[09:36] <wgrant> cjwatson: I concur, now that we have lots of PPA builders.
[09:36] <wgrant> And I wasn't looking forward to porting that bit of the query to the New World.
[09:36] <wgrant> :)
[09:36] <cjwatson> Which of our several new worlds?
[09:36] <wgrant> buildd-manager's.
[09:37] <wgrant> Having a separate scanner to keep a copy of the heads of the queues in memory, so we don't have to have three transactions per builder per cycle.
[09:37] <cjwatson> Hadn't realised you were rearranging the data model too.
[09:38] <wgrant> The data model isn't changing, but the per-builder scanners in buildd-manager will pull from an in-memory queue rather than directly from the database.
[09:38] <wgrant> So we'll pull the current queue for each (arch, virt) once every 15 seconds, rather than once for every builder every 15 seconds.
[09:39] <wgrant> So a no-op scan cycle should be quick and not hit the DB at all
[09:39] <cjwatson> Ah, right, I can see how that would break that query then.
[09:39] <wgrant> It might not break the query itself, but it makes it harder to reason about.
[09:40] <wgrant> Because a build can be disqualified based on the other builds that might be dispatched during that round.
[09:40] <wgrant> I really should also relax the delay for private builds.
[09:41] <wgrant> That bit of the query is relatively slow, and avoidable now we have the public restricted librarian.
[09:41] <lifeless> should really do the url mangling bugfix for that ;)
[09:42] <wgrant> Indeed, eventually :)
[09:48] <cjwatson> wgrant: Do you agree with my comment in the MP that it can probably be no-qa?
[09:49] <wgrant> cjwatson: Certainly.
[09:49] <cjwatson> OK, cool
[10:11] <stub> Anyone up for the usual thingami ?