- developers - lists.mariadb.org

Re: [Maria-developers] [Commits] b21e7af: MDEV-9969 mysql_install_db error processing ignore_db_dirs.
by Sergei Golubchik 16 Jun '16

16 Jun '16

Hi, Alexey! On Jun 16, Alexey Botchkov wrote: > revision-id: b21e7af20ed4e694376b6a2a246af9dd481d7bb0 (mariadb-10.0.25-7-gb21e7af) > parent(s): 0e50b924820d365bcc1659eec2cad606f6887597 > committer: Alexey Botchkov > timestamp: 2016-06-16 13:41:45 +0400 > message: > > MDEV-9969 mysql_install_db error processing ignore_db_dirs. > > Changes to the mysql_install_db scripts so they don't repeat > arguments twice. ok to push Regards, Sergei Chief Architect MariaDB and security(a)mariadb.org

1 0

[Maria-developers] Parallel replication slave stops with duplicate key errors
by Sergey Petrunia 16 Jun '16

16 Jun '16

Hi Kristian, I was trying out parallel replication and got an error that I am not sure what to do with. == Setup == I started a master with binlog_format=row, I ran sysbench there to load the dataset ("prepare" mode, about 10M rows IIRC) then I restarted the master and ran sysbench in "run" mode a few times. Then I restarted the master again an ran a "CREATE TABLE" statement manually. Just so that I have third binlog. The master state is: mysql> show binary logs; +-------------+-----------+ | Log_name | File_size | +-------------+-----------+ | pslp.000001 | 194918772 | | pslp.000002 | 28895350 | | pslp.000003 | 589 | +-------------+-----------+ == Non-parallel test run == Then I connected with a slave, with all default settings. The slave had empty datadir, so I made it a slave with this command: CHANGE MASTER TO MASTER_HOST='127.0.0.1', MASTER_PORT=3310, MASTER_USER='root'; All server settings were pretty much default. It took a slave one hour to catch up, but it did so, and encountered no errors. Ok, this was serial replication == Parallel test run == Then I killed the slave, deleted its data directory, put an empty directory instead I added this to slave.cnf file: slave_parallel_threads=4 slave_parallel_mode=aggressive and started the slave. It ran in parallel for a few minutes, I saw this: mysql> show processlist; +----+-------------+-----------+------+---------+------+-----------------------------------------------+------------------+----------+ | Id | User | Host | db | Command | Time | State | Info | Progress | +----+-------------+-----------+------+---------+------+-----------------------------------------------+------------------+----------+ | 3 | root | localhost | test | Query | 0 | init | show processlist | 0.000 | | 4 | system user | | NULL | Connect | 15 | Waiting for master to send event | NULL | 0.000 | | 5 | system user | | NULL | Connect | 0 | Write_rows_log_event::write_row(-1) | NULL | 0.000 | | 6 | system user | | NULL | Connect | 0 | Waiting for work from SQL thread | NULL | 0.000 | | 7 | system user | | NULL | Connect | 0 | Write_rows_log_event::write_row(-1) | NULL | 0.000 | | 8 | system user | | NULL | Connect | 0 | Waiting for work from SQL thread | NULL | 0.000 | | 9 | system user | | NULL | Connect | 14 | Waiting for room in worker thread event queue | NULL | 0.000 | +----+-------------+-----------+------+---------+------+-----------------------------------------------+------------------+----------+ 7 rows in set (0.00 sec) but then it stopped. The error log has this: 2016-06-15 15:09:58 140702586379008 [Note] Slave SQL thread initialized, starting replication in log 'FIRST' at position 0, relay log './psergey-desktop-relay-bin.000001' position: 4 2016-06-15 15:09:59 140702586985216 [Note] InnoDB: Online DDL : Start 2016-06-15 15:09:59 140702586985216 [Note] InnoDB: Online DDL : Start reading clustered index of the table and create temporary files 2016-06-15 15:09:59 140702586985216 [Note] InnoDB: Online DDL : End of reading clustered index of the table and create temporary files 2016-06-15 15:09:59 140702586985216 [Note] InnoDB: Online DDL : Start applying row log 2016-06-15 15:09:59 140702586985216 [Note] InnoDB: Online DDL : End of applying row log 2016-06-15 15:09:59 140702586985216 [Note] InnoDB: Online DDL : Completed 2016-06-15 15:11:18 140702586985216 [ERROR] Slave SQL: Error during XID COMMIT: failed to update GTID state in mysql.gtid_slave_pos: 1062: Duplicate entry '0-1021' for key 'PRIMARY', Gtid 0-1-1021, Internal MariaDB error code: 1942 2016-06-15 15:11:18 140702586985216 [ERROR] Slave (additional info): Duplicate entry '0-1021' for key 'PRIMARY' Error_code: 1062 2016-06-15 15:11:18 140702586985216 [Warning] Slave: Duplicate entry '0-1021' for key 'PRIMARY' Error_code: 1062 2016-06-15 15:11:18 140702586985216 [ERROR] Error running query, slave SQL thread aborted. Fix the problem, and restart the slave SQL thread with "SLAVE START". We stopped at log 'pslp.000002' position 8266 2016-06-15 15:11:18 140702586682112 [ERROR] Slave (additional info): Commit failed due to failure of an earlier commit on which this one depends Error_code: 1964 2016-06-15 15:11:18 140702586682112 [Warning] Slave: Duplicate entry '821827' for key 'PRIMARY' Error_code: 1062 2016-06-15 15:11:18 140702586682112 [Warning] Slave: Deadlock found when trying to get lock; try restarting transaction Error_code: 1213 2016-06-15 15:11:18 140702586682112 [Warning] Slave: Commit failed due to failure of an earlier commit on which this one depends Error_code: 1964 2016-06-15 15:11:18 140702586682112 [ERROR] Error running query, slave SQL thread aborted. Fix the problem, and restart the slave SQL thread with "SLAVE START". We stopped at log 'pslp.000002' position 8266 2016-06-15 15:11:18 140702602623744 [ERROR] Slave (additional info): Commit failed due to failure of an earlier commit on which this one depends Error_code: 1964 2016-06-15 15:11:18 140702602623744 [Warning] Slave: Commit failed due to failure of an earlier commit on which this one depends Error_code: 1964 2016-06-15 15:11:18 140702602623744 [ERROR] Error running query, slave SQL thread aborted. Fix the problem, and restart the slave SQL thread with "SLAVE START". We stopped at log 'pslp.000002' position 8266 2016-06-15 15:11:18 140702602926848 [ERROR] Slave (additional info): Commit failed due to failure of an earlier commit on which this one depends Error_code: 1964 2016-06-15 15:11:18 140702602926848 [Warning] Slave: Commit failed due to failure of an earlier commit on which this one depends Error_code: 1964 2016-06-15 15:11:18 140702602926848 [ERROR] Error running query, slave SQL thread aborted. Fix the problem, and restart the slave SQL thread with "SLAVE START". We stopped at log 'pslp.000002' position 8266 2016-06-15 15:11:18 140702586985216 [ERROR] Slave (additional info): Commit failed due to failure of an earlier commit on which this one depends Error_code: 1964 2016-06-15 15:11:18 140702586985216 [Warning] Slave: Duplicate entry '0-1021' for key 'PRIMARY' Error_code: 1062 2016-06-15 15:11:18 140702586985216 [Warning] Slave: Commit failed due to failure of an earlier commit on which this one depends Error_code: 1964 2016-06-15 15:11:18 140702586985216 [ERROR] Error running query, slave SQL thread aborted. Fix the problem, and restart the slave SQL thread with "SLAVE START". We stopped at log 'pslp.000002' position 8266 2016-06-15 15:11:18 140702586682112 [ERROR] Slave (additional info): Commit failed due to failure of an earlier commit on which this one depends Error_code: 1964 2016-06-15 15:11:18 140702586682112 [Warning] Slave: Duplicate entry '821827' for key 'PRIMARY' Error_code: 1062 ... and there were lots of these errors. Eventually it stopped with: 2016-06-15 15:11:22 140702586682112 [Warning] Slave: Commit failed due to failure of an earlier commit on which this one depends Error_code: 1964 2016-06-15 15:11:22 140702586682112 [Warning] Slave: Commit failed due to failure of an earlier commit on which this one depends Error_code: 1964 2016-06-15 15:11:22 140702586682112 [Warning] Slave: Commit failed due to failure of an earlier commit on which this one depends Error_code: 1964 2016-06-15 15:11:22 140702586682112 [ERROR] Error running query, slave SQL thread aborted. Fix the problem, and restart the slave SQL thread with "SLAVE START". We stopped at log 'pslp.000002' position 8266 2016-06-15 15:11:22 140702586379008 [Note] Slave SQL thread exiting, replication stopped in log 'pslp.000002' at position 8266 The status is: mysql> show slave status\G *************************** 1. row *************************** Slave_IO_State: Waiting for master to send event Master_Host: 127.0.0.1 Master_User: root Master_Port: 3310 Connect_Retry: 60 Master_Log_File: pslp.000003 Read_Master_Log_Pos: 589 Relay_Log_File: psergey-desktop-relay-bin.000003 Relay_Log_Pos: 8549 Relay_Master_Log_File: pslp.000002 Slave_IO_Running: Yes Slave_SQL_Running: No Replicate_Do_DB: Replicate_Ignore_DB: Replicate_Do_Table: Replicate_Ignore_Table: Replicate_Wild_Do_Table: Replicate_Wild_Ignore_Table: Last_Errno: 1942 Last_Error: Error during XID COMMIT: failed to update GTID state in mysql.gtid_slave_pos: 1062: Duplicate entry '0-1021' for key 'PRIMARY' Skip_Counter: 0 Exec_Master_Log_Pos: 8266 Relay_Log_Space: 223815659 Until_Condition: None Until_Log_File: Until_Log_Pos: 0 Master_SSL_Allowed: No Master_SSL_CA_File: Master_SSL_CA_Path: Master_SSL_Cert: Master_SSL_Cipher: Master_SSL_Key: Seconds_Behind_Master: NULL Master_SSL_Verify_Server_Cert: No Last_IO_Errno: 0 Last_IO_Error: Last_SQL_Errno: 1942 Last_SQL_Error: Error during XID COMMIT: failed to update GTID state in mysql.gtid_slave_pos: 1062: Duplicate entry '0-1021' for key 'PRIMARY' Replicate_Ignore_Server_Ids: Master_Server_Id: 1 Master_SSL_Crl: Master_SSL_Crlpath: Using_Gtid: No Gtid_IO_Pos: Replicate_Do_Domain_Ids: Replicate_Ignore_Domain_Ids: Parallel_Mode: aggressive Can you advise, what does this error mean? (is there any benefit in cc'ing this to maria-developers@ ? if yes, feel free to..) BR Sergei -- Sergei Petrunia, Software Developer MariaDB Corporation | Skype: sergefp | Blog: http://s.petrunia.net/blog

2 1

Re: [Maria-developers] [Commits] 607e22e: MDEV-8633: information_schema.index_statistics doesn't delete
by Sergei Golubchik 15 Jun '16

15 Jun '16

Hi, Jan! On Apr 26, Jan Lindström wrote: > revision-id: 607e22e1426ef3d7e11e3a2e9d7f884d36cf3573 (mariadb-10.0.24-43-g607e22e) > parent(s): ce38adddfa91949b30956abd0e4cab201effcdef > committer: Jan Lindström > timestamp: 2016-04-26 13:47:30 +0300 > message: > > MDEV-8633: information_schema.index_statistics doesn't delete > item when drop table indexes or drop table; > > Problem was that table and index statistics is removed from > persistent tables but not from memory cache. Added functions > to remove table and index statistics from memory cache. That's of course, ok. Just minor comments, see below: > diff --git a/sql/sql_show.cc b/sql/sql_show.cc > index b8926b9..ac4503d 100644 > --- a/sql/sql_show.cc > +++ b/sql/sql_show.cc > @@ -3442,6 +3442,100 @@ int fill_schema_table_stats(THD *thd, TABLE_LIST *tables, COND *cond) > DBUG_RETURN(0); > } > > +/* Remove all indexes from global index statistics */ "all indexes for a given table" > +static > +int del_global_index_stats(THD *thd, uchar* cache_key, uint cache_key_length) > +{ > + int res = 0; > + DBUG_ENTER("del_global_index_stats"); > + > + mysql_mutex_lock(&LOCK_global_index_stats); > + > + for (uint i= 0; i < global_index_stats.records && !res;) > + { > + INDEX_STATS *index_stats = > + (INDEX_STATS*) my_hash_element(&global_index_stats, i); > + > + /* We search correct db\0table_name\0 string */ > + if (index_stats && > + index_stats->index_name_length >= cache_key_length && > + !memcmp(index_stats->index, cache_key, cache_key_length)) > + { > + res= my_hash_delete(&global_index_stats, (uchar*)index_stats); > + /* > + In our HASH implementation on deletion one elements > + is moved into a place where a deleted element was, > + and the last element is moved into the empty space. > + Thus we need to re-examine the current element, but > + we don't have to restart the search from the beginning. > + */ > + } > + else > + i++; > + } > + > + mysql_mutex_unlock(&LOCK_global_index_stats); > + DBUG_RETURN(res); > +} > + > +/* Remove a table from global table statistics */ > + > +int del_global_table_stat(THD *thd, LEX_STRING *db, LEX_STRING *table) > +{ > + TABLE_STATS *table_stats; > + int res = 0; > + uchar *cache_key; > + uint cache_key_length; > + DBUG_ENTER("del_global_TABLE_stat"); correct letter case: "del_global_table_stat". > + > + cache_key_length= db->length + 1 + table->length + 1; > + > + if(!(cache_key= (uchar *)my_malloc(cache_key_length, > + MYF(MY_WME | MY_ZEROFILL)))) > + { > + /* Out of memory error already given */ > + res = 1; > + goto end; > + } > + > + memcpy(cache_key, db->str, db->length); > + memcpy(cache_key + db->length + 1, table->str, table->length); > + > + res= del_global_index_stats(thd, cache_key, cache_key_length); > + > + mysql_mutex_lock(&LOCK_global_table_stats); > + > + if(!res && (table_stats= (TABLE_STATS*) my_hash_search(&global_table_stats, here and below you, probably, don't want if (!res) - if the table is dropped, you should delete as many its entries as possible, even if something fails for an unknown reason. > + cache_key, > + cache_key_length))) > + res= my_hash_delete(&global_table_stats, (uchar*)table_stats); > + > + my_free(cache_key); > + mysql_mutex_unlock(&LOCK_global_table_stats); > + > +end: > + DBUG_RETURN(res); > +} > + > +/* Remove a index from global index statistics */ > + > +int del_global_index_stat(THD *thd, TABLE* table, KEY* key_info) The difference is del_global_index_stat and del_global_index_stats is rather subtle. Could you rename them to make it clearer? For example, you can rename the static one to del_global_index_stats_for_table. > +{ > + INDEX_STATS *index_stats; > + uint key_length= table->s->table_cache_key.length + key_info->name_length + 1; > + int res = 0; > + DBUG_ENTER("del_global_index_stat"); > + mysql_mutex_lock(&LOCK_global_index_stats); > + > + if((index_stats= (INDEX_STATS*) my_hash_search(&global_index_stats, > + key_info->cache_name, > + key_length))) > + res= my_hash_delete(&global_index_stats, (uchar*)index_stats); > + > + mysql_mutex_unlock(&LOCK_global_index_stats); > + DBUG_RETURN(res); > +} > > /* Fill information schema table with index statistics */ > > diff --git a/sql/sql_statistics.cc b/sql/sql_statistics.cc > index e86c840..644ecf2 100644 > --- a/sql/sql_statistics.cc > +++ b/sql/sql_statistics.cc > @@ -29,6 +29,7 @@ > #include "sql_statistics.h" > #include "opt_range.h" > #include "my_atomic.h" > +#include "sql_show.h" > > /* > The system variable 'use_stat_tables' can take one of the > @@ -3193,6 +3194,9 @@ int delete_statistics_for_table(THD *thd, LEX_STRING *db, LEX_STRING *tab) > rc= 1; > } > > + if (!rc) again here and below, better to do it without if (!rc) > + rc= del_global_table_stat(thd, db, tab); > + > thd->restore_stmt_binlog_format(save_binlog_format); > > close_system_tables(thd, &open_tables_backup); > @@ -3339,6 +3343,9 @@ int delete_statistics_for_index(THD *thd, TABLE *tab, KEY *key_info, > } > } > > + if (!rc) > + rc= del_global_index_stat(thd, tab, key_info); > + > thd->restore_stmt_binlog_format(save_binlog_format); > Regards, Sergei Chief Architect MariaDB and security(a)mariadb.org

1 0

Re: [Maria-developers] Sachin weekly report
by Sergei Golubchik 14 Jun '16

14 Jun '16

Hi, Sachin! Sounds good. Please continue with tests. It's difficult to see what is working and what isn't without them... What do you plan to do this week? Row comparison? Hidden columns? On Jun 14, Sachin Setia wrote: > Hello Sergei, > Weekly Report for third week of gsoc > > 1. Improved code as suggested by you. > 2.Update worked. > 3.Tried with with tests but ./mtr failed , so tried some bug fixes now it > is working. > 4.Tried unique index for text , and trying for varchar length > file-> > max_key_part_length() > > > Regards > sachin > Regards, Sergei Chief Architect MariaDB and security(a)mariadb.org

1 0

Re: [Maria-developers] GSoC weekly reports (Unique indexes for blobs)
by Shubham Barai 14 Jun '16

14 Jun '16

GSoC (week 2) Hello everyone, Finally, I solved the issue with a mismatch of a key number between sql layer and MyISAM file.I also found a simpler approach for this with the help of my mentor. One of the other issues that I had to deal with was compatibility issue of key definitions between .frm file and information stored in MyISAM storage engine. The issue was due to check_defination which compares .frm file and MyISAM definitions of a table.The previous prototype of check_definition didn't contain mi_uniquedef. I have modified this function to compare mi_uniquedef.The main problem was due to the difference in a number of keys which we pass as an argument in ha_myisam::open. I have also started working on InnoDB.I discussed some of the implementation approaches with my mentor. I made changes to some of the functions in ha_innobase.cc. Regards, Shubham On 31 May 2016 at 03:32, Sergei Golubchik <serg(a)mariadb.org> wrote: > Hi, Shubham! > > Good, thanks! > > On May 31, Shubham Barai wrote: > > GSoC (week 1) > > > > Hello everyone, > > > > After developing a prototype for MyISAM ,there were some issues to be > > resolved. One of the issues is related to error regarding a duplicate > > entry. So if there is a duplicate entry for a particular key in a row, it > > generates an error for another key. > > > > After debugging, I found out that it was due to mismatch of a key number > > between sql layer and MyISAM file. Keys are stored in the order of > user's > > input in sql layer. In table2myisam function, this order is changed as > each > > key is either stored as mi_keydef or mi_uniquedef. In mi_create, for each > > unique definition, an extra key is created in the form of mi_keydef and > > stored in MYI file which creates the issue of mismatch. > > > > I tried to solve the issue by adding extra member sql_key_no to mi_keydef > > and mi_uniquedef.I also modified mi_keydef_write and mi_keydef_read to > > support a new member but after compiling, I got some error related to a > key > > file. Also with the help of my mentor, I found out that it could result > in > > compatibility issues related to old MyISAM file. > > > > So currently, I am trying to solve this using another approach. > > Regards, > Sergei > Chief Architect MariaDB > and security(a)mariadb.org >

2 2

[Maria-developers] Please review MDEV-10181 Illegal mix of collation for a field and an ASCII string as a view field
by Alexander Barkov 10 Jun '16

10 Jun '16

Hi Sergei, Please review a patch for MDEV-10181. Thanks!

2 1

[Maria-developers] Myisam-repair-threads question
by Alexandre hadjinlian guerra 10 Jun '16

10 Jun '16

Hi Im currently doing a reload process to an empty myisam table with 4 indexes on Maria db 10 1 13 on win64 and noted that on the last step that the status reports that its using repair to rebuild the table. Myisam-repair-threads remain the default but noted that many threads were running .im not sure if all of them were for repair. Is there any recommendation which i should take using the referred setting with a greater value?would setting it to zero make maridb decide which is there best value? Would the indexes get too much fragmented ? Thanks!

2 1

Re: [Maria-developers] [Commits] 9a03e78: MDEV-5535: Cannot reopen temporary table
by Sergei Golubchik 08 Jun '16

08 Jun '16

Hi, Nirbhay! On Jun 08, Nirbhay Choubey wrote: > revision-id: 9a03e780f179b575dc0d7be8de0c1731adf8726e (mariadb-10.2.0-82-g9a03e78) > parent(s): 3826f59972f19791754c7ef542e9e4a9beb67004 > author: Nirbhay Choubey > committer: Nirbhay Choubey > timestamp: 2016-06-08 10:09:42 -0400 > message: > > MDEV-5535: Cannot reopen temporary table > > Fix for some test failures and clean up. > > diff --git a/sql/temporary_tables.cc b/sql/temporary_tables.cc > index acb9507..e62cc13 100644 > --- a/sql/temporary_tables.cc > +++ b/sql/temporary_tables.cc > @@ -30,6 +30,9 @@ > #define IS_USER_TABLE(A) ((A->tmp_table == TRANSACTIONAL_TMP_TABLE) || \ > (A->tmp_table == NON_TRANSACTIONAL_TMP_TABLE)) > > +#define HAS_TEMPORARY_TABLES ((!rgi_slave && has_temporary_tables()) || \ > + (rgi_slave && \ > + unlikely(has_slave_temporary_tables()))) Ok, now I'm confused. This begs for a function, like if (rgi_slave) return has_temporary_tables(); else return has_temporary_tables(); but we had it and have just split it because you always called it with a constant true or false, never with rgi_slave. So, a function is still needed? Or was it not needed before but is needed now? Oh, I see... It was bool Temporary_tables::is_empty(bool check_slave) { DBUG_ENTER("Temporary_tables::is_empty"); bool result; if (!m_thd) DBUG_RETURN(true); rpl_group_info *rgi_slave= m_thd->rgi_slave; if (check_slave && rgi_slave) result= (rgi_slave->rli->save_temp_table_shares == NULL) ? true : false; else result= (m_table_shares == NULL) ? true : false; DBUG_RETURN(result); } so you should've split it into: bool has_temporary_tables() { return m_table_shares; } bool has_slave_temporary_tables() { return rgi_slave ? rgi_slave->rli->save_temp_table_shares : has_temporary_tables(); } because your 'check_slave' was always true or always false, so you should've split this function in two. one for check_slave=true, and one for check_slave=false; > /** > Check whether temporary tables exist. The decision is made based on the > @@ -117,7 +117,7 @@ TABLE *THD::find_temporary_table(const char *db, > uint key_length; > bool locked; > > - if (!(has_temporary_tables() || (rgi_slave && has_slave_temporary_tables()))) > + if (!HAS_TEMPORARY_TABLES) > { > DBUG_RETURN(NULL); > } Regards, Sergei Chief Architect MariaDB and security(a)mariadb.org

1 0

[Maria-developers] Unsubscribe
by Abdul Hakeem 08 Jun '16

08 Jun '16

1 0

Re: [Maria-developers] 56a2835: MDEV-5535: Cannot reopen temporary table
by Sergei Golubchik 08 Jun '16

08 Jun '16

Hi, Nirbhay! Just a couple of questions/comments, see below. And please pay attention to Kristian's email about locking and parallel replication. I don't think I can confidently review it in a patch, so you may want to spend more time testing or analyzing that code. Ok to push, when you're ready! Thanks! On May 26, Nirbhay Choubey wrote: > revision-id: 56a2835872c4ac7296ec0ae2ff618822855b0fc0 (mariadb-10.1.8-82-g56a2835) > parent(s): 28c289626f318631d707f85b057a90af99018b06 > author: Nirbhay Choubey > committer: Nirbhay Choubey > timestamp: 2016-05-26 20:42:47 -0400 > message: > > MDEV-5535: Cannot reopen temporary table > > mysqld maintains a list of TABLE objects for all temporary > tables in THD, where each table is represented by a TABLE > object. A query referencing a temporary table for more than > once, however, failed with ER_CANT_REOPEN_TABLE error. "because a TABLE_SHARE was allocate together with the TABLE, so temporary tables always had one TABLE per TABLE_SHARE" > This patch lift this restriction by preserving the TABLE_SHARE > object of each created temporary table, so that multiple instances > of TABLE objects can be created. better "by separating TABLE and TABLE_SHARE objects and storing TABLE_SHARE's in the list in a THD, and TABLE's in the list in the corresponding TABLE_SHARE" > diff --git a/sql/rpl_rli.cc b/sql/rpl_rli.cc > index 7bdd9c1..77f458b 100644 > --- a/sql/rpl_rli.cc > +++ b/sql/rpl_rli.cc > @@ -1060,24 +1061,43 @@ void Relay_log_info::inc_group_relay_log_pos(ulonglong log_pos, > > void Relay_log_info::close_temporary_tables() > { > - TABLE *table,*next; > DBUG_ENTER("Relay_log_info::close_temporary_tables"); > > - for (table=save_temporary_tables ; table ; table=next) > + TMP_TABLE_SHARE *share; > + TABLE *table; > + > + while ((share= save_temporary_tables.pop_front())) > { > - next=table->next; > + /* > + Iterate over the list of tables for this TABLE_SHARE and close them. > + */ > + while ((table= share->all_tmp_tables.pop_front())) > + { > + DBUG_PRINT("tmptable", ("closing table: '%s'.'%s'", > + table->s->db.str, table->s->table_name.str)); > + > + /* Reset in_use as the table may have been created by another thd */ > + table->in_use= 0; > + free_io_cache(table); > + /* > + Lets not free TABLE_SHARE here as there could be multiple TABLEs opened > + for the same table (TABLE_SHARE). > + */ > + closefrm(table, false); > + my_free(table); > + } > > - /* Reset in_use as the table may have been created by another thd */ > - table->in_use=0; > /* > Don't ask for disk deletion. For now, anyway they will be deleted when > slave restarts, but it is a better intention to not delete them. > */ > - DBUG_PRINT("info", ("table: 0x%lx", (long) table)); > - close_temporary(table, 1, 0); > + > + free_table_share(share); > + my_free(share); > } > - save_temporary_tables= 0; > - slave_open_temp_tables= 0; > + > + save_temporary_tables.empty(); Is it needed? you've popped all shares from the list, it's empty. I'd rather add an assert instead > + > DBUG_VOID_RETURN; > } > > diff --git a/sql/sql_class.h b/sql/sql_class.h > index ae240ae..a70c4a9 100644 > --- a/sql/sql_class.h > +++ b/sql/sql_class.h > @@ -1247,6 +1247,61 @@ enum enum_locked_tables_mode > LTM_PRELOCKED_UNDER_LOCK_TABLES > }; > > +/** > + The following structure is an extension to TABLE_SHARE and is > + exclusively for temporary tables. > + > + @note: > + Although, TDC_element has data members (like next, prev & > + all_tables) to store the list of TABLE_SHARE & TABLE objects > + related to a particular TABLE_SHARE, they cannot be moved to > + TABLE_SHARE in order to be reused for temporary tables. This > + is because, as concurrent threads iterating through hash of > + TDC_element's may need access to all_tables, but if all_tables > + is made part of TABLE_SHARE, then TDC_element->share->all_tables > + is not always guaranteed to be valid, as TDC_element can live > + longer than TABLE_SHARE. > +*/ > +struct TMP_TABLE_SHARE : public TABLE_SHARE > +{ > +private: > + /* > + Link to all temporary table shares. Declared as private to > + avoid direct manipulation with those objects. One should > + use methods of I_P_List template instead. > + */ > + TMP_TABLE_SHARE *tmp_next; > + TMP_TABLE_SHARE **tmp_prev; > + > + friend struct All_tmp_table_shares; > + > +public: > + /* > + Doubly-linked (back-linked) lists of used and unused TABLE objects > + for this share. and unused? you don't free them but keep in the list? > + */ > + All_share_tables_list all_tmp_tables; > +}; > + > +/** > + Helper class which specifies which members of TMP_TABLE_SHARE are > + used for participation in the list of temporary tables. > +*/ > + > +struct All_tmp_table_shares > +{ > + static inline TMP_TABLE_SHARE **next_ptr(TMP_TABLE_SHARE *l) > + { > + return &l->tmp_next; > + } > + static inline TMP_TABLE_SHARE ***prev_ptr(TMP_TABLE_SHARE *l) > + { > + return &l->tmp_prev; > + } > +}; > + > +/* Also used in rpl_rli.h. */ > +typedef I_P_List <TMP_TABLE_SHARE, All_tmp_table_shares> All_tmp_tables_list; > > /** > Class that holds information about tables which were opened and locked > diff --git a/sql/temporary_tables.cc b/sql/temporary_tables.cc > new file mode 100644 > index 0000000..1514b72 > --- /dev/null > +++ b/sql/temporary_tables.cc ... > +TABLE *THD::find_temporary_table(const char *db, > + const char *table_name) > +{ > + DBUG_ENTER("THD::find_temporary_table"); > + > + TABLE *table; > + char key[MAX_DBKEY_LENGTH]; > + uint key_length; > + bool locked; > + > + if (!(has_temporary_tables() || (rgi_slave && has_slave_temporary_tables()))) hmm, really? you had here is_empty(true) before and your is_empty() was working like this: if (flag_is_true) { do rgi_slave } else { do thd } your old is_empty was not checking thd if the argument was true, your new code does, as if your is_empty() was do thd; if (flag_is_true) { do rgi_slave} > + { > + DBUG_RETURN(NULL); > + } > + > + key_length= create_tmp_table_def_key(key, db, table_name); Regards, Sergei Chief Architect MariaDB and security(a)mariadb.org

2 1