Add a retain flag to retain the broken nodes for debugging.

Signed-off-by: Jintao Zhang <zhangjintao9020@gmail.com>

cmd/kind/create/cluster/createcluster.go

@@ -30,6 +30,7 @@ type flags struct {
 	Name      string
 	Config    string
 	ImageName string
+	Retain    bool
 }
 
 // NewCommand returns a new cobra.Command for cluster creation
@@ -47,6 +48,7 @@ func NewCommand() *cobra.Command {
 	cmd.Flags().StringVar(&flags.Name, "name", "1", "the cluster context name")
 	cmd.Flags().StringVar(&flags.Config, "config", "", "path to a kind config file")
 	cmd.Flags().StringVar(&flags.ImageName, "image", "", "node docker image to use for booting the cluster")
+	cmd.Flags().BoolVar(&flags.Retain, "retain", false, "whether retain the broken nodes for debugging")
 	return cmd
 }
 
@@ -68,7 +70,7 @@ func run(flags *flags, cmd *cobra.Command, args []string) {
 		log.Fatal("Aborting due to invalid configuration.")
 	}
 	// create a cluster context and create the cluster
-	ctx, err := cluster.NewContext(flags.Name)
+	ctx, err := cluster.NewContext(flags.Name, flags.Retain)
 	if err != nil {
 		log.Fatalf("Failed to create cluster context! %v", err)
 	}

cmd/kind/delete/cluster/deletecluster.go

@@ -25,7 +25,8 @@ import (
 )
 
 type flags struct {
-	Name string
+	Name   string
+	Retain bool
 }
 
 // NewCommand returns a new cobra.Command for cluster creation
@@ -41,12 +42,13 @@ func NewCommand() *cobra.Command {
 		},
 	}
 	cmd.Flags().StringVar(&flags.Name, "name", "1", "the cluster name")
+	cmd.Flags().BoolVar(&flags.Retain, "retain", false, "whether retain the broken nodes for debugging")
 	return cmd
 }
 
 func run(flags *flags, cmd *cobra.Command, args []string) {
 	// TODO(bentheelder): make this more configurable
-	ctx, err := cluster.NewContext(flags.Name)
+	ctx, err := cluster.NewContext(flags.Name, flags.Retain)
 	if err != nil {
 		log.Fatalf("Failed to create cluster context! %v", err)
 	}

cmd/kind/get/kubeconfigpath/kubeconfigpath.go

@@ -51,7 +51,7 @@ func NewCommand() *cobra.Command {
 }
 
 func runE(flags *flags, cmd *cobra.Command, args []string) error {
-	ctx, err := cluster.NewContext(flags.Name)
+	ctx, err := cluster.NewContext(flags.Name, false)
 	if err != nil {
 		return fmt.Errorf("failed to create cluster context! %v", err)
 	}

pkg/cluster/clusters.go

@@ -30,7 +30,7 @@ func List() ([]Context, error) {
 	}
 	clusters := []Context{}
 	for name := range n {
-		clusters = append(clusters, *newContextNoValidation(name))
+		clusters = append(clusters, *newContextNoValidation(name, false))
 	}
 	return clusters, nil
 }

pkg/cluster/context.go

@@ -42,6 +42,7 @@ import (
 type Context struct {
 	name   string
 	status *logutil.Status
+	retain bool
 }
 
 // similar to valid docker container names, but since we will prefix
@@ -52,7 +53,7 @@ var validNameRE = regexp.MustCompile(`^[a-zA-Z0-9_.-]+$`)
 
 // NewContext returns a new cluster management context
 // if name is "" the default ("1") will be used
-func NewContext(name string) (ctx *Context, err error) {
+func NewContext(name string, retain bool) (ctx *Context, err error) {
 	// TODO(bentheelder): move validation out of NewContext and into create type
 	// calls, so that EG delete still works on previously valid, now invalid
 	// names if kind updates
@@ -66,16 +67,17 @@ func NewContext(name string) (ctx *Context, err error) {
 			name, validNameRE.String(),
 		)
 	}
-	return newContextNoValidation(name), nil
+	return newContextNoValidation(name, retain), nil
 }
 
 // internal helper that does the actual allocation consitently, but does not
 // validate the cluster name
 // we need this so that if we tighten the validation, other internal code
 // can still create contexts to existing clusters by name (see List())
-func newContextNoValidation(name string) *Context {
+func newContextNoValidation(name string, retain bool) *Context {
 	return &Context{
-		name: name,
+		name:   name,
+		retain: retain,
 	}
 }
 
@@ -180,8 +182,9 @@ func (c *Context) provisionControlPlane(
 	// that don't seem to be configurable, and we need that flag
 	if err := node.FixMounts(); err != nil {
 		// TODO(bentheelder): logging here
-		// TODO(bentheelder): add a flag to retain the broken nodes for debugging
-		nodes.Delete(node)
+		if !c.retain {
+			nodes.Delete(node)
+		}
 		return "", err
 	}
 
@@ -198,17 +201,19 @@ func (c *Context) provisionControlPlane(
 	// signal the node entrypoint to continue booting into systemd
 	if err := node.SignalStart(); err != nil {
 		// TODO(bentheelder): logging here
-		// TODO(bentheelder): add a flag to retain the broken nodes for debugging
-		nodes.Delete(node)
+		if !c.retain {
+			nodes.Delete(node)
+		}
 		return "", err
 	}
 
 	c.status.Start(fmt.Sprintf("[%s] Waiting for docker to be ready 🐋", nodeName))
 	// wait for docker to be ready
 	if !node.WaitForDocker(time.Now().Add(time.Second * 30)) {
 		// TODO(bentheelder): logging here
-		// TODO(bentheelder): add a flag to retain the broken nodes for debugging
-		nodes.Delete(node)
+		if !c.retain {
+			nodes.Delete(node)
+		}
 		return "", fmt.Errorf("timed out waiting for docker to be ready on node")
 	}
 
@@ -219,8 +224,9 @@ func (c *Context) provisionControlPlane(
 	kubeVersion, err := node.KubeVersion()
 	if err != nil {
 		// TODO(bentheelder): logging here
-		// TODO(bentheelder): add a flag to retain the broken nodes for debugging
-		nodes.Delete(node)
+		if !c.retain {
+			nodes.Delete(node)
+		}
 		return "", fmt.Errorf("failed to get kubernetes version from node: %v", err)
 	}
 
@@ -234,15 +240,18 @@ func (c *Context) provisionControlPlane(
 		},
 	)
 	if err != nil {
-		nodes.Delete(node)
+		if !c.retain {
+			nodes.Delete(node)
+		}
 		return "", fmt.Errorf("failed to create kubeadm config: %v", err)
 	}
 
 	// copy the config to the node
 	if err := node.CopyTo(kubeadmConfig, "/kind/kubeadm.conf"); err != nil {
 		// TODO(bentheelder): logging here
-		// TODO(bentheelder): add a flag to retain the broken nodes for debugging
-		nodes.Delete(node)
+		if !c.retain {
+			nodes.Delete(node)
+		}
 		return kubeadmConfig, errors.Wrap(err, "failed to copy kubeadm config to node")
 	}