I could reproduce it manually reliably, but when it's put in MTR, it turns out to be weirdly sporadic. I need to look more at it.

So, when the scenario is run manually, the problem is sporadic too.
Apparently, there is a race condition between two slave SQL threads executing CREATE nearly simultaneously, caused by the specific (maybe erroneous) locking algorithm when it comes to converting CREATE to CREATE OR REPLACE inside the slave with IDEMPOTENT DDL.

I wasn't able to produce a deterministic test case because I cannot trigger sync points inside slave SQL threads, and running regular CREATE OR REPLACE statements doesn't seem to cause the same problem since it's processed differently.

Below is a crude MTR test case to reproduce the problem (not to be included into the suite!).

--let $rpl_topology=1->2->3
--source include/rpl_init.inc
--source include/have_binlog_format_statement.inc

--enable_connect_log

connection server_3;
--source include/stop_slave.inc
reset slave all;
eval change master 'm1' to master_host='127.0.0.1', master_port=$SERVER_MYPORT_1, master_user='root';
eval change master 'm2' to master_host='127.0.0.1', master_port=$SERVER_MYPORT_2, master_user='root';
start all slaves;

connection server_1;

let $count = 100;
while ($count)
{

	connection server_1;
	eval create table t$count (i int);
	save_master_pos;

	connection server_2;
	sync_with_master;

	connection server_3;
	sync_with_master 0, 'm1';

	connection server_2;
	save_master_pos;

	connection server_3;
	sync_with_master 0, 'm2';

	dec $count;
}

connection server_1;
drop database test;
create database test;

--source include/rpl_end.inc

Hi!

I was not able to repeat the problem.
How should the above program fail ?

I was trying to test this in 10.0, but noticed that the required test.cnf file was missing.

I created the following file mysql-test/suite/rpl/t/skr.cnf

!include suite/rpl/my.cnf

[mysqld.1]
log-slave-updates
loose-innodb

[mysqld.2]
log-slave-updates
loose-innodb
replicate-ignore-table=db.t2

[mysqld.3]
log-slave-updates
loose-innodb

[ENV]
SERVER_MYPORT_3= @mysqld.3.port
SERVER_MYSOCK_3= @mysqld.3.socket

When I run the test case as:

mysql-test-run rpl.skr

This crashed in rpl_end with:
t line 149: sync_slave_with_master failed: 'select master_pos_wait('slave-bin.000001', 13044, 300, '')' returned NULL indicating slave SQL thread failure

However, I don't have any problems with create table failing (as far as I can see)

Here is my cnf file:

!include suite/rpl/my.cnf

[mysqld.1]
log-slave-updates
loose-innodb

[mysqld.2]
log-slave-updates
loose-innodb

[mysqld.3]
log-slave-updates
loose-innodb

[ENV]
SERVER_MYPORT_3=                @mysqld.3.port
SERVER_MYSOCK_3=                @mysqld.3.socket

For me, the test also fails like this:

analyze: sync_with_master
mysqltest: At line 37: sync_with_master failed: 'select master_pos_wait('master-bin.000001', 689, 300, 'm1')' returned NULL indicating slave SQL thread failure

And the error logs shows this:

140730 17:59:22 [ERROR] Master 'm1': Slave SQL: Error 'Table 't98' already exists' on query. Default database: 'test'. Query: 'create table t98 (i int)', Gtid 0-1-3, Internal MariaDB error code: 1050
140730 17:59:22 [Warning] Master 'm1': Slave: Table 't98' already exists Error_code: 1050
140730 17:59:22 [ERROR] Master 'm1': Error running query, slave SQL thread aborted. Fix the problem, and restart the slave SQL thread with "SLAVE START". We stopped at log 'master-bin.000001' position 564

The race condition was in lock_table_names() which failed if we tried to do a create table withing the replication thread if the table was in use.

Fixed by propagating the REPLACE flag to lock_table_names().

Pushed into 10.0 tree

Hi

I am encountering the same issue but with a VIEW.

Master is Server version: 10.0.13-MariaDB-log Homebrew
on OS X 10.10.3

Slave is Server version: 10.0.19-MariaDB-1~precise-log mariadb.org binary distribution
on Ubuntu 14.04.1 LTS

Jun 13 16:54:49 whktest mysqld: 150613 16:54:49 [Warning] Slave: Table
'listview1' already exists Error_code: 1050
Jun 13 16:54:49 whktest mysqld: 150613 16:54:49 [ERROR] Error running
query, slave SQL thread aborted. Fix the problem, and restart the slave
SQL thread with "SLAVE START". We stopped at log 'master-bin.000002'
position 55904540

The statement causing this is

CREATE ALGORITHM = UNDEFINED
DEFINER = `xxxxxxxx`@`localhost`
SQL SECURITY DEFINER
VIEW `listview1`
AS
SELECT id ,
from_fld ,
to_fld ,
subj_fld ,
date_fld ,
mailbox ,
body_ fld ,
numAttach ,
attachNames,
attachText ,
headings
FROM email
WHERE id IN(197854,
197853,
197852)

Stepping over the statement does not work.

greetings, el

Dr Eberhard W Lisse,

I am encountering the same issue but with a VIEW

By "the same issue" do you mean that you also have two masters executing the statement, and observe the error despite slave_ddl_exec_mode=IDEMPOTENT?
Or do you just mean that you are getting replication failure with a similar error text?

This bug report was about a very specific situation. The conflict was expected to happen in the given setup, but it was supposed to be suppressed by the slave configuration.

If you mean the same – that you indeed create a duplicate view, but don't expect it to cause a problem due to IDEMPOTENT mode, then I think the expectation is not valid, according to the variable description it does not affect views.

If you just mean that you are getting the error when you think there should be no conflict, we will need much more information to find out why it might be happening, and it's not related to this bug report. Please create another JIRA issue, attach your binary logs (or their fragments around this view creation, the wider the better), cnf files and error logs.