On Wed, 26 Oct 2016, Trygve Vea wrote:
> Hi,
> 
> We have two Ceph-clusters, one exposing pools both for RGW and RBD 
> (OpenStack/KVM) pools - and one only for RBD.
> 
> After upgrading both to Jewel, we have seen a significantly increased CPU 
> footprint on the OSDs that are a part of the cluster which includes RGW.
> 
> This graph illustrates this: http://i.imgur.com/Z81LW5Y.png

That looks pretty significant!

This doesn't ring any bells--I don't think it's something we've seen.  Can 
you do a 'perf top -p `pidof ceph-osd`' on one of the OSDs and grab a 
snapshot of the output?  It would be nice to compare to hammer but I 
expect you've long since upgraded all of the OSDs...

sage


> 
> 
> I wonder if anyone else have seen this behaviour, and if this is a symptom of 
> a regression --- or if this was to be expected after moving from hammer to 
> jewel.
> 
> I have also observed that an OSD will occasionally be marked as down, but 
> will recover by itself.
> 
> This manifests itself in the osd logs as a series of lines along this:
> 
> 2016-10-26 06:32:20.106602 7fa57a942700  1 heartbeat_map is_healthy 
> 'OSD::osd_op_tp thread 0x7fa575938700' had timed out after 15
> 
> Some slow requests may be observed:
> 
> 2016-10-26 06:32:35.899597 7fa5aa41b700  0 log_channel(cluster) log [WRN] : 1 
> slow requests, 1 included below; oldest blocked for > 30.905777 secs
> 2016-10-26 06:32:35.899605 7fa5aa41b700  0 log_channel(cluster) log [WRN] : 
> slow request 30.905777 seconds old, received at 2016-10-26 06:32:04.993791: 
> replica scrub(pg: 
> 3.2e,from:0'0,to:27810'772752,epoch:28538,start:3:74000000::::head,end:3:7400039b::::0,chunky:1,deep:1,seed:4294967295,version:6)
>  currently reached_pg
> 
> Some failing heartbeat_checks (usually only from a single osd):
> 
> 2016-10-26 06:32:39.323412 7fa56f92c700 -1 osd.19 28538 heartbeat_check: no 
> reply from osd.15 since back 2016-10-26 06:32:19.017249 front 2016-10-26 
> 06:32:19.017249 (cutoff 2016-10-26 06:32:19.323409)
> 
> 
> A bunch of these (with the remote address targetting different osds):
> 
> 2016-10-26 06:32:45.522391 7fa598ec0700  0 -- 169.254.169.254:6812/151031797 
> >> 169.254.169.255:6802/41700 pipe(0x7fa5ebba7400 sd=160 :6812 s=2 pgs=4298 
> cs=1 l=0 c=0x7fa5d7c26400).fault with nothing to send, going to standby
> 
> 2016-10-26 06:32:45.525524 7fa5a5158700  0 log_channel(cluster) log [WRN] : 
> map e28540 wrongly marked me down
> 
> Followed by repeering, and then everything is fine again.
> 
> 
> 
> I wonder if anyone have been suffering from similar behaviour, if this is a 
> bug (known or unknown).  One detail to keep in mind is that the osds for the 
> rgw pools store replicas on different physical sites.  However, we have no 
> reason to believe that saturation or high latency is a problem.
> 
> 
> 
> Regards
> -- 
> Trygve Vea
> Redpill Linpro AS
> _______________________________________________
> ceph-users mailing list
> [email protected]
> http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
> 
> 
_______________________________________________
ceph-users mailing list
[email protected]
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com

Reply via email to