[CLOUD-202] Feature/cloud 202 add maxunhealthy config to clusterconfig (

kubernetes-sigs#467) * version v0.18.0-alpha * update docs for v0.17.0 * fix kind version in readme * comments-update-buildcontext * Added maxUnhealthy to clusterconfig * rollback chart and image versions. ok-to-test * Added clusterconfig validations * fix clusterconfig validations * fix clusterconfig validations * added init clusterconfig by default * added clusterconfig init by default --------- Co-authored-by: Benjamin Elder <bentheelder@google.com> Co-authored-by: Daman <aroradaman@gmail.com> Co-authored-by: Kubernetes Prow Robot <k8s-ci-robot@users.noreply.github.com> Co-authored-by: dandres-stratio <101642757+dandres-stratio@users.noreply.github.com>
stg-0 · Feb 22, 2024 · 634ba72 · 634ba72
1 parent da47c03
commit 634ba72
Show file tree

Hide file tree

Showing 7 changed files with 74 additions and 28 deletions.
diff --git a/pkg/cluster/internal/create/actions/createworker/createworker.go b/pkg/cluster/internal/create/actions/createworker/createworker.go
@@ -544,7 +544,7 @@ func (a *action) Execute(ctx *actions.ActionContext) error {
 		ctx.Status.Start("Enabling workload cluster's self-healing 🏥")
 		defer ctx.Status.End(false)
 
-		err = enableSelfHealing(n, a.keosCluster, capiClustersNamespace)
+		err = enableSelfHealing(n, a.keosCluster, capiClustersNamespace, a.clusterConfig)
 		if err != nil {
 			return errors.Wrap(err, "failed to enable workload cluster's self-healing")
 		}

diff --git a/pkg/cluster/internal/create/actions/createworker/provider.go b/pkg/cluster/internal/create/actions/createworker/provider.go
@@ -369,19 +369,18 @@ func (p *Provider) deployClusterOperator(n nodes.Node, privateParams PrivatePara
 		}
 		keosCluster.Spec.Keos = commons.Keos{}
 
-		if clusterConfig != nil {
-			clusterConfigYAML, err := yaml.Marshal(clusterConfig)
-			if err != nil {
-				return err
-			}
-			// Write keoscluster file
-			c = "echo '" + string(clusterConfigYAML) + "' > " + manifestsPath + "/clusterconfig.yaml"
-			_, err = commons.ExecuteCommand(n, c, 5)
-			if err != nil {
-				return errors.Wrap(err, "failed to write the keoscluster file")
-			}
-			keosCluster.Spec.ClusterConfigRef.Name = clusterConfig.Metadata.Name
+		clusterConfigYAML, err := yaml.Marshal(clusterConfig)
+		if err != nil {
+			return err
+		}
+		// Write keoscluster file
+		c = "echo '" + string(clusterConfigYAML) + "' > " + manifestsPath + "/clusterconfig.yaml"
+		_, err = commons.ExecuteCommand(n, c, 5)
+		if err != nil {
+			return errors.Wrap(err, "failed to write the keoscluster file")
 		}
+		keosCluster.Spec.ClusterConfigRef.Name = clusterConfig.Metadata.Name
+
 		keosClusterYAML, err := yaml.Marshal(keosCluster)
 		if err != nil {
 			return err
@@ -860,13 +859,20 @@ func (p *Provider) installCAPXLocal(n nodes.Node) error {
 	return nil
 }
 
-func enableSelfHealing(n nodes.Node, keosCluster commons.KeosCluster, namespace string) error {
+func enableSelfHealing(n nodes.Node, keosCluster commons.KeosCluster, namespace string, clusterConfig *commons.ClusterConfig) error {
 	var c string
 	var err error
 
 	if !keosCluster.Spec.ControlPlane.Managed {
 		machineRole := "-control-plane-node"
-		generateMHCManifest(n, keosCluster.Metadata.Name, namespace, machineHealthCheckControlPlaneNodePath, machineRole)
+		controlplane_maxunhealty := 34
+		if clusterConfig != nil {
+			if clusterConfig.Spec.ControlplaneConfig.MaxUnhealthy != nil {
+				controlplane_maxunhealty = *clusterConfig.Spec.ControlplaneConfig.MaxUnhealthy
+			}
+		}
+
+		generateMHCManifest(n, keosCluster.Metadata.Name, namespace, machineHealthCheckControlPlaneNodePath, machineRole, controlplane_maxunhealty)
 
 		c = "kubectl -n " + namespace + " apply -f " + machineHealthCheckControlPlaneNodePath
 		_, err = commons.ExecuteCommand(n, c, 5)
@@ -876,7 +882,13 @@ func enableSelfHealing(n nodes.Node, keosCluster commons.KeosCluster, namespace
 	}
 
 	machineRole := "-worker-node"
-	generateMHCManifest(n, keosCluster.Metadata.Name, namespace, machineHealthCheckWorkerNodePath, machineRole)
+	workernode_maxunhealty := 34
+	if clusterConfig != nil {
+		if clusterConfig.Spec.WorkersConfig.MaxUnhealthy != nil {
+			workernode_maxunhealty = *clusterConfig.Spec.WorkersConfig.MaxUnhealthy
+		}
+	}
+	generateMHCManifest(n, keosCluster.Metadata.Name, namespace, machineHealthCheckWorkerNodePath, machineRole, workernode_maxunhealty)
 
 	c = "kubectl -n " + namespace + " apply -f " + machineHealthCheckWorkerNodePath
 	_, err = commons.ExecuteCommand(n, c, 5)
@@ -887,14 +899,11 @@ func enableSelfHealing(n nodes.Node, keosCluster commons.KeosCluster, namespace
 	return nil
 }
 
-func generateMHCManifest(n nodes.Node, clusterID string, namespace string, manifestPath string, machineRole string) error {
+func generateMHCManifest(n nodes.Node, clusterID string, namespace string, manifestPath string, machineRole string, maxunhealthy int) error {
 	var c string
 	var err error
-	var maxUnhealthy = "100%"
+	var maxUnhealthy = strconv.Itoa(maxunhealthy) + "%"
 
-	if strings.Contains(machineRole, "control-plane-node") {
-		maxUnhealthy = "34%"
-	}
 	var machineHealthCheck = `
 apiVersion: cluster.x-k8s.io/v1beta1
 kind: MachineHealthCheck

diff --git a/pkg/cluster/internal/validate/common.go b/pkg/cluster/internal/validate/common.go
@@ -34,7 +34,7 @@ const (
 
 var k8sVersionSupported = []string{"1.24", "1.25", "1.26", "1.27", "1.28"}
 
-func validateCommon(spec commons.KeosSpec) error {
+func validateCommon(spec commons.KeosSpec, clusterConfigSpec commons.ClusterConfigSpec) error {
 	var err error
 	if err = validateK8SVersion(spec.K8SVersion); err != nil {
 		return err
@@ -45,6 +45,18 @@ func validateCommon(spec commons.KeosSpec) error {
 	if err = validateVolumes(spec); err != nil {
 		return err
 	}
+	if err = validateClusterConfig(spec, clusterConfigSpec); err != nil {
+		return err
+	}
+	return nil
+}
+
+func validateClusterConfig(spec commons.KeosSpec, clusterConfigSpec commons.ClusterConfigSpec) error {
+	if spec.ControlPlane.Managed {
+		if clusterConfigSpec.ControlplaneConfig.MaxUnhealthy != nil {
+			return errors.New("spec: Invalid value: \"controlplane_config.max_unhealthy\" in clusterConfig: This field cannot be set with managed cluster")
+		}
+	}
 	return nil
 }
 

diff --git a/pkg/cluster/internal/validate/validate.go b/pkg/cluster/internal/validate/validate.go
@@ -22,6 +22,7 @@ import (
 
 type ValidateParams struct {
 	KeosCluster   commons.KeosCluster
+	ClusterConfig *commons.ClusterConfig
 	SecretsPath   string
 	VaultPassword string
 }
@@ -34,8 +35,11 @@ func Cluster(params *ValidateParams) (commons.ClusterCredentials, error) {
 	if err != nil {
 		return commons.ClusterCredentials{}, err
 	}
-
-	if err := validateCommon(params.KeosCluster.Spec); err != nil {
+	clusterConfigSpec := commons.ClusterConfigSpec{}
+	if params.ClusterConfig != nil {
+		clusterConfigSpec = params.ClusterConfig.Spec
+	}
+	if err := validateCommon(params.KeosCluster.Spec, clusterConfigSpec); err != nil {
 		return commons.ClusterCredentials{}, err
 	}
 

diff --git a/pkg/cluster/provider.go b/pkg/cluster/provider.go
@@ -255,9 +255,10 @@ func (p *Provider) CollectLogs(name, dir string) error {
 	return p.provider.CollectLogs(dir, n)
 }
 
-func (p *Provider) Validate(keosCluster commons.KeosCluster, secretsPath string, vaultPassword string) (commons.ClusterCredentials, error) {
+func (p *Provider) Validate(keosCluster commons.KeosCluster, clusterConfig *commons.ClusterConfig, secretsPath string, vaultPassword string) (commons.ClusterCredentials, error) {
 	params := &internalvalidate.ValidateParams{
 		KeosCluster:   keosCluster,
+		ClusterConfig: clusterConfig,
 		SecretsPath:   secretsPath,
 		VaultPassword: vaultPassword,
 	}

diff --git a/pkg/cmd/kind/create/cluster/createcluster.go b/pkg/cmd/kind/create/cluster/createcluster.go
@@ -180,6 +180,7 @@ func runE(logger log.Logger, streams cmd.IOStreams, flags *flagpole) error {
 
 	clusterCredentials, err := provider.Validate(
 		*keosCluster,
+		clusterConfig,
 		secretsDefaultPath,
 		flags.VaultPassword,
 	)

diff --git a/pkg/commons/cluster.go b/pkg/commons/cluster.go
@@ -58,9 +58,19 @@ type Metadata struct {
 }
 
 type ClusterConfigSpec struct {
-	Private                     bool   `yaml:"private_registry"`
-	ClusterOperatorVersion      string `yaml:"cluster_operator_version,omitempty"`
-	ClusterOperatorImageVersion string `yaml:"cluster_operator_image_version,omitempty"`
+	Private                     bool               `yaml:"private_registry"`
+	ControlplaneConfig          ControlplaneConfig `yaml:"controlplane_config"`
+	WorkersConfig               WorkersConfig      `yaml:"workers_config"`
+	ClusterOperatorVersion      string             `yaml:"cluster_operator_version,omitempty"`
+	ClusterOperatorImageVersion string             `yaml:"cluster_operator_image_version,omitempty"`
+}
+
+type ControlplaneConfig struct {
+	MaxUnhealthy *int `yaml:"max_unhealthy,omitempty" validate:"omitempty,numeric,gte=0,lte=100"`
+}
+
+type WorkersConfig struct {
+	MaxUnhealthy *int `yaml:"max_unhealthy,omitempty" validate:"omitempty,numeric,gte=0,lte=100"`
 }
 
 type ClusterConfigRef struct {
@@ -347,6 +357,7 @@ type SCParameters struct {
 
 func (s ClusterConfigSpec) Init() ClusterConfigSpec {
 	s.Private = false
+	s.WorkersConfig.MaxUnhealthy = ToPtr[int](100)
 	return s
 }
 
@@ -458,6 +469,9 @@ func GetClusterDescriptor(descriptorPath string) (*KeosCluster, *ClusterConfig,
 		clusterConfig.Metadata.Name = keosCluster.Spec.InfraProvider + "-config"
 		clusterConfig.Metadata.Namespace = "cluster-" + keosCluster.Metadata.Name
 		clusterConfig.Spec = new(ClusterConfigSpec).Init()
+		if !keosCluster.Spec.ControlPlane.Managed {
+			clusterConfig.Spec.ControlplaneConfig.MaxUnhealthy = ToPtr[int](34)
+		}
 	}
 
 	return &keosCluster, &clusterConfig, nil
@@ -592,3 +606,8 @@ func requireCheckFieldValue(fl validator.FieldLevel, param string, value string,
 	return false
 
 }
+
+// Ptr returns a pointer to the provided value.
+func ToPtr[T any](v T) *T {
+	return &v
+}