Home

Future thoughts

• See p4-apps for tf-based version
• Separate "entry point" playbooks (e.g. cluster-infra-configure) from functionality ones (e.g generic)
• Change some names to make linkages clearer: e.g. ophc has config=openhpc, infra-playbook=cluster-infra, setup-playbook=cluster-infra-configure
• (openhpc) Allow network fs for /etc/slurm.conf, with appropriate daemon startup
• change so running tf directly (so there's a prompt before infra gets changed/destroyed!)
• Remove need for /etc/hosts
• Do something like euclid does for setting slurm nodes: num_nodes: "{{ groups['euclid-cam_compute'] | length }}"
• Why do we need os-config to retrieve secrets when we could just copy it from the deployment host?
• Remove the need for the "alaska" group in group_vars/, or at least make it so we don't need things here to create the cluster.
• The "openhpc_packages" and "extra_packages" vars overlap; really we want to be able to install on specific (groups of) nodes too. E.g. git only on head node?