> On Nov. 21, 2014, 12:02 a.m., Ben Mahler wrote:
> > src/slave/slave.cpp, lines 4032-4035
> > <https://reviews.apache.org/r/28189/diff/1/?file=768102#file768102line4032>
> >
> >     Why the change to the logic here? Looks like a regression in some cases:
> >     
> >     For example, if there is an executor with port 31000, but no cpus and 
> > no memory, we'll launch it with no memory now!

Ian, I don't think changing executorInfo just here is enough? AFAICT, when the 
command task finishes, the slave calls containerizer->update() with zero 
resources which fails?


- Vinod


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/28189/#review62469
-----------------------------------------------------------


On Nov. 18, 2014, 10:54 p.m., Ian Downes wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/28189/
> -----------------------------------------------------------
> 
> (Updated Nov. 18, 2014, 10:54 p.m.)
> 
> 
> Review request for mesos, Ben Mahler and Vinod Kone.
> 
> 
> Bugs: MESOS-1718
>     https://issues.apache.org/jira/browse/MESOS-1718
> 
> 
> Repository: mesos-git
> 
> 
> Description
> -------
> 
> Use the first task's resources if the ExecutorInfo does not specify any 
> resources. This avoids the previous slight overcommit.
> 
> I decided to make this review to just removing the overcommit. We'll need to 
> deprecate the behavior of accepting executors without resources and we should 
> do the transfer from task to executor for command executors then. Note, as 
> Ben pointed on the other review, this transfer cannot be done at the slave 
> level because command executors are not re-registered with the master and we 
> need to send the true task resources.
> 
> 
> Diffs
> -----
> 
>   src/slave/slave.cpp 06b2e18ff9b202c30f8bf4378cdd35aef734337f 
> 
> Diff: https://reviews.apache.org/r/28189/diff/
> 
> 
> Testing
> -------
> 
> make check
> 
> 
> Thanks,
> 
> Ian Downes
> 
>

Reply via email to