|
If a slave is restarted when it's synchronized with the master, and there are no new events in the master binlog, the old-style replication coordinates on the slave are not set properly (point at the last GTID event).
As a result, MASTER_POS_WAIT for the current master position does not return.
Another potential problem is that if at this point the slave is reconfigured NOT to use GTID, the last event will be re-executed, thus causing data inconsistency.
Test case:
--source include/master-slave.inc
|
|
--connection slave
|
--source include/stop_slave.inc
|
eval CHANGE MASTER TO master_use_gtid=1;
|
--source include/start_slave.inc
|
|
--connection master
|
CREATE TABLE t1 (i INT);
|
INSERT INTO t1 VALUES (1);
|
|
--sync_slave_with_master
|
|
--enable_reconnect
|
--append_file $MYSQLTEST_VARDIR/tmp/mysqld.2.expect
|
restart: --skip-slave-start=0
|
EOF
|
--shutdown_server 60
|
--source include/wait_until_connected_again.inc
|
|
--connection master
|
|
--echo # If we do INSERT here, everything works okay.
|
--echo # But if we don't, sync_slave_with_master hangs
|
--echo # because MASTER_POS_WAIT for the current master position
|
--echo # never returns
|
# insert into t1 values (2);
|
|
show master status;
|
|
--sync_slave_with_master
|
SELECT * FROM t1;
|
bzr version-info
revision-id: knielsen@knielsen-hq.org-20130506123534-v8r8bhazhj5gr1he
|
revno: 3628
|
branch-nick: 10.0-base
|
|