[OpenAFS] re doing rm in /vice*

rader@hep.wisc.edu rader@hep.wisc.edu
Thu, 23 Aug 2012 14:13:49 -0500

 > > We have an AFS filserver running openafs-server-1.4.14-80.1.sl5.
 > > We had to salvage one of it's RW volumes after a reboot.
 > > But then it's backup volume that wouldn't attach, "needs to be salvaged"
 > > Yet salvage said "read-only volume; not salvaged"
 > > I couldn't vos remove nor vos zap.
 > > So I removed the .vol file and did "bos restart"
 > > Then I was able to revive the volume with "vos backup".
 > > I've been using this shootgun approach for horked volumes for over a decade.
 > > (mostly in "can't remove" or "can't clean up after ab-ended move" contexts.)
 > > So, I'm wondering: should I be a little more gentle?  How??
 > Gentleness is always good. As others have mentioned, you need to salvage the RW vol,
 > not the RO.

Right, the RW WAS salavaged... but only once.  I considered doing it again
but didn't want to take the volume offline.  I guess next time (if there is
one) I'll try that.

 > Since you were destroying and recreating the .backup volume as part of the process, did
 > you try doing a 'vos backup' on the volume?

Yes, it/they (vos backup and vos backupsys) was/were failing.

 > Why you couldn't vos zap a .backup volume seems odd to me. It works on our systems
 > (1.4.16). Doing 'vos remove <volname>.backup' also worked.
 > Removing the .vol file does have some downsides. In particular, it eventually convinces
 > the server that it doesn't have a copy of the volume, but you never get back whatever
 > space was actually consumed by files which older versions lying around. If that's a
 > small amount, you can ignore it. If it's a significant amount, you may want that back.
 > The only way I know of to get it is to vos move everything to another server, then
 > salvage the entire partition. Vague memory tells me that doesn't always help; 

Yes, that's been my experience: wacking volumes by removing the .vol file
and doing salvage on the file system won't get the disk space back.

 > in such
 > cases I had to take the partition offline and do a mkfs on it. I suppose one could
 > simply go in and manually delete everything in /vicepX/AFSIDat/* that isn't a
 > directory, but that idea always gives me hives.

I would think deleting everything in /vicepX/AFSIDat/ but leaving the .vol files 
around would give you more than just hives.