Fix agents removing configured supervisor address

We shouldn't be replacing the configured server address on agents. Doing
so breaks the agent's ability to fall back to the fixed registration
endpoint when all servers are down, since we replaced it with the first
discovered apiserver address. The fixed registration endpoint will be
restored as default when the service is restarted, but this is not the
correct behavior. This should have only been done on etcd-only nodes
that start up using their local supervisor, but need to switch to a
control-plane node as soon as one is available.

Signed-off-by: Brad Davidson <brad.davidson@rancher.com>

pkg/agent/config/config.go

@@ -103,7 +103,7 @@ func APIServers(ctx context.Context, node *config.Node, proxy proxy.Proxy) []str
 			return false, err
 		}
 		if len(addresses) == 0 {
-			logrus.Infof("Waiting for apiserver addresses")
+			logrus.Infof("Waiting for supervisor to provide apiserver addresses")
 			return false, nil
 		}
 		return true, nil

pkg/agent/run.go

@@ -530,19 +530,30 @@ func setupTunnelAndRunAgent(ctx context.Context, nodeConfig *daemonconfig.Node,
 }
 
 func waitForAPIServerAddresses(ctx context.Context, nodeConfig *daemonconfig.Node, cfg cmds.Agent, proxy proxy.Proxy) error {
+	var localSupervisorDefault bool
+	if addresses := proxy.SupervisorAddresses(); len(addresses) > 0 {
+		host, _, _ := net.SplitHostPort(addresses[0])
+		if host == "127.0.0.1" || host == "::1" {
+			localSupervisorDefault = true
+		}
+	}
+
 	for {
 		select {
 		case <-time.After(5 * time.Second):
-			logrus.Info("Waiting for apiserver addresses")
+			logrus.Info("Waiting for control-plane node to register apiserver addresses in etcd")
 		case addresses := <-cfg.APIAddressCh:
 			for i, a := range addresses {
 				host, _, err := net.SplitHostPort(a)
 				if err == nil {
 					addresses[i] = net.JoinHostPort(host, strconv.Itoa(nodeConfig.ServerHTTPSPort))
-					if i == 0 {
-						proxy.SetSupervisorDefault(addresses[i])
 				}
 			}
+			// If this is an etcd-only node that started up using its local supervisor,
+			// switch to using a control-plane node as the supervisor. Otherwise, leave the
+			// configured server address as the default.
+			if localSupervisorDefault && len(addresses) > 0 {
+				proxy.SetSupervisorDefault(addresses[0])
 			}
 			proxy.Update(addresses)
 			return nil

pkg/agent/tunnel/tunnel.go

@@ -124,18 +124,33 @@ func Setup(ctx context.Context, config *daemonconfig.Node, proxy proxy.Proxy) er
 	// The loadbalancer is only disabled when there is a local apiserver.  Servers without a local
 	// apiserver load-balance to themselves initially, then switch over to an apiserver node as soon
 	// as we get some addresses from the code below.
+	var localSupervisorDefault bool
+	if addresses := proxy.SupervisorAddresses(); len(addresses) > 0 {
+		host, _, _ := net.SplitHostPort(addresses[0])
+		if host == "127.0.0.1" || host == "::1" {
+			localSupervisorDefault = true
+		}
+	}
+
 	if proxy.IsSupervisorLBEnabled() && proxy.SupervisorURL() != "" {
 		logrus.Info("Getting list of apiserver endpoints from server")
 		// If not running an apiserver locally, try to get a list of apiservers from the server we're
 		// connecting to. If that fails, fall back to querying the endpoints list from Kubernetes. This
 		// fallback requires that the server we're joining be running an apiserver, but is the only safe
 		// thing to do if its supervisor is down-level and can't provide us with an endpoint list.
-		if addresses := agentconfig.APIServers(ctx, config, proxy); len(addresses) > 0 {
+		addresses := agentconfig.APIServers(ctx, config, proxy)
+		logrus.Infof("Got apiserver addresses from supervisor: %v", addresses)
+
+		if len(addresses) > 0 {
+			if localSupervisorDefault {
 				proxy.SetSupervisorDefault(addresses[0])
+			}
 			proxy.Update(addresses)
 		} else {
 			if endpoint, _ := client.CoreV1().Endpoints(metav1.NamespaceDefault).Get(ctx, "kubernetes", metav1.GetOptions{}); endpoint != nil {
-				if addresses := util.GetAddresses(endpoint); len(addresses) > 0 {
+				addresses = util.GetAddresses(endpoint)
+				logrus.Infof("Got apiserver addresses from kubernetes endpoints: %v", addresses)
+				if len(addresses) > 0 {
 					proxy.Update(addresses)
 				}
 			}