Here's what we do with yum: We use it to control which RPM format software is loaded on a machine. Given a list of RPM names, we tell yum to do an install and it figures out all the dependencies for us. (which is not possible to do without an interface to rpmlib) We have a script which parses the Red Hat comps.xml file (and deps.xml). You tell the script which package groups you want to install (or not install), and which individual packages you want or don't want. This lets you choose, for instance, the 'workstation-common' group, but then deselect the 'admin-tools' group which is implicitly included in the 'workstation-common' group. You can also select/deselect individual packages by name. The output of the script is a list of all RPMs chosen according to your input, with the dependencies pre-resolved (via deps.xml). We then dynamically generate a yum.conf file which points to various yum repositories (depending upon the configuration of the particular machine), and we create a yumgroups.xml file from the flat text file list of RPMs. Then we run 'yum groupinstall' to do the work. All of our yum repositories are on file servers, so we don't want yum to bother copying the RPMs locally before installing. Currently we have a gross hack where we make a fake cache directory which just has symlinks back to the real yum repositories on the file server. We run yum -C so it doesn't try to download anything but instead just follows the symlinks. (Sorry, I was too lazy to fix yum to not attempt to actually download for file:// URLs) We don't use machine-specific Kickstart files to configure the software load on our machines, instead we just use the yum procedure as described above. The daily update process works the same way, which is nice because then machine installs and updates use an identical procedure. On top of all this is a separate system which integrates with RPM and manages individual files. Based on the list of packages on the system, it knows the characteristics of each and every file and can verify them for correctness (we don't use rpm -V). Modified files can be substituted for the ones in the RPMs and copied into place. This process also runs nightly to check that every file on the system is correct. Thanks, Chris wingc@xxxxxxxxxxxxxxx On Wed, 1 Dec 2004, seth vidal wrote: > Now that actually makes sense for yum 2.0.X. > > You might consider trying out yum 2.1.X with the new metadata and seeing > how it differs in the same situation. > > -sv
