Merge pull request #3129 from cyclinder/cherry_pick_0.9

Spidercoordinator: It able to get CIDR from kubeadm-config

docs/usage/install/overlay/get-started-calico-zh_cn.md

@@ -88,10 +88,31 @@ status:
   serviceCIDR:
   - 10.233.0.0/18
 ```
+
+> 目前 Spiderpool 优先通过查询 `kube-system/kubeadm-config` ConfigMap 获取集群的 Pod 和 Service 子网。 如果 kubeadm-config 不存在导致无法获取集群子网，那么 Spiderpool 会从 Kube-controller-manager Pod 中获取集群 Pod 和 Service 的子网。 如果您集群的 Kube-controller-manager 组件以 `systemd` 方式而不是以静态 Pod 运行。那么 Spiderpool 仍然无法获取集群的子网信息。
 
-> 1.如果 phase 不为 Synced, 那么将会阻止 Pod 被创建
->
-> 2.如果 overlayPodCIDR 不正常, 可能会导致通信问题
+如果上面两种方式都失败，Spiderpool 会同步 status.phase 为 NotReady, 这将会阻止 Pod 被创建。我们可以通过下面解决异常情况:
+
+- 手动创建 kubeadm-config ConfigMap, 并正确配置集群的子网信息:
+
+```shell
+export POD_SUBNET=<YOUR_POD_SUBNET>
+export SERVICE_SUBNET=<YOUR_SERVICE_SUBNET>
+cat << EOF | kubectl apply -f -
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: kubeadm-config
+  namespace: kube-system
+data:
+  ClusterConfiguration: 
+    networking:
+      podSubnet: ${POD_SUBNET}
+      serviceSubnet: ${SERVICE_SUBNET}
+EOF
+```
+
+一旦创建完成，Spiderpool 将会自动同步其状态。
 
 ### 创建 SpiderIPPool
 

docs/usage/install/overlay/get-started-calico.md

@@ -84,9 +84,30 @@ status:
   - 10.233.0.0/18
 ```
 
-> 1.If the phase is not synced, the pod will be prevented from being created.
->
-> 2.If the overlayPodCIDR does not meet expectations, it may cause pod communication issue.
+> At present, Spiderpool prioritizes obtaining the cluster's Pod and Service subnets by querying the kube-system/kubeadm-config ConfigMap. If the kubeadm-config does not exist, causing the failure to obtain the cluster subnet, Spiderpool will attempt to retrieve the cluster Pod and Service subnets from the kube-controller-manager Pod. If the kube-controller-manager component in your cluster runs in systemd mode instead of as a static Pod, Spiderpool still cannot retrieve the cluster's subnet information.
+
+If both of the above methods fail, Spiderpool will synchronize the status.phase as NotReady, preventing Pod creation. To address such abnormal situations, we can take either of the following approaches:
+
+- Manually create the kubeadm-config ConfigMap and correctly configure the cluster's subnet information:
+
+```shell
+export POD_SUBNET=<YOUR_POD_SUBNET>
+export SERVICE_SUBNET=<YOUR_SERVICE_SUBNET>
+cat << EOF | kubectl apply -f -
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: kubeadm-config
+  namespace: kube-system
+data:
+  ClusterConfiguration: 
+    networking:
+      podSubnet: ${POD_SUBNET}
+      serviceSubnet: ${SERVICE_SUBNET}
+EOF
+```
+
+Once created, Spiderpool will automatically synchronize its status.
 
 ### Create SpiderIPPool
 

docs/usage/install/overlay/get-started-cilium-zh_cn.md

@@ -85,9 +85,30 @@ status:
   - 10.233.0.0/18
 ```
 
-> 1.如果 phase 不为 Synced, 那么将会阻止 Pod 被创建
->
-> 2.如果 overlayPodCIDR 不正常, 可能会导致通信问题
+> 目前 Spiderpool 优先通过查询 `kube-system/kubeadm-config` ConfigMap 获取集群的 Pod 和 Service 子网。 如果 kubeadm-config 不存在导致无法获取集群子网，那么 Spiderpool 会从 Kube-controller-manager Pod 中获取集群 Pod 和 Service 的子网。 如果您集群的 Kube-controller-manager 组件以 `systemd` 方式而不是以静态 Pod 运行。那么 Spiderpool 仍然无法获取集群的子网信息。
+
+如果上面两种方式都失败，Spiderpool 会同步 status.phase 为 NotReady, 这将会阻止 Pod 被创建。我们可以通过下面的方式解决异常情况:
+
+- 手动创建 kubeadm-config ConfigMap, 并正确配置集群的子网信息:
+
+```shell
+export POD_SUBNET=<YOUR_POD_SUBNET>
+export SERVICE_SUBNET=<YOUR_SERVICE_SUBNET>
+cat << EOF | kubectl apply -f -
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: kubeadm-config
+  namespace: kube-system
+data:
+  ClusterConfiguration: |
+  networking:
+    podSubnet: ${POD_SUBNET}
+    serviceSubnet: ${SERVICE_SUBNET}
+EOF
+```
+
+一旦创建完成，Spiderpool 将会自动同步其状态。
 
 ### 创建 SpiderIPPool
 

docs/usage/install/overlay/get-started-cilium.md

@@ -85,9 +85,30 @@ status:
   - 10.233.0.0/18
 ```
 
-> 1.If the phase is not synced, the pod will be prevented from being created.
->
-> 2.If the overlayPodCIDR does not meet expectations, it may cause pod communication issue.
+> At present, Spiderpool prioritizes obtaining the cluster's Pod and Service subnets by querying the kube-system/kubeadm-config ConfigMap. If the kubeadm-config does not exist, causing the failure to obtain the cluster subnet, Spiderpool will attempt to retrieve the cluster Pod and Service subnets from the kube-controller-manager Pod. If the kube-controller-manager component in your cluster runs in systemd mode instead of as a static Pod, Spiderpool still cannot retrieve the cluster's subnet information.
+
+If both of the above methods fail, Spiderpool will synchronize the status.phase as NotReady, preventing Pod creation. To address such abnormal situations, we can take either of the following approaches:
+
+- Manually create the kubeadm-config ConfigMap and correctly configure the cluster's subnet information:
+
+```shell
+export POD_SUBNET=<YOUR_POD_SUBNET>
+export SERVICE_SUBNET=<YOUR_SERVICE_SUBNET>
+cat << EOF | kubectl apply -f -
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: kubeadm-config
+  namespace: kube-system
+data:
+  ClusterConfiguration: 
+  networking:
+    podSubnet: ${POD_SUBNET}
+    serviceSubnet: ${SERVICE_SUBNET}
+EOF
+```
+
+Once created, Spiderpool will automatically synchronize its status.
 
 ### Create SpiderIPPool
 

pkg/coordinatormanager/coordinator_informer.go

@@ -334,36 +334,8 @@ func (cc *CoordinatorController) syncHandler(ctx context.Context, coordinatorNam
 }
 
 func (cc *CoordinatorController) fetchPodAndServerCIDR(ctx context.Context, logger *zap.Logger, coordCopy *spiderpoolv2beta1.SpiderCoordinator) (*spiderpoolv2beta1.SpiderCoordinator, error) {
-	var err error
-	var cmPodList corev1.PodList
-	if err := cc.APIReader.List(ctx, &cmPodList, client.MatchingLabels{"component": "kube-controller-manager"}); err != nil {
-		event.EventRecorder.Eventf(
-			coordCopy,
-			corev1.EventTypeWarning,
-			"ClusterNotReady",
-			err.Error(),
-		)
-
-		setStatus2NoReady(logger, coordCopy)
-		return coordCopy, err
-	}
-	if len(cmPodList.Items) == 0 {
-		msg := `Failed to get kube-controller-manager Pod with label "component: kube-controller-manager"`
-		event.EventRecorder.Eventf(
-			coordCopy,
-			corev1.EventTypeWarning,
-			"ClusterNotReady",
-			msg,
-		)
-
-		setStatus2NoReady(logger, coordCopy)
-		return coordCopy, err
-	}
-
-	k8sPodCIDR, k8sServiceCIDR := extractK8sCIDR(&cmPodList.Items[0])
 	if *coordCopy.Spec.PodCIDRType == auto {
-		var podCidrType string
-		podCidrType, err = fetchType(cc.DefaultCniConfDir)
+		podCidrType, err := fetchType(cc.DefaultCniConfDir)
 		if err != nil {
 			if apierrors.IsNotFound(err) {
 				event.EventRecorder.Eventf(
@@ -381,6 +353,30 @@ func (cc *CoordinatorController) fetchPodAndServerCIDR(ctx context.Context, logg
 		coordCopy.Spec.PodCIDRType = &podCidrType
 	}
 
+	var err error
+	var cm *corev1.ConfigMap
+	var k8sPodCIDR, k8sServiceCIDR []string
+	if err := cc.APIReader.Get(ctx, types.NamespacedName{Namespace: metav1.NamespaceSystem, Name: "kubeadm-config"}, cm); err == nil && cm != nil {
+		logger.Sugar().Infof("Trying to fetch the ClusterCIDR from kube-system/kubeadm-config")
+		k8sPodCIDR, k8sServiceCIDR = ExtractK8sCIDRFromKubeadmConfigMap(cm)
+	} else {
+		logger.Sugar().Warn("kube-system/kubeadm-config is no found, trying to fetch the ClusterCIDR from kube-controller-manager Pod")
+		var cmPodList corev1.PodList
+		err = cc.APIReader.List(ctx, &cmPodList, client.MatchingLabels{"component": "kube-controller-manager"})
+		if err != nil {
+			logger.Sugar().Errorf("failed to get kube-controller-manager Pod with label \"component: kube-controller-manager\": %v", err)
+			event.EventRecorder.Eventf(
+				coordCopy,
+				corev1.EventTypeWarning,
+				"ClusterNotReady",
+				"Neither kubeadm-config ConfigMap nor kube-controller-manager Pod can be found",
+			)
+			setStatus2NoReady(logger, coordCopy)
+			return coordCopy, err
+		}
+		k8sPodCIDR, k8sServiceCIDR = ExtractK8sCIDRFromKCMPod(&cmPodList.Items[0])
+	}
+
 	switch *coordCopy.Spec.PodCIDRType {
 	case cluster:
 		if cc.caliCtrlCanncel != nil {
@@ -538,7 +534,42 @@ func (cc *CoordinatorController) fetchCiliumCIDR(ctx context.Context, logger *za
 	return nil
 }
 
-func extractK8sCIDR(kcm *corev1.Pod) ([]string, []string) {
+func ExtractK8sCIDRFromKubeadmConfigMap(cm *corev1.ConfigMap) ([]string, []string) {
+	var podCIDR, serviceCIDR []string
+
+	podReg := regexp.MustCompile(`podSubnet: (.*)`)
+	serviceReg := regexp.MustCompile(`serviceSubnet: (.*)`)
+
+	var podSubnets, serviceSubnets []string
+	for _, data := range cm.Data {
+		podSubnets = podReg.FindStringSubmatch(data)
+		serviceSubnets = serviceReg.FindStringSubmatch(data)
+	}
+
+	if len(podSubnets) != 0 {
+		for _, cidr := range strings.Split(podSubnets[1], ",") {
+			_, _, err := net.ParseCIDR(cidr)
+			if err != nil {
+				continue
+			}
+			podCIDR = append(podCIDR, cidr)
+		}
+	}
+
+	if len(serviceSubnets) != 0 {
+		for _, cidr := range strings.Split(serviceSubnets[1], ",") {
+			_, _, err := net.ParseCIDR(cidr)
+			if err != nil {
+				continue
+			}
+			serviceCIDR = append(serviceCIDR, cidr)
+		}
+	}
+
+	return podCIDR, serviceCIDR
+}
+
+func ExtractK8sCIDRFromKCMPod(kcm *corev1.Pod) ([]string, []string) {
 	var podCIDR, serviceCIDR []string
 
 	podReg := regexp.MustCompile(`--cluster-cidr=(.*)`)

test/doc/spidercoodinator.md

@@ -10,3 +10,4 @@
 | V00006  | status.phase is not-ready, expect the cidr of status to be empty                                          | p3       |       |  done  |       |
 | V00007  | spidercoordinator has the lowest priority                                                                 | p3       |       |  done  |       |
 | V00008  | status.phase is not-ready, pods will fail to run                                                          | p3       |       |  done  |       |
+| V00009 | it can get the clusterCIDR from kubeadmConfig or kube-controller-manager pod | p3 |  | done|