From: Pasi K. <pa...@ik...> - 2009-04-19 13:21:40
|
On Sun, Apr 19, 2009 at 03:39:10PM +0300, Pasi Kärkkäinen wrote: > On Wed, Apr 15, 2009 at 05:53:18PM -0400, Ross S. W. Walker wrote: > > Pasi Kärkkäinen wrote: > > > On Tue, Apr 14, 2009 at 06:55:49PM -0400, Ross S. W. Walker wrote: > > > > On Apr 14, 2009, at 9:46 AM, "Mark Hunting" <ma...@ne...> wrote: > > > > > > > > >Hi all, > > > > > > > > > >I use IET to export LVM (well, LVM2) volumes as iSCSI devices. On the > > > > >initiator I use the iSCSI devices as disks for KVM (Kernel-based > > > > >Virtual > > > > >Machine). I can create a snapshot of a LVM volume on the target just > > > > >fine. But can I be 100% sure that the resulting snapshot contains > > > > >consistent data? For example, when a database on the KVM is just > > > > >writing > > > > >data when I make the snapshot, is there a change that the snapshot > > > > >will be broken/non-consistent? > > > > > > > > > >When using for example a local ext3 filesystem on a LVM volume the > > > > >filesystem halts for a moment when the snapshot is created. Does IET > > > > >also do this? > > > > > > > > > Actually this really has nothing to do with IET as IET simply acts as > > > > a disk drive and snapshots aren't done at the disk drive level. > > > > > > > > You would need an LVM process to communicate somehow with the > > > > initiators to tell them, ok, it's snapshot time and that can be very > > > > tricky. > > > > > > > > CLVM still doesn't support cluster snapshots to this day for this > > > > reason. > > > > > > > > Another idea is another target or local disk on the initiator for > > > > snapshots and doing all snapshots on the initiator where applications > > > > can get notification to quiesce disk activity. Local disks for > > > > snapshots would reduce network utilization and probably increase > > > > throughput. Having these on the target side would slow operations for > > > > all targets in the volume group due to the overhead of the extra COW > > > > operations. > > > > > > > > > > Hello! > > > > > > Ross: Are you aware of any Windows VSS style projects for Linux? > > > > No unfortunately not. I don't think Linux would actually use VSS > > specification because I don't think it is GPL'd. Somebody could > > write a userland daemon though that would interface to Linux's > > "whatever" library for LVM events and send those to Windows > > hosts using Microsoft's VDS specification. Then the VSS writers > > on those hosts could quiesce there services (File, SQL, Exchange). > > > > OK. Yeah, I didn't meant to implement exactly VSS, but something like that.. > > > Those individuals who write a VDS daemon could also conceivably > > have it listen for RPC/IPC commands from the Windows hosts per > > the VDS specification and act on those with backend scripts. > > > > Unfortunately you need both Linux daemons and Windows COM objects > > for this to work. > > > > Hmm.. is VDS requires for VSS? > > If I understood you correctly I was thinking of something different.. > I was mostly thinking of the following scenario: > > - Backup agent on the Linux server starts a backup, as requested by the backup server > - Backup agent on the Linux server calls VSS-style library/API to prepare for snapshot > - VSS-style library signals all the 'registered' applications on the Linux server to prepare for snapshot.. flush their buffers etc. > - VSS-style library flushes/freezes the filesystem > - VSS-style library calls the ietd-specific SAN-plugin to take a snapshot on the SAN (target) > - when the snapshot is taken, VSS-style library signals the applications they can continue running normally > - Backup agent (or server) continues the backup from the snapshot > Of course it would be nice to have VSS/VDS support for ietd target in Windows initiators aswell :) But yeah, I was mostly thinking of Linux initiators in this discussion.. -- Pasi |