> On April 18, 2014, 5:55 p.m., Ben Mahler wrote: > > 3rdparty/libprocess/src/process.cpp, lines 1556-1571 > > <https://reviews.apache.org/r/20047/diff/10/?file=562625#file562625line1556> > > > > Feel free to just remove this. > > Dominic Hamon wrote: > elsewhere in this file (597) this is being set to 1 day, which then > overrides the libprocess default of 2 weeks. Given our concern with memory > usage for statistics do you think it's worth changing the libprocess default > to 1 day before removing this? > > Are we sure this env var isn't being used? > > Ben Mahler wrote: > Keeping the default as 1 day for now and keeping an environment variable > SGTM! Alternatively, leave a TODO and we can add LIBPROCESS_METRICS_WINDOW in > a subsequent patch. > > It was using a default of 1 day through LIBPROCESS_STATISTICS_WINDOW, > which overrides TIME_SERIES_WINDOW for the old 'Statistics' abstraction. > While a lower window would reduce consumption for low frequency statistics, > keep in mind that memory consumption is _bounded_ by the capacity using the > "sparsification" technique described in TimeSeries. The move from 2 weeks to > 1 day occurred before we had "sparsification" and it was because we were > storing a lot of resource monitoring data. > > Dominic Hamon wrote: > in the current patch, it's commented out with a TODO to reimplement it. > > as an aside - the current metrics use a time window but leave the > capacity as the default which is 1000. This should make the memory usage > predictable (thanks to sparsification, as you say) but allow for plenty of > data points. > > Once all these parts have landed, i'll be going back to evaluate both > adding the env var back in (which should be a small patch) and checking that > capacity/time window shouldn't both be set. > > sound good? > > Ben Mahler wrote: > Following up with the environment variables SGTM, I'm not sure what you > meant by "shouldn't both be set"?
Sorry, that could have been clearer: To assess whether Metrics should set both the history time window and capacity explicitly as opposed to just the time window. - Dominic ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/20047/#review40835 ----------------------------------------------------------- On April 21, 2014, 10:13 a.m., Dominic Hamon wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/20047/ > ----------------------------------------------------------- > > (Updated April 21, 2014, 10:13 a.m.) > > > Review request for mesos, Benjamin Hindman and Ben Mahler. > > > Bugs: MESOS-1036 > https://issues.apache.org/jira/browse/MESOS-1036 > > > Repository: mesos-git > > > Description > ------- > > see summary > > > Diffs > ----- > > 3rdparty/libprocess/Makefile.am d707ad759dacd16e0177e14f1bf5ece9e4ce2491 > 3rdparty/libprocess/include/process/statistics.hpp > a4f1db3a8a219c39193a1d237477f0350e47e681 > 3rdparty/libprocess/src/process.cpp > 9654c0437edb43cff65dbefdf08dee9e18ef96ab > 3rdparty/libprocess/src/statistics.cpp > 75aac4074d33cb5054da6c8b0bd4a890c2eaf80e > 3rdparty/libprocess/src/tests/statistics_tests.cpp > 3521bd565dae8fcbba464f2539b3b14a37a037f0 > > Diff: https://reviews.apache.org/r/20047/diff/ > > > Testing > ------- > > make check > > > Thanks, > > Dominic Hamon > >
