----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/66449/#review200472 -----------------------------------------------------------
src/tests/containerizer/linux_capabilities_isolator_tests.cpp Lines 747 (patched) <https://reviews.apache.org/r/66449/#comment281217> Should we call `slave.get()->terminate()` before `reset()`? See SlaveRecoveryTest. - Gilbert Song On April 4, 2018, 5:17 a.m., Andrei Budnik wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/66449/ > ----------------------------------------------------------- > > (Updated April 4, 2018, 5:17 a.m.) > > > Review request for mesos, Alexander Rukletsov, Gilbert Song, and Jie Yu. > > > Bugs: MESOS-8489 > https://issues.apache.org/jira/browse/MESOS-8489 > > > Repository: mesos > > > Description > ------- > > Starting more than one agent simultaneously in tests leads to a race > condition between a linux launcher which calls `cgroups::prepare()` for > the first slave and `LinuxLauncherProcess::recover()` which iterates > over cgroups hierarchy for the second slave. Therefore, `mesos/test` > cgroup that is created to check if the kernel supports nested cgroups > can be detected by a recovery process as they use same cgroup hierarchy > path by default. That leads to orphan containers and causes flakiness > of `ROOT_IsolatorFlags` test. To fix the issue, this patch adds > termination of an agent before starting a new one. > > > Diffs > ----- > > src/tests/containerizer/linux_capabilities_isolator_tests.cpp > 147f2cc09307cf8c9cf6f71d0175f8a3593c0256 > > > Diff: https://reviews.apache.org/r/66449/diff/1/ > > > Testing > ------- > > internal CI > > > Thanks, > > Andrei Budnik > >
