Merge pull request #38 from Gerbenvandervries/develop

Slurm role bugfixes for dynamic resources.
rug-cit-hpc · Jan 24, 2019 · 6e27a81 · 6e27a81
2 parents bcf0a16 + 7b89237
commit 6e27a81
Show file tree

Hide file tree

Showing 7 changed files with 22 additions and 19 deletions.
diff --git a/gearshift_hosts.ini b/gearshift_hosts.ini
@@ -64,10 +64,10 @@ gs-vcompute[01:11]
 [cluster:children]
 compute-vm
 administration
-jumphost
 
 [gearshift-cluster:children]
 cluster
+jumphost
 
 [metal]
 gs-openstack

diff --git a/host_vars/hc-sai b/host_vars/hc-sai
@@ -0,0 +1,3 @@
+---
+mailhub: 192.168.0.5
+rewrite_domain: hc-sai.gcc.rug.nl
diff --git a/hyperchicken_hosts.ini b/hyperchicken_hosts.ini
@@ -24,7 +24,7 @@ hc-vcompute[01:05]
 [cluster:children]
 compute-vm
 administration
-jumphost
 
 [hyperchicken-cluster:children]
 cluster
+jumphost
diff --git a/roles/slurm-client/tasks/main.yml b/roles/slurm-client/tasks/main.yml
@@ -47,23 +47,23 @@
     - name: /etc/slurm
       owner: root
       group: root
-      mode: 0755
+      mode: '0755'
     - name: /etc/nhc
       owner: root
       group: root
-      mode: 0755
+      mode: '0755'
     - name: /var/log/slurm
       owner: root
       group: root
-      mode: 0750
+      mode: '0750'
     - name: /var/spool/slurm
       owner: slurm
       group: root
-      mode: 0750
+      mode: '0750'
     - name: /var/spool/slurmd
       owner: slurm
       group: root
-      mode: 0750
+      mode: '0750'
 
 - name: Deploy slurm.conf
   template:
@@ -97,7 +97,7 @@
 
 - name: Deploy nhc.conf
   template:
-    src: template/nhc.conf
+    src: templates/nhc.conf
     dest: /etc/nhc/nhc.conf
     owner: root
     group: root

diff --git a/roles/slurm/defaults/main.yml b/roles/slurm/defaults/main.yml
@@ -9,7 +9,7 @@ munge_gid: 498
 #  * number of vcompute nodes in inventory and
 #  * cores per node, mem per node, etc. as specified in group_vars for cluster.
 #
-vcompute_host_count: "{{ groups['compute-vm']|length }}"
+vcompute_host_count: "{{ groups['compute-vm']|list|length }}"
 cluster_cores_total: "{{ vcompute_host_count|int * vcompute_max_cpus_per_node|int }}"
 cluster_mem_total: "{{ vcompute_host_count|int * vcompute_max_mem_per_node|int }}"
 ...
diff --git a/roles/slurm/files/configure_slurm_accounting_db.bash b/roles/slurm/files/configure_slurm_accounting_db.bash
@@ -59,22 +59,22 @@ sacctmgr -i create qos set \
         Name='regular-short' \
         Priority=10 \
         Description='regular-short' \
-        GrpSubmit=30000 MaxSubmitJobsPU=5000  MaxWall=06:00:00 
+        GrpSubmit=30000 MaxSubmitJobsPU=5000  MaxWall=06:00:00
 
 sacctmgr -i create qos set \
         Name='regular-medium' \
         Priority=10 \
         Description='regular-medium' \
         GrpSubmit=30000 MaxSubmitJobsPU=5000  MaxWall=1-00:00:00 \
-        MaxTRESPU=cpu={{ (cluster_cores_total * 0.4) | int }},mem={{ (cluster_mem_total * 0.4) | int }}
+        MaxTRESPU=cpu={{ (cluster_cores_total | float * 0.4) | int }},mem={{ (cluster_mem_total | float * 0.4) | int }}
 
 sacctmgr -i create qos set \
         Name='regular-long' \
         Priority=10 \
         Description='regular-long' \
         GrpSubmit=3000 MaxSubmitJobsPU=1000  MaxWall=7-00:00:00 \
-        GrpTRES=cpu={{ (cluster_cores_total * 0.3) | int }},mem={{ (cluster_mem_total * 0.3) | int }} \
-        MaxTRESPU=cpu={{ (cluster_cores_total * 0.15) | int }},mem={{ (cluster_mem_total * 0.15) | int }}
+        GrpTRES=cpu={{ (cluster_cores_total | float * 0.3) | int }},mem={{ (cluster_mem_total | float * 0.3) | int }} \
+        MaxTRESPU=cpu={{ (cluster_cores_total | float * 0.15) | int }},mem={{ (cluster_mem_total | float * 0.15) | int }}
 
 #
 # QoS priority
@@ -93,25 +93,25 @@ sacctmgr -i create qos set \
         UsageFactor=2 \
         Description='priority-short' \
         GrpSubmit=5000  MaxSubmitJobsPU=1000   MaxWall=06:00:00 \
-        MaxTRESPU=cpu={{ (cluster_cores_total * 0.25) | int }},mem={{ (cluster_mem_total * 0.25) | int }}
+        MaxTRESPU=cpu={{ (cluster_cores_total | float * 0.25) | int }},mem={{ (cluster_mem_total | float * 0.25) | int }}
 
 sacctmgr -i create qos set \
         Name='priority-medium' \
         Priority=20 \
         UsageFactor=2 \
         Description='priority-medium' \
         GrpSubmit=2500  MaxSubmitJobsPU=500   MaxWall=1-00:00:00 \
-        GrpTRES=cpu={{ (cluster_cores_total * 0.5) | int }},mem={{ (cluster_mem_total * 0.5) | int }} \
-        MaxTRESPU=cpu={{ (cluster_cores_total * 0.2) | int }},mem={{ (cluster_mem_total * 0.2) | int }}
+        GrpTRES=cpu={{ (cluster_cores_total | float * 0.5) | int }},mem={{ (cluster_mem_total | float * 0.5) | int }} \
+        MaxTRESPU=cpu={{ (cluster_cores_total | float * 0.2) | int }},mem={{ (cluster_mem_total | float * 0.2) | int }}
 
 sacctmgr -i create qos set \
         Name='priority-long' \
         Priority=20 \
         UsageFactor=2 \
         Description='priority-long' \
         GrpSubmit=250   MaxSubmitJobsPU=50   MaxWall=7-00:00:00 \
-        GrpTRES=cpu={{ (cluster_cores_total * 0.2) | int }},mem={{ (cluster_mem_total * 0.2) | int }} \
-        MaxTRESPU=cpu={{ (cluster_cores_total * 0.1) | int }},mem={{ (cluster_mem_total * 0.1) | int }}
+        GrpTRES=cpu={{ (cluster_cores_total | float * 0.2) | int }},mem={{ (cluster_mem_total | float * 0.2) | int }} \
+        MaxTRESPU=cpu={{ (cluster_cores_total | float * 0.1) | int }},mem={{ (cluster_mem_total | float * 0.1) | int }}
 
 #
 # QoS ds

diff --git a/talos_hosts.ini b/talos_hosts.ini
@@ -24,7 +24,7 @@ tl-vcompute[01:03]
 [cluster:children]
 compute-vm
 administration
-jumphost
 
 [talos-cluster:children]
 cluster
+jumphost