At Fri, 24 Aug 2012 01:37:11 +0800, Yunkai Zhang wrote: > > If correct one have been removed, how can fsck fix that? (I'm not > familiar with this command, I just have some doubts.) In my opinion, either of replicas is correct one. As long as sheepdog provides block device semantics, the guest os filesystem should handle any kind of I/O errors with a journaling or fsck command. > > > > > I don't fully understand yet what situation you want to avoid. What > > kinds of problem happens to you if you fix consistency with wrong > > objects? > > If fsck can help me to fix, I have no question, and this patch isn't > very important for me. I really need a concrete example. - What is the scenario where the consistency problem you are trying to solve happen? - In the scenario, how do you decide which object is correct? IMO, either is correct in any cases. - What kind of problems are you expecting when VM reads the wrong object? Thanks, Kazutaka > > I just want to try our best to protect sheepdog's data integrity, I > haven't encountered any situation in real work, I didn't even > encountered VM's unexpected termination. > > If fix consistency with wrong objects is really not harm, I do not > mind to drop this patch. > > > > > > Thanks, > > > > Kazutaka > > > > -- > Yunkai Zhang > Work at Taobao > -- > sheepdog mailing list > sheepdog at lists.wpkg.org > http://lists.wpkg.org/mailman/listinfo/sheepdog |