[GPU Driver Upgrade] use common label to select DaemonSets for stateMap

tian-men · Sep 26, 2023 · bd806dd · bd806dd
1 parent edcaae9
commit bd806dd
Show file tree

Hide file tree

Showing 10 changed files with 24 additions and 9 deletions.
diff --git a/assets/state-driver/0500_daemonset.yaml b/assets/state-driver/0500_daemonset.yaml
@@ -4,6 +4,7 @@ metadata:
   labels:
     app: nvidia-driver-daemonset
     nvidia.com/precompiled: "false"
+    app.kubernetes.io/component: nvidia-driver
   name: nvidia-driver-daemonset
   namespace: "FILLED BY THE OPERATOR"
   annotations:
@@ -21,6 +22,7 @@ spec:
       labels:
         app: nvidia-driver-daemonset
         nvidia.com/precompiled: "false"
+        app.kubernetes.io/component: nvidia-driver
     spec:
       nodeSelector:
         nvidia.com/gpu.deploy.driver: "true"

diff --git a/controllers/upgrade_controller.go b/controllers/upgrade_controller.go
@@ -62,6 +62,10 @@ const (
 	DriverLabelValue = "nvidia-driver-daemonset"
 	// UpgradeSkipDrainLabelSelector indicates the pod selector label to skip with drain
 	UpgradeSkipDrainLabelSelector = "nvidia.com/gpu-driver-upgrade-drain.skip!=true"
+	// AppComponentLabelKey indicates the label key of the component
+	AppComponentLabelKey = "app.kubernetes.io/component"
+	// AppComponentLabelValue indicates the label values of the nvidia-gpu-driver component
+	AppComponentLabelValue = "nvidia-driver"
 )
 
 //nolint
@@ -119,15 +123,8 @@ func (r *UpgradeReconciler) Reconcile(ctx context.Context, req ctrl.Request) (ct
 		clusterPolicyCtrl.operatorMetrics.driverAutoUpgradeEnabled.Set(driverAutoUpgradeEnabled)
 	}
 
-	driverLabelKey := DriverLabelKey
-	driverLabelValue := DriverLabelValue
-	if clusterPolicyCtrl.openshift != "" && clusterPolicyCtrl.ocpDriverToolkit.enabled {
-		// For OCP, when DTK is enabled app=nvidia-driver-daemonset label is not constant and changes
-		// based on rhcos version. Hence use DTK label instead
-		driverLabelKey = ocpDriverToolkitIdentificationLabel
-		driverLabelValue = ocpDriverToolkitIdentificationValue
-	}
-	state, err := r.StateManager.BuildState(ctx, clusterPolicyCtrl.operatorNamespace, map[string]string{driverLabelKey: driverLabelValue})
+	state, err := r.StateManager.BuildState(ctx, clusterPolicyCtrl.operatorNamespace,
+		map[string]string{AppComponentLabelKey: AppComponentLabelValue})
 	if err != nil {
 		r.Log.Error(err, "Failed to build cluster upgrade state")
 		return ctrl.Result{}, err

diff --git a/internal/state/testdata/golden/driver-additional-configs.yaml b/internal/state/testdata/golden/driver-additional-configs.yaml
@@ -122,6 +122,7 @@ metadata:
     openshift.io/scc: nvidia-gpu-driver-ubuntu22.04
   labels:
     app: nvidia-driver-daemonset
+    app.kubernetes.io/component: nvidia-driver
     nvidia.com/precompiled: "false"
   name: nvidia-gpu-driver-ubuntu22.04
   namespace: test-operator
@@ -135,6 +136,7 @@ spec:
         kubectl.kubernetes.io/default-container: nvidia-driver-ctr
       labels:
         app: nvidia-driver-daemonset
+        app.kubernetes.io/component: nvidia-driver
         nvidia.com/precompiled: "false"
     spec:
       containers:

diff --git a/internal/state/testdata/golden/driver-gds.yaml b/internal/state/testdata/golden/driver-gds.yaml
@@ -122,6 +122,7 @@ metadata:
     openshift.io/scc: nvidia-gpu-driver-ubuntu22.04
   labels:
     app: nvidia-driver-daemonset
+    app.kubernetes.io/component: nvidia-driver
     nvidia.com/precompiled: "false"
   name: nvidia-gpu-driver-ubuntu22.04
   namespace: test-operator
@@ -135,6 +136,7 @@ spec:
         kubectl.kubernetes.io/default-container: nvidia-driver-ctr
       labels:
         app: nvidia-driver-daemonset
+        app.kubernetes.io/component: nvidia-driver
         nvidia.com/precompiled: "false"
     spec:
       containers:

diff --git a/internal/state/testdata/golden/driver-minimal.yaml b/internal/state/testdata/golden/driver-minimal.yaml
@@ -122,6 +122,7 @@ metadata:
     openshift.io/scc: nvidia-gpu-driver-ubuntu22.04
   labels:
     app: nvidia-driver-daemonset
+    app.kubernetes.io/component: nvidia-driver
     nvidia.com/precompiled: "false"
   name: nvidia-gpu-driver-ubuntu22.04
   namespace: test-operator
@@ -135,6 +136,7 @@ spec:
         kubectl.kubernetes.io/default-container: nvidia-driver-ctr
       labels:
         app: nvidia-driver-daemonset
+        app.kubernetes.io/component: nvidia-driver
         nvidia.com/precompiled: "false"
     spec:
       containers:

diff --git a/internal/state/testdata/golden/driver-openshift-drivertoolkit.yaml b/internal/state/testdata/golden/driver-openshift-drivertoolkit.yaml
@@ -175,6 +175,7 @@ metadata:
     openshift.io/scc: nvidia-gpu-driver-openshift
   labels:
     app: nvidia-driver-daemonset-413.92.202304252344-0
+    app.kubernetes.io/component: nvidia-driver
     nvidia.com/precompiled: "false"
     openshift.driver-toolkit: "true"
     openshift.driver-toolkit.rhcos: 413.92.202304252344-0
@@ -190,6 +191,7 @@ spec:
         kubectl.kubernetes.io/default-container: nvidia-driver-ctr
       labels:
         app: nvidia-driver-daemonset-413.92.202304252344-0
+        app.kubernetes.io/component: nvidia-driver
         nvidia.com/precompiled: "false"
         openshift.driver-toolkit: "true"
     spec:

diff --git a/internal/state/testdata/golden/driver-precompiled.yaml b/internal/state/testdata/golden/driver-precompiled.yaml
@@ -122,6 +122,7 @@ metadata:
     openshift.io/scc: nvidia-gpu-driver-ubuntu22.04
   labels:
     app: nvidia-driver-daemonset
+    app.kubernetes.io/component: nvidia-driver
     nvidia.com/precompiled: "true"
   name: nvidia-gpu-driver-ubuntu22.04-5.4.0-150-generic
   namespace: test-operator
@@ -135,6 +136,7 @@ spec:
         kubectl.kubernetes.io/default-container: nvidia-driver-ctr
       labels:
         app: nvidia-driver-daemonset
+        app.kubernetes.io/component: nvidia-driver
         nvidia.com/precompiled: "true"
     spec:
       containers:

diff --git a/internal/state/testdata/golden/driver-rdma-hostmofed.yaml b/internal/state/testdata/golden/driver-rdma-hostmofed.yaml
@@ -122,6 +122,7 @@ metadata:
     openshift.io/scc: nvidia-gpu-driver-ubuntu22.04
   labels:
     app: nvidia-driver-daemonset
+    app.kubernetes.io/component: nvidia-driver
     nvidia.com/precompiled: "false"
   name: nvidia-gpu-driver-ubuntu22.04
   namespace: test-operator
@@ -135,6 +136,7 @@ spec:
         kubectl.kubernetes.io/default-container: nvidia-driver-ctr
       labels:
         app: nvidia-driver-daemonset
+        app.kubernetes.io/component: nvidia-driver
         nvidia.com/precompiled: "false"
     spec:
       containers:

diff --git a/internal/state/testdata/golden/driver-rdma.yaml b/internal/state/testdata/golden/driver-rdma.yaml
@@ -122,6 +122,7 @@ metadata:
     openshift.io/scc: nvidia-gpu-driver-ubuntu22.04
   labels:
     app: nvidia-driver-daemonset
+    app.kubernetes.io/component: nvidia-driver
     nvidia.com/precompiled: "false"
   name: nvidia-gpu-driver-ubuntu22.04
   namespace: test-operator
@@ -135,6 +136,7 @@ spec:
         kubectl.kubernetes.io/default-container: nvidia-driver-ctr
       labels:
         app: nvidia-driver-daemonset
+        app.kubernetes.io/component: nvidia-driver
         nvidia.com/precompiled: "false"
     spec:
       containers:

diff --git a/manifests/state-driver/0500_daemonset.yaml b/manifests/state-driver/0500_daemonset.yaml
@@ -8,6 +8,7 @@ metadata:
     app: nvidia-driver-daemonset
     {{- end }}
     nvidia.com/precompiled: {{ toString (deref .Driver.Spec.UsePrecompiled) | quote }}
+    app.kubernetes.io/component: "nvidia-driver"
     {{- if and (.Openshift) (.Runtime.OpenshiftDriverToolkitEnabled) }}
     openshift.driver-toolkit.rhcos: {{ .Openshift.RHCOSVersion | quote }}
     openshift.driver-toolkit: "true"
@@ -50,6 +51,7 @@ spec:
         app: nvidia-driver-daemonset
         {{- end }}
         nvidia.com/precompiled: {{ toString (deref .Driver.Spec.UsePrecompiled) | quote }}
+        app.kubernetes.io/component: "nvidia-driver"
     spec:
       nodeSelector:
         nvidia.com/gpu.deploy.driver: "true"