> On June 23, 2015, 4:31 p.m., Niklas Nielsen wrote: > > Right now, the executor removed hook will be called implicitly during > > shutdown, which (most likely) caused the race in the first place. > > Can we split the two concerns by issueing a sched.killTask().
The sched.killTask() doesn't invoke executor shutdown, not for the MockExecutor anyways. The reason being that it's upto the executor to decide whether it wants to hang around after the last task has exited or if it wants to quit. The current mechanism guarantees that the executor shutdown will be called for the MockExecutor and thus the remove-executor hook will be executed by the Slave. - Kapil ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/35756/#review89046 ----------------------------------------------------------- On June 23, 2015, 5:47 p.m., Kapil Arya wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/35756/ > ----------------------------------------------------------- > > (Updated June 23, 2015, 5:47 p.m.) > > > Review request for mesos and Niklas Nielsen. > > > Bugs: MESOS-2226 > https://issues.apache.org/jira/browse/MESOS-2226 > > > Repository: mesos > > > Description > ------- > > Previously, the code was not checking for TASK_RUNNING status message > from the MockExecutor before stopping the scheduler driver. This caused > the Executor to be terminated prematurely (before the tasks were > launched) and thus the remove-executor hook was never called. The fix > was to wait for the TASK_RUNNING status update and then wait for the > shutdown() within MockExecutor. Only then we wait for the future from > remove-executor hook. > > > Diffs > ----- > > src/tests/hook_tests.cpp 3ffde6d6b2faeb5a8a40eb27c3b0a2b7f9ecd2b1 > > Diff: https://reviews.apache.org/r/35756/diff/ > > > Testing > ------- > > make check. > > > Thanks, > > Kapil Arya > >
