✨ wip implement .spec.progressDeadlineMinutes

api/v1/clusterextension_types.go

@@ -107,6 +107,17 @@ type ClusterExtensionSpec struct {
 	//
 	// +optional
 	Config *ClusterExtensionConfig `json:"config,omitempty"`
+
+	// progressDeadlineMinutes is an optional field that defines the maximum period
+	// of time in minutes after which an installation should be considered failed and
+	// require manual intervention. This functionality is disabled when no value
+	// is provided. The minimum period is 10 minutes, and the maximum is 720 minutes (12 hours).
+	//
+	// +kubebuilder:validation:Minimum:=10
+	// +kubebuilder:validation:Maximum:=720
+	// +optional
+	// <opcon:experimental>
+	ProgressDeadlineMinutes int32 `json:"progressDeadlineMinutes,omitempty"`
 }
 
 const SourceTypeCatalog = "Catalog"

api/v1/clusterextensionrevision_types.go

@@ -87,6 +87,17 @@ type ClusterExtensionRevisionSpec struct {
 	// +listMapKey=name
 	// +optional
 	Phases []ClusterExtensionRevisionPhase `json:"phases,omitempty"`
+
+	// progressDeadlineMinutes is an optional field that defines the maximum period
+	// of time in minutes after which an installation should be considered failed and
+	// require manual intervention. This functionality is disabled when no value
+	// is provided. The minimum period is 10 minutes, and the maximum is 720 minutes (12 hours).
+	//
+	// +kubebuilder:validation:Minimum:=10
+	// +kubebuilder:validation:Maximum:=720
+	// +optional
+	// <opcon:experimental>
+	ProgressDeadlineMinutes int32 `json:"progressDeadlineMinutes,omitempty"`
 }
 
 // ClusterExtensionRevisionLifecycleState specifies the lifecycle state of the ClusterExtensionRevision.

api/v1/common_types.go

@@ -32,6 +32,7 @@ const (
 	ReasonDeprecated = "Deprecated"
 
 	// Common reasons
-	ReasonSucceeded = "Succeeded"
-	ReasonFailed    = "Failed"
+	ReasonSucceeded                = "Succeeded"
+	ReasonFailed                   = "Failed"
+	ReasonProgressDeadlineExceeded = "ProgressDeadlineExceeded"
 )

docs/api-reference/olmv1-api-reference.md

@@ -344,6 +344,7 @@ _Appears in:_
 | `source` _[SourceConfig](#sourceconfig)_ | source is required and selects the installation source of content for this ClusterExtension.<br />Set the sourceType field to perform the selection.<br />Catalog is currently the only implemented sourceType.<br />Setting sourceType to "Catalog" requires the catalog field to also be defined.<br />Below is a minimal example of a source definition (in yaml):<br />source:<br />  sourceType: Catalog<br />  catalog:<br />    packageName: example-package |  | Required: \{\} <br /> |
 | `install` _[ClusterExtensionInstallConfig](#clusterextensioninstallconfig)_ | install is optional and configures installation options for the ClusterExtension,<br />such as the pre-flight check configuration. |  |  |
 | `config` _[ClusterExtensionConfig](#clusterextensionconfig)_ | config is optional and specifies bundle-specific configuration.<br />Configuration is bundle-specific and a bundle may provide a configuration schema.<br />When not specified, the default configuration of the resolved bundle is used.<br />config is validated against a configuration schema provided by the resolved bundle. If the bundle does not provide<br />a configuration schema the bundle is deemed to not be configurable. More information on how<br />to configure bundles can be found in the OLM documentation associated with your current OLM version. |  |  |
+| `progressDeadlineMinutes` _integer_ | progressDeadlineMinutes is an optional field that defines the maximum period<br />of time in minutes after which an installation should be considered failed and<br />require manual intervention. This functionality is disabled when no value<br />is provided. The minimum period is 10 minutes, and the maximum is 720 minutes (12 hours).<br /><opcon:experimental> |  | Maximum: 720 <br />Minimum: 10 <br /> |
 
 
 #### ClusterExtensionStatus

helm/olmv1/base/operator-controller/crd/experimental/olm.operatorframework.io_clusterextensionrevisions.yaml

@@ -166,6 +166,16 @@ spec:
                 x-kubernetes-validations:
                 - message: phases is immutable
                   rule: self == oldSelf || oldSelf.size() == 0
+              progressDeadlineMinutes:
+                description: |-
+                  progressDeadlineMinutes is an optional field that defines the maximum period
+                  of time in minutes after which an installation should be considered failed and
+                  require manual intervention. This functionality is disabled when no value
+                  is provided. The minimum period is 10 minutes, and the maximum is 720 minutes (12 hours).
+                format: int32
+                maximum: 720
+                minimum: 10
+                type: integer
               revision:
                 description: |-
                   revision is a required, immutable sequence number representing a specific revision

helm/olmv1/base/operator-controller/crd/experimental/olm.operatorframework.io_clusterextensions.yaml

@@ -165,6 +165,16 @@ spec:
                   rule: self == oldSelf
                 - message: namespace must be a valid DNS1123 label
                   rule: self.matches("^[a-z0-9]([-a-z0-9]*[a-z0-9])?$")
+              progressDeadlineMinutes:
+                description: |-
+                  progressDeadlineMinutes is an optional field that defines the maximum period
+                  of time in minutes after which an installation should be considered failed and
+                  require manual intervention. This functionality is disabled when no value
+                  is provided. The minimum period is 10 minutes, and the maximum is 720 minutes (12 hours).
+                format: int32
+                maximum: 720
+                minimum: 10
+                type: integer
               serviceAccount:
                 description: |-
                   serviceAccount specifies a ServiceAccount used to perform all interactions with the cluster

internal/operator-controller/applier/boxcutter.go

@@ -191,7 +191,7 @@ func (r *SimpleRevisionGenerator) buildClusterExtensionRevision(
 	annotations[labels.ServiceAccountNameKey] = ext.Spec.ServiceAccount.Name
 	annotations[labels.ServiceAccountNamespaceKey] = ext.Spec.Namespace
 
-	return &ocv1.ClusterExtensionRevision{
+	cer := &ocv1.ClusterExtensionRevision{
 		ObjectMeta: metav1.ObjectMeta{
 			Annotations: annotations,
 			Labels: map[string]string{
@@ -206,6 +206,10 @@ func (r *SimpleRevisionGenerator) buildClusterExtensionRevision(
 			Phases:         PhaseSort(objects),
 		},
 	}
+	if p := ext.Spec.ProgressDeadlineMinutes; p > 0 {
+		cer.Spec.ProgressDeadlineMinutes = p
+	}
+	return cer
 }
 
 // BoxcutterStorageMigrator migrates ClusterExtensions from Helm-based storage to

internal/operator-controller/conditionsets/conditionsets.go

@@ -41,4 +41,5 @@ var ConditionReasons = []string{
 	ocv1.ReasonRetrying,
 	ocv1.ReasonAbsent,
 	ocv1.ReasonRollingOut,
+	ocv1.ReasonProgressDeadlineExceeded,
 }

internal/operator-controller/controllers/clusterextensionrevision_controller.go

@@ -8,6 +8,7 @@ import (
 	"errors"
 	"fmt"
 	"strings"
+	"sync"
 	"time"
 
 	appsv1 "k8s.io/api/apps/v1"
@@ -43,9 +44,10 @@ const (
 // ClusterExtensionRevisionReconciler actions individual snapshots of ClusterExtensions,
 // as part of the boxcutter integration.
 type ClusterExtensionRevisionReconciler struct {
-	Client                client.Client
-	RevisionEngineFactory RevisionEngineFactory
-	TrackingCache         trackingCache
+	Client                 client.Client
+	RevisionEngineFactory  RevisionEngineFactory
+	TrackingCache          trackingCache
+	progressionRequeueOnce sync.Map
 }
 
 type trackingCache interface {
@@ -74,6 +76,17 @@ func (c *ClusterExtensionRevisionReconciler) Reconcile(ctx context.Context, req
 	reconciledRev := existingRev.DeepCopy()
 	res, reconcileErr := c.reconcile(ctx, reconciledRev)
 
+	if pd := existingRev.Spec.ProgressDeadlineMinutes; pd > 0 {
+		if cnd := meta.FindStatusCondition(reconciledRev.Status.Conditions, ocv1.ClusterExtensionRevisionTypeProgressing); cnd != nil && cnd.Status == metav1.ConditionTrue && cnd.Reason != ocv1.ReasonSucceeded {
+			timeout := time.Duration(pd) * time.Minute
+			if time.Since(existingRev.CreationTimestamp.Time) > timeout {
+				markAsNotProgressing(reconciledRev, ocv1.ReasonProgressDeadlineExceeded, fmt.Sprintf("Revision has not rolled out for %d minutes.", pd))
+				// reset any errors that may have occurred during reconciliation and stop any further reconciliation due to the reached timeout
+				reconcileErr = nil
+				res = ctrl.Result{}
+			}
+		}
+	}
 	// Do checks before any Update()s, as Update() may modify the resource structure!
 	updateStatus := !equality.Semantic.DeepEqual(existingRev.Status, reconciledRev.Status)
 
@@ -92,6 +105,10 @@ func (c *ClusterExtensionRevisionReconciler) Reconcile(ctx context.Context, req
 		}
 	}
 
+	if _, found := c.progressionRequeueOnce.Load(existingRev.GetUID()); !found && reconcileErr == nil && existingRev.Spec.ProgressDeadlineMinutes > 0 {
+		c.progressionRequeueOnce.Store(existingRev.GetUID(), true)
+		res = ctrl.Result{RequeueAfter: time.Duration(existingRev.Spec.ProgressDeadlineMinutes) * time.Minute}
+	}
 	return res, reconcileErr
 }
 

manifests/experimental-e2e.yaml

@@ -778,6 +778,16 @@ spec:
                 x-kubernetes-validations:
                 - message: phases is immutable
                   rule: self == oldSelf || oldSelf.size() == 0
+              progressDeadlineMinutes:
+                description: |-
+                  progressDeadlineMinutes is an optional field that defines the maximum period
+                  of time in minutes after which an installation should be considered failed and
+                  require manual intervention. This functionality is disabled when no value
+                  is provided. The minimum period is 10 minutes, and the maximum is 720 minutes (12 hours).
+                format: int32
+                maximum: 720
+                minimum: 10
+                type: integer
               revision:
                 description: |-
                   revision is a required, immutable sequence number representing a specific revision
@@ -1052,6 +1062,16 @@ spec:
                   rule: self == oldSelf
                 - message: namespace must be a valid DNS1123 label
                   rule: self.matches("^[a-z0-9]([-a-z0-9]*[a-z0-9])?$")
+              progressDeadlineMinutes:
+                description: |-
+                  progressDeadlineMinutes is an optional field that defines the maximum period
+                  of time in minutes after which an installation should be considered failed and
+                  require manual intervention. This functionality is disabled when no value
+                  is provided. The minimum period is 10 minutes, and the maximum is 720 minutes (12 hours).
+                format: int32
+                maximum: 720
+                minimum: 10
+                type: integer
               serviceAccount:
                 description: |-
                   serviceAccount specifies a ServiceAccount used to perform all interactions with the cluster

manifests/experimental.yaml

@@ -739,6 +739,16 @@ spec:
                 x-kubernetes-validations:
                 - message: phases is immutable
                   rule: self == oldSelf || oldSelf.size() == 0
+              progressDeadlineMinutes:
+                description: |-
+                  progressDeadlineMinutes is an optional field that defines the maximum period
+                  of time in minutes after which an installation should be considered failed and
+                  require manual intervention. This functionality is disabled when no value
+                  is provided. The minimum period is 10 minutes, and the maximum is 720 minutes (12 hours).
+                format: int32
+                maximum: 720
+                minimum: 10
+                type: integer
               revision:
                 description: |-
                   revision is a required, immutable sequence number representing a specific revision
@@ -1013,6 +1023,16 @@ spec:
                   rule: self == oldSelf
                 - message: namespace must be a valid DNS1123 label
                   rule: self.matches("^[a-z0-9]([-a-z0-9]*[a-z0-9])?$")
+              progressDeadlineMinutes:
+                description: |-
+                  progressDeadlineMinutes is an optional field that defines the maximum period
+                  of time in minutes after which an installation should be considered failed and
+                  require manual intervention. This functionality is disabled when no value
+                  is provided. The minimum period is 10 minutes, and the maximum is 720 minutes (12 hours).
+                format: int32
+                maximum: 720
+                minimum: 10
+                type: integer
               serviceAccount:
                 description: |-
                   serviceAccount specifies a ServiceAccount used to perform all interactions with the cluster

test/e2e/features/install.feature

@@ -297,3 +297,31 @@ Feature: Install ClusterExtension
         valid: true
         mutate: true
       """
+
+  @foo
+  @BoxcutterRuntime
+  Scenario: Progression deadline
+    Given min value for ClusterExtension .spec.progressDeadlineMinutes is set to 1
+    And min value for ClusterExtensionRevision .spec.progressDeadlineMinutes is set to 1
+    When ClusterExtension is applied
+      """
+      apiVersion: olm.operatorframework.io/v1
+      kind: ClusterExtension
+      metadata:
+        name: ${NAME}
+      spec:
+        namespace: ${TEST_NAMESPACE}
+        progressDeadlineMinutes: 1
+        serviceAccount:
+          name: olm-sa
+        source:
+          sourceType: Catalog
+          catalog:
+            packageName: test
+            version: 1.0.3
+            selector:
+              matchLabels:
+                "olm.operatorframework.io/metadata.name": test-catalog
+      """
+    Then ClusterExtensionRevision "${NAME}-1" reports Progressing as False with Reason ProgressDeadlineExceeded
+    And ClusterExtension reports Progressing as False with Reason ProgressDeadlineExceeded

test/e2e/steps/steps.go

@@ -88,6 +88,8 @@ func RegisterSteps(sc *godog.ScenarioContext) {
 
 	sc.Step(`^(?i)operator "([^"]+)" target namespace is "([^"]+)"$`, OperatorTargetNamespace)
 	sc.Step(`^(?i)Prometheus metrics are returned in the response$`, PrometheusMetricsAreReturned)
+
+	sc.Step(`^(?i)min value for (ClusterExtension|ClusterExtensionRevision) ((?:\.[a-zA-Z]+)+) is set to (\d+)$`, SetCRDFieldMinValue)
 }
 
 func init() {
@@ -730,3 +732,30 @@ func MarkTestOperatorNotReady(ctx context.Context, state string) error {
 	_, err = k8sClient("exec", podName, "-n", sc.namespace, "--", op, "/var/www/ready")
 	return err
 }
+
+// SetCRDFieldMinValue patches a CRD to set the minimum value for a field.
+// jsonPath is in the format ".spec.fieldName" and gets converted to the CRD schema path.
+func SetCRDFieldMinValue(_ context.Context, resourceType, jsonPath string, minValue int) error {
+	var crdName string
+	switch resourceType {
+	case "ClusterExtension":
+		crdName = "clusterextensions.olm.operatorframework.io"
+	case "ClusterExtensionRevision":
+		crdName = "clusterextensionrevisions.olm.operatorframework.io"
+	default:
+		return fmt.Errorf("unsupported resource type: %s", resourceType)
+	}
+
+	// Convert JSON path like ".spec.progressDeadlineMinutes" to CRD schema path
+	// e.g., ".spec.progressDeadlineMinutes" -> "properties/spec/properties/progressDeadlineMinutes"
+	parts := strings.Split(strings.TrimPrefix(jsonPath, "."), ".")
+	schemaParts := make([]string, 0, 2*len(parts))
+	for _, part := range parts {
+		schemaParts = append(schemaParts, "properties", part)
+	}
+	patchPath := fmt.Sprintf("/spec/versions/0/schema/openAPIV3Schema/%s/minimum", strings.Join(schemaParts, "/"))
+
+	patch := fmt.Sprintf(`[{"op": "replace", "path": "%s", "value": %d}]`, patchPath, minValue)
+	_, err := k8sClient("patch", "crd", crdName, "--type=json", "-p", patch)
+	return err
+}

testdata/images/bundles/test-operator/v1.0.3/manifests/bundle.configmap.yaml

@@ -0,0 +1,6 @@
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: wrong-test-configmap
+wrongfield:
+  name: "test-configmap"