On Thu, 2009-09-17 at 11:06 -0400, Stuart D. Gathman wrote: > Using dd from a snapshot does not help the OP. Right. > He says the dd will take > 5-6 hours, and he needs a backup of the system *just before* the upgrade, > not hours before. Which a snapshot will achieve. > The rollback also needs to be quick. Which a snapshot will achieve. > A snapshot with dd does the "instant" backup part ok, but rollback > requires 5-6 hours to copy the snapsnot to a normal LV. Why would you copy the snapshot anywhere? If you decided you needed a roll-back, you boot from the snapshot of the root filesystem and have the /etc/fstab in that snapshot mount any snapshots of other filesystems he might have made. Personally, I keep separate /, /usr, and /var and so snapshot them all before an upgrade and fix up the /etc/fstab in the snapshot-of-/ to mount the /usr and /var snapshots (rather than the origins). AFAICT, the only issue the OP has with snapshots is that he fears an upgrade might screw up his LVM configuration. Personally, I think that is a low-probability fear, but everyone has to have their own comfort level I guess. > The OP is correct that creating and breaking a RAID1 mirror does what > he needs. Indeed. > I always build my VGs on md* devices. I typically have not. I typically don't mirror anything except my backup volume, which fully (as in bare metal restore able) backs up every filesystem I have in my network. For that reason I feel that mirroring the disks in all the machines is overkill. But granted, building systems on broken mirrors is easy enough and allows one the flexibility of mirroring at a later date if one wants. In fact I always thought/felt, that given that the MD superblock is at the end of a device (or partition), shrinking the filesystem on a partition and then trying to create an MD mirror with that partition should be doable. Never tried though. b.
Description: This is a digitally signed message part