代码拉取完成,页面将自动刷新
/*
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 vsphere
import (
"fmt"
"strconv"
"time"
. "github.com/onsi/ginkgo"
. "github.com/onsi/gomega"
"k8s.io/api/core/v1"
"k8s.io/apimachinery/pkg/util/uuid"
clientset "k8s.io/client-go/kubernetes"
"k8s.io/kubernetes/test/e2e/framework"
"k8s.io/kubernetes/test/e2e/storage/utils"
)
var _ = utils.SIGDescribe("Volume Placement", func() {
f := framework.NewDefaultFramework("volume-placement")
const (
NodeLabelKey = "vsphere_e2e_label_volume_placement"
)
var (
c clientset.Interface
ns string
volumePaths []string
node1Name string
node1KeyValueLabel map[string]string
node2Name string
node2KeyValueLabel map[string]string
isNodeLabeled bool
nodeInfo *NodeInfo
vsp *VSphere
)
BeforeEach(func() {
framework.SkipUnlessProviderIs("vsphere")
Bootstrap(f)
c = f.ClientSet
ns = f.Namespace.Name
framework.ExpectNoError(framework.WaitForAllNodesSchedulable(c, framework.TestContext.NodeSchedulableTimeout))
if !isNodeLabeled {
node1Name, node1KeyValueLabel, node2Name, node2KeyValueLabel = testSetupVolumePlacement(c, ns)
isNodeLabeled = true
nodeInfo = TestContext.NodeMapper.GetNodeInfo(node1Name)
vsp = nodeInfo.VSphere
}
By("creating vmdk")
volumePath, err := vsp.CreateVolume(&VolumeOptions{}, nodeInfo.DataCenterRef)
Expect(err).NotTo(HaveOccurred())
volumePaths = append(volumePaths, volumePath)
})
AfterEach(func() {
for _, volumePath := range volumePaths {
vsp.DeleteVolume(volumePath, nodeInfo.DataCenterRef)
}
volumePaths = nil
})
/*
Steps
1. Remove labels assigned to node 1 and node 2
2. Delete VMDK volume
*/
framework.AddCleanupAction(func() {
// Cleanup actions will be called even when the tests are skipped and leaves namespace unset.
if len(ns) > 0 {
if len(node1KeyValueLabel) > 0 {
framework.RemoveLabelOffNode(c, node1Name, NodeLabelKey)
}
if len(node2KeyValueLabel) > 0 {
framework.RemoveLabelOffNode(c, node2Name, NodeLabelKey)
}
}
})
/*
Steps
1. Create pod Spec with volume path of the vmdk and NodeSelector set to label assigned to node1.
2. Create pod and wait for pod to become ready.
3. Verify volume is attached to the node1.
4. Create empty file on the volume to verify volume is writable.
5. Verify newly created file and previously created files exist on the volume.
6. Delete pod.
7. Wait for volume to be detached from the node1.
8. Repeat Step 1 to 7 and make sure back to back pod creation on same worker node with the same volume is working as expected.
*/
It("should create and delete pod with the same volume source on the same worker node", func() {
var volumeFiles []string
pod := createPodWithVolumeAndNodeSelector(c, ns, node1Name, node1KeyValueLabel, volumePaths)
// Create empty files on the mounted volumes on the pod to verify volume is writable
// Verify newly and previously created files present on the volume mounted on the pod
newEmptyFileName := fmt.Sprintf("/mnt/volume1/%v_1.txt", ns)
volumeFiles = append(volumeFiles, newEmptyFileName)
createAndVerifyFilesOnVolume(ns, pod.Name, []string{newEmptyFileName}, volumeFiles)
deletePodAndWaitForVolumeToDetach(f, c, pod, node1Name, volumePaths)
By(fmt.Sprintf("Creating pod on the same node: %v", node1Name))
pod = createPodWithVolumeAndNodeSelector(c, ns, node1Name, node1KeyValueLabel, volumePaths)
// Create empty files on the mounted volumes on the pod to verify volume is writable
// Verify newly and previously created files present on the volume mounted on the pod
newEmptyFileName = fmt.Sprintf("/mnt/volume1/%v_2.txt", ns)
volumeFiles = append(volumeFiles, newEmptyFileName)
createAndVerifyFilesOnVolume(ns, pod.Name, []string{newEmptyFileName}, volumeFiles)
deletePodAndWaitForVolumeToDetach(f, c, pod, node1Name, volumePaths)
})
/*
Steps
1. Create pod Spec with volume path of the vmdk1 and NodeSelector set to node1's label.
2. Create pod and wait for POD to become ready.
3. Verify volume is attached to the node1.
4. Create empty file on the volume to verify volume is writable.
5. Verify newly created file and previously created files exist on the volume.
6. Delete pod.
7. Wait for volume to be detached from the node1.
8. Create pod Spec with volume path of the vmdk1 and NodeSelector set to node2's label.
9. Create pod and wait for pod to become ready.
10. Verify volume is attached to the node2.
11. Create empty file on the volume to verify volume is writable.
12. Verify newly created file and previously created files exist on the volume.
13. Delete pod.
*/
It("should create and delete pod with the same volume source attach/detach to different worker nodes", func() {
var volumeFiles []string
pod := createPodWithVolumeAndNodeSelector(c, ns, node1Name, node1KeyValueLabel, volumePaths)
// Create empty files on the mounted volumes on the pod to verify volume is writable
// Verify newly and previously created files present on the volume mounted on the pod
newEmptyFileName := fmt.Sprintf("/mnt/volume1/%v_1.txt", ns)
volumeFiles = append(volumeFiles, newEmptyFileName)
createAndVerifyFilesOnVolume(ns, pod.Name, []string{newEmptyFileName}, volumeFiles)
deletePodAndWaitForVolumeToDetach(f, c, pod, node1Name, volumePaths)
By(fmt.Sprintf("Creating pod on the another node: %v", node2Name))
pod = createPodWithVolumeAndNodeSelector(c, ns, node2Name, node2KeyValueLabel, volumePaths)
newEmptyFileName = fmt.Sprintf("/mnt/volume1/%v_2.txt", ns)
volumeFiles = append(volumeFiles, newEmptyFileName)
// Create empty files on the mounted volumes on the pod to verify volume is writable
// Verify newly and previously created files present on the volume mounted on the pod
createAndVerifyFilesOnVolume(ns, pod.Name, []string{newEmptyFileName}, volumeFiles)
deletePodAndWaitForVolumeToDetach(f, c, pod, node2Name, volumePaths)
})
/*
Test multiple volumes from same datastore within the same pod
1. Create volumes - vmdk2
2. Create pod Spec with volume path of vmdk1 (vmdk1 is created in test setup) and vmdk2.
3. Create pod using spec created in step-2 and wait for pod to become ready.
4. Verify both volumes are attached to the node on which pod are created. Write some data to make sure volume are accessible.
5. Delete pod.
6. Wait for vmdk1 and vmdk2 to be detached from node.
7. Create pod using spec created in step-2 and wait for pod to become ready.
8. Verify both volumes are attached to the node on which PODs are created. Verify volume contents are matching with the content written in step 4.
9. Delete POD.
10. Wait for vmdk1 and vmdk2 to be detached from node.
*/
It("should create and delete pod with multiple volumes from same datastore", func() {
By("creating another vmdk")
volumePath, err := vsp.CreateVolume(&VolumeOptions{}, nodeInfo.DataCenterRef)
Expect(err).NotTo(HaveOccurred())
volumePaths = append(volumePaths, volumePath)
By(fmt.Sprintf("Creating pod on the node: %v with volume: %v and volume: %v", node1Name, volumePaths[0], volumePaths[1]))
pod := createPodWithVolumeAndNodeSelector(c, ns, node1Name, node1KeyValueLabel, volumePaths)
// Create empty files on the mounted volumes on the pod to verify volume is writable
// Verify newly and previously created files present on the volume mounted on the pod
volumeFiles := []string{
fmt.Sprintf("/mnt/volume1/%v_1.txt", ns),
fmt.Sprintf("/mnt/volume2/%v_1.txt", ns),
}
createAndVerifyFilesOnVolume(ns, pod.Name, volumeFiles, volumeFiles)
deletePodAndWaitForVolumeToDetach(f, c, pod, node1Name, volumePaths)
By(fmt.Sprintf("Creating pod on the node: %v with volume :%v and volume: %v", node1Name, volumePaths[0], volumePaths[1]))
pod = createPodWithVolumeAndNodeSelector(c, ns, node1Name, node1KeyValueLabel, volumePaths)
// Create empty files on the mounted volumes on the pod to verify volume is writable
// Verify newly and previously created files present on the volume mounted on the pod
newEmptyFilesNames := []string{
fmt.Sprintf("/mnt/volume1/%v_2.txt", ns),
fmt.Sprintf("/mnt/volume2/%v_2.txt", ns),
}
volumeFiles = append(volumeFiles, newEmptyFilesNames[0])
volumeFiles = append(volumeFiles, newEmptyFilesNames[1])
createAndVerifyFilesOnVolume(ns, pod.Name, newEmptyFilesNames, volumeFiles)
})
/*
Test multiple volumes from different datastore within the same pod
1. Create volumes - vmdk2 on non default shared datastore.
2. Create pod Spec with volume path of vmdk1 (vmdk1 is created in test setup on default datastore) and vmdk2.
3. Create pod using spec created in step-2 and wait for pod to become ready.
4. Verify both volumes are attached to the node on which pod are created. Write some data to make sure volume are accessible.
5. Delete pod.
6. Wait for vmdk1 and vmdk2 to be detached from node.
7. Create pod using spec created in step-2 and wait for pod to become ready.
8. Verify both volumes are attached to the node on which PODs are created. Verify volume contents are matching with the content written in step 4.
9. Delete POD.
10. Wait for vmdk1 and vmdk2 to be detached from node.
*/
It("should create and delete pod with multiple volumes from different datastore", func() {
By("creating another vmdk on non default shared datastore")
var volumeOptions *VolumeOptions
volumeOptions = new(VolumeOptions)
volumeOptions.CapacityKB = 2097152
volumeOptions.Name = "e2e-vmdk-" + strconv.FormatInt(time.Now().UnixNano(), 10)
volumeOptions.Datastore = GetAndExpectStringEnvVar(SecondSharedDatastore)
volumePath, err := vsp.CreateVolume(volumeOptions, nodeInfo.DataCenterRef)
Expect(err).NotTo(HaveOccurred())
volumePaths = append(volumePaths, volumePath)
By(fmt.Sprintf("Creating pod on the node: %v with volume :%v and volume: %v", node1Name, volumePaths[0], volumePaths[1]))
pod := createPodWithVolumeAndNodeSelector(c, ns, node1Name, node1KeyValueLabel, volumePaths)
// Create empty files on the mounted volumes on the pod to verify volume is writable
// Verify newly and previously created files present on the volume mounted on the pod
volumeFiles := []string{
fmt.Sprintf("/mnt/volume1/%v_1.txt", ns),
fmt.Sprintf("/mnt/volume2/%v_1.txt", ns),
}
createAndVerifyFilesOnVolume(ns, pod.Name, volumeFiles, volumeFiles)
deletePodAndWaitForVolumeToDetach(f, c, pod, node1Name, volumePaths)
By(fmt.Sprintf("Creating pod on the node: %v with volume :%v and volume: %v", node1Name, volumePaths[0], volumePaths[1]))
pod = createPodWithVolumeAndNodeSelector(c, ns, node1Name, node1KeyValueLabel, volumePaths)
// Create empty files on the mounted volumes on the pod to verify volume is writable
// Verify newly and previously created files present on the volume mounted on the pod
newEmptyFileNames := []string{
fmt.Sprintf("/mnt/volume1/%v_2.txt", ns),
fmt.Sprintf("/mnt/volume2/%v_2.txt", ns),
}
volumeFiles = append(volumeFiles, newEmptyFileNames[0])
volumeFiles = append(volumeFiles, newEmptyFileNames[1])
createAndVerifyFilesOnVolume(ns, pod.Name, newEmptyFileNames, volumeFiles)
deletePodAndWaitForVolumeToDetach(f, c, pod, node1Name, volumePaths)
})
/*
Test Back-to-back pod creation/deletion with different volume sources on the same worker node
1. Create volumes - vmdk2
2. Create pod Spec - pod-SpecA with volume path of vmdk1 and NodeSelector set to label assigned to node1.
3. Create pod Spec - pod-SpecB with volume path of vmdk2 and NodeSelector set to label assigned to node1.
4. Create pod-A using pod-SpecA and wait for pod to become ready.
5. Create pod-B using pod-SpecB and wait for POD to become ready.
6. Verify volumes are attached to the node.
7. Create empty file on the volume to make sure volume is accessible. (Perform this step on pod-A and pod-B)
8. Verify file created in step 5 is present on the volume. (perform this step on pod-A and pod-B)
9. Delete pod-A and pod-B
10. Repeatedly (5 times) perform step 4 to 9 and verify associated volume's content is matching.
11. Wait for vmdk1 and vmdk2 to be detached from node.
*/
It("test back to back pod creation and deletion with different volume sources on the same worker node", func() {
var (
podA *v1.Pod
podB *v1.Pod
testvolumePathsPodA []string
testvolumePathsPodB []string
podAFiles []string
podBFiles []string
)
defer func() {
By("clean up undeleted pods")
framework.ExpectNoError(framework.DeletePodWithWait(f, c, podA), "defer: Failed to delete pod ", podA.Name)
framework.ExpectNoError(framework.DeletePodWithWait(f, c, podB), "defer: Failed to delete pod ", podB.Name)
By(fmt.Sprintf("wait for volumes to be detached from the node: %v", node1Name))
for _, volumePath := range volumePaths {
framework.ExpectNoError(waitForVSphereDiskToDetach(volumePath, node1Name))
}
}()
testvolumePathsPodA = append(testvolumePathsPodA, volumePaths[0])
// Create another VMDK Volume
By("creating another vmdk")
volumePath, err := vsp.CreateVolume(&VolumeOptions{}, nodeInfo.DataCenterRef)
Expect(err).NotTo(HaveOccurred())
volumePaths = append(volumePaths, volumePath)
testvolumePathsPodB = append(testvolumePathsPodA, volumePath)
for index := 0; index < 5; index++ {
By(fmt.Sprintf("Creating pod-A on the node: %v with volume: %v", node1Name, testvolumePathsPodA[0]))
podA = createPodWithVolumeAndNodeSelector(c, ns, node1Name, node1KeyValueLabel, testvolumePathsPodA)
By(fmt.Sprintf("Creating pod-B on the node: %v with volume: %v", node1Name, testvolumePathsPodB[0]))
podB = createPodWithVolumeAndNodeSelector(c, ns, node1Name, node1KeyValueLabel, testvolumePathsPodB)
podAFileName := fmt.Sprintf("/mnt/volume1/podA_%v_%v.txt", ns, index+1)
podBFileName := fmt.Sprintf("/mnt/volume1/podB_%v_%v.txt", ns, index+1)
podAFiles = append(podAFiles, podAFileName)
podBFiles = append(podBFiles, podBFileName)
// Create empty files on the mounted volumes on the pod to verify volume is writable
By("Creating empty file on volume mounted on pod-A")
framework.CreateEmptyFileOnPod(ns, podA.Name, podAFileName)
By("Creating empty file volume mounted on pod-B")
framework.CreateEmptyFileOnPod(ns, podB.Name, podBFileName)
// Verify newly and previously created files present on the volume mounted on the pod
By("Verify newly Created file and previously created files present on volume mounted on pod-A")
verifyFilesExistOnVSphereVolume(ns, podA.Name, podAFiles)
By("Verify newly Created file and previously created files present on volume mounted on pod-B")
verifyFilesExistOnVSphereVolume(ns, podB.Name, podBFiles)
By("Deleting pod-A")
framework.ExpectNoError(framework.DeletePodWithWait(f, c, podA), "Failed to delete pod ", podA.Name)
By("Deleting pod-B")
framework.ExpectNoError(framework.DeletePodWithWait(f, c, podB), "Failed to delete pod ", podB.Name)
}
})
})
func testSetupVolumePlacement(client clientset.Interface, namespace string) (node1Name string, node1KeyValueLabel map[string]string, node2Name string, node2KeyValueLabel map[string]string) {
nodes := framework.GetReadySchedulableNodesOrDie(client)
if len(nodes.Items) < 2 {
framework.Skipf("Requires at least %d nodes (not %d)", 2, len(nodes.Items))
}
node1Name = nodes.Items[0].Name
node2Name = nodes.Items[1].Name
node1LabelValue := "vsphere_e2e_" + string(uuid.NewUUID())
node1KeyValueLabel = make(map[string]string)
node1KeyValueLabel[NodeLabelKey] = node1LabelValue
framework.AddOrUpdateLabelOnNode(client, node1Name, NodeLabelKey, node1LabelValue)
node2LabelValue := "vsphere_e2e_" + string(uuid.NewUUID())
node2KeyValueLabel = make(map[string]string)
node2KeyValueLabel[NodeLabelKey] = node2LabelValue
framework.AddOrUpdateLabelOnNode(client, node2Name, NodeLabelKey, node2LabelValue)
return node1Name, node1KeyValueLabel, node2Name, node2KeyValueLabel
}
func createPodWithVolumeAndNodeSelector(client clientset.Interface, namespace string, nodeName string, nodeKeyValueLabel map[string]string, volumePaths []string) *v1.Pod {
var pod *v1.Pod
var err error
By(fmt.Sprintf("Creating pod on the node: %v", nodeName))
podspec := getVSpherePodSpecWithVolumePaths(volumePaths, nodeKeyValueLabel, nil)
pod, err = client.CoreV1().Pods(namespace).Create(podspec)
Expect(err).NotTo(HaveOccurred())
By("Waiting for pod to be ready")
Expect(framework.WaitForPodNameRunningInNamespace(client, pod.Name, namespace)).To(Succeed())
By(fmt.Sprintf("Verify volume is attached to the node:%v", nodeName))
for _, volumePath := range volumePaths {
isAttached, err := diskIsAttached(volumePath, nodeName)
Expect(err).NotTo(HaveOccurred())
Expect(isAttached).To(BeTrue(), "disk:"+volumePath+" is not attached with the node")
}
return pod
}
func createAndVerifyFilesOnVolume(namespace string, podname string, newEmptyfilesToCreate []string, filesToCheck []string) {
// Create empty files on the mounted volumes on the pod to verify volume is writable
By(fmt.Sprintf("Creating empty file on volume mounted on: %v", podname))
createEmptyFilesOnVSphereVolume(namespace, podname, newEmptyfilesToCreate)
// Verify newly and previously created files present on the volume mounted on the pod
By(fmt.Sprintf("Verify newly Created file and previously created files present on volume mounted on: %v", podname))
verifyFilesExistOnVSphereVolume(namespace, podname, filesToCheck)
}
func deletePodAndWaitForVolumeToDetach(f *framework.Framework, c clientset.Interface, pod *v1.Pod, nodeName string, volumePaths []string) {
By("Deleting pod")
framework.ExpectNoError(framework.DeletePodWithWait(f, c, pod), "Failed to delete pod ", pod.Name)
By("Waiting for volume to be detached from the node")
for _, volumePath := range volumePaths {
framework.ExpectNoError(waitForVSphereDiskToDetach(volumePath, nodeName))
}
}
此处可能存在不合适展示的内容,页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容,可点击提交进行申诉,我们将尽快为您处理。