; + ibool stats_null_not_equal; ullint n_sample_pages; /* number of pages to sample */ ulint not_empty_flag = 0; ulint total_external_size = 0; @@ -3225,16 +3284,43 @@ btr_estimate_number_of_different_key_vals( ullint add_on; mtr_t mtr; mem_heap_t* heap = NULL; - ulint offsets_rec_[REC_OFFS_NORMAL_SIZE]; - ulint offsets_next_rec_[REC_OFFS_NORMAL_SIZE]; - ulint* offsets_rec = offsets_rec_; - ulint* offsets_next_rec= offsets_next_rec_; - rec_offs_init(offsets_rec_); - rec_offs_init(offsets_next_rec_); + ulint* offsets_rec = NULL; + ulint* offsets_next_rec = NULL; n_cols = dict_index_get_n_unique(index); - n_diff = mem_zalloc((n_cols + 1) * sizeof(ib_int64_t)); + heap = mem_heap_create((sizeof *n_diff + sizeof *n_not_null) + * (n_cols + 1) + + dict_index_get_n_fields(index) + * (sizeof *offsets_rec + + sizeof *offsets_next_rec)); + + n_diff = mem_heap_zalloc(heap, (n_cols + 1) * sizeof(ib_int64_t)); + + n_not_null = NULL; + + /* Check srv_innodb_stats_method setting, and decide whether we + need to record non-null value and also decide if NULL is + considered equal (by setting stats_null_not_equal value) */ + switch (srv_innodb_stats_method) { + case SRV_STATS_NULLS_IGNORED: + n_not_null = mem_heap_zalloc(heap, (n_cols + 1) + * sizeof *n_not_null); + /* fall through */ + + case SRV_STATS_NULLS_UNEQUAL: + /* for both SRV_STATS_NULLS_IGNORED and SRV_STATS_NULLS_UNEQUAL + case, we will treat NULLs as unequal value */ + stats_null_not_equal = TRUE; + break; + + case SRV_STATS_NULLS_EQUAL: + stats_null_not_equal = FALSE; + break; + + default: + ut_error; + } /* It makes no sense to test more pages than are contained in the index, thus we lower the number if it is too high */ @@ -3251,7 +3337,6 @@ btr_estimate_number_of_different_key_vals( /* We sample some pages in the index to get an estimate */ for (i = 0; i < n_sample_pages; i++) { - rec_t* supremum; mtr_start(&mtr); btr_cur_open_at_rnd_pos(index, BTR_SEARCH_LEAF, &cursor, &mtr); @@ -3264,18 +3349,25 @@ btr_estimate_number_of_different_key_vals( page = btr_cur_get_page(&cursor); - supremum = page_get_supremum_rec(page); rec = page_rec_get_next(page_get_infimum_rec(page)); - if (rec != supremum) { + if (!page_rec_is_supremum(rec)) { not_empty_flag = 1; offsets_rec = rec_get_offsets(rec, index, offsets_rec, ULINT_UNDEFINED, &heap); + + if (n_not_null) { + btr_record_not_null_field_in_rec( + rec, n_cols, offsets_rec, n_not_null); + } } - while (rec != supremum) { + while (!page_rec_is_supremum(rec)) { rec_t* next_rec = page_rec_get_next(rec); - if (next_rec == supremum) { + if (page_rec_is_supremum(next_rec)) { + total_external_size += + btr_rec_get_externally_stored_len( + rec, offsets_rec); break; } @@ -3283,11 +3375,13 @@ btr_estimate_number_of_different_key_vals( matched_bytes = 0; offsets_next_rec = rec_get_offsets(next_rec, index, offsets_next_rec, - n_cols, &heap); + ULINT_UNDEFINED, + &heap); cmp_rec_rec_with_match(rec, next_rec, offsets_rec, offsets_next_rec, - index, &matched_fields, + index, stats_null_not_equal, + &matched_fields, &matched_bytes); for (j = matched_fields + 1; j <= n_cols; j++) { @@ -3297,6 +3391,12 @@ btr_estimate_number_of_different_key_vals( n_diff[j]++; } + if (n_not_null) { + btr_record_not_null_field_in_rec( + next_rec, n_cols, offsets_next_rec, + n_not_null); + } + total_external_size += btr_rec_get_externally_stored_len( rec, offsets_rec); @@ -3331,10 +3431,6 @@ btr_estimate_number_of_different_key_vals( } } - offsets_rec = rec_get_offsets(rec, index, offsets_rec, - ULINT_UNDEFINED, &heap); - total_external_size += btr_rec_get_externally_stored_len( - rec, offsets_rec); mtr_commit(&mtr); } @@ -3348,13 +3444,9 @@ btr_estimate_number_of_different_key_vals( for (j = 0; j <= n_cols; j++) { index->stat_n_diff_key_vals[j] - = ((n_diff[j] - * (ib_int64_t)index->stat_n_leaf_pages - + n_sample_pages - 1 - + total_external_size - + not_empty_flag) - / (n_sample_pages - + total_external_size)); + = BTR_TABLE_STATS_FROM_SAMPLE( + n_diff[j], index, n_sample_pages, + total_external_size, not_empty_flag); /* If the tree is small, smaller than 10 * n_sample_pages + total_external_size, then @@ -3373,12 +3465,20 @@ btr_estimate_number_of_different_key_vals( } index->stat_n_diff_key_vals[j] += add_on; - } - mem_free(n_diff); - if (UNIV_LIKELY_NULL(heap)) { - mem_heap_free(heap); + /* Update the stat_n_non_null_key_vals[] with our + sampled result. stat_n_non_null_key_vals[] is created + and initialized to zero in dict_index_add_to_cache(), + along with stat_n_diff_key_vals[] array */ + if (n_not_null != NULL && (j < n_cols)) { + index->stat_n_non_null_key_vals[j] = + BTR_TABLE_STATS_FROM_SAMPLE( + n_not_null[j], index, n_sample_pages, + total_external_size, not_empty_flag); + } } + + mem_heap_free(heap); } /*================== EXTERNAL STORAGE OF BIG FIELDS ===================*/ @@ -3667,13 +3767,12 @@ btr_blob_free( && buf_block_get_space(block) == space && buf_block_get_page_no(block) == page_no) { - if (buf_LRU_free_block(&block->page, all, NULL) - != BUF_LRU_FREED + if (buf_LRU_free_block(&block->page, all) != BUF_LRU_FREED && all && block->page.zip.data) { /* Attempt to deallocate the uncompressed page if the whole block cannot be deallocted. */ - buf_LRU_free_block(&block->page, FALSE, NULL); + buf_LRU_free_block(&block->page, FALSE); } } diff --git a/storage/innodb_plugin/btr/btr0sea.c b/storage/innodb_plugin/btr/btr0sea.c index 035fdbb61d2..9835efcf712 100644 --- a/storage/innodb_plugin/btr/btr0sea.c +++ b/storage/innodb_plugin/btr/btr0sea.c @@ -141,7 +141,7 @@ btr_search_check_free_space_in_heap(void) be enough free space in the hash table. */ if (heap->free_block == NULL) { - buf_block_t* block = buf_block_alloc(0); + buf_block_t* block = buf_block_alloc(); rw_lock_x_lock(&btr_search_latch); diff --git a/storage/innodb_plugin/buf/buf0buddy.c b/storage/innodb_plugin/buf/buf0buddy.c index ee5a569c3ff..63c99571510 100644 --- a/storage/innodb_plugin/buf/buf0buddy.c +++ b/storage/innodb_plugin/buf/buf0buddy.c @@ -327,7 +327,7 @@ buf_buddy_alloc_low( /* Try replacing an uncompressed page in the buffer pool. */ buf_pool_mutex_exit(); - block = buf_LRU_get_free_block(0); + block = buf_LRU_get_free_block(); *lru = TRUE; buf_pool_mutex_enter(); diff --git a/storage/innodb_plugin/buf/buf0buf.c b/storage/innodb_plugin/buf/buf0buf.c index dac416f9472..6e76e4c65be 100644 --- a/storage/innodb_plugin/buf/buf0buf.c +++ b/storage/innodb_plugin/buf/buf0buf.c @@ -1283,7 +1283,7 @@ shrink_again: buf_LRU_make_block_old(&block->page); dirty++; - } else if (buf_LRU_free_block(&block->page, TRUE, NULL) + } else if (buf_LRU_free_block(&block->page, TRUE) != BUF_LRU_FREED) { nonfree++; } @@ -1729,8 +1729,7 @@ err_exit: mutex_enter(block_mutex); /* Discard the uncompressed page frame if possible. */ - if (buf_LRU_free_block(bpage, FALSE, NULL) - == BUF_LRU_FREED) { + if (buf_LRU_free_block(bpage, FALSE) == BUF_LRU_FREED) { mutex_exit(block_mutex); goto lookup; @@ -2165,7 +2164,7 @@ wait_until_unfixed: buf_pool_mutex_exit(); mutex_exit(&buf_pool_zip_mutex); - block = buf_LRU_get_free_block(0); + block = buf_LRU_get_free_block(); ut_a(block); buf_pool_mutex_enter(); @@ -2291,8 +2290,7 @@ wait_until_unfixed: /* Try to evict the block from the buffer pool, to use the insert buffer as much as possible. */ - if (buf_LRU_free_block(&block->page, TRUE, NULL) - == BUF_LRU_FREED) { + if (buf_LRU_free_block(&block->page, TRUE) == BUF_LRU_FREED) { buf_pool_mutex_exit(); mutex_exit(&block->mutex); fprintf(stderr, @@ -2829,7 +2827,7 @@ buf_page_init_for_read( && UNIV_LIKELY(!recv_recovery_is_on())) { block = NULL; } else { - block = buf_LRU_get_free_block(0); + block = buf_LRU_get_free_block(); ut_ad(block); } @@ -3001,7 +2999,7 @@ buf_page_create( ut_ad(mtr->state == MTR_ACTIVE); ut_ad(space || !zip_size); - free_block = buf_LRU_get_free_block(0); + free_block = buf_LRU_get_free_block(); buf_pool_mutex_enter(); diff --git a/storage/innodb_plugin/buf/buf0lru.c b/storage/innodb_plugin/buf/buf0lru.c index e4cf218bf2e..39feb06ff23 100644 --- a/storage/innodb_plugin/buf/buf0lru.c +++ b/storage/innodb_plugin/buf/buf0lru.c @@ -575,7 +575,7 @@ buf_LRU_free_from_unzip_LRU_list( ut_ad(block->page.in_LRU_list); mutex_enter(&block->mutex); - freed = buf_LRU_free_block(&block->page, FALSE, NULL); + freed = buf_LRU_free_block(&block->page, FALSE); mutex_exit(&block->mutex); switch (freed) { @@ -636,7 +636,7 @@ buf_LRU_free_from_common_LRU_list( mutex_enter(block_mutex); accessed = buf_page_is_accessed(bpage); - freed = buf_LRU_free_block(bpage, TRUE, NULL); + freed = buf_LRU_free_block(bpage, TRUE); mutex_exit(block_mutex); switch (freed) { @@ -798,10 +798,8 @@ LRU list to the free list. @return the free control block, in state BUF_BLOCK_READY_FOR_USE */ UNIV_INTERN buf_block_t* -buf_LRU_get_free_block( -/*===================*/ - ulint zip_size) /*!< in: compressed page size in bytes, - or 0 if uncompressed tablespace */ +buf_LRU_get_free_block(void) +/*========================*/ { buf_block_t* block = NULL; ibool freed; @@ -877,26 +875,10 @@ loop: /* If there is a block in the free list, take it */ block = buf_LRU_get_free_only(); - if (block) { - -#ifdef UNIV_DEBUG - block->page.zip.m_start = -#endif /* UNIV_DEBUG */ - block->page.zip.m_end = - block->page.zip.m_nonempty = - block->page.zip.n_blobs = 0; - - if (UNIV_UNLIKELY(zip_size)) { - ibool lru; - page_zip_set_size(&block->page.zip, zip_size); - block->page.zip.data = buf_buddy_alloc(zip_size, &lru); - UNIV_MEM_DESC(block->page.zip.data, zip_size, block); - } else { - page_zip_set_size(&block->page.zip, 0); - block->page.zip.data = NULL; - } + buf_pool_mutex_exit(); - buf_pool_mutex_exit(); + if (block) { + memset(&block->page.zip, 0, sizeof block->page.zip); if (started_monitor) { srv_print_innodb_monitor = mon_value_was; @@ -908,8 +890,6 @@ loop: /* If no block was in the free list, search from the end of the LRU list and try to free a block there */ - buf_pool_mutex_exit(); - freed = buf_LRU_search_and_free_block(n_iterations); if (freed > 0) { @@ -1378,12 +1358,8 @@ enum buf_lru_free_block_status buf_LRU_free_block( /*===============*/ buf_page_t* bpage, /*!< in: block to be freed */ - ibool zip, /*!< in: TRUE if should remove also the + ibool zip) /*!< in: TRUE if should remove also the compressed page of an uncompressed page */ - ibool* buf_pool_mutex_released) - /*!< in: pointer to a variable that will - be assigned TRUE if buf_pool_mutex - was temporarily released, or NULL */ { buf_page_t* b = NULL; mutex_t* block_mutex = buf_page_get_mutex(bpage); @@ -1554,10 +1530,6 @@ alloc: b->io_fix = BUF_IO_READ; } - if (buf_pool_mutex_released) { - *buf_pool_mutex_released = TRUE; - } - buf_pool_mutex_exit(); mutex_exit(block_mutex); diff --git a/storage/innodb_plugin/dict/dict0dict.c b/storage/innodb_plugin/dict/dict0dict.c index 67765555658..ff56e9cb76a 100644 --- a/storage/innodb_plugin/dict/dict0dict.c +++ b/storage/innodb_plugin/dict/dict0dict.c @@ -1669,6 +1669,12 @@ undo_size_ok: new_index->heap, (1 + dict_index_get_n_unique(new_index)) * sizeof(ib_int64_t)); + + new_index->stat_n_non_null_key_vals = mem_heap_zalloc( + new_index->heap, + (1 + dict_index_get_n_unique(new_index)) + * sizeof(*new_index->stat_n_non_null_key_vals)); + /* Give some sensible values to stat_n_... in case we do not calculate statistics quickly enough */ @@ -4291,6 +4297,10 @@ dict_update_statistics( for (i = dict_index_get_n_unique(index); i; ) { index->stat_n_diff_key_vals[i--] = 1; } + + memset(index->stat_n_non_null_key_vals, 0, + (1 + dict_index_get_n_unique(index)) + * sizeof(*index->stat_n_non_null_key_vals)); } index = dict_table_get_next_index(index); diff --git a/storage/innodb_plugin/handler/ha_innodb.cc b/storage/innodb_plugin/handler/ha_innodb.cc index 86168e2bc9b..2d60c7397b0 100644 --- a/storage/innodb_plugin/handler/ha_innodb.cc +++ b/storage/innodb_plugin/handler/ha_innodb.cc @@ -174,6 +174,25 @@ static char* internal_innobase_data_file_path = NULL; static char* innodb_version_str = (char*) INNODB_VERSION_STR; +/** Possible values for system variable "innodb_stats_method". The values +are defined the same as its corresponding MyISAM system variable +"myisam_stats_method"(see "myisam_stats_method_names"), for better usability */ +static const char* innodb_stats_method_names[] = { + "nulls_equal", + "nulls_unequal", + "nulls_ignored", + NullS +}; + +/** Used to define an enumerate type of the system variable innodb_stats_method. +This is the same as "myisam_stats_method_typelib" */ +static TYPELIB innodb_stats_method_typelib = { + array_elements(innodb_stats_method_names) - 1, + "innodb_stats_method_typelib", + innodb_stats_method_names, + NULL +}; + /* The following counter is used to convey information to InnoDB about server activity: in selects it is not sensible to call srv_active_wake_master_thread after each fetch or search, we only do @@ -7507,6 +7526,65 @@ innobase_get_mysql_key_number_for_index( return(0); } + +/*********************************************************************//** +Calculate Record Per Key value. Need to exclude the NULL value if +innodb_stats_method is set to "nulls_ignored" +@return estimated record per key value */ +static +ha_rows +innodb_rec_per_key( +/*===============*/ + dict_index_t* index, /*!< in: dict_index_t structure */ + ulint i, /*!< in: the column we are + calculating rec per key */ + ha_rows records) /*!< in: estimated total records */ +{ + ha_rows rec_per_key; + + ut_ad(i < dict_index_get_n_unique(index)); + + /* Note the stat_n_diff_key_vals[] stores the diff value with + n-prefix indexing, so it is always stat_n_diff_key_vals[i + 1] */ + if (index->stat_n_diff_key_vals[i + 1] == 0) { + + rec_per_key = records; + } else if (srv_innodb_stats_method == SRV_STATS_NULLS_IGNORED) { + ib_int64_t num_null; + + /* Number of rows with NULL value in this + field */ + num_null = records - index->stat_n_non_null_key_vals[i]; + + /* In theory, index->stat_n_non_null_key_vals[i] + should always be less than the number of records. + Since this is statistics value, the value could + have slight discrepancy. But we will make sure + the number of null values is not a negative number. */ + num_null = (num_null < 0) ? 0 : num_null; + + /* If the number of NULL values is the same as or + large than that of the distinct values, we could + consider that the table consists mostly of NULL value. + Set rec_per_key to 1. */ + if (index->stat_n_diff_key_vals[i + 1] <= num_null) { + rec_per_key = 1; + } else { + /* Need to exclude rows with NULL values from + rec_per_key calculation */ + rec_per_key = (ha_rows)( + (records - num_null) + / (index->stat_n_diff_key_vals[i + 1] + - num_null)); + } + } else { + rec_per_key = (ha_rows) + (records / index->stat_n_diff_key_vals[i + 1]); + } + + return(rec_per_key); +} + /*********************************************************************//** Returns statistics information of the table to the MySQL interpreter, in various fields of the handle object. */ @@ -7737,13 +7815,8 @@ ha_innobase::info_low( break; } - if (index->stat_n_diff_key_vals[j + 1] == 0) { - - rec_per_key = stats.records; - } else { - rec_per_key = (ha_rows)(stats.records / - index->stat_n_diff_key_vals[j + 1]); - } + rec_per_key = innodb_rec_per_key( + index, j, stats.records); /* Since MySQL seems to favor table scans too much over index searches, we pretend @@ -10934,6 +11007,13 @@ static MYSQL_SYSVAR_STR(change_buffering, innobase_change_buffering, innodb_change_buffering_validate, innodb_change_buffering_update, "inserts"); +static MYSQL_SYSVAR_ENUM(stats_method, srv_innodb_stats_method, + PLUGIN_VAR_RQCMDARG, + "Specifies how InnoDB index statistics collection code should " + "treat NULLs. Possible values are NULLS_EQUAL (default), " + "NULLS_UNEQUAL and NULLS_IGNORED", + NULL, NULL, SRV_STATS_NULLS_EQUAL, &innodb_stats_method_typelib); + #if defined UNIV_DEBUG || defined UNIV_IBUF_DEBUG static MYSQL_SYSVAR_UINT(change_buffering_debug, ibuf_debug, PLUGIN_VAR_RQCMDARG, @@ -10988,6 +11068,7 @@ static struct st_mysql_sys_var* innobase_system_variables[]= { MYSQL_SYSVAR(stats_on_metadata), MYSQL_SYSVAR(stats_sample_pages), MYSQL_SYSVAR(adaptive_hash_index), + MYSQL_SYSVAR(stats_method), MYSQL_SYSVAR(replication_delay), MYSQL_SYSVAR(status_file), MYSQL_SYSVAR(strict_mode), diff --git a/storage/innodb_plugin/include/btr0cur.h b/storage/innodb_plugin/include/btr0cur.h index b477ad0320a..cb8cb399715 100644 --- a/storage/innodb_plugin/include/btr0cur.h +++ b/storage/innodb_plugin/include/btr0cur.h @@ -478,7 +478,10 @@ btr_estimate_n_rows_in_range( /*******************************************************************//** Estimates the number of different key values in a given index, for each n-column prefix of the index where n <= dict_index_get_n_unique(index). -The estimates are stored in the array index->stat_n_diff_key_vals. */ +The estimates are stored in the array index->stat_n_diff_key_vals. +If innodb_stats_method is nulls_ignored, we also record the number of +non-null values for each prefix and stored the estimates in +array index->stat_n_non_null_key_vals. */ UNIV_INTERN void btr_estimate_number_of_different_key_vals( diff --git a/storage/innodb_plugin/include/buf0buf.h b/storage/innodb_plugin/include/buf0buf.h index cd4ee5906f0..d903b443920 100644 --- a/storage/innodb_plugin/include/buf0buf.h +++ b/storage/innodb_plugin/include/buf0buf.h @@ -165,10 +165,8 @@ Allocates a buffer block. @return own: the allocated block, in state BUF_BLOCK_MEMORY */ UNIV_INLINE buf_block_t* -buf_block_alloc( -/*============*/ - ulint zip_size); /*!< in: compressed page size in bytes, - or 0 if uncompressed tablespace */ +buf_block_alloc(void); +/*=================*/ /********************************************************************//** Frees a buffer block which does not contain a file page. */ UNIV_INLINE diff --git a/storage/innodb_plugin/include/buf0buf.ic b/storage/innodb_plugin/include/buf0buf.ic index 23db684806c..0025bef5aac 100644 --- a/storage/innodb_plugin/include/buf0buf.ic +++ b/storage/innodb_plugin/include/buf0buf.ic @@ -719,14 +719,12 @@ Allocates a buffer block. @return own: the allocated block, in state BUF_BLOCK_MEMORY */ UNIV_INLINE buf_block_t* -buf_block_alloc( -/*============*/ - ulint zip_size) /*!< in: compressed page size in bytes, - or 0 if uncompressed tablespace */ +buf_block_alloc(void) +/*=================*/ { buf_block_t* block; - block = buf_LRU_get_free_block(zip_size); + block = buf_LRU_get_free_block(); buf_block_set_state(block, BUF_BLOCK_MEMORY); diff --git a/storage/innodb_plugin/include/buf0lru.h b/storage/innodb_plugin/include/buf0lru.h index 5a9cfd059f3..d543bce53cd 100644 --- a/storage/innodb_plugin/include/buf0lru.h +++ b/storage/innodb_plugin/include/buf0lru.h @@ -110,12 +110,9 @@ enum buf_lru_free_block_status buf_LRU_free_block( /*===============*/ buf_page_t* bpage, /*!< in: block to be freed */ - ibool zip, /*!< in: TRUE if should remove also the + ibool zip) /*!< in: TRUE if should remove also the compressed page of an uncompressed page */ - ibool* buf_pool_mutex_released); - /*!< in: pointer to a variable that will - be assigned TRUE if buf_pool_mutex - was temporarily released, or NULL */ + __attribute__((nonnull)); /******************************************************************//** Try to free a replaceable block. @return TRUE if found and freed */ @@ -146,10 +143,9 @@ LRU list to the free list. @return the free control block, in state BUF_BLOCK_READY_FOR_USE */ UNIV_INTERN buf_block_t* -buf_LRU_get_free_block( -/*===================*/ - ulint zip_size); /*!< in: compressed page size in bytes, - or 0 if uncompressed tablespace */ +buf_LRU_get_free_block(void) +/*========================*/ + __attribute__((warn_unused_result)); /******************************************************************//** Puts a block back to the free list. */ diff --git a/storage/innodb_plugin/include/dict0mem.h b/storage/innodb_plugin/include/dict0mem.h index 19782c2e76a..09a068ccb93 100644 --- a/storage/innodb_plugin/include/dict0mem.h +++ b/storage/innodb_plugin/include/dict0mem.h @@ -321,6 +321,12 @@ struct dict_index_struct{ dict_get_n_unique(index); we periodically calculate new estimates */ + ib_int64_t* stat_n_non_null_key_vals; + /* approximate number of non-null key values + for this index, for each column where + n < dict_get_n_unique(index); This + is used when innodb_stats_method is + "nulls_ignored". */ ulint stat_index_size; /*!< approximate index size in database pages */ diff --git a/storage/innodb_plugin/include/dict0types.h b/storage/innodb_plugin/include/dict0types.h index 7ad69193cc9..f14b59a19d4 100644 --- a/storage/innodb_plugin/include/dict0types.h +++ b/storage/innodb_plugin/include/dict0types.h @@ -33,11 +33,6 @@ typedef struct dict_index_struct dict_index_t; typedef struct dict_table_struct dict_table_t; typedef struct dict_foreign_struct dict_foreign_t; -/* A cluster object is a table object with the type field set to -DICT_CLUSTERED */ - -typedef dict_table_t dict_cluster_t; - typedef struct ind_node_struct ind_node_t; typedef struct tab_node_struct tab_node_t; diff --git a/storage/innodb_plugin/include/rem0cmp.h b/storage/innodb_plugin/include/rem0cmp.h index 2f751a38864..a908521c9f7 100644 --- a/storage/innodb_plugin/include/rem0cmp.h +++ b/storage/innodb_plugin/include/rem0cmp.h @@ -165,6 +165,10 @@ cmp_rec_rec_with_match( const ulint* offsets1,/*!< in: rec_get_offsets(rec1, index) */ const ulint* offsets2,/*!< in: rec_get_offsets(rec2, index) */ dict_index_t* index, /*!< in: data dictionary index */ + ibool nulls_unequal, + /* in: TRUE if this is for index statistics + cardinality estimation, and innodb_stats_method + is "nulls_unequal" or "nulls_ignored" */ ulint* matched_fields, /*!< in/out: number of already completely matched fields; when the function returns, contains the value the for current diff --git a/storage/innodb_plugin/include/rem0cmp.ic b/storage/innodb_plugin/include/rem0cmp.ic index 39ef5f4fba3..63415fe7837 100644 --- a/storage/innodb_plugin/include/rem0cmp.ic +++ b/storage/innodb_plugin/include/rem0cmp.ic @@ -87,5 +87,5 @@ cmp_rec_rec( ulint match_b = 0; return(cmp_rec_rec_with_match(rec1, rec2, offsets1, offsets2, index, - &match_f, &match_b)); + FALSE, &match_f, &match_b)); } diff --git a/storage/innodb_plugin/include/srv0srv.h b/storage/innodb_plugin/include/srv0srv.h index 7aa2ce74720..91ae895040c 100644 --- a/storage/innodb_plugin/include/srv0srv.h +++ b/storage/innodb_plugin/include/srv0srv.h @@ -154,6 +154,11 @@ capacity. PCT_IO(5) -> returns the number of IO operations that is 5% of the max where max is srv_io_capacity. */ #define PCT_IO(p) ((ulong) (srv_io_capacity * ((double) p / 100.0))) +/* The "innodb_stats_method" setting, decides how InnoDB is going +to treat NULL value when collecting statistics. It is not defined +as enum type because the configure option takes unsigned integer type. */ +extern ulong srv_innodb_stats_method; + #ifdef UNIV_LOG_ARCHIVE extern ibool srv_log_archive_on; extern ibool srv_archive_recovery; @@ -363,6 +368,19 @@ enum { in connection with recovery */ }; +/* Alternatives for srv_innodb_stats_method, which could be changed by +setting innodb_stats_method */ +enum srv_stats_method_name_enum { + SRV_STATS_NULLS_EQUAL, /* All NULL values are treated as + equal. This is the default setting + for innodb_stats_method */ + SRV_STATS_NULLS_UNEQUAL, /* All NULL values are treated as + NOT equal. */ + SRV_STATS_NULLS_IGNORED /* NULL values are ignored */ +}; + +typedef enum srv_stats_method_name_enum srv_stats_method_name_t; + #ifndef UNIV_HOTBACKUP /** Types of threads existing in the system. */ enum srv_thread_type { diff --git a/storage/innodb_plugin/include/sync0rw.h b/storage/innodb_plugin/include/sync0rw.h index 175f3deb77c..47f7dbfe0eb 100644 --- a/storage/innodb_plugin/include/sync0rw.h +++ b/storage/innodb_plugin/include/sync0rw.h @@ -1,6 +1,6 @@ /***************************************************************************** -Copyright (c) 1995, 2010, Innobase Oy. All Rights Reserved. +Copyright (c) 1995, 2011, Oracle and/or its affiliates. All Rights Reserved. Copyright (c) 2008, Google Inc. Portions of this file contain modifications contributed and copyrighted by @@ -490,6 +490,7 @@ UNIV_INTERN void rw_lock_debug_print( /*================*/ + FILE* f, /*!< in: output stream */ rw_lock_debug_t* info); /*!< in: debug struct */ #endif /* UNIV_SYNC_DEBUG */ diff --git a/storage/innodb_plugin/include/trx0rseg.h b/storage/innodb_plugin/include/trx0rseg.h index a25d84f1e84..e3674089735 100644 --- a/storage/innodb_plugin/include/trx0rseg.h +++ b/storage/innodb_plugin/include/trx0rseg.h @@ -135,9 +135,7 @@ struct trx_rseg_struct{ ulint id; /*!< rollback segment id == the index of its slot in the trx system file copy */ mutex_t mutex; /*!< mutex protecting the fields in this - struct except id; NOTE that the latching - order must always be kernel mutex -> - rseg mutex */ + struct except id, which is constant */ ulint space; /*!< space where the rollback segment is header is placed */ ulint zip_size;/* compressed page size of space diff --git a/storage/innodb_plugin/include/trx0trx.h b/storage/innodb_plugin/include/trx0trx.h index abd175d365b..833bae4a4ff 100644 --- a/storage/innodb_plugin/include/trx0trx.h +++ b/storage/innodb_plugin/include/trx0trx.h @@ -214,12 +214,12 @@ trx_recover_for_mysql( /*******************************************************************//** This function is used to find one X/Open XA distributed transaction which is in the prepared state -@return trx or NULL */ +@return trx or NULL; on match, the trx->xid will be invalidated */ UNIV_INTERN trx_t * trx_get_trx_by_xid( /*===============*/ - XID* xid); /*!< in: X/Open XA transaction identification */ + const XID* xid); /*!< in: X/Open XA transaction identifier */ /**********************************************************************//** If required, flushes the log to disk if we called trx_commit_for_mysql() with trx->flush_log_later == TRUE. diff --git a/storage/innodb_plugin/include/univ.i b/storage/innodb_plugin/include/univ.i index bbff8ddf1e3..4425950748b 100644 --- a/storage/innodb_plugin/include/univ.i +++ b/storage/innodb_plugin/include/univ.i @@ -412,7 +412,7 @@ it is read or written. */ /* Use sun_prefetch when compile with Sun Studio */ # define UNIV_EXPECT(expr,value) (expr) # define UNIV_LIKELY_NULL(expr) (expr) -# define UNIV_PREFETCH_R(addr) sun_prefetch_read_many(addr) +# define UNIV_PREFETCH_R(addr) sun_prefetch_read_many((void*) addr) # define UNIV_PREFETCH_RW(addr) sun_prefetch_write_many(addr) #else /* Dummy versions of the macros */ diff --git a/storage/innodb_plugin/mem/mem0mem.c b/storage/innodb_plugin/mem/mem0mem.c index 1dd4db30841..86100b04fd6 100644 --- a/storage/innodb_plugin/mem/mem0mem.c +++ b/storage/innodb_plugin/mem/mem0mem.c @@ -347,7 +347,7 @@ mem_heap_create_block( return(NULL); } } else { - buf_block = buf_block_alloc(0); + buf_block = buf_block_alloc(); } block = (mem_block_t*) buf_block->frame; diff --git a/storage/innodb_plugin/mtr/mtr0log.c b/storage/innodb_plugin/mtr/mtr0log.c index 3f3dab36b76..3349036b5b3 100644 --- a/storage/innodb_plugin/mtr/mtr0log.c +++ b/storage/innodb_plugin/mtr/mtr0log.c @@ -408,7 +408,7 @@ mlog_parse_string( ptr += 2; if (UNIV_UNLIKELY(offset >= UNIV_PAGE_SIZE) - || UNIV_UNLIKELY(len + offset) > UNIV_PAGE_SIZE) { + || UNIV_UNLIKELY(len + offset > UNIV_PAGE_SIZE)) { recv_sys->found_corrupt_log = TRUE; return(NULL); diff --git a/storage/innodb_plugin/page/page0zip.c b/storage/innodb_plugin/page/page0zip.c index d3b1edefc6b..bb9b0995c72 100644 --- a/storage/innodb_plugin/page/page0zip.c +++ b/storage/innodb_plugin/page/page0zip.c @@ -4439,7 +4439,7 @@ page_zip_reorganize( log_mode = mtr_set_log_mode(mtr, MTR_LOG_NONE); #ifndef UNIV_HOTBACKUP - temp_block = buf_block_alloc(0); + temp_block = buf_block_alloc(); btr_search_drop_page_hash_index(block); block->check_index_page_at_flush = TRUE; #else /* !UNIV_HOTBACKUP */ diff --git a/storage/innodb_plugin/rem/rem0cmp.c b/storage/innodb_plugin/rem/rem0cmp.c index 35b67992558..04d2c15437b 100644 --- a/storage/innodb_plugin/rem/rem0cmp.c +++ b/storage/innodb_plugin/rem/rem0cmp.c @@ -862,6 +862,10 @@ cmp_rec_rec_with_match( const ulint* offsets1,/*!< in: rec_get_offsets(rec1, index) */ const ulint* offsets2,/*!< in: rec_get_offsets(rec2, index) */ dict_index_t* index, /*!< in: data dictionary index */ + ibool nulls_unequal, + /* in: TRUE if this is for index statistics + cardinality estimation, and innodb_stats_method + is "nulls_unequal" or "nulls_ignored" */ ulint* matched_fields, /*!< in/out: number of already completely matched fields; when the function returns, contains the value the for current @@ -961,9 +965,13 @@ cmp_rec_rec_with_match( || rec2_f_len == UNIV_SQL_NULL) { if (rec1_f_len == rec2_f_len) { - - goto next_field; - + /* This is limited to stats collection, + cannot use it for regular search */ + if (nulls_unequal) { + ret = -1; + } else { + goto next_field; + } } else if (rec2_f_len == UNIV_SQL_NULL) { /* We define the SQL null to be the diff --git a/storage/innodb_plugin/row/row0upd.c b/storage/innodb_plugin/row/row0upd.c index 4aa1474a25b..691d263e6ed 100644 --- a/storage/innodb_plugin/row/row0upd.c +++ b/storage/innodb_plugin/row/row0upd.c @@ -1252,6 +1252,10 @@ row_upd_changes_ord_field_binary( || dfield_is_null(dfield)) { /* do nothing special */ } else if (UNIV_LIKELY_NULL(ext)) { + /* Silence a compiler warning without + silencing a Valgrind error. */ + dfield_len = 0; + UNIV_MEM_INVALID(&dfield_len, sizeof dfield_len); /* See if the column is stored externally. */ buf = row_ext_lookup(ext, col_no, &dfield_len); diff --git a/storage/innodb_plugin/row/row0vers.c b/storage/innodb_plugin/row/row0vers.c index b6d35363f08..d4fde0b939b 100644 --- a/storage/innodb_plugin/row/row0vers.c +++ b/storage/innodb_plugin/row/row0vers.c @@ -669,11 +669,15 @@ row_vers_build_for_semi_consistent_read( mutex_enter(&kernel_mutex); version_trx = trx_get_on_id(version_trx_id); + if (version_trx + && (version_trx->conc_state == TRX_COMMITTED_IN_MEMORY + || version_trx->conc_state == TRX_NOT_STARTED)) { + + version_trx = NULL; + } mutex_exit(&kernel_mutex); - if (!version_trx - || version_trx->conc_state == TRX_NOT_STARTED - || version_trx->conc_state == TRX_COMMITTED_IN_MEMORY) { + if (!version_trx) { /* We found a version that belongs to a committed transaction: return it. */ diff --git a/storage/innodb_plugin/srv/srv0srv.c b/storage/innodb_plugin/srv/srv0srv.c index f7e7e351bdc..3cf17f33c40 100644 --- a/storage/innodb_plugin/srv/srv0srv.c +++ b/storage/innodb_plugin/srv/srv0srv.c @@ -243,6 +243,11 @@ UNIV_INTERN ulong srv_max_buf_pool_modified_pct = 75; /* variable counts amount of data read in total (in bytes) */ UNIV_INTERN ulint srv_data_read = 0; +/* Internal setting for "innodb_stats_method". Decides how InnoDB treats +NULL value when collecting statistics. By default, it is set to +SRV_STATS_NULLS_EQUAL(0), ie. all NULL value are treated equal */ +ulong srv_innodb_stats_method = SRV_STATS_NULLS_EQUAL; + /* here we count the amount of data written in total (in bytes) */ UNIV_INTERN ulint srv_data_written = 0; diff --git a/storage/innodb_plugin/sync/sync0arr.c b/storage/innodb_plugin/sync/sync0arr.c index 3c825e2202b..ad29b90d344 100644 --- a/storage/innodb_plugin/sync/sync0arr.c +++ b/storage/innodb_plugin/sync/sync0arr.c @@ -1,6 +1,6 @@ /***************************************************************************** -Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved. +Copyright (c) 1995, 2011, Oracle and/or its affiliates. All Rights Reserved. Copyright (c) 2008, Google Inc. Portions of this file contain modifications contributed and copyrighted by @@ -715,7 +715,7 @@ print: fprintf(stderr, "rw-lock %p ", (void*) lock); sync_array_cell_print(stderr, cell); - rw_lock_debug_print(debug); + rw_lock_debug_print(stderr, debug); return(TRUE); } } diff --git a/storage/innodb_plugin/sync/sync0rw.c b/storage/innodb_plugin/sync/sync0rw.c index 572c3690a7f..00e0324becd 100644 --- a/storage/innodb_plugin/sync/sync0rw.c +++ b/storage/innodb_plugin/sync/sync0rw.c @@ -1,6 +1,6 @@ /***************************************************************************** -Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved. +Copyright (c) 1995, 2011, Oracle and/or its affiliates. All Rights Reserved. Copyright (c) 2008, Google Inc. Portions of this file contain modifications contributed and copyrighted by @@ -925,7 +925,7 @@ rw_lock_list_print_info( info = UT_LIST_GET_FIRST(lock->debug_list); while (info != NULL) { - rw_lock_debug_print(info); + rw_lock_debug_print(file, info); info = UT_LIST_GET_NEXT(list, info); } } @@ -973,7 +973,7 @@ rw_lock_print( info = UT_LIST_GET_FIRST(lock->debug_list); while (info != NULL) { - rw_lock_debug_print(info); + rw_lock_debug_print(stderr, info); info = UT_LIST_GET_NEXT(list, info); } } @@ -985,28 +985,29 @@ UNIV_INTERN void rw_lock_debug_print( /*================*/ + FILE* f, /*!< in: output stream */ rw_lock_debug_t* info) /*!< in: debug struct */ { ulint rwt; rwt = info->lock_type; - fprintf(stderr, "Locked: thread %lu file %s line %lu ", + fprintf(f, "Locked: thread %lu file %s line %lu ", (ulong) os_thread_pf(info->thread_id), info->file_name, (ulong) info->line); if (rwt == RW_LOCK_SHARED) { - fputs("S-LOCK", stderr); + fputs("S-LOCK", f); } else if (rwt == RW_LOCK_EX) { - fputs("X-LOCK", stderr); + fputs("X-LOCK", f); } else if (rwt == RW_LOCK_WAIT_EX) { - fputs("WAIT X-LOCK", stderr); + fputs("WAIT X-LOCK", f); } else { ut_error; } if (info->pass != 0) { - fprintf(stderr, " pass value %lu", (ulong) info->pass); + fprintf(f, " pass value %lu", (ulong) info->pass); } - putc('\n', stderr); + putc('\n', f); } /***************************************************************//** diff --git a/storage/innodb_plugin/trx/trx0trx.c b/storage/innodb_plugin/trx/trx0trx.c index ee744fd58b1..f0bbf220815 100644 --- a/storage/innodb_plugin/trx/trx0trx.c +++ b/storage/innodb_plugin/trx/trx0trx.c @@ -2010,18 +2010,18 @@ trx_recover_for_mysql( /*******************************************************************//** This function is used to find one X/Open XA distributed transaction which is in the prepared state -@return trx or NULL */ +@return trx or NULL; on match, the trx->xid will be invalidated */ UNIV_INTERN trx_t* trx_get_trx_by_xid( /*===============*/ - XID* xid) /*!< in: X/Open XA transaction identification */ + const XID* xid) /*!< in: X/Open XA transaction identifier */ { trx_t* trx; if (xid == NULL) { - return (NULL); + return(NULL); } mutex_enter(&kernel_mutex); @@ -2034,10 +2034,16 @@ trx_get_trx_by_xid( of gtrid_lenght+bqual_length bytes should be the same */ - if (xid->gtrid_length == trx->xid.gtrid_length + if (trx->conc_state == TRX_PREPARED + && xid->gtrid_length == trx->xid.gtrid_length && xid->bqual_length == trx->xid.bqual_length && memcmp(xid->data, trx->xid.data, xid->gtrid_length + xid->bqual_length) == 0) { + + /* Invalidate the XID, so that subsequent calls + will not find it. */ + memset(&trx->xid, 0, sizeof(trx->xid)); + trx->xid.formatID = -1; break; } @@ -2046,14 +2052,5 @@ trx_get_trx_by_xid( mutex_exit(&kernel_mutex); - if (trx) { - if (trx->conc_state != TRX_PREPARED) { - - return(NULL); - } - - return(trx); - } else { - return(NULL); - } + return(trx); } |