FYI, I just managed to crash the TokuDB storage engine on analytics-store MariaDB while replicating some schema changes for mediawiki.

It's busy recovering from the transaction log now. Probably an hour or so delay for that, plus the replication lag catch up. s1-analytics-slave is unaffected if you need emergency access to m2 (eventlogging), s1, or s2.

Luckily we have a stack trace and a candidate upstream bug fix for next time.

--