Ansible & Slurm Fixes

[zacharygraber]

Implements fixes for #19 and #20.

Changes

Ansible packages/version/collections

• Install ansible-core PyPI package instead of ansible, and pin version to <2.17 to fix #20
• Instead of relying on the community-supported ansible package, use a specific version of ansible-core and pull the collections we need (OpenStack and POSIX).
  • Further, pin the version of the OpenStack collection so we aren't affected by future breaking changes.

slurm.conf

• Make a 1-character change for #19

Other (Ansible)

• Use the modern role name openstack.cloud.server instead of the deprecated/old os_server.
• Replace a static 90-second wait for SSH connectivity with a built-in Ansible role to wait for SSH. Also, slightly increase timeout (just in case).

---

Testing

I've tested this on Jetstream2/Exosphere by using these steps. Please feel free to validate/replicate:

1. In Exosphere, open the instance creation flow

2. Select Featured-RockyLinux8 as the image

3. Set Create your own SLURM cluster with this instance as the head node to yes

4. In the Boot Script section, replace {create-cluster-command} inline with: su - rocky -c "git clone --branch rocky-linux --single-branch --depth 1 https://github.com/zacharygraber/Jetstream_Cluster.git; cd Jetstream_Cluster; ./cluster_create_local.sh -d 2>&1 | tee local_create.log"

   • This does the exact same thing Exosphere is doing to create the cluster normally, except it points to my fork instead of this repo.

5. Wait for setup to complete, then SSH into

6. sudo su - rocky

7. cd Jetstream_Cluster

8. Inspect the output in local_create.log to make sure everything looks alright.

9. Run sbatch slurm_test.job, wait for it to finish, then verify the output.

[julianpistorius]

Works! Thank you @zacharygraber.