[OpenAFS] Re: Volumes lost on fileserver / vldb post fs crash

[Jeffrey Hutzelman]

On Tuesday, April 27, 2004 09:01:05 +0200 Hartmut Reuter 
<reuter@rzg.mpg.de> wrote:

>
> Generally the salvager in MR-AFS can recreate lost directories filling
> them with entries lost+found.<vnode>.<uniquifier>. This is bettter than
> nothing.

The stock OpenAFS salvager won't rebuild directories, but it can reconnect 
"orphaned" files and directories that are not referenced by any directory. 
These will be attached in the root directory as __ORPHANFILE__.vnode.uniq 
or __ORPHANDIR__.vnode.uniq, as appropriate.  To enable this feature, 
invoke the salvager with '--orphans attach'.

> However, if the root directory is lost for some reason that doesn't work.
> There was two or three months ago a case on this list where someone
> had lost the root directory. I could help him by creating a new empty
> volume and then restoring incrementally a dump of his broken volume into
> this new one. The new volume had an empty root directory, of course. Now
> the
> MR-AFS salvager was able to reconnect the whole tree.
> The user only had to rename the lost+found entries in the root directory
> to the original ones.

This rule applies to stock OpenAFS as well -- the salvager simply cannot 
cope with a volume which has no root vnode.  The technique described above 
should be effective.

-- Jeffrey T. Hutzelman (N3NHS) <jhutz+@cmu.edu>
   Sr. Research Systems Programmer
   School of Computer Science - Research Computing Facility
   Carnegie Mellon University - Pittsburgh, PA