On Wed, Sep 08, 2004 at 11:28:46AM +0930, Greg 'groggy' Lehey wrote: > >> [...] the parity surely is not correctly recalculated during > >> the revive. > > If that were the case, the parity would be incorrect at offset 0. > Yes, it is recalculated.
Of course -- I hadn't thought of that. > >> Greg, can you tell me if this is correct behaviour? > > Sorry for the slow response. I was at a conference last week. No, > it's not correct. No problem; this is still a volunteer project last time I checked. In a way I am glad to hear that it is not correct. > > While not having heard back yet, I had to rebuild another subdisk, > > but I decided to do it off-line this time. Turns out the parity was > > rebuilt ok. > > Yes, this is what I recommended. OK. > > Might there be a bug in the online rebuild code? > > Looks like it. > > The current version of Vinum is on its last legs. Lukas Ertl is > rewriting it, so don't expect much change in this version. For the > time being, just accept that you should umount before rebuilding a > plex. I will; it's just that somehow I was led to believe that I didn't need to do that. This has caused me some pain in the past. May I suggest applying the attached patch to /usr/src/sbin/vinum/vinum.8? At least it would prevent someone else from making the same mistakes as me. Thanks for your response, --Stijn -- The problem is that there are several people in design positions now who couldn't design the Next Big Thing(TM) unless it involved them taking a photocopier and someone else's design of The Next Big Thing(TM). -- 'Alkaiser' in a post on Slashdot on game originality
--- vinum.8.orig Wed Sep 8 06:47:46 2004 +++ vinum.8 Wed Sep 8 06:51:19 2004 @@ -441,6 +441,10 @@ .Ic checkparity prints a running progress report. .Pp +It is advisable to always check the parity of a RAID-4 or RAID-5 plex after +an unclean shutdown. Corrupt parity is as bad as degraded mode for such a +plex; if one of the subdisks of such a plex fails, data corruption will occur. +.Pp .It Xo .Ic concat .Op Fl f @@ -1046,6 +1050,11 @@ flag is specified, .Ic rebuildparity prints a running progress report. +.Pp +At present, a bug prevents rebuildparity from correctly completing its job +when the vinum volume is mounted and being accessed. You should only rebuild +the parity of plexes on unmounted volumes in order to guarantee correct parity +checks. .Pp .It Xo .Ic rename
pgp2yuf5apg1p.pgp
Description: PGP signature