Ensure probes from the autoscaler are not passed to the user-container (

#13506)

The autoscaler sends a probe to detect the activator is in the data
path. This probe was inadvertently passed to the user container which
was preventing scale to 0.

Co-authored-by: dprotaso <[email protected]>

cmd/queue/main.go

@@ -325,7 +325,7 @@ func buildServer(ctx context.Context, env config, probeContainer func() bool, st
 	drainer := &pkghandler.Drainer{
 		QuietPeriod: drainSleepDuration,
 		// Add Activator probe header to the drainer so it can handle probes directly from activator
-		HealthCheckUAPrefixes: []string{netheader.ActivatorUserAgent},
+		HealthCheckUAPrefixes: []string{netheader.ActivatorUserAgent, netheader.AutoscalingUserAgent},
 		Inner:                 composedHandler,
 		HealthCheck:           health.ProbeHandler(probeContainer, tracingEnabled),
 	}

test/e2e/autoscale_test.go

@@ -25,13 +25,15 @@ import (
 	"time"
 
 	"golang.org/x/sync/errgroup"
+	appsv1 "k8s.io/api/apps/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/util/wait"
-
 	netcfg "knative.dev/networking/pkg/config"
 	"knative.dev/pkg/system"
+	pkgtest "knative.dev/pkg/test"
 	"knative.dev/serving/pkg/apis/autoscaling"
 	"knative.dev/serving/pkg/networking"
+	revnames "knative.dev/serving/pkg/reconciler/revision/resources/names"
 	"knative.dev/serving/pkg/reconciler/serverlessservice/resources/names"
 	"knative.dev/serving/pkg/resources"
 	rtesting "knative.dev/serving/pkg/testing/v1"
@@ -224,6 +226,38 @@ func TestTargetBurstCapacityMinusOne(t *testing.T) {
 	}
 }
 
+// Explicitly setting this should cause the revision to scale down after ~10s
+func TestTargetBurstCapacityZero(t *testing.T) {
+	t.Parallel()
+
+	ctx := SetupSvc(t, autoscaling.KPA, autoscaling.Concurrency, 10 /* target concurrency*/, targetUtilization,
+		rtesting.WithConfigAnnotations(map[string]string{
+			autoscaling.TargetBurstCapacityKey: "0",
+			autoscaling.WindowAnnotationKey:    "10s", // scale faster
+		}))
+
+	test.EnsureTearDown(t, ctx.Clients(), ctx.Names())
+
+	deploymentName := revnames.Deployment(ctx.resources.Revision)
+
+	t.Log("waiting for scale down")
+	err := pkgtest.WaitForDeploymentState(
+		context.Background(),
+		ctx.Clients().KubeClient,
+		deploymentName,
+		func(d *appsv1.Deployment) (bool, error) {
+			return d.Status.ReadyReplicas == 0, nil
+		},
+		"DeploymentIsScaledDown",
+		test.ServingFlags.TestNamespace,
+		time.Minute,
+	)
+
+	if err != nil {
+		t.Error(err)
+	}
+}
+
 func TestFastScaleToZero(t *testing.T) {
 	t.Parallel()