> On Sept. 15, 2014, 4:38 p.m., Benjamin Hindman wrote: > > src/master/master.cpp, line 4010 > > <https://reviews.apache.org/r/25250/diff/4/?file=682254#file682254line4010> > > > > I suggest we use TASK_LOST here instead. We definitely want a terminal > > state like TASK_KILLED, but we've reserved TASK_KILLED for when a framework > > has actually intiated the kill itself, and thus I'd prefer not to overload > > the semantics. This might be a good candidate for a new task state, e.g., > > TASK_REMOVED, which has been discussed in the past, but I can't recall if > > there is a JIRA for that or not. If not, it would be great to have you > > create one Alex so we can have a discussion about how to introduce new task > > states (and maybe even a way to introduce sub-states that framework writers > > themselves could customize). > > Alexander Rukletsov wrote: > I used to have `TASK_LOST` here, but my understanding is that `TASK_LOST` > is used for abnormal situations, i.e. when the task is not finished not > because of scheduler's direct command, but because of some external reasons. > I agree, that a new task state is a very good solution. We have [this > ticket](https://issues.apache.org/jira/browse/MESOS-343), one solution for > which would be to introduce something like `TaskStatusExplained` or a > protobuf message for every task. But maybe for this situation something like > `TASK_ABANDONED` would be rather descriptive.
Keeping TASK_KILLED, but added comments for clarity. - Alexander ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/25250/#review53350 ----------------------------------------------------------- On Sept. 22, 2014, 4:25 p.m., Alexander Rukletsov wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/25250/ > ----------------------------------------------------------- > > (Updated Sept. 22, 2014, 4:25 p.m.) > > > Review request for mesos, Benjamin Hindman and Till Toenshoff. > > > Bugs: MESOS-1736 > https://issues.apache.org/jira/browse/MESOS-1736 > > > Repository: mesos-git > > > Description > ------- > > When a framework is shut down e.g. by calling driver.stop() from the > scheduler, running tasks are marked KILLED before migrating them to completed. > > > Diffs > ----- > > src/master/master.cpp e5d30e9 > src/tests/master_tests.cpp 8e4ec1d > > Diff: https://reviews.apache.org/r/25250/diff/ > > > Testing > ------- > > make check (OS X) > > > Thanks, > > Alexander Rukletsov > >