[OpenAFS] RO volumes disappearing?

Derrick J Brashear shadow@dementia.org
Fri, 12 Dec 2003 14:10:33 -0500 (EST)


On Fri, 12 Dec 2003, Christopher Arnold wrote:

>
> Recently I've noticed an odd problem in our environment and can't seem to
> track down the source.   We are running 1.2.10 servers/clients on redhat linux 7.3/8.0 in
> two locations connected via a T3.  We have 2 read-only sites created for
> each volume at both of our two locations, most of which get released at least once
> per day.  We are seeing that clients connected to us on the other
> end of the T3 (facility B) are randomly losing access to the RO volumes (i suspect during
> vos releases but not sure). If I cd into the readonly path from the clients that are at facility B
> the directory will be completely empty if I do an ls however the RW path will still look fine. At facility A
> (closest to the RW copy) the RO path and .readonly volumes display contents fine.
> If I then zap all the read-onlys,re-release, and fs flushvolume on clients at facility B
> the problem goes away only to reappear with another random volume later in
> the day. This even happens with volumes which had been released to all their RO sites
> successfully at some point in the past so even if vos release wasn't completing I thought
> the old data would still be accessible to remote clients closest to those RO copies.  Anyone have any
> ideas what might be causing this behavior or where I should look for clues?

Sure, apply this patch (you'll need to reformat it slightly to apply to
1.2.10) and rebuild:
http://www.openafs.org/cgi-bin/wdelta/MAIN/namei-ihandle-leak-20031114