Keep zones ready when DNS record pruning fails

api/v1/condition_types.go

@@ -20,6 +20,9 @@ const (
 	// ConditionTypeReady represents the fact that the object is ready.
 	ConditionTypeReady string = "Ready"
 
+	// ConditionTypePruned represents the result of the latest DNS record pruning attempt.
+	ConditionTypePruned string = "Pruned"
+
 	// ConditionReasonReady represents the fact that the object is ready.
 	ConditionReasonReady string = "Ready"
 
@@ -28,4 +31,10 @@ const (
 
 	// ConditionReasonFailed represents the fact that the object has failed.
 	ConditionReasonFailed string = "Failed"
+
+	// ConditionReasonPruneSucceeded represents the fact that DNS record pruning succeeded.
+	ConditionReasonPruneSucceeded string = "PruneSucceeded"
+
+	// ConditionReasonPruneFailed represents the fact that DNS record pruning failed.
+	ConditionReasonPruneFailed string = "PruneFailed"
 )

api/v1/zone_types.go

@@ -68,6 +68,7 @@ const (
 // +kubebuilder:printcolumn:name="Zone Name",type="string",JSONPath=".spec.name"
 // +kubebuilder:printcolumn:name="ID",type="string",JSONPath=".status.id"
 // +kubebuilder:printcolumn:name="Ready",type="string",JSONPath=`.status.conditions[?(@.type == "Ready")].status`
+// +kubebuilder:printcolumn:name="Pruned",type="string",JSONPath=`.status.conditions[?(@.type == "Pruned")].status`
 type Zone struct {
 	metav1.TypeMeta   `json:",inline"`
 	metav1.ObjectMeta `json:"metadata,omitempty"`

config/crd/bases/cloudflare-operator.io_zones.yaml

@@ -24,6 +24,9 @@ spec:
     - jsonPath: .status.conditions[?(@.type == "Ready")].status
       name: Ready
       type: string
+    - jsonPath: .status.conditions[?(@.type == "Pruned")].status
+      name: Pruned
+      type: string
     name: v1
     schema:
       openAPIV3Schema:

internal/controller/zone_controller.go

@@ -26,8 +26,10 @@ import (
 
 	cloudflare "github.com/cloudflare/cloudflare-go/v7"
 	"github.com/cloudflare/cloudflare-go/v7/dns"
+	"github.com/fluxcd/pkg/runtime/conditions"
 	corev1 "k8s.io/api/core/v1"
 	apierrors "k8s.io/apimachinery/pkg/api/errors"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	apierrutil "k8s.io/apimachinery/pkg/util/errors"
 	"sigs.k8s.io/controller-runtime/pkg/builder"
 	"sigs.k8s.io/controller-runtime/pkg/controller/controllerutil"
@@ -152,18 +154,45 @@ func (r *ZoneReconciler) reconcileZone(ctx context.Context, zone *cloudflareoper
 
 	zone.Status.ID = zoneID
 
+	var pruneErr error
 	if zone.Spec.Prune {
-		if err := r.handlePrune(ctx, cloudflareAPI, zone); err != nil {
-			intconditions.MarkFalse(zone, fmt.Errorf("failed to prune DNS records: %v", err))
-			return ctrl.Result{RequeueAfter: r.RetryInterval}, nil
-		}
+		pruneErr = r.handlePrune(ctx, cloudflareAPI, zone)
 	}
+	setZoneReadyConditions(zone, pruneErr)
 
-	intconditions.MarkTrue(zone, "Zone is ready")
+	if pruneErr != nil {
+		return ctrl.Result{RequeueAfter: r.RetryInterval}, nil
+	}
 
 	return ctrl.Result{RequeueAfter: zone.Spec.Interval.Duration}, nil
 }
 
+func setZoneReadyConditions(zone *cloudflareoperatoriov1.Zone, pruneErr error) {
+	intconditions.MarkTrue(zone, "Zone is ready")
+
+	if !zone.Spec.Prune {
+		conditions.Delete(zone, cloudflareoperatoriov1.ConditionTypePruned)
+		return
+	}
+
+	if pruneErr != nil {
+		conditions.Set(zone, &metav1.Condition{
+			Type:    cloudflareoperatoriov1.ConditionTypePruned,
+			Status:  metav1.ConditionFalse,
+			Reason:  cloudflareoperatoriov1.ConditionReasonPruneFailed,
+			Message: fmt.Sprintf("failed to prune DNS records: %v", pruneErr),
+		})
+		return
+	}
+
+	conditions.Set(zone, &metav1.Condition{
+		Type:    cloudflareoperatoriov1.ConditionTypePruned,
+		Status:  metav1.ConditionTrue,
+		Reason:  cloudflareoperatoriov1.ConditionReasonPruneSucceeded,
+		Message: "DNS record pruning succeeded",
+	})
+}
+
 // handlePrune deletes DNS records that are not managed by the operator if enabled
 func (r *ZoneReconciler) handlePrune(ctx context.Context, cloudflareAPI *cloudflare.Client, zone *cloudflareoperatoriov1.Zone) error {
 	log := ctrl.LoggerFrom(ctx)
@@ -182,7 +211,6 @@ func (r *ZoneReconciler) handlePrune(ctx context.Context, cloudflareAPI *cloudfl
 
 	cloudflareDNSRecords, err := listCloudflareDNSRecords(ctx, cloudflareAPI, zone.Status.ID, dns.RecordListParams{})
 	if err != nil {
-		intconditions.MarkFalse(zone, err)
 		return err
 	}
 

internal/controller/zone_controller_test.go

@@ -18,6 +18,7 @@ package controller
 
 import (
 	"context"
+	"errors"
 	"os"
 	"testing"
 
@@ -164,6 +165,51 @@ func TestZoneReconciler_reconcileZone(t *testing.T) {
 	})
 }
 
+func TestSetZoneReadyConditions(t *testing.T) {
+	t.Run("prune disabled", func(t *testing.T) {
+		g := NewWithT(t)
+		zone := &cloudflareoperatoriov1.Zone{}
+		conditions.Set(zone, conditions.FalseCondition(
+			cloudflareoperatoriov1.ConditionTypePruned,
+			cloudflareoperatoriov1.ConditionReasonPruneFailed,
+			"old prune failure",
+		))
+
+		setZoneReadyConditions(zone, nil)
+
+		g.Expect(conditions.IsTrue(zone, cloudflareoperatoriov1.ConditionTypeReady)).To(BeTrue())
+		g.Expect(conditions.Get(zone, cloudflareoperatoriov1.ConditionTypePruned)).To(BeNil())
+	})
+
+	t.Run("prune succeeded", func(t *testing.T) {
+		g := NewWithT(t)
+		zone := &cloudflareoperatoriov1.Zone{
+			Spec: cloudflareoperatoriov1.ZoneSpec{Prune: true},
+		}
+
+		setZoneReadyConditions(zone, nil)
+
+		g.Expect(conditions.IsTrue(zone, cloudflareoperatoriov1.ConditionTypeReady)).To(BeTrue())
+		g.Expect(conditions.IsTrue(zone, cloudflareoperatoriov1.ConditionTypePruned)).To(BeTrue())
+		g.Expect(conditions.GetReason(zone, cloudflareoperatoriov1.ConditionTypePruned)).To(Equal(cloudflareoperatoriov1.ConditionReasonPruneSucceeded))
+		g.Expect(conditions.GetMessage(zone, cloudflareoperatoriov1.ConditionTypePruned)).To(Equal("DNS record pruning succeeded"))
+	})
+
+	t.Run("prune failed", func(t *testing.T) {
+		g := NewWithT(t)
+		zone := &cloudflareoperatoriov1.Zone{
+			Spec: cloudflareoperatoriov1.ZoneSpec{Prune: true},
+		}
+
+		setZoneReadyConditions(zone, errors.New("Cloudflare API unavailable"))
+
+		g.Expect(conditions.IsTrue(zone, cloudflareoperatoriov1.ConditionTypeReady)).To(BeTrue())
+		g.Expect(conditions.IsFalse(zone, cloudflareoperatoriov1.ConditionTypePruned)).To(BeTrue())
+		g.Expect(conditions.GetReason(zone, cloudflareoperatoriov1.ConditionTypePruned)).To(Equal(cloudflareoperatoriov1.ConditionReasonPruneFailed))
+		g.Expect(conditions.GetMessage(zone, cloudflareoperatoriov1.ConditionTypePruned)).To(Equal("failed to prune DNS records: Cloudflare API unavailable"))
+	})
+}
+
 func TestManagedDNSRecordKeysForZone(t *testing.T) {
 	g := NewWithT(t)