On 01/05/2012 05:30 AM, MORITA Kazutaka wrote: > I confirmed some bugs. > > - If I took a cluster-wide snapshot just after cluster format, I > couldn't restore it. > > $ collie cluster format -b farm > $ collie cluster snapshot > $ qemu-img create sheepdog:test 128G > Formatting 'sheepdog:test', fmt=raw size=137438953472 > $ collie cluster snapshot -l > Index Snapshot Time > 1 Thu Jan 5 06:34:36 2012 > $ collie cluster snapshot -R 1 > Restore failed: I/O error > This is a real bug, we should refuse to perform snapshot operation when store is not formatted. > - 'collie vdi list' shows wrong information after restoring a > snapshot. > > $ collie cluster format -b farm > $ qemu-img create sheepdog:test 128G > Formatting 'sheepdog:test', fmt=raw size=137438953472 > $ collie cluster snapshot > $ qemu-img create sheepdog:test2 128G > Formatting 'sheepdog:test2', fmt=raw size=137438953472 > $ collie cluster snapshot -R 1 > Cluster restore to the snapshot 1 > $ collie vdi list > Name Id Size Used Shared Creation time VDI id > test 1 128 GB 0.0 MB 0.0 MB 2012-01-05 06:37 7c2b25 > Failed to read object 80fd381500000000 No object found > Failed to read inode header This is not a bug, we simply haven't implemented it yet. Currently, we just have basic snapshot support, that is, we can just restore farm's state, not sheep system state. I leave system state restore to future patch, since it is not trivial, because we have to take the membership change into consideration. We are supposed to try it after farm is merged. Thanks, Yuan |