Our application deletes old records from 2 tables, which have a foreign key shared between them. The queries apparently run in succession on our primary host, but deadlock on the down stream servers. The downstreams crash, leave the cluster such that the remaining host starts looping, trying to reconnect. The survivor downgrades to non-Cluster, so it is useless to our applications until it is rebooted. Restarting the downstreams is arduous – SST fails, and have to be recovered manually (only to fail again at 02:00 the next night, when the deletes are repeated).

I've attached the mysql-error.log from one of the downstream hosts, which reflects what happens on each of them, and survivor-mysql-error , which shows how our primary host logs the events.