feat: Add L4 vLLM deployment and KEDA scale-to-zero for A100

apps/base/customer1/openclaw/keda-a100.yaml

@@ -0,0 +1,18 @@
+apiVersion: http.keda.sh/v1alpha1
+kind: HTTPScaledObject
+metadata:
+  name: openclaw-brain-a100-scaling
+  namespace: customer1
+spec:
+  hosts:
+    - a100-vllm.internal.cluster # Replace with actual internal routing host if needed
+  scaleTargetRef:
+    name: openclaw-brain-vllm
+    kind: Deployment
+    apiVersion: apps/v1
+    service: openclaw-brain-service
+    port: 8000
+  replicas:
+    min: 0
+    max: 1
+  scaledownPeriod: 900 # 15 minutes of idle time before scaling to zero

apps/base/customer1/openclaw/kustomization.yaml

@@ -1,6 +1,8 @@
 apiVersion: kustomize.config.k8s.io/v1beta1
 kind: Kustomization
 resources:
+  - vllm-l4.yaml
+  - keda-a100.yaml
   - pvc.yaml
   - configmap.yaml
   - deployment.yaml

apps/base/customer1/openclaw/vllm-l4.yaml

@@ -0,0 +1,86 @@
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  name: openclaw-brain-vllm-l4
+  namespace: customer1
+  labels:
+    app: openclaw-brain-l4
+spec:
+  replicas: 1
+  selector:
+    matchLabels:
+      app: openclaw-brain-l4
+  template:
+    metadata:
+      labels:
+        app: openclaw-brain-l4
+    spec:
+      nodeSelector:
+        cloud.google.com/gke-accelerator: "nvidia-l4"
+      tolerations:
+        - key: "nvidia.com/gpu"
+          operator: "Exists"
+          effect: "NoSchedule"
+      containers:
+        - name: vllm-brain-l4
+          image: vllm/vllm-openai:latest
+          command: ["python3", "-m", "vllm.entrypoints.openai.api_server"]
+          env:
+            - name: HF_TOKEN
+              value: ""
+          args:
+            - --model=Qwen/Qwen2.5-Coder-7B-Instruct
+            - --host=0.0.0.0
+            - --port=8000
+            - --tensor-parallel-size=1
+            - --gpu-memory-utilization=0.90
+            - --max-model-len=16384
+            - --enable-auto-tool-choice 
+            - --trust-remote-code
+            - --dtype=auto
+            - --tool-call-parser=qwen3_coder
+            - --enable-prefix-caching
+          ports:
+            - containerPort: 8000
+          resources:
+            limits:
+              nvidia.com/gpu: 1
+              memory: "24Gi"
+              cpu: "8"
+            requests:
+              nvidia.com/gpu: 1
+              memory: "16Gi"
+              cpu: "4"
+          volumeMounts:
+            - name: model-cache
+              mountPath: /root/.cache/huggingface
+      volumes:
+        - name: model-cache
+          persistentVolumeClaim:
+            claimName: vllm-model-cache-pvc-l4
+---
+apiVersion: v1
+kind: PersistentVolumeClaim
+metadata:
+  name: vllm-model-cache-pvc-l4
+  namespace: customer1
+spec:
+  accessModes:
+    - ReadWriteOnce
+  resources:
+    requests:
+      storage: 50Gi
+---
+apiVersion: v1
+kind: Service
+metadata:
+  name: openclaw-brain-service-l4
+  namespace: customer1
+spec:
+  selector:
+    app: openclaw-brain-l4
+  ports:
+    - protocol: TCP
+      port: 8000
+      targetPort: 8000
+  type: ClusterIP