> On July 29, 2014, 10:15 p.m., Ben Mahler wrote: > > src/master/master.cpp, lines 4318-4322 > > <https://reviews.apache.org/r/22796/diff/10/?file=644109#file644109line4318> > > > > Do we need to store the Timers at all? OfferIDs are unique and so this > > should be correct without storing / canceling Timers. Given that, should we > > simplify it a bit and remove the Timer storage / cancellation? > > Timothy Chen wrote: > Hi Ben, I think it should be technically still correct but I wonder if we > should try to cancel the timer to avoid it being still registered in the time > period for the delay, and won't always call removeTimer in the normal path?
Ok, seems good to cancel these timers as we don't fully understand the scalability of having a ton of Timers within libprocess, and there will be a lot of offers for large clusters with high churn. But let's leave a comment here saying that canceling the Timers is only done to avoid having too many active Timers in libprocess? - Ben ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/22796/#review49026 ----------------------------------------------------------- On July 29, 2014, 1:08 a.m., Timothy Chen wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/22796/ > ----------------------------------------------------------- > > (Updated July 29, 2014, 1:08 a.m.) > > > Review request for mesos, Adam B, Ben Mahler, and Niklas Nielsen. > > > Bugs: MESOS-186 > https://issues.apache.org/jira/browse/MESOS-186 > > > Repository: mesos-git > > > Description > ------- > > Based on Kapil's patch (https://reviews.apache.org/r/22066/), adding timeout > for each offer from master to remove the offer when it's no longer used. > > > Diffs > ----- > > src/master/flags.hpp 32704ce > src/master/master.hpp d8a4d9e > src/master/master.cpp 273a516 > src/tests/master_tests.cpp 5a1cf7f > > Diff: https://reviews.apache.org/r/22796/diff/ > > > Testing > ------- > > Added three more unit tests from Kapil's patch: Testing offer not rescinded > after task launched, offer not rescinded when framework/slave unregistered. > The test exposed a race condition that can lead to a segfault if two remove > offers are called on the same offer. > > make check. > > > Thanks, > > Timothy Chen > >
