Fail updates quickly if Elasticsearch is down or otherwise broken. This should include a fail fast pool counter in case we have a bunch of workers and a timeout. While we're at it we need to make sure that this only effects web users. Maintenance scripts should just wait.
Version: unspecified
Severity: normal
Whiteboard: cirrus_reenable
Oh, and these failures should be invisible to users. They should be logged but we shouldn't bother users with this.
This will require a change to ElasticaConnection - that class seems to retry connection errors without regards to any overall timeout.
Change 96639 had a related patch set uploaded by Manybubbles:
Add timeouts to updates and better log failures