Refactor startup command to wait for node IP changes

pkg/controller/cluster/client.go

@@ -23,7 +23,7 @@ func newVirtualClient(ctx context.Context, hostClient ctrlruntimeclient.Client,
 	}
 
 	if err := hostClient.Get(ctx, kubeconfigSecretName, &clusterKubeConfig); err != nil {
-		return nil, fmt.Errorf("failed to get kubeconfig secret: %w", err)
+		return nil, err
 	}
 
 	restConfig, err := clientcmd.RESTConfigFromKubeConfig(clusterKubeConfig.Data["kubeconfig.yaml"])

pkg/controller/cluster/server/server.go

@@ -416,9 +416,11 @@ func (s *Server) setupDynamicPersistence() v1.PersistentVolumeClaim {
 func (s *Server) setupStartCommand() (string, error) {
 	var output bytes.Buffer
 
-	tmpl := singleServerTemplate
+	tmpl := StartupCommand
+
+	mode := "single"
 	if *s.cluster.Spec.Servers > 1 {
-		tmpl = HAServerTemplate
+		mode = "ha"
 	}
 
 	tmplCmd, err := template.New("").Parse(tmpl)
@@ -427,10 +429,12 @@ func (s *Server) setupStartCommand() (string, error) {
 	}
 
 	if err := tmplCmd.Execute(&output, map[string]string{
-		"ETCD_DIR":      "/var/lib/rancher/k3s/server/db/etcd",
-		"INIT_CONFIG":   "/opt/rancher/k3s/init/config.yaml",
-		"SERVER_CONFIG": "/opt/rancher/k3s/server/config.yaml",
-		"EXTRA_ARGS":    strings.Join(s.cluster.Spec.ServerArgs, " "),
+		"ETCD_DIR":           "/var/lib/rancher/k3s/server/db/etcd",
+		"INIT_CONFIG":        "/opt/rancher/k3s/init/config.yaml",
+		"SERVER_CONFIG":      "/opt/rancher/k3s/server/config.yaml",
+		"ORCHESTRATION_MODE": mode,
+		"K3K_MODE":           string(s.cluster.Spec.Mode),
+		"EXTRA_ARGS":         strings.Join(s.cluster.Spec.ServerArgs, " "),
 	}); err != nil {
 		return "", err
 	}

pkg/controller/cluster/server/template.go

@@ -1,16 +1,99 @@
 package server
 
-var singleServerTemplate string = `
-if [ -d "{{.ETCD_DIR}}" ]; then
-	# if directory exists then it means its not an initial run
-	/bin/k3s server --cluster-reset --config  {{.INIT_CONFIG}} {{.EXTRA_ARGS}} 2>&1 | tee /var/log/k3s.log
-fi
-rm -f /var/lib/rancher/k3s/server/db/reset-flag
-/bin/k3s server --config {{.INIT_CONFIG}} {{.EXTRA_ARGS}} 2>&1 | tee /var/log/k3s.log`
-
-var HAServerTemplate string = ` 
-if [ ${POD_NAME: -1} == 0 ] && [ ! -d "{{.ETCD_DIR}}" ]; then
+var StartupCommand string = `
+info()
+{
+    echo "[INFO] [$(date +"%c")]" "$@"
+}
+
+fatal()
+{
+    echo "[FATAL] [$(date +"%c")] " "$@" >&2
+    exit 1
+}
+
+# safe mode function to reset node IP after pod restarts
+safe_mode() {
+	CURRENT_IP=""
+    if [ -f /var/lib/rancher/k3s/k3k-node-ip ]; then
+        CURRENT_IP=$(cat /var/lib/rancher/k3s/k3k-node-ip)
+    fi
+
+    if [ -z "$CURRENT_IP" ] || [ "$CURRENT_IP" = "$POD_IP" ] || [ {{.K3K_MODE}} != "virtual" ]; then
+        return
+    fi
+
+	# skipping if the node is starting for the first time
+	if [ -d "{{.ETCD_DIR}}" ]; then
+
+		info "Starting K3s in Safe Mode (Network Policy Disabled) to patch Node IP from ${CURRENT_IP} to ${POD_IP}"
+		/bin/k3s server --disable-network-policy --config $1 {{.EXTRA_ARGS}} > /dev/null 2>&1 &
+		PID=$!
+
+		# Start the loop to wait for the nodeIP to change
+		info "Waiting for Node IP to update to ${POD_IP}."
+		count=0
+		until kubectl get nodes -o wide 2>/dev/null | grep -q "${POD_IP}"; do
+			if ! kill -0 $PID 2>/dev/null; then
+				fatal "safe Mode K3s process died unexpectedly!"
+			fi
+			sleep 2
+			count=$((count+1))
+
+			if [ $count -gt 60 ]; then
+				fatal "timed out waiting for node to change IP from $CURRENT_IP to $POD_IP"
+			fi
+		done
+		
+		info "Node IP is set to ${POD_IP} successfully. Stopping Safe Mode process..."
+		kill $PID
+		wait $PID 2>/dev/null || true
+	fi
+}
+
+single_server() {
+	info "Starting single node setup..."
+
+	# checking for existing data in single server if found we must perform reset
+	if [ -d "{{.ETCD_DIR}}" ]; then
+		info "Existing data found in single node setup. Performing cluster-reset to ensure quorum..."
+
+		if ! /bin/k3s server --cluster-reset --config {{.INIT_CONFIG}} {{.EXTRA_ARGS}} > /dev/null 2>&1; then
+			fatal "cluster reset failed!"
+		fi
+		info "Cluster reset complete. Removing Reset flag file."
+		rm -f /var/lib/rancher/k3s/server/db/reset-flag
+	fi
+
+	# entering safe mode to ensure correct NodeIP
+	safe_mode {{.INIT_CONFIG}}
+
+	info "Adding pod IP file."
+	echo $POD_IP > /var/lib/rancher/k3s/k3k-node-ip
+
 	/bin/k3s server --config {{.INIT_CONFIG}} {{.EXTRA_ARGS}} 2>&1 | tee /var/log/k3s.log
-else 
-	/bin/k3s server --config {{.SERVER_CONFIG}} {{.EXTRA_ARGS}} 2>&1 | tee /var/log/k3s.log 
+}
+
+ha_server() {
+	info "Starting pod $POD_NAME in HA node setup"
+
+	if [ ${POD_NAME: -1} == 0 ] && [ ! -d "{{.ETCD_DIR}}" ]; then
+		info "Adding pod IP file."
+		echo $POD_IP > /var/lib/rancher/k3s/k3k-node-ip
+
+		/bin/k3s server --config {{.INIT_CONFIG}} {{.EXTRA_ARGS}} 2>&1 | tee /var/log/k3s.log
+	else
+		safe_mode {{.SERVER_CONFIG}}
+
+		info "Adding pod IP file."
+		echo $POD_IP > /var/lib/rancher/k3s/k3k-node-ip
+
+		/bin/k3s server --config {{.SERVER_CONFIG}} {{.EXTRA_ARGS}} 2>&1 | tee /var/log/k3s.info 
+	fi
+}
+
+if [ {{.ORCHESTRATION_MODE}} == "ha" ]; then
+	ha_server
+else
+	single_server
 fi`

pkg/controller/cluster/statefulset.go

@@ -120,7 +120,7 @@ func (p *StatefulSetReconciler) handleServerPod(ctx context.Context, cluster v1b
 
 	if pod.DeletionTimestamp.IsZero() {
 		if controllerutil.AddFinalizer(pod, etcdPodFinalizerName) {
-			log.V(1).Info("Server Pod is being deleted. Removing finalizer", "pod", pod.Name, "namespace", pod.Namespace)
+			log.V(1).Info("Server Pod is being created. Adding finalizer", "pod", pod.Name, "namespace", pod.Namespace)
 
 			return p.Client.Update(ctx, pod)
 		}

tests/cluster_update_test.go

@@ -127,6 +127,11 @@ var _ = When("a shared mode cluster update its envs", Label(e2eTestLabel), Label
 			serverPods := listServerPods(ctx, virtualCluster)
 			g.Expect(len(serverPods)).To(Equal(1))
 
+			serverPod := serverPods[0]
+			_, cond := pod.GetPodCondition(&serverPod.Status, v1.PodReady)
+			g.Expect(cond).NotTo(BeNil())
+			g.Expect(cond.Status).To(BeEquivalentTo(metav1.ConditionTrue))
+
 			serverEnv1, ok := getEnv(&serverPods[0], "TEST_SERVER_ENV_1")
 			g.Expect(ok).To(BeTrue())
 			g.Expect(serverEnv1).To(Equal("upgraded"))
@@ -145,6 +150,11 @@ var _ = When("a shared mode cluster update its envs", Label(e2eTestLabel), Label
 			aPods := listAgentPods(ctx, virtualCluster)
 			g.Expect(aPods).To(HaveLen(len(nodes.Items)))
 
+			agentPod := aPods[0]
+			_, cond = pod.GetPodCondition(&agentPod.Status, v1.PodReady)
+			g.Expect(cond).NotTo(BeNil())
+			g.Expect(cond.Status).To(BeEquivalentTo(metav1.ConditionTrue))
+
 			agentEnv1, ok := getEnv(&aPods[0], "TEST_AGENT_ENV_1")
 			g.Expect(ok).To(BeTrue())
 			g.Expect(agentEnv1).To(Equal("upgraded"))
@@ -213,6 +223,11 @@ var _ = When("a shared mode cluster update its server args", Label(e2eTestLabel)
 			sPods := listServerPods(ctx, virtualCluster)
 			g.Expect(len(sPods)).To(Equal(1))
 
+			serverPod := sPods[0]
+			_, cond := pod.GetPodCondition(&serverPod.Status, v1.PodReady)
+			g.Expect(cond).NotTo(BeNil())
+			g.Expect(cond.Status).To(BeEquivalentTo(metav1.ConditionTrue))
+
 			g.Expect(isArgFound(&sPods[0], "--node-label=test_server=upgraded")).To(BeTrue())
 		}).
 			WithPolling(time.Second * 2).
@@ -330,6 +345,11 @@ var _ = When("a virtual mode cluster update its envs", Label(e2eTestLabel), Labe
 			serverPods := listServerPods(ctx, virtualCluster)
 			g.Expect(len(serverPods)).To(Equal(1))
 
+			serverPod := serverPods[0]
+			_, cond := pod.GetPodCondition(&serverPod.Status, v1.PodReady)
+			g.Expect(cond).NotTo(BeNil())
+			g.Expect(cond.Status).To(BeEquivalentTo(metav1.ConditionTrue))
+
 			serverEnv1, ok := getEnv(&serverPods[0], "TEST_SERVER_ENV_1")
 			g.Expect(ok).To(BeTrue())
 			g.Expect(serverEnv1).To(Equal("upgraded"))
@@ -345,6 +365,11 @@ var _ = When("a virtual mode cluster update its envs", Label(e2eTestLabel), Labe
 			aPods := listAgentPods(ctx, virtualCluster)
 			g.Expect(len(aPods)).To(Equal(1))
 
+			agentPod := aPods[0]
+			_, cond = pod.GetPodCondition(&agentPod.Status, v1.PodReady)
+			g.Expect(cond).NotTo(BeNil())
+			g.Expect(cond.Status).To(BeEquivalentTo(metav1.ConditionTrue))
+
 			agentEnv1, ok := getEnv(&aPods[0], "TEST_AGENT_ENV_1")
 			g.Expect(ok).To(BeTrue())
 			g.Expect(agentEnv1).To(Equal("upgraded"))
@@ -416,6 +441,11 @@ var _ = When("a virtual mode cluster update its server args", Label(e2eTestLabel
 			sPods := listServerPods(ctx, virtualCluster)
 			g.Expect(len(sPods)).To(Equal(1))
 
+			serverPod := sPods[0]
+			_, cond := pod.GetPodCondition(&serverPod.Status, v1.PodReady)
+			g.Expect(cond).NotTo(BeNil())
+			g.Expect(cond.Status).To(BeEquivalentTo(metav1.ConditionTrue))
+
 			g.Expect(isArgFound(&sPods[0], "--node-label=test_server=upgraded")).To(BeTrue())
 		}).
 			WithPolling(time.Second * 2).

tests/common_test.go

@@ -140,9 +140,6 @@ func NewCluster(namespace string) *v1beta1.Cluster {
 			Persistence: v1beta1.PersistenceConfig{
 				Type: v1beta1.EphemeralPersistenceMode,
 			},
-			ServerArgs: []string{
-				"--disable-network-policy",
-			},
 		},
 	}
 }