Add cluster version rollback support with rollbackConfig

go.mod

@@ -16,7 +16,7 @@ require (
 	github.com/aws/aws-sdk-go-v2/service/cloudwatchlogs v1.73.0
 	github.com/aws/aws-sdk-go-v2/service/cognitoidentityprovider v1.60.2
 	github.com/aws/aws-sdk-go-v2/service/ec2 v1.302.0
-	github.com/aws/aws-sdk-go-v2/service/eks v1.87.0
+	github.com/aws/aws-sdk-go-v2/service/eks v1.88.0
 	github.com/aws/aws-sdk-go-v2/service/elasticloadbalancing v1.33.25
 	github.com/aws/aws-sdk-go-v2/service/elasticloadbalancingv2 v1.54.12
 	github.com/aws/aws-sdk-go-v2/service/iam v1.53.10

go.sum

@@ -136,8 +136,8 @@ github.com/aws/aws-sdk-go-v2/service/cognitoidentityprovider v1.60.2 h1:pZxE29WA
 github.com/aws/aws-sdk-go-v2/service/cognitoidentityprovider v1.60.2/go.mod h1:nbe4Nf/HOY+e54Dl+yjv04scYTGTC+4ZthbfOuPTXQs=
 github.com/aws/aws-sdk-go-v2/service/ec2 v1.302.0 h1:7c0jQaj+QKYUo3pgtEm9fQIePJH6QJA3bVKIgCCLdvM=
 github.com/aws/aws-sdk-go-v2/service/ec2 v1.302.0/go.mod h1:Y95W0Hm6FYLPa6o0hbnJ+sWgmdc4ifcLFjGkdobWVhY=
-github.com/aws/aws-sdk-go-v2/service/eks v1.87.0 h1:bftLltXNWmNr9ed3CaQnVlzNPTNTFdHguNhIsZF6DxM=
-github.com/aws/aws-sdk-go-v2/service/eks v1.87.0/go.mod h1:rbIASs+SfCDUXx2EdfMkNpDGptlW8hvMZ9AawRiUBqE=
+github.com/aws/aws-sdk-go-v2/service/eks v1.88.0 h1:dP/bd/5AG73UJGXyZh6gQSafVS9T+mBQgIULtJvzEnE=
+github.com/aws/aws-sdk-go-v2/service/eks v1.88.0/go.mod h1:rbIASs+SfCDUXx2EdfMkNpDGptlW8hvMZ9AawRiUBqE=
 github.com/aws/aws-sdk-go-v2/service/elasticloadbalancing v1.33.25 h1:VzmoYPRbNSUqk3pA04ZyGZUg52yfX259XXRqwr1lns4=
 github.com/aws/aws-sdk-go-v2/service/elasticloadbalancing v1.33.25/go.mod h1:r7chQGimOmFs4oqawhO+i+o3ez2l69rzAco5KTb7bjY=
 github.com/aws/aws-sdk-go-v2/service/elasticloadbalancingv2 v1.54.12 h1:TJXv7kZjdXA2maPDaJFFEQPBrPmvPtMybN3qYDOpJ4Y=

pkg/actions/cluster/upgrade.go

@@ -10,6 +10,7 @@ import (
 	api "github.com/weaveworks/eksctl/pkg/apis/eksctl.io/v1alpha5"
 	"github.com/weaveworks/eksctl/pkg/ctl/cmdutils"
 	"github.com/weaveworks/eksctl/pkg/eks"
+	"github.com/weaveworks/eksctl/pkg/utils"
 )
 
 func upgrade(ctx context.Context, cfg *api.ClusterConfig, ctl *eks.ClusterProvider, dryRun bool) (bool, error) {
@@ -31,14 +32,24 @@ func upgrade(ctx context.Context, cfg *api.ClusterConfig, ctl *eks.ClusterProvid
 	}
 
 	if upgradeVersion != "" {
+		currentVersion := ctl.ControlPlaneVersion()
+		// A resolved target lower than the current version is a rollback (downgrade);
+		// use direction-aware wording so the logs don't say "upgrade" for a downgrade.
+		action, pastAction := "upgrade", "upgraded"
+		if c, err := utils.CompareVersions(upgradeVersion, currentVersion); err != nil {
+			return false, err
+		} else if c < 0 {
+			action, pastAction = "roll back", "rolled back"
+		}
+
 		msgNodeGroupsAndAddons := "you will need to follow the upgrade procedure for all of nodegroups and add-ons"
-		cmdutils.LogIntendedAction(dryRun, "upgrade cluster %q control plane from current version %q to %q", cfg.Metadata.Name, ctl.ControlPlaneVersion(), upgradeVersion)
+		cmdutils.LogIntendedAction(dryRun, "%s cluster %q control plane from current version %q to %q", action, cfg.Metadata.Name, currentVersion, upgradeVersion)
 		if !dryRun {
 			cfg.Metadata.Version = upgradeVersion
 			if err := ctl.UpdateClusterVersionBlocking(ctx, cfg); err != nil {
 				return false, err
 			}
-			logger.Success("cluster %q control plane has been upgraded to version %q", cfg.Metadata.Name, cfg.Metadata.Version)
+			logger.Success("cluster %q control plane has been %s to version %q", cfg.Metadata.Name, pastAction, cfg.Metadata.Version)
 			logger.Info(msgNodeGroupsAndAddons)
 		}
 	} else {

pkg/apis/eksctl.io/v1alpha5/assets/schema.json

@@ -1072,6 +1072,11 @@
           "description": "the AWS region hosting this cluster",
           "x-intellij-html-description": "the AWS region hosting this cluster"
         },
+        "rollbackConfig": {
+          "$ref": "#/definitions/RollbackConfig",
+          "description": "configures the automatic rollback behaviour when updating the cluster version (including downgrades).",
+          "x-intellij-html-description": "configures the automatic rollback behaviour when updating the cluster version (including downgrades)."
+        },
         "tags": {
           "additionalProperties": {
             "type": "string"
@@ -1092,6 +1097,7 @@
         "region",
         "version",
         "forceUpdateVersion",
+        "rollbackConfig",
         "tags",
         "annotations"
       ],
@@ -3053,6 +3059,21 @@
       ],
       "additionalProperties": false
     },
+    "RollbackConfig": {
+      "properties": {
+        "timeoutMinutes": {
+          "type": "integer",
+          "description": "length of time in minutes to wait before cancelling the update. The timeout is a minimum-bound: cancellation occurs no sooner than the time specified, but can occur shortly thereafter. Valid values are between 120 (2 hours) and 10080 (7 days). When unset, EKS defaults to 720 (12 hours).",
+          "x-intellij-html-description": "length of time in minutes to wait before cancelling the update. The timeout is a minimum-bound: cancellation occurs no sooner than the time specified, but can occur shortly thereafter. Valid values are between 120 (2 hours) and 10080 (7 days). When unset, EKS defaults to 720 (12 hours)."
+        }
+      },
+      "preferredOrder": [
+        "timeoutMinutes"
+      ],
+      "additionalProperties": false,
+      "description": "holds the rollback configuration used when updating the cluster version. EKS waits for the configured timeout before automatically cancelling an in-progress cluster version update.",
+      "x-intellij-html-description": "holds the rollback configuration used when updating the cluster version. EKS waits for the configured timeout before automatically cancelling an in-progress cluster version update."
+    },
     "SSOIdentity": {
       "required": [
         "id",

pkg/apis/eksctl.io/v1alpha5/types.go

@@ -708,6 +708,10 @@ type ClusterMeta struct {
 	// When updating cluster version, provide the force flag to override upgrade-blocking insights
 	// +optional
 	ForceUpdateVersion *bool `json:"forceUpdateVersion,omitempty"`
+	// RollbackConfig configures the automatic rollback behaviour when updating
+	// the cluster version (including downgrades).
+	// +optional
+	RollbackConfig *RollbackConfig `json:"rollbackConfig,omitempty"`
 	// Tags are used to tag AWS resources created by eksctl
 	// +optional
 	Tags map[string]string `json:"tags,omitempty"`
@@ -727,6 +731,19 @@ type UpgradePolicy struct {
 	SupportType string `json:"supportType,omitempty"`
 }
 
+// RollbackConfig holds the rollback configuration used when updating the
+// cluster version. EKS waits for the configured timeout before automatically
+// cancelling an in-progress cluster version update.
+type RollbackConfig struct {
+	// TimeoutMinutes is the length of time in minutes to wait before cancelling
+	// the update. The timeout is a minimum-bound: cancellation occurs no sooner
+	// than the time specified, but can occur shortly thereafter. Valid values
+	// are between 120 (2 hours) and 10080 (7 days). When unset, EKS defaults to
+	// 720 (12 hours).
+	// +optional
+	TimeoutMinutes *int `json:"timeoutMinutes,omitempty"`
+}
+
 // KubernetesNetworkConfig contains cluster networking options
 type KubernetesNetworkConfig struct {
 	// Valid variants are `IPFamily` constants

pkg/apis/eksctl.io/v1alpha5/validation.go

@@ -182,6 +182,23 @@ func validateSupportType(supportType string) error {
 	return nil
 }
 
+// RollbackTimeoutMinutesMin and RollbackTimeoutMinutesMax are the inclusive
+// bounds EKS accepts for metadata.rollbackConfig.timeoutMinutes.
+const (
+	RollbackTimeoutMinutesMin = 120
+	RollbackTimeoutMinutesMax = 10080
+)
+
+func validateRollbackConfig(rollbackConfig *RollbackConfig) error {
+	if rollbackConfig == nil || rollbackConfig.TimeoutMinutes == nil {
+		return nil
+	}
+	if timeout := *rollbackConfig.TimeoutMinutes; timeout < RollbackTimeoutMinutesMin || timeout > RollbackTimeoutMinutesMax {
+		return fmt.Errorf("metadata.rollbackConfig.timeoutMinutes must be between %d and %d, got %d", RollbackTimeoutMinutesMin, RollbackTimeoutMinutesMax, timeout)
+	}
+	return nil
+}
+
 // ValidateClusterConfig checks compatible fields of a given ClusterConfig
 func ValidateClusterConfig(cfg *ClusterConfig) error {
 	if cfg.UpgradePolicy != nil {
@@ -192,6 +209,12 @@ func ValidateClusterConfig(cfg *ClusterConfig) error {
 		}
 	}
 
+	if cfg.Metadata != nil {
+		if err := validateRollbackConfig(cfg.Metadata.RollbackConfig); err != nil {
+			return err
+		}
+	}
+
 	if IsDisabled(cfg.IAM.WithOIDC) && len(cfg.IAM.ServiceAccounts) > 0 {
 		return fmt.Errorf("iam.withOIDC must be enabled explicitly for iam.serviceAccounts to be created")
 	}

pkg/apis/eksctl.io/v1alpha5/validation_test.go

@@ -55,6 +55,42 @@ var _ = Describe("ClusterConfig validation", func() {
 		})
 	})
 
+	Describe("metadata.rollbackConfig validation", func() {
+		var cfg *api.ClusterConfig
+
+		BeforeEach(func() {
+			cfg = api.NewClusterConfig()
+		})
+
+		It("accepts a missing rollbackConfig", func() {
+			Expect(api.ValidateClusterConfig(cfg)).To(Succeed())
+		})
+
+		It("accepts a rollbackConfig without a timeout", func() {
+			cfg.Metadata.RollbackConfig = &api.RollbackConfig{}
+			Expect(api.ValidateClusterConfig(cfg)).To(Succeed())
+		})
+
+		DescribeTable("timeoutMinutes bounds",
+			func(timeoutMinutes int, expectValid bool) {
+				cfg.Metadata.RollbackConfig = &api.RollbackConfig{
+					TimeoutMinutes: aws.Int(timeoutMinutes),
+				}
+				err := api.ValidateClusterConfig(cfg)
+				if expectValid {
+					Expect(err).NotTo(HaveOccurred())
+				} else {
+					Expect(err).To(MatchError(ContainSubstring("metadata.rollbackConfig.timeoutMinutes must be between 120 and 10080")))
+				}
+			},
+			Entry("rejects below the minimum", 119, false),
+			Entry("accepts the minimum", 120, true),
+			Entry("accepts a value within range", 720, true),
+			Entry("accepts the maximum", 10080, true),
+			Entry("rejects above the maximum", 10081, false),
+		)
+	})
+
 	Describe("nodeGroups[*].name validation", func() {
 		var (
 			cfg *api.ClusterConfig

pkg/eks/update.go

@@ -236,6 +236,14 @@ func (c *ClusterProvider) UpdateClusterVersion(ctx context.Context, cfg *api.Clu
 	if cfg.Metadata.ForceUpdateVersion != nil {
 		input.Force = *cfg.Metadata.ForceUpdateVersion
 	}
+	// Only send a RollbackConfig when there is a value to set; an empty struct
+	// would serialize as "rollbackConfig": {}, which is redundant with omitting it
+	// (EKS applies its default timeout either way).
+	if rollbackConfig := cfg.Metadata.RollbackConfig; rollbackConfig != nil && rollbackConfig.TimeoutMinutes != nil {
+		input.RollbackConfig = &ekstypes.RollbackConfig{
+			TimeoutMinutes: aws.Int32(int32(*rollbackConfig.TimeoutMinutes)),
+		}
+	}
 	output, err := c.AWSProvider.EKS().UpdateClusterVersion(ctx, input)
 	if err != nil {
 		return nil, err