Add a nodeSelector option to enable the controller to manage only a subset of nodes.

Signed-off-by: bingshen.wbs <[email protected]>

Author: BSWANG

deploy/helm/templates/configmap.yaml

@@ -16,5 +16,6 @@ data:
       "enableDevicePlugin": {{ .Values.config.enableDevicePlugin }},
       "enableWebhook": {{ .Values.config.enableWebhook }},
       "smcInitImage": "{{ .Values.config.smcInitImage }}",
-      "enableInitContainerInject": {{ .Values.config.enableInitContainerInject }}
+      "enableInitContainerInject": {{ .Values.config.enableInitContainerInject }},
+      "nodeSelector": {{ .Values.nodeSelector | toJson }}
     }

deploy/helm/templates/daemonset.yaml

@@ -8,6 +8,7 @@ spec:
   selector:
     matchLabels:
       {{- include "alibabacloud-erdma-controller.selectorLabels" . | nindent 6 }}
+      tier: agent
   template:
     metadata:
       {{- with .Values.podAnnotations }}
@@ -16,6 +17,7 @@ spec:
       {{- end }}
       labels:
         {{- include "alibabacloud-erdma-controller.selectorLabels" . | nindent 8 }}
+        tier: agent
     spec:
       hostPID: true
       hostNetwork: true
@@ -83,6 +85,7 @@ spec:
       - name: var-run
         hostPath:
           path: /var/run/
+      priorityClassName: system-node-critical
       {{- with .Values.nodeSelector }}
       nodeSelector:
         {{- toYaml . | nindent 8 }}

deploy/helm/templates/deployment.yaml

@@ -9,6 +9,7 @@ spec:
   selector:
     matchLabels:
       {{- include "alibabacloud-erdma-controller.selectorLabels" . | nindent 6 }}
+      tier: controller
   template:
     metadata:
       {{- with .Values.podAnnotations }}
@@ -17,6 +18,7 @@ spec:
       {{- end }}
       labels:
         {{- include "alibabacloud-erdma-controller.selectorLabels" . | nindent 8 }}
+        tier: controller
     spec:
       {{- with .Values.controller.imagePullSecrets }}
       imagePullSecrets:
@@ -50,11 +52,11 @@ spec:
           secretName: {{ .Release.Name }}
       - name: webhook-vol
         emptyDir: { }
-      {{- with .Values.nodeSelector }}
+      {{- with .Values.controller.nodeSelector }}
       nodeSelector:
         {{- toYaml . | nindent 8 }}
       {{- end }}
-      {{- with .Values.affinity }}
+      {{- with .Values.controller.affinity }}
       affinity:
         {{- toYaml . | nindent 8 }}
       {{- end }}

deploy/helm/values.yaml

@@ -13,6 +13,8 @@ controller:
   imagePullSecrets: []
   nameOverride: ""
   fullnameOverride: ""
+  nodeSelector: {}
+  affinity: {}
 
 agent:
   image:
@@ -39,7 +41,8 @@ podAnnotations: {}
 
 nodeSelector: {}
 
-tolerations: []
+tolerations:
+  - operator: "Exists"
 
 affinity: {}
 

internal/controller/node_controller.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"time"
 
+	"github.com/AliyunContainerService/alibabacloud-erdma-controller/internal/config"
 	"github.com/AliyunContainerService/alibabacloud-erdma-controller/internal/types"
 	"github.com/samber/lo"
 	k8stypes "k8s.io/apimachinery/pkg/types"
@@ -173,23 +174,43 @@ func RemoveERdmaDevices(erdmaClient client.Client, ctx context.Context, nodeName
 	return ctrl.Result{}, nil
 }
 
+func (r *NodeReconciler) OwnNode(node *v1.Node) bool {
+	if node == nil {
+		return false
+	}
+	for k, v := range config.GetConfig().NodeSelector {
+		if node.Labels[k] != v {
+			return false
+		}
+	}
+	return true
+}
+
 // SetupWithManager sets up the controller with the Manager.
 func (r *NodeReconciler) SetupWithManager(mgr ctrl.Manager) error {
 	pred := predicate.TypedFuncs[*v1.Node]{
 		CreateFunc: func(e event.TypedCreateEvent[*v1.Node]) bool {
-			return true
+			return r.OwnNode(e.Object)
 		},
 		DeleteFunc: func(e event.TypedDeleteEvent[*v1.Node]) bool {
-			return true
+			return r.OwnNode(e.Object)
 		},
 		UpdateFunc: func(e event.TypedUpdateEvent[*v1.Node]) bool {
+			if !r.OwnNode(e.ObjectNew) {
+				return false
+			}
+			if !r.OwnNode(e.ObjectOld) && r.OwnNode(e.ObjectNew) {
+				return true
+			}
+
 			if e.ObjectNew.DeletionTimestamp != nil {
 				return true
 			}
+
 			return e.ObjectOld.Spec.ProviderID != e.ObjectNew.Spec.ProviderID
 		},
 		GenericFunc: func(e event.TypedGenericEvent[*v1.Node]) bool {
-			return true
+			return r.OwnNode(e.Object)
 		},
 	}
 	c, err := controller.New("node-controller", mgr, controller.Options{Reconciler: r})

internal/types/config.go

@@ -1,16 +1,17 @@
 package types
 
 type Config struct {
-	Region                    string `json:"region"`
-	ManageNonOwnedERIs        bool   `json:"manageNonOwnedENIs"`
-	ControllerNamespace       string `json:"controllerNamespace"`
-	ControllerName            string `json:"controllerName"`
-	ClusterDomain             string `json:"clusterDomain"`
-	CertDir                   string `json:"certDir"`
-	EnableDevicePlugin        *bool  `json:"enableDevicePlugin"`
-	EnableWebhook             *bool  `json:"enableWebhook"`
-	SMCInitImage              string `json:"smcInitImage"`
-	EnableInitContainerInject *bool  `json:"enableInitContainerInject"`
+	Region                    string            `json:"region"`
+	ManageNonOwnedERIs        bool              `json:"manageNonOwnedENIs"`
+	ControllerNamespace       string            `json:"controllerNamespace"`
+	ControllerName            string            `json:"controllerName"`
+	ClusterDomain             string            `json:"clusterDomain"`
+	CertDir                   string            `json:"certDir"`
+	EnableDevicePlugin        *bool             `json:"enableDevicePlugin"`
+	EnableWebhook             *bool             `json:"enableWebhook"`
+	SMCInitImage              string            `json:"smcInitImage"`
+	EnableInitContainerInject *bool             `json:"enableInitContainerInject"`
+	NodeSelector              map[string]string `json:"nodeSelector"`
 }
 
 type Sensitive string