diff options
30 files changed, 839 insertions, 234 deletions
diff --git a/Build-tools/Do-compile b/Build-tools/Do-compile index b3b401d7e8d..e74d7c818e8 100755 --- a/Build-tools/Do-compile +++ b/Build-tools/Do-compile @@ -3,30 +3,30 @@ use Getopt::Long; $opt_distribution=$opt_user=$opt_result=$opt_config_options=$opt_config_env=""; $opt_dbd_options=$opt_perl_options=$opt_suffix=""; -$opt_tmp=$version_suffix=""; +$opt_tmp=$opt_version_suffix=""; $opt_help=$opt_Information=$opt_delete=$opt_debug=$opt_stage=$opt_rsh_mail=$opt_no_test=$opt_no_perl=$opt_with_low_memory=$opt_fast_benchmark=$opt_static_client=$opt_static_server=$opt_static_perl=$opt_sur=$opt_with_small_disk=$opt_local_perl=$opt_tcpip=$opt_build_thread=$opt_no_mysqltest=$opt_use_old_distribution=$opt_enable_shared=$opt_no_crash_me=$opt_no_strip=0; $opt_innodb=$opt_bdb=0; -GetOptions("Information","help","distribution=s","user=s","result=s","delete","no-test","no-mysqltest","perl-files=s","debug","config-options=s","config-env=s","stage=i","rsh-mail","with-low-memory","fast-benchmark","tmp=s","static-client","static-server","static-perl","no-perl","local-perl","perl-options=s","sur","with-small-disk","dbd-options=s","tcpip","suffix=s","build-thread=i","innodb","bdb","use-old-distribution","enable-shared","no-crash-me","no-strip") || usage(); +GetOptions("Information","help","distribution=s","user=s","result=s","delete","no-test","no-mysqltest","perl-files=s","debug","config-options=s","config-env=s","stage=i","rsh-mail","with-low-memory","fast-benchmark","tmp=s","static-client","static-server","static-perl","no-perl","local-perl","perl-options=s","sur","with-small-disk","dbd-options=s","tcpip","suffix=s","build-thread=i","innodb","bdb","use-old-distribution","enable-shared","no-crash-me","no-strip","version-suffix=s") || usage(); usage() if ($opt_help || $opt_Information); usage() if (!$opt_distribution); -if ($opt_innodb || $opt_bdb) +if (($opt_innodb || $opt_bdb) && $opt_version_suffix eq "") { - $version_suffix="-max"; + $opt_version_suffix="-max"; } chomp($host=`hostname`); $full_host_name=$host; -info("Compiling MySQL$version_suffix at $host$suffix, stage: $opt_stage\n"); +info("Compiling MySQL$opt_version_suffix at $host$suffix, stage: $opt_stage\n"); $connect_option= ($opt_tcpip ? "--host=$host" : ""); $host =~ /^([^.-]*)/; $host=$1 . $opt_suffix; $email="$opt_user\@mysql.com"; $pwd = `pwd`; chomp($pwd); -$log="$pwd/Logs/$host$version_suffix.log"; -$opt_distribution =~ /(mysql-[^\/]*)\.tar/; +$log="$pwd/Logs/$host$opt_version_suffix.log"; +$opt_distribution =~ /(mysql[^\/]*)\.tar/; $ver=$1; $gcc_version=which("gcc"); if (defined($gcc_version) && ! $opt_config_env) @@ -108,7 +108,7 @@ $|=1; select STDOUT; $|=1; -safe_cd("$host"); +safe_cd($host); if ($opt_stage == 0 && ! $opt_use_old_distribution) { safe_system("gunzip < $opt_distribution | $tar xf -"); @@ -118,6 +118,7 @@ if ($opt_stage == 0 && ! $opt_use_old_distribution) system("touch timestamp; find . -newer timestamp -print | xargs touch; rm -f timestamp"); sleep(2); # Ensure that files we don't want to rebuild are newer than other files + safe_cd($ver); foreach $name ("configure", "Docs/include.texi", "Docs/*.html", "Docs/manual.txt", "Docs/mysql.info", @@ -125,9 +126,11 @@ if ($opt_stage == 0 && ! $opt_use_old_distribution) { system("touch $name"); } + # Fix some file modes in BDB tables that makes life harder. + system("chmod -R u+rw ."); } -safe_cd($ver); +safe_cd("$pwd/$host/$ver"); if ($opt_stage <= 1) { $opt_config_options.=" --with-low-memory" if ($opt_with_low_memory); @@ -156,7 +159,7 @@ if ($opt_stage <= 1) { $opt_config_options.= " --with-innodb" } - check_system("$opt_config_env ./configure --prefix=/usr/local/mysql \"--with-comment=Official MySQL$version_suffix binary\" --with-extra-charsets=complex \"--with-server-suffix=$version_suffix\" --enable-thread-safe-client $opt_config_options","Thank you for choosing MySQL"); + check_system("$opt_config_env ./configure --prefix=/usr/local/mysql \"--with-comment=Official MySQL$opt_version_suffix binary\" --with-extra-charsets=complex \"--with-server-suffix=$opt_version_suffix\" --enable-thread-safe-client $opt_config_options","Thank you for choosing MySQL"); if (-d "$pwd/$host/include-mysql") { safe_system("cp -r $pwd/$host/include-mysql/* $pwd/$host/$ver/include"); @@ -209,7 +212,7 @@ if ($opt_stage <= 4 && !$opt_no_test) safe_system("gunzip < $tar_file | $tar xf -"); } -$tar_file =~ /(mysql-[^\/]*)\.tar/; +$tar_file =~ /(mysql[^\/]*)\.tar/; $ver=$1; $test_dir="$pwd/$host/test/$ver"; $ENV{"LD_LIBRARY_PATH"}= "$test_dir/lib:" . $ENV{"LD_LIBRARY_PATH"}; @@ -361,6 +364,9 @@ To set up the environment, like 'CC=cc CXX=gcc CXXFLAGS=-O3' --dbd-options 'options' Options for Makefile.PL when configuring msql-mysql-modules. +--version-suffix suffix +Can be used to set a suffix (normally 'com' or '-max') for a distribution + --with-low-memory Use less memory when compiling. diff --git a/Build-tools/mysql-copyright b/Build-tools/mysql-copyright index df819b20fa1..0d49acac400 100755 --- a/Build-tools/mysql-copyright +++ b/Build-tools/mysql-copyright @@ -96,7 +96,7 @@ sub main # remove the 'PUBLIC' file from distribution and copy LICENSE # on the toplevel of the directory instead. file 'PUBLIC' shouldn't # exist in the new mysql distributions, but let's be sure.. - `rm -f $destdir/PUBLIC`; + `rm -f $destdir/PUBLIC $destdir/README`; `cp -p $WD/Docs/LICENSE $destdir/`; # fix file copyrights @@ -129,7 +129,7 @@ sub fix_usage_copyright foreach my $Cfile (@Cfiles) { chop $Cfile; - `replace \"This is free software,\\\\\\nand you are welcome to modify and redistribute it under the GPL license\" \"This is commercial software,\\\\nplease see the file LICENSE for details\" -- $Cfile`; + `replace "This is free software," "This is commercial software," "and you are welcome to modify and redistribute it under the GPL license" "please see the file LICENSE for details" -- $Cfile`; } } diff --git a/Docs/manual.texi b/Docs/manual.texi index 031e9d57205..d075e17ee56 100644 --- a/Docs/manual.texi +++ b/Docs/manual.texi @@ -48720,6 +48720,7 @@ users use this code as the rest of the code and because of this we are not yet 100% confident in this code. @menu +* News-3.23.49:: Changes in release 3.23.49 * News-3.23.48:: Changes in release 3.23.48 * News-3.23.47:: Changes in release 3.23.47 * News-3.23.46:: Changes in release 3.23.46 @@ -48772,14 +48773,64 @@ not yet 100% confident in this code. * News-3.23.0:: Changes in release 3.23.0 @end menu -@node News-3.23.48, News-3.23.47, News-3.23.x, News-3.23.x +@node News-3.23.49, News-3.23.48, News-3.23.x, News-3.23.x +@appendixsubsec Changes in release 3.23.49 +@itemize @bullet +@item +Fixed unlikely caching bug when doing a join without keys. In this case +the last used field for a table always returned @code{NULL}. +@end itemize + +@node News-3.23.48, News-3.23.47, News-3.23.49, News-3.23.x @appendixsubsec Changes in release 3.23.48 @itemize @bullet @item +Changed to use @code{autoconf} 2.52 (from @code{autoconf} 2.13) +@item Fixed bug in complicated join with @code{const} tables. @item Added internal safety checks for InnoDB. @item +Some InnoDB variables was always shown in @code{SHOW VARIABLES} as +@code{OFF} on high-byte-first systems (like sparc). +@item +Fixed problem with one thread using an InnoDB table and another +thread doing an @code{ALTER TABLE} on the same table. Before that, +mysqld could crash with an assertion failure in row0row.c, line 474. +@item +Tuned the InnoDB SQL optimizer to favor more often index searches +over table scans. +@item +Fixed a performance problem with InnoDB tables when several large SELECT +queries are run concurrently on a multiprocessor Linux computer. Large +CPU-bound SELECT queries will now also generally run faster on all +platforms. +@item +If MySQL binlogging is used, InnoDB now prints after crash recovery the +latest MySQL binlog name and the offset InnoDB was able to recover +to. This is useful, for example, when resynchronizing a master and a +slave database in replication. +@item +Added better error messages to help in installation problems of InnoDB tables. +@item +One can now recover also MySQL temporary tables which have become +orphaned inside the InnoDB tablespace. +@item +InnoDB now prevents a @code{FOREIGN KEY} declaration where the signedness +is not the same in the referencing and referenced integer columns. +@item +Calling @code{SHOW CREATE TABLE} or @code{SHOW TABLE STATUS} could cause +memory corruption and make mysqld to crash. Especially at risk was +@code{mysqldump}, because it calls frequently @code{SHOW CREATE TABLE}. +@item +If inserts to several tables containing an auto-inc column were wrapped +inside one @code{LOCK TABLES}, InnoDB asserted in lock0lock.c. +@item +In 3.23.47 we allowed several @code{NULLS} in a @code{UNIQUE} secondary +index for an InnoDB table. But @code{CHECK TABLE} was not relaxed: it +reports the table as corrupt. @code{CHECK TABLE} no longer complains in +this situation. +@item @code{SHOW GRANTS} now shows @code{REFERENCES} instead of @code{REFERENCE}. @end itemize diff --git a/innobase/btr/btr0cur.c b/innobase/btr/btr0cur.c index 9f3d02525d8..d6944ecadc8 100644 --- a/innobase/btr/btr0cur.c +++ b/innobase/btr/btr0cur.c @@ -2552,6 +2552,7 @@ btr_estimate_number_of_different_key_vals( ulint total_external_size = 0; ulint i; ulint j; + ulint add_on; mtr_t mtr; n_cols = dict_index_get_n_unique(index); @@ -2624,8 +2625,25 @@ btr_estimate_number_of_different_key_vals( + not_empty_flag) / (BTR_KEY_VAL_ESTIMATE_N_PAGES + total_external_size); - } + /* If the tree is small, smaller than < + 10 * BTR_KEY_VAL_ESTIMATE_N_PAGES + total_external_size, then + the above estimate is ok. For bigger trees it is common that we + do not see any borders between key values in the few pages + we pick. But still there may be BTR_KEY_VAL_ESTIMATE_N_PAGES + different key values, or even more. Let us try to approximate + that: */ + + add_on = index->stat_n_leaf_pages / + (10 * (BTR_KEY_VAL_ESTIMATE_N_PAGES + total_external_size)); + + if (add_on > BTR_KEY_VAL_ESTIMATE_N_PAGES) { + add_on = BTR_KEY_VAL_ESTIMATE_N_PAGES; + } + + index->stat_n_diff_key_vals[j] += add_on; + } + mem_free(n_diff); } diff --git a/innobase/buf/buf0lru.c b/innobase/buf/buf0lru.c index a3bb673dfc8..5ab045212c2 100644 --- a/innobase/buf/buf0lru.c +++ b/innobase/buf/buf0lru.c @@ -243,7 +243,6 @@ loop: if (n_iterations > 30) { ut_print_timestamp(stderr); fprintf(stderr, - " ***********************************************\n" "InnoDB: Warning: difficult to find free blocks from\n" "InnoDB: the buffer pool (%lu search iterations)! Consider\n" "InnoDB: increasing the buffer pool size.\n", diff --git a/innobase/dict/dict0crea.c b/innobase/dict/dict0crea.c index 7b9255b4e27..69b1e7c61fd 100644 --- a/innobase/dict/dict0crea.c +++ b/innobase/dict/dict0crea.c @@ -1235,16 +1235,23 @@ loop: if (error != DB_SUCCESS) { fprintf(stderr, - "InnoDB: foreign constraint creation failed;\n" + "InnoDB: Foreign key constraint creation failed:\n" "InnoDB: internal error number %lu\n", error); - ut_a(error == DB_OUT_OF_FILE_SPACE); - - fprintf(stderr, "InnoDB: tablespace is full\n"); - - trx_general_rollback_for_mysql(trx, FALSE, NULL); + if (error == DB_DUPLICATE_KEY) { + fprintf(stderr, + "InnoDB: Duplicate key error in system table %s index %s\n", + ((dict_index_t*)trx->error_info)->table_name, + ((dict_index_t*)trx->error_info)->name); - error = DB_MUST_GET_MORE_FILE_SPACE; + fprintf(stderr, "%s\n", buf); + + fprintf(stderr, + "InnoDB: Maybe the internal data dictionary of InnoDB is\n" + "InnoDB: out-of-sync from the .frm files of your tables.\n" + "InnoDB: See section 15.1 Troubleshooting data dictionary operations\n" + "InnoDB: at http://www.innodb.com/ibman.html\n"); + } return(error); } diff --git a/innobase/fil/fil0fil.c b/innobase/fil/fil0fil.c index f1f2bc4fa17..727e52707e8 100644 --- a/innobase/fil/fil0fil.c +++ b/innobase/fil/fil0fil.c @@ -1088,7 +1088,15 @@ loop: node = UT_LIST_GET_FIRST(space->chain); for (;;) { - ut_a(node); + if (node == NULL) { + fprintf(stderr, + "InnoDB: Error: trying to access page number %lu in space %lu\n" + "InnoDB: which is outside the tablespace bounds.\n" + "InnoDB: Byte offset %lu, len %lu, i/o type %lu\n", + block_offset, space_id, byte_offset, len, type); + + ut_a(0); + } if (node->size > block_offset) { /* Found! */ diff --git a/innobase/include/log0recv.h b/innobase/include/log0recv.h index 51f14393d38..8f896756db9 100644 --- a/innobase/include/log0recv.h +++ b/innobase/include/log0recv.h @@ -258,6 +258,7 @@ struct recv_sys_struct{ extern recv_sys_t* recv_sys; extern ibool recv_recovery_on; extern ibool recv_no_ibuf_operations; +extern ibool recv_needed_recovery; /* States of recv_addr_struct */ #define RECV_NOT_PROCESSED 71 diff --git a/innobase/include/mem0mem.h b/innobase/include/mem0mem.h index 57fac93d3ac..89c5428f054 100644 --- a/innobase/include/mem0mem.h +++ b/innobase/include/mem0mem.h @@ -269,13 +269,24 @@ mem_realloc( ulint n, /* in: desired number of bytes */ char* file_name,/* in: file name where called */ ulint line); /* in: line where called */ +#ifdef MEM_PERIODIC_CHECK +/********************************************************************** +Goes through the list of all allocated mem blocks, checks their magic +numbers, and reports possible corruption. */ +void +mem_validate_all_blocks(void); +/*=========================*/ +#endif /*#######################################################################*/ /* The info header of a block in a memory heap */ struct mem_block_info_struct { + ulint magic_n;/* magic number for debugging */ + char file_name[8];/* file name where the mem heap was created */ + ulint line; /* line number where the mem heap was created */ UT_LIST_BASE_NODE_T(mem_block_t) base; /* In the first block in the the list this is the base node of the list of blocks; in subsequent blocks this is undefined */ @@ -299,9 +310,11 @@ struct mem_block_info_struct { allocated buffer frame, which can be appended as a free block to the heap, if we need more space; otherwise, this is NULL */ - ulint magic_n;/* magic number for debugging */ - char file_name[8];/* file name where the mem heap was created */ - ulint line; /* line number where the mem heap was created */ +#ifdef MEM_PERIODIC_CHECK + UT_LIST_NODE_T(mem_block_t) mem_block_list; + /* List of all mem blocks allocated; protected + by the mem_comm_pool mutex */ +#endif }; #define MEM_BLOCK_MAGIC_N 764741555 diff --git a/innobase/include/mem0pool.h b/innobase/include/mem0pool.h index eb675b4a7f9..43707bd5f61 100644 --- a/innobase/include/mem0pool.h +++ b/innobase/include/mem0pool.h @@ -72,6 +72,18 @@ mem_pool_get_reserved( /* out: reserved mmeory in bytes */ mem_pool_t* pool); /* in: memory pool */ /************************************************************************ +Reserves the mem pool mutex. */ + +void +mem_pool_mutex_enter(void); +/*======================*/ +/************************************************************************ +Releases the mem pool mutex. */ + +void +mem_pool_mutex_exit(void); +/*=====================*/ +/************************************************************************ Validates a memory pool. */ ibool diff --git a/innobase/include/row0mysql.h b/innobase/include/row0mysql.h index 0346299bb10..48b6ba8a715 100644 --- a/innobase/include/row0mysql.h +++ b/innobase/include/row0mysql.h @@ -251,6 +251,24 @@ row_table_add_foreign_constraints( char* name); /* in: table full name in the normalized form database_name/table_name */ /************************************************************************* +The master thread in srv0srv.c calls this regularly to drop tables which +we must drop in background after queries to them have ended. Such lazy +dropping of tables is needed in ALTER TABLE on Unix. */ + +ulint +row_drop_tables_for_mysql_in_background(void); +/*=========================================*/ + /* out: how many tables dropped + + remaining tables in list */ +/************************************************************************* +Get the background drop list length. NOTE: the caller must own the kernel +mutex! */ + +ulint +row_get_background_drop_list_len_low(void); +/*======================================*/ + /* out: how many tables in list */ +/************************************************************************* Drops a table for MySQL. If the name of the dropped table ends to characters INNODB_MONITOR, then this also stops printing of monitor output by the master thread. */ @@ -426,7 +444,7 @@ struct row_prebuilt_struct { fetched row in fetch_cache */ ulint n_fetch_cached; /* number of not yet fetched rows in fetch_cache */ - mem_heap_t* blob_heap; /* in SELECTS BLOB fields are copied + mem_heap_t* blob_heap; /* in SELECTS BLOB fie lds are copied to this heap */ mem_heap_t* old_vers_heap; /* memory heap where a previous version is built in consistent read */ diff --git a/innobase/include/srv0srv.h b/innobase/include/srv0srv.h index 831b4979356..2b40852fe67 100644 --- a/innobase/include/srv0srv.h +++ b/innobase/include/srv0srv.h @@ -250,6 +250,12 @@ mutex, for performace reasons). */ void srv_active_wake_master_thread(void); /*===============================*/ +/*********************************************************************** +Wakes up the master thread if it is suspended or being suspended. */ + +void +srv_wake_master_thread(void); +/*========================*/ /************************************************************************* Puts an OS thread to wait if there are too many concurrent threads (>= srv_thread_concurrency) inside InnoDB. The threads wait in a FIFO queue. */ diff --git a/innobase/include/trx0trx.h b/innobase/include/trx0trx.h index d67628b8bad..26c9ace08b6 100644 --- a/innobase/include/trx0trx.h +++ b/innobase/include/trx0trx.h @@ -50,6 +50,13 @@ trx_allocate_for_mysql(void); /*========================*/ /* out, own: transaction object */ /************************************************************************ +Creates a transaction object for background operations by the master thread. */ + +trx_t* +trx_allocate_for_background(void); +/*=============================*/ + /* out, own: transaction object */ +/************************************************************************ Frees a transaction object. */ void @@ -63,6 +70,13 @@ void trx_free_for_mysql( /*===============*/ trx_t* trx); /* in, own: trx object */ +/************************************************************************ +Frees a transaction object of a background operation of the master thread. */ + +void +trx_free_for_background( +/*====================*/ + trx_t* trx); /* in, own: trx object */ /******************************************************************** Creates trx objects for transactions and initializes the trx list of trx_sys at database start. Rollback segment and undo log lists must @@ -266,11 +280,14 @@ struct trx_sig_struct{ transaction is waiting a reply */ }; +#define TRX_MAGIC_N 91118598 + /* The transaction handle; every session has a trx object which is freed only when the session is freed; in addition there may be session-less transactions rolling back after a database recovery */ struct trx_struct{ + ulint magic_n; /* All the next fields are protected by the kernel mutex, except the undo logs which are protected by undo_mutex */ char* op_info; /* English text describing the diff --git a/innobase/log/log0recv.c b/innobase/log/log0recv.c index 7be172685ae..29e87c7572b 100644 --- a/innobase/log/log0recv.c +++ b/innobase/log/log0recv.c @@ -1020,8 +1020,9 @@ loop: if (recv_addr->state == RECV_NOT_PROCESSED) { if (!has_printed) { + ut_print_timestamp(stderr); fprintf(stderr, -"InnoDB: Starting an apply batch of log records to the database...\n" +" InnoDB: Starting an apply batch of log records to the database...\n" "InnoDB: Progress in percents: "); has_printed = TRUE; } diff --git a/innobase/mem/mem0mem.c b/innobase/mem/mem0mem.c index 3e0a9c008f3..0680968a7eb 100644 --- a/innobase/mem/mem0mem.c +++ b/innobase/mem/mem0mem.c @@ -75,6 +75,14 @@ After freeing, all the blocks in the heap are set to random bytes to help us discover errors which result from the use of buffers in an already freed heap. */ +#ifdef MEM_PERIODIC_CHECK + +ibool mem_block_list_inited; +/* List of all mem blocks allocated; protected by the mem_comm_pool mutex */ +UT_LIST_BASE_NODE_T(mem_block_t) mem_block_list; + +#endif + /******************************************************************* NOTE: Use the corresponding macro instead of this function. Allocates a single buffer of memory from the dynamic memory of @@ -169,7 +177,19 @@ mem_heap_create_block( 7); block->file_name[7]='\0'; block->line = line; + +#ifdef MEM_PERIODIC_CHECK + mem_pool_mutex_enter(); + + if (!mem_block_list_inited) { + mem_block_list_inited = TRUE; + UT_LIST_INIT(mem_block_list); + } + UT_LIST_ADD_LAST(mem_block_list, mem_block_list, block); + + mem_pool_mutex_exit(); +#endif mem_block_set_len(block, len); mem_block_set_type(block, type); mem_block_set_free(block, MEM_BLOCK_HEADER_SIZE); @@ -261,6 +281,13 @@ mem_heap_block_free( UT_LIST_REMOVE(list, heap->base, block); +#ifdef MEM_PERIODIC_CHECK + mem_pool_mutex_enter(); + + UT_LIST_REMOVE(mem_block_list, mem_block_list, block); + + mem_pool_mutex_exit(); +#endif type = heap->type; len = block->len; init_block = block->init_block; @@ -306,3 +333,30 @@ mem_heap_free_block_free( heap->free_block = NULL; } } + +#ifdef MEM_PERIODIC_CHECK +/********************************************************************** +Goes through the list of all allocated mem blocks, checks their magic +numbers, and reports possible corruption. */ + +void +mem_validate_all_blocks(void) +/*=========================*/ +{ + mem_block_t* block; + + mem_pool_mutex_enter(); + + block = UT_LIST_GET_FIRST(mem_block_list); + + while (block) { + if (block->magic_n != MEM_BLOCK_MAGIC_N) { + mem_analyze_corruption((byte*)block); + } + + block = UT_LIST_GET_NEXT(mem_block_list, block); + } + + mem_pool_mutex_exit(); +} +#endif diff --git a/innobase/mem/mem0pool.c b/innobase/mem/mem0pool.c index 48e7e686953..3681c8ef779 100644 --- a/innobase/mem/mem0pool.c +++ b/innobase/mem/mem0pool.c @@ -78,9 +78,9 @@ pool, and after that its locks will grow into the buffer pool. */ /* The smallest memory area total size */ #define MEM_AREA_MIN_SIZE (2 * MEM_AREA_EXTRA_SIZE) + /* Data structure for a memory pool. The space is allocated using the buddy algorithm, where free list i contains areas of size 2 to power i. */ - struct mem_pool_struct{ byte* buf; /* memory pool */ ulint size; /* memory common pool size */ @@ -99,6 +99,26 @@ mem_pool_t* mem_comm_pool = NULL; ulint mem_out_of_mem_err_msg_count = 0; /************************************************************************ +Reserves the mem pool mutex. */ + +void +mem_pool_mutex_enter(void) +/*======================*/ +{ + mutex_enter(&(mem_comm_pool->mutex)); +} + +/************************************************************************ +Releases the mem pool mutex. */ + +void +mem_pool_mutex_exit(void) +/*=====================*/ +{ + mutex_exit(&(mem_comm_pool->mutex)); +} + +/************************************************************************ Returns memory area size. */ UNIV_INLINE ulint @@ -240,15 +260,15 @@ mem_pool_fill_free_list( if (mem_out_of_mem_err_msg_count % 1000000000 == 0) { /* We do not print the message every time: */ + + ut_print_timestamp(stderr); fprintf(stderr, - "Innobase: Warning: out of memory in additional memory pool.\n"); - fprintf(stderr, - "Innobase: Innobase will start allocating memory from the OS.\n"); - fprintf(stderr, - "Innobase: You should restart the database with a bigger value in\n"); - fprintf(stderr, - "Innobase: the MySQL .cnf file for innobase_additional_mem_pool_size.\n"); + " InnoDB: Out of memory in additional memory pool.\n" + "InnoDB: InnoDB will start allocating memory from the OS.\n" + "InnoDB: You may get better performance if you configure a bigger\n" + "InnoDB: value in the MySQL my.cnf file for\n" + "InnoDB: innodb_additional_mem_pool_size.\n"); } mem_out_of_mem_err_msg_count++; diff --git a/innobase/rem/rem0cmp.c b/innobase/rem/rem0cmp.c index 47b7021bf27..9a5a0eb6e8e 100644 --- a/innobase/rem/rem0cmp.c +++ b/innobase/rem/rem0cmp.c @@ -113,6 +113,16 @@ cmp_types_are_equal( return(FALSE); } + if (type1->mtype == DATA_INT + && (type1->prtype & DATA_UNSIGNED) + != (type2->prtype & DATA_UNSIGNED)) { + /* The storage format of an unsigned integer is different + from a signed integer: in a signed integer we OR + 0x8000... to the value of positive integers. */ + + return(FALSE); + } + if (type1->mtype == DATA_MYSQL || type1->mtype == DATA_VARMYSQL) { diff --git a/innobase/row/row0mysql.c b/innobase/row/row0mysql.c index 00184c13ae9..292f8c0c6c8 100644 --- a/innobase/row/row0mysql.c +++ b/innobase/row/row0mysql.c @@ -26,6 +26,19 @@ Created 9/17/2000 Heikki Tuuri #include "trx0purge.h" #include "lock0lock.h" #include "rem0cmp.h" +#include "log0log.h" + +/* List of tables we should drop in background. ALTER TABLE in MySQL requires +that the table handler can drop the table in background when there are no +queries to it any more. Protected by the kernel mutex. */ +typedef struct row_mysql_drop_struct row_mysql_drop_t; +struct row_mysql_drop_struct{ + char* table_name; + UT_LIST_NODE_T(row_mysql_drop_t) row_mysql_drop_list; +}; + +UT_LIST_BASE_NODE_T(row_mysql_drop_t) row_mysql_drop_list; +ibool row_mysql_drop_list_inited = FALSE; /*********************************************************************** Reads a MySQL format variable-length field (like VARCHAR) length and @@ -172,10 +185,22 @@ handle_new_error: trx_general_rollback_for_mysql(trx, TRUE, savept); } } else if (err == DB_TOO_BIG_RECORD) { + if (savept) { + /* Roll back the latest, possibly incomplete + insertion or update */ + + trx_general_rollback_for_mysql(trx, TRUE, savept); + } /* MySQL will roll back the latest SQL statement */ } else if (err == DB_ROW_IS_REFERENCED || err == DB_NO_REFERENCED_ROW || err == DB_CANNOT_ADD_CONSTRAINT) { + if (savept) { + /* Roll back the latest, possibly incomplete + insertion or update */ + + trx_general_rollback_for_mysql(trx, TRUE, savept); + } /* MySQL will roll back the latest SQL statement */ } else if (err == DB_LOCK_WAIT) { @@ -200,6 +225,12 @@ handle_new_error: trx_general_rollback_for_mysql(trx, FALSE, NULL); } else if (err == DB_OUT_OF_FILE_SPACE) { + if (savept) { + /* Roll back the latest, possibly incomplete + insertion or update */ + + trx_general_rollback_for_mysql(trx, TRUE, savept); + } /* MySQL will roll back the latest SQL statement */ } else if (err == DB_MUST_GET_MORE_FILE_SPACE) { @@ -375,13 +406,13 @@ row_update_prebuilt_trx( handle */ trx_t* trx) /* in: transaction handle */ { - if (prebuilt->magic_n != ROW_PREBUILT_ALLOCATED) { + if (trx->magic_n != TRX_MAGIC_N) { fprintf(stderr, - "InnoDB: Error: trying to free a corrupt\n" - "InnoDB: table handle. Magic n %lu, table name %s\n", - prebuilt->magic_n, prebuilt->table->name); + "InnoDB: Error: trying to use a corrupt\n" + "InnoDB: trx handle. Magic n %lu\n", + trx->magic_n); - mem_analyze_corruption((byte*)prebuilt); + mem_analyze_corruption((byte*)trx); ut_a(0); } @@ -1172,8 +1203,11 @@ row_create_table_for_mysql( row_drop_table_for_mysql(table->name, trx, TRUE); } else { ut_a(err == DB_DUPLICATE_KEY); + + ut_print_timestamp(stderr); + fprintf(stderr, - "InnoDB: Error: table %s already exists in InnoDB internal\n" + " InnoDB: Error: table %s already exists in InnoDB internal\n" "InnoDB: data dictionary. Have you deleted the .frm file\n" "InnoDB: and not used DROP TABLE? Have you used DROP DATABASE\n" "InnoDB: for InnoDB tables in MySQL version <= 3.23.43?\n" @@ -1352,6 +1386,164 @@ row_table_add_foreign_constraints( } /************************************************************************* +Drops a table for MySQL as a background operation. MySQL relies on Unix +in ALTER TABLE to the fact that the table handler does not remove the +table before all handles to it has been removed. Furhermore, the MySQL's +call to drop table must be non-blocking. Therefore we do the drop table +as a background operation, which is taken care of by the master thread +in srv0srv.c. */ +static +int +row_drop_table_for_mysql_in_background( +/*===================================*/ + /* out: error code or DB_SUCCESS */ + char* name) /* in: table name */ +{ + ulint error; + trx_t* trx; + + trx = trx_allocate_for_background(); + +/* fprintf(stderr, "InnoDB: Dropping table %s in background drop list\n", + name); */ + /* Drop the table in InnoDB */ + + error = row_drop_table_for_mysql(name, trx, FALSE); + + if (error != DB_SUCCESS) { + fprintf(stderr, + "InnoDB: Error: Dropping table %s in background drop list failed\n", + name); + } + + /* Flush the log to reduce probability that the .frm files and + the InnoDB data dictionary get out-of-sync if the user runs + with innodb_flush_log_at_trx_commit = 0 */ + + log_flush_up_to(ut_dulint_max, LOG_WAIT_ONE_GROUP); + + trx_commit_for_mysql(trx); + + trx_free_for_background(trx); + + return(DB_SUCCESS); +} + +/************************************************************************* +The master thread in srv0srv.c calls this regularly to drop tables which +we must drop in background after queries to them have ended. Such lazy +dropping of tables is needed in ALTER TABLE on Unix. */ + +ulint +row_drop_tables_for_mysql_in_background(void) +/*=========================================*/ + /* out: how many tables dropped + + remaining tables in list */ +{ + row_mysql_drop_t* drop; + dict_table_t* table; + ulint n_tables; + ulint n_tables_dropped = 0; +loop: + mutex_enter(&kernel_mutex); + + if (!row_mysql_drop_list_inited) { + + UT_LIST_INIT(row_mysql_drop_list); + row_mysql_drop_list_inited = TRUE; + } + + drop = UT_LIST_GET_FIRST(row_mysql_drop_list); + + n_tables = UT_LIST_GET_LEN(row_mysql_drop_list); + + mutex_exit(&kernel_mutex); + + if (drop == NULL) { + + return(n_tables + n_tables_dropped); + } + + mutex_enter(&(dict_sys->mutex)); + table = dict_table_get_low(drop->table_name); + mutex_exit(&(dict_sys->mutex)); + + if (table->n_mysql_handles_opened > 0) { + + return(n_tables + n_tables_dropped); + } + + n_tables_dropped++; + + row_drop_table_for_mysql_in_background(drop->table_name); + + mutex_enter(&kernel_mutex); + + UT_LIST_REMOVE(row_mysql_drop_list, row_mysql_drop_list, drop); + + mem_free(drop->table_name); + + mem_free(drop); + + mutex_exit(&kernel_mutex); + + goto loop; +} + +/************************************************************************* +Get the background drop list length. NOTE: the caller must own the kernel +mutex! */ + +ulint +row_get_background_drop_list_len_low(void) +/*======================================*/ + /* out: how many tables in list */ +{ + ut_ad(mutex_own(&kernel_mutex)); + + if (!row_mysql_drop_list_inited) { + + UT_LIST_INIT(row_mysql_drop_list); + row_mysql_drop_list_inited = TRUE; + } + + return(UT_LIST_GET_LEN(row_mysql_drop_list)); +} + +/************************************************************************* +Adds a table to the list of tables which the master thread drops in +background. We need this on Unix because in ALTER TABLE MySQL may call +drop table even if the table has running queries on it. */ +static +void +row_add_table_to_background_drop_list( +/*==================================*/ + dict_table_t* table) /* in: table */ +{ + row_mysql_drop_t* drop; + + drop = mem_alloc(sizeof(row_mysql_drop_t)); + + drop->table_name = mem_alloc(1 + ut_strlen(table->name)); + + ut_memcpy(drop->table_name, table->name, 1 + ut_strlen(table->name)); + + mutex_enter(&kernel_mutex); + + if (!row_mysql_drop_list_inited) { + + UT_LIST_INIT(row_mysql_drop_list); + row_mysql_drop_list_inited = TRUE; + } + + UT_LIST_ADD_LAST(row_mysql_drop_list, row_mysql_drop_list, drop); + +/* fprintf(stderr, "InnoDB: Adding table %s to background drop list\n", + drop->table_name); */ + mutex_exit(&kernel_mutex); +} + +/************************************************************************* Drops a table for MySQL. If the name of the dropped table ends to characters INNODB_MONITOR, then this also stops printing of monitor output by the master thread. */ @@ -1541,9 +1733,10 @@ row_drop_table_for_mysql( if (!table) { err = DB_TABLE_NOT_FOUND; + ut_print_timestamp(stderr); fprintf(stderr, - "InnoDB: Error: table %s does not exist in the InnoDB internal\n" + " InnoDB: Error: table %s does not exist in the InnoDB internal\n" "InnoDB: data dictionary though MySQL is trying to drop it.\n" "InnoDB: Have you copied the .frm file of the table to the\n" "InnoDB: MySQL database directory from another database?\n", @@ -1551,42 +1744,19 @@ row_drop_table_for_mysql( goto funct_exit; } - /* Remove any locks there are on the table or its records */ - - lock_reset_all_on_table(table); -loop: if (table->n_mysql_handles_opened > 0) { - rw_lock_s_unlock(&(purge_sys->purge_is_running)); - - rw_lock_x_unlock(&(dict_foreign_key_check_lock)); - - mutex_exit(&(dict_sys->mutex)); - - if (rounds > 60) { - fprintf(stderr, - "InnoDB: waiting for queries to table %s to end before dropping it\n", - name); - } - os_thread_sleep(1000000); + row_add_table_to_background_drop_list(table); - mutex_enter(&(dict_sys->mutex)); - - rw_lock_x_lock(&(dict_foreign_key_check_lock)); - - rw_lock_s_lock(&(purge_sys->purge_is_running)); + err = DB_SUCCESS; - rounds++; - - if (rounds > 120) { - fprintf(stderr, -"InnoDB: Warning: queries to table %s have not ended but we continue anyway\n", - name); - } else { - goto loop; - } + goto funct_exit; } + /* Remove any locks there are on the table or its records */ + + lock_reset_all_on_table(table); + trx->dict_operation = TRUE; trx->table_id = table->id; @@ -1622,6 +1792,8 @@ funct_exit: trx->op_info = (char *) ""; + srv_wake_master_thread(); + return((int) err); } @@ -1793,7 +1965,31 @@ row_rename_table_for_mysql( err = trx->error_state; if (err != DB_SUCCESS) { - row_mysql_handle_errors(&err, trx, thr, NULL); + if (err == DB_DUPLICATE_KEY) { + ut_print_timestamp(stderr); + + fprintf(stderr, + " InnoDB: Error: table %s exists in the InnoDB internal data\n" + "InnoDB: dictionary though MySQL is trying rename table %s to it.\n" + "InnoDB: Have you deleted the .frm file and not used DROP TABLE?\n", + new_name, old_name); + + fprintf(stderr, + "InnoDB: If table %s is a temporary table #sql..., then it can be that\n" + "InnoDB: there are still queries running on the table, and it will be\n" + "InnoDB: dropped automatically when the queries end.\n", new_name); + + fprintf(stderr, + "InnoDB: You can drop the orphaned table inside InnoDB by\n" + "InnoDB: creating an InnoDB table with the same name in another\n" + "InnoDB: database and moving the .frm file to the current database.\n" + "InnoDB: Then MySQL thinks the table exists, and DROP TABLE will\n" + "InnoDB: succeed.\n"); + } + + trx->error_state = DB_SUCCESS; + trx_general_rollback_for_mysql(trx, FALSE, NULL); + trx->error_state = DB_SUCCESS; } else { ut_a(dict_table_rename_in_cache(table, new_name)); } @@ -1950,7 +2146,7 @@ row_check_table_for_mysql( ulint ret = DB_SUCCESS; prebuilt->trx->op_info = (char *) "checking table"; - + index = dict_table_get_first_index(table); while (index != NULL) { diff --git a/innobase/srv/srv0srv.c b/innobase/srv/srv0srv.c index cbc22b9b28a..c3d3ed53058 100644 --- a/innobase/srv/srv0srv.c +++ b/innobase/srv/srv0srv.c @@ -20,7 +20,7 @@ Windows 2000 will have something called thread pooling Another possibility could be to use some very fast user space thread library. This might confuse NT though. -(c) 1995 InnoDB Oy +(c) 1995 Innobase Oy Created 10/8/1995 Heikki Tuuri *******************************************************/ @@ -49,6 +49,7 @@ Created 10/8/1995 Heikki Tuuri #include "btr0sea.h" #include "dict0load.h" #include "srv0start.h" +#include "row0mysql.h" /* Buffer which can be used in printing fatal error messages */ char srv_fatal_errbuf[5000]; @@ -91,8 +92,43 @@ ibool srv_log_archive_on = TRUE; ulint srv_log_buffer_size = ULINT_MAX; /* size in database pages */ ibool srv_flush_log_at_trx_commit = TRUE; -byte srv_latin1_ordering[256]; /* The sort order table of the latin1 - character set */ +byte srv_latin1_ordering[256] /* The sort order table of the latin1 + character set. The following table is + the MySQL order as of Feb 10th, 2002 */ += { + 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07 +, 0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, 0x0E, 0x0F +, 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17 +, 0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F +, 0x20, 0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27 +, 0x28, 0x29, 0x2A, 0x2B, 0x2C, 0x2D, 0x2E, 0x2F +, 0x30, 0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37 +, 0x38, 0x39, 0x3A, 0x3B, 0x3C, 0x3D, 0x3E, 0x3F +, 0x40, 0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47 +, 0x48, 0x49, 0x4A, 0x4B, 0x4C, 0x4D, 0x4E, 0x4F +, 0x50, 0x51, 0x52, 0x53, 0x54, 0x55, 0x56, 0x57 +, 0x58, 0x59, 0x5A, 0x5B, 0x5C, 0x5D, 0x5E, 0x5F +, 0x60, 0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47 +, 0x48, 0x49, 0x4A, 0x4B, 0x4C, 0x4D, 0x4E, 0x4F +, 0x50, 0x51, 0x52, 0x53, 0x54, 0x55, 0x56, 0x57 +, 0x58, 0x59, 0x5A, 0x7B, 0x7C, 0x7D, 0x7E, 0x7F +, 0x80, 0x81, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87 +, 0x88, 0x89, 0x8A, 0x8B, 0x8C, 0x8D, 0x8E, 0x8F +, 0x90, 0x91, 0x92, 0x93, 0x94, 0x95, 0x96, 0x97 +, 0x98, 0x99, 0x9A, 0x9B, 0x9C, 0x9D, 0x9E, 0x9F +, 0xA0, 0xA1, 0xA2, 0xA3, 0xA4, 0xA5, 0xA6, 0xA7 +, 0xA8, 0xA9, 0xAA, 0xAB, 0xAC, 0xAD, 0xAE, 0xAF +, 0xB0, 0xB1, 0xB2, 0xB3, 0xB4, 0xB5, 0xB6, 0xB7 +, 0xB8, 0xB9, 0xBA, 0xBB, 0xBC, 0xBD, 0xBE, 0xBF +, 0x41, 0x41, 0x41, 0x41, 0x5C, 0x5B, 0x5C, 0x43 +, 0x45, 0x45, 0x45, 0x45, 0x49, 0x49, 0x49, 0x49 +, 0x44, 0x4E, 0x4F, 0x4F, 0x4F, 0x4F, 0x5D, 0xD7 +, 0xD8, 0x55, 0x55, 0x55, 0x59, 0x59, 0xDE, 0xDF +, 0x41, 0x41, 0x41, 0x41, 0x5C, 0x5B, 0x5C, 0x43 +, 0x45, 0x45, 0x45, 0x45, 0x49, 0x49, 0x49, 0x49 +, 0x44, 0x4E, 0x4F, 0x4F, 0x4F, 0x4F, 0x5D, 0xF7 +, 0xD8, 0x55, 0x55, 0x55, 0x59, 0x59, 0xDE, 0xFF +}; ibool srv_use_native_aio = FALSE; @@ -1920,17 +1956,12 @@ srv_boot(void) srv_init(); - /* Reserve the first slot for the current thread, i.e., the master - thread */ - - srv_table_reserve_slot(SRV_MASTER); - return(DB_SUCCESS); } /************************************************************************* Reserves a slot in the thread table for the current MySQL OS thread. -NOTE! The server mutex has to be reserved by the caller! */ +NOTE! The kernel mutex has to be reserved by the caller! */ static srv_slot_t* srv_table_reserve_slot_for_mysql(void) @@ -1940,6 +1971,8 @@ srv_table_reserve_slot_for_mysql(void) srv_slot_t* slot; ulint i; + ut_ad(mutex_own(&kernel_mutex)); + i = 0; slot = srv_mysql_table + i; @@ -2361,6 +2394,22 @@ srv_active_wake_master_thread(void) } } +/*********************************************************************** +Wakes up the master thread if it is suspended or being suspended. */ + +void +srv_wake_master_thread(void) +/*========================*/ +{ + srv_activity_count++; + + mutex_enter(&kernel_mutex); + + srv_release_threads(SRV_MASTER, 1); + + mutex_exit(&kernel_mutex); +} + /************************************************************************* The master thread controlling the server. */ @@ -2383,6 +2432,7 @@ srv_master_thread( ulint n_bytes_merged; ulint n_pages_flushed; ulint n_bytes_archived; + ulint n_tables_to_drop; ulint n_ios; ulint n_ios_old; ulint n_ios_very_old; @@ -2424,7 +2474,11 @@ loop: can drop tables lazily after there no longer are SELECT queries to them. */ -/* row_drop_tables_for_mysql_in_background(); */ + srv_main_thread_op_info = "doing background drop tables"; + + row_drop_tables_for_mysql_in_background(); + + srv_main_thread_op_info = ""; if (srv_force_recovery >= SRV_FORCE_NO_BACKGROUND) { @@ -2475,6 +2529,11 @@ loop: printf("Master thread wakes up!\n"); } +#ifdef MEM_PERIODIC_CHECK + /* Check magic numbers of every allocated mem block once in 10 + seconds */ + mem_validate_all_blocks(); +#endif /* If there were less than 200 i/os during the 10 second period, we assume that there is free disk i/o capacity available, and it makes sense to do a buffer pool flush. */ @@ -2531,6 +2590,12 @@ background_loop: /* In this loop we run background operations when the server is quiet and we also come here about once in 10 seconds */ + srv_main_thread_op_info = "doing background drop tables"; + + n_tables_to_drop = row_drop_tables_for_mysql_in_background(); + + srv_main_thread_op_info = ""; + srv_main_thread_op_info = "flushing buffer pool pages"; /* Flush a few oldest pages to make the checkpoint younger */ @@ -2616,11 +2681,13 @@ background_loop: log_archive_do(FALSE, &n_bytes_archived); if (srv_fast_shutdown && srv_shutdown_state > 0) { - if (n_pages_flushed + n_bytes_archived != 0) { + if (n_tables_to_drop + n_pages_flushed + + n_bytes_archived != 0) { goto background_loop; } - } else if (n_pages_purged + n_bytes_merged + n_pages_flushed + } else if (n_tables_to_drop + + n_pages_purged + n_bytes_merged + n_pages_flushed + n_bytes_archived != 0) { goto background_loop; } @@ -2639,6 +2706,12 @@ suspend_thread: mutex_enter(&kernel_mutex); + if (row_get_background_drop_list_len_low() > 0) { + mutex_exit(&kernel_mutex); + + goto loop; + } + event = srv_suspend_thread(); mutex_exit(&kernel_mutex); diff --git a/innobase/srv/srv0start.c b/innobase/srv/srv0start.c index 70c85d7fbeb..bbb48331811 100644 --- a/innobase/srv/srv0start.c +++ b/innobase/srv/srv0start.c @@ -271,13 +271,18 @@ open_or_create_log_file( } else { *log_file_created = TRUE; + ut_print_timestamp(stderr); + fprintf(stderr, - "InnoDB: Log file %s did not exist: new to be created\n", + " InnoDB: Log file %s did not exist: new to be created\n", name); fprintf(stderr, "InnoDB: Setting log file %s size to %lu MB\n", name, srv_log_file_size >> (20 - UNIV_PAGE_SIZE_SHIFT)); + fprintf(stderr, + "InnoDB: Database physically writes the file full: wait...\n"); + ret = os_file_set_size(name, files[i], srv_calc_low32(srv_log_file_size), srv_calc_high32(srv_log_file_size)); @@ -456,8 +461,9 @@ open_or_create_data_files( one_created = TRUE; if (i > 0) { + ut_print_timestamp(stderr); fprintf(stderr, - "InnoDB: Data file %s did not exist: new to be created\n", + " InnoDB: Data file %s did not exist: new to be created\n", name); } else { fprintf(stderr, @@ -466,8 +472,9 @@ open_or_create_data_files( *create_new_db = TRUE; } + ut_print_timestamp(stderr); fprintf(stderr, - "InnoDB: Setting file %s size to %lu MB\n", + " InnoDB: Setting file %s size to %lu MB\n", name, (srv_data_file_sizes[i] >> (20 - UNIV_PAGE_SIZE_SHIFT))); @@ -911,6 +918,12 @@ innobase_start_or_create_for_mysql(void) mtr_commit(&mtr); } + if (recv_needed_recovery) { + ut_print_timestamp(stderr); + fprintf(stderr, + " InnoDB: Flushing modified pages from the buffer pool...\n"); + } + log_make_checkpoint_at(ut_dulint_max, TRUE); if (!srv_log_archive_on) { @@ -991,9 +1004,8 @@ innobase_shutdown_for_mysql(void) ut_print_timestamp(stderr); fprintf(stderr, " InnoDB: Warning: shutting down a not properly started\n"); - ut_print_timestamp(stderr); fprintf(stderr, - " InnoDB: or created database!\n"); + " InnoDB: or created database!\n"); } return(DB_SUCCESS); diff --git a/innobase/trx/trx0purge.c b/innobase/trx/trx0purge.c index 31a0c20ad14..97362d00b4b 100644 --- a/innobase/trx/trx0purge.c +++ b/innobase/trx/trx0purge.c @@ -226,9 +226,9 @@ trx_purge_sys_create(void) value */ purge_sys->sess = sess_open(com_endpoint, (byte*)"purge_system", 13); - purge_sys->trx = (purge_sys->sess)->trx; + purge_sys->trx = purge_sys->sess->trx; - (purge_sys->trx)->type = TRX_PURGE; + purge_sys->trx->type = TRX_PURGE; ut_a(trx_start_low(purge_sys->trx, ULINT_UNDEFINED)); diff --git a/innobase/trx/trx0trx.c b/innobase/trx/trx0trx.c index 315bb7030ee..607c80edd09 100644 --- a/innobase/trx/trx0trx.c +++ b/innobase/trx/trx0trx.c @@ -26,9 +26,9 @@ Created 3/26/1996 Heikki Tuuri /* Copy of the prototype for innobase_mysql_print_thd: this - copy must be equal to the one in mysql/sql/ha_innobase.cc ! */ -void innobase_mysql_print_thd(void* thd); +copy must be equal to the one in mysql/sql/ha_innobase.cc ! */ +void innobase_mysql_print_thd(void* thd); /* Dummy session used currently in MySQL interface */ sess_t* trx_dummy_sess = NULL; @@ -64,6 +64,8 @@ trx_create( trx = mem_alloc(sizeof(trx_t)); + trx->magic_n = TRX_MAGIC_N; + trx->op_info = (char *) ""; trx->type = TRX_USER; @@ -158,6 +160,32 @@ trx_allocate_for_mysql(void) } /************************************************************************ +Creates a transaction object for background operations by the master thread. */ + +trx_t* +trx_allocate_for_background(void) +/*=============================*/ + /* out, own: transaction object */ +{ + trx_t* trx; + + mutex_enter(&kernel_mutex); + + /* Open a dummy session */ + + if (!trx_dummy_sess) { + trx_dummy_sess = sess_open(NULL, (byte*)"Dummy sess", + ut_strlen("Dummy sess")); + } + + trx = trx_create(trx_dummy_sess); + + mutex_exit(&kernel_mutex); + + return(trx); +} + +/************************************************************************ Releases the search latch if trx has reserved it. */ void @@ -181,6 +209,11 @@ trx_free( trx_t* trx) /* in, own: trx object */ { ut_ad(mutex_own(&kernel_mutex)); + + ut_a(trx->magic_n == TRX_MAGIC_N); + + trx->magic_n = 11112222; + ut_a(trx->conc_state == TRX_NOT_STARTED); mutex_free(&(trx->undo_mutex)); @@ -242,6 +275,21 @@ trx_free_for_mysql( mutex_exit(&kernel_mutex); } +/************************************************************************ +Frees a transaction object of a background operation of the master thread. */ + +void +trx_free_for_background( +/*====================*/ + trx_t* trx) /* in, own: trx object */ +{ + mutex_enter(&kernel_mutex); + + trx_free(trx); + + mutex_exit(&kernel_mutex); +} + /******************************************************************** Inserts the trx handle in the trx system trx list in the right position. The list is sorted on the trx id so that the biggest id is at the list diff --git a/mysql-test/t/join.test b/mysql-test/t/join.test index cbd4898785a..3bd78350267 100644 --- a/mysql-test/t/join.test +++ b/mysql-test/t/join.test @@ -194,3 +194,27 @@ t3.Contractor_ID = '999999' OR t3.Contractor_ID = '1') AND t3.CanRead='1' AND t3.Active='1'; drop table t1,t2,t3; + +# +# Bug when doing full join and NULL fields. +# + +CREATE TABLE t1 ( + t1_id int(11) default NULL, + t2_id int(11) default NULL, + type enum('Cost','Percent') default NULL, + cost_unit enum('Cost','Unit') default NULL, + min_value double default NULL, + max_value double default NULL, + t3_id int(11) default NULL, + item_id int(11) default NULL +) TYPE=MyISAM; +INSERT INTO t1 VALUES (12,5,'Percent','Cost',-1,0,-1,-1),(14,4,'Percent','Cost',-1,0,-1,-1),(18,5,'Percent','Cost',-1,0,-1,-1),(19,4,'Percent','Cost',-1,0,-1,-1),(20,5,'Percent','Cost',100,-1,22,291),(21,5,'Percent','Cost',100,-1,18,291),(22,1,'Percent','Cost',100,-1,6,291),(23,1,'Percent','Cost',100,-1,21,291),(24,1,'Percent','Cost',100,-1,9,291),(25,1,'Percent','Cost',100,-1,4,291),(26,1,'Percent','Cost',100,-1,20,291),(27,4,'Percent','Cost',100,-1,7,202),(28,1,'Percent','Cost',50,-1,-1,137),(29,2,'Percent','Cost',100,-1,4,354),(30,2,'Percent','Cost',100,-1,9,137),(93,2,'Cost','Cost',-1,10000000,-1,-1); +CREATE TABLE t2 ( + id int(10) unsigned NOT NULL auto_increment, + name varchar(255) default NULL, + PRIMARY KEY (id) +) TYPE=MyISAM; +INSERT INTO t2 VALUES (1,'s1'),(2,'s2'),(3,'s3'),(4,'s4'),(5,'s5'); +select t1.*, t2.* from t1, t2 where t2.id=t1.t2_id limit 2; +drop table t1,t2; diff --git a/sql/ha_innodb.cc b/sql/ha_innodb.cc index f99e677913e..e2657e5cad7 100644 --- a/sql/ha_innodb.cc +++ b/sql/ha_innodb.cc @@ -1,4 +1,4 @@ -/* Copyright (C) 2000 MySQL AB & MySQL Finland AB & InnoDB Oy +/* Copyright (C) 2000 MySQL AB & InnoDB Oy This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by @@ -85,8 +85,8 @@ long innobase_mirrored_log_groups, innobase_log_files_in_group, char *innobase_data_home_dir; char *innobase_log_group_home_dir, *innobase_log_arch_dir; char *innobase_unix_file_flush_method; -bool innobase_flush_log_at_trx_commit, innobase_log_archive, - innobase_use_native_aio, innobase_fast_shutdown; +my_bool innobase_flush_log_at_trx_commit, innobase_log_archive, + innobase_use_native_aio, innobase_fast_shutdown; /* Set default InnoDB size to 64M, to let users use InnoDB without having @@ -235,7 +235,7 @@ convert_error_code_to_mysql( extern "C" { /***************************************************************** Prints info of a THD object (== user session thread) to the -standatd output. NOTE that mysql/innobase/trx/trx0trx.c must contain +standard output. NOTE that mysql/innobase/trx/trx0trx.c must contain the prototype for this function! */ void @@ -304,6 +304,8 @@ check_trx_exists( thd->transaction.stmt.innobase_tid = (void*)&innodb_dummy_stmt_trx_handle; + } else { + ut_a(trx->magic_n == TRX_MAGIC_N); } return(trx); @@ -841,6 +843,7 @@ innobase_close_connection( whose transaction should be rolled back */ { if (NULL != thd->transaction.all.innobase_tid) { + trx_rollback_for_mysql((trx_t*) (thd->transaction.all.innobase_tid)); trx_free_for_mysql((trx_t*) @@ -2465,44 +2468,6 @@ ha_innobase::position( ref_stored_len = len; } -/*********************************************************************** -Tells something additional to the handler about how to do things. */ - -int -ha_innobase::extra( -/*===============*/ - /* out: 0 or error number */ - enum ha_extra_function operation) - /* in: HA_EXTRA_DONT_USE_CURSOR_TO_UPDATE */ -{ - row_prebuilt_t* prebuilt = (row_prebuilt_t*) innobase_prebuilt; - - switch (operation) { - case HA_EXTRA_RESET: - case HA_EXTRA_RESET_STATE: - prebuilt->read_just_key = 0; - break; - case HA_EXTRA_NO_KEYREAD: - prebuilt->read_just_key = 0; - break; - case HA_EXTRA_DONT_USE_CURSOR_TO_UPDATE: - prebuilt->in_update_remember_pos = FALSE; - break; - case HA_EXTRA_KEYREAD: - prebuilt->read_just_key = 1; - break; - default:/* Do nothing */ - ; - } - - return(0); -} - -int ha_innobase::reset(void) -{ - return(0); -} - /********************************************************************* Creates a table definition to an InnoDB database. */ @@ -2981,9 +2946,9 @@ ha_innobase::records_in_range( DBUG_ENTER("records_in_range"); - if (prebuilt->trx) { - prebuilt->trx->op_info = (char*) "estimating range size"; - } + /* Warning: since it is not sure that MySQL calls external_lock + before calling this function, the trx field in prebuilt can be + obsolete! */ active_index = keynr; @@ -3017,10 +2982,6 @@ ha_innobase::records_in_range( my_free((char*) key_val_buff2, MYF(0)); - if (prebuilt->trx) { - prebuilt->trx->op_info = (char*) ""; - } - DBUG_RETURN((ha_rows) n_rows); } @@ -3041,10 +3002,9 @@ ha_innobase::estimate_number_of_rows(void) ulonglong estimate; ulonglong data_file_length; - if (prebuilt->trx) { - prebuilt->trx->op_info = - (char*) "estimating upper bound of table size"; - } + /* Warning: since it is not sure that MySQL calls external_lock + before calling this function, the trx field in prebuilt can be + obsolete! */ DBUG_ENTER("info"); @@ -3061,10 +3021,6 @@ ha_innobase::estimate_number_of_rows(void) estimate = 2 * data_file_length / dict_index_calc_min_rec_len(index); - if (prebuilt->trx) { - prebuilt->trx->op_info = (char*) ""; - } - DBUG_RETURN((ha_rows) estimate); } @@ -3080,10 +3036,12 @@ ha_innobase::scan_time() { row_prebuilt_t* prebuilt = (row_prebuilt_t*) innobase_prebuilt; - /* In the following formula we assume that scanning 10 pages - takes the same time as a disk seek: */ - - return((double) (prebuilt->table->stat_clustered_index_size / 10)); + /* Since MySQL seems to favor table scans too much over index + searches, we pretend that a sequential read takes the same time + as a random disk read, that is, we do not divide the following + by 10, which would be physically realistic. */ + + return((double) (prebuilt->table->stat_clustered_index_size)); } /************************************************************************* @@ -3104,9 +3062,9 @@ ha_innobase::info( DBUG_ENTER("info"); - if (prebuilt->trx) { - prebuilt->trx->op_info = (char*) "calculating table stats"; - } + /* Warning: since it is not sure that MySQL calls external_lock + before calling this function, the trx field in prebuilt can be + obsolete! */ ib_table = prebuilt->table; @@ -3154,25 +3112,17 @@ ha_innobase::info( index->stat_n_diff_key_vals[j + 1]); } + /* Since MySQL seems to favor table scans + too much over index searches, we pretend + index selectivity is 2 times better than + our estimate: */ + + rec_per_key = rec_per_key / 2; + if (rec_per_key == 0) { rec_per_key = 1; } - /* Since the MySQL optimizer is often too - pessimistic in the assumption that a table - does not fit in the buffer pool, we - increase the attractiveness of indexes - by assuming the selectivity of any prefix - of an index is 1 / 100 or better. - (Actually, we should look at the table - size, and if the table is smaller than - the buffer pool, we should uniformly - increase the attractiveness of indexes, - regardless of the estimated selectivity.) */ - - if (rec_per_key > records / 100) { - rec_per_key = records / 100; - } table->key_info[i].rec_per_key[j] = rec_per_key; } @@ -3188,15 +3138,13 @@ ha_innobase::info( pointer and cause a seg fault. */ if (flag & HA_STATUS_ERRKEY) { + ut_a(prebuilt->trx && prebuilt->trx->magic_n == TRX_MAGIC_N); + errkey = (unsigned int) row_get_mysql_key_number_for_index( (dict_index_t*) trx_get_error_info(prebuilt->trx)); } - if (prebuilt->trx) { - prebuilt->trx->op_info = (char*) ""; - } - DBUG_VOID_RETURN; } @@ -3217,6 +3165,8 @@ ha_innobase::check( row_prebuilt_t* prebuilt = (row_prebuilt_t*) innobase_prebuilt; ulint ret; + ut_a(prebuilt->trx && prebuilt->trx->magic_n == TRX_MAGIC_N); + if (prebuilt->mysql_template == NULL) { /* Build the template; we will use a dummy template in index scans done in checking */ @@ -3250,6 +3200,10 @@ ha_innobase::update_table_comment( char* str = my_malloc(length + 550, MYF(0)); char* pos; + /* Warning: since it is not sure that MySQL calls external_lock + before calling this function, the trx field in prebuilt can be + obsolete! */ + if (!str) { return((char*)comment); } @@ -3271,6 +3225,53 @@ ha_innobase::update_table_comment( return(str); } +/*********************************************************************** +Tells something additional to the handler about how to do things. */ + +int +ha_innobase::extra( +/*===============*/ + /* out: 0 or error number */ + enum ha_extra_function operation) + /* in: HA_EXTRA_DONT_USE_CURSOR_TO_UPDATE */ +{ + row_prebuilt_t* prebuilt = (row_prebuilt_t*) innobase_prebuilt; + + /* Warning: since it is not sure that MySQL calls external_lock + before calling this function, the trx field in prebuilt can be + obsolete! */ + + switch (operation) { + case HA_EXTRA_RESET: + case HA_EXTRA_RESET_STATE: + prebuilt->read_just_key = 0; + break; + case HA_EXTRA_NO_KEYREAD: + prebuilt->read_just_key = 0; + break; + case HA_EXTRA_DONT_USE_CURSOR_TO_UPDATE: + prebuilt->in_update_remember_pos = FALSE; + break; + case HA_EXTRA_KEYREAD: + prebuilt->read_just_key = 1; + break; + default:/* Do nothing */ + ; + } + + return(0); +} + +/********************************************************************** +????????????? */ + +int +ha_innobase::reset(void) +/*====================*/ +{ + return(0); +} + /********************************************************************** As MySQL will execute an external lock for every new table it uses when it starts to process an SQL statement, we can use this function to store the @@ -3496,5 +3497,4 @@ ha_innobase::get_auto_increment() return(nr); } - #endif /* HAVE_INNOBASE_DB */ diff --git a/sql/ha_innodb.h b/sql/ha_innodb.h index f926f303b26..faae42575b5 100644 --- a/sql/ha_innodb.h +++ b/sql/ha_innodb.h @@ -178,8 +178,9 @@ extern long innobase_force_recovery, innobase_thread_concurrency; extern char *innobase_data_home_dir, *innobase_data_file_path; extern char *innobase_log_group_home_dir, *innobase_log_arch_dir; extern char *innobase_unix_file_flush_method; -extern bool innobase_flush_log_at_trx_commit, innobase_log_archive, - innobase_use_native_aio, innobase_fast_shutdown; +/* The following variables have to be my_bool for SHOW VARIABLES to work */ +extern my_bool innobase_flush_log_at_trx_commit, innobase_log_archive, + innobase_use_native_aio, innobase_fast_shutdown; extern TYPELIB innobase_lock_typelib; diff --git a/sql/handler.cc b/sql/handler.cc index 507b77c977b..8447797442c 100644 --- a/sql/handler.cc +++ b/sql/handler.cc @@ -222,28 +222,31 @@ int ha_autocommit_or_rollback(THD *thd, int error) DBUG_RETURN(error); } -/* This function is called when MySQL writes the log segment of a transaction -to the binlog. It is called when the LOCK_log mutex is reserved. Here we -communicate to transactional table handlers whta binlog position corresponds -to the current transaction. The handler can store it and in recovery print -to the user, so that the user knows from what position in the binlog to -start possible roll-forward, for example, if the crashed server was a slave -in replication. This function also calls the commit of the table handler, -because the order of trasnactions in the log of the table handler must be -the same as in the binlog. */ - -int ha_report_binlog_offset_and_commit( - THD *thd, /* in: user thread */ - char *log_file_name, /* in: latest binlog file name */ - my_off_t end_offset) /* in: the offset in the binlog file - up to which we wrote */ +/* + This function is called when MySQL writes the log segment of a + transaction to the binlog. It is called when the LOCK_log mutex is + reserved. Here we communicate to transactional table handlers whta + binlog position corresponds to the current transaction. The handler + can store it and in recovery print to the user, so that the user + knows from what position in the binlog to start possible + roll-forward, for example, if the crashed server was a slave in + replication. This function also calls the commit of the table + handler, because the order of trasnactions in the log of the table + handler must be the same as in the binlog. + + arguments: + log_file_name: latest binlog file name + end_offset: the offset in the binlog file up to which we wrote +*/ + +int ha_report_binlog_offset_and_commit(THD *thd, + char *log_file_name, + my_off_t end_offset) { + int error= 0; +#ifdef HAVE_INNOBASE_DB THD_TRANS *trans; - int error = 0; - trans = &thd->transaction.all; - -#ifdef HAVE_INNOBASE_DB if (trans->innobase_tid) { if ((error=innobase_report_binlog_offset_and_commit(thd, @@ -257,10 +260,10 @@ int ha_report_binlog_offset_and_commit( trans->innodb_active_trans=0; } #endif - return error; } + int ha_commit_trans(THD *thd, THD_TRANS* trans) { int error=0; diff --git a/sql/my_lock.c b/sql/my_lock.c index 9b4ac502e57..4d451fcff22 100644 --- a/sql/my_lock.c +++ b/sql/my_lock.c @@ -26,10 +26,6 @@ #include <thr_alarm.h> #include <errno.h> -#ifdef HAVE_FCNTL -static struct flock lock; /* Must be static for sun-sparc */ -#endif - /* Lock a part of a file */ int my_lock(File fd,int locktype,my_off_t start,my_off_t length,myf MyFlags) @@ -37,24 +33,25 @@ int my_lock(File fd,int locktype,my_off_t start,my_off_t length,myf MyFlags) thr_alarm_t alarmed; ALARM alarm_buff; uint wait_for_alarm; + struct flock m_lock; DBUG_ENTER("my_lock"); DBUG_PRINT("my",("Fd: %d Op: %d start: %ld Length: %ld MyFlags: %d", fd,locktype,(ulong) start,(ulong) length,MyFlags)); if (my_disable_locking) DBUG_RETURN(0); /* purecov: inspected */ - lock.l_type=(short) locktype; - lock.l_whence=0L; - lock.l_start=(long) start; - lock.l_len=(long) length; + m_lock.l_type=(short) locktype; + m_lock.l_whence=0L; + m_lock.l_start=(long) start; + m_lock.l_len=(long) length; wait_for_alarm=(MyFlags & MY_DONT_WAIT ? MY_HOW_OFTEN_TO_ALARM : (uint) 12*60*60); - if (fcntl(fd,F_SETLK,&lock) != -1) /* Check if we can lock */ + if (fcntl(fd,F_SETLK,&m_lock) != -1) /* Check if we can lock */ DBUG_RETURN(0); /* Ok, file locked */ DBUG_PRINT("info",("Was locked, trying with alarm")); if (!thr_alarm(&alarmed,wait_for_alarm,&alarm_buff)) { int value; - while ((value=fcntl(fd,F_SETLKW,&lock)) && !thr_got_alarm(&alarmed) && + while ((value=fcntl(fd,F_SETLKW,&m_lock)) && !thr_got_alarm(&alarmed) && errno == EINTR) ; thr_end_alarm(&alarmed); if (value != -1) diff --git a/sql/mysqld.cc b/sql/mysqld.cc index d30d285d0b8..d1d3d6072d6 100644 --- a/sql/mysqld.cc +++ b/sql/mysqld.cc @@ -3309,10 +3309,11 @@ static void use_help(void) static void usage(void) { print_version(); - puts("Copyright (C) 2000 MySQL AB & MySQL Finland AB, by Monty and others"); - puts("This software comes with ABSOLUTELY NO WARRANTY. This is free software,"); - puts("and you are welcome to modify and redistribute it under the GPL license\n"); - puts("Starts the MySQL server\n"); + puts("\ +Copyright (C) 2000 MySQL AB, by Monty and others\n\ +This software comes with ABSOLUTELY NO WARRANTY. This is free software,\n\ +and you are welcome to modify and redistribute it under the GPL license\n\ +Starts the MySQL server\n"); printf("Usage: %s [OPTIONS]\n", my_progname); puts("\n\ diff --git a/sql/sql_select.cc b/sql/sql_select.cc index e2e5e4bf55a..69cac90e2be 100644 --- a/sql/sql_select.cc +++ b/sql/sql_select.cc @@ -5962,10 +5962,10 @@ SORT_FIELD *make_unireg_sortorder(ORDER *order, uint *length) /***************************************************************************** -** Fill join cache with packed records -** Records are stored in tab->cache.buffer and last record in -** last record is stored with pointers to blobs to support very big -** records + Fill join cache with packed records + Records are stored in tab->cache.buffer and last record in + last record is stored with pointers to blobs to support very big + records ******************************************************************************/ static int @@ -6027,7 +6027,7 @@ join_init_cache(THD *thd,JOIN_TAB *tables,uint table_count) if (null_fields && tables[i].table->null_fields) { /* must copy null bits */ copy->str=(char*) tables[i].table->null_flags; - copy->length=(tables[i].table->null_fields+7)/8; + copy->length=tables[i].table->null_bytes; copy->strip=0; copy->blob_field=0; length+=copy->length; diff --git a/sql/sql_table.cc b/sql/sql_table.cc index 0335c6f49cf..8013afa194b 100644 --- a/sql/sql_table.cc +++ b/sql/sql_table.cc @@ -825,7 +825,7 @@ bool close_cached_table(THD *thd,TABLE *table) /* Mark all tables that are in use as 'old' */ mysql_lock_abort(thd,table); // end threads waiting on lock -#ifdef REMOVE_LOCKS +#if defined(USING_TRANSACTIONS) || defined( __WIN__) || defined( __EMX__) || !defined(OS2) /* Wait until all there are no other threads that has this table open */ while (remove_table_from_cache(thd,table->table_cache_key, table->table_name)) @@ -1674,20 +1674,29 @@ int mysql_alter_table(THD *thd,char *new_db, char *new_name, } } -#if defined( __WIN__) || defined( __EMX__) || defined( OS2) - // Win32 can't rename an open table, so we must close the org table! - table_name=thd->strdup(table_name); // must be saved - if (close_cached_table(thd,table)) - { // Aborted - VOID(quick_rm_table(new_db_type,new_db,tmp_name)); - VOID(pthread_mutex_unlock(&LOCK_open)); - goto err; +#if (!defined( __WIN__) && !defined( __EMX__) && !defined( OS2)) + if (table->file->has_transactions()) +#endif + { + /* + Win32 and InnoDB can't drop a table that is in use, so we must + close all the original table at before doing the rename + */ + table_name=thd->strdup(table_name); // must be saved + if (close_cached_table(thd,table)) + { // Aborted + VOID(quick_rm_table(new_db_type,new_db,tmp_name)); + VOID(pthread_mutex_unlock(&LOCK_open)); + goto err; + } + table=0; // Marker that table is closed } - table=0; // Marker for win32 version -#else - table->file->extra(HA_EXTRA_FORCE_REOPEN); // Don't use this file anymore +#if (!defined( __WIN__) && !defined( __EMX__) && !defined( OS2)) + else + table->file->extra(HA_EXTRA_FORCE_REOPEN); // Don't use this file anymore #endif + error=0; if (mysql_rename_table(old_db_type,db,table_name,db,old_name)) { |