On Sat, 31 Aug 2013, Geraint Jones wrote: > Hi > > After the fun we had the other night we have also found that the SSD's > acting as the journal for our OSD's are not up to the task. So along with > adding some 10gbit I am going to add some more/faster SSD's. > > Having done some investigation it seems that the best way to do this is to > do the following > 1. Shutdown the server as normal, Add the SSD(s) and bring it back up. I > understand this can make things a little slow during the reboot etc, but > that should not take particularly long. > 2. $ ceph osd set noout <- this is confusing me, we have osd.0 -> osd.8 on > this sever, will it noout all the osd's ? This sets a global flag that prevents the monitor from marking OSDs out after they have been down for more than 5 minutes. Do this once, then redo all of the osd journals and restart the daemons, and then ceph osd unset noout at the very end of the process. sage > 3. $ service ceph stop osd.X > 4. $ ceph-osd ?i X ?flush-journal > 5. change the symlink for the sod's journal to point to the new SSD > 6. $ ceph-osd ?i X ?mkjournal > 7. $ service ceph start osd.X > 8. $ ceph osd unset noout / goto 2. <- Like above do I do this once for the > host, then do all work or is this done per osd? > If anyone has any thoughts on the above procedure I would be keen to hear > them. > >
_______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
