chore (helm): add service for tensor-fusion

Author: 0x5457

charts/tensor-fusion/templates/controller-deployment.yaml

@@ -41,18 +41,18 @@ spec:
             - name: cert
               readOnly: true
               mountPath: /tmp/k8s-webhook-server/serving-certs
-              items:
-              - key: tls.crt
-                path: cert
-              - key: tls.key
-                path: key
             - name: config
               mountPath: /etc/tensor-fusion
       volumes:
         - name: cert
           secret:
             secretName: {{.Values.controller.admissionWebhooks.secretName}}
             defaultMode: 420
+            items:
+              - key: cert
+                path: tls.crt
+              - key: key
+                path: tls.key
         - name: config
           configMap:
             name: {{ include "tensor-fusion.fullname" . }}-config

charts/tensor-fusion/templates/controller-service.yaml

@@ -0,0 +1,12 @@
+apiVersion: v1
+kind: Service
+metadata:
+  name: {{ include "tensor-fusion.fullname" . }}
+  namespace: {{ include "tensor-fusion.namespace" . }}
+spec:
+  ports:
+    - port: 8080
+      protocol: TCP
+      targetPort: 8080
+  selector:
+    {{- include "tensor-fusion.controller.selectorLabels" . | nindent 6 }}

charts/tensor-fusion/templates/hypervisor-daemonset.yaml

@@ -53,6 +53,12 @@ spec:
           imagePullPolicy: IfNotPresent
         - name: vector
           image: docker.io/timberio/vector:nightly-2025-01-07-debian
+          env:
+            - name: NODE_IP
+              valueFrom:
+                fieldRef:
+                  apiVersion: v1
+                  fieldPath: status.hostIP
           resources:
             requests:
               cpu: 50m

charts/tensor-fusion/templates/hypervisor-vector.yaml

@@ -28,6 +28,7 @@ data:
           - parse_influx
         source: |
           .namespace = "tf"
+          .tags.nodeip = "${NODE_IP}"
       
       log_to_metric:
         type: log_to_metric
@@ -36,6 +37,7 @@ data:
         all_metrics: true
         metrics: []
     sinks:
+      
       sink_greptimedb_metrics:
         type: greptimedb_metrics
         inputs:

charts/tensor-fusion/templates/tensor-fusion-config.yaml

@@ -4,4 +4,68 @@ metadata:
   name: {{ include "tensor-fusion.fullname" . }}-config
   namespace: {{ include "tensor-fusion.namespace" . }}
 data:
-  config.yaml: | {{ toYaml .Values.controller.config | nindent 4 }}
+  config.yaml: |-
+    worker:
+      template:
+        spec:
+          terminationGracePeriodSeconds: 0
+          runtimeClassName: nvidia
+          volumes:
+            - name: worker-sock
+              hostPath:
+                path: /tensor-fusion/worker/sock
+                type: DirectoryOrCreate
+          hostNetwork: true
+          hostPID: true
+          containers:
+            - name: tensor-fusion-worker
+              image: tensorfusion/tensor-fusion-worker:latest
+              env:
+                - name: POD_NAME
+                  valueFrom:
+                    fieldRef:
+                      apiVersion: v1
+                      fieldPath: metadata.name
+                - name: TF_ENABLE_LOG
+                  value: '1'
+              volumeMounts:
+                - name: worker-sock
+                  mountPath: /tensor-fusion/worker/sock
+              command:
+                - /home/app/tensor-fusion-worker
+                - -n
+                - native
+                - -p
+                - '$(TENSOR_FUSION_WORKER_PORT)'
+                - -a
+                - '0x1129'
+                - -l
+                - /tensor-fusion/worker/sock/$(POD_NAME).sock
+
+    podMutation:
+      operatorEndpoint: http://{{ include "tensor-fusion.fullname" . }}.{{ include "tensor-fusion.namespace" . }}:8080
+      patchToContainer:
+        volumeMounts:
+          - mountPath: /tensor-fusion
+            name: tf-libs
+        env:
+          - name: LD_PRELOAD
+            value: /tensor-fusion/libcuda.so
+          - name: TF_ENABLE_LOG
+            value: '1'
+      patchToPod:
+        spec:
+          volumes:
+            - name: tf-libs
+              emptyDir: {}
+          initContainers:
+            - name: inject-lib
+              image: tensorfusion/tensor-fusion-client:latest
+              command:
+                - sh
+                - -c
+                - cp /home/app/*.so /tensor-fusion/ && cp nvidia-smi-linux-amd64-550.54.15 /tensor-fusion/nvidia-smi
+              volumeMounts:
+                - mountPath: /tensor-fusion
+                  name: tf-libs
+          

charts/tensor-fusion/values.yaml

@@ -55,70 +55,3 @@ controller:
     secretName: tensor-fusion-webhook-secret
     patch:
       image: registry.k8s.io/ingress-nginx/kube-webhook-certgen:v1.5.0
-
-  config:
-    worker:
-      port: 9999
-      template:
-        spec:
-          terminationGracePeriodSeconds: 0
-          runtimeClassName: nvidia
-          volumes:
-            - name: worker-sock
-              hostPath:
-                path: /tensor-fusion/worker/sock
-                type: DirectoryOrCreate
-          hostNetwork: true
-          hostPID: true
-          containers:
-            - name: tensor-fusion-worker
-              image: tensorfusion/tensor-fusion-worker:latest
-              env:
-                - name: POD_NAME
-                  valueFrom:
-                    fieldRef:
-                      apiVersion: v1
-                      fieldPath: metadata.name
-                - name: TF_ENABLE_LOG
-                  value: '1'
-              volumeMounts:
-                - name: worker-sock
-                  mountPath: /tensor-fusion/worker/sock
-              command:
-                - /home/app/tensor-fusion-worker
-                - -n
-                - native
-                - -p
-                - '9999'
-                - -a
-                - '0x1129'
-                - -l
-                - /tensor-fusion/worker/sock/$(POD_NAME).sock
-
-    podMutation:
-      operatorEndpoint: http://tensor-fusion-controller.tensor-fusion:8080
-      patchToContainer:
-        volumeMounts:
-          - mountPath: /tensor-fusion
-            name: tf-libs
-        env:
-          - name: LD_PRELOAD
-            value: /tensor-fusion/libcuda.so
-          - name: TF_ENABLE_LOG
-            value: '1'
-      patchToPod:
-        spec:
-          volumes:
-            - name: tf-libs
-              emptyDir: {}
-          initContainers:
-            - name: inject-lib
-              image: tensorfusion/tensor-fusion-client:latest
-              command:
-                - sh
-                - -c
-                - cp /home/app/*.so /tensor-fusion/
-              volumeMounts:
-                - mountPath: /tensor-fusion
-                  name: tf-libs
-