current deploy process is less than ideal, using webops' deploy stuff. also, it doesn't handle v-c-t. it should also drain/undrain the web heads for restarting httpd, similar to what BMO does. :gps likely has other, probably better, ideas.
Something we also should probably consider when deploying is blocking pushes to the review repos and letting in progress pushes complete. The part of the system talking to Review Board during the push makes many requests - if Review Board goes down mid push it will be left in a partial state. The user *should* be able to discard the broken state, or just push overtop of it, but I'm not 100% sure that will cover all possible cases. We should at least investigate this.
We can easily have a hook on the Mercurial server that looks for the presence (or lack thereof) of a heartbeat file and deny all push actions until the check passes.
You did work on this recently, right? What did you do, and what else should we do?
No, we still need to invest in this. This overlaps strongly with work inside v-c-t for configuring the testing environment. I'll find time to talk to fubar sometime about making this all suck less.
You need to log in before you can comment on or make changes to this bug.