Elena Stepanova added a comment - 2012-05-23 21:42 Wrong binlog order on concurrent DROP schema and CREATE function Concurrent execution of DROP DATABASE and CREATE FUNCTION | PROCEDURE | EVENT causes a wrong order of events in the binary log and, consequently, replication failure. Provided MTR test case is non-deterministic, so I am also adding an RQG grammar to reproduce. The problem is that on master CREATE FUNCTION gets executed without an error, but in the binlog DROP DATABASE is written before CREATE FUNCTION. Reproducible on maria/5.5 revno 3413, mysql-5.5 revno 3737, mysql-trunk revno 3827. Could not reproduce on maria/5.3, although due to the non-deterministic nature of the test there is no guarantee the problem is not present there. MTR test case t1.test run as perl ./mtr - mysqld= -binlog-format=mixed t1 --source include/master-slave.inc DROP DATABASE IF EXISTS db1; CREATE DATABASE db1; CREATE FUNCTION db1.f1() RETURNS INT RETURN 1; --send DROP DATABASE db1; --connection master1 --error 0,ER_BAD_DB_ERROR CREATE FUNCTION db1.f2() RETURNS INT RETURN 1; --connection master --reap SHOW BINLOG EVENTS; --sync_slave_with_master End of MTR test case Binlog contents: Log_name Pos Event_type Server_id End_log_pos Info slave-relay-bin.000002 4 Format_desc 2 245 Server ver: 5.5.23-MariaDB-debug-log, Binlog ver: 4 slave-relay-bin.000002 245 Rotate 1 0 master-bin.000001;pos=4 slave-relay-bin.000002 289 Format_desc 1 245 Server ver: 5.5.23-MariaDB-debug-log, Binlog ver: 4 slave-relay-bin.000002 530 Query 1 334 DROP DATABASE IF EXISTS db1 slave-relay-bin.000002 619 Query 1 415 CREATE DATABASE db1 slave-relay-bin.000002 700 Query 1 574 use `test`; CREATE DEFINER=`root`@`localhost` FUNCTION `db1`.`f1`() RETURNS int(11) RETURN 1 slave-relay-bin.000002 859 Query 1 653 DROP DATABASE db1 slave-relay-bin.000002 938 Query 1 812 use `test`; CREATE DEFINER=`root`@`localhost` FUNCTION `db1`.`f2`() RETURNS int(11) RETURN 1 RQG grammar t1.yy query: CREATE DATABASE IF NOT EXISTS db1; CREATE FUNCTION db1.f1() RETURNS INT RETURN 1; DROP DATABASE db1; thread1: CREATE FUNCTION db1.f2() RETURNS INT RETURN 1; DROP FUNCTION db1.f2; End of RQG grammar RQG command line: perl runall.pl \ --queries=100M \ --threads=2 \ --duration=600 \ --grammar=t1.yy \ --rpl_mode=mixed \ --basedir1=<your basedir> \ --vardir1=<your vardir>

Rasmus Johansson added a comment - 2012-05-26 14:43 Launchpad bug id: 1003679

Sergei Golubchik added a comment - 2013-01-08 14:11 fixed in mysql-5.7.0

Elena Stepanova added a comment - 2014-11-09 18:54 - edited Still reproducible on current 10.0 (10.0.14+, revno 4471). The MTR test case should be run with --repeat=... to reproduce; or use the RQG test. Fixed in 5.7.0: revno: 3940 revision-id: thayumanavar.x.sachithanantha@oracle.com-20120702054613-d6e02ydictgn46ph parent: mysql-builder@oracle.com-20120701165403-bacrn8oguctyglw4 committer: Thayumanavar <thayumanavar.x.sachithanantha@oracle.com> branch nick: mysql-trunk timestamp: Mon 2012-07-02 11:16:13 +0530 message: Bug#14127220 - WRONG BINLOG ORDER ON CONCURRENT DROP SCHEMA AND CREATE FUNCTION. PROBLEM: Concurrent execution of DROP DATABASE and CREATE FUNCTION causes wrong order of events in the binary log and this results in an replication failure. FIX: The root cause of the issue is when DROP DATABASE takes metadata lock and is in progress through it's execution, a concurrently running CREATE FUNCTION checks for the existence of database which it succeeds and then it waits on the metadata lock. Once DROP DATABASE writes to BINLOG and finally releases the metadata lock on schema object, the CREATE FUNCTION waiting on metadata lock gets in it's code path and succeeds and writes to binlog. When on slave, the sequence of events in binlog cause a failure since after DROP DATABASE and we attempt to create a function under the dropped database object which causes the replication failure. The fix moves the condition to check for existence of database object under the metadata lock in the CREATE FUNCTION code path so that the check for existence of db name happens after the DROP DATABASE has completed it's execution. This patch fixes similar issues when concurrent execution of DROP DATABASE and CREATE EVENT or ALTER EVENT.