diff --git a/mysql-test/suite/innodb_plugin/r/innodb_bug56680.result b/mysql-test/suite/innodb_plugin/r/innodb_bug56680.result new file mode 100644 index 00000000000..5e798b69167 --- /dev/null +++ b/mysql-test/suite/innodb_plugin/r/innodb_bug56680.result @@ -0,0 +1,109 @@ +SET GLOBAL tx_isolation='REPEATABLE-READ'; +SET GLOBAL innodb_file_format=Barracuda; +SET GLOBAL innodb_file_per_table=on; +CREATE TABLE bug56680( +a INT AUTO_INCREMENT PRIMARY KEY, +b CHAR(1), +c INT, +INDEX(b)) +ENGINE=InnoDB; +INSERT INTO bug56680 VALUES(0,'x',1); +BEGIN; +SELECT b FROM bug56680; +b +x +BEGIN; +UPDATE bug56680 SET b='X'; +SELECT b FROM bug56680; +b +x +SELECT * FROM bug56680; +a b c +1 x 1 +ROLLBACK; +SELECT b FROM bug56680; +b +x +SET GLOBAL tx_isolation='READ-UNCOMMITTED'; +INSERT INTO bug56680 SELECT 0,b,c FROM bug56680; +INSERT INTO bug56680 SELECT 0,b,c FROM bug56680; +INSERT INTO bug56680 SELECT 0,b,c FROM bug56680; +INSERT INTO bug56680 SELECT 0,b,c FROM bug56680; +INSERT INTO bug56680 SELECT 0,b,c FROM bug56680; +INSERT INTO bug56680 SELECT 0,b,c FROM bug56680; +INSERT INTO bug56680 SELECT 0,b,c FROM bug56680; +INSERT INTO bug56680 SELECT 0,b,c FROM bug56680; +INSERT INTO bug56680 SELECT 0,b,c FROM bug56680; +INSERT INTO bug56680 SELECT 0,b,c FROM bug56680; +INSERT INTO bug56680 SELECT 0,b,c FROM bug56680; +BEGIN; +SELECT b FROM bug56680 LIMIT 2; +b +x +x +BEGIN; +DELETE FROM bug56680 WHERE a=1; +INSERT INTO bug56680 VALUES(1,'X',1); +SELECT b FROM bug56680 LIMIT 3; +b +X +x +x +SELECT b FROM bug56680 LIMIT 2; +b +x +x +CHECK TABLE bug56680; +Table Op Msg_type Msg_text +test.bug56680 check status OK +ROLLBACK; +SELECT b FROM bug56680 LIMIT 2; +b +x +x +CHECK TABLE bug56680; +Table Op Msg_type Msg_text +test.bug56680 check status OK +SELECT b FROM bug56680 LIMIT 2; +b +x +x +CREATE TABLE bug56680_2( +a INT AUTO_INCREMENT PRIMARY KEY, +b VARCHAR(2) CHARSET latin1 COLLATE latin1_german2_ci, +c INT, +INDEX(b)) +ENGINE=InnoDB; +INSERT INTO bug56680_2 SELECT 0,_latin1 0xdf,c FROM bug56680; +BEGIN; +SELECT HEX(b) FROM bug56680_2 LIMIT 2; +HEX(b) +DF +DF +DELETE FROM bug56680_2 WHERE a=1; +INSERT INTO bug56680_2 VALUES(1,'SS',1); +SELECT HEX(b) FROM bug56680_2 LIMIT 3; +HEX(b) +5353 +DF +DF +CHECK TABLE bug56680_2; +Table Op Msg_type Msg_text +test.bug56680_2 check status OK +ALTER TABLE bug56680_2 ROW_FORMAT=COMPRESSED KEY_BLOCK_SIZE=1; +SELECT HEX(b) FROM bug56680_2 LIMIT 2; +HEX(b) +5353 +DF +DELETE FROM bug56680_2 WHERE a=1; +INSERT INTO bug56680_2 VALUES(1,_latin1 0xdf,1); +SELECT HEX(b) FROM bug56680_2 LIMIT 3; +HEX(b) +DF +DF +DF +CHECK TABLE bug56680_2; +Table Op Msg_type Msg_text +test.bug56680_2 check status OK +DROP TABLE bug56680_2; +DROP TABLE bug56680; diff --git a/mysql-test/suite/innodb_plugin/t/innodb_bug56680.test b/mysql-test/suite/innodb_plugin/t/innodb_bug56680.test new file mode 100644 index 00000000000..8d0f685c723 --- /dev/null +++ b/mysql-test/suite/innodb_plugin/t/innodb_bug56680.test @@ -0,0 +1,142 @@ +# +# Bug #56680 InnoDB may return wrong results from a case-insensitive index +# +-- source include/have_innodb_plugin.inc + +-- disable_query_log +SET @tx_isolation_orig = @@tx_isolation; +SET @innodb_file_per_table_orig = @@innodb_file_per_table; +SET @innodb_file_format_orig = @@innodb_file_format; +SET @innodb_file_format_check_orig = @@innodb_file_format_check; +# The flag innodb_change_buffering_debug is only available in debug builds. +# It instructs InnoDB to try to evict pages from the buffer pool when +# change buffering is possible, so that the change buffer will be used +# whenever possible. +-- error 0,ER_UNKNOWN_SYSTEM_VARIABLE +SET @innodb_change_buffering_debug_orig = @@innodb_change_buffering_debug; +-- error 0,ER_UNKNOWN_SYSTEM_VARIABLE +SET GLOBAL innodb_change_buffering_debug = 1; +-- enable_query_log +SET GLOBAL tx_isolation='REPEATABLE-READ'; +SET GLOBAL innodb_file_format=Barracuda; +SET GLOBAL innodb_file_per_table=on; + +CREATE TABLE bug56680( + a INT AUTO_INCREMENT PRIMARY KEY, + b CHAR(1), + c INT, + INDEX(b)) +ENGINE=InnoDB; + +INSERT INTO bug56680 VALUES(0,'x',1); +BEGIN; +SELECT b FROM bug56680; + +connect (con1,localhost,root,,); +connection con1; +BEGIN; +UPDATE bug56680 SET b='X'; + +connection default; +# This should return the last committed value 'x', but would return 'X' +# due to a bug in row_search_for_mysql(). +SELECT b FROM bug56680; +# This would always return the last committed value 'x'. +SELECT * FROM bug56680; + +connection con1; +ROLLBACK; +disconnect con1; + +connection default; + +SELECT b FROM bug56680; + +# For the rest of this test, use the READ UNCOMMITTED isolation level +# to see what exists in the secondary index. +SET GLOBAL tx_isolation='READ-UNCOMMITTED'; + +# Create enough rows for the table, so that the insert buffer will be +# used for modifying the secondary index page. There must be multiple +# index pages, because changes to the root page are never buffered. + +INSERT INTO bug56680 SELECT 0,b,c FROM bug56680; +INSERT INTO bug56680 SELECT 0,b,c FROM bug56680; +INSERT INTO bug56680 SELECT 0,b,c FROM bug56680; +INSERT INTO bug56680 SELECT 0,b,c FROM bug56680; +INSERT INTO bug56680 SELECT 0,b,c FROM bug56680; +INSERT INTO bug56680 SELECT 0,b,c FROM bug56680; +INSERT INTO bug56680 SELECT 0,b,c FROM bug56680; +INSERT INTO bug56680 SELECT 0,b,c FROM bug56680; +INSERT INTO bug56680 SELECT 0,b,c FROM bug56680; +INSERT INTO bug56680 SELECT 0,b,c FROM bug56680; +INSERT INTO bug56680 SELECT 0,b,c FROM bug56680; + +BEGIN; +SELECT b FROM bug56680 LIMIT 2; + +connect (con1,localhost,root,,); +connection con1; +BEGIN; +DELETE FROM bug56680 WHERE a=1; +# This should be buffered, if innodb_change_buffering_debug = 1 is in effect. +INSERT INTO bug56680 VALUES(1,'X',1); + +# This should force an insert buffer merge, and return 'X' in the first row. +SELECT b FROM bug56680 LIMIT 3; + +connection default; +SELECT b FROM bug56680 LIMIT 2; +CHECK TABLE bug56680; + +connection con1; +ROLLBACK; +SELECT b FROM bug56680 LIMIT 2; +CHECK TABLE bug56680; + +connection default; +disconnect con1; + +SELECT b FROM bug56680 LIMIT 2; + +CREATE TABLE bug56680_2( + a INT AUTO_INCREMENT PRIMARY KEY, + b VARCHAR(2) CHARSET latin1 COLLATE latin1_german2_ci, + c INT, + INDEX(b)) +ENGINE=InnoDB; + +INSERT INTO bug56680_2 SELECT 0,_latin1 0xdf,c FROM bug56680; + +BEGIN; +SELECT HEX(b) FROM bug56680_2 LIMIT 2; +DELETE FROM bug56680_2 WHERE a=1; +# This should be buffered, if innodb_change_buffering_debug = 1 is in effect. +INSERT INTO bug56680_2 VALUES(1,'SS',1); + +# This should force an insert buffer merge, and return 'SS' in the first row. +SELECT HEX(b) FROM bug56680_2 LIMIT 3; +CHECK TABLE bug56680_2; + +# Test this with compressed tables. +ALTER TABLE bug56680_2 ROW_FORMAT=COMPRESSED KEY_BLOCK_SIZE=1; + +SELECT HEX(b) FROM bug56680_2 LIMIT 2; +DELETE FROM bug56680_2 WHERE a=1; +# This should be buffered, if innodb_change_buffering_debug = 1 is in effect. +INSERT INTO bug56680_2 VALUES(1,_latin1 0xdf,1); + +# This should force an insert buffer merge, and return 0xdf in the first row. +SELECT HEX(b) FROM bug56680_2 LIMIT 3; +CHECK TABLE bug56680_2; + +DROP TABLE bug56680_2; +DROP TABLE bug56680; + +-- disable_query_log +SET GLOBAL tx_isolation = @tx_isolation_orig; +SET GLOBAL innodb_file_per_table = @innodb_file_per_table_orig; +SET GLOBAL innodb_file_format = @innodb_file_format_orig; +SET GLOBAL innodb_file_format_check = @innodb_file_format_check_orig; +-- error 0, ER_UNKNOWN_SYSTEM_VARIABLE +SET GLOBAL innodb_change_buffering_debug = @innodb_change_buffering_debug_orig; diff --git a/storage/innodb_plugin/ChangeLog b/storage/innodb_plugin/ChangeLog index 488669346fd..db37fa2b4bb 100644 --- a/storage/innodb_plugin/ChangeLog +++ b/storage/innodb_plugin/ChangeLog @@ -1,3 +1,13 @@ +2010-10-19 The InnoDB Team + + * btr/btr0cur.c, buf/buf0buf.c, buf/buf0flu.c, handler/ha_innodb.cc, + ibuf/ibuf0ibuf.c, include/btr0cur.h, include/buf0flu.h, + include/ibuf0ibuf.h, include/row0mysql.h, + row/row0mysql.c, row/row0sel.c, + innodb_bug56680.test, innodb_bug56680.result: + Fix Bug #56680 InnoDB may return wrong results from a + case-insensitive covering index + 2010-10-18 The InnoDB Team * handler/ha_innodb.cc, handler/ha_innodb.h, innodb_bug57252.result, diff --git a/storage/innodb_plugin/btr/btr0cur.c b/storage/innodb_plugin/btr/btr0cur.c index 9812e91e8a5..79fe328a631 100644 --- a/storage/innodb_plugin/btr/btr0cur.c +++ b/storage/innodb_plugin/btr/btr0cur.c @@ -1626,7 +1626,7 @@ func_exit: See if there is enough place in the page modification log to log an update-in-place. @return TRUE if enough place */ -static +UNIV_INTERN ibool btr_cur_update_alloc_zip( /*=====================*/ diff --git a/storage/innodb_plugin/buf/buf0buf.c b/storage/innodb_plugin/buf/buf0buf.c index 660686bac1e..65d4311b840 100644 --- a/storage/innodb_plugin/buf/buf0buf.c +++ b/storage/innodb_plugin/buf/buf0buf.c @@ -2286,6 +2286,30 @@ wait_until_unfixed: bytes. */ UNIV_MEM_ASSERT_RW(&block->page, sizeof block->page); #endif +#if defined UNIV_DEBUG || defined UNIV_IBUF_DEBUG + if (mode == BUF_GET_IF_IN_POOL && ibuf_debug) { + /* 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) { + buf_pool_mutex_exit(); + mutex_exit(&block->mutex); + fprintf(stderr, + "innodb_change_buffering_debug evict %u %u\n", + (unsigned) space, (unsigned) offset); + return(NULL); + } else if (buf_flush_page_try(block)) { + fprintf(stderr, + "innodb_change_buffering_debug flush %u %u\n", + (unsigned) space, (unsigned) offset); + guess = block; + goto loop; + } + + /* Failed to evict the page; change it directly */ + } +#endif /* UNIV_DEBUG || UNIV_IBUF_DEBUG */ buf_block_buf_fix_inc(block, file, line); diff --git a/storage/innodb_plugin/buf/buf0flu.c b/storage/innodb_plugin/buf/buf0flu.c index 747ce65879d..2123f8a060d 100644 --- a/storage/innodb_plugin/buf/buf0flu.c +++ b/storage/innodb_plugin/buf/buf0flu.c @@ -1039,6 +1039,82 @@ buf_flush_write_block_low( } } +# if defined UNIV_DEBUG || defined UNIV_IBUF_DEBUG +/********************************************************************//** +Writes a flushable page asynchronously from the buffer pool to a file. +NOTE: buf_pool_mutex and block->mutex must be held upon entering this +function, and they will be released by this function after flushing. +This is loosely based on buf_flush_batch() and buf_flush_page(). +@return TRUE if the page was flushed and the mutexes released */ +UNIV_INTERN +ibool +buf_flush_page_try( +/*===============*/ + buf_block_t* block) /*!< in/out: buffer control block */ +{ + ut_ad(buf_pool_mutex_own()); + ut_ad(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE); + ut_ad(mutex_own(&block->mutex)); + + if (!buf_flush_ready_for_flush(&block->page, BUF_FLUSH_LRU)) { + return(FALSE); + } + + if (buf_pool->n_flush[BUF_FLUSH_LRU] > 0 + || buf_pool->init_flush[BUF_FLUSH_LRU]) { + /* There is already a flush batch of the same type running */ + return(FALSE); + } + + buf_pool->init_flush[BUF_FLUSH_LRU] = TRUE; + + buf_page_set_io_fix(&block->page, BUF_IO_WRITE); + + buf_page_set_flush_type(&block->page, BUF_FLUSH_LRU); + + if (buf_pool->n_flush[BUF_FLUSH_LRU]++ == 0) { + + os_event_reset(buf_pool->no_flush[BUF_FLUSH_LRU]); + } + + /* VERY IMPORTANT: + Because any thread may call the LRU flush, even when owning + locks on pages, to avoid deadlocks, we must make sure that the + s-lock is acquired on the page without waiting: this is + accomplished because buf_flush_ready_for_flush() must hold, + and that requires the page not to be bufferfixed. */ + + rw_lock_s_lock_gen(&block->lock, BUF_IO_WRITE); + + /* Note that the s-latch is acquired before releasing the + buf_pool mutex: this ensures that the latch is acquired + immediately. */ + + mutex_exit(&block->mutex); + buf_pool_mutex_exit(); + + /* Even though block is not protected by any mutex at this + point, it is safe to access block, because it is io_fixed and + oldest_modification != 0. Thus, it cannot be relocated in the + buffer pool or removed from flush_list or LRU_list. */ + + buf_flush_write_block_low(&block->page); + + buf_pool_mutex_enter(); + buf_pool->init_flush[BUF_FLUSH_LRU] = FALSE; + + if (buf_pool->n_flush[BUF_FLUSH_LRU] == 0) { + /* The running flush batch has ended */ + os_event_set(buf_pool->no_flush[BUF_FLUSH_LRU]); + } + + buf_pool_mutex_exit(); + buf_flush_buffered_writes(); + + return(TRUE); +} +# endif /* UNIV_DEBUG || UNIV_IBUF_DEBUG */ + /********************************************************************//** Writes a flushable page asynchronously from the buffer pool to a file. NOTE: in simulated aio we must call diff --git a/storage/innodb_plugin/handler/ha_innodb.cc b/storage/innodb_plugin/handler/ha_innodb.cc index 35f1a4974e2..99d4f294f81 100644 --- a/storage/innodb_plugin/handler/ha_innodb.cc +++ b/storage/innodb_plugin/handler/ha_innodb.cc @@ -4432,17 +4432,18 @@ include_field: n_requested_fields++; templ->col_no = i; + templ->clust_rec_field_no = dict_col_get_clust_pos( + &index->table->cols[i], clust_index); + ut_ad(templ->clust_rec_field_no != ULINT_UNDEFINED); if (index == clust_index) { - templ->rec_field_no = dict_col_get_clust_pos( - &index->table->cols[i], index); + templ->rec_field_no = templ->clust_rec_field_no; } else { templ->rec_field_no = dict_index_get_nth_col_pos( index, i); - } - - if (templ->rec_field_no == ULINT_UNDEFINED) { - prebuilt->need_to_access_clustered = TRUE; + if (templ->rec_field_no == ULINT_UNDEFINED) { + prebuilt->need_to_access_clustered = TRUE; + } } if (field->null_ptr) { @@ -4494,9 +4495,7 @@ skip_field: for (i = 0; i < n_requested_fields; i++) { templ = prebuilt->mysql_template + i; - templ->rec_field_no = dict_col_get_clust_pos( - &index->table->cols[templ->col_no], - clust_index); + templ->rec_field_no = templ->clust_rec_field_no; } } } @@ -10945,6 +10944,13 @@ static MYSQL_SYSVAR_STR(change_buffering, innobase_change_buffering, innodb_change_buffering_validate, innodb_change_buffering_update, "inserts"); +#if defined UNIV_DEBUG || defined UNIV_IBUF_DEBUG +static MYSQL_SYSVAR_UINT(change_buffering_debug, ibuf_debug, + PLUGIN_VAR_RQCMDARG, + "Debug flags for InnoDB change buffering (0=none)", + NULL, NULL, 0, 0, 1, 0); +#endif /* UNIV_DEBUG || UNIV_IBUF_DEBUG */ + static MYSQL_SYSVAR_ULONG(read_ahead_threshold, srv_read_ahead_threshold, PLUGIN_VAR_RQCMDARG, "Number of pages that must be accessed sequentially for InnoDB to " @@ -11005,6 +11011,9 @@ static struct st_mysql_sys_var* innobase_system_variables[]= { MYSQL_SYSVAR(version), MYSQL_SYSVAR(use_sys_malloc), MYSQL_SYSVAR(change_buffering), +#if defined UNIV_DEBUG || defined UNIV_IBUF_DEBUG + MYSQL_SYSVAR(change_buffering_debug), +#endif /* UNIV_DEBUG || UNIV_IBUF_DEBUG */ MYSQL_SYSVAR(read_ahead_threshold), MYSQL_SYSVAR(io_capacity), NULL diff --git a/storage/innodb_plugin/ibuf/ibuf0ibuf.c b/storage/innodb_plugin/ibuf/ibuf0ibuf.c index 5e9b4b27611..701e8f0ef04 100644 --- a/storage/innodb_plugin/ibuf/ibuf0ibuf.c +++ b/storage/innodb_plugin/ibuf/ibuf0ibuf.c @@ -49,6 +49,7 @@ Created 7/19/1997 Heikki Tuuri #include "btr0cur.h" #include "btr0pcur.h" #include "btr0btr.h" +#include "row0upd.h" #include "sync0sync.h" #include "dict0boot.h" #include "fut0lst.h" @@ -170,6 +171,11 @@ access order rules. */ /** Operations that can currently be buffered. */ UNIV_INTERN ibuf_use_t ibuf_use = IBUF_USE_INSERT; +#if defined UNIV_DEBUG || defined UNIV_IBUF_DEBUG +/** Flag to control insert buffer debugging. */ +UNIV_INTERN uint ibuf_debug; +#endif /* UNIV_DEBUG || UNIV_IBUF_DEBUG */ + /** The insert buffer control structure */ UNIV_INTERN ibuf_t* ibuf = NULL; @@ -2881,9 +2887,80 @@ During merge, inserts to an index page a secondary index entry extracted from the insert buffer. */ static void +ibuf_insert_to_index_page_low( +/*==========================*/ + const dtuple_t* entry, /*!< in: buffered entry to insert */ + buf_block_t* block, /*!< in/out: index page where the buffered + entry should be placed */ + dict_index_t* index, /*!< in: record descriptor */ + mtr_t* mtr, /*!< in/out: mtr */ + page_cur_t* page_cur)/*!< in/out: cursor positioned on the record + after which to insert the buffered entry */ +{ + const page_t* page; + ulint space; + ulint page_no; + ulint zip_size; + const page_t* bitmap_page; + ulint old_bits; + + if (UNIV_LIKELY + (page_cur_tuple_insert(page_cur, entry, index, 0, mtr) != NULL)) { + return; + } + + /* If the record did not fit, reorganize */ + + btr_page_reorganize(block, index, mtr); + page_cur_search(block, index, entry, PAGE_CUR_LE, page_cur); + + /* This time the record must fit */ + + if (UNIV_LIKELY + (page_cur_tuple_insert(page_cur, entry, index, 0, mtr) != NULL)) { + return; + } + + page = buf_block_get_frame(block); + + ut_print_timestamp(stderr); + + fprintf(stderr, + " InnoDB: Error: Insert buffer insert fails;" + " page free %lu, dtuple size %lu\n", + (ulong) page_get_max_insert_size(page, 1), + (ulong) rec_get_converted_size(index, entry, 0)); + fputs("InnoDB: Cannot insert index record ", stderr); + dtuple_print(stderr, entry); + fputs("\nInnoDB: The table where this index record belongs\n" + "InnoDB: is now probably corrupt. Please run CHECK TABLE on\n" + "InnoDB: that table.\n", stderr); + + space = page_get_space_id(page); + zip_size = buf_block_get_zip_size(block); + page_no = page_get_page_no(page); + + bitmap_page = ibuf_bitmap_get_map_page(space, page_no, zip_size, mtr); + old_bits = ibuf_bitmap_page_get_bits(bitmap_page, page_no, zip_size, + IBUF_BITMAP_FREE, mtr); + + fprintf(stderr, + "InnoDB: space %lu, page %lu, zip_size %lu, bitmap bits %lu\n", + (ulong) space, (ulong) page_no, + (ulong) zip_size, (ulong) old_bits); + + fputs("InnoDB: Submit a detailed bug report" + " to http://bugs.mysql.com\n", stderr); +} + +/************************************************************************ +During merge, inserts to an index page a secondary index entry extracted +from the insert buffer. */ +static +void ibuf_insert_to_index_page( /*======================*/ - dtuple_t* entry, /*!< in: buffered entry to insert */ + const dtuple_t* entry, /*!< in: buffered entry to insert */ buf_block_t* block, /*!< in/out: index page where the buffered entry should be placed */ dict_index_t* index, /*!< in: record descriptor */ @@ -2893,11 +2970,10 @@ ibuf_insert_to_index_page( ulint low_match; page_t* page = buf_block_get_frame(block); rec_t* rec; - page_t* bitmap_page; - ulint old_bits; ut_ad(ibuf_inside()); ut_ad(dtuple_check_typed(entry)); + ut_ad(!buf_block_align(page)->is_hashed); if (UNIV_UNLIKELY(dict_table_is_comp(index->table) != (ibool)!!page_is_comp(page))) { @@ -2935,71 +3011,86 @@ dump: low_match = page_cur_search(block, index, entry, PAGE_CUR_LE, &page_cur); - if (low_match == dtuple_get_n_fields(entry)) { + if (UNIV_UNLIKELY(low_match == dtuple_get_n_fields(entry))) { + mem_heap_t* heap; + upd_t* update; + ulint* offsets; page_zip_des_t* page_zip; rec = page_cur_get_rec(&page_cur); + + /* This is based on + row_ins_sec_index_entry_by_modify(BTR_MODIFY_LEAF). */ + ut_ad(rec_get_deleted_flag(rec, page_is_comp(page))); + + heap = mem_heap_create(1024); + + offsets = rec_get_offsets(rec, index, NULL, ULINT_UNDEFINED, + &heap); + update = row_upd_build_sec_rec_difference_binary( + index, entry, rec, NULL, heap); + page_zip = buf_block_get_page_zip(block); - btr_cur_del_unmark_for_ibuf(rec, page_zip, mtr); - } else { - rec = page_cur_tuple_insert(&page_cur, entry, index, 0, mtr); - - if (UNIV_LIKELY(rec != NULL)) { + if (update->n_fields == 0) { + /* The records only differ in the delete-mark. + Clear the delete-mark, like we did before + Bug #56680 was fixed. */ + btr_cur_del_unmark_for_ibuf(rec, page_zip, mtr); +updated_in_place: + mem_heap_free(heap); return; } - /* If the record did not fit, reorganize */ + /* Copy the info bits. Clear the delete-mark. */ + update->info_bits = rec_get_info_bits(rec, page_is_comp(page)); + update->info_bits &= ~REC_INFO_DELETED_FLAG; - btr_page_reorganize(block, index, mtr); - page_cur_search(block, index, entry, PAGE_CUR_LE, &page_cur); - - /* This time the record must fit */ - if (UNIV_UNLIKELY - (!page_cur_tuple_insert(&page_cur, entry, index, - 0, mtr))) { - ulint space; - ulint page_no; - ulint zip_size; - - ut_print_timestamp(stderr); - - fprintf(stderr, - " InnoDB: Error: Insert buffer insert" - " fails; page free %lu," - " dtuple size %lu\n", - (ulong) page_get_max_insert_size( - page, 1), - (ulong) rec_get_converted_size( - index, entry, 0)); - fputs("InnoDB: Cannot insert index record ", - stderr); - dtuple_print(stderr, entry); - fputs("\nInnoDB: The table where" - " this index record belongs\n" - "InnoDB: is now probably corrupt." - " Please run CHECK TABLE on\n" - "InnoDB: that table.\n", stderr); - - space = page_get_space_id(page); - zip_size = buf_block_get_zip_size(block); - page_no = page_get_page_no(page); - - bitmap_page = ibuf_bitmap_get_map_page( - space, page_no, zip_size, mtr); - old_bits = ibuf_bitmap_page_get_bits( - bitmap_page, page_no, zip_size, - IBUF_BITMAP_FREE, mtr); - - fprintf(stderr, - "InnoDB: space %lu, page %lu," - " zip_size %lu, bitmap bits %lu\n", - (ulong) space, (ulong) page_no, - (ulong) zip_size, (ulong) old_bits); - - fputs("InnoDB: Submit a detailed bug report" - " to http://bugs.mysql.com\n", stderr); + /* We cannot invoke btr_cur_optimistic_update() here, + because we do not have a btr_cur_t or que_thr_t, + as the insert buffer merge occurs at a very low level. */ + if (!row_upd_changes_field_size_or_external(index, offsets, + update) + && (!page_zip || btr_cur_update_alloc_zip( + page_zip, block, index, + rec_offs_size(offsets), FALSE, mtr))) { + /* This is the easy case. Do something similar + to btr_cur_update_in_place(). */ + row_upd_rec_in_place(rec, index, offsets, + update, page_zip); + goto updated_in_place; } + + /* A collation may identify values that differ in + storage length. + Some examples (1 or 2 bytes): + utf8_turkish_ci: I = U+0131 LATIN SMALL LETTER DOTLESS I + utf8_general_ci: S = U+00DF LATIN SMALL LETTER SHARP S + utf8_general_ci: A = U+00E4 LATIN SMALL LETTER A WITH DIAERESIS + + latin1_german2_ci: SS = U+00DF LATIN SMALL LETTER SHARP S + + Examples of a character (3-byte UTF-8 sequence) + identified with 2 or 4 characters (1-byte UTF-8 sequences): + + utf8_unicode_ci: 'II' = U+2171 SMALL ROMAN NUMERAL TWO + utf8_unicode_ci: '(10)' = U+247D PARENTHESIZED NUMBER TEN + */ + + /* Delete the different-length record, and insert the + buffered one. */ + + lock_rec_store_on_page_infimum(block, rec); + page_cur_delete_rec(&page_cur, index, offsets, mtr); + page_cur_move_to_prev(&page_cur); + mem_heap_free(heap); + + ibuf_insert_to_index_page_low(entry, block, index, mtr, + &page_cur); + lock_rec_restore_from_page_infimum(block, rec, block); + } else { + ibuf_insert_to_index_page_low(entry, block, index, mtr, + &page_cur); } } diff --git a/storage/innodb_plugin/include/btr0cur.h b/storage/innodb_plugin/include/btr0cur.h index e151fdcb563..7f6bff11f84 100644 --- a/storage/innodb_plugin/include/btr0cur.h +++ b/storage/innodb_plugin/include/btr0cur.h @@ -242,6 +242,22 @@ btr_cur_pessimistic_insert( que_thr_t* thr, /*!< in: query thread or NULL */ mtr_t* mtr); /*!< in: mtr */ /*************************************************************//** +See if there is enough place in the page modification log to log +an update-in-place. +@return TRUE if enough place */ +UNIV_INTERN +ibool +btr_cur_update_alloc_zip( +/*=====================*/ + page_zip_des_t* page_zip,/*!< in/out: compressed page */ + buf_block_t* block, /*!< in/out: buffer page */ + dict_index_t* index, /*!< in: the index corresponding to the block */ + ulint length, /*!< in: size needed */ + ibool create, /*!< in: TRUE=delete-and-insert, + FALSE=update-in-place */ + mtr_t* mtr) /*!< in: mini-transaction */ + __attribute__((nonnull, warn_unused_result)); +/*************************************************************//** Updates a record when the update causes no size changes in its fields. @return DB_SUCCESS or error number */ UNIV_INTERN diff --git a/storage/innodb_plugin/include/buf0flu.h b/storage/innodb_plugin/include/buf0flu.h index c996f6eaab4..cb8c49fde15 100644 --- a/storage/innodb_plugin/include/buf0flu.h +++ b/storage/innodb_plugin/include/buf0flu.h @@ -75,6 +75,20 @@ buf_flush_init_for_writing( ib_uint64_t newest_lsn); /*!< in: newest modification lsn to the page */ #ifndef UNIV_HOTBACKUP +# if defined UNIV_DEBUG || defined UNIV_IBUF_DEBUG +/********************************************************************//** +Writes a flushable page asynchronously from the buffer pool to a file. +NOTE: buf_pool_mutex and block->mutex must be held upon entering this +function, and they will be released by this function after flushing. +This is loosely based on buf_flush_batch() and buf_flush_page(). +@return TRUE if the page was flushed and the mutexes released */ +UNIV_INTERN +ibool +buf_flush_page_try( +/*===============*/ + buf_block_t* block) /*!< in/out: buffer control block */ + __attribute__((nonnull, warn_unused_result)); +# endif /* UNIV_DEBUG || UNIV_IBUF_DEBUG */ /*******************************************************************//** This utility flushes dirty blocks from the end of the LRU list or flush_list. NOTE 1: in the case of an LRU flush the calling thread may own latches to diff --git a/storage/innodb_plugin/include/ibuf0ibuf.h b/storage/innodb_plugin/include/ibuf0ibuf.h index 8aa21fb9d95..f8cc4471d43 100644 --- a/storage/innodb_plugin/include/ibuf0ibuf.h +++ b/storage/innodb_plugin/include/ibuf0ibuf.h @@ -48,6 +48,11 @@ typedef enum { /** Operations that can currently be buffered. */ extern ibuf_use_t ibuf_use; +#if defined UNIV_DEBUG || defined UNIV_IBUF_DEBUG +/** Flag to control insert buffer debugging. */ +extern uint ibuf_debug; +#endif /* UNIV_DEBUG || UNIV_IBUF_DEBUG */ + /** The insert buffer control structure */ extern ibuf_t* ibuf; diff --git a/storage/innodb_plugin/include/row0mysql.h b/storage/innodb_plugin/include/row0mysql.h index b69e657361b..42182cc0044 100644 --- a/storage/innodb_plugin/include/row0mysql.h +++ b/storage/innodb_plugin/include/row0mysql.h @@ -527,6 +527,10 @@ struct mysql_row_templ_struct { Innobase record in the current index; not defined if template_type is ROW_MYSQL_WHOLE_ROW */ + ulint clust_rec_field_no; /*!< field number of the column in an + Innobase record in the clustered index; + not defined if template_type is + ROW_MYSQL_WHOLE_ROW */ ulint mysql_col_offset; /*!< offset of the column in the MySQL row format */ ulint mysql_col_len; /*!< length of the column in the MySQL diff --git a/storage/innodb_plugin/include/row0upd.h b/storage/innodb_plugin/include/row0upd.h index 635d746d5a1..4e2de9bd2ec 100644 --- a/storage/innodb_plugin/include/row0upd.h +++ b/storage/innodb_plugin/include/row0upd.h @@ -167,8 +167,11 @@ row_upd_changes_field_size_or_external( const upd_t* update);/*!< in: update vector */ #endif /* !UNIV_HOTBACKUP */ /***********************************************************//** -Replaces the new column values stored in the update vector to the record -given. No field size changes are allowed. */ +Replaces the new column values stored in the update vector to the +record given. No field size changes are allowed. This function is +usually invoked on a clustered index. The only use case for a +secondary index is row_ins_sec_index_entry_by_modify() or its +counterpart in ibuf_insert_to_index_page(). */ UNIV_INTERN void row_upd_rec_in_place( diff --git a/storage/innodb_plugin/row/row0mysql.c b/storage/innodb_plugin/row/row0mysql.c index 78b3b2afdbf..107af481b79 100644 --- a/storage/innodb_plugin/row/row0mysql.c +++ b/storage/innodb_plugin/row/row0mysql.c @@ -444,7 +444,7 @@ row_mysql_convert_row_to_innobase( row is used, as row may contain pointers to this record! */ { - mysql_row_templ_t* templ; + const mysql_row_templ_t*templ; dfield_t* dfield; ulint i; diff --git a/storage/innodb_plugin/row/row0sel.c b/storage/innodb_plugin/row/row0sel.c index ea1a19eb82f..ac78a95839c 100644 --- a/storage/innodb_plugin/row/row0sel.c +++ b/storage/innodb_plugin/row/row0sel.c @@ -2675,21 +2675,22 @@ row_sel_store_mysql_rec( row_prebuilt_t* prebuilt, /*!< in: prebuilt struct */ const rec_t* rec, /*!< in: Innobase record in the index which was described in prebuilt's - template; must be protected by - a page latch */ + template, or in the clustered index; + must be protected by a page latch */ + ibool rec_clust, /*!< in: TRUE if rec is in the + clustered index instead of + prebuilt->index */ const ulint* offsets) /*!< in: array returned by - rec_get_offsets() */ + rec_get_offsets(rec) */ { - mysql_row_templ_t* templ; - mem_heap_t* extern_field_heap = NULL; - mem_heap_t* heap; - const byte* data; - ulint len; - ulint i; + mem_heap_t* extern_field_heap = NULL; + mem_heap_t* heap; + ulint i; ut_ad(prebuilt->mysql_template); ut_ad(prebuilt->default_rec); ut_ad(rec_offs_validate(rec, NULL, offsets)); + ut_ad(!rec_get_deleted_flag(rec, rec_offs_comp(offsets))); if (UNIV_LIKELY_NULL(prebuilt->blob_heap)) { mem_heap_free(prebuilt->blob_heap); @@ -2698,10 +2699,15 @@ row_sel_store_mysql_rec( for (i = 0; i < prebuilt->n_template; i++) { - templ = prebuilt->mysql_template + i; + const mysql_row_templ_t*templ = prebuilt->mysql_template + i; + const byte* data; + ulint len; + ulint field_no; - if (UNIV_UNLIKELY(rec_offs_nth_extern(offsets, - templ->rec_field_no))) { + field_no = rec_clust + ? templ->clust_rec_field_no : templ->rec_field_no; + + if (UNIV_UNLIKELY(rec_offs_nth_extern(offsets, field_no))) { /* Copy an externally stored field to the temporary heap */ @@ -2729,7 +2735,7 @@ row_sel_store_mysql_rec( data = btr_rec_copy_externally_stored_field( rec, offsets, dict_table_zip_size(prebuilt->table), - templ->rec_field_no, &len, heap); + field_no, &len, heap); if (UNIV_UNLIKELY(!data)) { /* The externally stored field @@ -2750,8 +2756,7 @@ row_sel_store_mysql_rec( } else { /* Field is stored in the row. */ - data = rec_get_nth_field(rec, offsets, - templ->rec_field_no, &len); + data = rec_get_nth_field(rec, offsets, field_no, &len); if (UNIV_UNLIKELY(templ->type == DATA_BLOB) && len != UNIV_SQL_NULL) { @@ -3113,7 +3118,7 @@ row_sel_pop_cached_row_for_mysql( row_prebuilt_t* prebuilt) /*!< in: prebuilt struct */ { ulint i; - mysql_row_templ_t* templ; + const mysql_row_templ_t*templ; byte* cached_rec; ut_ad(prebuilt->n_fetch_cached > 0); ut_ad(prebuilt->mysql_prefix_len <= prebuilt->mysql_row_len); @@ -3170,15 +3175,21 @@ ibool row_sel_push_cache_row_for_mysql( /*=============================*/ row_prebuilt_t* prebuilt, /*!< in: prebuilt struct */ - const rec_t* rec, /*!< in: record to push; must - be protected by a page latch */ - const ulint* offsets) /*!< in: rec_get_offsets() */ + const rec_t* rec, /*!< in: record to push, in the index + which was described in prebuilt's + template, or in the clustered index; + must be protected by a page latch */ + ibool rec_clust, /*!< in: TRUE if rec is in the + clustered index instead of + prebuilt->index */ + const ulint* offsets) /*!< in: rec_get_offsets(rec) */ { byte* buf; ulint i; ut_ad(prebuilt->n_fetch_cached < MYSQL_FETCH_CACHE_SIZE); ut_ad(rec_offs_validate(rec, NULL, offsets)); + ut_ad(!rec_get_deleted_flag(rec, rec_offs_comp(offsets))); ut_a(!prebuilt->templ_contains_blob); if (prebuilt->fetch_cache[0] == NULL) { @@ -3207,7 +3218,7 @@ row_sel_push_cache_row_for_mysql( if (UNIV_UNLIKELY(!row_sel_store_mysql_rec( prebuilt->fetch_cache[ prebuilt->n_fetch_cached], - prebuilt, rec, offsets))) { + prebuilt, rec, rec_clust, offsets))) { return(FALSE); } @@ -3608,7 +3619,8 @@ row_search_for_mysql( ut_ad(!rec_get_deleted_flag(rec, comp)); if (!row_sel_store_mysql_rec(buf, prebuilt, - rec, offsets)) { + rec, FALSE, + offsets)) { /* Only fresh inserts may contain incomplete externally stored columns. Pretend that such @@ -4242,7 +4254,6 @@ no_gap_lock: is necessary, because we can only get the undo information via the clustered index record. */ - ut_ad(index != clust_index); ut_ad(!dict_index_is_clust(index)); if (!lock_sec_rec_cons_read_sees( @@ -4358,26 +4369,10 @@ requires_clust_rec: goto next_rec; } - if (prebuilt->need_to_access_clustered) { - - result_rec = clust_rec; - - ut_ad(rec_offs_validate(result_rec, clust_index, - offsets)); - } else { - /* We used 'offsets' for the clust rec, recalculate - them for 'rec' */ - offsets = rec_get_offsets(rec, index, offsets, - ULINT_UNDEFINED, &heap); - result_rec = rec; - } - - /* result_rec can legitimately be delete-marked - now that it has been established that it points to a - clustered index record that exists in the read view. */ + result_rec = clust_rec; + ut_ad(rec_offs_validate(result_rec, clust_index, offsets)); } else { result_rec = rec; - ut_ad(!rec_get_deleted_flag(rec, comp)); } /* We found a qualifying record 'result_rec'. At this point, @@ -4386,6 +4381,7 @@ requires_clust_rec: ut_ad(rec_offs_validate(result_rec, result_rec != rec ? clust_index : index, offsets)); + ut_ad(!rec_get_deleted_flag(result_rec, comp)); /* At this point, the clustered index record is protected by a page latch that was acquired when pcur was positioned. @@ -4410,6 +4406,7 @@ requires_clust_rec: cursor. */ if (!row_sel_push_cache_row_for_mysql(prebuilt, result_rec, + result_rec != rec, offsets)) { /* Only fresh inserts may contain incomplete externally stored columns. Pretend that such @@ -4427,15 +4424,31 @@ requires_clust_rec: goto next_rec; } else { - if (prebuilt->template_type == ROW_MYSQL_DUMMY_TEMPLATE) { + if (UNIV_UNLIKELY + (prebuilt->template_type == ROW_MYSQL_DUMMY_TEMPLATE)) { + /* CHECK TABLE: fetch the row */ + + if (result_rec != rec + && !prebuilt->need_to_access_clustered) { + /* We used 'offsets' for the clust + rec, recalculate them for 'rec' */ + offsets = rec_get_offsets(rec, index, offsets, + ULINT_UNDEFINED, + &heap); + result_rec = rec; + } + memcpy(buf + 4, result_rec - rec_offs_extra_size(offsets), rec_offs_size(offsets)); mach_write_to_4(buf, rec_offs_extra_size(offsets) + 4); } else { - if (!row_sel_store_mysql_rec(buf, prebuilt, - result_rec, offsets)) { + /* Returning a row to MySQL */ + + if (!row_sel_store_mysql_rec(buf, prebuilt, result_rec, + result_rec != rec, + offsets)) { /* Only fresh inserts may contain incomplete externally stored columns. Pretend that such records do diff --git a/storage/innodb_plugin/row/row0upd.c b/storage/innodb_plugin/row/row0upd.c index 04c3139fcc7..444003ba3f0 100644 --- a/storage/innodb_plugin/row/row0upd.c +++ b/storage/innodb_plugin/row/row0upd.c @@ -466,8 +466,11 @@ row_upd_changes_field_size_or_external( #endif /* !UNIV_HOTBACKUP */ /***********************************************************//** -Replaces the new column values stored in the update vector to the record -given. No field size changes are allowed. */ +Replaces the new column values stored in the update vector to the +record given. No field size changes are allowed. This function is +usually invoked on a clustered index. The only use case for a +secondary index is row_ins_sec_index_entry_by_modify() or its +counterpart in ibuf_insert_to_index_page(). */ UNIV_INTERN void row_upd_rec_in_place(