Re: [ceph-users] Replace all monitors
Le jeudi 08 août 2013 à 18:04 -0700, Sage Weil a écrit : On Fri, 9 Aug 2013, Olivier Bonvalet wrote: Le jeudi 08 ao?t 2013 ? 09:43 -0700, Sage Weil a ?crit : On Thu, 8 Aug 2013, Olivier Bonvalet wrote: Hi, from now I have 5 monitors which share slow SSD with several OSD journal. As a result, each data migration operation (reweight, recovery, etc) is very slow and the cluster is near down. So I have to change that. I'm looking to replace this 5 monitors by 3 new monitors, which still share (very fast) SSD with several OSD. I suppose it's not a good idea, since monitors should have a dedicated storage. What do you think about that ? Is it a better practice to have dedicated storage, but share CPU with Xen VM ? I think it's okay, as long as you aren't wroried about the device filling up and the monitors are on different hosts. Not sure to understand : by ?dedicated storage?, I was talking of the monitor. Can I put monitors on Xen ?host?, if they have dedicated storage ? Yeah, Xen would work fine here, although I'm not sure it is necessary. Just putting /var/lib/mon on a different storage device will probably be the most important piece. It sounds like it is storage contention, and not CPU contention, that is the source of your problems. sage Yop, the transition worked fine, thanks ! Newer mon are really fasters, and now I can migrate data without downtime. Good job devs ! Thanks again. Olivier ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
[ceph-users] Replace all monitors
Hi, from now I have 5 monitors which share slow SSD with several OSD journal. As a result, each data migration operation (reweight, recovery, etc) is very slow and the cluster is near down. So I have to change that. I'm looking to replace this 5 monitors by 3 new monitors, which still share (very fast) SSD with several OSD. I suppose it's not a good idea, since monitors should have a dedicated storage. What do you think about that ? Is it a better practice to have dedicated storage, but share CPU with Xen VM ? Second point, I'm not sure how to do that migration, without downtime. I was hoping to add the 3 new monitors, then progressively remove the 5 old monitors, but in the doc [1] indicate a special procedure for unhealthy cluster, which seem to be for clusters with damaged monitors, right ? In my case I only have dead PG [2] (#5226), from which I can't recover, but monitors are fine. Can I use the standard procedure ? Thanks, Olivier [1] http://ceph.com/docs/master/rados/operations/add-or-rm-mons/#removing-monitors [2] http://tracker.ceph.com/issues/5226 ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
Re: [ceph-users] Replace all monitors
On Thu, 8 Aug 2013, Olivier Bonvalet wrote: Hi, from now I have 5 monitors which share slow SSD with several OSD journal. As a result, each data migration operation (reweight, recovery, etc) is very slow and the cluster is near down. So I have to change that. I'm looking to replace this 5 monitors by 3 new monitors, which still share (very fast) SSD with several OSD. I suppose it's not a good idea, since monitors should have a dedicated storage. What do you think about that ? Is it a better practice to have dedicated storage, but share CPU with Xen VM ? I think it's okay, as long as you aren't wroried about the device filling up and the monitors are on different hosts. Second point, I'm not sure how to do that migration, without downtime. I was hoping to add the 3 new monitors, then progressively remove the 5 old monitors, but in the doc [1] indicate a special procedure for unhealthy cluster, which seem to be for clusters with damaged monitors, right ? In my case I only have dead PG [2] (#5226), from which I can't recover, but monitors are fine. Can I use the standard procedure ? The 'healthy' caveat in this case is about the monitor cluster; teh special procedure is only needed if you don't have enough healthy mons to form a quorum. The normal procedure should work just fine. sage ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
Re: [ceph-users] Replace all monitors
Le jeudi 08 août 2013 à 09:43 -0700, Sage Weil a écrit : On Thu, 8 Aug 2013, Olivier Bonvalet wrote: Hi, from now I have 5 monitors which share slow SSD with several OSD journal. As a result, each data migration operation (reweight, recovery, etc) is very slow and the cluster is near down. So I have to change that. I'm looking to replace this 5 monitors by 3 new monitors, which still share (very fast) SSD with several OSD. I suppose it's not a good idea, since monitors should have a dedicated storage. What do you think about that ? Is it a better practice to have dedicated storage, but share CPU with Xen VM ? I think it's okay, as long as you aren't wroried about the device filling up and the monitors are on different hosts. Not sure to understand : by «dedicated storage», I was talking of the monitor. Can I put monitors on Xen «host», if they have dedicated storage ? Second point, I'm not sure how to do that migration, without downtime. I was hoping to add the 3 new monitors, then progressively remove the 5 old monitors, but in the doc [1] indicate a special procedure for unhealthy cluster, which seem to be for clusters with damaged monitors, right ? In my case I only have dead PG [2] (#5226), from which I can't recover, but monitors are fine. Can I use the standard procedure ? The 'healthy' caveat in this case is about the monitor cluster; teh special procedure is only needed if you don't have enough healthy mons to form a quorum. The normal procedure should work just fine. Great, thanks ! sage ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com