[OpenAFS] Re: Writing allowed where it's not expected

Andrew Deason adeason@sinenomine.net
Mon, 19 Sep 2011 10:39:13 -0400


On Mon, 19 Sep 2011 10:20:49 -0400
Jeffrey Altman <jaltman@your-file-system.com> wrote:

> On 9/19/2011 9:36 AM, Andrew Deason wrote:
> > On Sun, 18 Sep 2011 10:20:05 +0200
> > Dirk Heinrichs <dirk.heinrichs@altum.de> wrote:
> > 
> >> The only thing I did was to "vos release" _another_ volume that was
> >> mounted below .../sw and which showed up as "not released" in the
> >> output of "vos listvldb". Does this also count as "being on a
> >> read/write path"?
[...]
> The state of foo.readonly should not impact the evaluation of
> /afs/cell/sw.  There clearly is something weird going on here.

Yes, sorry, I read 'below' as 'above'. In this case, perhaps the client
still had old vldb information, which did not contain the RO site? The
'vos examine' info for the RO said:

    Creation    Sat Sep 17 09:41:04 2011
    Copy        Sat Sep 17 09:41:04 2011
    Backup      Never
    Last Access Sat Sep 17 09:40:59 2011
    Last Update Sat Sep 17 09:40:59 2011

And the original problem was seen around:

% touch sw/foo
% ll -g -n sw/foo             
-rw------- 1 100 0 2011-09-17 11:14 sw/foo

Which is less than two hours later. If around 9:40 on Saturday was the
first time that RO had existed, you need to wait about 2 hours to
guarantee all clients will "see" the new RO (or you can run 'fs checkv'
on specific clients, to not need to wait).

-- 
Andrew Deason
adeason@sinenomine.net