2006-08-16 14:58:49 +02:00
SET SESSION STORAGE_ENGINE = MyISAM;
2006-01-05 23:47:49 +01:00
drop table if exists t1,t3,t4,t5;
2001-09-28 07:05:54 +02:00
create table t1 (a int, b char(10), key a(a), key b(a,b));
insert into t1 values
(17,"ddd"),(18,"eee"),(19,"fff"),(19,"yyy"),
(14,"aaa"),(15,"bbb"),(16,"ccc"),(16,"xxx"),
(20,"ggg"),(21,"hhh"),(22,"iii");
handler t1 open as t2;
2003-01-10 16:36:59 +01:00
handler t2 read a=(SELECT 1);
2004-06-16 05:18:20 +02:00
ERROR 42000: You have an error in your SQL syntax; check the manual that corresponds to your MySQL server version for the right syntax to use near 'SELECT 1)' at line 1
2001-09-28 07:05:54 +02:00
handler t2 read a first;
2001-04-07 00:18:33 +02:00
a b
14 aaa
2001-09-28 07:05:54 +02:00
handler t2 read a next;
2001-04-07 00:18:33 +02:00
a b
15 bbb
2001-09-28 07:05:54 +02:00
handler t2 read a next;
2001-04-07 00:18:33 +02:00
a b
16 ccc
2001-09-28 07:05:54 +02:00
handler t2 read a prev;
2001-04-07 00:18:33 +02:00
a b
15 bbb
2001-09-28 07:05:54 +02:00
handler t2 read a last;
2001-04-07 00:18:33 +02:00
a b
22 iii
2001-09-28 07:05:54 +02:00
handler t2 read a prev;
2001-04-07 00:18:33 +02:00
a b
21 hhh
2001-09-28 07:05:54 +02:00
handler t2 read a prev;
2001-04-07 00:18:33 +02:00
a b
20 ggg
2001-09-28 07:05:54 +02:00
handler t2 read a first;
2001-04-07 00:18:33 +02:00
a b
14 aaa
2001-09-28 07:05:54 +02:00
handler t2 read a prev;
2001-04-07 00:18:33 +02:00
a b
2001-09-28 07:05:54 +02:00
handler t2 read a last;
2001-04-07 00:18:33 +02:00
a b
22 iii
2001-09-28 07:05:54 +02:00
handler t2 read a prev;
2001-04-07 00:18:33 +02:00
a b
21 hhh
2001-09-28 07:05:54 +02:00
handler t2 read a next;
2001-04-07 00:18:33 +02:00
a b
22 iii
2001-09-28 07:05:54 +02:00
handler t2 read a next;
2001-04-07 00:18:33 +02:00
a b
2001-09-28 07:05:54 +02:00
handler t2 read a=(15);
2001-04-07 00:18:33 +02:00
a b
15 bbb
2001-09-28 07:05:54 +02:00
handler t2 read a=(16);
2001-04-15 20:14:40 +02:00
a b
16 ccc
2001-09-28 07:05:54 +02:00
handler t2 read a=(19,"fff");
2004-06-15 22:38:36 +02:00
ERROR 42000: Too many key parts specified; max 1 parts allowed
2001-09-28 07:05:54 +02:00
handler t2 read b=(19,"fff");
2001-04-15 20:14:40 +02:00
a b
19 fff
2001-09-28 07:05:54 +02:00
handler t2 read b=(19,"yyy");
2001-04-15 20:14:40 +02:00
a b
19 yyy
2001-09-28 07:05:54 +02:00
handler t2 read b=(19);
2001-04-15 20:14:40 +02:00
a b
19 fff
2001-09-28 07:05:54 +02:00
handler t1 read a last;
2003-06-04 17:28:51 +02:00
ERROR 42S02: Unknown table 't1' in HANDLER
2001-09-28 07:05:54 +02:00
handler t2 read a=(11);
2001-04-15 20:14:40 +02:00
a b
2001-09-28 07:05:54 +02:00
handler t2 read a>=(11);
2001-04-15 20:14:40 +02:00
a b
14 aaa
2001-09-28 07:05:54 +02:00
handler t2 read a=(18);
2001-04-15 20:14:40 +02:00
a b
18 eee
2001-09-28 07:05:54 +02:00
handler t2 read a>=(18);
2001-04-15 20:14:40 +02:00
a b
18 eee
2001-09-28 07:05:54 +02:00
handler t2 read a>(18);
2001-04-15 20:14:40 +02:00
a b
19 fff
2001-09-28 07:05:54 +02:00
handler t2 read a<=(18);
2001-04-15 20:14:40 +02:00
a b
18 eee
2001-09-28 07:05:54 +02:00
handler t2 read a<(18);
2001-04-15 20:14:40 +02:00
a b
17 ddd
2001-09-28 07:05:54 +02:00
handler t2 read a first limit 5;
2001-04-15 20:14:40 +02:00
a b
14 aaa
15 bbb
16 ccc
16 xxx
17 ddd
2001-09-28 07:05:54 +02:00
handler t2 read a next limit 3;
2001-04-15 20:14:40 +02:00
a b
18 eee
19 fff
19 yyy
2001-09-28 07:05:54 +02:00
handler t2 read a prev limit 10;
2001-04-15 20:14:40 +02:00
a b
19 fff
18 eee
17 ddd
16 xxx
16 ccc
15 bbb
14 aaa
2001-09-28 07:05:54 +02:00
handler t2 read a>=(16) limit 4;
2001-04-15 20:14:40 +02:00
a b
16 ccc
16 xxx
17 ddd
18 eee
2001-09-28 07:05:54 +02:00
handler t2 read a>=(16) limit 2,2;
2001-04-15 20:14:40 +02:00
a b
17 ddd
18 eee
2001-09-28 07:05:54 +02:00
handler t2 read a last limit 3;
2001-04-15 20:14:40 +02:00
a b
22 iii
21 hhh
20 ggg
2001-09-28 07:05:54 +02:00
handler t2 read a=(19);
2001-04-15 20:14:40 +02:00
a b
19 fff
2001-09-28 07:05:54 +02:00
handler t2 read a=(19) where b="yyy";
2001-04-15 20:14:40 +02:00
a b
19 yyy
2001-09-28 07:05:54 +02:00
handler t2 read first;
2001-04-15 20:14:40 +02:00
a b
17 ddd
2001-09-28 07:05:54 +02:00
handler t2 read next;
2001-04-15 20:14:40 +02:00
a b
18 eee
2001-09-28 07:05:54 +02:00
handler t2 read next;
2001-04-15 20:14:40 +02:00
a b
19 fff
2001-09-28 07:05:54 +02:00
handler t2 read last;
2004-06-16 05:18:20 +02:00
ERROR 42000: You have an error in your SQL syntax; check the manual that corresponds to your MySQL server version for the right syntax to use near '' at line 1
2001-09-28 07:05:54 +02:00
handler t2 close;
2006-08-16 14:58:49 +02:00
handler t1 open;
handler t1 read a next;
a b
14 aaa
handler t1 read a next;
a b
15 bbb
handler t1 close;
handler t1 open;
handler t1 read a prev;
a b
22 iii
handler t1 read a prev;
a b
21 hhh
handler t1 close;
handler t1 open as t2;
handler t2 read first;
a b
17 ddd
alter table t1 engine = MyISAM;
handler t2 read first;
ERROR 42S02: Unknown table 't2' in HANDLER
2003-01-28 14:36:22 +01:00
handler t1 open as t2;
drop table t1;
create table t1 (a int);
insert into t1 values (17);
handler t2 read first;
2003-06-04 17:28:51 +02:00
ERROR 42S02: Unknown table 't2' in HANDLER
2003-02-22 18:22:39 +01:00
handler t1 open as t2;
2006-08-16 19:29:49 +02:00
alter table t1 engine=MEMORY;
2003-02-22 18:22:39 +01:00
handler t2 read first;
2003-06-04 17:28:51 +02:00
ERROR 42S02: Unknown table 't2' in HANDLER
2003-01-28 14:36:22 +01:00
drop table t1;
2003-07-04 11:41:01 +02:00
create table t1 (a int);
insert into t1 values (1),(2),(3),(4),(5),(6);
delete from t1 limit 2;
handler t1 open;
handler t1 read first;
a
3
handler t1 read first limit 1,1;
a
4
handler t1 read first limit 2,2;
a
5
6
delete from t1 limit 3;
handler t1 read first;
a
6
drop table t1;
2003-07-03 17:57:51 +02:00
create table t1(a int, index(a));
insert into t1 values (1), (2), (3);
handler t1 open;
handler t1 read a=(W);
2003-08-18 23:08:08 +02:00
ERROR 42S22: Unknown column 'W' in 'field list'
2003-07-04 12:55:25 +02:00
handler t1 read a=(a);
2004-06-15 22:38:36 +02:00
ERROR HY000: Incorrect arguments to HANDLER ... READ
2003-07-03 17:57:51 +02:00
drop table t1;
2004-01-13 12:31:25 +01:00
create table t1 (a char(5));
insert into t1 values ("Ok");
handler t1 open as t;
handler t read first;
a
Ok
use mysql;
handler t read first;
a
Ok
handler t close;
handler test.t1 open as t;
handler t read first;
a
Ok
handler t close;
use test;
drop table t1;
2004-05-18 20:59:43 +02:00
create table t1 ( a int, b int, INDEX a (a) );
insert into t1 values (1,2), (2,1);
handler t1 open;
handler t1 read a=(1) where b=2;
a b
1 2
handler t1 read a=(1) where b=3;
a b
handler t1 read a=(1) where b=1;
a b
handler t1 close;
drop table t1;
2004-09-24 18:39:25 +02:00
drop database if exists test_test;
create database test_test;
use test_test;
create table t1(table_id char(20) primary key);
insert into t1 values ('test_test.t1');
insert into t1 values ('');
handler t1 open;
handler t1 read first limit 9;
table_id
test_test.t1
create table t2(table_id char(20) primary key);
insert into t2 values ('test_test.t2');
insert into t2 values ('');
handler t2 open;
handler t2 read first limit 9;
table_id
test_test.t2
use test;
drop table if exists t1;
create table t1(table_id char(20) primary key);
insert into t1 values ('test.t1');
insert into t1 values ('');
handler t1 open;
2004-10-07 09:50:13 +02:00
ERROR 42000: Not unique table/alias: 't1'
2004-09-24 18:39:25 +02:00
use test;
handler test.t1 read first limit 9;
2004-10-07 09:50:13 +02:00
ERROR 42000: You have an error in your SQL syntax; check the manual that corresponds to your MySQL server version for the right syntax to use near 'read first limit 9' at line 1
2004-09-24 18:39:25 +02:00
handler test_test.t1 read first limit 9;
2004-10-07 09:50:13 +02:00
ERROR 42000: You have an error in your SQL syntax; check the manual that corresponds to your MySQL server version for the right syntax to use near 'read first limit 9' at line 1
2004-09-24 18:39:25 +02:00
handler t1 read first limit 9;
table_id
test_test.t1
handler test_test.t2 read first limit 9;
2004-10-07 09:50:13 +02:00
ERROR 42000: You have an error in your SQL syntax; check the manual that corresponds to your MySQL server version for the right syntax to use near 'read first limit 9' at line 1
2004-09-24 18:39:25 +02:00
handler t2 read first limit 9;
table_id
test_test.t2
handler test_test.t1 close;
2004-10-07 09:50:13 +02:00
ERROR 42000: You have an error in your SQL syntax; check the manual that corresponds to your MySQL server version for the right syntax to use near 'close' at line 1
2004-09-24 18:39:25 +02:00
handler t1 close;
drop table test_test.t1;
handler test_test.t2 close;
2004-10-07 09:50:13 +02:00
ERROR 42000: You have an error in your SQL syntax; check the manual that corresponds to your MySQL server version for the right syntax to use near 'close' at line 1
2004-09-24 18:39:25 +02:00
handler t2 close;
drop table test_test.t2;
drop database test_test;
use test;
handler test.t1 close;
2004-10-07 09:50:13 +02:00
ERROR 42000: You have an error in your SQL syntax; check the manual that corresponds to your MySQL server version for the right syntax to use near 'close' at line 1
2004-09-24 18:39:25 +02:00
handler t1 close;
2004-10-07 09:50:13 +02:00
ERROR 42S02: Unknown table 't1' in HANDLER
2004-09-24 18:39:25 +02:00
drop table test.t1;
drop database if exists test_test;
drop table if exists t1;
drop table if exists t2;
drop table if exists t3;
create database test_test;
use test_test;
create table t1 (c1 char(20));
insert into t1 values ('test_test.t1');
create table t3 (c1 char(20));
insert into t3 values ('test_test.t3');
handler t1 open;
handler t1 read first limit 9;
c1
test_test.t1
handler t1 open h1;
handler h1 read first limit 9;
c1
test_test.t1
use test;
create table t1 (c1 char(20));
create table t2 (c1 char(20));
create table t3 (c1 char(20));
insert into t1 values ('t1');
insert into t2 values ('t2');
insert into t3 values ('t3');
handler t1 open;
2004-10-07 09:50:13 +02:00
ERROR 42000: Not unique table/alias: 't1'
2004-09-24 18:39:25 +02:00
handler t2 open t1;
2004-10-07 09:50:13 +02:00
ERROR 42000: Not unique table/alias: 't1'
2004-09-24 18:39:25 +02:00
handler t3 open t1;
2004-10-07 09:50:13 +02:00
ERROR 42000: Not unique table/alias: 't1'
2004-09-24 18:39:25 +02:00
handler t1 read first limit 9;
c1
test_test.t1
handler test.t1 close;
2004-10-07 09:50:13 +02:00
ERROR 42000: You have an error in your SQL syntax; check the manual that corresponds to your MySQL server version for the right syntax to use near 'close' at line 1
2004-09-24 18:39:25 +02:00
handler test.t1 open h1;
2004-10-07 09:50:13 +02:00
ERROR 42000: Not unique table/alias: 'h1'
2004-09-24 18:39:25 +02:00
handler test_test.t1 open h1;
2004-10-07 09:50:13 +02:00
ERROR 42000: Not unique table/alias: 'h1'
2004-09-24 18:39:25 +02:00
handler test_test.t3 open h3;
handler test.t1 open h2;
handler t1 read first limit 9;
c1
test_test.t1
handler h1 read first limit 9;
c1
test_test.t1
handler h2 read first limit 9;
c1
t1
handler h3 read first limit 9;
c1
test_test.t3
2004-10-07 09:50:13 +02:00
handler h2 read first limit 9;
2004-09-24 18:39:25 +02:00
c1
t1
handler test.h1 close;
2004-10-07 09:50:13 +02:00
ERROR 42000: You have an error in your SQL syntax; check the manual that corresponds to your MySQL server version for the right syntax to use near 'close' at line 1
handler t1 close;
handler h1 close;
2004-09-24 18:39:25 +02:00
handler h2 close;
handler t1 read first limit 9;
2004-10-07 09:50:13 +02:00
ERROR 42S02: Unknown table 't1' in HANDLER
2004-09-24 18:39:25 +02:00
handler h1 read first limit 9;
2004-10-07 09:50:13 +02:00
ERROR 42S02: Unknown table 'h1' in HANDLER
2004-09-24 18:39:25 +02:00
handler h2 read first limit 9;
2004-10-07 09:50:13 +02:00
ERROR 42S02: Unknown table 'h2' in HANDLER
2004-09-24 18:39:25 +02:00
handler h3 read first limit 9;
c1
test_test.t3
2004-10-07 09:50:13 +02:00
handler h3 read first limit 9;
2004-09-24 18:39:25 +02:00
c1
test_test.t3
use test_test;
handler h3 read first limit 9;
c1
test_test.t3
handler test.h3 read first limit 9;
2004-10-07 09:50:13 +02:00
ERROR 42000: You have an error in your SQL syntax; check the manual that corresponds to your MySQL server version for the right syntax to use near 'read first limit 9' at line 1
handler h3 close;
2004-09-24 18:39:25 +02:00
use test;
drop table t3;
drop table t2;
drop table t1;
drop database test_test;
create table t1 (c1 char(20));
insert into t1 values ("t1");
handler t1 open as h1;
handler h1 read first limit 9;
c1
t1
create table t2 (c1 char(20));
insert into t2 values ("t2");
handler t2 open as h2;
handler h2 read first limit 9;
c1
t2
create table t3 (c1 char(20));
insert into t3 values ("t3");
handler t3 open as h3;
handler h3 read first limit 9;
c1
t3
create table t4 (c1 char(20));
insert into t4 values ("t4");
handler t4 open as h4;
handler h4 read first limit 9;
c1
t4
create table t5 (c1 char(20));
insert into t5 values ("t5");
handler t5 open as h5;
handler h5 read first limit 9;
c1
t5
alter table t1 engine=MyISAM;
handler h1 read first limit 9;
2004-10-07 09:50:13 +02:00
ERROR 42S02: Unknown table 'h1' in HANDLER
2004-09-24 18:39:25 +02:00
handler h2 read first limit 9;
c1
t2
handler h3 read first limit 9;
c1
t3
handler h4 read first limit 9;
c1
t4
handler h5 read first limit 9;
c1
t5
alter table t5 engine=MyISAM;
handler h1 read first limit 9;
2004-10-07 09:50:13 +02:00
ERROR 42S02: Unknown table 'h1' in HANDLER
2004-09-24 18:39:25 +02:00
handler h2 read first limit 9;
c1
t2
handler h3 read first limit 9;
c1
t3
handler h4 read first limit 9;
c1
t4
handler h5 read first limit 9;
2004-10-07 09:50:13 +02:00
ERROR 42S02: Unknown table 'h5' in HANDLER
2004-09-24 18:39:25 +02:00
alter table t3 engine=MyISAM;
handler h1 read first limit 9;
2004-10-07 09:50:13 +02:00
ERROR 42S02: Unknown table 'h1' in HANDLER
2004-09-24 18:39:25 +02:00
handler h2 read first limit 9;
c1
t2
handler h3 read first limit 9;
2004-10-07 09:50:13 +02:00
ERROR 42S02: Unknown table 'h3' in HANDLER
2004-09-24 18:39:25 +02:00
handler h4 read first limit 9;
c1
t4
handler h5 read first limit 9;
2004-10-07 09:50:13 +02:00
ERROR 42S02: Unknown table 'h5' in HANDLER
2004-09-24 18:39:25 +02:00
handler h2 close;
handler h4 close;
handler t1 open as h1_1;
handler t1 open as h1_2;
handler t1 open as h1_3;
handler h1_1 read first limit 9;
c1
t1
handler h1_2 read first limit 9;
c1
t1
handler h1_3 read first limit 9;
c1
t1
alter table t1 engine=MyISAM;
handler h1_1 read first limit 9;
2004-10-07 09:50:13 +02:00
ERROR 42S02: Unknown table 'h1_1' in HANDLER
2004-09-24 18:39:25 +02:00
handler h1_2 read first limit 9;
2004-10-07 09:50:13 +02:00
ERROR 42S02: Unknown table 'h1_2' in HANDLER
2004-09-24 18:39:25 +02:00
handler h1_3 read first limit 9;
2004-10-07 09:50:13 +02:00
ERROR 42S02: Unknown table 'h1_3' in HANDLER
2004-09-24 18:39:25 +02:00
drop table t1;
drop table t2;
drop table t3;
drop table t4;
drop table t5;
2005-11-15 21:57:02 +01:00
create table t1 (c1 int);
insert into t1 values (1);
handler t1 open;
handler t1 read first;
c1
1
send the below to another connection, do not wait for the result
Update tests and result files after running with new mysqltest that better detects problems with test files
mysql-test/r/csv.result:
Update after add of missing semicolon
mysql-test/r/drop.result:
Update result file, no space before commands that has been "sent"
mysql-test/r/flush.result:
Update result file, no space before commands that has been "sent"
mysql-test/r/flush_block_commit.result:
Update result file, no space before commands that has been "sent"
mysql-test/r/flush_read_lock_kill.result:
Update result file, no space before commands that has been "sent"
mysql-test/r/grant2.result:
Update result file, no space before commands that has been "sent"
mysql-test/r/handler.result:
Update result file, no space before commands that has been "sent"
mysql-test/r/innodb_notembedded.result:
Update result file, no space before commands that has been "sent"
mysql-test/r/kill.result:
Update result file, no space before commands that has been "sent"
mysql-test/r/lock_multi.result:
Update result file, no space before commands that has been "sent"
mysql-test/r/multi_update.result:
Update result file, no space before commands that has been "sent"
mysql-test/r/mysqltest.result:
Update result
mysql-test/r/query_cache.result:
Update after add of missing semicolon
mysql-test/r/query_cache_notembedded.result:
Update result file, no space before commands that has been "sent"
mysql-test/r/sp-threads.result:
Update result file, no space before commands that has been "sent"
mysql-test/r/sp_notembedded.result:
Update after add of missing semicolon
mysql-test/r/type_blob.result:
Remove extra drop table
mysql-test/t/csv.test:
Add missing semicolon
mysql-test/t/query_cache.test:
Add missing semicolon
mysql-test/t/sp-error.test:
Remove "tab" from end of error declaration
mysql-test/t/sp.test:
Wrong delimiter, used ; instead of |
mysql-test/t/sp_notembedded.test:
Wrong delimiter, used ; instead of |
mysql-test/t/view_grant.test:
An incomplete error name specification was used.
2006-10-04 13:09:37 +02:00
optimize table t1;
2005-11-15 21:57:02 +01:00
proceed with the normal connection
handler t1 read next;
c1
1
handler t1 close;
read the result from the other connection
Table Op Msg_type Msg_text
test.t1 optimize status OK
proceed with the normal connection
drop table t1;
2006-08-16 14:58:49 +02:00
CREATE TABLE t1 ( no1 smallint(5) NOT NULL default '0', no2 int(10) NOT NULL default '0', PRIMARY KEY (no1,no2));
INSERT INTO t1 VALUES (1,274),(1,275),(2,6),(2,8),(4,1),(4,2);
HANDLER t1 OPEN;
HANDLER t1 READ `primary` = (1, 1000);
no1 no2
HANDLER t1 READ `primary` PREV;
no1 no2
1 275
DROP TABLE t1;
2005-11-15 21:57:02 +01:00
create table t1 (c1 int);
insert into t1 values (14397);
flush tables with read lock;
drop table t1;
ERROR HY000: Can't execute the query because you have a conflicting read lock
send the below to another connection, do not wait for the result
Update tests and result files after running with new mysqltest that better detects problems with test files
mysql-test/r/csv.result:
Update after add of missing semicolon
mysql-test/r/drop.result:
Update result file, no space before commands that has been "sent"
mysql-test/r/flush.result:
Update result file, no space before commands that has been "sent"
mysql-test/r/flush_block_commit.result:
Update result file, no space before commands that has been "sent"
mysql-test/r/flush_read_lock_kill.result:
Update result file, no space before commands that has been "sent"
mysql-test/r/grant2.result:
Update result file, no space before commands that has been "sent"
mysql-test/r/handler.result:
Update result file, no space before commands that has been "sent"
mysql-test/r/innodb_notembedded.result:
Update result file, no space before commands that has been "sent"
mysql-test/r/kill.result:
Update result file, no space before commands that has been "sent"
mysql-test/r/lock_multi.result:
Update result file, no space before commands that has been "sent"
mysql-test/r/multi_update.result:
Update result file, no space before commands that has been "sent"
mysql-test/r/mysqltest.result:
Update result
mysql-test/r/query_cache.result:
Update after add of missing semicolon
mysql-test/r/query_cache_notembedded.result:
Update result file, no space before commands that has been "sent"
mysql-test/r/sp-threads.result:
Update result file, no space before commands that has been "sent"
mysql-test/r/sp_notembedded.result:
Update after add of missing semicolon
mysql-test/r/type_blob.result:
Remove extra drop table
mysql-test/t/csv.test:
Add missing semicolon
mysql-test/t/query_cache.test:
Add missing semicolon
mysql-test/t/sp-error.test:
Remove "tab" from end of error declaration
mysql-test/t/sp.test:
Wrong delimiter, used ; instead of |
mysql-test/t/sp_notembedded.test:
Wrong delimiter, used ; instead of |
mysql-test/t/view_grant.test:
An incomplete error name specification was used.
2006-10-04 13:09:37 +02:00
drop table t1;
2005-11-15 21:57:02 +01:00
proceed with the normal connection
select * from t1;
c1
14397
unlock tables;
read the result from the other connection
proceed with the normal connection
select * from t1;
ERROR 42S02: Table 'test.t1' doesn't exist
drop table if exists t1;
Warnings:
Note 1051 Unknown table 't1'
2007-08-15 17:13:34 +02:00
drop table if exists t1;
create table t1 (a int) ENGINE=MEMORY;
--> client 2
handler t1 open;
ERROR HY000: Table storage engine for 't1' doesn't have this option
--> client 1
drop table t1;
2007-08-27 15:37:12 +02:00
drop table if exists t1;
create table t1 (a int);
handler t1 open as t1_alias;
handler t1_alias read a next;
2007-08-30 00:00:49 +02:00
ERROR 42000: Key 'a' doesn't exist in table 't1_alias'
2007-08-27 15:37:12 +02:00
handler t1_alias READ a next where inexistent > 0;
ERROR 42S22: Unknown column 'inexistent' in 'field list'
handler t1_alias read a next;
2007-08-30 00:00:49 +02:00
ERROR 42000: Key 'a' doesn't exist in table 't1_alias'
2007-08-27 15:37:12 +02:00
handler t1_alias READ a next where inexistent > 0;
ERROR 42S22: Unknown column 'inexistent' in 'field list'
handler t1_alias close;
drop table t1;
2007-10-11 00:06:53 +02:00
drop table if exists t1,t2;
create table t1 (c1 int);
create table t2 (c1 int);
insert into t1 values (1);
insert into t2 values (2);
connection: default
handler t1 open;
handler t1 read first;
c1
1
connection: flush
flush tables;;
Initial import of WL#3726 "DDL locking for all metadata objects".
Backport of:
------------------------------------------------------------
revno: 2630.4.1
committer: Dmitry Lenev <dlenev@mysql.com>
branch nick: mysql-6.0-3726-w
timestamp: Fri 2008-05-23 17:54:03 +0400
message:
WL#3726 "DDL locking for all metadata objects".
After review fixes in progress.
------------------------------------------------------------
This is the first patch in series. It transforms the metadata
locking subsystem to use a dedicated module (mdl.h,cc). No
significant changes in the locking protocol.
The import passes the test suite with the exception of
deprecated/removed 6.0 features, and MERGE tables. The latter
are subject to a fix by WL#4144.
Unfortunately, the original changeset comments got lost in a merge,
thus this import has its own (largely insufficient) comments.
This patch fixes Bug#25144 "replication / binlog with view breaks".
Warning: this patch introduces an incompatible change:
Under LOCK TABLES, it's no longer possible to FLUSH a table that
was not locked for WRITE.
Under LOCK TABLES, it's no longer possible to DROP a table or
VIEW that was not locked for WRITE.
******
Backport of:
------------------------------------------------------------
revno: 2630.4.2
committer: Dmitry Lenev <dlenev@mysql.com>
branch nick: mysql-6.0-3726-w
timestamp: Sat 2008-05-24 14:03:45 +0400
message:
WL#3726 "DDL locking for all metadata objects".
After review fixes in progress.
******
Backport of:
------------------------------------------------------------
revno: 2630.4.3
committer: Dmitry Lenev <dlenev@mysql.com>
branch nick: mysql-6.0-3726-w
timestamp: Sat 2008-05-24 14:08:51 +0400
message:
WL#3726 "DDL locking for all metadata objects"
Fixed failing Windows builds by adding mdl.cc to the lists
of files needed to build server/libmysqld on Windows.
******
Backport of:
------------------------------------------------------------
revno: 2630.4.4
committer: Dmitry Lenev <dlenev@mysql.com>
branch nick: mysql-6.0-3726-w
timestamp: Sat 2008-05-24 21:57:58 +0400
message:
WL#3726 "DDL locking for all metadata objects".
Fix for assert failures in kill.test which occured when one
tried to kill ALTER TABLE statement on merge table while it
was waiting in wait_while_table_is_used() for other connections
to close this table.
These assert failures stemmed from the fact that cleanup code
in this case assumed that temporary table representing new
version of table was open with adding to THD::temporary_tables
list while code which were opening this temporary table wasn't
always fulfilling this.
This patch changes code that opens new version of table to
always do this linking in. It also streamlines cleanup process
for cases when error occurs while we have new version of table
open.
******
WL#3726 "DDL locking for all metadata objects"
Add libmysqld/mdl.cc to .bzrignore.
******
Backport of:
------------------------------------------------------------
revno: 2630.4.6
committer: Dmitry Lenev <dlenev@mysql.com>
branch nick: mysql-6.0-3726-w
timestamp: Sun 2008-05-25 00:33:22 +0400
message:
WL#3726 "DDL locking for all metadata objects".
Addition to the fix of assert failures in kill.test caused by
changes for this worklog.
Make sure we close the new table only once.
.bzrignore:
Add libmysqld/mdl.cc
libmysqld/CMakeLists.txt:
Added mdl.cc to the list of files needed for building of libmysqld.
libmysqld/Makefile.am:
Added files implementing new meta-data locking subsystem to the server.
mysql-test/include/handler.inc:
Use separate connection for waiting while threads performing DDL
operations conflicting with open HANDLER tables reach blocked
state. This is required because now we check and close tables open
by HANDLER statements in this connection conflicting with DDL in
another each time open_tables() is called and thus select from I_S
which is used for waiting will unblock DDL operations if issued
from connection with open HANDLERs.
mysql-test/r/create.result:
Adjusted test case after change in implementation of CREATE TABLE
... SELECT. We no longer have special check in open_table() which
catches the case when we select from the table created. Instead we
rely on unique_table() call which happens after opening and
locking all tables.
mysql-test/r/flush.result:
FLUSH TABLES WITH READ LOCK can no longer happen under LOCK
TABLES. Updated test accordingly.
mysql-test/r/flush_table.result:
Under LOCK TABLES we no longer allow to do FLUSH TABLES for tables
locked for read. Updated test accordingly.
mysql-test/r/handler_innodb.result:
Use separate connection for waiting while threads performing DDL
operations conflicting with open HANDLER tables reach blocked
state. This is required because now we check and close tables open
by HANDLER statements in this connection conflicting with DDL in
another each time open_tables() is called and thus select from I_S
which is used for waiting will unblock DDL operations if issued
from connection with open HANDLERs.
mysql-test/r/handler_myisam.result:
Use separate connection for waiting while threads performing DDL
operations conflicting with open HANDLER tables reach blocked
state. This is required because now we check and close tables open
by HANDLER statements in this connection conflicting with DDL in
another each time open_tables() is called and thus select from I_S
which is used for waiting will unblock DDL operations if issued
from connection with open HANDLERs.
mysql-test/r/information_schema.result:
Additional test for WL#3726 "DDL locking for all metadata
objects". Check that we use high-priority metadata lock requests
when filling I_S tables.
Rearrange tests to match 6.0 better (fewer merge conflicts).
mysql-test/r/kill.result:
Added tests checking that DDL and DML statements waiting for
metadata locks can be interrupted by KILL command.
mysql-test/r/lock.result:
One no longer is allowed to do DROP VIEW under LOCK TABLES even if
this view is locked by LOCK TABLES. The problem is that in such
situation write locks on view are not mutually exclusive so
upgrading metadata lock which is required for dropping of view
will lead to deadlock.
mysql-test/r/partition_column_prune.result:
Update results (same results in 6.0), WL#3726
mysql-test/r/partition_pruning.result:
Update results (same results in 6.0), WL#3726
mysql-test/r/ps_ddl.result:
We no longer invalidate prepared CREATE TABLE ... SELECT statement
if target table changes. This is OK since it is not strictly
necessary.
The first change is wrong, is caused by FLUSH TABLE
now flushing all unused tables. This is a regression that
Dmitri fixed in 6.0 in a follow up patch.
mysql-test/r/sp.result:
Under LOCK TABLES we no longer allow accessing views which were
not explicitly locked. To access view we need to obtain metadata
lock on it and doing this under LOCK TABLES may lead to deadlocks.
mysql-test/r/view.result:
One no longer is allowed to do DROP VIEW under LOCK TABLES even if
this view is locked by LOCK TABLES. The problem is that in such
situation even "write locks" on view are not mutually exclusive so
upgrading metadata lock which is required for dropping of view
will lead to deadlock
mysql-test/r/view_grant.result:
ALTER VIEW implementation was changed to open a view only after
checking that user which does alter has appropriate privileges on
it. This means that in case when user's privileges are
insufficient for this we won't check that new view definer is the
same as original one or user performing alter has SUPER privilege.
Adjusted test case accordingly.
mysql-test/r/view_multi.result:
Added test case for bug#25144 "replication / binlog with view
breaks".
mysql-test/suite/rpl/t/disabled.def:
Disable test for deprecated features (they don't work with new MDL).
mysql-test/t/create.test:
Adjusted test case after change in implementation of CREATE TABLE
... SELECT. We no longer have special check in open_table() which
catches the case when we select from the table created. Instead we
rely on unique_table() call which happens after opening and
locking all tables.
mysql-test/t/disabled.def:
Disable merge.test, subject of WL#4144
mysql-test/t/flush.test:
FLUSH TABLES WITH READ LOCK can no longer happen under LOCK
TABLES. Updated test accordingly.
mysql-test/t/flush_table.test:
Under LOCK TABLES we no longer allow to do FLUSH TABLES for tables
locked for read. Updated test accordingly.
mysql-test/t/information_schema.test:
Additional test for WL#3726 "DDL locking for all metadata
objects". Check that we use high-priority metadata lock requests
when filling I_S tables.
Rearrange the results for easier merges with 6.0.
mysql-test/t/kill.test:
Added tests checking that DDL and DML statements waiting for
metadata locks can be interrupted by KILL command.
mysql-test/t/lock.test:
One no longer is allowed to do DROP VIEW under LOCK TABLES even if
this view is locked by LOCK TABLES. The problem is that in such
situation write locks on view are not mutually exclusive so
upgrading metadata lock which is required for dropping of view
will lead to deadlock.
mysql-test/t/lock_multi.test:
Adjusted test case to the changes of status in various places
caused by change in implementation FLUSH TABLES WITH READ LOCK,
which is now takes global metadata lock before flushing tables and
therefore waits on at these places.
mysql-test/t/ps_ddl.test:
We no longer invalidate prepared CREATE TABLE ... SELECT statement
if target table changes. This is OK since it is not strictly
necessary.
The first change is wrong, is caused by FLUSH TABLE
now flushing all unused tables. This is a regression that
Dmitri fixed in 6.0 in a follow up patch.
mysql-test/t/sp.test:
Under LOCK TABLES we no longer allow accessing views which were
not explicitly locked. To access view we need to obtain metadata
lock on it and doing this under LOCK TABLES may lead to deadlocks.
mysql-test/t/trigger_notembedded.test:
Adjusted test case to the changes of status in various places
caused by change in implementation FLUSH TABLES WITH READ LOCK,
which is now takes global metadata lock before flushing tables and
therefore waits on at these places.
mysql-test/t/view.test:
One no longer is allowed to do DROP VIEW under LOCK TABLES even if
this view is locked by LOCK TABLES. The problem is that in such
situation even "write locks" on view are not mutually exclusive so
upgrading metadata lock which is required for dropping of view
will lead to deadlock.
mysql-test/t/view_grant.test:
ALTER VIEW implementation was changed to open a view only after
checking that user which does alter has appropriate privileges on
it. This means that in case when user's privileges are
insufficient for this we won't check that new view definer is the
same as original one or user performing alter has SUPER privilege.
Adjusted test case accordingly.
mysql-test/t/view_multi.test:
Added test case for bug#25144 "replication / binlog with view
breaks".
sql/CMakeLists.txt:
Added mdl.cc to the list of files needed for building of server.
sql/Makefile.am:
Added files implementing new meta-data locking subsystem to the
server.
sql/event_db_repository.cc:
Allocate metadata lock requests objects (MDL_LOCK) on execution
memory root in cases when TABLE_LIST objects is also allocated
there or on stack.
sql/ha_ndbcluster.cc:
Adjusted code to work nicely with new metadata locking subsystem.
close_cached_tables() no longer has wait_for_placeholder argument.
Instead of relying on this parameter and related behavior FLUSH
TABLES WITH READ LOCK now takes global shared metadata lock.
sql/ha_ndbcluster_binlog.cc:
Adjusted code to work with new metadata locking subsystem.
close_cached_tables() no longer has wait_for_placeholder argument.
Instead of relying on this parameter and related behavior FLUSH
TABLES WITH READ LOCK now takes global shared metadata lock.
sql/handler.cc:
update_frm_version():
Directly update TABLE_SHARE::mysql_version member instead of
going through all TABLE instances for this table (old code was a
legacy from pre-table-definition-cache days).
sql/lock.cc:
Use new metadata locking subsystem. Threw away most of functions
related to name locking as now one is supposed to use metadata
locking API instead. In lock_global_read_lock() and
unlock_global_read_lock() in order to avoid problems with global
read lock sneaking in at the moment when we perform FLUSH TABLES
or ALTER TABLE under LOCK TABLES and when tables being reopened
are protected only by metadata locks we also have to take global
shared meta data lock.
sql/log_event.cc:
Adjusted code to work with new metadata locking subsystem. For
tables open by slave thread for applying RBR events allocate
memory for lock request object in the same chunk of memory as
TABLE_LIST objects for them. In order to ensure that we keep these
objects around until tables are open always close tables before
calling Relay_log_info::clear_tables_to_lock(). Use new auxiliary
Relay_log_info::slave_close_thread_tables() method to enforce
this.
sql/log_event_old.cc:
Adjusted code to work with new metadata locking subsystem. Since
for tables open by slave thread for applying RBR events memory for
lock request object is allocated in the same chunk of memory as
TABLE_LIST objects for them we have to ensure that we keep these
objects around until tables are open. To ensure this we always
close tables before calling
Relay_log_info::clear_tables_to_lock(). To enfore this we use
new auxiliary Relay_log_info::slave_close_thread_tables()
method.
sql/mdl.cc:
Implemented new metadata locking subsystem and API described in
WL3726 "DDL locking for all metadata objects".
sql/mdl.h:
Implemented new metadata locking subsystem and API described in
WL3726 "DDL locking for all metadata objects".
sql/mysql_priv.h:
- close_thread_tables()/close_tables_for_reopen() now has one more
argument which indicates that metadata locks should be released
but not removed from the context in order to be used later in
mdl_wait_for_locks() and tdc_wait_for_old_version().
- close_cached_table() routine is no longer public.
- Thread waiting in wait_while_table_is_used() can be now killed
so this function returns boolean to make caller aware of such
situation.
- We no longer have table cache as separate entity instead used
and unused TABLE instances are linked to TABLE_SHARE objects in
table definition cache.
- Now third argument of open_table() is also used for requesting
table repair or auto-discovery of table's new definition. So its
type was changed from bool to enum.
- Added tdc_open_view() function for opening view by getting its
definition from disk (and table cache in future).
- reopen_name_locked_table() no longer needs "link_in" argument as
now we have exclusive metadata locks instead of dummy TABLE
instances when this function is called.
- find_locked_table() now takes head of list of TABLE instances
instead of always scanning through THD::open_tables list. Also
added find_write_locked_table() auxiliary.
- reopen_tables(), close_cached_tables() no longer have
mark_share_as_old and wait_for_placeholder arguments. Instead of
relying on this parameters and related behavior FLUSH TABLES
WITH READ LOCK now takes global shared metadata lock.
- We no longer need drop_locked_tables() and
abort_locked_tables().
- mysql_ha_rm_tables() now always assume that LOCK_open is not
acquired by caller.
- Added notify_thread_having_shared_lock() callback invoked by
metadata locking subsystem when acquiring an exclusive lock, for
each thread that has a conflicting shared metadata lock.
- Introduced expel_table_from_cache() as replacement for
remove_table_from_cache() (the main difference is that this new
function assumes that caller follows metadata locking protocol
and never waits).
- Threw away most of functions related to name locking. One should
use new metadata locking subsystem and API instead.
sql/mysqld.cc:
Got rid of call initializing/deinitializing table cache since now
it is embedded into table definition cache. Added calls for
initializing/ deinitializing metadata locking subsystem.
sql/rpl_rli.cc:
Introduced auxiliary Relay_log_info::slave_close_thread_tables()
method which is used for enforcing that we always close tables
open for RBR before deallocating TABLE_LIST elements and MDL_LOCK
objects for them.
sql/rpl_rli.h:
Introduced auxiliary Relay_log_info::slave_close_thread_tables()
method which is used for enforcing that we always close tables
open for RBR before deallocating TABLE_LIST elements and MDL_LOCK
objects for them.
sql/set_var.cc:
close_cached_tables() no longer has wait_for_placeholder argument.
Instead of relying on this parameter and related behavior FLUSH
TABLES WITH READ LOCK now takes global shared metadata lock.
sql/sp_head.cc:
For tables added to the statement's table list by prelocking
algorithm we allocate these objects either on the same memory as
corresponding table list elements or on THD::locked_tables_root
(if we are building table list for LOCK TABLES).
sql/sql_acl.cc:
Allocate metadata lock requests objects (MDL_LOCK) on execution
memory root in cases when we use stack TABLE_LIST objects to open
tables. Got rid of redundant code by using unlock_locked_tables()
function.
sql/sql_base.cc:
Changed code to use new MDL subsystem. Got rid of separate table
cache. Now used and unused TABLE instances are linked to the
TABLE_SHAREs in table definition cache.
check_unused():
Adjusted code to the fact that we no longer have separate table
cache. Removed dead code.
table_def_free():
Free TABLE instances referenced from TABLE_SHARE objects before
destroying table definition cache.
get_table_share():
Added assert which ensures that noone will be able to access
table (and its share) without acquiring some kind of metadata
lock first.
close_handle_and_leave_table_as_lock():
Adjusted code to the fact that TABLE instances now are linked to
list in TABLE_SHARE.
list_open_tables():
Changed this function to use table definition cache instead of
table cache.
free_cache_entry():
Unlink freed TABLE elements from the list of all TABLE instances
for the table in TABLE_SHARE.
kill_delayed_thread_for_table():
Added auxiliary for killing delayed insert threads for
particular table.
close_cached_tables():
Got rid of wait_for_refresh argument as we now rely on global
shared metadata lock to prevent FLUSH WITH READ LOCK sneaking in
when we are reopening tables. Heavily reworked this function to
use new MDL code and not to rely on separate table cache entity.
close_open_tables():
We no longer have separate table cache.
close_thread_tables():
Release metadata locks after closing all tables. Added skip_mdl
argument which allows us not to remove metadata lock requests
from the context in case when we are going to use this requests
later in mdl_wait_for_locks() and tdc_wait_for_old_versions().
close_thread_table()/close_table_for_reopen():
Since we no longer have separate table cache and all TABLE
instances are linked to TABLE_SHARE objects in table definition
cache we have to link/unlink TABLE object to/from appropriate
lists in the share.
name_lock_locked_table():
Moved redundant code to find_write_locked_table() function and
adjusted code to the fact that wait_while_table_is_used() can
now return with an error if our thread is killed.
reopen_table_entry():
We no longer need "link_in" argument as with MDL we no longer
call this function with dummy TABLE object pre-allocated and
added to the THD::open_tables. Also now we add newly-open TABLE
instance to the list of share's used TABLE instances.
table_cache_insert_placeholder():
Got rid of name-locking legacy.
lock_table_name_if_not_cached():
Moved to sql_table.cc the only place where it is used. It was
also reimplemented using new MDL API.
open_table():
- Reworked this function to use new MDL subsystem.
- Changed code to deal with table definition cache directly
instead of going through separate table cache.
- Now third argument is also used for requesting table repair
or auto-discovery of table's new definition. So its type was
changed from bool to enum.
find_locked_table()/find_write_locked_table():
Accept head of list of TABLE objects as first argument and use
this list instead of always searching in THD::open_tables list.
Also added auxiliary for finding write-locked locked tables.
reopen_table():
Adjusted function to work with new MDL subsystem and to properly
manuipulate with lists of used/unused TABLE instaces in
TABLE_SHARE.
reopen_tables():
Removed mark_share_as_old parameter. Instead of relying on it
and related behavior FLUSH TABLES WITH READ LOCK now takes
global shared metadata lock. Changed code after removing
separate table cache.
drop_locked_tables()/abort_locked_tables():
Got rid of functions which are no longer needed.
unlock_locked_tables():
Moved this function from sql_parse.cc and changed it to release
memory which was used for allocating metadata lock requests for
tables open and locked by LOCK TABLES.
tdc_open_view():
Intoduced function for opening a view by getting its definition
from disk (and table cache in future).
reopen_table_entry():
Introduced function for opening table definitions while holding
exclusive metatadata lock on it.
open_unireg_entry():
Got rid of this function. Most of its functionality is relocated
to open_table() and open_table_fini() functions, and some of it
to reopen_table_entry() and tdc_open_view(). Also code
resposible for auto-repair and auto-discovery of tables was
moved to separate function.
open_table_entry_fini():
Introduced function which contains common actions which finalize
process of TABLE object creation.
auto_repair_table():
Moved code responsible for auto-repair of table being opened
here.
handle_failed_open_table_attempt()
Moved code responsible for handling failing attempt to open
table to one place (retry due to lock conflict/old version,
auto-discovery and repair).
open_tables():
- Flush open HANDLER tables if they have old version of if there
is conflicting metadata lock against them (before this moment
we had this code in open_table()).
- When we open view which should be processed via derived table
on the second execution of prepared statement or stored
routine we still should call open_table() for it in order to
obtain metadata lock on it and prepare its security context.
- In cases when we discover that some special handling of
failure to open table is needed call
handle_failed_open_table_attempt() which handles all such
scenarios.
open_ltable():
Handling of various special scenarios of failure to open a table
was moved to separate handle_failed_open_table_attempt()
function.
remove_db_from_cache():
Removed this function as it is no longer used.
notify_thread_having_shared_lock():
Added callback which is invoked by MDL subsystem when acquiring
an exclusive lock, for each thread that has a conflicting shared
metadata lock.
expel_table_from_cache():
Introduced function for removing unused TABLE instances. Unlike
remove_table_from_cache() it relies on caller following MDL
protocol and having appropriate locks when calling it and thus
does not do any waiting if table is still in use.
tdc_wait_for_old_version():
Added function which allows open_tables() to wait in cases when
we discover that we should back-off due to presence of old
version of table.
abort_and_upgrade_lock():
Use new MDL calls.
mysql_wait_completed_table():
Got rid of unused function.
open_system_tables_for_read/for_update()/performance_schema_table():
Allocate MDL_LOCK objects on execution memory root in cases when
TABLE_LIST objects for corresponding tables is allocated on
stack.
close_performance_schema_table():
Release metadata locks after closing tables.
******
Use I_P_List for free/used tables list in the table share.
sql/sql_binlog.cc:
Use Relay_log_info::slave_close_thread_tables() method to enforce
that we always close tables open for RBR before deallocating
TABLE_LIST elements and MDL_LOCK objects for them.
sql/sql_class.cc:
Added meta-data locking contexts as part of Open_tables_state
context. Also introduced THD::locked_tables_root memory root
which is to be used for allocating MDL_LOCK objects for tables in
LOCK TABLES statement (end of lifetime for such objects is UNLOCK
TABLES so we can't use statement or execution root for them).
sql/sql_class.h:
Added meta-data locking contexts as part of Open_tables_state
context. Also introduced THD::locked_tables_root memory root
which is to be used for allocating MDL_LOCK objects for tables in
LOCK TABLES statement (end of lifetime for such objects is UNLOCK
TABLES so we can't use statement or execution root for them).
Note: handler_mdl_context and locked_tables_root and
mdl_el_root will be removed by subsequent patches.
sql/sql_db.cc:
mysql_rm_db() does not really need to call remove_db_from_cache()
as it drops each table in the database using
mysql_rm_table_part2(), which performs all necessary operations on
table (definition) cache.
sql/sql_delete.cc:
Use the new metadata locking API for TRUNCATE.
sql/sql_handler.cc:
Changed HANDLER implementation to use new metadata locking
subsystem. Note that MDL_LOCK objects for HANDLER tables are
allocated in the same chunk of heap memory as TABLE_LIST object
for those tables.
sql/sql_insert.cc:
mysql_insert():
find_locked_table() now takes head of list of TABLE object as
its argument instead of always scanning through THD::open_tables
list.
handle_delayed_insert():
Allocate metadata lock request object for table open by delayed
insert thread on execution memroot. create_table_from_items():
We no longer allocate dummy TABLE objects for tables being
created if they don't exist. As consequence
reopen_name_locked_table() no longer has link_in argument.
open_table() now has one more argument which is not relevant for
temporary tables.
sql/sql_parse.cc:
- Moved unlock_locked_tables() routine to sql_base.cc and made
available it in other files. Got rid of some redundant code by
using this function.
- Replaced boolean TABLE_LIST::create member with enum
open_table_type member.
- Use special memory root for allocating MDL_LOCK objects for
tables open and locked by LOCK TABLES (these object should live
till UNLOCK TABLES so we can't allocate them on statement nor
execution memory root). Also properly set metadata lock
upgradability attribure for those tables.
- Under LOCK TABLES it is no longer allowed to flush tables which
are not write-locked as this breaks metadata locking protocol
and thus potentially might lead to deadlock.
- Added auxiliary adjust_mdl_locks_upgradability() function.
sql/sql_partition.cc:
Adjusted code to the fact that reopen_tables() no longer has
"mark_share_as_old" argument. Got rid of comments which are no
longer true.
sql/sql_plist.h:
Added I_P_List template class for parametrized intrusive doubly
linked lists and I_P_List_iterator for corresponding iterator.
Unlike for I_List<> list elements of such list can participate in
several lists. Unlike List<> such lists are doubly-linked and
intrusive.
sql/sql_plugin.cc:
Allocate metadata lock requests objects (MDL_LOCK) on execution
memory root in cases when we use stack TABLE_LIST objects to open
tables.
sql/sql_prepare.cc:
Replaced boolean TABLE_LIST::create member with enum
open_table_type member. This allows easily handle situation in
which instead of opening the table we want only to take exclusive
metadata lock on it.
sql/sql_rename.cc:
Use new metadata locking subsystem in implementation of RENAME
TABLE.
sql/sql_servers.cc:
Allocate metadata lock requests objects (MDL_LOCK) on execution
memory root in cases when we use stack TABLE_LIST objects to open
tables. Got rid of redundant code by using unlock_locked_tables()
function.
sql/sql_show.cc:
Acquire shared metadata lock when we are getting information for
I_S table directly from TABLE_SHARE without doing full-blown table
open. We use high priority lock request in this situation in
order to avoid deadlocks.
Also allocate metadata lock requests objects (MDL_LOCK) on
execution memory root in cases when TABLE_LIST objects are also
allocated there
sql/sql_table.cc:
mysql_rm_table():
Removed comment which is no longer relevant.
mysql_rm_table_part2():
Now caller of mysql_ha_rm_tables() should not own LOCK_open.
Adjusted code to use new metadata locking subsystem instead of
name-locks.
lock_table_name_if_not_cached():
Moved this function from sql_base.cc to this file and
reimplemented it using metadata locking API.
mysql_create_table():
Adjusted code to use new MDL API.
wait_while_table_is_used():
Changed function to use new MDL subsystem. Made thread waiting
in it killable (this also led to introduction of return value so
caller can distinguish successful executions from situations
when waiting was aborted).
close_cached_tables():
Thread waiting in this function is killable now. As result it
has return value for distinguishing between succes and failure.
Got rid of redundant boradcast_refresh() call.
prepare_for_repair():
Use MDL subsystem instead of name-locks.
mysql_admin_table():
mysql_ha_rm_tables() now always assumes that caller doesn't own
LOCK_open.
mysql_repair_table():
We should mark all elements of table list as requiring
upgradable metadata locks.
mysql_create_table_like():
Use new MDL subsystem instead of name-locks.
create_temporary_tables():
We don't need to obtain metadata locks when creating temporary
table.
mysql_fast_or_online_alter_table():
Thread waiting in wait_while_table_is_used() is now killable.
mysql_alter_table():
Adjusted code to work with new MDL subsystem and to the fact
that threads waiting in what_while_table_is_used() and
close_cached_table() are now killable.
sql/sql_test.cc:
We no longer have separate table cache. TABLE instances are now
associated with/linked to TABLE_SHARE objects in table definition
cache.
sql/sql_trigger.cc:
Adjusted code to work with new metadata locking subsystem. Also
reopen_tables() no longer has mark_share_as_old argument (Instead
of relying on this parameter and related behavior FLUSH TABLES
WITH READ LOCK now takes global shared metadata lock).
sql/sql_udf.cc:
Allocate metadata lock requests objects (MDL_LOCK) on execution
memory root in cases when we use stack TABLE_LIST objects to open
tables.
sql/sql_update.cc:
Adjusted code to work with new meta-data locking subsystem.
sql/sql_view.cc:
Added proper meta-data locking to implementations of
CREATE/ALTER/DROP VIEW statements. Now we obtain exclusive
meta-data lock on a view before creating/ changing/dropping it.
This ensures that all concurrent statements that use this view
will finish before our statement will proceed and therefore we
will get correct order of statements in the binary log.
Also ensure that TABLE_LIST::mdl_upgradable attribute is properly
propagated for underlying tables of view.
sql/table.cc:
Added auxiliary alloc_mdl_locks() function for allocating metadata
lock request objects for all elements of table list.
sql/table.h:
TABLE_SHARE:
Got rid of unused members. Introduced members for storing lists
of used and unused TABLE objects for this share.
TABLE:
Added members for linking TABLE objects into per-share lists of
used and unused TABLE instances. Added member for holding
pointer to metadata lock for this table.
TABLE_LIST:
Replaced boolean TABLE_LIST::create member with enum
open_table_type member. This allows easily handle situation in
which instead of opening the table we want only to take
exclusive meta-data lock on it (we need this in order to handle
ALTER VIEW and CREATE VIEW statements).
Introduced new mdl_upgradable member for marking elements of
table list for which we need to take upgradable shared metadata
lock instead of plain shared metadata lock. Added pointer for
holding pointer to MDL_LOCK for the table.
Added auxiliary alloc_mdl_locks() function for allocating metadata
lock requests objects for all elements of table list. Added
auxiliary set_all_mdl_upgradable() function for marking all
elements in table list as requiring upgradable metadata locks.
storage/myisammrg/ha_myisammrg.cc:
Allocate MDL_LOCK objects for underlying tables of MERGE table.
To be reworked once Ingo pushes his patch for WL4144.
2009-11-30 16:55:03 +01:00
connection: waiter
2007-10-11 00:06:53 +02:00
connection: default
handler t2 open;
handler t2 read first;
c1
2
handler t1 read next;
c1
2007-10-11 02:39:22 +02:00
1
2007-10-11 00:06:53 +02:00
handler t1 close;
handler t2 close;
drop table t1,t2;
Patch that changes metadata locking subsystem to use mutex per lock and
condition variable per context instead of one mutex and one conditional
variable for the whole subsystem.
This should increase concurrency in this subsystem.
It also opens the way for further changes which are necessary to solve
such bugs as bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock"
and bug #37346 "innodb does not detect deadlock between update and alter
table".
Two other notable changes done by this patch:
- MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object metadata lock is acquired. Now this has
to be done by explicit calls outside of MDL subsystem.
- Instead of using separate MDL_context for opening system tables/tables
for purposes of I_S we now create MDL savepoint in the main context
before opening tables and rollback to this savepoint after closing
them. This means that it is now possible to get ER_LOCK_DEADLOCK error
even not inside a transaction. This might happen in unlikely case when
one runs DDL on one of system tables while also running DDL on some
other tables. Cases when this ER_LOCK_DEADLOCK error is not justified
will be addressed by advanced deadlock detector for MDL subsystem which
we plan to implement.
mysql-test/include/handler.inc:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/handler_innodb.result:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/handler_myisam.result:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/mdl_sync.result:
Adjusted mdl_sync.test to the fact that exclusive metadata locks on
tables are now acquired according to alphabetical order of fully
qualified table names instead of order in which tables are mentioned
in statement.
mysql-test/t/mdl_sync.test:
Adjusted mdl_sync.test to the fact that exclusive metadata locks on
tables are now acquired according to alphabetical order of fully
qualified table names instead of order in which tables are mentioned
in statement.
sql/events.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/ha_ndbcluster.cc:
Since manipulations with open table state no longer install proxy
MDL_context it does not make sense to perform them in order to
satisfy assert in mysql_rm_tables_part2(). Removed them per agreement
with Cluster team. This has not broken test suite since scenario in
which deadlock can occur and assertion fails is not covered by tests.
sql/lock.cc:
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object exclusive metadata lock is acquired.
Now this has to be done by explicit calls outside of MDL subsystem.
sql/log.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/mdl.cc:
Changed metadata locking subsystem to use mutex per lock and condition
variable per context instead of one mutex and one conditional variable
for the whole subsystem.
Changed approach to handling of global metadata locks. Instead of
implicitly acquiring intention exclusive locks when user requests
per-object upgradeable or exclusive locks now we require them to be
acquired explicitly in the same way as ordinary metadata locks.
In fact global lock are now ordinary metadata locks in new GLOBAL
namespace.
To implement these changes:
- Removed LOCK_mdl mutex and COND_mdl condition variable.
- Introduced MDL_lock::m_mutex mutexes which protect individual lock
objects.
- Replaced mdl_locks hash with MDL_map class, which has hash for
MDL_lock objects as a member and separate mutex which protects this
hash. Methods of this class allow to find(), find_or_create() or
remove() MDL_lock objects in concurrency-friendly fashion (i.e.
for most common operation, find_or_create(), we don't acquire
MDL_lock::m_mutex while holding MDL_map::m_mutex. Thanks to MikaelR
for this idea and benchmarks!). Added three auxiliary members to
MDL_lock class (m_is_destroyed, m_ref_usage, m_ref_release) to
support this concurrency-friendly behavior.
- Introduced MDL_context::m_ctx_wakeup_cond condition variable to be
used for waiting until this context's pending request can be
satisfied or its thread has to perform actions to resolve potential
deadlock. Context which want to wait add ticket corresponding to the
request to an appropriate queue of waiters in MDL_lock object so
they can be noticed when other contexts change state of lock and be
awaken by them by signalling on MDL_context::m_ctx_wakeup_cond.
As consequence MDL_ticket objects has to be used for any waiting
in metadata locking subsystem including one which happens in
MDL_context::wait_for_locks() method.
Another consequence is that MDL_context is no longer copyable and
can't be saved/restored when working with system tables.
- Made MDL_lock an abstract class, which delegates specifying exact
compatibility matrix to its descendants. Added MDL_global_lock child
class for global lock (The old is_lock_type_compatible() method
became can_grant_lock() method of this class). Added MDL_object_lock
class to represent per-object lock (The old MDL_lock::can_grant_lock()
became its method). Choice between two classes happens based on MDL
namespace in MDL_lock::create() method.
- Got rid of MDL_lock::type member as its meaning became ambigous for
global locks.
- To simplify waking up of contexts waiting for lock split waiting queue
in MDL_lock class in two queues. One for pending requests for exclusive
(including intention exclusive) locks and another for requests for
shared locks.
- Added virtual wake_up_waiters() method to MDL_lock, MDL_global_lock and
MDL_object_lock classes which allows to wake up waiting contexts after
state of lock changes. Replaced old duplicated code with calls to this
method.
- Adjusted MDL_context::try_acquire_shared_lock()/exclusive_lock()/
global_shared_lock(), MDL_ticket::upgrade_shared_lock_to_exclusive_lock()
and MDL_context::release_ticket() methods to use MDL_map and
MDL_lock::m_mutex instead of single LOCK_mdl mutex and wake up
waiters according to the approach described above. The latter method
also was renamed to MDL_context::release_lock().
- Changed MDL_context::try_acquire_shared_lock()/exclusive_lock() and
release_lock() not to handle global locks. They are now supposed to
be taken explicitly like ordinary metadata locks.
- Added helper MDL_context::try_acquire_global_intention_exclusive_lock()
and acquire_global_intention_exclusive_lock() methods.
- Moved common code from MDL_context::acquire_global_shared_lock() and
acquire_global_intention_exclusive_lock() to new method -
MDL_context::acquire_lock_impl().
- Moved common code from MDL_context::try_acquire_shared_lock(),
try_acquire_global_intention_exclusive_lock()/exclusive_lock()
to MDL_context::try_acquire_lock_impl().
- Since acquiring of several exclusive locks can no longer happen under
single LOCK_mdl mutex the approach to it had to be changed. Now we do
it in one by one fashion. This is done in alphabetical order to avoid
deadlocks. Changed MDL_context::acquire_exclusive_locks() accordingly
(as part of this change moved code responsible for acquiring single
exclusive lock to new MDL_context::acquire_exclusive_lock_impl()
method).
- Since we no longer have single LOCK_mdl mutex which protects all
MDL_context::m_is_waiting_in_mdl members using these members to
determine if we have really awaken context holding conflicting
shared lock became inconvinient. Got rid of this member and changed
notify_shared_lock() helper function and process of acquiring
of/upgrading to exclusive lock not to rely on such information.
Now in MDL_context::acquire_exclusive_lock_impl() and
MDL_ticket::upgrade_shared_lock_to_exclusive_lock() we simply
re-try to wake up threads holding conflicting shared locks after
small time out.
- Adjusted MDL_context::can_wait_lead_to_deadlock() and
MDL_ticket::has_pending_conflicting_lock() to use per-lock
mutexes instead of LOCK_mdl. To do this introduced
MDL_lock::has_pending_exclusive_lock() method.
sql/mdl.h:
Changed metadata locking subsystem to use mutex per lock and condition
variable per context instead of one mutex and one conditional variable
for the whole subsystem. In order to implement this change:
- Added MDL_key::cmp() method to be able to sort MDL_key objects
alphabetically. Changed length fields in MDL_key class to uint16
as 16-bit is enough for length of any key.
- Changed MDL_ticket::get_ctx() to return pointer to non-const
object in order to be able to use MDL_context::awake() method
for such contexts.
- Got rid of unlocked versions of can_wait_lead_to_deadlock()/
has_pending_conflicting_lock() methods in MDL_context and
MDL_ticket. We no longer has single mutex which protects all
locks. Thus one always has to use versions of these methods
which acquire per-lock mutexes.
- MDL_request_list type of list now counts its elements.
- Added MDL_context::m_ctx_wakeup_cond condition variable to be used
for waiting until this context's pending request can be satisfied
or its thread has to perform actions to resolve potential deadlock.
Added awake() method to wake up context from such wait.
Addition of condition variable made MDL_context uncopyable.
As result we no longer can save/restore MDL_context when working
with system tables. Instead we create MDL savepoint before opening
those tables and rollback to it once they are closed.
- MDL_context::release_ticket() became release_lock() method.
- Added auxiliary MDL_context::acquire_exclusive_lock_impl() method
which does all necessary work to acquire exclusive lock on one object
but should not be used directly as it does not enforce any asserts
ensuring that no deadlocks are possible.
- Since we no longer need to know if thread trying to acquire exclusive
lock managed to wake up any threads having conflicting shared locks
(as, anyway, we will try to wake up such threads again shortly)
- MDL_context::m_is_waiting_in_mdl member became unnecessary and
notify_shared_lock() no longer needs to be friend of MDL_context.
Changed approach to handling of global metadata locks. Instead of
implicitly acquiring intention exclusive locks when user requests
per-object upgradeable or exclusive locks now we require them to be
acquired explicitly in the same way as ordinary metadata locks.
- Added new GLOBAL namespace for such locks.
- Added new type of lock to be requested MDL_INTENTION_EXCLISIVE.
- Added MDL_context::try_acquire_global_intention_exclusive_lock()
and acquire_global_intention_exclusive_lock() methods.
- Moved common code from MDL_context::acquire_global_shared_lock()
and acquire_global_intention_exclusive_lock() to new method -
MDL_context::acquire_lock_impl().
- Moved common code from MDL_context::try_acquire_shared_lock(),
try_acquire_global_intention_exclusive_lock()/exclusive_lock()
to MDL_context::try_acquire_lock_impl().
- Added helper MDL_context::is_global_lock_owner() method to be
able easily to find what kind of global lock this context holds.
- MDL_context::m_has_global_shared_lock became unnecessary as
global read lock is now represented by ordinary ticket.
- Removed assert in MDL_context::set_lt_or_ha_sentinel() which became
false for cases when we execute LOCK TABLES under global read lock
mode.
sql/mysql_priv.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result calls
opening/closing system tables were changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sp.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sp.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sql_base.cc:
close_thread_tables():
Since we no longer use separate MDL_context for opening system
tables we need to avoid releasing all transaction locks when
closing system table. Releasing metadata lock on system table
is now responsibility of THD::restore_backup_open_tables_state().
open_table_get_mdl_lock(),
Open_table_context::recover_from_failed_open():
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object upgradable or exclusive metadata lock is
acquired. So this have to be done explicitly from these calls.
Changed Open_table_context class to store MDL_request object for
global intention exclusive lock acquired when opening tables.
open_table():
Do not release metadata lock if we have failed to open table as
this lock might have been acquired by one of previous statements
in transaction, and therefore should not be released.
open_system_tables_for_read()/close_system_tables()/
open_performance_schema_table():
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
close_performance_schema_table():
Got rid of duplicated code.
sql/sql_class.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. Also releasing
metadata lock on system table is now responsibility of
THD::restore_backup_open_tables_state().
Adjusted assert in THD::cleanup() to take into account fact that now
we also use MDL sentinel for global read lock.
sql/sql_class.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. As result:
- 'mdl_context' member was moved out of Open_tables_state to THD class.
enter_locked_tables_mode()/leave_locked_tables_mode() had to follow.
- Methods of THD responsible for saving/restoring open table state were
changed to use Open_tables_backup class which in addition to
Open_table_state has a member for this savepoint.
Changed Open_table_context class to store MDL_request object for
global intention exclusive lock acquired when opening tables.
sql/sql_delete.cc:
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object exclusive metadata lock is acquired.
Now this has to be done by explicit calls outside of MDL subsystem.
sql/sql_help.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sql_parse.cc:
Adjusted assert reload_acl_and_cache() to the fact that global read
lock now takes full-blown metadata lock.
sql/sql_plist.h:
Added support for element counting to I_P_List list template.
One can use policy classes to specify if such counting is needed
or not needed for particular list.
sql/sql_show.cc:
Instead of using separate MDL_context for opening tables for I_S
purposes we now create MDL savepoint in the main context before
opening tables and rollback to this savepoint after closing them.
To support this and similar change for system tables methods of
THD responsible for saving/restoring open table state were changed
to use Open_tables_backup class which in addition to Open_table_state
has a member for this savepoint. As result code opening/closing tables
for I_S purposes was changed to use Open_tables_backup instead of
Open_table_state class as well.
sql/sql_table.cc:
mysql_rm_tables_part2():
Since now global intention exclusive metadata lock is ordinary
metadata lock we no longer can rely that by releasing MDL locks
on all tables we will release all locks acquired by this routine.
So in non-LOCK-TABLES mode we have to release all locks acquired
explicitly.
prepare_for_repair(), mysql_alter_table():
MDL subsystem no longer implicitly acquires global intention
exclusive metadata lock when per-object exclusive metadata lock
is acquired. Now this has to be done by explicit calls outside of
MDL subsystem.
sql/tztime.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
Also changed code not to use special mechanism for open system tables
when it is not really necessary.
2010-01-21 21:43:03 +01:00
drop table if exists t1, t0;
2007-10-11 02:39:22 +02:00
create table t1 (c1 int);
connection: default
handler t1 open;
handler t1 read first;
c1
connection: flush
Patch that changes metadata locking subsystem to use mutex per lock and
condition variable per context instead of one mutex and one conditional
variable for the whole subsystem.
This should increase concurrency in this subsystem.
It also opens the way for further changes which are necessary to solve
such bugs as bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock"
and bug #37346 "innodb does not detect deadlock between update and alter
table".
Two other notable changes done by this patch:
- MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object metadata lock is acquired. Now this has
to be done by explicit calls outside of MDL subsystem.
- Instead of using separate MDL_context for opening system tables/tables
for purposes of I_S we now create MDL savepoint in the main context
before opening tables and rollback to this savepoint after closing
them. This means that it is now possible to get ER_LOCK_DEADLOCK error
even not inside a transaction. This might happen in unlikely case when
one runs DDL on one of system tables while also running DDL on some
other tables. Cases when this ER_LOCK_DEADLOCK error is not justified
will be addressed by advanced deadlock detector for MDL subsystem which
we plan to implement.
mysql-test/include/handler.inc:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/handler_innodb.result:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/handler_myisam.result:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/mdl_sync.result:
Adjusted mdl_sync.test to the fact that exclusive metadata locks on
tables are now acquired according to alphabetical order of fully
qualified table names instead of order in which tables are mentioned
in statement.
mysql-test/t/mdl_sync.test:
Adjusted mdl_sync.test to the fact that exclusive metadata locks on
tables are now acquired according to alphabetical order of fully
qualified table names instead of order in which tables are mentioned
in statement.
sql/events.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/ha_ndbcluster.cc:
Since manipulations with open table state no longer install proxy
MDL_context it does not make sense to perform them in order to
satisfy assert in mysql_rm_tables_part2(). Removed them per agreement
with Cluster team. This has not broken test suite since scenario in
which deadlock can occur and assertion fails is not covered by tests.
sql/lock.cc:
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object exclusive metadata lock is acquired.
Now this has to be done by explicit calls outside of MDL subsystem.
sql/log.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/mdl.cc:
Changed metadata locking subsystem to use mutex per lock and condition
variable per context instead of one mutex and one conditional variable
for the whole subsystem.
Changed approach to handling of global metadata locks. Instead of
implicitly acquiring intention exclusive locks when user requests
per-object upgradeable or exclusive locks now we require them to be
acquired explicitly in the same way as ordinary metadata locks.
In fact global lock are now ordinary metadata locks in new GLOBAL
namespace.
To implement these changes:
- Removed LOCK_mdl mutex and COND_mdl condition variable.
- Introduced MDL_lock::m_mutex mutexes which protect individual lock
objects.
- Replaced mdl_locks hash with MDL_map class, which has hash for
MDL_lock objects as a member and separate mutex which protects this
hash. Methods of this class allow to find(), find_or_create() or
remove() MDL_lock objects in concurrency-friendly fashion (i.e.
for most common operation, find_or_create(), we don't acquire
MDL_lock::m_mutex while holding MDL_map::m_mutex. Thanks to MikaelR
for this idea and benchmarks!). Added three auxiliary members to
MDL_lock class (m_is_destroyed, m_ref_usage, m_ref_release) to
support this concurrency-friendly behavior.
- Introduced MDL_context::m_ctx_wakeup_cond condition variable to be
used for waiting until this context's pending request can be
satisfied or its thread has to perform actions to resolve potential
deadlock. Context which want to wait add ticket corresponding to the
request to an appropriate queue of waiters in MDL_lock object so
they can be noticed when other contexts change state of lock and be
awaken by them by signalling on MDL_context::m_ctx_wakeup_cond.
As consequence MDL_ticket objects has to be used for any waiting
in metadata locking subsystem including one which happens in
MDL_context::wait_for_locks() method.
Another consequence is that MDL_context is no longer copyable and
can't be saved/restored when working with system tables.
- Made MDL_lock an abstract class, which delegates specifying exact
compatibility matrix to its descendants. Added MDL_global_lock child
class for global lock (The old is_lock_type_compatible() method
became can_grant_lock() method of this class). Added MDL_object_lock
class to represent per-object lock (The old MDL_lock::can_grant_lock()
became its method). Choice between two classes happens based on MDL
namespace in MDL_lock::create() method.
- Got rid of MDL_lock::type member as its meaning became ambigous for
global locks.
- To simplify waking up of contexts waiting for lock split waiting queue
in MDL_lock class in two queues. One for pending requests for exclusive
(including intention exclusive) locks and another for requests for
shared locks.
- Added virtual wake_up_waiters() method to MDL_lock, MDL_global_lock and
MDL_object_lock classes which allows to wake up waiting contexts after
state of lock changes. Replaced old duplicated code with calls to this
method.
- Adjusted MDL_context::try_acquire_shared_lock()/exclusive_lock()/
global_shared_lock(), MDL_ticket::upgrade_shared_lock_to_exclusive_lock()
and MDL_context::release_ticket() methods to use MDL_map and
MDL_lock::m_mutex instead of single LOCK_mdl mutex and wake up
waiters according to the approach described above. The latter method
also was renamed to MDL_context::release_lock().
- Changed MDL_context::try_acquire_shared_lock()/exclusive_lock() and
release_lock() not to handle global locks. They are now supposed to
be taken explicitly like ordinary metadata locks.
- Added helper MDL_context::try_acquire_global_intention_exclusive_lock()
and acquire_global_intention_exclusive_lock() methods.
- Moved common code from MDL_context::acquire_global_shared_lock() and
acquire_global_intention_exclusive_lock() to new method -
MDL_context::acquire_lock_impl().
- Moved common code from MDL_context::try_acquire_shared_lock(),
try_acquire_global_intention_exclusive_lock()/exclusive_lock()
to MDL_context::try_acquire_lock_impl().
- Since acquiring of several exclusive locks can no longer happen under
single LOCK_mdl mutex the approach to it had to be changed. Now we do
it in one by one fashion. This is done in alphabetical order to avoid
deadlocks. Changed MDL_context::acquire_exclusive_locks() accordingly
(as part of this change moved code responsible for acquiring single
exclusive lock to new MDL_context::acquire_exclusive_lock_impl()
method).
- Since we no longer have single LOCK_mdl mutex which protects all
MDL_context::m_is_waiting_in_mdl members using these members to
determine if we have really awaken context holding conflicting
shared lock became inconvinient. Got rid of this member and changed
notify_shared_lock() helper function and process of acquiring
of/upgrading to exclusive lock not to rely on such information.
Now in MDL_context::acquire_exclusive_lock_impl() and
MDL_ticket::upgrade_shared_lock_to_exclusive_lock() we simply
re-try to wake up threads holding conflicting shared locks after
small time out.
- Adjusted MDL_context::can_wait_lead_to_deadlock() and
MDL_ticket::has_pending_conflicting_lock() to use per-lock
mutexes instead of LOCK_mdl. To do this introduced
MDL_lock::has_pending_exclusive_lock() method.
sql/mdl.h:
Changed metadata locking subsystem to use mutex per lock and condition
variable per context instead of one mutex and one conditional variable
for the whole subsystem. In order to implement this change:
- Added MDL_key::cmp() method to be able to sort MDL_key objects
alphabetically. Changed length fields in MDL_key class to uint16
as 16-bit is enough for length of any key.
- Changed MDL_ticket::get_ctx() to return pointer to non-const
object in order to be able to use MDL_context::awake() method
for such contexts.
- Got rid of unlocked versions of can_wait_lead_to_deadlock()/
has_pending_conflicting_lock() methods in MDL_context and
MDL_ticket. We no longer has single mutex which protects all
locks. Thus one always has to use versions of these methods
which acquire per-lock mutexes.
- MDL_request_list type of list now counts its elements.
- Added MDL_context::m_ctx_wakeup_cond condition variable to be used
for waiting until this context's pending request can be satisfied
or its thread has to perform actions to resolve potential deadlock.
Added awake() method to wake up context from such wait.
Addition of condition variable made MDL_context uncopyable.
As result we no longer can save/restore MDL_context when working
with system tables. Instead we create MDL savepoint before opening
those tables and rollback to it once they are closed.
- MDL_context::release_ticket() became release_lock() method.
- Added auxiliary MDL_context::acquire_exclusive_lock_impl() method
which does all necessary work to acquire exclusive lock on one object
but should not be used directly as it does not enforce any asserts
ensuring that no deadlocks are possible.
- Since we no longer need to know if thread trying to acquire exclusive
lock managed to wake up any threads having conflicting shared locks
(as, anyway, we will try to wake up such threads again shortly)
- MDL_context::m_is_waiting_in_mdl member became unnecessary and
notify_shared_lock() no longer needs to be friend of MDL_context.
Changed approach to handling of global metadata locks. Instead of
implicitly acquiring intention exclusive locks when user requests
per-object upgradeable or exclusive locks now we require them to be
acquired explicitly in the same way as ordinary metadata locks.
- Added new GLOBAL namespace for such locks.
- Added new type of lock to be requested MDL_INTENTION_EXCLISIVE.
- Added MDL_context::try_acquire_global_intention_exclusive_lock()
and acquire_global_intention_exclusive_lock() methods.
- Moved common code from MDL_context::acquire_global_shared_lock()
and acquire_global_intention_exclusive_lock() to new method -
MDL_context::acquire_lock_impl().
- Moved common code from MDL_context::try_acquire_shared_lock(),
try_acquire_global_intention_exclusive_lock()/exclusive_lock()
to MDL_context::try_acquire_lock_impl().
- Added helper MDL_context::is_global_lock_owner() method to be
able easily to find what kind of global lock this context holds.
- MDL_context::m_has_global_shared_lock became unnecessary as
global read lock is now represented by ordinary ticket.
- Removed assert in MDL_context::set_lt_or_ha_sentinel() which became
false for cases when we execute LOCK TABLES under global read lock
mode.
sql/mysql_priv.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result calls
opening/closing system tables were changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sp.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sp.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sql_base.cc:
close_thread_tables():
Since we no longer use separate MDL_context for opening system
tables we need to avoid releasing all transaction locks when
closing system table. Releasing metadata lock on system table
is now responsibility of THD::restore_backup_open_tables_state().
open_table_get_mdl_lock(),
Open_table_context::recover_from_failed_open():
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object upgradable or exclusive metadata lock is
acquired. So this have to be done explicitly from these calls.
Changed Open_table_context class to store MDL_request object for
global intention exclusive lock acquired when opening tables.
open_table():
Do not release metadata lock if we have failed to open table as
this lock might have been acquired by one of previous statements
in transaction, and therefore should not be released.
open_system_tables_for_read()/close_system_tables()/
open_performance_schema_table():
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
close_performance_schema_table():
Got rid of duplicated code.
sql/sql_class.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. Also releasing
metadata lock on system table is now responsibility of
THD::restore_backup_open_tables_state().
Adjusted assert in THD::cleanup() to take into account fact that now
we also use MDL sentinel for global read lock.
sql/sql_class.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. As result:
- 'mdl_context' member was moved out of Open_tables_state to THD class.
enter_locked_tables_mode()/leave_locked_tables_mode() had to follow.
- Methods of THD responsible for saving/restoring open table state were
changed to use Open_tables_backup class which in addition to
Open_table_state has a member for this savepoint.
Changed Open_table_context class to store MDL_request object for
global intention exclusive lock acquired when opening tables.
sql/sql_delete.cc:
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object exclusive metadata lock is acquired.
Now this has to be done by explicit calls outside of MDL subsystem.
sql/sql_help.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sql_parse.cc:
Adjusted assert reload_acl_and_cache() to the fact that global read
lock now takes full-blown metadata lock.
sql/sql_plist.h:
Added support for element counting to I_P_List list template.
One can use policy classes to specify if such counting is needed
or not needed for particular list.
sql/sql_show.cc:
Instead of using separate MDL_context for opening tables for I_S
purposes we now create MDL savepoint in the main context before
opening tables and rollback to this savepoint after closing them.
To support this and similar change for system tables methods of
THD responsible for saving/restoring open table state were changed
to use Open_tables_backup class which in addition to Open_table_state
has a member for this savepoint. As result code opening/closing tables
for I_S purposes was changed to use Open_tables_backup instead of
Open_table_state class as well.
sql/sql_table.cc:
mysql_rm_tables_part2():
Since now global intention exclusive metadata lock is ordinary
metadata lock we no longer can rely that by releasing MDL locks
on all tables we will release all locks acquired by this routine.
So in non-LOCK-TABLES mode we have to release all locks acquired
explicitly.
prepare_for_repair(), mysql_alter_table():
MDL subsystem no longer implicitly acquires global intention
exclusive metadata lock when per-object exclusive metadata lock
is acquired. Now this has to be done by explicit calls outside of
MDL subsystem.
sql/tztime.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
Also changed code not to use special mechanism for open system tables
when it is not really necessary.
2010-01-21 21:43:03 +01:00
rename table t1 to t0;;
Initial import of WL#3726 "DDL locking for all metadata objects".
Backport of:
------------------------------------------------------------
revno: 2630.4.1
committer: Dmitry Lenev <dlenev@mysql.com>
branch nick: mysql-6.0-3726-w
timestamp: Fri 2008-05-23 17:54:03 +0400
message:
WL#3726 "DDL locking for all metadata objects".
After review fixes in progress.
------------------------------------------------------------
This is the first patch in series. It transforms the metadata
locking subsystem to use a dedicated module (mdl.h,cc). No
significant changes in the locking protocol.
The import passes the test suite with the exception of
deprecated/removed 6.0 features, and MERGE tables. The latter
are subject to a fix by WL#4144.
Unfortunately, the original changeset comments got lost in a merge,
thus this import has its own (largely insufficient) comments.
This patch fixes Bug#25144 "replication / binlog with view breaks".
Warning: this patch introduces an incompatible change:
Under LOCK TABLES, it's no longer possible to FLUSH a table that
was not locked for WRITE.
Under LOCK TABLES, it's no longer possible to DROP a table or
VIEW that was not locked for WRITE.
******
Backport of:
------------------------------------------------------------
revno: 2630.4.2
committer: Dmitry Lenev <dlenev@mysql.com>
branch nick: mysql-6.0-3726-w
timestamp: Sat 2008-05-24 14:03:45 +0400
message:
WL#3726 "DDL locking for all metadata objects".
After review fixes in progress.
******
Backport of:
------------------------------------------------------------
revno: 2630.4.3
committer: Dmitry Lenev <dlenev@mysql.com>
branch nick: mysql-6.0-3726-w
timestamp: Sat 2008-05-24 14:08:51 +0400
message:
WL#3726 "DDL locking for all metadata objects"
Fixed failing Windows builds by adding mdl.cc to the lists
of files needed to build server/libmysqld on Windows.
******
Backport of:
------------------------------------------------------------
revno: 2630.4.4
committer: Dmitry Lenev <dlenev@mysql.com>
branch nick: mysql-6.0-3726-w
timestamp: Sat 2008-05-24 21:57:58 +0400
message:
WL#3726 "DDL locking for all metadata objects".
Fix for assert failures in kill.test which occured when one
tried to kill ALTER TABLE statement on merge table while it
was waiting in wait_while_table_is_used() for other connections
to close this table.
These assert failures stemmed from the fact that cleanup code
in this case assumed that temporary table representing new
version of table was open with adding to THD::temporary_tables
list while code which were opening this temporary table wasn't
always fulfilling this.
This patch changes code that opens new version of table to
always do this linking in. It also streamlines cleanup process
for cases when error occurs while we have new version of table
open.
******
WL#3726 "DDL locking for all metadata objects"
Add libmysqld/mdl.cc to .bzrignore.
******
Backport of:
------------------------------------------------------------
revno: 2630.4.6
committer: Dmitry Lenev <dlenev@mysql.com>
branch nick: mysql-6.0-3726-w
timestamp: Sun 2008-05-25 00:33:22 +0400
message:
WL#3726 "DDL locking for all metadata objects".
Addition to the fix of assert failures in kill.test caused by
changes for this worklog.
Make sure we close the new table only once.
.bzrignore:
Add libmysqld/mdl.cc
libmysqld/CMakeLists.txt:
Added mdl.cc to the list of files needed for building of libmysqld.
libmysqld/Makefile.am:
Added files implementing new meta-data locking subsystem to the server.
mysql-test/include/handler.inc:
Use separate connection for waiting while threads performing DDL
operations conflicting with open HANDLER tables reach blocked
state. This is required because now we check and close tables open
by HANDLER statements in this connection conflicting with DDL in
another each time open_tables() is called and thus select from I_S
which is used for waiting will unblock DDL operations if issued
from connection with open HANDLERs.
mysql-test/r/create.result:
Adjusted test case after change in implementation of CREATE TABLE
... SELECT. We no longer have special check in open_table() which
catches the case when we select from the table created. Instead we
rely on unique_table() call which happens after opening and
locking all tables.
mysql-test/r/flush.result:
FLUSH TABLES WITH READ LOCK can no longer happen under LOCK
TABLES. Updated test accordingly.
mysql-test/r/flush_table.result:
Under LOCK TABLES we no longer allow to do FLUSH TABLES for tables
locked for read. Updated test accordingly.
mysql-test/r/handler_innodb.result:
Use separate connection for waiting while threads performing DDL
operations conflicting with open HANDLER tables reach blocked
state. This is required because now we check and close tables open
by HANDLER statements in this connection conflicting with DDL in
another each time open_tables() is called and thus select from I_S
which is used for waiting will unblock DDL operations if issued
from connection with open HANDLERs.
mysql-test/r/handler_myisam.result:
Use separate connection for waiting while threads performing DDL
operations conflicting with open HANDLER tables reach blocked
state. This is required because now we check and close tables open
by HANDLER statements in this connection conflicting with DDL in
another each time open_tables() is called and thus select from I_S
which is used for waiting will unblock DDL operations if issued
from connection with open HANDLERs.
mysql-test/r/information_schema.result:
Additional test for WL#3726 "DDL locking for all metadata
objects". Check that we use high-priority metadata lock requests
when filling I_S tables.
Rearrange tests to match 6.0 better (fewer merge conflicts).
mysql-test/r/kill.result:
Added tests checking that DDL and DML statements waiting for
metadata locks can be interrupted by KILL command.
mysql-test/r/lock.result:
One no longer is allowed to do DROP VIEW under LOCK TABLES even if
this view is locked by LOCK TABLES. The problem is that in such
situation write locks on view are not mutually exclusive so
upgrading metadata lock which is required for dropping of view
will lead to deadlock.
mysql-test/r/partition_column_prune.result:
Update results (same results in 6.0), WL#3726
mysql-test/r/partition_pruning.result:
Update results (same results in 6.0), WL#3726
mysql-test/r/ps_ddl.result:
We no longer invalidate prepared CREATE TABLE ... SELECT statement
if target table changes. This is OK since it is not strictly
necessary.
The first change is wrong, is caused by FLUSH TABLE
now flushing all unused tables. This is a regression that
Dmitri fixed in 6.0 in a follow up patch.
mysql-test/r/sp.result:
Under LOCK TABLES we no longer allow accessing views which were
not explicitly locked. To access view we need to obtain metadata
lock on it and doing this under LOCK TABLES may lead to deadlocks.
mysql-test/r/view.result:
One no longer is allowed to do DROP VIEW under LOCK TABLES even if
this view is locked by LOCK TABLES. The problem is that in such
situation even "write locks" on view are not mutually exclusive so
upgrading metadata lock which is required for dropping of view
will lead to deadlock
mysql-test/r/view_grant.result:
ALTER VIEW implementation was changed to open a view only after
checking that user which does alter has appropriate privileges on
it. This means that in case when user's privileges are
insufficient for this we won't check that new view definer is the
same as original one or user performing alter has SUPER privilege.
Adjusted test case accordingly.
mysql-test/r/view_multi.result:
Added test case for bug#25144 "replication / binlog with view
breaks".
mysql-test/suite/rpl/t/disabled.def:
Disable test for deprecated features (they don't work with new MDL).
mysql-test/t/create.test:
Adjusted test case after change in implementation of CREATE TABLE
... SELECT. We no longer have special check in open_table() which
catches the case when we select from the table created. Instead we
rely on unique_table() call which happens after opening and
locking all tables.
mysql-test/t/disabled.def:
Disable merge.test, subject of WL#4144
mysql-test/t/flush.test:
FLUSH TABLES WITH READ LOCK can no longer happen under LOCK
TABLES. Updated test accordingly.
mysql-test/t/flush_table.test:
Under LOCK TABLES we no longer allow to do FLUSH TABLES for tables
locked for read. Updated test accordingly.
mysql-test/t/information_schema.test:
Additional test for WL#3726 "DDL locking for all metadata
objects". Check that we use high-priority metadata lock requests
when filling I_S tables.
Rearrange the results for easier merges with 6.0.
mysql-test/t/kill.test:
Added tests checking that DDL and DML statements waiting for
metadata locks can be interrupted by KILL command.
mysql-test/t/lock.test:
One no longer is allowed to do DROP VIEW under LOCK TABLES even if
this view is locked by LOCK TABLES. The problem is that in such
situation write locks on view are not mutually exclusive so
upgrading metadata lock which is required for dropping of view
will lead to deadlock.
mysql-test/t/lock_multi.test:
Adjusted test case to the changes of status in various places
caused by change in implementation FLUSH TABLES WITH READ LOCK,
which is now takes global metadata lock before flushing tables and
therefore waits on at these places.
mysql-test/t/ps_ddl.test:
We no longer invalidate prepared CREATE TABLE ... SELECT statement
if target table changes. This is OK since it is not strictly
necessary.
The first change is wrong, is caused by FLUSH TABLE
now flushing all unused tables. This is a regression that
Dmitri fixed in 6.0 in a follow up patch.
mysql-test/t/sp.test:
Under LOCK TABLES we no longer allow accessing views which were
not explicitly locked. To access view we need to obtain metadata
lock on it and doing this under LOCK TABLES may lead to deadlocks.
mysql-test/t/trigger_notembedded.test:
Adjusted test case to the changes of status in various places
caused by change in implementation FLUSH TABLES WITH READ LOCK,
which is now takes global metadata lock before flushing tables and
therefore waits on at these places.
mysql-test/t/view.test:
One no longer is allowed to do DROP VIEW under LOCK TABLES even if
this view is locked by LOCK TABLES. The problem is that in such
situation even "write locks" on view are not mutually exclusive so
upgrading metadata lock which is required for dropping of view
will lead to deadlock.
mysql-test/t/view_grant.test:
ALTER VIEW implementation was changed to open a view only after
checking that user which does alter has appropriate privileges on
it. This means that in case when user's privileges are
insufficient for this we won't check that new view definer is the
same as original one or user performing alter has SUPER privilege.
Adjusted test case accordingly.
mysql-test/t/view_multi.test:
Added test case for bug#25144 "replication / binlog with view
breaks".
sql/CMakeLists.txt:
Added mdl.cc to the list of files needed for building of server.
sql/Makefile.am:
Added files implementing new meta-data locking subsystem to the
server.
sql/event_db_repository.cc:
Allocate metadata lock requests objects (MDL_LOCK) on execution
memory root in cases when TABLE_LIST objects is also allocated
there or on stack.
sql/ha_ndbcluster.cc:
Adjusted code to work nicely with new metadata locking subsystem.
close_cached_tables() no longer has wait_for_placeholder argument.
Instead of relying on this parameter and related behavior FLUSH
TABLES WITH READ LOCK now takes global shared metadata lock.
sql/ha_ndbcluster_binlog.cc:
Adjusted code to work with new metadata locking subsystem.
close_cached_tables() no longer has wait_for_placeholder argument.
Instead of relying on this parameter and related behavior FLUSH
TABLES WITH READ LOCK now takes global shared metadata lock.
sql/handler.cc:
update_frm_version():
Directly update TABLE_SHARE::mysql_version member instead of
going through all TABLE instances for this table (old code was a
legacy from pre-table-definition-cache days).
sql/lock.cc:
Use new metadata locking subsystem. Threw away most of functions
related to name locking as now one is supposed to use metadata
locking API instead. In lock_global_read_lock() and
unlock_global_read_lock() in order to avoid problems with global
read lock sneaking in at the moment when we perform FLUSH TABLES
or ALTER TABLE under LOCK TABLES and when tables being reopened
are protected only by metadata locks we also have to take global
shared meta data lock.
sql/log_event.cc:
Adjusted code to work with new metadata locking subsystem. For
tables open by slave thread for applying RBR events allocate
memory for lock request object in the same chunk of memory as
TABLE_LIST objects for them. In order to ensure that we keep these
objects around until tables are open always close tables before
calling Relay_log_info::clear_tables_to_lock(). Use new auxiliary
Relay_log_info::slave_close_thread_tables() method to enforce
this.
sql/log_event_old.cc:
Adjusted code to work with new metadata locking subsystem. Since
for tables open by slave thread for applying RBR events memory for
lock request object is allocated in the same chunk of memory as
TABLE_LIST objects for them we have to ensure that we keep these
objects around until tables are open. To ensure this we always
close tables before calling
Relay_log_info::clear_tables_to_lock(). To enfore this we use
new auxiliary Relay_log_info::slave_close_thread_tables()
method.
sql/mdl.cc:
Implemented new metadata locking subsystem and API described in
WL3726 "DDL locking for all metadata objects".
sql/mdl.h:
Implemented new metadata locking subsystem and API described in
WL3726 "DDL locking for all metadata objects".
sql/mysql_priv.h:
- close_thread_tables()/close_tables_for_reopen() now has one more
argument which indicates that metadata locks should be released
but not removed from the context in order to be used later in
mdl_wait_for_locks() and tdc_wait_for_old_version().
- close_cached_table() routine is no longer public.
- Thread waiting in wait_while_table_is_used() can be now killed
so this function returns boolean to make caller aware of such
situation.
- We no longer have table cache as separate entity instead used
and unused TABLE instances are linked to TABLE_SHARE objects in
table definition cache.
- Now third argument of open_table() is also used for requesting
table repair or auto-discovery of table's new definition. So its
type was changed from bool to enum.
- Added tdc_open_view() function for opening view by getting its
definition from disk (and table cache in future).
- reopen_name_locked_table() no longer needs "link_in" argument as
now we have exclusive metadata locks instead of dummy TABLE
instances when this function is called.
- find_locked_table() now takes head of list of TABLE instances
instead of always scanning through THD::open_tables list. Also
added find_write_locked_table() auxiliary.
- reopen_tables(), close_cached_tables() no longer have
mark_share_as_old and wait_for_placeholder arguments. Instead of
relying on this parameters and related behavior FLUSH TABLES
WITH READ LOCK now takes global shared metadata lock.
- We no longer need drop_locked_tables() and
abort_locked_tables().
- mysql_ha_rm_tables() now always assume that LOCK_open is not
acquired by caller.
- Added notify_thread_having_shared_lock() callback invoked by
metadata locking subsystem when acquiring an exclusive lock, for
each thread that has a conflicting shared metadata lock.
- Introduced expel_table_from_cache() as replacement for
remove_table_from_cache() (the main difference is that this new
function assumes that caller follows metadata locking protocol
and never waits).
- Threw away most of functions related to name locking. One should
use new metadata locking subsystem and API instead.
sql/mysqld.cc:
Got rid of call initializing/deinitializing table cache since now
it is embedded into table definition cache. Added calls for
initializing/ deinitializing metadata locking subsystem.
sql/rpl_rli.cc:
Introduced auxiliary Relay_log_info::slave_close_thread_tables()
method which is used for enforcing that we always close tables
open for RBR before deallocating TABLE_LIST elements and MDL_LOCK
objects for them.
sql/rpl_rli.h:
Introduced auxiliary Relay_log_info::slave_close_thread_tables()
method which is used for enforcing that we always close tables
open for RBR before deallocating TABLE_LIST elements and MDL_LOCK
objects for them.
sql/set_var.cc:
close_cached_tables() no longer has wait_for_placeholder argument.
Instead of relying on this parameter and related behavior FLUSH
TABLES WITH READ LOCK now takes global shared metadata lock.
sql/sp_head.cc:
For tables added to the statement's table list by prelocking
algorithm we allocate these objects either on the same memory as
corresponding table list elements or on THD::locked_tables_root
(if we are building table list for LOCK TABLES).
sql/sql_acl.cc:
Allocate metadata lock requests objects (MDL_LOCK) on execution
memory root in cases when we use stack TABLE_LIST objects to open
tables. Got rid of redundant code by using unlock_locked_tables()
function.
sql/sql_base.cc:
Changed code to use new MDL subsystem. Got rid of separate table
cache. Now used and unused TABLE instances are linked to the
TABLE_SHAREs in table definition cache.
check_unused():
Adjusted code to the fact that we no longer have separate table
cache. Removed dead code.
table_def_free():
Free TABLE instances referenced from TABLE_SHARE objects before
destroying table definition cache.
get_table_share():
Added assert which ensures that noone will be able to access
table (and its share) without acquiring some kind of metadata
lock first.
close_handle_and_leave_table_as_lock():
Adjusted code to the fact that TABLE instances now are linked to
list in TABLE_SHARE.
list_open_tables():
Changed this function to use table definition cache instead of
table cache.
free_cache_entry():
Unlink freed TABLE elements from the list of all TABLE instances
for the table in TABLE_SHARE.
kill_delayed_thread_for_table():
Added auxiliary for killing delayed insert threads for
particular table.
close_cached_tables():
Got rid of wait_for_refresh argument as we now rely on global
shared metadata lock to prevent FLUSH WITH READ LOCK sneaking in
when we are reopening tables. Heavily reworked this function to
use new MDL code and not to rely on separate table cache entity.
close_open_tables():
We no longer have separate table cache.
close_thread_tables():
Release metadata locks after closing all tables. Added skip_mdl
argument which allows us not to remove metadata lock requests
from the context in case when we are going to use this requests
later in mdl_wait_for_locks() and tdc_wait_for_old_versions().
close_thread_table()/close_table_for_reopen():
Since we no longer have separate table cache and all TABLE
instances are linked to TABLE_SHARE objects in table definition
cache we have to link/unlink TABLE object to/from appropriate
lists in the share.
name_lock_locked_table():
Moved redundant code to find_write_locked_table() function and
adjusted code to the fact that wait_while_table_is_used() can
now return with an error if our thread is killed.
reopen_table_entry():
We no longer need "link_in" argument as with MDL we no longer
call this function with dummy TABLE object pre-allocated and
added to the THD::open_tables. Also now we add newly-open TABLE
instance to the list of share's used TABLE instances.
table_cache_insert_placeholder():
Got rid of name-locking legacy.
lock_table_name_if_not_cached():
Moved to sql_table.cc the only place where it is used. It was
also reimplemented using new MDL API.
open_table():
- Reworked this function to use new MDL subsystem.
- Changed code to deal with table definition cache directly
instead of going through separate table cache.
- Now third argument is also used for requesting table repair
or auto-discovery of table's new definition. So its type was
changed from bool to enum.
find_locked_table()/find_write_locked_table():
Accept head of list of TABLE objects as first argument and use
this list instead of always searching in THD::open_tables list.
Also added auxiliary for finding write-locked locked tables.
reopen_table():
Adjusted function to work with new MDL subsystem and to properly
manuipulate with lists of used/unused TABLE instaces in
TABLE_SHARE.
reopen_tables():
Removed mark_share_as_old parameter. Instead of relying on it
and related behavior FLUSH TABLES WITH READ LOCK now takes
global shared metadata lock. Changed code after removing
separate table cache.
drop_locked_tables()/abort_locked_tables():
Got rid of functions which are no longer needed.
unlock_locked_tables():
Moved this function from sql_parse.cc and changed it to release
memory which was used for allocating metadata lock requests for
tables open and locked by LOCK TABLES.
tdc_open_view():
Intoduced function for opening a view by getting its definition
from disk (and table cache in future).
reopen_table_entry():
Introduced function for opening table definitions while holding
exclusive metatadata lock on it.
open_unireg_entry():
Got rid of this function. Most of its functionality is relocated
to open_table() and open_table_fini() functions, and some of it
to reopen_table_entry() and tdc_open_view(). Also code
resposible for auto-repair and auto-discovery of tables was
moved to separate function.
open_table_entry_fini():
Introduced function which contains common actions which finalize
process of TABLE object creation.
auto_repair_table():
Moved code responsible for auto-repair of table being opened
here.
handle_failed_open_table_attempt()
Moved code responsible for handling failing attempt to open
table to one place (retry due to lock conflict/old version,
auto-discovery and repair).
open_tables():
- Flush open HANDLER tables if they have old version of if there
is conflicting metadata lock against them (before this moment
we had this code in open_table()).
- When we open view which should be processed via derived table
on the second execution of prepared statement or stored
routine we still should call open_table() for it in order to
obtain metadata lock on it and prepare its security context.
- In cases when we discover that some special handling of
failure to open table is needed call
handle_failed_open_table_attempt() which handles all such
scenarios.
open_ltable():
Handling of various special scenarios of failure to open a table
was moved to separate handle_failed_open_table_attempt()
function.
remove_db_from_cache():
Removed this function as it is no longer used.
notify_thread_having_shared_lock():
Added callback which is invoked by MDL subsystem when acquiring
an exclusive lock, for each thread that has a conflicting shared
metadata lock.
expel_table_from_cache():
Introduced function for removing unused TABLE instances. Unlike
remove_table_from_cache() it relies on caller following MDL
protocol and having appropriate locks when calling it and thus
does not do any waiting if table is still in use.
tdc_wait_for_old_version():
Added function which allows open_tables() to wait in cases when
we discover that we should back-off due to presence of old
version of table.
abort_and_upgrade_lock():
Use new MDL calls.
mysql_wait_completed_table():
Got rid of unused function.
open_system_tables_for_read/for_update()/performance_schema_table():
Allocate MDL_LOCK objects on execution memory root in cases when
TABLE_LIST objects for corresponding tables is allocated on
stack.
close_performance_schema_table():
Release metadata locks after closing tables.
******
Use I_P_List for free/used tables list in the table share.
sql/sql_binlog.cc:
Use Relay_log_info::slave_close_thread_tables() method to enforce
that we always close tables open for RBR before deallocating
TABLE_LIST elements and MDL_LOCK objects for them.
sql/sql_class.cc:
Added meta-data locking contexts as part of Open_tables_state
context. Also introduced THD::locked_tables_root memory root
which is to be used for allocating MDL_LOCK objects for tables in
LOCK TABLES statement (end of lifetime for such objects is UNLOCK
TABLES so we can't use statement or execution root for them).
sql/sql_class.h:
Added meta-data locking contexts as part of Open_tables_state
context. Also introduced THD::locked_tables_root memory root
which is to be used for allocating MDL_LOCK objects for tables in
LOCK TABLES statement (end of lifetime for such objects is UNLOCK
TABLES so we can't use statement or execution root for them).
Note: handler_mdl_context and locked_tables_root and
mdl_el_root will be removed by subsequent patches.
sql/sql_db.cc:
mysql_rm_db() does not really need to call remove_db_from_cache()
as it drops each table in the database using
mysql_rm_table_part2(), which performs all necessary operations on
table (definition) cache.
sql/sql_delete.cc:
Use the new metadata locking API for TRUNCATE.
sql/sql_handler.cc:
Changed HANDLER implementation to use new metadata locking
subsystem. Note that MDL_LOCK objects for HANDLER tables are
allocated in the same chunk of heap memory as TABLE_LIST object
for those tables.
sql/sql_insert.cc:
mysql_insert():
find_locked_table() now takes head of list of TABLE object as
its argument instead of always scanning through THD::open_tables
list.
handle_delayed_insert():
Allocate metadata lock request object for table open by delayed
insert thread on execution memroot. create_table_from_items():
We no longer allocate dummy TABLE objects for tables being
created if they don't exist. As consequence
reopen_name_locked_table() no longer has link_in argument.
open_table() now has one more argument which is not relevant for
temporary tables.
sql/sql_parse.cc:
- Moved unlock_locked_tables() routine to sql_base.cc and made
available it in other files. Got rid of some redundant code by
using this function.
- Replaced boolean TABLE_LIST::create member with enum
open_table_type member.
- Use special memory root for allocating MDL_LOCK objects for
tables open and locked by LOCK TABLES (these object should live
till UNLOCK TABLES so we can't allocate them on statement nor
execution memory root). Also properly set metadata lock
upgradability attribure for those tables.
- Under LOCK TABLES it is no longer allowed to flush tables which
are not write-locked as this breaks metadata locking protocol
and thus potentially might lead to deadlock.
- Added auxiliary adjust_mdl_locks_upgradability() function.
sql/sql_partition.cc:
Adjusted code to the fact that reopen_tables() no longer has
"mark_share_as_old" argument. Got rid of comments which are no
longer true.
sql/sql_plist.h:
Added I_P_List template class for parametrized intrusive doubly
linked lists and I_P_List_iterator for corresponding iterator.
Unlike for I_List<> list elements of such list can participate in
several lists. Unlike List<> such lists are doubly-linked and
intrusive.
sql/sql_plugin.cc:
Allocate metadata lock requests objects (MDL_LOCK) on execution
memory root in cases when we use stack TABLE_LIST objects to open
tables.
sql/sql_prepare.cc:
Replaced boolean TABLE_LIST::create member with enum
open_table_type member. This allows easily handle situation in
which instead of opening the table we want only to take exclusive
metadata lock on it.
sql/sql_rename.cc:
Use new metadata locking subsystem in implementation of RENAME
TABLE.
sql/sql_servers.cc:
Allocate metadata lock requests objects (MDL_LOCK) on execution
memory root in cases when we use stack TABLE_LIST objects to open
tables. Got rid of redundant code by using unlock_locked_tables()
function.
sql/sql_show.cc:
Acquire shared metadata lock when we are getting information for
I_S table directly from TABLE_SHARE without doing full-blown table
open. We use high priority lock request in this situation in
order to avoid deadlocks.
Also allocate metadata lock requests objects (MDL_LOCK) on
execution memory root in cases when TABLE_LIST objects are also
allocated there
sql/sql_table.cc:
mysql_rm_table():
Removed comment which is no longer relevant.
mysql_rm_table_part2():
Now caller of mysql_ha_rm_tables() should not own LOCK_open.
Adjusted code to use new metadata locking subsystem instead of
name-locks.
lock_table_name_if_not_cached():
Moved this function from sql_base.cc to this file and
reimplemented it using metadata locking API.
mysql_create_table():
Adjusted code to use new MDL API.
wait_while_table_is_used():
Changed function to use new MDL subsystem. Made thread waiting
in it killable (this also led to introduction of return value so
caller can distinguish successful executions from situations
when waiting was aborted).
close_cached_tables():
Thread waiting in this function is killable now. As result it
has return value for distinguishing between succes and failure.
Got rid of redundant boradcast_refresh() call.
prepare_for_repair():
Use MDL subsystem instead of name-locks.
mysql_admin_table():
mysql_ha_rm_tables() now always assumes that caller doesn't own
LOCK_open.
mysql_repair_table():
We should mark all elements of table list as requiring
upgradable metadata locks.
mysql_create_table_like():
Use new MDL subsystem instead of name-locks.
create_temporary_tables():
We don't need to obtain metadata locks when creating temporary
table.
mysql_fast_or_online_alter_table():
Thread waiting in wait_while_table_is_used() is now killable.
mysql_alter_table():
Adjusted code to work with new MDL subsystem and to the fact
that threads waiting in what_while_table_is_used() and
close_cached_table() are now killable.
sql/sql_test.cc:
We no longer have separate table cache. TABLE instances are now
associated with/linked to TABLE_SHARE objects in table definition
cache.
sql/sql_trigger.cc:
Adjusted code to work with new metadata locking subsystem. Also
reopen_tables() no longer has mark_share_as_old argument (Instead
of relying on this parameter and related behavior FLUSH TABLES
WITH READ LOCK now takes global shared metadata lock).
sql/sql_udf.cc:
Allocate metadata lock requests objects (MDL_LOCK) on execution
memory root in cases when we use stack TABLE_LIST objects to open
tables.
sql/sql_update.cc:
Adjusted code to work with new meta-data locking subsystem.
sql/sql_view.cc:
Added proper meta-data locking to implementations of
CREATE/ALTER/DROP VIEW statements. Now we obtain exclusive
meta-data lock on a view before creating/ changing/dropping it.
This ensures that all concurrent statements that use this view
will finish before our statement will proceed and therefore we
will get correct order of statements in the binary log.
Also ensure that TABLE_LIST::mdl_upgradable attribute is properly
propagated for underlying tables of view.
sql/table.cc:
Added auxiliary alloc_mdl_locks() function for allocating metadata
lock request objects for all elements of table list.
sql/table.h:
TABLE_SHARE:
Got rid of unused members. Introduced members for storing lists
of used and unused TABLE objects for this share.
TABLE:
Added members for linking TABLE objects into per-share lists of
used and unused TABLE instances. Added member for holding
pointer to metadata lock for this table.
TABLE_LIST:
Replaced boolean TABLE_LIST::create member with enum
open_table_type member. This allows easily handle situation in
which instead of opening the table we want only to take
exclusive meta-data lock on it (we need this in order to handle
ALTER VIEW and CREATE VIEW statements).
Introduced new mdl_upgradable member for marking elements of
table list for which we need to take upgradable shared metadata
lock instead of plain shared metadata lock. Added pointer for
holding pointer to MDL_LOCK for the table.
Added auxiliary alloc_mdl_locks() function for allocating metadata
lock requests objects for all elements of table list. Added
auxiliary set_all_mdl_upgradable() function for marking all
elements in table list as requiring upgradable metadata locks.
storage/myisammrg/ha_myisammrg.cc:
Allocate MDL_LOCK objects for underlying tables of MERGE table.
To be reworked once Ingo pushes his patch for WL4144.
2009-11-30 16:55:03 +01:00
connection: waiter
2007-10-11 02:39:22 +02:00
connection: default
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
#
# RENAME placed two pending locks and waits.
Patch that changes metadata locking subsystem to use mutex per lock and
condition variable per context instead of one mutex and one conditional
variable for the whole subsystem.
This should increase concurrency in this subsystem.
It also opens the way for further changes which are necessary to solve
such bugs as bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock"
and bug #37346 "innodb does not detect deadlock between update and alter
table".
Two other notable changes done by this patch:
- MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object metadata lock is acquired. Now this has
to be done by explicit calls outside of MDL subsystem.
- Instead of using separate MDL_context for opening system tables/tables
for purposes of I_S we now create MDL savepoint in the main context
before opening tables and rollback to this savepoint after closing
them. This means that it is now possible to get ER_LOCK_DEADLOCK error
even not inside a transaction. This might happen in unlikely case when
one runs DDL on one of system tables while also running DDL on some
other tables. Cases when this ER_LOCK_DEADLOCK error is not justified
will be addressed by advanced deadlock detector for MDL subsystem which
we plan to implement.
mysql-test/include/handler.inc:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/handler_innodb.result:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/handler_myisam.result:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/mdl_sync.result:
Adjusted mdl_sync.test to the fact that exclusive metadata locks on
tables are now acquired according to alphabetical order of fully
qualified table names instead of order in which tables are mentioned
in statement.
mysql-test/t/mdl_sync.test:
Adjusted mdl_sync.test to the fact that exclusive metadata locks on
tables are now acquired according to alphabetical order of fully
qualified table names instead of order in which tables are mentioned
in statement.
sql/events.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/ha_ndbcluster.cc:
Since manipulations with open table state no longer install proxy
MDL_context it does not make sense to perform them in order to
satisfy assert in mysql_rm_tables_part2(). Removed them per agreement
with Cluster team. This has not broken test suite since scenario in
which deadlock can occur and assertion fails is not covered by tests.
sql/lock.cc:
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object exclusive metadata lock is acquired.
Now this has to be done by explicit calls outside of MDL subsystem.
sql/log.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/mdl.cc:
Changed metadata locking subsystem to use mutex per lock and condition
variable per context instead of one mutex and one conditional variable
for the whole subsystem.
Changed approach to handling of global metadata locks. Instead of
implicitly acquiring intention exclusive locks when user requests
per-object upgradeable or exclusive locks now we require them to be
acquired explicitly in the same way as ordinary metadata locks.
In fact global lock are now ordinary metadata locks in new GLOBAL
namespace.
To implement these changes:
- Removed LOCK_mdl mutex and COND_mdl condition variable.
- Introduced MDL_lock::m_mutex mutexes which protect individual lock
objects.
- Replaced mdl_locks hash with MDL_map class, which has hash for
MDL_lock objects as a member and separate mutex which protects this
hash. Methods of this class allow to find(), find_or_create() or
remove() MDL_lock objects in concurrency-friendly fashion (i.e.
for most common operation, find_or_create(), we don't acquire
MDL_lock::m_mutex while holding MDL_map::m_mutex. Thanks to MikaelR
for this idea and benchmarks!). Added three auxiliary members to
MDL_lock class (m_is_destroyed, m_ref_usage, m_ref_release) to
support this concurrency-friendly behavior.
- Introduced MDL_context::m_ctx_wakeup_cond condition variable to be
used for waiting until this context's pending request can be
satisfied or its thread has to perform actions to resolve potential
deadlock. Context which want to wait add ticket corresponding to the
request to an appropriate queue of waiters in MDL_lock object so
they can be noticed when other contexts change state of lock and be
awaken by them by signalling on MDL_context::m_ctx_wakeup_cond.
As consequence MDL_ticket objects has to be used for any waiting
in metadata locking subsystem including one which happens in
MDL_context::wait_for_locks() method.
Another consequence is that MDL_context is no longer copyable and
can't be saved/restored when working with system tables.
- Made MDL_lock an abstract class, which delegates specifying exact
compatibility matrix to its descendants. Added MDL_global_lock child
class for global lock (The old is_lock_type_compatible() method
became can_grant_lock() method of this class). Added MDL_object_lock
class to represent per-object lock (The old MDL_lock::can_grant_lock()
became its method). Choice between two classes happens based on MDL
namespace in MDL_lock::create() method.
- Got rid of MDL_lock::type member as its meaning became ambigous for
global locks.
- To simplify waking up of contexts waiting for lock split waiting queue
in MDL_lock class in two queues. One for pending requests for exclusive
(including intention exclusive) locks and another for requests for
shared locks.
- Added virtual wake_up_waiters() method to MDL_lock, MDL_global_lock and
MDL_object_lock classes which allows to wake up waiting contexts after
state of lock changes. Replaced old duplicated code with calls to this
method.
- Adjusted MDL_context::try_acquire_shared_lock()/exclusive_lock()/
global_shared_lock(), MDL_ticket::upgrade_shared_lock_to_exclusive_lock()
and MDL_context::release_ticket() methods to use MDL_map and
MDL_lock::m_mutex instead of single LOCK_mdl mutex and wake up
waiters according to the approach described above. The latter method
also was renamed to MDL_context::release_lock().
- Changed MDL_context::try_acquire_shared_lock()/exclusive_lock() and
release_lock() not to handle global locks. They are now supposed to
be taken explicitly like ordinary metadata locks.
- Added helper MDL_context::try_acquire_global_intention_exclusive_lock()
and acquire_global_intention_exclusive_lock() methods.
- Moved common code from MDL_context::acquire_global_shared_lock() and
acquire_global_intention_exclusive_lock() to new method -
MDL_context::acquire_lock_impl().
- Moved common code from MDL_context::try_acquire_shared_lock(),
try_acquire_global_intention_exclusive_lock()/exclusive_lock()
to MDL_context::try_acquire_lock_impl().
- Since acquiring of several exclusive locks can no longer happen under
single LOCK_mdl mutex the approach to it had to be changed. Now we do
it in one by one fashion. This is done in alphabetical order to avoid
deadlocks. Changed MDL_context::acquire_exclusive_locks() accordingly
(as part of this change moved code responsible for acquiring single
exclusive lock to new MDL_context::acquire_exclusive_lock_impl()
method).
- Since we no longer have single LOCK_mdl mutex which protects all
MDL_context::m_is_waiting_in_mdl members using these members to
determine if we have really awaken context holding conflicting
shared lock became inconvinient. Got rid of this member and changed
notify_shared_lock() helper function and process of acquiring
of/upgrading to exclusive lock not to rely on such information.
Now in MDL_context::acquire_exclusive_lock_impl() and
MDL_ticket::upgrade_shared_lock_to_exclusive_lock() we simply
re-try to wake up threads holding conflicting shared locks after
small time out.
- Adjusted MDL_context::can_wait_lead_to_deadlock() and
MDL_ticket::has_pending_conflicting_lock() to use per-lock
mutexes instead of LOCK_mdl. To do this introduced
MDL_lock::has_pending_exclusive_lock() method.
sql/mdl.h:
Changed metadata locking subsystem to use mutex per lock and condition
variable per context instead of one mutex and one conditional variable
for the whole subsystem. In order to implement this change:
- Added MDL_key::cmp() method to be able to sort MDL_key objects
alphabetically. Changed length fields in MDL_key class to uint16
as 16-bit is enough for length of any key.
- Changed MDL_ticket::get_ctx() to return pointer to non-const
object in order to be able to use MDL_context::awake() method
for such contexts.
- Got rid of unlocked versions of can_wait_lead_to_deadlock()/
has_pending_conflicting_lock() methods in MDL_context and
MDL_ticket. We no longer has single mutex which protects all
locks. Thus one always has to use versions of these methods
which acquire per-lock mutexes.
- MDL_request_list type of list now counts its elements.
- Added MDL_context::m_ctx_wakeup_cond condition variable to be used
for waiting until this context's pending request can be satisfied
or its thread has to perform actions to resolve potential deadlock.
Added awake() method to wake up context from such wait.
Addition of condition variable made MDL_context uncopyable.
As result we no longer can save/restore MDL_context when working
with system tables. Instead we create MDL savepoint before opening
those tables and rollback to it once they are closed.
- MDL_context::release_ticket() became release_lock() method.
- Added auxiliary MDL_context::acquire_exclusive_lock_impl() method
which does all necessary work to acquire exclusive lock on one object
but should not be used directly as it does not enforce any asserts
ensuring that no deadlocks are possible.
- Since we no longer need to know if thread trying to acquire exclusive
lock managed to wake up any threads having conflicting shared locks
(as, anyway, we will try to wake up such threads again shortly)
- MDL_context::m_is_waiting_in_mdl member became unnecessary and
notify_shared_lock() no longer needs to be friend of MDL_context.
Changed approach to handling of global metadata locks. Instead of
implicitly acquiring intention exclusive locks when user requests
per-object upgradeable or exclusive locks now we require them to be
acquired explicitly in the same way as ordinary metadata locks.
- Added new GLOBAL namespace for such locks.
- Added new type of lock to be requested MDL_INTENTION_EXCLISIVE.
- Added MDL_context::try_acquire_global_intention_exclusive_lock()
and acquire_global_intention_exclusive_lock() methods.
- Moved common code from MDL_context::acquire_global_shared_lock()
and acquire_global_intention_exclusive_lock() to new method -
MDL_context::acquire_lock_impl().
- Moved common code from MDL_context::try_acquire_shared_lock(),
try_acquire_global_intention_exclusive_lock()/exclusive_lock()
to MDL_context::try_acquire_lock_impl().
- Added helper MDL_context::is_global_lock_owner() method to be
able easily to find what kind of global lock this context holds.
- MDL_context::m_has_global_shared_lock became unnecessary as
global read lock is now represented by ordinary ticket.
- Removed assert in MDL_context::set_lt_or_ha_sentinel() which became
false for cases when we execute LOCK TABLES under global read lock
mode.
sql/mysql_priv.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result calls
opening/closing system tables were changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sp.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sp.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sql_base.cc:
close_thread_tables():
Since we no longer use separate MDL_context for opening system
tables we need to avoid releasing all transaction locks when
closing system table. Releasing metadata lock on system table
is now responsibility of THD::restore_backup_open_tables_state().
open_table_get_mdl_lock(),
Open_table_context::recover_from_failed_open():
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object upgradable or exclusive metadata lock is
acquired. So this have to be done explicitly from these calls.
Changed Open_table_context class to store MDL_request object for
global intention exclusive lock acquired when opening tables.
open_table():
Do not release metadata lock if we have failed to open table as
this lock might have been acquired by one of previous statements
in transaction, and therefore should not be released.
open_system_tables_for_read()/close_system_tables()/
open_performance_schema_table():
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
close_performance_schema_table():
Got rid of duplicated code.
sql/sql_class.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. Also releasing
metadata lock on system table is now responsibility of
THD::restore_backup_open_tables_state().
Adjusted assert in THD::cleanup() to take into account fact that now
we also use MDL sentinel for global read lock.
sql/sql_class.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. As result:
- 'mdl_context' member was moved out of Open_tables_state to THD class.
enter_locked_tables_mode()/leave_locked_tables_mode() had to follow.
- Methods of THD responsible for saving/restoring open table state were
changed to use Open_tables_backup class which in addition to
Open_table_state has a member for this savepoint.
Changed Open_table_context class to store MDL_request object for
global intention exclusive lock acquired when opening tables.
sql/sql_delete.cc:
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object exclusive metadata lock is acquired.
Now this has to be done by explicit calls outside of MDL subsystem.
sql/sql_help.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sql_parse.cc:
Adjusted assert reload_acl_and_cache() to the fact that global read
lock now takes full-blown metadata lock.
sql/sql_plist.h:
Added support for element counting to I_P_List list template.
One can use policy classes to specify if such counting is needed
or not needed for particular list.
sql/sql_show.cc:
Instead of using separate MDL_context for opening tables for I_S
purposes we now create MDL savepoint in the main context before
opening tables and rollback to this savepoint after closing them.
To support this and similar change for system tables methods of
THD responsible for saving/restoring open table state were changed
to use Open_tables_backup class which in addition to Open_table_state
has a member for this savepoint. As result code opening/closing tables
for I_S purposes was changed to use Open_tables_backup instead of
Open_table_state class as well.
sql/sql_table.cc:
mysql_rm_tables_part2():
Since now global intention exclusive metadata lock is ordinary
metadata lock we no longer can rely that by releasing MDL locks
on all tables we will release all locks acquired by this routine.
So in non-LOCK-TABLES mode we have to release all locks acquired
explicitly.
prepare_for_repair(), mysql_alter_table():
MDL subsystem no longer implicitly acquires global intention
exclusive metadata lock when per-object exclusive metadata lock
is acquired. Now this has to be done by explicit calls outside of
MDL subsystem.
sql/tztime.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
Also changed code not to use special mechanism for open system tables
when it is not really necessary.
2010-01-21 21:43:03 +01:00
# When HANDLER t0 OPEN does open_tables(), it calls
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
# mysql_ha_flush(), which in turn closes the open HANDLER for t1.
# RENAME TABLE gets unblocked. If it gets scheduled quickly
# and manages to complete before open_tables()
Patch that changes metadata locking subsystem to use mutex per lock and
condition variable per context instead of one mutex and one conditional
variable for the whole subsystem.
This should increase concurrency in this subsystem.
It also opens the way for further changes which are necessary to solve
such bugs as bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock"
and bug #37346 "innodb does not detect deadlock between update and alter
table".
Two other notable changes done by this patch:
- MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object metadata lock is acquired. Now this has
to be done by explicit calls outside of MDL subsystem.
- Instead of using separate MDL_context for opening system tables/tables
for purposes of I_S we now create MDL savepoint in the main context
before opening tables and rollback to this savepoint after closing
them. This means that it is now possible to get ER_LOCK_DEADLOCK error
even not inside a transaction. This might happen in unlikely case when
one runs DDL on one of system tables while also running DDL on some
other tables. Cases when this ER_LOCK_DEADLOCK error is not justified
will be addressed by advanced deadlock detector for MDL subsystem which
we plan to implement.
mysql-test/include/handler.inc:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/handler_innodb.result:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/handler_myisam.result:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/mdl_sync.result:
Adjusted mdl_sync.test to the fact that exclusive metadata locks on
tables are now acquired according to alphabetical order of fully
qualified table names instead of order in which tables are mentioned
in statement.
mysql-test/t/mdl_sync.test:
Adjusted mdl_sync.test to the fact that exclusive metadata locks on
tables are now acquired according to alphabetical order of fully
qualified table names instead of order in which tables are mentioned
in statement.
sql/events.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/ha_ndbcluster.cc:
Since manipulations with open table state no longer install proxy
MDL_context it does not make sense to perform them in order to
satisfy assert in mysql_rm_tables_part2(). Removed them per agreement
with Cluster team. This has not broken test suite since scenario in
which deadlock can occur and assertion fails is not covered by tests.
sql/lock.cc:
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object exclusive metadata lock is acquired.
Now this has to be done by explicit calls outside of MDL subsystem.
sql/log.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/mdl.cc:
Changed metadata locking subsystem to use mutex per lock and condition
variable per context instead of one mutex and one conditional variable
for the whole subsystem.
Changed approach to handling of global metadata locks. Instead of
implicitly acquiring intention exclusive locks when user requests
per-object upgradeable or exclusive locks now we require them to be
acquired explicitly in the same way as ordinary metadata locks.
In fact global lock are now ordinary metadata locks in new GLOBAL
namespace.
To implement these changes:
- Removed LOCK_mdl mutex and COND_mdl condition variable.
- Introduced MDL_lock::m_mutex mutexes which protect individual lock
objects.
- Replaced mdl_locks hash with MDL_map class, which has hash for
MDL_lock objects as a member and separate mutex which protects this
hash. Methods of this class allow to find(), find_or_create() or
remove() MDL_lock objects in concurrency-friendly fashion (i.e.
for most common operation, find_or_create(), we don't acquire
MDL_lock::m_mutex while holding MDL_map::m_mutex. Thanks to MikaelR
for this idea and benchmarks!). Added three auxiliary members to
MDL_lock class (m_is_destroyed, m_ref_usage, m_ref_release) to
support this concurrency-friendly behavior.
- Introduced MDL_context::m_ctx_wakeup_cond condition variable to be
used for waiting until this context's pending request can be
satisfied or its thread has to perform actions to resolve potential
deadlock. Context which want to wait add ticket corresponding to the
request to an appropriate queue of waiters in MDL_lock object so
they can be noticed when other contexts change state of lock and be
awaken by them by signalling on MDL_context::m_ctx_wakeup_cond.
As consequence MDL_ticket objects has to be used for any waiting
in metadata locking subsystem including one which happens in
MDL_context::wait_for_locks() method.
Another consequence is that MDL_context is no longer copyable and
can't be saved/restored when working with system tables.
- Made MDL_lock an abstract class, which delegates specifying exact
compatibility matrix to its descendants. Added MDL_global_lock child
class for global lock (The old is_lock_type_compatible() method
became can_grant_lock() method of this class). Added MDL_object_lock
class to represent per-object lock (The old MDL_lock::can_grant_lock()
became its method). Choice between two classes happens based on MDL
namespace in MDL_lock::create() method.
- Got rid of MDL_lock::type member as its meaning became ambigous for
global locks.
- To simplify waking up of contexts waiting for lock split waiting queue
in MDL_lock class in two queues. One for pending requests for exclusive
(including intention exclusive) locks and another for requests for
shared locks.
- Added virtual wake_up_waiters() method to MDL_lock, MDL_global_lock and
MDL_object_lock classes which allows to wake up waiting contexts after
state of lock changes. Replaced old duplicated code with calls to this
method.
- Adjusted MDL_context::try_acquire_shared_lock()/exclusive_lock()/
global_shared_lock(), MDL_ticket::upgrade_shared_lock_to_exclusive_lock()
and MDL_context::release_ticket() methods to use MDL_map and
MDL_lock::m_mutex instead of single LOCK_mdl mutex and wake up
waiters according to the approach described above. The latter method
also was renamed to MDL_context::release_lock().
- Changed MDL_context::try_acquire_shared_lock()/exclusive_lock() and
release_lock() not to handle global locks. They are now supposed to
be taken explicitly like ordinary metadata locks.
- Added helper MDL_context::try_acquire_global_intention_exclusive_lock()
and acquire_global_intention_exclusive_lock() methods.
- Moved common code from MDL_context::acquire_global_shared_lock() and
acquire_global_intention_exclusive_lock() to new method -
MDL_context::acquire_lock_impl().
- Moved common code from MDL_context::try_acquire_shared_lock(),
try_acquire_global_intention_exclusive_lock()/exclusive_lock()
to MDL_context::try_acquire_lock_impl().
- Since acquiring of several exclusive locks can no longer happen under
single LOCK_mdl mutex the approach to it had to be changed. Now we do
it in one by one fashion. This is done in alphabetical order to avoid
deadlocks. Changed MDL_context::acquire_exclusive_locks() accordingly
(as part of this change moved code responsible for acquiring single
exclusive lock to new MDL_context::acquire_exclusive_lock_impl()
method).
- Since we no longer have single LOCK_mdl mutex which protects all
MDL_context::m_is_waiting_in_mdl members using these members to
determine if we have really awaken context holding conflicting
shared lock became inconvinient. Got rid of this member and changed
notify_shared_lock() helper function and process of acquiring
of/upgrading to exclusive lock not to rely on such information.
Now in MDL_context::acquire_exclusive_lock_impl() and
MDL_ticket::upgrade_shared_lock_to_exclusive_lock() we simply
re-try to wake up threads holding conflicting shared locks after
small time out.
- Adjusted MDL_context::can_wait_lead_to_deadlock() and
MDL_ticket::has_pending_conflicting_lock() to use per-lock
mutexes instead of LOCK_mdl. To do this introduced
MDL_lock::has_pending_exclusive_lock() method.
sql/mdl.h:
Changed metadata locking subsystem to use mutex per lock and condition
variable per context instead of one mutex and one conditional variable
for the whole subsystem. In order to implement this change:
- Added MDL_key::cmp() method to be able to sort MDL_key objects
alphabetically. Changed length fields in MDL_key class to uint16
as 16-bit is enough for length of any key.
- Changed MDL_ticket::get_ctx() to return pointer to non-const
object in order to be able to use MDL_context::awake() method
for such contexts.
- Got rid of unlocked versions of can_wait_lead_to_deadlock()/
has_pending_conflicting_lock() methods in MDL_context and
MDL_ticket. We no longer has single mutex which protects all
locks. Thus one always has to use versions of these methods
which acquire per-lock mutexes.
- MDL_request_list type of list now counts its elements.
- Added MDL_context::m_ctx_wakeup_cond condition variable to be used
for waiting until this context's pending request can be satisfied
or its thread has to perform actions to resolve potential deadlock.
Added awake() method to wake up context from such wait.
Addition of condition variable made MDL_context uncopyable.
As result we no longer can save/restore MDL_context when working
with system tables. Instead we create MDL savepoint before opening
those tables and rollback to it once they are closed.
- MDL_context::release_ticket() became release_lock() method.
- Added auxiliary MDL_context::acquire_exclusive_lock_impl() method
which does all necessary work to acquire exclusive lock on one object
but should not be used directly as it does not enforce any asserts
ensuring that no deadlocks are possible.
- Since we no longer need to know if thread trying to acquire exclusive
lock managed to wake up any threads having conflicting shared locks
(as, anyway, we will try to wake up such threads again shortly)
- MDL_context::m_is_waiting_in_mdl member became unnecessary and
notify_shared_lock() no longer needs to be friend of MDL_context.
Changed approach to handling of global metadata locks. Instead of
implicitly acquiring intention exclusive locks when user requests
per-object upgradeable or exclusive locks now we require them to be
acquired explicitly in the same way as ordinary metadata locks.
- Added new GLOBAL namespace for such locks.
- Added new type of lock to be requested MDL_INTENTION_EXCLISIVE.
- Added MDL_context::try_acquire_global_intention_exclusive_lock()
and acquire_global_intention_exclusive_lock() methods.
- Moved common code from MDL_context::acquire_global_shared_lock()
and acquire_global_intention_exclusive_lock() to new method -
MDL_context::acquire_lock_impl().
- Moved common code from MDL_context::try_acquire_shared_lock(),
try_acquire_global_intention_exclusive_lock()/exclusive_lock()
to MDL_context::try_acquire_lock_impl().
- Added helper MDL_context::is_global_lock_owner() method to be
able easily to find what kind of global lock this context holds.
- MDL_context::m_has_global_shared_lock became unnecessary as
global read lock is now represented by ordinary ticket.
- Removed assert in MDL_context::set_lt_or_ha_sentinel() which became
false for cases when we execute LOCK TABLES under global read lock
mode.
sql/mysql_priv.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result calls
opening/closing system tables were changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sp.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sp.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sql_base.cc:
close_thread_tables():
Since we no longer use separate MDL_context for opening system
tables we need to avoid releasing all transaction locks when
closing system table. Releasing metadata lock on system table
is now responsibility of THD::restore_backup_open_tables_state().
open_table_get_mdl_lock(),
Open_table_context::recover_from_failed_open():
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object upgradable or exclusive metadata lock is
acquired. So this have to be done explicitly from these calls.
Changed Open_table_context class to store MDL_request object for
global intention exclusive lock acquired when opening tables.
open_table():
Do not release metadata lock if we have failed to open table as
this lock might have been acquired by one of previous statements
in transaction, and therefore should not be released.
open_system_tables_for_read()/close_system_tables()/
open_performance_schema_table():
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
close_performance_schema_table():
Got rid of duplicated code.
sql/sql_class.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. Also releasing
metadata lock on system table is now responsibility of
THD::restore_backup_open_tables_state().
Adjusted assert in THD::cleanup() to take into account fact that now
we also use MDL sentinel for global read lock.
sql/sql_class.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. As result:
- 'mdl_context' member was moved out of Open_tables_state to THD class.
enter_locked_tables_mode()/leave_locked_tables_mode() had to follow.
- Methods of THD responsible for saving/restoring open table state were
changed to use Open_tables_backup class which in addition to
Open_table_state has a member for this savepoint.
Changed Open_table_context class to store MDL_request object for
global intention exclusive lock acquired when opening tables.
sql/sql_delete.cc:
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object exclusive metadata lock is acquired.
Now this has to be done by explicit calls outside of MDL subsystem.
sql/sql_help.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sql_parse.cc:
Adjusted assert reload_acl_and_cache() to the fact that global read
lock now takes full-blown metadata lock.
sql/sql_plist.h:
Added support for element counting to I_P_List list template.
One can use policy classes to specify if such counting is needed
or not needed for particular list.
sql/sql_show.cc:
Instead of using separate MDL_context for opening tables for I_S
purposes we now create MDL savepoint in the main context before
opening tables and rollback to this savepoint after closing them.
To support this and similar change for system tables methods of
THD responsible for saving/restoring open table state were changed
to use Open_tables_backup class which in addition to Open_table_state
has a member for this savepoint. As result code opening/closing tables
for I_S purposes was changed to use Open_tables_backup instead of
Open_table_state class as well.
sql/sql_table.cc:
mysql_rm_tables_part2():
Since now global intention exclusive metadata lock is ordinary
metadata lock we no longer can rely that by releasing MDL locks
on all tables we will release all locks acquired by this routine.
So in non-LOCK-TABLES mode we have to release all locks acquired
explicitly.
prepare_for_repair(), mysql_alter_table():
MDL subsystem no longer implicitly acquires global intention
exclusive metadata lock when per-object exclusive metadata lock
is acquired. Now this has to be done by explicit calls outside of
MDL subsystem.
sql/tztime.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
Also changed code not to use special mechanism for open system tables
when it is not really necessary.
2010-01-21 21:43:03 +01:00
# of HANDLER t0 OPEN, open_tables() and therefore the whole
# HANDLER t0 OPEN succeeds. Otherwise open_tables()
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
# notices a pending or active exclusive metadata lock on t2
Patch that changes metadata locking subsystem to use mutex per lock and
condition variable per context instead of one mutex and one conditional
variable for the whole subsystem.
This should increase concurrency in this subsystem.
It also opens the way for further changes which are necessary to solve
such bugs as bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock"
and bug #37346 "innodb does not detect deadlock between update and alter
table".
Two other notable changes done by this patch:
- MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object metadata lock is acquired. Now this has
to be done by explicit calls outside of MDL subsystem.
- Instead of using separate MDL_context for opening system tables/tables
for purposes of I_S we now create MDL savepoint in the main context
before opening tables and rollback to this savepoint after closing
them. This means that it is now possible to get ER_LOCK_DEADLOCK error
even not inside a transaction. This might happen in unlikely case when
one runs DDL on one of system tables while also running DDL on some
other tables. Cases when this ER_LOCK_DEADLOCK error is not justified
will be addressed by advanced deadlock detector for MDL subsystem which
we plan to implement.
mysql-test/include/handler.inc:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/handler_innodb.result:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/handler_myisam.result:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/mdl_sync.result:
Adjusted mdl_sync.test to the fact that exclusive metadata locks on
tables are now acquired according to alphabetical order of fully
qualified table names instead of order in which tables are mentioned
in statement.
mysql-test/t/mdl_sync.test:
Adjusted mdl_sync.test to the fact that exclusive metadata locks on
tables are now acquired according to alphabetical order of fully
qualified table names instead of order in which tables are mentioned
in statement.
sql/events.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/ha_ndbcluster.cc:
Since manipulations with open table state no longer install proxy
MDL_context it does not make sense to perform them in order to
satisfy assert in mysql_rm_tables_part2(). Removed them per agreement
with Cluster team. This has not broken test suite since scenario in
which deadlock can occur and assertion fails is not covered by tests.
sql/lock.cc:
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object exclusive metadata lock is acquired.
Now this has to be done by explicit calls outside of MDL subsystem.
sql/log.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/mdl.cc:
Changed metadata locking subsystem to use mutex per lock and condition
variable per context instead of one mutex and one conditional variable
for the whole subsystem.
Changed approach to handling of global metadata locks. Instead of
implicitly acquiring intention exclusive locks when user requests
per-object upgradeable or exclusive locks now we require them to be
acquired explicitly in the same way as ordinary metadata locks.
In fact global lock are now ordinary metadata locks in new GLOBAL
namespace.
To implement these changes:
- Removed LOCK_mdl mutex and COND_mdl condition variable.
- Introduced MDL_lock::m_mutex mutexes which protect individual lock
objects.
- Replaced mdl_locks hash with MDL_map class, which has hash for
MDL_lock objects as a member and separate mutex which protects this
hash. Methods of this class allow to find(), find_or_create() or
remove() MDL_lock objects in concurrency-friendly fashion (i.e.
for most common operation, find_or_create(), we don't acquire
MDL_lock::m_mutex while holding MDL_map::m_mutex. Thanks to MikaelR
for this idea and benchmarks!). Added three auxiliary members to
MDL_lock class (m_is_destroyed, m_ref_usage, m_ref_release) to
support this concurrency-friendly behavior.
- Introduced MDL_context::m_ctx_wakeup_cond condition variable to be
used for waiting until this context's pending request can be
satisfied or its thread has to perform actions to resolve potential
deadlock. Context which want to wait add ticket corresponding to the
request to an appropriate queue of waiters in MDL_lock object so
they can be noticed when other contexts change state of lock and be
awaken by them by signalling on MDL_context::m_ctx_wakeup_cond.
As consequence MDL_ticket objects has to be used for any waiting
in metadata locking subsystem including one which happens in
MDL_context::wait_for_locks() method.
Another consequence is that MDL_context is no longer copyable and
can't be saved/restored when working with system tables.
- Made MDL_lock an abstract class, which delegates specifying exact
compatibility matrix to its descendants. Added MDL_global_lock child
class for global lock (The old is_lock_type_compatible() method
became can_grant_lock() method of this class). Added MDL_object_lock
class to represent per-object lock (The old MDL_lock::can_grant_lock()
became its method). Choice between two classes happens based on MDL
namespace in MDL_lock::create() method.
- Got rid of MDL_lock::type member as its meaning became ambigous for
global locks.
- To simplify waking up of contexts waiting for lock split waiting queue
in MDL_lock class in two queues. One for pending requests for exclusive
(including intention exclusive) locks and another for requests for
shared locks.
- Added virtual wake_up_waiters() method to MDL_lock, MDL_global_lock and
MDL_object_lock classes which allows to wake up waiting contexts after
state of lock changes. Replaced old duplicated code with calls to this
method.
- Adjusted MDL_context::try_acquire_shared_lock()/exclusive_lock()/
global_shared_lock(), MDL_ticket::upgrade_shared_lock_to_exclusive_lock()
and MDL_context::release_ticket() methods to use MDL_map and
MDL_lock::m_mutex instead of single LOCK_mdl mutex and wake up
waiters according to the approach described above. The latter method
also was renamed to MDL_context::release_lock().
- Changed MDL_context::try_acquire_shared_lock()/exclusive_lock() and
release_lock() not to handle global locks. They are now supposed to
be taken explicitly like ordinary metadata locks.
- Added helper MDL_context::try_acquire_global_intention_exclusive_lock()
and acquire_global_intention_exclusive_lock() methods.
- Moved common code from MDL_context::acquire_global_shared_lock() and
acquire_global_intention_exclusive_lock() to new method -
MDL_context::acquire_lock_impl().
- Moved common code from MDL_context::try_acquire_shared_lock(),
try_acquire_global_intention_exclusive_lock()/exclusive_lock()
to MDL_context::try_acquire_lock_impl().
- Since acquiring of several exclusive locks can no longer happen under
single LOCK_mdl mutex the approach to it had to be changed. Now we do
it in one by one fashion. This is done in alphabetical order to avoid
deadlocks. Changed MDL_context::acquire_exclusive_locks() accordingly
(as part of this change moved code responsible for acquiring single
exclusive lock to new MDL_context::acquire_exclusive_lock_impl()
method).
- Since we no longer have single LOCK_mdl mutex which protects all
MDL_context::m_is_waiting_in_mdl members using these members to
determine if we have really awaken context holding conflicting
shared lock became inconvinient. Got rid of this member and changed
notify_shared_lock() helper function and process of acquiring
of/upgrading to exclusive lock not to rely on such information.
Now in MDL_context::acquire_exclusive_lock_impl() and
MDL_ticket::upgrade_shared_lock_to_exclusive_lock() we simply
re-try to wake up threads holding conflicting shared locks after
small time out.
- Adjusted MDL_context::can_wait_lead_to_deadlock() and
MDL_ticket::has_pending_conflicting_lock() to use per-lock
mutexes instead of LOCK_mdl. To do this introduced
MDL_lock::has_pending_exclusive_lock() method.
sql/mdl.h:
Changed metadata locking subsystem to use mutex per lock and condition
variable per context instead of one mutex and one conditional variable
for the whole subsystem. In order to implement this change:
- Added MDL_key::cmp() method to be able to sort MDL_key objects
alphabetically. Changed length fields in MDL_key class to uint16
as 16-bit is enough for length of any key.
- Changed MDL_ticket::get_ctx() to return pointer to non-const
object in order to be able to use MDL_context::awake() method
for such contexts.
- Got rid of unlocked versions of can_wait_lead_to_deadlock()/
has_pending_conflicting_lock() methods in MDL_context and
MDL_ticket. We no longer has single mutex which protects all
locks. Thus one always has to use versions of these methods
which acquire per-lock mutexes.
- MDL_request_list type of list now counts its elements.
- Added MDL_context::m_ctx_wakeup_cond condition variable to be used
for waiting until this context's pending request can be satisfied
or its thread has to perform actions to resolve potential deadlock.
Added awake() method to wake up context from such wait.
Addition of condition variable made MDL_context uncopyable.
As result we no longer can save/restore MDL_context when working
with system tables. Instead we create MDL savepoint before opening
those tables and rollback to it once they are closed.
- MDL_context::release_ticket() became release_lock() method.
- Added auxiliary MDL_context::acquire_exclusive_lock_impl() method
which does all necessary work to acquire exclusive lock on one object
but should not be used directly as it does not enforce any asserts
ensuring that no deadlocks are possible.
- Since we no longer need to know if thread trying to acquire exclusive
lock managed to wake up any threads having conflicting shared locks
(as, anyway, we will try to wake up such threads again shortly)
- MDL_context::m_is_waiting_in_mdl member became unnecessary and
notify_shared_lock() no longer needs to be friend of MDL_context.
Changed approach to handling of global metadata locks. Instead of
implicitly acquiring intention exclusive locks when user requests
per-object upgradeable or exclusive locks now we require them to be
acquired explicitly in the same way as ordinary metadata locks.
- Added new GLOBAL namespace for such locks.
- Added new type of lock to be requested MDL_INTENTION_EXCLISIVE.
- Added MDL_context::try_acquire_global_intention_exclusive_lock()
and acquire_global_intention_exclusive_lock() methods.
- Moved common code from MDL_context::acquire_global_shared_lock()
and acquire_global_intention_exclusive_lock() to new method -
MDL_context::acquire_lock_impl().
- Moved common code from MDL_context::try_acquire_shared_lock(),
try_acquire_global_intention_exclusive_lock()/exclusive_lock()
to MDL_context::try_acquire_lock_impl().
- Added helper MDL_context::is_global_lock_owner() method to be
able easily to find what kind of global lock this context holds.
- MDL_context::m_has_global_shared_lock became unnecessary as
global read lock is now represented by ordinary ticket.
- Removed assert in MDL_context::set_lt_or_ha_sentinel() which became
false for cases when we execute LOCK TABLES under global read lock
mode.
sql/mysql_priv.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result calls
opening/closing system tables were changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sp.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sp.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sql_base.cc:
close_thread_tables():
Since we no longer use separate MDL_context for opening system
tables we need to avoid releasing all transaction locks when
closing system table. Releasing metadata lock on system table
is now responsibility of THD::restore_backup_open_tables_state().
open_table_get_mdl_lock(),
Open_table_context::recover_from_failed_open():
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object upgradable or exclusive metadata lock is
acquired. So this have to be done explicitly from these calls.
Changed Open_table_context class to store MDL_request object for
global intention exclusive lock acquired when opening tables.
open_table():
Do not release metadata lock if we have failed to open table as
this lock might have been acquired by one of previous statements
in transaction, and therefore should not be released.
open_system_tables_for_read()/close_system_tables()/
open_performance_schema_table():
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
close_performance_schema_table():
Got rid of duplicated code.
sql/sql_class.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. Also releasing
metadata lock on system table is now responsibility of
THD::restore_backup_open_tables_state().
Adjusted assert in THD::cleanup() to take into account fact that now
we also use MDL sentinel for global read lock.
sql/sql_class.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. As result:
- 'mdl_context' member was moved out of Open_tables_state to THD class.
enter_locked_tables_mode()/leave_locked_tables_mode() had to follow.
- Methods of THD responsible for saving/restoring open table state were
changed to use Open_tables_backup class which in addition to
Open_table_state has a member for this savepoint.
Changed Open_table_context class to store MDL_request object for
global intention exclusive lock acquired when opening tables.
sql/sql_delete.cc:
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object exclusive metadata lock is acquired.
Now this has to be done by explicit calls outside of MDL subsystem.
sql/sql_help.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sql_parse.cc:
Adjusted assert reload_acl_and_cache() to the fact that global read
lock now takes full-blown metadata lock.
sql/sql_plist.h:
Added support for element counting to I_P_List list template.
One can use policy classes to specify if such counting is needed
or not needed for particular list.
sql/sql_show.cc:
Instead of using separate MDL_context for opening tables for I_S
purposes we now create MDL savepoint in the main context before
opening tables and rollback to this savepoint after closing them.
To support this and similar change for system tables methods of
THD responsible for saving/restoring open table state were changed
to use Open_tables_backup class which in addition to Open_table_state
has a member for this savepoint. As result code opening/closing tables
for I_S purposes was changed to use Open_tables_backup instead of
Open_table_state class as well.
sql/sql_table.cc:
mysql_rm_tables_part2():
Since now global intention exclusive metadata lock is ordinary
metadata lock we no longer can rely that by releasing MDL locks
on all tables we will release all locks acquired by this routine.
So in non-LOCK-TABLES mode we have to release all locks acquired
explicitly.
prepare_for_repair(), mysql_alter_table():
MDL subsystem no longer implicitly acquires global intention
exclusive metadata lock when per-object exclusive metadata lock
is acquired. Now this has to be done by explicit calls outside of
MDL subsystem.
sql/tztime.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
Also changed code not to use special mechanism for open system tables
when it is not really necessary.
2010-01-21 21:43:03 +01:00
# and the whole HANDLER t0 OPEN fails with ER_LOCK_DEADLOCK
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
# error.
#
Patch that changes metadata locking subsystem to use mutex per lock and
condition variable per context instead of one mutex and one conditional
variable for the whole subsystem.
This should increase concurrency in this subsystem.
It also opens the way for further changes which are necessary to solve
such bugs as bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock"
and bug #37346 "innodb does not detect deadlock between update and alter
table".
Two other notable changes done by this patch:
- MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object metadata lock is acquired. Now this has
to be done by explicit calls outside of MDL subsystem.
- Instead of using separate MDL_context for opening system tables/tables
for purposes of I_S we now create MDL savepoint in the main context
before opening tables and rollback to this savepoint after closing
them. This means that it is now possible to get ER_LOCK_DEADLOCK error
even not inside a transaction. This might happen in unlikely case when
one runs DDL on one of system tables while also running DDL on some
other tables. Cases when this ER_LOCK_DEADLOCK error is not justified
will be addressed by advanced deadlock detector for MDL subsystem which
we plan to implement.
mysql-test/include/handler.inc:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/handler_innodb.result:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/handler_myisam.result:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/mdl_sync.result:
Adjusted mdl_sync.test to the fact that exclusive metadata locks on
tables are now acquired according to alphabetical order of fully
qualified table names instead of order in which tables are mentioned
in statement.
mysql-test/t/mdl_sync.test:
Adjusted mdl_sync.test to the fact that exclusive metadata locks on
tables are now acquired according to alphabetical order of fully
qualified table names instead of order in which tables are mentioned
in statement.
sql/events.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/ha_ndbcluster.cc:
Since manipulations with open table state no longer install proxy
MDL_context it does not make sense to perform them in order to
satisfy assert in mysql_rm_tables_part2(). Removed them per agreement
with Cluster team. This has not broken test suite since scenario in
which deadlock can occur and assertion fails is not covered by tests.
sql/lock.cc:
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object exclusive metadata lock is acquired.
Now this has to be done by explicit calls outside of MDL subsystem.
sql/log.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/mdl.cc:
Changed metadata locking subsystem to use mutex per lock and condition
variable per context instead of one mutex and one conditional variable
for the whole subsystem.
Changed approach to handling of global metadata locks. Instead of
implicitly acquiring intention exclusive locks when user requests
per-object upgradeable or exclusive locks now we require them to be
acquired explicitly in the same way as ordinary metadata locks.
In fact global lock are now ordinary metadata locks in new GLOBAL
namespace.
To implement these changes:
- Removed LOCK_mdl mutex and COND_mdl condition variable.
- Introduced MDL_lock::m_mutex mutexes which protect individual lock
objects.
- Replaced mdl_locks hash with MDL_map class, which has hash for
MDL_lock objects as a member and separate mutex which protects this
hash. Methods of this class allow to find(), find_or_create() or
remove() MDL_lock objects in concurrency-friendly fashion (i.e.
for most common operation, find_or_create(), we don't acquire
MDL_lock::m_mutex while holding MDL_map::m_mutex. Thanks to MikaelR
for this idea and benchmarks!). Added three auxiliary members to
MDL_lock class (m_is_destroyed, m_ref_usage, m_ref_release) to
support this concurrency-friendly behavior.
- Introduced MDL_context::m_ctx_wakeup_cond condition variable to be
used for waiting until this context's pending request can be
satisfied or its thread has to perform actions to resolve potential
deadlock. Context which want to wait add ticket corresponding to the
request to an appropriate queue of waiters in MDL_lock object so
they can be noticed when other contexts change state of lock and be
awaken by them by signalling on MDL_context::m_ctx_wakeup_cond.
As consequence MDL_ticket objects has to be used for any waiting
in metadata locking subsystem including one which happens in
MDL_context::wait_for_locks() method.
Another consequence is that MDL_context is no longer copyable and
can't be saved/restored when working with system tables.
- Made MDL_lock an abstract class, which delegates specifying exact
compatibility matrix to its descendants. Added MDL_global_lock child
class for global lock (The old is_lock_type_compatible() method
became can_grant_lock() method of this class). Added MDL_object_lock
class to represent per-object lock (The old MDL_lock::can_grant_lock()
became its method). Choice between two classes happens based on MDL
namespace in MDL_lock::create() method.
- Got rid of MDL_lock::type member as its meaning became ambigous for
global locks.
- To simplify waking up of contexts waiting for lock split waiting queue
in MDL_lock class in two queues. One for pending requests for exclusive
(including intention exclusive) locks and another for requests for
shared locks.
- Added virtual wake_up_waiters() method to MDL_lock, MDL_global_lock and
MDL_object_lock classes which allows to wake up waiting contexts after
state of lock changes. Replaced old duplicated code with calls to this
method.
- Adjusted MDL_context::try_acquire_shared_lock()/exclusive_lock()/
global_shared_lock(), MDL_ticket::upgrade_shared_lock_to_exclusive_lock()
and MDL_context::release_ticket() methods to use MDL_map and
MDL_lock::m_mutex instead of single LOCK_mdl mutex and wake up
waiters according to the approach described above. The latter method
also was renamed to MDL_context::release_lock().
- Changed MDL_context::try_acquire_shared_lock()/exclusive_lock() and
release_lock() not to handle global locks. They are now supposed to
be taken explicitly like ordinary metadata locks.
- Added helper MDL_context::try_acquire_global_intention_exclusive_lock()
and acquire_global_intention_exclusive_lock() methods.
- Moved common code from MDL_context::acquire_global_shared_lock() and
acquire_global_intention_exclusive_lock() to new method -
MDL_context::acquire_lock_impl().
- Moved common code from MDL_context::try_acquire_shared_lock(),
try_acquire_global_intention_exclusive_lock()/exclusive_lock()
to MDL_context::try_acquire_lock_impl().
- Since acquiring of several exclusive locks can no longer happen under
single LOCK_mdl mutex the approach to it had to be changed. Now we do
it in one by one fashion. This is done in alphabetical order to avoid
deadlocks. Changed MDL_context::acquire_exclusive_locks() accordingly
(as part of this change moved code responsible for acquiring single
exclusive lock to new MDL_context::acquire_exclusive_lock_impl()
method).
- Since we no longer have single LOCK_mdl mutex which protects all
MDL_context::m_is_waiting_in_mdl members using these members to
determine if we have really awaken context holding conflicting
shared lock became inconvinient. Got rid of this member and changed
notify_shared_lock() helper function and process of acquiring
of/upgrading to exclusive lock not to rely on such information.
Now in MDL_context::acquire_exclusive_lock_impl() and
MDL_ticket::upgrade_shared_lock_to_exclusive_lock() we simply
re-try to wake up threads holding conflicting shared locks after
small time out.
- Adjusted MDL_context::can_wait_lead_to_deadlock() and
MDL_ticket::has_pending_conflicting_lock() to use per-lock
mutexes instead of LOCK_mdl. To do this introduced
MDL_lock::has_pending_exclusive_lock() method.
sql/mdl.h:
Changed metadata locking subsystem to use mutex per lock and condition
variable per context instead of one mutex and one conditional variable
for the whole subsystem. In order to implement this change:
- Added MDL_key::cmp() method to be able to sort MDL_key objects
alphabetically. Changed length fields in MDL_key class to uint16
as 16-bit is enough for length of any key.
- Changed MDL_ticket::get_ctx() to return pointer to non-const
object in order to be able to use MDL_context::awake() method
for such contexts.
- Got rid of unlocked versions of can_wait_lead_to_deadlock()/
has_pending_conflicting_lock() methods in MDL_context and
MDL_ticket. We no longer has single mutex which protects all
locks. Thus one always has to use versions of these methods
which acquire per-lock mutexes.
- MDL_request_list type of list now counts its elements.
- Added MDL_context::m_ctx_wakeup_cond condition variable to be used
for waiting until this context's pending request can be satisfied
or its thread has to perform actions to resolve potential deadlock.
Added awake() method to wake up context from such wait.
Addition of condition variable made MDL_context uncopyable.
As result we no longer can save/restore MDL_context when working
with system tables. Instead we create MDL savepoint before opening
those tables and rollback to it once they are closed.
- MDL_context::release_ticket() became release_lock() method.
- Added auxiliary MDL_context::acquire_exclusive_lock_impl() method
which does all necessary work to acquire exclusive lock on one object
but should not be used directly as it does not enforce any asserts
ensuring that no deadlocks are possible.
- Since we no longer need to know if thread trying to acquire exclusive
lock managed to wake up any threads having conflicting shared locks
(as, anyway, we will try to wake up such threads again shortly)
- MDL_context::m_is_waiting_in_mdl member became unnecessary and
notify_shared_lock() no longer needs to be friend of MDL_context.
Changed approach to handling of global metadata locks. Instead of
implicitly acquiring intention exclusive locks when user requests
per-object upgradeable or exclusive locks now we require them to be
acquired explicitly in the same way as ordinary metadata locks.
- Added new GLOBAL namespace for such locks.
- Added new type of lock to be requested MDL_INTENTION_EXCLISIVE.
- Added MDL_context::try_acquire_global_intention_exclusive_lock()
and acquire_global_intention_exclusive_lock() methods.
- Moved common code from MDL_context::acquire_global_shared_lock()
and acquire_global_intention_exclusive_lock() to new method -
MDL_context::acquire_lock_impl().
- Moved common code from MDL_context::try_acquire_shared_lock(),
try_acquire_global_intention_exclusive_lock()/exclusive_lock()
to MDL_context::try_acquire_lock_impl().
- Added helper MDL_context::is_global_lock_owner() method to be
able easily to find what kind of global lock this context holds.
- MDL_context::m_has_global_shared_lock became unnecessary as
global read lock is now represented by ordinary ticket.
- Removed assert in MDL_context::set_lt_or_ha_sentinel() which became
false for cases when we execute LOCK TABLES under global read lock
mode.
sql/mysql_priv.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result calls
opening/closing system tables were changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sp.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sp.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sql_base.cc:
close_thread_tables():
Since we no longer use separate MDL_context for opening system
tables we need to avoid releasing all transaction locks when
closing system table. Releasing metadata lock on system table
is now responsibility of THD::restore_backup_open_tables_state().
open_table_get_mdl_lock(),
Open_table_context::recover_from_failed_open():
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object upgradable or exclusive metadata lock is
acquired. So this have to be done explicitly from these calls.
Changed Open_table_context class to store MDL_request object for
global intention exclusive lock acquired when opening tables.
open_table():
Do not release metadata lock if we have failed to open table as
this lock might have been acquired by one of previous statements
in transaction, and therefore should not be released.
open_system_tables_for_read()/close_system_tables()/
open_performance_schema_table():
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
close_performance_schema_table():
Got rid of duplicated code.
sql/sql_class.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. Also releasing
metadata lock on system table is now responsibility of
THD::restore_backup_open_tables_state().
Adjusted assert in THD::cleanup() to take into account fact that now
we also use MDL sentinel for global read lock.
sql/sql_class.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. As result:
- 'mdl_context' member was moved out of Open_tables_state to THD class.
enter_locked_tables_mode()/leave_locked_tables_mode() had to follow.
- Methods of THD responsible for saving/restoring open table state were
changed to use Open_tables_backup class which in addition to
Open_table_state has a member for this savepoint.
Changed Open_table_context class to store MDL_request object for
global intention exclusive lock acquired when opening tables.
sql/sql_delete.cc:
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object exclusive metadata lock is acquired.
Now this has to be done by explicit calls outside of MDL subsystem.
sql/sql_help.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sql_parse.cc:
Adjusted assert reload_acl_and_cache() to the fact that global read
lock now takes full-blown metadata lock.
sql/sql_plist.h:
Added support for element counting to I_P_List list template.
One can use policy classes to specify if such counting is needed
or not needed for particular list.
sql/sql_show.cc:
Instead of using separate MDL_context for opening tables for I_S
purposes we now create MDL savepoint in the main context before
opening tables and rollback to this savepoint after closing them.
To support this and similar change for system tables methods of
THD responsible for saving/restoring open table state were changed
to use Open_tables_backup class which in addition to Open_table_state
has a member for this savepoint. As result code opening/closing tables
for I_S purposes was changed to use Open_tables_backup instead of
Open_table_state class as well.
sql/sql_table.cc:
mysql_rm_tables_part2():
Since now global intention exclusive metadata lock is ordinary
metadata lock we no longer can rely that by releasing MDL locks
on all tables we will release all locks acquired by this routine.
So in non-LOCK-TABLES mode we have to release all locks acquired
explicitly.
prepare_for_repair(), mysql_alter_table():
MDL subsystem no longer implicitly acquires global intention
exclusive metadata lock when per-object exclusive metadata lock
is acquired. Now this has to be done by explicit calls outside of
MDL subsystem.
sql/tztime.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
Also changed code not to use special mechanism for open system tables
when it is not really necessary.
2010-01-21 21:43:03 +01:00
handler t0 open;
handler t0 close;
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
connection: flush
2007-10-11 02:39:22 +02:00
handler t1 read next;
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
ERROR 42S02: Unknown table 't1' in HANDLER
2007-10-11 02:39:22 +02:00
handler t1 close;
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
ERROR 42S02: Unknown table 't1' in HANDLER
Patch that changes metadata locking subsystem to use mutex per lock and
condition variable per context instead of one mutex and one conditional
variable for the whole subsystem.
This should increase concurrency in this subsystem.
It also opens the way for further changes which are necessary to solve
such bugs as bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock"
and bug #37346 "innodb does not detect deadlock between update and alter
table".
Two other notable changes done by this patch:
- MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object metadata lock is acquired. Now this has
to be done by explicit calls outside of MDL subsystem.
- Instead of using separate MDL_context for opening system tables/tables
for purposes of I_S we now create MDL savepoint in the main context
before opening tables and rollback to this savepoint after closing
them. This means that it is now possible to get ER_LOCK_DEADLOCK error
even not inside a transaction. This might happen in unlikely case when
one runs DDL on one of system tables while also running DDL on some
other tables. Cases when this ER_LOCK_DEADLOCK error is not justified
will be addressed by advanced deadlock detector for MDL subsystem which
we plan to implement.
mysql-test/include/handler.inc:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/handler_innodb.result:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/handler_myisam.result:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/mdl_sync.result:
Adjusted mdl_sync.test to the fact that exclusive metadata locks on
tables are now acquired according to alphabetical order of fully
qualified table names instead of order in which tables are mentioned
in statement.
mysql-test/t/mdl_sync.test:
Adjusted mdl_sync.test to the fact that exclusive metadata locks on
tables are now acquired according to alphabetical order of fully
qualified table names instead of order in which tables are mentioned
in statement.
sql/events.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/ha_ndbcluster.cc:
Since manipulations with open table state no longer install proxy
MDL_context it does not make sense to perform them in order to
satisfy assert in mysql_rm_tables_part2(). Removed them per agreement
with Cluster team. This has not broken test suite since scenario in
which deadlock can occur and assertion fails is not covered by tests.
sql/lock.cc:
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object exclusive metadata lock is acquired.
Now this has to be done by explicit calls outside of MDL subsystem.
sql/log.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/mdl.cc:
Changed metadata locking subsystem to use mutex per lock and condition
variable per context instead of one mutex and one conditional variable
for the whole subsystem.
Changed approach to handling of global metadata locks. Instead of
implicitly acquiring intention exclusive locks when user requests
per-object upgradeable or exclusive locks now we require them to be
acquired explicitly in the same way as ordinary metadata locks.
In fact global lock are now ordinary metadata locks in new GLOBAL
namespace.
To implement these changes:
- Removed LOCK_mdl mutex and COND_mdl condition variable.
- Introduced MDL_lock::m_mutex mutexes which protect individual lock
objects.
- Replaced mdl_locks hash with MDL_map class, which has hash for
MDL_lock objects as a member and separate mutex which protects this
hash. Methods of this class allow to find(), find_or_create() or
remove() MDL_lock objects in concurrency-friendly fashion (i.e.
for most common operation, find_or_create(), we don't acquire
MDL_lock::m_mutex while holding MDL_map::m_mutex. Thanks to MikaelR
for this idea and benchmarks!). Added three auxiliary members to
MDL_lock class (m_is_destroyed, m_ref_usage, m_ref_release) to
support this concurrency-friendly behavior.
- Introduced MDL_context::m_ctx_wakeup_cond condition variable to be
used for waiting until this context's pending request can be
satisfied or its thread has to perform actions to resolve potential
deadlock. Context which want to wait add ticket corresponding to the
request to an appropriate queue of waiters in MDL_lock object so
they can be noticed when other contexts change state of lock and be
awaken by them by signalling on MDL_context::m_ctx_wakeup_cond.
As consequence MDL_ticket objects has to be used for any waiting
in metadata locking subsystem including one which happens in
MDL_context::wait_for_locks() method.
Another consequence is that MDL_context is no longer copyable and
can't be saved/restored when working with system tables.
- Made MDL_lock an abstract class, which delegates specifying exact
compatibility matrix to its descendants. Added MDL_global_lock child
class for global lock (The old is_lock_type_compatible() method
became can_grant_lock() method of this class). Added MDL_object_lock
class to represent per-object lock (The old MDL_lock::can_grant_lock()
became its method). Choice between two classes happens based on MDL
namespace in MDL_lock::create() method.
- Got rid of MDL_lock::type member as its meaning became ambigous for
global locks.
- To simplify waking up of contexts waiting for lock split waiting queue
in MDL_lock class in two queues. One for pending requests for exclusive
(including intention exclusive) locks and another for requests for
shared locks.
- Added virtual wake_up_waiters() method to MDL_lock, MDL_global_lock and
MDL_object_lock classes which allows to wake up waiting contexts after
state of lock changes. Replaced old duplicated code with calls to this
method.
- Adjusted MDL_context::try_acquire_shared_lock()/exclusive_lock()/
global_shared_lock(), MDL_ticket::upgrade_shared_lock_to_exclusive_lock()
and MDL_context::release_ticket() methods to use MDL_map and
MDL_lock::m_mutex instead of single LOCK_mdl mutex and wake up
waiters according to the approach described above. The latter method
also was renamed to MDL_context::release_lock().
- Changed MDL_context::try_acquire_shared_lock()/exclusive_lock() and
release_lock() not to handle global locks. They are now supposed to
be taken explicitly like ordinary metadata locks.
- Added helper MDL_context::try_acquire_global_intention_exclusive_lock()
and acquire_global_intention_exclusive_lock() methods.
- Moved common code from MDL_context::acquire_global_shared_lock() and
acquire_global_intention_exclusive_lock() to new method -
MDL_context::acquire_lock_impl().
- Moved common code from MDL_context::try_acquire_shared_lock(),
try_acquire_global_intention_exclusive_lock()/exclusive_lock()
to MDL_context::try_acquire_lock_impl().
- Since acquiring of several exclusive locks can no longer happen under
single LOCK_mdl mutex the approach to it had to be changed. Now we do
it in one by one fashion. This is done in alphabetical order to avoid
deadlocks. Changed MDL_context::acquire_exclusive_locks() accordingly
(as part of this change moved code responsible for acquiring single
exclusive lock to new MDL_context::acquire_exclusive_lock_impl()
method).
- Since we no longer have single LOCK_mdl mutex which protects all
MDL_context::m_is_waiting_in_mdl members using these members to
determine if we have really awaken context holding conflicting
shared lock became inconvinient. Got rid of this member and changed
notify_shared_lock() helper function and process of acquiring
of/upgrading to exclusive lock not to rely on such information.
Now in MDL_context::acquire_exclusive_lock_impl() and
MDL_ticket::upgrade_shared_lock_to_exclusive_lock() we simply
re-try to wake up threads holding conflicting shared locks after
small time out.
- Adjusted MDL_context::can_wait_lead_to_deadlock() and
MDL_ticket::has_pending_conflicting_lock() to use per-lock
mutexes instead of LOCK_mdl. To do this introduced
MDL_lock::has_pending_exclusive_lock() method.
sql/mdl.h:
Changed metadata locking subsystem to use mutex per lock and condition
variable per context instead of one mutex and one conditional variable
for the whole subsystem. In order to implement this change:
- Added MDL_key::cmp() method to be able to sort MDL_key objects
alphabetically. Changed length fields in MDL_key class to uint16
as 16-bit is enough for length of any key.
- Changed MDL_ticket::get_ctx() to return pointer to non-const
object in order to be able to use MDL_context::awake() method
for such contexts.
- Got rid of unlocked versions of can_wait_lead_to_deadlock()/
has_pending_conflicting_lock() methods in MDL_context and
MDL_ticket. We no longer has single mutex which protects all
locks. Thus one always has to use versions of these methods
which acquire per-lock mutexes.
- MDL_request_list type of list now counts its elements.
- Added MDL_context::m_ctx_wakeup_cond condition variable to be used
for waiting until this context's pending request can be satisfied
or its thread has to perform actions to resolve potential deadlock.
Added awake() method to wake up context from such wait.
Addition of condition variable made MDL_context uncopyable.
As result we no longer can save/restore MDL_context when working
with system tables. Instead we create MDL savepoint before opening
those tables and rollback to it once they are closed.
- MDL_context::release_ticket() became release_lock() method.
- Added auxiliary MDL_context::acquire_exclusive_lock_impl() method
which does all necessary work to acquire exclusive lock on one object
but should not be used directly as it does not enforce any asserts
ensuring that no deadlocks are possible.
- Since we no longer need to know if thread trying to acquire exclusive
lock managed to wake up any threads having conflicting shared locks
(as, anyway, we will try to wake up such threads again shortly)
- MDL_context::m_is_waiting_in_mdl member became unnecessary and
notify_shared_lock() no longer needs to be friend of MDL_context.
Changed approach to handling of global metadata locks. Instead of
implicitly acquiring intention exclusive locks when user requests
per-object upgradeable or exclusive locks now we require them to be
acquired explicitly in the same way as ordinary metadata locks.
- Added new GLOBAL namespace for such locks.
- Added new type of lock to be requested MDL_INTENTION_EXCLISIVE.
- Added MDL_context::try_acquire_global_intention_exclusive_lock()
and acquire_global_intention_exclusive_lock() methods.
- Moved common code from MDL_context::acquire_global_shared_lock()
and acquire_global_intention_exclusive_lock() to new method -
MDL_context::acquire_lock_impl().
- Moved common code from MDL_context::try_acquire_shared_lock(),
try_acquire_global_intention_exclusive_lock()/exclusive_lock()
to MDL_context::try_acquire_lock_impl().
- Added helper MDL_context::is_global_lock_owner() method to be
able easily to find what kind of global lock this context holds.
- MDL_context::m_has_global_shared_lock became unnecessary as
global read lock is now represented by ordinary ticket.
- Removed assert in MDL_context::set_lt_or_ha_sentinel() which became
false for cases when we execute LOCK TABLES under global read lock
mode.
sql/mysql_priv.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result calls
opening/closing system tables were changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sp.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sp.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sql_base.cc:
close_thread_tables():
Since we no longer use separate MDL_context for opening system
tables we need to avoid releasing all transaction locks when
closing system table. Releasing metadata lock on system table
is now responsibility of THD::restore_backup_open_tables_state().
open_table_get_mdl_lock(),
Open_table_context::recover_from_failed_open():
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object upgradable or exclusive metadata lock is
acquired. So this have to be done explicitly from these calls.
Changed Open_table_context class to store MDL_request object for
global intention exclusive lock acquired when opening tables.
open_table():
Do not release metadata lock if we have failed to open table as
this lock might have been acquired by one of previous statements
in transaction, and therefore should not be released.
open_system_tables_for_read()/close_system_tables()/
open_performance_schema_table():
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
close_performance_schema_table():
Got rid of duplicated code.
sql/sql_class.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. Also releasing
metadata lock on system table is now responsibility of
THD::restore_backup_open_tables_state().
Adjusted assert in THD::cleanup() to take into account fact that now
we also use MDL sentinel for global read lock.
sql/sql_class.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. As result:
- 'mdl_context' member was moved out of Open_tables_state to THD class.
enter_locked_tables_mode()/leave_locked_tables_mode() had to follow.
- Methods of THD responsible for saving/restoring open table state were
changed to use Open_tables_backup class which in addition to
Open_table_state has a member for this savepoint.
Changed Open_table_context class to store MDL_request object for
global intention exclusive lock acquired when opening tables.
sql/sql_delete.cc:
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object exclusive metadata lock is acquired.
Now this has to be done by explicit calls outside of MDL subsystem.
sql/sql_help.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sql_parse.cc:
Adjusted assert reload_acl_and_cache() to the fact that global read
lock now takes full-blown metadata lock.
sql/sql_plist.h:
Added support for element counting to I_P_List list template.
One can use policy classes to specify if such counting is needed
or not needed for particular list.
sql/sql_show.cc:
Instead of using separate MDL_context for opening tables for I_S
purposes we now create MDL savepoint in the main context before
opening tables and rollback to this savepoint after closing them.
To support this and similar change for system tables methods of
THD responsible for saving/restoring open table state were changed
to use Open_tables_backup class which in addition to Open_table_state
has a member for this savepoint. As result code opening/closing tables
for I_S purposes was changed to use Open_tables_backup instead of
Open_table_state class as well.
sql/sql_table.cc:
mysql_rm_tables_part2():
Since now global intention exclusive metadata lock is ordinary
metadata lock we no longer can rely that by releasing MDL locks
on all tables we will release all locks acquired by this routine.
So in non-LOCK-TABLES mode we have to release all locks acquired
explicitly.
prepare_for_repair(), mysql_alter_table():
MDL subsystem no longer implicitly acquires global intention
exclusive metadata lock when per-object exclusive metadata lock
is acquired. Now this has to be done by explicit calls outside of
MDL subsystem.
sql/tztime.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
Also changed code not to use special mechanism for open system tables
when it is not really necessary.
2010-01-21 21:43:03 +01:00
drop table t0;
2007-11-01 21:52:56 +01:00
drop table if exists t1;
create temporary table t1 (a int, b char(1), key a(a), key b(a,b));
insert into t1 values (0,"a"),(1,"b"),(2,"c"),(3,"d"),(4,"e"),
(5,"f"),(6,"g"),(7,"h"),(8,"i"),(9,"j");
select a,b from t1;
a b
0 a
1 b
2 c
3 d
4 e
5 f
6 g
7 h
8 i
9 j
handler t1 open as a1;
handler a1 read a first;
a b
0 a
handler a1 read a next;
a b
1 b
handler a1 read a next;
a b
2 c
select a,b from t1;
ERROR HY000: Can't reopen table: 'a1'
handler a1 read a prev;
a b
1 b
handler a1 read a prev;
a b
0 a
handler a1 read a=(6) where b="g";
a b
6 g
handler a1 close;
select a,b from t1;
a b
0 a
1 b
2 c
3 d
4 e
5 f
6 g
7 h
8 i
9 j
handler t1 open as a2;
handler a2 read a first;
a b
0 a
handler a2 read a last;
a b
9 j
handler a2 read a prev;
a b
8 i
handler a2 close;
drop table t1;
2007-11-20 18:17:53 +01:00
drop table if exists t1,t2;
create table t1 (a int);
handler t1 open as t1_alias;
drop table t1;
create table t1 (a int);
handler t1 open as t1_alias;
flush tables;
drop table t1;
create table t1 (a int);
handler t1 open as t1_alias;
handler t1_alias close;
drop table t1;
create table t1 (a int);
handler t1 open as t1_alias;
handler t1_alias read first;
a
drop table t1;
handler t1_alias read next;
ERROR 42S02: Unknown table 't1_alias' in HANDLER
create table t1 (a int);
create temporary table t2 (a int, key(a));
handler t1 open as a1;
handler t2 open as a2;
handler a2 read a first;
a
drop table t1, t2;
handler a2 read a next;
ERROR 42S02: Unknown table 'a2' in HANDLER
handler a1 close;
ERROR 42S02: Unknown table 'a1' in HANDLER
create table t1 (a int, key(a));
create table t2 like t1;
handler t1 open as a1;
handler t2 open as a2;
handler a1 read a first;
a
handler a2 read a first;
a
alter table t1 add b int;
handler a1 close;
ERROR 42S02: Unknown table 'a1' in HANDLER
handler a2 close;
drop table t1, t2;
create table t1 (a int, key(a));
handler t1 open as a1;
handler a1 read a first;
a
rename table t1 to t2;
handler a1 read a first;
ERROR 42S02: Unknown table 'a1' in HANDLER
drop table t2;
create table t1 (a int, key(a));
create table t2 like t1;
handler t1 open as a1;
handler t2 open as a2;
handler a1 read a first;
a
handler a2 read a first;
a
optimize table t1;
Table Op Msg_type Msg_text
test.t1 optimize status Table is already up to date
handler a1 close;
ERROR 42S02: Unknown table 'a1' in HANDLER
handler a2 close;
drop table t1, t2;
create table t1 (a int, b char(1), key a(a), key b(a,b));
insert into t1 values (0,"a"),(1,"b"),(2,"c"),(3,"d"),(4,"e"),
(5,"f"),(6,"g"),(7,"h"),(8,"i"),(9,"j");
handler t1 open;
handler t1 read a first;
a b
0 a
handler t1 read a next;
a b
1 b
flush tables;
handler t1 read a next;
a b
0 a
handler t1 read a next;
a b
1 b
flush tables with read lock;
handler t1 read a next;
a b
0 a
unlock tables;
drop table t1;
handler t1 read a next;
ERROR 42S02: Unknown table 't1' in HANDLER
2009-02-24 10:49:18 +01:00
drop table if exists t1;
Implement new type-of-operation-aware metadata locks.
Add a wait-for graph based deadlock detector to the
MDL subsystem.
Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and
bug #37346 "innodb does not detect deadlock between update and
alter table".
The first bug manifested itself as an unwarranted abort of a
transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER
statement, when this transaction tried to repeat use of a
table, which it has already used in a similar fashion before
ALTER started.
The second bug showed up as a deadlock between table-level
locks and InnoDB row locks, which was "detected" only after
innodb_lock_wait_timeout timeout.
A transaction would start using the table and modify a few
rows.
Then ALTER TABLE would come in, and start copying rows
into a temporary table. Eventually it would stumble on
the modified records and get blocked on a row lock.
The first transaction would try to do more updates, and get
blocked on thr_lock.c lock.
This situation of circular wait would only get resolved
by a timeout.
Both these bugs stemmed from inadequate solutions to the
problem of deadlocks occurring between different
locking subsystems.
In the first case we tried to avoid deadlocks between metadata
locking and table-level locking subsystems, when upgrading shared
metadata lock to exclusive one.
Transactions holding the shared lock on the table and waiting for
some table-level lock used to be aborted too aggressively.
We also allowed ALTER TABLE to start in presence of transactions
that modify the subject table. ALTER TABLE acquires
TL_WRITE_ALLOW_READ lock at start, and that block all writes
against the table (naturally, we don't want any writes to be lost
when switching the old and the new table). TL_WRITE_ALLOW_READ
lock, in turn, would block the started transaction on thr_lock.c
lock, should they do more updates. This, again, lead to the need
to abort such transactions.
The second bug occurred simply because we didn't have any
mechanism to detect deadlocks between the table-level locks
in thr_lock.c and row-level locks in InnoDB, other than
innodb_lock_wait_timeout.
This patch solves both these problems by moving lock conflicts
which are causing these deadlocks into the metadata locking
subsystem, thus making it possible to avoid or detect such
deadlocks inside MDL.
To do this we introduce new type-of-operation-aware metadata
locks, which allow MDL subsystem to know not only the fact that
transaction has used or is going to use some object but also what
kind of operation it has carried out or going to carry out on the
object.
This, along with the addition of a special kind of upgradable
metadata lock, allows ALTER TABLE to wait until all
transactions which has updated the table to go away.
This solves the second issue.
Another special type of upgradable metadata lock is acquired
by LOCK TABLE WRITE. This second lock type allows to solve the
first issue, since abortion of table-level locks in event of
DDL under LOCK TABLES becomes also unnecessary.
Below follows the list of incompatible changes introduced by
this patch:
- From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those
statements that acquire TL_WRITE_ALLOW_READ lock)
wait for all transactions which has *updated* the table to
complete.
- From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE
(i.e. all statements which acquire TL_WRITE table-level lock) wait
for all transaction which *updated or read* from the table
to complete.
As a consequence, innodb_table_locks=0 option no longer applies
to LOCK TABLES ... WRITE.
- DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort
statements or transactions which use tables being dropped or
renamed, and instead wait for these transactions to complete.
- Since LOCK TABLES WRITE now takes a special metadata lock,
not compatible with with reads or writes against the subject table
and transaction-wide, thr_lock.c deadlock avoidance algorithm
that used to ensure absence of deadlocks between LOCK TABLES
WRITE and other statements is no longer sufficient, even for
MyISAM. The wait-for graph based deadlock detector of MDL
subsystem may sometimes be necessary and is involved. This may
lead to ER_LOCK_DEADLOCK error produced for multi-statement
transactions even if these only use MyISAM:
session 1: session 2:
begin;
update t1 ... lock table t2 write, t1 write;
-- gets a lock on t2, blocks on t1
update t2 ...
(ER_LOCK_DEADLOCK)
- Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE
was abandoned.
LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same
priority as the usual LOCK TABLE ... WRITE.
SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in
the wait queue.
- We do not take upgradable metadata locks on implicitly
locked tables. So if one has, say, a view v1 that uses
table t1, and issues:
LOCK TABLE v1 WRITE;
FLUSH TABLE t1; -- (or just 'FLUSH TABLES'),
an error is produced.
In order to be able to perform DDL on a table under LOCK TABLES,
the table must be locked explicitly in the LOCK TABLES list.
mysql-test/include/handler.inc:
Adjusted test case to trigger an execution path on which bug 41110
"crash with handler command when used concurrently with alter
table" and bug 41112 "crash in mysql_ha_close_table/get_lock_data
with alter table" were originally discovered. Left old test case
which no longer triggers this execution path for the sake of
coverage.
Added test coverage for HANDLER SQL statements and type-aware
metadata locks.
Added a test for the global shared lock and HANDLER SQL.
Updated tests to take into account that the old simple deadlock
detection heuristics was replaced with a graph-based deadlock
detector.
mysql-test/r/debug_sync.result:
Updated results (see debug_sync.test).
mysql-test/r/handler_innodb.result:
Updated results (see handler.inc test).
mysql-test/r/handler_myisam.result:
Updated results (see handler.inc test).
mysql-test/r/innodb-lock.result:
Updated results (see innodb-lock.test).
mysql-test/r/innodb_mysql_lock.result:
Updated results (see innodb_mysql_lock.test).
mysql-test/r/lock.result:
Updated results (see lock.test).
mysql-test/r/lock_multi.result:
Updated results (see lock_multi.test).
mysql-test/r/lock_sync.result:
Updated results (see lock_sync.test).
mysql-test/r/mdl_sync.result:
Updated results (see mdl_sync.test).
mysql-test/r/sp-threads.result:
SHOW PROCESSLIST output has changed due to the fact that waiting
for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/r/truncate_coverage.result:
Updated results (see truncate_coverage.test).
mysql-test/suite/funcs_1/datadict/processlist_val.inc:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/funcs_1/r/processlist_val_no_prot.result:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/rpl/t/rpl_sp.test:
Updated to a new SHOW PROCESSLIST state name.
mysql-test/t/debug_sync.test:
Use LOCK TABLES READ instead of LOCK TABLES WRITE as the latter
no longer allows to trigger execution path involving waiting on
thr_lock.c lock and therefore reaching debug sync-point covered
by this test.
mysql-test/t/innodb-lock.test:
Adjusted test case to the fact that innodb_table_locks=0 option is
no longer supported, since LOCK TABLES WRITE handles all its
conflicts within MDL subsystem.
mysql-test/t/innodb_mysql_lock.test:
Added test for bug #37346 "innodb does not detect deadlock between
update and alter table".
mysql-test/t/lock.test:
Added test coverage which checks the fact that we no longer support
DDL under LOCK TABLES on tables which were locked implicitly.
Adjusted existing test cases accordingly.
mysql-test/t/lock_multi.test:
Added test for bug #46272 "MySQL 5.4.4, new MDL: unnecessary
deadlock". Adjusted other test cases to take into account the
fact that waiting for LOCK TABLES ... WRITE now happens within MDL
subsystem.
mysql-test/t/lock_sync.test:
Since LOCK TABLES ... WRITE now takes SNRW metadata lock for
tables locked explicitly we have to implicitly lock InnoDB tables
(through view) to trigger the table-level lock conflict between
TL_WRITE and TL_WRITE_ALLOW_WRITE.
mysql-test/t/mdl_sync.test:
Added basic test coverage for type-of-operation-aware metadata
locks. Also covered with tests some use cases involving HANDLER
statements in which a deadlock could arise.
Adjusted existing tests to take type-of-operation-aware MDL into
account.
mysql-test/t/multi_update.test:
Update to a new SHOW PROCESSLIST state name.
mysql-test/t/truncate_coverage.test:
Adjusted test case after making LOCK TABLES WRITE to wait until
transactions that use the table to be locked are completed.
Updated to the changed name of DEBUG_SYNC point.
sql/handler.cc:
Global read lock functionality has been
moved into a class.
sql/lock.cc:
Global read lock functionality has been
moved into a class.
Updated code to use the new MDL API.
sql/mdl.cc:
Introduced new type-of-operation aware metadata locks.
To do this:
- Changed MDL_lock to use one list for waiting requests and one
list for granted requests. For each list, added a bitmap
that holds information what lock types a list contains.
Added a helper class MDL_lock::List to manipulate with granted
and waited lists while keeping the bitmaps in sync
with list contents.
- Changed lock-compatibility functions to use bitmaps that
define compatibility.
- Introduced a graph based deadlock detector inspired by
waiting_threads.c from Maria implementation.
- Now that we have a deadlock detector, and no longer have
a global lock to protect individual lock objects, but rather
use an rw lock per object, removed redundant code for upgrade,
and the global read lock. Changed the MDL API to
no longer require the caller to acquire the global
intention exclusive lock by means of a separate method.
Removed a few more methods that became redundant.
- Removed deadlock detection heuristic, it has been made
obsolete by the deadlock detector.
- With operation-type-aware metadata locks, MDL subsystem has
become aware of potential conflicts between DDL and open
transactions. This made it possible to remove calls to
mysql_abort_transactions_with_shared_lock() from acquisition
paths for exclusive lock and lock upgrade. Now we can simply
wait for these transactions to complete without fear of
deadlock. Function mysql_lock_abort() has also become
unnecessary for all conflicting cases except when a DDL
conflicts with a connection that has an open HANDLER.
sql/mdl.h:
Introduced new type-of-operation aware metadata locks.
Introduced a graph based deadlock detector and supporting
methods.
Added comments.
God rid of redundant API calls.
Renamed m_lt_or_ha_sentinel to m_trans_sentinel,
since now it guards the global read lock as well as
LOCK TABLES and HANDLER locks.
sql/mysql_priv.h:
Moved the global read lock functionality into a
class.
Added MYSQL_OPEN_FORCE_SHARED_MDL flag which forces
open_tables() to take MDL_SHARED on tables instead of
metadata locks specified in the parser. We use this to
allow PREPARE run concurrently in presence of
LOCK TABLES ... WRITE.
Added signature for find_table_for_mdl_ugprade().
sql/set_var.cc:
Global read lock functionality has been
moved into a class.
sql/sp_head.cc:
When creating TABLE_LIST elements for prelocking or
system tables set the type of request for metadata
lock according to the operation that will be performed
on the table.
sql/sql_base.cc:
- Updated code to use the new MDL API.
- In order to avoid locks starvation we take upgradable
locks all at once. As result implicitly locked tables no
longer get an upgradable lock. Consequently DDL and FLUSH
TABLES for such tables is prohibited.
find_write_locked_table() was replaced by
find_table_for_mdl_upgrade() function.
open_table() was adjusted to return TABLE instance with
upgradable ticket when necessary.
- We no longer wait for all locks on OT_WAIT back off
action -- only on the lock that caused the wait
conflict. Moreover, now we distinguish cases when we
have to wait due to conflict in MDL and old version
of table in TDC.
- Upate mysql_notify_threads_having_share_locks()
to only abort thr_lock.c waits of threads that
have open HANDLERs, since lock conflicts with only
these threads now can lead to deadlocks not detectable
by the MDL deadlock detector.
- Remove mysql_abort_transactions_with_shared_locks()
which is no longer needed.
sql/sql_class.cc:
Global read lock functionality has been moved into a class.
Re-arranged code in THD::cleanup() to simplify assert.
sql/sql_class.h:
Introduced class to incapsulate global read lock
functionality.
Now sentinel in MDL subsystem guards the global read lock
as well as LOCK TABLES and HANDLER locks. Adjusted code
accordingly.
sql/sql_db.cc:
Global read lock functionality has been moved into a class.
sql/sql_delete.cc:
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
TRUNCATE TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_handler.cc:
Inform MDL_context about presence of open HANDLERs.
Since HANLDERs break MDL protocol by acquiring table-level
lock while holding only S metadata lock on a table MDL
subsystem should take special care about such contexts (Now
this is the only case when mysql_lock_abort() is used).
sql/sql_parse.cc:
Global read lock functionality has been moved into a class.
Do not take upgradable metadata locks when opening tables
for CREATE TABLE SELECT as it is not necessary and limits
concurrency.
When initializing TABLE_LIST objects before adding them
to the table list set the type of request for metadata lock
according to the operation that will be performed on the
table.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result FLUSH
TABLES is no longer allowed for such tables.
sql/sql_prepare.cc:
Use MYSQL_OPEN_FORCE_SHARED_MDL flag when opening
tables during PREPARE. This allows PREPARE to run
concurrently in presence of LOCK TABLES ... WRITE.
sql/sql_rename.cc:
Global read lock functionality has been moved into a class.
sql/sql_show.cc:
Updated code to use the new MDL API.
sql/sql_table.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result DROP
TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_trigger.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
CREATE/DROP TRIGGER is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_view.cc:
Global read lock functionality has been moved into a class.
Fixed results of wrong merge that led to misuse of GLR API.
CREATE VIEW statement is not a commit statement.
sql/table.cc:
When resetting TABLE_LIST objects for PS or SP re-execution
set the type of request for metadata lock according to the
operation that will be performed on the table. Do the same
in auxiliary function initializing metadata lock requests
in a table list.
sql/table.h:
When initializing TABLE_LIST objects set the type of request
for metadata lock according to the operation that will be
performed on the table.
sql/transaction.cc:
Global read lock functionality has been moved into a class.
2010-02-01 12:43:06 +01:00
# First test case which is supposed trigger the execution
# path on which problem was discovered.
create table t1 (a int);
2009-02-24 10:49:18 +01:00
insert into t1 values (1);
handler t1 open;
Implement new type-of-operation-aware metadata locks.
Add a wait-for graph based deadlock detector to the
MDL subsystem.
Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and
bug #37346 "innodb does not detect deadlock between update and
alter table".
The first bug manifested itself as an unwarranted abort of a
transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER
statement, when this transaction tried to repeat use of a
table, which it has already used in a similar fashion before
ALTER started.
The second bug showed up as a deadlock between table-level
locks and InnoDB row locks, which was "detected" only after
innodb_lock_wait_timeout timeout.
A transaction would start using the table and modify a few
rows.
Then ALTER TABLE would come in, and start copying rows
into a temporary table. Eventually it would stumble on
the modified records and get blocked on a row lock.
The first transaction would try to do more updates, and get
blocked on thr_lock.c lock.
This situation of circular wait would only get resolved
by a timeout.
Both these bugs stemmed from inadequate solutions to the
problem of deadlocks occurring between different
locking subsystems.
In the first case we tried to avoid deadlocks between metadata
locking and table-level locking subsystems, when upgrading shared
metadata lock to exclusive one.
Transactions holding the shared lock on the table and waiting for
some table-level lock used to be aborted too aggressively.
We also allowed ALTER TABLE to start in presence of transactions
that modify the subject table. ALTER TABLE acquires
TL_WRITE_ALLOW_READ lock at start, and that block all writes
against the table (naturally, we don't want any writes to be lost
when switching the old and the new table). TL_WRITE_ALLOW_READ
lock, in turn, would block the started transaction on thr_lock.c
lock, should they do more updates. This, again, lead to the need
to abort such transactions.
The second bug occurred simply because we didn't have any
mechanism to detect deadlocks between the table-level locks
in thr_lock.c and row-level locks in InnoDB, other than
innodb_lock_wait_timeout.
This patch solves both these problems by moving lock conflicts
which are causing these deadlocks into the metadata locking
subsystem, thus making it possible to avoid or detect such
deadlocks inside MDL.
To do this we introduce new type-of-operation-aware metadata
locks, which allow MDL subsystem to know not only the fact that
transaction has used or is going to use some object but also what
kind of operation it has carried out or going to carry out on the
object.
This, along with the addition of a special kind of upgradable
metadata lock, allows ALTER TABLE to wait until all
transactions which has updated the table to go away.
This solves the second issue.
Another special type of upgradable metadata lock is acquired
by LOCK TABLE WRITE. This second lock type allows to solve the
first issue, since abortion of table-level locks in event of
DDL under LOCK TABLES becomes also unnecessary.
Below follows the list of incompatible changes introduced by
this patch:
- From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those
statements that acquire TL_WRITE_ALLOW_READ lock)
wait for all transactions which has *updated* the table to
complete.
- From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE
(i.e. all statements which acquire TL_WRITE table-level lock) wait
for all transaction which *updated or read* from the table
to complete.
As a consequence, innodb_table_locks=0 option no longer applies
to LOCK TABLES ... WRITE.
- DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort
statements or transactions which use tables being dropped or
renamed, and instead wait for these transactions to complete.
- Since LOCK TABLES WRITE now takes a special metadata lock,
not compatible with with reads or writes against the subject table
and transaction-wide, thr_lock.c deadlock avoidance algorithm
that used to ensure absence of deadlocks between LOCK TABLES
WRITE and other statements is no longer sufficient, even for
MyISAM. The wait-for graph based deadlock detector of MDL
subsystem may sometimes be necessary and is involved. This may
lead to ER_LOCK_DEADLOCK error produced for multi-statement
transactions even if these only use MyISAM:
session 1: session 2:
begin;
update t1 ... lock table t2 write, t1 write;
-- gets a lock on t2, blocks on t1
update t2 ...
(ER_LOCK_DEADLOCK)
- Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE
was abandoned.
LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same
priority as the usual LOCK TABLE ... WRITE.
SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in
the wait queue.
- We do not take upgradable metadata locks on implicitly
locked tables. So if one has, say, a view v1 that uses
table t1, and issues:
LOCK TABLE v1 WRITE;
FLUSH TABLE t1; -- (or just 'FLUSH TABLES'),
an error is produced.
In order to be able to perform DDL on a table under LOCK TABLES,
the table must be locked explicitly in the LOCK TABLES list.
mysql-test/include/handler.inc:
Adjusted test case to trigger an execution path on which bug 41110
"crash with handler command when used concurrently with alter
table" and bug 41112 "crash in mysql_ha_close_table/get_lock_data
with alter table" were originally discovered. Left old test case
which no longer triggers this execution path for the sake of
coverage.
Added test coverage for HANDLER SQL statements and type-aware
metadata locks.
Added a test for the global shared lock and HANDLER SQL.
Updated tests to take into account that the old simple deadlock
detection heuristics was replaced with a graph-based deadlock
detector.
mysql-test/r/debug_sync.result:
Updated results (see debug_sync.test).
mysql-test/r/handler_innodb.result:
Updated results (see handler.inc test).
mysql-test/r/handler_myisam.result:
Updated results (see handler.inc test).
mysql-test/r/innodb-lock.result:
Updated results (see innodb-lock.test).
mysql-test/r/innodb_mysql_lock.result:
Updated results (see innodb_mysql_lock.test).
mysql-test/r/lock.result:
Updated results (see lock.test).
mysql-test/r/lock_multi.result:
Updated results (see lock_multi.test).
mysql-test/r/lock_sync.result:
Updated results (see lock_sync.test).
mysql-test/r/mdl_sync.result:
Updated results (see mdl_sync.test).
mysql-test/r/sp-threads.result:
SHOW PROCESSLIST output has changed due to the fact that waiting
for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/r/truncate_coverage.result:
Updated results (see truncate_coverage.test).
mysql-test/suite/funcs_1/datadict/processlist_val.inc:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/funcs_1/r/processlist_val_no_prot.result:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/rpl/t/rpl_sp.test:
Updated to a new SHOW PROCESSLIST state name.
mysql-test/t/debug_sync.test:
Use LOCK TABLES READ instead of LOCK TABLES WRITE as the latter
no longer allows to trigger execution path involving waiting on
thr_lock.c lock and therefore reaching debug sync-point covered
by this test.
mysql-test/t/innodb-lock.test:
Adjusted test case to the fact that innodb_table_locks=0 option is
no longer supported, since LOCK TABLES WRITE handles all its
conflicts within MDL subsystem.
mysql-test/t/innodb_mysql_lock.test:
Added test for bug #37346 "innodb does not detect deadlock between
update and alter table".
mysql-test/t/lock.test:
Added test coverage which checks the fact that we no longer support
DDL under LOCK TABLES on tables which were locked implicitly.
Adjusted existing test cases accordingly.
mysql-test/t/lock_multi.test:
Added test for bug #46272 "MySQL 5.4.4, new MDL: unnecessary
deadlock". Adjusted other test cases to take into account the
fact that waiting for LOCK TABLES ... WRITE now happens within MDL
subsystem.
mysql-test/t/lock_sync.test:
Since LOCK TABLES ... WRITE now takes SNRW metadata lock for
tables locked explicitly we have to implicitly lock InnoDB tables
(through view) to trigger the table-level lock conflict between
TL_WRITE and TL_WRITE_ALLOW_WRITE.
mysql-test/t/mdl_sync.test:
Added basic test coverage for type-of-operation-aware metadata
locks. Also covered with tests some use cases involving HANDLER
statements in which a deadlock could arise.
Adjusted existing tests to take type-of-operation-aware MDL into
account.
mysql-test/t/multi_update.test:
Update to a new SHOW PROCESSLIST state name.
mysql-test/t/truncate_coverage.test:
Adjusted test case after making LOCK TABLES WRITE to wait until
transactions that use the table to be locked are completed.
Updated to the changed name of DEBUG_SYNC point.
sql/handler.cc:
Global read lock functionality has been
moved into a class.
sql/lock.cc:
Global read lock functionality has been
moved into a class.
Updated code to use the new MDL API.
sql/mdl.cc:
Introduced new type-of-operation aware metadata locks.
To do this:
- Changed MDL_lock to use one list for waiting requests and one
list for granted requests. For each list, added a bitmap
that holds information what lock types a list contains.
Added a helper class MDL_lock::List to manipulate with granted
and waited lists while keeping the bitmaps in sync
with list contents.
- Changed lock-compatibility functions to use bitmaps that
define compatibility.
- Introduced a graph based deadlock detector inspired by
waiting_threads.c from Maria implementation.
- Now that we have a deadlock detector, and no longer have
a global lock to protect individual lock objects, but rather
use an rw lock per object, removed redundant code for upgrade,
and the global read lock. Changed the MDL API to
no longer require the caller to acquire the global
intention exclusive lock by means of a separate method.
Removed a few more methods that became redundant.
- Removed deadlock detection heuristic, it has been made
obsolete by the deadlock detector.
- With operation-type-aware metadata locks, MDL subsystem has
become aware of potential conflicts between DDL and open
transactions. This made it possible to remove calls to
mysql_abort_transactions_with_shared_lock() from acquisition
paths for exclusive lock and lock upgrade. Now we can simply
wait for these transactions to complete without fear of
deadlock. Function mysql_lock_abort() has also become
unnecessary for all conflicting cases except when a DDL
conflicts with a connection that has an open HANDLER.
sql/mdl.h:
Introduced new type-of-operation aware metadata locks.
Introduced a graph based deadlock detector and supporting
methods.
Added comments.
God rid of redundant API calls.
Renamed m_lt_or_ha_sentinel to m_trans_sentinel,
since now it guards the global read lock as well as
LOCK TABLES and HANDLER locks.
sql/mysql_priv.h:
Moved the global read lock functionality into a
class.
Added MYSQL_OPEN_FORCE_SHARED_MDL flag which forces
open_tables() to take MDL_SHARED on tables instead of
metadata locks specified in the parser. We use this to
allow PREPARE run concurrently in presence of
LOCK TABLES ... WRITE.
Added signature for find_table_for_mdl_ugprade().
sql/set_var.cc:
Global read lock functionality has been
moved into a class.
sql/sp_head.cc:
When creating TABLE_LIST elements for prelocking or
system tables set the type of request for metadata
lock according to the operation that will be performed
on the table.
sql/sql_base.cc:
- Updated code to use the new MDL API.
- In order to avoid locks starvation we take upgradable
locks all at once. As result implicitly locked tables no
longer get an upgradable lock. Consequently DDL and FLUSH
TABLES for such tables is prohibited.
find_write_locked_table() was replaced by
find_table_for_mdl_upgrade() function.
open_table() was adjusted to return TABLE instance with
upgradable ticket when necessary.
- We no longer wait for all locks on OT_WAIT back off
action -- only on the lock that caused the wait
conflict. Moreover, now we distinguish cases when we
have to wait due to conflict in MDL and old version
of table in TDC.
- Upate mysql_notify_threads_having_share_locks()
to only abort thr_lock.c waits of threads that
have open HANDLERs, since lock conflicts with only
these threads now can lead to deadlocks not detectable
by the MDL deadlock detector.
- Remove mysql_abort_transactions_with_shared_locks()
which is no longer needed.
sql/sql_class.cc:
Global read lock functionality has been moved into a class.
Re-arranged code in THD::cleanup() to simplify assert.
sql/sql_class.h:
Introduced class to incapsulate global read lock
functionality.
Now sentinel in MDL subsystem guards the global read lock
as well as LOCK TABLES and HANDLER locks. Adjusted code
accordingly.
sql/sql_db.cc:
Global read lock functionality has been moved into a class.
sql/sql_delete.cc:
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
TRUNCATE TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_handler.cc:
Inform MDL_context about presence of open HANDLERs.
Since HANLDERs break MDL protocol by acquiring table-level
lock while holding only S metadata lock on a table MDL
subsystem should take special care about such contexts (Now
this is the only case when mysql_lock_abort() is used).
sql/sql_parse.cc:
Global read lock functionality has been moved into a class.
Do not take upgradable metadata locks when opening tables
for CREATE TABLE SELECT as it is not necessary and limits
concurrency.
When initializing TABLE_LIST objects before adding them
to the table list set the type of request for metadata lock
according to the operation that will be performed on the
table.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result FLUSH
TABLES is no longer allowed for such tables.
sql/sql_prepare.cc:
Use MYSQL_OPEN_FORCE_SHARED_MDL flag when opening
tables during PREPARE. This allows PREPARE to run
concurrently in presence of LOCK TABLES ... WRITE.
sql/sql_rename.cc:
Global read lock functionality has been moved into a class.
sql/sql_show.cc:
Updated code to use the new MDL API.
sql/sql_table.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result DROP
TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_trigger.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
CREATE/DROP TRIGGER is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_view.cc:
Global read lock functionality has been moved into a class.
Fixed results of wrong merge that led to misuse of GLR API.
CREATE VIEW statement is not a commit statement.
sql/table.cc:
When resetting TABLE_LIST objects for PS or SP re-execution
set the type of request for metadata lock according to the
operation that will be performed on the table. Do the same
in auxiliary function initializing metadata lock requests
in a table list.
sql/table.h:
When initializing TABLE_LIST objects set the type of request
for metadata lock according to the operation that will be
performed on the table.
sql/transaction.cc:
Global read lock functionality has been moved into a class.
2010-02-01 12:43:06 +01:00
lock table t1 write;
2009-02-24 10:49:18 +01:00
alter table t1 engine=memory;
handler t1 read a next;
Implement new type-of-operation-aware metadata locks.
Add a wait-for graph based deadlock detector to the
MDL subsystem.
Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and
bug #37346 "innodb does not detect deadlock between update and
alter table".
The first bug manifested itself as an unwarranted abort of a
transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER
statement, when this transaction tried to repeat use of a
table, which it has already used in a similar fashion before
ALTER started.
The second bug showed up as a deadlock between table-level
locks and InnoDB row locks, which was "detected" only after
innodb_lock_wait_timeout timeout.
A transaction would start using the table and modify a few
rows.
Then ALTER TABLE would come in, and start copying rows
into a temporary table. Eventually it would stumble on
the modified records and get blocked on a row lock.
The first transaction would try to do more updates, and get
blocked on thr_lock.c lock.
This situation of circular wait would only get resolved
by a timeout.
Both these bugs stemmed from inadequate solutions to the
problem of deadlocks occurring between different
locking subsystems.
In the first case we tried to avoid deadlocks between metadata
locking and table-level locking subsystems, when upgrading shared
metadata lock to exclusive one.
Transactions holding the shared lock on the table and waiting for
some table-level lock used to be aborted too aggressively.
We also allowed ALTER TABLE to start in presence of transactions
that modify the subject table. ALTER TABLE acquires
TL_WRITE_ALLOW_READ lock at start, and that block all writes
against the table (naturally, we don't want any writes to be lost
when switching the old and the new table). TL_WRITE_ALLOW_READ
lock, in turn, would block the started transaction on thr_lock.c
lock, should they do more updates. This, again, lead to the need
to abort such transactions.
The second bug occurred simply because we didn't have any
mechanism to detect deadlocks between the table-level locks
in thr_lock.c and row-level locks in InnoDB, other than
innodb_lock_wait_timeout.
This patch solves both these problems by moving lock conflicts
which are causing these deadlocks into the metadata locking
subsystem, thus making it possible to avoid or detect such
deadlocks inside MDL.
To do this we introduce new type-of-operation-aware metadata
locks, which allow MDL subsystem to know not only the fact that
transaction has used or is going to use some object but also what
kind of operation it has carried out or going to carry out on the
object.
This, along with the addition of a special kind of upgradable
metadata lock, allows ALTER TABLE to wait until all
transactions which has updated the table to go away.
This solves the second issue.
Another special type of upgradable metadata lock is acquired
by LOCK TABLE WRITE. This second lock type allows to solve the
first issue, since abortion of table-level locks in event of
DDL under LOCK TABLES becomes also unnecessary.
Below follows the list of incompatible changes introduced by
this patch:
- From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those
statements that acquire TL_WRITE_ALLOW_READ lock)
wait for all transactions which has *updated* the table to
complete.
- From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE
(i.e. all statements which acquire TL_WRITE table-level lock) wait
for all transaction which *updated or read* from the table
to complete.
As a consequence, innodb_table_locks=0 option no longer applies
to LOCK TABLES ... WRITE.
- DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort
statements or transactions which use tables being dropped or
renamed, and instead wait for these transactions to complete.
- Since LOCK TABLES WRITE now takes a special metadata lock,
not compatible with with reads or writes against the subject table
and transaction-wide, thr_lock.c deadlock avoidance algorithm
that used to ensure absence of deadlocks between LOCK TABLES
WRITE and other statements is no longer sufficient, even for
MyISAM. The wait-for graph based deadlock detector of MDL
subsystem may sometimes be necessary and is involved. This may
lead to ER_LOCK_DEADLOCK error produced for multi-statement
transactions even if these only use MyISAM:
session 1: session 2:
begin;
update t1 ... lock table t2 write, t1 write;
-- gets a lock on t2, blocks on t1
update t2 ...
(ER_LOCK_DEADLOCK)
- Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE
was abandoned.
LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same
priority as the usual LOCK TABLE ... WRITE.
SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in
the wait queue.
- We do not take upgradable metadata locks on implicitly
locked tables. So if one has, say, a view v1 that uses
table t1, and issues:
LOCK TABLE v1 WRITE;
FLUSH TABLE t1; -- (or just 'FLUSH TABLES'),
an error is produced.
In order to be able to perform DDL on a table under LOCK TABLES,
the table must be locked explicitly in the LOCK TABLES list.
mysql-test/include/handler.inc:
Adjusted test case to trigger an execution path on which bug 41110
"crash with handler command when used concurrently with alter
table" and bug 41112 "crash in mysql_ha_close_table/get_lock_data
with alter table" were originally discovered. Left old test case
which no longer triggers this execution path for the sake of
coverage.
Added test coverage for HANDLER SQL statements and type-aware
metadata locks.
Added a test for the global shared lock and HANDLER SQL.
Updated tests to take into account that the old simple deadlock
detection heuristics was replaced with a graph-based deadlock
detector.
mysql-test/r/debug_sync.result:
Updated results (see debug_sync.test).
mysql-test/r/handler_innodb.result:
Updated results (see handler.inc test).
mysql-test/r/handler_myisam.result:
Updated results (see handler.inc test).
mysql-test/r/innodb-lock.result:
Updated results (see innodb-lock.test).
mysql-test/r/innodb_mysql_lock.result:
Updated results (see innodb_mysql_lock.test).
mysql-test/r/lock.result:
Updated results (see lock.test).
mysql-test/r/lock_multi.result:
Updated results (see lock_multi.test).
mysql-test/r/lock_sync.result:
Updated results (see lock_sync.test).
mysql-test/r/mdl_sync.result:
Updated results (see mdl_sync.test).
mysql-test/r/sp-threads.result:
SHOW PROCESSLIST output has changed due to the fact that waiting
for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/r/truncate_coverage.result:
Updated results (see truncate_coverage.test).
mysql-test/suite/funcs_1/datadict/processlist_val.inc:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/funcs_1/r/processlist_val_no_prot.result:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/rpl/t/rpl_sp.test:
Updated to a new SHOW PROCESSLIST state name.
mysql-test/t/debug_sync.test:
Use LOCK TABLES READ instead of LOCK TABLES WRITE as the latter
no longer allows to trigger execution path involving waiting on
thr_lock.c lock and therefore reaching debug sync-point covered
by this test.
mysql-test/t/innodb-lock.test:
Adjusted test case to the fact that innodb_table_locks=0 option is
no longer supported, since LOCK TABLES WRITE handles all its
conflicts within MDL subsystem.
mysql-test/t/innodb_mysql_lock.test:
Added test for bug #37346 "innodb does not detect deadlock between
update and alter table".
mysql-test/t/lock.test:
Added test coverage which checks the fact that we no longer support
DDL under LOCK TABLES on tables which were locked implicitly.
Adjusted existing test cases accordingly.
mysql-test/t/lock_multi.test:
Added test for bug #46272 "MySQL 5.4.4, new MDL: unnecessary
deadlock". Adjusted other test cases to take into account the
fact that waiting for LOCK TABLES ... WRITE now happens within MDL
subsystem.
mysql-test/t/lock_sync.test:
Since LOCK TABLES ... WRITE now takes SNRW metadata lock for
tables locked explicitly we have to implicitly lock InnoDB tables
(through view) to trigger the table-level lock conflict between
TL_WRITE and TL_WRITE_ALLOW_WRITE.
mysql-test/t/mdl_sync.test:
Added basic test coverage for type-of-operation-aware metadata
locks. Also covered with tests some use cases involving HANDLER
statements in which a deadlock could arise.
Adjusted existing tests to take type-of-operation-aware MDL into
account.
mysql-test/t/multi_update.test:
Update to a new SHOW PROCESSLIST state name.
mysql-test/t/truncate_coverage.test:
Adjusted test case after making LOCK TABLES WRITE to wait until
transactions that use the table to be locked are completed.
Updated to the changed name of DEBUG_SYNC point.
sql/handler.cc:
Global read lock functionality has been
moved into a class.
sql/lock.cc:
Global read lock functionality has been
moved into a class.
Updated code to use the new MDL API.
sql/mdl.cc:
Introduced new type-of-operation aware metadata locks.
To do this:
- Changed MDL_lock to use one list for waiting requests and one
list for granted requests. For each list, added a bitmap
that holds information what lock types a list contains.
Added a helper class MDL_lock::List to manipulate with granted
and waited lists while keeping the bitmaps in sync
with list contents.
- Changed lock-compatibility functions to use bitmaps that
define compatibility.
- Introduced a graph based deadlock detector inspired by
waiting_threads.c from Maria implementation.
- Now that we have a deadlock detector, and no longer have
a global lock to protect individual lock objects, but rather
use an rw lock per object, removed redundant code for upgrade,
and the global read lock. Changed the MDL API to
no longer require the caller to acquire the global
intention exclusive lock by means of a separate method.
Removed a few more methods that became redundant.
- Removed deadlock detection heuristic, it has been made
obsolete by the deadlock detector.
- With operation-type-aware metadata locks, MDL subsystem has
become aware of potential conflicts between DDL and open
transactions. This made it possible to remove calls to
mysql_abort_transactions_with_shared_lock() from acquisition
paths for exclusive lock and lock upgrade. Now we can simply
wait for these transactions to complete without fear of
deadlock. Function mysql_lock_abort() has also become
unnecessary for all conflicting cases except when a DDL
conflicts with a connection that has an open HANDLER.
sql/mdl.h:
Introduced new type-of-operation aware metadata locks.
Introduced a graph based deadlock detector and supporting
methods.
Added comments.
God rid of redundant API calls.
Renamed m_lt_or_ha_sentinel to m_trans_sentinel,
since now it guards the global read lock as well as
LOCK TABLES and HANDLER locks.
sql/mysql_priv.h:
Moved the global read lock functionality into a
class.
Added MYSQL_OPEN_FORCE_SHARED_MDL flag which forces
open_tables() to take MDL_SHARED on tables instead of
metadata locks specified in the parser. We use this to
allow PREPARE run concurrently in presence of
LOCK TABLES ... WRITE.
Added signature for find_table_for_mdl_ugprade().
sql/set_var.cc:
Global read lock functionality has been
moved into a class.
sql/sp_head.cc:
When creating TABLE_LIST elements for prelocking or
system tables set the type of request for metadata
lock according to the operation that will be performed
on the table.
sql/sql_base.cc:
- Updated code to use the new MDL API.
- In order to avoid locks starvation we take upgradable
locks all at once. As result implicitly locked tables no
longer get an upgradable lock. Consequently DDL and FLUSH
TABLES for such tables is prohibited.
find_write_locked_table() was replaced by
find_table_for_mdl_upgrade() function.
open_table() was adjusted to return TABLE instance with
upgradable ticket when necessary.
- We no longer wait for all locks on OT_WAIT back off
action -- only on the lock that caused the wait
conflict. Moreover, now we distinguish cases when we
have to wait due to conflict in MDL and old version
of table in TDC.
- Upate mysql_notify_threads_having_share_locks()
to only abort thr_lock.c waits of threads that
have open HANDLERs, since lock conflicts with only
these threads now can lead to deadlocks not detectable
by the MDL deadlock detector.
- Remove mysql_abort_transactions_with_shared_locks()
which is no longer needed.
sql/sql_class.cc:
Global read lock functionality has been moved into a class.
Re-arranged code in THD::cleanup() to simplify assert.
sql/sql_class.h:
Introduced class to incapsulate global read lock
functionality.
Now sentinel in MDL subsystem guards the global read lock
as well as LOCK TABLES and HANDLER locks. Adjusted code
accordingly.
sql/sql_db.cc:
Global read lock functionality has been moved into a class.
sql/sql_delete.cc:
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
TRUNCATE TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_handler.cc:
Inform MDL_context about presence of open HANDLERs.
Since HANLDERs break MDL protocol by acquiring table-level
lock while holding only S metadata lock on a table MDL
subsystem should take special care about such contexts (Now
this is the only case when mysql_lock_abort() is used).
sql/sql_parse.cc:
Global read lock functionality has been moved into a class.
Do not take upgradable metadata locks when opening tables
for CREATE TABLE SELECT as it is not necessary and limits
concurrency.
When initializing TABLE_LIST objects before adding them
to the table list set the type of request for metadata lock
according to the operation that will be performed on the
table.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result FLUSH
TABLES is no longer allowed for such tables.
sql/sql_prepare.cc:
Use MYSQL_OPEN_FORCE_SHARED_MDL flag when opening
tables during PREPARE. This allows PREPARE to run
concurrently in presence of LOCK TABLES ... WRITE.
sql/sql_rename.cc:
Global read lock functionality has been moved into a class.
sql/sql_show.cc:
Updated code to use the new MDL API.
sql/sql_table.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result DROP
TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_trigger.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
CREATE/DROP TRIGGER is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_view.cc:
Global read lock functionality has been moved into a class.
Fixed results of wrong merge that led to misuse of GLR API.
CREATE VIEW statement is not a commit statement.
sql/table.cc:
When resetting TABLE_LIST objects for PS or SP re-execution
set the type of request for metadata lock according to the
operation that will be performed on the table. Do the same
in auxiliary function initializing metadata lock requests
in a table list.
sql/table.h:
When initializing TABLE_LIST objects set the type of request
for metadata lock according to the operation that will be
performed on the table.
sql/transaction.cc:
Global read lock functionality has been moved into a class.
2010-02-01 12:43:06 +01:00
ERROR HY000: Table storage engine for 't1' doesn't have this option
handler t1 close;
unlock tables;
drop table t1;
# Now test case which was reported originally but which no longer
# triggers execution path which has caused the problem.
create table t1 (a int, key(a));
insert into t1 values (1);
handler t1 open;
alter table t1 engine=memory;
# Since S metadata lock was already acquired at HANDLER OPEN time
# and TL_READ lock requested by HANDLER READ is compatible with
# ALTER's TL_WRITE_ALLOW_READ the below statement should succeed
# without waiting. The old version of table should be used in it.
handler t1 read a next;
2009-12-30 18:53:30 +01:00
a
1
2009-02-24 10:49:18 +01:00
handler t1 close;
drop table t1;
2009-04-17 09:41:15 +02:00
USE information_schema;
HANDLER COLUMNS OPEN;
ERROR HY000: Incorrect usage of HANDLER OPEN and information_schema
USE test;
2009-08-21 07:55:35 +02:00
#
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
# Add test coverage for HANDLER and LOCK TABLES, HANDLER and DDL.
#
drop table if exists t1, t2, t3;
create table t1 (a int, key a (a));
insert into t1 (a) values (1), (2), (3), (4), (5);
create table t2 (a int, key a (a)) select * from t1;
create temporary table t3 (a int, key a (a)) select * from t2;
handler t1 open;
handler t2 open;
handler t3 open;
#
2010-02-12 08:05:43 +01:00
# No HANDLER sql is allowed under LOCK TABLES.
# But it does not implicitly closes all handlers.
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
#
2010-02-12 08:05:43 +01:00
lock table t1 read;
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
handler t1 open;
ERROR HY000: Can't execute the given command because you have active locked tables or an active transaction
handler t1 read next;
ERROR HY000: Can't execute the given command because you have active locked tables or an active transaction
handler t2 close;
ERROR HY000: Can't execute the given command because you have active locked tables or an active transaction
handler t3 open;
ERROR HY000: Can't execute the given command because you have active locked tables or an active transaction
2010-02-12 08:05:43 +01:00
# After UNLOCK TABLES handlers should be around and
# we should be able to continue reading through them.
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
unlock tables;
handler t1 read next;
2010-02-12 08:05:43 +01:00
a
1
handler t1 close;
handler t2 read next;
a
1
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
handler t2 close;
handler t3 read next;
2010-02-12 08:05:43 +01:00
a
1
handler t3 close;
drop temporary table t3;
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
#
2010-02-12 08:05:43 +01:00
# Other operations that implicitly close handler:
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
#
# TRUNCATE
#
handler t1 open;
truncate table t1;
handler t1 read next;
ERROR 42S02: Unknown table 't1' in HANDLER
handler t1 open;
#
# CREATE TRIGGER
#
create trigger t1_ai after insert on t1 for each row set @a=1;
handler t1 read next;
ERROR 42S02: Unknown table 't1' in HANDLER
#
# DROP TRIGGER
#
handler t1 open;
drop trigger t1_ai;
handler t1 read next;
ERROR 42S02: Unknown table 't1' in HANDLER
#
# ALTER TABLE
#
handler t1 open;
alter table t1 add column b int;
handler t1 read next;
ERROR 42S02: Unknown table 't1' in HANDLER
#
# ANALYZE TABLE
#
handler t1 open;
analyze table t1;
Table Op Msg_type Msg_text
test.t1 analyze status Table is already up to date
handler t1 read next;
ERROR 42S02: Unknown table 't1' in HANDLER
#
# OPTIMIZE TABLE
#
handler t1 open;
optimize table t1;
Table Op Msg_type Msg_text
test.t1 optimize status OK
handler t1 read next;
ERROR 42S02: Unknown table 't1' in HANDLER
#
# REPAIR TABLE
#
handler t1 open;
repair table t1;
Table Op Msg_type Msg_text
test.t1 repair status OK
handler t1 read next;
ERROR 42S02: Unknown table 't1' in HANDLER
#
# DROP TABLE, naturally.
#
handler t1 open;
drop table t1;
handler t1 read next;
ERROR 42S02: Unknown table 't1' in HANDLER
create table t1 (a int, b int, key a (a)) select a from t2;
#
# RENAME TABLE, naturally
#
handler t1 open;
rename table t1 to t3;
handler t1 read next;
ERROR 42S02: Unknown table 't1' in HANDLER
#
# CREATE TABLE (even with IF NOT EXISTS clause,
# and the table exists).
#
handler t2 open;
create table if not exists t2 (a int);
Warnings:
Note 1050 Table 't2' already exists
handler t2 read next;
ERROR 42S02: Unknown table 't2' in HANDLER
rename table t3 to t1;
drop table t2;
#
# FLUSH TABLE doesn't close the table but loses the position
#
handler t1 open;
handler t1 read a prev;
b a
NULL 5
flush table t1;
handler t1 read a prev;
b a
NULL 5
handler t1 close;
#
# FLUSH TABLES WITH READ LOCK behaves like FLUSH TABLE.
#
handler t1 open;
handler t1 read a prev;
b a
NULL 5
flush tables with read lock;
handler t1 read a prev;
b a
NULL 5
handler t1 close;
unlock tables;
#
2010-02-12 08:05:43 +01:00
# Let us also check that these operations behave in similar
# way under LOCK TABLES.
#
# TRUNCATE under LOCK TABLES.
#
handler t1 open;
lock tables t1 write;
truncate table t1;
unlock tables;
handler t1 read next;
ERROR 42S02: Unknown table 't1' in HANDLER
handler t1 open;
#
# CREATE TRIGGER under LOCK TABLES.
#
lock tables t1 write;
create trigger t1_ai after insert on t1 for each row set @a=1;
unlock tables;
handler t1 read next;
ERROR 42S02: Unknown table 't1' in HANDLER
#
# DROP TRIGGER under LOCK TABLES.
#
handler t1 open;
lock tables t1 write;
drop trigger t1_ai;
unlock tables;
handler t1 read next;
ERROR 42S02: Unknown table 't1' in HANDLER
#
# ALTER TABLE under LOCK TABLES.
#
handler t1 open;
lock tables t1 write;
alter table t1 drop column b;
unlock tables;
handler t1 read next;
ERROR 42S02: Unknown table 't1' in HANDLER
#
# ANALYZE TABLE under LOCK TABLES.
#
handler t1 open;
lock tables t1 write;
analyze table t1;
Table Op Msg_type Msg_text
test.t1 analyze status Table is already up to date
unlock tables;
handler t1 read next;
ERROR 42S02: Unknown table 't1' in HANDLER
#
# OPTIMIZE TABLE under LOCK TABLES.
#
handler t1 open;
lock tables t1 write;
optimize table t1;
Table Op Msg_type Msg_text
test.t1 optimize status OK
unlock tables;
handler t1 read next;
ERROR 42S02: Unknown table 't1' in HANDLER
#
# REPAIR TABLE under LOCK TABLES.
#
handler t1 open;
lock tables t1 write;
repair table t1;
Table Op Msg_type Msg_text
test.t1 repair status OK
unlock tables;
handler t1 read next;
ERROR 42S02: Unknown table 't1' in HANDLER
#
# DROP TABLE under LOCK TABLES, naturally.
#
handler t1 open;
lock tables t1 write;
drop table t1;
unlock tables;
handler t1 read next;
ERROR 42S02: Unknown table 't1' in HANDLER
create table t1 (a int, b int, key a (a));
insert into t1 (a) values (1), (2), (3), (4), (5);
#
# FLUSH TABLE doesn't close the table but loses the position
#
handler t1 open;
handler t1 read a prev;
a b
5 NULL
lock tables t1 write;
flush table t1;
unlock tables;
handler t1 read a prev;
a b
5 NULL
handler t1 close;
#
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
# Explore the effect of HANDLER locks on concurrent DDL
#
handler t1 open;
# Establishing auxiliary connections con1, con2, con3
# --> connection con1;
# Sending:
drop table t1 ;
# We can't use connection 'default' as wait_condition will
# autoclose handlers.
# --> connection con2
# Waitng for 'drop table t1' to get blocked...
# --> connection default
handler t1 read a prev;
2010-02-12 08:05:43 +01:00
a b
5 NULL
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
handler t1 read a prev;
2010-02-12 08:05:43 +01:00
a b
4 NULL
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
handler t1 close;
# --> connection con1
# Reaping 'drop table t1'...
# --> connection default
#
# Explore the effect of HANDLER locks in parallel with SELECT
#
create table t1 (a int, key a (a));
insert into t1 (a) values (1), (2), (3), (4), (5);
begin;
select * from t1;
a
1
2
3
4
5
handler t1 open;
handler t1 read a prev;
a
5
handler t1 read a prev;
a
4
handler t1 close;
# --> connection con1;
# Sending:
drop table t1 ;
# --> connection con2
# Waiting for 'drop table t1' to get blocked...
# --> connection default
# We can still use the table, it's part of the transaction
select * from t1;
a
1
2
3
4
5
# Such are the circumstances that t1 is a part of transaction,
# thus we can reopen it in the handler
handler t1 open;
# We can commit the transaction, it doesn't close the handler
# and doesn't let DROP to proceed.
commit;
handler t1 read a prev;
a
5
handler t1 read a prev;
a
4
handler t1 read a prev;
a
3
handler t1 close;
# --> connection con1
# Now drop can proceed
# Reaping 'drop table t1'...
# --> connection default
#
# Demonstrate that HANDLER locks and transaction locks
# reside in the same context, and we don't back-off
# when have transaction or handler locks.
#
create table t1 (a int, key a (a));
insert into t1 (a) values (1), (2), (3), (4), (5);
Patch that changes metadata locking subsystem to use mutex per lock and
condition variable per context instead of one mutex and one conditional
variable for the whole subsystem.
This should increase concurrency in this subsystem.
It also opens the way for further changes which are necessary to solve
such bugs as bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock"
and bug #37346 "innodb does not detect deadlock between update and alter
table".
Two other notable changes done by this patch:
- MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object metadata lock is acquired. Now this has
to be done by explicit calls outside of MDL subsystem.
- Instead of using separate MDL_context for opening system tables/tables
for purposes of I_S we now create MDL savepoint in the main context
before opening tables and rollback to this savepoint after closing
them. This means that it is now possible to get ER_LOCK_DEADLOCK error
even not inside a transaction. This might happen in unlikely case when
one runs DDL on one of system tables while also running DDL on some
other tables. Cases when this ER_LOCK_DEADLOCK error is not justified
will be addressed by advanced deadlock detector for MDL subsystem which
we plan to implement.
mysql-test/include/handler.inc:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/handler_innodb.result:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/handler_myisam.result:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/mdl_sync.result:
Adjusted mdl_sync.test to the fact that exclusive metadata locks on
tables are now acquired according to alphabetical order of fully
qualified table names instead of order in which tables are mentioned
in statement.
mysql-test/t/mdl_sync.test:
Adjusted mdl_sync.test to the fact that exclusive metadata locks on
tables are now acquired according to alphabetical order of fully
qualified table names instead of order in which tables are mentioned
in statement.
sql/events.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/ha_ndbcluster.cc:
Since manipulations with open table state no longer install proxy
MDL_context it does not make sense to perform them in order to
satisfy assert in mysql_rm_tables_part2(). Removed them per agreement
with Cluster team. This has not broken test suite since scenario in
which deadlock can occur and assertion fails is not covered by tests.
sql/lock.cc:
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object exclusive metadata lock is acquired.
Now this has to be done by explicit calls outside of MDL subsystem.
sql/log.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/mdl.cc:
Changed metadata locking subsystem to use mutex per lock and condition
variable per context instead of one mutex and one conditional variable
for the whole subsystem.
Changed approach to handling of global metadata locks. Instead of
implicitly acquiring intention exclusive locks when user requests
per-object upgradeable or exclusive locks now we require them to be
acquired explicitly in the same way as ordinary metadata locks.
In fact global lock are now ordinary metadata locks in new GLOBAL
namespace.
To implement these changes:
- Removed LOCK_mdl mutex and COND_mdl condition variable.
- Introduced MDL_lock::m_mutex mutexes which protect individual lock
objects.
- Replaced mdl_locks hash with MDL_map class, which has hash for
MDL_lock objects as a member and separate mutex which protects this
hash. Methods of this class allow to find(), find_or_create() or
remove() MDL_lock objects in concurrency-friendly fashion (i.e.
for most common operation, find_or_create(), we don't acquire
MDL_lock::m_mutex while holding MDL_map::m_mutex. Thanks to MikaelR
for this idea and benchmarks!). Added three auxiliary members to
MDL_lock class (m_is_destroyed, m_ref_usage, m_ref_release) to
support this concurrency-friendly behavior.
- Introduced MDL_context::m_ctx_wakeup_cond condition variable to be
used for waiting until this context's pending request can be
satisfied or its thread has to perform actions to resolve potential
deadlock. Context which want to wait add ticket corresponding to the
request to an appropriate queue of waiters in MDL_lock object so
they can be noticed when other contexts change state of lock and be
awaken by them by signalling on MDL_context::m_ctx_wakeup_cond.
As consequence MDL_ticket objects has to be used for any waiting
in metadata locking subsystem including one which happens in
MDL_context::wait_for_locks() method.
Another consequence is that MDL_context is no longer copyable and
can't be saved/restored when working with system tables.
- Made MDL_lock an abstract class, which delegates specifying exact
compatibility matrix to its descendants. Added MDL_global_lock child
class for global lock (The old is_lock_type_compatible() method
became can_grant_lock() method of this class). Added MDL_object_lock
class to represent per-object lock (The old MDL_lock::can_grant_lock()
became its method). Choice between two classes happens based on MDL
namespace in MDL_lock::create() method.
- Got rid of MDL_lock::type member as its meaning became ambigous for
global locks.
- To simplify waking up of contexts waiting for lock split waiting queue
in MDL_lock class in two queues. One for pending requests for exclusive
(including intention exclusive) locks and another for requests for
shared locks.
- Added virtual wake_up_waiters() method to MDL_lock, MDL_global_lock and
MDL_object_lock classes which allows to wake up waiting contexts after
state of lock changes. Replaced old duplicated code with calls to this
method.
- Adjusted MDL_context::try_acquire_shared_lock()/exclusive_lock()/
global_shared_lock(), MDL_ticket::upgrade_shared_lock_to_exclusive_lock()
and MDL_context::release_ticket() methods to use MDL_map and
MDL_lock::m_mutex instead of single LOCK_mdl mutex and wake up
waiters according to the approach described above. The latter method
also was renamed to MDL_context::release_lock().
- Changed MDL_context::try_acquire_shared_lock()/exclusive_lock() and
release_lock() not to handle global locks. They are now supposed to
be taken explicitly like ordinary metadata locks.
- Added helper MDL_context::try_acquire_global_intention_exclusive_lock()
and acquire_global_intention_exclusive_lock() methods.
- Moved common code from MDL_context::acquire_global_shared_lock() and
acquire_global_intention_exclusive_lock() to new method -
MDL_context::acquire_lock_impl().
- Moved common code from MDL_context::try_acquire_shared_lock(),
try_acquire_global_intention_exclusive_lock()/exclusive_lock()
to MDL_context::try_acquire_lock_impl().
- Since acquiring of several exclusive locks can no longer happen under
single LOCK_mdl mutex the approach to it had to be changed. Now we do
it in one by one fashion. This is done in alphabetical order to avoid
deadlocks. Changed MDL_context::acquire_exclusive_locks() accordingly
(as part of this change moved code responsible for acquiring single
exclusive lock to new MDL_context::acquire_exclusive_lock_impl()
method).
- Since we no longer have single LOCK_mdl mutex which protects all
MDL_context::m_is_waiting_in_mdl members using these members to
determine if we have really awaken context holding conflicting
shared lock became inconvinient. Got rid of this member and changed
notify_shared_lock() helper function and process of acquiring
of/upgrading to exclusive lock not to rely on such information.
Now in MDL_context::acquire_exclusive_lock_impl() and
MDL_ticket::upgrade_shared_lock_to_exclusive_lock() we simply
re-try to wake up threads holding conflicting shared locks after
small time out.
- Adjusted MDL_context::can_wait_lead_to_deadlock() and
MDL_ticket::has_pending_conflicting_lock() to use per-lock
mutexes instead of LOCK_mdl. To do this introduced
MDL_lock::has_pending_exclusive_lock() method.
sql/mdl.h:
Changed metadata locking subsystem to use mutex per lock and condition
variable per context instead of one mutex and one conditional variable
for the whole subsystem. In order to implement this change:
- Added MDL_key::cmp() method to be able to sort MDL_key objects
alphabetically. Changed length fields in MDL_key class to uint16
as 16-bit is enough for length of any key.
- Changed MDL_ticket::get_ctx() to return pointer to non-const
object in order to be able to use MDL_context::awake() method
for such contexts.
- Got rid of unlocked versions of can_wait_lead_to_deadlock()/
has_pending_conflicting_lock() methods in MDL_context and
MDL_ticket. We no longer has single mutex which protects all
locks. Thus one always has to use versions of these methods
which acquire per-lock mutexes.
- MDL_request_list type of list now counts its elements.
- Added MDL_context::m_ctx_wakeup_cond condition variable to be used
for waiting until this context's pending request can be satisfied
or its thread has to perform actions to resolve potential deadlock.
Added awake() method to wake up context from such wait.
Addition of condition variable made MDL_context uncopyable.
As result we no longer can save/restore MDL_context when working
with system tables. Instead we create MDL savepoint before opening
those tables and rollback to it once they are closed.
- MDL_context::release_ticket() became release_lock() method.
- Added auxiliary MDL_context::acquire_exclusive_lock_impl() method
which does all necessary work to acquire exclusive lock on one object
but should not be used directly as it does not enforce any asserts
ensuring that no deadlocks are possible.
- Since we no longer need to know if thread trying to acquire exclusive
lock managed to wake up any threads having conflicting shared locks
(as, anyway, we will try to wake up such threads again shortly)
- MDL_context::m_is_waiting_in_mdl member became unnecessary and
notify_shared_lock() no longer needs to be friend of MDL_context.
Changed approach to handling of global metadata locks. Instead of
implicitly acquiring intention exclusive locks when user requests
per-object upgradeable or exclusive locks now we require them to be
acquired explicitly in the same way as ordinary metadata locks.
- Added new GLOBAL namespace for such locks.
- Added new type of lock to be requested MDL_INTENTION_EXCLISIVE.
- Added MDL_context::try_acquire_global_intention_exclusive_lock()
and acquire_global_intention_exclusive_lock() methods.
- Moved common code from MDL_context::acquire_global_shared_lock()
and acquire_global_intention_exclusive_lock() to new method -
MDL_context::acquire_lock_impl().
- Moved common code from MDL_context::try_acquire_shared_lock(),
try_acquire_global_intention_exclusive_lock()/exclusive_lock()
to MDL_context::try_acquire_lock_impl().
- Added helper MDL_context::is_global_lock_owner() method to be
able easily to find what kind of global lock this context holds.
- MDL_context::m_has_global_shared_lock became unnecessary as
global read lock is now represented by ordinary ticket.
- Removed assert in MDL_context::set_lt_or_ha_sentinel() which became
false for cases when we execute LOCK TABLES under global read lock
mode.
sql/mysql_priv.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result calls
opening/closing system tables were changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sp.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sp.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sql_base.cc:
close_thread_tables():
Since we no longer use separate MDL_context for opening system
tables we need to avoid releasing all transaction locks when
closing system table. Releasing metadata lock on system table
is now responsibility of THD::restore_backup_open_tables_state().
open_table_get_mdl_lock(),
Open_table_context::recover_from_failed_open():
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object upgradable or exclusive metadata lock is
acquired. So this have to be done explicitly from these calls.
Changed Open_table_context class to store MDL_request object for
global intention exclusive lock acquired when opening tables.
open_table():
Do not release metadata lock if we have failed to open table as
this lock might have been acquired by one of previous statements
in transaction, and therefore should not be released.
open_system_tables_for_read()/close_system_tables()/
open_performance_schema_table():
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
close_performance_schema_table():
Got rid of duplicated code.
sql/sql_class.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. Also releasing
metadata lock on system table is now responsibility of
THD::restore_backup_open_tables_state().
Adjusted assert in THD::cleanup() to take into account fact that now
we also use MDL sentinel for global read lock.
sql/sql_class.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. As result:
- 'mdl_context' member was moved out of Open_tables_state to THD class.
enter_locked_tables_mode()/leave_locked_tables_mode() had to follow.
- Methods of THD responsible for saving/restoring open table state were
changed to use Open_tables_backup class which in addition to
Open_table_state has a member for this savepoint.
Changed Open_table_context class to store MDL_request object for
global intention exclusive lock acquired when opening tables.
sql/sql_delete.cc:
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object exclusive metadata lock is acquired.
Now this has to be done by explicit calls outside of MDL subsystem.
sql/sql_help.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sql_parse.cc:
Adjusted assert reload_acl_and_cache() to the fact that global read
lock now takes full-blown metadata lock.
sql/sql_plist.h:
Added support for element counting to I_P_List list template.
One can use policy classes to specify if such counting is needed
or not needed for particular list.
sql/sql_show.cc:
Instead of using separate MDL_context for opening tables for I_S
purposes we now create MDL savepoint in the main context before
opening tables and rollback to this savepoint after closing them.
To support this and similar change for system tables methods of
THD responsible for saving/restoring open table state were changed
to use Open_tables_backup class which in addition to Open_table_state
has a member for this savepoint. As result code opening/closing tables
for I_S purposes was changed to use Open_tables_backup instead of
Open_table_state class as well.
sql/sql_table.cc:
mysql_rm_tables_part2():
Since now global intention exclusive metadata lock is ordinary
metadata lock we no longer can rely that by releasing MDL locks
on all tables we will release all locks acquired by this routine.
So in non-LOCK-TABLES mode we have to release all locks acquired
explicitly.
prepare_for_repair(), mysql_alter_table():
MDL subsystem no longer implicitly acquires global intention
exclusive metadata lock when per-object exclusive metadata lock
is acquired. Now this has to be done by explicit calls outside of
MDL subsystem.
sql/tztime.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
Also changed code not to use special mechanism for open system tables
when it is not really necessary.
2010-01-21 21:43:03 +01:00
create table t0 (a int, key a (a));
insert into t0 (a) values (1), (2), (3), (4), (5);
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
begin;
select * from t1;
a
1
2
3
4
5
# --> connection con2
# Sending:
Patch that changes metadata locking subsystem to use mutex per lock and
condition variable per context instead of one mutex and one conditional
variable for the whole subsystem.
This should increase concurrency in this subsystem.
It also opens the way for further changes which are necessary to solve
such bugs as bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock"
and bug #37346 "innodb does not detect deadlock between update and alter
table".
Two other notable changes done by this patch:
- MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object metadata lock is acquired. Now this has
to be done by explicit calls outside of MDL subsystem.
- Instead of using separate MDL_context for opening system tables/tables
for purposes of I_S we now create MDL savepoint in the main context
before opening tables and rollback to this savepoint after closing
them. This means that it is now possible to get ER_LOCK_DEADLOCK error
even not inside a transaction. This might happen in unlikely case when
one runs DDL on one of system tables while also running DDL on some
other tables. Cases when this ER_LOCK_DEADLOCK error is not justified
will be addressed by advanced deadlock detector for MDL subsystem which
we plan to implement.
mysql-test/include/handler.inc:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/handler_innodb.result:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/handler_myisam.result:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/mdl_sync.result:
Adjusted mdl_sync.test to the fact that exclusive metadata locks on
tables are now acquired according to alphabetical order of fully
qualified table names instead of order in which tables are mentioned
in statement.
mysql-test/t/mdl_sync.test:
Adjusted mdl_sync.test to the fact that exclusive metadata locks on
tables are now acquired according to alphabetical order of fully
qualified table names instead of order in which tables are mentioned
in statement.
sql/events.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/ha_ndbcluster.cc:
Since manipulations with open table state no longer install proxy
MDL_context it does not make sense to perform them in order to
satisfy assert in mysql_rm_tables_part2(). Removed them per agreement
with Cluster team. This has not broken test suite since scenario in
which deadlock can occur and assertion fails is not covered by tests.
sql/lock.cc:
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object exclusive metadata lock is acquired.
Now this has to be done by explicit calls outside of MDL subsystem.
sql/log.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/mdl.cc:
Changed metadata locking subsystem to use mutex per lock and condition
variable per context instead of one mutex and one conditional variable
for the whole subsystem.
Changed approach to handling of global metadata locks. Instead of
implicitly acquiring intention exclusive locks when user requests
per-object upgradeable or exclusive locks now we require them to be
acquired explicitly in the same way as ordinary metadata locks.
In fact global lock are now ordinary metadata locks in new GLOBAL
namespace.
To implement these changes:
- Removed LOCK_mdl mutex and COND_mdl condition variable.
- Introduced MDL_lock::m_mutex mutexes which protect individual lock
objects.
- Replaced mdl_locks hash with MDL_map class, which has hash for
MDL_lock objects as a member and separate mutex which protects this
hash. Methods of this class allow to find(), find_or_create() or
remove() MDL_lock objects in concurrency-friendly fashion (i.e.
for most common operation, find_or_create(), we don't acquire
MDL_lock::m_mutex while holding MDL_map::m_mutex. Thanks to MikaelR
for this idea and benchmarks!). Added three auxiliary members to
MDL_lock class (m_is_destroyed, m_ref_usage, m_ref_release) to
support this concurrency-friendly behavior.
- Introduced MDL_context::m_ctx_wakeup_cond condition variable to be
used for waiting until this context's pending request can be
satisfied or its thread has to perform actions to resolve potential
deadlock. Context which want to wait add ticket corresponding to the
request to an appropriate queue of waiters in MDL_lock object so
they can be noticed when other contexts change state of lock and be
awaken by them by signalling on MDL_context::m_ctx_wakeup_cond.
As consequence MDL_ticket objects has to be used for any waiting
in metadata locking subsystem including one which happens in
MDL_context::wait_for_locks() method.
Another consequence is that MDL_context is no longer copyable and
can't be saved/restored when working with system tables.
- Made MDL_lock an abstract class, which delegates specifying exact
compatibility matrix to its descendants. Added MDL_global_lock child
class for global lock (The old is_lock_type_compatible() method
became can_grant_lock() method of this class). Added MDL_object_lock
class to represent per-object lock (The old MDL_lock::can_grant_lock()
became its method). Choice between two classes happens based on MDL
namespace in MDL_lock::create() method.
- Got rid of MDL_lock::type member as its meaning became ambigous for
global locks.
- To simplify waking up of contexts waiting for lock split waiting queue
in MDL_lock class in two queues. One for pending requests for exclusive
(including intention exclusive) locks and another for requests for
shared locks.
- Added virtual wake_up_waiters() method to MDL_lock, MDL_global_lock and
MDL_object_lock classes which allows to wake up waiting contexts after
state of lock changes. Replaced old duplicated code with calls to this
method.
- Adjusted MDL_context::try_acquire_shared_lock()/exclusive_lock()/
global_shared_lock(), MDL_ticket::upgrade_shared_lock_to_exclusive_lock()
and MDL_context::release_ticket() methods to use MDL_map and
MDL_lock::m_mutex instead of single LOCK_mdl mutex and wake up
waiters according to the approach described above. The latter method
also was renamed to MDL_context::release_lock().
- Changed MDL_context::try_acquire_shared_lock()/exclusive_lock() and
release_lock() not to handle global locks. They are now supposed to
be taken explicitly like ordinary metadata locks.
- Added helper MDL_context::try_acquire_global_intention_exclusive_lock()
and acquire_global_intention_exclusive_lock() methods.
- Moved common code from MDL_context::acquire_global_shared_lock() and
acquire_global_intention_exclusive_lock() to new method -
MDL_context::acquire_lock_impl().
- Moved common code from MDL_context::try_acquire_shared_lock(),
try_acquire_global_intention_exclusive_lock()/exclusive_lock()
to MDL_context::try_acquire_lock_impl().
- Since acquiring of several exclusive locks can no longer happen under
single LOCK_mdl mutex the approach to it had to be changed. Now we do
it in one by one fashion. This is done in alphabetical order to avoid
deadlocks. Changed MDL_context::acquire_exclusive_locks() accordingly
(as part of this change moved code responsible for acquiring single
exclusive lock to new MDL_context::acquire_exclusive_lock_impl()
method).
- Since we no longer have single LOCK_mdl mutex which protects all
MDL_context::m_is_waiting_in_mdl members using these members to
determine if we have really awaken context holding conflicting
shared lock became inconvinient. Got rid of this member and changed
notify_shared_lock() helper function and process of acquiring
of/upgrading to exclusive lock not to rely on such information.
Now in MDL_context::acquire_exclusive_lock_impl() and
MDL_ticket::upgrade_shared_lock_to_exclusive_lock() we simply
re-try to wake up threads holding conflicting shared locks after
small time out.
- Adjusted MDL_context::can_wait_lead_to_deadlock() and
MDL_ticket::has_pending_conflicting_lock() to use per-lock
mutexes instead of LOCK_mdl. To do this introduced
MDL_lock::has_pending_exclusive_lock() method.
sql/mdl.h:
Changed metadata locking subsystem to use mutex per lock and condition
variable per context instead of one mutex and one conditional variable
for the whole subsystem. In order to implement this change:
- Added MDL_key::cmp() method to be able to sort MDL_key objects
alphabetically. Changed length fields in MDL_key class to uint16
as 16-bit is enough for length of any key.
- Changed MDL_ticket::get_ctx() to return pointer to non-const
object in order to be able to use MDL_context::awake() method
for such contexts.
- Got rid of unlocked versions of can_wait_lead_to_deadlock()/
has_pending_conflicting_lock() methods in MDL_context and
MDL_ticket. We no longer has single mutex which protects all
locks. Thus one always has to use versions of these methods
which acquire per-lock mutexes.
- MDL_request_list type of list now counts its elements.
- Added MDL_context::m_ctx_wakeup_cond condition variable to be used
for waiting until this context's pending request can be satisfied
or its thread has to perform actions to resolve potential deadlock.
Added awake() method to wake up context from such wait.
Addition of condition variable made MDL_context uncopyable.
As result we no longer can save/restore MDL_context when working
with system tables. Instead we create MDL savepoint before opening
those tables and rollback to it once they are closed.
- MDL_context::release_ticket() became release_lock() method.
- Added auxiliary MDL_context::acquire_exclusive_lock_impl() method
which does all necessary work to acquire exclusive lock on one object
but should not be used directly as it does not enforce any asserts
ensuring that no deadlocks are possible.
- Since we no longer need to know if thread trying to acquire exclusive
lock managed to wake up any threads having conflicting shared locks
(as, anyway, we will try to wake up such threads again shortly)
- MDL_context::m_is_waiting_in_mdl member became unnecessary and
notify_shared_lock() no longer needs to be friend of MDL_context.
Changed approach to handling of global metadata locks. Instead of
implicitly acquiring intention exclusive locks when user requests
per-object upgradeable or exclusive locks now we require them to be
acquired explicitly in the same way as ordinary metadata locks.
- Added new GLOBAL namespace for such locks.
- Added new type of lock to be requested MDL_INTENTION_EXCLISIVE.
- Added MDL_context::try_acquire_global_intention_exclusive_lock()
and acquire_global_intention_exclusive_lock() methods.
- Moved common code from MDL_context::acquire_global_shared_lock()
and acquire_global_intention_exclusive_lock() to new method -
MDL_context::acquire_lock_impl().
- Moved common code from MDL_context::try_acquire_shared_lock(),
try_acquire_global_intention_exclusive_lock()/exclusive_lock()
to MDL_context::try_acquire_lock_impl().
- Added helper MDL_context::is_global_lock_owner() method to be
able easily to find what kind of global lock this context holds.
- MDL_context::m_has_global_shared_lock became unnecessary as
global read lock is now represented by ordinary ticket.
- Removed assert in MDL_context::set_lt_or_ha_sentinel() which became
false for cases when we execute LOCK TABLES under global read lock
mode.
sql/mysql_priv.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result calls
opening/closing system tables were changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sp.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sp.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sql_base.cc:
close_thread_tables():
Since we no longer use separate MDL_context for opening system
tables we need to avoid releasing all transaction locks when
closing system table. Releasing metadata lock on system table
is now responsibility of THD::restore_backup_open_tables_state().
open_table_get_mdl_lock(),
Open_table_context::recover_from_failed_open():
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object upgradable or exclusive metadata lock is
acquired. So this have to be done explicitly from these calls.
Changed Open_table_context class to store MDL_request object for
global intention exclusive lock acquired when opening tables.
open_table():
Do not release metadata lock if we have failed to open table as
this lock might have been acquired by one of previous statements
in transaction, and therefore should not be released.
open_system_tables_for_read()/close_system_tables()/
open_performance_schema_table():
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
close_performance_schema_table():
Got rid of duplicated code.
sql/sql_class.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. Also releasing
metadata lock on system table is now responsibility of
THD::restore_backup_open_tables_state().
Adjusted assert in THD::cleanup() to take into account fact that now
we also use MDL sentinel for global read lock.
sql/sql_class.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. As result:
- 'mdl_context' member was moved out of Open_tables_state to THD class.
enter_locked_tables_mode()/leave_locked_tables_mode() had to follow.
- Methods of THD responsible for saving/restoring open table state were
changed to use Open_tables_backup class which in addition to
Open_table_state has a member for this savepoint.
Changed Open_table_context class to store MDL_request object for
global intention exclusive lock acquired when opening tables.
sql/sql_delete.cc:
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object exclusive metadata lock is acquired.
Now this has to be done by explicit calls outside of MDL subsystem.
sql/sql_help.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sql_parse.cc:
Adjusted assert reload_acl_and_cache() to the fact that global read
lock now takes full-blown metadata lock.
sql/sql_plist.h:
Added support for element counting to I_P_List list template.
One can use policy classes to specify if such counting is needed
or not needed for particular list.
sql/sql_show.cc:
Instead of using separate MDL_context for opening tables for I_S
purposes we now create MDL savepoint in the main context before
opening tables and rollback to this savepoint after closing them.
To support this and similar change for system tables methods of
THD responsible for saving/restoring open table state were changed
to use Open_tables_backup class which in addition to Open_table_state
has a member for this savepoint. As result code opening/closing tables
for I_S purposes was changed to use Open_tables_backup instead of
Open_table_state class as well.
sql/sql_table.cc:
mysql_rm_tables_part2():
Since now global intention exclusive metadata lock is ordinary
metadata lock we no longer can rely that by releasing MDL locks
on all tables we will release all locks acquired by this routine.
So in non-LOCK-TABLES mode we have to release all locks acquired
explicitly.
prepare_for_repair(), mysql_alter_table():
MDL subsystem no longer implicitly acquires global intention
exclusive metadata lock when per-object exclusive metadata lock
is acquired. Now this has to be done by explicit calls outside of
MDL subsystem.
sql/tztime.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
Also changed code not to use special mechanism for open system tables
when it is not really necessary.
2010-01-21 21:43:03 +01:00
rename table t0 to t3, t1 to t0, t3 to t1;
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
# --> connection con1
2009-12-30 18:53:30 +01:00
# Waiting for 'rename table ...' to get blocked...
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
# --> connection default
Patch that changes metadata locking subsystem to use mutex per lock and
condition variable per context instead of one mutex and one conditional
variable for the whole subsystem.
This should increase concurrency in this subsystem.
It also opens the way for further changes which are necessary to solve
such bugs as bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock"
and bug #37346 "innodb does not detect deadlock between update and alter
table".
Two other notable changes done by this patch:
- MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object metadata lock is acquired. Now this has
to be done by explicit calls outside of MDL subsystem.
- Instead of using separate MDL_context for opening system tables/tables
for purposes of I_S we now create MDL savepoint in the main context
before opening tables and rollback to this savepoint after closing
them. This means that it is now possible to get ER_LOCK_DEADLOCK error
even not inside a transaction. This might happen in unlikely case when
one runs DDL on one of system tables while also running DDL on some
other tables. Cases when this ER_LOCK_DEADLOCK error is not justified
will be addressed by advanced deadlock detector for MDL subsystem which
we plan to implement.
mysql-test/include/handler.inc:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/handler_innodb.result:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/handler_myisam.result:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/mdl_sync.result:
Adjusted mdl_sync.test to the fact that exclusive metadata locks on
tables are now acquired according to alphabetical order of fully
qualified table names instead of order in which tables are mentioned
in statement.
mysql-test/t/mdl_sync.test:
Adjusted mdl_sync.test to the fact that exclusive metadata locks on
tables are now acquired according to alphabetical order of fully
qualified table names instead of order in which tables are mentioned
in statement.
sql/events.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/ha_ndbcluster.cc:
Since manipulations with open table state no longer install proxy
MDL_context it does not make sense to perform them in order to
satisfy assert in mysql_rm_tables_part2(). Removed them per agreement
with Cluster team. This has not broken test suite since scenario in
which deadlock can occur and assertion fails is not covered by tests.
sql/lock.cc:
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object exclusive metadata lock is acquired.
Now this has to be done by explicit calls outside of MDL subsystem.
sql/log.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/mdl.cc:
Changed metadata locking subsystem to use mutex per lock and condition
variable per context instead of one mutex and one conditional variable
for the whole subsystem.
Changed approach to handling of global metadata locks. Instead of
implicitly acquiring intention exclusive locks when user requests
per-object upgradeable or exclusive locks now we require them to be
acquired explicitly in the same way as ordinary metadata locks.
In fact global lock are now ordinary metadata locks in new GLOBAL
namespace.
To implement these changes:
- Removed LOCK_mdl mutex and COND_mdl condition variable.
- Introduced MDL_lock::m_mutex mutexes which protect individual lock
objects.
- Replaced mdl_locks hash with MDL_map class, which has hash for
MDL_lock objects as a member and separate mutex which protects this
hash. Methods of this class allow to find(), find_or_create() or
remove() MDL_lock objects in concurrency-friendly fashion (i.e.
for most common operation, find_or_create(), we don't acquire
MDL_lock::m_mutex while holding MDL_map::m_mutex. Thanks to MikaelR
for this idea and benchmarks!). Added three auxiliary members to
MDL_lock class (m_is_destroyed, m_ref_usage, m_ref_release) to
support this concurrency-friendly behavior.
- Introduced MDL_context::m_ctx_wakeup_cond condition variable to be
used for waiting until this context's pending request can be
satisfied or its thread has to perform actions to resolve potential
deadlock. Context which want to wait add ticket corresponding to the
request to an appropriate queue of waiters in MDL_lock object so
they can be noticed when other contexts change state of lock and be
awaken by them by signalling on MDL_context::m_ctx_wakeup_cond.
As consequence MDL_ticket objects has to be used for any waiting
in metadata locking subsystem including one which happens in
MDL_context::wait_for_locks() method.
Another consequence is that MDL_context is no longer copyable and
can't be saved/restored when working with system tables.
- Made MDL_lock an abstract class, which delegates specifying exact
compatibility matrix to its descendants. Added MDL_global_lock child
class for global lock (The old is_lock_type_compatible() method
became can_grant_lock() method of this class). Added MDL_object_lock
class to represent per-object lock (The old MDL_lock::can_grant_lock()
became its method). Choice between two classes happens based on MDL
namespace in MDL_lock::create() method.
- Got rid of MDL_lock::type member as its meaning became ambigous for
global locks.
- To simplify waking up of contexts waiting for lock split waiting queue
in MDL_lock class in two queues. One for pending requests for exclusive
(including intention exclusive) locks and another for requests for
shared locks.
- Added virtual wake_up_waiters() method to MDL_lock, MDL_global_lock and
MDL_object_lock classes which allows to wake up waiting contexts after
state of lock changes. Replaced old duplicated code with calls to this
method.
- Adjusted MDL_context::try_acquire_shared_lock()/exclusive_lock()/
global_shared_lock(), MDL_ticket::upgrade_shared_lock_to_exclusive_lock()
and MDL_context::release_ticket() methods to use MDL_map and
MDL_lock::m_mutex instead of single LOCK_mdl mutex and wake up
waiters according to the approach described above. The latter method
also was renamed to MDL_context::release_lock().
- Changed MDL_context::try_acquire_shared_lock()/exclusive_lock() and
release_lock() not to handle global locks. They are now supposed to
be taken explicitly like ordinary metadata locks.
- Added helper MDL_context::try_acquire_global_intention_exclusive_lock()
and acquire_global_intention_exclusive_lock() methods.
- Moved common code from MDL_context::acquire_global_shared_lock() and
acquire_global_intention_exclusive_lock() to new method -
MDL_context::acquire_lock_impl().
- Moved common code from MDL_context::try_acquire_shared_lock(),
try_acquire_global_intention_exclusive_lock()/exclusive_lock()
to MDL_context::try_acquire_lock_impl().
- Since acquiring of several exclusive locks can no longer happen under
single LOCK_mdl mutex the approach to it had to be changed. Now we do
it in one by one fashion. This is done in alphabetical order to avoid
deadlocks. Changed MDL_context::acquire_exclusive_locks() accordingly
(as part of this change moved code responsible for acquiring single
exclusive lock to new MDL_context::acquire_exclusive_lock_impl()
method).
- Since we no longer have single LOCK_mdl mutex which protects all
MDL_context::m_is_waiting_in_mdl members using these members to
determine if we have really awaken context holding conflicting
shared lock became inconvinient. Got rid of this member and changed
notify_shared_lock() helper function and process of acquiring
of/upgrading to exclusive lock not to rely on such information.
Now in MDL_context::acquire_exclusive_lock_impl() and
MDL_ticket::upgrade_shared_lock_to_exclusive_lock() we simply
re-try to wake up threads holding conflicting shared locks after
small time out.
- Adjusted MDL_context::can_wait_lead_to_deadlock() and
MDL_ticket::has_pending_conflicting_lock() to use per-lock
mutexes instead of LOCK_mdl. To do this introduced
MDL_lock::has_pending_exclusive_lock() method.
sql/mdl.h:
Changed metadata locking subsystem to use mutex per lock and condition
variable per context instead of one mutex and one conditional variable
for the whole subsystem. In order to implement this change:
- Added MDL_key::cmp() method to be able to sort MDL_key objects
alphabetically. Changed length fields in MDL_key class to uint16
as 16-bit is enough for length of any key.
- Changed MDL_ticket::get_ctx() to return pointer to non-const
object in order to be able to use MDL_context::awake() method
for such contexts.
- Got rid of unlocked versions of can_wait_lead_to_deadlock()/
has_pending_conflicting_lock() methods in MDL_context and
MDL_ticket. We no longer has single mutex which protects all
locks. Thus one always has to use versions of these methods
which acquire per-lock mutexes.
- MDL_request_list type of list now counts its elements.
- Added MDL_context::m_ctx_wakeup_cond condition variable to be used
for waiting until this context's pending request can be satisfied
or its thread has to perform actions to resolve potential deadlock.
Added awake() method to wake up context from such wait.
Addition of condition variable made MDL_context uncopyable.
As result we no longer can save/restore MDL_context when working
with system tables. Instead we create MDL savepoint before opening
those tables and rollback to it once they are closed.
- MDL_context::release_ticket() became release_lock() method.
- Added auxiliary MDL_context::acquire_exclusive_lock_impl() method
which does all necessary work to acquire exclusive lock on one object
but should not be used directly as it does not enforce any asserts
ensuring that no deadlocks are possible.
- Since we no longer need to know if thread trying to acquire exclusive
lock managed to wake up any threads having conflicting shared locks
(as, anyway, we will try to wake up such threads again shortly)
- MDL_context::m_is_waiting_in_mdl member became unnecessary and
notify_shared_lock() no longer needs to be friend of MDL_context.
Changed approach to handling of global metadata locks. Instead of
implicitly acquiring intention exclusive locks when user requests
per-object upgradeable or exclusive locks now we require them to be
acquired explicitly in the same way as ordinary metadata locks.
- Added new GLOBAL namespace for such locks.
- Added new type of lock to be requested MDL_INTENTION_EXCLISIVE.
- Added MDL_context::try_acquire_global_intention_exclusive_lock()
and acquire_global_intention_exclusive_lock() methods.
- Moved common code from MDL_context::acquire_global_shared_lock()
and acquire_global_intention_exclusive_lock() to new method -
MDL_context::acquire_lock_impl().
- Moved common code from MDL_context::try_acquire_shared_lock(),
try_acquire_global_intention_exclusive_lock()/exclusive_lock()
to MDL_context::try_acquire_lock_impl().
- Added helper MDL_context::is_global_lock_owner() method to be
able easily to find what kind of global lock this context holds.
- MDL_context::m_has_global_shared_lock became unnecessary as
global read lock is now represented by ordinary ticket.
- Removed assert in MDL_context::set_lt_or_ha_sentinel() which became
false for cases when we execute LOCK TABLES under global read lock
mode.
sql/mysql_priv.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result calls
opening/closing system tables were changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sp.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sp.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sql_base.cc:
close_thread_tables():
Since we no longer use separate MDL_context for opening system
tables we need to avoid releasing all transaction locks when
closing system table. Releasing metadata lock on system table
is now responsibility of THD::restore_backup_open_tables_state().
open_table_get_mdl_lock(),
Open_table_context::recover_from_failed_open():
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object upgradable or exclusive metadata lock is
acquired. So this have to be done explicitly from these calls.
Changed Open_table_context class to store MDL_request object for
global intention exclusive lock acquired when opening tables.
open_table():
Do not release metadata lock if we have failed to open table as
this lock might have been acquired by one of previous statements
in transaction, and therefore should not be released.
open_system_tables_for_read()/close_system_tables()/
open_performance_schema_table():
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
close_performance_schema_table():
Got rid of duplicated code.
sql/sql_class.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. Also releasing
metadata lock on system table is now responsibility of
THD::restore_backup_open_tables_state().
Adjusted assert in THD::cleanup() to take into account fact that now
we also use MDL sentinel for global read lock.
sql/sql_class.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. As result:
- 'mdl_context' member was moved out of Open_tables_state to THD class.
enter_locked_tables_mode()/leave_locked_tables_mode() had to follow.
- Methods of THD responsible for saving/restoring open table state were
changed to use Open_tables_backup class which in addition to
Open_table_state has a member for this savepoint.
Changed Open_table_context class to store MDL_request object for
global intention exclusive lock acquired when opening tables.
sql/sql_delete.cc:
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object exclusive metadata lock is acquired.
Now this has to be done by explicit calls outside of MDL subsystem.
sql/sql_help.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sql_parse.cc:
Adjusted assert reload_acl_and_cache() to the fact that global read
lock now takes full-blown metadata lock.
sql/sql_plist.h:
Added support for element counting to I_P_List list template.
One can use policy classes to specify if such counting is needed
or not needed for particular list.
sql/sql_show.cc:
Instead of using separate MDL_context for opening tables for I_S
purposes we now create MDL savepoint in the main context before
opening tables and rollback to this savepoint after closing them.
To support this and similar change for system tables methods of
THD responsible for saving/restoring open table state were changed
to use Open_tables_backup class which in addition to Open_table_state
has a member for this savepoint. As result code opening/closing tables
for I_S purposes was changed to use Open_tables_backup instead of
Open_table_state class as well.
sql/sql_table.cc:
mysql_rm_tables_part2():
Since now global intention exclusive metadata lock is ordinary
metadata lock we no longer can rely that by releasing MDL locks
on all tables we will release all locks acquired by this routine.
So in non-LOCK-TABLES mode we have to release all locks acquired
explicitly.
prepare_for_repair(), mysql_alter_table():
MDL subsystem no longer implicitly acquires global intention
exclusive metadata lock when per-object exclusive metadata lock
is acquired. Now this has to be done by explicit calls outside of
MDL subsystem.
sql/tztime.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
Also changed code not to use special mechanism for open system tables
when it is not really necessary.
2010-01-21 21:43:03 +01:00
handler t0 open;
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
ERROR 40001: Deadlock found when trying to get lock; try restarting transaction
Patch that changes metadata locking subsystem to use mutex per lock and
condition variable per context instead of one mutex and one conditional
variable for the whole subsystem.
This should increase concurrency in this subsystem.
It also opens the way for further changes which are necessary to solve
such bugs as bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock"
and bug #37346 "innodb does not detect deadlock between update and alter
table".
Two other notable changes done by this patch:
- MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object metadata lock is acquired. Now this has
to be done by explicit calls outside of MDL subsystem.
- Instead of using separate MDL_context for opening system tables/tables
for purposes of I_S we now create MDL savepoint in the main context
before opening tables and rollback to this savepoint after closing
them. This means that it is now possible to get ER_LOCK_DEADLOCK error
even not inside a transaction. This might happen in unlikely case when
one runs DDL on one of system tables while also running DDL on some
other tables. Cases when this ER_LOCK_DEADLOCK error is not justified
will be addressed by advanced deadlock detector for MDL subsystem which
we plan to implement.
mysql-test/include/handler.inc:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/handler_innodb.result:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/handler_myisam.result:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/mdl_sync.result:
Adjusted mdl_sync.test to the fact that exclusive metadata locks on
tables are now acquired according to alphabetical order of fully
qualified table names instead of order in which tables are mentioned
in statement.
mysql-test/t/mdl_sync.test:
Adjusted mdl_sync.test to the fact that exclusive metadata locks on
tables are now acquired according to alphabetical order of fully
qualified table names instead of order in which tables are mentioned
in statement.
sql/events.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/ha_ndbcluster.cc:
Since manipulations with open table state no longer install proxy
MDL_context it does not make sense to perform them in order to
satisfy assert in mysql_rm_tables_part2(). Removed them per agreement
with Cluster team. This has not broken test suite since scenario in
which deadlock can occur and assertion fails is not covered by tests.
sql/lock.cc:
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object exclusive metadata lock is acquired.
Now this has to be done by explicit calls outside of MDL subsystem.
sql/log.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/mdl.cc:
Changed metadata locking subsystem to use mutex per lock and condition
variable per context instead of one mutex and one conditional variable
for the whole subsystem.
Changed approach to handling of global metadata locks. Instead of
implicitly acquiring intention exclusive locks when user requests
per-object upgradeable or exclusive locks now we require them to be
acquired explicitly in the same way as ordinary metadata locks.
In fact global lock are now ordinary metadata locks in new GLOBAL
namespace.
To implement these changes:
- Removed LOCK_mdl mutex and COND_mdl condition variable.
- Introduced MDL_lock::m_mutex mutexes which protect individual lock
objects.
- Replaced mdl_locks hash with MDL_map class, which has hash for
MDL_lock objects as a member and separate mutex which protects this
hash. Methods of this class allow to find(), find_or_create() or
remove() MDL_lock objects in concurrency-friendly fashion (i.e.
for most common operation, find_or_create(), we don't acquire
MDL_lock::m_mutex while holding MDL_map::m_mutex. Thanks to MikaelR
for this idea and benchmarks!). Added three auxiliary members to
MDL_lock class (m_is_destroyed, m_ref_usage, m_ref_release) to
support this concurrency-friendly behavior.
- Introduced MDL_context::m_ctx_wakeup_cond condition variable to be
used for waiting until this context's pending request can be
satisfied or its thread has to perform actions to resolve potential
deadlock. Context which want to wait add ticket corresponding to the
request to an appropriate queue of waiters in MDL_lock object so
they can be noticed when other contexts change state of lock and be
awaken by them by signalling on MDL_context::m_ctx_wakeup_cond.
As consequence MDL_ticket objects has to be used for any waiting
in metadata locking subsystem including one which happens in
MDL_context::wait_for_locks() method.
Another consequence is that MDL_context is no longer copyable and
can't be saved/restored when working with system tables.
- Made MDL_lock an abstract class, which delegates specifying exact
compatibility matrix to its descendants. Added MDL_global_lock child
class for global lock (The old is_lock_type_compatible() method
became can_grant_lock() method of this class). Added MDL_object_lock
class to represent per-object lock (The old MDL_lock::can_grant_lock()
became its method). Choice between two classes happens based on MDL
namespace in MDL_lock::create() method.
- Got rid of MDL_lock::type member as its meaning became ambigous for
global locks.
- To simplify waking up of contexts waiting for lock split waiting queue
in MDL_lock class in two queues. One for pending requests for exclusive
(including intention exclusive) locks and another for requests for
shared locks.
- Added virtual wake_up_waiters() method to MDL_lock, MDL_global_lock and
MDL_object_lock classes which allows to wake up waiting contexts after
state of lock changes. Replaced old duplicated code with calls to this
method.
- Adjusted MDL_context::try_acquire_shared_lock()/exclusive_lock()/
global_shared_lock(), MDL_ticket::upgrade_shared_lock_to_exclusive_lock()
and MDL_context::release_ticket() methods to use MDL_map and
MDL_lock::m_mutex instead of single LOCK_mdl mutex and wake up
waiters according to the approach described above. The latter method
also was renamed to MDL_context::release_lock().
- Changed MDL_context::try_acquire_shared_lock()/exclusive_lock() and
release_lock() not to handle global locks. They are now supposed to
be taken explicitly like ordinary metadata locks.
- Added helper MDL_context::try_acquire_global_intention_exclusive_lock()
and acquire_global_intention_exclusive_lock() methods.
- Moved common code from MDL_context::acquire_global_shared_lock() and
acquire_global_intention_exclusive_lock() to new method -
MDL_context::acquire_lock_impl().
- Moved common code from MDL_context::try_acquire_shared_lock(),
try_acquire_global_intention_exclusive_lock()/exclusive_lock()
to MDL_context::try_acquire_lock_impl().
- Since acquiring of several exclusive locks can no longer happen under
single LOCK_mdl mutex the approach to it had to be changed. Now we do
it in one by one fashion. This is done in alphabetical order to avoid
deadlocks. Changed MDL_context::acquire_exclusive_locks() accordingly
(as part of this change moved code responsible for acquiring single
exclusive lock to new MDL_context::acquire_exclusive_lock_impl()
method).
- Since we no longer have single LOCK_mdl mutex which protects all
MDL_context::m_is_waiting_in_mdl members using these members to
determine if we have really awaken context holding conflicting
shared lock became inconvinient. Got rid of this member and changed
notify_shared_lock() helper function and process of acquiring
of/upgrading to exclusive lock not to rely on such information.
Now in MDL_context::acquire_exclusive_lock_impl() and
MDL_ticket::upgrade_shared_lock_to_exclusive_lock() we simply
re-try to wake up threads holding conflicting shared locks after
small time out.
- Adjusted MDL_context::can_wait_lead_to_deadlock() and
MDL_ticket::has_pending_conflicting_lock() to use per-lock
mutexes instead of LOCK_mdl. To do this introduced
MDL_lock::has_pending_exclusive_lock() method.
sql/mdl.h:
Changed metadata locking subsystem to use mutex per lock and condition
variable per context instead of one mutex and one conditional variable
for the whole subsystem. In order to implement this change:
- Added MDL_key::cmp() method to be able to sort MDL_key objects
alphabetically. Changed length fields in MDL_key class to uint16
as 16-bit is enough for length of any key.
- Changed MDL_ticket::get_ctx() to return pointer to non-const
object in order to be able to use MDL_context::awake() method
for such contexts.
- Got rid of unlocked versions of can_wait_lead_to_deadlock()/
has_pending_conflicting_lock() methods in MDL_context and
MDL_ticket. We no longer has single mutex which protects all
locks. Thus one always has to use versions of these methods
which acquire per-lock mutexes.
- MDL_request_list type of list now counts its elements.
- Added MDL_context::m_ctx_wakeup_cond condition variable to be used
for waiting until this context's pending request can be satisfied
or its thread has to perform actions to resolve potential deadlock.
Added awake() method to wake up context from such wait.
Addition of condition variable made MDL_context uncopyable.
As result we no longer can save/restore MDL_context when working
with system tables. Instead we create MDL savepoint before opening
those tables and rollback to it once they are closed.
- MDL_context::release_ticket() became release_lock() method.
- Added auxiliary MDL_context::acquire_exclusive_lock_impl() method
which does all necessary work to acquire exclusive lock on one object
but should not be used directly as it does not enforce any asserts
ensuring that no deadlocks are possible.
- Since we no longer need to know if thread trying to acquire exclusive
lock managed to wake up any threads having conflicting shared locks
(as, anyway, we will try to wake up such threads again shortly)
- MDL_context::m_is_waiting_in_mdl member became unnecessary and
notify_shared_lock() no longer needs to be friend of MDL_context.
Changed approach to handling of global metadata locks. Instead of
implicitly acquiring intention exclusive locks when user requests
per-object upgradeable or exclusive locks now we require them to be
acquired explicitly in the same way as ordinary metadata locks.
- Added new GLOBAL namespace for such locks.
- Added new type of lock to be requested MDL_INTENTION_EXCLISIVE.
- Added MDL_context::try_acquire_global_intention_exclusive_lock()
and acquire_global_intention_exclusive_lock() methods.
- Moved common code from MDL_context::acquire_global_shared_lock()
and acquire_global_intention_exclusive_lock() to new method -
MDL_context::acquire_lock_impl().
- Moved common code from MDL_context::try_acquire_shared_lock(),
try_acquire_global_intention_exclusive_lock()/exclusive_lock()
to MDL_context::try_acquire_lock_impl().
- Added helper MDL_context::is_global_lock_owner() method to be
able easily to find what kind of global lock this context holds.
- MDL_context::m_has_global_shared_lock became unnecessary as
global read lock is now represented by ordinary ticket.
- Removed assert in MDL_context::set_lt_or_ha_sentinel() which became
false for cases when we execute LOCK TABLES under global read lock
mode.
sql/mysql_priv.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result calls
opening/closing system tables were changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sp.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sp.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sql_base.cc:
close_thread_tables():
Since we no longer use separate MDL_context for opening system
tables we need to avoid releasing all transaction locks when
closing system table. Releasing metadata lock on system table
is now responsibility of THD::restore_backup_open_tables_state().
open_table_get_mdl_lock(),
Open_table_context::recover_from_failed_open():
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object upgradable or exclusive metadata lock is
acquired. So this have to be done explicitly from these calls.
Changed Open_table_context class to store MDL_request object for
global intention exclusive lock acquired when opening tables.
open_table():
Do not release metadata lock if we have failed to open table as
this lock might have been acquired by one of previous statements
in transaction, and therefore should not be released.
open_system_tables_for_read()/close_system_tables()/
open_performance_schema_table():
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
close_performance_schema_table():
Got rid of duplicated code.
sql/sql_class.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. Also releasing
metadata lock on system table is now responsibility of
THD::restore_backup_open_tables_state().
Adjusted assert in THD::cleanup() to take into account fact that now
we also use MDL sentinel for global read lock.
sql/sql_class.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. As result:
- 'mdl_context' member was moved out of Open_tables_state to THD class.
enter_locked_tables_mode()/leave_locked_tables_mode() had to follow.
- Methods of THD responsible for saving/restoring open table state were
changed to use Open_tables_backup class which in addition to
Open_table_state has a member for this savepoint.
Changed Open_table_context class to store MDL_request object for
global intention exclusive lock acquired when opening tables.
sql/sql_delete.cc:
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object exclusive metadata lock is acquired.
Now this has to be done by explicit calls outside of MDL subsystem.
sql/sql_help.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sql_parse.cc:
Adjusted assert reload_acl_and_cache() to the fact that global read
lock now takes full-blown metadata lock.
sql/sql_plist.h:
Added support for element counting to I_P_List list template.
One can use policy classes to specify if such counting is needed
or not needed for particular list.
sql/sql_show.cc:
Instead of using separate MDL_context for opening tables for I_S
purposes we now create MDL savepoint in the main context before
opening tables and rollback to this savepoint after closing them.
To support this and similar change for system tables methods of
THD responsible for saving/restoring open table state were changed
to use Open_tables_backup class which in addition to Open_table_state
has a member for this savepoint. As result code opening/closing tables
for I_S purposes was changed to use Open_tables_backup instead of
Open_table_state class as well.
sql/sql_table.cc:
mysql_rm_tables_part2():
Since now global intention exclusive metadata lock is ordinary
metadata lock we no longer can rely that by releasing MDL locks
on all tables we will release all locks acquired by this routine.
So in non-LOCK-TABLES mode we have to release all locks acquired
explicitly.
prepare_for_repair(), mysql_alter_table():
MDL subsystem no longer implicitly acquires global intention
exclusive metadata lock when per-object exclusive metadata lock
is acquired. Now this has to be done by explicit calls outside of
MDL subsystem.
sql/tztime.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
Also changed code not to use special mechanism for open system tables
when it is not really necessary.
2010-01-21 21:43:03 +01:00
select * from t0;
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
ERROR 40001: Deadlock found when trying to get lock; try restarting transaction
handler t1 open;
commit;
handler t1 close;
# --> connection con2
2009-12-30 18:53:30 +01:00
# Reaping 'rename table ...'...
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
# --> connection default
handler t1 open;
handler t1 read a prev;
a
5
handler t1 close;
Patch that changes metadata locking subsystem to use mutex per lock and
condition variable per context instead of one mutex and one conditional
variable for the whole subsystem.
This should increase concurrency in this subsystem.
It also opens the way for further changes which are necessary to solve
such bugs as bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock"
and bug #37346 "innodb does not detect deadlock between update and alter
table".
Two other notable changes done by this patch:
- MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object metadata lock is acquired. Now this has
to be done by explicit calls outside of MDL subsystem.
- Instead of using separate MDL_context for opening system tables/tables
for purposes of I_S we now create MDL savepoint in the main context
before opening tables and rollback to this savepoint after closing
them. This means that it is now possible to get ER_LOCK_DEADLOCK error
even not inside a transaction. This might happen in unlikely case when
one runs DDL on one of system tables while also running DDL on some
other tables. Cases when this ER_LOCK_DEADLOCK error is not justified
will be addressed by advanced deadlock detector for MDL subsystem which
we plan to implement.
mysql-test/include/handler.inc:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/handler_innodb.result:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/handler_myisam.result:
Adjusted handler_myisam.test and handler_innodb.test to the fact that
exclusive metadata locks on tables are now acquired according to
alphabetical order of fully qualified table names instead of order
in which tables are mentioned in statement.
mysql-test/r/mdl_sync.result:
Adjusted mdl_sync.test to the fact that exclusive metadata locks on
tables are now acquired according to alphabetical order of fully
qualified table names instead of order in which tables are mentioned
in statement.
mysql-test/t/mdl_sync.test:
Adjusted mdl_sync.test to the fact that exclusive metadata locks on
tables are now acquired according to alphabetical order of fully
qualified table names instead of order in which tables are mentioned
in statement.
sql/events.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/ha_ndbcluster.cc:
Since manipulations with open table state no longer install proxy
MDL_context it does not make sense to perform them in order to
satisfy assert in mysql_rm_tables_part2(). Removed them per agreement
with Cluster team. This has not broken test suite since scenario in
which deadlock can occur and assertion fails is not covered by tests.
sql/lock.cc:
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object exclusive metadata lock is acquired.
Now this has to be done by explicit calls outside of MDL subsystem.
sql/log.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/mdl.cc:
Changed metadata locking subsystem to use mutex per lock and condition
variable per context instead of one mutex and one conditional variable
for the whole subsystem.
Changed approach to handling of global metadata locks. Instead of
implicitly acquiring intention exclusive locks when user requests
per-object upgradeable or exclusive locks now we require them to be
acquired explicitly in the same way as ordinary metadata locks.
In fact global lock are now ordinary metadata locks in new GLOBAL
namespace.
To implement these changes:
- Removed LOCK_mdl mutex and COND_mdl condition variable.
- Introduced MDL_lock::m_mutex mutexes which protect individual lock
objects.
- Replaced mdl_locks hash with MDL_map class, which has hash for
MDL_lock objects as a member and separate mutex which protects this
hash. Methods of this class allow to find(), find_or_create() or
remove() MDL_lock objects in concurrency-friendly fashion (i.e.
for most common operation, find_or_create(), we don't acquire
MDL_lock::m_mutex while holding MDL_map::m_mutex. Thanks to MikaelR
for this idea and benchmarks!). Added three auxiliary members to
MDL_lock class (m_is_destroyed, m_ref_usage, m_ref_release) to
support this concurrency-friendly behavior.
- Introduced MDL_context::m_ctx_wakeup_cond condition variable to be
used for waiting until this context's pending request can be
satisfied or its thread has to perform actions to resolve potential
deadlock. Context which want to wait add ticket corresponding to the
request to an appropriate queue of waiters in MDL_lock object so
they can be noticed when other contexts change state of lock and be
awaken by them by signalling on MDL_context::m_ctx_wakeup_cond.
As consequence MDL_ticket objects has to be used for any waiting
in metadata locking subsystem including one which happens in
MDL_context::wait_for_locks() method.
Another consequence is that MDL_context is no longer copyable and
can't be saved/restored when working with system tables.
- Made MDL_lock an abstract class, which delegates specifying exact
compatibility matrix to its descendants. Added MDL_global_lock child
class for global lock (The old is_lock_type_compatible() method
became can_grant_lock() method of this class). Added MDL_object_lock
class to represent per-object lock (The old MDL_lock::can_grant_lock()
became its method). Choice between two classes happens based on MDL
namespace in MDL_lock::create() method.
- Got rid of MDL_lock::type member as its meaning became ambigous for
global locks.
- To simplify waking up of contexts waiting for lock split waiting queue
in MDL_lock class in two queues. One for pending requests for exclusive
(including intention exclusive) locks and another for requests for
shared locks.
- Added virtual wake_up_waiters() method to MDL_lock, MDL_global_lock and
MDL_object_lock classes which allows to wake up waiting contexts after
state of lock changes. Replaced old duplicated code with calls to this
method.
- Adjusted MDL_context::try_acquire_shared_lock()/exclusive_lock()/
global_shared_lock(), MDL_ticket::upgrade_shared_lock_to_exclusive_lock()
and MDL_context::release_ticket() methods to use MDL_map and
MDL_lock::m_mutex instead of single LOCK_mdl mutex and wake up
waiters according to the approach described above. The latter method
also was renamed to MDL_context::release_lock().
- Changed MDL_context::try_acquire_shared_lock()/exclusive_lock() and
release_lock() not to handle global locks. They are now supposed to
be taken explicitly like ordinary metadata locks.
- Added helper MDL_context::try_acquire_global_intention_exclusive_lock()
and acquire_global_intention_exclusive_lock() methods.
- Moved common code from MDL_context::acquire_global_shared_lock() and
acquire_global_intention_exclusive_lock() to new method -
MDL_context::acquire_lock_impl().
- Moved common code from MDL_context::try_acquire_shared_lock(),
try_acquire_global_intention_exclusive_lock()/exclusive_lock()
to MDL_context::try_acquire_lock_impl().
- Since acquiring of several exclusive locks can no longer happen under
single LOCK_mdl mutex the approach to it had to be changed. Now we do
it in one by one fashion. This is done in alphabetical order to avoid
deadlocks. Changed MDL_context::acquire_exclusive_locks() accordingly
(as part of this change moved code responsible for acquiring single
exclusive lock to new MDL_context::acquire_exclusive_lock_impl()
method).
- Since we no longer have single LOCK_mdl mutex which protects all
MDL_context::m_is_waiting_in_mdl members using these members to
determine if we have really awaken context holding conflicting
shared lock became inconvinient. Got rid of this member and changed
notify_shared_lock() helper function and process of acquiring
of/upgrading to exclusive lock not to rely on such information.
Now in MDL_context::acquire_exclusive_lock_impl() and
MDL_ticket::upgrade_shared_lock_to_exclusive_lock() we simply
re-try to wake up threads holding conflicting shared locks after
small time out.
- Adjusted MDL_context::can_wait_lead_to_deadlock() and
MDL_ticket::has_pending_conflicting_lock() to use per-lock
mutexes instead of LOCK_mdl. To do this introduced
MDL_lock::has_pending_exclusive_lock() method.
sql/mdl.h:
Changed metadata locking subsystem to use mutex per lock and condition
variable per context instead of one mutex and one conditional variable
for the whole subsystem. In order to implement this change:
- Added MDL_key::cmp() method to be able to sort MDL_key objects
alphabetically. Changed length fields in MDL_key class to uint16
as 16-bit is enough for length of any key.
- Changed MDL_ticket::get_ctx() to return pointer to non-const
object in order to be able to use MDL_context::awake() method
for such contexts.
- Got rid of unlocked versions of can_wait_lead_to_deadlock()/
has_pending_conflicting_lock() methods in MDL_context and
MDL_ticket. We no longer has single mutex which protects all
locks. Thus one always has to use versions of these methods
which acquire per-lock mutexes.
- MDL_request_list type of list now counts its elements.
- Added MDL_context::m_ctx_wakeup_cond condition variable to be used
for waiting until this context's pending request can be satisfied
or its thread has to perform actions to resolve potential deadlock.
Added awake() method to wake up context from such wait.
Addition of condition variable made MDL_context uncopyable.
As result we no longer can save/restore MDL_context when working
with system tables. Instead we create MDL savepoint before opening
those tables and rollback to it once they are closed.
- MDL_context::release_ticket() became release_lock() method.
- Added auxiliary MDL_context::acquire_exclusive_lock_impl() method
which does all necessary work to acquire exclusive lock on one object
but should not be used directly as it does not enforce any asserts
ensuring that no deadlocks are possible.
- Since we no longer need to know if thread trying to acquire exclusive
lock managed to wake up any threads having conflicting shared locks
(as, anyway, we will try to wake up such threads again shortly)
- MDL_context::m_is_waiting_in_mdl member became unnecessary and
notify_shared_lock() no longer needs to be friend of MDL_context.
Changed approach to handling of global metadata locks. Instead of
implicitly acquiring intention exclusive locks when user requests
per-object upgradeable or exclusive locks now we require them to be
acquired explicitly in the same way as ordinary metadata locks.
- Added new GLOBAL namespace for such locks.
- Added new type of lock to be requested MDL_INTENTION_EXCLISIVE.
- Added MDL_context::try_acquire_global_intention_exclusive_lock()
and acquire_global_intention_exclusive_lock() methods.
- Moved common code from MDL_context::acquire_global_shared_lock()
and acquire_global_intention_exclusive_lock() to new method -
MDL_context::acquire_lock_impl().
- Moved common code from MDL_context::try_acquire_shared_lock(),
try_acquire_global_intention_exclusive_lock()/exclusive_lock()
to MDL_context::try_acquire_lock_impl().
- Added helper MDL_context::is_global_lock_owner() method to be
able easily to find what kind of global lock this context holds.
- MDL_context::m_has_global_shared_lock became unnecessary as
global read lock is now represented by ordinary ticket.
- Removed assert in MDL_context::set_lt_or_ha_sentinel() which became
false for cases when we execute LOCK TABLES under global read lock
mode.
sql/mysql_priv.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result calls
opening/closing system tables were changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sp.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sp.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sql_base.cc:
close_thread_tables():
Since we no longer use separate MDL_context for opening system
tables we need to avoid releasing all transaction locks when
closing system table. Releasing metadata lock on system table
is now responsibility of THD::restore_backup_open_tables_state().
open_table_get_mdl_lock(),
Open_table_context::recover_from_failed_open():
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object upgradable or exclusive metadata lock is
acquired. So this have to be done explicitly from these calls.
Changed Open_table_context class to store MDL_request object for
global intention exclusive lock acquired when opening tables.
open_table():
Do not release metadata lock if we have failed to open table as
this lock might have been acquired by one of previous statements
in transaction, and therefore should not be released.
open_system_tables_for_read()/close_system_tables()/
open_performance_schema_table():
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
close_performance_schema_table():
Got rid of duplicated code.
sql/sql_class.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. Also releasing
metadata lock on system table is now responsibility of
THD::restore_backup_open_tables_state().
Adjusted assert in THD::cleanup() to take into account fact that now
we also use MDL sentinel for global read lock.
sql/sql_class.h:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. As result:
- 'mdl_context' member was moved out of Open_tables_state to THD class.
enter_locked_tables_mode()/leave_locked_tables_mode() had to follow.
- Methods of THD responsible for saving/restoring open table state were
changed to use Open_tables_backup class which in addition to
Open_table_state has a member for this savepoint.
Changed Open_table_context class to store MDL_request object for
global intention exclusive lock acquired when opening tables.
sql/sql_delete.cc:
MDL subsystem no longer implicitly acquires global intention exclusive
metadata lock when per-object exclusive metadata lock is acquired.
Now this has to be done by explicit calls outside of MDL subsystem.
sql/sql_help.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
sql/sql_parse.cc:
Adjusted assert reload_acl_and_cache() to the fact that global read
lock now takes full-blown metadata lock.
sql/sql_plist.h:
Added support for element counting to I_P_List list template.
One can use policy classes to specify if such counting is needed
or not needed for particular list.
sql/sql_show.cc:
Instead of using separate MDL_context for opening tables for I_S
purposes we now create MDL savepoint in the main context before
opening tables and rollback to this savepoint after closing them.
To support this and similar change for system tables methods of
THD responsible for saving/restoring open table state were changed
to use Open_tables_backup class which in addition to Open_table_state
has a member for this savepoint. As result code opening/closing tables
for I_S purposes was changed to use Open_tables_backup instead of
Open_table_state class as well.
sql/sql_table.cc:
mysql_rm_tables_part2():
Since now global intention exclusive metadata lock is ordinary
metadata lock we no longer can rely that by releasing MDL locks
on all tables we will release all locks acquired by this routine.
So in non-LOCK-TABLES mode we have to release all locks acquired
explicitly.
prepare_for_repair(), mysql_alter_table():
MDL subsystem no longer implicitly acquires global intention
exclusive metadata lock when per-object exclusive metadata lock
is acquired. Now this has to be done by explicit calls outside of
MDL subsystem.
sql/tztime.cc:
Instead of using separate MDL_context for opening system tables we now
create MDL savepoint in the main context before opening such tables
and rollback to this savepoint after closing them. To support this
change methods of THD responsible for saving/restoring open table
state were changed to use Open_tables_backup class which in addition
to Open_table_state has a member for this savepoint. As result code
opening/closing system tables was changed to use Open_tables_backup
instead of Open_table_state class as well.
Also changed code not to use special mechanism for open system tables
when it is not really necessary.
2010-01-21 21:43:03 +01:00
drop table t0;
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
#
2009-12-30 18:53:30 +01:00
# Originally there was a deadlock error in this test.
# With implementation of deadlock detector
# we no longer deadlock, but block and wait on a lock.
# The HANDLER is auto-closed as soon as the connection
# sees a pending conflicting lock against it.
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
#
create table t2 (a int, key a (a));
handler t1 open;
# --> connection con1
lock tables t2 read;
# --> connection con2
# Sending 'drop table t2'...
drop table t2;
# --> connection con1
# Waiting for 'drop table t2' to get blocked...
# --> connection default
2009-12-30 18:53:30 +01:00
# Sending 'select * from t2'
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
select * from t2;
# --> connection con1
2009-12-30 18:53:30 +01:00
# Waiting for 'select * from t2' to get blocked...
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
unlock tables;
# --> connection con2
# Reaping 'drop table t2'...
# --> connection default
2009-12-30 18:53:30 +01:00
# Reaping 'select * from t2'
ERROR 42S02: Table 'test.t2' doesn't exist
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
handler t1 close;
#
# ROLLBACK TO SAVEPOINT releases transactional locks,
# but has no effect on open HANDLERs
#
create table t2 like t1;
create table t3 like t1;
begin;
# Have something before the savepoint
select * from t3;
a
savepoint sv;
handler t1 open;
handler t1 read a first;
a
1
handler t1 read a next;
a
2
select * from t2;
a
# --> connection con1
# Sending:
drop table t1;
# --> connection con2
# Sending:
drop table t2;
# --> connection default
# Let DROP TABLE statements sync in. We must use
# a separate connection for that, because otherwise SELECT
# will auto-close the HANDLERs, becaues there are pending
# exclusive locks against them.
# --> connection con3
# Waiting for 'drop table t1' to get blocked...
# Waiting for 'drop table t2' to get blocked...
# Demonstrate that t2 lock was released and t2 was dropped
# after ROLLBACK TO SAVEPOINT
# --> connection default
rollback to savepoint sv;
# --> connection con2
# Reaping 'drop table t2'...
# Demonstrate that ROLLBACK TO SAVEPOINT didn't release the handler
# lock.
# --> connection default
handler t1 read a next;
a
3
handler t1 read a next;
a
4
# Demonstrate that the drop will go through as soon as we close the
# HANDLER
handler t1 close;
# connection con1
# Reaping 'drop table t1'...
# --> connection default
commit;
drop table t3;
#
# A few special cases when using SAVEPOINT/ROLLBACK TO
# SAVEPOINT and HANDLER.
#
# Show that rollback to the savepoint taken in the beginning
# of the transaction doesn't release mdl lock on
# the HANDLER that was opened later.
#
create table t1 (a int, key a(a));
insert into t1 (a) values (1), (2), (3), (4), (5);
create table t2 like t1;
begin;
savepoint sv;
handler t1 open;
handler t1 read a first;
a
1
handler t1 read a next;
a
2
select * from t2;
a
# --> connection con1
# Sending:
drop table t1;
# --> connection con2
# Sending:
drop table t2;
# --> connection default
# Let DROP TABLE statements sync in. We must use
# a separate connection for that, because otherwise SELECT
# will auto-close the HANDLERs, becaues there are pending
# exclusive locks against them.
# --> connection con3
# Waiting for 'drop table t1' to get blocked...
# Waiting for 'drop table t2' to get blocked...
# Demonstrate that t2 lock was released and t2 was dropped
# after ROLLBACK TO SAVEPOINT
# --> connection default
rollback to savepoint sv;
# --> connection con2
# Reaping 'drop table t2'...
# Demonstrate that ROLLBACK TO SAVEPOINT didn't release the handler
# lock.
# --> connection default
handler t1 read a next;
a
3
handler t1 read a next;
a
4
# Demonstrate that the drop will go through as soon as we close the
# HANDLER
handler t1 close;
# connection con1
# Reaping 'drop table t1'...
# --> connection default
commit;
#
# Show that rollback to the savepoint taken in the beginning
# of the transaction works properly (no valgrind warnins, etc),
# even though it's done after the HANDLER mdl lock that was there
# at the beginning is released and added again.
#
create table t1 (a int, key a(a));
insert into t1 (a) values (1), (2), (3), (4), (5);
create table t2 like t1;
create table t3 like t1;
insert into t3 (a) select a from t1;
begin;
handler t1 open;
savepoint sv;
handler t1 read a first;
a
1
select * from t2;
a
handler t1 close;
handler t3 open;
handler t3 read a first;
a
1
rollback to savepoint sv;
# --> connection con1
drop table t1, t2;
# Sending:
drop table t3;
# Let DROP TABLE statement sync in.
# --> connection con2
# Waiting for 'drop table t3' to get blocked...
# Demonstrate that ROLLBACK TO SAVEPOINT didn't release the handler
# lock.
# --> connection default
handler t3 read a next;
a
2
# Demonstrate that the drop will go through as soon as we close the
# HANDLER
handler t3 close;
# connection con1
# Reaping 'drop table t3'...
# --> connection default
commit;
#
# If we have to wait on an exclusive locks while having
# an open HANDLER, ER_LOCK_DEADLOCK is reported.
#
create table t1 (a int, key a(a));
create table t2 like t1;
handler t1 open;
# --> connection con1
Implement new type-of-operation-aware metadata locks.
Add a wait-for graph based deadlock detector to the
MDL subsystem.
Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and
bug #37346 "innodb does not detect deadlock between update and
alter table".
The first bug manifested itself as an unwarranted abort of a
transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER
statement, when this transaction tried to repeat use of a
table, which it has already used in a similar fashion before
ALTER started.
The second bug showed up as a deadlock between table-level
locks and InnoDB row locks, which was "detected" only after
innodb_lock_wait_timeout timeout.
A transaction would start using the table and modify a few
rows.
Then ALTER TABLE would come in, and start copying rows
into a temporary table. Eventually it would stumble on
the modified records and get blocked on a row lock.
The first transaction would try to do more updates, and get
blocked on thr_lock.c lock.
This situation of circular wait would only get resolved
by a timeout.
Both these bugs stemmed from inadequate solutions to the
problem of deadlocks occurring between different
locking subsystems.
In the first case we tried to avoid deadlocks between metadata
locking and table-level locking subsystems, when upgrading shared
metadata lock to exclusive one.
Transactions holding the shared lock on the table and waiting for
some table-level lock used to be aborted too aggressively.
We also allowed ALTER TABLE to start in presence of transactions
that modify the subject table. ALTER TABLE acquires
TL_WRITE_ALLOW_READ lock at start, and that block all writes
against the table (naturally, we don't want any writes to be lost
when switching the old and the new table). TL_WRITE_ALLOW_READ
lock, in turn, would block the started transaction on thr_lock.c
lock, should they do more updates. This, again, lead to the need
to abort such transactions.
The second bug occurred simply because we didn't have any
mechanism to detect deadlocks between the table-level locks
in thr_lock.c and row-level locks in InnoDB, other than
innodb_lock_wait_timeout.
This patch solves both these problems by moving lock conflicts
which are causing these deadlocks into the metadata locking
subsystem, thus making it possible to avoid or detect such
deadlocks inside MDL.
To do this we introduce new type-of-operation-aware metadata
locks, which allow MDL subsystem to know not only the fact that
transaction has used or is going to use some object but also what
kind of operation it has carried out or going to carry out on the
object.
This, along with the addition of a special kind of upgradable
metadata lock, allows ALTER TABLE to wait until all
transactions which has updated the table to go away.
This solves the second issue.
Another special type of upgradable metadata lock is acquired
by LOCK TABLE WRITE. This second lock type allows to solve the
first issue, since abortion of table-level locks in event of
DDL under LOCK TABLES becomes also unnecessary.
Below follows the list of incompatible changes introduced by
this patch:
- From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those
statements that acquire TL_WRITE_ALLOW_READ lock)
wait for all transactions which has *updated* the table to
complete.
- From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE
(i.e. all statements which acquire TL_WRITE table-level lock) wait
for all transaction which *updated or read* from the table
to complete.
As a consequence, innodb_table_locks=0 option no longer applies
to LOCK TABLES ... WRITE.
- DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort
statements or transactions which use tables being dropped or
renamed, and instead wait for these transactions to complete.
- Since LOCK TABLES WRITE now takes a special metadata lock,
not compatible with with reads or writes against the subject table
and transaction-wide, thr_lock.c deadlock avoidance algorithm
that used to ensure absence of deadlocks between LOCK TABLES
WRITE and other statements is no longer sufficient, even for
MyISAM. The wait-for graph based deadlock detector of MDL
subsystem may sometimes be necessary and is involved. This may
lead to ER_LOCK_DEADLOCK error produced for multi-statement
transactions even if these only use MyISAM:
session 1: session 2:
begin;
update t1 ... lock table t2 write, t1 write;
-- gets a lock on t2, blocks on t1
update t2 ...
(ER_LOCK_DEADLOCK)
- Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE
was abandoned.
LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same
priority as the usual LOCK TABLE ... WRITE.
SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in
the wait queue.
- We do not take upgradable metadata locks on implicitly
locked tables. So if one has, say, a view v1 that uses
table t1, and issues:
LOCK TABLE v1 WRITE;
FLUSH TABLE t1; -- (or just 'FLUSH TABLES'),
an error is produced.
In order to be able to perform DDL on a table under LOCK TABLES,
the table must be locked explicitly in the LOCK TABLES list.
mysql-test/include/handler.inc:
Adjusted test case to trigger an execution path on which bug 41110
"crash with handler command when used concurrently with alter
table" and bug 41112 "crash in mysql_ha_close_table/get_lock_data
with alter table" were originally discovered. Left old test case
which no longer triggers this execution path for the sake of
coverage.
Added test coverage for HANDLER SQL statements and type-aware
metadata locks.
Added a test for the global shared lock and HANDLER SQL.
Updated tests to take into account that the old simple deadlock
detection heuristics was replaced with a graph-based deadlock
detector.
mysql-test/r/debug_sync.result:
Updated results (see debug_sync.test).
mysql-test/r/handler_innodb.result:
Updated results (see handler.inc test).
mysql-test/r/handler_myisam.result:
Updated results (see handler.inc test).
mysql-test/r/innodb-lock.result:
Updated results (see innodb-lock.test).
mysql-test/r/innodb_mysql_lock.result:
Updated results (see innodb_mysql_lock.test).
mysql-test/r/lock.result:
Updated results (see lock.test).
mysql-test/r/lock_multi.result:
Updated results (see lock_multi.test).
mysql-test/r/lock_sync.result:
Updated results (see lock_sync.test).
mysql-test/r/mdl_sync.result:
Updated results (see mdl_sync.test).
mysql-test/r/sp-threads.result:
SHOW PROCESSLIST output has changed due to the fact that waiting
for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/r/truncate_coverage.result:
Updated results (see truncate_coverage.test).
mysql-test/suite/funcs_1/datadict/processlist_val.inc:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/funcs_1/r/processlist_val_no_prot.result:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/rpl/t/rpl_sp.test:
Updated to a new SHOW PROCESSLIST state name.
mysql-test/t/debug_sync.test:
Use LOCK TABLES READ instead of LOCK TABLES WRITE as the latter
no longer allows to trigger execution path involving waiting on
thr_lock.c lock and therefore reaching debug sync-point covered
by this test.
mysql-test/t/innodb-lock.test:
Adjusted test case to the fact that innodb_table_locks=0 option is
no longer supported, since LOCK TABLES WRITE handles all its
conflicts within MDL subsystem.
mysql-test/t/innodb_mysql_lock.test:
Added test for bug #37346 "innodb does not detect deadlock between
update and alter table".
mysql-test/t/lock.test:
Added test coverage which checks the fact that we no longer support
DDL under LOCK TABLES on tables which were locked implicitly.
Adjusted existing test cases accordingly.
mysql-test/t/lock_multi.test:
Added test for bug #46272 "MySQL 5.4.4, new MDL: unnecessary
deadlock". Adjusted other test cases to take into account the
fact that waiting for LOCK TABLES ... WRITE now happens within MDL
subsystem.
mysql-test/t/lock_sync.test:
Since LOCK TABLES ... WRITE now takes SNRW metadata lock for
tables locked explicitly we have to implicitly lock InnoDB tables
(through view) to trigger the table-level lock conflict between
TL_WRITE and TL_WRITE_ALLOW_WRITE.
mysql-test/t/mdl_sync.test:
Added basic test coverage for type-of-operation-aware metadata
locks. Also covered with tests some use cases involving HANDLER
statements in which a deadlock could arise.
Adjusted existing tests to take type-of-operation-aware MDL into
account.
mysql-test/t/multi_update.test:
Update to a new SHOW PROCESSLIST state name.
mysql-test/t/truncate_coverage.test:
Adjusted test case after making LOCK TABLES WRITE to wait until
transactions that use the table to be locked are completed.
Updated to the changed name of DEBUG_SYNC point.
sql/handler.cc:
Global read lock functionality has been
moved into a class.
sql/lock.cc:
Global read lock functionality has been
moved into a class.
Updated code to use the new MDL API.
sql/mdl.cc:
Introduced new type-of-operation aware metadata locks.
To do this:
- Changed MDL_lock to use one list for waiting requests and one
list for granted requests. For each list, added a bitmap
that holds information what lock types a list contains.
Added a helper class MDL_lock::List to manipulate with granted
and waited lists while keeping the bitmaps in sync
with list contents.
- Changed lock-compatibility functions to use bitmaps that
define compatibility.
- Introduced a graph based deadlock detector inspired by
waiting_threads.c from Maria implementation.
- Now that we have a deadlock detector, and no longer have
a global lock to protect individual lock objects, but rather
use an rw lock per object, removed redundant code for upgrade,
and the global read lock. Changed the MDL API to
no longer require the caller to acquire the global
intention exclusive lock by means of a separate method.
Removed a few more methods that became redundant.
- Removed deadlock detection heuristic, it has been made
obsolete by the deadlock detector.
- With operation-type-aware metadata locks, MDL subsystem has
become aware of potential conflicts between DDL and open
transactions. This made it possible to remove calls to
mysql_abort_transactions_with_shared_lock() from acquisition
paths for exclusive lock and lock upgrade. Now we can simply
wait for these transactions to complete without fear of
deadlock. Function mysql_lock_abort() has also become
unnecessary for all conflicting cases except when a DDL
conflicts with a connection that has an open HANDLER.
sql/mdl.h:
Introduced new type-of-operation aware metadata locks.
Introduced a graph based deadlock detector and supporting
methods.
Added comments.
God rid of redundant API calls.
Renamed m_lt_or_ha_sentinel to m_trans_sentinel,
since now it guards the global read lock as well as
LOCK TABLES and HANDLER locks.
sql/mysql_priv.h:
Moved the global read lock functionality into a
class.
Added MYSQL_OPEN_FORCE_SHARED_MDL flag which forces
open_tables() to take MDL_SHARED on tables instead of
metadata locks specified in the parser. We use this to
allow PREPARE run concurrently in presence of
LOCK TABLES ... WRITE.
Added signature for find_table_for_mdl_ugprade().
sql/set_var.cc:
Global read lock functionality has been
moved into a class.
sql/sp_head.cc:
When creating TABLE_LIST elements for prelocking or
system tables set the type of request for metadata
lock according to the operation that will be performed
on the table.
sql/sql_base.cc:
- Updated code to use the new MDL API.
- In order to avoid locks starvation we take upgradable
locks all at once. As result implicitly locked tables no
longer get an upgradable lock. Consequently DDL and FLUSH
TABLES for such tables is prohibited.
find_write_locked_table() was replaced by
find_table_for_mdl_upgrade() function.
open_table() was adjusted to return TABLE instance with
upgradable ticket when necessary.
- We no longer wait for all locks on OT_WAIT back off
action -- only on the lock that caused the wait
conflict. Moreover, now we distinguish cases when we
have to wait due to conflict in MDL and old version
of table in TDC.
- Upate mysql_notify_threads_having_share_locks()
to only abort thr_lock.c waits of threads that
have open HANDLERs, since lock conflicts with only
these threads now can lead to deadlocks not detectable
by the MDL deadlock detector.
- Remove mysql_abort_transactions_with_shared_locks()
which is no longer needed.
sql/sql_class.cc:
Global read lock functionality has been moved into a class.
Re-arranged code in THD::cleanup() to simplify assert.
sql/sql_class.h:
Introduced class to incapsulate global read lock
functionality.
Now sentinel in MDL subsystem guards the global read lock
as well as LOCK TABLES and HANDLER locks. Adjusted code
accordingly.
sql/sql_db.cc:
Global read lock functionality has been moved into a class.
sql/sql_delete.cc:
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
TRUNCATE TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_handler.cc:
Inform MDL_context about presence of open HANDLERs.
Since HANLDERs break MDL protocol by acquiring table-level
lock while holding only S metadata lock on a table MDL
subsystem should take special care about such contexts (Now
this is the only case when mysql_lock_abort() is used).
sql/sql_parse.cc:
Global read lock functionality has been moved into a class.
Do not take upgradable metadata locks when opening tables
for CREATE TABLE SELECT as it is not necessary and limits
concurrency.
When initializing TABLE_LIST objects before adding them
to the table list set the type of request for metadata lock
according to the operation that will be performed on the
table.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result FLUSH
TABLES is no longer allowed for such tables.
sql/sql_prepare.cc:
Use MYSQL_OPEN_FORCE_SHARED_MDL flag when opening
tables during PREPARE. This allows PREPARE to run
concurrently in presence of LOCK TABLES ... WRITE.
sql/sql_rename.cc:
Global read lock functionality has been moved into a class.
sql/sql_show.cc:
Updated code to use the new MDL API.
sql/sql_table.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result DROP
TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_trigger.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
CREATE/DROP TRIGGER is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_view.cc:
Global read lock functionality has been moved into a class.
Fixed results of wrong merge that led to misuse of GLR API.
CREATE VIEW statement is not a commit statement.
sql/table.cc:
When resetting TABLE_LIST objects for PS or SP re-execution
set the type of request for metadata lock according to the
operation that will be performed on the table. Do the same
in auxiliary function initializing metadata lock requests
in a table list.
sql/table.h:
When initializing TABLE_LIST objects set the type of request
for metadata lock according to the operation that will be
performed on the table.
sql/transaction.cc:
Global read lock functionality has been moved into a class.
2010-02-01 12:43:06 +01:00
lock table t1 write, t2 write;
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
# --> connection default
drop table t2;
Implement new type-of-operation-aware metadata locks.
Add a wait-for graph based deadlock detector to the
MDL subsystem.
Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and
bug #37346 "innodb does not detect deadlock between update and
alter table".
The first bug manifested itself as an unwarranted abort of a
transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER
statement, when this transaction tried to repeat use of a
table, which it has already used in a similar fashion before
ALTER started.
The second bug showed up as a deadlock between table-level
locks and InnoDB row locks, which was "detected" only after
innodb_lock_wait_timeout timeout.
A transaction would start using the table and modify a few
rows.
Then ALTER TABLE would come in, and start copying rows
into a temporary table. Eventually it would stumble on
the modified records and get blocked on a row lock.
The first transaction would try to do more updates, and get
blocked on thr_lock.c lock.
This situation of circular wait would only get resolved
by a timeout.
Both these bugs stemmed from inadequate solutions to the
problem of deadlocks occurring between different
locking subsystems.
In the first case we tried to avoid deadlocks between metadata
locking and table-level locking subsystems, when upgrading shared
metadata lock to exclusive one.
Transactions holding the shared lock on the table and waiting for
some table-level lock used to be aborted too aggressively.
We also allowed ALTER TABLE to start in presence of transactions
that modify the subject table. ALTER TABLE acquires
TL_WRITE_ALLOW_READ lock at start, and that block all writes
against the table (naturally, we don't want any writes to be lost
when switching the old and the new table). TL_WRITE_ALLOW_READ
lock, in turn, would block the started transaction on thr_lock.c
lock, should they do more updates. This, again, lead to the need
to abort such transactions.
The second bug occurred simply because we didn't have any
mechanism to detect deadlocks between the table-level locks
in thr_lock.c and row-level locks in InnoDB, other than
innodb_lock_wait_timeout.
This patch solves both these problems by moving lock conflicts
which are causing these deadlocks into the metadata locking
subsystem, thus making it possible to avoid or detect such
deadlocks inside MDL.
To do this we introduce new type-of-operation-aware metadata
locks, which allow MDL subsystem to know not only the fact that
transaction has used or is going to use some object but also what
kind of operation it has carried out or going to carry out on the
object.
This, along with the addition of a special kind of upgradable
metadata lock, allows ALTER TABLE to wait until all
transactions which has updated the table to go away.
This solves the second issue.
Another special type of upgradable metadata lock is acquired
by LOCK TABLE WRITE. This second lock type allows to solve the
first issue, since abortion of table-level locks in event of
DDL under LOCK TABLES becomes also unnecessary.
Below follows the list of incompatible changes introduced by
this patch:
- From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those
statements that acquire TL_WRITE_ALLOW_READ lock)
wait for all transactions which has *updated* the table to
complete.
- From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE
(i.e. all statements which acquire TL_WRITE table-level lock) wait
for all transaction which *updated or read* from the table
to complete.
As a consequence, innodb_table_locks=0 option no longer applies
to LOCK TABLES ... WRITE.
- DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort
statements or transactions which use tables being dropped or
renamed, and instead wait for these transactions to complete.
- Since LOCK TABLES WRITE now takes a special metadata lock,
not compatible with with reads or writes against the subject table
and transaction-wide, thr_lock.c deadlock avoidance algorithm
that used to ensure absence of deadlocks between LOCK TABLES
WRITE and other statements is no longer sufficient, even for
MyISAM. The wait-for graph based deadlock detector of MDL
subsystem may sometimes be necessary and is involved. This may
lead to ER_LOCK_DEADLOCK error produced for multi-statement
transactions even if these only use MyISAM:
session 1: session 2:
begin;
update t1 ... lock table t2 write, t1 write;
-- gets a lock on t2, blocks on t1
update t2 ...
(ER_LOCK_DEADLOCK)
- Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE
was abandoned.
LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same
priority as the usual LOCK TABLE ... WRITE.
SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in
the wait queue.
- We do not take upgradable metadata locks on implicitly
locked tables. So if one has, say, a view v1 that uses
table t1, and issues:
LOCK TABLE v1 WRITE;
FLUSH TABLE t1; -- (or just 'FLUSH TABLES'),
an error is produced.
In order to be able to perform DDL on a table under LOCK TABLES,
the table must be locked explicitly in the LOCK TABLES list.
mysql-test/include/handler.inc:
Adjusted test case to trigger an execution path on which bug 41110
"crash with handler command when used concurrently with alter
table" and bug 41112 "crash in mysql_ha_close_table/get_lock_data
with alter table" were originally discovered. Left old test case
which no longer triggers this execution path for the sake of
coverage.
Added test coverage for HANDLER SQL statements and type-aware
metadata locks.
Added a test for the global shared lock and HANDLER SQL.
Updated tests to take into account that the old simple deadlock
detection heuristics was replaced with a graph-based deadlock
detector.
mysql-test/r/debug_sync.result:
Updated results (see debug_sync.test).
mysql-test/r/handler_innodb.result:
Updated results (see handler.inc test).
mysql-test/r/handler_myisam.result:
Updated results (see handler.inc test).
mysql-test/r/innodb-lock.result:
Updated results (see innodb-lock.test).
mysql-test/r/innodb_mysql_lock.result:
Updated results (see innodb_mysql_lock.test).
mysql-test/r/lock.result:
Updated results (see lock.test).
mysql-test/r/lock_multi.result:
Updated results (see lock_multi.test).
mysql-test/r/lock_sync.result:
Updated results (see lock_sync.test).
mysql-test/r/mdl_sync.result:
Updated results (see mdl_sync.test).
mysql-test/r/sp-threads.result:
SHOW PROCESSLIST output has changed due to the fact that waiting
for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/r/truncate_coverage.result:
Updated results (see truncate_coverage.test).
mysql-test/suite/funcs_1/datadict/processlist_val.inc:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/funcs_1/r/processlist_val_no_prot.result:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/rpl/t/rpl_sp.test:
Updated to a new SHOW PROCESSLIST state name.
mysql-test/t/debug_sync.test:
Use LOCK TABLES READ instead of LOCK TABLES WRITE as the latter
no longer allows to trigger execution path involving waiting on
thr_lock.c lock and therefore reaching debug sync-point covered
by this test.
mysql-test/t/innodb-lock.test:
Adjusted test case to the fact that innodb_table_locks=0 option is
no longer supported, since LOCK TABLES WRITE handles all its
conflicts within MDL subsystem.
mysql-test/t/innodb_mysql_lock.test:
Added test for bug #37346 "innodb does not detect deadlock between
update and alter table".
mysql-test/t/lock.test:
Added test coverage which checks the fact that we no longer support
DDL under LOCK TABLES on tables which were locked implicitly.
Adjusted existing test cases accordingly.
mysql-test/t/lock_multi.test:
Added test for bug #46272 "MySQL 5.4.4, new MDL: unnecessary
deadlock". Adjusted other test cases to take into account the
fact that waiting for LOCK TABLES ... WRITE now happens within MDL
subsystem.
mysql-test/t/lock_sync.test:
Since LOCK TABLES ... WRITE now takes SNRW metadata lock for
tables locked explicitly we have to implicitly lock InnoDB tables
(through view) to trigger the table-level lock conflict between
TL_WRITE and TL_WRITE_ALLOW_WRITE.
mysql-test/t/mdl_sync.test:
Added basic test coverage for type-of-operation-aware metadata
locks. Also covered with tests some use cases involving HANDLER
statements in which a deadlock could arise.
Adjusted existing tests to take type-of-operation-aware MDL into
account.
mysql-test/t/multi_update.test:
Update to a new SHOW PROCESSLIST state name.
mysql-test/t/truncate_coverage.test:
Adjusted test case after making LOCK TABLES WRITE to wait until
transactions that use the table to be locked are completed.
Updated to the changed name of DEBUG_SYNC point.
sql/handler.cc:
Global read lock functionality has been
moved into a class.
sql/lock.cc:
Global read lock functionality has been
moved into a class.
Updated code to use the new MDL API.
sql/mdl.cc:
Introduced new type-of-operation aware metadata locks.
To do this:
- Changed MDL_lock to use one list for waiting requests and one
list for granted requests. For each list, added a bitmap
that holds information what lock types a list contains.
Added a helper class MDL_lock::List to manipulate with granted
and waited lists while keeping the bitmaps in sync
with list contents.
- Changed lock-compatibility functions to use bitmaps that
define compatibility.
- Introduced a graph based deadlock detector inspired by
waiting_threads.c from Maria implementation.
- Now that we have a deadlock detector, and no longer have
a global lock to protect individual lock objects, but rather
use an rw lock per object, removed redundant code for upgrade,
and the global read lock. Changed the MDL API to
no longer require the caller to acquire the global
intention exclusive lock by means of a separate method.
Removed a few more methods that became redundant.
- Removed deadlock detection heuristic, it has been made
obsolete by the deadlock detector.
- With operation-type-aware metadata locks, MDL subsystem has
become aware of potential conflicts between DDL and open
transactions. This made it possible to remove calls to
mysql_abort_transactions_with_shared_lock() from acquisition
paths for exclusive lock and lock upgrade. Now we can simply
wait for these transactions to complete without fear of
deadlock. Function mysql_lock_abort() has also become
unnecessary for all conflicting cases except when a DDL
conflicts with a connection that has an open HANDLER.
sql/mdl.h:
Introduced new type-of-operation aware metadata locks.
Introduced a graph based deadlock detector and supporting
methods.
Added comments.
God rid of redundant API calls.
Renamed m_lt_or_ha_sentinel to m_trans_sentinel,
since now it guards the global read lock as well as
LOCK TABLES and HANDLER locks.
sql/mysql_priv.h:
Moved the global read lock functionality into a
class.
Added MYSQL_OPEN_FORCE_SHARED_MDL flag which forces
open_tables() to take MDL_SHARED on tables instead of
metadata locks specified in the parser. We use this to
allow PREPARE run concurrently in presence of
LOCK TABLES ... WRITE.
Added signature for find_table_for_mdl_ugprade().
sql/set_var.cc:
Global read lock functionality has been
moved into a class.
sql/sp_head.cc:
When creating TABLE_LIST elements for prelocking or
system tables set the type of request for metadata
lock according to the operation that will be performed
on the table.
sql/sql_base.cc:
- Updated code to use the new MDL API.
- In order to avoid locks starvation we take upgradable
locks all at once. As result implicitly locked tables no
longer get an upgradable lock. Consequently DDL and FLUSH
TABLES for such tables is prohibited.
find_write_locked_table() was replaced by
find_table_for_mdl_upgrade() function.
open_table() was adjusted to return TABLE instance with
upgradable ticket when necessary.
- We no longer wait for all locks on OT_WAIT back off
action -- only on the lock that caused the wait
conflict. Moreover, now we distinguish cases when we
have to wait due to conflict in MDL and old version
of table in TDC.
- Upate mysql_notify_threads_having_share_locks()
to only abort thr_lock.c waits of threads that
have open HANDLERs, since lock conflicts with only
these threads now can lead to deadlocks not detectable
by the MDL deadlock detector.
- Remove mysql_abort_transactions_with_shared_locks()
which is no longer needed.
sql/sql_class.cc:
Global read lock functionality has been moved into a class.
Re-arranged code in THD::cleanup() to simplify assert.
sql/sql_class.h:
Introduced class to incapsulate global read lock
functionality.
Now sentinel in MDL subsystem guards the global read lock
as well as LOCK TABLES and HANDLER locks. Adjusted code
accordingly.
sql/sql_db.cc:
Global read lock functionality has been moved into a class.
sql/sql_delete.cc:
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
TRUNCATE TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_handler.cc:
Inform MDL_context about presence of open HANDLERs.
Since HANLDERs break MDL protocol by acquiring table-level
lock while holding only S metadata lock on a table MDL
subsystem should take special care about such contexts (Now
this is the only case when mysql_lock_abort() is used).
sql/sql_parse.cc:
Global read lock functionality has been moved into a class.
Do not take upgradable metadata locks when opening tables
for CREATE TABLE SELECT as it is not necessary and limits
concurrency.
When initializing TABLE_LIST objects before adding them
to the table list set the type of request for metadata lock
according to the operation that will be performed on the
table.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result FLUSH
TABLES is no longer allowed for such tables.
sql/sql_prepare.cc:
Use MYSQL_OPEN_FORCE_SHARED_MDL flag when opening
tables during PREPARE. This allows PREPARE to run
concurrently in presence of LOCK TABLES ... WRITE.
sql/sql_rename.cc:
Global read lock functionality has been moved into a class.
sql/sql_show.cc:
Updated code to use the new MDL API.
sql/sql_table.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result DROP
TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_trigger.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
CREATE/DROP TRIGGER is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_view.cc:
Global read lock functionality has been moved into a class.
Fixed results of wrong merge that led to misuse of GLR API.
CREATE VIEW statement is not a commit statement.
sql/table.cc:
When resetting TABLE_LIST objects for PS or SP re-execution
set the type of request for metadata lock according to the
operation that will be performed on the table. Do the same
in auxiliary function initializing metadata lock requests
in a table list.
sql/table.h:
When initializing TABLE_LIST objects set the type of request
for metadata lock according to the operation that will be
performed on the table.
sql/transaction.cc:
Global read lock functionality has been moved into a class.
2010-02-01 12:43:06 +01:00
# --> connection con2
# Waiting for 'drop table t2' to get blocked...
# --> connection con1
drop table t1;
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
ERROR 40001: Deadlock found when trying to get lock; try restarting transaction
Implement new type-of-operation-aware metadata locks.
Add a wait-for graph based deadlock detector to the
MDL subsystem.
Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and
bug #37346 "innodb does not detect deadlock between update and
alter table".
The first bug manifested itself as an unwarranted abort of a
transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER
statement, when this transaction tried to repeat use of a
table, which it has already used in a similar fashion before
ALTER started.
The second bug showed up as a deadlock between table-level
locks and InnoDB row locks, which was "detected" only after
innodb_lock_wait_timeout timeout.
A transaction would start using the table and modify a few
rows.
Then ALTER TABLE would come in, and start copying rows
into a temporary table. Eventually it would stumble on
the modified records and get blocked on a row lock.
The first transaction would try to do more updates, and get
blocked on thr_lock.c lock.
This situation of circular wait would only get resolved
by a timeout.
Both these bugs stemmed from inadequate solutions to the
problem of deadlocks occurring between different
locking subsystems.
In the first case we tried to avoid deadlocks between metadata
locking and table-level locking subsystems, when upgrading shared
metadata lock to exclusive one.
Transactions holding the shared lock on the table and waiting for
some table-level lock used to be aborted too aggressively.
We also allowed ALTER TABLE to start in presence of transactions
that modify the subject table. ALTER TABLE acquires
TL_WRITE_ALLOW_READ lock at start, and that block all writes
against the table (naturally, we don't want any writes to be lost
when switching the old and the new table). TL_WRITE_ALLOW_READ
lock, in turn, would block the started transaction on thr_lock.c
lock, should they do more updates. This, again, lead to the need
to abort such transactions.
The second bug occurred simply because we didn't have any
mechanism to detect deadlocks between the table-level locks
in thr_lock.c and row-level locks in InnoDB, other than
innodb_lock_wait_timeout.
This patch solves both these problems by moving lock conflicts
which are causing these deadlocks into the metadata locking
subsystem, thus making it possible to avoid or detect such
deadlocks inside MDL.
To do this we introduce new type-of-operation-aware metadata
locks, which allow MDL subsystem to know not only the fact that
transaction has used or is going to use some object but also what
kind of operation it has carried out or going to carry out on the
object.
This, along with the addition of a special kind of upgradable
metadata lock, allows ALTER TABLE to wait until all
transactions which has updated the table to go away.
This solves the second issue.
Another special type of upgradable metadata lock is acquired
by LOCK TABLE WRITE. This second lock type allows to solve the
first issue, since abortion of table-level locks in event of
DDL under LOCK TABLES becomes also unnecessary.
Below follows the list of incompatible changes introduced by
this patch:
- From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those
statements that acquire TL_WRITE_ALLOW_READ lock)
wait for all transactions which has *updated* the table to
complete.
- From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE
(i.e. all statements which acquire TL_WRITE table-level lock) wait
for all transaction which *updated or read* from the table
to complete.
As a consequence, innodb_table_locks=0 option no longer applies
to LOCK TABLES ... WRITE.
- DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort
statements or transactions which use tables being dropped or
renamed, and instead wait for these transactions to complete.
- Since LOCK TABLES WRITE now takes a special metadata lock,
not compatible with with reads or writes against the subject table
and transaction-wide, thr_lock.c deadlock avoidance algorithm
that used to ensure absence of deadlocks between LOCK TABLES
WRITE and other statements is no longer sufficient, even for
MyISAM. The wait-for graph based deadlock detector of MDL
subsystem may sometimes be necessary and is involved. This may
lead to ER_LOCK_DEADLOCK error produced for multi-statement
transactions even if these only use MyISAM:
session 1: session 2:
begin;
update t1 ... lock table t2 write, t1 write;
-- gets a lock on t2, blocks on t1
update t2 ...
(ER_LOCK_DEADLOCK)
- Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE
was abandoned.
LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same
priority as the usual LOCK TABLE ... WRITE.
SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in
the wait queue.
- We do not take upgradable metadata locks on implicitly
locked tables. So if one has, say, a view v1 that uses
table t1, and issues:
LOCK TABLE v1 WRITE;
FLUSH TABLE t1; -- (or just 'FLUSH TABLES'),
an error is produced.
In order to be able to perform DDL on a table under LOCK TABLES,
the table must be locked explicitly in the LOCK TABLES list.
mysql-test/include/handler.inc:
Adjusted test case to trigger an execution path on which bug 41110
"crash with handler command when used concurrently with alter
table" and bug 41112 "crash in mysql_ha_close_table/get_lock_data
with alter table" were originally discovered. Left old test case
which no longer triggers this execution path for the sake of
coverage.
Added test coverage for HANDLER SQL statements and type-aware
metadata locks.
Added a test for the global shared lock and HANDLER SQL.
Updated tests to take into account that the old simple deadlock
detection heuristics was replaced with a graph-based deadlock
detector.
mysql-test/r/debug_sync.result:
Updated results (see debug_sync.test).
mysql-test/r/handler_innodb.result:
Updated results (see handler.inc test).
mysql-test/r/handler_myisam.result:
Updated results (see handler.inc test).
mysql-test/r/innodb-lock.result:
Updated results (see innodb-lock.test).
mysql-test/r/innodb_mysql_lock.result:
Updated results (see innodb_mysql_lock.test).
mysql-test/r/lock.result:
Updated results (see lock.test).
mysql-test/r/lock_multi.result:
Updated results (see lock_multi.test).
mysql-test/r/lock_sync.result:
Updated results (see lock_sync.test).
mysql-test/r/mdl_sync.result:
Updated results (see mdl_sync.test).
mysql-test/r/sp-threads.result:
SHOW PROCESSLIST output has changed due to the fact that waiting
for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/r/truncate_coverage.result:
Updated results (see truncate_coverage.test).
mysql-test/suite/funcs_1/datadict/processlist_val.inc:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/funcs_1/r/processlist_val_no_prot.result:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/rpl/t/rpl_sp.test:
Updated to a new SHOW PROCESSLIST state name.
mysql-test/t/debug_sync.test:
Use LOCK TABLES READ instead of LOCK TABLES WRITE as the latter
no longer allows to trigger execution path involving waiting on
thr_lock.c lock and therefore reaching debug sync-point covered
by this test.
mysql-test/t/innodb-lock.test:
Adjusted test case to the fact that innodb_table_locks=0 option is
no longer supported, since LOCK TABLES WRITE handles all its
conflicts within MDL subsystem.
mysql-test/t/innodb_mysql_lock.test:
Added test for bug #37346 "innodb does not detect deadlock between
update and alter table".
mysql-test/t/lock.test:
Added test coverage which checks the fact that we no longer support
DDL under LOCK TABLES on tables which were locked implicitly.
Adjusted existing test cases accordingly.
mysql-test/t/lock_multi.test:
Added test for bug #46272 "MySQL 5.4.4, new MDL: unnecessary
deadlock". Adjusted other test cases to take into account the
fact that waiting for LOCK TABLES ... WRITE now happens within MDL
subsystem.
mysql-test/t/lock_sync.test:
Since LOCK TABLES ... WRITE now takes SNRW metadata lock for
tables locked explicitly we have to implicitly lock InnoDB tables
(through view) to trigger the table-level lock conflict between
TL_WRITE and TL_WRITE_ALLOW_WRITE.
mysql-test/t/mdl_sync.test:
Added basic test coverage for type-of-operation-aware metadata
locks. Also covered with tests some use cases involving HANDLER
statements in which a deadlock could arise.
Adjusted existing tests to take type-of-operation-aware MDL into
account.
mysql-test/t/multi_update.test:
Update to a new SHOW PROCESSLIST state name.
mysql-test/t/truncate_coverage.test:
Adjusted test case after making LOCK TABLES WRITE to wait until
transactions that use the table to be locked are completed.
Updated to the changed name of DEBUG_SYNC point.
sql/handler.cc:
Global read lock functionality has been
moved into a class.
sql/lock.cc:
Global read lock functionality has been
moved into a class.
Updated code to use the new MDL API.
sql/mdl.cc:
Introduced new type-of-operation aware metadata locks.
To do this:
- Changed MDL_lock to use one list for waiting requests and one
list for granted requests. For each list, added a bitmap
that holds information what lock types a list contains.
Added a helper class MDL_lock::List to manipulate with granted
and waited lists while keeping the bitmaps in sync
with list contents.
- Changed lock-compatibility functions to use bitmaps that
define compatibility.
- Introduced a graph based deadlock detector inspired by
waiting_threads.c from Maria implementation.
- Now that we have a deadlock detector, and no longer have
a global lock to protect individual lock objects, but rather
use an rw lock per object, removed redundant code for upgrade,
and the global read lock. Changed the MDL API to
no longer require the caller to acquire the global
intention exclusive lock by means of a separate method.
Removed a few more methods that became redundant.
- Removed deadlock detection heuristic, it has been made
obsolete by the deadlock detector.
- With operation-type-aware metadata locks, MDL subsystem has
become aware of potential conflicts between DDL and open
transactions. This made it possible to remove calls to
mysql_abort_transactions_with_shared_lock() from acquisition
paths for exclusive lock and lock upgrade. Now we can simply
wait for these transactions to complete without fear of
deadlock. Function mysql_lock_abort() has also become
unnecessary for all conflicting cases except when a DDL
conflicts with a connection that has an open HANDLER.
sql/mdl.h:
Introduced new type-of-operation aware metadata locks.
Introduced a graph based deadlock detector and supporting
methods.
Added comments.
God rid of redundant API calls.
Renamed m_lt_or_ha_sentinel to m_trans_sentinel,
since now it guards the global read lock as well as
LOCK TABLES and HANDLER locks.
sql/mysql_priv.h:
Moved the global read lock functionality into a
class.
Added MYSQL_OPEN_FORCE_SHARED_MDL flag which forces
open_tables() to take MDL_SHARED on tables instead of
metadata locks specified in the parser. We use this to
allow PREPARE run concurrently in presence of
LOCK TABLES ... WRITE.
Added signature for find_table_for_mdl_ugprade().
sql/set_var.cc:
Global read lock functionality has been
moved into a class.
sql/sp_head.cc:
When creating TABLE_LIST elements for prelocking or
system tables set the type of request for metadata
lock according to the operation that will be performed
on the table.
sql/sql_base.cc:
- Updated code to use the new MDL API.
- In order to avoid locks starvation we take upgradable
locks all at once. As result implicitly locked tables no
longer get an upgradable lock. Consequently DDL and FLUSH
TABLES for such tables is prohibited.
find_write_locked_table() was replaced by
find_table_for_mdl_upgrade() function.
open_table() was adjusted to return TABLE instance with
upgradable ticket when necessary.
- We no longer wait for all locks on OT_WAIT back off
action -- only on the lock that caused the wait
conflict. Moreover, now we distinguish cases when we
have to wait due to conflict in MDL and old version
of table in TDC.
- Upate mysql_notify_threads_having_share_locks()
to only abort thr_lock.c waits of threads that
have open HANDLERs, since lock conflicts with only
these threads now can lead to deadlocks not detectable
by the MDL deadlock detector.
- Remove mysql_abort_transactions_with_shared_locks()
which is no longer needed.
sql/sql_class.cc:
Global read lock functionality has been moved into a class.
Re-arranged code in THD::cleanup() to simplify assert.
sql/sql_class.h:
Introduced class to incapsulate global read lock
functionality.
Now sentinel in MDL subsystem guards the global read lock
as well as LOCK TABLES and HANDLER locks. Adjusted code
accordingly.
sql/sql_db.cc:
Global read lock functionality has been moved into a class.
sql/sql_delete.cc:
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
TRUNCATE TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_handler.cc:
Inform MDL_context about presence of open HANDLERs.
Since HANLDERs break MDL protocol by acquiring table-level
lock while holding only S metadata lock on a table MDL
subsystem should take special care about such contexts (Now
this is the only case when mysql_lock_abort() is used).
sql/sql_parse.cc:
Global read lock functionality has been moved into a class.
Do not take upgradable metadata locks when opening tables
for CREATE TABLE SELECT as it is not necessary and limits
concurrency.
When initializing TABLE_LIST objects before adding them
to the table list set the type of request for metadata lock
according to the operation that will be performed on the
table.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result FLUSH
TABLES is no longer allowed for such tables.
sql/sql_prepare.cc:
Use MYSQL_OPEN_FORCE_SHARED_MDL flag when opening
tables during PREPARE. This allows PREPARE to run
concurrently in presence of LOCK TABLES ... WRITE.
sql/sql_rename.cc:
Global read lock functionality has been moved into a class.
sql/sql_show.cc:
Updated code to use the new MDL API.
sql/sql_table.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result DROP
TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_trigger.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
CREATE/DROP TRIGGER is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_view.cc:
Global read lock functionality has been moved into a class.
Fixed results of wrong merge that led to misuse of GLR API.
CREATE VIEW statement is not a commit statement.
sql/table.cc:
When resetting TABLE_LIST objects for PS or SP re-execution
set the type of request for metadata lock according to the
operation that will be performed on the table. Do the same
in auxiliary function initializing metadata lock requests
in a table list.
sql/table.h:
When initializing TABLE_LIST objects set the type of request
for metadata lock according to the operation that will be
performed on the table.
sql/transaction.cc:
Global read lock functionality has been moved into a class.
2010-02-01 12:43:06 +01:00
unlock tables;
# --> connection default
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
# Demonstrate that there is no deadlock with FLUSH TABLE,
# even though it is waiting for the other table to go away
Implement new type-of-operation-aware metadata locks.
Add a wait-for graph based deadlock detector to the
MDL subsystem.
Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and
bug #37346 "innodb does not detect deadlock between update and
alter table".
The first bug manifested itself as an unwarranted abort of a
transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER
statement, when this transaction tried to repeat use of a
table, which it has already used in a similar fashion before
ALTER started.
The second bug showed up as a deadlock between table-level
locks and InnoDB row locks, which was "detected" only after
innodb_lock_wait_timeout timeout.
A transaction would start using the table and modify a few
rows.
Then ALTER TABLE would come in, and start copying rows
into a temporary table. Eventually it would stumble on
the modified records and get blocked on a row lock.
The first transaction would try to do more updates, and get
blocked on thr_lock.c lock.
This situation of circular wait would only get resolved
by a timeout.
Both these bugs stemmed from inadequate solutions to the
problem of deadlocks occurring between different
locking subsystems.
In the first case we tried to avoid deadlocks between metadata
locking and table-level locking subsystems, when upgrading shared
metadata lock to exclusive one.
Transactions holding the shared lock on the table and waiting for
some table-level lock used to be aborted too aggressively.
We also allowed ALTER TABLE to start in presence of transactions
that modify the subject table. ALTER TABLE acquires
TL_WRITE_ALLOW_READ lock at start, and that block all writes
against the table (naturally, we don't want any writes to be lost
when switching the old and the new table). TL_WRITE_ALLOW_READ
lock, in turn, would block the started transaction on thr_lock.c
lock, should they do more updates. This, again, lead to the need
to abort such transactions.
The second bug occurred simply because we didn't have any
mechanism to detect deadlocks between the table-level locks
in thr_lock.c and row-level locks in InnoDB, other than
innodb_lock_wait_timeout.
This patch solves both these problems by moving lock conflicts
which are causing these deadlocks into the metadata locking
subsystem, thus making it possible to avoid or detect such
deadlocks inside MDL.
To do this we introduce new type-of-operation-aware metadata
locks, which allow MDL subsystem to know not only the fact that
transaction has used or is going to use some object but also what
kind of operation it has carried out or going to carry out on the
object.
This, along with the addition of a special kind of upgradable
metadata lock, allows ALTER TABLE to wait until all
transactions which has updated the table to go away.
This solves the second issue.
Another special type of upgradable metadata lock is acquired
by LOCK TABLE WRITE. This second lock type allows to solve the
first issue, since abortion of table-level locks in event of
DDL under LOCK TABLES becomes also unnecessary.
Below follows the list of incompatible changes introduced by
this patch:
- From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those
statements that acquire TL_WRITE_ALLOW_READ lock)
wait for all transactions which has *updated* the table to
complete.
- From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE
(i.e. all statements which acquire TL_WRITE table-level lock) wait
for all transaction which *updated or read* from the table
to complete.
As a consequence, innodb_table_locks=0 option no longer applies
to LOCK TABLES ... WRITE.
- DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort
statements or transactions which use tables being dropped or
renamed, and instead wait for these transactions to complete.
- Since LOCK TABLES WRITE now takes a special metadata lock,
not compatible with with reads or writes against the subject table
and transaction-wide, thr_lock.c deadlock avoidance algorithm
that used to ensure absence of deadlocks between LOCK TABLES
WRITE and other statements is no longer sufficient, even for
MyISAM. The wait-for graph based deadlock detector of MDL
subsystem may sometimes be necessary and is involved. This may
lead to ER_LOCK_DEADLOCK error produced for multi-statement
transactions even if these only use MyISAM:
session 1: session 2:
begin;
update t1 ... lock table t2 write, t1 write;
-- gets a lock on t2, blocks on t1
update t2 ...
(ER_LOCK_DEADLOCK)
- Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE
was abandoned.
LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same
priority as the usual LOCK TABLE ... WRITE.
SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in
the wait queue.
- We do not take upgradable metadata locks on implicitly
locked tables. So if one has, say, a view v1 that uses
table t1, and issues:
LOCK TABLE v1 WRITE;
FLUSH TABLE t1; -- (or just 'FLUSH TABLES'),
an error is produced.
In order to be able to perform DDL on a table under LOCK TABLES,
the table must be locked explicitly in the LOCK TABLES list.
mysql-test/include/handler.inc:
Adjusted test case to trigger an execution path on which bug 41110
"crash with handler command when used concurrently with alter
table" and bug 41112 "crash in mysql_ha_close_table/get_lock_data
with alter table" were originally discovered. Left old test case
which no longer triggers this execution path for the sake of
coverage.
Added test coverage for HANDLER SQL statements and type-aware
metadata locks.
Added a test for the global shared lock and HANDLER SQL.
Updated tests to take into account that the old simple deadlock
detection heuristics was replaced with a graph-based deadlock
detector.
mysql-test/r/debug_sync.result:
Updated results (see debug_sync.test).
mysql-test/r/handler_innodb.result:
Updated results (see handler.inc test).
mysql-test/r/handler_myisam.result:
Updated results (see handler.inc test).
mysql-test/r/innodb-lock.result:
Updated results (see innodb-lock.test).
mysql-test/r/innodb_mysql_lock.result:
Updated results (see innodb_mysql_lock.test).
mysql-test/r/lock.result:
Updated results (see lock.test).
mysql-test/r/lock_multi.result:
Updated results (see lock_multi.test).
mysql-test/r/lock_sync.result:
Updated results (see lock_sync.test).
mysql-test/r/mdl_sync.result:
Updated results (see mdl_sync.test).
mysql-test/r/sp-threads.result:
SHOW PROCESSLIST output has changed due to the fact that waiting
for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/r/truncate_coverage.result:
Updated results (see truncate_coverage.test).
mysql-test/suite/funcs_1/datadict/processlist_val.inc:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/funcs_1/r/processlist_val_no_prot.result:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/rpl/t/rpl_sp.test:
Updated to a new SHOW PROCESSLIST state name.
mysql-test/t/debug_sync.test:
Use LOCK TABLES READ instead of LOCK TABLES WRITE as the latter
no longer allows to trigger execution path involving waiting on
thr_lock.c lock and therefore reaching debug sync-point covered
by this test.
mysql-test/t/innodb-lock.test:
Adjusted test case to the fact that innodb_table_locks=0 option is
no longer supported, since LOCK TABLES WRITE handles all its
conflicts within MDL subsystem.
mysql-test/t/innodb_mysql_lock.test:
Added test for bug #37346 "innodb does not detect deadlock between
update and alter table".
mysql-test/t/lock.test:
Added test coverage which checks the fact that we no longer support
DDL under LOCK TABLES on tables which were locked implicitly.
Adjusted existing test cases accordingly.
mysql-test/t/lock_multi.test:
Added test for bug #46272 "MySQL 5.4.4, new MDL: unnecessary
deadlock". Adjusted other test cases to take into account the
fact that waiting for LOCK TABLES ... WRITE now happens within MDL
subsystem.
mysql-test/t/lock_sync.test:
Since LOCK TABLES ... WRITE now takes SNRW metadata lock for
tables locked explicitly we have to implicitly lock InnoDB tables
(through view) to trigger the table-level lock conflict between
TL_WRITE and TL_WRITE_ALLOW_WRITE.
mysql-test/t/mdl_sync.test:
Added basic test coverage for type-of-operation-aware metadata
locks. Also covered with tests some use cases involving HANDLER
statements in which a deadlock could arise.
Adjusted existing tests to take type-of-operation-aware MDL into
account.
mysql-test/t/multi_update.test:
Update to a new SHOW PROCESSLIST state name.
mysql-test/t/truncate_coverage.test:
Adjusted test case after making LOCK TABLES WRITE to wait until
transactions that use the table to be locked are completed.
Updated to the changed name of DEBUG_SYNC point.
sql/handler.cc:
Global read lock functionality has been
moved into a class.
sql/lock.cc:
Global read lock functionality has been
moved into a class.
Updated code to use the new MDL API.
sql/mdl.cc:
Introduced new type-of-operation aware metadata locks.
To do this:
- Changed MDL_lock to use one list for waiting requests and one
list for granted requests. For each list, added a bitmap
that holds information what lock types a list contains.
Added a helper class MDL_lock::List to manipulate with granted
and waited lists while keeping the bitmaps in sync
with list contents.
- Changed lock-compatibility functions to use bitmaps that
define compatibility.
- Introduced a graph based deadlock detector inspired by
waiting_threads.c from Maria implementation.
- Now that we have a deadlock detector, and no longer have
a global lock to protect individual lock objects, but rather
use an rw lock per object, removed redundant code for upgrade,
and the global read lock. Changed the MDL API to
no longer require the caller to acquire the global
intention exclusive lock by means of a separate method.
Removed a few more methods that became redundant.
- Removed deadlock detection heuristic, it has been made
obsolete by the deadlock detector.
- With operation-type-aware metadata locks, MDL subsystem has
become aware of potential conflicts between DDL and open
transactions. This made it possible to remove calls to
mysql_abort_transactions_with_shared_lock() from acquisition
paths for exclusive lock and lock upgrade. Now we can simply
wait for these transactions to complete without fear of
deadlock. Function mysql_lock_abort() has also become
unnecessary for all conflicting cases except when a DDL
conflicts with a connection that has an open HANDLER.
sql/mdl.h:
Introduced new type-of-operation aware metadata locks.
Introduced a graph based deadlock detector and supporting
methods.
Added comments.
God rid of redundant API calls.
Renamed m_lt_or_ha_sentinel to m_trans_sentinel,
since now it guards the global read lock as well as
LOCK TABLES and HANDLER locks.
sql/mysql_priv.h:
Moved the global read lock functionality into a
class.
Added MYSQL_OPEN_FORCE_SHARED_MDL flag which forces
open_tables() to take MDL_SHARED on tables instead of
metadata locks specified in the parser. We use this to
allow PREPARE run concurrently in presence of
LOCK TABLES ... WRITE.
Added signature for find_table_for_mdl_ugprade().
sql/set_var.cc:
Global read lock functionality has been
moved into a class.
sql/sp_head.cc:
When creating TABLE_LIST elements for prelocking or
system tables set the type of request for metadata
lock according to the operation that will be performed
on the table.
sql/sql_base.cc:
- Updated code to use the new MDL API.
- In order to avoid locks starvation we take upgradable
locks all at once. As result implicitly locked tables no
longer get an upgradable lock. Consequently DDL and FLUSH
TABLES for such tables is prohibited.
find_write_locked_table() was replaced by
find_table_for_mdl_upgrade() function.
open_table() was adjusted to return TABLE instance with
upgradable ticket when necessary.
- We no longer wait for all locks on OT_WAIT back off
action -- only on the lock that caused the wait
conflict. Moreover, now we distinguish cases when we
have to wait due to conflict in MDL and old version
of table in TDC.
- Upate mysql_notify_threads_having_share_locks()
to only abort thr_lock.c waits of threads that
have open HANDLERs, since lock conflicts with only
these threads now can lead to deadlocks not detectable
by the MDL deadlock detector.
- Remove mysql_abort_transactions_with_shared_locks()
which is no longer needed.
sql/sql_class.cc:
Global read lock functionality has been moved into a class.
Re-arranged code in THD::cleanup() to simplify assert.
sql/sql_class.h:
Introduced class to incapsulate global read lock
functionality.
Now sentinel in MDL subsystem guards the global read lock
as well as LOCK TABLES and HANDLER locks. Adjusted code
accordingly.
sql/sql_db.cc:
Global read lock functionality has been moved into a class.
sql/sql_delete.cc:
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
TRUNCATE TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_handler.cc:
Inform MDL_context about presence of open HANDLERs.
Since HANLDERs break MDL protocol by acquiring table-level
lock while holding only S metadata lock on a table MDL
subsystem should take special care about such contexts (Now
this is the only case when mysql_lock_abort() is used).
sql/sql_parse.cc:
Global read lock functionality has been moved into a class.
Do not take upgradable metadata locks when opening tables
for CREATE TABLE SELECT as it is not necessary and limits
concurrency.
When initializing TABLE_LIST objects before adding them
to the table list set the type of request for metadata lock
according to the operation that will be performed on the
table.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result FLUSH
TABLES is no longer allowed for such tables.
sql/sql_prepare.cc:
Use MYSQL_OPEN_FORCE_SHARED_MDL flag when opening
tables during PREPARE. This allows PREPARE to run
concurrently in presence of LOCK TABLES ... WRITE.
sql/sql_rename.cc:
Global read lock functionality has been moved into a class.
sql/sql_show.cc:
Updated code to use the new MDL API.
sql/sql_table.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result DROP
TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_trigger.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
CREATE/DROP TRIGGER is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_view.cc:
Global read lock functionality has been moved into a class.
Fixed results of wrong merge that led to misuse of GLR API.
CREATE VIEW statement is not a commit statement.
sql/table.cc:
When resetting TABLE_LIST objects for PS or SP re-execution
set the type of request for metadata lock according to the
operation that will be performed on the table. Do the same
in auxiliary function initializing metadata lock requests
in a table list.
sql/table.h:
When initializing TABLE_LIST objects set the type of request
for metadata lock according to the operation that will be
performed on the table.
sql/transaction.cc:
Global read lock functionality has been moved into a class.
2010-02-01 12:43:06 +01:00
create table t2 like t1;
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
# Sending:
flush table t2;
# --> connection con2
drop table t1;
# --> connection con1
unlock tables;
# --> connection default
# Reaping 'flush table t2'...
drop table t2;
#
# Bug #46224 HANDLER statements within a transaction might
# lead to deadlocks
#
create table t1 (a int, key a(a));
Implement new type-of-operation-aware metadata locks.
Add a wait-for graph based deadlock detector to the
MDL subsystem.
Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and
bug #37346 "innodb does not detect deadlock between update and
alter table".
The first bug manifested itself as an unwarranted abort of a
transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER
statement, when this transaction tried to repeat use of a
table, which it has already used in a similar fashion before
ALTER started.
The second bug showed up as a deadlock between table-level
locks and InnoDB row locks, which was "detected" only after
innodb_lock_wait_timeout timeout.
A transaction would start using the table and modify a few
rows.
Then ALTER TABLE would come in, and start copying rows
into a temporary table. Eventually it would stumble on
the modified records and get blocked on a row lock.
The first transaction would try to do more updates, and get
blocked on thr_lock.c lock.
This situation of circular wait would only get resolved
by a timeout.
Both these bugs stemmed from inadequate solutions to the
problem of deadlocks occurring between different
locking subsystems.
In the first case we tried to avoid deadlocks between metadata
locking and table-level locking subsystems, when upgrading shared
metadata lock to exclusive one.
Transactions holding the shared lock on the table and waiting for
some table-level lock used to be aborted too aggressively.
We also allowed ALTER TABLE to start in presence of transactions
that modify the subject table. ALTER TABLE acquires
TL_WRITE_ALLOW_READ lock at start, and that block all writes
against the table (naturally, we don't want any writes to be lost
when switching the old and the new table). TL_WRITE_ALLOW_READ
lock, in turn, would block the started transaction on thr_lock.c
lock, should they do more updates. This, again, lead to the need
to abort such transactions.
The second bug occurred simply because we didn't have any
mechanism to detect deadlocks between the table-level locks
in thr_lock.c and row-level locks in InnoDB, other than
innodb_lock_wait_timeout.
This patch solves both these problems by moving lock conflicts
which are causing these deadlocks into the metadata locking
subsystem, thus making it possible to avoid or detect such
deadlocks inside MDL.
To do this we introduce new type-of-operation-aware metadata
locks, which allow MDL subsystem to know not only the fact that
transaction has used or is going to use some object but also what
kind of operation it has carried out or going to carry out on the
object.
This, along with the addition of a special kind of upgradable
metadata lock, allows ALTER TABLE to wait until all
transactions which has updated the table to go away.
This solves the second issue.
Another special type of upgradable metadata lock is acquired
by LOCK TABLE WRITE. This second lock type allows to solve the
first issue, since abortion of table-level locks in event of
DDL under LOCK TABLES becomes also unnecessary.
Below follows the list of incompatible changes introduced by
this patch:
- From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those
statements that acquire TL_WRITE_ALLOW_READ lock)
wait for all transactions which has *updated* the table to
complete.
- From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE
(i.e. all statements which acquire TL_WRITE table-level lock) wait
for all transaction which *updated or read* from the table
to complete.
As a consequence, innodb_table_locks=0 option no longer applies
to LOCK TABLES ... WRITE.
- DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort
statements or transactions which use tables being dropped or
renamed, and instead wait for these transactions to complete.
- Since LOCK TABLES WRITE now takes a special metadata lock,
not compatible with with reads or writes against the subject table
and transaction-wide, thr_lock.c deadlock avoidance algorithm
that used to ensure absence of deadlocks between LOCK TABLES
WRITE and other statements is no longer sufficient, even for
MyISAM. The wait-for graph based deadlock detector of MDL
subsystem may sometimes be necessary and is involved. This may
lead to ER_LOCK_DEADLOCK error produced for multi-statement
transactions even if these only use MyISAM:
session 1: session 2:
begin;
update t1 ... lock table t2 write, t1 write;
-- gets a lock on t2, blocks on t1
update t2 ...
(ER_LOCK_DEADLOCK)
- Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE
was abandoned.
LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same
priority as the usual LOCK TABLE ... WRITE.
SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in
the wait queue.
- We do not take upgradable metadata locks on implicitly
locked tables. So if one has, say, a view v1 that uses
table t1, and issues:
LOCK TABLE v1 WRITE;
FLUSH TABLE t1; -- (or just 'FLUSH TABLES'),
an error is produced.
In order to be able to perform DDL on a table under LOCK TABLES,
the table must be locked explicitly in the LOCK TABLES list.
mysql-test/include/handler.inc:
Adjusted test case to trigger an execution path on which bug 41110
"crash with handler command when used concurrently with alter
table" and bug 41112 "crash in mysql_ha_close_table/get_lock_data
with alter table" were originally discovered. Left old test case
which no longer triggers this execution path for the sake of
coverage.
Added test coverage for HANDLER SQL statements and type-aware
metadata locks.
Added a test for the global shared lock and HANDLER SQL.
Updated tests to take into account that the old simple deadlock
detection heuristics was replaced with a graph-based deadlock
detector.
mysql-test/r/debug_sync.result:
Updated results (see debug_sync.test).
mysql-test/r/handler_innodb.result:
Updated results (see handler.inc test).
mysql-test/r/handler_myisam.result:
Updated results (see handler.inc test).
mysql-test/r/innodb-lock.result:
Updated results (see innodb-lock.test).
mysql-test/r/innodb_mysql_lock.result:
Updated results (see innodb_mysql_lock.test).
mysql-test/r/lock.result:
Updated results (see lock.test).
mysql-test/r/lock_multi.result:
Updated results (see lock_multi.test).
mysql-test/r/lock_sync.result:
Updated results (see lock_sync.test).
mysql-test/r/mdl_sync.result:
Updated results (see mdl_sync.test).
mysql-test/r/sp-threads.result:
SHOW PROCESSLIST output has changed due to the fact that waiting
for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/r/truncate_coverage.result:
Updated results (see truncate_coverage.test).
mysql-test/suite/funcs_1/datadict/processlist_val.inc:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/funcs_1/r/processlist_val_no_prot.result:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/rpl/t/rpl_sp.test:
Updated to a new SHOW PROCESSLIST state name.
mysql-test/t/debug_sync.test:
Use LOCK TABLES READ instead of LOCK TABLES WRITE as the latter
no longer allows to trigger execution path involving waiting on
thr_lock.c lock and therefore reaching debug sync-point covered
by this test.
mysql-test/t/innodb-lock.test:
Adjusted test case to the fact that innodb_table_locks=0 option is
no longer supported, since LOCK TABLES WRITE handles all its
conflicts within MDL subsystem.
mysql-test/t/innodb_mysql_lock.test:
Added test for bug #37346 "innodb does not detect deadlock between
update and alter table".
mysql-test/t/lock.test:
Added test coverage which checks the fact that we no longer support
DDL under LOCK TABLES on tables which were locked implicitly.
Adjusted existing test cases accordingly.
mysql-test/t/lock_multi.test:
Added test for bug #46272 "MySQL 5.4.4, new MDL: unnecessary
deadlock". Adjusted other test cases to take into account the
fact that waiting for LOCK TABLES ... WRITE now happens within MDL
subsystem.
mysql-test/t/lock_sync.test:
Since LOCK TABLES ... WRITE now takes SNRW metadata lock for
tables locked explicitly we have to implicitly lock InnoDB tables
(through view) to trigger the table-level lock conflict between
TL_WRITE and TL_WRITE_ALLOW_WRITE.
mysql-test/t/mdl_sync.test:
Added basic test coverage for type-of-operation-aware metadata
locks. Also covered with tests some use cases involving HANDLER
statements in which a deadlock could arise.
Adjusted existing tests to take type-of-operation-aware MDL into
account.
mysql-test/t/multi_update.test:
Update to a new SHOW PROCESSLIST state name.
mysql-test/t/truncate_coverage.test:
Adjusted test case after making LOCK TABLES WRITE to wait until
transactions that use the table to be locked are completed.
Updated to the changed name of DEBUG_SYNC point.
sql/handler.cc:
Global read lock functionality has been
moved into a class.
sql/lock.cc:
Global read lock functionality has been
moved into a class.
Updated code to use the new MDL API.
sql/mdl.cc:
Introduced new type-of-operation aware metadata locks.
To do this:
- Changed MDL_lock to use one list for waiting requests and one
list for granted requests. For each list, added a bitmap
that holds information what lock types a list contains.
Added a helper class MDL_lock::List to manipulate with granted
and waited lists while keeping the bitmaps in sync
with list contents.
- Changed lock-compatibility functions to use bitmaps that
define compatibility.
- Introduced a graph based deadlock detector inspired by
waiting_threads.c from Maria implementation.
- Now that we have a deadlock detector, and no longer have
a global lock to protect individual lock objects, but rather
use an rw lock per object, removed redundant code for upgrade,
and the global read lock. Changed the MDL API to
no longer require the caller to acquire the global
intention exclusive lock by means of a separate method.
Removed a few more methods that became redundant.
- Removed deadlock detection heuristic, it has been made
obsolete by the deadlock detector.
- With operation-type-aware metadata locks, MDL subsystem has
become aware of potential conflicts between DDL and open
transactions. This made it possible to remove calls to
mysql_abort_transactions_with_shared_lock() from acquisition
paths for exclusive lock and lock upgrade. Now we can simply
wait for these transactions to complete without fear of
deadlock. Function mysql_lock_abort() has also become
unnecessary for all conflicting cases except when a DDL
conflicts with a connection that has an open HANDLER.
sql/mdl.h:
Introduced new type-of-operation aware metadata locks.
Introduced a graph based deadlock detector and supporting
methods.
Added comments.
God rid of redundant API calls.
Renamed m_lt_or_ha_sentinel to m_trans_sentinel,
since now it guards the global read lock as well as
LOCK TABLES and HANDLER locks.
sql/mysql_priv.h:
Moved the global read lock functionality into a
class.
Added MYSQL_OPEN_FORCE_SHARED_MDL flag which forces
open_tables() to take MDL_SHARED on tables instead of
metadata locks specified in the parser. We use this to
allow PREPARE run concurrently in presence of
LOCK TABLES ... WRITE.
Added signature for find_table_for_mdl_ugprade().
sql/set_var.cc:
Global read lock functionality has been
moved into a class.
sql/sp_head.cc:
When creating TABLE_LIST elements for prelocking or
system tables set the type of request for metadata
lock according to the operation that will be performed
on the table.
sql/sql_base.cc:
- Updated code to use the new MDL API.
- In order to avoid locks starvation we take upgradable
locks all at once. As result implicitly locked tables no
longer get an upgradable lock. Consequently DDL and FLUSH
TABLES for such tables is prohibited.
find_write_locked_table() was replaced by
find_table_for_mdl_upgrade() function.
open_table() was adjusted to return TABLE instance with
upgradable ticket when necessary.
- We no longer wait for all locks on OT_WAIT back off
action -- only on the lock that caused the wait
conflict. Moreover, now we distinguish cases when we
have to wait due to conflict in MDL and old version
of table in TDC.
- Upate mysql_notify_threads_having_share_locks()
to only abort thr_lock.c waits of threads that
have open HANDLERs, since lock conflicts with only
these threads now can lead to deadlocks not detectable
by the MDL deadlock detector.
- Remove mysql_abort_transactions_with_shared_locks()
which is no longer needed.
sql/sql_class.cc:
Global read lock functionality has been moved into a class.
Re-arranged code in THD::cleanup() to simplify assert.
sql/sql_class.h:
Introduced class to incapsulate global read lock
functionality.
Now sentinel in MDL subsystem guards the global read lock
as well as LOCK TABLES and HANDLER locks. Adjusted code
accordingly.
sql/sql_db.cc:
Global read lock functionality has been moved into a class.
sql/sql_delete.cc:
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
TRUNCATE TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_handler.cc:
Inform MDL_context about presence of open HANDLERs.
Since HANLDERs break MDL protocol by acquiring table-level
lock while holding only S metadata lock on a table MDL
subsystem should take special care about such contexts (Now
this is the only case when mysql_lock_abort() is used).
sql/sql_parse.cc:
Global read lock functionality has been moved into a class.
Do not take upgradable metadata locks when opening tables
for CREATE TABLE SELECT as it is not necessary and limits
concurrency.
When initializing TABLE_LIST objects before adding them
to the table list set the type of request for metadata lock
according to the operation that will be performed on the
table.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result FLUSH
TABLES is no longer allowed for such tables.
sql/sql_prepare.cc:
Use MYSQL_OPEN_FORCE_SHARED_MDL flag when opening
tables during PREPARE. This allows PREPARE to run
concurrently in presence of LOCK TABLES ... WRITE.
sql/sql_rename.cc:
Global read lock functionality has been moved into a class.
sql/sql_show.cc:
Updated code to use the new MDL API.
sql/sql_table.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result DROP
TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_trigger.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
CREATE/DROP TRIGGER is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_view.cc:
Global read lock functionality has been moved into a class.
Fixed results of wrong merge that led to misuse of GLR API.
CREATE VIEW statement is not a commit statement.
sql/table.cc:
When resetting TABLE_LIST objects for PS or SP re-execution
set the type of request for metadata lock according to the
operation that will be performed on the table. Do the same
in auxiliary function initializing metadata lock requests
in a table list.
sql/table.h:
When initializing TABLE_LIST objects set the type of request
for metadata lock according to the operation that will be
performed on the table.
sql/transaction.cc:
Global read lock functionality has been moved into a class.
2010-02-01 12:43:06 +01:00
insert into t1 values (1), (2);
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
# --> connection default
begin;
select * from t1;
a
Implement new type-of-operation-aware metadata locks.
Add a wait-for graph based deadlock detector to the
MDL subsystem.
Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and
bug #37346 "innodb does not detect deadlock between update and
alter table".
The first bug manifested itself as an unwarranted abort of a
transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER
statement, when this transaction tried to repeat use of a
table, which it has already used in a similar fashion before
ALTER started.
The second bug showed up as a deadlock between table-level
locks and InnoDB row locks, which was "detected" only after
innodb_lock_wait_timeout timeout.
A transaction would start using the table and modify a few
rows.
Then ALTER TABLE would come in, and start copying rows
into a temporary table. Eventually it would stumble on
the modified records and get blocked on a row lock.
The first transaction would try to do more updates, and get
blocked on thr_lock.c lock.
This situation of circular wait would only get resolved
by a timeout.
Both these bugs stemmed from inadequate solutions to the
problem of deadlocks occurring between different
locking subsystems.
In the first case we tried to avoid deadlocks between metadata
locking and table-level locking subsystems, when upgrading shared
metadata lock to exclusive one.
Transactions holding the shared lock on the table and waiting for
some table-level lock used to be aborted too aggressively.
We also allowed ALTER TABLE to start in presence of transactions
that modify the subject table. ALTER TABLE acquires
TL_WRITE_ALLOW_READ lock at start, and that block all writes
against the table (naturally, we don't want any writes to be lost
when switching the old and the new table). TL_WRITE_ALLOW_READ
lock, in turn, would block the started transaction on thr_lock.c
lock, should they do more updates. This, again, lead to the need
to abort such transactions.
The second bug occurred simply because we didn't have any
mechanism to detect deadlocks between the table-level locks
in thr_lock.c and row-level locks in InnoDB, other than
innodb_lock_wait_timeout.
This patch solves both these problems by moving lock conflicts
which are causing these deadlocks into the metadata locking
subsystem, thus making it possible to avoid or detect such
deadlocks inside MDL.
To do this we introduce new type-of-operation-aware metadata
locks, which allow MDL subsystem to know not only the fact that
transaction has used or is going to use some object but also what
kind of operation it has carried out or going to carry out on the
object.
This, along with the addition of a special kind of upgradable
metadata lock, allows ALTER TABLE to wait until all
transactions which has updated the table to go away.
This solves the second issue.
Another special type of upgradable metadata lock is acquired
by LOCK TABLE WRITE. This second lock type allows to solve the
first issue, since abortion of table-level locks in event of
DDL under LOCK TABLES becomes also unnecessary.
Below follows the list of incompatible changes introduced by
this patch:
- From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those
statements that acquire TL_WRITE_ALLOW_READ lock)
wait for all transactions which has *updated* the table to
complete.
- From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE
(i.e. all statements which acquire TL_WRITE table-level lock) wait
for all transaction which *updated or read* from the table
to complete.
As a consequence, innodb_table_locks=0 option no longer applies
to LOCK TABLES ... WRITE.
- DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort
statements or transactions which use tables being dropped or
renamed, and instead wait for these transactions to complete.
- Since LOCK TABLES WRITE now takes a special metadata lock,
not compatible with with reads or writes against the subject table
and transaction-wide, thr_lock.c deadlock avoidance algorithm
that used to ensure absence of deadlocks between LOCK TABLES
WRITE and other statements is no longer sufficient, even for
MyISAM. The wait-for graph based deadlock detector of MDL
subsystem may sometimes be necessary and is involved. This may
lead to ER_LOCK_DEADLOCK error produced for multi-statement
transactions even if these only use MyISAM:
session 1: session 2:
begin;
update t1 ... lock table t2 write, t1 write;
-- gets a lock on t2, blocks on t1
update t2 ...
(ER_LOCK_DEADLOCK)
- Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE
was abandoned.
LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same
priority as the usual LOCK TABLE ... WRITE.
SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in
the wait queue.
- We do not take upgradable metadata locks on implicitly
locked tables. So if one has, say, a view v1 that uses
table t1, and issues:
LOCK TABLE v1 WRITE;
FLUSH TABLE t1; -- (or just 'FLUSH TABLES'),
an error is produced.
In order to be able to perform DDL on a table under LOCK TABLES,
the table must be locked explicitly in the LOCK TABLES list.
mysql-test/include/handler.inc:
Adjusted test case to trigger an execution path on which bug 41110
"crash with handler command when used concurrently with alter
table" and bug 41112 "crash in mysql_ha_close_table/get_lock_data
with alter table" were originally discovered. Left old test case
which no longer triggers this execution path for the sake of
coverage.
Added test coverage for HANDLER SQL statements and type-aware
metadata locks.
Added a test for the global shared lock and HANDLER SQL.
Updated tests to take into account that the old simple deadlock
detection heuristics was replaced with a graph-based deadlock
detector.
mysql-test/r/debug_sync.result:
Updated results (see debug_sync.test).
mysql-test/r/handler_innodb.result:
Updated results (see handler.inc test).
mysql-test/r/handler_myisam.result:
Updated results (see handler.inc test).
mysql-test/r/innodb-lock.result:
Updated results (see innodb-lock.test).
mysql-test/r/innodb_mysql_lock.result:
Updated results (see innodb_mysql_lock.test).
mysql-test/r/lock.result:
Updated results (see lock.test).
mysql-test/r/lock_multi.result:
Updated results (see lock_multi.test).
mysql-test/r/lock_sync.result:
Updated results (see lock_sync.test).
mysql-test/r/mdl_sync.result:
Updated results (see mdl_sync.test).
mysql-test/r/sp-threads.result:
SHOW PROCESSLIST output has changed due to the fact that waiting
for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/r/truncate_coverage.result:
Updated results (see truncate_coverage.test).
mysql-test/suite/funcs_1/datadict/processlist_val.inc:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/funcs_1/r/processlist_val_no_prot.result:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/rpl/t/rpl_sp.test:
Updated to a new SHOW PROCESSLIST state name.
mysql-test/t/debug_sync.test:
Use LOCK TABLES READ instead of LOCK TABLES WRITE as the latter
no longer allows to trigger execution path involving waiting on
thr_lock.c lock and therefore reaching debug sync-point covered
by this test.
mysql-test/t/innodb-lock.test:
Adjusted test case to the fact that innodb_table_locks=0 option is
no longer supported, since LOCK TABLES WRITE handles all its
conflicts within MDL subsystem.
mysql-test/t/innodb_mysql_lock.test:
Added test for bug #37346 "innodb does not detect deadlock between
update and alter table".
mysql-test/t/lock.test:
Added test coverage which checks the fact that we no longer support
DDL under LOCK TABLES on tables which were locked implicitly.
Adjusted existing test cases accordingly.
mysql-test/t/lock_multi.test:
Added test for bug #46272 "MySQL 5.4.4, new MDL: unnecessary
deadlock". Adjusted other test cases to take into account the
fact that waiting for LOCK TABLES ... WRITE now happens within MDL
subsystem.
mysql-test/t/lock_sync.test:
Since LOCK TABLES ... WRITE now takes SNRW metadata lock for
tables locked explicitly we have to implicitly lock InnoDB tables
(through view) to trigger the table-level lock conflict between
TL_WRITE and TL_WRITE_ALLOW_WRITE.
mysql-test/t/mdl_sync.test:
Added basic test coverage for type-of-operation-aware metadata
locks. Also covered with tests some use cases involving HANDLER
statements in which a deadlock could arise.
Adjusted existing tests to take type-of-operation-aware MDL into
account.
mysql-test/t/multi_update.test:
Update to a new SHOW PROCESSLIST state name.
mysql-test/t/truncate_coverage.test:
Adjusted test case after making LOCK TABLES WRITE to wait until
transactions that use the table to be locked are completed.
Updated to the changed name of DEBUG_SYNC point.
sql/handler.cc:
Global read lock functionality has been
moved into a class.
sql/lock.cc:
Global read lock functionality has been
moved into a class.
Updated code to use the new MDL API.
sql/mdl.cc:
Introduced new type-of-operation aware metadata locks.
To do this:
- Changed MDL_lock to use one list for waiting requests and one
list for granted requests. For each list, added a bitmap
that holds information what lock types a list contains.
Added a helper class MDL_lock::List to manipulate with granted
and waited lists while keeping the bitmaps in sync
with list contents.
- Changed lock-compatibility functions to use bitmaps that
define compatibility.
- Introduced a graph based deadlock detector inspired by
waiting_threads.c from Maria implementation.
- Now that we have a deadlock detector, and no longer have
a global lock to protect individual lock objects, but rather
use an rw lock per object, removed redundant code for upgrade,
and the global read lock. Changed the MDL API to
no longer require the caller to acquire the global
intention exclusive lock by means of a separate method.
Removed a few more methods that became redundant.
- Removed deadlock detection heuristic, it has been made
obsolete by the deadlock detector.
- With operation-type-aware metadata locks, MDL subsystem has
become aware of potential conflicts between DDL and open
transactions. This made it possible to remove calls to
mysql_abort_transactions_with_shared_lock() from acquisition
paths for exclusive lock and lock upgrade. Now we can simply
wait for these transactions to complete without fear of
deadlock. Function mysql_lock_abort() has also become
unnecessary for all conflicting cases except when a DDL
conflicts with a connection that has an open HANDLER.
sql/mdl.h:
Introduced new type-of-operation aware metadata locks.
Introduced a graph based deadlock detector and supporting
methods.
Added comments.
God rid of redundant API calls.
Renamed m_lt_or_ha_sentinel to m_trans_sentinel,
since now it guards the global read lock as well as
LOCK TABLES and HANDLER locks.
sql/mysql_priv.h:
Moved the global read lock functionality into a
class.
Added MYSQL_OPEN_FORCE_SHARED_MDL flag which forces
open_tables() to take MDL_SHARED on tables instead of
metadata locks specified in the parser. We use this to
allow PREPARE run concurrently in presence of
LOCK TABLES ... WRITE.
Added signature for find_table_for_mdl_ugprade().
sql/set_var.cc:
Global read lock functionality has been
moved into a class.
sql/sp_head.cc:
When creating TABLE_LIST elements for prelocking or
system tables set the type of request for metadata
lock according to the operation that will be performed
on the table.
sql/sql_base.cc:
- Updated code to use the new MDL API.
- In order to avoid locks starvation we take upgradable
locks all at once. As result implicitly locked tables no
longer get an upgradable lock. Consequently DDL and FLUSH
TABLES for such tables is prohibited.
find_write_locked_table() was replaced by
find_table_for_mdl_upgrade() function.
open_table() was adjusted to return TABLE instance with
upgradable ticket when necessary.
- We no longer wait for all locks on OT_WAIT back off
action -- only on the lock that caused the wait
conflict. Moreover, now we distinguish cases when we
have to wait due to conflict in MDL and old version
of table in TDC.
- Upate mysql_notify_threads_having_share_locks()
to only abort thr_lock.c waits of threads that
have open HANDLERs, since lock conflicts with only
these threads now can lead to deadlocks not detectable
by the MDL deadlock detector.
- Remove mysql_abort_transactions_with_shared_locks()
which is no longer needed.
sql/sql_class.cc:
Global read lock functionality has been moved into a class.
Re-arranged code in THD::cleanup() to simplify assert.
sql/sql_class.h:
Introduced class to incapsulate global read lock
functionality.
Now sentinel in MDL subsystem guards the global read lock
as well as LOCK TABLES and HANDLER locks. Adjusted code
accordingly.
sql/sql_db.cc:
Global read lock functionality has been moved into a class.
sql/sql_delete.cc:
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
TRUNCATE TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_handler.cc:
Inform MDL_context about presence of open HANDLERs.
Since HANLDERs break MDL protocol by acquiring table-level
lock while holding only S metadata lock on a table MDL
subsystem should take special care about such contexts (Now
this is the only case when mysql_lock_abort() is used).
sql/sql_parse.cc:
Global read lock functionality has been moved into a class.
Do not take upgradable metadata locks when opening tables
for CREATE TABLE SELECT as it is not necessary and limits
concurrency.
When initializing TABLE_LIST objects before adding them
to the table list set the type of request for metadata lock
according to the operation that will be performed on the
table.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result FLUSH
TABLES is no longer allowed for such tables.
sql/sql_prepare.cc:
Use MYSQL_OPEN_FORCE_SHARED_MDL flag when opening
tables during PREPARE. This allows PREPARE to run
concurrently in presence of LOCK TABLES ... WRITE.
sql/sql_rename.cc:
Global read lock functionality has been moved into a class.
sql/sql_show.cc:
Updated code to use the new MDL API.
sql/sql_table.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result DROP
TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_trigger.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
CREATE/DROP TRIGGER is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_view.cc:
Global read lock functionality has been moved into a class.
Fixed results of wrong merge that led to misuse of GLR API.
CREATE VIEW statement is not a commit statement.
sql/table.cc:
When resetting TABLE_LIST objects for PS or SP re-execution
set the type of request for metadata lock according to the
operation that will be performed on the table. Do the same
in auxiliary function initializing metadata lock requests
in a table list.
sql/table.h:
When initializing TABLE_LIST objects set the type of request
for metadata lock according to the operation that will be
performed on the table.
sql/transaction.cc:
Global read lock functionality has been moved into a class.
2010-02-01 12:43:06 +01:00
1
2
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
handler t1 open;
# --> connection con1
Implement new type-of-operation-aware metadata locks.
Add a wait-for graph based deadlock detector to the
MDL subsystem.
Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and
bug #37346 "innodb does not detect deadlock between update and
alter table".
The first bug manifested itself as an unwarranted abort of a
transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER
statement, when this transaction tried to repeat use of a
table, which it has already used in a similar fashion before
ALTER started.
The second bug showed up as a deadlock between table-level
locks and InnoDB row locks, which was "detected" only after
innodb_lock_wait_timeout timeout.
A transaction would start using the table and modify a few
rows.
Then ALTER TABLE would come in, and start copying rows
into a temporary table. Eventually it would stumble on
the modified records and get blocked on a row lock.
The first transaction would try to do more updates, and get
blocked on thr_lock.c lock.
This situation of circular wait would only get resolved
by a timeout.
Both these bugs stemmed from inadequate solutions to the
problem of deadlocks occurring between different
locking subsystems.
In the first case we tried to avoid deadlocks between metadata
locking and table-level locking subsystems, when upgrading shared
metadata lock to exclusive one.
Transactions holding the shared lock on the table and waiting for
some table-level lock used to be aborted too aggressively.
We also allowed ALTER TABLE to start in presence of transactions
that modify the subject table. ALTER TABLE acquires
TL_WRITE_ALLOW_READ lock at start, and that block all writes
against the table (naturally, we don't want any writes to be lost
when switching the old and the new table). TL_WRITE_ALLOW_READ
lock, in turn, would block the started transaction on thr_lock.c
lock, should they do more updates. This, again, lead to the need
to abort such transactions.
The second bug occurred simply because we didn't have any
mechanism to detect deadlocks between the table-level locks
in thr_lock.c and row-level locks in InnoDB, other than
innodb_lock_wait_timeout.
This patch solves both these problems by moving lock conflicts
which are causing these deadlocks into the metadata locking
subsystem, thus making it possible to avoid or detect such
deadlocks inside MDL.
To do this we introduce new type-of-operation-aware metadata
locks, which allow MDL subsystem to know not only the fact that
transaction has used or is going to use some object but also what
kind of operation it has carried out or going to carry out on the
object.
This, along with the addition of a special kind of upgradable
metadata lock, allows ALTER TABLE to wait until all
transactions which has updated the table to go away.
This solves the second issue.
Another special type of upgradable metadata lock is acquired
by LOCK TABLE WRITE. This second lock type allows to solve the
first issue, since abortion of table-level locks in event of
DDL under LOCK TABLES becomes also unnecessary.
Below follows the list of incompatible changes introduced by
this patch:
- From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those
statements that acquire TL_WRITE_ALLOW_READ lock)
wait for all transactions which has *updated* the table to
complete.
- From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE
(i.e. all statements which acquire TL_WRITE table-level lock) wait
for all transaction which *updated or read* from the table
to complete.
As a consequence, innodb_table_locks=0 option no longer applies
to LOCK TABLES ... WRITE.
- DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort
statements or transactions which use tables being dropped or
renamed, and instead wait for these transactions to complete.
- Since LOCK TABLES WRITE now takes a special metadata lock,
not compatible with with reads or writes against the subject table
and transaction-wide, thr_lock.c deadlock avoidance algorithm
that used to ensure absence of deadlocks between LOCK TABLES
WRITE and other statements is no longer sufficient, even for
MyISAM. The wait-for graph based deadlock detector of MDL
subsystem may sometimes be necessary and is involved. This may
lead to ER_LOCK_DEADLOCK error produced for multi-statement
transactions even if these only use MyISAM:
session 1: session 2:
begin;
update t1 ... lock table t2 write, t1 write;
-- gets a lock on t2, blocks on t1
update t2 ...
(ER_LOCK_DEADLOCK)
- Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE
was abandoned.
LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same
priority as the usual LOCK TABLE ... WRITE.
SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in
the wait queue.
- We do not take upgradable metadata locks on implicitly
locked tables. So if one has, say, a view v1 that uses
table t1, and issues:
LOCK TABLE v1 WRITE;
FLUSH TABLE t1; -- (or just 'FLUSH TABLES'),
an error is produced.
In order to be able to perform DDL on a table under LOCK TABLES,
the table must be locked explicitly in the LOCK TABLES list.
mysql-test/include/handler.inc:
Adjusted test case to trigger an execution path on which bug 41110
"crash with handler command when used concurrently with alter
table" and bug 41112 "crash in mysql_ha_close_table/get_lock_data
with alter table" were originally discovered. Left old test case
which no longer triggers this execution path for the sake of
coverage.
Added test coverage for HANDLER SQL statements and type-aware
metadata locks.
Added a test for the global shared lock and HANDLER SQL.
Updated tests to take into account that the old simple deadlock
detection heuristics was replaced with a graph-based deadlock
detector.
mysql-test/r/debug_sync.result:
Updated results (see debug_sync.test).
mysql-test/r/handler_innodb.result:
Updated results (see handler.inc test).
mysql-test/r/handler_myisam.result:
Updated results (see handler.inc test).
mysql-test/r/innodb-lock.result:
Updated results (see innodb-lock.test).
mysql-test/r/innodb_mysql_lock.result:
Updated results (see innodb_mysql_lock.test).
mysql-test/r/lock.result:
Updated results (see lock.test).
mysql-test/r/lock_multi.result:
Updated results (see lock_multi.test).
mysql-test/r/lock_sync.result:
Updated results (see lock_sync.test).
mysql-test/r/mdl_sync.result:
Updated results (see mdl_sync.test).
mysql-test/r/sp-threads.result:
SHOW PROCESSLIST output has changed due to the fact that waiting
for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/r/truncate_coverage.result:
Updated results (see truncate_coverage.test).
mysql-test/suite/funcs_1/datadict/processlist_val.inc:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/funcs_1/r/processlist_val_no_prot.result:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/rpl/t/rpl_sp.test:
Updated to a new SHOW PROCESSLIST state name.
mysql-test/t/debug_sync.test:
Use LOCK TABLES READ instead of LOCK TABLES WRITE as the latter
no longer allows to trigger execution path involving waiting on
thr_lock.c lock and therefore reaching debug sync-point covered
by this test.
mysql-test/t/innodb-lock.test:
Adjusted test case to the fact that innodb_table_locks=0 option is
no longer supported, since LOCK TABLES WRITE handles all its
conflicts within MDL subsystem.
mysql-test/t/innodb_mysql_lock.test:
Added test for bug #37346 "innodb does not detect deadlock between
update and alter table".
mysql-test/t/lock.test:
Added test coverage which checks the fact that we no longer support
DDL under LOCK TABLES on tables which were locked implicitly.
Adjusted existing test cases accordingly.
mysql-test/t/lock_multi.test:
Added test for bug #46272 "MySQL 5.4.4, new MDL: unnecessary
deadlock". Adjusted other test cases to take into account the
fact that waiting for LOCK TABLES ... WRITE now happens within MDL
subsystem.
mysql-test/t/lock_sync.test:
Since LOCK TABLES ... WRITE now takes SNRW metadata lock for
tables locked explicitly we have to implicitly lock InnoDB tables
(through view) to trigger the table-level lock conflict between
TL_WRITE and TL_WRITE_ALLOW_WRITE.
mysql-test/t/mdl_sync.test:
Added basic test coverage for type-of-operation-aware metadata
locks. Also covered with tests some use cases involving HANDLER
statements in which a deadlock could arise.
Adjusted existing tests to take type-of-operation-aware MDL into
account.
mysql-test/t/multi_update.test:
Update to a new SHOW PROCESSLIST state name.
mysql-test/t/truncate_coverage.test:
Adjusted test case after making LOCK TABLES WRITE to wait until
transactions that use the table to be locked are completed.
Updated to the changed name of DEBUG_SYNC point.
sql/handler.cc:
Global read lock functionality has been
moved into a class.
sql/lock.cc:
Global read lock functionality has been
moved into a class.
Updated code to use the new MDL API.
sql/mdl.cc:
Introduced new type-of-operation aware metadata locks.
To do this:
- Changed MDL_lock to use one list for waiting requests and one
list for granted requests. For each list, added a bitmap
that holds information what lock types a list contains.
Added a helper class MDL_lock::List to manipulate with granted
and waited lists while keeping the bitmaps in sync
with list contents.
- Changed lock-compatibility functions to use bitmaps that
define compatibility.
- Introduced a graph based deadlock detector inspired by
waiting_threads.c from Maria implementation.
- Now that we have a deadlock detector, and no longer have
a global lock to protect individual lock objects, but rather
use an rw lock per object, removed redundant code for upgrade,
and the global read lock. Changed the MDL API to
no longer require the caller to acquire the global
intention exclusive lock by means of a separate method.
Removed a few more methods that became redundant.
- Removed deadlock detection heuristic, it has been made
obsolete by the deadlock detector.
- With operation-type-aware metadata locks, MDL subsystem has
become aware of potential conflicts between DDL and open
transactions. This made it possible to remove calls to
mysql_abort_transactions_with_shared_lock() from acquisition
paths for exclusive lock and lock upgrade. Now we can simply
wait for these transactions to complete without fear of
deadlock. Function mysql_lock_abort() has also become
unnecessary for all conflicting cases except when a DDL
conflicts with a connection that has an open HANDLER.
sql/mdl.h:
Introduced new type-of-operation aware metadata locks.
Introduced a graph based deadlock detector and supporting
methods.
Added comments.
God rid of redundant API calls.
Renamed m_lt_or_ha_sentinel to m_trans_sentinel,
since now it guards the global read lock as well as
LOCK TABLES and HANDLER locks.
sql/mysql_priv.h:
Moved the global read lock functionality into a
class.
Added MYSQL_OPEN_FORCE_SHARED_MDL flag which forces
open_tables() to take MDL_SHARED on tables instead of
metadata locks specified in the parser. We use this to
allow PREPARE run concurrently in presence of
LOCK TABLES ... WRITE.
Added signature for find_table_for_mdl_ugprade().
sql/set_var.cc:
Global read lock functionality has been
moved into a class.
sql/sp_head.cc:
When creating TABLE_LIST elements for prelocking or
system tables set the type of request for metadata
lock according to the operation that will be performed
on the table.
sql/sql_base.cc:
- Updated code to use the new MDL API.
- In order to avoid locks starvation we take upgradable
locks all at once. As result implicitly locked tables no
longer get an upgradable lock. Consequently DDL and FLUSH
TABLES for such tables is prohibited.
find_write_locked_table() was replaced by
find_table_for_mdl_upgrade() function.
open_table() was adjusted to return TABLE instance with
upgradable ticket when necessary.
- We no longer wait for all locks on OT_WAIT back off
action -- only on the lock that caused the wait
conflict. Moreover, now we distinguish cases when we
have to wait due to conflict in MDL and old version
of table in TDC.
- Upate mysql_notify_threads_having_share_locks()
to only abort thr_lock.c waits of threads that
have open HANDLERs, since lock conflicts with only
these threads now can lead to deadlocks not detectable
by the MDL deadlock detector.
- Remove mysql_abort_transactions_with_shared_locks()
which is no longer needed.
sql/sql_class.cc:
Global read lock functionality has been moved into a class.
Re-arranged code in THD::cleanup() to simplify assert.
sql/sql_class.h:
Introduced class to incapsulate global read lock
functionality.
Now sentinel in MDL subsystem guards the global read lock
as well as LOCK TABLES and HANDLER locks. Adjusted code
accordingly.
sql/sql_db.cc:
Global read lock functionality has been moved into a class.
sql/sql_delete.cc:
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
TRUNCATE TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_handler.cc:
Inform MDL_context about presence of open HANDLERs.
Since HANLDERs break MDL protocol by acquiring table-level
lock while holding only S metadata lock on a table MDL
subsystem should take special care about such contexts (Now
this is the only case when mysql_lock_abort() is used).
sql/sql_parse.cc:
Global read lock functionality has been moved into a class.
Do not take upgradable metadata locks when opening tables
for CREATE TABLE SELECT as it is not necessary and limits
concurrency.
When initializing TABLE_LIST objects before adding them
to the table list set the type of request for metadata lock
according to the operation that will be performed on the
table.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result FLUSH
TABLES is no longer allowed for such tables.
sql/sql_prepare.cc:
Use MYSQL_OPEN_FORCE_SHARED_MDL flag when opening
tables during PREPARE. This allows PREPARE to run
concurrently in presence of LOCK TABLES ... WRITE.
sql/sql_rename.cc:
Global read lock functionality has been moved into a class.
sql/sql_show.cc:
Updated code to use the new MDL API.
sql/sql_table.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result DROP
TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_trigger.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
CREATE/DROP TRIGGER is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_view.cc:
Global read lock functionality has been moved into a class.
Fixed results of wrong merge that led to misuse of GLR API.
CREATE VIEW statement is not a commit statement.
sql/table.cc:
When resetting TABLE_LIST objects for PS or SP re-execution
set the type of request for metadata lock according to the
operation that will be performed on the table. Do the same
in auxiliary function initializing metadata lock requests
in a table list.
sql/table.h:
When initializing TABLE_LIST objects set the type of request
for metadata lock according to the operation that will be
performed on the table.
sql/transaction.cc:
Global read lock functionality has been moved into a class.
2010-02-01 12:43:06 +01:00
# Sending:
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
lock tables t1 write;
Implement new type-of-operation-aware metadata locks.
Add a wait-for graph based deadlock detector to the
MDL subsystem.
Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and
bug #37346 "innodb does not detect deadlock between update and
alter table".
The first bug manifested itself as an unwarranted abort of a
transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER
statement, when this transaction tried to repeat use of a
table, which it has already used in a similar fashion before
ALTER started.
The second bug showed up as a deadlock between table-level
locks and InnoDB row locks, which was "detected" only after
innodb_lock_wait_timeout timeout.
A transaction would start using the table and modify a few
rows.
Then ALTER TABLE would come in, and start copying rows
into a temporary table. Eventually it would stumble on
the modified records and get blocked on a row lock.
The first transaction would try to do more updates, and get
blocked on thr_lock.c lock.
This situation of circular wait would only get resolved
by a timeout.
Both these bugs stemmed from inadequate solutions to the
problem of deadlocks occurring between different
locking subsystems.
In the first case we tried to avoid deadlocks between metadata
locking and table-level locking subsystems, when upgrading shared
metadata lock to exclusive one.
Transactions holding the shared lock on the table and waiting for
some table-level lock used to be aborted too aggressively.
We also allowed ALTER TABLE to start in presence of transactions
that modify the subject table. ALTER TABLE acquires
TL_WRITE_ALLOW_READ lock at start, and that block all writes
against the table (naturally, we don't want any writes to be lost
when switching the old and the new table). TL_WRITE_ALLOW_READ
lock, in turn, would block the started transaction on thr_lock.c
lock, should they do more updates. This, again, lead to the need
to abort such transactions.
The second bug occurred simply because we didn't have any
mechanism to detect deadlocks between the table-level locks
in thr_lock.c and row-level locks in InnoDB, other than
innodb_lock_wait_timeout.
This patch solves both these problems by moving lock conflicts
which are causing these deadlocks into the metadata locking
subsystem, thus making it possible to avoid or detect such
deadlocks inside MDL.
To do this we introduce new type-of-operation-aware metadata
locks, which allow MDL subsystem to know not only the fact that
transaction has used or is going to use some object but also what
kind of operation it has carried out or going to carry out on the
object.
This, along with the addition of a special kind of upgradable
metadata lock, allows ALTER TABLE to wait until all
transactions which has updated the table to go away.
This solves the second issue.
Another special type of upgradable metadata lock is acquired
by LOCK TABLE WRITE. This second lock type allows to solve the
first issue, since abortion of table-level locks in event of
DDL under LOCK TABLES becomes also unnecessary.
Below follows the list of incompatible changes introduced by
this patch:
- From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those
statements that acquire TL_WRITE_ALLOW_READ lock)
wait for all transactions which has *updated* the table to
complete.
- From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE
(i.e. all statements which acquire TL_WRITE table-level lock) wait
for all transaction which *updated or read* from the table
to complete.
As a consequence, innodb_table_locks=0 option no longer applies
to LOCK TABLES ... WRITE.
- DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort
statements or transactions which use tables being dropped or
renamed, and instead wait for these transactions to complete.
- Since LOCK TABLES WRITE now takes a special metadata lock,
not compatible with with reads or writes against the subject table
and transaction-wide, thr_lock.c deadlock avoidance algorithm
that used to ensure absence of deadlocks between LOCK TABLES
WRITE and other statements is no longer sufficient, even for
MyISAM. The wait-for graph based deadlock detector of MDL
subsystem may sometimes be necessary and is involved. This may
lead to ER_LOCK_DEADLOCK error produced for multi-statement
transactions even if these only use MyISAM:
session 1: session 2:
begin;
update t1 ... lock table t2 write, t1 write;
-- gets a lock on t2, blocks on t1
update t2 ...
(ER_LOCK_DEADLOCK)
- Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE
was abandoned.
LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same
priority as the usual LOCK TABLE ... WRITE.
SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in
the wait queue.
- We do not take upgradable metadata locks on implicitly
locked tables. So if one has, say, a view v1 that uses
table t1, and issues:
LOCK TABLE v1 WRITE;
FLUSH TABLE t1; -- (or just 'FLUSH TABLES'),
an error is produced.
In order to be able to perform DDL on a table under LOCK TABLES,
the table must be locked explicitly in the LOCK TABLES list.
mysql-test/include/handler.inc:
Adjusted test case to trigger an execution path on which bug 41110
"crash with handler command when used concurrently with alter
table" and bug 41112 "crash in mysql_ha_close_table/get_lock_data
with alter table" were originally discovered. Left old test case
which no longer triggers this execution path for the sake of
coverage.
Added test coverage for HANDLER SQL statements and type-aware
metadata locks.
Added a test for the global shared lock and HANDLER SQL.
Updated tests to take into account that the old simple deadlock
detection heuristics was replaced with a graph-based deadlock
detector.
mysql-test/r/debug_sync.result:
Updated results (see debug_sync.test).
mysql-test/r/handler_innodb.result:
Updated results (see handler.inc test).
mysql-test/r/handler_myisam.result:
Updated results (see handler.inc test).
mysql-test/r/innodb-lock.result:
Updated results (see innodb-lock.test).
mysql-test/r/innodb_mysql_lock.result:
Updated results (see innodb_mysql_lock.test).
mysql-test/r/lock.result:
Updated results (see lock.test).
mysql-test/r/lock_multi.result:
Updated results (see lock_multi.test).
mysql-test/r/lock_sync.result:
Updated results (see lock_sync.test).
mysql-test/r/mdl_sync.result:
Updated results (see mdl_sync.test).
mysql-test/r/sp-threads.result:
SHOW PROCESSLIST output has changed due to the fact that waiting
for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/r/truncate_coverage.result:
Updated results (see truncate_coverage.test).
mysql-test/suite/funcs_1/datadict/processlist_val.inc:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/funcs_1/r/processlist_val_no_prot.result:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/rpl/t/rpl_sp.test:
Updated to a new SHOW PROCESSLIST state name.
mysql-test/t/debug_sync.test:
Use LOCK TABLES READ instead of LOCK TABLES WRITE as the latter
no longer allows to trigger execution path involving waiting on
thr_lock.c lock and therefore reaching debug sync-point covered
by this test.
mysql-test/t/innodb-lock.test:
Adjusted test case to the fact that innodb_table_locks=0 option is
no longer supported, since LOCK TABLES WRITE handles all its
conflicts within MDL subsystem.
mysql-test/t/innodb_mysql_lock.test:
Added test for bug #37346 "innodb does not detect deadlock between
update and alter table".
mysql-test/t/lock.test:
Added test coverage which checks the fact that we no longer support
DDL under LOCK TABLES on tables which were locked implicitly.
Adjusted existing test cases accordingly.
mysql-test/t/lock_multi.test:
Added test for bug #46272 "MySQL 5.4.4, new MDL: unnecessary
deadlock". Adjusted other test cases to take into account the
fact that waiting for LOCK TABLES ... WRITE now happens within MDL
subsystem.
mysql-test/t/lock_sync.test:
Since LOCK TABLES ... WRITE now takes SNRW metadata lock for
tables locked explicitly we have to implicitly lock InnoDB tables
(through view) to trigger the table-level lock conflict between
TL_WRITE and TL_WRITE_ALLOW_WRITE.
mysql-test/t/mdl_sync.test:
Added basic test coverage for type-of-operation-aware metadata
locks. Also covered with tests some use cases involving HANDLER
statements in which a deadlock could arise.
Adjusted existing tests to take type-of-operation-aware MDL into
account.
mysql-test/t/multi_update.test:
Update to a new SHOW PROCESSLIST state name.
mysql-test/t/truncate_coverage.test:
Adjusted test case after making LOCK TABLES WRITE to wait until
transactions that use the table to be locked are completed.
Updated to the changed name of DEBUG_SYNC point.
sql/handler.cc:
Global read lock functionality has been
moved into a class.
sql/lock.cc:
Global read lock functionality has been
moved into a class.
Updated code to use the new MDL API.
sql/mdl.cc:
Introduced new type-of-operation aware metadata locks.
To do this:
- Changed MDL_lock to use one list for waiting requests and one
list for granted requests. For each list, added a bitmap
that holds information what lock types a list contains.
Added a helper class MDL_lock::List to manipulate with granted
and waited lists while keeping the bitmaps in sync
with list contents.
- Changed lock-compatibility functions to use bitmaps that
define compatibility.
- Introduced a graph based deadlock detector inspired by
waiting_threads.c from Maria implementation.
- Now that we have a deadlock detector, and no longer have
a global lock to protect individual lock objects, but rather
use an rw lock per object, removed redundant code for upgrade,
and the global read lock. Changed the MDL API to
no longer require the caller to acquire the global
intention exclusive lock by means of a separate method.
Removed a few more methods that became redundant.
- Removed deadlock detection heuristic, it has been made
obsolete by the deadlock detector.
- With operation-type-aware metadata locks, MDL subsystem has
become aware of potential conflicts between DDL and open
transactions. This made it possible to remove calls to
mysql_abort_transactions_with_shared_lock() from acquisition
paths for exclusive lock and lock upgrade. Now we can simply
wait for these transactions to complete without fear of
deadlock. Function mysql_lock_abort() has also become
unnecessary for all conflicting cases except when a DDL
conflicts with a connection that has an open HANDLER.
sql/mdl.h:
Introduced new type-of-operation aware metadata locks.
Introduced a graph based deadlock detector and supporting
methods.
Added comments.
God rid of redundant API calls.
Renamed m_lt_or_ha_sentinel to m_trans_sentinel,
since now it guards the global read lock as well as
LOCK TABLES and HANDLER locks.
sql/mysql_priv.h:
Moved the global read lock functionality into a
class.
Added MYSQL_OPEN_FORCE_SHARED_MDL flag which forces
open_tables() to take MDL_SHARED on tables instead of
metadata locks specified in the parser. We use this to
allow PREPARE run concurrently in presence of
LOCK TABLES ... WRITE.
Added signature for find_table_for_mdl_ugprade().
sql/set_var.cc:
Global read lock functionality has been
moved into a class.
sql/sp_head.cc:
When creating TABLE_LIST elements for prelocking or
system tables set the type of request for metadata
lock according to the operation that will be performed
on the table.
sql/sql_base.cc:
- Updated code to use the new MDL API.
- In order to avoid locks starvation we take upgradable
locks all at once. As result implicitly locked tables no
longer get an upgradable lock. Consequently DDL and FLUSH
TABLES for such tables is prohibited.
find_write_locked_table() was replaced by
find_table_for_mdl_upgrade() function.
open_table() was adjusted to return TABLE instance with
upgradable ticket when necessary.
- We no longer wait for all locks on OT_WAIT back off
action -- only on the lock that caused the wait
conflict. Moreover, now we distinguish cases when we
have to wait due to conflict in MDL and old version
of table in TDC.
- Upate mysql_notify_threads_having_share_locks()
to only abort thr_lock.c waits of threads that
have open HANDLERs, since lock conflicts with only
these threads now can lead to deadlocks not detectable
by the MDL deadlock detector.
- Remove mysql_abort_transactions_with_shared_locks()
which is no longer needed.
sql/sql_class.cc:
Global read lock functionality has been moved into a class.
Re-arranged code in THD::cleanup() to simplify assert.
sql/sql_class.h:
Introduced class to incapsulate global read lock
functionality.
Now sentinel in MDL subsystem guards the global read lock
as well as LOCK TABLES and HANDLER locks. Adjusted code
accordingly.
sql/sql_db.cc:
Global read lock functionality has been moved into a class.
sql/sql_delete.cc:
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
TRUNCATE TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_handler.cc:
Inform MDL_context about presence of open HANDLERs.
Since HANLDERs break MDL protocol by acquiring table-level
lock while holding only S metadata lock on a table MDL
subsystem should take special care about such contexts (Now
this is the only case when mysql_lock_abort() is used).
sql/sql_parse.cc:
Global read lock functionality has been moved into a class.
Do not take upgradable metadata locks when opening tables
for CREATE TABLE SELECT as it is not necessary and limits
concurrency.
When initializing TABLE_LIST objects before adding them
to the table list set the type of request for metadata lock
according to the operation that will be performed on the
table.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result FLUSH
TABLES is no longer allowed for such tables.
sql/sql_prepare.cc:
Use MYSQL_OPEN_FORCE_SHARED_MDL flag when opening
tables during PREPARE. This allows PREPARE to run
concurrently in presence of LOCK TABLES ... WRITE.
sql/sql_rename.cc:
Global read lock functionality has been moved into a class.
sql/sql_show.cc:
Updated code to use the new MDL API.
sql/sql_table.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result DROP
TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_trigger.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
CREATE/DROP TRIGGER is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_view.cc:
Global read lock functionality has been moved into a class.
Fixed results of wrong merge that led to misuse of GLR API.
CREATE VIEW statement is not a commit statement.
sql/table.cc:
When resetting TABLE_LIST objects for PS or SP re-execution
set the type of request for metadata lock according to the
operation that will be performed on the table. Do the same
in auxiliary function initializing metadata lock requests
in a table list.
sql/table.h:
When initializing TABLE_LIST objects set the type of request
for metadata lock according to the operation that will be
performed on the table.
sql/transaction.cc:
Global read lock functionality has been moved into a class.
2010-02-01 12:43:06 +01:00
# --> connection con2
# Check that 'lock tables t1 write' waits until transaction which
# has read from the table commits.
# --> connection default
# The below 'handler t1 read ...' should not be blocked as
# 'lock tables t1 write' has not succeeded yet.
handler t1 read a next;
a
1
# Unblock 'lock tables t1 write'.
commit;
# --> connection con1
# Reap 'lock tables t1 write'.
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
# --> connection default
# Sending:
handler t1 read a next;
# --> connection con1
# Waiting for 'handler t1 read a next' to get blocked...
Implement new type-of-operation-aware metadata locks.
Add a wait-for graph based deadlock detector to the
MDL subsystem.
Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and
bug #37346 "innodb does not detect deadlock between update and
alter table".
The first bug manifested itself as an unwarranted abort of a
transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER
statement, when this transaction tried to repeat use of a
table, which it has already used in a similar fashion before
ALTER started.
The second bug showed up as a deadlock between table-level
locks and InnoDB row locks, which was "detected" only after
innodb_lock_wait_timeout timeout.
A transaction would start using the table and modify a few
rows.
Then ALTER TABLE would come in, and start copying rows
into a temporary table. Eventually it would stumble on
the modified records and get blocked on a row lock.
The first transaction would try to do more updates, and get
blocked on thr_lock.c lock.
This situation of circular wait would only get resolved
by a timeout.
Both these bugs stemmed from inadequate solutions to the
problem of deadlocks occurring between different
locking subsystems.
In the first case we tried to avoid deadlocks between metadata
locking and table-level locking subsystems, when upgrading shared
metadata lock to exclusive one.
Transactions holding the shared lock on the table and waiting for
some table-level lock used to be aborted too aggressively.
We also allowed ALTER TABLE to start in presence of transactions
that modify the subject table. ALTER TABLE acquires
TL_WRITE_ALLOW_READ lock at start, and that block all writes
against the table (naturally, we don't want any writes to be lost
when switching the old and the new table). TL_WRITE_ALLOW_READ
lock, in turn, would block the started transaction on thr_lock.c
lock, should they do more updates. This, again, lead to the need
to abort such transactions.
The second bug occurred simply because we didn't have any
mechanism to detect deadlocks between the table-level locks
in thr_lock.c and row-level locks in InnoDB, other than
innodb_lock_wait_timeout.
This patch solves both these problems by moving lock conflicts
which are causing these deadlocks into the metadata locking
subsystem, thus making it possible to avoid or detect such
deadlocks inside MDL.
To do this we introduce new type-of-operation-aware metadata
locks, which allow MDL subsystem to know not only the fact that
transaction has used or is going to use some object but also what
kind of operation it has carried out or going to carry out on the
object.
This, along with the addition of a special kind of upgradable
metadata lock, allows ALTER TABLE to wait until all
transactions which has updated the table to go away.
This solves the second issue.
Another special type of upgradable metadata lock is acquired
by LOCK TABLE WRITE. This second lock type allows to solve the
first issue, since abortion of table-level locks in event of
DDL under LOCK TABLES becomes also unnecessary.
Below follows the list of incompatible changes introduced by
this patch:
- From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those
statements that acquire TL_WRITE_ALLOW_READ lock)
wait for all transactions which has *updated* the table to
complete.
- From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE
(i.e. all statements which acquire TL_WRITE table-level lock) wait
for all transaction which *updated or read* from the table
to complete.
As a consequence, innodb_table_locks=0 option no longer applies
to LOCK TABLES ... WRITE.
- DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort
statements or transactions which use tables being dropped or
renamed, and instead wait for these transactions to complete.
- Since LOCK TABLES WRITE now takes a special metadata lock,
not compatible with with reads or writes against the subject table
and transaction-wide, thr_lock.c deadlock avoidance algorithm
that used to ensure absence of deadlocks between LOCK TABLES
WRITE and other statements is no longer sufficient, even for
MyISAM. The wait-for graph based deadlock detector of MDL
subsystem may sometimes be necessary and is involved. This may
lead to ER_LOCK_DEADLOCK error produced for multi-statement
transactions even if these only use MyISAM:
session 1: session 2:
begin;
update t1 ... lock table t2 write, t1 write;
-- gets a lock on t2, blocks on t1
update t2 ...
(ER_LOCK_DEADLOCK)
- Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE
was abandoned.
LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same
priority as the usual LOCK TABLE ... WRITE.
SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in
the wait queue.
- We do not take upgradable metadata locks on implicitly
locked tables. So if one has, say, a view v1 that uses
table t1, and issues:
LOCK TABLE v1 WRITE;
FLUSH TABLE t1; -- (or just 'FLUSH TABLES'),
an error is produced.
In order to be able to perform DDL on a table under LOCK TABLES,
the table must be locked explicitly in the LOCK TABLES list.
mysql-test/include/handler.inc:
Adjusted test case to trigger an execution path on which bug 41110
"crash with handler command when used concurrently with alter
table" and bug 41112 "crash in mysql_ha_close_table/get_lock_data
with alter table" were originally discovered. Left old test case
which no longer triggers this execution path for the sake of
coverage.
Added test coverage for HANDLER SQL statements and type-aware
metadata locks.
Added a test for the global shared lock and HANDLER SQL.
Updated tests to take into account that the old simple deadlock
detection heuristics was replaced with a graph-based deadlock
detector.
mysql-test/r/debug_sync.result:
Updated results (see debug_sync.test).
mysql-test/r/handler_innodb.result:
Updated results (see handler.inc test).
mysql-test/r/handler_myisam.result:
Updated results (see handler.inc test).
mysql-test/r/innodb-lock.result:
Updated results (see innodb-lock.test).
mysql-test/r/innodb_mysql_lock.result:
Updated results (see innodb_mysql_lock.test).
mysql-test/r/lock.result:
Updated results (see lock.test).
mysql-test/r/lock_multi.result:
Updated results (see lock_multi.test).
mysql-test/r/lock_sync.result:
Updated results (see lock_sync.test).
mysql-test/r/mdl_sync.result:
Updated results (see mdl_sync.test).
mysql-test/r/sp-threads.result:
SHOW PROCESSLIST output has changed due to the fact that waiting
for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/r/truncate_coverage.result:
Updated results (see truncate_coverage.test).
mysql-test/suite/funcs_1/datadict/processlist_val.inc:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/funcs_1/r/processlist_val_no_prot.result:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/rpl/t/rpl_sp.test:
Updated to a new SHOW PROCESSLIST state name.
mysql-test/t/debug_sync.test:
Use LOCK TABLES READ instead of LOCK TABLES WRITE as the latter
no longer allows to trigger execution path involving waiting on
thr_lock.c lock and therefore reaching debug sync-point covered
by this test.
mysql-test/t/innodb-lock.test:
Adjusted test case to the fact that innodb_table_locks=0 option is
no longer supported, since LOCK TABLES WRITE handles all its
conflicts within MDL subsystem.
mysql-test/t/innodb_mysql_lock.test:
Added test for bug #37346 "innodb does not detect deadlock between
update and alter table".
mysql-test/t/lock.test:
Added test coverage which checks the fact that we no longer support
DDL under LOCK TABLES on tables which were locked implicitly.
Adjusted existing test cases accordingly.
mysql-test/t/lock_multi.test:
Added test for bug #46272 "MySQL 5.4.4, new MDL: unnecessary
deadlock". Adjusted other test cases to take into account the
fact that waiting for LOCK TABLES ... WRITE now happens within MDL
subsystem.
mysql-test/t/lock_sync.test:
Since LOCK TABLES ... WRITE now takes SNRW metadata lock for
tables locked explicitly we have to implicitly lock InnoDB tables
(through view) to trigger the table-level lock conflict between
TL_WRITE and TL_WRITE_ALLOW_WRITE.
mysql-test/t/mdl_sync.test:
Added basic test coverage for type-of-operation-aware metadata
locks. Also covered with tests some use cases involving HANDLER
statements in which a deadlock could arise.
Adjusted existing tests to take type-of-operation-aware MDL into
account.
mysql-test/t/multi_update.test:
Update to a new SHOW PROCESSLIST state name.
mysql-test/t/truncate_coverage.test:
Adjusted test case after making LOCK TABLES WRITE to wait until
transactions that use the table to be locked are completed.
Updated to the changed name of DEBUG_SYNC point.
sql/handler.cc:
Global read lock functionality has been
moved into a class.
sql/lock.cc:
Global read lock functionality has been
moved into a class.
Updated code to use the new MDL API.
sql/mdl.cc:
Introduced new type-of-operation aware metadata locks.
To do this:
- Changed MDL_lock to use one list for waiting requests and one
list for granted requests. For each list, added a bitmap
that holds information what lock types a list contains.
Added a helper class MDL_lock::List to manipulate with granted
and waited lists while keeping the bitmaps in sync
with list contents.
- Changed lock-compatibility functions to use bitmaps that
define compatibility.
- Introduced a graph based deadlock detector inspired by
waiting_threads.c from Maria implementation.
- Now that we have a deadlock detector, and no longer have
a global lock to protect individual lock objects, but rather
use an rw lock per object, removed redundant code for upgrade,
and the global read lock. Changed the MDL API to
no longer require the caller to acquire the global
intention exclusive lock by means of a separate method.
Removed a few more methods that became redundant.
- Removed deadlock detection heuristic, it has been made
obsolete by the deadlock detector.
- With operation-type-aware metadata locks, MDL subsystem has
become aware of potential conflicts between DDL and open
transactions. This made it possible to remove calls to
mysql_abort_transactions_with_shared_lock() from acquisition
paths for exclusive lock and lock upgrade. Now we can simply
wait for these transactions to complete without fear of
deadlock. Function mysql_lock_abort() has also become
unnecessary for all conflicting cases except when a DDL
conflicts with a connection that has an open HANDLER.
sql/mdl.h:
Introduced new type-of-operation aware metadata locks.
Introduced a graph based deadlock detector and supporting
methods.
Added comments.
God rid of redundant API calls.
Renamed m_lt_or_ha_sentinel to m_trans_sentinel,
since now it guards the global read lock as well as
LOCK TABLES and HANDLER locks.
sql/mysql_priv.h:
Moved the global read lock functionality into a
class.
Added MYSQL_OPEN_FORCE_SHARED_MDL flag which forces
open_tables() to take MDL_SHARED on tables instead of
metadata locks specified in the parser. We use this to
allow PREPARE run concurrently in presence of
LOCK TABLES ... WRITE.
Added signature for find_table_for_mdl_ugprade().
sql/set_var.cc:
Global read lock functionality has been
moved into a class.
sql/sp_head.cc:
When creating TABLE_LIST elements for prelocking or
system tables set the type of request for metadata
lock according to the operation that will be performed
on the table.
sql/sql_base.cc:
- Updated code to use the new MDL API.
- In order to avoid locks starvation we take upgradable
locks all at once. As result implicitly locked tables no
longer get an upgradable lock. Consequently DDL and FLUSH
TABLES for such tables is prohibited.
find_write_locked_table() was replaced by
find_table_for_mdl_upgrade() function.
open_table() was adjusted to return TABLE instance with
upgradable ticket when necessary.
- We no longer wait for all locks on OT_WAIT back off
action -- only on the lock that caused the wait
conflict. Moreover, now we distinguish cases when we
have to wait due to conflict in MDL and old version
of table in TDC.
- Upate mysql_notify_threads_having_share_locks()
to only abort thr_lock.c waits of threads that
have open HANDLERs, since lock conflicts with only
these threads now can lead to deadlocks not detectable
by the MDL deadlock detector.
- Remove mysql_abort_transactions_with_shared_locks()
which is no longer needed.
sql/sql_class.cc:
Global read lock functionality has been moved into a class.
Re-arranged code in THD::cleanup() to simplify assert.
sql/sql_class.h:
Introduced class to incapsulate global read lock
functionality.
Now sentinel in MDL subsystem guards the global read lock
as well as LOCK TABLES and HANDLER locks. Adjusted code
accordingly.
sql/sql_db.cc:
Global read lock functionality has been moved into a class.
sql/sql_delete.cc:
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
TRUNCATE TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_handler.cc:
Inform MDL_context about presence of open HANDLERs.
Since HANLDERs break MDL protocol by acquiring table-level
lock while holding only S metadata lock on a table MDL
subsystem should take special care about such contexts (Now
this is the only case when mysql_lock_abort() is used).
sql/sql_parse.cc:
Global read lock functionality has been moved into a class.
Do not take upgradable metadata locks when opening tables
for CREATE TABLE SELECT as it is not necessary and limits
concurrency.
When initializing TABLE_LIST objects before adding them
to the table list set the type of request for metadata lock
according to the operation that will be performed on the
table.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result FLUSH
TABLES is no longer allowed for such tables.
sql/sql_prepare.cc:
Use MYSQL_OPEN_FORCE_SHARED_MDL flag when opening
tables during PREPARE. This allows PREPARE to run
concurrently in presence of LOCK TABLES ... WRITE.
sql/sql_rename.cc:
Global read lock functionality has been moved into a class.
sql/sql_show.cc:
Updated code to use the new MDL API.
sql/sql_table.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result DROP
TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_trigger.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
CREATE/DROP TRIGGER is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_view.cc:
Global read lock functionality has been moved into a class.
Fixed results of wrong merge that led to misuse of GLR API.
CREATE VIEW statement is not a commit statement.
sql/table.cc:
When resetting TABLE_LIST objects for PS or SP re-execution
set the type of request for metadata lock according to the
operation that will be performed on the table. Do the same
in auxiliary function initializing metadata lock requests
in a table list.
sql/table.h:
When initializing TABLE_LIST objects set the type of request
for metadata lock according to the operation that will be
performed on the table.
sql/transaction.cc:
Global read lock functionality has been moved into a class.
2010-02-01 12:43:06 +01:00
# The below 'drop table t1' should be able to proceed without
# waiting as it will force HANDLER to be closed.
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
drop table t1;
Implement new type-of-operation-aware metadata locks.
Add a wait-for graph based deadlock detector to the
MDL subsystem.
Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and
bug #37346 "innodb does not detect deadlock between update and
alter table".
The first bug manifested itself as an unwarranted abort of a
transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER
statement, when this transaction tried to repeat use of a
table, which it has already used in a similar fashion before
ALTER started.
The second bug showed up as a deadlock between table-level
locks and InnoDB row locks, which was "detected" only after
innodb_lock_wait_timeout timeout.
A transaction would start using the table and modify a few
rows.
Then ALTER TABLE would come in, and start copying rows
into a temporary table. Eventually it would stumble on
the modified records and get blocked on a row lock.
The first transaction would try to do more updates, and get
blocked on thr_lock.c lock.
This situation of circular wait would only get resolved
by a timeout.
Both these bugs stemmed from inadequate solutions to the
problem of deadlocks occurring between different
locking subsystems.
In the first case we tried to avoid deadlocks between metadata
locking and table-level locking subsystems, when upgrading shared
metadata lock to exclusive one.
Transactions holding the shared lock on the table and waiting for
some table-level lock used to be aborted too aggressively.
We also allowed ALTER TABLE to start in presence of transactions
that modify the subject table. ALTER TABLE acquires
TL_WRITE_ALLOW_READ lock at start, and that block all writes
against the table (naturally, we don't want any writes to be lost
when switching the old and the new table). TL_WRITE_ALLOW_READ
lock, in turn, would block the started transaction on thr_lock.c
lock, should they do more updates. This, again, lead to the need
to abort such transactions.
The second bug occurred simply because we didn't have any
mechanism to detect deadlocks between the table-level locks
in thr_lock.c and row-level locks in InnoDB, other than
innodb_lock_wait_timeout.
This patch solves both these problems by moving lock conflicts
which are causing these deadlocks into the metadata locking
subsystem, thus making it possible to avoid or detect such
deadlocks inside MDL.
To do this we introduce new type-of-operation-aware metadata
locks, which allow MDL subsystem to know not only the fact that
transaction has used or is going to use some object but also what
kind of operation it has carried out or going to carry out on the
object.
This, along with the addition of a special kind of upgradable
metadata lock, allows ALTER TABLE to wait until all
transactions which has updated the table to go away.
This solves the second issue.
Another special type of upgradable metadata lock is acquired
by LOCK TABLE WRITE. This second lock type allows to solve the
first issue, since abortion of table-level locks in event of
DDL under LOCK TABLES becomes also unnecessary.
Below follows the list of incompatible changes introduced by
this patch:
- From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those
statements that acquire TL_WRITE_ALLOW_READ lock)
wait for all transactions which has *updated* the table to
complete.
- From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE
(i.e. all statements which acquire TL_WRITE table-level lock) wait
for all transaction which *updated or read* from the table
to complete.
As a consequence, innodb_table_locks=0 option no longer applies
to LOCK TABLES ... WRITE.
- DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort
statements or transactions which use tables being dropped or
renamed, and instead wait for these transactions to complete.
- Since LOCK TABLES WRITE now takes a special metadata lock,
not compatible with with reads or writes against the subject table
and transaction-wide, thr_lock.c deadlock avoidance algorithm
that used to ensure absence of deadlocks between LOCK TABLES
WRITE and other statements is no longer sufficient, even for
MyISAM. The wait-for graph based deadlock detector of MDL
subsystem may sometimes be necessary and is involved. This may
lead to ER_LOCK_DEADLOCK error produced for multi-statement
transactions even if these only use MyISAM:
session 1: session 2:
begin;
update t1 ... lock table t2 write, t1 write;
-- gets a lock on t2, blocks on t1
update t2 ...
(ER_LOCK_DEADLOCK)
- Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE
was abandoned.
LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same
priority as the usual LOCK TABLE ... WRITE.
SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in
the wait queue.
- We do not take upgradable metadata locks on implicitly
locked tables. So if one has, say, a view v1 that uses
table t1, and issues:
LOCK TABLE v1 WRITE;
FLUSH TABLE t1; -- (or just 'FLUSH TABLES'),
an error is produced.
In order to be able to perform DDL on a table under LOCK TABLES,
the table must be locked explicitly in the LOCK TABLES list.
mysql-test/include/handler.inc:
Adjusted test case to trigger an execution path on which bug 41110
"crash with handler command when used concurrently with alter
table" and bug 41112 "crash in mysql_ha_close_table/get_lock_data
with alter table" were originally discovered. Left old test case
which no longer triggers this execution path for the sake of
coverage.
Added test coverage for HANDLER SQL statements and type-aware
metadata locks.
Added a test for the global shared lock and HANDLER SQL.
Updated tests to take into account that the old simple deadlock
detection heuristics was replaced with a graph-based deadlock
detector.
mysql-test/r/debug_sync.result:
Updated results (see debug_sync.test).
mysql-test/r/handler_innodb.result:
Updated results (see handler.inc test).
mysql-test/r/handler_myisam.result:
Updated results (see handler.inc test).
mysql-test/r/innodb-lock.result:
Updated results (see innodb-lock.test).
mysql-test/r/innodb_mysql_lock.result:
Updated results (see innodb_mysql_lock.test).
mysql-test/r/lock.result:
Updated results (see lock.test).
mysql-test/r/lock_multi.result:
Updated results (see lock_multi.test).
mysql-test/r/lock_sync.result:
Updated results (see lock_sync.test).
mysql-test/r/mdl_sync.result:
Updated results (see mdl_sync.test).
mysql-test/r/sp-threads.result:
SHOW PROCESSLIST output has changed due to the fact that waiting
for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/r/truncate_coverage.result:
Updated results (see truncate_coverage.test).
mysql-test/suite/funcs_1/datadict/processlist_val.inc:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/funcs_1/r/processlist_val_no_prot.result:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/rpl/t/rpl_sp.test:
Updated to a new SHOW PROCESSLIST state name.
mysql-test/t/debug_sync.test:
Use LOCK TABLES READ instead of LOCK TABLES WRITE as the latter
no longer allows to trigger execution path involving waiting on
thr_lock.c lock and therefore reaching debug sync-point covered
by this test.
mysql-test/t/innodb-lock.test:
Adjusted test case to the fact that innodb_table_locks=0 option is
no longer supported, since LOCK TABLES WRITE handles all its
conflicts within MDL subsystem.
mysql-test/t/innodb_mysql_lock.test:
Added test for bug #37346 "innodb does not detect deadlock between
update and alter table".
mysql-test/t/lock.test:
Added test coverage which checks the fact that we no longer support
DDL under LOCK TABLES on tables which were locked implicitly.
Adjusted existing test cases accordingly.
mysql-test/t/lock_multi.test:
Added test for bug #46272 "MySQL 5.4.4, new MDL: unnecessary
deadlock". Adjusted other test cases to take into account the
fact that waiting for LOCK TABLES ... WRITE now happens within MDL
subsystem.
mysql-test/t/lock_sync.test:
Since LOCK TABLES ... WRITE now takes SNRW metadata lock for
tables locked explicitly we have to implicitly lock InnoDB tables
(through view) to trigger the table-level lock conflict between
TL_WRITE and TL_WRITE_ALLOW_WRITE.
mysql-test/t/mdl_sync.test:
Added basic test coverage for type-of-operation-aware metadata
locks. Also covered with tests some use cases involving HANDLER
statements in which a deadlock could arise.
Adjusted existing tests to take type-of-operation-aware MDL into
account.
mysql-test/t/multi_update.test:
Update to a new SHOW PROCESSLIST state name.
mysql-test/t/truncate_coverage.test:
Adjusted test case after making LOCK TABLES WRITE to wait until
transactions that use the table to be locked are completed.
Updated to the changed name of DEBUG_SYNC point.
sql/handler.cc:
Global read lock functionality has been
moved into a class.
sql/lock.cc:
Global read lock functionality has been
moved into a class.
Updated code to use the new MDL API.
sql/mdl.cc:
Introduced new type-of-operation aware metadata locks.
To do this:
- Changed MDL_lock to use one list for waiting requests and one
list for granted requests. For each list, added a bitmap
that holds information what lock types a list contains.
Added a helper class MDL_lock::List to manipulate with granted
and waited lists while keeping the bitmaps in sync
with list contents.
- Changed lock-compatibility functions to use bitmaps that
define compatibility.
- Introduced a graph based deadlock detector inspired by
waiting_threads.c from Maria implementation.
- Now that we have a deadlock detector, and no longer have
a global lock to protect individual lock objects, but rather
use an rw lock per object, removed redundant code for upgrade,
and the global read lock. Changed the MDL API to
no longer require the caller to acquire the global
intention exclusive lock by means of a separate method.
Removed a few more methods that became redundant.
- Removed deadlock detection heuristic, it has been made
obsolete by the deadlock detector.
- With operation-type-aware metadata locks, MDL subsystem has
become aware of potential conflicts between DDL and open
transactions. This made it possible to remove calls to
mysql_abort_transactions_with_shared_lock() from acquisition
paths for exclusive lock and lock upgrade. Now we can simply
wait for these transactions to complete without fear of
deadlock. Function mysql_lock_abort() has also become
unnecessary for all conflicting cases except when a DDL
conflicts with a connection that has an open HANDLER.
sql/mdl.h:
Introduced new type-of-operation aware metadata locks.
Introduced a graph based deadlock detector and supporting
methods.
Added comments.
God rid of redundant API calls.
Renamed m_lt_or_ha_sentinel to m_trans_sentinel,
since now it guards the global read lock as well as
LOCK TABLES and HANDLER locks.
sql/mysql_priv.h:
Moved the global read lock functionality into a
class.
Added MYSQL_OPEN_FORCE_SHARED_MDL flag which forces
open_tables() to take MDL_SHARED on tables instead of
metadata locks specified in the parser. We use this to
allow PREPARE run concurrently in presence of
LOCK TABLES ... WRITE.
Added signature for find_table_for_mdl_ugprade().
sql/set_var.cc:
Global read lock functionality has been
moved into a class.
sql/sp_head.cc:
When creating TABLE_LIST elements for prelocking or
system tables set the type of request for metadata
lock according to the operation that will be performed
on the table.
sql/sql_base.cc:
- Updated code to use the new MDL API.
- In order to avoid locks starvation we take upgradable
locks all at once. As result implicitly locked tables no
longer get an upgradable lock. Consequently DDL and FLUSH
TABLES for such tables is prohibited.
find_write_locked_table() was replaced by
find_table_for_mdl_upgrade() function.
open_table() was adjusted to return TABLE instance with
upgradable ticket when necessary.
- We no longer wait for all locks on OT_WAIT back off
action -- only on the lock that caused the wait
conflict. Moreover, now we distinguish cases when we
have to wait due to conflict in MDL and old version
of table in TDC.
- Upate mysql_notify_threads_having_share_locks()
to only abort thr_lock.c waits of threads that
have open HANDLERs, since lock conflicts with only
these threads now can lead to deadlocks not detectable
by the MDL deadlock detector.
- Remove mysql_abort_transactions_with_shared_locks()
which is no longer needed.
sql/sql_class.cc:
Global read lock functionality has been moved into a class.
Re-arranged code in THD::cleanup() to simplify assert.
sql/sql_class.h:
Introduced class to incapsulate global read lock
functionality.
Now sentinel in MDL subsystem guards the global read lock
as well as LOCK TABLES and HANDLER locks. Adjusted code
accordingly.
sql/sql_db.cc:
Global read lock functionality has been moved into a class.
sql/sql_delete.cc:
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
TRUNCATE TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_handler.cc:
Inform MDL_context about presence of open HANDLERs.
Since HANLDERs break MDL protocol by acquiring table-level
lock while holding only S metadata lock on a table MDL
subsystem should take special care about such contexts (Now
this is the only case when mysql_lock_abort() is used).
sql/sql_parse.cc:
Global read lock functionality has been moved into a class.
Do not take upgradable metadata locks when opening tables
for CREATE TABLE SELECT as it is not necessary and limits
concurrency.
When initializing TABLE_LIST objects before adding them
to the table list set the type of request for metadata lock
according to the operation that will be performed on the
table.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result FLUSH
TABLES is no longer allowed for such tables.
sql/sql_prepare.cc:
Use MYSQL_OPEN_FORCE_SHARED_MDL flag when opening
tables during PREPARE. This allows PREPARE to run
concurrently in presence of LOCK TABLES ... WRITE.
sql/sql_rename.cc:
Global read lock functionality has been moved into a class.
sql/sql_show.cc:
Updated code to use the new MDL API.
sql/sql_table.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result DROP
TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_trigger.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
CREATE/DROP TRIGGER is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_view.cc:
Global read lock functionality has been moved into a class.
Fixed results of wrong merge that led to misuse of GLR API.
CREATE VIEW statement is not a commit statement.
sql/table.cc:
When resetting TABLE_LIST objects for PS or SP re-execution
set the type of request for metadata lock according to the
operation that will be performed on the table. Do the same
in auxiliary function initializing metadata lock requests
in a table list.
sql/table.h:
When initializing TABLE_LIST objects set the type of request
for metadata lock according to the operation that will be
performed on the table.
sql/transaction.cc:
Global read lock functionality has been moved into a class.
2010-02-01 12:43:06 +01:00
unlock tables;
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
# --> connection default
# Reaping 'handler t1 read a next'...
Implement new type-of-operation-aware metadata locks.
Add a wait-for graph based deadlock detector to the
MDL subsystem.
Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and
bug #37346 "innodb does not detect deadlock between update and
alter table".
The first bug manifested itself as an unwarranted abort of a
transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER
statement, when this transaction tried to repeat use of a
table, which it has already used in a similar fashion before
ALTER started.
The second bug showed up as a deadlock between table-level
locks and InnoDB row locks, which was "detected" only after
innodb_lock_wait_timeout timeout.
A transaction would start using the table and modify a few
rows.
Then ALTER TABLE would come in, and start copying rows
into a temporary table. Eventually it would stumble on
the modified records and get blocked on a row lock.
The first transaction would try to do more updates, and get
blocked on thr_lock.c lock.
This situation of circular wait would only get resolved
by a timeout.
Both these bugs stemmed from inadequate solutions to the
problem of deadlocks occurring between different
locking subsystems.
In the first case we tried to avoid deadlocks between metadata
locking and table-level locking subsystems, when upgrading shared
metadata lock to exclusive one.
Transactions holding the shared lock on the table and waiting for
some table-level lock used to be aborted too aggressively.
We also allowed ALTER TABLE to start in presence of transactions
that modify the subject table. ALTER TABLE acquires
TL_WRITE_ALLOW_READ lock at start, and that block all writes
against the table (naturally, we don't want any writes to be lost
when switching the old and the new table). TL_WRITE_ALLOW_READ
lock, in turn, would block the started transaction on thr_lock.c
lock, should they do more updates. This, again, lead to the need
to abort such transactions.
The second bug occurred simply because we didn't have any
mechanism to detect deadlocks between the table-level locks
in thr_lock.c and row-level locks in InnoDB, other than
innodb_lock_wait_timeout.
This patch solves both these problems by moving lock conflicts
which are causing these deadlocks into the metadata locking
subsystem, thus making it possible to avoid or detect such
deadlocks inside MDL.
To do this we introduce new type-of-operation-aware metadata
locks, which allow MDL subsystem to know not only the fact that
transaction has used or is going to use some object but also what
kind of operation it has carried out or going to carry out on the
object.
This, along with the addition of a special kind of upgradable
metadata lock, allows ALTER TABLE to wait until all
transactions which has updated the table to go away.
This solves the second issue.
Another special type of upgradable metadata lock is acquired
by LOCK TABLE WRITE. This second lock type allows to solve the
first issue, since abortion of table-level locks in event of
DDL under LOCK TABLES becomes also unnecessary.
Below follows the list of incompatible changes introduced by
this patch:
- From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those
statements that acquire TL_WRITE_ALLOW_READ lock)
wait for all transactions which has *updated* the table to
complete.
- From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE
(i.e. all statements which acquire TL_WRITE table-level lock) wait
for all transaction which *updated or read* from the table
to complete.
As a consequence, innodb_table_locks=0 option no longer applies
to LOCK TABLES ... WRITE.
- DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort
statements or transactions which use tables being dropped or
renamed, and instead wait for these transactions to complete.
- Since LOCK TABLES WRITE now takes a special metadata lock,
not compatible with with reads or writes against the subject table
and transaction-wide, thr_lock.c deadlock avoidance algorithm
that used to ensure absence of deadlocks between LOCK TABLES
WRITE and other statements is no longer sufficient, even for
MyISAM. The wait-for graph based deadlock detector of MDL
subsystem may sometimes be necessary and is involved. This may
lead to ER_LOCK_DEADLOCK error produced for multi-statement
transactions even if these only use MyISAM:
session 1: session 2:
begin;
update t1 ... lock table t2 write, t1 write;
-- gets a lock on t2, blocks on t1
update t2 ...
(ER_LOCK_DEADLOCK)
- Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE
was abandoned.
LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same
priority as the usual LOCK TABLE ... WRITE.
SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in
the wait queue.
- We do not take upgradable metadata locks on implicitly
locked tables. So if one has, say, a view v1 that uses
table t1, and issues:
LOCK TABLE v1 WRITE;
FLUSH TABLE t1; -- (or just 'FLUSH TABLES'),
an error is produced.
In order to be able to perform DDL on a table under LOCK TABLES,
the table must be locked explicitly in the LOCK TABLES list.
mysql-test/include/handler.inc:
Adjusted test case to trigger an execution path on which bug 41110
"crash with handler command when used concurrently with alter
table" and bug 41112 "crash in mysql_ha_close_table/get_lock_data
with alter table" were originally discovered. Left old test case
which no longer triggers this execution path for the sake of
coverage.
Added test coverage for HANDLER SQL statements and type-aware
metadata locks.
Added a test for the global shared lock and HANDLER SQL.
Updated tests to take into account that the old simple deadlock
detection heuristics was replaced with a graph-based deadlock
detector.
mysql-test/r/debug_sync.result:
Updated results (see debug_sync.test).
mysql-test/r/handler_innodb.result:
Updated results (see handler.inc test).
mysql-test/r/handler_myisam.result:
Updated results (see handler.inc test).
mysql-test/r/innodb-lock.result:
Updated results (see innodb-lock.test).
mysql-test/r/innodb_mysql_lock.result:
Updated results (see innodb_mysql_lock.test).
mysql-test/r/lock.result:
Updated results (see lock.test).
mysql-test/r/lock_multi.result:
Updated results (see lock_multi.test).
mysql-test/r/lock_sync.result:
Updated results (see lock_sync.test).
mysql-test/r/mdl_sync.result:
Updated results (see mdl_sync.test).
mysql-test/r/sp-threads.result:
SHOW PROCESSLIST output has changed due to the fact that waiting
for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/r/truncate_coverage.result:
Updated results (see truncate_coverage.test).
mysql-test/suite/funcs_1/datadict/processlist_val.inc:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/funcs_1/r/processlist_val_no_prot.result:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/rpl/t/rpl_sp.test:
Updated to a new SHOW PROCESSLIST state name.
mysql-test/t/debug_sync.test:
Use LOCK TABLES READ instead of LOCK TABLES WRITE as the latter
no longer allows to trigger execution path involving waiting on
thr_lock.c lock and therefore reaching debug sync-point covered
by this test.
mysql-test/t/innodb-lock.test:
Adjusted test case to the fact that innodb_table_locks=0 option is
no longer supported, since LOCK TABLES WRITE handles all its
conflicts within MDL subsystem.
mysql-test/t/innodb_mysql_lock.test:
Added test for bug #37346 "innodb does not detect deadlock between
update and alter table".
mysql-test/t/lock.test:
Added test coverage which checks the fact that we no longer support
DDL under LOCK TABLES on tables which were locked implicitly.
Adjusted existing test cases accordingly.
mysql-test/t/lock_multi.test:
Added test for bug #46272 "MySQL 5.4.4, new MDL: unnecessary
deadlock". Adjusted other test cases to take into account the
fact that waiting for LOCK TABLES ... WRITE now happens within MDL
subsystem.
mysql-test/t/lock_sync.test:
Since LOCK TABLES ... WRITE now takes SNRW metadata lock for
tables locked explicitly we have to implicitly lock InnoDB tables
(through view) to trigger the table-level lock conflict between
TL_WRITE and TL_WRITE_ALLOW_WRITE.
mysql-test/t/mdl_sync.test:
Added basic test coverage for type-of-operation-aware metadata
locks. Also covered with tests some use cases involving HANDLER
statements in which a deadlock could arise.
Adjusted existing tests to take type-of-operation-aware MDL into
account.
mysql-test/t/multi_update.test:
Update to a new SHOW PROCESSLIST state name.
mysql-test/t/truncate_coverage.test:
Adjusted test case after making LOCK TABLES WRITE to wait until
transactions that use the table to be locked are completed.
Updated to the changed name of DEBUG_SYNC point.
sql/handler.cc:
Global read lock functionality has been
moved into a class.
sql/lock.cc:
Global read lock functionality has been
moved into a class.
Updated code to use the new MDL API.
sql/mdl.cc:
Introduced new type-of-operation aware metadata locks.
To do this:
- Changed MDL_lock to use one list for waiting requests and one
list for granted requests. For each list, added a bitmap
that holds information what lock types a list contains.
Added a helper class MDL_lock::List to manipulate with granted
and waited lists while keeping the bitmaps in sync
with list contents.
- Changed lock-compatibility functions to use bitmaps that
define compatibility.
- Introduced a graph based deadlock detector inspired by
waiting_threads.c from Maria implementation.
- Now that we have a deadlock detector, and no longer have
a global lock to protect individual lock objects, but rather
use an rw lock per object, removed redundant code for upgrade,
and the global read lock. Changed the MDL API to
no longer require the caller to acquire the global
intention exclusive lock by means of a separate method.
Removed a few more methods that became redundant.
- Removed deadlock detection heuristic, it has been made
obsolete by the deadlock detector.
- With operation-type-aware metadata locks, MDL subsystem has
become aware of potential conflicts between DDL and open
transactions. This made it possible to remove calls to
mysql_abort_transactions_with_shared_lock() from acquisition
paths for exclusive lock and lock upgrade. Now we can simply
wait for these transactions to complete without fear of
deadlock. Function mysql_lock_abort() has also become
unnecessary for all conflicting cases except when a DDL
conflicts with a connection that has an open HANDLER.
sql/mdl.h:
Introduced new type-of-operation aware metadata locks.
Introduced a graph based deadlock detector and supporting
methods.
Added comments.
God rid of redundant API calls.
Renamed m_lt_or_ha_sentinel to m_trans_sentinel,
since now it guards the global read lock as well as
LOCK TABLES and HANDLER locks.
sql/mysql_priv.h:
Moved the global read lock functionality into a
class.
Added MYSQL_OPEN_FORCE_SHARED_MDL flag which forces
open_tables() to take MDL_SHARED on tables instead of
metadata locks specified in the parser. We use this to
allow PREPARE run concurrently in presence of
LOCK TABLES ... WRITE.
Added signature for find_table_for_mdl_ugprade().
sql/set_var.cc:
Global read lock functionality has been
moved into a class.
sql/sp_head.cc:
When creating TABLE_LIST elements for prelocking or
system tables set the type of request for metadata
lock according to the operation that will be performed
on the table.
sql/sql_base.cc:
- Updated code to use the new MDL API.
- In order to avoid locks starvation we take upgradable
locks all at once. As result implicitly locked tables no
longer get an upgradable lock. Consequently DDL and FLUSH
TABLES for such tables is prohibited.
find_write_locked_table() was replaced by
find_table_for_mdl_upgrade() function.
open_table() was adjusted to return TABLE instance with
upgradable ticket when necessary.
- We no longer wait for all locks on OT_WAIT back off
action -- only on the lock that caused the wait
conflict. Moreover, now we distinguish cases when we
have to wait due to conflict in MDL and old version
of table in TDC.
- Upate mysql_notify_threads_having_share_locks()
to only abort thr_lock.c waits of threads that
have open HANDLERs, since lock conflicts with only
these threads now can lead to deadlocks not detectable
by the MDL deadlock detector.
- Remove mysql_abort_transactions_with_shared_locks()
which is no longer needed.
sql/sql_class.cc:
Global read lock functionality has been moved into a class.
Re-arranged code in THD::cleanup() to simplify assert.
sql/sql_class.h:
Introduced class to incapsulate global read lock
functionality.
Now sentinel in MDL subsystem guards the global read lock
as well as LOCK TABLES and HANDLER locks. Adjusted code
accordingly.
sql/sql_db.cc:
Global read lock functionality has been moved into a class.
sql/sql_delete.cc:
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
TRUNCATE TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_handler.cc:
Inform MDL_context about presence of open HANDLERs.
Since HANLDERs break MDL protocol by acquiring table-level
lock while holding only S metadata lock on a table MDL
subsystem should take special care about such contexts (Now
this is the only case when mysql_lock_abort() is used).
sql/sql_parse.cc:
Global read lock functionality has been moved into a class.
Do not take upgradable metadata locks when opening tables
for CREATE TABLE SELECT as it is not necessary and limits
concurrency.
When initializing TABLE_LIST objects before adding them
to the table list set the type of request for metadata lock
according to the operation that will be performed on the
table.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result FLUSH
TABLES is no longer allowed for such tables.
sql/sql_prepare.cc:
Use MYSQL_OPEN_FORCE_SHARED_MDL flag when opening
tables during PREPARE. This allows PREPARE to run
concurrently in presence of LOCK TABLES ... WRITE.
sql/sql_rename.cc:
Global read lock functionality has been moved into a class.
sql/sql_show.cc:
Updated code to use the new MDL API.
sql/sql_table.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result DROP
TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_trigger.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
CREATE/DROP TRIGGER is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_view.cc:
Global read lock functionality has been moved into a class.
Fixed results of wrong merge that led to misuse of GLR API.
CREATE VIEW statement is not a commit statement.
sql/table.cc:
When resetting TABLE_LIST objects for PS or SP re-execution
set the type of request for metadata lock according to the
operation that will be performed on the table. Do the same
in auxiliary function initializing metadata lock requests
in a table list.
sql/table.h:
When initializing TABLE_LIST objects set the type of request
for metadata lock according to the operation that will be
performed on the table.
sql/transaction.cc:
Global read lock functionality has been moved into a class.
2010-02-01 12:43:06 +01:00
ERROR 42S02: Table 'test.t1' doesn't exist
A prerequisite patch for the fix for Bug#46224
"HANDLER statements within a transaction might lead to deadlocks".
Introduce a notion of a sentinel to MDL_context. A sentinel
is a ticket that separates all tickets in the context into two
groups: before and after it. Currently we can have (and need) only
one designated sentinel -- it separates all locks taken by LOCK
TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK
and all other locks, which must be released at COMMIT or ROLLBACK.
The tricky part is maintaining the sentinel up to date when
someone release its corresponding ticket. This can happen, e.g.
if someone issues DROP TABLE under LOCK TABLES (generally,
see all calls to release_all_locks_for_name()).
MDL_context::release_ticket() is modified to take care of it.
******
A fix and a test case for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
An attempt to mix HANDLER SQL statements, which are transaction-
agnostic, an open multi-statement transaction,
and DDL against the involved tables (in a concurrent connection)
could lead to a deadlock. The deadlock would occur when
HANDLER OPEN or HANDLER READ would have to wait on a conflicting
metadata lock. If the connection that issued HANDLER statement
also had other metadata locks (say, acquired in scope of a
transaction), a classical deadlock situation of mutual wait
could occur.
Incompatible change: entering LOCK TABLES mode automatically
closes all open HANDLERs in the current connection.
Incompatible change: previously an attempt to wait on a lock
in a connection that has an open HANDLER statement could wait
indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK
is produced.
The idea of the fix is to merge thd->handler_mdl_context
with the main mdl_context of the connection, used for transactional
locks. This makes deadlock detection possible, since all waits
with locks are "visible" and available to analysis in a single
MDL context of the connection.
Since HANDLER locks and transactional locks have a different life
cycle -- HANDLERs are explicitly open and closed, and so
are HANDLER locks, explicitly acquired and released, whereas
transactional locks "accumulate" till the end of a transaction
and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT,
a concept of "sentinel" was introduced to MDL_context.
All locks, HANDLER and others, reside in the same linked list.
However, a selected element of the list separates locks with
different life cycle. HANDLER locks always reside at the
end of the list, after the sentinel. Transactional locks are
prepended to the beginning of the list, before the sentinel.
Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only
release those locks that reside before the sentinel. HANDLER locks
must be released explicitly as part of HANDLER CLOSE statement,
or an implicit close.
The same approach with sentinel
is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES
statement has never worked together, the implementation is
made simple and only maintains one sentinel, which is used either
for HANDLER locks, or for LOCK TABLES locks.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
2009-12-22 17:09:15 +01:00
handler t1 close;
# --> connection con1
# --> connection con2
# --> connection con3
#
# A temporary table test.
# Check that we don't loose positions of HANDLER opened
# against a temporary table.
#
create table t1 (a int, b int, key a (a));
insert into t1 (a) values (1), (2), (3), (4), (5);
create temporary table t2 (a int, b int, key a (a));
insert into t2 (a) select a from t1;
handler t1 open;
handler t1 read a next;
a b
1 NULL
handler t2 open;
handler t2 read a next;
a b
1 NULL
flush table t1;
handler t2 read a next;
a b
2 NULL
# Sic: the position is lost
handler t1 read a next;
a b
1 NULL
select * from t1;
a b
1 NULL
2 NULL
3 NULL
4 NULL
5 NULL
# Sic: the position is not lost
handler t2 read a next;
a b
3 NULL
select * from t2;
ERROR HY000: Can't reopen table: 't2'
handler t2 read a next;
a b
4 NULL
drop table t1;
drop temporary table t2;
#
# A test for lock_table_names()/unlock_table_names() function.
# It should work properly in presence of open HANDLER.
#
create table t1 (a int, b int, key a (a));
create table t2 like t1;
create table t3 like t1;
create table t4 like t1;
handler t1 open;
handler t2 open;
rename table t4 to t5, t3 to t4, t5 to t3;
handler t1 read first;
a b
handler t2 read first;
a b
drop table t1, t2, t3, t4;
#
Implement new type-of-operation-aware metadata locks.
Add a wait-for graph based deadlock detector to the
MDL subsystem.
Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and
bug #37346 "innodb does not detect deadlock between update and
alter table".
The first bug manifested itself as an unwarranted abort of a
transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER
statement, when this transaction tried to repeat use of a
table, which it has already used in a similar fashion before
ALTER started.
The second bug showed up as a deadlock between table-level
locks and InnoDB row locks, which was "detected" only after
innodb_lock_wait_timeout timeout.
A transaction would start using the table and modify a few
rows.
Then ALTER TABLE would come in, and start copying rows
into a temporary table. Eventually it would stumble on
the modified records and get blocked on a row lock.
The first transaction would try to do more updates, and get
blocked on thr_lock.c lock.
This situation of circular wait would only get resolved
by a timeout.
Both these bugs stemmed from inadequate solutions to the
problem of deadlocks occurring between different
locking subsystems.
In the first case we tried to avoid deadlocks between metadata
locking and table-level locking subsystems, when upgrading shared
metadata lock to exclusive one.
Transactions holding the shared lock on the table and waiting for
some table-level lock used to be aborted too aggressively.
We also allowed ALTER TABLE to start in presence of transactions
that modify the subject table. ALTER TABLE acquires
TL_WRITE_ALLOW_READ lock at start, and that block all writes
against the table (naturally, we don't want any writes to be lost
when switching the old and the new table). TL_WRITE_ALLOW_READ
lock, in turn, would block the started transaction on thr_lock.c
lock, should they do more updates. This, again, lead to the need
to abort such transactions.
The second bug occurred simply because we didn't have any
mechanism to detect deadlocks between the table-level locks
in thr_lock.c and row-level locks in InnoDB, other than
innodb_lock_wait_timeout.
This patch solves both these problems by moving lock conflicts
which are causing these deadlocks into the metadata locking
subsystem, thus making it possible to avoid or detect such
deadlocks inside MDL.
To do this we introduce new type-of-operation-aware metadata
locks, which allow MDL subsystem to know not only the fact that
transaction has used or is going to use some object but also what
kind of operation it has carried out or going to carry out on the
object.
This, along with the addition of a special kind of upgradable
metadata lock, allows ALTER TABLE to wait until all
transactions which has updated the table to go away.
This solves the second issue.
Another special type of upgradable metadata lock is acquired
by LOCK TABLE WRITE. This second lock type allows to solve the
first issue, since abortion of table-level locks in event of
DDL under LOCK TABLES becomes also unnecessary.
Below follows the list of incompatible changes introduced by
this patch:
- From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those
statements that acquire TL_WRITE_ALLOW_READ lock)
wait for all transactions which has *updated* the table to
complete.
- From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE
(i.e. all statements which acquire TL_WRITE table-level lock) wait
for all transaction which *updated or read* from the table
to complete.
As a consequence, innodb_table_locks=0 option no longer applies
to LOCK TABLES ... WRITE.
- DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort
statements or transactions which use tables being dropped or
renamed, and instead wait for these transactions to complete.
- Since LOCK TABLES WRITE now takes a special metadata lock,
not compatible with with reads or writes against the subject table
and transaction-wide, thr_lock.c deadlock avoidance algorithm
that used to ensure absence of deadlocks between LOCK TABLES
WRITE and other statements is no longer sufficient, even for
MyISAM. The wait-for graph based deadlock detector of MDL
subsystem may sometimes be necessary and is involved. This may
lead to ER_LOCK_DEADLOCK error produced for multi-statement
transactions even if these only use MyISAM:
session 1: session 2:
begin;
update t1 ... lock table t2 write, t1 write;
-- gets a lock on t2, blocks on t1
update t2 ...
(ER_LOCK_DEADLOCK)
- Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE
was abandoned.
LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same
priority as the usual LOCK TABLE ... WRITE.
SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in
the wait queue.
- We do not take upgradable metadata locks on implicitly
locked tables. So if one has, say, a view v1 that uses
table t1, and issues:
LOCK TABLE v1 WRITE;
FLUSH TABLE t1; -- (or just 'FLUSH TABLES'),
an error is produced.
In order to be able to perform DDL on a table under LOCK TABLES,
the table must be locked explicitly in the LOCK TABLES list.
mysql-test/include/handler.inc:
Adjusted test case to trigger an execution path on which bug 41110
"crash with handler command when used concurrently with alter
table" and bug 41112 "crash in mysql_ha_close_table/get_lock_data
with alter table" were originally discovered. Left old test case
which no longer triggers this execution path for the sake of
coverage.
Added test coverage for HANDLER SQL statements and type-aware
metadata locks.
Added a test for the global shared lock and HANDLER SQL.
Updated tests to take into account that the old simple deadlock
detection heuristics was replaced with a graph-based deadlock
detector.
mysql-test/r/debug_sync.result:
Updated results (see debug_sync.test).
mysql-test/r/handler_innodb.result:
Updated results (see handler.inc test).
mysql-test/r/handler_myisam.result:
Updated results (see handler.inc test).
mysql-test/r/innodb-lock.result:
Updated results (see innodb-lock.test).
mysql-test/r/innodb_mysql_lock.result:
Updated results (see innodb_mysql_lock.test).
mysql-test/r/lock.result:
Updated results (see lock.test).
mysql-test/r/lock_multi.result:
Updated results (see lock_multi.test).
mysql-test/r/lock_sync.result:
Updated results (see lock_sync.test).
mysql-test/r/mdl_sync.result:
Updated results (see mdl_sync.test).
mysql-test/r/sp-threads.result:
SHOW PROCESSLIST output has changed due to the fact that waiting
for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/r/truncate_coverage.result:
Updated results (see truncate_coverage.test).
mysql-test/suite/funcs_1/datadict/processlist_val.inc:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/funcs_1/r/processlist_val_no_prot.result:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/rpl/t/rpl_sp.test:
Updated to a new SHOW PROCESSLIST state name.
mysql-test/t/debug_sync.test:
Use LOCK TABLES READ instead of LOCK TABLES WRITE as the latter
no longer allows to trigger execution path involving waiting on
thr_lock.c lock and therefore reaching debug sync-point covered
by this test.
mysql-test/t/innodb-lock.test:
Adjusted test case to the fact that innodb_table_locks=0 option is
no longer supported, since LOCK TABLES WRITE handles all its
conflicts within MDL subsystem.
mysql-test/t/innodb_mysql_lock.test:
Added test for bug #37346 "innodb does not detect deadlock between
update and alter table".
mysql-test/t/lock.test:
Added test coverage which checks the fact that we no longer support
DDL under LOCK TABLES on tables which were locked implicitly.
Adjusted existing test cases accordingly.
mysql-test/t/lock_multi.test:
Added test for bug #46272 "MySQL 5.4.4, new MDL: unnecessary
deadlock". Adjusted other test cases to take into account the
fact that waiting for LOCK TABLES ... WRITE now happens within MDL
subsystem.
mysql-test/t/lock_sync.test:
Since LOCK TABLES ... WRITE now takes SNRW metadata lock for
tables locked explicitly we have to implicitly lock InnoDB tables
(through view) to trigger the table-level lock conflict between
TL_WRITE and TL_WRITE_ALLOW_WRITE.
mysql-test/t/mdl_sync.test:
Added basic test coverage for type-of-operation-aware metadata
locks. Also covered with tests some use cases involving HANDLER
statements in which a deadlock could arise.
Adjusted existing tests to take type-of-operation-aware MDL into
account.
mysql-test/t/multi_update.test:
Update to a new SHOW PROCESSLIST state name.
mysql-test/t/truncate_coverage.test:
Adjusted test case after making LOCK TABLES WRITE to wait until
transactions that use the table to be locked are completed.
Updated to the changed name of DEBUG_SYNC point.
sql/handler.cc:
Global read lock functionality has been
moved into a class.
sql/lock.cc:
Global read lock functionality has been
moved into a class.
Updated code to use the new MDL API.
sql/mdl.cc:
Introduced new type-of-operation aware metadata locks.
To do this:
- Changed MDL_lock to use one list for waiting requests and one
list for granted requests. For each list, added a bitmap
that holds information what lock types a list contains.
Added a helper class MDL_lock::List to manipulate with granted
and waited lists while keeping the bitmaps in sync
with list contents.
- Changed lock-compatibility functions to use bitmaps that
define compatibility.
- Introduced a graph based deadlock detector inspired by
waiting_threads.c from Maria implementation.
- Now that we have a deadlock detector, and no longer have
a global lock to protect individual lock objects, but rather
use an rw lock per object, removed redundant code for upgrade,
and the global read lock. Changed the MDL API to
no longer require the caller to acquire the global
intention exclusive lock by means of a separate method.
Removed a few more methods that became redundant.
- Removed deadlock detection heuristic, it has been made
obsolete by the deadlock detector.
- With operation-type-aware metadata locks, MDL subsystem has
become aware of potential conflicts between DDL and open
transactions. This made it possible to remove calls to
mysql_abort_transactions_with_shared_lock() from acquisition
paths for exclusive lock and lock upgrade. Now we can simply
wait for these transactions to complete without fear of
deadlock. Function mysql_lock_abort() has also become
unnecessary for all conflicting cases except when a DDL
conflicts with a connection that has an open HANDLER.
sql/mdl.h:
Introduced new type-of-operation aware metadata locks.
Introduced a graph based deadlock detector and supporting
methods.
Added comments.
God rid of redundant API calls.
Renamed m_lt_or_ha_sentinel to m_trans_sentinel,
since now it guards the global read lock as well as
LOCK TABLES and HANDLER locks.
sql/mysql_priv.h:
Moved the global read lock functionality into a
class.
Added MYSQL_OPEN_FORCE_SHARED_MDL flag which forces
open_tables() to take MDL_SHARED on tables instead of
metadata locks specified in the parser. We use this to
allow PREPARE run concurrently in presence of
LOCK TABLES ... WRITE.
Added signature for find_table_for_mdl_ugprade().
sql/set_var.cc:
Global read lock functionality has been
moved into a class.
sql/sp_head.cc:
When creating TABLE_LIST elements for prelocking or
system tables set the type of request for metadata
lock according to the operation that will be performed
on the table.
sql/sql_base.cc:
- Updated code to use the new MDL API.
- In order to avoid locks starvation we take upgradable
locks all at once. As result implicitly locked tables no
longer get an upgradable lock. Consequently DDL and FLUSH
TABLES for such tables is prohibited.
find_write_locked_table() was replaced by
find_table_for_mdl_upgrade() function.
open_table() was adjusted to return TABLE instance with
upgradable ticket when necessary.
- We no longer wait for all locks on OT_WAIT back off
action -- only on the lock that caused the wait
conflict. Moreover, now we distinguish cases when we
have to wait due to conflict in MDL and old version
of table in TDC.
- Upate mysql_notify_threads_having_share_locks()
to only abort thr_lock.c waits of threads that
have open HANDLERs, since lock conflicts with only
these threads now can lead to deadlocks not detectable
by the MDL deadlock detector.
- Remove mysql_abort_transactions_with_shared_locks()
which is no longer needed.
sql/sql_class.cc:
Global read lock functionality has been moved into a class.
Re-arranged code in THD::cleanup() to simplify assert.
sql/sql_class.h:
Introduced class to incapsulate global read lock
functionality.
Now sentinel in MDL subsystem guards the global read lock
as well as LOCK TABLES and HANDLER locks. Adjusted code
accordingly.
sql/sql_db.cc:
Global read lock functionality has been moved into a class.
sql/sql_delete.cc:
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
TRUNCATE TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_handler.cc:
Inform MDL_context about presence of open HANDLERs.
Since HANLDERs break MDL protocol by acquiring table-level
lock while holding only S metadata lock on a table MDL
subsystem should take special care about such contexts (Now
this is the only case when mysql_lock_abort() is used).
sql/sql_parse.cc:
Global read lock functionality has been moved into a class.
Do not take upgradable metadata locks when opening tables
for CREATE TABLE SELECT as it is not necessary and limits
concurrency.
When initializing TABLE_LIST objects before adding them
to the table list set the type of request for metadata lock
according to the operation that will be performed on the
table.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result FLUSH
TABLES is no longer allowed for such tables.
sql/sql_prepare.cc:
Use MYSQL_OPEN_FORCE_SHARED_MDL flag when opening
tables during PREPARE. This allows PREPARE to run
concurrently in presence of LOCK TABLES ... WRITE.
sql/sql_rename.cc:
Global read lock functionality has been moved into a class.
sql/sql_show.cc:
Updated code to use the new MDL API.
sql/sql_table.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result DROP
TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_trigger.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
CREATE/DROP TRIGGER is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_view.cc:
Global read lock functionality has been moved into a class.
Fixed results of wrong merge that led to misuse of GLR API.
CREATE VIEW statement is not a commit statement.
sql/table.cc:
When resetting TABLE_LIST objects for PS or SP re-execution
set the type of request for metadata lock according to the
operation that will be performed on the table. Do the same
in auxiliary function initializing metadata lock requests
in a table list.
sql/table.h:
When initializing TABLE_LIST objects set the type of request
for metadata lock according to the operation that will be
performed on the table.
sql/transaction.cc:
Global read lock functionality has been moved into a class.
2010-02-01 12:43:06 +01:00
# A test for FLUSH TABLES WITH READ LOCK and HANDLER statements.
#
set autocommit=0;
create table t1 (a int, b int, key a (a));
insert into t1 (a, b) values (1, 1), (2, 1), (3, 2), (4, 2), (5, 5);
create table t2 like t1;
insert into t2 (a, b) select a, b from t1;
create table t3 like t1;
insert into t3 (a, b) select a, b from t1;
commit;
flush tables with read lock;
handler t1 open;
lock table t1 read;
handler t1 read next;
ERROR HY000: Can't execute the given command because you have active locked tables or an active transaction
# This implicitly leaves LOCK TABLES but doesn't drop the GLR
lock table not_exists_write read;
ERROR 42S02: Table 'test.not_exists_write' doesn't exist
# We still have the read lock.
drop table t1;
ERROR HY000: Can't execute the query because you have a conflicting read lock
2010-02-12 08:05:43 +01:00
handler t1 read next;
a b
1 1
handler t1 close;
Implement new type-of-operation-aware metadata locks.
Add a wait-for graph based deadlock detector to the
MDL subsystem.
Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and
bug #37346 "innodb does not detect deadlock between update and
alter table".
The first bug manifested itself as an unwarranted abort of a
transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER
statement, when this transaction tried to repeat use of a
table, which it has already used in a similar fashion before
ALTER started.
The second bug showed up as a deadlock between table-level
locks and InnoDB row locks, which was "detected" only after
innodb_lock_wait_timeout timeout.
A transaction would start using the table and modify a few
rows.
Then ALTER TABLE would come in, and start copying rows
into a temporary table. Eventually it would stumble on
the modified records and get blocked on a row lock.
The first transaction would try to do more updates, and get
blocked on thr_lock.c lock.
This situation of circular wait would only get resolved
by a timeout.
Both these bugs stemmed from inadequate solutions to the
problem of deadlocks occurring between different
locking subsystems.
In the first case we tried to avoid deadlocks between metadata
locking and table-level locking subsystems, when upgrading shared
metadata lock to exclusive one.
Transactions holding the shared lock on the table and waiting for
some table-level lock used to be aborted too aggressively.
We also allowed ALTER TABLE to start in presence of transactions
that modify the subject table. ALTER TABLE acquires
TL_WRITE_ALLOW_READ lock at start, and that block all writes
against the table (naturally, we don't want any writes to be lost
when switching the old and the new table). TL_WRITE_ALLOW_READ
lock, in turn, would block the started transaction on thr_lock.c
lock, should they do more updates. This, again, lead to the need
to abort such transactions.
The second bug occurred simply because we didn't have any
mechanism to detect deadlocks between the table-level locks
in thr_lock.c and row-level locks in InnoDB, other than
innodb_lock_wait_timeout.
This patch solves both these problems by moving lock conflicts
which are causing these deadlocks into the metadata locking
subsystem, thus making it possible to avoid or detect such
deadlocks inside MDL.
To do this we introduce new type-of-operation-aware metadata
locks, which allow MDL subsystem to know not only the fact that
transaction has used or is going to use some object but also what
kind of operation it has carried out or going to carry out on the
object.
This, along with the addition of a special kind of upgradable
metadata lock, allows ALTER TABLE to wait until all
transactions which has updated the table to go away.
This solves the second issue.
Another special type of upgradable metadata lock is acquired
by LOCK TABLE WRITE. This second lock type allows to solve the
first issue, since abortion of table-level locks in event of
DDL under LOCK TABLES becomes also unnecessary.
Below follows the list of incompatible changes introduced by
this patch:
- From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those
statements that acquire TL_WRITE_ALLOW_READ lock)
wait for all transactions which has *updated* the table to
complete.
- From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE
(i.e. all statements which acquire TL_WRITE table-level lock) wait
for all transaction which *updated or read* from the table
to complete.
As a consequence, innodb_table_locks=0 option no longer applies
to LOCK TABLES ... WRITE.
- DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort
statements or transactions which use tables being dropped or
renamed, and instead wait for these transactions to complete.
- Since LOCK TABLES WRITE now takes a special metadata lock,
not compatible with with reads or writes against the subject table
and transaction-wide, thr_lock.c deadlock avoidance algorithm
that used to ensure absence of deadlocks between LOCK TABLES
WRITE and other statements is no longer sufficient, even for
MyISAM. The wait-for graph based deadlock detector of MDL
subsystem may sometimes be necessary and is involved. This may
lead to ER_LOCK_DEADLOCK error produced for multi-statement
transactions even if these only use MyISAM:
session 1: session 2:
begin;
update t1 ... lock table t2 write, t1 write;
-- gets a lock on t2, blocks on t1
update t2 ...
(ER_LOCK_DEADLOCK)
- Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE
was abandoned.
LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same
priority as the usual LOCK TABLE ... WRITE.
SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in
the wait queue.
- We do not take upgradable metadata locks on implicitly
locked tables. So if one has, say, a view v1 that uses
table t1, and issues:
LOCK TABLE v1 WRITE;
FLUSH TABLE t1; -- (or just 'FLUSH TABLES'),
an error is produced.
In order to be able to perform DDL on a table under LOCK TABLES,
the table must be locked explicitly in the LOCK TABLES list.
mysql-test/include/handler.inc:
Adjusted test case to trigger an execution path on which bug 41110
"crash with handler command when used concurrently with alter
table" and bug 41112 "crash in mysql_ha_close_table/get_lock_data
with alter table" were originally discovered. Left old test case
which no longer triggers this execution path for the sake of
coverage.
Added test coverage for HANDLER SQL statements and type-aware
metadata locks.
Added a test for the global shared lock and HANDLER SQL.
Updated tests to take into account that the old simple deadlock
detection heuristics was replaced with a graph-based deadlock
detector.
mysql-test/r/debug_sync.result:
Updated results (see debug_sync.test).
mysql-test/r/handler_innodb.result:
Updated results (see handler.inc test).
mysql-test/r/handler_myisam.result:
Updated results (see handler.inc test).
mysql-test/r/innodb-lock.result:
Updated results (see innodb-lock.test).
mysql-test/r/innodb_mysql_lock.result:
Updated results (see innodb_mysql_lock.test).
mysql-test/r/lock.result:
Updated results (see lock.test).
mysql-test/r/lock_multi.result:
Updated results (see lock_multi.test).
mysql-test/r/lock_sync.result:
Updated results (see lock_sync.test).
mysql-test/r/mdl_sync.result:
Updated results (see mdl_sync.test).
mysql-test/r/sp-threads.result:
SHOW PROCESSLIST output has changed due to the fact that waiting
for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/r/truncate_coverage.result:
Updated results (see truncate_coverage.test).
mysql-test/suite/funcs_1/datadict/processlist_val.inc:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/funcs_1/r/processlist_val_no_prot.result:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/rpl/t/rpl_sp.test:
Updated to a new SHOW PROCESSLIST state name.
mysql-test/t/debug_sync.test:
Use LOCK TABLES READ instead of LOCK TABLES WRITE as the latter
no longer allows to trigger execution path involving waiting on
thr_lock.c lock and therefore reaching debug sync-point covered
by this test.
mysql-test/t/innodb-lock.test:
Adjusted test case to the fact that innodb_table_locks=0 option is
no longer supported, since LOCK TABLES WRITE handles all its
conflicts within MDL subsystem.
mysql-test/t/innodb_mysql_lock.test:
Added test for bug #37346 "innodb does not detect deadlock between
update and alter table".
mysql-test/t/lock.test:
Added test coverage which checks the fact that we no longer support
DDL under LOCK TABLES on tables which were locked implicitly.
Adjusted existing test cases accordingly.
mysql-test/t/lock_multi.test:
Added test for bug #46272 "MySQL 5.4.4, new MDL: unnecessary
deadlock". Adjusted other test cases to take into account the
fact that waiting for LOCK TABLES ... WRITE now happens within MDL
subsystem.
mysql-test/t/lock_sync.test:
Since LOCK TABLES ... WRITE now takes SNRW metadata lock for
tables locked explicitly we have to implicitly lock InnoDB tables
(through view) to trigger the table-level lock conflict between
TL_WRITE and TL_WRITE_ALLOW_WRITE.
mysql-test/t/mdl_sync.test:
Added basic test coverage for type-of-operation-aware metadata
locks. Also covered with tests some use cases involving HANDLER
statements in which a deadlock could arise.
Adjusted existing tests to take type-of-operation-aware MDL into
account.
mysql-test/t/multi_update.test:
Update to a new SHOW PROCESSLIST state name.
mysql-test/t/truncate_coverage.test:
Adjusted test case after making LOCK TABLES WRITE to wait until
transactions that use the table to be locked are completed.
Updated to the changed name of DEBUG_SYNC point.
sql/handler.cc:
Global read lock functionality has been
moved into a class.
sql/lock.cc:
Global read lock functionality has been
moved into a class.
Updated code to use the new MDL API.
sql/mdl.cc:
Introduced new type-of-operation aware metadata locks.
To do this:
- Changed MDL_lock to use one list for waiting requests and one
list for granted requests. For each list, added a bitmap
that holds information what lock types a list contains.
Added a helper class MDL_lock::List to manipulate with granted
and waited lists while keeping the bitmaps in sync
with list contents.
- Changed lock-compatibility functions to use bitmaps that
define compatibility.
- Introduced a graph based deadlock detector inspired by
waiting_threads.c from Maria implementation.
- Now that we have a deadlock detector, and no longer have
a global lock to protect individual lock objects, but rather
use an rw lock per object, removed redundant code for upgrade,
and the global read lock. Changed the MDL API to
no longer require the caller to acquire the global
intention exclusive lock by means of a separate method.
Removed a few more methods that became redundant.
- Removed deadlock detection heuristic, it has been made
obsolete by the deadlock detector.
- With operation-type-aware metadata locks, MDL subsystem has
become aware of potential conflicts between DDL and open
transactions. This made it possible to remove calls to
mysql_abort_transactions_with_shared_lock() from acquisition
paths for exclusive lock and lock upgrade. Now we can simply
wait for these transactions to complete without fear of
deadlock. Function mysql_lock_abort() has also become
unnecessary for all conflicting cases except when a DDL
conflicts with a connection that has an open HANDLER.
sql/mdl.h:
Introduced new type-of-operation aware metadata locks.
Introduced a graph based deadlock detector and supporting
methods.
Added comments.
God rid of redundant API calls.
Renamed m_lt_or_ha_sentinel to m_trans_sentinel,
since now it guards the global read lock as well as
LOCK TABLES and HANDLER locks.
sql/mysql_priv.h:
Moved the global read lock functionality into a
class.
Added MYSQL_OPEN_FORCE_SHARED_MDL flag which forces
open_tables() to take MDL_SHARED on tables instead of
metadata locks specified in the parser. We use this to
allow PREPARE run concurrently in presence of
LOCK TABLES ... WRITE.
Added signature for find_table_for_mdl_ugprade().
sql/set_var.cc:
Global read lock functionality has been
moved into a class.
sql/sp_head.cc:
When creating TABLE_LIST elements for prelocking or
system tables set the type of request for metadata
lock according to the operation that will be performed
on the table.
sql/sql_base.cc:
- Updated code to use the new MDL API.
- In order to avoid locks starvation we take upgradable
locks all at once. As result implicitly locked tables no
longer get an upgradable lock. Consequently DDL and FLUSH
TABLES for such tables is prohibited.
find_write_locked_table() was replaced by
find_table_for_mdl_upgrade() function.
open_table() was adjusted to return TABLE instance with
upgradable ticket when necessary.
- We no longer wait for all locks on OT_WAIT back off
action -- only on the lock that caused the wait
conflict. Moreover, now we distinguish cases when we
have to wait due to conflict in MDL and old version
of table in TDC.
- Upate mysql_notify_threads_having_share_locks()
to only abort thr_lock.c waits of threads that
have open HANDLERs, since lock conflicts with only
these threads now can lead to deadlocks not detectable
by the MDL deadlock detector.
- Remove mysql_abort_transactions_with_shared_locks()
which is no longer needed.
sql/sql_class.cc:
Global read lock functionality has been moved into a class.
Re-arranged code in THD::cleanup() to simplify assert.
sql/sql_class.h:
Introduced class to incapsulate global read lock
functionality.
Now sentinel in MDL subsystem guards the global read lock
as well as LOCK TABLES and HANDLER locks. Adjusted code
accordingly.
sql/sql_db.cc:
Global read lock functionality has been moved into a class.
sql/sql_delete.cc:
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
TRUNCATE TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_handler.cc:
Inform MDL_context about presence of open HANDLERs.
Since HANLDERs break MDL protocol by acquiring table-level
lock while holding only S metadata lock on a table MDL
subsystem should take special care about such contexts (Now
this is the only case when mysql_lock_abort() is used).
sql/sql_parse.cc:
Global read lock functionality has been moved into a class.
Do not take upgradable metadata locks when opening tables
for CREATE TABLE SELECT as it is not necessary and limits
concurrency.
When initializing TABLE_LIST objects before adding them
to the table list set the type of request for metadata lock
according to the operation that will be performed on the
table.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result FLUSH
TABLES is no longer allowed for such tables.
sql/sql_prepare.cc:
Use MYSQL_OPEN_FORCE_SHARED_MDL flag when opening
tables during PREPARE. This allows PREPARE to run
concurrently in presence of LOCK TABLES ... WRITE.
sql/sql_rename.cc:
Global read lock functionality has been moved into a class.
sql/sql_show.cc:
Updated code to use the new MDL API.
sql/sql_table.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result DROP
TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_trigger.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
CREATE/DROP TRIGGER is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_view.cc:
Global read lock functionality has been moved into a class.
Fixed results of wrong merge that led to misuse of GLR API.
CREATE VIEW statement is not a commit statement.
sql/table.cc:
When resetting TABLE_LIST objects for PS or SP re-execution
set the type of request for metadata lock according to the
operation that will be performed on the table. Do the same
in auxiliary function initializing metadata lock requests
in a table list.
sql/table.h:
When initializing TABLE_LIST objects set the type of request
for metadata lock according to the operation that will be
performed on the table.
sql/transaction.cc:
Global read lock functionality has been moved into a class.
2010-02-01 12:43:06 +01:00
handler t1 open;
select a from t2;
a
1
2
3
4
5
handler t1 read next;
a b
1 1
flush tables with read lock;
handler t2 open;
flush tables with read lock;
handler t1 read next;
a b
1 1
select a from t3;
a
1
2
3
4
5
handler t2 read next;
a b
1 1
handler t1 close;
rollback;
handler t2 close;
drop table t1;
ERROR HY000: Can't execute the query because you have a conflicting read lock
commit;
flush tables;
drop table t1;
ERROR HY000: Can't execute the query because you have a conflicting read lock
unlock tables;
drop table t1;
set autocommit=default;
drop table t2, t3;
#
# HANDLER statement and operation-type aware metadata locks.
# Check that when we clone a ticket for HANDLER we downrade
# the lock.
#
# Establish an auxiliary connection con1.
# -> connection default
create table t1 (a int, b int, key a (a));
insert into t1 (a, b) values (1, 1), (2, 1), (3, 2), (4, 2), (5, 5);
begin;
insert into t1 (a, b) values (6, 6);
handler t1 open;
handler t1 read a last;
a b
6 6
insert into t1 (a, b) values (7, 7);
handler t1 read a last;
a b
7 7
commit;
# -> connection con1
# Demonstrate that the HANDLER doesn't hold MDL_SHARED_WRITE.
lock table t1 write;
unlock tables;
# -> connection default
handler t1 read a prev;
a b
6 6
handler t1 close;
# Cleanup.
drop table t1;
# -> connection con1
# -> connection default
#
2010-02-02 14:58:15 +01:00
# A test for Bug#50555 "handler commands crash server in
# my_hash_first()".
#
handler no_such_table read no_such_index first;
ERROR 42S02: Unknown table 'no_such_table' in HANDLER
handler no_such_table close;
ERROR 42S02: Unknown table 'no_such_table' in HANDLER
#
2010-02-05 15:52:17 +01:00
# Bug#50907 Assertion `hash_tables->table->next == __null' on
# HANDLER OPEN
#
DROP TABLE IF EXISTS t1, t2;
CREATE TEMPORARY TABLE t1 (i INT);
CREATE TEMPORARY TABLE t2 (i INT);
HANDLER t2 OPEN;
HANDLER t2 READ FIRST;
i
HANDLER t2 CLOSE;
DROP TABLE t1, t2;
#
2010-02-06 10:44:03 +01:00
# Bug#50912 Assertion `ticket->m_type >= mdl_request->type'
# failed on HANDLER + I_S
#
DROP TABLE IF EXISTS t1;
CREATE TABLE t1 (id INT);
HANDLER t1 OPEN;
SELECT table_name, table_comment FROM information_schema.tables
WHERE table_schema= 'test' AND table_name= 't1';
table_name table_comment
t1
HANDLER t1 CLOSE;
DROP TABLE t1;
#
2010-02-12 08:05:43 +01:00
# Test for bug #50908 "Assertion `handler_tables_hash.records == 0'
# failed in enter_locked_tables_mode".
#
drop tables if exists t1, t2;
drop function if exists f1;
create table t1 (i int);
insert into t1 values (1), (2);
create table t2 (j int);
insert into t2 values (1);
create function f1() returns int return (select count(*) from t2);
# Check that open HANDLER survives statement executed in
# prelocked mode.
handler t1 open;
handler t1 read next;
i
1
# The below statement were aborted due to an assertion failure.
select f1() from t2;
f1()
1
handler t1 read next;
i
2
handler t1 close;
# Check that the same happens under GLOBAL READ LOCK.
flush tables with read lock;
handler t1 open;
handler t1 read next;
i
1
select f1() from t2;
f1()
1
handler t1 read next;
i
2
unlock tables;
handler t1 close;
# Now, check that the same happens if LOCK TABLES is executed.
handler t1 open;
handler t1 read next;
i
1
lock table t2 read;
select * from t2;
j
1
unlock tables;
handler t1 read next;
i
2
handler t1 close;
# Finally, check scenario with GRL and LOCK TABLES.
flush tables with read lock;
handler t1 open;
handler t1 read next;
i
1
lock table t2 read;
select * from t2;
j
1
# This unlocks both tables and GRL.
unlock tables;
handler t1 read next;
i
2
handler t1 close;
# Clean-up.
drop function f1;
drop tables t1, t2;
#
2009-08-21 07:55:35 +02:00
# BUG #46456: HANDLER OPEN + TRUNCATE + DROP (temporary) TABLE, crash
#
CREATE TABLE t1 AS SELECT 1 AS f1;
HANDLER t1 OPEN;
TRUNCATE t1;
HANDLER t1 READ FIRST;
ERROR 42S02: Unknown table 't1' in HANDLER
DROP TABLE t1;
CREATE TEMPORARY TABLE t1 AS SELECT 1 AS f1;
HANDLER t1 OPEN;
TRUNCATE t1;
HANDLER t1 READ FIRST;
ERROR 42S02: Unknown table 't1' in HANDLER
DROP TABLE t1;
End of 5.1 tests