> On Feb. 18, 2016, 1:44 a.m., Qian Zhang wrote: > > I see you send TASK_KILLING once the killTask() is invoked. However, > > according to the description of MESOS-4140, we may need to send such status > > update after SIGTERM is sent to the task and before SIGKILL is sent. So > > maybe you should send TASK_KILLING in shutdown() instead? > > Abhishek Dasgupta wrote: > Okay, i see. but in the case of docker executor, TASK_KILLING should come > after docker stop. Does it sound good?? > > Abhishek Dasgupta wrote: > Actually, to set the status, we need to send the task_id also. Task_id is > not available inside shutdown. Moreover, I don't think it will make any > difference if it is called inside shutdown instead of killTask. It makes > sense to change the status a TASK_KILLING as soon as killTask is issued. > > Qian Zhang wrote: > shutdown() will be not only called by killTask(), but also called when > the executor itself is asked to shutdown, so if you send the status in > killTask(), that means for the latter case (executor shutdown), the > TASK_KILLING status will not be sent? > > Abhishek Dasgupta wrote: > So, here is the dubious scenario. If executor shuts down, you can't > exepect to provide task_id to shutdown() as there might be lots of tasks. > That's why the only argument of shutdown() is executordriver. But to update > TASK_KILLING status, you must need task_id. So, in my understanding, shutdown > by executor shut down should not require TASK_KILLING status as in that case, > all the tasks are unhealthy.
Recall that executors can manage many tasks, the command/docker executors are implementations provided by mesos and they manage a single task only. It is up to the executor to decide how to shutdown, in the case of the command executor we kill the active task and report TASK_KILLED (in both the killTask and shutdown cases): https://github.com/apache/mesos/blob/0.27.0/src/launcher/executor.cpp#L514-L516 Since we'd like to send a TASK_KILLING when we start to kill a task, we need to send it in both the killTask and shutdown. Make sense? - Ben ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/43489/#review119568 ----------------------------------------------------------- On Feb. 16, 2016, 10:23 a.m., Abhishek Dasgupta wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/43489/ > ----------------------------------------------------------- > > (Updated Feb. 16, 2016, 10:23 a.m.) > > > Review request for mesos, Ben Mahler and Qian Zhang. > > > Bugs: MESOS-4547 > https://issues.apache.org/jira/browse/MESOS-4547 > > > Repository: mesos > > > Description > ------- > > KillTask introduces TASK_KILLING state. > > > Diffs > ----- > > src/docker/executor.cpp 654a41db843a85d953880d5145bc95ada9ed2920 > src/launcher/executor.cpp c27e0792e4b9bcec0829a46be4232c013d965cf9 > > Diff: https://reviews.apache.org/r/43489/diff/ > > > Testing > ------- > > make check > > > Thanks, > > Abhishek Dasgupta > >