Merge pull request #5373 from ykakarap/clusterctl-better-errors

✨clusterctl: generate returns better error message in case there is no management cluster available

Author: k8s-ci-robot

cmd/clusterctl/client/cluster/client.go

@@ -21,14 +21,11 @@ import (
 	"time"
 
 	"github.com/pkg/errors"
-	"k8s.io/apimachinery/pkg/apis/meta/v1/unstructured"
 	"k8s.io/apimachinery/pkg/util/wait"
-	"k8s.io/client-go/rest"
 	"sigs.k8s.io/cluster-api/cmd/clusterctl/client/config"
 	"sigs.k8s.io/cluster-api/cmd/clusterctl/client/repository"
 	yaml "sigs.k8s.io/cluster-api/cmd/clusterctl/client/yamlprocessor"
 	logf "sigs.k8s.io/cluster-api/cmd/clusterctl/log"
-	"sigs.k8s.io/controller-runtime/pkg/client"
 )
 
 const (
@@ -61,15 +58,15 @@ type Client interface {
 	// Proxy return the Proxy used for operating objects in the management cluster.
 	Proxy() Proxy
 
-	// CertManager returns a CertManagerClient that can be user for
+	// CertManager returns a CertManagerClient that can be used for
 	// operating the cert-manager components in the cluster.
 	CertManager() CertManagerClient
 
-	// ProviderComponents returns a ComponentsClient object that can be user for
+	// ProviderComponents returns a ComponentsClient object that can be used for
 	// operating provider components objects in the management cluster (e.g. the CRDs, controllers, RBAC).
 	ProviderComponents() ComponentsClient
 
-	// ProviderInventory returns a InventoryClient object that can be user for
+	// ProviderInventory returns a InventoryClient object that can be used for
 	// operating provider inventory stored in the management cluster (e.g. the list of installed providers/versions).
 	ProviderInventory() InventoryClient
 
@@ -219,30 +216,6 @@ func newClusterClient(kubeconfig Kubeconfig, configClient config.Client, options
 	return client
 }
 
-// Proxy defines a client proxy interface.
-type Proxy interface {
-	// GetConfig returns the rest.Config
-	GetConfig() (*rest.Config, error)
-
-	// CurrentNamespace returns the namespace from the current context in the kubeconfig file
-	CurrentNamespace() (string, error)
-
-	// ValidateKubernetesVersion returns an error if management cluster version less than minimumKubernetesVersion
-	ValidateKubernetesVersion() error
-
-	// NewClient returns a new controller runtime Client object for working on the management cluster
-	NewClient() (client.Client, error)
-
-	// ListResources returns all the Kubernetes objects with the given labels existing the listed namespaces.
-	ListResources(labels map[string]string, namespaces ...string) ([]unstructured.Unstructured, error)
-
-	// GetContexts returns the list of contexts in kubeconfig which begin with prefix.
-	GetContexts(prefix string) ([]string, error)
-
-	// GetResourceNames returns the list of resource names which begin with prefix.
-	GetResourceNames(groupVersion, kind string, options []client.ListOption, prefix string) ([]string, error)
-}
-
 // retryWithExponentialBackoff repeats an operation until it passes or the exponential backoff times out.
 func retryWithExponentialBackoff(opts wait.Backoff, operation func() error) error {
 	log := logf.Log
@@ -291,6 +264,20 @@ func newConnectBackoff() wait.Backoff {
 	}
 }
 
+// newShortConnectBackoff creates a new API Machinery backoff parameter set suitable for use when clusterctl connect to a cluster.
+// Preferred over newConnectBackoff() only when used to perform quick checks to check if a cluster is reachable.
+func newShortConnectBackoff() wait.Backoff {
+	// Return a exponential backoff configuration which returns durations for a total time of ~5s.
+	// Example: 0, .25s, .6s, 1.2, 2.1s, 3.4s, 5.5s.
+	// Jitter is added as a random fraction of the duration multiplied by the jitter factor.
+	return wait.Backoff{
+		Duration: 250 * time.Millisecond,
+		Factor:   1.5,
+		Steps:    7,
+		Jitter:   0.1,
+	}
+}
+
 // newReadBackoff creates a new API Machinery backoff parameter set suitable for use with clusterctl read operations.
 func newReadBackoff() wait.Backoff {
 	// Return a exponential backoff configuration which returns durations for a total time of ~15s.

cmd/clusterctl/client/cluster/proxy.go

@@ -39,6 +39,33 @@ var (
 	localScheme = scheme.Scheme
 )
 
+// Proxy defines a client proxy interface.
+type Proxy interface {
+	// GetConfig returns the rest.Config
+	GetConfig() (*rest.Config, error)
+
+	// CurrentNamespace returns the namespace from the current context in the kubeconfig file.
+	CurrentNamespace() (string, error)
+
+	// ValidateKubernetesVersion returns an error if management cluster version less than minimumKubernetesVersion.
+	ValidateKubernetesVersion() error
+
+	// NewClient returns a new controller runtime Client object for working on the management cluster.
+	NewClient() (client.Client, error)
+
+	// CheckClusterAvailable checks if a a cluster is available and reachable.
+	CheckClusterAvailable() error
+
+	// ListResources returns all the Kubernetes objects with the given labels existing the listed namespaces.
+	ListResources(labels map[string]string, namespaces ...string) ([]unstructured.Unstructured, error)
+
+	// GetContexts returns the list of contexts in kubeconfig which begin with prefix.
+	GetContexts(prefix string) ([]string, error)
+
+	// GetResourceNames returns the list of resource names which begin with prefix.
+	GetResourceNames(groupVersion, kind string, options []client.ListOption, prefix string) ([]string, error)
+}
+
 type proxy struct {
 	kubeconfig         Kubeconfig
 	timeout            time.Duration
@@ -150,6 +177,27 @@ func (k *proxy) NewClient() (client.Client, error) {
 	return c, nil
 }
 
+func (k *proxy) CheckClusterAvailable() error {
+	// Check if the cluster is available by creating a client to the cluster.
+	// If creating the client times out and never established we assume that
+	// the cluster does not exist or is not reachable.
+	// For the purposes of clusterctl operations non-existent clusters and
+	// non-reachable clusters can be treated as the same.
+	config, err := k.GetConfig()
+	if err != nil {
+		return err
+	}
+
+	connectBackoff := newShortConnectBackoff()
+	if err := retryWithExponentialBackoff(connectBackoff, func() error {
+		_, err := client.New(config, client.Options{Scheme: localScheme})
+		return err
+	}); err != nil {
+		return err
+	}
+	return nil
+}
+
 func (k *proxy) ListResources(labels map[string]string, namespaces ...string) ([]unstructured.Unstructured, error) {
 	cs, err := k.newClientSet()
 	if err != nil {

cmd/clusterctl/client/config.go

@@ -224,6 +224,9 @@ func (c *clusterctlClient) GetClusterTemplate(options GetClusterTemplateOptions)
 
 	// If the option specifying the targetNamespace is empty, try to detect it.
 	if options.TargetNamespace == "" {
+		if err := clusterClient.Proxy().CheckClusterAvailable(); err != nil {
+			return nil, errors.Wrap(err, "management cluster not available. Cannot auto-discover target namespace. Please specify a target namespace")
+		}
 		currentNamespace, err := clusterClient.Proxy().CurrentNamespace()
 		if err != nil {
 			return nil, err
@@ -273,6 +276,9 @@ func (c *clusterctlClient) getTemplateFromRepository(cluster cluster.Client, opt
 	provider := source.InfrastructureProvider
 	ensureCustomResourceDefinitions := false
 	if provider == "" {
+		if err := cluster.Proxy().CheckClusterAvailable(); err != nil {
+			return nil, errors.Wrap(err, "management cluster not available. Cannot auto-discover default infrastructure provider. Please specify an infrastructure provider")
+		}
 		// ensure the custom resource definitions required by clusterctl are in place
 		if err := cluster.ProviderInventory().EnsureCustomResourceDefinitions(); err != nil {
 			return nil, errors.Wrapf(err, "provider custom resource definitions (CRDs) are not installed")
@@ -298,6 +304,9 @@ func (c *clusterctlClient) getTemplateFromRepository(cluster cluster.Client, opt
 
 	// If the version of the infrastructure provider to get templates from is empty, try to detect it.
 	if version == "" {
+		if err := cluster.Proxy().CheckClusterAvailable(); err != nil {
+			return nil, errors.Wrapf(err, "management cluster not available. Cannot auto-discover version for the provider %q automatically. Please specify a version", name)
+		}
 		// ensure the custom resource definitions required by clusterctl are in place (if not already done)
 		if !ensureCustomResourceDefinitions {
 			if err := cluster.ProviderInventory().EnsureCustomResourceDefinitions(); err != nil {

cmd/clusterctl/internal/test/fake_proxy.go

@@ -17,12 +17,15 @@ limitations under the License.
 package test
 
 import (
+	"errors"
+
 	apiextensionsv1 "k8s.io/apiextensions-apiserver/pkg/apis/apiextensions/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/apis/meta/v1/unstructured"
 	"k8s.io/apimachinery/pkg/runtime"
 	clientgoscheme "k8s.io/client-go/kubernetes/scheme"
 	"k8s.io/client-go/rest"
+	"k8s.io/utils/pointer"
 	clusterv1 "sigs.k8s.io/cluster-api/api/v1beta1"
 	clusterctlv1 "sigs.k8s.io/cluster-api/cmd/clusterctl/api/v1alpha3"
 	fakebootstrap "sigs.k8s.io/cluster-api/cmd/clusterctl/internal/test/providers/bootstrap"
@@ -39,6 +42,7 @@ type FakeProxy struct {
 	cs        client.Client
 	namespace string
 	objs      []client.Object
+	available *bool
 }
 
 var (
@@ -79,6 +83,15 @@ func (f *FakeProxy) NewClient() (client.Client, error) {
 	return f.cs, nil
 }
 
+func (f *FakeProxy) CheckClusterAvailable() error {
+	// default to considering the cluster as available unless explicitly set to be
+	// unavailable.
+	if f.available == nil || *f.available {
+		return nil
+	}
+	return errors.New("cluster is not available")
+}
+
 // ListResources returns all the resources known by the FakeProxy.
 func (f *FakeProxy) ListResources(labels map[string]string, namespaces ...string) ([]unstructured.Unstructured, error) {
 	var ret []unstructured.Unstructured //nolint
@@ -185,6 +198,11 @@ func (f *FakeProxy) WithFakeCAPISetup() *FakeProxy {
 	return f
 }
 
+func (f *FakeProxy) WithClusterAvailable(available bool) *FakeProxy {
+	f.available = pointer.Bool(available)
+	return f
+}
+
 // FakeCAPISetupObjects return required objects in order to make kubeadm pass checks
 // ensuring that management cluster has a proper release of Cluster API installed.
 func FakeCAPISetupObjects() []client.Object {