在 2012-8-24 AM12:34，"Yunkai Zhang" <yunkai.me at gmail.com>写道： > > On Thu, Aug 23, 2012 at 10:20 PM, MORITA Kazutaka > <morita.kazutaka at lab.ntt.co.jp> wrote: > > At Thu, 23 Aug 2012 21:46:55 +0800, > > Yunkai Zhang wrote: > >> > >> When inconsistency happens, in the worst case, we just need to make > >> replicas with the same value but needn't to care about which replica > >> is correct. > >> > >> But in the best case, only several objects were polluted, then this > >> patch can give us a chance to verify them, This patch is used to work > >> with 'collie vdi check' command, that is my original intention when > >> developed it. > > > > Well, replica inconsistency happens only when the VM terminates > > unexpectedly. If the vm image needs a fix, the guest os filesystem > > should fix it. What sheepdog should do is only keeping consistency so > > Can you explain how the guest os files system fix it? > > I had discussed with Yuan, he told me that these inconsistent replicas maybe you misunderstand me. I didn't say it's useless. We can't remove them at all. I think what Ive said was similar to Kazum Yuan > is useless for VM. If so, I will have another question: why not delete > it directly? Or if these replicas are useful data, How can we randomly > select a replica? > > > that the VMs doesn't read the wrong data as a block storage, isn't it? > > Have you ever encountered a situation where you need to specify a > > replica to be recovered? > > > > Thanks, > > > > Kazutaka > > > > -- > Yunkai Zhang > Work at Taobao > -- > sheepdog mailing list > sheepdog at lists.wpkg.org > http://lists.wpkg.org/mailman/listinfo/sheepdog -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.wpkg.org/pipermail/sheepdog/attachments/20120824/ac99c7e3/attachment.html> |