After the primary locked up and was restarted, it was brought back up with the wrong filesystem, causing a global outage. See also HN discussion.