> On Nov. 19, 2013, 11:33 p.m., Ben Mahler wrote: > > 3rdparty/libprocess/3rdparty/stout/include/stout/os.hpp, lines 839-841 > > <https://reviews.apache.org/r/15653/diff/2/?file=388562#file388562line839> > > > > Hey Nik, I see your gist here: > > https://gist.github.com/nqn/7493244 > > > > More interesting than node-wide load average will be the total cpu time > > for the master, can we expose the same cpu time information as what we do > > in ProcessIsolator::usage instead of the load average? > > Niklas Nielsen wrote: > If I understand correctly, get and aggregate cpu times for the master's > process tree instead? > I think that is valuable too and we could totally do that, but I still > think the system load average is a good indicator (and well-known convention) > of the general machine load. > > Can overcommitment be detected from the cpu time? AFAIK, you get from cpu > time to cpu usage % from the fraction (cpu time / measured total time) which > is at most 1. The load average is based on the process queue length, which > along with the number of cores captures overcommitment. > I might be totally off - maybe there is a good reason for having both? > > Thoughts?
Having both sounds good, just want to make sure we have cpu time as well, so that we can detect CPU consumption spikes in the Master :) - Ben ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/15653/#review29151 ----------------------------------------------------------- On Nov. 19, 2013, 11:08 p.m., Niklas Nielsen wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/15653/ > ----------------------------------------------------------- > > (Updated Nov. 19, 2013, 11:08 p.m.) > > > Review request for mesos, Benjamin Hindman, Ben Mahler, and Vinod Kone. > > > Repository: mesos-git > > > Description > ------- > > This patch includes a wrapper to get system load averages in uptime(1) > format. This is used by an upcoming patch which expose these averages > over master and slave stats.json endpoints. > > > Diffs > ----- > > 3rdparty/libprocess/3rdparty/stout/include/stout/os.hpp f6bbf5e > > Diff: https://reviews.apache.org/r/15653/diff/ > > > Testing > ------- > > make check and functional testing with endpoints. > > > Thanks, > > Niklas Nielsen > >
