Ignore inactive hosts

cluster/cluster.go

@@ -29,6 +29,7 @@ type Cluster struct {
 	EtcdHosts                        []*hosts.Host
 	WorkerHosts                      []*hosts.Host
 	ControlPlaneHosts                []*hosts.Host
+	InactiveHosts                    []*hosts.Host
 	KubeClient                       *kubernetes.Clientset
 	KubernetesServiceIP              net.IP
 	Certificates                     map[string]pki.CertificatePKI

cluster/hosts.go

@@ -9,6 +9,7 @@ import (
 	"github.com/rancher/rke/log"
 	"github.com/rancher/rke/pki"
 	"github.com/rancher/rke/services"
+	"github.com/rancher/types/apis/management.cattle.io/v3"
 	"github.com/sirupsen/logrus"
 	"golang.org/x/sync/errgroup"
 )
@@ -26,23 +27,23 @@ func (c *Cluster) TunnelHosts(ctx context.Context, local bool) error {
 		}
 		return nil
 	}
-	for i := range c.EtcdHosts {
-		if err := c.EtcdHosts[i].TunnelUp(ctx, c.DockerDialerFactory); err != nil {
-			return fmt.Errorf("Failed to set up SSH tunneling for Etcd host [%s]: %v", c.EtcdHosts[i].Address, err)
+	c.InactiveHosts = make([]*hosts.Host, 0)
+	uniqueHosts := hosts.GetUniqueHostList(c.EtcdHosts, c.ControlPlaneHosts, c.WorkerHosts)
+	for i := range uniqueHosts {
+		if err := uniqueHosts[i].TunnelUp(ctx, c.DockerDialerFactory); err != nil {
+			log.Warnf(ctx, "Failed to set up SSH tunneling for host [%s]: %v", uniqueHosts[i].Address, err)
+			c.InactiveHosts = append(c.InactiveHosts, uniqueHosts[i])
 		}
 	}
-	for i := range c.ControlPlaneHosts {
-		err := c.ControlPlaneHosts[i].TunnelUp(ctx, c.DockerDialerFactory)
-		if err != nil {
-			return fmt.Errorf("Failed to set up SSH tunneling for Control host [%s]: %v", c.ControlPlaneHosts[i].Address, err)
-		}
+	for _, host := range c.InactiveHosts {
+		log.Warnf(ctx, "Removing host [%s] from node lists", host.Address)
+		c.EtcdHosts = removeFromHosts(host, c.EtcdHosts)
+		c.ControlPlaneHosts = removeFromHosts(host, c.ControlPlaneHosts)
+		c.WorkerHosts = removeFromHosts(host, c.WorkerHosts)
+		c.RancherKubernetesEngineConfig.Nodes = removeFromRKENodes(host.RKEConfigNode, c.RancherKubernetesEngineConfig.Nodes)
 	}
-	for i := range c.WorkerHosts {
-		if err := c.WorkerHosts[i].TunnelUp(ctx, c.DockerDialerFactory); err != nil {
-			return fmt.Errorf("Failed to set up SSH tunneling for Worker host [%s]: %v", c.WorkerHosts[i].Address, err)
-		}
-	}
-	return nil
+	return ValidateHostCount(c)
+
 }
 
 func (c *Cluster) InvertIndexHosts() error {
@@ -127,3 +128,21 @@ func CheckEtcdHostsChanged(kubeCluster, currentCluster *Cluster) error {
 	}
 	return nil
 }
+
+func removeFromHosts(hostToRemove *hosts.Host, hostList []*hosts.Host) []*hosts.Host {
+	for i := range hostList {
+		if hostToRemove.Address == hostList[i].Address {
+			return append(hostList[:i], hostList[i+1:]...)
+		}
+	}
+	return hostList
+}
+
+func removeFromRKENodes(nodeToRemove v3.RKEConfigNode, nodeList []v3.RKEConfigNode) []v3.RKEConfigNode {
+	for i := range nodeList {
+		if nodeToRemove.Address == nodeList[i].Address {
+			return append(nodeList[:i], nodeList[i+1:]...)
+		}
+	}
+	return nodeList
+}

cluster/reconcile.go

@@ -52,7 +52,7 @@ func ReconcileCluster(ctx context.Context, kubeCluster, currentCluster *Cluster)
 func reconcileWorker(ctx context.Context, currentCluster, kubeCluster *Cluster, kubeClient *kubernetes.Clientset) error {
 	// worker deleted first to avoid issues when worker+controller on same host
 	logrus.Debugf("[reconcile] Check worker hosts to be deleted")
-	wpToDelete := hosts.GetToDeleteHosts(currentCluster.WorkerHosts, kubeCluster.WorkerHosts)
+	wpToDelete := hosts.GetToDeleteHosts(currentCluster.WorkerHosts, kubeCluster.WorkerHosts, kubeCluster.InactiveHosts)
 	for _, toDeleteHost := range wpToDelete {
 		toDeleteHost.IsWorker = false
 		if err := hosts.DeleteNode(ctx, toDeleteHost, kubeClient, toDeleteHost.IsControl); err != nil {
@@ -80,7 +80,7 @@ func reconcileControl(ctx context.Context, currentCluster, kubeCluster *Cluster,
 	if err != nil {
 		return err
 	}
-	cpToDelete := hosts.GetToDeleteHosts(currentCluster.ControlPlaneHosts, kubeCluster.ControlPlaneHosts)
+	cpToDelete := hosts.GetToDeleteHosts(currentCluster.ControlPlaneHosts, kubeCluster.ControlPlaneHosts, kubeCluster.InactiveHosts)
 	// move the current host in local kubeconfig to the end of the list
 	for i, toDeleteHost := range cpToDelete {
 		if toDeleteHost.Address == selfDeleteAddress {
@@ -152,7 +152,7 @@ func reconcileEtcd(ctx context.Context, currentCluster, kubeCluster *Cluster, ku
 	clientCert := cert.EncodeCertPEM(currentCluster.Certificates[pki.KubeNodeCertName].Certificate)
 	clientkey := cert.EncodePrivateKeyPEM(currentCluster.Certificates[pki.KubeNodeCertName].Key)
 
-	etcdToDelete := hosts.GetToDeleteHosts(currentCluster.EtcdHosts, kubeCluster.EtcdHosts)
+	etcdToDelete := hosts.GetToDeleteHosts(currentCluster.EtcdHosts, kubeCluster.EtcdHosts, kubeCluster.InactiveHosts)
 	for _, etcdHost := range etcdToDelete {
 		if err := services.RemoveEtcdMember(ctx, etcdHost, kubeCluster.EtcdHosts, currentCluster.LocalConnDialerFactory, clientCert, clientkey); err != nil {
 			log.Warnf(ctx, "[reconcile] %v", err)

cluster/validation.go

@@ -9,12 +9,8 @@ import (
 
 func (c *Cluster) ValidateCluster() error {
 	// make sure cluster has at least one controlplane/etcd host
-
-	if len(c.EtcdHosts) == 0 && len(c.Services.Etcd.ExternalURLs) == 0 {
-		return fmt.Errorf("Cluster must have at least one etcd plane host")
-	}
-	if len(c.EtcdHosts) > 0 && len(c.Services.Etcd.ExternalURLs) > 0 {
-		return fmt.Errorf("Cluster can't have both internal and external etcd")
+	if err := ValidateHostCount(c); err != nil {
+		return err
 	}
 
 	// validate hosts options
@@ -120,3 +116,13 @@ func validateIngressOptions(c *Cluster) error {
 	}
 	return nil
 }
+
+func ValidateHostCount(c *Cluster) error {
+	if len(c.EtcdHosts) == 0 && len(c.Services.Etcd.ExternalURLs) == 0 {
+		return fmt.Errorf("Cluster must have at least one etcd plane host")
+	}
+	if len(c.EtcdHosts) > 0 && len(c.Services.Etcd.ExternalURLs) > 0 {
+		return fmt.Errorf("Cluster can't have both internal and external etcd")
+	}
+	return nil
+}

hosts/hosts.go

@@ -159,7 +159,7 @@ func RemoveTaintFromHost(ctx context.Context, host *Host, taintKey string, kubeC
 	return nil
 }
 
-func GetToDeleteHosts(currentHosts, configHosts []*Host) []*Host {
+func GetToDeleteHosts(currentHosts, configHosts, inactiveHosts []*Host) []*Host {
 	toDeleteHosts := []*Host{}
 	for _, currentHost := range currentHosts {
 		found := false
@@ -169,7 +169,16 @@ func GetToDeleteHosts(currentHosts, configHosts []*Host) []*Host {
 			}
 		}
 		if !found {
-			toDeleteHosts = append(toDeleteHosts, currentHost)
+			inactive := false
+			for _, inactiveHost := range inactiveHosts {
+				if inactiveHost.Address == currentHost.Address {
+					inactive = true
+					break
+				}
+			}
+			if !inactive {
+				toDeleteHosts = append(toDeleteHosts, currentHost)
+			}
 		}
 	}
 	return toDeleteHosts