Low memory ? Running the rules fill up the ast cache with nearly one entry per compiled method instance. (this are thousands instances of Array/OrderedCollections/RB and OC classes) My pharo images runs from ~ 50 MB to ~ 350-400 MB after running all rules for all classes!
2015-09-02 13:42 GMT+02:00 Marcus Denker <[email protected]>: > I have the feeling that is is an artefact of the virtualisation somehow. > > On 02 Sep 2015, at 10:24, Guillermo Polito <[email protected]> > wrote: > > I saw also that one of the linux slaves is much much slower than the > others. Just check this: > > https://ci.inria.fr/pharo/job/30-Bootstrap/buildTimeTrend > > In one of the slaves the build takes one hour. In the other it takes 6. > > El mié., 2 de sept. de 2015 a la(s) 9:11 a. m., Marcus Denker < > [email protected]> escribió: > >> I don’t think so… the job used to run in 15 minutes. But we added more >> tests, >> Rules got slower… >> >> On 02 Sep 2015, at 08:34, Nicolai Hess <[email protected]> wrote: >> >> It is possible to see from the console output which rule or test takes a >> lot time? >> >> 2015-09-02 8:09 GMT+02:00 Marcus Denker <[email protected]>: >> >>> >>> On 01 Sep 2015, at 19:15, Ben Coman <[email protected]> wrote: >>> >>> Do you want to try a post-load... >>> Delay delaySchedulerClass: DelayExperimentalSemaphoreScheduler >>> >>> which fixes the "super fast delay" on virtual machines reported here... >>> http://comments.gmane.org/gmane.comp.lang.smalltalk.pharo.devel/100873 >>> >>> Its named "experimental" only since it was integrated just before >>> Pharo 4 release and did not have wide spread community testing, but >>> its complete and working. There is some further overall streamlining >>> of the DelayScheduler hierarchy I'd like to do later, but it would be >>> good to give this one a run as the default for a while to see how it >>> impacts the issue. (It would be good to get rough "CIAndRule: Timeout" >>> before and after statistics.) >>> >>> btw, is that CI error indicating it times out after one hour? Are >>> there any other timestamps present in the log that indicate if 1 hour >>> really passed? >>> >>> >>> Yes, I think so… >>> >>> https://ci.inria.fr/pharo/job/Pharo-5.0-Issue-Validator/buildTimeTrend >>> >>> Now looking at it (no idea why it is chinese), it seems that we have some >>> slowdown recently and all runs take 40-50 minutes, a slowdown to 60 can >>> happen >>> just by having less compute power available (or running multiple things >>> in parallel). >>> >>> We should >>> >>> -> do a pass over slow tests >>> -> do a pass over slow rules >>> >>> I will put the limit higher than 60 min for now. >>> >>> Marcus >>> >>> >>> >> >> >
