For Giant, we have changed the default librbd caching options to: rbd cache = true rbd cache writethrough until flush = true
The second option enables the cache for reads but does writethrough until we observe a FLUSH command come through, which implies that the guest OS is issuing barriers. This doesn't guarantee they are doing it properly, of course, but it means they are at least trying. Once we see a flush, we infer that writeback is safe. sage On Sat, 23 Aug 2014, Alexandre DERUMIER wrote: > >>But what about Windows? Does NTFS support barriers too? > > Windows > 2003 support FUA (like in newer linux kernel). So it's safe. > > virtio-win driver support it too since 1 or 2 year. > > I have had a discuss about it some year ago, see : > > https://github.com/YanVugenfirer/kvm-guest-drivers-windows/issues/3 > > > ----- Mail original ----- > > De: "Yufang" <yufang521...@gmail.com> > ?: "Alexandre DERUMIER" <aderum...@odiso.com> > Cc: ceph-users@lists.ceph.com > Envoy?: Vendredi 22 Ao?t 2014 18:05:32 > Objet: Re: [ceph-users] Is it safe to enable rbd cache with qemu? > > Thanks, Alexandre. But what about Windows? Does NTFS support barriers too? > Should I have confident that win2k3 guest could survive from data loss on > host/guest crash? > ???? iPad > > > ? 2014?8?22??23:07?Alexandre DERUMIER <aderum...@odiso.com> ??? > > > > Hi, > > for RHEL5, I'm not sure > > > > be barriers supported is maybe not implemented in virtio devices,lvm,dm > > raid and some filesystem, > > depend of the kernel version. > > > > Not sure what is backported in rhel5 kernel > > > > > > see > > http://monolight.cc/2011/06/barriers-caches-filesystems/ > > > > > > > > ----- Mail original ----- > > > > De: "Yufang Zhang" <yufang521...@gmail.com> > > ?: ceph-users@lists.ceph.com > > Envoy?: Vendredi 22 Ao?t 2014 13:05:02 > > Objet: [ceph-users] Is it safe to enable rbd cache with qemu? > > > > > > Hi guys, > > > > > > Apologize if this question has been asked before. I'd like to know if it is > > safe to enable rbd cache with qemu (cache mode set as writeback) in > > production? Currently, there are 4 types of guest os supported in our > > production: REHL5, RHEL6, Win2k3, Win2k8. Our host is RHEL6.2 on which qemu > > supports ' barrier-passing'. Thus we have confident that RHEL6 guests(with > > barrier enabled by default ) could work well with rbd cache enabled. But as > > for REHL5, Win2k3 and Win2k8, I am not sure if it is 100% safe on scenarios > > such as guest crash, host crash or power loss. Could anybody give some > > suggestion? Really appreciate your help. > > > > > > Yufang > > _______________________________________________ > > ceph-users mailing list > > ceph-users@lists.ceph.com > > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com > _______________________________________________ > ceph-users mailing list > ceph-users@lists.ceph.com > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com > _______________________________________________ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
