Bug 1832923: Forget bootstrap etcd member IP after bootstrap

go.mod

@@ -24,4 +24,5 @@ require (
 	k8s.io/client-go v0.18.3
 	k8s.io/component-base v0.18.3
 	k8s.io/klog v1.0.0
+	k8s.io/utils v0.0.0-20200324210504-a9aa75ae1b89
 )

pkg/etcdcli/etcdcli.go

@@ -74,28 +74,24 @@ func (g *etcdClientGetter) getEtcdClient() (*clientv3.Client, error) {
 
 	network, err := g.networkLister.Get("cluster")
 	if err != nil {
-		return nil, err
+		return nil, fmt.Errorf("failed to list cluster network: %w", err)
 	}
 
 	etcdEndpoints := []string{}
 	nodes, err := g.nodeLister.List(labels.Set{"node-role.kubernetes.io/master": ""}.AsSelector())
 	for _, node := range nodes {
 		internalIP, err := dnshelpers.GetEscapedPreferredInternalIPAddressForNodeName(network, node)
 		if err != nil {
-			return nil, err
+			return nil, fmt.Errorf("failed to get internal IP for node: %w", err)
 		}
 		etcdEndpoints = append(etcdEndpoints, fmt.Sprintf("https://%s:2379", internalIP))
 	}
 
-	hostEtcd, err := g.configmapsLister.ConfigMaps(operatorclient.TargetNamespace).Get("etcd-endpoints")
+	configmap, err := g.configmapsLister.ConfigMaps(operatorclient.TargetNamespace).Get("etcd-endpoints")
 	if err != nil {
-		return nil, err
-	}
-	bootstrapIP, ok := hostEtcd.Annotations[BootstrapIPAnnotationKey]
-	if !ok {
-		klog.V(2).Infof("configmaps/etcd-endpoints is missing annotation %s", BootstrapIPAnnotationKey)
+		return nil, fmt.Errorf("failed to list endpoints: %w", err)
 	}
-	if bootstrapIP != "" {
+	if bootstrapIP, ok := configmap.Annotations[BootstrapIPAnnotationKey]; ok && bootstrapIP != "" {
 		// escape if IPv6
 		if net.ParseIP(bootstrapIP).To4() == nil {
 			bootstrapIP = "[" + bootstrapIP + "]"
@@ -112,11 +108,11 @@ func (g *etcdClientGetter) getEtcdClient() (*clientv3.Client, error) {
 
 	c, err := getEtcdClient(etcdEndpoints)
 	if err != nil {
-		return nil, err
+		return nil, fmt.Errorf("failed to get etcd client: %w", err)
 	}
 	if g.cachedClient != nil {
 		if err := g.cachedClient.Close(); err != nil {
-			utilruntime.HandleError(err)
+			utilruntime.HandleError(fmt.Errorf("failed to close cached client: %w", err))
 		}
 	}
 	g.cachedClient = c
@@ -146,7 +142,7 @@ func getEtcdClient(endpoints []string) (*clientv3.Client, error) {
 
 	cli, err := clientv3.New(*cfg)
 	if err != nil {
-		return nil, err
+		return nil, fmt.Errorf("failed to make etcd client for endpoints %v: %w", endpoints, err)
 	}
 	return cli, err
 }

pkg/operator/etcdendpointscontroller/etcdendpointscontroller.go

@@ -13,50 +13,50 @@ import (
 	"github.com/openshift/library-go/pkg/operator/v1helpers"
 	operatorv1helpers "github.com/openshift/library-go/pkg/operator/v1helpers"
 	corev1 "k8s.io/api/core/v1"
+	"k8s.io/apimachinery/pkg/api/errors"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/labels"
 	"k8s.io/client-go/kubernetes"
 	corev1client "k8s.io/client-go/kubernetes/typed/core/v1"
 	corev1listers "k8s.io/client-go/listers/core/v1"
+	"k8s.io/klog"
 
+	"github.com/openshift/cluster-etcd-operator/pkg/etcdcli"
 	"github.com/openshift/cluster-etcd-operator/pkg/operator/operatorclient"
 )
 
 // EtcdEndpointsController maintains a configmap resource with
 // IP addresses for etcd. It should never depend on DNS directly or transitively.
 type EtcdEndpointsController struct {
-	operatorClient  v1helpers.OperatorClient
+	operatorClient  v1helpers.StaticPodOperatorClient
 	nodeLister      corev1listers.NodeLister
 	configmapLister corev1listers.ConfigMapLister
 	configmapClient corev1client.ConfigMapsGetter
 }
 
 func NewEtcdEndpointsController(
-	operatorClient v1helpers.OperatorClient,
+	operatorClient v1helpers.StaticPodOperatorClient,
 	eventRecorder events.Recorder,
 	kubeClient kubernetes.Interface,
 	kubeInformers operatorv1helpers.KubeInformersForNamespaces,
 ) factory.Controller {
-	kubeInformersForTargetNamespace := kubeInformers.InformersFor(operatorclient.TargetNamespace)
-	configmapsInformer := kubeInformersForTargetNamespace.Core().V1().ConfigMaps()
-	kubeInformersForCluster := kubeInformers.InformersFor("")
-	nodeInformer := kubeInformersForCluster.Core().V1().Nodes()
+	nodeInformer := kubeInformers.InformersFor("").Core().V1().Nodes()
 
 	c := &EtcdEndpointsController{
 		operatorClient:  operatorClient,
 		nodeLister:      nodeInformer.Lister(),
-		configmapLister: configmapsInformer.Lister(),
+		configmapLister: kubeInformers.ConfigMapLister(),
 		configmapClient: kubeClient.CoreV1(),
 	}
 	return factory.New().ResyncEvery(time.Minute).WithInformers(
 		operatorClient.Informer(),
-		configmapsInformer.Informer(),
+		kubeInformers.InformersFor(operatorclient.TargetNamespace).Core().V1().ConfigMaps().Informer(),
 		nodeInformer.Informer(),
 	).WithSync(c.sync).ToController("EtcdEndpointsController", eventRecorder.WithComponentSuffix("etcd-endpoints-controller"))
 }
 
 func (c *EtcdEndpointsController) sync(ctx context.Context, syncCtx factory.SyncContext) error {
-	err := c.syncConfigMap(ctx, syncCtx.Recorder())
+	err := c.syncConfigMap(syncCtx.Recorder())
 
 	if err != nil {
 		_, _, updateErr := v1helpers.UpdateStatus(c.operatorClient, v1helpers.UpdateConditionFn(operatorv1.OperatorCondition{
@@ -83,9 +83,29 @@ func (c *EtcdEndpointsController) sync(ctx context.Context, syncCtx factory.Sync
 	return nil
 }
 
-func (c *EtcdEndpointsController) syncConfigMap(ctx context.Context, recorder events.Recorder) error {
+func (c *EtcdEndpointsController) syncConfigMap(recorder events.Recorder) error {
+	bootstrapComplete, err := isBootstrapComplete(c.configmapLister, c.operatorClient)
+	if err != nil {
+		return fmt.Errorf("couldn't determine bootstrap status: %w", err)
+	}
+
 	required := configMapAsset()
 
+	// If the bootstrap IP is present on the existing configmap, either copy it
+	// forward or remove it if possible so clients can forget about it.
+	if existing, err := c.configmapLister.ConfigMaps(operatorclient.TargetNamespace).Get("etcd-endpoints"); err == nil {
+		if existingIP, hasExistingIP := existing.Annotations[etcdcli.BootstrapIPAnnotationKey]; hasExistingIP {
+			if bootstrapComplete {
+				// remove the annotation
+				required.Annotations[etcdcli.BootstrapIPAnnotationKey+"-"] = existingIP
+			} else {
+				required.Annotations[etcdcli.BootstrapIPAnnotationKey] = existingIP
+			}
+		}
+	} else if !errors.IsNotFound(err) {
+		return fmt.Errorf("couldn't get configmap %s/%s: %w", operatorclient.TargetNamespace, "etcd-endpoints", err)
+	}
+
 	// create endpoint addresses for each node
 	nodes, err := c.nodeLister.List(labels.Set{"node-role.kubernetes.io/master": ""}.AsSelector())
 	if err != nil {
@@ -112,15 +132,60 @@ func (c *EtcdEndpointsController) syncConfigMap(ctx context.Context, recorder ev
 
 	required.Data = endpointAddresses
 
-	_, _, err = resourceapply.ApplyConfigMap(c.configmapClient, recorder, required)
-	return err
+	// Apply endpoint updates
+	if _, _, err := resourceapply.ApplyConfigMap(c.configmapClient, recorder, required); err != nil {
+		return err
+	}
+
+	return nil
 }
 
 func configMapAsset() *corev1.ConfigMap {
 	return &corev1.ConfigMap{
 		ObjectMeta: metav1.ObjectMeta{
-			Name:      "etcd-endpoints",
-			Namespace: operatorclient.TargetNamespace,
+			Name:        "etcd-endpoints",
+			Namespace:   operatorclient.TargetNamespace,
+			Annotations: map[string]string{},
 		},
 	}
 }
+
+// isBootstrapComplete returns true if bootstrap has completed. This is used to
+// indicate whether it's safe for clients to forget about the bootstrap member IP.
+func isBootstrapComplete(configMapClient corev1listers.ConfigMapLister, staticPodClient v1helpers.StaticPodOperatorClient) (bool, error) {
+	// do a cheap check to see if the annotation is already gone.
+	// check to see if bootstrapping is complete
+	bootstrapFinishedConfigMap, err := configMapClient.ConfigMaps("kube-system").Get("bootstrap")
+	if err != nil {
+		if errors.IsNotFound(err) {
+			// If the resource was deleted (e.g. by an admin) after bootstrap is actually complete,
+			// this is a false negative.
+			klog.V(4).Infof("bootstrap considered incomplete because the kube-system/bootstrap configmap wasn't found")
+			return false, nil
+		}
+		// We don't know, give up quickly.
+		return false, fmt.Errorf("failed to get configmap %s/%s: %w", "kube-system", "bootstrap", err)
+	}
+
+	if status, ok := bootstrapFinishedConfigMap.Data["status"]; !ok || status != "complete" {
+		// do nothing, not torn down
+		klog.V(4).Infof("bootstrap considered incomplete because status is %q", status)
+		return false, nil
+	}
+
+	// now run check to stability of revisions
+	_, status, _, err := staticPodClient.GetStaticPodOperatorState()
+	if err != nil {
+		return false, fmt.Errorf("failed to get static pod operator state: %w", err)
+	}
+	if status.LatestAvailableRevision == 0 {
+		return false, nil
+	}
+	for _, curr := range status.NodeStatuses {
+		if curr.CurrentRevision != status.LatestAvailableRevision {
+			klog.V(4).Infof("bootstrap considered incomplete because revision %d is still in progress", status.LatestAvailableRevision)
+			return false, nil
+		}
+	}
+	return true, nil
+}