Apparent annotation service restart in prod

[gfr10598]

Looks like there was a restart on Jan 21 at 15:18 UTC. Instance names were not changed, suggesting this may have been crashes of some kind that caused the instances to restart. Very odd that all the instances would have restarted, though. Perhaps indicates that a particular data date caused the crashes. But they didn't crash loop, so that is odd too.