[OpenAFS] Re: (minor) VLDB corruption - how to clean up/repair?

Michael Meffie mmeffie@sinenomine.net
Fri, 7 Nov 2014 16:13:32 -0500


On Fri, 7 Nov 2014 18:40:13 +0100
Jan Iven <jan.iven@cern.ch> wrote:

> On 11/07/2014 05:13 PM, Andrew Deason wrote:
> > On Fri, 7 Nov 2014 16:40:51 +0100
> > Jan Iven <jan.iven@cern.ch> wrote:
> >
> >> Are there any other "just do it"-level commands or flags that might
> >> allow me to get unstuck via command line tools, or is this potentially
> >> "hexedit time" (in which case I might be inclined to let sleeping dogs
> >> rest a while)?
> >
> > There is 'vldb_check -fix', which retries to rebuild the hash chains etc
> > to give you a more useful db. But make copies of the db and verify the
> > changes it made afterwards; it's not a well-exercised code path. You
> > also may want to grab the vldb_check from 1.6.10; iirc there have been
> > some fixes for it in the last few versions.
> 
> Many thanks. Indeed "-fix" does several changes, but also to other 
> volumes/accounts (so will have to check whether these were 
> accessible/reported as corrupted before or not, and dump+compare the 
> VLDB entries just in case..).

Yes, that seems likely, since the hash chains were rebuilt, serveral
vl entries could be changed.
 
> 1.6.10-2 does not fix things "differently" in our case (same md5sum on 
> the resulting DB file)

The recent changes to check_vldb were so sort out the internal references
to server information, so may not have effected your situation.

Thanks,
Mike

-- 
Michael Meffie <mmeffie@sinenomine.net>