On Thu, Jul 9, 2009 at 14:12, Justin Mason<j...@jmason.org> wrote: > On Thu, Jul 9, 2009 at 13:29, Jukka Zitting<jukka.zitt...@gmail.com> wrote: >> Hi, >> >> On Thu, Jul 9, 2009 at 2:19 PM, Justin Mason<j...@jmason.org> wrote: >>> FWIW, my experience over the last few days of monitoring has been that >>> our build backlogs on the Hudson machine are due to contention for the >>> limited number of executors; particularly the 2 on the main instance. >>> There are a few projects that perform 1.5-hour deployments from this. >>> IMO we need to come up with a way to accomodate this. >> >> Should we add a "long build" lock that all builds that normally take >> more than say 60 minutes should synchronize on? That way we'd never >> have situations where two long builds block both executors at the same >> time. > > hmm, that's a good idea. WDYT, Nigel?
Nigel's on holidays. I've gone ahead and done this anyway ;) There are now two Locks: "Long-running jobs on hudson.zones.apache.org" and "Long-running jobs on minerva.apache.org". (I haven't created one for the Lucene/Hadoop hosts.) Any builds that seem to be taking a very long time (fsvo "very long") will be changed to synch on those locks, in order to leave one of the executors free on those hosts for shorter builds. I've changed all the builds that seem to be consistently running for longer than 1 hour to sync on those locks. --j.
