Merge pull request #1026 from NVIDIA/nvidia-driver-ctr-node-envars

tariq1890 · web-flow · commit 34e4dc2b2b07 · 2024-10-07T14:02:32.000-07:00
[state-driver] add downward API envars to fetch node name and IP
diff --git a/assets/state-driver/0500_daemonset.yaml b/assets/state-driver/0500_daemonset.yaml
@@ -94,6 +94,15 @@ spec:
         name: nvidia-driver-ctr
         command: ["nvidia-driver"]
         args: ["init"]
+        env:
+          - name: NODE_NAME
+            valueFrom:
+              fieldRef:
+                fieldPath: spec.nodeName
+          - name: NODE_IP
+            valueFrom:
+              fieldRef:
+                fieldPath: status.hostIP
         securityContext:
           privileged: true
           seLinuxOptions:
diff --git a/internal/state/testdata/golden/driver-additional-configs.yaml b/internal/state/testdata/golden/driver-additional-configs.yaml
@@ -132,6 +132,14 @@ spec:
         env:
         - name: NVIDIA_VISIBLE_DEVICES
           value: void
+        - name: NODE_NAME
+          valueFrom:
+            fieldRef:
+              fieldPath: spec.nodeName
+        - name: NODE_IP
+          valueFrom:
+            fieldRef:
+              fieldPath: status.hostIP
         image: nvcr.io/nvidia/driver:525.85.03-ubuntu22.04
         imagePullPolicy: IfNotPresent
         lifecycle:
diff --git a/internal/state/testdata/golden/driver-full-spec.yaml b/internal/state/testdata/golden/driver-full-spec.yaml
@@ -138,6 +138,14 @@ spec:
         env:
         - name: NVIDIA_VISIBLE_DEVICES
           value: void
+        - name: NODE_NAME
+          valueFrom:
+            fieldRef:
+              fieldPath: spec.nodeName
+        - name: NODE_IP
+          valueFrom:
+            fieldRef:
+              fieldPath: status.hostIP
         - name: OPEN_KERNEL_MODULES_ENABLED
           value: "true"
         - name: FOO
diff --git a/internal/state/testdata/golden/driver-gdrcopy-openshift.yaml b/internal/state/testdata/golden/driver-gdrcopy-openshift.yaml
@@ -188,6 +188,14 @@ spec:
         env:
         - name: NVIDIA_VISIBLE_DEVICES
           value: void
+        - name: NODE_NAME
+          valueFrom:
+            fieldRef:
+              fieldPath: spec.nodeName
+        - name: NODE_IP
+          valueFrom:
+            fieldRef:
+              fieldPath: status.hostIP
         - name: OPENSHIFT_VERSION
           value: "4.13"
         - name: HTTP_PROXY
diff --git a/internal/state/testdata/golden/driver-gdrcopy.yaml b/internal/state/testdata/golden/driver-gdrcopy.yaml
@@ -132,6 +132,14 @@ spec:
         env:
         - name: NVIDIA_VISIBLE_DEVICES
           value: void
+        - name: NODE_NAME
+          valueFrom:
+            fieldRef:
+              fieldPath: spec.nodeName
+        - name: NODE_IP
+          valueFrom:
+            fieldRef:
+              fieldPath: status.hostIP
         image: nvcr.io/nvidia/driver:525.85.03-ubuntu22.04
         imagePullPolicy: IfNotPresent
         lifecycle:
diff --git a/internal/state/testdata/golden/driver-gds.yaml b/internal/state/testdata/golden/driver-gds.yaml
@@ -132,6 +132,14 @@ spec:
         env:
         - name: NVIDIA_VISIBLE_DEVICES
           value: void
+        - name: NODE_NAME
+          valueFrom:
+            fieldRef:
+              fieldPath: spec.nodeName
+        - name: NODE_IP
+          valueFrom:
+            fieldRef:
+              fieldPath: status.hostIP
         image: nvcr.io/nvidia/driver:525.85.03-ubuntu22.04
         imagePullPolicy: IfNotPresent
         lifecycle:
diff --git a/internal/state/testdata/golden/driver-minimal.yaml b/internal/state/testdata/golden/driver-minimal.yaml
@@ -132,6 +132,14 @@ spec:
         env:
         - name: NVIDIA_VISIBLE_DEVICES
           value: void
+        - name: NODE_NAME
+          valueFrom:
+            fieldRef:
+              fieldPath: spec.nodeName
+        - name: NODE_IP
+          valueFrom:
+            fieldRef:
+              fieldPath: status.hostIP
         image: nvcr.io/nvidia/driver:525.85.03-ubuntu22.04
         imagePullPolicy: IfNotPresent
         lifecycle:
diff --git a/internal/state/testdata/golden/driver-openshift-drivertoolkit.yaml b/internal/state/testdata/golden/driver-openshift-drivertoolkit.yaml
@@ -188,6 +188,14 @@ spec:
         env:
         - name: NVIDIA_VISIBLE_DEVICES
           value: void
+        - name: NODE_NAME
+          valueFrom:
+            fieldRef:
+              fieldPath: spec.nodeName
+        - name: NODE_IP
+          valueFrom:
+            fieldRef:
+              fieldPath: status.hostIP
         - name: OPENSHIFT_VERSION
           value: "4.13"
         - name: HTTP_PROXY
diff --git a/internal/state/testdata/golden/driver-precompiled.yaml b/internal/state/testdata/golden/driver-precompiled.yaml
@@ -134,6 +134,14 @@ spec:
         env:
         - name: NVIDIA_VISIBLE_DEVICES
           value: void
+        - name: NODE_NAME
+          valueFrom:
+            fieldRef:
+              fieldPath: spec.nodeName
+        - name: NODE_IP
+          valueFrom:
+            fieldRef:
+              fieldPath: status.hostIP
         image: nvcr.io/nvidia/driver:535-5.4.0-150-generic-ubuntu22.04
         imagePullPolicy: IfNotPresent
         lifecycle:
diff --git a/internal/state/testdata/golden/driver-rdma-hostmofed.yaml b/internal/state/testdata/golden/driver-rdma-hostmofed.yaml
@@ -132,6 +132,14 @@ spec:
         env:
         - name: NVIDIA_VISIBLE_DEVICES
           value: void
+        - name: NODE_NAME
+          valueFrom:
+            fieldRef:
+              fieldPath: spec.nodeName
+        - name: NODE_IP
+          valueFrom:
+            fieldRef:
+              fieldPath: status.hostIP
         - name: GPU_DIRECT_RDMA_ENABLED
           value: "true"
         - name: USE_HOST_MOFED
diff --git a/internal/state/testdata/golden/driver-rdma.yaml b/internal/state/testdata/golden/driver-rdma.yaml
@@ -132,6 +132,14 @@ spec:
         env:
         - name: NVIDIA_VISIBLE_DEVICES
           value: void
+        - name: NODE_NAME
+          valueFrom:
+            fieldRef:
+              fieldPath: spec.nodeName
+        - name: NODE_IP
+          valueFrom:
+            fieldRef:
+              fieldPath: status.hostIP
         - name: GPU_DIRECT_RDMA_ENABLED
           value: "true"
         image: nvcr.io/nvidia/driver:525.85.03-ubuntu22.04
diff --git a/internal/state/testdata/golden/driver-vgpu-host-manager-openshift.yaml b/internal/state/testdata/golden/driver-vgpu-host-manager-openshift.yaml
@@ -178,6 +178,14 @@ spec:
         env:
         - name: NVIDIA_VISIBLE_DEVICES
           value: void
+        - name: NODE_NAME
+          valueFrom:
+            fieldRef:
+              fieldPath: spec.nodeName
+        - name: NODE_IP
+          valueFrom:
+            fieldRef:
+              fieldPath: status.hostIP
         - name: OPENSHIFT_VERSION
           value: "4.13"
         image: nvcr.io/nvidia/vgpu-manager:525.85.03-rhel8.0
diff --git a/internal/state/testdata/golden/driver-vgpu-host-manager.yaml b/internal/state/testdata/golden/driver-vgpu-host-manager.yaml
@@ -132,6 +132,14 @@ spec:
         env:
         - name: NVIDIA_VISIBLE_DEVICES
           value: void
+        - name: NODE_NAME
+          valueFrom:
+            fieldRef:
+              fieldPath: spec.nodeName
+        - name: NODE_IP
+          valueFrom:
+            fieldRef:
+              fieldPath: status.hostIP
         image: nvcr.io/nvidia/vgpu-manager:525.85.03-ubuntu22.04
         imagePullPolicy: IfNotPresent
         name: nvidia-driver-ctr
diff --git a/internal/state/testdata/golden/driver-vgpu-licensing.yaml b/internal/state/testdata/golden/driver-vgpu-licensing.yaml
@@ -132,6 +132,14 @@ spec:
         env:
         - name: NVIDIA_VISIBLE_DEVICES
           value: void
+        - name: NODE_NAME
+          valueFrom:
+            fieldRef:
+              fieldPath: spec.nodeName
+        - name: NODE_IP
+          valueFrom:
+            fieldRef:
+              fieldPath: status.hostIP
         image: nvcr.io/nvidia/driver:525.85.03-ubuntu22.04
         imagePullPolicy: IfNotPresent
         lifecycle:
diff --git a/manifests/state-driver/0500_daemonset.yaml b/manifests/state-driver/0500_daemonset.yaml
@@ -205,6 +205,14 @@ spec:
         # always use runc for driver containers
         - name: NVIDIA_VISIBLE_DEVICES
           value: void
+        - name: NODE_NAME
+          valueFrom:
+            fieldRef:
+              fieldPath: spec.nodeName
+        - name: NODE_IP
+          valueFrom:
+            fieldRef:
+              fieldPath: status.hostIP
       {{- if deref .Driver.Spec.UseOpenKernelModules }}
         - name: OPEN_KERNEL_MODULES_ENABLED
           value: "true"
