Eric V. Smith wrote: > >I think I'll go with a solution like Aaron's, where each host grabs its >config files from a central server, which generates the config files as >needed. > > That's what LCFG does (http://www.lcfg.org) - you define a profile per node: #include <site_specific.h> /* things like domain names, network details etc */ boot.services httpd postfix httpd.port 80 portfix.smarthost mail.mydomain.com ... etc. .. then the central server compiles these values (lists are also allowed) into an XML version. A notification is sent to the client which then picks up the new profile and rewrites the new config file on the fly - optionally restarting the service too. The config fire re-writing is based on a template engine so that centrally you dont' care about config file formats. We're got this tied into our install system so that we can even do things like: fstab.partitions hda1 hda2 fstab.partitions_hda1 100M fstab.partitions_hda2 30G and things like: profile.packages @workstation.rpms @teaching_labs.rpms .. you shove in a CD or PXE boot an the machine comes up with all the packages you want and fully configured (including NFS, LDAP, Kerberos etc.) Any changes to the profile database are picked up by the server and notifications sent out automatically. Adding an rpm to all the machines would be as simple as adding it to the file workstations.rpms - removing it just means deleting it. It's quite handy as a systems developer to be able to rpm -Uvh/-e and mangle stuff in /etc all day on my workstation testing things and know that the next morning (optionally) that my machine and all it's config will have been reverted to the baseline config. Carwyn
