Releases/Firefox 3.0.6/Post Mortem

The following are the meeting notes from the Firefox 3.0.6 post-mortem held on Friday, February 6, 2009 at 12:00pm PST.

Development

Bouncer slave database was disabled, causing updates to fail when the master was under high load (bug 476753). Had to pull/push updates a few times while debugging.
1. tm-bouncer01-slave02 was disabled in the load balancer during a previous maintenance window and never re-enabled.
2. Actions:
  1. Need to monitor backend service status on the load balancer (bug 476764) in the same way we monitor origin web servers. Nagios would have alerted after maintenance window that the slave was disabled.
  2. Bouncer needs three databases to withstand a failure of one during release (bug 477183).
Had to throttle bits after release (bug 476875) because mirrors couldn't handle load.
1. Mirror weighting needs to automatically change when mirrors come and go. See bug 454023 ("Sentry should automatically decrease weights on mirrors that get dropped and re-added") for more details.
2. Process for adding new mirrors gated on one person, IT's added more subscribers to the list
Throttling remained on long after necessary
1. Need some monitoring to alert/remind when throttling is still configured (bug 477184).