> On June 4, 2014, 4:37 p.m., Vinod Kone wrote: > > src/slave/slave.cpp, line 2304 > > <https://reviews.apache.org/r/22251/diff/1/?file=603699#file603699line2304> > > > > pull these out into constants. > > > > DEFAULT_EXECUTOR_CPUS > > DEFAULT_EXECUTOR_MEM > > > > Also these values seem to be too high? what is the minimum value > > allowed by cgroups?
Technically, zero is valid for both but it needs to be something sensible to avoid starving or OOMing the executor. I looked at a command executor running in a memory cgroup and its memory usage was ~ 4 MB. The numbers don't really mean much because the slave doesn't account for them; they are upper bounds to ensure a borked executor doesn't overtake the host. - Ian ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/22251/#review44776 ----------------------------------------------------------- On June 4, 2014, 4:36 p.m., Ian Downes wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/22251/ > ----------------------------------------------------------- > > (Updated June 4, 2014, 4:36 p.m.) > > > Review request for mesos, Ben Mahler, Jie Yu, and Vinod Kone. > > > Bugs: MESOS-1417 > https://issues.apache.org/jira/browse/MESOS-1417 > > > Repository: mesos-git > > > Description > ------- > > Give the command executor some cpu and memory. > > This is necessary because after a terminal task status update the slave > updates the container's resources, which would be zero if the executor didn't > have its own resources. > > This does lead to a small overcommit but the command executor is only > (mainly?) used for tests. > > > Diffs > ----- > > src/slave/slave.cpp 643c0882a4bab1b612b3fb6fd1004e09edf5f368 > > Diff: https://reviews.apache.org/r/22251/diff/ > > > Testing > ------- > > make check > > > Thanks, > > Ian Downes > >
