Thanks Bill, I'll keep an eye on that. We'll definitely try to switch over to using this method once we understand the best practice.
Josh On Wed, Oct 1, 2014 at 3:47 PM, Bill Farner <wfar...@apache.org> wrote: > Also side-note: in case you're not following AURORA-703 [1], there's a > draft [2] of dedicated machines documentation up. > > > [1] https://issues.apache.org/jira/browse/AURORA-703 > [2] https://reviews.apache.org/r/26244/ > > -=Bill > > On Wed, Oct 1, 2014 at 3:07 PM, Josh Adams <j...@gmail.com> wrote: > > > Yep, makes sense. We will probably be able to rely on these coarse > metrics > > once we homogenize more of our task and slave configurations. We'll > > definitely JIRA if that doesn't work out though! > > > > Best, > > Josh > > > > On Wed, Oct 1, 2014 at 3:04 PM, Kevin Sweeney <kevi...@apache.org> > wrote: > > > > > Hi Josh, > > > > > > Those numbers are only rough estimates, based on ideal T-shirt-sized > > > instances that can be placed anywhere in the cluster. In practice we've > > > found this to be sufficient for high-level monitoring, combined with > the > > > error messages the scheduler generates for pending tasks of individual > > > services. > > > > > > We've experimented with a more fine-grained simulator but have not > > > published any of that code. If you're convinced you need this feature > > > please feel free to file a ticket against us. > > > > > > Cheers, > > > Kevin > > > > > > On Wed, Oct 1, 2014 at 2:41 PM, Josh Adams <j...@gmail.com> wrote: > > > > > > > Hey Kevin, thanks for the fast reply. Currently we're using > scheduling > > > > constraints for dedicated resources (which we'd like to convert to > the > > > > actual "dedicated" resource feature once it's documented). These > > > available > > > > slot numbers are useful at a high level but unfortunately not > detailed > > > > enough for our slightly more fine-grained capacity estimation right > > now. > > > > > > > > Best, > > > > Josh > > > > > > > > On Wed, Oct 1, 2014 at 2:29 PM, Kevin Sweeney <kevi...@apache.org> > > > wrote: > > > > > > > > > There is (currently undocumented) under the scheduler's /vars or > > > > /vars.json > > > > > endpoint. > > > > > > > > > > empty_slots_* > > > > > > > > > > See > > > > > > > > > > > > > > > > > > > > https://github.com/apache/incubator-aurora/blob/master/src/main/java/org/apache/aurora/scheduler/stats/SlotSizeCounter.java > > > > > > > > > > The constants are defined here: > > > > > > > > > > > > > > > > > > > > https://github.com/apache/incubator-aurora/blob/master/src/main/java/org/apache/aurora/scheduler/base/ResourceAggregates.java#L29-L39 > > > > > > > > > > On Wed, Oct 1, 2014 at 2:24 PM, Josh Adams <j...@gmail.com> wrote: > > > > > > > > > > > Hi there, > > > > > > > > > > > > Is there a way to ask the Scheduler for a current headroom > estimate > > > (in > > > > > > number of potential task "slots" available) for a given > TaskConfig? > > > > > > > > > > > > I understand that I could do this manually by processing > > /state.json > > > on > > > > > the > > > > > > Mesos master and joining that with info from a getJobsResult, but > > I'd > > > > > > rather not reinvent the wheel if there's already something that > > does > > > > > this. > > > > > > > > > > > > Thanks! > > > > > > Josh > > > > > > > > > > > > > > > > > > > > >
