>> > > Just out of interest do any of your journals or disks look like they are getting maxed out? > > Your latency breakdown seems to indicate that the bulk of requests are being serviced in reasonable time, but around 5% (or less) are taking excessively long for some reason. > > I'm wondering if something is causing a backlog on the journals? > > In regards to your query about the delay between the applied and done I wonder if this is just showing the time that the data was sitting in journal waiting to be flushed? Once the write has been committed by the journal, from a clients perspective the write is complete. Any thing in particular I should be looking at here? I have the collectd ceph plugin running and am watching the max journal latency and it doesn’t go very high. Journals exist on the OSD disks rather than on separate disks. And I agree, most of the latencies are pretty low, unless I start hitting the cluster with all 7 load generators in which case I get a lot of very high latencies. _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |