fix: Genai helm service fix (#8885)

* wip trying new service for redis queue

* update the genai helm chart integration to enforce services for all hosts

* revert the changes from testing

* use clusterip instead

helm/charts/determined/templates/genai/genai-service.yaml → helm/charts/determined/templates/genai/genai-backend-service.yaml

@@ -3,7 +3,7 @@
 apiVersion: v1
 kind: Service
 metadata:
-  name: genai-service-{{ .Release.Name }}
+  name: genai-backend-service-{{ .Release.Name }}
   namespace: {{ .Release.Namespace }}
   labels:
     app: genai-{{ .Release.Name }}
@@ -13,7 +13,7 @@ spec:
   - port: {{ required "A valid Values.genai.port entry required!" .Values.genai.port }}
     targetPort: {{ .Values.genai.port }}
     protocol: TCP
-  type: LoadBalancer
+  type: ClusterIP
   selector:
     app: genai-{{ .Release.Name }}
 {{ end }}

helm/charts/determined/templates/genai/genai-deployment.yaml

@@ -46,6 +46,14 @@ spec:
             value: {{ include "genai.PVCName" . }}
           - name: RESOURCE_POOL_EXTRA_METADATA_PATH
             value: /run/determined/workdir/rp_config/resource_pool_metadata.yaml
+          - name: MESSAGE_QUEUE_HOST
+            value: genai-queue-service-{{ .Release.Name }}
+          - name: MESSAGE_QUEUE_PORT
+            value: {{ .Values.genai.messageQueuePort | quote }}
+          - name: GENAI_HOST
+            value: genai-backend-service-{{ .Release.Name }}
+          - name: GENAI_HOST_PORT
+            value: {{ .Values.genai.port | quote }}
         volumeMounts:
           - name: genai-pvc-storage
             mountPath: /run/determined/workdir/shared_fs

helm/charts/determined/templates/genai/genai-queue-service.yaml

@@ -0,0 +1,20 @@
+{{- if .Values.genai }}
+{{- if .Values.genai.version }}
+apiVersion: v1
+kind: Service
+metadata:
+  name: genai-queue-service-{{ .Release.Name }}
+  namespace: {{ .Release.Namespace }}
+  labels:
+    app: genai-{{ .Release.Name }}
+    release: {{ .Release.Name }}
+spec:
+  ports:
+  - port: {{ required "A valid Values.genai.messageQueuePort entry required!" .Values.genai.messageQueuePort }}
+    targetPort: {{ .Values.genai.messageQueuePort }}
+    protocol: TCP
+  type: ClusterIP
+  selector:
+    app: genai-{{ .Release.Name }}
+{{ end }}
+{{ end }}

helm/charts/determined/templates/master-config.yaml

@@ -140,9 +140,9 @@ stringData:
     {{- $port := (required "A valid .Values.genai.port entry required!" .Values.genai.port) }}
     __internal:
       proxied_servers:
-        - destination: "http://genai-service-{{ .Release.Name }}:{{ $port }}/lore"
+        - destination: "http://genai-backend-service-{{ .Release.Name }}:{{ $port }}/lore"
           path_prefix: /lore
-        - destination: "http://genai-service-{{ .Release.Name }}:{{ $port }}/genai"
+        - destination: "http://genai-backend-service-{{ .Release.Name }}:{{ $port }}/genai"
           path_prefix: /genai
     {{- end }}
     {{- end }}