[OpenAFS] Systems switching to read/write

[Russ Allbery]

Has anyone encountered a problem before with systems suddenly deciding to
look at the read/write versions of all of the volumes in one's cell rather
than the read-only version?

We have a large cluster of Unix systems running Solaris 8, and
periodically a large chunk of those systems (but not all of them) suddenly
decide that root.cell isn't replicated and decide that
/afs/ir.stanford.edu is root.cell and not root.cell.readonly.  Running:

    cd /afs
    fs flush .
    fs flushmount ir.stanford.edu
    fs flush .
    fs lq ir.stanford.edu

fixes the problem (I'm not sure what of all that is necessary).  I'm not
sure what's causing it; I thought it was related to a power outage that
brought down a lot of our cell at the same time a while back, but a bunch
of the systems just did this again and there was no similar event this
time.

afsd is using the options:

-dynroot -afsdb -nosettime -stat 2000 -dcache 800 -daemons 3 -volumes 70

Could this be some sort of bug in the -dynroot support?  We only recently
switched to using -dynroot everywhere.

These systems are running OpenAFS 1.2.8 (please let me know if this is a
known bug fixed in 1.2.9).

-- 
Russ Allbery (rra@stanford.edu)             <http://www.eyrie.org/~eagle/>