We are running a similar 2-site environment, too, and we had the same
ideas about AFS database availability already when we installed AFS
more than 3 years ago. I am sure I mentioned these ideas to Transarc,
too, where they obviously disappeared in /dev/null. Luckily, our
network connection between the 2 sites proved stable enough so that
this never became a real problem for us. Otherwise, we today would
probably have 2 AFS cells with identical KAS and PTS entries, some
obscure mechanism to shadow RW volumes between both cells, and some
hack to automatically obtain tokens for both cells at login.
You might try a workaround that detects partitioning of your cell and
in that case restarts your database servers with a modified
/usr/afs/etc/CellServDB, making them believe that there are no servers
but those at the same site. When the connection comes back, you would
have to somehow resynchronize the databases on both sites, or restore
backups from before the partitioning. The latter might be tolerable for
KAS and PTS, but if someone created, released or removed volumes during
the partitioning, you will certainly end up with a mess in your VLDB.
You might try to mount /usr/afs/db read-only during the partitioning,
but I doubt that the database servers will tolerate this.
--
Michael Niksch TEL: +41-1-7248-913
IBM Zurich Research Laboratory FAX: +41-1-7103608
Saeumerstrasse 4 BITNET/VNET: NIK at ZURICH
CH-8803 Rueschlikon / Switzerland INTERNET/IBMIPNET: [EMAIL PROTECTED]
