Things which need to be done on boot to make the fat image usable for a compute node immediately (i.e. no ansible):
munge key: written to file by ansible
control name/addr for slurm - written to file by ansible
FreeIPA: We run the renrolement stuff from Ansible at the moment.
services start:
cuda role (not enabled fat image anyway) would run in full during fat image build sets nvidia-persistenced enabled
munge
slurmd
node-exporter: is already enabled in image build
etc_hosts: At least for arcus where internal dns doesn't work. And this is hard cause we don’t know the IPs until we have created the hosts (or at least ports)
also potentially resolv_conf, proxy
also need to think about systemd role / common defaults
Things which need to be done on boot to make the fat image usable for a compute node immediately (i.e. no ansible):
See also https://github.com/stackhpc/ansible-slurm-appliance/compare/main...experiment/bootable-compute