We don't test explicitly for this, but I'm surprised to hear about a jump of that magnitude. Do you have any more detailed profiling? Can you generate some? (With the tcmalloc heap dumps.) -Greg Software Engineer #42 @ http://inktank.com | http://ceph.com On Mon, Jul 7, 2014 at 3:03 AM, Sylvain Munaut <s.munaut at whatever-company.com> wrote: > Hi, > > >> We actually saw a decrease in memory usage after upgrading to Firefly, >> though we did reboot the nodes after the upgrade while we had the >> maintenance window. This is with 216 OSDs total (32-40 per node): >> http://i.imgur.com/BC7RuXJ.png > > > Interesting. Is that cluster for RBD or RGW ? My RBD OSDs are a bit > better behaved but still had this 25% bump in mem usage ... > > > > Here the memory pretty much just grows continually. > > This is the log over the last year. > > http://i.imgur.com/0NUFjpz.png > > At the very beginning (~250M per process) those OSD were empty, just > added. Then we changed the crushmap to map all the RGW pools we have > to them, then it just grows slowly with a bump at pretty much each > update. > > And this is a pretty small pool of OSDs, for theses there is only 8 > OSD processes over 4 nodes, storing barely 1 To in 2.5 millions > objects, split into 7 pools and 5376 PGs (some pools have size=3, > other size=2) > > 1.5 Go per OSD process seems a bit big to me. > > > Cheers, > > Sylvain > _______________________________________________ > ceph-users mailing list > ceph-users at lists.ceph.com > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |