On Thu, Oct 18, 2001 at 12:20:22PM -0600, Andreas Dilger wrote: > This may be possible to do in the short term with writable snapshots, > but then you start getting into a very ugly situation where the "good" > volume (i.e. the writable snapshot) depends on the snapshot, which > depends on the original "bad" volume and you can't get rid of any of them. I suppose the ideal solution is to allow people to choose which way round a snapshot is taken depending on whether they thinks it's likely they'll rollback, or commit to the snapshot. ie. You probably want to rollback; the snapshot lv contains new data written to the snapshot (as it is now). Rollback is quick, commit (merge) is slow. You probably want to commit to the snapshot; the snapshot lv contains copies of the original data. Rollback (merge) is slow, commit is quick. - Joe
