feat: return cluster error messages on wait handler

Author: h3adex

CHANGELOG.md

@@ -31,6 +31,11 @@
   - Add field `Labels` (type `*map[string]string`) to structs `LoadBalancer`, `CreateLoadBalancerPayload`, `UpdateLoadBalancerPayload`
 - `git`: [v0.8.0](services/git/CHANGELOG.md#v080)
   - **Feature:** Add support for the instance patch operation
+- `ske`
+  - [v1.12.0](services/ske/CHANGELOG.md#v141)
+    - **Feature:** Exit `CreateOrUpdateClusterWaitHandler` early when the cluster reports structured errors and is in a failure state.
+
+
 
 ## Release (2025-09-11)
 - `cdn`: [v1.5.0](services/cdn/CHANGELOG.md#v150)

services/ske/CHANGELOG.md

@@ -1,3 +1,6 @@
+## v1.4.1
+- **Feature:** Exit `CreateOrUpdateClusterWaitHandler` early when the cluster reports structured errors and is in a failure state.
+
 ## v1.4.0
 - **Feature:** Add new field `Kubernetes` to `Nodepool` model
 

services/ske/VERSION

@@ -1 +1 @@
-v1.4.0
+v1.4.1

services/ske/wait/wait.go

@@ -33,6 +33,8 @@ const (
 	CredentialsRotationStateCompleted = "COMPLETED"
 	// Deprecated: InvalidArgusInstanceErrorCode is deprecated and will be removed after 14th November 2025. Use [ske.RUNTIMEERRORCODE_OBSERVABILITY_INSTANCE_NOT_FOUND] instead.
 	InvalidArgusInstanceErrorCode = "SKE_ARGUS_INSTANCE_NOT_FOUND"
+
+	ClusterNoValidHostFound = "SKE_NODE_NO_VALID_HOST_FOUND"
 )
 
 type APIClientClusterInterface interface {
@@ -42,6 +44,8 @@ type APIClientClusterInterface interface {
 
 // CreateOrUpdateClusterWaitHandler will wait for cluster creation or update
 func CreateOrUpdateClusterWaitHandler(ctx context.Context, a APIClientClusterInterface, projectId, region, name string) *wait.AsyncActionHandler[ske.Cluster] {
+	startTime := time.Now()
+
 	handler := wait.New(func() (waitFinished bool, response *ske.Cluster, err error) {
 		s, err := a.GetClusterExecute(ctx, projectId, region, name)
 		if err != nil {
@@ -56,6 +60,26 @@ func CreateOrUpdateClusterWaitHandler(ctx context.Context, a APIClientClusterInt
 			return true, s, nil
 		}
 
+		// If cluster is UNSPECIFIED or UNHEALTHY and has structured errors, exit early
+		hasStructuredErrors := s.Status.Errors != nil && len(*s.Status.Errors) > 0
+		if (state == ske.CLUSTERSTATUSSTATE_UNSPECIFIED || state == ske.CLUSTERSTATUSSTATE_UNHEALTHY) && hasStructuredErrors {
+			for _, clusterError := range *s.Status.Errors {
+				if clusterError.Code != nil && clusterError.Message != nil {
+					return true, s, nil
+				}
+			}
+		}
+
+		// Waiter has been running more than 15 minutes and cluster is still in CREATING or RECONCILING state with errors
+		if time.Since(startTime) > 15*time.Minute &&
+			(state == ske.CLUSTERSTATUSSTATE_CREATING || state == ske.CLUSTERSTATUSSTATE_RECONCILING) && hasStructuredErrors {
+			for _, clusterError := range *s.Status.Errors {
+				if clusterError.Code != nil && clusterError.Message != nil {
+					return true, s, nil
+				}
+			}
+		}
+
 		if state == ske.CLUSTERSTATUSSTATE_HEALTHY || state == ske.CLUSTERSTATUSSTATE_HIBERNATED {
 			return true, s, nil
 		}

services/ske/wait/wait_test.go

@@ -18,6 +18,7 @@ type apiClientClusterMocked struct {
 	name                 string
 	resourceState        ske.ClusterStatusState
 	invalidArgusInstance bool
+	errorList            *[]ske.ClusterError
 }
 
 const testRegion = "eu01"
@@ -45,7 +46,17 @@ func (a *apiClientClusterMocked) GetClusterExecute(_ context.Context, _, _, _ st
 	return &ske.Cluster{
 		Name: utils.Ptr("cluster"),
 		Status: &ske.ClusterStatus{
-			Aggregated: &rs,
+			Aggregated: utils.Ptr(rs),
+			Error: func() *ske.RuntimeError {
+				if a.invalidArgusInstance {
+					return &ske.RuntimeError{
+						Code:    utils.Ptr(ske.RUNTIMEERRORCODE_OBSERVABILITY_INSTANCE_NOT_FOUND),
+						Message: utils.Ptr("invalid argus instance"),
+					}
+				}
+				return nil
+			}(),
+			Errors: a.errorList,
 		},
 	}, nil
 }
@@ -77,6 +88,7 @@ func TestCreateOrUpdateClusterWaitHandler(t *testing.T) {
 		invalidArgusInstance bool
 		wantErr              bool
 		wantResp             bool
+		errorList            *[]ske.ClusterError
 	}{
 		{
 			desc:          "create_succeeded",
@@ -120,6 +132,40 @@ func TestCreateOrUpdateClusterWaitHandler(t *testing.T) {
 			wantErr:       true,
 			wantResp:      false,
 		},
+		{
+			desc:          "status_errors_present_state_unhealthy",
+			getFails:      false,
+			resourceState: ske.CLUSTERSTATUSSTATE_UNHEALTHY,
+			errorList: &[]ske.ClusterError{
+				{
+					Code:    utils.Ptr("ERR_CODE"),
+					Message: utils.Ptr("Error 1"),
+				},
+				{
+					Code:    utils.Ptr("ERR_OTHER"),
+					Message: utils.Ptr("Error 2"),
+				},
+			},
+			wantErr:  false,
+			wantResp: true,
+		},
+		{
+			desc:          "status_errors_present_state_unspecified",
+			getFails:      false,
+			resourceState: ske.CLUSTERSTATUSSTATE_UNSPECIFIED,
+			errorList: &[]ske.ClusterError{
+				{
+					Code:    utils.Ptr("ERR_CODE"),
+					Message: utils.Ptr("Error 1"),
+				},
+				{
+					Code:    utils.Ptr("ERR_OTHER"),
+					Message: utils.Ptr("Error 2"),
+				},
+			},
+			wantErr:  false,
+			wantResp: true,
+		},
 	}
 	for _, tt := range tests {
 		t.Run(tt.desc, func(t *testing.T) {
@@ -130,6 +176,7 @@ func TestCreateOrUpdateClusterWaitHandler(t *testing.T) {
 				name:                 name,
 				resourceState:        tt.resourceState,
 				invalidArgusInstance: tt.invalidArgusInstance,
+				errorList:            tt.errorList,
 			}
 			var wantRes *ske.Cluster
 			rs := ske.ClusterStatusState(tt.resourceState)
@@ -147,6 +194,10 @@ func TestCreateOrUpdateClusterWaitHandler(t *testing.T) {
 						Message: utils.Ptr("invalid argus instance"),
 					}
 				}
+
+				if tt.errorList != nil && len(*tt.errorList) > 0 {
+					wantRes.Status.Errors = tt.errorList
+				}
 			}
 
 			handler := CreateOrUpdateClusterWaitHandler(context.Background(), apiClient, "", testRegion, name)