Update slurm play for nodes

files/default.slurmd → files/default.slurmd.computenode

@@ -8,4 +8,4 @@
 #SLURMCTLD_OPTIONS=""
 
 # Additional options that are passed to the slurmd daemon
-SLURMD_OPTIONS="-Z --conf-server cicus03.douglas.rtss.qc.ca --conf 'Gres=gpu:geforce Feature=workstation'"
+SLURMD_OPTIONS="-Z --conf-server dnpus01.douglas.rtss.qc.ca --conf 'Feature=computenode Weight=1'"

files/default.slurmd.workstation

@@ -0,0 +1,11 @@
+# Defaults for slurmd initscript
+# sourced by /etc/init.d/slurmd
+# installed at /etc/default/slurmd by the maintainer scripts
+#
+# This is a POSIX shell fragment
+#
+# Additional options that are passed to the slurmctld daemon
+#SLURMCTLD_OPTIONS=""
+
+# Additional options that are passed to the slurmd daemon
+SLURMD_OPTIONS="-Z --conf-server dnpus01.douglas.rtss.qc.ca --conf 'Gres=gpu:geforce Feature=workstation Weight=50'"

inventory

@@ -19,6 +19,7 @@ cicws16
 cicws17
 cicws18
 cicws19
+cicws23
 cicws24
 cicws25
 cicws26
@@ -56,7 +57,6 @@ dnpws13
 dnpws14
 dnpws15
 dnpws16
-cichm01
 dnpws17
 dnpws18
 dnpws19
@@ -66,12 +66,23 @@ dnpws22
 dnpws23
 dnpws24
 dnpws25
+dnpws26
+cichm01
 
 [computenodes]
+ciccs01
+ciccs02
+ciccs03
+ciccs04
+ciccs05
 ciccs06
 ciccs07
 ciccs08
 ciccs09
 ciccs10
 
+[userservers]
+cicus03 nis_role=master
+dnpus01 nis_role=slave
+
 [servers]

roles/common/tasks/main.yml

@@ -15,7 +15,7 @@
 - {import_tasks: qbatch.yml, tags: [software,qbatch] }
 - {import_tasks: manual-deb.yml, tags: [software,manualdeb]}
 - {import_tasks: file-config.yml, tags: files}
-#- {import_tasks: slurm.yml, tags: slurm}
+- {import_tasks: slurm.yml, tags: slurm}
 - {import_tasks: system-config.yml, tags: system}
 - {import_tasks: motd.yml, tags: motd}
 

roles/common/tasks/slurm.yml

@@ -20,6 +20,7 @@
       - liblz4-dev
       - libmunge2
       - libmunge-dev
+      - libncurses6
       - libncurses-dev
       - libpam0g-dev
       - libperl-dev
@@ -36,6 +37,7 @@
 
 
 - name: Check if current version is installed
+  become: yes
   stat:
     path: "{{ slurm_prefix_path}}/sbin/slurmd"
   register: slurmd_installed
@@ -102,11 +104,32 @@
 - name: restart munge
   systemd: state=restarted enabled=true name=munge daemon_reload=yes
 
-- name: install slurm default config
-  copy: src=files/default.slurmd dest=/etc/default/slurmd
+- name: copy slurm default configuration based on node type
+  block:
+    - name: Install slurm default config on workstations
+      copy:
+        src: files/default.slurmd.workstation
+        dest: /etc/default/slurmd
+      when: "'workstations' in group_names"
+
+    - name: Install slurm default config on compute nodes
+      copy:
+        src: files/default.slurmd.computenode
+        dest: /etc/default/slurmd
+      when: "'computenodes' in group_names"
 
 - name: install slurm path
   copy: src=files/99slurm_path.sh dest=/etc/profile.d/
 
+- name: update slurmd ExecStart to use the highest nice value
+  lineinfile:
+    path: /etc/systemd/system/slurmd.service
+    regexp: '^ExecStart='
+    line: "ExecStart=/opt/slurm/current/sbin/slurmd --systemd $SLURMD_OPTIONS -n 19"
+  tags:
+    - execstart
+
 - name: enable slurmd
   systemd: state=restarted enabled=true name=slurmd daemon_reload=yes
+  tags:
+    - execstart

roles/common/templates/hosts.j2

@@ -27,7 +27,7 @@ ff02::2 ip6-allrouters
 172.16.67.215 cicus01.douglas.rtss.qc.ca cicus01
 172.16.67.216 cicus02.douglas.rtss.qc.ca cicus02
 172.16.67.230 cicus03.douglas.rtss.qc.ca cicus03
-172.16.67.231 cicus04.douglas.rtss.qc.ca cicus04
+172.16.67.231 dnpus01.douglas.rtss.qc.ca dnpus01
 
 #Login/User Servers IPMI
 172.16.69.50 ipmi-us01 ipmi-cicus01

vars/slurm.yml

@@ -1,4 +1,4 @@
-slurm_version: 23.02.2
+slurm_version: 24.05.1
 slurm_tarball_url: "https://download.schedmd.com/slurm/slurm-{{ slurm_version }}.tar.bz2"
 slurm_src_dir: "/tmp/slurm-{{ slurm_version }}"
 slurm_sbin_path: "/opt/slurm/current/sbin/slurmd"