Add maintenance state for customer action needed (Azure#3294)

ventifus · Feb 7, 2024 · 90cd002 · 90cd002
1 parent 7768311
commit 90cd002
Show file tree

Hide file tree

Showing 8 changed files with 570 additions and 17 deletions.
diff --git a/pkg/api/admin/openshiftcluster.go b/pkg/api/admin/openshiftcluster.go
@@ -87,24 +87,37 @@ const (
 type MaintenanceState string
 
 const (
-	MaintenanceStateNone      MaintenanceState = "None"
-	MaintenanceStatePending   MaintenanceState = "Pending"
-	MaintenanceStatePlanned   MaintenanceState = "Planned"
-	MaintenanceStateUnplanned MaintenanceState = "Unplanned"
+	MaintenanceStateNone                 MaintenanceState = "None"
+	MaintenanceStatePending              MaintenanceState = "Pending"
+	MaintenanceStatePlanned              MaintenanceState = "Planned"
+	MaintenanceStateUnplanned            MaintenanceState = "Unplanned"
+	MaintenanceStateCustomerActionNeeded MaintenanceState = "CustomerActionNeeded"
 )
 
 type MaintenanceTask string
 
 const (
+	//
+	// Maintenance tasks that perform work on the cluster
+	//
+
 	MaintenanceTaskEverything MaintenanceTask = "Everything"
 	MaintenanceTaskOperator   MaintenanceTask = "OperatorUpdate"
 	MaintenanceTaskRenewCerts MaintenanceTask = "CertificatesRenewal"
 
+	//
 	// Maintenance tasks for updating customer maintenance signals
+	//
+
+	MaintenanceTaskPending MaintenanceTask = "Pending"
+
 	// None signal should only be used when (1) admin update fails and (2) SRE fixes the failed admin update without running another admin updates
 	// Admin update success should automatically set the cluster into None state
-	MaintenanceTaskPending MaintenanceTask = "Pending"
-	MaintenanceTaskNone    MaintenanceTask = "None"
+	MaintenanceTaskNone MaintenanceTask = "None"
+
+	// Customer action needed signal should only be used when (1) admin update fails and (2) customer needs to take action to resolve the failure
+	// To remove the signal after customer takes action, use maintenance task None
+	MaintenanceTaskCustomerActionNeeded MaintenanceTask = "CustomerActionNeeded"
 )
 
 // Operator feature flags

diff --git a/pkg/api/admin/openshiftcluster_validatestatic.go b/pkg/api/admin/openshiftcluster_validatestatic.go
@@ -38,7 +38,8 @@ func validateMaintenanceTask(task MaintenanceTask) error {
 		task == MaintenanceTaskOperator ||
 		task == MaintenanceTaskRenewCerts ||
 		task == MaintenanceTaskPending ||
-		task == MaintenanceTaskNone) {
+		task == MaintenanceTaskNone ||
+		task == MaintenanceTaskCustomerActionNeeded) {
 		return api.NewCloudError(http.StatusBadRequest, api.CloudErrorCodeInvalidParameter, "properties.maintenanceTask", "Invalid enum parameter.")
 	}
 

diff --git a/pkg/api/admin/openshiftcluster_validatestatic_test.go b/pkg/api/admin/openshiftcluster_validatestatic_test.go
@@ -702,6 +702,19 @@ func TestOpenShiftClusterStaticValidateDelta(t *testing.T) {
 				oc.Properties.MaintenanceTask = ""
 			},
 		},
+		{
+			name: "maintenanceTask change to customer action needed allowed",
+			oc: func() *OpenShiftCluster {
+				return &OpenShiftCluster{
+					Properties: OpenShiftClusterProperties{
+						MaintenanceTask: MaintenanceTaskCustomerActionNeeded,
+					},
+				}
+			},
+			modify: func(oc *OpenShiftCluster) {
+				oc.Properties.MaintenanceTask = ""
+			},
+		},
 		{
 			name: "maintenanceTask change to other values is disallowed",
 			oc: func() *OpenShiftCluster {

diff --git a/pkg/api/openshiftcluster.go b/pkg/api/openshiftcluster.go
@@ -185,24 +185,37 @@ const (
 type MaintenanceState string
 
 const (
-	MaintenanceStateNone      MaintenanceState = "None"
-	MaintenanceStatePending   MaintenanceState = "Pending"
-	MaintenanceStatePlanned   MaintenanceState = "Planned"
-	MaintenanceStateUnplanned MaintenanceState = "Unplanned"
+	MaintenanceStateNone                 MaintenanceState = "None"
+	MaintenanceStatePending              MaintenanceState = "Pending"
+	MaintenanceStatePlanned              MaintenanceState = "Planned"
+	MaintenanceStateUnplanned            MaintenanceState = "Unplanned"
+	MaintenanceStateCustomerActionNeeded MaintenanceState = "CustomerActionNeeded"
 )
 
 type MaintenanceTask string
 
 const (
+	//
+	// Maintenance tasks that perform work on the cluster
+	//
+
 	MaintenanceTaskEverything MaintenanceTask = "Everything"
 	MaintenanceTaskOperator   MaintenanceTask = "OperatorUpdate"
 	MaintenanceTaskRenewCerts MaintenanceTask = "CertificatesRenewal"
 
+	//
 	// Maintenance tasks for updating customer maintenance signals
+	//
+
+	MaintenanceTaskPending MaintenanceTask = "Pending"
+
 	// None signal should only be used when (1) admin update fails and (2) SRE fixes the failed admin update without running another admin updates
 	// Admin update success should automatically set the cluster into None state
-	MaintenanceTaskPending MaintenanceTask = "Pending"
-	MaintenanceTaskNone    MaintenanceTask = "None"
+	MaintenanceTaskNone MaintenanceTask = "None"
+
+	// Customer action needed signal should only be used when (1) admin update fails and (2) customer needs to take action to resolve the failure
+	// To remove the signal after customer takes action, use maintenance task None
+	MaintenanceTaskCustomerActionNeeded MaintenanceTask = "CustomerActionNeeded"
 )
 
 // IsMaintenanceOngoingTask returns true if the maintenance task should change state to maintenance ongoing (planned/unplanned)

diff --git a/pkg/frontend/openshiftcluster_putorpatch.go b/pkg/frontend/openshiftcluster_putorpatch.go
@@ -345,6 +345,8 @@ func adminUpdateProvisioningState(doc *api.OpenShiftClusterDocument) {
 			doc.OpenShiftCluster.Properties.MaintenanceState = api.MaintenanceStatePending
 		case api.MaintenanceTaskNone:
 			doc.OpenShiftCluster.Properties.MaintenanceState = api.MaintenanceStateNone
+		case api.MaintenanceTaskCustomerActionNeeded:
+			doc.OpenShiftCluster.Properties.MaintenanceState = api.MaintenanceStateCustomerActionNeeded
 		}
 
 		// This enables future admin update actions with body `{}` to succeed