fix tg config deletion when gw is stuck in programmed = false

Author: zac-nixon

controllers/gateway/gateway_controller.go

@@ -277,7 +277,6 @@ func (r *gatewayReconciler) reconcileHelper(ctx context.Context, req reconcile.R
 			r.logger.Error(err, "Failed to process gateway delete")
 			return err
 		}
-		r.serviceReferenceCounter.UpdateRelations([]types.NamespacedName{}, k8s.NamespacedName(gw), true)
 		return nil
 	}
 	r.serviceReferenceCounter.UpdateRelations(getServicesFromRoutes(allRoutes), k8s.NamespacedName(gw), false)
@@ -313,6 +312,7 @@ func (r *gatewayReconciler) reconcileDelete(ctx context.Context, gw *gwv1.Gatewa
 		if err := r.backendSGProvider.Release(ctx, networking.ResourceTypeGateway, []types.NamespacedName{k8s.NamespacedName(gw)}); err != nil {
 			return err
 		}
+		r.serviceReferenceCounter.UpdateRelations([]types.NamespacedName{}, k8s.NamespacedName(gw), true)
 		// remove gateway finalizer
 		if err := r.finalizerManager.RemoveFinalizers(ctx, gw, r.finalizer); err != nil {
 			r.eventRecorder.Event(gw, corev1.EventTypeWarning, k8s.GatewayEventReasonFailedRemoveFinalizer, fmt.Sprintf("Failed remove gateway finalizer due to %v", err))

controllers/gateway/targetgroup_configuration_controller.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"fmt"
 	"github.com/go-logr/logr"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/types"
 	"k8s.io/client-go/kubernetes"
 	"k8s.io/client-go/tools/record"
@@ -98,7 +99,14 @@ func (r *targetgroupConfigurationReconciler) handleDelete(tgConf *elbv2gw.Target
 		}
 
 		for _, gw := range partial {
-			allGateways = append(allGateways, k8s.NamespacedName(gw))
+
+			if gw.Status.Conditions != nil {
+				for _, cond := range gw.Status.Conditions {
+					if cond.Type == string(gwv1.GatewayReasonAccepted) && cond.Status == metav1.ConditionTrue {
+						allGateways = append(allGateways, k8s.NamespacedName(gw))
+					}
+				}
+			}
 		}
 	}
 

pkg/gateway/referencecounter/service_reference_counter.go

@@ -1,6 +1,7 @@
 package referencecounter
 
 import (
+	"github.com/go-logr/logr"
 	"k8s.io/apimachinery/pkg/types"
 	"k8s.io/apimachinery/pkg/util/sets"
 	"sync"
@@ -13,7 +14,8 @@ type ServiceReferenceCounter interface {
 }
 
 type serviceReferenceCounter struct {
-	mutex sync.RWMutex
+	logger logr.Logger
+	mutex  sync.RWMutex
 	// key: gateway, value: set of svc
 	relations map[types.NamespacedName]sets.Set[types.NamespacedName]
 	refCount  map[types.NamespacedName]int
@@ -65,17 +67,19 @@ func (t *serviceReferenceCounter) updateRefCount(svcs sets.Set[types.NamespacedN
 func (t *serviceReferenceCounter) IsEligibleForRemoval(svcName types.NamespacedName, expectedGateways []types.NamespacedName) bool {
 	t.mutex.RLock()
 	defer t.mutex.RUnlock()
-	_, ok := t.refCount[svcName]
+	v, ok := t.refCount[svcName]
 	// If we have a ref count for this service, we can't remove it.
 	// updateRefCount should always remove 0 entry services.
 	if ok {
+		t.logger.Info("Got a ref count!", "ref", v)
 		return false
 	}
 
 	// Next we check if the Gateway cache is correctly populated. This prevents premature removal of items
 	// when the cache is not warm.
 	for _, gw := range expectedGateways {
 		if _, exists := t.relations[gw]; !exists {
+			t.logger.Info("Missing relation", "gateway", gw)
 			return false
 		}
 	}