m/test/e2e: split out tests into subpackages
The end-to-end tests have grown large enough that they merit their own
test targets. To make this more Go-idiomatic, we split away the tests
not just int separate Bazel targets, but also Go packages.
We also add per-test resource requests for Bazel, including a new
resource kind (iops). This makes the tests more deterministic and allows
use to eg. use --runs_per_test=10 to deflake test logic without hitting
resource contention issues.
//metropolis/test/e2e/suites/core:core_test PASSED in 35.1s
Stats over 10 runs: max = 35.1s, min = 26.6s, avg = 31.9s, dev = 2.6s
//metropolis/test/e2e/suites/ha:ha_test PASSED in 114.6s
Stats over 10 runs: max = 114.6s, min = 90.1s, avg = 100.9s, dev = 7.6s
//metropolis/test/e2e/suites/ha_cold:ha_cold_test PASSED in 67.8s
Stats over 10 runs: max = 67.8s, min = 55.5s, avg = 62.0s, dev = 4.1s
//metropolis/test/e2e/suites/kubernetes:kubernetes_test PASSED in 80.9s
Stats over 10 runs: max = 80.9s, min = 58.8s, avg = 68.6s, dev = 6.0s
Change-Id: I8f31e09f599fd90c9941e2b69f36789817fa90ce
Reviewed-on: https://review.monogon.dev/c/monogon/+/3086
Reviewed-by: Jan Schär <jan@monogon.tech>
Tested-by: Jenkins CI
diff --git a/metropolis/test/e2e/suites/kubernetes/run_test.go b/metropolis/test/e2e/suites/kubernetes/run_test.go
new file mode 100644
index 0000000..ec38aa3
--- /dev/null
+++ b/metropolis/test/e2e/suites/kubernetes/run_test.go
@@ -0,0 +1,368 @@
+package kubernetes
+
+import (
+ "context"
+ "crypto/tls"
+ "crypto/x509"
+ "errors"
+ "fmt"
+ "io"
+ "net"
+ "net/http"
+ _ "net/http/pprof"
+ "net/url"
+ "os"
+ "strings"
+ "testing"
+ "time"
+
+ "github.com/bazelbuild/rules_go/go/runfiles"
+ corev1 "k8s.io/api/core/v1"
+ kerrors "k8s.io/apimachinery/pkg/api/errors"
+ "k8s.io/apimachinery/pkg/api/resource"
+ metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+ podv1 "k8s.io/kubernetes/pkg/api/v1/pod"
+
+ "source.monogon.dev/metropolis/pkg/localregistry"
+ "source.monogon.dev/metropolis/test/launch/cluster"
+ "source.monogon.dev/metropolis/test/util"
+
+ common "source.monogon.dev/metropolis/node"
+)
+
+const (
+ // Timeout for the global test context.
+ //
+ // Bazel would eventually time out the test after 900s ("large") if, for
+ // some reason, the context cancellation fails to abort it.
+ globalTestTimeout = 600 * time.Second
+
+ // Timeouts for individual end-to-end tests of different sizes.
+ smallTestTimeout = 60 * time.Second
+ largeTestTimeout = 120 * time.Second
+)
+
+// TestE2EKubernetes exercises the Kubernetes functionality of Metropolis.
+//
+// The tests are performed against an in-memory cluster.
+func TestE2EKubernetes(t *testing.T) {
+ // Set a global timeout to make sure this terminates
+ ctx, cancel := context.WithTimeout(context.Background(), globalTestTimeout)
+ defer cancel()
+
+ rPath, err := runfiles.Rlocation("_main/metropolis/test/e2e/testimages_manifest.prototxt")
+ if err != nil {
+ t.Fatalf("Resolving registry manifest failed: %v", err)
+ }
+ df, err := os.ReadFile(rPath)
+ if err != nil {
+ t.Fatalf("Reading registry manifest failed: %v", err)
+ }
+ lr, err := localregistry.FromBazelManifest(df)
+ if err != nil {
+ t.Fatalf("Creating test image registry failed: %v", err)
+ }
+
+ // Launch cluster.
+ clusterOptions := cluster.ClusterOptions{
+ NumNodes: 2,
+ LocalRegistry: lr,
+ }
+ cluster, err := cluster.LaunchCluster(ctx, clusterOptions)
+ if err != nil {
+ t.Fatalf("LaunchCluster failed: %v", err)
+ }
+ defer func() {
+ err := cluster.Close()
+ if err != nil {
+ t.Fatalf("cluster Close failed: %v", err)
+ }
+ }()
+
+ clientSet, err := cluster.GetKubeClientSet()
+ if err != nil {
+ t.Fatal(err)
+ }
+ util.TestEventual(t, "Add KubernetesWorker roles", ctx, smallTestTimeout, func(ctx context.Context) error {
+ // Make everything but the first node into KubernetesWorkers.
+ for i := 1; i < clusterOptions.NumNodes; i++ {
+ err := cluster.MakeKubernetesWorker(ctx, cluster.NodeIDs[i])
+ if err != nil {
+ return util.Permanent(fmt.Errorf("MakeKubernetesWorker: %w", err))
+ }
+ }
+ return nil
+ })
+ util.TestEventual(t, "Node is registered and ready", ctx, largeTestTimeout, func(ctx context.Context) error {
+ nodes, err := clientSet.CoreV1().Nodes().List(ctx, metav1.ListOptions{})
+ if err != nil {
+ return err
+ }
+ if len(nodes.Items) < 1 {
+ return errors.New("node not yet registered")
+ }
+ node := nodes.Items[0]
+ for _, cond := range node.Status.Conditions {
+ if cond.Type != corev1.NodeReady {
+ continue
+ }
+ if cond.Status != corev1.ConditionTrue {
+ return fmt.Errorf("node not ready: %v", cond.Message)
+ }
+ }
+ return nil
+ })
+ util.TestEventual(t, "Simple deployment", ctx, largeTestTimeout, func(ctx context.Context) error {
+ _, err := clientSet.AppsV1().Deployments("default").Create(ctx, makeTestDeploymentSpec("test-deploy-1"), metav1.CreateOptions{})
+ return err
+ })
+ util.TestEventual(t, "Simple deployment is running", ctx, largeTestTimeout, func(ctx context.Context) error {
+ res, err := clientSet.CoreV1().Pods("default").List(ctx, metav1.ListOptions{LabelSelector: "name=test-deploy-1"})
+ if err != nil {
+ return err
+ }
+ if len(res.Items) == 0 {
+ return errors.New("pod didn't get created")
+ }
+ pod := res.Items[0]
+ if podv1.IsPodAvailable(&pod, 1, metav1.NewTime(time.Now())) {
+ return nil
+ }
+ events, err := clientSet.CoreV1().Events("default").List(ctx, metav1.ListOptions{FieldSelector: fmt.Sprintf("involvedObject.name=%s,involvedObject.namespace=default", pod.Name)})
+ if err != nil || len(events.Items) == 0 {
+ return fmt.Errorf("pod is not ready: %v", pod.Status.Phase)
+ } else {
+ return fmt.Errorf("pod is not ready: %v", events.Items[0].Message)
+ }
+ })
+ util.TestEventual(t, "Simple deployment with gvisor", ctx, largeTestTimeout, func(ctx context.Context) error {
+ deployment := makeTestDeploymentSpec("test-deploy-2")
+ gvisorStr := "gvisor"
+ deployment.Spec.Template.Spec.RuntimeClassName = &gvisorStr
+ _, err := clientSet.AppsV1().Deployments("default").Create(ctx, deployment, metav1.CreateOptions{})
+ return err
+ })
+ util.TestEventual(t, "Simple deployment is running on gvisor", ctx, largeTestTimeout, func(ctx context.Context) error {
+ res, err := clientSet.CoreV1().Pods("default").List(ctx, metav1.ListOptions{LabelSelector: "name=test-deploy-2"})
+ if err != nil {
+ return err
+ }
+ if len(res.Items) == 0 {
+ return errors.New("pod didn't get created")
+ }
+ pod := res.Items[0]
+ if podv1.IsPodAvailable(&pod, 1, metav1.NewTime(time.Now())) {
+ return nil
+ }
+ events, err := clientSet.CoreV1().Events("default").List(ctx, metav1.ListOptions{FieldSelector: fmt.Sprintf("involvedObject.name=%s,involvedObject.namespace=default", pod.Name)})
+ if err != nil || len(events.Items) == 0 {
+ return fmt.Errorf("pod is not ready: %v", pod.Status.Phase)
+ } else {
+ var errorMsg strings.Builder
+ for _, msg := range events.Items {
+ errorMsg.WriteString(" | ")
+ errorMsg.WriteString(msg.Message)
+ }
+ return fmt.Errorf("pod is not ready: %v", errorMsg.String())
+ }
+ })
+ util.TestEventual(t, "Simple StatefulSet with PVC", ctx, largeTestTimeout, func(ctx context.Context) error {
+ _, err := clientSet.AppsV1().StatefulSets("default").Create(ctx, makeTestStatefulSet("test-statefulset-1", corev1.PersistentVolumeFilesystem), metav1.CreateOptions{})
+ return err
+ })
+ util.TestEventual(t, "Simple StatefulSet with PVC is running", ctx, largeTestTimeout, func(ctx context.Context) error {
+ res, err := clientSet.CoreV1().Pods("default").List(ctx, metav1.ListOptions{LabelSelector: "name=test-statefulset-1"})
+ if err != nil {
+ return err
+ }
+ if len(res.Items) == 0 {
+ return errors.New("pod didn't get created")
+ }
+ pod := res.Items[0]
+ if podv1.IsPodAvailable(&pod, 1, metav1.NewTime(time.Now())) {
+ return nil
+ }
+ events, err := clientSet.CoreV1().Events("default").List(ctx, metav1.ListOptions{FieldSelector: fmt.Sprintf("involvedObject.name=%s,involvedObject.namespace=default", pod.Name)})
+ if err != nil || len(events.Items) == 0 {
+ return fmt.Errorf("pod is not ready: %v", pod.Status.Phase)
+ } else {
+ return fmt.Errorf("pod is not ready: %v", events.Items[0].Message)
+ }
+ })
+ util.TestEventual(t, "Simple StatefulSet with Block PVC", ctx, largeTestTimeout, func(ctx context.Context) error {
+ _, err := clientSet.AppsV1().StatefulSets("default").Create(ctx, makeTestStatefulSet("test-statefulset-2", corev1.PersistentVolumeBlock), metav1.CreateOptions{})
+ return err
+ })
+ util.TestEventual(t, "Simple StatefulSet with Block PVC is running", ctx, largeTestTimeout, func(ctx context.Context) error {
+ res, err := clientSet.CoreV1().Pods("default").List(ctx, metav1.ListOptions{LabelSelector: "name=test-statefulset-2"})
+ if err != nil {
+ return err
+ }
+ if len(res.Items) == 0 {
+ return errors.New("pod didn't get created")
+ }
+ pod := res.Items[0]
+ if podv1.IsPodAvailable(&pod, 1, metav1.NewTime(time.Now())) {
+ return nil
+ }
+ events, err := clientSet.CoreV1().Events("default").List(ctx, metav1.ListOptions{FieldSelector: fmt.Sprintf("involvedObject.name=%s,involvedObject.namespace=default", pod.Name)})
+ if err != nil || len(events.Items) == 0 {
+ return fmt.Errorf("pod is not ready: %v", pod.Status.Phase)
+ } else {
+ return fmt.Errorf("pod is not ready: %v", events.Items[0].Message)
+ }
+ })
+ util.TestEventual(t, "In-cluster self-test job", ctx, smallTestTimeout, func(ctx context.Context) error {
+ _, err := clientSet.BatchV1().Jobs("default").Create(ctx, makeSelftestSpec("selftest"), metav1.CreateOptions{})
+ return err
+ })
+ util.TestEventual(t, "In-cluster self-test job passed", ctx, smallTestTimeout, func(ctx context.Context) error {
+ res, err := clientSet.BatchV1().Jobs("default").Get(ctx, "selftest", metav1.GetOptions{})
+ if err != nil {
+ return err
+ }
+ if res.Status.Failed > 0 {
+ pods, err := clientSet.CoreV1().Pods("default").List(ctx, metav1.ListOptions{
+ LabelSelector: "job-name=selftest",
+ })
+ if err != nil {
+ return util.Permanent(fmt.Errorf("job failed but failed to find pod: %w", err))
+ }
+ if len(pods.Items) < 1 {
+ return fmt.Errorf("job failed but pod does not exist")
+ }
+ lines, err := getPodLogLines(ctx, clientSet, pods.Items[0].Name, 1)
+ if err != nil {
+ return fmt.Errorf("job failed but could not get logs: %w", err)
+ }
+ if len(lines) > 0 {
+ return util.Permanent(fmt.Errorf("job failed, last log line: %s", lines[0]))
+ }
+ return util.Permanent(fmt.Errorf("job failed, empty log"))
+ }
+ if res.Status.Succeeded > 0 {
+ return nil
+ }
+ return fmt.Errorf("job still running")
+ })
+ util.TestEventual(t, "Start NodePort test setup", ctx, smallTestTimeout, func(ctx context.Context) error {
+ _, err := clientSet.AppsV1().Deployments("default").Create(ctx, makeHTTPServerDeploymentSpec("nodeport-server"), metav1.CreateOptions{})
+ if err != nil && !kerrors.IsAlreadyExists(err) {
+ return err
+ }
+ _, err = clientSet.CoreV1().Services("default").Create(ctx, makeHTTPServerNodePortService("nodeport-server"), metav1.CreateOptions{})
+ if err != nil && !kerrors.IsAlreadyExists(err) {
+ return err
+ }
+ return nil
+ })
+ util.TestEventual(t, "NodePort accessible from all nodes", ctx, smallTestTimeout, func(ctx context.Context) error {
+ nodes, err := clientSet.CoreV1().Nodes().List(ctx, metav1.ListOptions{})
+ if err != nil {
+ return err
+ }
+ // Use a new client for each attempt
+ hc := http.Client{
+ Timeout: 2 * time.Second,
+ Transport: &http.Transport{
+ Dial: cluster.SOCKSDialer.Dial,
+ },
+ }
+ for _, n := range nodes.Items {
+ var addr string
+ for _, a := range n.Status.Addresses {
+ if a.Type == corev1.NodeInternalIP {
+ addr = a.Address
+ }
+ }
+ u := url.URL{Scheme: "http", Host: addr, Path: "/"}
+ res, err := hc.Get(u.String())
+ if err != nil {
+ return fmt.Errorf("failed getting from node %q: %w", n.Name, err)
+ }
+ if res.StatusCode != http.StatusOK {
+ return fmt.Errorf("getting from node %q: HTTP %d", n.Name, res.StatusCode)
+ }
+ t.Logf("Got response from %q", n.Name)
+ }
+ return nil
+ })
+ util.TestEventual(t, "containerd metrics retrieved", ctx, smallTestTimeout, func(ctx context.Context) error {
+ pool := x509.NewCertPool()
+ pool.AddCert(cluster.CACertificate)
+ cl := http.Client{
+ Transport: &http.Transport{
+ TLSClientConfig: &tls.Config{
+ Certificates: []tls.Certificate{cluster.Owner},
+ RootCAs: pool,
+ },
+ DialContext: func(ctx context.Context, _, addr string) (net.Conn, error) {
+ return cluster.DialNode(ctx, addr)
+ },
+ },
+ }
+ u := url.URL{
+ Scheme: "https",
+ Host: net.JoinHostPort(cluster.NodeIDs[1], common.MetricsPort.PortString()),
+ Path: "/metrics/containerd",
+ }
+ res, err := cl.Get(u.String())
+ if err != nil {
+ return err
+ }
+ defer res.Body.Close()
+ if res.StatusCode != 200 {
+ return fmt.Errorf("status code %d", res.StatusCode)
+ }
+
+ body, err := io.ReadAll(res.Body)
+ if err != nil {
+ return err
+ }
+ needle := "containerd_build_info_total"
+ if !strings.Contains(string(body), needle) {
+ return util.Permanent(fmt.Errorf("could not find %q in returned response", needle))
+ }
+ return nil
+ })
+ if os.Getenv("HAVE_NESTED_KVM") != "" {
+ util.TestEventual(t, "Pod for KVM/QEMU smoke test", ctx, smallTestTimeout, func(ctx context.Context) error {
+ runcRuntimeClass := "runc"
+ _, err := clientSet.CoreV1().Pods("default").Create(ctx, &corev1.Pod{
+ ObjectMeta: metav1.ObjectMeta{
+ Name: "vm-smoketest",
+ },
+ Spec: corev1.PodSpec{
+ Containers: []corev1.Container{{
+ Name: "vm-smoketest",
+ ImagePullPolicy: corev1.PullNever,
+ Image: "test.monogon.internal/metropolis/vm/smoketest:smoketest_container",
+ Resources: corev1.ResourceRequirements{
+ Limits: corev1.ResourceList{
+ "devices.monogon.dev/kvm": *resource.NewQuantity(1, ""),
+ },
+ },
+ }},
+ RuntimeClassName: &runcRuntimeClass,
+ RestartPolicy: corev1.RestartPolicyNever,
+ },
+ }, metav1.CreateOptions{})
+ return err
+ })
+ util.TestEventual(t, "KVM/QEMU smoke test completion", ctx, smallTestTimeout, func(ctx context.Context) error {
+ pod, err := clientSet.CoreV1().Pods("default").Get(ctx, "vm-smoketest", metav1.GetOptions{})
+ if err != nil {
+ return fmt.Errorf("failed to get pod: %w", err)
+ }
+ if pod.Status.Phase == corev1.PodSucceeded {
+ return nil
+ }
+ events, err := clientSet.CoreV1().Events("default").List(ctx, metav1.ListOptions{FieldSelector: fmt.Sprintf("involvedObject.name=%s,involvedObject.namespace=default", pod.Name)})
+ if err != nil || len(events.Items) == 0 {
+ return fmt.Errorf("pod is not ready: %v", pod.Status.Phase)
+ } else {
+ return fmt.Errorf("pod is not ready: %v", events.Items[len(events.Items)-1].Message)
+ }
+ })
+ }
+}