new-pod-scale-up-delay=0 should effectively disable filtering out young pods

cluster-autoscaler/core/static_autoscaler.go

@@ -984,7 +984,10 @@ func (a *StaticAutoscaler) filterOutYoungPods(allUnschedulablePods []*apiv1.Pod,
 			}
 		}
 
-		if podAge > podScaleUpDelay {
+		// newPodScaleUpDelay <= 0 means that young pod filtering out is disabled. 
+		// As pod age is calculated from CA loop start time, it is technically possible that some pods will have negative
+		// age and would be unnecesarily skipped by the current loop.
+		if podScaleUpDelay <= 0 || podAge > podScaleUpDelay {
 			oldUnschedulablePods = append(oldUnschedulablePods, pod)
 		} else {
 			klog.V(3).Infof("Pod %s is %.3f seconds old, too new to consider unschedulable", pod.Name, podAge.Seconds())

cluster-autoscaler/core/static_autoscaler_test.go

@@ -2836,6 +2836,8 @@ func TestFilterOutYoungPods(t *testing.T) {
 	p4.Annotations = map[string]string{
 		annotations.PodScaleUpDelayAnnotationKey: "error",
 	}
+	p5 := BuildTestPod("p5", 500, 1000)
+	p5.CreationTimestamp = metav1.NewTime(now)
 
 	tests := []struct {
 		name               string
@@ -2882,6 +2884,20 @@ func TestFilterOutYoungPods(t *testing.T) {
 			expectedPods:       []*apiv1.Pod{p1, p4},
 			expectedError:      "Failed to parse pod",
 		},
+		{
+			name:               "future pods included when podScaleUpDelay is 0",
+			newPodScaleUpDelay: 0,
+			runTime:            now,
+			pods:               []*apiv1.Pod{p1, p5},
+			expectedPods:       []*apiv1.Pod{p1, p5},
+		},
+		{
+			name:               "future pods excluded when podScaleUpDelay is above 0",
+			newPodScaleUpDelay: 1 * time.Second,
+			runTime:            now,
+			pods:               []*apiv1.Pod{p1, p5},
+			expectedPods:       []*apiv1.Pod{p1},
+		},
 	}
 
 	for _, tt := range tests {