On 01/23/2014 06:39 AM, Sylvain Munaut wrote:
Hi,
I have a cluster that contains 16 OSDs spread over 4 physical
machines. Each machines runs 4 OSD process.
Among those, one isue periodically using 100% of the CPU.
I finally tracked this down. CPU usage was mostly from leveldb calls
Andrey Korolyov (xdeller on IRC) pointed out they had a similar issue
and upgrading leveldb stable to >= 1.14 solved the issue.
I was rebuilding my own packages and the debian/control file just
depends on leveldb-dev without any particular version requirement so
it was using the one that comes from ubuntu precise which is an old
2012 version. I backported 1.14 from debian to precise and rebuilt
against that and this seems to have fixed my issues as well.
Thanks for checking this out Sylvain. We've seen leveldb use quite a
bit of CPU in the past too at times, specifically during object
iteration and memory compaction.
Cheers,
Sylvain
_______________________________________________
ceph-users mailing list
ceph-users@lists.ceph.com
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
_______________________________________________
ceph-users mailing list
ceph-users@lists.ceph.com
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
