mesos-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Joseph Wu <jos...@mesosphere.io>
Subject Re: Review Request 70577: Changed Agent V1 GET_STATE for any completed executor's tasks.
Date Wed, 01 May 2019 19:24:43 GMT


> On April 30, 2019, 9:13 p.m., Benjamin Mahler wrote:
> > Hm.. at first glance it seems odd to put the special logic in the http response
handler instead of just having the internal state reflect that the task is terminal (after
all.. it is terminated). Cant't the agnet move the tasks to an appropriate terminal state
prior to "completing" the executor?
> > 
> > That way, we also will know that the endpoints are accurately reflecting the in-memory
state.

Your suggestion leads to a neater change, and I've posted the revision.  The test needed no
changes :)


- Joseph


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/70577/#review214974
-----------------------------------------------------------


On May 1, 2019, 12:21 p.m., Joseph Wu wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/70577/
> -----------------------------------------------------------
> 
> (Updated May 1, 2019, 12:21 p.m.)
> 
> 
> Review request for mesos, Gilbert Song, Greg Mann, and Vinod Kone.
> 
> 
> Bugs: MESOS-9750
>     https://issues.apache.org/jira/browse/MESOS-9750
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> It is possible for a completed executor to have a non-terminal task
> (based on last status update).  For example, during graceful shutdown
> of an agent, graceful shutdown of an executor will race with the
> agent's shutdown grace period.  If the executor does not send a
> TASK_KILLED in time, the agent will still mark the executor as complete
> and kill it.
> 
> After agent recovery, these completed executors will show up in an
> agent's /state and GET_STATE responses.  In GET_STATE however, any
> non-terminal tasks will appear under `launched_tasks`.  This may
> provide misleading information about the total number of tasks running.
> 
> This commit adds extra logic to place these non-terminal tasks under
> the `terminated_tasks` category, and adds a regression test.
> 
> 
> Diffs
> -----
> 
>   src/slave/slave.cpp 95f05a18c7905d5032de1cd35726ac3a17f0b682 
>   src/tests/api_tests.cpp bc19d7e9661f091aebf3072967548c7c3196239d 
> 
> 
> Diff: https://reviews.apache.org/r/70577/diff/2/
> 
> 
> Testing
> -------
> 
> ```
> make check
> src/mesos-tests --gtest_filter="*GetStateWithNonTerminalCompletedTask*" --verbose
> ```
> 
> 
> Thanks,
> 
> Joseph Wu
> 
>


Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message