[Infrastructures] Entire linux hierarchy maintained in CVS

[Paul Dlug]

On Mar 18, 2005, at 10:23 AM, Jordan Curzon wrote:

> How you any of you mix radmind with the system package management? Do
> you only use radmind for configuration, and local packages for
> software? Do you ditch the package management all together? Or do you
> use package management to manage the radmind loadsets? What are you
> experiences with the different approches?

I can share some of my experiences with this, my group is currently 
managing about 150 mac os x desktops, 40 freebsd servers, 10 solaris 
servers and 5 linux servers with radmind.

We're using radmind for everything including base OS, software packages 
and configuration. Our approach is to install a package on a server 
just containing the minimal OS loadsets using the normal packaging 
tools (FreeBSD ports, linux rpm, etc.) then generate the radmind 
loadset for it. Configuration files for that package go into a separate 
loadset. To perform an upgrade the loadset is applied to the minimal 
server again, upgraded with the native package tools and regenerated.

I don't really see a way to use radmind for the base OS and config 
files while using another package management tool. You would have to 
put much of the filesystem in the negative space which creates 
problems. I once had all config files negative and used cfengine for 
managing them, this was less than ideal as well. I've found that 
avoiding negatives as much as possible and creating smaller loadsets 
gives me the most flexibility.

As much as I like radmind there are somethings I'd like to be able to 
do much better such as version management, I'd like to be able to put 
config files in CVS/subversion and still use radmind as the transport. 
Right now we just have that set as a policy that anyone updating a 
major config file must check it back in.


--Paul