Elena Stepanova added a comment - 2014-11-23 15:55 The sequence that triggers the failure is perl ./mtr --noreorder main.mdev-504 perfschema.global_read_lock

Elena Stepanova added a comment - 2014-11-23 19:06 - edited Here is the essential parts of the test put together to get the same failure: --let $trial = 10000   while ($trial) { --connect (con3,localhost,root,,) --disconnect con3 --dec $trial }   --connection default   use performance_schema;   update performance_schema.setup_instruments set enabled='YES';   connect (con1, localhost, root, , test);   lock tables performance_schema.setup_instruments read; --disable_result_log select * from performance_schema.setup_instruments; --enable_result_log unlock tables;   lock tables performance_schema.setup_instruments write; update performance_schema.setup_instruments set enabled='NO'; update performance_schema.setup_instruments set enabled='YES'; unlock tables;   --echo connection default; connection default;   flush tables with read lock;   connection con1;   --send lock tables performance_schema.setup_instruments write;   connection default;   let $wait_condition= select 1 from performance_schema.events_waits_current where event_name like "wait/synch/cond/sql/MDL_context::COND_wait_status"; let $wait_timeout= 5;   --source include/wait_condition.inc   unlock tables; It goes all right on 10.0, but on 10.1 it causes a timeout in wait_condition. That's because instead of the expected wait/synch/cond/sql/MDL_context::COND_wait_status we are now getting wait/synch/mutex/sql/MDL_wait::LOCK_wait_status . For me, it starts happening on 10.1 tree with revision 3f2d9a902ec93327515ae94ae0c8c0c2c485d15f . On the previous revision f1afc003eefe0aafd3e070c7453d9e029d8445a8 there is no timeout. I am not sure whether it's expected or not, because my further attempt to investigate got lost in git magic. If I look at the revision 3f2d9a902ec93327515ae94ae0c8c0c2c485d15f , it appears to be just a tiny change in an unrelated test case, which couldn't possibly make a difference. But if I do a git diff between f1afc003eefe0aafd3e070c7453d9e029d8445a8 and 3f2d9a902ec93327515ae94ae0c8c0c2c485d15f , I get a huge diff (like 400K lines).

Elena Stepanova added a comment - 2014-12-20 01:20 After resolving git mystery with serg's help, I got another suspect for breaking the test: commit ab150128ce78fd363f6041277862686a61730b2b Merge: 9534fd8 20e20f6 Author: Jan Lindström <jan.lindstrom@skysql.com> Date: Wed Aug 27 13:15:37 2014 +0300   MDEV-6247: Merge 10.0-galera to 10.1. Merged lp:maria/maria-10.0-galera up to revision 3880. Added a new functions to handler API to forcefully abort_transaction, producing fake_trx_id, get_checkpoint and set_checkpoint for XA. These were added for future possiblity to add more storage engines that could use galera replication.