Add sackd file override to restart on-failure (#204)

* Add sackd file override to restart on-failure * Update ansible/roles/slurm/tasks/service.yml Co-authored-by: Tom Downes <[email protected]> * Update ansible/roles/slurm/tasks/service.yml Co-authored-by: Tom Downes <[email protected]> --------- Co-authored-by: Tom Downes <[email protected]>
GoogleCloudPlatform · Sep 3, 2024 · 0041af4 · 0041af4
1 parent 75f7375
commit 0041af4
Show file tree

Hide file tree

Showing 3 changed files with 16 additions and 0 deletions.
diff --git a/ansible/roles/slurm/defaults/main.yml b/ansible/roles/slurm/defaults/main.yml
@@ -40,5 +40,6 @@ slurmrestd_user:
   gid: 982
 
 slurmd_override_path: /etc/systemd/system/slurmd.service.d/overrides.conf
+sackd_override_path: /etc/systemd/system/sackd.service.d/override.conf
 slurmcmd_timeout: 30
 handle_services: true
diff --git a/ansible/roles/slurm/tasks/service.yml b/ansible/roles/slurm/tasks/service.yml
@@ -59,3 +59,15 @@
     dest: '{{ slurmd_override_path }}'
     mode: 0o644
   notify: Reload SystemD configuration
+
+- name: Create sackd override directory
+  file:
+    path: '{{ sackd_override_path | dirname }}'
+    state: directory
+
+- name: Sackd SystemD overrides
+  template:
+    src: systemd/sackd_overrides.j2
+    dest: '{{ sackd_override_path }}'
+    mode: 0o644
+  notify: Reload SystemD configuration
diff --git a/ansible/roles/slurm/templates/systemd/sackd_overrides.j2 b/ansible/roles/slurm/templates/systemd/sackd_overrides.j2
@@ -0,0 +1,3 @@
+[Service]
+RestartSec=15s
+Restart=on-failure