> On Oct. 22, 2013, 10:44 p.m., Ben Mahler wrote: > > We can also expose this as a new statistic "mem_usage_bytes", leaving > > "mem_rss_bytes" unchanged. Do you think differentiating the two would be > > beneficial or is rss not measured correctly? > > Vinod Kone wrote: > +1
RSS is an abbreviation of resident set size, which is just memory management speak for how much memory something is using. Which is exactly the memory usage in bytes. memory.stat:rss measures something completely different. For a trivial application like bash it will deviate from the resident set size by as much as 50% of the memory mapped and in use by the process. memory.stat:rss is even worse because it does not even total up the entire memory cgroup. So memory.stat:rss does not seem to be a value we want to use for anything. Furthermore mem_rss_bytes is the multiplatform generic value and as such we should populate it with the most reasonable value we can find of the rss. Allowing for some code that works across Mesos on different OS's. - Eric ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/14825/#review27340 ----------------------------------------------------------- On Oct. 22, 2013, 8:07 a.m., Eric Biederman wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/14825/ > ----------------------------------------------------------- > > (Updated Oct. 22, 2013, 8:07 a.m.) > > > Review request for mesos, Ben Mahler, Ian Downes, and Vinod Kone. > > > Bugs: Mesos-758 > https://issues.apache.org/jira/browse/Mesos-758 > > > Repository: mesos-git > > > Description > ------- > > cgroup_isolator: Report the actual amount of memory used > > Report the number of bytes used by the memory cgroup, and stop > reporting memory with no file backed pages taken into account. > > > Diffs > ----- > > src/slave/cgroups_isolator.cpp fc36f1f > > Diff: https://reviews.apache.org/r/14825/diff/ > > > Testing > ------- > > Verified that memory.stat:rss is the wrong value. > Verified that memory.usage_in_bytes is the correct value. > > make check > > Verified the code is idiomatic and trivial. > > > > Thanks, > > Eric Biederman > >
