> On Feb. 28, 2014, 6:25 p.m., Ben Mahler wrote:
> > src/launcher/executor.cpp, lines 215-216
> > <https://reviews.apache.org/r/18604/diff/1/?file=506639#file506639line215>
> >
> >     Do we still need this TODO, and/or should we document the fact that the 
> > slave will take care of kill escalation?
> 
> Niklas Nielsen wrote:
>     Ah - shoot. I should have removed that comment.
>     
>     How about leaving a comment along the lines of the RR description:
>     
>     // If the processes doesn't respond to SIGTERM, the slave will call
>     // containerizer->destroy() after EXECUTOR_SHUTDOWN_GRACE_PERIOD seconds.
>     // MesosContainerizer::destroy() calls SIGKILL on the process tree which
>     // effectively is the necessary signal escalation.
> 
> Ben Mahler wrote:
>     I think there's a bug here!
>     
>     Consider killTask() when the child process swallows SIGTERM. The reaped() 
> callback will never be invoked, and the driver will never be stopped. There 
> is no grace period for killTask() as far as the slave is concerned so the 
> command executor becomes effectively 'stuck'.
> 
> Niklas Nielsen wrote:
>     Missed that :s Good catch. Would you recommend rolling this change back 
> and reconsider https://reviews.apache.org/r/18597/ ?

How about we commit a fix (s/SIGTERM/SIGKILL) to master, instead of reverting?


- Ben


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/18604/#review35807
-----------------------------------------------------------


On Feb. 28, 2014, 2:07 a.m., Niklas Nielsen wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/18604/
> -----------------------------------------------------------
> 
> (Updated Feb. 28, 2014, 2:07 a.m.)
> 
> 
> Review request for mesos.
> 
> 
> Repository: mesos-git
> 
> 
> Description
> -------
> 
> If command executor processes doesn't respond to SIGTERM during shutdown, the 
> slave will call containerizer->destroy() after EXECUTOR_SHUTDOWN_GRACE_PERIOD 
> seconds.
> MesosContainerizer::destroy() calls SIGKILL on the process tree which 
> effectively is the necessary signal escalation.
> 
> 
> Diffs
> -----
> 
>   src/launcher/executor.cpp e30d77a 
> 
> Diff: https://reviews.apache.org/r/18604/diff/
> 
> 
> Testing
> -------
> 
> make check
> 
> 
> Thanks,
> 
> Niklas Nielsen
> 
>

Reply via email to