On 17-01-04 03:16, Gregory Farnum wrote:
On Fri, Dec 23, 2016 at 12:04 AM, Henrik Korkuc <lists@xxxxxxxxx> wrote:
Hello,
I wondered if Ceph can emit stats (via perf counters, statsd or in some
other way) IO and bandwidth stats per Ceph user? I was unable to find such
stats. I know that we can get at least some of these stats from RGW, but I'd
like to have something like that for RBD and CephFS. Example usage could be
figuring out who is hammering CephFS with IO requests.
Maybe someone could provide basic guidance where to dig in if I'd like to
make this feature myself? I there any plans/blueprints for such stats?
This really isn't feasible right now as a starter project: doing it
"for real" requires all OSDs track all clients and IOs and then a
central location to receive that data and correlate it. We discussed a
sampling implementation for ceph-mgr recently in the Ceph Dev Monthly
and I think work is upcoming to enable it, but I'm not sure what kind
of timeline it's on.
-Greg
I am thinking about implementing it myself. My high level idea is to
have additional thread per OSD to collect these stats and periodically
emit them to statsd for aggregation by metrics system. As I am not
familiar with Ceph code base it may take a while to implement.. :)
If there are some docs/plans for ceph-mgr metrics maybe I could
contribute to that instead doing it my way? This way we would be more
aligned and it would be easier to contribute it back to Ceph.
_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
