I've been testing out the latest OpenAFS snapshots for Linux 2.6 related issues and just came across a default configuration issue that will cause people a lot of trouble. With the 1.3.7x branch, the default Linux configuration file (src/afsd/afs.conf.linux) now specifies the CACHESIZE= directive in the default install as 'CACHESIZE=AUTOMATIC'. However the logic in the init script (src/afsd/afs.rc.linux) will exit with an error if CACHESIZE=AUTOMATIC and $CACHE is on / or /usr. This will result in an inoperable AFS because the startup script will never allow AFS to start. Where this will become a problem is a managed environment such as a lab. Pushing out an update (or a rebuild) of workstations that do not have a separate /usr/vice/cache, will force the administrator to visit each machine and hand-fix /etc/sysconfig/afs to set an appropriate cache size, or else force administrators to hand-maintain their own installation packages. The argument could be made that your configuration management system would fix this, but if your config management is based out of AFS....
I would propose that the init script be altered to start AFS with a very small cache, either disk or memory, if the conditions are CACHESIZE=AUTOMATIC && cache is on /. Then at least the upgrade will succeed (i.e. AFS starts), allowing admins to have their management systems replace the config file with the site-specifc one. Any thoughts about this? -- Jason McCormick <[EMAIL PROTECTED]> CERT Infrastructure Team _______________________________________________ OpenAFS-info mailing list [email protected] https://lists.openafs.org/mailman/listinfo/openafs-info
