At Tue, 12 Nov 2013 14:28:17 +0800, Liu Yuan wrote: > > On Tue, Nov 12, 2013 at 03:22:27PM +0900, Hitoshi Mitake wrote: > > At Tue, 12 Nov 2013 11:09:03 +0800, > > Liu Yuan wrote: > > > > > > On Tue, Nov 12, 2013 at 11:45:23AM +0900, Hitoshi Mitake wrote: > > > > Current object repair strategy of "dog vdi check" doesn't work well if > > > > objects are corrupted. The first one adds a test for the case, and the > > > > second patch adds a mechanism for majority voting in the command. > > > > > > > > v5: drop ignore change in v4 > > > > > > Applied after fixing new blank line at EOF for 077 > > > > Oops, thanks for your fix. > > > > > > > > I'd like to see patche(s) that deal with the case when majority object > > > isn't > > > found. What in my head to deal with this is: > > > > > > - add an option 'depth' for 'dog vdi track' command to get the > > > information of > > > targeted object placement. > > > > > > With this information, people can do a manual copy\overwrite to > > > missing/inconsistent > > > objects. > > > > > > > I agree. In addition, corrupted or lost inode objects should be > > repaired by dog vdi repair. I'll implement them later. > > > > I guess you mean 'vdi check'. > > Current code indeed fix inode object, no?
Ah, yes, I missed the line in do_vdi_check(). But it seems that we need a new test for corrupted/lost inode object. I'll write it later. Thanks, Hitoshi -- sheepdog mailing list [email protected] http://lists.wpkg.org/mailman/listinfo/sheepdog
