Merge pull request #32 from ualberta-rcg/main

update and cleanup

slurmctld/slurmctld.yaml

@@ -0,0 +1,74 @@
+apiVersion: apps/v1
+kind: Deployment  
+metadata:
+  name: slurmctld
+  namespace: slurm
+spec:
+  replicas: 1  
+  selector:
+    matchLabels:
+      app: slurmctld
+  template:
+    metadata:
+      labels:
+        app: slurmctld
+    spec:
+      containers:
+      - name: slurmctld
+        image: rkhoja/slurm:slurmctld  # Adjust image as needed
+        imagePullPolicy: Always
+        ports:
+        - containerPort: 6817
+        volumeMounts:
+        - name: munge-key
+          mountPath: /etc/munge/.secret/
+        - name: slurm-config
+          mountPath: /etc/slurm
+        - name: slurmctld-state
+          mountPath: /var/spool/slurmctld
+#         command: ["/bin/bash", "-c"]
+#         args:
+#           - |
+#             tail -f /dev/null
+      volumes:
+      - name: munge-key
+        secret:
+          secretName: munge-key
+          items:
+          - key: munge-key
+            path: munge.keyfile
+      - name: slurm-config
+        nfs:
+          server: 192.168.1.1
+          path: /disks/slurm/etc/slurm
+      - name: slurmctld-state
+        nfs:
+          server: 192.168.1.1
+          path: /disks/slurm/var/spool/slurmctld/
+---
+apiVersion: v1
+kind: Service
+metadata:
+  name: slurmctld-service
+  namespace: slurm
+  labels:
+    app: slurmctld-service
+spec:
+  selector:
+    app: slurmctld
+  ports:
+  - name: slurmctld
+    port: 6817
+    protocol: TCP
+    targetPort: 6817
+  sessionAffinity: ClientIP
+  sessionAffinityConfig:
+    clientIP:
+      timeoutSeconds: 10800
+  type: LoadBalancer
+  loadBalancerIP: 192.168.1.20
+  externalTrafficPolicy: Cluster
+  internalTrafficPolicy: Cluster
+  ipFamilies:
+  - IPv4
+  ipFamilyPolicy: SingleStack

slurmdbd/slurmdbd.yaml

@@ -34,18 +34,22 @@ spec:
               name: paice-db-pxc-db-secrets
               key: root
         - name: STORAGE_LOC
-          value: "slurm_acct_db" # Or your desired database name
+          value: "slurm_acct_db" 
         - name: AUTH_TYPE
           value: "auth/munge"
         volumeMounts:
         - name: munge-key
           mountPath: /etc/munge/.secret/
+#        command: ["/bin/bash", "-c"]
+#        args:
+#          - |
+#            tail -f /dev/null
       volumes:
       - name: munge-key
         secret:
-          secretName: slurm
+          secretName: munge-key
           items:
-          - key: munge
+          - key: munge-key
             path: munge.keyfile
       restartPolicy: Always
 ---
@@ -54,12 +58,25 @@ kind: Service
 metadata:
   name: slurmdbd-service
   namespace: slurm
+  labels:
+    app: slurmdbd-service
 spec:
+  allocateLoadBalancerNodePorts: true
   selector:
     app: slurmdbd
   ports:
-  - port: 6819 # The port used by slurmdbd
-    targetPort: 6819
+  - name: slurmdbd
+    port: 6819
     protocol: TCP
-  sessionAffinity: ClientIP # Enable sticky sessions
-  type: LoadBalancer # Or NodePort for bare-metal, depending on your setup
+    targetPort: 6819
+  sessionAffinity: ClientIP
+  sessionAffinityConfig:
+    clientIP:
+      timeoutSeconds: 10800  # 3 hours
+  type: LoadBalancer
+  loadBalancerIP: 192.168.1.20
+  externalTrafficPolicy: Cluster
+  internalTrafficPolicy: Cluster
+  ipFamilies:
+  - IPv4
+  ipFamilyPolicy: SingleStack