> On March 2, 2017, 9:26 a.m., Jiang Yan Xu wrote: > > src/slave/slave.cpp > > Lines 1830-1840 (original), 1818-1835 (patched) > > <https://reviews.apache.org/r/55887/diff/6/?file=1652719#file1652719line1831> > > > > This results in a behavior change: if tasks are killed during > > unscheduling of directories, we now send TASK_DROPPED instead of > > TASK_KILLED. I think this is a better behavior because this is race anyways > > and the TASK_DROPPED here better reveals the error condition on the host. > > > > Nevertheless we could check with the @anand and @vinod who last changed > > this. > > Anindya Sinha wrote: > I reinstated the original behavior. We can followup with @anand and > @vinod later. > > Jiang Yan Xu wrote: > So I checked with @anandm on slack: > http://mesos.slackarchive.io/dev/-/1487591716.002096/1488847311.002217/1488846253002216/ > > He convinced me that it makes sense to consistently send `TASK_KILLED` in > this case: > > It takes multiple async steps to launch a task: 1) unscheudling GC, (2, > new) authorize task, 3) launch executor and so far if a task kill lands on > the agent in the other two cases it would result in a `TASK_KILLED`. We > should probalby send `TASK_KILLED` for 2) as well. > > Agreed?
Yes, that is the case in this version. - Anindya ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/55887/#review167654 ----------------------------------------------------------- On March 7, 2017, 12:51 a.m., Anindya Sinha wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/55887/ > ----------------------------------------------------------- > > (Updated March 7, 2017, 12:51 a.m.) > > > Review request for mesos, Adam B, Anand Mazumdar, Alexander Rojas, and Jiang > Yan Xu. > > > Bugs: MESOS-6953 > https://issues.apache.org/jira/browse/MESOS-6953 > > > Repository: mesos > > > Description > ------- > > Added support for action `run_tasks` on the agent's flag `acl`. Based on > the ACL configured for `run_tasks`, a task to be launched on the agent > can be (dis)allowed to launch on the agent. > If a task or task group cannot be launched due to failed authorization, > a `TASK_ERROR` Status Update shall be sent with a reason code of > `REASON_TASK_UNAUTHORIZED` or `REASON_TASK_GROUP_UNAUTHORIZED` as > applicable. > Note that in case of a task group, all tasks fail if any of the tasks > within the task group encounter the authorization error. > > > Diffs > ----- > > src/slave/slave.hpp 857338cc54b341873c338967223c6b0138e9dc3e > src/slave/slave.cpp 775f43b449ec53aa51cc5d3de448ddc7c2059bff > > > Diff: https://reviews.apache.org/r/55887/diff/7/ > > > Testing > ------- > > All tests passed. > > > Thanks, > > Anindya Sinha > >
