Merge pull request #724 from stackhpc/fix/only-slurm-rebuild

Fix cuda install when building without lustre first

Author: jovial

.github/workflows/extra.yml

@@ -8,6 +8,7 @@ on:
       - 'environments/.stackhpc/tofu/cluster_image.auto.tfvars.json'
       - 'ansible/roles/doca/**'
       - 'ansible/roles/cuda/**'
+      - 'ansible/roles/slurm_recompile/**' # runs on cuda group
       - 'ansible/roles/lustre/**'
       - '.github/workflows/extra.yml'
   pull_request:

ansible/.gitignore

@@ -90,7 +90,7 @@ roles/*
 !roles/gateway/**
 !roles/alertmanager/
 !roles/alertmanager/**
-!roles/slurm_recompile/**
+!roles/slurm_recompile/
 !roles/slurm_recompile/**
 !roles/nhc/
 !roles/nhc/**

ansible/roles/slurm_recompile/defaults/main.yml

@@ -2,3 +2,5 @@
 # Whether to link slurm against the NVIDIA management library
 slurm_recompile_with_nvml: false
 
+slurm_recompile_build_packages:
+  - rpm-build

ansible/roles/slurm_recompile/tasks/main.yml

@@ -12,6 +12,10 @@
   set_fact:
     slurm_package: "{{ ansible_facts.packages['slurm-slurmd-ohpc'].0 }}"
 
+- name: Install build packages
+  ansible.builtin.dnf:
+    name: "{{ slurm_recompile_build_packages }}"
+
 - name: Recompile and install slurm packages
   shell: |
     #!/bin/bash