Comment 1 for bug 1715924

I suspect the reported problem is not linked to the number of file descriptor available to the DB, nor to the configured LimitNOFILE for pacemaker.

When pacemaker start the galera resource, the galera resource agent explicitly runs mysqld_safe with an file descriptor override giving it 16384 fd, which is plenty enough to not cause fd starvation IMHO.

Maybe the original deadlock message is due to concurrent access to the DB, which is normally handled by retrying DB operation on the client side?

Anyway, I don't have any easy means to reproduce the issue right now, and I miss logs for further analysis.

If anything, I would suggest to retest against recent tripleo release, and close the bug if this doesn't reproduce.