Hello, I have read much about ZFS and I find it great, especially the checksums against silent data corruption and the COW writing policy and the snapshots and of course the storage pooling. But there are some points I have problems with: - What happens to the pools if the machine is shut down and rebooted? Are the pools automatically exported and imported again on boot up? Where is the information about the pools kept? What component is responsible for automatic export/import? - There seem to be a lack of tools or options for recovering when ZFS refuses to import a pool for some reason. I have seen many forum posts about problems of that kind. I''m worried about that. Perhaps it might also be a good idea if you could force ZFS to forget about a device and to deal with the remaining disks. I think it should be always possible to access the data which lies on fully functional disks. Wasn''t it also a feature of ZFS that it can be used even when parts of the metadata are missing or broken? - What about data wiping? Simply overriding does not work because of the COW policy. So how can it be achieved? I think it should be considered to integrate some mechanism to intentionally override data. - I think ZFS needs more flexibility in vdev management. It bothers me that you cannot remove a disk from a (non-mirror) pool although there is enough space to migrate the data stripes. Well, I heared that this feature is planned for the future. - A flexible encryption support would also be great. (I know that it is planned.) This message posted from opensolaris.org
not available wrote:> - What about data wiping? Simply overriding does not work because of the COW policy. So how can it be achieved? I think it should be considered to integrate some mechanism to intentionally override data.See the archives of this list, there was a very long discussion about this almost a year ago now. http://mail.opensolaris.org/pipermail/zfs-discuss/2006-December/036650.html -- Darren J Moffat
not available wrote:> But there are some points I have problems with: > - What happens to the pools if the machine is shut down and rebooted? Are the pools automatically exported and imported again on boot up? Where is the information about the pools kept? What component is responsible for automatic export/import?Yes the pools are properly dealt with on a reboot. What makes you think this wouldn''t be the case ? Do you have a specific case where you believe it has failed ? There is a private state file that caches the pools held in /etc/zfs/zpool.cache so that zfs doesn''t have to look on every attached disk. -- Darren J Moffat
Thanks for your answers so far.> Yes the pools are properly dealt with on a reboot. > What makes you think > this wouldn''t be the case ? Do you have a specific > case where you > believe it has failed ?Well, I have to admit I only played around a little bit with zfs-fuse so far. And I don''t manage to re-import my pools after a reboot. I don''t know what''s going wrong. Last time I tried a mirrored pool using two files I got this: pool: testpool id: 15927372805316771887 state: FAULTED action: The pool cannot be imported due to damaged devices or data. config: testpool UNAVAIL insufficient replicas mirror UNAVAIL corrupted data /mnt/data/linux/zfstest1.img ONLINE /mnt/data/linux/zfstest2.img ONLINE I find that strange because both "disks" are online but zpool says "insufficient replicas". Of course, I''m sure that those things work fine on Solaris. :) And how would I recover my data in such a case when the disks might be fine but zpool refuses to import? This message posted from opensolaris.org
> What do *you* mean by flexible in this context ?> ZFS Crypto is by design extensible to new crypto algorithms and modes. > It is also designed to allow for multiple different key management > strategies and implementations, and in fact is explicitly a multiple > phase project for this reason.Yeah, by "flexible" I meant just that what you said: an extensible design to support different crypto algorithms and so on. Sorry, I have no overview about the current state of zfs development. This message posted from opensolaris.org