So, I had a fun ZFS learning experience a few months ago. A server of mine suddenly dropped off the network, or so it seemed. It was an OpenSolaris 2008.05 box serving up samba shares from a ZFS pool, but it noticed too many checksum errors and so decided it was time to take the pool down so as to save the (apparently) dying disk from further damage. Seemed inconvenient at the time, but a in hindsight that''s a cool feature. Haven''t actually found any problems with the drive (an SSD), which has worked fine ever since. Bit rot? Power failure (we had a lot of those for a while)? Who knows. At first I was afraid my ZFS pool had corrupted itself until I realized that it was a unique feature of ZFS actually protecting me from further damage rather than ZFS itself being the problem. At any rate, in this case, the corruption managed to make it over to my backup server replicated with SNDR. One of the corrupted blocks happened to be referenced by every single one of my daily snapshots going back nearly a year. I had no mirrored storage and copies set to 1. Arguably a bad setup, I''m sure, but that''s why I had a replicated server. At any rate, I didn''t care about the file referencing the corrupt block. I would just as well have deleted it, but it was still referenced by all the snapshots. It was a crisis at the time, so I just switched over to my replicated server (in case the drive on the primary server actually was bad) and deleted the files containing corrupt blocks and then deleted all the snapshots so zfs would quit unmounting the pool and just to get going again. Things have been fine ever since, but I still wonder - is there something different that I could have done to get rid of the corrupt blocks without losing all my snapshots (could have restored them from backup, but it would have taken forever). I guess I could just do clones and then have the capability of deleting stuff, but then I don''t believe I''d be able to back the thing up - if I don''t do incremental zfs send/recv, the backup takes over 24 hours since there''s so many snapshots, and I wouldn''t think clones work with incremental zfs send/recv (especially if you start deleting files willy-nilly). Am I just missing something altogether, or is restoring from backup the only option? -- This message posted from opensolaris.org