Merge pull request #865 from NVIDIA/fix-volumes-for-vgpu-manager

[vgpu-manager] mount necessary directories so that GSP firmware can be used
NVIDIA · Jul 25, 2024 · 7848af2 · 7848af2
2 parents bfefa4d + 270bed7
commit 7848af2
Show file tree

Hide file tree

Showing 7 changed files with 446 additions and 1 deletion.
diff --git a/assets/state-vgpu-manager/0210_clusterrole.yaml b/assets/state-vgpu-manager/0210_clusterrole.yaml
@@ -14,7 +14,6 @@ rules:
   - ""
   resources:
   - nodes
-  - pods
   verbs:
   - get
   - list
@@ -23,6 +22,18 @@ rules:
 - apiGroups:
   - ""
   resources:
+  - pods
   - pods/eviction
   verbs:
+  - create
+  - delete
+  - get
+  - list
+  - patch
+  - update
+- apiGroups:
+  - apps
+  resources:
+  - daemonsets
+  verbs:
   - get
diff --git a/assets/state-vgpu-manager/0500_daemonset.yaml b/assets/state-vgpu-manager/0500_daemonset.yaml
@@ -84,6 +84,10 @@ spec:
               mountPath: /var/log
             - name: dev-log
               mountPath: /dev/log
+            - name: firmware-search-path
+              mountPath: /sys/module/firmware_class/parameters/path
+            - name: nv-firmware
+              mountPath: /lib/firmware
         # Only kept when OpenShift DriverToolkit side-car is enabled.
         - image: "FILLED BY THE OPERATOR"
           imagePullPolicy: IfNotPresent
@@ -114,6 +118,10 @@ spec:
               mountPath: /var/log
             - name: dev-log
               mountPath: /dev/log
+            - name: firmware-search-path
+              mountPath: /sys/module/firmware_class/parameters/path
+            - name: nv-firmware
+              mountPath: /lib/firmware
       volumes:
         - name: run-nvidia
           hostPath:
@@ -144,3 +152,10 @@ spec:
         - name: dev-log
           hostPath:
             path: /dev/log
+        - name: firmware-search-path
+          hostPath:
+            path: /sys/module/firmware_class/parameters/path
+        - name: nv-firmware
+          hostPath:
+            path: /run/nvidia/driver/lib/firmware
+            type: DirectoryOrCreate
diff --git a/internal/state/driver_test.go b/internal/state/driver_test.go
@@ -576,6 +576,45 @@ func TestVGPUHostManagerDaemonset(t *testing.T) {
 	require.Equal(t, string(o), actual)
 }
 
+func TestVGPUHostManagerDaemonsetOpenShift(t *testing.T) {
+	const (
+		testName     = "driver-vgpu-host-manager-openshift"
+		rhcosVersion = "413.92.202304252344-0"
+		toolkitImage = "quay.io/openshift-release-dev/ocp-v4.0-art-dev@sha256:7fecaebc1d51b28bc3548171907e4d91823a031d7a6a694ab686999be2b4d867"
+	)
+	state, err := NewStateDriver(nil, nil, manifestDir)
+	require.Nil(t, err)
+	stateDriver, ok := state.(*stateDriver)
+	require.True(t, ok)
+
+	renderData := getMinimalDriverRenderData()
+	renderData.Driver.Spec.DriverType = nvidiav1alpha1.VGPUHostManager
+	renderData.Driver.Name = "nvidia-vgpu-manager-openshift"
+	renderData.Driver.AppName = "nvidia-vgpu-manager-openshift-7c6d7bd86b"
+	renderData.Driver.ImagePath = "nvcr.io/nvidia/vgpu-manager:525.85.03-rhel8.0"
+	renderData.Driver.OSVersion = "rhel8.0"
+	renderData.Openshift = &openshiftSpec{
+		ToolkitImage: toolkitImage,
+		RHCOSVersion: rhcosVersion,
+	}
+	renderData.Runtime.OpenshiftDriverToolkitEnabled = true
+	renderData.Runtime.OpenshiftVersion = "4.13"
+
+	objs, err := stateDriver.renderer.RenderObjects(
+		&render.TemplatingData{
+			Data: renderData,
+		})
+	require.Nil(t, err)
+
+	actual, err := getYAMLString(objs)
+	require.Nil(t, err)
+
+	o, err := os.ReadFile(filepath.Join(manifestResultDir, testName+".yaml"))
+	require.Nil(t, err)
+
+	require.Equal(t, string(o), actual)
+}
+
 func getMinimalDriverRenderData() *driverRenderData {
 	return &driverRenderData{
 		Driver: &driverSpec{

diff --git a/internal/state/testdata/golden/driver-gdrcopy-openshift.yaml b/internal/state/testdata/golden/driver-gdrcopy-openshift.yaml
@@ -351,6 +351,10 @@ spec:
         - mountPath: /host-etc/os-release
           name: host-os-release
           readOnly: true
+        - mountPath: /sys/module/firmware_class/parameters/path
+          name: firmware-search-path
+        - mountPath: /lib/firmware
+          name: nv-firmware
       hostPID: true
       imagePullSecrets:
       - name: ngc-secret

diff --git a/internal/state/testdata/golden/driver-openshift-drivertoolkit.yaml b/internal/state/testdata/golden/driver-openshift-drivertoolkit.yaml
@@ -287,6 +287,10 @@ spec:
         - mountPath: /host-etc/os-release
           name: host-os-release
           readOnly: true
+        - mountPath: /sys/module/firmware_class/parameters/path
+          name: firmware-search-path
+        - mountPath: /lib/firmware
+          name: nv-firmware
       hostPID: true
       initContainers:
       - args: