Description:
Start group_replication to rejoin cluster failure as clone plugin is waiting for table metadata lock
46 mysql.session localhost NULL Query 71 Waiting for table metadata lock PLUGIN: DROP TABLE `test`.`t1`
TABLE test t1 NULL 139876890984720 EXCLUSIVE TRANSACTION PENDING sql_parse.cc:6093 88 138596
TABLE test t1 NULL 139876817266448 SHARED_WRITE TRANSACTION GRANTED mdl.cc:3694 57 4513
How to repeat:
1.See the steps in the uploaded test case gr_xa_clone_rejoin_failure.test and Run it.
2.Check the output of peformance_schema.metadata_locks in the logfile
./mtr gr_xa_clone_rejoin_failure.test
Logging: ./mtr gr_xa_clone_rejoin_failure.test
MySQL Version 8.0.32
Checking supported features
Using 'all' suites
Collecting tests
Checking leftover processes
Removing old var directory
gr_xa_clone_rejoin_failure.log shows:
**** SHOW PROCESSLIST on server_2 ****
46 mysql.session localhost NULL Query 70 Waiting for table metadata lock PLUGIN: DROP TABLE `test`.`t1`
select * from performance_schema.metadata_locks;
TABLE test t1 NULL 140632676629552 EXCLUSIVE TRANSACTION PENDING sql_parse.cc:6093 88 138690
TABLE test t1 NULL 140632328857360 SHARED_WRITE TRANSACTION GRANTED mdl.cc:3694 57 4511
**** SELECT * FROM performance_schema.threads on server_2 ****
88 thread/group_rpl/THD_plugin_server_session FOREGROUND 46 root localhost NULL Query 70 Waiting for table metadata lock NULL 87 NULL YES YES Plugin 1337840 SYS_default PRIMARY 16720 29056 3276218 3366295
**** replication_group_members on server_2 ****
MEMBER_STATE RECOVERING
Suggested fix:
WL#9335: Enable MDL Locking for Recovered and Detached Prepared XA Transactions
https://dev.mysql.com/worklog/task/?id=9335
According to the WL#9335 and the uploaded testcase, Detached Prepared XA Transactions hold SHARED_WRITE MDL lock, But Clone also needs EXCLUSIVE MDL lock to drop table when node rejoin to cluster
Is this a bug?
Otherwise, it should be stated in the document that it is a limitation of XA and clone, even though this is very unfriendly to use.
