[ExternalNode]Create/Delete AntreaAgentInfo based on ExternalNode (#3757

)

Antrea controller watches ExternalNode create and delete event.
It creates AntreaAgentInfo whose name is the same as ExternalNode name
after ExternalNode is created and delete AntreaAgentInfo when
ExternalNode is deleted.

The change also refactors monitoring part for Node case and lets controller
create AntreaAgentInfo for Node too. With this change,
for both Node and ExternalNode cases, it is always
Antrea controller to create/delete AntreaAgentInfo and
it is always Antrea agent to update AntreaAgentInfo.

Signed-off-by: Mengdie Song <[email protected]>

build/charts/antrea/templates/agent/clusterrole.yaml

@@ -57,9 +57,7 @@ rules:
       - antreaagentinfos
     verbs:
       - get
-      - create
       - update
-      - delete
   - apiGroups:
       - controlplane.antrea.io
     resources:

build/charts/antrea/templates/controller/clusterrole.yaml

@@ -143,6 +143,7 @@ rules:
       - antreaagentinfos
     verbs:
       - list
+      - create
       - delete
   - apiGroups:
       - crd.antrea.io

build/yamls/antrea-aks.yml

@@ -2940,9 +2940,7 @@ rules:
       - antreaagentinfos
     verbs:
       - get
-      - create
       - update
-      - delete
   - apiGroups:
       - controlplane.antrea.io
     resources:
@@ -3287,6 +3285,7 @@ rules:
       - antreaagentinfos
     verbs:
       - list
+      - create
       - delete
   - apiGroups:
       - crd.antrea.io

build/yamls/antrea-eks.yml

@@ -2940,9 +2940,7 @@ rules:
       - antreaagentinfos
     verbs:
       - get
-      - create
       - update
-      - delete
   - apiGroups:
       - controlplane.antrea.io
     resources:
@@ -3287,6 +3285,7 @@ rules:
       - antreaagentinfos
     verbs:
       - list
+      - create
       - delete
   - apiGroups:
       - crd.antrea.io

build/yamls/antrea-gke.yml

@@ -2940,9 +2940,7 @@ rules:
       - antreaagentinfos
     verbs:
       - get
-      - create
       - update
-      - delete
   - apiGroups:
       - controlplane.antrea.io
     resources:
@@ -3287,6 +3285,7 @@ rules:
       - antreaagentinfos
     verbs:
       - list
+      - create
       - delete
   - apiGroups:
       - crd.antrea.io

build/yamls/antrea-ipsec.yml

@@ -2953,9 +2953,7 @@ rules:
       - antreaagentinfos
     verbs:
       - get
-      - create
       - update
-      - delete
   - apiGroups:
       - controlplane.antrea.io
     resources:
@@ -3300,6 +3298,7 @@ rules:
       - antreaagentinfos
     verbs:
       - list
+      - create
       - delete
   - apiGroups:
       - crd.antrea.io

build/yamls/antrea.yml

@@ -2940,9 +2940,7 @@ rules:
       - antreaagentinfos
     verbs:
       - get
-      - create
       - update
-      - delete
   - apiGroups:
       - controlplane.antrea.io
     resources:
@@ -3287,6 +3285,7 @@ rules:
       - antreaagentinfos
     verbs:
       - list
+      - create
       - delete
   - apiGroups:
       - crd.antrea.io

cmd/antrea-controller/controller.go

@@ -172,7 +172,8 @@ func run(o *Options) error {
 
 	controllerQuerier := querier.NewControllerQuerier(networkPolicyController, o.config.APIPort)
 
-	controllerMonitor := monitor.NewControllerMonitor(crdClient, nodeInformer, controllerQuerier)
+	externalNodeEnabled := features.DefaultFeatureGate.Enabled(features.ExternalNode)
+	controllerMonitor := monitor.NewControllerMonitor(crdClient, nodeInformer, externalNodeInformer, controllerQuerier, externalNodeEnabled)
 
 	var egressController *egress.EgressController
 	var externalIPPoolController *externalippool.ExternalIPPoolController

pkg/monitor/agent.go

@@ -18,7 +18,6 @@ import (
 	"context"
 	"time"
 
-	"k8s.io/apimachinery/pkg/api/errors"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/util/wait"
 	"k8s.io/klog/v2"
@@ -59,30 +58,20 @@ func (monitor *agentMonitor) syncAgentCRD() {
 		if monitor.agentCRD, err = monitor.updateAgentCRD(true); err == nil {
 			return
 		}
-		klog.Errorf("Failed to partially update agent monitoring CRD: %v", err)
+		klog.ErrorS(err, "Failed to partially update agent monitoring CRD")
 		monitor.agentCRD = nil
 	}
 
 	monitor.agentCRD, err = monitor.getAgentCRD()
-
-	if errors.IsNotFound(err) {
-		monitor.agentCRD, err = monitor.createAgentCRD()
-		if err != nil {
-			klog.Errorf("Failed to create agent monitoring CRD: %v", err)
-			monitor.agentCRD = nil
-		}
-		return
-	}
-
 	if err != nil {
-		klog.Errorf("Failed to get agent monitoring CRD: %v", err)
+		klog.ErrorS(err, "Failed to get agent monitoring CRD")
 		monitor.agentCRD = nil
 		return
 	}
 
 	monitor.agentCRD, err = monitor.updateAgentCRD(false)
 	if err != nil {
-		klog.Errorf("Failed to entirely update agent monitoring CRD: %v", err)
+		klog.ErrorS(err, "Failed to entirely update agent monitoring CRD")
 		monitor.agentCRD = nil
 	}
 }
@@ -91,18 +80,10 @@ func (monitor *agentMonitor) syncAgentCRD() {
 // So when the pod restarts, it will update this monitoring CRD instead of creating a new one.
 func (monitor *agentMonitor) getAgentCRD() (*v1beta1.AntreaAgentInfo, error) {
 	crdName := monitor.querier.GetNodeConfig().Name
-	klog.V(2).Infof("Getting agent monitoring CRD %+v", crdName)
+	klog.V(2).InfoS("Getting agent monitoring CRD", "name", crdName)
 	return monitor.client.CrdV1beta1().AntreaAgentInfos().Get(context.TODO(), crdName, metav1.GetOptions{})
 }
 
-// createAgentCRD creates a new agent CRD.
-func (monitor *agentMonitor) createAgentCRD() (*v1beta1.AntreaAgentInfo, error) {
-	agentCRD := new(v1beta1.AntreaAgentInfo)
-	monitor.querier.GetAgentInfo(agentCRD, false)
-	klog.V(2).Infof("Creating agent monitoring CRD %+v", agentCRD)
-	return monitor.client.CrdV1beta1().AntreaAgentInfos().Create(context.TODO(), agentCRD, metav1.CreateOptions{})
-}
-
 // updateAgentCRD updates the monitoring CRD.
 func (monitor *agentMonitor) updateAgentCRD(partial bool) (*v1beta1.AntreaAgentInfo, error) {
 	monitor.querier.GetAgentInfo(monitor.agentCRD, partial)