vendor files

This commit is contained in:
Serguei Bezverkhi
2018-01-09 13:57:14 -05:00
parent 558bc6c02a
commit 7b24313bd6
16547 changed files with 4527373 additions and 0 deletions

167
vendor/k8s.io/kubernetes/test/e2e/framework/BUILD generated vendored Normal file
View File

@ -0,0 +1,167 @@
package(default_visibility = ["//visibility:public"])
load(
"@io_bazel_rules_go//go:def.bzl",
"go_library",
)
go_library(
name = "go_default_library",
srcs = [
"authorizer_util.go",
"cleanup.go",
"deployment_util.go",
"exec_util.go",
"firewall_util.go",
"framework.go",
"get-kubemark-resource-usage.go",
"google_compute.go",
"gpu_util.go",
"ingress_utils.go",
"jobs_util.go",
"kubelet_stats.go",
"log_size_monitoring.go",
"metrics_util.go",
"networking_utils.go",
"nodes_util.go",
"perf_util.go",
"pods.go",
"psp_util.go",
"pv_util.go",
"rc_util.go",
"resource_usage_gatherer.go",
"rs_util.go",
"service_util.go",
"size.go",
"statefulset_utils.go",
"test_context.go",
"upgrade_util.go",
"util.go",
"volume_util.go",
],
importpath = "k8s.io/kubernetes/test/e2e/framework",
deps = [
"//pkg/api/legacyscheme:go_default_library",
"//pkg/api/testapi:go_default_library",
"//pkg/api/v1/pod:go_default_library",
"//pkg/apis/apps:go_default_library",
"//pkg/apis/batch:go_default_library",
"//pkg/apis/core:go_default_library",
"//pkg/apis/core/v1/helper:go_default_library",
"//pkg/apis/extensions:go_default_library",
"//pkg/client/clientset_generated/internalclientset:go_default_library",
"//pkg/client/conditions:go_default_library",
"//pkg/cloudprovider:go_default_library",
"//pkg/cloudprovider/providers/aws:go_default_library",
"//pkg/cloudprovider/providers/azure:go_default_library",
"//pkg/cloudprovider/providers/gce:go_default_library",
"//pkg/controller:go_default_library",
"//pkg/controller/deployment/util:go_default_library",
"//pkg/controller/node:go_default_library",
"//pkg/features:go_default_library",
"//pkg/kubectl:go_default_library",
"//pkg/kubelet/apis/kubeletconfig:go_default_library",
"//pkg/kubelet/apis/stats/v1alpha1:go_default_library",
"//pkg/kubelet/dockershim/metrics:go_default_library",
"//pkg/kubelet/events:go_default_library",
"//pkg/kubelet/metrics:go_default_library",
"//pkg/kubelet/sysctl:go_default_library",
"//pkg/kubelet/util/format:go_default_library",
"//pkg/kubemark:go_default_library",
"//pkg/master/ports:go_default_library",
"//pkg/security/podsecuritypolicy/seccomp:go_default_library",
"//pkg/ssh:go_default_library",
"//pkg/util/file:go_default_library",
"//pkg/util/system:go_default_library",
"//pkg/util/taints:go_default_library",
"//pkg/util/version:go_default_library",
"//pkg/volume/util/volumehelper:go_default_library",
"//plugin/pkg/scheduler/algorithm/predicates:go_default_library",
"//plugin/pkg/scheduler/schedulercache:go_default_library",
"//test/e2e/framework/ginkgowrapper:go_default_library",
"//test/e2e/framework/metrics:go_default_library",
"//test/e2e/manifest:go_default_library",
"//test/e2e/perftype:go_default_library",
"//test/utils:go_default_library",
"//test/utils/image:go_default_library",
"//vendor/github.com/aws/aws-sdk-go/aws:go_default_library",
"//vendor/github.com/aws/aws-sdk-go/aws/awserr:go_default_library",
"//vendor/github.com/aws/aws-sdk-go/aws/session:go_default_library",
"//vendor/github.com/aws/aws-sdk-go/service/autoscaling:go_default_library",
"//vendor/github.com/aws/aws-sdk-go/service/ec2:go_default_library",
"//vendor/github.com/golang/glog:go_default_library",
"//vendor/github.com/onsi/ginkgo:go_default_library",
"//vendor/github.com/onsi/ginkgo/config:go_default_library",
"//vendor/github.com/onsi/gomega:go_default_library",
"//vendor/github.com/onsi/gomega/types:go_default_library",
"//vendor/github.com/prometheus/common/expfmt:go_default_library",
"//vendor/github.com/prometheus/common/model:go_default_library",
"//vendor/github.com/spf13/viper:go_default_library",
"//vendor/golang.org/x/crypto/ssh:go_default_library",
"//vendor/golang.org/x/net/websocket:go_default_library",
"//vendor/google.golang.org/api/compute/v1:go_default_library",
"//vendor/google.golang.org/api/googleapi:go_default_library",
"//vendor/k8s.io/api/apps/v1beta1:go_default_library",
"//vendor/k8s.io/api/apps/v1beta2:go_default_library",
"//vendor/k8s.io/api/authorization/v1beta1:go_default_library",
"//vendor/k8s.io/api/batch/v1:go_default_library",
"//vendor/k8s.io/api/core/v1:go_default_library",
"//vendor/k8s.io/api/extensions/v1beta1:go_default_library",
"//vendor/k8s.io/api/policy/v1beta1:go_default_library",
"//vendor/k8s.io/api/rbac/v1beta1:go_default_library",
"//vendor/k8s.io/apimachinery/pkg/api/errors:go_default_library",
"//vendor/k8s.io/apimachinery/pkg/api/resource:go_default_library",
"//vendor/k8s.io/apimachinery/pkg/apis/meta/v1:go_default_library",
"//vendor/k8s.io/apimachinery/pkg/apis/meta/v1/unstructured:go_default_library",
"//vendor/k8s.io/apimachinery/pkg/fields:go_default_library",
"//vendor/k8s.io/apimachinery/pkg/labels:go_default_library",
"//vendor/k8s.io/apimachinery/pkg/runtime:go_default_library",
"//vendor/k8s.io/apimachinery/pkg/runtime/schema:go_default_library",
"//vendor/k8s.io/apimachinery/pkg/types:go_default_library",
"//vendor/k8s.io/apimachinery/pkg/util/errors:go_default_library",
"//vendor/k8s.io/apimachinery/pkg/util/intstr:go_default_library",
"//vendor/k8s.io/apimachinery/pkg/util/net:go_default_library",
"//vendor/k8s.io/apimachinery/pkg/util/rand:go_default_library",
"//vendor/k8s.io/apimachinery/pkg/util/runtime:go_default_library",
"//vendor/k8s.io/apimachinery/pkg/util/sets:go_default_library",
"//vendor/k8s.io/apimachinery/pkg/util/uuid:go_default_library",
"//vendor/k8s.io/apimachinery/pkg/util/wait:go_default_library",
"//vendor/k8s.io/apimachinery/pkg/util/yaml:go_default_library",
"//vendor/k8s.io/apimachinery/pkg/version:go_default_library",
"//vendor/k8s.io/apimachinery/pkg/watch:go_default_library",
"//vendor/k8s.io/apiserver/pkg/authentication/serviceaccount:go_default_library",
"//vendor/k8s.io/apiserver/pkg/util/feature:go_default_library",
"//vendor/k8s.io/client-go/discovery:go_default_library",
"//vendor/k8s.io/client-go/dynamic:go_default_library",
"//vendor/k8s.io/client-go/informers:go_default_library",
"//vendor/k8s.io/client-go/kubernetes:go_default_library",
"//vendor/k8s.io/client-go/kubernetes/typed/authorization/v1beta1:go_default_library",
"//vendor/k8s.io/client-go/kubernetes/typed/core/v1:go_default_library",
"//vendor/k8s.io/client-go/kubernetes/typed/rbac/v1beta1:go_default_library",
"//vendor/k8s.io/client-go/rest:go_default_library",
"//vendor/k8s.io/client-go/tools/clientcmd:go_default_library",
"//vendor/k8s.io/client-go/tools/clientcmd/api:go_default_library",
"//vendor/k8s.io/client-go/tools/remotecommand:go_default_library",
"//vendor/k8s.io/client-go/util/retry:go_default_library",
"//vendor/k8s.io/kube-aggregator/pkg/client/clientset_generated/clientset:go_default_library",
"//vendor/k8s.io/utils/exec:go_default_library",
],
)
filegroup(
name = "package-srcs",
srcs = glob(["**"]),
tags = ["automanaged"],
visibility = ["//visibility:private"],
)
filegroup(
name = "all-srcs",
srcs = [
":package-srcs",
"//test/e2e/framework/ginkgowrapper:all-srcs",
"//test/e2e/framework/metrics:all-srcs",
"//test/e2e/framework/timer:all-srcs",
],
tags = ["automanaged"],
)

View File

@ -0,0 +1,151 @@
/*
Copyright 2016 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 framework
import (
"fmt"
"sync"
"time"
authorizationv1beta1 "k8s.io/api/authorization/v1beta1"
rbacv1beta1 "k8s.io/api/rbac/v1beta1"
apierrors "k8s.io/apimachinery/pkg/api/errors"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/runtime/schema"
"k8s.io/apimachinery/pkg/util/wait"
v1beta1authorization "k8s.io/client-go/kubernetes/typed/authorization/v1beta1"
v1beta1rbac "k8s.io/client-go/kubernetes/typed/rbac/v1beta1"
)
const (
policyCachePollInterval = 100 * time.Millisecond
policyCachePollTimeout = 5 * time.Second
)
// WaitForAuthorizationUpdate checks if the given user can perform the named verb and action.
// If policyCachePollTimeout is reached without the expected condition matching, an error is returned
func WaitForAuthorizationUpdate(c v1beta1authorization.SubjectAccessReviewsGetter, user, namespace, verb string, resource schema.GroupResource, allowed bool) error {
return WaitForNamedAuthorizationUpdate(c, user, namespace, verb, "", resource, allowed)
}
// WaitForAuthorizationUpdate checks if the given user can perform the named verb and action on the named resource.
// If policyCachePollTimeout is reached without the expected condition matching, an error is returned
func WaitForNamedAuthorizationUpdate(c v1beta1authorization.SubjectAccessReviewsGetter, user, namespace, verb, resourceName string, resource schema.GroupResource, allowed bool) error {
review := &authorizationv1beta1.SubjectAccessReview{
Spec: authorizationv1beta1.SubjectAccessReviewSpec{
ResourceAttributes: &authorizationv1beta1.ResourceAttributes{
Group: resource.Group,
Verb: verb,
Resource: resource.Resource,
Namespace: namespace,
Name: resourceName,
},
User: user,
},
}
err := wait.Poll(policyCachePollInterval, policyCachePollTimeout, func() (bool, error) {
response, err := c.SubjectAccessReviews().Create(review)
// GKE doesn't enable the SAR endpoint. Without this endpoint, we cannot determine if the policy engine
// has adjusted as expected. In this case, simply wait one second and hope it's up to date
if apierrors.IsNotFound(err) {
fmt.Printf("SubjectAccessReview endpoint is missing\n")
time.Sleep(1 * time.Second)
return true, nil
}
if err != nil {
return false, err
}
if response.Status.Allowed != allowed {
return false, nil
}
return true, nil
})
return err
}
// BindClusterRole binds the cluster role at the cluster scope
func BindClusterRole(c v1beta1rbac.ClusterRoleBindingsGetter, clusterRole, ns string, subjects ...rbacv1beta1.Subject) {
// Since the namespace names are unique, we can leave this lying around so we don't have to race any caches
_, err := c.ClusterRoleBindings().Create(&rbacv1beta1.ClusterRoleBinding{
ObjectMeta: metav1.ObjectMeta{
Name: ns + "--" + clusterRole,
},
RoleRef: rbacv1beta1.RoleRef{
APIGroup: "rbac.authorization.k8s.io",
Kind: "ClusterRole",
Name: clusterRole,
},
Subjects: subjects,
})
// if we failed, don't fail the entire test because it may still work. RBAC may simply be disabled.
if err != nil {
fmt.Printf("Error binding clusterrole/%s for %q for %v\n", clusterRole, ns, subjects)
}
}
// BindClusterRoleInNamespace binds the cluster role at the namespace scope
func BindClusterRoleInNamespace(c v1beta1rbac.RoleBindingsGetter, clusterRole, ns string, subjects ...rbacv1beta1.Subject) {
bindInNamespace(c, "ClusterRole", clusterRole, ns, subjects...)
}
// BindRoleInNamespace binds the role at the namespace scope
func BindRoleInNamespace(c v1beta1rbac.RoleBindingsGetter, role, ns string, subjects ...rbacv1beta1.Subject) {
bindInNamespace(c, "Role", role, ns, subjects...)
}
func bindInNamespace(c v1beta1rbac.RoleBindingsGetter, roleType, role, ns string, subjects ...rbacv1beta1.Subject) {
// Since the namespace names are unique, we can leave this lying around so we don't have to race any caches
_, err := c.RoleBindings(ns).Create(&rbacv1beta1.RoleBinding{
ObjectMeta: metav1.ObjectMeta{
Name: ns + "--" + role,
},
RoleRef: rbacv1beta1.RoleRef{
APIGroup: "rbac.authorization.k8s.io",
Kind: roleType,
Name: role,
},
Subjects: subjects,
})
// if we failed, don't fail the entire test because it may still work. RBAC may simply be disabled.
if err != nil {
fmt.Printf("Error binding %s/%s into %q for %v\n", roleType, role, ns, subjects)
}
}
var (
isRBACEnabledOnce sync.Once
isRBACEnabled bool
)
func IsRBACEnabled(f *Framework) bool {
isRBACEnabledOnce.Do(func() {
crs, err := f.ClientSet.RbacV1().ClusterRoles().List(metav1.ListOptions{})
if err != nil {
Logf("Error listing ClusterRoles; assuming RBAC is disabled: %v", err)
isRBACEnabled = false
} else if crs == nil || len(crs.Items) == 0 {
Logf("No ClusteRoles found; assuming RBAC is disabled.")
isRBACEnabled = false
} else {
Logf("Found ClusterRoles; assuming RBAC is enabled.")
isRBACEnabled = true
}
})
return isRBACEnabled
}

61
vendor/k8s.io/kubernetes/test/e2e/framework/cleanup.go generated vendored Normal file
View File

@ -0,0 +1,61 @@
/*
Copyright 2016 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 framework
import "sync"
type CleanupActionHandle *int
var cleanupActionsLock sync.Mutex
var cleanupActions = map[CleanupActionHandle]func(){}
// AddCleanupAction installs a function that will be called in the event of the
// whole test being terminated. This allows arbitrary pieces of the overall
// test to hook into SynchronizedAfterSuite().
func AddCleanupAction(fn func()) CleanupActionHandle {
p := CleanupActionHandle(new(int))
cleanupActionsLock.Lock()
defer cleanupActionsLock.Unlock()
cleanupActions[p] = fn
return p
}
// RemoveCleanupAction removes a function that was installed by
// AddCleanupAction.
func RemoveCleanupAction(p CleanupActionHandle) {
cleanupActionsLock.Lock()
defer cleanupActionsLock.Unlock()
delete(cleanupActions, p)
}
// RunCleanupActions runs all functions installed by AddCleanupAction. It does
// not remove them (see RemoveCleanupAction) but it does run unlocked, so they
// may remove themselves.
func RunCleanupActions() {
list := []func(){}
func() {
cleanupActionsLock.Lock()
defer cleanupActionsLock.Unlock()
for _, fn := range cleanupActions {
list = append(list, fn)
}
}()
// Run unlocked.
for _, fn := range list {
fn()
}
}

View File

@ -0,0 +1,297 @@
/*
Copyright 2017 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 framework
import (
"fmt"
"time"
. "github.com/onsi/ginkgo"
"k8s.io/api/core/v1"
extensions "k8s.io/api/extensions/v1beta1"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/util/uuid"
"k8s.io/apimachinery/pkg/util/wait"
"k8s.io/apimachinery/pkg/watch"
clientset "k8s.io/client-go/kubernetes"
extensionsinternal "k8s.io/kubernetes/pkg/apis/extensions"
"k8s.io/kubernetes/pkg/client/clientset_generated/internalclientset"
deploymentutil "k8s.io/kubernetes/pkg/controller/deployment/util"
testutils "k8s.io/kubernetes/test/utils"
)
func UpdateDeploymentWithRetries(c clientset.Interface, namespace, name string, applyUpdate testutils.UpdateDeploymentFunc) (*extensions.Deployment, error) {
return testutils.UpdateDeploymentWithRetries(c, namespace, name, applyUpdate, Logf, Poll, pollShortTimeout)
}
// Waits for the deployment to clean up old rcs.
func WaitForDeploymentOldRSsNum(c clientset.Interface, ns, deploymentName string, desiredRSNum int) error {
var oldRSs []*extensions.ReplicaSet
var d *extensions.Deployment
pollErr := wait.PollImmediate(Poll, 5*time.Minute, func() (bool, error) {
deployment, err := c.ExtensionsV1beta1().Deployments(ns).Get(deploymentName, metav1.GetOptions{})
if err != nil {
return false, err
}
d = deployment
_, oldRSs, err = deploymentutil.GetOldReplicaSets(deployment, c.ExtensionsV1beta1())
if err != nil {
return false, err
}
return len(oldRSs) == desiredRSNum, nil
})
if pollErr == wait.ErrWaitTimeout {
pollErr = fmt.Errorf("%d old replica sets were not cleaned up for deployment %q", len(oldRSs)-desiredRSNum, deploymentName)
logReplicaSetsOfDeployment(d, oldRSs, nil)
}
return pollErr
}
func logReplicaSetsOfDeployment(deployment *extensions.Deployment, allOldRSs []*extensions.ReplicaSet, newRS *extensions.ReplicaSet) {
testutils.LogReplicaSetsOfDeployment(deployment, allOldRSs, newRS, Logf)
}
func WaitForObservedDeployment(c clientset.Interface, ns, deploymentName string, desiredGeneration int64) error {
return testutils.WaitForObservedDeployment(c, ns, deploymentName, desiredGeneration)
}
func WaitForDeploymentWithCondition(c clientset.Interface, ns, deploymentName, reason string, condType extensions.DeploymentConditionType) error {
return testutils.WaitForDeploymentWithCondition(c, ns, deploymentName, reason, condType, Logf, Poll, pollLongTimeout)
}
// WaitForDeploymentRevisionAndImage waits for the deployment's and its new RS's revision and container image to match the given revision and image.
// Note that deployment revision and its new RS revision should be updated shortly most of the time, but an overwhelmed RS controller
// may result in taking longer to relabel a RS.
func WaitForDeploymentRevisionAndImage(c clientset.Interface, ns, deploymentName string, revision, image string) error {
return testutils.WaitForDeploymentRevisionAndImage(c, ns, deploymentName, revision, image, Logf, Poll, pollLongTimeout)
}
func NewDeployment(deploymentName string, replicas int32, podLabels map[string]string, imageName, image string, strategyType extensions.DeploymentStrategyType) *extensions.Deployment {
zero := int64(0)
return &extensions.Deployment{
ObjectMeta: metav1.ObjectMeta{
Name: deploymentName,
},
Spec: extensions.DeploymentSpec{
Replicas: &replicas,
Selector: &metav1.LabelSelector{MatchLabels: podLabels},
Strategy: extensions.DeploymentStrategy{
Type: strategyType,
},
Template: v1.PodTemplateSpec{
ObjectMeta: metav1.ObjectMeta{
Labels: podLabels,
},
Spec: v1.PodSpec{
TerminationGracePeriodSeconds: &zero,
Containers: []v1.Container{
{
Name: imageName,
Image: image,
},
},
},
},
},
}
}
// Waits for the deployment to complete, and don't check if rolling update strategy is broken.
// Rolling update strategy is used only during a rolling update, and can be violated in other situations,
// such as shortly after a scaling event or the deployment is just created.
func WaitForDeploymentComplete(c clientset.Interface, d *extensions.Deployment) error {
return testutils.WaitForDeploymentComplete(c, d, Logf, Poll, pollLongTimeout)
}
// Waits for the deployment to complete, and check rolling update strategy isn't broken at any times.
// Rolling update strategy should not be broken during a rolling update.
func WaitForDeploymentCompleteAndCheckRolling(c clientset.Interface, d *extensions.Deployment) error {
return testutils.WaitForDeploymentCompleteAndCheckRolling(c, d, Logf, Poll, pollLongTimeout)
}
// WaitForDeploymentUpdatedReplicasLTE waits for given deployment to be observed by the controller and has at least a number of updatedReplicas
func WaitForDeploymentUpdatedReplicasLTE(c clientset.Interface, ns, deploymentName string, minUpdatedReplicas int32, desiredGeneration int64) error {
return testutils.WaitForDeploymentUpdatedReplicasLTE(c, ns, deploymentName, minUpdatedReplicas, desiredGeneration, Poll, pollLongTimeout)
}
// WaitForDeploymentRollbackCleared waits for given deployment either started rolling back or doesn't need to rollback.
// Note that rollback should be cleared shortly, so we only wait for 1 minute here to fail early.
func WaitForDeploymentRollbackCleared(c clientset.Interface, ns, deploymentName string) error {
return testutils.WaitForDeploymentRollbackCleared(c, ns, deploymentName, Poll, pollShortTimeout)
}
// WatchRecreateDeployment watches Recreate deployments and ensures no new pods will run at the same time with
// old pods.
func WatchRecreateDeployment(c clientset.Interface, d *extensions.Deployment) error {
if d.Spec.Strategy.Type != extensions.RecreateDeploymentStrategyType {
return fmt.Errorf("deployment %q does not use a Recreate strategy: %s", d.Name, d.Spec.Strategy.Type)
}
w, err := c.ExtensionsV1beta1().Deployments(d.Namespace).Watch(metav1.SingleObject(metav1.ObjectMeta{Name: d.Name, ResourceVersion: d.ResourceVersion}))
if err != nil {
return err
}
status := d.Status
condition := func(event watch.Event) (bool, error) {
d := event.Object.(*extensions.Deployment)
status = d.Status
if d.Status.UpdatedReplicas > 0 && d.Status.Replicas != d.Status.UpdatedReplicas {
_, allOldRSs, err := deploymentutil.GetOldReplicaSets(d, c.ExtensionsV1beta1())
newRS, nerr := deploymentutil.GetNewReplicaSet(d, c.ExtensionsV1beta1())
if err == nil && nerr == nil {
Logf("%+v", d)
logReplicaSetsOfDeployment(d, allOldRSs, newRS)
logPodsOfDeployment(c, d, append(allOldRSs, newRS))
}
return false, fmt.Errorf("deployment %q is running new pods alongside old pods: %#v", d.Name, status)
}
return *(d.Spec.Replicas) == d.Status.Replicas &&
*(d.Spec.Replicas) == d.Status.UpdatedReplicas &&
d.Generation <= d.Status.ObservedGeneration, nil
}
_, err = watch.Until(2*time.Minute, w, condition)
if err == wait.ErrWaitTimeout {
err = fmt.Errorf("deployment %q never completed: %#v", d.Name, status)
}
return err
}
func ScaleDeployment(clientset clientset.Interface, internalClientset internalclientset.Interface, ns, name string, size uint, wait bool) error {
return ScaleResource(clientset, internalClientset, ns, name, size, wait, extensionsinternal.Kind("Deployment"))
}
func RunDeployment(config testutils.DeploymentConfig) error {
By(fmt.Sprintf("creating deployment %s in namespace %s", config.Name, config.Namespace))
config.NodeDumpFunc = DumpNodeDebugInfo
config.ContainerDumpFunc = LogFailedContainers
return testutils.RunDeployment(config)
}
func logPodsOfDeployment(c clientset.Interface, deployment *extensions.Deployment, rsList []*extensions.ReplicaSet) {
testutils.LogPodsOfDeployment(c, deployment, rsList, Logf)
}
func WaitForDeploymentRevision(c clientset.Interface, d *extensions.Deployment, targetRevision string) error {
err := wait.PollImmediate(Poll, pollLongTimeout, func() (bool, error) {
deployment, err := c.ExtensionsV1beta1().Deployments(d.Namespace).Get(d.Name, metav1.GetOptions{})
if err != nil {
return false, err
}
revision := deployment.Annotations[deploymentutil.RevisionAnnotation]
return revision == targetRevision, nil
})
if err != nil {
return fmt.Errorf("error waiting for revision to become %q for deployment %q: %v", targetRevision, d.Name, err)
}
return nil
}
// CheckDeploymentRevisionAndImage checks if the input deployment's and its new replica set's revision and image are as expected.
func CheckDeploymentRevisionAndImage(c clientset.Interface, ns, deploymentName, revision, image string) error {
return testutils.CheckDeploymentRevisionAndImage(c, ns, deploymentName, revision, image)
}
func CreateDeployment(client clientset.Interface, replicas int32, podLabels map[string]string, namespace string, pvclaims []*v1.PersistentVolumeClaim, command string) (*extensions.Deployment, error) {
deploymentSpec := MakeDeployment(replicas, podLabels, namespace, pvclaims, false, command)
deployment, err := client.Extensions().Deployments(namespace).Create(deploymentSpec)
if err != nil {
return nil, fmt.Errorf("deployment %q Create API error: %v", deploymentSpec.Name, err)
}
Logf("Waiting deployment %q to complete", deploymentSpec.Name)
err = WaitForDeploymentComplete(client, deployment)
if err != nil {
return nil, fmt.Errorf("deployment %q failed to complete: %v", deploymentSpec.Name, err)
}
return deployment, nil
}
// MakeDeployment creates a deployment definition based on the namespace. The deployment references the PVC's
// name. A slice of BASH commands can be supplied as args to be run by the pod
func MakeDeployment(replicas int32, podLabels map[string]string, namespace string, pvclaims []*v1.PersistentVolumeClaim, isPrivileged bool, command string) *extensions.Deployment {
if len(command) == 0 {
command = "while true; do sleep 1; done"
}
zero := int64(0)
deploymentName := "deployment-" + string(uuid.NewUUID())
deploymentSpec := &extensions.Deployment{
ObjectMeta: metav1.ObjectMeta{
Name: deploymentName,
Namespace: namespace,
},
Spec: extensions.DeploymentSpec{
Replicas: &replicas,
Template: v1.PodTemplateSpec{
ObjectMeta: metav1.ObjectMeta{
Labels: podLabels,
},
Spec: v1.PodSpec{
TerminationGracePeriodSeconds: &zero,
Containers: []v1.Container{
{
Name: "write-pod",
Image: "busybox",
Command: []string{"/bin/sh"},
Args: []string{"-c", command},
SecurityContext: &v1.SecurityContext{
Privileged: &isPrivileged,
},
},
},
RestartPolicy: v1.RestartPolicyAlways,
},
},
},
}
var volumeMounts = make([]v1.VolumeMount, len(pvclaims))
var volumes = make([]v1.Volume, len(pvclaims))
for index, pvclaim := range pvclaims {
volumename := fmt.Sprintf("volume%v", index+1)
volumeMounts[index] = v1.VolumeMount{Name: volumename, MountPath: "/mnt/" + volumename}
volumes[index] = v1.Volume{Name: volumename, VolumeSource: v1.VolumeSource{PersistentVolumeClaim: &v1.PersistentVolumeClaimVolumeSource{ClaimName: pvclaim.Name, ReadOnly: false}}}
}
deploymentSpec.Spec.Template.Spec.Containers[0].VolumeMounts = volumeMounts
deploymentSpec.Spec.Template.Spec.Volumes = volumes
return deploymentSpec
}
// GetPodsForDeployment gets pods for the given deployment
func GetPodsForDeployment(client clientset.Interface, deployment *extensions.Deployment) (*v1.PodList, error) {
replicaSet, err := deploymentutil.GetNewReplicaSet(deployment, client.ExtensionsV1beta1())
if err != nil {
return nil, fmt.Errorf("Failed to get new replica set for deployment %q: %v", deployment.Name, err)
}
if replicaSet == nil {
return nil, fmt.Errorf("expected a new replica set for deployment %q, found none", deployment.Name)
}
podListFunc := func(namespace string, options metav1.ListOptions) (*v1.PodList, error) {
return client.Core().Pods(namespace).List(options)
}
rsList := []*extensions.ReplicaSet{replicaSet}
podList, err := deploymentutil.ListPods(deployment, rsList, podListFunc)
if err != nil {
return nil, fmt.Errorf("Failed to list Pods of Deployment %q: %v", deployment.Name, err)
}
return podList, nil
}

View File

@ -0,0 +1,147 @@
/*
Copyright 2016 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 framework
import (
"bytes"
"io"
"net/url"
"strings"
"k8s.io/api/core/v1"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
restclient "k8s.io/client-go/rest"
"k8s.io/client-go/tools/remotecommand"
"k8s.io/kubernetes/pkg/api/legacyscheme"
. "github.com/onsi/gomega"
)
// ExecOptions passed to ExecWithOptions
type ExecOptions struct {
Command []string
Namespace string
PodName string
ContainerName string
Stdin io.Reader
CaptureStdout bool
CaptureStderr bool
// If false, whitespace in std{err,out} will be removed.
PreserveWhitespace bool
}
// ExecWithOptions executes a command in the specified container,
// returning stdout, stderr and error. `options` allowed for
// additional parameters to be passed.
func (f *Framework) ExecWithOptions(options ExecOptions) (string, string, error) {
Logf("ExecWithOptions %+v", options)
config, err := LoadConfig()
Expect(err).NotTo(HaveOccurred(), "failed to load restclient config")
const tty = false
req := f.ClientSet.CoreV1().RESTClient().Post().
Resource("pods").
Name(options.PodName).
Namespace(options.Namespace).
SubResource("exec").
Param("container", options.ContainerName)
req.VersionedParams(&v1.PodExecOptions{
Container: options.ContainerName,
Command: options.Command,
Stdin: options.Stdin != nil,
Stdout: options.CaptureStdout,
Stderr: options.CaptureStderr,
TTY: tty,
}, legacyscheme.ParameterCodec)
var stdout, stderr bytes.Buffer
err = execute("POST", req.URL(), config, options.Stdin, &stdout, &stderr, tty)
if options.PreserveWhitespace {
return stdout.String(), stderr.String(), err
}
return strings.TrimSpace(stdout.String()), strings.TrimSpace(stderr.String()), err
}
// ExecCommandInContainerWithFullOutput executes a command in the
// specified container and return stdout, stderr and error
func (f *Framework) ExecCommandInContainerWithFullOutput(podName, containerName string, cmd ...string) (string, string, error) {
return f.ExecWithOptions(ExecOptions{
Command: cmd,
Namespace: f.Namespace.Name,
PodName: podName,
ContainerName: containerName,
Stdin: nil,
CaptureStdout: true,
CaptureStderr: true,
PreserveWhitespace: false,
})
}
// ExecCommandInContainer executes a command in the specified container.
func (f *Framework) ExecCommandInContainer(podName, containerName string, cmd ...string) string {
stdout, stderr, err := f.ExecCommandInContainerWithFullOutput(podName, containerName, cmd...)
Logf("Exec stderr: %q", stderr)
Expect(err).NotTo(HaveOccurred(),
"failed to execute command in pod %v, container %v: %v",
podName, containerName, err)
return stdout
}
func (f *Framework) ExecShellInContainer(podName, containerName string, cmd string) string {
return f.ExecCommandInContainer(podName, containerName, "/bin/sh", "-c", cmd)
}
func (f *Framework) ExecCommandInPod(podName string, cmd ...string) string {
pod, err := f.PodClient().Get(podName, metav1.GetOptions{})
Expect(err).NotTo(HaveOccurred(), "failed to get pod")
Expect(pod.Spec.Containers).NotTo(BeEmpty())
return f.ExecCommandInContainer(podName, pod.Spec.Containers[0].Name, cmd...)
}
func (f *Framework) ExecCommandInPodWithFullOutput(podName string, cmd ...string) (string, string, error) {
pod, err := f.PodClient().Get(podName, metav1.GetOptions{})
Expect(err).NotTo(HaveOccurred(), "failed to get pod")
Expect(pod.Spec.Containers).NotTo(BeEmpty())
return f.ExecCommandInContainerWithFullOutput(podName, pod.Spec.Containers[0].Name, cmd...)
}
func (f *Framework) ExecShellInPod(podName string, cmd string) string {
return f.ExecCommandInPod(podName, "/bin/sh", "-c", cmd)
}
func (f *Framework) ExecShellInPodWithFullOutput(podName string, cmd string) (string, string, error) {
return f.ExecCommandInPodWithFullOutput(podName, "/bin/sh", "-c", cmd)
}
func execute(method string, url *url.URL, config *restclient.Config, stdin io.Reader, stdout, stderr io.Writer, tty bool) error {
exec, err := remotecommand.NewSPDYExecutor(config, method, url)
if err != nil {
return err
}
return exec.Stream(remotecommand.StreamOptions{
Stdin: stdin,
Stdout: stdout,
Stderr: stderr,
Tty: tty,
})
}

View File

@ -0,0 +1,389 @@
/*
Copyright 2016 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 framework
import (
"fmt"
"net/http"
"strconv"
"strings"
"time"
"k8s.io/api/core/v1"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/util/sets"
"k8s.io/apimachinery/pkg/util/wait"
clientset "k8s.io/client-go/kubernetes"
"k8s.io/kubernetes/pkg/cloudprovider"
gcecloud "k8s.io/kubernetes/pkg/cloudprovider/providers/gce"
. "github.com/onsi/gomega"
compute "google.golang.org/api/compute/v1"
)
const (
FirewallTimeoutDefault = 3 * time.Minute
FirewallTestTcpTimeout = time.Duration(1 * time.Second)
// Set ports outside of 30000-32767, 80 and 8080 to avoid being whitelisted by the e2e cluster
FirewallTestHttpPort = int32(29999)
FirewallTestUdpPort = int32(29998)
)
// MakeFirewallNameForLBService return the expected firewall name for a LB service.
// This should match the formatting of makeFirewallName() in pkg/cloudprovider/providers/gce/gce_loadbalancer.go
func MakeFirewallNameForLBService(name string) string {
return fmt.Sprintf("k8s-fw-%s", name)
}
// ConstructFirewallForLBService returns the expected GCE firewall rule for a loadbalancer type service
func ConstructFirewallForLBService(svc *v1.Service, nodeTag string) *compute.Firewall {
if svc.Spec.Type != v1.ServiceTypeLoadBalancer {
Failf("can not construct firewall rule for non-loadbalancer type service")
}
fw := compute.Firewall{}
fw.Name = MakeFirewallNameForLBService(cloudprovider.GetLoadBalancerName(svc))
fw.TargetTags = []string{nodeTag}
if svc.Spec.LoadBalancerSourceRanges == nil {
fw.SourceRanges = []string{"0.0.0.0/0"}
} else {
fw.SourceRanges = svc.Spec.LoadBalancerSourceRanges
}
for _, sp := range svc.Spec.Ports {
fw.Allowed = append(fw.Allowed, &compute.FirewallAllowed{
IPProtocol: strings.ToLower(string(sp.Protocol)),
Ports: []string{strconv.Itoa(int(sp.Port))},
})
}
return &fw
}
func MakeHealthCheckFirewallNameForLBService(clusterID, name string, isNodesHealthCheck bool) string {
return gcecloud.MakeHealthCheckFirewallName(clusterID, name, isNodesHealthCheck)
}
// ConstructHealthCheckFirewallForLBService returns the expected GCE firewall rule for a loadbalancer type service
func ConstructHealthCheckFirewallForLBService(clusterID string, svc *v1.Service, nodeTag string, isNodesHealthCheck bool) *compute.Firewall {
if svc.Spec.Type != v1.ServiceTypeLoadBalancer {
Failf("can not construct firewall rule for non-loadbalancer type service")
}
fw := compute.Firewall{}
fw.Name = MakeHealthCheckFirewallNameForLBService(clusterID, cloudprovider.GetLoadBalancerName(svc), isNodesHealthCheck)
fw.TargetTags = []string{nodeTag}
fw.SourceRanges = gcecloud.LoadBalancerSrcRanges()
healthCheckPort := gcecloud.GetNodesHealthCheckPort()
if !isNodesHealthCheck {
healthCheckPort = svc.Spec.HealthCheckNodePort
}
fw.Allowed = []*compute.FirewallAllowed{
{
IPProtocol: "tcp",
Ports: []string{fmt.Sprintf("%d", healthCheckPort)},
},
}
return &fw
}
// GetInstanceTags gets tags from GCE instance with given name.
func GetInstanceTags(cloudConfig CloudConfig, instanceName string) *compute.Tags {
gceCloud := cloudConfig.Provider.(*gcecloud.GCECloud)
res, err := gceCloud.GetComputeService().Instances.Get(cloudConfig.ProjectID, cloudConfig.Zone,
instanceName).Do()
if err != nil {
Failf("Failed to get instance tags for %v: %v", instanceName, err)
}
return res.Tags
}
// SetInstanceTags sets tags on GCE instance with given name.
func SetInstanceTags(cloudConfig CloudConfig, instanceName, zone string, tags []string) []string {
gceCloud := cloudConfig.Provider.(*gcecloud.GCECloud)
// Re-get instance everytime because we need the latest fingerprint for updating metadata
resTags := GetInstanceTags(cloudConfig, instanceName)
_, err := gceCloud.GetComputeService().Instances.SetTags(
cloudConfig.ProjectID, zone, instanceName,
&compute.Tags{Fingerprint: resTags.Fingerprint, Items: tags}).Do()
if err != nil {
Failf("failed to set instance tags: %v", err)
}
Logf("Sent request to set tags %v on instance: %v", tags, instanceName)
return resTags.Items
}
// GetNodeTags gets k8s node tag from one of the nodes
func GetNodeTags(c clientset.Interface, cloudConfig CloudConfig) []string {
nodes := GetReadySchedulableNodesOrDie(c)
if len(nodes.Items) == 0 {
Logf("GetNodeTags: Found 0 node.")
return []string{}
}
return GetInstanceTags(cloudConfig, nodes.Items[0].Name).Items
}
// GetInstancePrefix returns the INSTANCE_PREFIX env we set for e2e cluster.
// From cluster/gce/config-test.sh, master name is set up using below format:
// MASTER_NAME="${INSTANCE_PREFIX}-master"
func GetInstancePrefix(masterName string) (string, error) {
if !strings.HasSuffix(masterName, "-master") {
return "", fmt.Errorf("unexpected master name format: %v", masterName)
}
return masterName[:len(masterName)-7], nil
}
// GetClusterName returns the CLUSTER_NAME env we set for e2e cluster.
// From cluster/gce/config-test.sh, cluster name is set up using below format:
// CLUSTER_NAME="${CLUSTER_NAME:-${INSTANCE_PREFIX}}"
func GetClusterName(instancePrefix string) string {
return instancePrefix
}
// GetE2eFirewalls returns all firewall rules we create for an e2e cluster.
// From cluster/gce/util.sh, all firewall rules should be consistent with the ones created by startup scripts.
func GetE2eFirewalls(masterName, masterTag, nodeTag, network, clusterIpRange string) []*compute.Firewall {
instancePrefix, err := GetInstancePrefix(masterName)
Expect(err).NotTo(HaveOccurred())
clusterName := GetClusterName(instancePrefix)
fws := []*compute.Firewall{}
fws = append(fws, &compute.Firewall{
Name: clusterName + "-default-internal-master",
SourceRanges: []string{"10.0.0.0/8"},
TargetTags: []string{masterTag},
Allowed: []*compute.FirewallAllowed{
{
IPProtocol: "tcp",
Ports: []string{"1-2379"},
},
{
IPProtocol: "tcp",
Ports: []string{"2382-65535"},
},
{
IPProtocol: "udp",
Ports: []string{"1-65535"},
},
{
IPProtocol: "icmp",
},
},
})
fws = append(fws, &compute.Firewall{
Name: clusterName + "-default-internal-node",
SourceRanges: []string{"10.0.0.0/8"},
TargetTags: []string{nodeTag},
Allowed: []*compute.FirewallAllowed{
{
IPProtocol: "tcp",
Ports: []string{"1-65535"},
},
{
IPProtocol: "udp",
Ports: []string{"1-65535"},
},
{
IPProtocol: "icmp",
},
},
})
fws = append(fws, &compute.Firewall{
Name: network + "-default-ssh",
SourceRanges: []string{"0.0.0.0/0"},
Allowed: []*compute.FirewallAllowed{
{
IPProtocol: "tcp",
Ports: []string{"22"},
},
},
})
fws = append(fws, &compute.Firewall{
Name: masterName + "-etcd",
SourceTags: []string{masterTag},
TargetTags: []string{masterTag},
Allowed: []*compute.FirewallAllowed{
{
IPProtocol: "tcp",
Ports: []string{"2380"},
},
{
IPProtocol: "tcp",
Ports: []string{"2381"},
},
},
})
fws = append(fws, &compute.Firewall{
Name: masterName + "-https",
SourceRanges: []string{"0.0.0.0/0"},
TargetTags: []string{masterTag},
Allowed: []*compute.FirewallAllowed{
{
IPProtocol: "tcp",
Ports: []string{"443"},
},
},
})
fws = append(fws, &compute.Firewall{
Name: nodeTag + "-all",
SourceRanges: []string{clusterIpRange},
TargetTags: []string{nodeTag},
Allowed: []*compute.FirewallAllowed{
{
IPProtocol: "tcp",
},
{
IPProtocol: "udp",
},
{
IPProtocol: "icmp",
},
{
IPProtocol: "esp",
},
{
IPProtocol: "ah",
},
{
IPProtocol: "sctp",
},
},
})
fws = append(fws, &compute.Firewall{
Name: nodeTag + "-" + instancePrefix + "-http-alt",
SourceRanges: []string{"0.0.0.0/0"},
TargetTags: []string{nodeTag},
Allowed: []*compute.FirewallAllowed{
{
IPProtocol: "tcp",
Ports: []string{"80"},
},
{
IPProtocol: "tcp",
Ports: []string{"8080"},
},
},
})
fws = append(fws, &compute.Firewall{
Name: nodeTag + "-" + instancePrefix + "-nodeports",
SourceRanges: []string{"0.0.0.0/0"},
TargetTags: []string{nodeTag},
Allowed: []*compute.FirewallAllowed{
{
IPProtocol: "tcp",
Ports: []string{"30000-32767"},
},
{
IPProtocol: "udp",
Ports: []string{"30000-32767"},
},
},
})
return fws
}
// PackProtocolsPortsFromFirewall packs protocols and ports in an unified way for verification.
func PackProtocolsPortsFromFirewall(alloweds []*compute.FirewallAllowed) []string {
protocolPorts := []string{}
for _, allowed := range alloweds {
for _, port := range allowed.Ports {
protocolPorts = append(protocolPorts, strings.ToLower(allowed.IPProtocol+"/"+port))
}
}
return protocolPorts
}
// SameStringArray verifies whether two string arrays have the same strings, return error if not.
// Order does not matter.
// When `include` is set to true, verifies whether result includes all elements from expected.
func SameStringArray(result, expected []string, include bool) error {
res := sets.NewString(result...)
exp := sets.NewString(expected...)
if !include {
diff := res.Difference(exp)
if len(diff) != 0 {
return fmt.Errorf("found differences: %v", diff)
}
} else {
if !res.IsSuperset(exp) {
return fmt.Errorf("some elements are missing: expected %v, got %v", expected, result)
}
}
return nil
}
// VerifyFirewallRule verifies whether the result firewall is consistent with the expected firewall.
// When `portsSubset` is false, match given ports exactly. Otherwise, only check ports are included.
func VerifyFirewallRule(res, exp *compute.Firewall, network string, portsSubset bool) error {
if res == nil || exp == nil {
return fmt.Errorf("res and exp must not be nil")
}
if res.Name != exp.Name {
return fmt.Errorf("incorrect name: %v, expected %v", res.Name, exp.Name)
}
// Sample Network value: https://www.googleapis.com/compute/v1/projects/{project-id}/global/networks/e2e
if !strings.HasSuffix(res.Network, "/"+network) {
return fmt.Errorf("incorrect network: %v, expected ends with: %v", res.Network, "/"+network)
}
if err := SameStringArray(PackProtocolsPortsFromFirewall(res.Allowed),
PackProtocolsPortsFromFirewall(exp.Allowed), portsSubset); err != nil {
return fmt.Errorf("incorrect allowed protocols ports: %v", err)
}
if err := SameStringArray(res.SourceRanges, exp.SourceRanges, false); err != nil {
return fmt.Errorf("incorrect source ranges %v, expected %v: %v", res.SourceRanges, exp.SourceRanges, err)
}
if err := SameStringArray(res.SourceTags, exp.SourceTags, false); err != nil {
return fmt.Errorf("incorrect source tags %v, expected %v: %v", res.SourceTags, exp.SourceTags, err)
}
if err := SameStringArray(res.TargetTags, exp.TargetTags, false); err != nil {
return fmt.Errorf("incorrect target tags %v, expected %v: %v", res.TargetTags, exp.TargetTags, err)
}
return nil
}
func WaitForFirewallRule(gceCloud *gcecloud.GCECloud, fwName string, exist bool, timeout time.Duration) (*compute.Firewall, error) {
Logf("Waiting up to %v for firewall %v exist=%v", timeout, fwName, exist)
var fw *compute.Firewall
var err error
condition := func() (bool, error) {
fw, err = gceCloud.GetFirewall(fwName)
if err != nil && exist ||
err == nil && !exist ||
err != nil && !exist && !IsGoogleAPIHTTPErrorCode(err, http.StatusNotFound) {
return false, nil
}
return true, nil
}
if err := wait.PollImmediate(5*time.Second, timeout, condition); err != nil {
return nil, fmt.Errorf("error waiting for firewall %v exist=%v", fwName, exist)
}
return fw, nil
}
func GetClusterID(c clientset.Interface) (string, error) {
cm, err := c.CoreV1().ConfigMaps(metav1.NamespaceSystem).Get(gcecloud.UIDConfigMapName, metav1.GetOptions{})
if err != nil || cm == nil {
return "", fmt.Errorf("error getting cluster ID: %v", err)
}
clusterID, clusterIDExists := cm.Data[gcecloud.UIDCluster]
providerID, providerIDExists := cm.Data[gcecloud.UIDProvider]
if !clusterIDExists {
return "", fmt.Errorf("cluster ID not set")
}
if providerIDExists {
return providerID, nil
}
return clusterID, nil
}

View File

@ -0,0 +1,835 @@
/*
Copyright 2015 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 framework
import (
"bufio"
"bytes"
"fmt"
"os"
"path"
"strings"
"sync"
"time"
"k8s.io/api/core/v1"
apierrors "k8s.io/apimachinery/pkg/api/errors"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/labels"
"k8s.io/apimachinery/pkg/runtime/schema"
"k8s.io/apimachinery/pkg/util/intstr"
"k8s.io/apimachinery/pkg/util/wait"
"k8s.io/client-go/dynamic"
"k8s.io/client-go/informers"
clientset "k8s.io/client-go/kubernetes"
"k8s.io/client-go/tools/clientcmd"
aggregatorclient "k8s.io/kube-aggregator/pkg/client/clientset_generated/clientset"
"k8s.io/kubernetes/pkg/api/legacyscheme"
"k8s.io/kubernetes/pkg/client/clientset_generated/internalclientset"
"k8s.io/kubernetes/pkg/kubemark"
"k8s.io/kubernetes/test/e2e/framework/metrics"
testutils "k8s.io/kubernetes/test/utils"
. "github.com/onsi/ginkgo"
. "github.com/onsi/gomega"
)
const (
maxKubectlExecRetries = 5
// TODO(mikedanese): reset this to 5 minutes once #47135 is resolved.
// ref https://github.com/kubernetes/kubernetes/issues/47135
DefaultNamespaceDeletionTimeout = 10 * time.Minute
)
// Framework supports common operations used by e2e tests; it will keep a client & a namespace for you.
// Eventual goal is to merge this with integration test framework.
type Framework struct {
BaseName string
ClientSet clientset.Interface
KubemarkExternalClusterClientSet clientset.Interface
InternalClientset *internalclientset.Clientset
AggregatorClient *aggregatorclient.Clientset
ClientPool dynamic.ClientPool
SkipNamespaceCreation bool // Whether to skip creating a namespace
Namespace *v1.Namespace // Every test has at least one namespace unless creation is skipped
namespacesToDelete []*v1.Namespace // Some tests have more than one.
NamespaceDeletionTimeout time.Duration
SkipPrivilegedPSPBinding bool // Whether to skip creating a binding to the privileged PSP in the test namespace
gatherer *containerResourceGatherer
// Constraints that passed to a check which is executed after data is gathered to
// see if 99% of results are within acceptable bounds. It has to be injected in the test,
// as expectations vary greatly. Constraints are grouped by the container names.
AddonResourceConstraints map[string]ResourceConstraint
logsSizeWaitGroup sync.WaitGroup
logsSizeCloseChannel chan bool
logsSizeVerifier *LogsSizeVerifier
// To make sure that this framework cleans up after itself, no matter what,
// we install a Cleanup action before each test and clear it after. If we
// should abort, the AfterSuite hook should run all Cleanup actions.
cleanupHandle CleanupActionHandle
// configuration for framework's client
Options FrameworkOptions
// Place where various additional data is stored during test run to be printed to ReportDir,
// or stdout if ReportDir is not set once test ends.
TestSummaries []TestDataSummary
kubemarkControllerCloseChannel chan struct{}
// Place to keep ClusterAutoscaler metrics from before test in order to compute delta.
clusterAutoscalerMetricsBeforeTest metrics.MetricsCollection
}
type TestDataSummary interface {
SummaryKind() string
PrintHumanReadable() string
PrintJSON() string
}
type FrameworkOptions struct {
ClientQPS float32
ClientBurst int
GroupVersion *schema.GroupVersion
}
// NewFramework makes a new framework and sets up a BeforeEach/AfterEach for
// you (you can write additional before/after each functions).
func NewDefaultFramework(baseName string) *Framework {
options := FrameworkOptions{
ClientQPS: 20,
ClientBurst: 50,
}
return NewFramework(baseName, options, nil)
}
func NewFramework(baseName string, options FrameworkOptions, client clientset.Interface) *Framework {
f := &Framework{
BaseName: baseName,
AddonResourceConstraints: make(map[string]ResourceConstraint),
Options: options,
ClientSet: client,
}
BeforeEach(f.BeforeEach)
AfterEach(f.AfterEach)
return f
}
// BeforeEach gets a client and makes a namespace.
func (f *Framework) BeforeEach() {
// The fact that we need this feels like a bug in ginkgo.
// https://github.com/onsi/ginkgo/issues/222
f.cleanupHandle = AddCleanupAction(f.AfterEach)
if f.ClientSet == nil {
By("Creating a kubernetes client")
config, err := LoadConfig()
Expect(err).NotTo(HaveOccurred())
config.QPS = f.Options.ClientQPS
config.Burst = f.Options.ClientBurst
if f.Options.GroupVersion != nil {
config.GroupVersion = f.Options.GroupVersion
}
if TestContext.KubeAPIContentType != "" {
config.ContentType = TestContext.KubeAPIContentType
}
f.ClientSet, err = clientset.NewForConfig(config)
Expect(err).NotTo(HaveOccurred())
f.InternalClientset, err = internalclientset.NewForConfig(config)
Expect(err).NotTo(HaveOccurred())
f.AggregatorClient, err = aggregatorclient.NewForConfig(config)
Expect(err).NotTo(HaveOccurred())
f.ClientPool = dynamic.NewClientPool(config, legacyscheme.Registry.RESTMapper(), dynamic.LegacyAPIPathResolverFunc)
if ProviderIs("kubemark") && TestContext.KubemarkExternalKubeConfig != "" && TestContext.CloudConfig.KubemarkController == nil {
externalConfig, err := clientcmd.BuildConfigFromFlags("", TestContext.KubemarkExternalKubeConfig)
externalConfig.QPS = f.Options.ClientQPS
externalConfig.Burst = f.Options.ClientBurst
Expect(err).NotTo(HaveOccurred())
externalClient, err := clientset.NewForConfig(externalConfig)
Expect(err).NotTo(HaveOccurred())
f.KubemarkExternalClusterClientSet = externalClient
f.kubemarkControllerCloseChannel = make(chan struct{})
externalInformerFactory := informers.NewSharedInformerFactory(externalClient, 0)
kubemarkInformerFactory := informers.NewSharedInformerFactory(f.ClientSet, 0)
kubemarkNodeInformer := kubemarkInformerFactory.Core().V1().Nodes()
go kubemarkNodeInformer.Informer().Run(f.kubemarkControllerCloseChannel)
TestContext.CloudConfig.KubemarkController, err = kubemark.NewKubemarkController(f.KubemarkExternalClusterClientSet, externalInformerFactory, f.ClientSet, kubemarkNodeInformer)
Expect(err).NotTo(HaveOccurred())
externalInformerFactory.Start(f.kubemarkControllerCloseChannel)
Expect(TestContext.CloudConfig.KubemarkController.WaitForCacheSync(f.kubemarkControllerCloseChannel)).To(BeTrue())
go TestContext.CloudConfig.KubemarkController.Run(f.kubemarkControllerCloseChannel)
}
}
if !f.SkipNamespaceCreation {
By("Building a namespace api object")
namespace, err := f.CreateNamespace(f.BaseName, map[string]string{
"e2e-framework": f.BaseName,
})
Expect(err).NotTo(HaveOccurred())
f.Namespace = namespace
if TestContext.VerifyServiceAccount {
By("Waiting for a default service account to be provisioned in namespace")
err = WaitForDefaultServiceAccountInNamespace(f.ClientSet, namespace.Name)
Expect(err).NotTo(HaveOccurred())
} else {
Logf("Skipping waiting for service account")
}
}
if TestContext.GatherKubeSystemResourceUsageData != "false" && TestContext.GatherKubeSystemResourceUsageData != "none" {
var err error
f.gatherer, err = NewResourceUsageGatherer(f.ClientSet, ResourceGathererOptions{
InKubemark: ProviderIs("kubemark"),
MasterOnly: TestContext.GatherKubeSystemResourceUsageData == "master",
ResourceDataGatheringPeriod: 60 * time.Second,
ProbeDuration: 15 * time.Second,
PrintVerboseLogs: false,
}, nil)
if err != nil {
Logf("Error while creating NewResourceUsageGatherer: %v", err)
} else {
go f.gatherer.StartGatheringData()
}
}
if TestContext.GatherLogsSizes {
f.logsSizeWaitGroup = sync.WaitGroup{}
f.logsSizeWaitGroup.Add(1)
f.logsSizeCloseChannel = make(chan bool)
f.logsSizeVerifier = NewLogsVerifier(f.ClientSet, f.logsSizeCloseChannel)
go func() {
f.logsSizeVerifier.Run()
f.logsSizeWaitGroup.Done()
}()
}
gatherMetricsAfterTest := TestContext.GatherMetricsAfterTest == "true" || TestContext.GatherMetricsAfterTest == "master"
if gatherMetricsAfterTest && TestContext.IncludeClusterAutoscalerMetrics {
grabber, err := metrics.NewMetricsGrabber(f.ClientSet, f.KubemarkExternalClusterClientSet, !ProviderIs("kubemark"), false, false, false, TestContext.IncludeClusterAutoscalerMetrics)
if err != nil {
Logf("Failed to create MetricsGrabber (skipping ClusterAutoscaler metrics gathering before test): %v", err)
} else {
f.clusterAutoscalerMetricsBeforeTest, err = grabber.Grab()
if err != nil {
Logf("MetricsGrabber failed to grab CA metrics before test (skipping metrics gathering): %v", err)
} else {
Logf("Gathered ClusterAutoscaler metrics before test")
}
}
}
}
// AfterEach deletes the namespace, after reading its events.
func (f *Framework) AfterEach() {
RemoveCleanupAction(f.cleanupHandle)
// DeleteNamespace at the very end in defer, to avoid any
// expectation failures preventing deleting the namespace.
defer func() {
nsDeletionErrors := map[string]error{}
// Whether to delete namespace is determined by 3 factors: delete-namespace flag, delete-namespace-on-failure flag and the test result
// if delete-namespace set to false, namespace will always be preserved.
// if delete-namespace is true and delete-namespace-on-failure is false, namespace will be preserved if test failed.
if TestContext.DeleteNamespace && (TestContext.DeleteNamespaceOnFailure || !CurrentGinkgoTestDescription().Failed) {
for _, ns := range f.namespacesToDelete {
By(fmt.Sprintf("Destroying namespace %q for this suite.", ns.Name))
timeout := DefaultNamespaceDeletionTimeout
if f.NamespaceDeletionTimeout != 0 {
timeout = f.NamespaceDeletionTimeout
}
if err := deleteNS(f.ClientSet, f.ClientPool, ns.Name, timeout); err != nil {
if !apierrors.IsNotFound(err) {
nsDeletionErrors[ns.Name] = err
} else {
Logf("Namespace %v was already deleted", ns.Name)
}
}
}
} else {
if !TestContext.DeleteNamespace {
Logf("Found DeleteNamespace=false, skipping namespace deletion!")
} else {
Logf("Found DeleteNamespaceOnFailure=false and current test failed, skipping namespace deletion!")
}
}
// Paranoia-- prevent reuse!
f.Namespace = nil
f.ClientSet = nil
f.namespacesToDelete = nil
// if we had errors deleting, report them now.
if len(nsDeletionErrors) != 0 {
messages := []string{}
for namespaceKey, namespaceErr := range nsDeletionErrors {
messages = append(messages, fmt.Sprintf("Couldn't delete ns: %q: %s (%#v)", namespaceKey, namespaceErr, namespaceErr))
}
Failf(strings.Join(messages, ","))
}
}()
// Print events if the test failed.
if CurrentGinkgoTestDescription().Failed && TestContext.DumpLogsOnFailure {
// Pass both unversioned client and and versioned clientset, till we have removed all uses of the unversioned client.
if !f.SkipNamespaceCreation {
DumpAllNamespaceInfo(f.ClientSet, f.Namespace.Name)
}
logFunc := Logf
if TestContext.ReportDir != "" {
filePath := path.Join(TestContext.ReportDir, "image-puller.txt")
file, err := os.Create(filePath)
if err != nil {
By(fmt.Sprintf("Failed to create a file with image-puller data %v: %v\nPrinting to stdout", filePath, err))
} else {
By(fmt.Sprintf("Dumping a list of prepulled images on each node to file %v", filePath))
defer file.Close()
if err = file.Chmod(0644); err != nil {
Logf("Failed to chmod to 644 of %v: %v", filePath, err)
}
logFunc = GetLogToFileFunc(file)
}
} else {
By("Dumping a list of prepulled images on each node...")
}
LogContainersInPodsWithLabels(f.ClientSet, metav1.NamespaceSystem, ImagePullerLabels, "image-puller", logFunc)
}
if TestContext.GatherKubeSystemResourceUsageData != "false" && TestContext.GatherKubeSystemResourceUsageData != "none" && f.gatherer != nil {
By("Collecting resource usage data")
summary, resourceViolationError := f.gatherer.StopAndSummarize([]int{90, 99, 100}, f.AddonResourceConstraints)
defer ExpectNoError(resourceViolationError)
f.TestSummaries = append(f.TestSummaries, summary)
}
if TestContext.GatherLogsSizes {
By("Gathering log sizes data")
close(f.logsSizeCloseChannel)
f.logsSizeWaitGroup.Wait()
f.TestSummaries = append(f.TestSummaries, f.logsSizeVerifier.GetSummary())
}
if TestContext.GatherMetricsAfterTest != "false" {
By("Gathering metrics")
// Grab apiserver, scheduler, controller-manager metrics and (optionally) nodes' kubelet metrics.
grabMetricsFromKubelets := TestContext.GatherMetricsAfterTest != "master" && !ProviderIs("kubemark")
grabber, err := metrics.NewMetricsGrabber(f.ClientSet, f.KubemarkExternalClusterClientSet, grabMetricsFromKubelets, true, true, true, TestContext.IncludeClusterAutoscalerMetrics)
if err != nil {
Logf("Failed to create MetricsGrabber (skipping metrics gathering): %v", err)
} else {
received, err := grabber.Grab()
if err != nil {
Logf("MetricsGrabber failed to grab some of the metrics: %v", err)
}
(*MetricsForE2E)(&received).computeClusterAutoscalerMetricsDelta(f.clusterAutoscalerMetricsBeforeTest)
f.TestSummaries = append(f.TestSummaries, (*MetricsForE2E)(&received))
}
}
if TestContext.CloudConfig.KubemarkController != nil {
close(f.kubemarkControllerCloseChannel)
}
PrintSummaries(f.TestSummaries, f.BaseName)
// Check whether all nodes are ready after the test.
// This is explicitly done at the very end of the test, to avoid
// e.g. not removing namespace in case of this failure.
if err := AllNodesReady(f.ClientSet, 3*time.Minute); err != nil {
Failf("All nodes should be ready after test, %v", err)
}
}
func (f *Framework) CreateNamespace(baseName string, labels map[string]string) (*v1.Namespace, error) {
createTestingNS := TestContext.CreateTestingNS
if createTestingNS == nil {
createTestingNS = CreateTestingNS
}
ns, err := createTestingNS(baseName, f.ClientSet, labels)
// check ns instead of err to see if it's nil as we may
// fail to create serviceAccount in it.
// In this case, we should not forget to delete the namespace.
if ns != nil {
f.namespacesToDelete = append(f.namespacesToDelete, ns)
}
if !f.SkipPrivilegedPSPBinding {
CreatePrivilegedPSPBinding(f, ns.Name)
}
return ns, err
}
// WaitForPodTerminated waits for the pod to be terminated with the given reason.
func (f *Framework) WaitForPodTerminated(podName, reason string) error {
return waitForPodTerminatedInNamespace(f.ClientSet, podName, reason, f.Namespace.Name)
}
// WaitForPodNotFound waits for the pod to be completely terminated (not "Get-able").
func (f *Framework) WaitForPodNotFound(podName string, timeout time.Duration) error {
return waitForPodNotFoundInNamespace(f.ClientSet, podName, f.Namespace.Name, timeout)
}
// WaitForPodRunning waits for the pod to run in the namespace.
func (f *Framework) WaitForPodRunning(podName string) error {
return WaitForPodNameRunningInNamespace(f.ClientSet, podName, f.Namespace.Name)
}
// WaitForPodReady waits for the pod to flip to ready in the namespace.
func (f *Framework) WaitForPodReady(podName string) error {
return waitTimeoutForPodReadyInNamespace(f.ClientSet, podName, f.Namespace.Name, PodStartTimeout)
}
// WaitForPodRunningSlow waits for the pod to run in the namespace.
// It has a longer timeout then WaitForPodRunning (util.slowPodStartTimeout).
func (f *Framework) WaitForPodRunningSlow(podName string) error {
return waitForPodRunningInNamespaceSlow(f.ClientSet, podName, f.Namespace.Name)
}
// WaitForPodNoLongerRunning waits for the pod to no longer be running in the namespace, for either
// success or failure.
func (f *Framework) WaitForPodNoLongerRunning(podName string) error {
return WaitForPodNoLongerRunningInNamespace(f.ClientSet, podName, f.Namespace.Name)
}
// TestContainerOutput runs the given pod in the given namespace and waits
// for all of the containers in the podSpec to move into the 'Success' status, and tests
// the specified container log against the given expected output using a substring matcher.
func (f *Framework) TestContainerOutput(scenarioName string, pod *v1.Pod, containerIndex int, expectedOutput []string) {
f.testContainerOutputMatcher(scenarioName, pod, containerIndex, expectedOutput, ContainSubstring)
}
// TestContainerOutputRegexp runs the given pod in the given namespace and waits
// for all of the containers in the podSpec to move into the 'Success' status, and tests
// the specified container log against the given expected output using a regexp matcher.
func (f *Framework) TestContainerOutputRegexp(scenarioName string, pod *v1.Pod, containerIndex int, expectedOutput []string) {
f.testContainerOutputMatcher(scenarioName, pod, containerIndex, expectedOutput, MatchRegexp)
}
// Write a file using kubectl exec echo <contents> > <path> via specified container
// Because of the primitive technique we're using here, we only allow ASCII alphanumeric characters
func (f *Framework) WriteFileViaContainer(podName, containerName string, path string, contents string) error {
By("writing a file in the container")
allowedCharacters := "0123456789abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ"
for _, c := range contents {
if !strings.ContainsRune(allowedCharacters, c) {
return fmt.Errorf("Unsupported character in string to write: %v", c)
}
}
command := fmt.Sprintf("echo '%s' > '%s'", contents, path)
stdout, stderr, err := kubectlExecWithRetry(f.Namespace.Name, podName, containerName, "--", "/bin/sh", "-c", command)
if err != nil {
Logf("error running kubectl exec to write file: %v\nstdout=%v\nstderr=%v)", err, string(stdout), string(stderr))
}
return err
}
// Read a file using kubectl exec cat <path>
func (f *Framework) ReadFileViaContainer(podName, containerName string, path string) (string, error) {
By("reading a file in the container")
stdout, stderr, err := kubectlExecWithRetry(f.Namespace.Name, podName, containerName, "--", "cat", path)
if err != nil {
Logf("error running kubectl exec to read file: %v\nstdout=%v\nstderr=%v)", err, string(stdout), string(stderr))
}
return string(stdout), err
}
func (f *Framework) CheckFileSizeViaContainer(podName, containerName, path string) (string, error) {
By("checking a file size in the container")
stdout, stderr, err := kubectlExecWithRetry(f.Namespace.Name, podName, containerName, "--", "ls", "-l", path)
if err != nil {
Logf("error running kubectl exec to read file: %v\nstdout=%v\nstderr=%v)", err, string(stdout), string(stderr))
}
return string(stdout), err
}
// CreateServiceForSimpleAppWithPods is a convenience wrapper to create a service and its matching pods all at once.
func (f *Framework) CreateServiceForSimpleAppWithPods(contPort int, svcPort int, appName string, podSpec func(n v1.Node) v1.PodSpec, count int, block bool) (error, *v1.Service) {
var err error = nil
theService := f.CreateServiceForSimpleApp(contPort, svcPort, appName)
f.CreatePodsPerNodeForSimpleApp(appName, podSpec, count)
if block {
err = testutils.WaitForPodsWithLabelRunning(f.ClientSet, f.Namespace.Name, labels.SelectorFromSet(labels.Set(theService.Spec.Selector)))
}
return err, theService
}
// CreateServiceForSimpleApp returns a service that selects/exposes pods (send -1 ports if no exposure needed) with an app label.
func (f *Framework) CreateServiceForSimpleApp(contPort, svcPort int, appName string) *v1.Service {
if appName == "" {
panic(fmt.Sprintf("no app name provided"))
}
serviceSelector := map[string]string{
"app": appName + "-pod",
}
// For convenience, user sending ports are optional.
portsFunc := func() []v1.ServicePort {
if contPort < 1 || svcPort < 1 {
return nil
} else {
return []v1.ServicePort{{
Protocol: "TCP",
Port: int32(svcPort),
TargetPort: intstr.FromInt(contPort),
}}
}
}
Logf("Creating a service-for-%v for selecting app=%v-pod", appName, appName)
service, err := f.ClientSet.CoreV1().Services(f.Namespace.Name).Create(&v1.Service{
ObjectMeta: metav1.ObjectMeta{
Name: "service-for-" + appName,
Labels: map[string]string{
"app": appName + "-service",
},
},
Spec: v1.ServiceSpec{
Ports: portsFunc(),
Selector: serviceSelector,
},
})
ExpectNoError(err)
return service
}
// CreatePodsPerNodeForSimpleApp Creates pods w/ labels. Useful for tests which make a bunch of pods w/o any networking.
func (f *Framework) CreatePodsPerNodeForSimpleApp(appName string, podSpec func(n v1.Node) v1.PodSpec, maxCount int) map[string]string {
nodes := GetReadySchedulableNodesOrDie(f.ClientSet)
labels := map[string]string{
"app": appName + "-pod",
}
for i, node := range nodes.Items {
// one per node, but no more than maxCount.
if i <= maxCount {
Logf("%v/%v : Creating container with label app=%v-pod", i, maxCount, appName)
_, err := f.ClientSet.CoreV1().Pods(f.Namespace.Name).Create(&v1.Pod{
ObjectMeta: metav1.ObjectMeta{
Name: fmt.Sprintf(appName+"-pod-%v", i),
Labels: labels,
},
Spec: podSpec(node),
})
ExpectNoError(err)
}
}
return labels
}
type KubeUser struct {
Name string `yaml:"name"`
User struct {
Username string `yaml:"username"`
Password string `yaml:"password"`
Token string `yaml:"token"`
} `yaml:"user"`
}
type KubeCluster struct {
Name string `yaml:"name"`
Cluster struct {
CertificateAuthorityData string `yaml:"certificate-authority-data"`
Server string `yaml:"server"`
} `yaml:"cluster"`
}
type KubeConfig struct {
Contexts []struct {
Name string `yaml:"name"`
Context struct {
Cluster string `yaml:"cluster"`
User string
} `yaml:"context"`
} `yaml:"contexts"`
Clusters []KubeCluster `yaml:"clusters"`
Users []KubeUser `yaml:"users"`
}
func (kc *KubeConfig) FindUser(name string) *KubeUser {
for _, user := range kc.Users {
if user.Name == name {
return &user
}
}
return nil
}
func (kc *KubeConfig) FindCluster(name string) *KubeCluster {
for _, cluster := range kc.Clusters {
if cluster.Name == name {
return &cluster
}
}
return nil
}
func kubectlExecWithRetry(namespace string, podName, containerName string, args ...string) ([]byte, []byte, error) {
for numRetries := 0; numRetries < maxKubectlExecRetries; numRetries++ {
if numRetries > 0 {
Logf("Retrying kubectl exec (retry count=%v/%v)", numRetries+1, maxKubectlExecRetries)
}
stdOutBytes, stdErrBytes, err := kubectlExec(namespace, podName, containerName, args...)
if err != nil {
if strings.Contains(strings.ToLower(string(stdErrBytes)), "i/o timeout") {
// Retry on "i/o timeout" errors
Logf("Warning: kubectl exec encountered i/o timeout.\nerr=%v\nstdout=%v\nstderr=%v)", err, string(stdOutBytes), string(stdErrBytes))
continue
}
if strings.Contains(strings.ToLower(string(stdErrBytes)), "container not found") {
// Retry on "container not found" errors
Logf("Warning: kubectl exec encountered container not found.\nerr=%v\nstdout=%v\nstderr=%v)", err, string(stdOutBytes), string(stdErrBytes))
time.Sleep(2 * time.Second)
continue
}
}
return stdOutBytes, stdErrBytes, err
}
err := fmt.Errorf("Failed: kubectl exec failed %d times with \"i/o timeout\". Giving up.", maxKubectlExecRetries)
return nil, nil, err
}
func kubectlExec(namespace string, podName, containerName string, args ...string) ([]byte, []byte, error) {
var stdout, stderr bytes.Buffer
cmdArgs := []string{
"exec",
fmt.Sprintf("--namespace=%v", namespace),
podName,
fmt.Sprintf("-c=%v", containerName),
}
cmdArgs = append(cmdArgs, args...)
cmd := KubectlCmd(cmdArgs...)
cmd.Stdout, cmd.Stderr = &stdout, &stderr
Logf("Running '%s %s'", cmd.Path, strings.Join(cmdArgs, " "))
err := cmd.Run()
return stdout.Bytes(), stderr.Bytes(), err
}
// Wrapper function for ginkgo describe. Adds namespacing.
// TODO: Support type safe tagging as well https://github.com/kubernetes/kubernetes/pull/22401.
func KubeDescribe(text string, body func()) bool {
return Describe("[k8s.io] "+text, body)
}
// Wrapper function for ginkgo It. Adds "[Conformance]" tag and makes static analysis easier.
func ConformanceIt(text string, body interface{}, timeout ...float64) bool {
return It(text+" [Conformance]", body, timeout...)
}
// PodStateVerification represents a verification of pod state.
// Any time you have a set of pods that you want to operate against or query,
// this struct can be used to declaratively identify those pods.
type PodStateVerification struct {
// Optional: only pods that have k=v labels will pass this filter.
Selectors map[string]string
// Required: The phases which are valid for your pod.
ValidPhases []v1.PodPhase
// Optional: only pods passing this function will pass the filter
// Verify a pod.
// As an optimization, in addition to specfying filter (boolean),
// this function allows specifying an error as well.
// The error indicates that the polling of the pod spectrum should stop.
Verify func(v1.Pod) (bool, error)
// Optional: only pods with this name will pass the filter.
PodName string
}
type ClusterVerification struct {
client clientset.Interface
namespace *v1.Namespace // pointer rather than string, since ns isn't created until before each.
podState PodStateVerification
}
func (f *Framework) NewClusterVerification(namespace *v1.Namespace, filter PodStateVerification) *ClusterVerification {
return &ClusterVerification{
f.ClientSet,
namespace,
filter,
}
}
func passesPodNameFilter(pod v1.Pod, name string) bool {
return name == "" || strings.Contains(pod.Name, name)
}
func passesVerifyFilter(pod v1.Pod, verify func(p v1.Pod) (bool, error)) (bool, error) {
if verify == nil {
return true, nil
} else {
verified, err := verify(pod)
// If an error is returned, by definition, pod verification fails
if err != nil {
return false, err
} else {
return verified, nil
}
}
}
func passesPhasesFilter(pod v1.Pod, validPhases []v1.PodPhase) bool {
passesPhaseFilter := false
for _, phase := range validPhases {
if pod.Status.Phase == phase {
passesPhaseFilter = true
}
}
return passesPhaseFilter
}
// filterLabels returns a list of pods which have labels.
func filterLabels(selectors map[string]string, cli clientset.Interface, ns string) (*v1.PodList, error) {
var err error
var selector labels.Selector
var pl *v1.PodList
// List pods based on selectors. This might be a tiny optimization rather then filtering
// everything manually.
if len(selectors) > 0 {
selector = labels.SelectorFromSet(labels.Set(selectors))
options := metav1.ListOptions{LabelSelector: selector.String()}
pl, err = cli.CoreV1().Pods(ns).List(options)
} else {
pl, err = cli.CoreV1().Pods(ns).List(metav1.ListOptions{})
}
return pl, err
}
// filter filters pods which pass a filter. It can be used to compose
// the more useful abstractions like ForEach, WaitFor, and so on, which
// can be used directly by tests.
func (p *PodStateVerification) filter(c clientset.Interface, namespace *v1.Namespace) ([]v1.Pod, error) {
if len(p.ValidPhases) == 0 || namespace == nil {
panic(fmt.Errorf("Need to specify a valid pod phases (%v) and namespace (%v). ", p.ValidPhases, namespace))
}
ns := namespace.Name
pl, err := filterLabels(p.Selectors, c, ns) // Build an v1.PodList to operate against.
Logf("Selector matched %v pods for %v", len(pl.Items), p.Selectors)
if len(pl.Items) == 0 || err != nil {
return pl.Items, err
}
unfilteredPods := pl.Items
filteredPods := []v1.Pod{}
ReturnPodsSoFar:
// Next: Pod must match at least one of the states that the user specified
for _, pod := range unfilteredPods {
if !(passesPhasesFilter(pod, p.ValidPhases) && passesPodNameFilter(pod, p.PodName)) {
continue
}
passesVerify, err := passesVerifyFilter(pod, p.Verify)
if err != nil {
Logf("Error detected on %v : %v !", pod.Name, err)
break ReturnPodsSoFar
}
if passesVerify {
filteredPods = append(filteredPods, pod)
}
}
return filteredPods, err
}
// WaitFor waits for some minimum number of pods to be verified, according to the PodStateVerification
// definition.
func (cl *ClusterVerification) WaitFor(atLeast int, timeout time.Duration) ([]v1.Pod, error) {
pods := []v1.Pod{}
var returnedErr error
err := wait.Poll(1*time.Second, timeout, func() (bool, error) {
pods, returnedErr = cl.podState.filter(cl.client, cl.namespace)
// Failure
if returnedErr != nil {
Logf("Cutting polling short: We got an error from the pod filtering layer.")
// stop polling if the pod filtering returns an error. that should never happen.
// it indicates, for example, that the client is broken or something non-pod related.
return false, returnedErr
}
Logf("Found %v / %v", len(pods), atLeast)
// Success
if len(pods) >= atLeast {
return true, nil
}
// Keep trying...
return false, nil
})
Logf("WaitFor completed with timeout %v. Pods found = %v out of %v", timeout, len(pods), atLeast)
return pods, err
}
// WaitForOrFail provides a shorthand WaitFor with failure as an option if anything goes wrong.
func (cl *ClusterVerification) WaitForOrFail(atLeast int, timeout time.Duration) {
pods, err := cl.WaitFor(atLeast, timeout)
if err != nil || len(pods) < atLeast {
Failf("Verified %v of %v pods , error : %v", len(pods), atLeast, err)
}
}
// ForEach runs a function against every verifiable pod. Be warned that this doesn't wait for "n" pods to verifiy,
// so it may return very quickly if you have strict pod state requirements.
//
// For example, if you require at least 5 pods to be running before your test will pass,
// its smart to first call "clusterVerification.WaitFor(5)" before you call clusterVerification.ForEach.
func (cl *ClusterVerification) ForEach(podFunc func(v1.Pod)) error {
pods, err := cl.podState.filter(cl.client, cl.namespace)
if err == nil {
if len(pods) == 0 {
Failf("No pods matched the filter.")
}
Logf("ForEach: Found %v pods from the filter. Now looping through them.", len(pods))
for _, p := range pods {
podFunc(p)
}
} else {
Logf("ForEach: Something went wrong when filtering pods to execute against: %v", err)
}
return err
}
// GetLogToFileFunc is a convenience function that returns a function that have the same interface as
// Logf, but writes to a specified file.
func GetLogToFileFunc(file *os.File) func(format string, args ...interface{}) {
return func(format string, args ...interface{}) {
writer := bufio.NewWriter(file)
if _, err := fmt.Fprintf(writer, format, args...); err != nil {
Logf("Failed to write file %v with test performance data: %v", file.Name(), err)
}
writer.Flush()
}
}

View File

@ -0,0 +1,84 @@
/*
Copyright 2016 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 framework
import (
"bufio"
"fmt"
"strings"
)
type KubemarkResourceUsage struct {
Name string
MemoryWorkingSetInBytes uint64
CPUUsageInCores float64
}
func getMasterUsageByPrefix(prefix string) (string, error) {
sshResult, err := SSH(fmt.Sprintf("ps ax -o %%cpu,rss,command | tail -n +2 | grep %v | sed 's/\\s+/ /g'", prefix), GetMasterHost()+":22", TestContext.Provider)
if err != nil {
return "", err
}
return sshResult.Stdout, nil
}
// TODO: figure out how to move this to kubemark directory (need to factor test SSH out of e2e framework)
func GetKubemarkMasterComponentsResourceUsage() map[string]*KubemarkResourceUsage {
result := make(map[string]*KubemarkResourceUsage)
// Get kuberenetes component resource usage
sshResult, err := getMasterUsageByPrefix("kube")
if err != nil {
Logf("Error when trying to SSH to master machine. Skipping probe. %v", err)
return nil
}
scanner := bufio.NewScanner(strings.NewReader(sshResult))
for scanner.Scan() {
var cpu float64
var mem uint64
var name string
fmt.Sscanf(strings.TrimSpace(scanner.Text()), "%f %d /usr/local/bin/kube-%s", &cpu, &mem, &name)
if name != "" {
// Gatherer expects pod_name/container_name format
fullName := name + "/" + name
result[fullName] = &KubemarkResourceUsage{Name: fullName, MemoryWorkingSetInBytes: mem * 1024, CPUUsageInCores: cpu / 100}
}
}
// Get etcd resource usage
sshResult, err = getMasterUsageByPrefix("bin/etcd")
if err != nil {
Logf("Error when trying to SSH to master machine. Skipping probe")
return nil
}
scanner = bufio.NewScanner(strings.NewReader(sshResult))
for scanner.Scan() {
var cpu float64
var mem uint64
var etcdKind string
fmt.Sscanf(strings.TrimSpace(scanner.Text()), "%f %d /bin/sh -c /usr/local/bin/etcd", &cpu, &mem)
dataDirStart := strings.Index(scanner.Text(), "--data-dir")
if dataDirStart < 0 {
continue
}
fmt.Sscanf(scanner.Text()[dataDirStart:], "--data-dir=/var/%s", &etcdKind)
if etcdKind != "" {
// Gatherer expects pod_name/container_name format
fullName := "etcd/" + etcdKind
result[fullName] = &KubemarkResourceUsage{Name: fullName, MemoryWorkingSetInBytes: mem * 1024, CPUUsageInCores: cpu / 100}
}
}
return result
}

View File

@ -0,0 +1,26 @@
package(default_visibility = ["//visibility:public"])
load(
"@io_bazel_rules_go//go:def.bzl",
"go_library",
)
go_library(
name = "go_default_library",
srcs = ["wrapper.go"],
importpath = "k8s.io/kubernetes/test/e2e/framework/ginkgowrapper",
deps = ["//vendor/github.com/onsi/ginkgo:go_default_library"],
)
filegroup(
name = "package-srcs",
srcs = glob(["**"]),
tags = ["automanaged"],
visibility = ["//visibility:private"],
)
filegroup(
name = "all-srcs",
srcs = [":package-srcs"],
tags = ["automanaged"],
)

View File

@ -0,0 +1,134 @@
/*
Copyright 2017 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 ginkgowrapper wraps Ginkgo Fail and Skip functions to panic
// with structured data instead of a constant string.
package ginkgowrapper
import (
"bufio"
"bytes"
"regexp"
"runtime"
"runtime/debug"
"strings"
"github.com/onsi/ginkgo"
)
// FailurePanic is the value that will be panicked from Fail.
type FailurePanic struct {
Message string // The failure message passed to Fail
Filename string // The filename that is the source of the failure
Line int // The line number of the filename that is the source of the failure
FullStackTrace string // A full stack trace starting at the source of the failure
}
// String makes FailurePanic look like the old Ginkgo panic when printed.
func (FailurePanic) String() string { return ginkgo.GINKGO_PANIC }
// Fail wraps ginkgo.Fail so that it panics with more useful
// information about the failure. This function will panic with a
// FailurePanic.
func Fail(message string, callerSkip ...int) {
skip := 1
if len(callerSkip) > 0 {
skip += callerSkip[0]
}
_, file, line, _ := runtime.Caller(skip)
fp := FailurePanic{
Message: message,
Filename: file,
Line: line,
FullStackTrace: pruneStack(skip),
}
defer func() {
e := recover()
if e != nil {
panic(fp)
}
}()
ginkgo.Fail(message, skip)
}
// SkipPanic is the value that will be panicked from Skip.
type SkipPanic struct {
Message string // The failure message passed to Fail
Filename string // The filename that is the source of the failure
Line int // The line number of the filename that is the source of the failure
FullStackTrace string // A full stack trace starting at the source of the failure
}
// String makes SkipPanic look like the old Ginkgo panic when printed.
func (SkipPanic) String() string { return ginkgo.GINKGO_PANIC }
// Skip wraps ginkgo.Skip so that it panics with more useful
// information about why the test is being skipped. This function will
// panic with a SkipPanic.
func Skip(message string, callerSkip ...int) {
skip := 1
if len(callerSkip) > 0 {
skip += callerSkip[0]
}
_, file, line, _ := runtime.Caller(skip)
sp := SkipPanic{
Message: message,
Filename: file,
Line: line,
FullStackTrace: pruneStack(skip),
}
defer func() {
e := recover()
if e != nil {
panic(sp)
}
}()
ginkgo.Skip(message, skip)
}
// ginkgo adds a lot of test running infrastructure to the stack, so
// we filter those out
var stackSkipPattern = regexp.MustCompile(`onsi/ginkgo`)
func pruneStack(skip int) string {
skip += 2 // one for pruneStack and one for debug.Stack
stack := debug.Stack()
scanner := bufio.NewScanner(bytes.NewBuffer(stack))
var prunedStack []string
// skip the top of the stack
for i := 0; i < 2*skip+1; i++ {
scanner.Scan()
}
for scanner.Scan() {
if stackSkipPattern.Match(scanner.Bytes()) {
scanner.Scan() // these come in pairs
} else {
prunedStack = append(prunedStack, scanner.Text())
scanner.Scan() // these come in pairs
prunedStack = append(prunedStack, scanner.Text())
}
}
return strings.Join(prunedStack, "\n")
}

View File

@ -0,0 +1,198 @@
/*
Copyright 2014 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 framework
import (
"encoding/json"
"fmt"
"io/ioutil"
"os/exec"
"path/filepath"
"regexp"
"strings"
"time"
"github.com/golang/glog"
"k8s.io/kubernetes/pkg/cloudprovider/providers/gce"
)
// TODO: These should really just use the GCE API client library or at least use
// better formatted output from the --format flag.
func CreateGCEStaticIP(name string) (string, error) {
// gcloud compute --project "abshah-kubernetes-001" addresses create "test-static-ip" --region "us-central1"
// abshah@abhidesk:~/go/src/code.google.com/p/google-api-go-client/compute/v1$ gcloud compute --project "abshah-kubernetes-001" addresses create "test-static-ip" --region "us-central1"
// Created [https://www.googleapis.com/compute/v1/projects/abshah-kubernetes-001/regions/us-central1/addresses/test-static-ip].
// NAME REGION ADDRESS STATUS
// test-static-ip us-central1 104.197.143.7 RESERVED
var outputBytes []byte
var err error
region, err := gce.GetGCERegion(TestContext.CloudConfig.Zone)
if err != nil {
return "", fmt.Errorf("failed to convert zone to region: %v", err)
}
glog.Infof("Creating static IP with name %q in project %q in region %q", name, TestContext.CloudConfig.ProjectID, region)
for attempts := 0; attempts < 4; attempts++ {
outputBytes, err = exec.Command("gcloud", "compute", "addresses", "create",
name, "--project", TestContext.CloudConfig.ProjectID,
"--region", region, "-q", "--format=yaml").CombinedOutput()
if err == nil {
break
}
glog.Errorf("output from failed attempt to create static IP: %s", outputBytes)
time.Sleep(time.Duration(5*attempts) * time.Second)
}
if err != nil {
// Ditch the error, since the stderr in the output is what actually contains
// any useful info.
return "", fmt.Errorf("failed to create static IP: %s", outputBytes)
}
output := string(outputBytes)
if strings.Contains(output, "RESERVED") {
r, _ := regexp.Compile("[0-9]+\\.[0-9]+\\.[0-9]+\\.[0-9]+")
staticIP := r.FindString(output)
if staticIP == "" {
return "", fmt.Errorf("static IP not found in gcloud command output: %v", output)
} else {
return staticIP, nil
}
} else {
return "", fmt.Errorf("static IP %q could not be reserved: %v", name, output)
}
}
func DeleteGCEStaticIP(name string) error {
// gcloud compute --project "abshah-kubernetes-001" addresses create "test-static-ip" --region "us-central1"
// abshah@abhidesk:~/go/src/code.google.com/p/google-api-go-client/compute/v1$ gcloud compute --project "abshah-kubernetes-001" addresses create "test-static-ip" --region "us-central1"
// Created [https://www.googleapis.com/compute/v1/projects/abshah-kubernetes-001/regions/us-central1/addresses/test-static-ip].
// NAME REGION ADDRESS STATUS
// test-static-ip us-central1 104.197.143.7 RESERVED
region, err := gce.GetGCERegion(TestContext.CloudConfig.Zone)
if err != nil {
return fmt.Errorf("failed to convert zone to region: %v", err)
}
glog.Infof("Deleting static IP with name %q in project %q in region %q", name, TestContext.CloudConfig.ProjectID, region)
outputBytes, err := exec.Command("gcloud", "compute", "addresses", "delete",
name, "--project", TestContext.CloudConfig.ProjectID,
"--region", region, "-q").CombinedOutput()
if err != nil {
// Ditch the error, since the stderr in the output is what actually contains
// any useful info.
return fmt.Errorf("failed to delete static IP %q: %v", name, string(outputBytes))
}
return nil
}
// Returns master & node image string, or error
func lookupClusterImageSources() (string, string, error) {
// Given args for a gcloud compute command, run it with other args, and return the values,
// whether separated by newlines, commas or semicolons.
gcloudf := func(argv ...string) ([]string, error) {
args := []string{"compute"}
args = append(args, argv...)
args = append(args, "--project", TestContext.CloudConfig.ProjectID,
"--zone", TestContext.CloudConfig.Zone)
outputBytes, err := exec.Command("gcloud", args...).CombinedOutput()
str := strings.Replace(string(outputBytes), ",", "\n", -1)
str = strings.Replace(str, ";", "\n", -1)
lines := strings.Split(str, "\n")
if err != nil {
Logf("lookupDiskImageSources: gcloud error with [%#v]; err:%v", argv, err)
for _, l := range lines {
Logf(" > %s", l)
}
}
return lines, err
}
// Given a GCE instance, look through its disks, finding one that has a sourceImage
host2image := func(instance string) (string, error) {
// gcloud compute instances describe {INSTANCE} --format="get(disks[].source)"
// gcloud compute disks describe {DISKURL} --format="get(sourceImage)"
disks, err := gcloudf("instances", "describe", instance, "--format=get(disks[].source)")
if err != nil {
return "", err
} else if len(disks) == 0 {
return "", fmt.Errorf("instance %q had no findable disks", instance)
}
// Loop over disks, looking for the boot disk
for _, disk := range disks {
lines, err := gcloudf("disks", "describe", disk, "--format=get(sourceImage)")
if err != nil {
return "", err
} else if len(lines) > 0 && lines[0] != "" {
return lines[0], nil // break, we're done
}
}
return "", fmt.Errorf("instance %q had no disk with a sourceImage", instance)
}
// gcloud compute instance-groups list-instances {GROUPNAME} --format="get(instance)"
nodeName := ""
instGroupName := strings.Split(TestContext.CloudConfig.NodeInstanceGroup, ",")[0]
if lines, err := gcloudf("instance-groups", "list-instances", instGroupName, "--format=get(instance)"); err != nil {
return "", "", err
} else if len(lines) == 0 {
return "", "", fmt.Errorf("no instances inside instance-group %q", instGroupName)
} else {
nodeName = lines[0]
}
nodeImg, err := host2image(nodeName)
if err != nil {
return "", "", err
}
frags := strings.Split(nodeImg, "/")
nodeImg = frags[len(frags)-1]
// For GKE clusters, MasterName will not be defined; we just leave masterImg blank.
masterImg := ""
if masterName := TestContext.CloudConfig.MasterName; masterName != "" {
img, err := host2image(masterName)
if err != nil {
return "", "", err
}
frags = strings.Split(img, "/")
masterImg = frags[len(frags)-1]
}
return masterImg, nodeImg, nil
}
func LogClusterImageSources() {
masterImg, nodeImg, err := lookupClusterImageSources()
if err != nil {
Logf("Cluster image sources lookup failed: %v\n", err)
return
}
Logf("cluster-master-image: %s", masterImg)
Logf("cluster-node-image: %s", nodeImg)
images := map[string]string{
"master_os_image": masterImg,
"node_os_image": nodeImg,
}
outputBytes, _ := json.MarshalIndent(images, "", " ")
filePath := filepath.Join(TestContext.ReportDir, "images.json")
if err := ioutil.WriteFile(filePath, outputBytes, 0644); err != nil {
Logf("cluster images sources, could not write to %q: %v", filePath, err)
}
}

View File

@ -0,0 +1,76 @@
/*
Copyright 2017 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 framework
import (
"k8s.io/api/core/v1"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/util/uuid"
. "github.com/onsi/gomega"
)
const (
// GPUResourceName is the extended name of the GPU resource since v1.8
// this uses the device plugin mechanism
NVIDIAGPUResourceName = "nvidia.com/gpu"
// TODO: Parametrize it by making it a feature in TestFramework.
// so we can override the daemonset in other setups (non COS).
// GPUDevicePluginDSYAML is the official Google Device Plugin Daemonset NVIDIA GPU manifest for GKE
GPUDevicePluginDSYAML = "https://raw.githubusercontent.com/kubernetes/kubernetes/master/cluster/addons/device-plugins/nvidia-gpu/daemonset.yaml"
)
// TODO make this generic and not linked to COS only
// NumberOfGPUs returs the number of GPUs advertised by a node
// This is based on the Device Plugin system and expected to run on a COS based node
// After the NVIDIA drivers were installed
func NumberOfNVIDIAGPUs(node *v1.Node) int64 {
val, ok := node.Status.Capacity[NVIDIAGPUResourceName]
if !ok {
return 0
}
return val.Value()
}
// NVIDIADevicePlugin returns the official Google Device Plugin pod for NVIDIA GPU in GKE
func NVIDIADevicePlugin(ns string) *v1.Pod {
ds, err := DsFromManifest(GPUDevicePluginDSYAML)
Expect(err).NotTo(HaveOccurred())
p := &v1.Pod{
ObjectMeta: metav1.ObjectMeta{
Name: "device-plugin-nvidia-gpu-" + string(uuid.NewUUID()),
Namespace: ns,
},
Spec: ds.Spec.Template.Spec,
}
// Remove node affinity
p.Spec.Affinity = nil
return p
}
func GetGPUDevicePluginImage() string {
ds, err := DsFromManifest(GPUDevicePluginDSYAML)
if err != nil || ds == nil || len(ds.Spec.Template.Spec.Containers) < 1 {
return ""
}
return ds.Spec.Template.Spec.Containers[0].Image
}

File diff suppressed because it is too large Load Diff

View File

@ -0,0 +1,259 @@
/*
Copyright 2017 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 framework
import (
"fmt"
"time"
batch "k8s.io/api/batch/v1"
"k8s.io/api/core/v1"
"k8s.io/apimachinery/pkg/api/errors"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/labels"
"k8s.io/apimachinery/pkg/util/wait"
clientset "k8s.io/client-go/kubernetes"
)
const (
// How long to wait for a job to finish.
JobTimeout = 15 * time.Minute
// Job selector name
JobSelectorKey = "job"
)
// NewTestJob returns a Job which does one of several testing behaviors. notTerminate starts a Job that will run
// effectively forever. fail starts a Job that will fail immediately. succeed starts a Job that will succeed
// immediately. randomlySucceedOrFail starts a Job that will succeed or fail randomly. failOnce fails the Job the
// first time it is run and succeeds subsequently. name is the Name of the Job. RestartPolicy indicates the restart
// policy of the containers in which the Pod is running. Parallelism is the Job's parallelism, and completions is the
// Job's required number of completions.
func NewTestJob(behavior, name string, rPol v1.RestartPolicy, parallelism, completions int32, activeDeadlineSeconds *int64, backoffLimit int32) *batch.Job {
job := &batch.Job{
ObjectMeta: metav1.ObjectMeta{
Name: name,
},
TypeMeta: metav1.TypeMeta{
Kind: "Job",
},
Spec: batch.JobSpec{
ActiveDeadlineSeconds: activeDeadlineSeconds,
Parallelism: &parallelism,
Completions: &completions,
BackoffLimit: &backoffLimit,
ManualSelector: newBool(false),
Template: v1.PodTemplateSpec{
ObjectMeta: metav1.ObjectMeta{
Labels: map[string]string{JobSelectorKey: name},
},
Spec: v1.PodSpec{
RestartPolicy: rPol,
Volumes: []v1.Volume{
{
Name: "data",
VolumeSource: v1.VolumeSource{
EmptyDir: &v1.EmptyDirVolumeSource{},
},
},
},
Containers: []v1.Container{
{
Name: "c",
Image: BusyBoxImage,
Command: []string{},
VolumeMounts: []v1.VolumeMount{
{
MountPath: "/data",
Name: "data",
},
},
},
},
},
},
},
}
switch behavior {
case "notTerminate":
job.Spec.Template.Spec.Containers[0].Command = []string{"sleep", "1000000"}
case "fail":
job.Spec.Template.Spec.Containers[0].Command = []string{"/bin/sh", "-c", "exit 1"}
case "succeed":
job.Spec.Template.Spec.Containers[0].Command = []string{"/bin/sh", "-c", "exit 0"}
case "randomlySucceedOrFail":
// Bash's $RANDOM generates pseudorandom int in range 0 - 32767.
// Dividing by 16384 gives roughly 50/50 chance of success.
job.Spec.Template.Spec.Containers[0].Command = []string{"/bin/sh", "-c", "exit $(( $RANDOM / 16384 ))"}
case "failOnce":
// Fail the first the container of the pod is run, and
// succeed the second time. Checks for file on emptydir.
// If present, succeed. If not, create but fail.
// Note that this cannot be used with RestartNever because
// it always fails the first time for a pod.
job.Spec.Template.Spec.Containers[0].Command = []string{"/bin/sh", "-c", "if [[ -r /data/foo ]] ; then exit 0 ; else touch /data/foo ; exit 1 ; fi"}
}
return job
}
// GetJob uses c to get the Job in namespace ns named name. If the returned error is nil, the returned Job is valid.
func GetJob(c clientset.Interface, ns, name string) (*batch.Job, error) {
return c.BatchV1().Jobs(ns).Get(name, metav1.GetOptions{})
}
// CreateJob uses c to create job in namespace ns. If the returned error is nil, the returned Job is valid and has
// been created.
func CreateJob(c clientset.Interface, ns string, job *batch.Job) (*batch.Job, error) {
return c.BatchV1().Jobs(ns).Create(job)
}
// UpdateJob uses c to updated job in namespace ns. If the returned error is nil, the returned Job is valid and has
// been updated.
func UpdateJob(c clientset.Interface, ns string, job *batch.Job) (*batch.Job, error) {
return c.BatchV1().Jobs(ns).Update(job)
}
// UpdateJobFunc updates the job object. It retries if there is a conflict, throw out error if
// there is any other errors. name is the job name, updateFn is the function updating the
// job object.
func UpdateJobFunc(c clientset.Interface, ns, name string, updateFn func(job *batch.Job)) {
ExpectNoError(wait.Poll(time.Millisecond*500, time.Second*30, func() (bool, error) {
job, err := GetJob(c, ns, name)
if err != nil {
return false, fmt.Errorf("failed to get pod %q: %v", name, err)
}
updateFn(job)
_, err = UpdateJob(c, ns, job)
if err == nil {
Logf("Successfully updated job %q", name)
return true, nil
}
if errors.IsConflict(err) {
Logf("Conflicting update to job %q, re-get and re-update: %v", name, err)
return false, nil
}
return false, fmt.Errorf("failed to update job %q: %v", name, err)
}))
}
// DeleteJob uses c to delete the Job named name in namespace ns. If the returned error is nil, the Job has been
// deleted.
func DeleteJob(c clientset.Interface, ns, name string) error {
return c.BatchV1().Jobs(ns).Delete(name, nil)
}
// GetJobPods returns a list of Pods belonging to a Job.
func GetJobPods(c clientset.Interface, ns, jobName string) (*v1.PodList, error) {
label := labels.SelectorFromSet(labels.Set(map[string]string{JobSelectorKey: jobName}))
options := metav1.ListOptions{LabelSelector: label.String()}
return c.CoreV1().Pods(ns).List(options)
}
// WaitForAllJobPodsRunning wait for all pods for the Job named JobName in namespace ns to become Running. Only use
// when pods will run for a long time, or it will be racy.
func WaitForAllJobPodsRunning(c clientset.Interface, ns, jobName string, parallelism int32) error {
return wait.Poll(Poll, JobTimeout, func() (bool, error) {
pods, err := GetJobPods(c, ns, jobName)
if err != nil {
return false, err
}
count := int32(0)
for _, p := range pods.Items {
if p.Status.Phase == v1.PodRunning {
count++
}
}
return count == parallelism, nil
})
}
// WaitForJobFinish uses c to wait for compeletions to complete for the Job jobName in namespace ns.
func WaitForJobFinish(c clientset.Interface, ns, jobName string, completions int32) error {
return wait.Poll(Poll, JobTimeout, func() (bool, error) {
curr, err := c.BatchV1().Jobs(ns).Get(jobName, metav1.GetOptions{})
if err != nil {
return false, err
}
return curr.Status.Succeeded == completions, nil
})
}
// WaitForJobFailure uses c to wait for up to timeout for the Job named jobName in namespace ns to fail.
func WaitForJobFailure(c clientset.Interface, ns, jobName string, timeout time.Duration, reason string) error {
return wait.Poll(Poll, timeout, func() (bool, error) {
curr, err := c.BatchV1().Jobs(ns).Get(jobName, metav1.GetOptions{})
if err != nil {
return false, err
}
for _, c := range curr.Status.Conditions {
if c.Type == batch.JobFailed && c.Status == v1.ConditionTrue {
if reason == "" || reason == c.Reason {
return true, nil
}
}
}
return false, nil
})
}
// CheckForAllJobPodsRunning uses c to check in the Job named jobName in ns is running. If the returned error is not
// nil the returned bool is true if the Job is running.
func CheckForAllJobPodsRunning(c clientset.Interface, ns, jobName string, parallelism int32) (bool, error) {
label := labels.SelectorFromSet(labels.Set(map[string]string{JobSelectorKey: jobName}))
options := metav1.ListOptions{LabelSelector: label.String()}
pods, err := c.CoreV1().Pods(ns).List(options)
if err != nil {
return false, err
}
count := int32(0)
for _, p := range pods.Items {
if p.Status.Phase == v1.PodRunning {
count++
}
}
return count == parallelism, nil
}
func newBool(val bool) *bool {
p := new(bool)
*p = val
return p
}
type updateJobFunc func(*batch.Job)
func UpdateJobWithRetries(c clientset.Interface, namespace, name string, applyUpdate updateJobFunc) (job *batch.Job, err error) {
jobs := c.BatchV1().Jobs(namespace)
var updateErr error
pollErr := wait.PollImmediate(10*time.Millisecond, 1*time.Minute, func() (bool, error) {
if job, err = jobs.Get(name, metav1.GetOptions{}); err != nil {
return false, err
}
// Apply the update, then attempt to push it to the apiserver.
applyUpdate(job)
if job, err = jobs.Update(job); err == nil {
Logf("Updating job %s", name)
return true, nil
}
updateErr = err
return false, nil
})
if pollErr == wait.ErrWaitTimeout {
pollErr = fmt.Errorf("couldn't apply the provided updated to job %q: %v", name, updateErr)
}
return job, pollErr
}

View File

@ -0,0 +1,852 @@
/*
Copyright 2014 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 framework
import (
"bytes"
"context"
"encoding/json"
"fmt"
"sort"
"strconv"
"strings"
"sync"
"text/tabwriter"
"time"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
utilerrors "k8s.io/apimachinery/pkg/util/errors"
"k8s.io/apimachinery/pkg/util/sets"
"k8s.io/apimachinery/pkg/util/wait"
clientset "k8s.io/client-go/kubernetes"
stats "k8s.io/kubernetes/pkg/kubelet/apis/stats/v1alpha1"
dockermetrics "k8s.io/kubernetes/pkg/kubelet/dockershim/metrics"
kubeletmetrics "k8s.io/kubernetes/pkg/kubelet/metrics"
"k8s.io/kubernetes/pkg/master/ports"
"k8s.io/kubernetes/test/e2e/framework/metrics"
"github.com/prometheus/common/model"
)
// KubeletMetric stores metrics scraped from the kubelet server's /metric endpoint.
// TODO: Get some more structure around the metrics and this type
type KubeletLatencyMetric struct {
// eg: list, info, create
Operation string
// eg: sync_pods, pod_worker
Method string
// 0 <= quantile <=1, e.g. 0.95 is 95%tile, 0.5 is median.
Quantile float64
Latency time.Duration
}
// KubeletMetricByLatency implements sort.Interface for []KubeletMetric based on
// the latency field.
type KubeletLatencyMetrics []KubeletLatencyMetric
func (a KubeletLatencyMetrics) Len() int { return len(a) }
func (a KubeletLatencyMetrics) Swap(i, j int) { a[i], a[j] = a[j], a[i] }
func (a KubeletLatencyMetrics) Less(i, j int) bool { return a[i].Latency > a[j].Latency }
// If a apiserver client is passed in, the function will try to get kubelet metrics from metrics grabber;
// or else, the function will try to get kubelet metrics directly from the node.
func getKubeletMetricsFromNode(c clientset.Interface, nodeName string) (metrics.KubeletMetrics, error) {
if c == nil {
return metrics.GrabKubeletMetricsWithoutProxy(nodeName)
}
grabber, err := metrics.NewMetricsGrabber(c, nil, true, false, false, false, false)
if err != nil {
return metrics.KubeletMetrics{}, err
}
return grabber.GrabFromKubelet(nodeName)
}
// getKubeletMetrics gets all metrics in kubelet subsystem from specified node and trims
// the subsystem prefix.
func getKubeletMetrics(c clientset.Interface, nodeName string) (metrics.KubeletMetrics, error) {
ms, err := getKubeletMetricsFromNode(c, nodeName)
if err != nil {
return metrics.KubeletMetrics{}, err
}
kubeletMetrics := make(metrics.KubeletMetrics)
for name, samples := range ms {
const prefix = kubeletmetrics.KubeletSubsystem + "_"
if !strings.HasPrefix(name, prefix) {
// Not a kubelet metric.
continue
}
method := strings.TrimPrefix(name, prefix)
kubeletMetrics[method] = samples
}
return kubeletMetrics, nil
}
// GetKubeletLatencyMetrics gets all latency related kubelet metrics. Note that the KubeletMetrcis
// passed in should not contain subsystem prefix.
func GetKubeletLatencyMetrics(ms metrics.KubeletMetrics) KubeletLatencyMetrics {
latencyMethods := sets.NewString(
kubeletmetrics.PodWorkerLatencyKey,
kubeletmetrics.PodWorkerStartLatencyKey,
kubeletmetrics.PodStartLatencyKey,
kubeletmetrics.CgroupManagerOperationsKey,
dockermetrics.DockerOperationsLatencyKey,
kubeletmetrics.PodWorkerStartLatencyKey,
kubeletmetrics.PLEGRelistLatencyKey,
)
return GetKubeletMetrics(ms, latencyMethods)
}
func GetKubeletMetrics(ms metrics.KubeletMetrics, methods sets.String) KubeletLatencyMetrics {
var latencyMetrics KubeletLatencyMetrics
for method, samples := range ms {
if !methods.Has(method) {
continue
}
for _, sample := range samples {
latency := sample.Value
operation := string(sample.Metric["operation_type"])
var quantile float64
if val, ok := sample.Metric[model.QuantileLabel]; ok {
var err error
if quantile, err = strconv.ParseFloat(string(val), 64); err != nil {
continue
}
}
latencyMetrics = append(latencyMetrics, KubeletLatencyMetric{
Operation: operation,
Method: method,
Quantile: quantile,
Latency: time.Duration(int64(latency)) * time.Microsecond,
})
}
}
return latencyMetrics
}
// RuntimeOperationMonitor is the tool getting and parsing docker operation metrics.
type RuntimeOperationMonitor struct {
client clientset.Interface
nodesRuntimeOps map[string]NodeRuntimeOperationErrorRate
}
// NodeRuntimeOperationErrorRate is the runtime operation error rate on one node.
type NodeRuntimeOperationErrorRate map[string]*RuntimeOperationErrorRate
// RuntimeOperationErrorRate is the error rate of a specified runtime operation.
type RuntimeOperationErrorRate struct {
TotalNumber float64
ErrorRate float64
TimeoutRate float64
}
func NewRuntimeOperationMonitor(c clientset.Interface) *RuntimeOperationMonitor {
m := &RuntimeOperationMonitor{
client: c,
nodesRuntimeOps: make(map[string]NodeRuntimeOperationErrorRate),
}
nodes, err := m.client.CoreV1().Nodes().List(metav1.ListOptions{})
if err != nil {
Failf("RuntimeOperationMonitor: unable to get list of nodes: %v", err)
}
for _, node := range nodes.Items {
m.nodesRuntimeOps[node.Name] = make(NodeRuntimeOperationErrorRate)
}
// Initialize the runtime operation error rate
m.GetRuntimeOperationErrorRate()
return m
}
// GetRuntimeOperationErrorRate gets runtime operation records from kubelet metrics and calculate
// error rates of all runtime operations.
func (m *RuntimeOperationMonitor) GetRuntimeOperationErrorRate() map[string]NodeRuntimeOperationErrorRate {
for node := range m.nodesRuntimeOps {
nodeResult, err := getNodeRuntimeOperationErrorRate(m.client, node)
if err != nil {
Logf("GetRuntimeOperationErrorRate: unable to get kubelet metrics from node %q: %v", node, err)
continue
}
m.nodesRuntimeOps[node] = nodeResult
}
return m.nodesRuntimeOps
}
// GetLatestRuntimeOperationErrorRate gets latest error rate and timeout rate from last observed RuntimeOperationErrorRate.
func (m *RuntimeOperationMonitor) GetLatestRuntimeOperationErrorRate() map[string]NodeRuntimeOperationErrorRate {
result := make(map[string]NodeRuntimeOperationErrorRate)
for node := range m.nodesRuntimeOps {
result[node] = make(NodeRuntimeOperationErrorRate)
oldNodeResult := m.nodesRuntimeOps[node]
curNodeResult, err := getNodeRuntimeOperationErrorRate(m.client, node)
if err != nil {
Logf("GetLatestRuntimeOperationErrorRate: unable to get kubelet metrics from node %q: %v", node, err)
continue
}
for op, cur := range curNodeResult {
t := *cur
if old, found := oldNodeResult[op]; found {
t.ErrorRate = (t.ErrorRate*t.TotalNumber - old.ErrorRate*old.TotalNumber) / (t.TotalNumber - old.TotalNumber)
t.TimeoutRate = (t.TimeoutRate*t.TotalNumber - old.TimeoutRate*old.TotalNumber) / (t.TotalNumber - old.TotalNumber)
t.TotalNumber -= old.TotalNumber
}
result[node][op] = &t
}
m.nodesRuntimeOps[node] = curNodeResult
}
return result
}
// FormatRuntimeOperationErrorRate formats the runtime operation error rate to string.
func FormatRuntimeOperationErrorRate(nodesResult map[string]NodeRuntimeOperationErrorRate) string {
lines := []string{}
for node, nodeResult := range nodesResult {
lines = append(lines, fmt.Sprintf("node %q runtime operation error rate:", node))
for op, result := range nodeResult {
line := fmt.Sprintf("operation %q: total - %.0f; error rate - %f; timeout rate - %f", op,
result.TotalNumber, result.ErrorRate, result.TimeoutRate)
lines = append(lines, line)
}
lines = append(lines, fmt.Sprintln())
}
return strings.Join(lines, "\n")
}
// getNodeRuntimeOperationErrorRate gets runtime operation error rate from specified node.
func getNodeRuntimeOperationErrorRate(c clientset.Interface, node string) (NodeRuntimeOperationErrorRate, error) {
result := make(NodeRuntimeOperationErrorRate)
ms, err := getKubeletMetrics(c, node)
if err != nil {
return result, err
}
// If no corresponding metrics are found, the returned samples will be empty. Then the following
// loop will be skipped automatically.
allOps := ms[dockermetrics.DockerOperationsKey]
errOps := ms[dockermetrics.DockerOperationsErrorsKey]
timeoutOps := ms[dockermetrics.DockerOperationsTimeoutKey]
for _, sample := range allOps {
operation := string(sample.Metric["operation_type"])
result[operation] = &RuntimeOperationErrorRate{TotalNumber: float64(sample.Value)}
}
for _, sample := range errOps {
operation := string(sample.Metric["operation_type"])
// Should always find the corresponding item, just in case
if _, found := result[operation]; found {
result[operation].ErrorRate = float64(sample.Value) / result[operation].TotalNumber
}
}
for _, sample := range timeoutOps {
operation := string(sample.Metric["operation_type"])
if _, found := result[operation]; found {
result[operation].TimeoutRate = float64(sample.Value) / result[operation].TotalNumber
}
}
return result, nil
}
// HighLatencyKubeletOperations logs and counts the high latency metrics exported by the kubelet server via /metrics.
func HighLatencyKubeletOperations(c clientset.Interface, threshold time.Duration, nodeName string, logFunc func(fmt string, args ...interface{})) (KubeletLatencyMetrics, error) {
ms, err := getKubeletMetrics(c, nodeName)
if err != nil {
return KubeletLatencyMetrics{}, err
}
latencyMetrics := GetKubeletLatencyMetrics(ms)
sort.Sort(latencyMetrics)
var badMetrics KubeletLatencyMetrics
logFunc("\nLatency metrics for node %v", nodeName)
for _, m := range latencyMetrics {
if m.Latency > threshold {
badMetrics = append(badMetrics, m)
Logf("%+v", m)
}
}
return badMetrics, nil
}
// getStatsSummary contacts kubelet for the container information.
func getStatsSummary(c clientset.Interface, nodeName string) (*stats.Summary, error) {
ctx, cancel := context.WithTimeout(context.Background(), SingleCallTimeout)
defer cancel()
data, err := c.CoreV1().RESTClient().Get().
Context(ctx).
Resource("nodes").
SubResource("proxy").
Name(fmt.Sprintf("%v:%v", nodeName, ports.KubeletPort)).
Suffix("stats/summary").
Do().Raw()
if err != nil {
return nil, err
}
summary := stats.Summary{}
err = json.Unmarshal(data, &summary)
if err != nil {
return nil, err
}
return &summary, nil
}
func removeUint64Ptr(ptr *uint64) uint64 {
if ptr == nil {
return 0
}
return *ptr
}
// getOneTimeResourceUsageOnNode queries the node's /stats/summary endpoint
// and returns the resource usage of all containerNames for the past
// cpuInterval.
// The acceptable range of the interval is 2s~120s. Be warned that as the
// interval (and #containers) increases, the size of kubelet's response
// could be significant. E.g., the 60s interval stats for ~20 containers is
// ~1.5MB. Don't hammer the node with frequent, heavy requests.
//
// cadvisor records cumulative cpu usage in nanoseconds, so we need to have two
// stats points to compute the cpu usage over the interval. Assuming cadvisor
// polls every second, we'd need to get N stats points for N-second interval.
// Note that this is an approximation and may not be accurate, hence we also
// write the actual interval used for calculation (based on the timestamps of
// the stats points in ContainerResourceUsage.CPUInterval.
//
// containerNames is a function returning a collection of container names in which
// user is interested in.
func getOneTimeResourceUsageOnNode(
c clientset.Interface,
nodeName string,
cpuInterval time.Duration,
containerNames func() []string,
) (ResourceUsagePerContainer, error) {
const (
// cadvisor records stats about every second.
cadvisorStatsPollingIntervalInSeconds float64 = 1.0
// cadvisor caches up to 2 minutes of stats (configured by kubelet).
maxNumStatsToRequest int = 120
)
numStats := int(float64(cpuInterval.Seconds()) / cadvisorStatsPollingIntervalInSeconds)
if numStats < 2 || numStats > maxNumStatsToRequest {
return nil, fmt.Errorf("numStats needs to be > 1 and < %d", maxNumStatsToRequest)
}
// Get information of all containers on the node.
summary, err := getStatsSummary(c, nodeName)
if err != nil {
return nil, err
}
f := func(name string, newStats *stats.ContainerStats) *ContainerResourceUsage {
if newStats == nil || newStats.CPU == nil || newStats.Memory == nil {
return nil
}
return &ContainerResourceUsage{
Name: name,
Timestamp: newStats.StartTime.Time,
CPUUsageInCores: float64(removeUint64Ptr(newStats.CPU.UsageNanoCores)) / 1000000000,
MemoryUsageInBytes: removeUint64Ptr(newStats.Memory.UsageBytes),
MemoryWorkingSetInBytes: removeUint64Ptr(newStats.Memory.WorkingSetBytes),
MemoryRSSInBytes: removeUint64Ptr(newStats.Memory.RSSBytes),
CPUInterval: 0,
}
}
// Process container infos that are relevant to us.
containers := containerNames()
usageMap := make(ResourceUsagePerContainer, len(containers))
observedContainers := []string{}
for _, pod := range summary.Pods {
for _, container := range pod.Containers {
isInteresting := false
for _, interestingContainerName := range containers {
if container.Name == interestingContainerName {
isInteresting = true
observedContainers = append(observedContainers, container.Name)
break
}
}
if !isInteresting {
continue
}
if usage := f(pod.PodRef.Name+"/"+container.Name, &container); usage != nil {
usageMap[pod.PodRef.Name+"/"+container.Name] = usage
}
}
}
return usageMap, nil
}
func getNodeStatsSummary(c clientset.Interface, nodeName string) (*stats.Summary, error) {
data, err := c.CoreV1().RESTClient().Get().
Resource("nodes").
SubResource("proxy").
Name(fmt.Sprintf("%v:%v", nodeName, ports.KubeletPort)).
Suffix("stats/summary").
SetHeader("Content-Type", "application/json").
Do().Raw()
if err != nil {
return nil, err
}
var summary *stats.Summary
err = json.Unmarshal(data, &summary)
if err != nil {
return nil, err
}
return summary, nil
}
func getSystemContainerStats(summary *stats.Summary) map[string]*stats.ContainerStats {
statsList := summary.Node.SystemContainers
statsMap := make(map[string]*stats.ContainerStats)
for i := range statsList {
statsMap[statsList[i].Name] = &statsList[i]
}
// Create a root container stats using information available in
// stats.NodeStats. This is necessary since it is a different type.
statsMap[rootContainerName] = &stats.ContainerStats{
CPU: summary.Node.CPU,
Memory: summary.Node.Memory,
}
return statsMap
}
const (
rootContainerName = "/"
)
// A list of containers for which we want to collect resource usage.
func TargetContainers() []string {
return []string{
rootContainerName,
stats.SystemContainerRuntime,
stats.SystemContainerKubelet,
}
}
type ContainerResourceUsage struct {
Name string
Timestamp time.Time
CPUUsageInCores float64
MemoryUsageInBytes uint64
MemoryWorkingSetInBytes uint64
MemoryRSSInBytes uint64
// The interval used to calculate CPUUsageInCores.
CPUInterval time.Duration
}
func (r *ContainerResourceUsage) isStrictlyGreaterThan(rhs *ContainerResourceUsage) bool {
return r.CPUUsageInCores > rhs.CPUUsageInCores && r.MemoryWorkingSetInBytes > rhs.MemoryWorkingSetInBytes
}
type ResourceUsagePerContainer map[string]*ContainerResourceUsage
type ResourceUsagePerNode map[string]ResourceUsagePerContainer
func formatResourceUsageStats(nodeName string, containerStats ResourceUsagePerContainer) string {
// Example output:
//
// Resource usage for node "e2e-test-foo-node-abcde":
// container cpu(cores) memory(MB)
// "/" 0.363 2942.09
// "/docker-daemon" 0.088 521.80
// "/kubelet" 0.086 424.37
// "/system" 0.007 119.88
buf := &bytes.Buffer{}
w := tabwriter.NewWriter(buf, 1, 0, 1, ' ', 0)
fmt.Fprintf(w, "container\tcpu(cores)\tmemory_working_set(MB)\tmemory_rss(MB)\n")
for name, s := range containerStats {
fmt.Fprintf(w, "%q\t%.3f\t%.2f\t%.2f\n", name, s.CPUUsageInCores, float64(s.MemoryWorkingSetInBytes)/(1024*1024), float64(s.MemoryRSSInBytes)/(1024*1024))
}
w.Flush()
return fmt.Sprintf("Resource usage on node %q:\n%s", nodeName, buf.String())
}
type uint64arr []uint64
func (a uint64arr) Len() int { return len(a) }
func (a uint64arr) Swap(i, j int) { a[i], a[j] = a[j], a[i] }
func (a uint64arr) Less(i, j int) bool { return a[i] < a[j] }
type usageDataPerContainer struct {
cpuData []float64
memUseData []uint64
memWorkSetData []uint64
}
func GetKubeletHeapStats(c clientset.Interface, nodeName string) (string, error) {
client, err := NodeProxyRequest(c, nodeName, "debug/pprof/heap")
if err != nil {
return "", err
}
raw, errRaw := client.Raw()
if errRaw != nil {
return "", err
}
stats := string(raw)
// Only dumping the runtime.MemStats numbers to avoid polluting the log.
numLines := 23
lines := strings.Split(stats, "\n")
return strings.Join(lines[len(lines)-numLines:], "\n"), nil
}
func PrintAllKubeletPods(c clientset.Interface, nodeName string) {
podList, err := GetKubeletPods(c, nodeName)
if err != nil {
Logf("Unable to retrieve kubelet pods for node %v: %v", nodeName, err)
return
}
for _, p := range podList.Items {
Logf("%v from %v started at %v (%d container statuses recorded)", p.Name, p.Namespace, p.Status.StartTime, len(p.Status.ContainerStatuses))
for _, c := range p.Status.ContainerStatuses {
Logf("\tContainer %v ready: %v, restart count %v",
c.Name, c.Ready, c.RestartCount)
}
}
}
func computeContainerResourceUsage(name string, oldStats, newStats *stats.ContainerStats) *ContainerResourceUsage {
return &ContainerResourceUsage{
Name: name,
Timestamp: newStats.CPU.Time.Time,
CPUUsageInCores: float64(*newStats.CPU.UsageCoreNanoSeconds-*oldStats.CPU.UsageCoreNanoSeconds) / float64(newStats.CPU.Time.Time.Sub(oldStats.CPU.Time.Time).Nanoseconds()),
MemoryUsageInBytes: *newStats.Memory.UsageBytes,
MemoryWorkingSetInBytes: *newStats.Memory.WorkingSetBytes,
MemoryRSSInBytes: *newStats.Memory.RSSBytes,
CPUInterval: newStats.CPU.Time.Time.Sub(oldStats.CPU.Time.Time),
}
}
// resourceCollector periodically polls the node, collect stats for a given
// list of containers, computes and cache resource usage up to
// maxEntriesPerContainer for each container.
type resourceCollector struct {
lock sync.RWMutex
node string
containers []string
client clientset.Interface
buffers map[string][]*ContainerResourceUsage
pollingInterval time.Duration
stopCh chan struct{}
}
func newResourceCollector(c clientset.Interface, nodeName string, containerNames []string, pollingInterval time.Duration) *resourceCollector {
buffers := make(map[string][]*ContainerResourceUsage)
return &resourceCollector{
node: nodeName,
containers: containerNames,
client: c,
buffers: buffers,
pollingInterval: pollingInterval,
}
}
// Start starts a goroutine to Poll the node every pollingInterval.
func (r *resourceCollector) Start() {
r.stopCh = make(chan struct{}, 1)
// Keep the last observed stats for comparison.
oldStats := make(map[string]*stats.ContainerStats)
go wait.Until(func() { r.collectStats(oldStats) }, r.pollingInterval, r.stopCh)
}
// Stop sends a signal to terminate the stats collecting goroutine.
func (r *resourceCollector) Stop() {
close(r.stopCh)
}
// collectStats gets the latest stats from kubelet stats summary API, computes
// the resource usage, and pushes it to the buffer.
func (r *resourceCollector) collectStats(oldStatsMap map[string]*stats.ContainerStats) {
summary, err := getNodeStatsSummary(r.client, r.node)
if err != nil {
Logf("Error getting node stats summary on %q, err: %v", r.node, err)
return
}
cStatsMap := getSystemContainerStats(summary)
r.lock.Lock()
defer r.lock.Unlock()
for _, name := range r.containers {
cStats, ok := cStatsMap[name]
if !ok {
Logf("Missing info/stats for container %q on node %q", name, r.node)
return
}
if oldStats, ok := oldStatsMap[name]; ok {
if oldStats.CPU.Time.Equal(&cStats.CPU.Time) {
// No change -> skip this stat.
continue
}
r.buffers[name] = append(r.buffers[name], computeContainerResourceUsage(name, oldStats, cStats))
}
// Update the old stats.
oldStatsMap[name] = cStats
}
}
func (r *resourceCollector) GetLatest() (ResourceUsagePerContainer, error) {
r.lock.RLock()
defer r.lock.RUnlock()
stats := make(ResourceUsagePerContainer)
for _, name := range r.containers {
contStats, ok := r.buffers[name]
if !ok || len(contStats) == 0 {
return nil, fmt.Errorf("Resource usage on node %q is not ready yet", r.node)
}
stats[name] = contStats[len(contStats)-1]
}
return stats, nil
}
// Reset frees the stats and start over.
func (r *resourceCollector) Reset() {
r.lock.Lock()
defer r.lock.Unlock()
for _, name := range r.containers {
r.buffers[name] = []*ContainerResourceUsage{}
}
}
type resourceUsageByCPU []*ContainerResourceUsage
func (r resourceUsageByCPU) Len() int { return len(r) }
func (r resourceUsageByCPU) Swap(i, j int) { r[i], r[j] = r[j], r[i] }
func (r resourceUsageByCPU) Less(i, j int) bool { return r[i].CPUUsageInCores < r[j].CPUUsageInCores }
// The percentiles to report.
var percentiles = [...]float64{0.05, 0.20, 0.50, 0.70, 0.90, 0.95, 0.99}
// GetBasicCPUStats returns the percentiles the cpu usage in cores for
// containerName. This method examines all data currently in the buffer.
func (r *resourceCollector) GetBasicCPUStats(containerName string) map[float64]float64 {
r.lock.RLock()
defer r.lock.RUnlock()
result := make(map[float64]float64, len(percentiles))
usages := r.buffers[containerName]
sort.Sort(resourceUsageByCPU(usages))
for _, q := range percentiles {
index := int(float64(len(usages))*q) - 1
if index < 0 {
// We don't have enough data.
result[q] = 0
continue
}
result[q] = usages[index].CPUUsageInCores
}
return result
}
// ResourceMonitor manages a resourceCollector per node.
type ResourceMonitor struct {
client clientset.Interface
containers []string
pollingInterval time.Duration
collectors map[string]*resourceCollector
}
func NewResourceMonitor(c clientset.Interface, containerNames []string, pollingInterval time.Duration) *ResourceMonitor {
return &ResourceMonitor{
containers: containerNames,
client: c,
pollingInterval: pollingInterval,
}
}
func (r *ResourceMonitor) Start() {
// It should be OK to monitor unschedulable Nodes
nodes, err := r.client.CoreV1().Nodes().List(metav1.ListOptions{})
if err != nil {
Failf("ResourceMonitor: unable to get list of nodes: %v", err)
}
r.collectors = make(map[string]*resourceCollector, 0)
for _, node := range nodes.Items {
collector := newResourceCollector(r.client, node.Name, r.containers, r.pollingInterval)
r.collectors[node.Name] = collector
collector.Start()
}
}
func (r *ResourceMonitor) Stop() {
for _, collector := range r.collectors {
collector.Stop()
}
}
func (r *ResourceMonitor) Reset() {
for _, collector := range r.collectors {
collector.Reset()
}
}
func (r *ResourceMonitor) LogLatest() {
summary, err := r.GetLatest()
if err != nil {
Logf("%v", err)
}
Logf("%s", r.FormatResourceUsage(summary))
}
func (r *ResourceMonitor) FormatResourceUsage(s ResourceUsagePerNode) string {
summary := []string{}
for node, usage := range s {
summary = append(summary, formatResourceUsageStats(node, usage))
}
return strings.Join(summary, "\n")
}
func (r *ResourceMonitor) GetLatest() (ResourceUsagePerNode, error) {
result := make(ResourceUsagePerNode)
errs := []error{}
for key, collector := range r.collectors {
s, err := collector.GetLatest()
if err != nil {
errs = append(errs, err)
continue
}
result[key] = s
}
return result, utilerrors.NewAggregate(errs)
}
func (r *ResourceMonitor) GetMasterNodeLatest(usagePerNode ResourceUsagePerNode) ResourceUsagePerNode {
result := make(ResourceUsagePerNode)
var masterUsage ResourceUsagePerContainer
var nodesUsage []ResourceUsagePerContainer
for node, usage := range usagePerNode {
if strings.HasSuffix(node, "master") {
masterUsage = usage
} else {
nodesUsage = append(nodesUsage, usage)
}
}
nodeAvgUsage := make(ResourceUsagePerContainer)
for _, nodeUsage := range nodesUsage {
for c, usage := range nodeUsage {
if _, found := nodeAvgUsage[c]; !found {
nodeAvgUsage[c] = &ContainerResourceUsage{Name: usage.Name}
}
nodeAvgUsage[c].CPUUsageInCores += usage.CPUUsageInCores
nodeAvgUsage[c].MemoryUsageInBytes += usage.MemoryUsageInBytes
nodeAvgUsage[c].MemoryWorkingSetInBytes += usage.MemoryWorkingSetInBytes
nodeAvgUsage[c].MemoryRSSInBytes += usage.MemoryRSSInBytes
}
}
for c := range nodeAvgUsage {
nodeAvgUsage[c].CPUUsageInCores /= float64(len(nodesUsage))
nodeAvgUsage[c].MemoryUsageInBytes /= uint64(len(nodesUsage))
nodeAvgUsage[c].MemoryWorkingSetInBytes /= uint64(len(nodesUsage))
nodeAvgUsage[c].MemoryRSSInBytes /= uint64(len(nodesUsage))
}
result["master"] = masterUsage
result["node"] = nodeAvgUsage
return result
}
// ContainersCPUSummary is indexed by the container name with each entry a
// (percentile, value) map.
type ContainersCPUSummary map[string]map[float64]float64
// NodesCPUSummary is indexed by the node name with each entry a
// ContainersCPUSummary map.
type NodesCPUSummary map[string]ContainersCPUSummary
func (r *ResourceMonitor) FormatCPUSummary(summary NodesCPUSummary) string {
// Example output for a node (the percentiles may differ):
// CPU usage of containers on node "e2e-test-foo-node-0vj7":
// container 5th% 50th% 90th% 95th%
// "/" 0.051 0.159 0.387 0.455
// "/runtime 0.000 0.000 0.146 0.166
// "/kubelet" 0.036 0.053 0.091 0.154
// "/misc" 0.001 0.001 0.001 0.002
var summaryStrings []string
var header []string
header = append(header, "container")
for _, p := range percentiles {
header = append(header, fmt.Sprintf("%.0fth%%", p*100))
}
for nodeName, containers := range summary {
buf := &bytes.Buffer{}
w := tabwriter.NewWriter(buf, 1, 0, 1, ' ', 0)
fmt.Fprintf(w, "%s\n", strings.Join(header, "\t"))
for _, containerName := range TargetContainers() {
var s []string
s = append(s, fmt.Sprintf("%q", containerName))
data, ok := containers[containerName]
for _, p := range percentiles {
value := "N/A"
if ok {
value = fmt.Sprintf("%.3f", data[p])
}
s = append(s, value)
}
fmt.Fprintf(w, "%s\n", strings.Join(s, "\t"))
}
w.Flush()
summaryStrings = append(summaryStrings, fmt.Sprintf("CPU usage of containers on node %q\n:%s", nodeName, buf.String()))
}
return strings.Join(summaryStrings, "\n")
}
func (r *ResourceMonitor) LogCPUSummary() {
summary := r.GetCPUSummary()
Logf("%s", r.FormatCPUSummary(summary))
}
func (r *ResourceMonitor) GetCPUSummary() NodesCPUSummary {
result := make(NodesCPUSummary)
for nodeName, collector := range r.collectors {
result[nodeName] = make(ContainersCPUSummary)
for _, containerName := range TargetContainers() {
data := collector.GetBasicCPUStats(containerName)
result[nodeName][containerName] = data
}
}
return result
}
func (r *ResourceMonitor) GetMasterNodeCPUSummary(summaryPerNode NodesCPUSummary) NodesCPUSummary {
result := make(NodesCPUSummary)
var masterSummary ContainersCPUSummary
var nodesSummaries []ContainersCPUSummary
for node, summary := range summaryPerNode {
if strings.HasSuffix(node, "master") {
masterSummary = summary
} else {
nodesSummaries = append(nodesSummaries, summary)
}
}
nodeAvgSummary := make(ContainersCPUSummary)
for _, nodeSummary := range nodesSummaries {
for c, summary := range nodeSummary {
if _, found := nodeAvgSummary[c]; !found {
nodeAvgSummary[c] = map[float64]float64{}
}
for perc, value := range summary {
nodeAvgSummary[c][perc] += value
}
}
}
for c := range nodeAvgSummary {
for perc := range nodeAvgSummary[c] {
nodeAvgSummary[c][perc] /= float64(len(nodesSummaries))
}
}
result["master"] = masterSummary
result["node"] = nodeAvgSummary
return result
}

View File

@ -0,0 +1,277 @@
/*
Copyright 2015 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 framework
import (
"bytes"
"fmt"
"strconv"
"strings"
"sync"
"text/tabwriter"
"time"
clientset "k8s.io/client-go/kubernetes"
)
const (
// Minimal period between polling log sizes from components
pollingPeriod = 60 * time.Second
workersNo = 5
kubeletLogsPath = "/var/log/kubelet.log"
kubeProxyLogsPath = "/var/log/kube-proxy.log"
kubeAddonsLogsPath = "/var/log/kube-addons.log"
kubeMasterAddonsLogsPath = "/var/log/kube-master-addons.log"
apiServerLogsPath = "/var/log/kube-apiserver.log"
controllersLogsPath = "/var/log/kube-controller-manager.log"
schedulerLogsPath = "/var/log/kube-scheduler.log"
)
var (
nodeLogsToCheck = []string{kubeletLogsPath, kubeProxyLogsPath}
masterLogsToCheck = []string{kubeletLogsPath, kubeAddonsLogsPath, kubeMasterAddonsLogsPath,
apiServerLogsPath, controllersLogsPath, schedulerLogsPath}
)
// TimestampedSize contains a size together with a time of measurement.
type TimestampedSize struct {
timestamp time.Time
size int
}
// LogSizeGatherer is a worker which grabs a WorkItem from the channel and does assigned work.
type LogSizeGatherer struct {
stopChannel chan bool
data *LogsSizeData
wg *sync.WaitGroup
workChannel chan WorkItem
}
// LogsSizeVerifier gathers data about log files sizes from master and node machines.
// It oversees a <workersNo> workers which do the gathering.
type LogsSizeVerifier struct {
client clientset.Interface
stopChannel chan bool
// data stores LogSizeData groupped per IP and log_path
data *LogsSizeData
masterAddress string
nodeAddresses []string
wg sync.WaitGroup
workChannel chan WorkItem
workers []*LogSizeGatherer
}
type SingleLogSummary struct {
AverageGenerationRate int
NumberOfProbes int
}
type LogSizeDataTimeseries map[string]map[string][]TimestampedSize
// node -> file -> data
type LogsSizeDataSummary map[string]map[string]SingleLogSummary
// TODO: make sure that we don't need locking here
func (s *LogsSizeDataSummary) PrintHumanReadable() string {
buf := &bytes.Buffer{}
w := tabwriter.NewWriter(buf, 1, 0, 1, ' ', 0)
fmt.Fprintf(w, "host\tlog_file\taverage_rate (B/s)\tnumber_of_probes\n")
for k, v := range *s {
fmt.Fprintf(w, "%v\t\t\t\n", k)
for path, data := range v {
fmt.Fprintf(w, "\t%v\t%v\t%v\n", path, data.AverageGenerationRate, data.NumberOfProbes)
}
}
w.Flush()
return buf.String()
}
func (s *LogsSizeDataSummary) PrintJSON() string {
return PrettyPrintJSON(*s)
}
func (s *LogsSizeDataSummary) SummaryKind() string {
return "LogSizeSummary"
}
type LogsSizeData struct {
data LogSizeDataTimeseries
lock sync.Mutex
}
// WorkItem is a command for a worker that contains an IP of machine from which we want to
// gather data and paths to all files we're interested in.
type WorkItem struct {
ip string
paths []string
backoffMultiplier int
}
func prepareData(masterAddress string, nodeAddresses []string) *LogsSizeData {
data := make(LogSizeDataTimeseries)
ips := append(nodeAddresses, masterAddress)
for _, ip := range ips {
data[ip] = make(map[string][]TimestampedSize)
}
return &LogsSizeData{
data: data,
lock: sync.Mutex{},
}
}
func (d *LogsSizeData) AddNewData(ip, path string, timestamp time.Time, size int) {
d.lock.Lock()
defer d.lock.Unlock()
d.data[ip][path] = append(
d.data[ip][path],
TimestampedSize{
timestamp: timestamp,
size: size,
},
)
}
// NewLogsVerifier creates a new LogsSizeVerifier which will stop when stopChannel is closed
func NewLogsVerifier(c clientset.Interface, stopChannel chan bool) *LogsSizeVerifier {
nodeAddresses, err := NodeSSHHosts(c)
ExpectNoError(err)
masterAddress := GetMasterHost() + ":22"
workChannel := make(chan WorkItem, len(nodeAddresses)+1)
workers := make([]*LogSizeGatherer, workersNo)
verifier := &LogsSizeVerifier{
client: c,
stopChannel: stopChannel,
data: prepareData(masterAddress, nodeAddresses),
masterAddress: masterAddress,
nodeAddresses: nodeAddresses,
wg: sync.WaitGroup{},
workChannel: workChannel,
workers: workers,
}
verifier.wg.Add(workersNo)
for i := 0; i < workersNo; i++ {
workers[i] = &LogSizeGatherer{
stopChannel: stopChannel,
data: verifier.data,
wg: &verifier.wg,
workChannel: workChannel,
}
}
return verifier
}
// GetSummary returns a summary (average generation rate and number of probes) of the data gathered by LogSizeVerifier
func (s *LogsSizeVerifier) GetSummary() *LogsSizeDataSummary {
result := make(LogsSizeDataSummary)
for k, v := range s.data.data {
result[k] = make(map[string]SingleLogSummary)
for path, data := range v {
if len(data) > 1 {
last := data[len(data)-1]
first := data[0]
rate := (last.size - first.size) / int(last.timestamp.Sub(first.timestamp)/time.Second)
result[k][path] = SingleLogSummary{
AverageGenerationRate: rate,
NumberOfProbes: len(data),
}
}
}
}
return &result
}
// Run starts log size gathering. It starts a gorouting for every worker and then blocks until stopChannel is closed
func (v *LogsSizeVerifier) Run() {
v.workChannel <- WorkItem{
ip: v.masterAddress,
paths: masterLogsToCheck,
backoffMultiplier: 1,
}
for _, node := range v.nodeAddresses {
v.workChannel <- WorkItem{
ip: node,
paths: nodeLogsToCheck,
backoffMultiplier: 1,
}
}
for _, worker := range v.workers {
go worker.Run()
}
<-v.stopChannel
v.wg.Wait()
}
func (g *LogSizeGatherer) Run() {
for g.Work() {
}
}
func (g *LogSizeGatherer) pushWorkItem(workItem WorkItem) {
select {
case <-time.After(time.Duration(workItem.backoffMultiplier) * pollingPeriod):
g.workChannel <- workItem
case <-g.stopChannel:
return
}
}
// Work does a single unit of work: tries to take out a WorkItem from the queue, ssh-es into a given machine,
// gathers data, writes it to the shared <data> map, and creates a gorouting which reinserts work item into
// the queue with a <pollingPeriod> delay. Returns false if worker should exit.
func (g *LogSizeGatherer) Work() bool {
var workItem WorkItem
select {
case <-g.stopChannel:
g.wg.Done()
return false
case workItem = <-g.workChannel:
}
sshResult, err := SSH(
fmt.Sprintf("ls -l %v | awk '{print $9, $5}' | tr '\n' ' '", strings.Join(workItem.paths, " ")),
workItem.ip,
TestContext.Provider,
)
if err != nil {
Logf("Error while trying to SSH to %v, skipping probe. Error: %v", workItem.ip, err)
// In case of repeated error give up.
if workItem.backoffMultiplier >= 128 {
Logf("Failed to ssh to a node %v multiple times in a row. Giving up.", workItem.ip)
g.wg.Done()
return false
}
workItem.backoffMultiplier *= 2
go g.pushWorkItem(workItem)
return true
}
workItem.backoffMultiplier = 1
results := strings.Split(sshResult.Stdout, " ")
now := time.Now()
for i := 0; i+1 < len(results); i = i + 2 {
path := results[i]
size, err := strconv.Atoi(results[i+1])
if err != nil {
Logf("Error during conversion to int: %v, skipping data. Error: %v", results[i+1], err)
continue
}
g.data.AddNewData(workItem.ip, path, now, size)
}
go g.pushWorkItem(workItem)
return true
}

View File

@ -0,0 +1,44 @@
package(default_visibility = ["//visibility:public"])
load(
"@io_bazel_rules_go//go:def.bzl",
"go_library",
)
go_library(
name = "go_default_library",
srcs = [
"api_server_metrics.go",
"cluster_autoscaler_metrics.go",
"controller_manager_metrics.go",
"generic_metrics.go",
"kubelet_metrics.go",
"metrics_grabber.go",
"scheduler_metrics.go",
],
importpath = "k8s.io/kubernetes/test/e2e/framework/metrics",
deps = [
"//pkg/apis/core:go_default_library",
"//pkg/master/ports:go_default_library",
"//pkg/util/system:go_default_library",
"//vendor/github.com/golang/glog:go_default_library",
"//vendor/github.com/prometheus/common/expfmt:go_default_library",
"//vendor/github.com/prometheus/common/model:go_default_library",
"//vendor/k8s.io/apimachinery/pkg/apis/meta/v1:go_default_library",
"//vendor/k8s.io/apimachinery/pkg/fields:go_default_library",
"//vendor/k8s.io/client-go/kubernetes:go_default_library",
],
)
filegroup(
name = "package-srcs",
srcs = glob(["**"]),
tags = ["automanaged"],
visibility = ["//visibility:private"],
)
filegroup(
name = "all-srcs",
srcs = [":package-srcs"],
tags = ["automanaged"],
)

View File

@ -0,0 +1,44 @@
/*
Copyright 2015 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 metrics
type ApiServerMetrics Metrics
func (m *ApiServerMetrics) Equal(o ApiServerMetrics) bool {
return (*Metrics)(m).Equal(Metrics(o))
}
func NewApiServerMetrics() ApiServerMetrics {
result := NewMetrics()
return ApiServerMetrics(result)
}
func parseApiServerMetrics(data string) (ApiServerMetrics, error) {
result := NewApiServerMetrics()
if err := parseMetrics(data, (*Metrics)(&result)); err != nil {
return ApiServerMetrics{}, err
}
return result, nil
}
func (g *MetricsGrabber) getMetricsFromApiServer() (string, error) {
rawOutput, err := g.client.CoreV1().RESTClient().Get().RequestURI("/metrics").Do().Raw()
if err != nil {
return "", err
}
return string(rawOutput), nil
}

View File

@ -0,0 +1,36 @@
/*
Copyright 2015 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 metrics
type ClusterAutoscalerMetrics Metrics
func (m *ClusterAutoscalerMetrics) Equal(o ClusterAutoscalerMetrics) bool {
return (*Metrics)(m).Equal(Metrics(o))
}
func NewClusterAutoscalerMetrics() ClusterAutoscalerMetrics {
result := NewMetrics()
return ClusterAutoscalerMetrics(result)
}
func parseClusterAutoscalerMetrics(data string) (ClusterAutoscalerMetrics, error) {
result := NewClusterAutoscalerMetrics()
if err := parseMetrics(data, (*Metrics)(&result)); err != nil {
return ClusterAutoscalerMetrics{}, err
}
return result, nil
}

View File

@ -0,0 +1,36 @@
/*
Copyright 2015 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 metrics
type ControllerManagerMetrics Metrics
func (m *ControllerManagerMetrics) Equal(o ControllerManagerMetrics) bool {
return (*Metrics)(m).Equal(Metrics(o))
}
func NewControllerManagerMetrics() ControllerManagerMetrics {
result := NewMetrics()
return ControllerManagerMetrics(result)
}
func parseControllerManagerMetrics(data string) (ControllerManagerMetrics, error) {
result := NewControllerManagerMetrics()
if err := parseMetrics(data, (*Metrics)(&result)); err != nil {
return ControllerManagerMetrics{}, err
}
return result, nil
}

View File

@ -0,0 +1,99 @@
/*
Copyright 2015 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 metrics
import (
"fmt"
"io"
"reflect"
"strings"
"github.com/golang/glog"
"github.com/prometheus/common/expfmt"
"github.com/prometheus/common/model"
)
type Metrics map[string]model.Samples
func (m *Metrics) Equal(o Metrics) bool {
leftKeySet := []string{}
rightKeySet := []string{}
for k := range *m {
leftKeySet = append(leftKeySet, k)
}
for k := range o {
rightKeySet = append(rightKeySet, k)
}
if !reflect.DeepEqual(leftKeySet, rightKeySet) {
return false
}
for _, k := range leftKeySet {
if !(*m)[k].Equal(o[k]) {
return false
}
}
return true
}
func PrintSample(sample *model.Sample) string {
buf := make([]string, 0)
// Id is a VERY special label. For 'normal' container it's useless, but it's necessary
// for 'system' containers (e.g. /docker-daemon, /kubelet, etc.). We know if that's the
// case by checking if there's a label "kubernetes_container_name" present. It's hacky
// but it works...
_, normalContainer := sample.Metric["kubernetes_container_name"]
for k, v := range sample.Metric {
if strings.HasPrefix(string(k), "__") {
continue
}
if string(k) == "id" && normalContainer {
continue
}
buf = append(buf, fmt.Sprintf("%v=%v", string(k), v))
}
return fmt.Sprintf("[%v] = %v", strings.Join(buf, ","), sample.Value)
}
func NewMetrics() Metrics {
result := make(Metrics)
return result
}
func parseMetrics(data string, output *Metrics) error {
dec := expfmt.NewDecoder(strings.NewReader(data), expfmt.FmtText)
decoder := expfmt.SampleDecoder{
Dec: dec,
Opts: &expfmt.DecodeOptions{},
}
for {
var v model.Vector
if err := decoder.Decode(&v); err != nil {
if err == io.EOF {
// Expected loop termination condition.
return nil
}
glog.Warningf("Invalid Decode. Skipping.")
continue
}
for _, metric := range v {
name := string(metric.Metric[model.MetricNameLabel])
(*output)[name] = append((*output)[name], metric)
}
}
}

View File

@ -0,0 +1,85 @@
/*
Copyright 2015 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 metrics
import (
"fmt"
"io/ioutil"
"net/http"
"time"
)
type KubeletMetrics Metrics
func (m *KubeletMetrics) Equal(o KubeletMetrics) bool {
return (*Metrics)(m).Equal(Metrics(o))
}
func NewKubeletMetrics() KubeletMetrics {
result := NewMetrics()
return KubeletMetrics(result)
}
// GrabKubeletMetricsWithoutProxy retrieve metrics from the kubelet on the given node using a simple GET over http.
// Currently only used in integration tests.
func GrabKubeletMetricsWithoutProxy(nodeName string) (KubeletMetrics, error) {
metricsEndpoint := "http://%s/metrics"
resp, err := http.Get(fmt.Sprintf(metricsEndpoint, nodeName))
if err != nil {
return KubeletMetrics{}, err
}
defer resp.Body.Close()
body, err := ioutil.ReadAll(resp.Body)
if err != nil {
return KubeletMetrics{}, err
}
return parseKubeletMetrics(string(body))
}
func parseKubeletMetrics(data string) (KubeletMetrics, error) {
result := NewKubeletMetrics()
if err := parseMetrics(data, (*Metrics)(&result)); err != nil {
return KubeletMetrics{}, err
}
return result, nil
}
func (g *MetricsGrabber) getMetricsFromNode(nodeName string, kubeletPort int) (string, error) {
// There's a problem with timing out during proxy. Wrapping this in a goroutine to prevent deadlock.
// Hanging goroutine will be leaked.
finished := make(chan struct{})
var err error
var rawOutput []byte
go func() {
rawOutput, err = g.client.CoreV1().RESTClient().Get().
Resource("nodes").
SubResource("proxy").
Name(fmt.Sprintf("%v:%v", nodeName, kubeletPort)).
Suffix("metrics").
Do().Raw()
finished <- struct{}{}
}()
select {
case <-time.After(ProxyTimeout):
return "", fmt.Errorf("Timed out when waiting for proxy to gather metrics from %v", nodeName)
case <-finished:
if err != nil {
return "", err
}
return string(rawOutput), nil
}
}

View File

@ -0,0 +1,245 @@
/*
Copyright 2015 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 metrics
import (
"fmt"
"time"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/fields"
clientset "k8s.io/client-go/kubernetes"
api "k8s.io/kubernetes/pkg/apis/core"
"k8s.io/kubernetes/pkg/master/ports"
"k8s.io/kubernetes/pkg/util/system"
"github.com/golang/glog"
)
const (
ProxyTimeout = 2 * time.Minute
)
type MetricsCollection struct {
ApiServerMetrics ApiServerMetrics
ControllerManagerMetrics ControllerManagerMetrics
KubeletMetrics map[string]KubeletMetrics
SchedulerMetrics SchedulerMetrics
ClusterAutoscalerMetrics ClusterAutoscalerMetrics
}
type MetricsGrabber struct {
client clientset.Interface
externalClient clientset.Interface
grabFromApiServer bool
grabFromControllerManager bool
grabFromKubelets bool
grabFromScheduler bool
grabFromClusterAutoscaler bool
masterName string
registeredMaster bool
}
func NewMetricsGrabber(c clientset.Interface, ec clientset.Interface, kubelets bool, scheduler bool, controllers bool, apiServer bool, clusterAutoscaler bool) (*MetricsGrabber, error) {
registeredMaster := false
masterName := ""
nodeList, err := c.CoreV1().Nodes().List(metav1.ListOptions{})
if err != nil {
return nil, err
}
if len(nodeList.Items) < 1 {
glog.Warning("Can't find any Nodes in the API server to grab metrics from")
}
for _, node := range nodeList.Items {
if system.IsMasterNode(node.Name) {
registeredMaster = true
masterName = node.Name
break
}
}
if !registeredMaster {
scheduler = false
controllers = false
clusterAutoscaler = ec != nil
if clusterAutoscaler {
glog.Warningf("Master node is not registered. Grabbing metrics from Scheduler, ControllerManager is disabled.")
} else {
glog.Warningf("Master node is not registered. Grabbing metrics from Scheduler, ControllerManager and ClusterAutoscaler is disabled.")
}
}
return &MetricsGrabber{
client: c,
externalClient: ec,
grabFromApiServer: apiServer,
grabFromControllerManager: controllers,
grabFromKubelets: kubelets,
grabFromScheduler: scheduler,
grabFromClusterAutoscaler: clusterAutoscaler,
masterName: masterName,
registeredMaster: registeredMaster,
}, nil
}
// HasRegisteredMaster returns if metrics grabber was able to find a master node
func (g *MetricsGrabber) HasRegisteredMaster() bool {
return g.registeredMaster
}
func (g *MetricsGrabber) GrabFromKubelet(nodeName string) (KubeletMetrics, error) {
nodes, err := g.client.CoreV1().Nodes().List(metav1.ListOptions{FieldSelector: fields.Set{api.ObjectNameField: nodeName}.AsSelector().String()})
if err != nil {
return KubeletMetrics{}, err
}
if len(nodes.Items) != 1 {
return KubeletMetrics{}, fmt.Errorf("Error listing nodes with name %v, got %v", nodeName, nodes.Items)
}
kubeletPort := nodes.Items[0].Status.DaemonEndpoints.KubeletEndpoint.Port
return g.grabFromKubeletInternal(nodeName, int(kubeletPort))
}
func (g *MetricsGrabber) grabFromKubeletInternal(nodeName string, kubeletPort int) (KubeletMetrics, error) {
if kubeletPort <= 0 || kubeletPort > 65535 {
return KubeletMetrics{}, fmt.Errorf("Invalid Kubelet port %v. Skipping Kubelet's metrics gathering.", kubeletPort)
}
output, err := g.getMetricsFromNode(nodeName, int(kubeletPort))
if err != nil {
return KubeletMetrics{}, err
}
return parseKubeletMetrics(output)
}
func (g *MetricsGrabber) GrabFromScheduler() (SchedulerMetrics, error) {
if !g.registeredMaster {
return SchedulerMetrics{}, fmt.Errorf("Master's Kubelet is not registered. Skipping Scheduler's metrics gathering.")
}
output, err := g.getMetricsFromPod(g.client, fmt.Sprintf("%v-%v", "kube-scheduler", g.masterName), metav1.NamespaceSystem, ports.SchedulerPort)
if err != nil {
return SchedulerMetrics{}, err
}
return parseSchedulerMetrics(output)
}
func (g *MetricsGrabber) GrabFromClusterAutoscaler() (ClusterAutoscalerMetrics, error) {
if !g.registeredMaster && g.externalClient == nil {
return ClusterAutoscalerMetrics{}, fmt.Errorf("Master's Kubelet is not registered. Skipping ClusterAutoscaler's metrics gathering.")
}
var client clientset.Interface
var namespace string
if g.externalClient != nil {
client = g.externalClient
namespace = "kubemark"
} else {
client = g.client
namespace = metav1.NamespaceSystem
}
output, err := g.getMetricsFromPod(client, "cluster-autoscaler", namespace, 8085)
if err != nil {
return ClusterAutoscalerMetrics{}, err
}
return parseClusterAutoscalerMetrics(output)
}
func (g *MetricsGrabber) GrabFromControllerManager() (ControllerManagerMetrics, error) {
if !g.registeredMaster {
return ControllerManagerMetrics{}, fmt.Errorf("Master's Kubelet is not registered. Skipping ControllerManager's metrics gathering.")
}
output, err := g.getMetricsFromPod(g.client, fmt.Sprintf("%v-%v", "kube-controller-manager", g.masterName), metav1.NamespaceSystem, ports.ControllerManagerPort)
if err != nil {
return ControllerManagerMetrics{}, err
}
return parseControllerManagerMetrics(output)
}
func (g *MetricsGrabber) GrabFromApiServer() (ApiServerMetrics, error) {
output, err := g.getMetricsFromApiServer()
if err != nil {
return ApiServerMetrics{}, nil
}
return parseApiServerMetrics(output)
}
func (g *MetricsGrabber) Grab() (MetricsCollection, error) {
result := MetricsCollection{}
var errs []error
if g.grabFromApiServer {
metrics, err := g.GrabFromApiServer()
if err != nil {
errs = append(errs, err)
} else {
result.ApiServerMetrics = metrics
}
}
if g.grabFromScheduler {
metrics, err := g.GrabFromScheduler()
if err != nil {
errs = append(errs, err)
} else {
result.SchedulerMetrics = metrics
}
}
if g.grabFromControllerManager {
metrics, err := g.GrabFromControllerManager()
if err != nil {
errs = append(errs, err)
} else {
result.ControllerManagerMetrics = metrics
}
}
if g.grabFromClusterAutoscaler {
metrics, err := g.GrabFromClusterAutoscaler()
if err != nil {
errs = append(errs, err)
} else {
result.ClusterAutoscalerMetrics = metrics
}
}
if g.grabFromKubelets {
result.KubeletMetrics = make(map[string]KubeletMetrics)
nodes, err := g.client.CoreV1().Nodes().List(metav1.ListOptions{})
if err != nil {
errs = append(errs, err)
} else {
for _, node := range nodes.Items {
kubeletPort := node.Status.DaemonEndpoints.KubeletEndpoint.Port
metrics, err := g.grabFromKubeletInternal(node.Name, int(kubeletPort))
if err != nil {
errs = append(errs, err)
}
result.KubeletMetrics[node.Name] = metrics
}
}
}
if len(errs) > 0 {
return result, fmt.Errorf("Errors while grabbing metrics: %v", errs)
}
return result, nil
}
func (g *MetricsGrabber) getMetricsFromPod(client clientset.Interface, podName string, namespace string, port int) (string, error) {
rawOutput, err := client.CoreV1().RESTClient().Get().
Namespace(namespace).
Resource("pods").
SubResource("proxy").
Name(fmt.Sprintf("%v:%v", podName, port)).
Suffix("metrics").
Do().Raw()
if err != nil {
return "", err
}
return string(rawOutput), nil
}

View File

@ -0,0 +1,36 @@
/*
Copyright 2015 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 metrics
type SchedulerMetrics Metrics
func (m *SchedulerMetrics) Equal(o SchedulerMetrics) bool {
return (*Metrics)(m).Equal(Metrics(o))
}
func NewSchedulerMetrics() SchedulerMetrics {
result := NewMetrics()
return SchedulerMetrics(result)
}
func parseSchedulerMetrics(data string) (SchedulerMetrics, error) {
result := NewSchedulerMetrics()
if err := parseMetrics(data, (*Metrics)(&result)); err != nil {
return SchedulerMetrics{}, err
}
return result, nil
}

View File

@ -0,0 +1,621 @@
/*
Copyright 2015 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 framework
import (
"bytes"
"context"
"encoding/json"
"fmt"
"io"
"math"
"sort"
"strconv"
"strings"
"time"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/util/sets"
clientset "k8s.io/client-go/kubernetes"
"k8s.io/kubernetes/pkg/master/ports"
"k8s.io/kubernetes/pkg/util/system"
"k8s.io/kubernetes/test/e2e/framework/metrics"
"github.com/prometheus/common/expfmt"
"github.com/prometheus/common/model"
)
const (
// NodeStartupThreshold is a rough estimate of the time allocated for a pod to start on a node.
NodeStartupThreshold = 4 * time.Second
podStartupThreshold time.Duration = 5 * time.Second
// We are setting 1s threshold for apicalls even in small clusters to avoid flakes.
// The problem is that if long GC is happening in small clusters (where we have e.g.
// 1-core master machines) and tests are pretty short, it may consume significant
// portion of CPU and basically stop all the real work.
// Increasing threshold to 1s is within our SLO and should solve this problem.
apiCallLatencyThreshold time.Duration = 1 * time.Second
// We use a higher threshold for list apicalls if the cluster is big (i.e having > 500 nodes)
// as list response sizes are bigger in general for big clusters. We also use a higher threshold
// for list calls at cluster scope (this includes non-namespaced and all-namespaced calls).
apiListCallLatencyThreshold time.Duration = 5 * time.Second
apiClusterScopeListCallThreshold time.Duration = 10 * time.Second
bigClusterNodeCountThreshold = 500
// Cluster Autoscaler metrics names
caFunctionMetric = "cluster_autoscaler_function_duration_seconds_bucket"
caFunctionMetricLabel = "function"
)
type MetricsForE2E metrics.MetricsCollection
func (m *MetricsForE2E) filterMetrics() {
interestingApiServerMetrics := make(metrics.ApiServerMetrics)
for _, metric := range InterestingApiServerMetrics {
interestingApiServerMetrics[metric] = (*m).ApiServerMetrics[metric]
}
interestingControllerManagerMetrics := make(metrics.ControllerManagerMetrics)
for _, metric := range InterestingControllerManagerMetrics {
interestingControllerManagerMetrics[metric] = (*m).ControllerManagerMetrics[metric]
}
interestingClusterAutoscalerMetrics := make(metrics.ClusterAutoscalerMetrics)
for _, metric := range InterestingClusterAutoscalerMetrics {
interestingClusterAutoscalerMetrics[metric] = (*m).ClusterAutoscalerMetrics[metric]
}
interestingKubeletMetrics := make(map[string]metrics.KubeletMetrics)
for kubelet, grabbed := range (*m).KubeletMetrics {
interestingKubeletMetrics[kubelet] = make(metrics.KubeletMetrics)
for _, metric := range InterestingKubeletMetrics {
interestingKubeletMetrics[kubelet][metric] = grabbed[metric]
}
}
(*m).ApiServerMetrics = interestingApiServerMetrics
(*m).ControllerManagerMetrics = interestingControllerManagerMetrics
(*m).KubeletMetrics = interestingKubeletMetrics
}
func (m *MetricsForE2E) PrintHumanReadable() string {
buf := bytes.Buffer{}
for _, interestingMetric := range InterestingApiServerMetrics {
buf.WriteString(fmt.Sprintf("For %v:\n", interestingMetric))
for _, sample := range (*m).ApiServerMetrics[interestingMetric] {
buf.WriteString(fmt.Sprintf("\t%v\n", metrics.PrintSample(sample)))
}
}
for _, interestingMetric := range InterestingControllerManagerMetrics {
buf.WriteString(fmt.Sprintf("For %v:\n", interestingMetric))
for _, sample := range (*m).ControllerManagerMetrics[interestingMetric] {
buf.WriteString(fmt.Sprintf("\t%v\n", metrics.PrintSample(sample)))
}
}
for _, interestingMetric := range InterestingClusterAutoscalerMetrics {
buf.WriteString(fmt.Sprintf("For %v:\n", interestingMetric))
for _, sample := range (*m).ClusterAutoscalerMetrics[interestingMetric] {
buf.WriteString(fmt.Sprintf("\t%v\n", metrics.PrintSample(sample)))
}
}
for kubelet, grabbed := range (*m).KubeletMetrics {
buf.WriteString(fmt.Sprintf("For %v:\n", kubelet))
for _, interestingMetric := range InterestingKubeletMetrics {
buf.WriteString(fmt.Sprintf("\tFor %v:\n", interestingMetric))
for _, sample := range grabbed[interestingMetric] {
buf.WriteString(fmt.Sprintf("\t\t%v\n", metrics.PrintSample(sample)))
}
}
}
return buf.String()
}
func (m *MetricsForE2E) PrintJSON() string {
m.filterMetrics()
return PrettyPrintJSON(m)
}
func (m *MetricsForE2E) SummaryKind() string {
return "MetricsForE2E"
}
var InterestingApiServerMetrics = []string{
"apiserver_request_count",
"apiserver_request_latencies_summary",
"etcd_helper_cache_entry_count",
"etcd_helper_cache_hit_count",
"etcd_helper_cache_miss_count",
"etcd_request_cache_add_latencies_summary",
"etcd_request_cache_get_latencies_summary",
"etcd_request_latencies_summary",
}
var InterestingControllerManagerMetrics = []string{
"garbage_collector_attempt_to_delete_queue_latency",
"garbage_collector_attempt_to_delete_work_duration",
"garbage_collector_attempt_to_orphan_queue_latency",
"garbage_collector_attempt_to_orphan_work_duration",
"garbage_collector_dirty_processing_latency_microseconds",
"garbage_collector_event_processing_latency_microseconds",
"garbage_collector_graph_changes_queue_latency",
"garbage_collector_graph_changes_work_duration",
"garbage_collector_orphan_processing_latency_microseconds",
"namespace_queue_latency",
"namespace_queue_latency_sum",
"namespace_queue_latency_count",
"namespace_retries",
"namespace_work_duration",
"namespace_work_duration_sum",
"namespace_work_duration_count",
}
var InterestingKubeletMetrics = []string{
"kubelet_container_manager_latency_microseconds",
"kubelet_docker_errors",
"kubelet_docker_operations_latency_microseconds",
"kubelet_generate_pod_status_latency_microseconds",
"kubelet_pod_start_latency_microseconds",
"kubelet_pod_worker_latency_microseconds",
"kubelet_pod_worker_start_latency_microseconds",
"kubelet_sync_pods_latency_microseconds",
}
var InterestingClusterAutoscalerMetrics = []string{
"function_duration_seconds",
"errors_total",
"evicted_pods_total",
}
// Dashboard metrics
type LatencyMetric struct {
Perc50 time.Duration `json:"Perc50"`
Perc90 time.Duration `json:"Perc90"`
Perc99 time.Duration `json:"Perc99"`
Perc100 time.Duration `json:"Perc100"`
}
type PodStartupLatency struct {
Latency LatencyMetric `json:"latency"`
}
func (l *PodStartupLatency) SummaryKind() string {
return "PodStartupLatency"
}
func (l *PodStartupLatency) PrintHumanReadable() string {
return PrettyPrintJSON(l)
}
func (l *PodStartupLatency) PrintJSON() string {
return PrettyPrintJSON(PodStartupLatencyToPerfData(l))
}
type SchedulingLatency struct {
Scheduling LatencyMetric `json:"scheduling"`
Binding LatencyMetric `json:"binding"`
Total LatencyMetric `json:"total"`
}
func (l *SchedulingLatency) SummaryKind() string {
return "SchedulingLatency"
}
func (l *SchedulingLatency) PrintHumanReadable() string {
return PrettyPrintJSON(l)
}
func (l *SchedulingLatency) PrintJSON() string {
return PrettyPrintJSON(l)
}
type SaturationTime struct {
TimeToSaturate time.Duration `json:"timeToSaturate"`
NumberOfNodes int `json:"numberOfNodes"`
NumberOfPods int `json:"numberOfPods"`
Throughput float32 `json:"throughput"`
}
type APICall struct {
Resource string `json:"resource"`
Subresource string `json:"subresource"`
Verb string `json:"verb"`
Scope string `json:"scope"`
Latency LatencyMetric `json:"latency"`
Count int `json:"count"`
}
type APIResponsiveness struct {
APICalls []APICall `json:"apicalls"`
}
func (a *APIResponsiveness) SummaryKind() string {
return "APIResponsiveness"
}
func (a *APIResponsiveness) PrintHumanReadable() string {
return PrettyPrintJSON(a)
}
func (a *APIResponsiveness) PrintJSON() string {
return PrettyPrintJSON(ApiCallToPerfData(a))
}
func (a *APIResponsiveness) Len() int { return len(a.APICalls) }
func (a *APIResponsiveness) Swap(i, j int) {
a.APICalls[i], a.APICalls[j] = a.APICalls[j], a.APICalls[i]
}
func (a *APIResponsiveness) Less(i, j int) bool {
return a.APICalls[i].Latency.Perc99 < a.APICalls[j].Latency.Perc99
}
// Set request latency for a particular quantile in the APICall metric entry (creating one if necessary).
// 0 <= quantile <=1 (e.g. 0.95 is 95%tile, 0.5 is median)
// Only 0.5, 0.9 and 0.99 quantiles are supported.
func (a *APIResponsiveness) addMetricRequestLatency(resource, subresource, verb, scope string, quantile float64, latency time.Duration) {
for i, apicall := range a.APICalls {
if apicall.Resource == resource && apicall.Subresource == subresource && apicall.Verb == verb && apicall.Scope == scope {
a.APICalls[i] = setQuantileAPICall(apicall, quantile, latency)
return
}
}
apicall := setQuantileAPICall(APICall{Resource: resource, Subresource: subresource, Verb: verb, Scope: scope}, quantile, latency)
a.APICalls = append(a.APICalls, apicall)
}
// 0 <= quantile <=1 (e.g. 0.95 is 95%tile, 0.5 is median)
// Only 0.5, 0.9 and 0.99 quantiles are supported.
func setQuantileAPICall(apicall APICall, quantile float64, latency time.Duration) APICall {
setQuantile(&apicall.Latency, quantile, latency)
return apicall
}
// Only 0.5, 0.9 and 0.99 quantiles are supported.
func setQuantile(metric *LatencyMetric, quantile float64, latency time.Duration) {
switch quantile {
case 0.5:
metric.Perc50 = latency
case 0.9:
metric.Perc90 = latency
case 0.99:
metric.Perc99 = latency
}
}
// Add request count to the APICall metric entry (creating one if necessary).
func (a *APIResponsiveness) addMetricRequestCount(resource, subresource, verb, scope string, count int) {
for i, apicall := range a.APICalls {
if apicall.Resource == resource && apicall.Subresource == subresource && apicall.Verb == verb && apicall.Scope == scope {
a.APICalls[i].Count += count
return
}
}
apicall := APICall{Resource: resource, Subresource: subresource, Verb: verb, Count: count, Scope: scope}
a.APICalls = append(a.APICalls, apicall)
}
func readLatencyMetrics(c clientset.Interface) (*APIResponsiveness, error) {
var a APIResponsiveness
body, err := getMetrics(c)
if err != nil {
return nil, err
}
samples, err := extractMetricSamples(body)
if err != nil {
return nil, err
}
ignoredResources := sets.NewString("events")
// TODO: figure out why we're getting non-capitalized proxy and fix this.
ignoredVerbs := sets.NewString("WATCH", "WATCHLIST", "PROXY", "proxy", "CONNECT")
for _, sample := range samples {
// Example line:
// apiserver_request_latencies_summary{resource="namespaces",verb="LIST",quantile="0.99"} 908
// apiserver_request_count{resource="pods",verb="LIST",client="kubectl",code="200",contentType="json"} 233
if sample.Metric[model.MetricNameLabel] != "apiserver_request_latencies_summary" &&
sample.Metric[model.MetricNameLabel] != "apiserver_request_count" {
continue
}
resource := string(sample.Metric["resource"])
subresource := string(sample.Metric["subresource"])
verb := string(sample.Metric["verb"])
scope := string(sample.Metric["scope"])
if ignoredResources.Has(resource) || ignoredVerbs.Has(verb) {
continue
}
switch sample.Metric[model.MetricNameLabel] {
case "apiserver_request_latencies_summary":
latency := sample.Value
quantile, err := strconv.ParseFloat(string(sample.Metric[model.QuantileLabel]), 64)
if err != nil {
return nil, err
}
a.addMetricRequestLatency(resource, subresource, verb, scope, quantile, time.Duration(int64(latency))*time.Microsecond)
case "apiserver_request_count":
count := sample.Value
a.addMetricRequestCount(resource, subresource, verb, scope, int(count))
}
}
return &a, err
}
// Prints top five summary metrics for request types with latency and returns
// number of such request types above threshold. We use a higher threshold for
// list calls if nodeCount is above a given threshold (i.e. cluster is big).
func HighLatencyRequests(c clientset.Interface, nodeCount int) (int, *APIResponsiveness, error) {
isBigCluster := (nodeCount > bigClusterNodeCountThreshold)
metrics, err := readLatencyMetrics(c)
if err != nil {
return 0, metrics, err
}
sort.Sort(sort.Reverse(metrics))
badMetrics := 0
top := 5
for i := range metrics.APICalls {
latency := metrics.APICalls[i].Latency.Perc99
isListCall := (metrics.APICalls[i].Verb == "LIST")
isClusterScopedCall := (metrics.APICalls[i].Scope == "cluster")
isBad := false
latencyThreshold := apiCallLatencyThreshold
if isListCall && isBigCluster {
latencyThreshold = apiListCallLatencyThreshold
if isClusterScopedCall {
latencyThreshold = apiClusterScopeListCallThreshold
}
}
if latency > latencyThreshold {
isBad = true
badMetrics++
}
if top > 0 || isBad {
top--
prefix := ""
if isBad {
prefix = "WARNING "
}
Logf("%vTop latency metric: %+v", prefix, metrics.APICalls[i])
}
}
return badMetrics, metrics, nil
}
// Verifies whether 50, 90 and 99th percentiles of PodStartupLatency are
// within the threshold.
func VerifyPodStartupLatency(latency *PodStartupLatency) error {
if latency.Latency.Perc50 > podStartupThreshold {
return fmt.Errorf("too high pod startup latency 50th percentile: %v", latency.Latency.Perc50)
}
if latency.Latency.Perc90 > podStartupThreshold {
return fmt.Errorf("too high pod startup latency 90th percentile: %v", latency.Latency.Perc90)
}
if latency.Latency.Perc99 > podStartupThreshold {
return fmt.Errorf("too high pod startup latency 99th percentile: %v", latency.Latency.Perc99)
}
return nil
}
// Resets latency metrics in apiserver.
func ResetMetrics(c clientset.Interface) error {
Logf("Resetting latency metrics in apiserver...")
body, err := c.CoreV1().RESTClient().Delete().AbsPath("/metrics").DoRaw()
if err != nil {
return err
}
if string(body) != "metrics reset\n" {
return fmt.Errorf("Unexpected response: %q", string(body))
}
return nil
}
// Retrieves metrics information.
func getMetrics(c clientset.Interface) (string, error) {
body, err := c.CoreV1().RESTClient().Get().AbsPath("/metrics").DoRaw()
if err != nil {
return "", err
}
return string(body), nil
}
// Retrieves scheduler metrics information.
func getSchedulingLatency(c clientset.Interface) (*SchedulingLatency, error) {
result := SchedulingLatency{}
// Check if master Node is registered
nodes, err := c.CoreV1().Nodes().List(metav1.ListOptions{})
ExpectNoError(err)
var data string
var masterRegistered = false
for _, node := range nodes.Items {
if system.IsMasterNode(node.Name) {
masterRegistered = true
}
}
if masterRegistered {
ctx, cancel := context.WithTimeout(context.Background(), SingleCallTimeout)
defer cancel()
var rawData []byte
rawData, err = c.CoreV1().RESTClient().Get().
Context(ctx).
Namespace(metav1.NamespaceSystem).
Resource("pods").
Name(fmt.Sprintf("kube-scheduler-%v:%v", TestContext.CloudConfig.MasterName, ports.SchedulerPort)).
SubResource("proxy").
Suffix("metrics").
Do().Raw()
ExpectNoError(err)
data = string(rawData)
} else {
// If master is not registered fall back to old method of using SSH.
if TestContext.Provider == "gke" {
Logf("Not grabbing scheduler metrics through master SSH: unsupported for gke")
return nil, nil
}
cmd := "curl http://localhost:10251/metrics"
sshResult, err := SSH(cmd, GetMasterHost()+":22", TestContext.Provider)
if err != nil || sshResult.Code != 0 {
return &result, fmt.Errorf("unexpected error (code: %d) in ssh connection to master: %#v", sshResult.Code, err)
}
data = sshResult.Stdout
}
samples, err := extractMetricSamples(data)
if err != nil {
return nil, err
}
for _, sample := range samples {
var metric *LatencyMetric = nil
switch sample.Metric[model.MetricNameLabel] {
case "scheduler_scheduling_algorithm_latency_microseconds":
metric = &result.Scheduling
case "scheduler_binding_latency_microseconds":
metric = &result.Binding
case "scheduler_e2e_scheduling_latency_microseconds":
metric = &result.Total
}
if metric == nil {
continue
}
latency := sample.Value
quantile, err := strconv.ParseFloat(string(sample.Metric[model.QuantileLabel]), 64)
if err != nil {
return nil, err
}
setQuantile(metric, quantile, time.Duration(int64(latency))*time.Microsecond)
}
return &result, nil
}
// Verifies (currently just by logging them) the scheduling latencies.
func VerifySchedulerLatency(c clientset.Interface) (*SchedulingLatency, error) {
latency, err := getSchedulingLatency(c)
if err != nil {
return nil, err
}
return latency, nil
}
func PrettyPrintJSON(metrics interface{}) string {
output := &bytes.Buffer{}
if err := json.NewEncoder(output).Encode(metrics); err != nil {
Logf("Error building encoder: %v", err)
return ""
}
formatted := &bytes.Buffer{}
if err := json.Indent(formatted, output.Bytes(), "", " "); err != nil {
Logf("Error indenting: %v", err)
return ""
}
return string(formatted.Bytes())
}
// extractMetricSamples parses the prometheus metric samples from the input string.
func extractMetricSamples(metricsBlob string) ([]*model.Sample, error) {
dec := expfmt.NewDecoder(strings.NewReader(metricsBlob), expfmt.FmtText)
decoder := expfmt.SampleDecoder{
Dec: dec,
Opts: &expfmt.DecodeOptions{},
}
var samples []*model.Sample
for {
var v model.Vector
if err := decoder.Decode(&v); err != nil {
if err == io.EOF {
// Expected loop termination condition.
return samples, nil
}
return nil, err
}
samples = append(samples, v...)
}
}
// PodLatencyData encapsulates pod startup latency information.
type PodLatencyData struct {
// Name of the pod
Name string
// Node this pod was running on
Node string
// Latency information related to pod startuptime
Latency time.Duration
}
type LatencySlice []PodLatencyData
func (a LatencySlice) Len() int { return len(a) }
func (a LatencySlice) Swap(i, j int) { a[i], a[j] = a[j], a[i] }
func (a LatencySlice) Less(i, j int) bool { return a[i].Latency < a[j].Latency }
func ExtractLatencyMetrics(latencies []PodLatencyData) LatencyMetric {
length := len(latencies)
perc50 := latencies[int(math.Ceil(float64(length*50)/100))-1].Latency
perc90 := latencies[int(math.Ceil(float64(length*90)/100))-1].Latency
perc99 := latencies[int(math.Ceil(float64(length*99)/100))-1].Latency
perc100 := latencies[length-1].Latency
return LatencyMetric{Perc50: perc50, Perc90: perc90, Perc99: perc99, Perc100: perc100}
}
// LogSuspiciousLatency logs metrics/docker errors from all nodes that had slow startup times
// If latencyDataLag is nil then it will be populated from latencyData
func LogSuspiciousLatency(latencyData []PodLatencyData, latencyDataLag []PodLatencyData, nodeCount int, c clientset.Interface) {
if latencyDataLag == nil {
latencyDataLag = latencyData
}
for _, l := range latencyData {
if l.Latency > NodeStartupThreshold {
HighLatencyKubeletOperations(c, 1*time.Second, l.Node, Logf)
}
}
Logf("Approx throughput: %v pods/min",
float64(nodeCount)/(latencyDataLag[len(latencyDataLag)-1].Latency.Minutes()))
}
func PrintLatencies(latencies []PodLatencyData, header string) {
metrics := ExtractLatencyMetrics(latencies)
Logf("10%% %s: %v", header, latencies[(len(latencies)*9)/10:])
Logf("perc50: %v, perc90: %v, perc99: %v", metrics.Perc50, metrics.Perc90, metrics.Perc99)
}
func (m *MetricsForE2E) computeClusterAutoscalerMetricsDelta(before metrics.MetricsCollection) {
if beforeSamples, found := before.ClusterAutoscalerMetrics[caFunctionMetric]; found {
if afterSamples, found := m.ClusterAutoscalerMetrics[caFunctionMetric]; found {
beforeSamplesMap := make(map[string]*model.Sample)
for _, bSample := range beforeSamples {
beforeSamplesMap[makeKey(bSample.Metric[caFunctionMetricLabel], bSample.Metric["le"])] = bSample
}
for _, aSample := range afterSamples {
if bSample, found := beforeSamplesMap[makeKey(aSample.Metric[caFunctionMetricLabel], aSample.Metric["le"])]; found {
aSample.Value = aSample.Value - bSample.Value
}
}
}
}
}
func makeKey(a, b model.LabelValue) string {
return string(a) + "___" + string(b)
}

View File

@ -0,0 +1,977 @@
/*
Copyright 2014 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 framework
import (
"bytes"
"encoding/json"
"fmt"
"io/ioutil"
"net"
"net/http"
"strconv"
"strings"
"time"
. "github.com/onsi/ginkgo"
. "github.com/onsi/gomega"
"k8s.io/api/core/v1"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/labels"
"k8s.io/apimachinery/pkg/util/intstr"
utilnet "k8s.io/apimachinery/pkg/util/net"
"k8s.io/apimachinery/pkg/util/rand"
"k8s.io/apimachinery/pkg/util/sets"
"k8s.io/apimachinery/pkg/util/uuid"
"k8s.io/apimachinery/pkg/util/wait"
clientset "k8s.io/client-go/kubernetes"
coreclientset "k8s.io/client-go/kubernetes/typed/core/v1"
"k8s.io/kubernetes/pkg/api/testapi"
imageutils "k8s.io/kubernetes/test/utils/image"
)
const (
EndpointHttpPort = 8080
EndpointUdpPort = 8081
TestContainerHttpPort = 8080
ClusterHttpPort = 80
ClusterUdpPort = 90
testPodName = "test-container-pod"
hostTestPodName = "host-test-container-pod"
nodePortServiceName = "node-port-service"
sessionAffinityServiceName = "session-affinity-service"
// wait time between poll attempts of a Service vip and/or nodePort.
// coupled with testTries to produce a net timeout value.
hitEndpointRetryDelay = 2 * time.Second
// Number of retries to hit a given set of endpoints. Needs to be high
// because we verify iptables statistical rr loadbalancing.
testTries = 30
// Maximum number of pods in a test, to make test work in large clusters.
maxNetProxyPodsCount = 10
// Number of checks to hit a given set of endpoints when enable session affinity.
SessionAffinityChecks = 10
)
var NetexecImageName = imageutils.GetE2EImage(imageutils.Netexec)
// NewNetworkingTestConfig creates and sets up a new test config helper.
func NewNetworkingTestConfig(f *Framework) *NetworkingTestConfig {
config := &NetworkingTestConfig{f: f, Namespace: f.Namespace.Name}
By(fmt.Sprintf("Performing setup for networking test in namespace %v", config.Namespace))
config.setup(getServiceSelector())
return config
}
// NewNetworkingTestNodeE2EConfig creates and sets up a new test config helper for Node E2E.
func NewCoreNetworkingTestConfig(f *Framework) *NetworkingTestConfig {
config := &NetworkingTestConfig{f: f, Namespace: f.Namespace.Name}
By(fmt.Sprintf("Performing setup for networking test in namespace %v", config.Namespace))
config.setupCore(getServiceSelector())
return config
}
func getServiceSelector() map[string]string {
By("creating a selector")
selectorName := "selector-" + string(uuid.NewUUID())
serviceSelector := map[string]string{
selectorName: "true",
}
return serviceSelector
}
// NetworkingTestConfig is a convenience class around some utility methods
// for testing kubeproxy/networking/services/endpoints.
type NetworkingTestConfig struct {
// TestContaienrPod is a test pod running the netexec image. It is capable
// of executing tcp/udp requests against ip:port.
TestContainerPod *v1.Pod
// HostTestContainerPod is a pod running with hostNetworking=true, and the
// hostexec image.
HostTestContainerPod *v1.Pod
// EndpointPods are the pods belonging to the Service created by this
// test config. Each invocation of `setup` creates a service with
// 1 pod per node running the netexecImage.
EndpointPods []*v1.Pod
f *Framework
podClient *PodClient
// NodePortService is a Service with Type=NodePort spanning over all
// endpointPods.
NodePortService *v1.Service
// SessionAffinityService is a Service with SessionAffinity=ClientIP
// spanning over all endpointPods.
SessionAffinityService *v1.Service
// ExternalAddrs is a list of external IPs of nodes in the cluster.
ExternalAddrs []string
// Nodes is a list of nodes in the cluster.
Nodes []v1.Node
// MaxTries is the number of retries tolerated for tests run against
// endpoints and services created by this config.
MaxTries int
// The ClusterIP of the Service reated by this test config.
ClusterIP string
// External ip of first node for use in nodePort testing.
NodeIP string
// The http/udp nodePorts of the Service.
NodeHttpPort int
NodeUdpPort int
// The kubernetes namespace within which all resources for this
// config are created
Namespace string
}
func (config *NetworkingTestConfig) DialFromEndpointContainer(protocol, targetIP string, targetPort, maxTries, minTries int, expectedEps sets.String) {
config.DialFromContainer(protocol, config.EndpointPods[0].Status.PodIP, targetIP, EndpointHttpPort, targetPort, maxTries, minTries, expectedEps)
}
func (config *NetworkingTestConfig) DialFromTestContainer(protocol, targetIP string, targetPort, maxTries, minTries int, expectedEps sets.String) {
config.DialFromContainer(protocol, config.TestContainerPod.Status.PodIP, targetIP, TestContainerHttpPort, targetPort, maxTries, minTries, expectedEps)
}
// diagnoseMissingEndpoints prints debug information about the endpoints that
// are NOT in the given list of foundEndpoints. These are the endpoints we
// expected a response from.
func (config *NetworkingTestConfig) diagnoseMissingEndpoints(foundEndpoints sets.String) {
for _, e := range config.EndpointPods {
if foundEndpoints.Has(e.Name) {
continue
}
Logf("\nOutput of kubectl describe pod %v/%v:\n", e.Namespace, e.Name)
desc, _ := RunKubectl(
"describe", "pod", e.Name, fmt.Sprintf("--namespace=%v", e.Namespace))
Logf(desc)
}
}
// EndpointHostnames returns a set of hostnames for existing endpoints.
func (config *NetworkingTestConfig) EndpointHostnames() sets.String {
expectedEps := sets.NewString()
for _, p := range config.EndpointPods {
expectedEps.Insert(p.Name)
}
return expectedEps
}
// DialFromContainers executes a curl via kubectl exec in a test container,
// which might then translate to a tcp or udp request based on the protocol
// argument in the url.
// - minTries is the minimum number of curl attempts required before declaring
// success. Set to 0 if you'd like to return as soon as all endpoints respond
// at least once.
// - maxTries is the maximum number of curl attempts. If this many attempts pass
// and we don't see all expected endpoints, the test fails.
// - expectedEps is the set of endpointnames to wait for. Typically this is also
// the hostname reported by each pod in the service through /hostName.
// maxTries == minTries will confirm that we see the expected endpoints and no
// more for maxTries. Use this if you want to eg: fail a readiness check on a
// pod and confirm it doesn't show up as an endpoint.
func (config *NetworkingTestConfig) DialFromContainer(protocol, containerIP, targetIP string, containerHttpPort, targetPort, maxTries, minTries int, expectedEps sets.String) {
ipPort := net.JoinHostPort(containerIP, strconv.Itoa(containerHttpPort))
// The current versions of curl included in CentOS and RHEL distros
// misinterpret square brackets around IPv6 as globbing, so use the -g
// argument to disable globbing to handle the IPv6 case.
cmd := fmt.Sprintf("curl -g -q -s 'http://%s/dial?request=hostName&protocol=%s&host=%s&port=%d&tries=1'",
ipPort,
protocol,
targetIP,
targetPort)
eps := sets.NewString()
for i := 0; i < maxTries; i++ {
stdout, stderr, err := config.f.ExecShellInPodWithFullOutput(config.HostTestContainerPod.Name, cmd)
if err != nil {
// A failure to kubectl exec counts as a try, not a hard fail.
// Also note that we will keep failing for maxTries in tests where
// we confirm unreachability.
Logf("Failed to execute %q: %v, stdout: %q, stderr %q", cmd, err, stdout, stderr)
} else {
var output map[string][]string
if err := json.Unmarshal([]byte(stdout), &output); err != nil {
Logf("WARNING: Failed to unmarshal curl response. Cmd %v run in %v, output: %s, err: %v",
cmd, config.HostTestContainerPod.Name, stdout, err)
continue
}
for _, hostName := range output["responses"] {
trimmed := strings.TrimSpace(hostName)
if trimmed != "" {
eps.Insert(trimmed)
}
}
}
Logf("Waiting for endpoints: %v", expectedEps.Difference(eps))
// Check against i+1 so we exit if minTries == maxTries.
if (eps.Equal(expectedEps) || eps.Len() == 0 && expectedEps.Len() == 0) && i+1 >= minTries {
return
}
// TODO: get rid of this delay #36281
time.Sleep(hitEndpointRetryDelay)
}
config.diagnoseMissingEndpoints(eps)
Failf("Failed to find expected endpoints:\nTries %d\nCommand %v\nretrieved %v\nexpected %v\n", maxTries, cmd, eps, expectedEps)
}
func (config *NetworkingTestConfig) GetEndpointsFromTestContainer(protocol, targetIP string, targetPort, tries int) (sets.String, error) {
return config.GetEndpointsFromContainer(protocol, config.TestContainerPod.Status.PodIP, targetIP, TestContainerHttpPort, targetPort, tries)
}
// GetEndpointsFromContainer executes a curl via kubectl exec in a test container,
// which might then translate to a tcp or udp request based on the protocol argument
// in the url. It returns all different endpoints from multiple retries.
// - tries is the number of curl attempts. If this many attempts pass and
// we don't see any endpoints, the test fails.
func (config *NetworkingTestConfig) GetEndpointsFromContainer(protocol, containerIP, targetIP string, containerHttpPort, targetPort, tries int) (sets.String, error) {
ipPort := net.JoinHostPort(containerIP, strconv.Itoa(containerHttpPort))
// The current versions of curl included in CentOS and RHEL distros
// misinterpret square brackets around IPv6 as globbing, so use the -g
// argument to disable globbing to handle the IPv6 case.
cmd := fmt.Sprintf("curl -g -q -s 'http://%s/dial?request=hostName&protocol=%s&host=%s&port=%d&tries=1'",
ipPort,
protocol,
targetIP,
targetPort)
eps := sets.NewString()
for i := 0; i < tries; i++ {
stdout, stderr, err := config.f.ExecShellInPodWithFullOutput(config.HostTestContainerPod.Name, cmd)
if err != nil {
// A failure to kubectl exec counts as a try, not a hard fail.
// Also note that we will keep failing for maxTries in tests where
// we confirm unreachability.
Logf("Failed to execute %q: %v, stdout: %q, stderr: %q", cmd, err, stdout, stderr)
} else {
Logf("Tries: %d, in try: %d, stdout: %v, stderr: %v, command run in: %#v", tries, i, stdout, stderr, config.HostTestContainerPod)
var output map[string][]string
if err := json.Unmarshal([]byte(stdout), &output); err != nil {
Logf("WARNING: Failed to unmarshal curl response. Cmd %v run in %v, output: %s, err: %v",
cmd, config.HostTestContainerPod.Name, stdout, err)
continue
}
for _, hostName := range output["responses"] {
trimmed := strings.TrimSpace(hostName)
if trimmed != "" {
eps.Insert(trimmed)
}
}
// TODO: get rid of this delay #36281
time.Sleep(hitEndpointRetryDelay)
}
}
return eps, nil
}
// DialFromNode executes a tcp or udp request based on protocol via kubectl exec
// in a test container running with host networking.
// - minTries is the minimum number of curl attempts required before declaring
// success. Set to 0 if you'd like to return as soon as all endpoints respond
// at least once.
// - maxTries is the maximum number of curl attempts. If this many attempts pass
// and we don't see all expected endpoints, the test fails.
// maxTries == minTries will confirm that we see the expected endpoints and no
// more for maxTries. Use this if you want to eg: fail a readiness check on a
// pod and confirm it doesn't show up as an endpoint.
func (config *NetworkingTestConfig) DialFromNode(protocol, targetIP string, targetPort, maxTries, minTries int, expectedEps sets.String) {
var cmd string
if protocol == "udp" {
// TODO: It would be enough to pass 1s+epsilon to timeout, but unfortunately
// busybox timeout doesn't support non-integer values.
cmd = fmt.Sprintf("echo 'hostName' | timeout -t 2 nc -w 1 -u %s %d", targetIP, targetPort)
} else {
ipPort := net.JoinHostPort(targetIP, strconv.Itoa(targetPort))
// The current versions of curl included in CentOS and RHEL distros
// misinterpret square brackets around IPv6 as globbing, so use the -g
// argument to disable globbing to handle the IPv6 case.
cmd = fmt.Sprintf("timeout -t 15 curl -g -q -s --connect-timeout 1 http://%s/hostName", ipPort)
}
// TODO: This simply tells us that we can reach the endpoints. Check that
// the probability of hitting a specific endpoint is roughly the same as
// hitting any other.
eps := sets.NewString()
filterCmd := fmt.Sprintf("%s | grep -v '^\\s*$'", cmd)
for i := 0; i < maxTries; i++ {
stdout, stderr, err := config.f.ExecShellInPodWithFullOutput(config.HostTestContainerPod.Name, filterCmd)
if err != nil || len(stderr) > 0 {
// A failure to exec command counts as a try, not a hard fail.
// Also note that we will keep failing for maxTries in tests where
// we confirm unreachability.
Logf("Failed to execute %q: %v, stdout: %q, stderr: %q", filterCmd, err, stdout, stderr)
} else {
trimmed := strings.TrimSpace(stdout)
if trimmed != "" {
eps.Insert(trimmed)
}
}
// Check against i+1 so we exit if minTries == maxTries.
if eps.Equal(expectedEps) && i+1 >= minTries {
Logf("Found all expected endpoints: %+v", eps.List())
return
}
Logf("Waiting for %+v endpoints (expected=%+v, actual=%+v)", expectedEps.Difference(eps).List(), expectedEps.List(), eps.List())
// TODO: get rid of this delay #36281
time.Sleep(hitEndpointRetryDelay)
}
config.diagnoseMissingEndpoints(eps)
Failf("Failed to find expected endpoints:\nTries %d\nCommand %v\nretrieved %v\nexpected %v\n", maxTries, cmd, eps, expectedEps)
}
// GetSelfURL executes a curl against the given path via kubectl exec into a
// test container running with host networking, and fails if the output
// doesn't match the expected string.
func (config *NetworkingTestConfig) GetSelfURL(port int32, path string, expected string) {
cmd := fmt.Sprintf("curl -i -q -s --connect-timeout 1 http://localhost:%d%s", port, path)
By(fmt.Sprintf("Getting kube-proxy self URL %s", path))
config.executeCurlCmd(cmd, expected)
}
// GetSelfStatusCode executes a curl against the given path via kubectl exec into a
// test container running with host networking, and fails if the returned status
// code doesn't match the expected string.
func (config *NetworkingTestConfig) GetSelfURLStatusCode(port int32, path string, expected string) {
// check status code
cmd := fmt.Sprintf("curl -o /dev/null -i -q -s -w %%{http_code} --connect-timeout 1 http://localhost:%d%s", port, path)
By(fmt.Sprintf("Checking status code against http://localhost:%d%s", port, path))
config.executeCurlCmd(cmd, expected)
}
func (config *NetworkingTestConfig) executeCurlCmd(cmd string, expected string) {
// These are arbitrary timeouts. The curl command should pass on first try,
// unless remote server is starved/bootstrapping/restarting etc.
const retryInterval = 1 * time.Second
const retryTimeout = 30 * time.Second
podName := config.HostTestContainerPod.Name
var msg string
if pollErr := wait.PollImmediate(retryInterval, retryTimeout, func() (bool, error) {
stdout, err := RunHostCmd(config.Namespace, podName, cmd)
if err != nil {
msg = fmt.Sprintf("failed executing cmd %v in %v/%v: %v", cmd, config.Namespace, podName, err)
Logf(msg)
return false, nil
}
if !strings.Contains(stdout, expected) {
msg = fmt.Sprintf("successfully executed %v in %v/%v, but output '%v' doesn't contain expected string '%v'", cmd, config.Namespace, podName, stdout, expected)
Logf(msg)
return false, nil
}
return true, nil
}); pollErr != nil {
Logf("\nOutput of kubectl describe pod %v/%v:\n", config.Namespace, podName)
desc, _ := RunKubectl(
"describe", "pod", podName, fmt.Sprintf("--namespace=%v", config.Namespace))
Logf("%s", desc)
Failf("Timed out in %v: %v", retryTimeout, msg)
}
}
func (config *NetworkingTestConfig) createNetShellPodSpec(podName, hostname string) *v1.Pod {
probe := &v1.Probe{
InitialDelaySeconds: 10,
TimeoutSeconds: 30,
PeriodSeconds: 10,
SuccessThreshold: 1,
FailureThreshold: 3,
Handler: v1.Handler{
HTTPGet: &v1.HTTPGetAction{
Path: "/healthz",
Port: intstr.IntOrString{IntVal: EndpointHttpPort},
},
},
}
pod := &v1.Pod{
TypeMeta: metav1.TypeMeta{
Kind: "Pod",
APIVersion: testapi.Groups[v1.GroupName].GroupVersion().String(),
},
ObjectMeta: metav1.ObjectMeta{
Name: podName,
Namespace: config.Namespace,
},
Spec: v1.PodSpec{
Containers: []v1.Container{
{
Name: "webserver",
Image: NetexecImageName,
ImagePullPolicy: v1.PullIfNotPresent,
Command: []string{
"/netexec",
fmt.Sprintf("--http-port=%d", EndpointHttpPort),
fmt.Sprintf("--udp-port=%d", EndpointUdpPort),
},
Ports: []v1.ContainerPort{
{
Name: "http",
ContainerPort: EndpointHttpPort,
},
{
Name: "udp",
ContainerPort: EndpointUdpPort,
Protocol: v1.ProtocolUDP,
},
},
LivenessProbe: probe,
ReadinessProbe: probe,
},
},
NodeSelector: map[string]string{
"kubernetes.io/hostname": hostname,
},
},
}
return pod
}
func (config *NetworkingTestConfig) createTestPodSpec() *v1.Pod {
pod := &v1.Pod{
TypeMeta: metav1.TypeMeta{
Kind: "Pod",
APIVersion: testapi.Groups[v1.GroupName].GroupVersion().String(),
},
ObjectMeta: metav1.ObjectMeta{
Name: testPodName,
Namespace: config.Namespace,
},
Spec: v1.PodSpec{
Containers: []v1.Container{
{
Name: "webserver",
Image: NetexecImageName,
ImagePullPolicy: v1.PullIfNotPresent,
Command: []string{
"/netexec",
fmt.Sprintf("--http-port=%d", EndpointHttpPort),
fmt.Sprintf("--udp-port=%d", EndpointUdpPort),
},
Ports: []v1.ContainerPort{
{
Name: "http",
ContainerPort: TestContainerHttpPort,
},
},
},
},
},
}
return pod
}
func (config *NetworkingTestConfig) createNodePortServiceSpec(svcName string, selector map[string]string, enableSessionAffinity bool) *v1.Service {
sessionAffinity := v1.ServiceAffinityNone
if enableSessionAffinity {
sessionAffinity = v1.ServiceAffinityClientIP
}
return &v1.Service{
ObjectMeta: metav1.ObjectMeta{
Name: svcName,
},
Spec: v1.ServiceSpec{
Type: v1.ServiceTypeNodePort,
Ports: []v1.ServicePort{
{Port: ClusterHttpPort, Name: "http", Protocol: v1.ProtocolTCP, TargetPort: intstr.FromInt(EndpointHttpPort)},
{Port: ClusterUdpPort, Name: "udp", Protocol: v1.ProtocolUDP, TargetPort: intstr.FromInt(EndpointUdpPort)},
},
Selector: selector,
SessionAffinity: sessionAffinity,
},
}
}
func (config *NetworkingTestConfig) createNodePortService(selector map[string]string) {
config.NodePortService = config.createService(config.createNodePortServiceSpec(nodePortServiceName, selector, false))
}
func (config *NetworkingTestConfig) createSessionAffinityService(selector map[string]string) {
config.SessionAffinityService = config.createService(config.createNodePortServiceSpec(sessionAffinityServiceName, selector, true))
}
func (config *NetworkingTestConfig) DeleteNodePortService() {
err := config.getServiceClient().Delete(config.NodePortService.Name, nil)
Expect(err).NotTo(HaveOccurred(), "error while deleting NodePortService. err:%v)", err)
time.Sleep(15 * time.Second) // wait for kube-proxy to catch up with the service being deleted.
}
func (config *NetworkingTestConfig) createTestPods() {
testContainerPod := config.createTestPodSpec()
hostTestContainerPod := NewHostExecPodSpec(config.Namespace, hostTestPodName)
config.createPod(testContainerPod)
config.createPod(hostTestContainerPod)
ExpectNoError(config.f.WaitForPodRunning(testContainerPod.Name))
ExpectNoError(config.f.WaitForPodRunning(hostTestContainerPod.Name))
var err error
config.TestContainerPod, err = config.getPodClient().Get(testContainerPod.Name, metav1.GetOptions{})
if err != nil {
Failf("Failed to retrieve %s pod: %v", testContainerPod.Name, err)
}
config.HostTestContainerPod, err = config.getPodClient().Get(hostTestContainerPod.Name, metav1.GetOptions{})
if err != nil {
Failf("Failed to retrieve %s pod: %v", hostTestContainerPod.Name, err)
}
}
func (config *NetworkingTestConfig) createService(serviceSpec *v1.Service) *v1.Service {
_, err := config.getServiceClient().Create(serviceSpec)
Expect(err).NotTo(HaveOccurred(), fmt.Sprintf("Failed to create %s service: %v", serviceSpec.Name, err))
err = WaitForService(config.f.ClientSet, config.Namespace, serviceSpec.Name, true, 5*time.Second, 45*time.Second)
Expect(err).NotTo(HaveOccurred(), fmt.Sprintf("error while waiting for service:%s err: %v", serviceSpec.Name, err))
createdService, err := config.getServiceClient().Get(serviceSpec.Name, metav1.GetOptions{})
Expect(err).NotTo(HaveOccurred(), fmt.Sprintf("Failed to create %s service: %v", serviceSpec.Name, err))
return createdService
}
// setupCore sets up the pods and core test config
// mainly for simplified node e2e setup
func (config *NetworkingTestConfig) setupCore(selector map[string]string) {
By("Creating the service pods in kubernetes")
podName := "netserver"
config.EndpointPods = config.createNetProxyPods(podName, selector)
By("Creating test pods")
config.createTestPods()
epCount := len(config.EndpointPods)
config.MaxTries = epCount*epCount + testTries
}
// setup includes setupCore and also sets up services
func (config *NetworkingTestConfig) setup(selector map[string]string) {
config.setupCore(selector)
By("Getting node addresses")
ExpectNoError(WaitForAllNodesSchedulable(config.f.ClientSet, 10*time.Minute))
nodeList := GetReadySchedulableNodesOrDie(config.f.ClientSet)
config.ExternalAddrs = NodeAddresses(nodeList, v1.NodeExternalIP)
SkipUnlessNodeCountIsAtLeast(2)
config.Nodes = nodeList.Items
By("Creating the service on top of the pods in kubernetes")
config.createNodePortService(selector)
config.createSessionAffinityService(selector)
for _, p := range config.NodePortService.Spec.Ports {
switch p.Protocol {
case v1.ProtocolUDP:
config.NodeUdpPort = int(p.NodePort)
case v1.ProtocolTCP:
config.NodeHttpPort = int(p.NodePort)
default:
continue
}
}
config.ClusterIP = config.NodePortService.Spec.ClusterIP
if len(config.ExternalAddrs) != 0 {
config.NodeIP = config.ExternalAddrs[0]
} else {
internalAddrs := NodeAddresses(nodeList, v1.NodeInternalIP)
config.NodeIP = internalAddrs[0]
}
}
func (config *NetworkingTestConfig) cleanup() {
nsClient := config.getNamespacesClient()
nsList, err := nsClient.List(metav1.ListOptions{})
if err == nil {
for _, ns := range nsList.Items {
if strings.Contains(ns.Name, config.f.BaseName) && ns.Name != config.Namespace {
nsClient.Delete(ns.Name, nil)
}
}
}
}
// shuffleNodes copies nodes from the specified slice into a copy in random
// order. It returns a new slice.
func shuffleNodes(nodes []v1.Node) []v1.Node {
shuffled := make([]v1.Node, len(nodes))
perm := rand.Perm(len(nodes))
for i, j := range perm {
shuffled[j] = nodes[i]
}
return shuffled
}
func (config *NetworkingTestConfig) createNetProxyPods(podName string, selector map[string]string) []*v1.Pod {
ExpectNoError(WaitForAllNodesSchedulable(config.f.ClientSet, 10*time.Minute))
nodeList := GetReadySchedulableNodesOrDie(config.f.ClientSet)
// To make this test work reasonably fast in large clusters,
// we limit the number of NetProxyPods to no more than
// maxNetProxyPodsCount on random nodes.
nodes := shuffleNodes(nodeList.Items)
if len(nodes) > maxNetProxyPodsCount {
nodes = nodes[:maxNetProxyPodsCount]
}
// create pods, one for each node
createdPods := make([]*v1.Pod, 0, len(nodes))
for i, n := range nodes {
podName := fmt.Sprintf("%s-%d", podName, i)
hostname, _ := n.Labels["kubernetes.io/hostname"]
pod := config.createNetShellPodSpec(podName, hostname)
pod.ObjectMeta.Labels = selector
createdPod := config.createPod(pod)
createdPods = append(createdPods, createdPod)
}
// wait that all of them are up
runningPods := make([]*v1.Pod, 0, len(nodes))
for _, p := range createdPods {
ExpectNoError(config.f.WaitForPodReady(p.Name))
rp, err := config.getPodClient().Get(p.Name, metav1.GetOptions{})
ExpectNoError(err)
runningPods = append(runningPods, rp)
}
return runningPods
}
func (config *NetworkingTestConfig) DeleteNetProxyPod() {
pod := config.EndpointPods[0]
config.getPodClient().Delete(pod.Name, metav1.NewDeleteOptions(0))
config.EndpointPods = config.EndpointPods[1:]
// wait for pod being deleted.
err := WaitForPodToDisappear(config.f.ClientSet, config.Namespace, pod.Name, labels.Everything(), time.Second, wait.ForeverTestTimeout)
if err != nil {
Failf("Failed to delete %s pod: %v", pod.Name, err)
}
// wait for endpoint being removed.
err = WaitForServiceEndpointsNum(config.f.ClientSet, config.Namespace, nodePortServiceName, len(config.EndpointPods), time.Second, wait.ForeverTestTimeout)
if err != nil {
Failf("Failed to remove endpoint from service: %s", nodePortServiceName)
}
// wait for kube-proxy to catch up with the pod being deleted.
time.Sleep(5 * time.Second)
}
func (config *NetworkingTestConfig) createPod(pod *v1.Pod) *v1.Pod {
return config.getPodClient().Create(pod)
}
func (config *NetworkingTestConfig) getPodClient() *PodClient {
if config.podClient == nil {
config.podClient = config.f.PodClient()
}
return config.podClient
}
func (config *NetworkingTestConfig) getServiceClient() coreclientset.ServiceInterface {
return config.f.ClientSet.CoreV1().Services(config.Namespace)
}
func (config *NetworkingTestConfig) getNamespacesClient() coreclientset.NamespaceInterface {
return config.f.ClientSet.CoreV1().Namespaces()
}
func CheckReachabilityFromPod(expectToBeReachable bool, timeout time.Duration, namespace, pod, target string) {
cmd := fmt.Sprintf("wget -T 5 -qO- %q", target)
err := wait.PollImmediate(Poll, timeout, func() (bool, error) {
_, err := RunHostCmd(namespace, pod, cmd)
if expectToBeReachable && err != nil {
Logf("Expect target to be reachable. But got err: %v. Retry until timeout", err)
return false, nil
}
if !expectToBeReachable && err == nil {
Logf("Expect target NOT to be reachable. But it is reachable. Retry until timeout")
return false, nil
}
return true, nil
})
Expect(err).NotTo(HaveOccurred())
}
// Does an HTTP GET, but does not reuse TCP connections
// This masks problems where the iptables rule has changed, but we don't see it
// This is intended for relatively quick requests (status checks), so we set a short (5 seconds) timeout
func httpGetNoConnectionPool(url string) (*http.Response, error) {
return httpGetNoConnectionPoolTimeout(url, 5*time.Second)
}
func httpGetNoConnectionPoolTimeout(url string, timeout time.Duration) (*http.Response, error) {
tr := utilnet.SetTransportDefaults(&http.Transport{
DisableKeepAlives: true,
})
client := &http.Client{
Transport: tr,
Timeout: timeout,
}
return client.Get(url)
}
func TestReachableHTTP(ip string, port int, request string, expect string) (bool, error) {
return TestReachableHTTPWithContent(ip, port, request, expect, nil)
}
func TestReachableHTTPWithRetriableErrorCodes(ip string, port int, request string, expect string, retriableErrCodes []int) (bool, error) {
return TestReachableHTTPWithContentTimeoutWithRetriableErrorCodes(ip, port, request, expect, nil, retriableErrCodes, time.Second*5)
}
func TestReachableHTTPWithContent(ip string, port int, request string, expect string, content *bytes.Buffer) (bool, error) {
return TestReachableHTTPWithContentTimeout(ip, port, request, expect, content, 5*time.Second)
}
func TestReachableHTTPWithContentTimeout(ip string, port int, request string, expect string, content *bytes.Buffer, timeout time.Duration) (bool, error) {
return TestReachableHTTPWithContentTimeoutWithRetriableErrorCodes(ip, port, request, expect, content, []int{}, timeout)
}
func TestReachableHTTPWithContentTimeoutWithRetriableErrorCodes(ip string, port int, request string, expect string, content *bytes.Buffer, retriableErrCodes []int, timeout time.Duration) (bool, error) {
ipPort := net.JoinHostPort(ip, strconv.Itoa(port))
url := fmt.Sprintf("http://%s%s", ipPort, request)
if ip == "" {
Failf("Got empty IP for reachability check (%s)", url)
return false, nil
}
if port == 0 {
Failf("Got port==0 for reachability check (%s)", url)
return false, nil
}
Logf("Testing HTTP reachability of %v", url)
resp, err := httpGetNoConnectionPoolTimeout(url, timeout)
if err != nil {
Logf("Got error testing for reachability of %s: %v", url, err)
return false, nil
}
defer resp.Body.Close()
body, err := ioutil.ReadAll(resp.Body)
if err != nil {
Logf("Got error reading response from %s: %v", url, err)
return false, nil
}
if resp.StatusCode != 200 {
for _, code := range retriableErrCodes {
if resp.StatusCode == code {
Logf("Got non-success status %q when trying to access %s, but the error code is retriable", resp.Status, url)
return false, nil
}
}
return false, fmt.Errorf("received non-success return status %q trying to access %s; got body: %s",
resp.Status, url, string(body))
}
if !strings.Contains(string(body), expect) {
return false, fmt.Errorf("received response body without expected substring %q: %s", expect, string(body))
}
if content != nil {
content.Write(body)
}
return true, nil
}
func TestNotReachableHTTP(ip string, port int) (bool, error) {
return TestNotReachableHTTPTimeout(ip, port, 5*time.Second)
}
func TestNotReachableHTTPTimeout(ip string, port int, timeout time.Duration) (bool, error) {
ipPort := net.JoinHostPort(ip, strconv.Itoa(port))
url := fmt.Sprintf("http://%s", ipPort)
if ip == "" {
Failf("Got empty IP for non-reachability check (%s)", url)
return false, nil
}
if port == 0 {
Failf("Got port==0 for non-reachability check (%s)", url)
return false, nil
}
Logf("Testing HTTP non-reachability of %v", url)
resp, err := httpGetNoConnectionPoolTimeout(url, timeout)
if err != nil {
Logf("Confirmed that %s is not reachable", url)
return true, nil
}
resp.Body.Close()
return false, nil
}
func TestReachableUDP(ip string, port int, request string, expect string) (bool, error) {
ipPort := net.JoinHostPort(ip, strconv.Itoa(port))
uri := fmt.Sprintf("udp://%s", ipPort)
if ip == "" {
Failf("Got empty IP for reachability check (%s)", uri)
return false, nil
}
if port == 0 {
Failf("Got port==0 for reachability check (%s)", uri)
return false, nil
}
Logf("Testing UDP reachability of %v", uri)
con, err := net.Dial("udp", ipPort)
if err != nil {
return false, fmt.Errorf("Failed to dial %s: %v", ipPort, err)
}
_, err = con.Write([]byte(fmt.Sprintf("%s\n", request)))
if err != nil {
return false, fmt.Errorf("Failed to send request: %v", err)
}
var buf []byte = make([]byte, len(expect)+1)
err = con.SetDeadline(time.Now().Add(3 * time.Second))
if err != nil {
return false, fmt.Errorf("Failed to set deadline: %v", err)
}
_, err = con.Read(buf)
if err != nil {
return false, nil
}
if !strings.Contains(string(buf), expect) {
return false, fmt.Errorf("Failed to retrieve %q, got %q", expect, string(buf))
}
Logf("Successfully reached %v", uri)
return true, nil
}
func TestNotReachableUDP(ip string, port int, request string) (bool, error) {
ipPort := net.JoinHostPort(ip, strconv.Itoa(port))
uri := fmt.Sprintf("udp://%s", ipPort)
if ip == "" {
Failf("Got empty IP for reachability check (%s)", uri)
return false, nil
}
if port == 0 {
Failf("Got port==0 for reachability check (%s)", uri)
return false, nil
}
Logf("Testing UDP non-reachability of %v", uri)
con, err := net.Dial("udp", ipPort)
if err != nil {
Logf("Confirmed that %s is not reachable", uri)
return true, nil
}
_, err = con.Write([]byte(fmt.Sprintf("%s\n", request)))
if err != nil {
Logf("Confirmed that %s is not reachable", uri)
return true, nil
}
var buf []byte = make([]byte, 1)
err = con.SetDeadline(time.Now().Add(3 * time.Second))
if err != nil {
return false, fmt.Errorf("Failed to set deadline: %v", err)
}
_, err = con.Read(buf)
if err != nil {
Logf("Confirmed that %s is not reachable", uri)
return true, nil
}
return false, nil
}
func TestHitNodesFromOutside(externalIP string, httpPort int32, timeout time.Duration, expectedHosts sets.String) error {
return TestHitNodesFromOutsideWithCount(externalIP, httpPort, timeout, expectedHosts, 1)
}
func TestHitNodesFromOutsideWithCount(externalIP string, httpPort int32, timeout time.Duration, expectedHosts sets.String,
countToSucceed int) error {
Logf("Waiting up to %v for satisfying expectedHosts for %v times", timeout, countToSucceed)
hittedHosts := sets.NewString()
count := 0
condition := func() (bool, error) {
var respBody bytes.Buffer
reached, err := TestReachableHTTPWithContentTimeout(externalIP, int(httpPort), "/hostname", "", &respBody,
1*time.Second)
if err != nil || !reached {
return false, nil
}
hittedHost := strings.TrimSpace(respBody.String())
if !expectedHosts.Has(hittedHost) {
Logf("Error hitting unexpected host: %v, reset counter: %v", hittedHost, count)
count = 0
return false, nil
}
if !hittedHosts.Has(hittedHost) {
hittedHosts.Insert(hittedHost)
Logf("Missing %+v, got %+v", expectedHosts.Difference(hittedHosts), hittedHosts)
}
if hittedHosts.Equal(expectedHosts) {
count++
if count >= countToSucceed {
return true, nil
}
}
return false, nil
}
if err := wait.Poll(time.Second, timeout, condition); err != nil {
return fmt.Errorf("error waiting for expectedHosts: %v, hittedHosts: %v, count: %v, expected count: %v",
expectedHosts, hittedHosts, count, countToSucceed)
}
return nil
}
// Blocks outgoing network traffic on 'node'. Then runs testFunc and returns its status.
// At the end (even in case of errors), the network traffic is brought back to normal.
// This function executes commands on a node so it will work only for some
// environments.
func TestUnderTemporaryNetworkFailure(c clientset.Interface, ns string, node *v1.Node, testFunc func()) {
host := GetNodeExternalIP(node)
master := GetMasterAddress(c)
By(fmt.Sprintf("block network traffic from node %s to the master", node.Name))
defer func() {
// This code will execute even if setting the iptables rule failed.
// It is on purpose because we may have an error even if the new rule
// had been inserted. (yes, we could look at the error code and ssh error
// separately, but I prefer to stay on the safe side).
By(fmt.Sprintf("Unblock network traffic from node %s to the master", node.Name))
UnblockNetwork(host, master)
}()
Logf("Waiting %v to ensure node %s is ready before beginning test...", resizeNodeReadyTimeout, node.Name)
if !WaitForNodeToBe(c, node.Name, v1.NodeReady, true, resizeNodeReadyTimeout) {
Failf("Node %s did not become ready within %v", node.Name, resizeNodeReadyTimeout)
}
BlockNetwork(host, master)
Logf("Waiting %v for node %s to be not ready after simulated network failure", resizeNodeNotReadyTimeout, node.Name)
if !WaitForNodeToBe(c, node.Name, v1.NodeReady, false, resizeNodeNotReadyTimeout) {
Failf("Node %s did not become not-ready within %v", node.Name, resizeNodeNotReadyTimeout)
}
testFunc()
// network traffic is unblocked in a deferred function
}

View File

@ -0,0 +1,344 @@
/*
Copyright 2014 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 framework
import (
"fmt"
"os"
"path"
"path/filepath"
"strings"
"time"
"k8s.io/api/core/v1"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/fields"
"k8s.io/apimachinery/pkg/util/wait"
clientset "k8s.io/client-go/kubernetes"
)
func EtcdUpgrade(target_storage, target_version string) error {
switch TestContext.Provider {
case "gce":
return etcdUpgradeGCE(target_storage, target_version)
default:
return fmt.Errorf("EtcdUpgrade() is not implemented for provider %s", TestContext.Provider)
}
}
func MasterUpgrade(v string) error {
switch TestContext.Provider {
case "gce":
return masterUpgradeGCE(v, false)
case "gke":
return masterUpgradeGKE(v)
case "kubernetes-anywhere":
return masterUpgradeKubernetesAnywhere(v)
default:
return fmt.Errorf("MasterUpgrade() is not implemented for provider %s", TestContext.Provider)
}
}
func etcdUpgradeGCE(target_storage, target_version string) error {
env := append(
os.Environ(),
"TEST_ETCD_VERSION="+target_version,
"STORAGE_BACKEND="+target_storage,
"TEST_ETCD_IMAGE=3.1.10")
_, _, err := RunCmdEnv(env, gceUpgradeScript(), "-l", "-M")
return err
}
// TODO(mrhohn): Remove this function when kube-proxy is run as a DaemonSet by default.
func MasterUpgradeGCEWithKubeProxyDaemonSet(v string, enableKubeProxyDaemonSet bool) error {
return masterUpgradeGCE(v, enableKubeProxyDaemonSet)
}
// TODO(mrhohn): Remove 'enableKubeProxyDaemonSet' when kube-proxy is run as a DaemonSet by default.
func masterUpgradeGCE(rawV string, enableKubeProxyDaemonSet bool) error {
env := append(os.Environ(), fmt.Sprintf("KUBE_PROXY_DAEMONSET=%v", enableKubeProxyDaemonSet))
// TODO: Remove these variables when they're no longer needed for downgrades.
if TestContext.EtcdUpgradeVersion != "" && TestContext.EtcdUpgradeStorage != "" {
env = append(env,
"TEST_ETCD_VERSION="+TestContext.EtcdUpgradeVersion,
"STORAGE_BACKEND="+TestContext.EtcdUpgradeStorage,
"TEST_ETCD_IMAGE=3.1.10")
} else {
// In e2e tests, we skip the confirmation prompt about
// implicit etcd upgrades to simulate the user entering "y".
env = append(env, "TEST_ALLOW_IMPLICIT_ETCD_UPGRADE=true")
}
v := "v" + rawV
_, _, err := RunCmdEnv(env, gceUpgradeScript(), "-M", v)
return err
}
func masterUpgradeGKE(v string) error {
Logf("Upgrading master to %q", v)
_, _, err := RunCmd("gcloud", "container",
"clusters",
fmt.Sprintf("--project=%s", TestContext.CloudConfig.ProjectID),
fmt.Sprintf("--zone=%s", TestContext.CloudConfig.Zone),
"upgrade",
TestContext.CloudConfig.Cluster,
"--master",
fmt.Sprintf("--cluster-version=%s", v),
"--quiet")
if err != nil {
return err
}
waitForSSHTunnels()
return nil
}
func masterUpgradeKubernetesAnywhere(v string) error {
Logf("Upgrading master to %q", v)
kaPath := TestContext.KubernetesAnywherePath
originalConfigPath := filepath.Join(kaPath, ".config")
backupConfigPath := filepath.Join(kaPath, ".config.bak")
updatedConfigPath := filepath.Join(kaPath, fmt.Sprintf(".config-%s", v))
// modify config with specified k8s version
if _, _, err := RunCmd("sed",
"-i.bak", // writes original to .config.bak
fmt.Sprintf(`s/kubernetes_version=.*$/kubernetes_version=%q/`, v),
originalConfigPath); err != nil {
return err
}
defer func() {
// revert .config.bak to .config
if err := os.Rename(backupConfigPath, originalConfigPath); err != nil {
Logf("Could not rename %s back to %s", backupConfigPath, originalConfigPath)
}
}()
// invoke ka upgrade
if _, _, err := RunCmd("make", "-C", TestContext.KubernetesAnywherePath,
"WAIT_FOR_KUBECONFIG=y", "upgrade-master"); err != nil {
return err
}
// move .config to .config.<version>
if err := os.Rename(originalConfigPath, updatedConfigPath); err != nil {
return err
}
return nil
}
func NodeUpgrade(f *Framework, v string, img string) error {
// Perform the upgrade.
var err error
switch TestContext.Provider {
case "gce":
err = nodeUpgradeGCE(v, img, false)
case "gke":
err = nodeUpgradeGKE(v, img)
default:
err = fmt.Errorf("NodeUpgrade() is not implemented for provider %s", TestContext.Provider)
}
if err != nil {
return err
}
// Wait for it to complete and validate nodes are healthy.
//
// TODO(ihmccreery) We shouldn't have to wait for nodes to be ready in
// GKE; the operation shouldn't return until they all are.
Logf("Waiting up to %v for all nodes to be ready after the upgrade", RestartNodeReadyAgainTimeout)
if _, err := CheckNodesReady(f.ClientSet, RestartNodeReadyAgainTimeout, TestContext.CloudConfig.NumNodes); err != nil {
return err
}
return nil
}
// TODO(mrhohn): Remove this function when kube-proxy is run as a DaemonSet by default.
func NodeUpgradeGCEWithKubeProxyDaemonSet(f *Framework, v string, img string, enableKubeProxyDaemonSet bool) error {
// Perform the upgrade.
if err := nodeUpgradeGCE(v, img, enableKubeProxyDaemonSet); err != nil {
return err
}
// Wait for it to complete and validate nodes are healthy.
Logf("Waiting up to %v for all nodes to be ready after the upgrade", RestartNodeReadyAgainTimeout)
if _, err := CheckNodesReady(f.ClientSet, RestartNodeReadyAgainTimeout, TestContext.CloudConfig.NumNodes); err != nil {
return err
}
return nil
}
// TODO(mrhohn): Remove 'enableKubeProxyDaemonSet' when kube-proxy is run as a DaemonSet by default.
func nodeUpgradeGCE(rawV, img string, enableKubeProxyDaemonSet bool) error {
v := "v" + rawV
env := append(os.Environ(), fmt.Sprintf("KUBE_PROXY_DAEMONSET=%v", enableKubeProxyDaemonSet))
if img != "" {
env = append(env, "KUBE_NODE_OS_DISTRIBUTION="+img)
_, _, err := RunCmdEnv(env, gceUpgradeScript(), "-N", "-o", v)
return err
}
_, _, err := RunCmdEnv(env, gceUpgradeScript(), "-N", v)
return err
}
func nodeUpgradeGKE(v string, img string) error {
Logf("Upgrading nodes to version %q and image %q", v, img)
args := []string{
"container",
"clusters",
fmt.Sprintf("--project=%s", TestContext.CloudConfig.ProjectID),
fmt.Sprintf("--zone=%s", TestContext.CloudConfig.Zone),
"upgrade",
TestContext.CloudConfig.Cluster,
fmt.Sprintf("--cluster-version=%s", v),
"--quiet",
}
if len(img) > 0 {
args = append(args, fmt.Sprintf("--image-type=%s", img))
}
_, _, err := RunCmd("gcloud", args...)
if err != nil {
return err
}
waitForSSHTunnels()
return nil
}
// CheckNodesReady waits up to nt for expect nodes accessed by c to be ready,
// returning an error if this doesn't happen in time. It returns the names of
// nodes it finds.
func CheckNodesReady(c clientset.Interface, nt time.Duration, expect int) ([]string, error) {
// First, keep getting all of the nodes until we get the number we expect.
var nodeList *v1.NodeList
var errLast error
start := time.Now()
found := wait.Poll(Poll, nt, func() (bool, error) {
// A rolling-update (GCE/GKE implementation of restart) can complete before the apiserver
// knows about all of the nodes. Thus, we retry the list nodes call
// until we get the expected number of nodes.
nodeList, errLast = c.CoreV1().Nodes().List(metav1.ListOptions{
FieldSelector: fields.Set{"spec.unschedulable": "false"}.AsSelector().String()})
if errLast != nil {
return false, nil
}
if len(nodeList.Items) != expect {
errLast = fmt.Errorf("expected to find %d nodes but found only %d (%v elapsed)",
expect, len(nodeList.Items), time.Since(start))
Logf("%v", errLast)
return false, nil
}
return true, nil
}) == nil
nodeNames := make([]string, len(nodeList.Items))
for i, n := range nodeList.Items {
nodeNames[i] = n.ObjectMeta.Name
}
if !found {
return nodeNames, fmt.Errorf("couldn't find %d nodes within %v; last error: %v",
expect, nt, errLast)
}
Logf("Successfully found %d nodes", expect)
// Next, ensure in parallel that all the nodes are ready. We subtract the
// time we spent waiting above.
timeout := nt - time.Since(start)
result := make(chan bool, len(nodeList.Items))
for _, n := range nodeNames {
n := n
go func() { result <- WaitForNodeToBeReady(c, n, timeout) }()
}
failed := false
// TODO(mbforbes): Change to `for range` syntax once we support only Go
// >= 1.4.
for i := range nodeList.Items {
_ = i
if !<-result {
failed = true
}
}
if failed {
return nodeNames, fmt.Errorf("at least one node failed to be ready")
}
return nodeNames, nil
}
// MigTemplate (GCE-only) returns the name of the MIG template that the
// nodes of the cluster use.
func MigTemplate() (string, error) {
var errLast error
var templ string
key := "instanceTemplate"
if wait.Poll(Poll, SingleCallTimeout, func() (bool, error) {
// TODO(mikedanese): make this hit the compute API directly instead of
// shelling out to gcloud.
// An `instance-groups managed describe` call outputs what we want to stdout.
output, _, err := retryCmd("gcloud", "compute", "instance-groups", "managed",
fmt.Sprintf("--project=%s", TestContext.CloudConfig.ProjectID),
"describe",
fmt.Sprintf("--zone=%s", TestContext.CloudConfig.Zone),
TestContext.CloudConfig.NodeInstanceGroup)
if err != nil {
errLast = fmt.Errorf("gcloud compute instance-groups managed describe call failed with err: %v", err)
return false, nil
}
// The 'describe' call probably succeeded; parse the output and try to
// find the line that looks like "instanceTemplate: url/to/<templ>" and
// return <templ>.
if val := ParseKVLines(output, key); len(val) > 0 {
url := strings.Split(val, "/")
templ = url[len(url)-1]
Logf("MIG group %s using template: %s", TestContext.CloudConfig.NodeInstanceGroup, templ)
return true, nil
}
errLast = fmt.Errorf("couldn't find %s in output to get MIG template. Output: %s", key, output)
return false, nil
}) != nil {
return "", fmt.Errorf("MigTemplate() failed with last error: %v", errLast)
}
return templ, nil
}
func gceUpgradeScript() string {
if len(TestContext.GCEUpgradeScript) == 0 {
return path.Join(TestContext.RepoRoot, "cluster/gce/upgrade.sh")
}
return TestContext.GCEUpgradeScript
}
func waitForSSHTunnels() {
Logf("Waiting for SSH tunnels to establish")
RunKubectl("run", "ssh-tunnel-test",
"--image=busybox",
"--restart=Never",
"--command", "--",
"echo", "Hello")
defer RunKubectl("delete", "pod", "ssh-tunnel-test")
// allow up to a minute for new ssh tunnels to establish
wait.PollImmediate(5*time.Second, time.Minute, func() (bool, error) {
_, err := RunKubectl("logs", "ssh-tunnel-test")
return err == nil, nil
})
}

View File

@ -0,0 +1,158 @@
/*
Copyright 2016 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 framework
import (
"fmt"
"k8s.io/kubernetes/test/e2e/perftype"
)
// TODO(random-liu): Change the tests to actually use PerfData from the beginning instead of
// translating one to the other here.
// currentApiCallMetricsVersion is the current apicall performance metrics version. We should
// bump up the version each time we make incompatible change to the metrics.
const currentApiCallMetricsVersion = "v1"
// ApiCallToPerfData transforms APIResponsiveness to PerfData.
func ApiCallToPerfData(apicalls *APIResponsiveness) *perftype.PerfData {
perfData := &perftype.PerfData{Version: currentApiCallMetricsVersion}
for _, apicall := range apicalls.APICalls {
item := perftype.DataItem{
Data: map[string]float64{
"Perc50": float64(apicall.Latency.Perc50) / 1000000, // us -> ms
"Perc90": float64(apicall.Latency.Perc90) / 1000000,
"Perc99": float64(apicall.Latency.Perc99) / 1000000,
},
Unit: "ms",
Labels: map[string]string{
"Verb": apicall.Verb,
"Resource": apicall.Resource,
"Subresource": apicall.Subresource,
"Scope": apicall.Scope,
"Count": fmt.Sprintf("%v", apicall.Count),
},
}
perfData.DataItems = append(perfData.DataItems, item)
}
return perfData
}
// PodStartupLatencyToPerfData transforms PodStartupLatency to PerfData.
func PodStartupLatencyToPerfData(latency *PodStartupLatency) *perftype.PerfData {
perfData := &perftype.PerfData{Version: currentApiCallMetricsVersion}
item := perftype.DataItem{
Data: map[string]float64{
"Perc50": float64(latency.Latency.Perc50) / 1000000, // us -> ms
"Perc90": float64(latency.Latency.Perc90) / 1000000,
"Perc99": float64(latency.Latency.Perc99) / 1000000,
"Perc100": float64(latency.Latency.Perc100) / 1000000,
},
Unit: "ms",
Labels: map[string]string{
"Metric": "pod_startup",
},
}
perfData.DataItems = append(perfData.DataItems, item)
return perfData
}
// CurrentKubeletPerfMetricsVersion is the current kubelet performance metrics
// version. This is used by mutiple perf related data structures. We should
// bump up the version each time we make an incompatible change to the metrics.
const CurrentKubeletPerfMetricsVersion = "v2"
// ResourceUsageToPerfData transforms ResourceUsagePerNode to PerfData. Notice that this function
// only cares about memory usage, because cpu usage information will be extracted from NodesCPUSummary.
func ResourceUsageToPerfData(usagePerNode ResourceUsagePerNode) *perftype.PerfData {
return ResourceUsageToPerfDataWithLabels(usagePerNode, nil)
}
// CPUUsageToPerfData transforms NodesCPUSummary to PerfData.
func CPUUsageToPerfData(usagePerNode NodesCPUSummary) *perftype.PerfData {
return CPUUsageToPerfDataWithLabels(usagePerNode, nil)
}
// PrintPerfData prints the perfdata in json format with PerfResultTag prefix.
// If an error occurs, nothing will be printed.
func PrintPerfData(p *perftype.PerfData) {
// Notice that we must make sure the perftype.PerfResultEnd is in a new line.
if str := PrettyPrintJSON(p); str != "" {
Logf("%s %s\n%s", perftype.PerfResultTag, str, perftype.PerfResultEnd)
}
}
// ResourceUsageToPerfDataWithLabels transforms ResourceUsagePerNode to PerfData with additional labels.
// Notice that this function only cares about memory usage, because cpu usage information will be extracted from NodesCPUSummary.
func ResourceUsageToPerfDataWithLabels(usagePerNode ResourceUsagePerNode, labels map[string]string) *perftype.PerfData {
items := []perftype.DataItem{}
for node, usages := range usagePerNode {
for c, usage := range usages {
item := perftype.DataItem{
Data: map[string]float64{
"memory": float64(usage.MemoryUsageInBytes) / (1024 * 1024),
"workingset": float64(usage.MemoryWorkingSetInBytes) / (1024 * 1024),
"rss": float64(usage.MemoryRSSInBytes) / (1024 * 1024),
},
Unit: "MB",
Labels: map[string]string{
"node": node,
"container": c,
"datatype": "resource",
"resource": "memory",
},
}
items = append(items, item)
}
}
return &perftype.PerfData{
Version: CurrentKubeletPerfMetricsVersion,
DataItems: items,
Labels: labels,
}
}
// CPUUsageToPerfDataWithLabels transforms NodesCPUSummary to PerfData with additional labels.
func CPUUsageToPerfDataWithLabels(usagePerNode NodesCPUSummary, labels map[string]string) *perftype.PerfData {
items := []perftype.DataItem{}
for node, usages := range usagePerNode {
for c, usage := range usages {
data := map[string]float64{}
for perc, value := range usage {
data[fmt.Sprintf("Perc%02.0f", perc*100)] = value * 1000
}
item := perftype.DataItem{
Data: data,
Unit: "mCPU",
Labels: map[string]string{
"node": node,
"container": c,
"datatype": "resource",
"resource": "cpu",
},
}
items = append(items, item)
}
}
return &perftype.PerfData{
Version: CurrentKubeletPerfMetricsVersion,
DataItems: items,
Labels: labels,
}
}

261
vendor/k8s.io/kubernetes/test/e2e/framework/pods.go generated vendored Normal file
View File

@ -0,0 +1,261 @@
/*
Copyright 2016 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 framework
import (
"fmt"
"regexp"
"sync"
"time"
"k8s.io/api/core/v1"
"k8s.io/apimachinery/pkg/api/errors"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/labels"
"k8s.io/apimachinery/pkg/util/sets"
"k8s.io/apimachinery/pkg/util/wait"
v1core "k8s.io/client-go/kubernetes/typed/core/v1"
"k8s.io/kubernetes/pkg/api/legacyscheme"
"k8s.io/kubernetes/pkg/kubelet/events"
"k8s.io/kubernetes/pkg/kubelet/sysctl"
. "github.com/onsi/ginkgo"
. "github.com/onsi/gomega"
)
const DefaultPodDeletionTimeout = 3 * time.Minute
// ImageWhiteList is the images used in the current test suite. It should be initialized in test suite and
// the images in the white list should be pre-pulled in the test suite. Currently, this is only used by
// node e2e test.
var ImageWhiteList sets.String
// Convenience method for getting a pod client interface in the framework's namespace,
// possibly applying test-suite specific transformations to the pod spec, e.g. for
// node e2e pod scheduling.
func (f *Framework) PodClient() *PodClient {
return &PodClient{
f: f,
PodInterface: f.ClientSet.CoreV1().Pods(f.Namespace.Name),
}
}
// Convenience method for getting a pod client interface in an alternative namespace,
// possibly applying test-suite specific transformations to the pod spec, e.g. for
// node e2e pod scheduling.
func (f *Framework) PodClientNS(namespace string) *PodClient {
return &PodClient{
f: f,
PodInterface: f.ClientSet.CoreV1().Pods(namespace),
}
}
type PodClient struct {
f *Framework
v1core.PodInterface
}
// Create creates a new pod according to the framework specifications (don't wait for it to start).
func (c *PodClient) Create(pod *v1.Pod) *v1.Pod {
c.mungeSpec(pod)
p, err := c.PodInterface.Create(pod)
ExpectNoError(err, "Error creating Pod")
return p
}
// CreateSync creates a new pod according to the framework specifications in the given namespace, and waits for it to start.
func (c *PodClient) CreateSyncInNamespace(pod *v1.Pod, namespace string) *v1.Pod {
p := c.Create(pod)
ExpectNoError(WaitForPodNameRunningInNamespace(c.f.ClientSet, p.Name, namespace))
// Get the newest pod after it becomes running, some status may change after pod created, such as pod ip.
p, err := c.Get(p.Name, metav1.GetOptions{})
ExpectNoError(err)
return p
}
// CreateSync creates a new pod according to the framework specifications, and wait for it to start.
func (c *PodClient) CreateSync(pod *v1.Pod) *v1.Pod {
return c.CreateSyncInNamespace(pod, c.f.Namespace.Name)
}
// CreateBatch create a batch of pods. All pods are created before waiting.
func (c *PodClient) CreateBatch(pods []*v1.Pod) []*v1.Pod {
ps := make([]*v1.Pod, len(pods))
var wg sync.WaitGroup
for i, pod := range pods {
wg.Add(1)
go func(i int, pod *v1.Pod) {
defer wg.Done()
defer GinkgoRecover()
ps[i] = c.CreateSync(pod)
}(i, pod)
}
wg.Wait()
return ps
}
// Update updates the pod object. It retries if there is a conflict, throw out error if
// there is any other errors. name is the pod name, updateFn is the function updating the
// pod object.
func (c *PodClient) Update(name string, updateFn func(pod *v1.Pod)) {
ExpectNoError(wait.Poll(time.Millisecond*500, time.Second*30, func() (bool, error) {
pod, err := c.PodInterface.Get(name, metav1.GetOptions{})
if err != nil {
return false, fmt.Errorf("failed to get pod %q: %v", name, err)
}
updateFn(pod)
_, err = c.PodInterface.Update(pod)
if err == nil {
Logf("Successfully updated pod %q", name)
return true, nil
}
if errors.IsConflict(err) {
Logf("Conflicting update to pod %q, re-get and re-update: %v", name, err)
return false, nil
}
return false, fmt.Errorf("failed to update pod %q: %v", name, err)
}))
}
// DeleteSync deletes the pod and wait for the pod to disappear for `timeout`. If the pod doesn't
// disappear before the timeout, it will fail the test.
func (c *PodClient) DeleteSync(name string, options *metav1.DeleteOptions, timeout time.Duration) {
c.DeleteSyncInNamespace(name, c.f.Namespace.Name, options, timeout)
}
// DeleteSyncInNamespace deletes the pod from the namespace and wait for the pod to disappear for `timeout`. If the pod doesn't
// disappear before the timeout, it will fail the test.
func (c *PodClient) DeleteSyncInNamespace(name string, namespace string, options *metav1.DeleteOptions, timeout time.Duration) {
err := c.Delete(name, options)
if err != nil && !errors.IsNotFound(err) {
Failf("Failed to delete pod %q: %v", name, err)
}
Expect(WaitForPodToDisappear(c.f.ClientSet, namespace, name, labels.Everything(),
2*time.Second, timeout)).To(Succeed(), "wait for pod %q to disappear", name)
}
// mungeSpec apply test-suite specific transformations to the pod spec.
func (c *PodClient) mungeSpec(pod *v1.Pod) {
if !TestContext.NodeE2E {
return
}
Expect(pod.Spec.NodeName).To(Or(BeZero(), Equal(TestContext.NodeName)), "Test misconfigured")
pod.Spec.NodeName = TestContext.NodeName
// Node e2e does not support the default DNSClusterFirst policy. Set
// the policy to DNSDefault, which is configured per node.
pod.Spec.DNSPolicy = v1.DNSDefault
// PrepullImages only works for node e2e now. For cluster e2e, image prepull is not enforced,
// we should not munge ImagePullPolicy for cluster e2e pods.
if !TestContext.PrepullImages {
return
}
// If prepull is enabled, munge the container spec to make sure the images are not pulled
// during the test.
for i := range pod.Spec.Containers {
c := &pod.Spec.Containers[i]
if c.ImagePullPolicy == v1.PullAlways {
// If the image pull policy is PullAlways, the image doesn't need to be in
// the white list or pre-pulled, because the image is expected to be pulled
// in the test anyway.
continue
}
// If the image policy is not PullAlways, the image must be in the white list and
// pre-pulled.
Expect(ImageWhiteList.Has(c.Image)).To(BeTrue(), "Image %q is not in the white list, consider adding it to CommonImageWhiteList in test/e2e/common/util.go or NodeImageWhiteList in test/e2e_node/image_list.go", c.Image)
// Do not pull images during the tests because the images in white list should have
// been prepulled.
c.ImagePullPolicy = v1.PullNever
}
}
// TODO(random-liu): Move pod wait function into this file
// WaitForSuccess waits for pod to succeed.
func (c *PodClient) WaitForSuccess(name string, timeout time.Duration) {
f := c.f
Expect(WaitForPodCondition(f.ClientSet, f.Namespace.Name, name, "success or failure", timeout,
func(pod *v1.Pod) (bool, error) {
switch pod.Status.Phase {
case v1.PodFailed:
return true, fmt.Errorf("pod %q failed with reason: %q, message: %q", name, pod.Status.Reason, pod.Status.Message)
case v1.PodSucceeded:
return true, nil
default:
return false, nil
}
},
)).To(Succeed(), "wait for pod %q to success", name)
}
// WaitForFailure waits for pod to fail.
func (c *PodClient) WaitForFailure(name string, timeout time.Duration) {
f := c.f
Expect(WaitForPodCondition(f.ClientSet, f.Namespace.Name, name, "success or failure", timeout,
func(pod *v1.Pod) (bool, error) {
switch pod.Status.Phase {
case v1.PodFailed:
return true, nil
case v1.PodSucceeded:
return true, fmt.Errorf("pod %q successed with reason: %q, message: %q", name, pod.Status.Reason, pod.Status.Message)
default:
return false, nil
}
},
)).To(Succeed(), "wait for pod %q to fail", name)
}
// WaitForSuccess waits for pod to succeed or an error event for that pod.
func (c *PodClient) WaitForErrorEventOrSuccess(pod *v1.Pod) (*v1.Event, error) {
var ev *v1.Event
err := wait.Poll(Poll, PodStartTimeout, func() (bool, error) {
evnts, err := c.f.ClientSet.CoreV1().Events(pod.Namespace).Search(legacyscheme.Scheme, pod)
if err != nil {
return false, fmt.Errorf("error in listing events: %s", err)
}
for _, e := range evnts.Items {
switch e.Reason {
case events.KillingContainer, events.FailedToCreateContainer, sysctl.UnsupportedReason, sysctl.ForbiddenReason:
ev = &e
return true, nil
case events.StartedContainer:
return true, nil
default:
// ignore all other errors
}
}
return false, nil
})
return ev, err
}
// MatchContainerOutput gets output of a container and match expected regexp in the output.
func (c *PodClient) MatchContainerOutput(name string, containerName string, expectedRegexp string) error {
f := c.f
output, err := GetPodLogs(f.ClientSet, f.Namespace.Name, name, containerName)
if err != nil {
return fmt.Errorf("failed to get output for container %q of pod %q", containerName, name)
}
regex, err := regexp.Compile(expectedRegexp)
if err != nil {
return fmt.Errorf("failed to compile regexp %q: %v", expectedRegexp, err)
}
if !regex.MatchString(output) {
return fmt.Errorf("failed to match regexp %q in output %q", expectedRegexp, output)
}
return nil
}

143
vendor/k8s.io/kubernetes/test/e2e/framework/psp_util.go generated vendored Normal file
View File

@ -0,0 +1,143 @@
/*
Copyright 2017 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 framework
import (
"fmt"
"sync"
corev1 "k8s.io/api/core/v1"
extensionsv1beta1 "k8s.io/api/extensions/v1beta1"
rbacv1beta1 "k8s.io/api/rbac/v1beta1"
apierrs "k8s.io/apimachinery/pkg/api/errors"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/runtime/schema"
"k8s.io/apiserver/pkg/authentication/serviceaccount"
"k8s.io/kubernetes/pkg/security/podsecuritypolicy/seccomp"
. "github.com/onsi/ginkgo"
)
const (
podSecurityPolicyPrivileged = "e2e-test-privileged-psp"
)
var (
isPSPEnabledOnce sync.Once
isPSPEnabled bool
)
// Creates a PodSecurityPolicy that allows everything.
func PrivilegedPSP(name string) *extensionsv1beta1.PodSecurityPolicy {
allowPrivilegeEscalation := true
return &extensionsv1beta1.PodSecurityPolicy{
ObjectMeta: metav1.ObjectMeta{
Name: name,
Annotations: map[string]string{seccomp.AllowedProfilesAnnotationKey: seccomp.AllowAny},
},
Spec: extensionsv1beta1.PodSecurityPolicySpec{
Privileged: true,
AllowPrivilegeEscalation: &allowPrivilegeEscalation,
AllowedCapabilities: []corev1.Capability{"*"},
Volumes: []extensionsv1beta1.FSType{extensionsv1beta1.All},
HostNetwork: true,
HostPorts: []extensionsv1beta1.HostPortRange{{Min: 0, Max: 65535}},
HostIPC: true,
HostPID: true,
RunAsUser: extensionsv1beta1.RunAsUserStrategyOptions{
Rule: extensionsv1beta1.RunAsUserStrategyRunAsAny,
},
SELinux: extensionsv1beta1.SELinuxStrategyOptions{
Rule: extensionsv1beta1.SELinuxStrategyRunAsAny,
},
SupplementalGroups: extensionsv1beta1.SupplementalGroupsStrategyOptions{
Rule: extensionsv1beta1.SupplementalGroupsStrategyRunAsAny,
},
FSGroup: extensionsv1beta1.FSGroupStrategyOptions{
Rule: extensionsv1beta1.FSGroupStrategyRunAsAny,
},
ReadOnlyRootFilesystem: false,
},
}
}
func IsPodSecurityPolicyEnabled(f *Framework) bool {
isPSPEnabledOnce.Do(func() {
psps, err := f.ClientSet.ExtensionsV1beta1().PodSecurityPolicies().List(metav1.ListOptions{})
if err != nil {
Logf("Error listing PodSecurityPolicies; assuming PodSecurityPolicy is disabled: %v", err)
isPSPEnabled = false
} else if psps == nil || len(psps.Items) == 0 {
Logf("No PodSecurityPolicies found; assuming PodSecurityPolicy is disabled.")
isPSPEnabled = false
} else {
Logf("Found PodSecurityPolicies; assuming PodSecurityPolicy is enabled.")
isPSPEnabled = true
}
})
return isPSPEnabled
}
var (
privilegedPSPOnce sync.Once
)
func CreatePrivilegedPSPBinding(f *Framework, namespace string) {
if !IsPodSecurityPolicyEnabled(f) {
return
}
// Create the privileged PSP & role
privilegedPSPOnce.Do(func() {
_, err := f.ClientSet.ExtensionsV1beta1().PodSecurityPolicies().Get(
podSecurityPolicyPrivileged, metav1.GetOptions{})
if !apierrs.IsNotFound(err) {
// Privileged PSP was already created.
ExpectNoError(err, "Failed to get PodSecurityPolicy %s", podSecurityPolicyPrivileged)
return
}
psp := PrivilegedPSP(podSecurityPolicyPrivileged)
psp, err = f.ClientSet.ExtensionsV1beta1().PodSecurityPolicies().Create(psp)
ExpectNoError(err, "Failed to create PSP %s", podSecurityPolicyPrivileged)
// Create the Role to bind it to the namespace.
_, err = f.ClientSet.RbacV1beta1().ClusterRoles().Create(&rbacv1beta1.ClusterRole{
ObjectMeta: metav1.ObjectMeta{Name: podSecurityPolicyPrivileged},
Rules: []rbacv1beta1.PolicyRule{{
APIGroups: []string{"extensions"},
Resources: []string{"podsecuritypolicies"},
ResourceNames: []string{podSecurityPolicyPrivileged},
Verbs: []string{"use"},
}},
})
ExpectNoError(err, "Failed to create PSP role")
})
By(fmt.Sprintf("Binding the %s PodSecurityPolicy to the default service account in %s",
podSecurityPolicyPrivileged, namespace))
BindClusterRoleInNamespace(f.ClientSet.RbacV1beta1(),
podSecurityPolicyPrivileged,
namespace,
rbacv1beta1.Subject{
Kind: rbacv1beta1.ServiceAccountKind,
Namespace: namespace,
Name: "default",
})
ExpectNoError(WaitForNamedAuthorizationUpdate(f.ClientSet.AuthorizationV1beta1(),
serviceaccount.MakeUsername(namespace, "default"), namespace, "use", podSecurityPolicyPrivileged,
schema.GroupResource{Group: "extensions", Resource: "podsecuritypolicies"}, true))
}

985
vendor/k8s.io/kubernetes/test/e2e/framework/pv_util.go generated vendored Normal file
View File

@ -0,0 +1,985 @@
/*
Copyright 2015 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 framework
import (
"fmt"
"strings"
"time"
"github.com/aws/aws-sdk-go/aws"
"github.com/aws/aws-sdk-go/aws/awserr"
"github.com/aws/aws-sdk-go/aws/session"
"github.com/aws/aws-sdk-go/service/ec2"
. "github.com/onsi/ginkgo"
"google.golang.org/api/googleapi"
"k8s.io/api/core/v1"
apierrs "k8s.io/apimachinery/pkg/api/errors"
"k8s.io/apimachinery/pkg/api/resource"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/labels"
"k8s.io/apimachinery/pkg/types"
"k8s.io/apimachinery/pkg/util/uuid"
clientset "k8s.io/client-go/kubernetes"
"k8s.io/kubernetes/pkg/api/testapi"
"k8s.io/kubernetes/pkg/apis/core/v1/helper"
awscloud "k8s.io/kubernetes/pkg/cloudprovider/providers/aws"
gcecloud "k8s.io/kubernetes/pkg/cloudprovider/providers/gce"
"k8s.io/kubernetes/pkg/volume/util/volumehelper"
)
const (
PDRetryTimeout = 5 * time.Minute
PDRetryPollTime = 5 * time.Second
VolumeSelectorKey = "e2e-pv-pool"
)
// Map of all PVs used in the multi pv-pvc tests. The key is the PV's name, which is
// guaranteed to be unique. The value is {} (empty struct) since we're only interested
// in the PV's name and if it is present. We must always Get the pv object before
// referencing any of its values, eg its ClaimRef.
type pvval struct{}
type PVMap map[string]pvval
// Map of all PVCs used in the multi pv-pvc tests. The key is "namespace/pvc.Name". The
// value is {} (empty struct) since we're only interested in the PVC's name and if it is
// present. We must always Get the pvc object before referencing any of its values, eg.
// its VolumeName.
// Note: It's unsafe to add keys to a map in a loop. Their insertion in the map is
// unpredictable and can result in the same key being iterated over again.
type pvcval struct{}
type PVCMap map[types.NamespacedName]pvcval
// PersistentVolumeConfig is consumed by MakePersistentVolume() to generate a PV object
// for varying storage options (NFS, ceph, glusterFS, etc.).
// (+optional) prebind holds a pre-bound PVC
// Example pvSource:
// pvSource: api.PersistentVolumeSource{
// NFS: &api.NFSVolumeSource{
// ...
// },
// }
type PersistentVolumeConfig struct {
PVSource v1.PersistentVolumeSource
Prebind *v1.PersistentVolumeClaim
ReclaimPolicy v1.PersistentVolumeReclaimPolicy
NamePrefix string
Labels labels.Set
StorageClassName string
NodeAffinity *v1.NodeAffinity
}
// PersistentVolumeClaimConfig is consumed by MakePersistentVolumeClaim() to generate a PVC object.
// AccessModes defaults to all modes (RWO, RWX, ROX) if left empty
// (+optional) Annotations defines the PVC's annotations
type PersistentVolumeClaimConfig struct {
AccessModes []v1.PersistentVolumeAccessMode
Annotations map[string]string
Selector *metav1.LabelSelector
StorageClassName *string
}
// Clean up a pv and pvc in a single pv/pvc test case.
// Note: delete errors are appended to []error so that we can attempt to delete both the pvc and pv.
func PVPVCCleanup(c clientset.Interface, ns string, pv *v1.PersistentVolume, pvc *v1.PersistentVolumeClaim) []error {
var errs []error
if pvc != nil {
err := DeletePersistentVolumeClaim(c, pvc.Name, ns)
if err != nil {
errs = append(errs, fmt.Errorf("failed to delete PVC %q: %v", pvc.Name, err))
}
} else {
Logf("pvc is nil")
}
if pv != nil {
err := DeletePersistentVolume(c, pv.Name)
if err != nil {
errs = append(errs, fmt.Errorf("failed to delete PV %q: %v", pv.Name, err))
}
} else {
Logf("pv is nil")
}
return errs
}
// Clean up pvs and pvcs in multi-pv-pvc test cases. Entries found in the pv and claim maps are
// deleted as long as the Delete api call succeeds.
// Note: delete errors are appended to []error so that as many pvcs and pvs as possible are deleted.
func PVPVCMapCleanup(c clientset.Interface, ns string, pvols PVMap, claims PVCMap) []error {
var errs []error
for pvcKey := range claims {
err := DeletePersistentVolumeClaim(c, pvcKey.Name, ns)
if err != nil {
errs = append(errs, fmt.Errorf("failed to delete PVC %q: %v", pvcKey.Name, err))
} else {
delete(claims, pvcKey)
}
}
for pvKey := range pvols {
err := DeletePersistentVolume(c, pvKey)
if err != nil {
errs = append(errs, fmt.Errorf("failed to delete PV %q: %v", pvKey, err))
} else {
delete(pvols, pvKey)
}
}
return errs
}
// Delete the PV.
func DeletePersistentVolume(c clientset.Interface, pvName string) error {
if c != nil && len(pvName) > 0 {
Logf("Deleting PersistentVolume %q", pvName)
err := c.CoreV1().PersistentVolumes().Delete(pvName, nil)
if err != nil && !apierrs.IsNotFound(err) {
return fmt.Errorf("PV Delete API error: %v", err)
}
}
return nil
}
// Delete the Claim
func DeletePersistentVolumeClaim(c clientset.Interface, pvcName string, ns string) error {
if c != nil && len(pvcName) > 0 {
Logf("Deleting PersistentVolumeClaim %q", pvcName)
err := c.CoreV1().PersistentVolumeClaims(ns).Delete(pvcName, nil)
if err != nil && !apierrs.IsNotFound(err) {
return fmt.Errorf("PVC Delete API error: %v", err)
}
}
return nil
}
// Delete the PVC and wait for the PV to enter its expected phase. Validate that the PV
// has been reclaimed (assumption here about reclaimPolicy). Caller tells this func which
// phase value to expect for the pv bound to the to-be-deleted claim.
func DeletePVCandValidatePV(c clientset.Interface, ns string, pvc *v1.PersistentVolumeClaim, pv *v1.PersistentVolume, expectPVPhase v1.PersistentVolumePhase) error {
pvname := pvc.Spec.VolumeName
Logf("Deleting PVC %v to trigger reclamation of PV %v", pvc.Name, pvname)
err := DeletePersistentVolumeClaim(c, pvc.Name, ns)
if err != nil {
return err
}
// Wait for the PV's phase to return to be `expectPVPhase`
Logf("Waiting for reclaim process to complete.")
err = WaitForPersistentVolumePhase(expectPVPhase, c, pv.Name, 1*time.Second, 300*time.Second)
if err != nil {
return fmt.Errorf("pv %q phase did not become %v: %v", pv.Name, expectPVPhase, err)
}
// examine the pv's ClaimRef and UID and compare to expected values
pv, err = c.CoreV1().PersistentVolumes().Get(pv.Name, metav1.GetOptions{})
if err != nil {
return fmt.Errorf("PV Get API error: %v", err)
}
cr := pv.Spec.ClaimRef
if expectPVPhase == v1.VolumeAvailable {
if cr != nil && len(cr.UID) > 0 {
return fmt.Errorf("PV is 'Available' but ClaimRef.UID is not empty")
}
} else if expectPVPhase == v1.VolumeBound {
if cr == nil {
return fmt.Errorf("PV is 'Bound' but ClaimRef is nil")
}
if len(cr.UID) == 0 {
return fmt.Errorf("PV is 'Bound' but ClaimRef.UID is empty")
}
}
Logf("PV %v now in %q phase", pv.Name, expectPVPhase)
return nil
}
// Wraps deletePVCandValidatePV() by calling the function in a loop over the PV map. Only bound PVs
// are deleted. Validates that the claim was deleted and the PV is in the expected Phase (Released,
// Available, Bound).
// Note: if there are more claims than pvs then some of the remaining claims may bind to just made
// available pvs.
func DeletePVCandValidatePVGroup(c clientset.Interface, ns string, pvols PVMap, claims PVCMap, expectPVPhase v1.PersistentVolumePhase) error {
var boundPVs, deletedPVCs int
for pvName := range pvols {
pv, err := c.CoreV1().PersistentVolumes().Get(pvName, metav1.GetOptions{})
if err != nil {
return fmt.Errorf("PV Get API error: %v", err)
}
cr := pv.Spec.ClaimRef
// if pv is bound then delete the pvc it is bound to
if cr != nil && len(cr.Name) > 0 {
boundPVs++
// Assert bound PVC is tracked in this test. Failing this might
// indicate external PVCs interfering with the test.
pvcKey := makePvcKey(ns, cr.Name)
if _, found := claims[pvcKey]; !found {
return fmt.Errorf("internal: claims map is missing pvc %q", pvcKey)
}
// get the pvc for the delete call below
pvc, err := c.CoreV1().PersistentVolumeClaims(ns).Get(cr.Name, metav1.GetOptions{})
if err == nil {
if err = DeletePVCandValidatePV(c, ns, pvc, pv, expectPVPhase); err != nil {
return err
}
} else if !apierrs.IsNotFound(err) {
return fmt.Errorf("PVC Get API error: %v", err)
}
// delete pvckey from map even if apierrs.IsNotFound above is true and thus the
// claim was not actually deleted here
delete(claims, pvcKey)
deletedPVCs++
}
}
if boundPVs != deletedPVCs {
return fmt.Errorf("expect number of bound PVs (%v) to equal number of deleted PVCs (%v)", boundPVs, deletedPVCs)
}
return nil
}
// create the PV resource. Fails test on error.
func createPV(c clientset.Interface, pv *v1.PersistentVolume) (*v1.PersistentVolume, error) {
pv, err := c.CoreV1().PersistentVolumes().Create(pv)
if err != nil {
return nil, fmt.Errorf("PV Create API error: %v", err)
}
return pv, nil
}
// create the PVC resource. Fails test on error.
func CreatePVC(c clientset.Interface, ns string, pvc *v1.PersistentVolumeClaim) (*v1.PersistentVolumeClaim, error) {
pvc, err := c.CoreV1().PersistentVolumeClaims(ns).Create(pvc)
if err != nil {
return nil, fmt.Errorf("PVC Create API error: %v", err)
}
return pvc, nil
}
// Create a PVC followed by the PV based on the passed in nfs-server ip and
// namespace. If the "preBind" bool is true then pre-bind the PV to the PVC
// via the PV's ClaimRef. Return the pv and pvc to reflect the created objects.
// Note: in the pre-bind case the real PVC name, which is generated, is not
// known until after the PVC is instantiated. This is why the pvc is created
// before the pv.
func CreatePVCPV(c clientset.Interface, pvConfig PersistentVolumeConfig, pvcConfig PersistentVolumeClaimConfig, ns string, preBind bool) (*v1.PersistentVolume, *v1.PersistentVolumeClaim, error) {
// make the pvc spec
pvc := MakePersistentVolumeClaim(pvcConfig, ns)
preBindMsg := ""
if preBind {
preBindMsg = " pre-bound"
pvConfig.Prebind = pvc
}
// make the pv spec
pv := MakePersistentVolume(pvConfig)
By(fmt.Sprintf("Creating a PVC followed by a%s PV", preBindMsg))
pvc, err := CreatePVC(c, ns, pvc)
if err != nil {
return nil, nil, err
}
// instantiate the pv, handle pre-binding by ClaimRef if needed
if preBind {
pv.Spec.ClaimRef.Name = pvc.Name
}
pv, err = createPV(c, pv)
if err != nil {
return nil, pvc, err
}
return pv, pvc, nil
}
// Create a PV followed by the PVC based on the passed in nfs-server ip and
// namespace. If the "preBind" bool is true then pre-bind the PVC to the PV
// via the PVC's VolumeName. Return the pv and pvc to reflect the created
// objects.
// Note: in the pre-bind case the real PV name, which is generated, is not
// known until after the PV is instantiated. This is why the pv is created
// before the pvc.
func CreatePVPVC(c clientset.Interface, pvConfig PersistentVolumeConfig, pvcConfig PersistentVolumeClaimConfig, ns string, preBind bool) (*v1.PersistentVolume, *v1.PersistentVolumeClaim, error) {
preBindMsg := ""
if preBind {
preBindMsg = " pre-bound"
}
Logf("Creating a PV followed by a%s PVC", preBindMsg)
// make the pv and pvc definitions
pv := MakePersistentVolume(pvConfig)
pvc := MakePersistentVolumeClaim(pvcConfig, ns)
// instantiate the pv
pv, err := createPV(c, pv)
if err != nil {
return nil, nil, err
}
// instantiate the pvc, handle pre-binding by VolumeName if needed
if preBind {
pvc.Spec.VolumeName = pv.Name
}
pvc, err = CreatePVC(c, ns, pvc)
if err != nil {
return pv, nil, err
}
return pv, pvc, nil
}
// Create the desired number of PVs and PVCs and return them in separate maps. If the
// number of PVs != the number of PVCs then the min of those two counts is the number of
// PVs expected to bind. If a Create error occurs, the returned maps may contain pv and pvc
// entries for the resources that were successfully created. In other words, when the caller
// sees an error returned, it needs to decide what to do about entries in the maps.
// Note: when the test suite deletes the namespace orphaned pvcs and pods are deleted. However,
// orphaned pvs are not deleted and will remain after the suite completes.
func CreatePVsPVCs(numpvs, numpvcs int, c clientset.Interface, ns string, pvConfig PersistentVolumeConfig, pvcConfig PersistentVolumeClaimConfig) (PVMap, PVCMap, error) {
pvMap := make(PVMap, numpvs)
pvcMap := make(PVCMap, numpvcs)
extraPVCs := 0
extraPVs := numpvs - numpvcs
if extraPVs < 0 {
extraPVCs = -extraPVs
extraPVs = 0
}
pvsToCreate := numpvs - extraPVs // want the min(numpvs, numpvcs)
// create pvs and pvcs
for i := 0; i < pvsToCreate; i++ {
pv, pvc, err := CreatePVPVC(c, pvConfig, pvcConfig, ns, false)
if err != nil {
return pvMap, pvcMap, err
}
pvMap[pv.Name] = pvval{}
pvcMap[makePvcKey(ns, pvc.Name)] = pvcval{}
}
// create extra pvs or pvcs as needed
for i := 0; i < extraPVs; i++ {
pv := MakePersistentVolume(pvConfig)
pv, err := createPV(c, pv)
if err != nil {
return pvMap, pvcMap, err
}
pvMap[pv.Name] = pvval{}
}
for i := 0; i < extraPVCs; i++ {
pvc := MakePersistentVolumeClaim(pvcConfig, ns)
pvc, err := CreatePVC(c, ns, pvc)
if err != nil {
return pvMap, pvcMap, err
}
pvcMap[makePvcKey(ns, pvc.Name)] = pvcval{}
}
return pvMap, pvcMap, nil
}
// Wait for the pv and pvc to bind to each other.
func WaitOnPVandPVC(c clientset.Interface, ns string, pv *v1.PersistentVolume, pvc *v1.PersistentVolumeClaim) error {
// Wait for newly created PVC to bind to the PV
Logf("Waiting for PV %v to bind to PVC %v", pv.Name, pvc.Name)
err := WaitForPersistentVolumeClaimPhase(v1.ClaimBound, c, ns, pvc.Name, 3*time.Second, 300*time.Second)
if err != nil {
return fmt.Errorf("PVC %q did not become Bound: %v", pvc.Name, err)
}
// Wait for PersistentVolume.Status.Phase to be Bound, which it should be
// since the PVC is already bound.
err = WaitForPersistentVolumePhase(v1.VolumeBound, c, pv.Name, 3*time.Second, 300*time.Second)
if err != nil {
return fmt.Errorf("PV %q did not become Bound: %v", pv.Name, err)
}
// Re-get the pv and pvc objects
pv, err = c.CoreV1().PersistentVolumes().Get(pv.Name, metav1.GetOptions{})
if err != nil {
return fmt.Errorf("PV Get API error: %v", err)
}
pvc, err = c.CoreV1().PersistentVolumeClaims(ns).Get(pvc.Name, metav1.GetOptions{})
if err != nil {
return fmt.Errorf("PVC Get API error: %v", err)
}
// The pv and pvc are both bound, but to each other?
// Check that the PersistentVolume.ClaimRef matches the PVC
if pv.Spec.ClaimRef == nil {
return fmt.Errorf("PV %q ClaimRef is nil", pv.Name)
}
if pv.Spec.ClaimRef.Name != pvc.Name {
return fmt.Errorf("PV %q ClaimRef's name (%q) should be %q", pv.Name, pv.Spec.ClaimRef.Name, pvc.Name)
}
if pvc.Spec.VolumeName != pv.Name {
return fmt.Errorf("PVC %q VolumeName (%q) should be %q", pvc.Name, pvc.Spec.VolumeName, pv.Name)
}
if pv.Spec.ClaimRef.UID != pvc.UID {
return fmt.Errorf("PV %q ClaimRef's UID (%q) should be %q", pv.Name, pv.Spec.ClaimRef.UID, pvc.UID)
}
return nil
}
// Search for bound PVs and PVCs by examining pvols for non-nil claimRefs.
// NOTE: Each iteration waits for a maximum of 3 minutes per PV and, if the PV is bound,
// up to 3 minutes for the PVC. When the number of PVs != number of PVCs, this can lead
// to situations where the maximum wait times are reached several times in succession,
// extending test time. Thus, it is recommended to keep the delta between PVs and PVCs
// small.
func WaitAndVerifyBinds(c clientset.Interface, ns string, pvols PVMap, claims PVCMap, testExpected bool) error {
var actualBinds int
expectedBinds := len(pvols)
if expectedBinds > len(claims) { // want the min of # pvs or #pvcs
expectedBinds = len(claims)
}
for pvName := range pvols {
err := WaitForPersistentVolumePhase(v1.VolumeBound, c, pvName, 3*time.Second, 180*time.Second)
if err != nil && len(pvols) > len(claims) {
Logf("WARN: pv %v is not bound after max wait", pvName)
Logf(" This may be ok since there are more pvs than pvcs")
continue
}
if err != nil {
return fmt.Errorf("PV %q did not become Bound: %v", pvName, err)
}
pv, err := c.CoreV1().PersistentVolumes().Get(pvName, metav1.GetOptions{})
if err != nil {
return fmt.Errorf("PV Get API error: %v", err)
}
cr := pv.Spec.ClaimRef
if cr != nil && len(cr.Name) > 0 {
// Assert bound pvc is a test resource. Failing assertion could
// indicate non-test PVC interference or a bug in the test
pvcKey := makePvcKey(ns, cr.Name)
if _, found := claims[pvcKey]; !found {
return fmt.Errorf("internal: claims map is missing pvc %q", pvcKey)
}
err := WaitForPersistentVolumeClaimPhase(v1.ClaimBound, c, ns, cr.Name, 3*time.Second, 180*time.Second)
if err != nil {
return fmt.Errorf("PVC %q did not become Bound: %v", cr.Name, err)
}
actualBinds++
}
}
if testExpected && actualBinds != expectedBinds {
return fmt.Errorf("expect number of bound PVs (%v) to equal number of claims (%v)", actualBinds, expectedBinds)
}
return nil
}
// Test the pod's exit code to be zero.
func testPodSuccessOrFail(c clientset.Interface, ns string, pod *v1.Pod) error {
By("Pod should terminate with exitcode 0 (success)")
if err := WaitForPodSuccessInNamespace(c, pod.Name, ns); err != nil {
return fmt.Errorf("pod %q failed to reach Success: %v", pod.Name, err)
}
Logf("Pod %v succeeded ", pod.Name)
return nil
}
// Deletes the passed-in pod and waits for the pod to be terminated. Resilient to the pod
// not existing.
func DeletePodWithWait(f *Framework, c clientset.Interface, pod *v1.Pod) error {
const maxWait = 5 * time.Minute
if pod == nil {
return nil
}
Logf("Deleting pod %q in namespace %q", pod.Name, pod.Namespace)
err := c.CoreV1().Pods(pod.Namespace).Delete(pod.Name, nil)
if err != nil {
if apierrs.IsNotFound(err) {
return nil // assume pod was already deleted
}
return fmt.Errorf("pod Delete API error: %v", err)
}
Logf("Wait up to %v for pod %q to be fully deleted", maxWait, pod.Name)
err = f.WaitForPodNotFound(pod.Name, maxWait)
if err != nil {
return fmt.Errorf("pod %q was not deleted: %v", pod.Name, err)
}
return nil
}
// Create the test pod, wait for (hopefully) success, and then delete the pod.
// Note: need named return value so that the err assignment in the defer sets the returned error.
// Has been shown to be necessary using Go 1.7.
func CreateWaitAndDeletePod(f *Framework, c clientset.Interface, ns string, pvc *v1.PersistentVolumeClaim) (err error) {
Logf("Creating nfs test pod")
pod := MakeWritePod(ns, pvc)
runPod, err := c.CoreV1().Pods(ns).Create(pod)
if err != nil {
return fmt.Errorf("pod Create API error: %v", err)
}
defer func() {
delErr := DeletePodWithWait(f, c, runPod)
if err == nil { // don't override previous err value
err = delErr // assign to returned err, can be nil
}
}()
err = testPodSuccessOrFail(c, ns, runPod)
if err != nil {
return fmt.Errorf("pod %q did not exit with Success: %v", runPod.Name, err)
}
return // note: named return value
}
// Sanity check for GCE testing. Verify the persistent disk attached to the node.
func VerifyGCEDiskAttached(diskName string, nodeName types.NodeName) (bool, error) {
gceCloud, err := GetGCECloud()
if err != nil {
return false, fmt.Errorf("GetGCECloud error: %v", err)
}
isAttached, err := gceCloud.DiskIsAttached(diskName, nodeName)
if err != nil {
return false, fmt.Errorf("cannot verify if GCE disk is attached: %v", err)
}
return isAttached, nil
}
// Return a pvckey struct.
func makePvcKey(ns, name string) types.NamespacedName {
return types.NamespacedName{Namespace: ns, Name: name}
}
// Returns a PV definition based on the nfs server IP. If the PVC is not nil
// then the PV is defined with a ClaimRef which includes the PVC's namespace.
// If the PVC is nil then the PV is not defined with a ClaimRef. If no reclaimPolicy
// is assigned, assumes "Retain". Specs are expected to match the test's PVC.
// Note: the passed-in claim does not have a name until it is created and thus the PV's
// ClaimRef cannot be completely filled-in in this func. Therefore, the ClaimRef's name
// is added later in CreatePVCPV.
func MakePersistentVolume(pvConfig PersistentVolumeConfig) *v1.PersistentVolume {
var claimRef *v1.ObjectReference
// If the reclaimPolicy is not provided, assume Retain
if pvConfig.ReclaimPolicy == "" {
Logf("PV ReclaimPolicy unspecified, default: Retain")
pvConfig.ReclaimPolicy = v1.PersistentVolumeReclaimRetain
}
if pvConfig.Prebind != nil {
claimRef = &v1.ObjectReference{
Name: pvConfig.Prebind.Name,
Namespace: pvConfig.Prebind.Namespace,
}
}
pv := &v1.PersistentVolume{
ObjectMeta: metav1.ObjectMeta{
GenerateName: pvConfig.NamePrefix,
Labels: pvConfig.Labels,
Annotations: map[string]string{
volumehelper.VolumeGidAnnotationKey: "777",
},
},
Spec: v1.PersistentVolumeSpec{
PersistentVolumeReclaimPolicy: pvConfig.ReclaimPolicy,
Capacity: v1.ResourceList{
v1.ResourceName(v1.ResourceStorage): resource.MustParse("2Gi"),
},
PersistentVolumeSource: pvConfig.PVSource,
AccessModes: []v1.PersistentVolumeAccessMode{
v1.ReadWriteOnce,
v1.ReadOnlyMany,
v1.ReadWriteMany,
},
ClaimRef: claimRef,
StorageClassName: pvConfig.StorageClassName,
},
}
err := helper.StorageNodeAffinityToAlphaAnnotation(pv.Annotations, pvConfig.NodeAffinity)
if err != nil {
Logf("Setting storage node affinity failed: %v", err)
return nil
}
return pv
}
// Returns a PVC definition based on the namespace.
// Note: if this PVC is intended to be pre-bound to a PV, whose name is not
// known until the PV is instantiated, then the func CreatePVPVC will add
// pvc.Spec.VolumeName to this claim.
func MakePersistentVolumeClaim(cfg PersistentVolumeClaimConfig, ns string) *v1.PersistentVolumeClaim {
// Specs are expected to match this test's PersistentVolume
if len(cfg.AccessModes) == 0 {
Logf("AccessModes unspecified, default: all modes (RWO, RWX, ROX).")
cfg.AccessModes = append(cfg.AccessModes, v1.ReadWriteOnce, v1.ReadOnlyMany, v1.ReadOnlyMany)
}
return &v1.PersistentVolumeClaim{
ObjectMeta: metav1.ObjectMeta{
GenerateName: "pvc-",
Namespace: ns,
Annotations: cfg.Annotations,
},
Spec: v1.PersistentVolumeClaimSpec{
Selector: cfg.Selector,
AccessModes: cfg.AccessModes,
Resources: v1.ResourceRequirements{
Requests: v1.ResourceList{
v1.ResourceName(v1.ResourceStorage): resource.MustParse("1Gi"),
},
},
StorageClassName: cfg.StorageClassName,
},
}
}
func createPDWithRetry(zone string) (string, error) {
var err error
for start := time.Now(); time.Since(start) < PDRetryTimeout; time.Sleep(PDRetryPollTime) {
newDiskName, err := createPD(zone)
if err != nil {
Logf("Couldn't create a new PD, sleeping 5 seconds: %v", err)
continue
}
Logf("Successfully created a new PD: %q.", newDiskName)
return newDiskName, nil
}
return "", err
}
func CreatePDWithRetry() (string, error) {
return createPDWithRetry("")
}
func CreatePDWithRetryAndZone(zone string) (string, error) {
return createPDWithRetry(zone)
}
func DeletePDWithRetry(diskName string) error {
var err error
for start := time.Now(); time.Since(start) < PDRetryTimeout; time.Sleep(PDRetryPollTime) {
err = deletePD(diskName)
if err != nil {
Logf("Couldn't delete PD %q, sleeping %v: %v", diskName, PDRetryPollTime, err)
continue
}
Logf("Successfully deleted PD %q.", diskName)
return nil
}
return fmt.Errorf("unable to delete PD %q: %v", diskName, err)
}
func createPD(zone string) (string, error) {
if zone == "" {
zone = TestContext.CloudConfig.Zone
}
if TestContext.Provider == "gce" || TestContext.Provider == "gke" {
pdName := fmt.Sprintf("%s-%s", TestContext.Prefix, string(uuid.NewUUID()))
gceCloud, err := GetGCECloud()
if err != nil {
return "", err
}
tags := map[string]string{}
err = gceCloud.CreateDisk(pdName, gcecloud.DiskTypeSSD, zone, 10 /* sizeGb */, tags)
if err != nil {
return "", err
}
return pdName, nil
} else if TestContext.Provider == "aws" {
client := ec2.New(session.New())
request := &ec2.CreateVolumeInput{}
request.AvailabilityZone = aws.String(zone)
request.Size = aws.Int64(10)
request.VolumeType = aws.String(awscloud.DefaultVolumeType)
response, err := client.CreateVolume(request)
if err != nil {
return "", err
}
az := aws.StringValue(response.AvailabilityZone)
awsID := aws.StringValue(response.VolumeId)
volumeName := "aws://" + az + "/" + awsID
return volumeName, nil
} else if TestContext.Provider == "azure" {
pdName := fmt.Sprintf("%s-%s", TestContext.Prefix, string(uuid.NewUUID()))
azureCloud, err := GetAzureCloud()
if err != nil {
return "", err
}
_, diskURI, _, err := azureCloud.CreateVolume(pdName, "" /* account */, "" /* sku */, "" /* location */, 1 /* sizeGb */)
if err != nil {
return "", err
}
return diskURI, nil
} else {
return "", fmt.Errorf("provider does not support volume creation")
}
}
func deletePD(pdName string) error {
if TestContext.Provider == "gce" || TestContext.Provider == "gke" {
gceCloud, err := GetGCECloud()
if err != nil {
return err
}
err = gceCloud.DeleteDisk(pdName)
if err != nil {
if gerr, ok := err.(*googleapi.Error); ok && len(gerr.Errors) > 0 && gerr.Errors[0].Reason == "notFound" {
// PD already exists, ignore error.
return nil
}
Logf("error deleting PD %q: %v", pdName, err)
}
return err
} else if TestContext.Provider == "aws" {
client := ec2.New(session.New())
tokens := strings.Split(pdName, "/")
awsVolumeID := tokens[len(tokens)-1]
request := &ec2.DeleteVolumeInput{VolumeId: aws.String(awsVolumeID)}
_, err := client.DeleteVolume(request)
if err != nil {
if awsError, ok := err.(awserr.Error); ok && awsError.Code() == "InvalidVolume.NotFound" {
Logf("volume deletion implicitly succeeded because volume %q does not exist.", pdName)
} else {
return fmt.Errorf("error deleting EBS volumes: %v", err)
}
}
return nil
} else if TestContext.Provider == "azure" {
azureCloud, err := GetAzureCloud()
if err != nil {
return err
}
err = azureCloud.DeleteVolume(pdName)
if err != nil {
Logf("failed to delete Azure volume %q: %v", pdName, err)
return err
}
return nil
} else {
return fmt.Errorf("provider does not support volume deletion")
}
}
// Returns a pod definition based on the namespace. The pod references the PVC's
// name.
func MakeWritePod(ns string, pvc *v1.PersistentVolumeClaim) *v1.Pod {
return MakePod(ns, nil, []*v1.PersistentVolumeClaim{pvc}, true, "touch /mnt/volume1/SUCCESS && (id -G | grep -E '\\b777\\b')")
}
// Returns a pod definition based on the namespace. The pod references the PVC's
// name. A slice of BASH commands can be supplied as args to be run by the pod
func MakePod(ns string, nodeSelector map[string]string, pvclaims []*v1.PersistentVolumeClaim, isPrivileged bool, command string) *v1.Pod {
if len(command) == 0 {
command = "while true; do sleep 1; done"
}
podSpec := &v1.Pod{
TypeMeta: metav1.TypeMeta{
Kind: "Pod",
APIVersion: testapi.Groups[v1.GroupName].GroupVersion().String(),
},
ObjectMeta: metav1.ObjectMeta{
GenerateName: "pvc-tester-",
Namespace: ns,
},
Spec: v1.PodSpec{
Containers: []v1.Container{
{
Name: "write-pod",
Image: BusyBoxImage,
Command: []string{"/bin/sh"},
Args: []string{"-c", command},
SecurityContext: &v1.SecurityContext{
Privileged: &isPrivileged,
},
},
},
RestartPolicy: v1.RestartPolicyOnFailure,
},
}
var volumeMounts = make([]v1.VolumeMount, len(pvclaims))
var volumes = make([]v1.Volume, len(pvclaims))
for index, pvclaim := range pvclaims {
volumename := fmt.Sprintf("volume%v", index+1)
volumeMounts[index] = v1.VolumeMount{Name: volumename, MountPath: "/mnt/" + volumename}
volumes[index] = v1.Volume{Name: volumename, VolumeSource: v1.VolumeSource{PersistentVolumeClaim: &v1.PersistentVolumeClaimVolumeSource{ClaimName: pvclaim.Name, ReadOnly: false}}}
}
podSpec.Spec.Containers[0].VolumeMounts = volumeMounts
podSpec.Spec.Volumes = volumes
if nodeSelector != nil {
podSpec.Spec.NodeSelector = nodeSelector
}
return podSpec
}
// Returns a pod definition based on the namespace. The pod references the PVC's
// name. A slice of BASH commands can be supplied as args to be run by the pod.
// SELinux testing requires to pass HostIPC and HostPID as booleansi arguments.
func MakeSecPod(ns string, pvclaims []*v1.PersistentVolumeClaim, isPrivileged bool, command string, hostIPC bool, hostPID bool, seLinuxLabel *v1.SELinuxOptions) *v1.Pod {
if len(command) == 0 {
command = "while true; do sleep 1; done"
}
podName := "security-context-" + string(uuid.NewUUID())
fsGroup := int64(1000)
podSpec := &v1.Pod{
TypeMeta: metav1.TypeMeta{
Kind: "Pod",
APIVersion: testapi.Groups[v1.GroupName].GroupVersion().String(),
},
ObjectMeta: metav1.ObjectMeta{
Name: podName,
Namespace: ns,
},
Spec: v1.PodSpec{
HostIPC: hostIPC,
HostPID: hostPID,
SecurityContext: &v1.PodSecurityContext{
FSGroup: &fsGroup,
},
Containers: []v1.Container{
{
Name: "write-pod",
Image: "gcr.io/google_containers/busybox:1.24",
Command: []string{"/bin/sh"},
Args: []string{"-c", command},
SecurityContext: &v1.SecurityContext{
Privileged: &isPrivileged,
},
},
},
RestartPolicy: v1.RestartPolicyOnFailure,
},
}
var volumeMounts = make([]v1.VolumeMount, len(pvclaims))
var volumes = make([]v1.Volume, len(pvclaims))
for index, pvclaim := range pvclaims {
volumename := fmt.Sprintf("volume%v", index+1)
volumeMounts[index] = v1.VolumeMount{Name: volumename, MountPath: "/mnt/" + volumename}
volumes[index] = v1.Volume{Name: volumename, VolumeSource: v1.VolumeSource{PersistentVolumeClaim: &v1.PersistentVolumeClaimVolumeSource{ClaimName: pvclaim.Name, ReadOnly: false}}}
}
podSpec.Spec.Containers[0].VolumeMounts = volumeMounts
podSpec.Spec.Volumes = volumes
podSpec.Spec.SecurityContext.SELinuxOptions = seLinuxLabel
return podSpec
}
// CreatePod with given claims based on node selector
func CreatePod(client clientset.Interface, namespace string, nodeSelector map[string]string, pvclaims []*v1.PersistentVolumeClaim, isPrivileged bool, command string) (*v1.Pod, error) {
pod := MakePod(namespace, nodeSelector, pvclaims, isPrivileged, command)
pod, err := client.CoreV1().Pods(namespace).Create(pod)
if err != nil {
return nil, fmt.Errorf("pod Create API error: %v", err)
}
// Waiting for pod to be running
err = WaitForPodNameRunningInNamespace(client, pod.Name, namespace)
if err != nil {
return pod, fmt.Errorf("pod %q is not Running: %v", pod.Name, err)
}
// get fresh pod info
pod, err = client.CoreV1().Pods(namespace).Get(pod.Name, metav1.GetOptions{})
if err != nil {
return pod, fmt.Errorf("pod Get API error: %v", err)
}
return pod, nil
}
// create security pod with given claims
func CreateSecPod(client clientset.Interface, namespace string, pvclaims []*v1.PersistentVolumeClaim, isPrivileged bool, command string, hostIPC bool, hostPID bool, seLinuxLabel *v1.SELinuxOptions) (*v1.Pod, error) {
pod := MakeSecPod(namespace, pvclaims, isPrivileged, command, hostIPC, hostPID, seLinuxLabel)
pod, err := client.CoreV1().Pods(namespace).Create(pod)
if err != nil {
return nil, fmt.Errorf("pod Create API error: %v", err)
}
// Waiting for pod to be running
err = WaitForPodNameRunningInNamespace(client, pod.Name, namespace)
if err != nil {
return pod, fmt.Errorf("pod %q is not Running: %v", pod.Name, err)
}
// get fresh pod info
pod, err = client.CoreV1().Pods(namespace).Get(pod.Name, metav1.GetOptions{})
if err != nil {
return pod, fmt.Errorf("pod Get API error: %v", err)
}
return pod, nil
}
// Define and create a pod with a mounted PV. Pod runs infinite loop until killed.
func CreateClientPod(c clientset.Interface, ns string, pvc *v1.PersistentVolumeClaim) (*v1.Pod, error) {
return CreatePod(c, ns, nil, []*v1.PersistentVolumeClaim{pvc}, true, "")
}
// wait until all pvcs phase set to bound
func WaitForPVClaimBoundPhase(client clientset.Interface, pvclaims []*v1.PersistentVolumeClaim, timeout time.Duration) ([]*v1.PersistentVolume, error) {
persistentvolumes := make([]*v1.PersistentVolume, len(pvclaims))
for index, claim := range pvclaims {
err := WaitForPersistentVolumeClaimPhase(v1.ClaimBound, client, claim.Namespace, claim.Name, Poll, timeout)
if err != nil {
return persistentvolumes, err
}
// Get new copy of the claim
claim, err = client.CoreV1().PersistentVolumeClaims(claim.Namespace).Get(claim.Name, metav1.GetOptions{})
if err != nil {
return persistentvolumes, fmt.Errorf("PVC Get API error: %v", err)
}
// Get the bounded PV
persistentvolumes[index], err = client.CoreV1().PersistentVolumes().Get(claim.Spec.VolumeName, metav1.GetOptions{})
if err != nil {
return persistentvolumes, fmt.Errorf("PV Get API error: %v", err)
}
}
return persistentvolumes, nil
}
func CreatePVSource(zone string) (*v1.PersistentVolumeSource, error) {
diskName, err := CreatePDWithRetryAndZone(zone)
if err != nil {
return nil, err
}
if TestContext.Provider == "gce" || TestContext.Provider == "gke" {
return &v1.PersistentVolumeSource{
GCEPersistentDisk: &v1.GCEPersistentDiskVolumeSource{
PDName: diskName,
FSType: "ext3",
ReadOnly: false,
},
}, nil
} else if TestContext.Provider == "aws" {
return &v1.PersistentVolumeSource{
AWSElasticBlockStore: &v1.AWSElasticBlockStoreVolumeSource{
VolumeID: diskName,
FSType: "ext3",
},
}, nil
} else {
return nil, fmt.Errorf("Provider not supported")
}
}
func DeletePVSource(pvSource *v1.PersistentVolumeSource) error {
if TestContext.Provider == "gce" || TestContext.Provider == "gke" {
return DeletePDWithRetry(pvSource.GCEPersistentDisk.PDName)
} else if TestContext.Provider == "aws" {
return DeletePDWithRetry(pvSource.AWSElasticBlockStore.VolumeID)
} else {
return fmt.Errorf("Provider not supported")
}
}

287
vendor/k8s.io/kubernetes/test/e2e/framework/rc_util.go generated vendored Normal file
View File

@ -0,0 +1,287 @@
/*
Copyright 2017 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 framework
import (
"fmt"
"strings"
"time"
. "github.com/onsi/ginkgo"
"k8s.io/api/core/v1"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/labels"
"k8s.io/apimachinery/pkg/util/wait"
clientset "k8s.io/client-go/kubernetes"
"k8s.io/kubernetes/pkg/api/testapi"
api "k8s.io/kubernetes/pkg/apis/core"
"k8s.io/kubernetes/pkg/client/clientset_generated/internalclientset"
testutils "k8s.io/kubernetes/test/utils"
)
// RcByNamePort returns a ReplicationController with specified name and port
func RcByNamePort(name string, replicas int32, image string, port int, protocol v1.Protocol,
labels map[string]string, gracePeriod *int64) *v1.ReplicationController {
return RcByNameContainer(name, replicas, image, labels, v1.Container{
Name: name,
Image: image,
Ports: []v1.ContainerPort{{ContainerPort: int32(port), Protocol: protocol}},
}, gracePeriod)
}
// RcByNameContainer returns a ReplicationControoler with specified name and container
func RcByNameContainer(name string, replicas int32, image string, labels map[string]string, c v1.Container,
gracePeriod *int64) *v1.ReplicationController {
zeroGracePeriod := int64(0)
// Add "name": name to the labels, overwriting if it exists.
labels["name"] = name
if gracePeriod == nil {
gracePeriod = &zeroGracePeriod
}
return &v1.ReplicationController{
TypeMeta: metav1.TypeMeta{
Kind: "ReplicationController",
APIVersion: testapi.Groups[v1.GroupName].GroupVersion().String(),
},
ObjectMeta: metav1.ObjectMeta{
Name: name,
},
Spec: v1.ReplicationControllerSpec{
Replicas: func(i int32) *int32 { return &i }(replicas),
Selector: map[string]string{
"name": name,
},
Template: &v1.PodTemplateSpec{
ObjectMeta: metav1.ObjectMeta{
Labels: labels,
},
Spec: v1.PodSpec{
Containers: []v1.Container{c},
TerminationGracePeriodSeconds: gracePeriod,
},
},
},
}
}
// ScaleRCByLabels scales an RC via ns/label lookup. If replicas == 0 it waits till
// none are running, otherwise it does what a synchronous scale operation would do.
func ScaleRCByLabels(clientset clientset.Interface, internalClientset internalclientset.Interface, ns string, l map[string]string, replicas uint) error {
listOpts := metav1.ListOptions{LabelSelector: labels.SelectorFromSet(labels.Set(l)).String()}
rcs, err := clientset.CoreV1().ReplicationControllers(ns).List(listOpts)
if err != nil {
return err
}
if len(rcs.Items) == 0 {
return fmt.Errorf("RC with labels %v not found in ns %v", l, ns)
}
Logf("Scaling %v RCs with labels %v in ns %v to %v replicas.", len(rcs.Items), l, ns, replicas)
for _, labelRC := range rcs.Items {
name := labelRC.Name
if err := ScaleRC(clientset, internalClientset, ns, name, replicas, false); err != nil {
return err
}
rc, err := clientset.CoreV1().ReplicationControllers(ns).Get(name, metav1.GetOptions{})
if err != nil {
return err
}
if replicas == 0 {
ps, err := podStoreForSelector(clientset, rc.Namespace, labels.SelectorFromSet(rc.Spec.Selector))
if err != nil {
return err
}
defer ps.Stop()
if err = waitForPodsGone(ps, 10*time.Second, 10*time.Minute); err != nil {
return fmt.Errorf("error while waiting for pods gone %s: %v", name, err)
}
} else {
if err := testutils.WaitForPodsWithLabelRunning(
clientset, ns, labels.SelectorFromSet(labels.Set(rc.Spec.Selector))); err != nil {
return err
}
}
}
return nil
}
type updateRcFunc func(d *v1.ReplicationController)
func UpdateReplicationControllerWithRetries(c clientset.Interface, namespace, name string, applyUpdate updateRcFunc) (*v1.ReplicationController, error) {
var rc *v1.ReplicationController
var updateErr error
pollErr := wait.PollImmediate(10*time.Millisecond, 1*time.Minute, func() (bool, error) {
var err error
if rc, err = c.CoreV1().ReplicationControllers(namespace).Get(name, metav1.GetOptions{}); err != nil {
return false, err
}
// Apply the update, then attempt to push it to the apiserver.
applyUpdate(rc)
if rc, err = c.CoreV1().ReplicationControllers(namespace).Update(rc); err == nil {
Logf("Updating replication controller %q", name)
return true, nil
}
updateErr = err
return false, nil
})
if pollErr == wait.ErrWaitTimeout {
pollErr = fmt.Errorf("couldn't apply the provided updated to rc %q: %v", name, updateErr)
}
return rc, pollErr
}
// DeleteRCAndWaitForGC deletes only the Replication Controller and waits for GC to delete the pods.
func DeleteRCAndWaitForGC(c clientset.Interface, ns, name string) error {
return DeleteResourceAndWaitForGC(c, api.Kind("ReplicationController"), ns, name)
}
func DeleteRCAndPods(clientset clientset.Interface, internalClientset internalclientset.Interface, ns, name string) error {
return DeleteResourceAndPods(clientset, internalClientset, api.Kind("ReplicationController"), ns, name)
}
func ScaleRC(clientset clientset.Interface, internalClientset internalclientset.Interface, ns, name string, size uint, wait bool) error {
return ScaleResource(clientset, internalClientset, ns, name, size, wait, api.Kind("ReplicationController"))
}
func RunRC(config testutils.RCConfig) error {
By(fmt.Sprintf("creating replication controller %s in namespace %s", config.Name, config.Namespace))
config.NodeDumpFunc = DumpNodeDebugInfo
config.ContainerDumpFunc = LogFailedContainers
return testutils.RunRC(config)
}
// WaitForRCPodToDisappear returns nil if the pod from the given replication controller (described by rcName) no longer exists.
// In case of failure or too long waiting time, an error is returned.
func WaitForRCPodToDisappear(c clientset.Interface, ns, rcName, podName string) error {
label := labels.SelectorFromSet(labels.Set(map[string]string{"name": rcName}))
// NodeController evicts pod after 5 minutes, so we need timeout greater than that to observe effects.
// The grace period must be set to 0 on the pod for it to be deleted during the partition.
// Otherwise, it goes to the 'Terminating' state till the kubelet confirms deletion.
return WaitForPodToDisappear(c, ns, podName, label, 20*time.Second, 10*time.Minute)
}
// WaitForReplicationController waits until the RC appears (exist == true), or disappears (exist == false)
func WaitForReplicationController(c clientset.Interface, namespace, name string, exist bool, interval, timeout time.Duration) error {
err := wait.PollImmediate(interval, timeout, func() (bool, error) {
_, err := c.CoreV1().ReplicationControllers(namespace).Get(name, metav1.GetOptions{})
if err != nil {
Logf("Get ReplicationController %s in namespace %s failed (%v).", name, namespace, err)
return !exist, nil
} else {
Logf("ReplicationController %s in namespace %s found.", name, namespace)
return exist, nil
}
})
if err != nil {
stateMsg := map[bool]string{true: "to appear", false: "to disappear"}
return fmt.Errorf("error waiting for ReplicationController %s/%s %s: %v", namespace, name, stateMsg[exist], err)
}
return nil
}
// WaitForReplicationControllerwithSelector waits until any RC with given selector appears (exist == true), or disappears (exist == false)
func WaitForReplicationControllerwithSelector(c clientset.Interface, namespace string, selector labels.Selector, exist bool, interval,
timeout time.Duration) error {
err := wait.PollImmediate(interval, timeout, func() (bool, error) {
rcs, err := c.CoreV1().ReplicationControllers(namespace).List(metav1.ListOptions{LabelSelector: selector.String()})
switch {
case len(rcs.Items) != 0:
Logf("ReplicationController with %s in namespace %s found.", selector.String(), namespace)
return exist, nil
case len(rcs.Items) == 0:
Logf("ReplicationController with %s in namespace %s disappeared.", selector.String(), namespace)
return !exist, nil
default:
Logf("List ReplicationController with %s in namespace %s failed: %v", selector.String(), namespace, err)
return false, nil
}
})
if err != nil {
stateMsg := map[bool]string{true: "to appear", false: "to disappear"}
return fmt.Errorf("error waiting for ReplicationControllers with %s in namespace %s %s: %v", selector.String(), namespace, stateMsg[exist], err)
}
return nil
}
// validatorFn is the function which is individual tests will implement.
// we may want it to return more than just an error, at some point.
type validatorFn func(c clientset.Interface, podID string) error
// ValidateController is a generic mechanism for testing RC's that are running.
// It takes a container name, a test name, and a validator function which is plugged in by a specific test.
// "containername": this is grepped for.
// "containerImage" : this is the name of the image we expect to be launched. Not to confuse w/ images (kitten.jpg) which are validated.
// "testname": which gets bubbled up to the logging/failure messages if errors happen.
// "validator" function: This function is given a podID and a client, and it can do some specific validations that way.
func ValidateController(c clientset.Interface, containerImage string, replicas int, containername string, testname string, validator validatorFn, ns string) {
getPodsTemplate := "--template={{range.items}}{{.metadata.name}} {{end}}"
// NB: kubectl adds the "exists" function to the standard template functions.
// This lets us check to see if the "running" entry exists for each of the containers
// we care about. Exists will never return an error and it's safe to check a chain of
// things, any one of which may not exist. In the below template, all of info,
// containername, and running might be nil, so the normal index function isn't very
// helpful.
// This template is unit-tested in kubectl, so if you change it, update the unit test.
// You can read about the syntax here: http://golang.org/pkg/text/template/.
getContainerStateTemplate := fmt.Sprintf(`--template={{if (exists . "status" "containerStatuses")}}{{range .status.containerStatuses}}{{if (and (eq .name "%s") (exists . "state" "running"))}}true{{end}}{{end}}{{end}}`, containername)
getImageTemplate := fmt.Sprintf(`--template={{if (exists . "status" "containerStatuses")}}{{range .status.containerStatuses}}{{if eq .name "%s"}}{{.image}}{{end}}{{end}}{{end}}`, containername)
By(fmt.Sprintf("waiting for all containers in %s pods to come up.", testname)) //testname should be selector
waitLoop:
for start := time.Now(); time.Since(start) < PodStartTimeout; time.Sleep(5 * time.Second) {
getPodsOutput := RunKubectlOrDie("get", "pods", "-o", "template", getPodsTemplate, "-l", testname, fmt.Sprintf("--namespace=%v", ns))
pods := strings.Fields(getPodsOutput)
if numPods := len(pods); numPods != replicas {
By(fmt.Sprintf("Replicas for %s: expected=%d actual=%d", testname, replicas, numPods))
continue
}
var runningPods []string
for _, podID := range pods {
running := RunKubectlOrDie("get", "pods", podID, "-o", "template", getContainerStateTemplate, fmt.Sprintf("--namespace=%v", ns))
if running != "true" {
Logf("%s is created but not running", podID)
continue waitLoop
}
currentImage := RunKubectlOrDie("get", "pods", podID, "-o", "template", getImageTemplate, fmt.Sprintf("--namespace=%v", ns))
if currentImage != containerImage {
Logf("%s is created but running wrong image; expected: %s, actual: %s", podID, containerImage, currentImage)
continue waitLoop
}
// Call the generic validator function here.
// This might validate for example, that (1) getting a url works and (2) url is serving correct content.
if err := validator(c, podID); err != nil {
Logf("%s is running right image but validator function failed: %v", podID, err)
continue waitLoop
}
Logf("%s is verified up and running", podID)
runningPods = append(runningPods, podID)
}
// If we reach here, then all our checks passed.
if len(runningPods) == replicas {
return
}
}
// Reaching here means that one of more checks failed multiple times. Assuming its not a race condition, something is broken.
Failf("Timed out after %v seconds waiting for %s pods to reach valid state", PodStartTimeout.Seconds(), testname)
}

View File

@ -0,0 +1,379 @@
/*
Copyright 2015 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 framework
import (
"bytes"
"fmt"
"math"
"sort"
"strconv"
"strings"
"sync"
"text/tabwriter"
"time"
"k8s.io/api/core/v1"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
utilruntime "k8s.io/apimachinery/pkg/util/runtime"
clientset "k8s.io/client-go/kubernetes"
"k8s.io/kubernetes/pkg/util/system"
)
type ResourceConstraint struct {
CPUConstraint float64
MemoryConstraint uint64
}
type SingleContainerSummary struct {
Name string
Cpu float64
Mem uint64
}
// we can't have int here, as JSON does not accept integer keys.
type ResourceUsageSummary map[string][]SingleContainerSummary
func (s *ResourceUsageSummary) PrintHumanReadable() string {
buf := &bytes.Buffer{}
w := tabwriter.NewWriter(buf, 1, 0, 1, ' ', 0)
for perc, summaries := range *s {
buf.WriteString(fmt.Sprintf("%v percentile:\n", perc))
fmt.Fprintf(w, "container\tcpu(cores)\tmemory(MB)\n")
for _, summary := range summaries {
fmt.Fprintf(w, "%q\t%.3f\t%.2f\n", summary.Name, summary.Cpu, float64(summary.Mem)/(1024*1024))
}
w.Flush()
}
return buf.String()
}
func (s *ResourceUsageSummary) PrintJSON() string {
return PrettyPrintJSON(*s)
}
func (s *ResourceUsageSummary) SummaryKind() string {
return "ResourceUsageSummary"
}
func computePercentiles(timeSeries []ResourceUsagePerContainer, percentilesToCompute []int) map[int]ResourceUsagePerContainer {
if len(timeSeries) == 0 {
return make(map[int]ResourceUsagePerContainer)
}
dataMap := make(map[string]*usageDataPerContainer)
for i := range timeSeries {
for name, data := range timeSeries[i] {
if dataMap[name] == nil {
dataMap[name] = &usageDataPerContainer{
cpuData: make([]float64, 0, len(timeSeries)),
memUseData: make([]uint64, 0, len(timeSeries)),
memWorkSetData: make([]uint64, 0, len(timeSeries)),
}
}
dataMap[name].cpuData = append(dataMap[name].cpuData, data.CPUUsageInCores)
dataMap[name].memUseData = append(dataMap[name].memUseData, data.MemoryUsageInBytes)
dataMap[name].memWorkSetData = append(dataMap[name].memWorkSetData, data.MemoryWorkingSetInBytes)
}
}
for _, v := range dataMap {
sort.Float64s(v.cpuData)
sort.Sort(uint64arr(v.memUseData))
sort.Sort(uint64arr(v.memWorkSetData))
}
result := make(map[int]ResourceUsagePerContainer)
for _, perc := range percentilesToCompute {
data := make(ResourceUsagePerContainer)
for k, v := range dataMap {
percentileIndex := int(math.Ceil(float64(len(v.cpuData)*perc)/100)) - 1
data[k] = &ContainerResourceUsage{
Name: k,
CPUUsageInCores: v.cpuData[percentileIndex],
MemoryUsageInBytes: v.memUseData[percentileIndex],
MemoryWorkingSetInBytes: v.memWorkSetData[percentileIndex],
}
}
result[perc] = data
}
return result
}
func leftMergeData(left, right map[int]ResourceUsagePerContainer) map[int]ResourceUsagePerContainer {
result := make(map[int]ResourceUsagePerContainer)
for percentile, data := range left {
result[percentile] = data
if _, ok := right[percentile]; !ok {
continue
}
for k, v := range right[percentile] {
result[percentile][k] = v
}
}
return result
}
type resourceGatherWorker struct {
c clientset.Interface
nodeName string
wg *sync.WaitGroup
containerIDs []string
stopCh chan struct{}
dataSeries []ResourceUsagePerContainer
finished bool
inKubemark bool
resourceDataGatheringPeriod time.Duration
probeDuration time.Duration
printVerboseLogs bool
}
func (w *resourceGatherWorker) singleProbe() {
data := make(ResourceUsagePerContainer)
if w.inKubemark {
kubemarkData := GetKubemarkMasterComponentsResourceUsage()
if data == nil {
return
}
for k, v := range kubemarkData {
data[k] = &ContainerResourceUsage{
Name: v.Name,
MemoryWorkingSetInBytes: v.MemoryWorkingSetInBytes,
CPUUsageInCores: v.CPUUsageInCores,
}
}
} else {
nodeUsage, err := getOneTimeResourceUsageOnNode(w.c, w.nodeName, w.probeDuration, func() []string { return w.containerIDs })
if err != nil {
Logf("Error while reading data from %v: %v", w.nodeName, err)
return
}
for k, v := range nodeUsage {
data[k] = v
if w.printVerboseLogs {
Logf("Get container %v usage on node %v. CPUUsageInCores: %v, MemoryUsageInBytes: %v, MemoryWorkingSetInBytes: %v", k, w.nodeName, v.CPUUsageInCores, v.MemoryUsageInBytes, v.MemoryWorkingSetInBytes)
}
}
}
w.dataSeries = append(w.dataSeries, data)
}
func (w *resourceGatherWorker) gather(initialSleep time.Duration) {
defer utilruntime.HandleCrash()
defer w.wg.Done()
defer Logf("Closing worker for %v", w.nodeName)
defer func() { w.finished = true }()
select {
case <-time.After(initialSleep):
w.singleProbe()
for {
select {
case <-time.After(w.resourceDataGatheringPeriod):
w.singleProbe()
case <-w.stopCh:
return
}
}
case <-w.stopCh:
return
}
}
type containerResourceGatherer struct {
client clientset.Interface
stopCh chan struct{}
workers []resourceGatherWorker
workerWg sync.WaitGroup
containerIDs []string
options ResourceGathererOptions
}
type ResourceGathererOptions struct {
InKubemark bool
MasterOnly bool
ResourceDataGatheringPeriod time.Duration
ProbeDuration time.Duration
PrintVerboseLogs bool
}
func NewResourceUsageGatherer(c clientset.Interface, options ResourceGathererOptions, pods *v1.PodList) (*containerResourceGatherer, error) {
g := containerResourceGatherer{
client: c,
stopCh: make(chan struct{}),
containerIDs: make([]string, 0),
options: options,
}
if options.InKubemark {
g.workerWg.Add(1)
g.workers = append(g.workers, resourceGatherWorker{
inKubemark: true,
stopCh: g.stopCh,
wg: &g.workerWg,
finished: false,
resourceDataGatheringPeriod: options.ResourceDataGatheringPeriod,
probeDuration: options.ProbeDuration,
printVerboseLogs: options.PrintVerboseLogs,
})
} else {
// Tracks kube-system pods if no valid PodList is passed in.
var err error
if pods == nil {
pods, err = c.CoreV1().Pods("kube-system").List(metav1.ListOptions{})
if err != nil {
Logf("Error while listing Pods: %v", err)
return nil, err
}
}
for _, pod := range pods.Items {
for _, container := range pod.Status.InitContainerStatuses {
g.containerIDs = append(g.containerIDs, container.Name)
}
for _, container := range pod.Status.ContainerStatuses {
g.containerIDs = append(g.containerIDs, container.Name)
}
}
nodeList, err := c.CoreV1().Nodes().List(metav1.ListOptions{})
if err != nil {
Logf("Error while listing Nodes: %v", err)
return nil, err
}
for _, node := range nodeList.Items {
if !options.MasterOnly || system.IsMasterNode(node.Name) {
g.workerWg.Add(1)
g.workers = append(g.workers, resourceGatherWorker{
c: c,
nodeName: node.Name,
wg: &g.workerWg,
containerIDs: g.containerIDs,
stopCh: g.stopCh,
finished: false,
inKubemark: false,
resourceDataGatheringPeriod: options.ResourceDataGatheringPeriod,
probeDuration: options.ProbeDuration,
printVerboseLogs: options.PrintVerboseLogs,
})
if options.MasterOnly {
break
}
}
}
}
return &g, nil
}
// StartGatheringData starts a stat gathering worker blocks for each node to track,
// and blocks until StopAndSummarize is called.
func (g *containerResourceGatherer) StartGatheringData() {
if len(g.workers) == 0 {
return
}
delayPeriod := g.options.ResourceDataGatheringPeriod / time.Duration(len(g.workers))
delay := time.Duration(0)
for i := range g.workers {
go g.workers[i].gather(delay)
delay += delayPeriod
}
g.workerWg.Wait()
}
// StopAndSummarize stops stat gathering workers, processes the collected stats,
// generates resource summary for the passed-in percentiles, and returns the summary.
// It returns an error if the resource usage at any percentile is beyond the
// specified resource constraints.
func (g *containerResourceGatherer) StopAndSummarize(percentiles []int, constraints map[string]ResourceConstraint) (*ResourceUsageSummary, error) {
close(g.stopCh)
Logf("Closed stop channel. Waiting for %v workers", len(g.workers))
finished := make(chan struct{})
go func() {
g.workerWg.Wait()
finished <- struct{}{}
}()
select {
case <-finished:
Logf("Waitgroup finished.")
case <-time.After(2 * time.Minute):
unfinished := make([]string, 0)
for i := range g.workers {
if !g.workers[i].finished {
unfinished = append(unfinished, g.workers[i].nodeName)
}
}
Logf("Timed out while waiting for waitgroup, some workers failed to finish: %v", unfinished)
}
if len(percentiles) == 0 {
Logf("Warning! Empty percentile list for stopAndPrintData.")
return &ResourceUsageSummary{}, fmt.Errorf("Failed to get any resource usage data")
}
data := make(map[int]ResourceUsagePerContainer)
for i := range g.workers {
if g.workers[i].finished {
stats := computePercentiles(g.workers[i].dataSeries, percentiles)
data = leftMergeData(stats, data)
}
}
// Workers has been stopped. We need to gather data stored in them.
sortedKeys := []string{}
for name := range data[percentiles[0]] {
sortedKeys = append(sortedKeys, name)
}
sort.Strings(sortedKeys)
violatedConstraints := make([]string, 0)
summary := make(ResourceUsageSummary)
for _, perc := range percentiles {
for _, name := range sortedKeys {
usage := data[perc][name]
summary[strconv.Itoa(perc)] = append(summary[strconv.Itoa(perc)], SingleContainerSummary{
Name: name,
Cpu: usage.CPUUsageInCores,
Mem: usage.MemoryWorkingSetInBytes,
})
// Verifying 99th percentile of resource usage
if perc == 99 {
// Name has a form: <pod_name>/<container_name>
containerName := strings.Split(name, "/")[1]
if constraint, ok := constraints[containerName]; ok {
if usage.CPUUsageInCores > constraint.CPUConstraint {
violatedConstraints = append(
violatedConstraints,
fmt.Sprintf("Container %v is using %v/%v CPU",
name,
usage.CPUUsageInCores,
constraint.CPUConstraint,
),
)
}
if usage.MemoryWorkingSetInBytes > constraint.MemoryConstraint {
violatedConstraints = append(
violatedConstraints,
fmt.Sprintf("Container %v is using %v/%v MB of memory",
name,
float64(usage.MemoryWorkingSetInBytes)/(1024*1024),
float64(constraint.MemoryConstraint)/(1024*1024),
),
)
}
}
}
}
}
if len(violatedConstraints) > 0 {
return &summary, fmt.Errorf(strings.Join(violatedConstraints, "\n"))
}
return &summary, nil
}

110
vendor/k8s.io/kubernetes/test/e2e/framework/rs_util.go generated vendored Normal file
View File

@ -0,0 +1,110 @@
/*
Copyright 2017 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 framework
import (
"fmt"
. "github.com/onsi/ginkgo"
"k8s.io/api/core/v1"
extensions "k8s.io/api/extensions/v1beta1"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/util/wait"
clientset "k8s.io/client-go/kubernetes"
deploymentutil "k8s.io/kubernetes/pkg/controller/deployment/util"
testutils "k8s.io/kubernetes/test/utils"
)
type updateRsFunc func(d *extensions.ReplicaSet)
func UpdateReplicaSetWithRetries(c clientset.Interface, namespace, name string, applyUpdate testutils.UpdateReplicaSetFunc) (*extensions.ReplicaSet, error) {
return testutils.UpdateReplicaSetWithRetries(c, namespace, name, applyUpdate, Logf, Poll, pollShortTimeout)
}
// CheckNewRSAnnotations check if the new RS's annotation is as expected
func CheckNewRSAnnotations(c clientset.Interface, ns, deploymentName string, expectedAnnotations map[string]string) error {
deployment, err := c.ExtensionsV1beta1().Deployments(ns).Get(deploymentName, metav1.GetOptions{})
if err != nil {
return err
}
newRS, err := deploymentutil.GetNewReplicaSet(deployment, c.ExtensionsV1beta1())
if err != nil {
return err
}
for k, v := range expectedAnnotations {
// Skip checking revision annotations
if k != deploymentutil.RevisionAnnotation && v != newRS.Annotations[k] {
return fmt.Errorf("Expected new RS annotations = %+v, got %+v", expectedAnnotations, newRS.Annotations)
}
}
return nil
}
// WaitForReadyReplicaSet waits until the replicaset has all of its replicas ready.
func WaitForReadyReplicaSet(c clientset.Interface, ns, name string) error {
err := wait.Poll(Poll, pollShortTimeout, func() (bool, error) {
rs, err := c.ExtensionsV1beta1().ReplicaSets(ns).Get(name, metav1.GetOptions{})
if err != nil {
return false, err
}
return *(rs.Spec.Replicas) == rs.Status.Replicas && *(rs.Spec.Replicas) == rs.Status.ReadyReplicas, nil
})
if err == wait.ErrWaitTimeout {
err = fmt.Errorf("replicaset %q never became ready", name)
}
return err
}
func RunReplicaSet(config testutils.ReplicaSetConfig) error {
By(fmt.Sprintf("creating replicaset %s in namespace %s", config.Name, config.Namespace))
config.NodeDumpFunc = DumpNodeDebugInfo
config.ContainerDumpFunc = LogFailedContainers
return testutils.RunReplicaSet(config)
}
func NewReplicaSet(name, namespace string, replicas int32, podLabels map[string]string, imageName, image string) *extensions.ReplicaSet {
return &extensions.ReplicaSet{
TypeMeta: metav1.TypeMeta{
Kind: "ReplicaSet",
APIVersion: "extensions/v1beta1",
},
ObjectMeta: metav1.ObjectMeta{
Namespace: namespace,
Name: name,
},
Spec: extensions.ReplicaSetSpec{
Selector: &metav1.LabelSelector{
MatchLabels: podLabels,
},
Replicas: &replicas,
Template: v1.PodTemplateSpec{
ObjectMeta: metav1.ObjectMeta{
Labels: podLabels,
},
Spec: v1.PodSpec{
Containers: []v1.Container{
{
Name: imageName,
Image: image,
},
},
},
},
},
}
}

File diff suppressed because it is too large Load Diff

154
vendor/k8s.io/kubernetes/test/e2e/framework/size.go generated vendored Normal file
View File

@ -0,0 +1,154 @@
/*
Copyright 2015 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 framework
import (
"fmt"
"os/exec"
"regexp"
"strings"
"time"
"github.com/aws/aws-sdk-go/aws/session"
"github.com/aws/aws-sdk-go/service/autoscaling"
awscloud "k8s.io/kubernetes/pkg/cloudprovider/providers/aws"
)
const (
resizeNodeReadyTimeout = 2 * time.Minute
resizeNodeNotReadyTimeout = 2 * time.Minute
)
func ResizeGroup(group string, size int32) error {
if TestContext.ReportDir != "" {
CoreDump(TestContext.ReportDir)
defer CoreDump(TestContext.ReportDir)
}
if TestContext.Provider == "gce" || TestContext.Provider == "gke" {
// TODO: make this hit the compute API directly instead of shelling out to gcloud.
// TODO: make gce/gke implement InstanceGroups, so we can eliminate the per-provider logic
zone, err := getGCEZoneForGroup(group)
if err != nil {
return err
}
output, err := exec.Command("gcloud", "compute", "instance-groups", "managed", "resize",
group, fmt.Sprintf("--size=%v", size),
"--project="+TestContext.CloudConfig.ProjectID, "--zone="+zone).CombinedOutput()
if err != nil {
return fmt.Errorf("Failed to resize node instance group %s: %s", group, output)
}
return nil
} else if TestContext.Provider == "aws" {
client := autoscaling.New(session.New())
return awscloud.ResizeInstanceGroup(client, group, int(size))
} else if TestContext.Provider == "kubemark" {
return TestContext.CloudConfig.KubemarkController.SetNodeGroupSize(group, int(size))
} else {
return fmt.Errorf("Provider does not support InstanceGroups")
}
}
func GetGroupNodes(group string) ([]string, error) {
if TestContext.Provider == "gce" || TestContext.Provider == "gke" {
// TODO: make this hit the compute API directly instead of shelling out to gcloud.
// TODO: make gce/gke implement InstanceGroups, so we can eliminate the per-provider logic
zone, err := getGCEZoneForGroup(group)
if err != nil {
return nil, err
}
output, err := exec.Command("gcloud", "compute", "instance-groups", "managed",
"list-instances", group, "--project="+TestContext.CloudConfig.ProjectID,
"--zone="+zone).CombinedOutput()
if err != nil {
return nil, fmt.Errorf("Failed to get nodes in instance group %s: %s", group, output)
}
re := regexp.MustCompile(".*RUNNING")
lines := re.FindAllString(string(output), -1)
for i, line := range lines {
lines[i] = line[:strings.Index(line, " ")]
}
return lines, nil
} else if TestContext.Provider == "kubemark" {
return TestContext.CloudConfig.KubemarkController.GetNodeNamesForNodeGroup(group)
} else {
return nil, fmt.Errorf("provider does not support InstanceGroups")
}
}
func GroupSize(group string) (int, error) {
if TestContext.Provider == "gce" || TestContext.Provider == "gke" {
// TODO: make this hit the compute API directly instead of shelling out to gcloud.
// TODO: make gce/gke implement InstanceGroups, so we can eliminate the per-provider logic
zone, err := getGCEZoneForGroup(group)
if err != nil {
return -1, err
}
output, err := exec.Command("gcloud", "compute", "instance-groups", "managed",
"list-instances", group, "--project="+TestContext.CloudConfig.ProjectID,
"--zone="+zone).CombinedOutput()
if err != nil {
return -1, fmt.Errorf("Failed to get group size for group %s: %s", group, output)
}
re := regexp.MustCompile("RUNNING")
return len(re.FindAllString(string(output), -1)), nil
} else if TestContext.Provider == "aws" {
client := autoscaling.New(session.New())
instanceGroup, err := awscloud.DescribeInstanceGroup(client, group)
if err != nil {
return -1, fmt.Errorf("error describing instance group: %v", err)
}
if instanceGroup == nil {
return -1, fmt.Errorf("instance group not found: %s", group)
}
return instanceGroup.CurrentSize()
} else if TestContext.Provider == "kubemark" {
return TestContext.CloudConfig.KubemarkController.GetNodeGroupSize(group)
} else {
return -1, fmt.Errorf("provider does not support InstanceGroups")
}
}
func WaitForGroupSize(group string, size int32) error {
timeout := 30 * time.Minute
for start := time.Now(); time.Since(start) < timeout; time.Sleep(20 * time.Second) {
currentSize, err := GroupSize(group)
if err != nil {
Logf("Failed to get node instance group size: %v", err)
continue
}
if currentSize != int(size) {
Logf("Waiting for node instance group size %d, current size %d", size, currentSize)
continue
}
Logf("Node instance group has reached the desired size %d", size)
return nil
}
return fmt.Errorf("timeout waiting %v for node instance group size to be %d", timeout, size)
}
func getGCEZoneForGroup(group string) (string, error) {
zone := TestContext.CloudConfig.Zone
if TestContext.CloudConfig.MultiZone {
output, err := exec.Command("gcloud", "compute", "instance-groups", "managed", "list",
"--project="+TestContext.CloudConfig.ProjectID, "--format=value(zone)", "--filter=name="+group).CombinedOutput()
if err != nil {
return "", fmt.Errorf("Failed to get zone for node group %s: %s", group, output)
}
zone = strings.TrimSpace(string(output))
}
return zone, nil
}

View File

@ -0,0 +1,894 @@
/*
Copyright 2014 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 framework
import (
"fmt"
"path/filepath"
"reflect"
"regexp"
"sort"
"strconv"
"strings"
"time"
. "github.com/onsi/gomega"
apps "k8s.io/api/apps/v1beta1"
appsV1beta2 "k8s.io/api/apps/v1beta2"
"k8s.io/api/core/v1"
apierrs "k8s.io/apimachinery/pkg/api/errors"
"k8s.io/apimachinery/pkg/api/resource"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/labels"
"k8s.io/apimachinery/pkg/util/intstr"
"k8s.io/apimachinery/pkg/util/sets"
"k8s.io/apimachinery/pkg/util/wait"
clientset "k8s.io/client-go/kubernetes"
podutil "k8s.io/kubernetes/pkg/api/v1/pod"
"k8s.io/kubernetes/test/e2e/manifest"
imageutils "k8s.io/kubernetes/test/utils/image"
)
const (
// Poll interval for StatefulSet tests
StatefulSetPoll = 10 * time.Second
// Timeout interval for StatefulSet operations
StatefulSetTimeout = 10 * time.Minute
// Timeout for stateful pods to change state
StatefulPodTimeout = 5 * time.Minute
)
// CreateStatefulSetService creates a Headless Service with Name name and Selector set to match labels.
func CreateStatefulSetService(name string, labels map[string]string) *v1.Service {
headlessService := &v1.Service{
ObjectMeta: metav1.ObjectMeta{
Name: name,
},
Spec: v1.ServiceSpec{
Selector: labels,
},
}
headlessService.Spec.Ports = []v1.ServicePort{
{Port: 80, Name: "http", Protocol: "TCP"},
}
headlessService.Spec.ClusterIP = "None"
return headlessService
}
// StatefulSetTester is a struct that contains utility methods for testing StatefulSet related functionality. It uses a
// clientset.Interface to communicate with the API server.
type StatefulSetTester struct {
c clientset.Interface
}
// NewStatefulSetTester creates a StatefulSetTester that uses c to interact with the API server.
func NewStatefulSetTester(c clientset.Interface) *StatefulSetTester {
return &StatefulSetTester{c}
}
// GetStatefulSet gets the StatefulSet named name in namespace.
func (s *StatefulSetTester) GetStatefulSet(namespace, name string) *apps.StatefulSet {
ss, err := s.c.AppsV1beta1().StatefulSets(namespace).Get(name, metav1.GetOptions{})
if err != nil {
Failf("Failed to get StatefulSet %s/%s: %v", namespace, name, err)
}
return ss
}
// CreateStatefulSet creates a StatefulSet from the manifest at manifestPath in the Namespace ns using kubectl create.
func (s *StatefulSetTester) CreateStatefulSet(manifestPath, ns string) *apps.StatefulSet {
mkpath := func(file string) string {
return filepath.Join(manifestPath, file)
}
Logf("Parsing statefulset from %v", mkpath("statefulset.yaml"))
ss, err := manifest.StatefulSetFromManifest(mkpath("statefulset.yaml"), ns)
Expect(err).NotTo(HaveOccurred())
Logf("Parsing service from %v", mkpath("service.yaml"))
svc, err := manifest.SvcFromManifest(mkpath("service.yaml"))
Expect(err).NotTo(HaveOccurred())
Logf(fmt.Sprintf("creating " + ss.Name + " service"))
_, err = s.c.CoreV1().Services(ns).Create(svc)
Expect(err).NotTo(HaveOccurred())
Logf(fmt.Sprintf("creating statefulset %v/%v with %d replicas and selector %+v", ss.Namespace, ss.Name, *(ss.Spec.Replicas), ss.Spec.Selector))
_, err = s.c.AppsV1beta1().StatefulSets(ns).Create(ss)
Expect(err).NotTo(HaveOccurred())
s.WaitForRunningAndReady(*ss.Spec.Replicas, ss)
return ss
}
// CheckMount checks that the mount at mountPath is valid for all Pods in ss.
func (s *StatefulSetTester) CheckMount(ss *apps.StatefulSet, mountPath string) error {
for _, cmd := range []string{
// Print inode, size etc
fmt.Sprintf("ls -idlh %v", mountPath),
// Print subdirs
fmt.Sprintf("find %v", mountPath),
// Try writing
fmt.Sprintf("touch %v", filepath.Join(mountPath, fmt.Sprintf("%v", time.Now().UnixNano()))),
} {
if err := s.ExecInStatefulPods(ss, cmd); err != nil {
return fmt.Errorf("failed to execute %v, error: %v", cmd, err)
}
}
return nil
}
// ExecInStatefulPods executes cmd in all Pods in ss. If a error occurs it is returned and cmd is not execute in any subsequent Pods.
func (s *StatefulSetTester) ExecInStatefulPods(ss *apps.StatefulSet, cmd string) error {
podList := s.GetPodList(ss)
for _, statefulPod := range podList.Items {
stdout, err := RunHostCmdWithRetries(statefulPod.Namespace, statefulPod.Name, cmd, StatefulSetPoll, StatefulPodTimeout)
Logf("stdout of %v on %v: %v", cmd, statefulPod.Name, stdout)
if err != nil {
return err
}
}
return nil
}
// CheckHostname verifies that all Pods in ss have the correct Hostname. If the returned error is not nil than verification failed.
func (s *StatefulSetTester) CheckHostname(ss *apps.StatefulSet) error {
cmd := "printf $(hostname)"
podList := s.GetPodList(ss)
for _, statefulPod := range podList.Items {
hostname, err := RunHostCmdWithRetries(statefulPod.Namespace, statefulPod.Name, cmd, StatefulSetPoll, StatefulPodTimeout)
if err != nil {
return err
}
if hostname != statefulPod.Name {
return fmt.Errorf("unexpected hostname (%s) and stateful pod name (%s) not equal", hostname, statefulPod.Name)
}
}
return nil
}
// Saturate waits for all Pods in ss to become Running and Ready.
func (s *StatefulSetTester) Saturate(ss *apps.StatefulSet) {
var i int32
for i = 0; i < *(ss.Spec.Replicas); i++ {
Logf("Waiting for stateful pod at index %v to enter Running", i)
s.WaitForRunning(i+1, i, ss)
Logf("Resuming stateful pod at index %v", i)
s.ResumeNextPod(ss)
}
}
// DeleteStatefulPodAtIndex deletes the Pod with ordinal index in ss.
func (s *StatefulSetTester) DeleteStatefulPodAtIndex(index int, ss *apps.StatefulSet) {
name := getStatefulSetPodNameAtIndex(index, ss)
noGrace := int64(0)
if err := s.c.CoreV1().Pods(ss.Namespace).Delete(name, &metav1.DeleteOptions{GracePeriodSeconds: &noGrace}); err != nil {
Failf("Failed to delete stateful pod %v for StatefulSet %v/%v: %v", name, ss.Namespace, ss.Name, err)
}
}
// VerifyStatefulPodFunc is a func that examines a StatefulSetPod.
type VerifyStatefulPodFunc func(*v1.Pod)
// VerifyPodAtIndex applies a visitor patter to the Pod at index in ss. verify is is applied to the Pod to "visit" it.
func (s *StatefulSetTester) VerifyPodAtIndex(index int, ss *apps.StatefulSet, verify VerifyStatefulPodFunc) {
name := getStatefulSetPodNameAtIndex(index, ss)
pod, err := s.c.CoreV1().Pods(ss.Namespace).Get(name, metav1.GetOptions{})
Expect(err).NotTo(HaveOccurred(), fmt.Sprintf("Failed to get stateful pod %s for StatefulSet %s/%s", name, ss.Namespace, ss.Name))
verify(pod)
}
func getStatefulSetPodNameAtIndex(index int, ss *apps.StatefulSet) string {
// TODO: we won't use "-index" as the name strategy forever,
// pull the name out from an identity mapper.
return fmt.Sprintf("%v-%v", ss.Name, index)
}
// Scale scales ss to count replicas.
func (s *StatefulSetTester) Scale(ss *apps.StatefulSet, count int32) (*apps.StatefulSet, error) {
name := ss.Name
ns := ss.Namespace
Logf("Scaling statefulset %s to %d", name, count)
ss = s.update(ns, name, func(ss *apps.StatefulSet) { *(ss.Spec.Replicas) = count })
var statefulPodList *v1.PodList
pollErr := wait.PollImmediate(StatefulSetPoll, StatefulSetTimeout, func() (bool, error) {
statefulPodList = s.GetPodList(ss)
if int32(len(statefulPodList.Items)) == count {
return true, nil
}
return false, nil
})
if pollErr != nil {
unhealthy := []string{}
for _, statefulPod := range statefulPodList.Items {
delTs, phase, readiness := statefulPod.DeletionTimestamp, statefulPod.Status.Phase, podutil.IsPodReady(&statefulPod)
if delTs != nil || phase != v1.PodRunning || !readiness {
unhealthy = append(unhealthy, fmt.Sprintf("%v: deletion %v, phase %v, readiness %v", statefulPod.Name, delTs, phase, readiness))
}
}
return ss, fmt.Errorf("Failed to scale statefulset to %d in %v. Remaining pods:\n%v", count, StatefulSetTimeout, unhealthy)
}
return ss, nil
}
// UpdateReplicas updates the replicas of ss to count.
func (s *StatefulSetTester) UpdateReplicas(ss *apps.StatefulSet, count int32) {
s.update(ss.Namespace, ss.Name, func(ss *apps.StatefulSet) { *(ss.Spec.Replicas) = count })
}
// Restart scales ss to 0 and then back to its previous number of replicas.
func (s *StatefulSetTester) Restart(ss *apps.StatefulSet) {
oldReplicas := *(ss.Spec.Replicas)
ss, err := s.Scale(ss, 0)
ExpectNoError(err)
// Wait for controller to report the desired number of Pods.
// This way we know the controller has observed all Pod deletions
// before we scale it back up.
s.WaitForStatusReplicas(ss, 0)
s.update(ss.Namespace, ss.Name, func(ss *apps.StatefulSet) { *(ss.Spec.Replicas) = oldReplicas })
}
func (s *StatefulSetTester) update(ns, name string, update func(ss *apps.StatefulSet)) *apps.StatefulSet {
for i := 0; i < 3; i++ {
ss, err := s.c.AppsV1beta1().StatefulSets(ns).Get(name, metav1.GetOptions{})
if err != nil {
Failf("failed to get statefulset %q: %v", name, err)
}
update(ss)
ss, err = s.c.AppsV1beta1().StatefulSets(ns).Update(ss)
if err == nil {
return ss
}
if !apierrs.IsConflict(err) && !apierrs.IsServerTimeout(err) {
Failf("failed to update statefulset %q: %v", name, err)
}
}
Failf("too many retries draining statefulset %q", name)
return nil
}
// GetPodList gets the current Pods in ss.
func (s *StatefulSetTester) GetPodList(ss *apps.StatefulSet) *v1.PodList {
selector, err := metav1.LabelSelectorAsSelector(ss.Spec.Selector)
ExpectNoError(err)
podList, err := s.c.CoreV1().Pods(ss.Namespace).List(metav1.ListOptions{LabelSelector: selector.String()})
ExpectNoError(err)
return podList
}
// ConfirmStatefulPodCount asserts that the current number of Pods in ss is count waiting up to timeout for ss to
// to scale to count.
func (s *StatefulSetTester) ConfirmStatefulPodCount(count int, ss *apps.StatefulSet, timeout time.Duration, hard bool) {
start := time.Now()
deadline := start.Add(timeout)
for t := time.Now(); t.Before(deadline); t = time.Now() {
podList := s.GetPodList(ss)
statefulPodCount := len(podList.Items)
if statefulPodCount != count {
logPodStates(podList.Items)
if hard {
Failf("StatefulSet %v scaled unexpectedly scaled to %d -> %d replicas", ss.Name, count, len(podList.Items))
} else {
Logf("StatefulSet %v has not reached scale %d, at %d", ss.Name, count, statefulPodCount)
}
time.Sleep(1 * time.Second)
continue
}
Logf("Verifying statefulset %v doesn't scale past %d for another %+v", ss.Name, count, deadline.Sub(t))
time.Sleep(1 * time.Second)
}
}
// WaitForRunning waits for numPodsRunning in ss to be Running and for the first
// numPodsReady ordinals to be Ready.
func (s *StatefulSetTester) WaitForRunning(numPodsRunning, numPodsReady int32, ss *apps.StatefulSet) {
pollErr := wait.PollImmediate(StatefulSetPoll, StatefulSetTimeout,
func() (bool, error) {
podList := s.GetPodList(ss)
s.SortStatefulPods(podList)
if int32(len(podList.Items)) < numPodsRunning {
Logf("Found %d stateful pods, waiting for %d", len(podList.Items), numPodsRunning)
return false, nil
}
if int32(len(podList.Items)) > numPodsRunning {
return false, fmt.Errorf("Too many pods scheduled, expected %d got %d", numPodsRunning, len(podList.Items))
}
for _, p := range podList.Items {
shouldBeReady := getStatefulPodOrdinal(&p) < int(numPodsReady)
isReady := podutil.IsPodReady(&p)
desiredReadiness := shouldBeReady == isReady
Logf("Waiting for pod %v to enter %v - Ready=%v, currently %v - Ready=%v", p.Name, v1.PodRunning, shouldBeReady, p.Status.Phase, isReady)
if p.Status.Phase != v1.PodRunning || !desiredReadiness {
return false, nil
}
}
return true, nil
})
if pollErr != nil {
Failf("Failed waiting for pods to enter running: %v", pollErr)
}
}
// WaitForState periodically polls for the ss and its pods until the until function returns either true or an error
func (s *StatefulSetTester) WaitForState(ss *apps.StatefulSet, until func(*apps.StatefulSet, *v1.PodList) (bool, error)) {
pollErr := wait.PollImmediate(StatefulSetPoll, StatefulSetTimeout,
func() (bool, error) {
ssGet, err := s.c.AppsV1beta1().StatefulSets(ss.Namespace).Get(ss.Name, metav1.GetOptions{})
if err != nil {
return false, err
}
podList := s.GetPodList(ssGet)
return until(ssGet, podList)
})
if pollErr != nil {
Failf("Failed waiting for state update: %v", pollErr)
}
}
// WaitForStatus waits for the StatefulSetStatus's ObservedGeneration to be greater than or equal to set's Generation.
// The returned StatefulSet contains such a StatefulSetStatus
func (s *StatefulSetTester) WaitForStatus(set *apps.StatefulSet) *apps.StatefulSet {
s.WaitForState(set, func(set2 *apps.StatefulSet, pods *v1.PodList) (bool, error) {
if set2.Status.ObservedGeneration != nil && *set2.Status.ObservedGeneration >= set.Generation {
set = set2
return true, nil
}
return false, nil
})
return set
}
// WaitForRunningAndReady waits for numStatefulPods in ss to be Running and Ready.
func (s *StatefulSetTester) WaitForRunningAndReady(numStatefulPods int32, ss *apps.StatefulSet) {
s.WaitForRunning(numStatefulPods, numStatefulPods, ss)
}
// WaitForPodReady waits for the Pod named podName in set to exist and have a Ready condition.
func (s *StatefulSetTester) WaitForPodReady(set *apps.StatefulSet, podName string) (*apps.StatefulSet, *v1.PodList) {
var pods *v1.PodList
s.WaitForState(set, func(set2 *apps.StatefulSet, pods2 *v1.PodList) (bool, error) {
set = set2
pods = pods2
for i := range pods.Items {
if pods.Items[i].Name == podName {
return podutil.IsPodReady(&pods.Items[i]), nil
}
}
return false, nil
})
return set, pods
}
// WaitForPodNotReady waist for the Pod named podName in set to exist and to not have a Ready condition.
func (s *StatefulSetTester) WaitForPodNotReady(set *apps.StatefulSet, podName string) (*apps.StatefulSet, *v1.PodList) {
var pods *v1.PodList
s.WaitForState(set, func(set2 *apps.StatefulSet, pods2 *v1.PodList) (bool, error) {
set = set2
pods = pods2
for i := range pods.Items {
if pods.Items[i].Name == podName {
return !podutil.IsPodReady(&pods.Items[i]), nil
}
}
return false, nil
})
return set, pods
}
// WaitForRollingUpdate waits for all Pods in set to exist and have the correct revision and for the RollingUpdate to
// complete. set must have a RollingUpdateStatefulSetStrategyType.
func (s *StatefulSetTester) WaitForRollingUpdate(set *apps.StatefulSet) (*apps.StatefulSet, *v1.PodList) {
var pods *v1.PodList
if set.Spec.UpdateStrategy.Type != apps.RollingUpdateStatefulSetStrategyType {
Failf("StatefulSet %s/%s attempt to wait for rolling update with updateStrategy %s",
set.Namespace,
set.Name,
set.Spec.UpdateStrategy.Type)
}
s.WaitForState(set, func(set2 *apps.StatefulSet, pods2 *v1.PodList) (bool, error) {
set = set2
pods = pods2
if len(pods.Items) < int(*set.Spec.Replicas) {
return false, nil
}
if set.Status.UpdateRevision != set.Status.CurrentRevision {
Logf("Waiting for StatefulSet %s/%s to complete update",
set.Namespace,
set.Name,
)
s.SortStatefulPods(pods)
for i := range pods.Items {
if pods.Items[i].Labels[apps.StatefulSetRevisionLabel] != set.Status.UpdateRevision {
Logf("Waiting for Pod %s/%s to have revision %s update revision %s",
pods.Items[i].Namespace,
pods.Items[i].Name,
set.Status.UpdateRevision,
pods.Items[i].Labels[apps.StatefulSetRevisionLabel])
}
}
return false, nil
}
return true, nil
})
return set, pods
}
// WaitForPartitionedRollingUpdate waits for all Pods in set to exist and have the correct revision. set must have
// a RollingUpdateStatefulSetStrategyType with a non-nil RollingUpdate and Partition. All Pods with ordinals less
// than or equal to the Partition are expected to be at set's current revision. All other Pods are expected to be
// at its update revision.
func (s *StatefulSetTester) WaitForPartitionedRollingUpdate(set *apps.StatefulSet) (*apps.StatefulSet, *v1.PodList) {
var pods *v1.PodList
if set.Spec.UpdateStrategy.Type != apps.RollingUpdateStatefulSetStrategyType {
Failf("StatefulSet %s/%s attempt to wait for partitioned update with updateStrategy %s",
set.Namespace,
set.Name,
set.Spec.UpdateStrategy.Type)
}
if set.Spec.UpdateStrategy.RollingUpdate == nil || set.Spec.UpdateStrategy.RollingUpdate.Partition == nil {
Failf("StatefulSet %s/%s attempt to wait for partitioned update with nil RollingUpdate or nil Partition",
set.Namespace,
set.Name)
}
s.WaitForState(set, func(set2 *apps.StatefulSet, pods2 *v1.PodList) (bool, error) {
set = set2
pods = pods2
partition := int(*set.Spec.UpdateStrategy.RollingUpdate.Partition)
if len(pods.Items) < int(*set.Spec.Replicas) {
return false, nil
}
if partition <= 0 && set.Status.UpdateRevision != set.Status.CurrentRevision {
Logf("Waiting for StatefulSet %s/%s to complete update",
set.Namespace,
set.Name,
)
s.SortStatefulPods(pods)
for i := range pods.Items {
if pods.Items[i].Labels[apps.StatefulSetRevisionLabel] != set.Status.UpdateRevision {
Logf("Waiting for Pod %s/%s to have revision %s update revision %s",
pods.Items[i].Namespace,
pods.Items[i].Name,
set.Status.UpdateRevision,
pods.Items[i].Labels[apps.StatefulSetRevisionLabel])
}
}
return false, nil
} else {
for i := int(*set.Spec.Replicas) - 1; i >= partition; i-- {
if pods.Items[i].Labels[apps.StatefulSetRevisionLabel] != set.Status.UpdateRevision {
Logf("Waiting for Pod %s/%s to have revision %s update revision %s",
pods.Items[i].Namespace,
pods.Items[i].Name,
set.Status.UpdateRevision,
pods.Items[i].Labels[apps.StatefulSetRevisionLabel])
return false, nil
}
}
}
return true, nil
})
return set, pods
}
// WaitForRunningAndReady waits for numStatefulPods in ss to be Running and not Ready.
func (s *StatefulSetTester) WaitForRunningAndNotReady(numStatefulPods int32, ss *apps.StatefulSet) {
s.WaitForRunning(numStatefulPods, 0, ss)
}
var httpProbe = &v1.Probe{
Handler: v1.Handler{
HTTPGet: &v1.HTTPGetAction{
Path: "/index.html",
Port: intstr.IntOrString{IntVal: 80},
},
},
PeriodSeconds: 1,
SuccessThreshold: 1,
FailureThreshold: 1,
}
// SetHttpProbe sets the pod template's ReadinessProbe for Nginx StatefulSet containers.
// This probe can then be controlled with BreakHttpProbe() and RestoreHttpProbe().
// Note that this cannot be used together with PauseNewPods().
func (s *StatefulSetTester) SetHttpProbe(ss *apps.StatefulSet) {
ss.Spec.Template.Spec.Containers[0].ReadinessProbe = httpProbe
}
// BreakHttpProbe breaks the readiness probe for Nginx StatefulSet containers in ss.
func (s *StatefulSetTester) BreakHttpProbe(ss *apps.StatefulSet) error {
path := httpProbe.HTTPGet.Path
if path == "" {
return fmt.Errorf("Path expected to be not empty: %v", path)
}
// Ignore 'mv' errors to make this idempotent.
cmd := fmt.Sprintf("mv -v /usr/share/nginx/html%v /tmp/ || true", path)
return s.ExecInStatefulPods(ss, cmd)
}
// BreakPodHttpProbe breaks the readiness probe for Nginx StatefulSet containers in one pod.
func (s *StatefulSetTester) BreakPodHttpProbe(ss *apps.StatefulSet, pod *v1.Pod) error {
path := httpProbe.HTTPGet.Path
if path == "" {
return fmt.Errorf("Path expected to be not empty: %v", path)
}
// Ignore 'mv' errors to make this idempotent.
cmd := fmt.Sprintf("mv -v /usr/share/nginx/html%v /tmp/ || true", path)
stdout, err := RunHostCmdWithRetries(pod.Namespace, pod.Name, cmd, StatefulSetPoll, StatefulPodTimeout)
Logf("stdout of %v on %v: %v", cmd, pod.Name, stdout)
return err
}
// RestoreHttpProbe restores the readiness probe for Nginx StatefulSet containers in ss.
func (s *StatefulSetTester) RestoreHttpProbe(ss *apps.StatefulSet) error {
path := httpProbe.HTTPGet.Path
if path == "" {
return fmt.Errorf("Path expected to be not empty: %v", path)
}
// Ignore 'mv' errors to make this idempotent.
cmd := fmt.Sprintf("mv -v /tmp%v /usr/share/nginx/html/ || true", path)
return s.ExecInStatefulPods(ss, cmd)
}
// RestorePodHttpProbe restores the readiness probe for Nginx StatefulSet containers in pod.
func (s *StatefulSetTester) RestorePodHttpProbe(ss *apps.StatefulSet, pod *v1.Pod) error {
path := httpProbe.HTTPGet.Path
if path == "" {
return fmt.Errorf("Path expected to be not empty: %v", path)
}
// Ignore 'mv' errors to make this idempotent.
cmd := fmt.Sprintf("mv -v /tmp%v /usr/share/nginx/html/ || true", path)
stdout, err := RunHostCmdWithRetries(pod.Namespace, pod.Name, cmd, StatefulSetPoll, StatefulPodTimeout)
Logf("stdout of %v on %v: %v", cmd, pod.Name, stdout)
return err
}
var pauseProbe = &v1.Probe{
Handler: v1.Handler{
Exec: &v1.ExecAction{Command: []string{"test", "-f", "/data/statefulset-continue"}},
},
PeriodSeconds: 1,
SuccessThreshold: 1,
FailureThreshold: 1,
}
func hasPauseProbe(pod *v1.Pod) bool {
probe := pod.Spec.Containers[0].ReadinessProbe
return probe != nil && reflect.DeepEqual(probe.Exec.Command, pauseProbe.Exec.Command)
}
// PauseNewPods adds an always-failing ReadinessProbe to the StatefulSet PodTemplate.
// This causes all newly-created Pods to stay Unready until they are manually resumed
// with ResumeNextPod().
// Note that this cannot be used together with SetHttpProbe().
func (s *StatefulSetTester) PauseNewPods(ss *apps.StatefulSet) {
ss.Spec.Template.Spec.Containers[0].ReadinessProbe = pauseProbe
}
// ResumeNextPod allows the next Pod in the StatefulSet to continue by removing the ReadinessProbe
// added by PauseNewPods(), if it's still there.
// It fails the test if it finds any pods that are not in phase Running,
// or if it finds more than one paused Pod existing at the same time.
// This is a no-op if there are no paused pods.
func (s *StatefulSetTester) ResumeNextPod(ss *apps.StatefulSet) {
podList := s.GetPodList(ss)
resumedPod := ""
for _, pod := range podList.Items {
if pod.Status.Phase != v1.PodRunning {
Failf("Found pod in phase %q, cannot resume", pod.Status.Phase)
}
if podutil.IsPodReady(&pod) || !hasPauseProbe(&pod) {
continue
}
if resumedPod != "" {
Failf("Found multiple paused stateful pods: %v and %v", pod.Name, resumedPod)
}
_, err := RunHostCmdWithRetries(pod.Namespace, pod.Name, "touch /data/statefulset-continue; sync", StatefulSetPoll, StatefulPodTimeout)
ExpectNoError(err)
Logf("Resumed pod %v", pod.Name)
resumedPod = pod.Name
}
}
// WaitForStatusReadyReplicas waits for the ss.Status.ReadyReplicas to be equal to expectedReplicas
func (s *StatefulSetTester) WaitForStatusReadyReplicas(ss *apps.StatefulSet, expectedReplicas int32) {
Logf("Waiting for statefulset status.replicas updated to %d", expectedReplicas)
ns, name := ss.Namespace, ss.Name
pollErr := wait.PollImmediate(StatefulSetPoll, StatefulSetTimeout,
func() (bool, error) {
ssGet, err := s.c.AppsV1beta1().StatefulSets(ns).Get(name, metav1.GetOptions{})
if err != nil {
return false, err
}
if *ssGet.Status.ObservedGeneration < ss.Generation {
return false, nil
}
if ssGet.Status.ReadyReplicas != expectedReplicas {
Logf("Waiting for stateful set status.readyReplicas to become %d, currently %d", expectedReplicas, ssGet.Status.ReadyReplicas)
return false, nil
}
return true, nil
})
if pollErr != nil {
Failf("Failed waiting for stateful set status.readyReplicas updated to %d: %v", expectedReplicas, pollErr)
}
}
// WaitForStatusReplicas waits for the ss.Status.Replicas to be equal to expectedReplicas
func (s *StatefulSetTester) WaitForStatusReplicas(ss *apps.StatefulSet, expectedReplicas int32) {
Logf("Waiting for statefulset status.replicas updated to %d", expectedReplicas)
ns, name := ss.Namespace, ss.Name
pollErr := wait.PollImmediate(StatefulSetPoll, StatefulSetTimeout,
func() (bool, error) {
ssGet, err := s.c.AppsV1beta1().StatefulSets(ns).Get(name, metav1.GetOptions{})
if err != nil {
return false, err
}
if *ssGet.Status.ObservedGeneration < ss.Generation {
return false, nil
}
if ssGet.Status.Replicas != expectedReplicas {
Logf("Waiting for stateful set status.replicas to become %d, currently %d", expectedReplicas, ssGet.Status.Replicas)
return false, nil
}
return true, nil
})
if pollErr != nil {
Failf("Failed waiting for stateful set status.replicas updated to %d: %v", expectedReplicas, pollErr)
}
}
// CheckServiceName asserts that the ServiceName for ss is equivalent to expectedServiceName.
func (p *StatefulSetTester) CheckServiceName(ss *apps.StatefulSet, expectedServiceName string) error {
Logf("Checking if statefulset spec.serviceName is %s", expectedServiceName)
if expectedServiceName != ss.Spec.ServiceName {
return fmt.Errorf("Wrong service name governing statefulset. Expected %s got %s",
expectedServiceName, ss.Spec.ServiceName)
}
return nil
}
// SortStatefulPods sorts pods by their ordinals
func (s *StatefulSetTester) SortStatefulPods(pods *v1.PodList) {
sort.Sort(statefulPodsByOrdinal(pods.Items))
}
// DeleteAllStatefulSets deletes all StatefulSet API Objects in Namespace ns.
func DeleteAllStatefulSets(c clientset.Interface, ns string) {
sst := &StatefulSetTester{c: c}
ssList, err := c.AppsV1beta1().StatefulSets(ns).List(metav1.ListOptions{LabelSelector: labels.Everything().String()})
ExpectNoError(err)
// Scale down each statefulset, then delete it completely.
// Deleting a pvc without doing this will leak volumes, #25101.
errList := []string{}
for i := range ssList.Items {
ss := &ssList.Items[i]
var err error
if ss, err = sst.Scale(ss, 0); err != nil {
errList = append(errList, fmt.Sprintf("%v", err))
}
sst.WaitForStatusReplicas(ss, 0)
Logf("Deleting statefulset %v", ss.Name)
// Use OrphanDependents=false so it's deleted synchronously.
// We already made sure the Pods are gone inside Scale().
if err := c.AppsV1beta1().StatefulSets(ss.Namespace).Delete(ss.Name, &metav1.DeleteOptions{OrphanDependents: new(bool)}); err != nil {
errList = append(errList, fmt.Sprintf("%v", err))
}
}
// pvs are global, so we need to wait for the exact ones bound to the statefulset pvcs.
pvNames := sets.NewString()
// TODO: Don't assume all pvcs in the ns belong to a statefulset
pvcPollErr := wait.PollImmediate(StatefulSetPoll, StatefulSetTimeout, func() (bool, error) {
pvcList, err := c.CoreV1().PersistentVolumeClaims(ns).List(metav1.ListOptions{LabelSelector: labels.Everything().String()})
if err != nil {
Logf("WARNING: Failed to list pvcs, retrying %v", err)
return false, nil
}
for _, pvc := range pvcList.Items {
pvNames.Insert(pvc.Spec.VolumeName)
// TODO: Double check that there are no pods referencing the pvc
Logf("Deleting pvc: %v with volume %v", pvc.Name, pvc.Spec.VolumeName)
if err := c.CoreV1().PersistentVolumeClaims(ns).Delete(pvc.Name, nil); err != nil {
return false, nil
}
}
return true, nil
})
if pvcPollErr != nil {
errList = append(errList, fmt.Sprintf("Timeout waiting for pvc deletion."))
}
pollErr := wait.PollImmediate(StatefulSetPoll, StatefulSetTimeout, func() (bool, error) {
pvList, err := c.CoreV1().PersistentVolumes().List(metav1.ListOptions{LabelSelector: labels.Everything().String()})
if err != nil {
Logf("WARNING: Failed to list pvs, retrying %v", err)
return false, nil
}
waitingFor := []string{}
for _, pv := range pvList.Items {
if pvNames.Has(pv.Name) {
waitingFor = append(waitingFor, fmt.Sprintf("%v: %+v", pv.Name, pv.Status))
}
}
if len(waitingFor) == 0 {
return true, nil
}
Logf("Still waiting for pvs of statefulset to disappear:\n%v", strings.Join(waitingFor, "\n"))
return false, nil
})
if pollErr != nil {
errList = append(errList, fmt.Sprintf("Timeout waiting for pv provisioner to delete pvs, this might mean the test leaked pvs."))
}
if len(errList) != 0 {
ExpectNoError(fmt.Errorf("%v", strings.Join(errList, "\n")))
}
}
// NewStatefulSetPVC returns a PersistentVolumeClaim named name, for testing StatefulSets.
func NewStatefulSetPVC(name string) v1.PersistentVolumeClaim {
return v1.PersistentVolumeClaim{
ObjectMeta: metav1.ObjectMeta{
Name: name,
},
Spec: v1.PersistentVolumeClaimSpec{
AccessModes: []v1.PersistentVolumeAccessMode{
v1.ReadWriteOnce,
},
Resources: v1.ResourceRequirements{
Requests: v1.ResourceList{
v1.ResourceStorage: *resource.NewQuantity(1, resource.BinarySI),
},
},
},
}
}
// NewStatefulSet creates a new NGINX StatefulSet for testing. The StatefulSet is named name, is in namespace ns,
// statefulPodsMounts are the mounts that will be backed by PVs. podsMounts are the mounts that are mounted directly
// to the Pod. labels are the labels that will be usd for the StatefulSet selector.
func NewStatefulSet(name, ns, governingSvcName string, replicas int32, statefulPodMounts []v1.VolumeMount, podMounts []v1.VolumeMount, labels map[string]string) *apps.StatefulSet {
mounts := append(statefulPodMounts, podMounts...)
claims := []v1.PersistentVolumeClaim{}
for _, m := range statefulPodMounts {
claims = append(claims, NewStatefulSetPVC(m.Name))
}
vols := []v1.Volume{}
for _, m := range podMounts {
vols = append(vols, v1.Volume{
Name: m.Name,
VolumeSource: v1.VolumeSource{
HostPath: &v1.HostPathVolumeSource{
Path: fmt.Sprintf("/tmp/%v", m.Name),
},
},
})
}
return &apps.StatefulSet{
TypeMeta: metav1.TypeMeta{
Kind: "StatefulSet",
APIVersion: "apps/v1beta1",
},
ObjectMeta: metav1.ObjectMeta{
Name: name,
Namespace: ns,
},
Spec: apps.StatefulSetSpec{
Selector: &metav1.LabelSelector{
MatchLabels: labels,
},
Replicas: func(i int32) *int32 { return &i }(replicas),
Template: v1.PodTemplateSpec{
ObjectMeta: metav1.ObjectMeta{
Labels: labels,
Annotations: map[string]string{},
},
Spec: v1.PodSpec{
Containers: []v1.Container{
{
Name: "nginx",
Image: imageutils.GetE2EImage(imageutils.NginxSlim),
VolumeMounts: mounts,
},
},
Volumes: vols,
},
},
UpdateStrategy: apps.StatefulSetUpdateStrategy{Type: apps.RollingUpdateStatefulSetStrategyType},
VolumeClaimTemplates: claims,
ServiceName: governingSvcName,
},
}
}
// NewStatefulSetScale creates a new StatefulSet scale subresource and returns it
func NewStatefulSetScale(ss *apps.StatefulSet) *appsV1beta2.Scale {
return &appsV1beta2.Scale{
// TODO: Create a variant of ObjectMeta type that only contains the fields below.
ObjectMeta: metav1.ObjectMeta{
Name: ss.Name,
Namespace: ss.Namespace,
},
Spec: appsV1beta2.ScaleSpec{
Replicas: *(ss.Spec.Replicas),
},
Status: appsV1beta2.ScaleStatus{
Replicas: ss.Status.Replicas,
},
}
}
var statefulPodRegex = regexp.MustCompile("(.*)-([0-9]+)$")
func getStatefulPodOrdinal(pod *v1.Pod) int {
ordinal := -1
subMatches := statefulPodRegex.FindStringSubmatch(pod.Name)
if len(subMatches) < 3 {
return ordinal
}
if i, err := strconv.ParseInt(subMatches[2], 10, 32); err == nil {
ordinal = int(i)
}
return ordinal
}
type statefulPodsByOrdinal []v1.Pod
func (sp statefulPodsByOrdinal) Len() int {
return len(sp)
}
func (sp statefulPodsByOrdinal) Swap(i, j int) {
sp[i], sp[j] = sp[j], sp[i]
}
func (sp statefulPodsByOrdinal) Less(i, j int) bool {
return getStatefulPodOrdinal(&sp[i]) < getStatefulPodOrdinal(&sp[j])
}
type updateStatefulSetFunc func(*apps.StatefulSet)
func UpdateStatefulSetWithRetries(c clientset.Interface, namespace, name string, applyUpdate updateStatefulSetFunc) (statefulSet *apps.StatefulSet, err error) {
statefulSets := c.AppsV1beta1().StatefulSets(namespace)
var updateErr error
pollErr := wait.Poll(10*time.Millisecond, 1*time.Minute, func() (bool, error) {
if statefulSet, err = statefulSets.Get(name, metav1.GetOptions{}); err != nil {
return false, err
}
// Apply the update, then attempt to push it to the apiserver.
applyUpdate(statefulSet)
if statefulSet, err = statefulSets.Update(statefulSet); err == nil {
Logf("Updating stateful set %s", name)
return true, nil
}
updateErr = err
return false, nil
})
if pollErr == wait.ErrWaitTimeout {
pollErr = fmt.Errorf("couldn't apply the provided updated to stateful set %q: %v", name, updateErr)
}
return statefulSet, pollErr
}

View File

@ -0,0 +1,357 @@
/*
Copyright 2016 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 framework
import (
"flag"
"fmt"
"io/ioutil"
"os"
"time"
"github.com/golang/glog"
"github.com/onsi/ginkgo/config"
"github.com/spf13/viper"
restclient "k8s.io/client-go/rest"
"k8s.io/client-go/tools/clientcmd"
clientcmdapi "k8s.io/client-go/tools/clientcmd/api"
"k8s.io/kubernetes/pkg/cloudprovider"
"k8s.io/kubernetes/pkg/kubelet/apis/kubeletconfig"
"k8s.io/kubernetes/pkg/kubemark"
)
const defaultHost = "http://127.0.0.1:8080"
type TestContextType struct {
KubeConfig string
KubemarkExternalKubeConfig string
KubeContext string
KubeAPIContentType string
KubeVolumeDir string
CertDir string
Host string
// TODO: Deprecating this over time... instead just use gobindata_util.go , see #23987.
RepoRoot string
DockershimCheckpointDir string
Provider string
CloudConfig CloudConfig
KubectlPath string
OutputDir string
ReportDir string
ReportPrefix string
Prefix string
MinStartupPods int
// Timeout for waiting for system pods to be running
SystemPodsStartupTimeout time.Duration
UpgradeTarget string
EtcdUpgradeStorage string
EtcdUpgradeVersion string
UpgradeImage string
GCEUpgradeScript string
ContainerRuntime string
ContainerRuntimeEndpoint string
// SystemdServices are comma separated list of systemd services the test framework
// will dump logs for.
SystemdServices string
ImageServiceEndpoint string
MasterOSDistro string
NodeOSDistro string
VerifyServiceAccount bool
DeleteNamespace bool
DeleteNamespaceOnFailure bool
AllowedNotReadyNodes int
CleanStart bool
// If set to 'true' or 'all' framework will start a goroutine monitoring resource usage of system add-ons.
// It will read the data every 30 seconds from all Nodes and print summary during afterEach. If set to 'master'
// only master Node will be monitored.
GatherKubeSystemResourceUsageData string
GatherLogsSizes bool
GatherMetricsAfterTest string
GatherSuiteMetricsAfterTest bool
// If set to 'true' framework will gather ClusterAutoscaler metrics when gathering them for other components.
IncludeClusterAutoscalerMetrics bool
// Currently supported values are 'hr' for human-readable and 'json'. It's a comma separated list.
OutputPrintType string
// NodeSchedulableTimeout is the timeout for waiting for all nodes to be schedulable.
NodeSchedulableTimeout time.Duration
// CreateTestingNS is responsible for creating namespace used for executing e2e tests.
// It accepts namespace base name, which will be prepended with e2e prefix, kube client
// and labels to be applied to a namespace.
CreateTestingNS CreateTestingNSFn
// If set to true test will dump data about the namespace in which test was running.
DumpLogsOnFailure bool
// Disables dumping cluster log from master and nodes after all tests.
DisableLogDump bool
// Path to the GCS artifacts directory to dump logs from nodes. Logexporter gets enabled if this is non-empty.
LogexporterGCSPath string
// If the garbage collector is enabled in the kube-apiserver and kube-controller-manager.
GarbageCollectorEnabled bool
// FeatureGates is a set of key=value pairs that describe feature gates for alpha/experimental features.
FeatureGates string
// Node e2e specific test context
NodeTestContextType
// Monitoring solution that is used in current cluster.
ClusterMonitoringMode string
// Indicates what path the kubernetes-anywhere is installed on
KubernetesAnywherePath string
// Viper-only parameters. These will in time replace all flags.
// Example: Create a file 'e2e.json' with the following:
// "Cadvisor":{
// "MaxRetries":"6"
// }
Viper string
Cadvisor struct {
MaxRetries int
SleepDurationMS int
}
LoggingSoak struct {
Scale int
MilliSecondsBetweenWaves int
}
}
// NodeTestContextType is part of TestContextType, it is shared by all node e2e test.
type NodeTestContextType struct {
// NodeE2E indicates whether it is running node e2e.
NodeE2E bool
// Name of the node to run tests on.
NodeName string
// NodeConformance indicates whether the test is running in node conformance mode.
NodeConformance bool
// PrepullImages indicates whether node e2e framework should prepull images.
PrepullImages bool
// KubeletConfig is the kubelet configuration the test is running against.
KubeletConfig kubeletconfig.KubeletConfiguration
// ImageDescription is the description of the image on which the test is running.
ImageDescription string
// SystemSpecName is the name of the system spec (e.g., gke) that's used in
// the node e2e test. If empty, the default one (system.DefaultSpec) is
// used. The system specs are in test/e2e_node/system/specs/.
SystemSpecName string
}
type CloudConfig struct {
ApiEndpoint string
ProjectID string
Zone string // for multizone tests, arbitrarily chosen zone
Region string
MultiZone bool
Cluster string
MasterName string
NodeInstanceGroup string // comma-delimited list of groups' names
NumNodes int
ClusterIPRange string
ClusterTag string
Network string
ConfigFile string // for azure and openstack
NodeTag string
MasterTag string
Provider cloudprovider.Interface
KubemarkController *kubemark.KubemarkController
}
var TestContext TestContextType
// Register flags common to all e2e test suites.
func RegisterCommonFlags() {
// Turn on verbose by default to get spec names
config.DefaultReporterConfig.Verbose = true
// Turn on EmitSpecProgress to get spec progress (especially on interrupt)
config.GinkgoConfig.EmitSpecProgress = true
// Randomize specs as well as suites
config.GinkgoConfig.RandomizeAllSpecs = true
flag.StringVar(&TestContext.GatherKubeSystemResourceUsageData, "gather-resource-usage", "false", "If set to 'true' or 'all' framework will be monitoring resource usage of system all add-ons in (some) e2e tests, if set to 'master' framework will be monitoring master node only, if set to 'none' of 'false' monitoring will be turned off.")
flag.BoolVar(&TestContext.GatherLogsSizes, "gather-logs-sizes", false, "If set to true framework will be monitoring logs sizes on all machines running e2e tests.")
flag.StringVar(&TestContext.GatherMetricsAfterTest, "gather-metrics-at-teardown", "false", "If set to 'true' framework will gather metrics from all components after each test. If set to 'master' only master component metrics would be gathered.")
flag.BoolVar(&TestContext.GatherSuiteMetricsAfterTest, "gather-suite-metrics-at-teardown", false, "If set to true framwork will gather metrics from all components after the whole test suite completes.")
flag.BoolVar(&TestContext.IncludeClusterAutoscalerMetrics, "include-cluster-autoscaler", false, "If set to true, framework will include Cluster Autoscaler when gathering metrics.")
flag.StringVar(&TestContext.OutputPrintType, "output-print-type", "json", "Format in which summaries should be printed: 'hr' for human readable, 'json' for JSON ones.")
flag.BoolVar(&TestContext.DumpLogsOnFailure, "dump-logs-on-failure", true, "If set to true test will dump data about the namespace in which test was running.")
flag.BoolVar(&TestContext.DisableLogDump, "disable-log-dump", false, "If set to true, logs from master and nodes won't be gathered after test run.")
flag.StringVar(&TestContext.LogexporterGCSPath, "logexporter-gcs-path", "", "Path to the GCS artifacts directory to dump logs from nodes. Logexporter gets enabled if this is non-empty.")
flag.BoolVar(&TestContext.DeleteNamespace, "delete-namespace", true, "If true tests will delete namespace after completion. It is only designed to make debugging easier, DO NOT turn it off by default.")
flag.BoolVar(&TestContext.DeleteNamespaceOnFailure, "delete-namespace-on-failure", true, "If true, framework will delete test namespace on failure. Used only during test debugging.")
flag.IntVar(&TestContext.AllowedNotReadyNodes, "allowed-not-ready-nodes", 0, "If non-zero, framework will allow for that many non-ready nodes when checking for all ready nodes.")
flag.StringVar(&TestContext.Host, "host", "", fmt.Sprintf("The host, or apiserver, to connect to. Will default to %s if this argument and --kubeconfig are not set", defaultHost))
flag.StringVar(&TestContext.ReportPrefix, "report-prefix", "", "Optional prefix for JUnit XML reports. Default is empty, which doesn't prepend anything to the default name.")
flag.StringVar(&TestContext.ReportDir, "report-dir", "", "Path to the directory where the JUnit XML reports should be saved. Default is empty, which doesn't generate these reports.")
flag.StringVar(&TestContext.FeatureGates, "feature-gates", "", "A set of key=value pairs that describe feature gates for alpha/experimental features.")
flag.StringVar(&TestContext.Viper, "viper-config", "e2e", "The name of the viper config i.e. 'e2e' will read values from 'e2e.json' locally. All e2e parameters are meant to be configurable by viper.")
flag.StringVar(&TestContext.ContainerRuntime, "container-runtime", "docker", "The container runtime of cluster VM instances (docker/rkt/remote).")
flag.StringVar(&TestContext.ContainerRuntimeEndpoint, "container-runtime-endpoint", "", "The container runtime endpoint of cluster VM instances.")
flag.StringVar(&TestContext.SystemdServices, "systemd-services", "docker", "The comma separated list of systemd services the framework will dump logs for.")
flag.StringVar(&TestContext.ImageServiceEndpoint, "image-service-endpoint", "", "The image service endpoint of cluster VM instances.")
flag.StringVar(&TestContext.DockershimCheckpointDir, "dockershim-checkpoint-dir", "/var/lib/dockershim/sandbox", "The directory for dockershim to store sandbox checkpoints.")
flag.StringVar(&TestContext.KubernetesAnywherePath, "kubernetes-anywhere-path", "/workspace/kubernetes-anywhere", "Which directory kubernetes-anywhere is installed to.")
}
// Register flags specific to the cluster e2e test suite.
func RegisterClusterFlags() {
flag.BoolVar(&TestContext.VerifyServiceAccount, "e2e-verify-service-account", true, "If true tests will verify the service account before running.")
flag.StringVar(&TestContext.KubeConfig, clientcmd.RecommendedConfigPathFlag, os.Getenv(clientcmd.RecommendedConfigPathEnvVar), "Path to kubeconfig containing embedded authinfo.")
flag.StringVar(&TestContext.KubemarkExternalKubeConfig, fmt.Sprintf("%s-%s", "kubemark-external", clientcmd.RecommendedConfigPathFlag), "", "Path to kubeconfig containing embedded authinfo for external cluster.")
flag.StringVar(&TestContext.KubeContext, clientcmd.FlagContext, "", "kubeconfig context to use/override. If unset, will use value from 'current-context'")
flag.StringVar(&TestContext.KubeAPIContentType, "kube-api-content-type", "application/vnd.kubernetes.protobuf", "ContentType used to communicate with apiserver")
flag.StringVar(&TestContext.KubeVolumeDir, "volume-dir", "/var/lib/kubelet", "Path to the directory containing the kubelet volumes.")
flag.StringVar(&TestContext.CertDir, "cert-dir", "", "Path to the directory containing the certs. Default is empty, which doesn't use certs.")
flag.StringVar(&TestContext.RepoRoot, "repo-root", "../../", "Root directory of kubernetes repository, for finding test files.")
flag.StringVar(&TestContext.Provider, "provider", "", "The name of the Kubernetes provider (gce, gke, local, vagrant, etc.)")
flag.StringVar(&TestContext.KubectlPath, "kubectl-path", "kubectl", "The kubectl binary to use. For development, you might use 'cluster/kubectl.sh' here.")
flag.StringVar(&TestContext.OutputDir, "e2e-output-dir", "/tmp", "Output directory for interesting/useful test data, like performance data, benchmarks, and other metrics.")
flag.StringVar(&TestContext.Prefix, "prefix", "e2e", "A prefix to be added to cloud resources created during testing.")
flag.StringVar(&TestContext.MasterOSDistro, "master-os-distro", "debian", "The OS distribution of cluster master (debian, trusty, or coreos).")
flag.StringVar(&TestContext.NodeOSDistro, "node-os-distro", "debian", "The OS distribution of cluster VM instances (debian, trusty, or coreos).")
flag.StringVar(&TestContext.ClusterMonitoringMode, "cluster-monitoring-mode", "influxdb", "The monitoring solution that is used in the cluster.")
// TODO: Flags per provider? Rename gce-project/gce-zone?
cloudConfig := &TestContext.CloudConfig
flag.StringVar(&cloudConfig.MasterName, "kube-master", "", "Name of the kubernetes master. Only required if provider is gce or gke")
flag.StringVar(&cloudConfig.ApiEndpoint, "gce-api-endpoint", "", "The GCE ApiEndpoint being used, if applicable")
flag.StringVar(&cloudConfig.ProjectID, "gce-project", "", "The GCE project being used, if applicable")
flag.StringVar(&cloudConfig.Zone, "gce-zone", "", "GCE zone being used, if applicable")
flag.StringVar(&cloudConfig.Region, "gce-region", "", "GCE region being used, if applicable")
flag.BoolVar(&cloudConfig.MultiZone, "gce-multizone", false, "If true, start GCE cloud provider with multizone support.")
flag.StringVar(&cloudConfig.Cluster, "gke-cluster", "", "GKE name of cluster being used, if applicable")
flag.StringVar(&cloudConfig.NodeInstanceGroup, "node-instance-group", "", "Name of the managed instance group for nodes. Valid only for gce, gke or aws. If there is more than one group: comma separated list of groups.")
flag.StringVar(&cloudConfig.Network, "network", "e2e", "The cloud provider network for this e2e cluster.")
flag.IntVar(&cloudConfig.NumNodes, "num-nodes", -1, "Number of nodes in the cluster")
flag.StringVar(&cloudConfig.ClusterIPRange, "cluster-ip-range", "10.64.0.0/14", "A CIDR notation IP range from which to assign IPs in the cluster.")
flag.StringVar(&cloudConfig.NodeTag, "node-tag", "", "Network tags used on node instances. Valid only for gce, gke")
flag.StringVar(&cloudConfig.MasterTag, "master-tag", "", "Network tags used on master instances. Valid only for gce, gke")
flag.StringVar(&cloudConfig.ClusterTag, "cluster-tag", "", "Tag used to identify resources. Only required if provider is aws.")
flag.StringVar(&cloudConfig.ConfigFile, "cloud-config-file", "", "Cloud config file. Only required if provider is azure.")
flag.IntVar(&TestContext.MinStartupPods, "minStartupPods", 0, "The number of pods which we need to see in 'Running' state with a 'Ready' condition of true, before we try running tests. This is useful in any cluster which needs some base pod-based services running before it can be used.")
flag.DurationVar(&TestContext.SystemPodsStartupTimeout, "system-pods-startup-timeout", 10*time.Minute, "Timeout for waiting for all system pods to be running before starting tests.")
flag.DurationVar(&TestContext.NodeSchedulableTimeout, "node-schedulable-timeout", 4*time.Hour, "Timeout for waiting for all nodes to be schedulable.")
flag.StringVar(&TestContext.UpgradeTarget, "upgrade-target", "ci/latest", "Version to upgrade to (e.g. 'release/stable', 'release/latest', 'ci/latest', '0.19.1', '0.19.1-669-gabac8c8') if doing an upgrade test.")
flag.StringVar(&TestContext.EtcdUpgradeStorage, "etcd-upgrade-storage", "", "The storage version to upgrade to (either 'etcdv2' or 'etcdv3') if doing an etcd upgrade test.")
flag.StringVar(&TestContext.EtcdUpgradeVersion, "etcd-upgrade-version", "", "The etcd binary version to upgrade to (e.g., '3.0.14', '2.3.7') if doing an etcd upgrade test.")
flag.StringVar(&TestContext.UpgradeImage, "upgrade-image", "", "Image to upgrade to (e.g. 'container_vm' or 'gci') if doing an upgrade test.")
flag.StringVar(&TestContext.GCEUpgradeScript, "gce-upgrade-script", "", "Script to use to upgrade a GCE cluster.")
flag.BoolVar(&TestContext.CleanStart, "clean-start", false, "If true, purge all namespaces except default and system before running tests. This serves to Cleanup test namespaces from failed/interrupted e2e runs in a long-lived cluster.")
flag.BoolVar(&TestContext.GarbageCollectorEnabled, "garbage-collector-enabled", true, "Set to true if the garbage collector is enabled in the kube-apiserver and kube-controller-manager, then some tests will rely on the garbage collector to delete dependent resources.")
}
// Register flags specific to the node e2e test suite.
func RegisterNodeFlags() {
// Mark the test as node e2e when node flags are api.Registry.
TestContext.NodeE2E = true
flag.StringVar(&TestContext.NodeName, "node-name", "", "Name of the node to run tests on.")
// TODO(random-liu): Move kubelet start logic out of the test.
// TODO(random-liu): Move log fetch logic out of the test.
// There are different ways to start kubelet (systemd, initd, docker, rkt, manually started etc.)
// and manage logs (journald, upstart etc.).
// For different situation we need to mount different things into the container, run different commands.
// It is hard and unnecessary to deal with the complexity inside the test suite.
flag.BoolVar(&TestContext.NodeConformance, "conformance", false, "If true, the test suite will not start kubelet, and fetch system log (kernel, docker, kubelet log etc.) to the report directory.")
flag.BoolVar(&TestContext.PrepullImages, "prepull-images", true, "If true, prepull images so image pull failures do not cause test failures.")
flag.StringVar(&TestContext.ImageDescription, "image-description", "", "The description of the image which the test will be running on.")
flag.StringVar(&TestContext.SystemSpecName, "system-spec-name", "", "The name of the system spec (e.g., gke) that's used in the node e2e test. The system specs are in test/e2e_node/system/specs/. This is used by the test framework to determine which tests to run for validating the system requirements.")
}
// ViperizeFlags sets up all flag and config processing. Future configuration info should be added to viper, not to flags.
func ViperizeFlags() {
// Part 1: Set regular flags.
// TODO: Future, lets eliminate e2e 'flag' deps entirely in favor of viper only,
// since go test 'flag's are sort of incompatible w/ flag, glog, etc.
RegisterCommonFlags()
RegisterClusterFlags()
flag.Parse()
// Part 2: Set Viper provided flags.
// This must be done after common flags are registered, since Viper is a flag option.
viper.SetConfigName(TestContext.Viper)
viper.AddConfigPath(".")
viper.ReadInConfig()
// TODO Consider wether or not we want to use overwriteFlagsWithViperConfig().
viper.Unmarshal(&TestContext)
AfterReadingAllFlags(&TestContext)
}
func createKubeConfig(clientCfg *restclient.Config) *clientcmdapi.Config {
clusterNick := "cluster"
userNick := "user"
contextNick := "context"
config := clientcmdapi.NewConfig()
credentials := clientcmdapi.NewAuthInfo()
credentials.Token = clientCfg.BearerToken
credentials.ClientCertificate = clientCfg.TLSClientConfig.CertFile
if len(credentials.ClientCertificate) == 0 {
credentials.ClientCertificateData = clientCfg.TLSClientConfig.CertData
}
credentials.ClientKey = clientCfg.TLSClientConfig.KeyFile
if len(credentials.ClientKey) == 0 {
credentials.ClientKeyData = clientCfg.TLSClientConfig.KeyData
}
config.AuthInfos[userNick] = credentials
cluster := clientcmdapi.NewCluster()
cluster.Server = clientCfg.Host
cluster.CertificateAuthority = clientCfg.CAFile
if len(cluster.CertificateAuthority) == 0 {
cluster.CertificateAuthorityData = clientCfg.CAData
}
cluster.InsecureSkipTLSVerify = clientCfg.Insecure
config.Clusters[clusterNick] = cluster
context := clientcmdapi.NewContext()
context.Cluster = clusterNick
context.AuthInfo = userNick
config.Contexts[contextNick] = context
config.CurrentContext = contextNick
return config
}
// AfterReadingAllFlags makes changes to the context after all flags
// have been read.
func AfterReadingAllFlags(t *TestContextType) {
// Only set a default host if one won't be supplied via kubeconfig
if len(t.Host) == 0 && len(t.KubeConfig) == 0 {
// Check if we can use the in-cluster config
if clusterConfig, err := restclient.InClusterConfig(); err == nil {
if tempFile, err := ioutil.TempFile(os.TempDir(), "kubeconfig-"); err == nil {
kubeConfig := createKubeConfig(clusterConfig)
clientcmd.WriteToFile(*kubeConfig, tempFile.Name())
t.KubeConfig = tempFile.Name()
glog.Infof("Using a temporary kubeconfig file from in-cluster config : %s", tempFile.Name())
}
}
if len(t.KubeConfig) == 0 {
glog.Warningf("Unable to find in-cluster config, using default host : %s", defaultHost)
t.Host = defaultHost
}
}
}

View File

@ -0,0 +1,34 @@
load("@io_bazel_rules_go//go:def.bzl", "go_library", "go_test")
go_library(
name = "go_default_library",
srcs = ["timer.go"],
importpath = "k8s.io/kubernetes/test/e2e/framework/timer",
visibility = ["//visibility:public"],
deps = [
"//test/e2e/framework:go_default_library",
"//test/e2e/perftype:go_default_library",
],
)
go_test(
name = "go_default_test",
srcs = ["timer_test.go"],
importpath = "k8s.io/kubernetes/test/e2e/framework/timer",
library = ":go_default_library",
deps = ["//vendor/github.com/onsi/gomega:go_default_library"],
)
filegroup(
name = "package-srcs",
srcs = glob(["**"]),
tags = ["automanaged"],
visibility = ["//visibility:private"],
)
filegroup(
name = "all-srcs",
srcs = [":package-srcs"],
tags = ["automanaged"],
visibility = ["//visibility:public"],
)

View File

@ -0,0 +1,126 @@
/*
Copyright 2017 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 timer
import (
"time"
"bytes"
"fmt"
"k8s.io/kubernetes/test/e2e/framework"
"k8s.io/kubernetes/test/e2e/perftype"
"sync"
)
var now = time.Now
// Represents a phase of a test. Phases can overlap.
type Phase struct {
sequenceNumber int
name string
startTime time.Time
endTime time.Time
}
func (phase *Phase) ended() bool {
return !phase.endTime.IsZero()
}
// End marks the phase as ended, unless it had already been ended before.
func (phase *Phase) End() {
if !phase.ended() {
phase.endTime = now()
}
}
func (phase *Phase) label() string {
return fmt.Sprintf("%03d-%s", phase.sequenceNumber, phase.name)
}
func (phase *Phase) duration() time.Duration {
endTime := phase.endTime
if !phase.ended() {
endTime = now()
}
return endTime.Sub(phase.startTime)
}
func (phase *Phase) humanReadable() string {
if phase.ended() {
return fmt.Sprintf("Phase %s: %v\n", phase.label(), phase.duration())
} else {
return fmt.Sprintf("Phase %s: %v so far\n", phase.label(), phase.duration())
}
}
// A TestPhaseTimer groups phases and provides a way to export their measurements as JSON or human-readable text.
// It is safe to use concurrently.
type TestPhaseTimer struct {
lock sync.Mutex
phases []*Phase
}
// NewTestPhaseTimer creates a new TestPhaseTimer.
func NewTestPhaseTimer() *TestPhaseTimer {
return &TestPhaseTimer{}
}
// StartPhase starts a new phase.
// sequenceNumber is an integer prepended to phaseName in the output, such that lexicographic sorting
// of phases in perfdash reconstructs the order of execution. Unfortunately it needs to be
// provided manually, since a simple incrementing counter would have the effect that inserting
// a new phase would renumber subsequent phases, breaking the continuity of historical records.
func (timer *TestPhaseTimer) StartPhase(sequenceNumber int, phaseName string) *Phase {
timer.lock.Lock()
defer timer.lock.Unlock()
newPhase := &Phase{sequenceNumber: sequenceNumber, name: phaseName, startTime: now()}
timer.phases = append(timer.phases, newPhase)
return newPhase
}
func (timer *TestPhaseTimer) SummaryKind() string {
return "TestPhaseTimer"
}
func (timer *TestPhaseTimer) PrintHumanReadable() string {
buf := bytes.Buffer{}
timer.lock.Lock()
defer timer.lock.Unlock()
for _, phase := range timer.phases {
buf.WriteString(phase.humanReadable())
}
return buf.String()
}
func (timer *TestPhaseTimer) PrintJSON() string {
data := perftype.PerfData{
Version: "v1",
DataItems: []perftype.DataItem{{
Unit: "s",
Labels: map[string]string{"test": "phases"},
Data: make(map[string]float64)}}}
timer.lock.Lock()
defer timer.lock.Unlock()
for _, phase := range timer.phases {
data.DataItems[0].Data[phase.label()] = phase.duration().Seconds()
if !phase.ended() {
data.DataItems[0].Labels["ended"] = "false"
}
}
return framework.PrettyPrintJSON(data)
}

View File

@ -0,0 +1,92 @@
/*
Copyright 2017 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 timer
import (
"testing"
"time"
. "github.com/onsi/gomega"
)
var currentTime time.Time
func init() {
setCurrentTimeSinceEpoch(0)
now = func() time.Time { return currentTime }
}
func setCurrentTimeSinceEpoch(duration time.Duration) {
currentTime = time.Unix(0, duration.Nanoseconds())
}
func testUsageWithDefer(timer *TestPhaseTimer) {
defer timer.StartPhase(33, "two").End()
setCurrentTimeSinceEpoch(6*time.Second + 500*time.Millisecond)
}
func TestTimer(t *testing.T) {
RegisterTestingT(t)
timer := NewTestPhaseTimer()
setCurrentTimeSinceEpoch(1 * time.Second)
phaseOne := timer.StartPhase(1, "one")
setCurrentTimeSinceEpoch(3 * time.Second)
testUsageWithDefer(timer)
Expect(timer.PrintJSON()).To(MatchJSON(`{
"version": "v1",
"dataItems": [
{
"data": {
"001-one": 5.5,
"033-two": 3.5
},
"unit": "s",
"labels": {
"test": "phases",
"ended": "false"
}
}
]
}`))
Expect(timer.PrintHumanReadable()).To(Equal(`Phase 001-one: 5.5s so far
Phase 033-two: 3.5s
`))
setCurrentTimeSinceEpoch(7*time.Second + 500*time.Millisecond)
phaseOne.End()
Expect(timer.PrintJSON()).To(MatchJSON(`{
"version": "v1",
"dataItems": [
{
"data": {
"001-one": 6.5,
"033-two": 3.5
},
"unit": "s",
"labels": {
"test": "phases"
}
}
]
}`))
Expect(timer.PrintHumanReadable()).To(Equal(`Phase 001-one: 6.5s
Phase 033-two: 3.5s
`))
}

View File

@ -0,0 +1,105 @@
/*
Copyright 2017 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 framework
import (
"fmt"
"os/exec"
"path"
"strings"
"time"
"k8s.io/apimachinery/pkg/util/wait"
"k8s.io/apimachinery/pkg/version"
clientset "k8s.io/client-go/kubernetes"
)
// RealVersion turns a version constants into a version string deployable on
// GKE. See hack/get-build.sh for more information.
func RealVersion(s string) (string, error) {
Logf("Getting real version for %q", s)
v, _, err := RunCmd(path.Join(TestContext.RepoRoot, "hack/get-build.sh"), "-v", s)
if err != nil {
return v, err
}
Logf("Version for %q is %q", s, v)
return strings.TrimPrefix(strings.TrimSpace(v), "v"), nil
}
func traceRouteToMaster() {
path, err := exec.LookPath("traceroute")
if err != nil {
Logf("Could not find traceroute program")
return
}
cmd := exec.Command(path, "-I", GetMasterHost())
out, err := cmd.Output()
if len(out) != 0 {
Logf(string(out))
}
if exiterr, ok := err.(*exec.ExitError); err != nil && ok {
Logf("error while running traceroute: %s", exiterr.Stderr)
}
}
func CheckMasterVersion(c clientset.Interface, want string) error {
Logf("Checking master version")
var err error
var v *version.Info
waitErr := wait.PollImmediate(5*time.Second, 2*time.Minute, func() (bool, error) {
v, err = c.Discovery().ServerVersion()
if err != nil {
traceRouteToMaster()
return false, nil
}
return true, nil
})
if waitErr != nil {
return fmt.Errorf("CheckMasterVersion() couldn't get the master version: %v", err)
}
// We do prefix trimming and then matching because:
// want looks like: 0.19.3-815-g50e67d4
// got looks like: v0.19.3-815-g50e67d4034e858-dirty
got := strings.TrimPrefix(v.GitVersion, "v")
if !strings.HasPrefix(got, want) {
return fmt.Errorf("master had kube-apiserver version %s which does not start with %s",
got, want)
}
Logf("Master is at version %s", want)
return nil
}
func CheckNodesVersions(cs clientset.Interface, want string) error {
l := GetReadySchedulableNodesOrDie(cs)
for _, n := range l.Items {
// We do prefix trimming and then matching because:
// want looks like: 0.19.3-815-g50e67d4
// kv/kvp look like: v0.19.3-815-g50e67d4034e858-dirty
kv, kpv := strings.TrimPrefix(n.Status.NodeInfo.KubeletVersion, "v"),
strings.TrimPrefix(n.Status.NodeInfo.KubeProxyVersion, "v")
if !strings.HasPrefix(kv, want) {
return fmt.Errorf("node %s had kubelet version %s which does not start with %s",
n.ObjectMeta.Name, kv, want)
}
if !strings.HasPrefix(kpv, want) {
return fmt.Errorf("node %s had kube-proxy version %s which does not start with %s",
n.ObjectMeta.Name, kpv, want)
}
}
return nil
}

5127
vendor/k8s.io/kubernetes/test/e2e/framework/util.go generated vendored Normal file

File diff suppressed because it is too large Load Diff

View File

@ -0,0 +1,508 @@
/*
Copyright 2017 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.
*/
/*
* This test checks that various VolumeSources are working.
*
* There are two ways, how to test the volumes:
* 1) With containerized server (NFS, Ceph, Gluster, iSCSI, ...)
* The test creates a server pod, exporting simple 'index.html' file.
* Then it uses appropriate VolumeSource to import this file into a client pod
* and checks that the pod can see the file. It does so by importing the file
* into web server root and loadind the index.html from it.
*
* These tests work only when privileged containers are allowed, exporting
* various filesystems (NFS, GlusterFS, ...) usually needs some mounting or
* other privileged magic in the server pod.
*
* Note that the server containers are for testing purposes only and should not
* be used in production.
*
* 2) With server outside of Kubernetes (Cinder, ...)
* Appropriate server (e.g. OpenStack Cinder) must exist somewhere outside
* the tested Kubernetes cluster. The test itself creates a new volume,
* and checks, that Kubernetes can use it as a volume.
*/
package framework
import (
"fmt"
"strconv"
"time"
"k8s.io/api/core/v1"
apierrs "k8s.io/apimachinery/pkg/api/errors"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
clientset "k8s.io/client-go/kubernetes"
"github.com/golang/glog"
. "github.com/onsi/ginkgo"
. "github.com/onsi/gomega"
)
// Current supported images for e2e volume testing to be assigned to VolumeTestConfig.serverImage
const (
NfsServerImage string = "gcr.io/google_containers/volume-nfs:0.8"
IscsiServerImage string = "gcr.io/google_containers/volume-iscsi:0.1"
GlusterfsServerImage string = "gcr.io/google_containers/volume-gluster:0.2"
CephServerImage string = "gcr.io/google_containers/volume-ceph:0.1"
RbdServerImage string = "gcr.io/google_containers/volume-rbd:0.1"
)
const (
Kb int64 = 1000
Mb int64 = 1000 * Kb
Gb int64 = 1000 * Mb
Tb int64 = 1000 * Gb
KiB int64 = 1024
MiB int64 = 1024 * KiB
GiB int64 = 1024 * MiB
TiB int64 = 1024 * GiB
)
// Configuration of one tests. The test consist of:
// - server pod - runs serverImage, exports ports[]
// - client pod - does not need any special configuration
type VolumeTestConfig struct {
Namespace string
// Prefix of all pods. Typically the test name.
Prefix string
// Name of container image for the server pod.
ServerImage string
// Ports to export from the server pod. TCP only.
ServerPorts []int
// Commands to run in the container image.
ServerCmds []string
// Arguments to pass to the container image.
ServerArgs []string
// Volumes needed to be mounted to the server container from the host
// map <host (source) path> -> <container (dst.) path>
ServerVolumes map[string]string
// Wait for the pod to terminate successfully
// False indicates that the pod is long running
WaitForCompletion bool
// ServerNodeName is the spec.nodeName to run server pod on. Default is any node.
ServerNodeName string
// ClientNodeName is the spec.nodeName to run client pod on. Default is any node.
ClientNodeName string
// NodeSelector to use in pod spec (server, client and injector pods).
NodeSelector map[string]string
}
// VolumeTest contains a volume to mount into a client pod and its
// expected content.
type VolumeTest struct {
Volume v1.VolumeSource
File string
ExpectedContent string
}
// NFS-specific wrapper for CreateStorageServer.
func NewNFSServer(cs clientset.Interface, namespace string, args []string) (config VolumeTestConfig, pod *v1.Pod, ip string) {
config = VolumeTestConfig{
Namespace: namespace,
Prefix: "nfs",
ServerImage: NfsServerImage,
ServerPorts: []int{2049},
}
if len(args) > 0 {
config.ServerArgs = args
}
pod, ip = CreateStorageServer(cs, config)
return config, pod, ip
}
// GlusterFS-specific wrapper for CreateStorageServer. Also creates the gluster endpoints object.
func NewGlusterfsServer(cs clientset.Interface, namespace string) (config VolumeTestConfig, pod *v1.Pod, ip string) {
config = VolumeTestConfig{
Namespace: namespace,
Prefix: "gluster",
ServerImage: GlusterfsServerImage,
ServerPorts: []int{24007, 24008, 49152},
}
pod, ip = CreateStorageServer(cs, config)
By("creating Gluster endpoints")
endpoints := &v1.Endpoints{
TypeMeta: metav1.TypeMeta{
Kind: "Endpoints",
APIVersion: "v1",
},
ObjectMeta: metav1.ObjectMeta{
Name: config.Prefix + "-server",
},
Subsets: []v1.EndpointSubset{
{
Addresses: []v1.EndpointAddress{
{
IP: ip,
},
},
Ports: []v1.EndpointPort{
{
Name: "gluster",
Port: 24007,
Protocol: v1.ProtocolTCP,
},
},
},
},
}
endpoints, err := cs.CoreV1().Endpoints(namespace).Create(endpoints)
Expect(err).NotTo(HaveOccurred(), "failed to create endpoints for Gluster server")
return config, pod, ip
}
// iSCSI-specific wrapper for CreateStorageServer.
func NewISCSIServer(cs clientset.Interface, namespace string) (config VolumeTestConfig, pod *v1.Pod, ip string) {
config = VolumeTestConfig{
Namespace: namespace,
Prefix: "iscsi",
ServerImage: IscsiServerImage,
ServerPorts: []int{3260},
ServerVolumes: map[string]string{
// iSCSI container needs to insert modules from the host
"/lib/modules": "/lib/modules",
},
}
pod, ip = CreateStorageServer(cs, config)
return config, pod, ip
}
// CephRBD-specific wrapper for CreateStorageServer.
func NewRBDServer(cs clientset.Interface, namespace string) (config VolumeTestConfig, pod *v1.Pod, ip string) {
config = VolumeTestConfig{
Namespace: namespace,
Prefix: "rbd",
ServerImage: RbdServerImage,
ServerPorts: []int{6789},
ServerVolumes: map[string]string{
"/lib/modules": "/lib/modules",
},
}
pod, ip = CreateStorageServer(cs, config)
return config, pod, ip
}
// Wrapper for StartVolumeServer(). A storage server config is passed in, and a pod pointer
// and ip address string are returned.
// Note: Expect() is called so no error is returned.
func CreateStorageServer(cs clientset.Interface, config VolumeTestConfig) (pod *v1.Pod, ip string) {
pod = StartVolumeServer(cs, config)
Expect(pod).NotTo(BeNil(), "storage server pod should not be nil")
ip = pod.Status.PodIP
Expect(len(ip)).NotTo(BeZero(), fmt.Sprintf("pod %s's IP should not be empty", pod.Name))
Logf("%s server pod IP address: %s", config.Prefix, ip)
return pod, ip
}
// Starts a container specified by config.serverImage and exports all
// config.serverPorts from it. The returned pod should be used to get the server
// IP address and create appropriate VolumeSource.
func StartVolumeServer(client clientset.Interface, config VolumeTestConfig) *v1.Pod {
podClient := client.CoreV1().Pods(config.Namespace)
portCount := len(config.ServerPorts)
serverPodPorts := make([]v1.ContainerPort, portCount)
for i := 0; i < portCount; i++ {
portName := fmt.Sprintf("%s-%d", config.Prefix, i)
serverPodPorts[i] = v1.ContainerPort{
Name: portName,
ContainerPort: int32(config.ServerPorts[i]),
Protocol: v1.ProtocolTCP,
}
}
volumeCount := len(config.ServerVolumes)
volumes := make([]v1.Volume, volumeCount)
mounts := make([]v1.VolumeMount, volumeCount)
i := 0
for src, dst := range config.ServerVolumes {
mountName := fmt.Sprintf("path%d", i)
volumes[i].Name = mountName
volumes[i].VolumeSource.HostPath = &v1.HostPathVolumeSource{
Path: src,
}
mounts[i].Name = mountName
mounts[i].ReadOnly = false
mounts[i].MountPath = dst
i++
}
serverPodName := fmt.Sprintf("%s-server", config.Prefix)
By(fmt.Sprint("creating ", serverPodName, " pod"))
privileged := new(bool)
*privileged = true
restartPolicy := v1.RestartPolicyAlways
if config.WaitForCompletion {
restartPolicy = v1.RestartPolicyNever
}
serverPod := &v1.Pod{
TypeMeta: metav1.TypeMeta{
Kind: "Pod",
APIVersion: "v1",
},
ObjectMeta: metav1.ObjectMeta{
Name: serverPodName,
Labels: map[string]string{
"role": serverPodName,
},
},
Spec: v1.PodSpec{
Containers: []v1.Container{
{
Name: serverPodName,
Image: config.ServerImage,
SecurityContext: &v1.SecurityContext{
Privileged: privileged,
},
Command: config.ServerCmds,
Args: config.ServerArgs,
Ports: serverPodPorts,
VolumeMounts: mounts,
},
},
Volumes: volumes,
RestartPolicy: restartPolicy,
NodeName: config.ServerNodeName,
NodeSelector: config.NodeSelector,
},
}
var pod *v1.Pod
serverPod, err := podClient.Create(serverPod)
// ok if the server pod already exists. TODO: make this controllable by callers
if err != nil {
if apierrs.IsAlreadyExists(err) {
Logf("Ignore \"already-exists\" error, re-get pod...")
By(fmt.Sprintf("re-getting the %q server pod", serverPodName))
serverPod, err = podClient.Get(serverPodName, metav1.GetOptions{})
ExpectNoError(err, "Cannot re-get the server pod %q: %v", serverPodName, err)
pod = serverPod
} else {
ExpectNoError(err, "Failed to create %q pod: %v", serverPodName, err)
}
}
if config.WaitForCompletion {
ExpectNoError(WaitForPodSuccessInNamespace(client, serverPod.Name, serverPod.Namespace))
ExpectNoError(podClient.Delete(serverPod.Name, nil))
} else {
ExpectNoError(WaitForPodRunningInNamespace(client, serverPod))
if pod == nil {
By(fmt.Sprintf("locating the %q server pod", serverPodName))
pod, err = podClient.Get(serverPodName, metav1.GetOptions{})
ExpectNoError(err, "Cannot locate the server pod %q: %v", serverPodName, err)
}
}
return pod
}
// Clean both server and client pods.
func VolumeTestCleanup(f *Framework, config VolumeTestConfig) {
By(fmt.Sprint("cleaning the environment after ", config.Prefix))
defer GinkgoRecover()
client := f.ClientSet
podClient := client.CoreV1().Pods(config.Namespace)
err := podClient.Delete(config.Prefix+"-client", nil)
if err != nil {
// Log the error before failing test: if the test has already failed,
// framework.ExpectNoError() won't print anything to logs!
glog.Warningf("Failed to delete client pod: %v", err)
ExpectNoError(err, "Failed to delete client pod: %v", err)
}
if config.ServerImage != "" {
if err := f.WaitForPodTerminated(config.Prefix+"-client", ""); !apierrs.IsNotFound(err) {
ExpectNoError(err, "Failed to wait client pod terminated: %v", err)
}
// See issue #24100.
// Prevent umount errors by making sure making sure the client pod exits cleanly *before* the volume server pod exits.
By("sleeping a bit so client can stop and unmount")
time.Sleep(20 * time.Second)
err = podClient.Delete(config.Prefix+"-server", nil)
if err != nil {
glog.Warningf("Failed to delete server pod: %v", err)
ExpectNoError(err, "Failed to delete server pod: %v", err)
}
}
}
// Start a client pod using given VolumeSource (exported by startVolumeServer())
// and check that the pod sees expected data, e.g. from the server pod.
// Multiple VolumeTests can be specified to mount multiple volumes to a single
// pod.
func TestVolumeClient(client clientset.Interface, config VolumeTestConfig, fsGroup *int64, tests []VolumeTest) {
By(fmt.Sprint("starting ", config.Prefix, " client"))
var gracePeriod int64 = 1
clientPod := &v1.Pod{
TypeMeta: metav1.TypeMeta{
Kind: "Pod",
APIVersion: "v1",
},
ObjectMeta: metav1.ObjectMeta{
Name: config.Prefix + "-client",
Labels: map[string]string{
"role": config.Prefix + "-client",
},
},
Spec: v1.PodSpec{
Containers: []v1.Container{
{
Name: config.Prefix + "-client",
Image: BusyBoxImage,
WorkingDir: "/opt",
// An imperative and easily debuggable container which reads vol contents for
// us to scan in the tests or by eye.
// We expect that /opt is empty in the minimal containers which we use in this test.
Command: []string{
"/bin/sh",
"-c",
"while true ; do cat /opt/0/index.html ; sleep 2 ; ls -altrh /opt/ ; sleep 2 ; done ",
},
VolumeMounts: []v1.VolumeMount{},
},
},
TerminationGracePeriodSeconds: &gracePeriod,
SecurityContext: &v1.PodSecurityContext{
SELinuxOptions: &v1.SELinuxOptions{
Level: "s0:c0,c1",
},
},
Volumes: []v1.Volume{},
NodeName: config.ClientNodeName,
NodeSelector: config.NodeSelector,
},
}
podsNamespacer := client.CoreV1().Pods(config.Namespace)
if fsGroup != nil {
clientPod.Spec.SecurityContext.FSGroup = fsGroup
}
for i, test := range tests {
volumeName := fmt.Sprintf("%s-%s-%d", config.Prefix, "volume", i)
clientPod.Spec.Containers[0].VolumeMounts = append(clientPod.Spec.Containers[0].VolumeMounts, v1.VolumeMount{
Name: volumeName,
MountPath: fmt.Sprintf("/opt/%d", i),
})
clientPod.Spec.Volumes = append(clientPod.Spec.Volumes, v1.Volume{
Name: volumeName,
VolumeSource: test.Volume,
})
}
clientPod, err := podsNamespacer.Create(clientPod)
if err != nil {
Failf("Failed to create %s pod: %v", clientPod.Name, err)
}
ExpectNoError(WaitForPodRunningInNamespace(client, clientPod))
By("Checking that text file contents are perfect.")
for i, test := range tests {
fileName := fmt.Sprintf("/opt/%d/%s", i, test.File)
_, err = LookForStringInPodExec(config.Namespace, clientPod.Name, []string{"cat", fileName}, test.ExpectedContent, time.Minute)
Expect(err).NotTo(HaveOccurred(), "failed: finding the contents of the mounted file %s.", fileName)
}
if fsGroup != nil {
By("Checking fsGroup is correct.")
_, err = LookForStringInPodExec(config.Namespace, clientPod.Name, []string{"ls", "-ld", "/opt/0"}, strconv.Itoa(int(*fsGroup)), time.Minute)
Expect(err).NotTo(HaveOccurred(), "failed: getting the right priviliges in the file %v", int(*fsGroup))
}
}
// Insert index.html with given content into given volume. It does so by
// starting and auxiliary pod which writes the file there.
// The volume must be writable.
func InjectHtml(client clientset.Interface, config VolumeTestConfig, volume v1.VolumeSource, content string) {
By(fmt.Sprint("starting ", config.Prefix, " injector"))
podClient := client.CoreV1().Pods(config.Namespace)
injectPod := &v1.Pod{
TypeMeta: metav1.TypeMeta{
Kind: "Pod",
APIVersion: "v1",
},
ObjectMeta: metav1.ObjectMeta{
Name: config.Prefix + "-injector",
Labels: map[string]string{
"role": config.Prefix + "-injector",
},
},
Spec: v1.PodSpec{
Containers: []v1.Container{
{
Name: config.Prefix + "-injector",
Image: BusyBoxImage,
Command: []string{"/bin/sh"},
Args: []string{"-c", "echo '" + content + "' > /mnt/index.html && chmod o+rX /mnt /mnt/index.html"},
VolumeMounts: []v1.VolumeMount{
{
Name: config.Prefix + "-volume",
MountPath: "/mnt",
},
},
},
},
SecurityContext: &v1.PodSecurityContext{
SELinuxOptions: &v1.SELinuxOptions{
Level: "s0:c0,c1",
},
},
RestartPolicy: v1.RestartPolicyNever,
Volumes: []v1.Volume{
{
Name: config.Prefix + "-volume",
VolumeSource: volume,
},
},
NodeSelector: config.NodeSelector,
},
}
defer func() {
podClient.Delete(config.Prefix+"-injector", nil)
}()
injectPod, err := podClient.Create(injectPod)
ExpectNoError(err, "Failed to create injector pod: %v", err)
err = WaitForPodSuccessInNamespace(client, injectPod.Name, injectPod.Namespace)
Expect(err).NotTo(HaveOccurred())
}
func CreateGCEVolume() (*v1.PersistentVolumeSource, string) {
diskName, err := CreatePDWithRetry()
ExpectNoError(err)
return &v1.PersistentVolumeSource{
GCEPersistentDisk: &v1.GCEPersistentDiskVolumeSource{
PDName: diskName,
FSType: "ext3",
ReadOnly: false,
},
}, diskName
}