2016-05-23 10:31 GMT+08:00 Sharuzzaman Ahmat Raslan <sharuzzaman@xxxxxxxxx>: > is your service only have one instance? > are your service running on vm? about twenty krbd instance.they are map and mount in three osd machine. > On May 23, 2016 10:23 AM, "lin zhou" <hnuzhoulin2@xxxxxxxxx> wrote: >> >> Christian Balzer,thanks for your reply. >> >> you say,kernel upgrade will interrupt service.but AFAIK,umount rbd and >> strim >> it on another machine will interrupt service too. >> >> In my environment,all krbd are online business,service interruption is >> now allow. >> >> 2016-05-21 11:58 GMT+08:00 Christian Balzer <chibi@xxxxxxx>: >> > On Fri, 20 May 2016 17:00:03 +0800 lin zhou wrote: >> > >> >> Hi,cephers >> >> we only using krbd in ceph.and it works well near two yeas.but now I >> >> face a size problem. >> >> >> >> I have 7 nodes with 10 3T osd each.we using ceph 0.67.9 in >> >> ubuntu12.04.I know it is too old,but update is beyond my control. >> >> >> >> and now we use 80% size,so we start to delete historic unneeded >> >> data,but free size do not increase. >> >> >> >> and then I found total assigned size of rbds are much bigger than >> >> ceph size,so it means if we do not do something,it will reach the >> >> limit. >> >> >> >> but the true data in the user side,we only use 40%. >> >> >> >> I read the blog of Sebastien and some maillist.I know the command of >> >> fstrim in kernel 3.18 can deal with it. and is it risky that update >> >> kernel to 3.18 in ubuntu12.04? >> >> >> >> I try to add discard option in mount command,but it do not work. >> >> >> >> so what way do you recommend to free krbd size in ceph 0.67.9 in >> >> ubuntu12.04 >> >> >> > >> > If you're considering a kernel upgrade (and thus a service >> > interruption), >> > another option would be to unmount that image on your old machine, mount >> > it on a newer machine and/or with librbd and fuse and then run fstrim. >> > >> > However trim is a pretty costly activity in Ceph, so it may >> > a) impact your cluster performance and >> > b) take a while, depending on how much data we're talking about. >> > >> > Lastly, while having a sparse storage serice like Ceph is very nice I >> > always try to have enough actual space available to handle all >> > commitments. >> > >> > Christian >> > -- >> > Christian Balzer Network/Systems Engineer >> > chibi@xxxxxxx Global OnLine Japan/Rakuten Communications >> > http://www.gol.com/ >> _______________________________________________ >> ceph-users mailing list >> ceph-users@xxxxxxxxxxxxxx >> http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |