feat: add CF stackNameOverrides in cluster.yaml (kubernetes-retired#1484

)

* feat: add CF stackNameOverride in cluster.yaml

Add option in cluster.yaml to override the name of CF controlPlane,
network and etcd stacks.

This is useful for migration of CF stacks from kube-aws <0.11.x without
creating a new VPC.

* wrap stackNameOverrides into cloudformation config

Because it is part of the cloudformation config in cluster.yaml.

core/controlplane/cluster/cluster.go

@@ -137,7 +137,7 @@ func NewCluster(cfgRef *config.Cluster, opts config.StackTemplateOptions, plugin
 	clusterRef := newClusterRef(cfg, session)
 	// TODO Do this in a cleaner way e.g. in config.go
 	clusterRef.KubeResourcesAutosave.S3Path = model.NewS3Folders(cfg.DeploymentSettings.S3URI, clusterRef.ClusterName).ClusterBackups().Path()
-	stackConfig, err := clusterRef.StackConfig(config.ControlPlaneStackName, opts, session, plugins)
+	stackConfig, err := clusterRef.StackConfig(clusterRef.ControlPlaneStackName(), opts, session, plugins)
 	if err != nil {
 		return nil, err
 	}
@@ -279,15 +279,15 @@ func (c *Cluster) Validate() error {
 
 // NestedStackName returns a sanitized name of this control-plane which is usable as a valid cloudformation nested stack name
 func (c Cluster) NestedStackName() string {
-	return naming.FromStackToCfnResource(config.ControlPlaneStackName)
+	return naming.FromStackToCfnResource(c.ControlPlaneStackName())
 }
 
 func (c *Cluster) String() string {
 	return fmt.Sprintf("{Config:%+v}", *c.StackConfig.Config)
 }
 
 func (c *ClusterRef) Destroy() error {
-	return cfnstack.NewDestroyer(config.ControlPlaneStackName, c.session, c.CloudFormation.RoleARN).Destroy()
+	return cfnstack.NewDestroyer(c.Cluster.ControlPlaneStackName(), c.session, c.CloudFormation.RoleARN).Destroy()
 }
 
 func (c *ClusterRef) validateKeyPair(ec2Svc ec2Service) error {

core/controlplane/cluster/cluster_test.go

@@ -169,6 +169,39 @@ func (svc dummyEC2Service) DescribeKeyPairs(input *ec2.DescribeKeyPairsInput) (*
 	return output, nil
 }
 
+func TestStackNameDefaults(t *testing.T) {
+	configBody := defaultConfigValues(t, "")
+	clusterConfig, err := config.ClusterFromBytes([]byte(configBody))
+	if err != nil {
+		t.Errorf("could not get valid cluster config in `%s`: %v", configBody, err)
+	}
+	c := &ClusterRef{Cluster: clusterConfig}
+
+	assert.Equal(t, "control-plane", c.ControlPlaneStackName(), "Invalid ControlPlane Stackname, should be set to 'control-plane' if no override is provided.")
+	assert.Equal(t, "network", c.NetworkStackName(), "Invalid Network Stackname, should be set to 'network' if no override is provided.")
+	assert.Equal(t, "etcd", c.EtcdStackName(), "Invalid Etcd Stackname, should be set to 'etcd' if no override is provided.")
+}
+
+func TestStackNameOverrides(t *testing.T) {
+	stackNameOverrideConfig := `
+cloudformation:
+  stackNameOverrides:
+    controlPlane: "control-plane-override"
+    network: "network-override"
+    etcd: "etcd-override"
+`
+	configBody := defaultConfigValues(t, stackNameOverrideConfig)
+	clusterConfig, err := config.ClusterFromBytes([]byte(configBody))
+	if err != nil {
+		t.Errorf("could not get valid cluster config in `%s`: %v", configBody, err)
+	}
+	c := &ClusterRef{Cluster: clusterConfig}
+
+	assert.Equal(t, "control-plane-override", c.ControlPlaneStackName(), "Invalid ControlPlane Stackname, should be overridden with 'control-plane-override'.")
+	assert.Equal(t, "network-override", c.NetworkStackName(), "Invalid Network Stackname, should overridden to 'network-override'.")
+	assert.Equal(t, "etcd-override", c.EtcdStackName(), "Invalid Etcd Stackname, should be overridden with 'etcd-override'.")
+}
+
 func TestExistingVPCValidation(t *testing.T) {
 
 	goodExistingVPCConfigs := []string{

core/controlplane/config/config.go

@@ -45,7 +45,10 @@ const (
 	// This is not needed to be unique in an AWS account because the actual name of a nested stack is generated randomly
 	// by CloudFormation by including the logical name.
 	// This is NOT intended to be used to reference stack name from cloud-config as the target of awscli or cfn-bootstrap-tools commands e.g. `cfn-init` and `cfn-signal`
-	ControlPlaneStackName = "control-plane"
+	// These names can be overridden in cluster.yaml
+	controlPlaneStackName = "control-plane"
+	networkStackName      = "network"
+	etcdStackName         = "etcd"
 )
 
 func NewDefaultCluster() *Cluster {
@@ -1097,6 +1100,27 @@ type Config struct {
 	ControllerFlags  pluginmodel.ControllerFlags
 }
 
+func (c Cluster) ControlPlaneStackName() string {
+	if c.CloudFormation.StackNameOverrides.ControlPlane != "" {
+		return c.CloudFormation.StackNameOverrides.ControlPlane
+	}
+	return controlPlaneStackName
+}
+
+func (c Cluster) NetworkStackName() string {
+	if c.CloudFormation.StackNameOverrides.Network != "" {
+		return c.CloudFormation.StackNameOverrides.Network
+	}
+	return networkStackName
+}
+
+func (c Cluster) EtcdStackName() string {
+	if c.CloudFormation.StackNameOverrides.Etcd != "" {
+		return c.CloudFormation.StackNameOverrides.Etcd
+	}
+	return etcdStackName
+}
+
 func (c Cluster) StackNameEnvFileName() string {
 	return "/etc/environment"
 }
@@ -1277,7 +1301,7 @@ func (c Cluster) validate() error {
 
 	clusterNamePlaceholder := "<my-cluster-name>"
 	nestedStackNamePlaceHolder := "<my-nested-stack-name>"
-	replacer := strings.NewReplacer(clusterNamePlaceholder, "", nestedStackNamePlaceHolder, ControlPlaneStackName)
+	replacer := strings.NewReplacer(clusterNamePlaceholder, "", nestedStackNamePlaceHolder, c.ControlPlaneStackName())
 	simulatedLcName := fmt.Sprintf("%s-%s-1N2C4K3LLBEDZ-%sLC-BC2S9P3JG2QD", clusterNamePlaceholder, nestedStackNamePlaceHolder, c.Controller.LogicalName())
 	limit := 63 - len(replacer.Replace(simulatedLcName))
 	if c.Experimental.AwsNodeLabels.Enabled && len(c.ClusterName) > limit {
@@ -1293,7 +1317,7 @@ func (c Cluster) validate() error {
 			return e
 		}
 	} else {
-		if e := cfnresource.ValidateUnstableRoleNameLength(c.ClusterName, naming.FromStackToCfnResource(ControlPlaneStackName), c.Controller.IAMConfig.Role.Name, c.Region.String()); e != nil {
+		if e := cfnresource.ValidateUnstableRoleNameLength(c.ClusterName, naming.FromStackToCfnResource(c.ControlPlaneStackName()), c.Controller.IAMConfig.Role.Name, c.Region.String()); e != nil {
 			return e
 		}
 	}

core/etcd/cluster/cluster.go

@@ -143,7 +143,7 @@ func NewCluster(cfgRef *controlplaneconfig.Cluster, opts controlplaneconfig.Stac
 	clusterRef.KubeAWSVersion = controlplanecluster.VERSION
 	clusterRef.HostOS = cfgRef.HostOS
 
-	cpStackConfig, err := clusterRef.StackConfig("etcd", opts, session, plugins)
+	cpStackConfig, err := clusterRef.StackConfig(cfgRef.EtcdStackName(), opts, session, plugins)
 	if err != nil {
 		return nil, err
 	}
@@ -309,7 +309,7 @@ func (c *Cluster) String() string {
 }
 
 func (c *ClusterRef) Destroy() error {
-	return cfnstack.NewDestroyer("etcd", c.session, c.CloudFormation.RoleARN).Destroy()
+	return cfnstack.NewDestroyer(c.EtcdStackName(), c.session, c.CloudFormation.RoleARN).Destroy()
 }
 
 // lookupExistingEtcdEndpoints supports the migration from embedded etcd servers to their own stack

core/network/cluster/cluster.go

@@ -119,7 +119,7 @@ func NewCluster(cfgRef *config.Cluster, opts config.StackTemplateOptions, plugin
 	// TODO Do this in a cleaner way e.g. in config.go
 	clusterRef.KubeResourcesAutosave.S3Path = model.NewS3Folders(cfg.DeploymentSettings.S3URI, clusterRef.ClusterName).ClusterBackups().Path()
 
-	stackConfig, err := clusterRef.StackConfig("network", opts, session, plugins)
+	stackConfig, err := clusterRef.StackConfig(cfgRef.NetworkStackName(), opts, session, plugins)
 	if err != nil {
 		return nil, err
 	}
@@ -215,5 +215,5 @@ func (c *Cluster) String() string {
 }
 
 func (c *ClusterRef) Destroy() error {
-	return cfnstack.NewDestroyer("network", c.session, c.CloudFormation.RoleARN).Destroy()
+	return cfnstack.NewDestroyer(c.NetworkStackName(), c.session, c.CloudFormation.RoleARN).Destroy()
 }

core/root/config/templates/cluster.yaml

@@ -19,6 +19,22 @@ amiId: "{{.AmiId}}"
 # Container Linux has automatic updates https://coreos.com/os/docs/latest/update-strategies.html. This can be a risk in certain situations and this is why is disabled by default and you can enable it by setting this param to false.
 disableContainerLinuxAutomaticUpdates: true
 
+# Override the CloudFormation logical sub-stack names of control plane, etcd and/or network.
+# Changing these names causes CF to delete the old and create a new stack and thus deleting and recreating all components in the stack.
+#
+# From kube-aws 0.11.x and onwards, separate networking and etcd CF stacks have been introduced.
+# Because CF does not support migration of resources between stacks (or renaming stacks), a new VPC will be created if you upgrade an existing 0.10.x cluster.
+# Creating a new VPC and deleting the old one can cause a lot of issues and might not be desired.
+# To make this migration more feasible it is posible to change the name of the network stack to the old controleplane stack to keep the vpc components in the same stack. This way CF will not create a new vpc.
+# In this scenario the name of the networkstack becomes "ControlPlane". (which is the name of the stack which holds the vpc components in a kube-aws pre 0.11 cluster)
+# For clearity it is advised also rename the stack of the controlPlane components to, for example, "masters"
+# Note that this will keep some legacy naming around in your configs and CF stacks...
+#cloudformation:
+#  stackNameOverrides:
+#    controlPlane: "control-plane"
+#    network: "network"
+#    etcd: "etcd"
+
 # The ID of hosted zone to add the externalDNSName to.
 # Either specify hostedZoneId or hostedZone, but not both
 #hostedZoneId: ""

model/cloudformation.go

@@ -1,5 +1,6 @@
 package model
 
 type CloudFormation struct {
-	RoleARN string `yaml:"roleARN,omitempty"`
+	RoleARN            string             `yaml:"roleARN,omitempty"`
+	StackNameOverrides StackNameOverrides `yaml:"stackNameOverrides,omitempty"`
 }

model/stack_name_overrides.go

@@ -0,0 +1,7 @@
+package model
+
+type StackNameOverrides struct {
+	ControlPlane string `yaml:"controlPlane,omitempty"`
+	Network      string `yaml:"network,omitempty"`
+	Etcd         string `yaml:"etcd,omitempty"`
+}