Fix flaky pod Watcher test in target allocator

The root cause of the flakiness was that the fake client we're using
doesn't support resource versions. Informers rely on these to avoid race
conditions when processing the initial object list. The flaky tests were
caused by this race condition. I've fixed this by having the tests check
whether the informer event handlers have synced before changing objects
using the client.

I've also had the test pass around a pointer to the watcher rather than
the whole struct.

Author: swiatekm

cmd/otel-allocator/internal/collector/collector.go

@@ -27,6 +27,7 @@ type Watcher struct {
 	log                          logr.Logger
 	k8sClient                    kubernetes.Interface
 	close                        chan struct{}
+	eventHandlerReg              cache.ResourceEventHandlerRegistration
 	minUpdateInterval            time.Duration
 	collectorNotReadyGracePeriod time.Duration
 	collectorsDiscovered         metric.Int64Gauge
@@ -63,7 +64,7 @@ func (k *Watcher) Watch(
 	}
 	informerFactory := informers.NewSharedInformerFactoryWithOptions(
 		k.k8sClient,
-		2*k.minUpdateInterval,
+		30*time.Second,
 		informers.WithNamespace(collectorNamespace),
 		informers.WithTweakListOptions(listOptionsFunc))
 	informer := informerFactory.Core().V1().Pods().Informer()
@@ -77,7 +78,7 @@ func (k *Watcher) Watch(
 		default:
 		}
 	}
-	_, err = informer.AddEventHandler(cache.ResourceEventHandlerFuncs{
+	k.eventHandlerReg, err = informer.AddEventHandler(cache.ResourceEventHandlerFuncs{
 		AddFunc: notifyFunc,
 		UpdateFunc: func(oldObj, newObj interface{}) {
 			notifyFunc(newObj)
@@ -92,6 +93,13 @@ func (k *Watcher) Watch(
 	return nil
 }
 
+func (k *Watcher) isSynced() bool {
+	if k.eventHandlerReg == nil {
+		return false
+	}
+	return k.eventHandlerReg.HasSynced()
+}
+
 // rateLimitedCollectorHandler runs fn on collectors present in the store whenever it gets a notification on the notify channel,
 // but not more frequently than once per k.eventPeriod.
 func (k *Watcher) rateLimitedCollectorHandler(notify chan struct{}, store cache.Store, fn func(collectors map[string]*allocation.Collector)) {

cmd/otel-allocator/internal/collector/collector_test.go

@@ -40,7 +40,7 @@ func (r *reportingGauge) Record(_ context.Context, value int64, _ ...metric.Reco
 	r.value.Store(value)
 }
 
-func getTestPodWatcher(collectorNotReadyGracePeriod time.Duration) Watcher {
+func getTestPodWatcher(collectorNotReadyGracePeriod time.Duration) *Watcher {
 	podWatcher := Watcher{
 		k8sClient:                    fake.NewClientset(),
 		close:                        make(chan struct{}),
@@ -49,7 +49,7 @@ func getTestPodWatcher(collectorNotReadyGracePeriod time.Duration) Watcher {
 		collectorNotReadyGracePeriod: collectorNotReadyGracePeriod,
 		collectorsDiscovered:         &reportingGauge{},
 	}
-	return podWatcher
+	return &podWatcher
 }
 
 func pod(name string) *v1.Pod {
@@ -105,7 +105,7 @@ func Test_runWatch(t *testing.T) {
 	namespace := "test-ns"
 	type args struct {
 		collectorNotReadyGracePeriod time.Duration
-		kubeFn                       func(t *testing.T, podWatcher Watcher)
+		kubeFn                       func(t *testing.T, podWatcher *Watcher)
 		collectorMap                 map[string]*allocation.Collector
 	}
 	tests := []struct {
@@ -117,7 +117,7 @@ func Test_runWatch(t *testing.T) {
 			name: "pod add",
 			args: args{
 				collectorNotReadyGracePeriod: 0 * time.Second,
-				kubeFn: func(t *testing.T, podWatcher Watcher) {
+				kubeFn: func(t *testing.T, podWatcher *Watcher) {
 					for _, k := range []string{"test-pod1", "test-pod2", "test-pod3"} {
 						p := pod(k)
 						_, err := podWatcher.k8sClient.CoreV1().Pods(namespace).Create(context.Background(), p, metav1.CreateOptions{})
@@ -145,7 +145,7 @@ func Test_runWatch(t *testing.T) {
 			name: "pod delete",
 			args: args{
 				collectorNotReadyGracePeriod: 0 * time.Second,
-				kubeFn: func(t *testing.T, podWatcher Watcher) {
+				kubeFn: func(t *testing.T, podWatcher *Watcher) {
 					for _, k := range []string{"test-pod2", "test-pod3"} {
 						err := podWatcher.k8sClient.CoreV1().Pods(namespace).Delete(context.Background(), k, metav1.DeleteOptions{})
 						assert.NoError(t, err)
@@ -187,14 +187,15 @@ func Test_runWatch(t *testing.T) {
 				_, err := podWatcher.k8sClient.CoreV1().Pods("test-ns").Create(context.Background(), p, metav1.CreateOptions{})
 				assert.NoError(t, err)
 			}
-			go func(podWatcher Watcher) {
+			go func() {
 				err := podWatcher.Watch(namespace, &labelSelector, func(colMap map[string]*allocation.Collector) {
 					mapMutex.Lock()
 					defer mapMutex.Unlock()
 					actual = colMap
 				})
 				require.NoError(t, err)
-			}(podWatcher)
+			}()
+			assert.Eventually(t, podWatcher.isSynced, time.Second*30, time.Millisecond*100)
 
 			tt.args.kubeFn(t, podWatcher)
 
@@ -311,14 +312,14 @@ func Test_gracePeriodWithNonRunningPodPhase(t *testing.T) {
 				assert.NoError(t, err)
 			}
 
-			go func(podWatcher Watcher) {
+			go func() {
 				err := podWatcher.Watch(namespace, &labelSelector, func(colMap map[string]*allocation.Collector) {
 					mapMutex.Lock()
 					defer mapMutex.Unlock()
 					actual = colMap
 				})
 				require.NoError(t, err)
-			}(podWatcher)
+			}()
 
 			assert.EventuallyWithT(t, func(collect *assert.CollectT) {
 				mapMutex.Lock()
@@ -434,14 +435,14 @@ func Test_gracePeriodWithNonReadyPodCondition(t *testing.T) {
 				assert.NoError(t, err)
 			}
 
-			go func(podWatcher Watcher) {
+			go func() {
 				err := podWatcher.Watch(namespace, &labelSelector, func(colMap map[string]*allocation.Collector) {
 					mapMutex.Lock()
 					defer mapMutex.Unlock()
 					actual = colMap
 				})
 				require.NoError(t, err)
-			}(podWatcher)
+			}()
 
 			assert.EventuallyWithT(t, func(collect *assert.CollectT) {
 				mapMutex.Lock()
@@ -461,11 +462,11 @@ func Test_closeChannel(t *testing.T) {
 	var wg sync.WaitGroup
 	wg.Add(1)
 
-	go func(podWatcher Watcher) {
+	go func() {
 		defer wg.Done()
 		err := podWatcher.Watch("default", &labelSelector, func(colMap map[string]*allocation.Collector) {})
 		require.NoError(t, err)
-	}(podWatcher)
+	}()
 
 	podWatcher.Close()
 	wg.Wait()