One mitigation is to reduce innodb_lock_wait_timeout (from default 50 secs to 1 sec in our case): the deadlock gets detected faster, retries are triggered and deletions finish quicker. The underlying problem is of course still there.
One mitigation is to reduce innodb_ lock_wait_ timeout (from default 50 secs to 1 sec in our case): the deadlock gets detected faster, retries are triggered and deletions finish quicker. The underlying problem is of course still there.