[OpenAFS] Re: OpenAFS algorithm

Andrew Deason adeason@sinenomine.net
Tue, 20 Apr 2010 10:14:53 -0500

On Tue, 20 Apr 2010 11:36:12 +0200
Fabien COMBERNOUS <fcombernous@kezia.com> wrote:

> I have two servers mpt and lyn. The two servers are file server and db
> server.  The two servers hosts a RO copy of my volume. Only mpt have a
> RW copy.
> On a client near lyn, fs where give me only mpt, and fs gets also.
> How to add the other server lyn ?

You need to access the RO volume, but for some reason, your client is
accessing the RW (which only exists at mpt).

One of the most common reasons this occurs is that you have an RW volume
somewhere in the path to the volume mount point. For example, say you
are looking at /afs/localcell/vol.foo/. If the /afs/localcell mountpoint
is an RW mountpoint, /afs/localcell/vol.foo will give you RW data, even
if /afs/localcell/vol.foo is a normal mountpoint and you have RO copies
of vol.foo.

Look at every path component to the mountpoint of the volume (look at
/afs/, /afs/cell.name/, /afs/cell.name/path/to/vol, etc), and run
'fs exam <path>' on each one. Where do you stop getting volume names
that look like 'foo.readonly'?

Andrew Deason