----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/65695/#review198841 -----------------------------------------------------------
Ship it! LGTM. I wonder if we could repurpose/extend any of our existing tests to cover this case... The DefaultExecutorTests already cover a couple of kill-cases, but none where the `KILL` call itself fails. - Joseph Wu On Feb. 16, 2018, 4:27 p.m., Gaston Kleiman wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/65695/ > ----------------------------------------------------------- > > (Updated Feb. 16, 2018, 4:27 p.m.) > > > Review request for mesos, Joseph Wu, Qian Zhang, and Vinod Kone. > > > Bugs: MESOS-8530 > https://issues.apache.org/jira/browse/MESOS-8530 > > > Repository: mesos > > > Description > ------- > > The default executor transitions a task to `TASK_KILLING` and marks its > child container as being killed before posting a `KILL` call to the > agent. > > The executor ignores kill requests for containers that are marked as > being killed, and it doesn't remove this mark if the `KILL` call fails. > This means that it's possible for tasks to get stuck in a `TASK_KILLING` > state. > > This patch makes the default executor remove the killing mark if a > `KILL` call fails. That way a scheduler can retry a kill. > > > Diffs > ----- > > src/launcher/default_executor.cpp 8720dada8bc6ca66f9e0fec6dc265eda3dcc7407 > > > Diff: https://reviews.apache.org/r/65695/diff/1/ > > > Testing > ------- > > `sudo bin/mesos-tests.sh` on GNU/Linux > > > Thanks, > > Gaston Kleiman > >