On Wednesday, 8 December 2004 at 11:52:55 +0100, Stijn Hoop wrote: > Scott, your procedure is what I have used, except for: > > On Wed, Dec 08, 2004 at 10:09:05AM +0000, Scott Mitchell wrote: >> 6. Tell vinum to restart the failed subdisk: >> >> # vinum start raid.p0.s0 >> >> 7. Wait ages while the new disk is 'revived'. >> >> I was quite impressed that the volume remained available with users >> accessing it throughout this procedure :-) > > Yes I was too -- however I wasn't as impressed with the fact that I had parity > errors afterwards. Have you run 'vinum checkparity' after these rebuilds? In > my case I suffered data corruption...
*sigh* Yes, there's some problem there. > AFAIK the only way to guarantee a consistent rebuild is to do it > offline (at least in 4.x, haven't tested gvinum in 5.x yet). > >> To play it safe you might want to unmount the volume before starting. > > I *have* to. The issue is contention round where stripes are being written. The code *should* avoid the contention, but it appears that there's a bug there somewhere. I certainly agree with you that you should umount the file system first. There's no reason to believe that this problem exists in gvinum: I believe the code has been completely rewritten. Greg -- See complete headers for address and phone numbers.
pgpIdH4cerh1f.pgp
Description: PGP signature