There are two kinds of deadlocks:
* natural InnoDB deadlocks, that would happen with single MySQL server
* cluster conflicts, which are caused by simultaneous writes to same
data elements in separate cluster nodes.
Only natural InnoDB deadlocks are logged in INNODB STATUS output.
Currently there is not much diagnostics from cluster conflicts. Only
way to monitor them in more detail is by enabling wsrep_debug. However
that will cause other superfluous log messages. We are working on ways
to improve cluster conflict diagnostics. First step in this is to add
new variable: wsrep_log_conflicts, which will enable cluster conflict
reporting in mysql error log. We will probably get this feature for
the next 2.2 release.
wsrep_retry_autocommit can be tried to help with deadlocks, if
application cannot deal with deadlock errors.
-seppo
> --