Allow disabling version availability check at the start of the upgrade (#177)

Adds a new field `.spec.desiredVersionCheckAvailability`, defaulting to true, to the `UpgradeJob` definition.

Author: bastjan

api/v1beta1/upgradejob_types.go

@@ -72,10 +72,26 @@ type UpgradeJobSpec struct {
 	// +optional
 	DesiredVersion *configv1.Update `json:"desiredVersion,omitempty"`
 
+	// DesiredVersionCheckAvailability defines whether to check if the desired version is available at the start of the upgrade.
+	// This is done by searching the ClusterVersion status for the desired version.
+	// If set to true, the upgrade job will fail with reason UpgradeJobReasonUpgradeWithdrawn if the desired version is not available.
+	// If set to false, the upgrade job will not check for the availability of the desired version and will apply the DesiredVersion directly.
+	// Defaults to true.
+	// +optional
+	DesiredVersionCheckAvailability *bool `json:"desiredVersionCheckAvailability,omitempty"`
+
 	// UpgradeJobConfig defines the configuration for the upgrade job
 	UpgradeJobConfig `json:"config"`
 }
 
+// GetDesiredVersionCheckAvailability returns DesiredVersionCheckAvailability with a default value of true if nil.
+func (s UpgradeJobSpec) GetDesiredVersionCheckAvailability() bool {
+	if s.DesiredVersionCheckAvailability == nil {
+		return true
+	}
+	return *s.DesiredVersionCheckAvailability
+}
+
 // UpgradeJobConfig defines the configuration for the upgrade job
 type UpgradeJobConfig struct {
 	// UpgradeTimeout defines the timeout after which the upgrade is considered failed.

api/v1beta1/zz_generated.deepcopy.go

@@ -214,6 +214,14 @@ spec:
                 - message: Version must be set if Architecture is set
                   rule: 'has(self.architecture) && self.architecture != '''' ? self.version
                     != '''' : true'
+              desiredVersionCheckAvailability:
+                description: |-
+                  DesiredVersionCheckAvailability defines whether to check if the desired version is available at the start of the upgrade.
+                  This is done by searching the ClusterVersion status for the desired version.
+                  If set to true, the upgrade job will fail with reason UpgradeJobReasonUpgradeWithdrawn if the desired version is not available.
+                  If set to false, the upgrade job will not check for the availability of the desired version and will apply the DesiredVersion directly.
+                  Defaults to true.
+                type: boolean
               startAfter:
                 description: StartAfter defines the time after which the upgrade job
                   should start

config/crd/bases/managedupgrade.appuio.io_upgradejobs.yaml

@@ -226,8 +226,8 @@ func (r *UpgradeJobReconciler) reconcileStartedJob(ctx context.Context, uj *mana
 	if uj.Spec.DesiredVersion != nil {
 		// Check if the desired version is already set
 		if version.Spec.DesiredUpdate == nil || *version.Spec.DesiredUpdate != *uj.Spec.DesiredVersion {
-			update := clusterversion.FindAvailableUpdate(version, uj.Spec.DesiredVersion.Image, uj.Spec.DesiredVersion.Version)
-			if update == nil {
+			if uj.Spec.GetDesiredVersionCheckAvailability() &&
+				clusterversion.FindAvailableUpdate(version, uj.Spec.DesiredVersion.Image, uj.Spec.DesiredVersion.Version) == nil {
 				r.setStatusCondition(&uj.Status.Conditions, metav1.Condition{
 					Type:    managedupgradev1beta1.UpgradeJobConditionFailed,
 					Status:  metav1.ConditionTrue,

controllers/upgradejob_controller.go

@@ -19,6 +19,7 @@ import (
 	"k8s.io/apimachinery/pkg/runtime"
 	"k8s.io/apimachinery/pkg/types"
 	clientgoscheme "k8s.io/client-go/kubernetes/scheme"
+	"k8s.io/utils/ptr"
 	ctrl "sigs.k8s.io/controller-runtime"
 	"sigs.k8s.io/controller-runtime/pkg/client"
 	"sigs.k8s.io/controller-runtime/pkg/client/fake"
@@ -977,6 +978,66 @@ func Test_UpgradeJobReconciler_Reconcile_UpgradeWithdrawn(t *testing.T) {
 	require.Empty(t, ucv.Annotations[JobLockAnnotation], "should clear lock annotation")
 }
 
+func Test_UpgradeJobReconciler_Reconcile_UpgradeWithdrawn_NoCheckAvailability(t *testing.T) {
+	ctx := log.IntoContext(t.Context(), testr.New(t))
+	clock := mockClock{now: time.Date(2022, 12, 4, 22, 45, 0, 0, time.UTC)}
+
+	ucv := &configv1.ClusterVersion{
+		ObjectMeta: metav1.ObjectMeta{
+			Name: "version",
+		},
+		Spec: configv1.ClusterVersionSpec{
+			ClusterID: "9b588658-9671-429c-a762-34106da5795f",
+			DesiredUpdate: &configv1.Update{
+				Version: "4.5.12",
+				Image:   "quay.io/openshift-release-dev/ocp-release@sha256:d732fee6462de7f04f9432f1bb3925f57554db1d8c8d6f3138eea70e5787c7ae",
+			},
+		},
+		Status: configv1.ClusterVersionStatus{},
+	}
+
+	upgradeJob := &managedupgradev1beta1.UpgradeJob{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      "upgrade-1234-4-5-13",
+			Namespace: "appuio-openshift-upgrade-controller",
+		},
+		Spec: managedupgradev1beta1.UpgradeJobSpec{
+			StartBefore:                     metav1.NewTime(clock.Now().Add(time.Hour)),
+			StartAfter:                      metav1.NewTime(clock.Now().Add(-time.Hour)),
+			DesiredVersionCheckAvailability: ptr.To(false),
+			DesiredVersion: &configv1.Update{
+				Version: "4.5.13",
+				Image:   "quay.io/openshift-release-dev/ocp-release@sha256:d094f1952995b3c5fd8e0b19b128905931e1e8fdb4b6cb377857ab0dfddcff47",
+			},
+			UpgradeJobConfig: managedupgradev1beta1.UpgradeJobConfig{
+				UpgradeTimeout: metav1.Duration{Duration: 12 * time.Hour},
+				PreUpgradeHealthChecks: managedupgradev1beta1.UpgradeJobHealthCheck{
+					SkipDegradedOperatorsCheck: true,
+				},
+			},
+		},
+	}
+
+	client := controllerClient(t, ucv, upgradeJob)
+
+	subject := &UpgradeJobReconciler{
+		Client: client,
+		Scheme: client.Scheme(),
+
+		Clock: &clock,
+
+		ManagedUpstreamClusterVersionName: "version",
+	}
+
+	reconcileNTimes(t, subject, ctx, requestForObject(upgradeJob), 10)
+
+	require.NoError(t, client.Get(ctx, requestForObject(upgradeJob).NamespacedName, upgradeJob))
+	failedCond := apimeta.FindStatusCondition(upgradeJob.Status.Conditions, managedupgradev1beta1.UpgradeJobConditionFailed)
+	require.Nil(t, failedCond, "should not fail")
+	require.NoError(t, client.Get(ctx, requestForObject(ucv).NamespacedName, ucv))
+	require.Equal(t, upgradeJob.Spec.DesiredVersion.Version, ucv.Spec.DesiredUpdate.Version, "should update cluster version")
+}
+
 func Test_UpgradeJobReconciler_Reconcile_Timeout(t *testing.T) {
 	ctx := log.IntoContext(t.Context(), testr.New(t))
 	clock := mockClock{now: time.Date(2022, 12, 4, 22, 45, 0, 0, time.UTC)}