Merge pull request #123343 from haircommander/image-gc-e2e-2

KEP-4210: add e2e tests and add small fix for ImageGCMaxAge

test/e2e_node/image_gc_test.go

@@ -0,0 +1,117 @@
+/*
+Copyright 2024 The Kubernetes Authors.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+package e2enode
+
+import (
+	"context"
+	"time"
+
+	v1 "k8s.io/api/core/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	internalapi "k8s.io/cri-api/pkg/apis"
+	runtimeapi "k8s.io/cri-api/pkg/apis/runtime/v1"
+	kubefeatures "k8s.io/kubernetes/pkg/features"
+	kubeletconfig "k8s.io/kubernetes/pkg/kubelet/apis/config"
+	"k8s.io/kubernetes/test/e2e/framework"
+	e2epod "k8s.io/kubernetes/test/e2e/framework/pod"
+	admissionapi "k8s.io/pod-security-admission/api"
+
+	"github.com/onsi/ginkgo/v2"
+	"github.com/onsi/gomega"
+)
+
+const (
+	// Kubelet GC's on a frequency of once every 5 minutes
+	// Add a little leeway to give it time.
+	checkGCUntil time.Duration = 6 * time.Minute
+	checkGCFreq  time.Duration = 30 * time.Second
+)
+
+var _ = SIGDescribe("ImageGarbageCollect", framework.WithSerial(), framework.WithNodeFeature("GarbageCollect"), func() {
+	f := framework.NewDefaultFramework("image-garbage-collect-test")
+	f.NamespacePodSecurityLevel = admissionapi.LevelPrivileged
+	var is internalapi.ImageManagerService
+	ginkgo.BeforeEach(func() {
+		var err error
+		_, is, err = getCRIClient()
+		framework.ExpectNoError(err)
+	})
+	ginkgo.AfterEach(func() {
+		framework.ExpectNoError(PrePullAllImages())
+	})
+	ginkgo.Context("when ImageMaximumGCAge is set", func() {
+		tempSetCurrentKubeletConfig(f, func(ctx context.Context, initialConfig *kubeletconfig.KubeletConfiguration) {
+			initialConfig.ImageMaximumGCAge = metav1.Duration{Duration: time.Duration(time.Minute * 1)}
+			initialConfig.ImageMinimumGCAge = metav1.Duration{Duration: time.Duration(time.Second * 1)}
+			if initialConfig.FeatureGates == nil {
+				initialConfig.FeatureGates = make(map[string]bool)
+			}
+			initialConfig.FeatureGates[string(kubefeatures.ImageMaximumGCAge)] = true
+		})
+		ginkgo.It("should GC unused images", func(ctx context.Context) {
+			pod := innocentPod()
+			e2epod.NewPodClient(f).CreateBatch(ctx, []*v1.Pod{pod})
+
+			_, err := is.PullImage(context.Background(), &runtimeapi.ImageSpec{Image: agnhostImage}, nil, nil)
+			framework.ExpectNoError(err)
+
+			allImages, err := is.ListImages(context.Background(), &runtimeapi.ImageFilter{})
+			framework.ExpectNoError(err)
+
+			e2epod.NewPodClient(f).DeleteSync(ctx, pod.ObjectMeta.Name, metav1.DeleteOptions{}, e2epod.DefaultPodDeletionTimeout)
+
+			// Even though the image gc max timing is less, we are bound by the kubelet's
+			// ImageGCPeriod, which is hardcoded to 5 minutes.
+			gomega.Eventually(ctx, func() int {
+				gcdImageList, err := is.ListImages(context.Background(), &runtimeapi.ImageFilter{})
+				framework.ExpectNoError(err)
+				return len(gcdImageList)
+			}, checkGCUntil, checkGCFreq).Should(gomega.BeNumerically("<", len(allImages)))
+		})
+		ginkgo.It("should not GC unused images prematurely", func(ctx context.Context) {
+			pod := innocentPod()
+			e2epod.NewPodClient(f).CreateBatch(ctx, []*v1.Pod{pod})
+
+			_, err := is.PullImage(context.Background(), &runtimeapi.ImageSpec{Image: agnhostImage}, nil, nil)
+			framework.ExpectNoError(err)
+
+			allImages, err := is.ListImages(context.Background(), &runtimeapi.ImageFilter{})
+			framework.ExpectNoError(err)
+
+			e2epod.NewPodClient(f).DeleteSync(ctx, pod.ObjectMeta.Name, metav1.DeleteOptions{}, e2epod.DefaultPodDeletionTimeout)
+
+			restartKubelet(true)
+			waitForKubeletToStart(ctx, f)
+
+			// Wait until the maxAge of the image after the kubelet is restarted to ensure it doesn't
+			// GC too early.
+			gomega.Consistently(ctx, func() int {
+				gcdImageList, err := is.ListImages(context.Background(), &runtimeapi.ImageFilter{})
+				framework.ExpectNoError(err)
+				return len(gcdImageList)
+			}, 50*time.Second, 10*time.Second).Should(gomega.Equal(len(allImages)))
+
+			// Even though the image gc max timing is less, we are bound by the kubelet's
+			// ImageGCPeriod, which is hardcoded to 5 minutes.
+			gomega.Eventually(ctx, func() int {
+				gcdImageList, err := is.ListImages(context.Background(), &runtimeapi.ImageFilter{})
+				framework.ExpectNoError(err)
+				return len(gcdImageList)
+			}, checkGCUntil, checkGCFreq).Should(gomega.BeNumerically("<", len(allImages)))
+		})
+	})
+})

test/e2e_node/util.go

@@ -74,6 +74,7 @@ import (
 var startServices = flag.Bool("start-services", true, "If true, start local node services")
 var stopServices = flag.Bool("stop-services", true, "If true, stop local node services after running tests")
 var busyboxImage = imageutils.GetE2EImage(imageutils.BusyBox)
+var agnhostImage = imageutils.GetE2EImage(imageutils.Agnhost)
 
 const (
 	// Kubelet internal cgroup name for node allocatable cgroup.
@@ -233,7 +234,10 @@ func updateKubeletConfig(ctx context.Context, f *framework.Framework, kubeletCon
 
 	ginkgo.By("Starting the kubelet")
 	startKubelet()
+	waitForKubeletToStart(ctx, f)
+}
 
+func waitForKubeletToStart(ctx context.Context, f *framework.Framework) {
 	// wait until the kubelet health check will succeed
 	gomega.Eventually(ctx, func() bool {
 		return kubeletHealthCheck(kubeletHealthCheckURL)