> On March 15, 2016, 10:25 p.m., Ben Mahler wrote:
> > src/launcher/executor.cpp, line 499
> > <https://reviews.apache.org/r/44657/diff/4/?file=1299823#file1299823line499>
> >
> >     I'd expect that the buffer is in addition to dealing with the reap 
> > interval, since the reap interval is not really a buffer but a time that we 
> > are expecting to wait. Could you add maybe an additional 1s buffer on top 
> > of the reap interval?
> >     
> >     Note that we could improve the reaper to block auxiliary threads when 
> > the process is a child, rather than polling for children. I had a TODO for 
> > this:
> >     
> >     ```
> >     // TODO(bmahler): This can be optimized to use a thread per pid, where
> >     // each thread makes a blocking call to waitpid. This eliminates the
> >     // unfortunate poll delay.
> >     ```
> 
> Alexander Rukletsov wrote:
>     I think refactoring the reaper is a great idea but maybe not now. I will 
> add an extra time buffer and leave a todo to remove it once the reaper is 
> updated.

Definitely not now :) just justifying why we need a buffer on top of the reap 
interval.


- Ben


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/44657/#review123765
-----------------------------------------------------------


On March 18, 2016, 5:19 p.m., Alexander Rukletsov wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/44657/
> -----------------------------------------------------------
> 
> (Updated March 18, 2016, 5:19 p.m.)
> 
> 
> Review request for mesos, Ben Mahler and Gilbert Song.
> 
> 
> Bugs: MESOS-4909
>     https://issues.apache.org/jira/browse/MESOS-4909
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> The command executor determines how much time it allots the
> underlying task to clean up (effectively how long to wait for
> the task to comply to SIGTERM before sending SIGKILL) based
> on both optional task's `KillPolicy` and optional
> `shutdown_grace_period` field in `ExecutorInfo`.
> 
> 
> Diffs
> -----
> 
>   src/launcher/executor.cpp 4149f084b0b234fb995df04e9d1ca704feec082e 
>   src/slave/slave.cpp 840534ff0687e82ed063c386e36bbabada230697 
> 
> Diff: https://reviews.apache.org/r/44657/diff/
> 
> 
> Testing
> -------
> 
> The complete chain was tested. See https://reviews.apache.org/r/44662/.
> 
> 
> Thanks,
> 
> Alexander Rukletsov
> 
>

Reply via email to