Closed jabl closed 7 years ago
Having a look at maybe also removing it via an RPM deprecation. I guess we should also remove this from the compute nodes? ( examples/group_vars/fgci-default-packages )
One issue we found is that openmpi package from EPEL wants the environment-modules package :/ Would perhaps be nice to get rid of this with an rpm somehow, but not clear yet what is the best way to do it.
Yes, I think we should remove it everywhere, that's why I put it under group_vars/all/. Though maybe it needs to be in other places too in case those places override that one..
And yes, removing the environment-modules package gets rid of openmpi and mvapich2 packages too. But that's Ok, since those are provided by the OS and lack slurm integration (IIRC I earlier made another pull request to remove them already), so we should use the MPI libraries provided via cvmfs/local modules anyway.
I pushed a commit to your branch. It removes environment-modules also on compute nodes and in addition a few openmpi packages which also installs environment-modules. I couldn't find any trace of mvapich2.
Set one of our nodes to draining so I can try a reinstall.
p.s. Travis and ansible 2.2.2 fails because https://github.com/ansible/ansible/issues/23016
Oh crap this was against the master branch. I'll try to sort this out tomorrow when I sync devel/master. I've enabled "force review" of PRs to master - maybe in the future that will make me check it.. Also I forgot to reinstall a node and see how that goes.. doing that now.
Did a few reinstalls and after removing the openmpi packages then environment-modules is on longer installed. After updating the ansible variables one needs to push group_vars to the install node and also update the kickstart files.
Since FGCI is using Lmod, make sure that the old Tcl environment modules package is not installed, in order to avoid potential confusion.