pbacsko commented on code in PR #608: URL: https://github.com/apache/yunikorn-k8shim/pull/608#discussion_r1218202600
########## test/e2e/preemption/preemption_test.go: ########## @@ -0,0 +1,337 @@ +/* + Licensed to the Apache Software Foundation (ASF) under one + or more contributor license agreements. See the NOTICE file + distributed with this work for additional information + regarding copyright ownership. The ASF licenses this file + to you 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 preemption_test + +import ( + "fmt" + "strings" + "time" + + "github.com/onsi/ginkgo/v2" + "github.com/onsi/gomega" + v1 "k8s.io/api/core/v1" + k8serrors "k8s.io/apimachinery/pkg/api/errors" + metav1 "k8s.io/apimachinery/pkg/apis/meta/v1" + + "github.com/apache/yunikorn-k8shim/pkg/common/constants" + tests "github.com/apache/yunikorn-k8shim/test/e2e" + "github.com/apache/yunikorn-k8shim/test/e2e/framework/configmanager" + "github.com/apache/yunikorn-k8shim/test/e2e/framework/helpers/common" + "github.com/apache/yunikorn-k8shim/test/e2e/framework/helpers/k8s" + "github.com/apache/yunikorn-k8shim/test/e2e/framework/helpers/yunikorn" +) + +var kClient k8s.KubeCtl +var restClient yunikorn.RClient +var ns *v1.Namespace +var dev = "dev" + common.RandSeq(5) +var oldConfigMap = new(v1.ConfigMap) +var annotation = "ann-" + common.RandSeq(10) + +// Nodes +var Worker = "" +var WorkerMemRes int64 +var sleepPodMemLimit int64 +var taintKey = "e2e_test_preemption" +var nodesToTaint []string + +// Queues +var preemptionQueueYamlFormat = ` +partitions: + - name: default + queues: + - name: root + submitacl: '*' + queues: + - name: sandbox1 + submitacl: '*' + properties: + preemption.delay: 1s + resources: + guaranteed: + memory: %dM + - name: sandbox2 + submitacl: '*' + properties: + preemption.policy: %s + preemption.delay: 1s + resources: + guaranteed: + memory: %dM +` + +var _ = ginkgo.BeforeSuite(func() { + // Initializing kubectl client + kClient = k8s.KubeCtl{} + Ω(kClient.SetClient()).To(gomega.BeNil()) + // Initializing rest client + restClient = yunikorn.RClient{} + Ω(restClient).NotTo(gomega.BeNil()) + + annotation = "ann-" + common.RandSeq(10) + yunikorn.EnsureYuniKornConfigsPresent() + yunikorn.UpdateConfigMapWrapper(oldConfigMap, "", annotation) + + ginkgo.By("Port-forward the scheduler pod") + var err = kClient.PortForwardYkSchedulerPod() + Ω(err).NotTo(gomega.HaveOccurred()) + + ginkgo.By("create development namespace") + ns, err = kClient.CreateNamespace(dev, nil) + gomega.Ω(err).NotTo(gomega.HaveOccurred()) + gomega.Ω(ns.Status.Phase).To(gomega.Equal(v1.NamespaceActive)) + + var nodes *v1.NodeList + nodes, err = kClient.GetNodes() + Ω(err).NotTo(gomega.HaveOccurred()) + Ω(len(nodes.Items)).NotTo(gomega.BeZero(), "Nodes cant be empty") + + // Extract node allocatable resources + for _, node := range nodes.Items { + // skip master if it's marked as such + node := node + if k8s.IsMasterNode(&node) || !k8s.IsComputeNode(&node) { + continue + } + if Worker == "" { + Worker = node.Name + } else { + nodesToTaint = append(nodesToTaint, node.Name) + } + } + Ω(Worker).NotTo(gomega.BeEmpty(), "Worker node not found") + + ginkgo.By("Tainting some nodes..") + for _, nodeName := range nodesToTaint { + ginkgo.By("Tainting node " + nodeName) + err = kClient.TaintNode(nodeName, taintKey, "value", v1.TaintEffectNoSchedule) + Ω(err).NotTo(gomega.HaveOccurred()) + } + + nodesDAOInfo, err := restClient.GetNodes(constants.DefaultPartition) + Ω(err).NotTo(gomega.HaveOccurred()) + Ω(nodesDAOInfo).NotTo(gomega.BeNil()) + + for _, node := range *nodesDAOInfo { + if node.NodeID == Worker { + WorkerMemRes = node.Available["memory"] + } + } + sleepPodMemLimit = int64(float64(WorkerMemRes)/3) / (1000 * 1000) + Ω(sleepPodMemLimit).NotTo(gomega.BeZero(), "Sleep pod memory limit cannot be zero") +}) + +var _ = ginkgo.AfterSuite(func() { + + ginkgo.By("Untainting some nodes") + for _, nodeName := range nodesToTaint { + ginkgo.By("Untainting node " + nodeName) + err := kClient.UntaintNode(nodeName, taintKey) + Ω(err).NotTo(gomega.HaveOccurred(), "Could not remove taint from node "+nodeName) + } + + ginkgo.By("Check Yunikorn's health") + checks, err := yunikorn.GetFailedHealthChecks() + Ω(err).NotTo(gomega.HaveOccurred()) + Ω(checks).To(gomega.Equal(""), checks) + + testDescription := ginkgo.CurrentSpecReport() + if testDescription.Failed() { + tests.LogTestClusterInfoWrapper(testDescription.FailureMessage(), []string{ns.Name}) + tests.LogYunikornContainer(testDescription.FailureMessage()) + } + ginkgo.By("Tearing down namespace: " + ns.Name) + err = kClient.TearDownNamespace(ns.Name) + Ω(err).NotTo(gomega.HaveOccurred()) + + yunikorn.RestoreConfigMapWrapper(oldConfigMap, annotation) +}) + +var _ = ginkgo.Describe("Preemption", func() { + ginkgo.It("Verify_basic_preemption", func() { + ginkgo.By("A queue uses resource more than the guaranteed value even after removing one of the pods. The cluster doesn't have enough resource to deploy a pod in another queue which uses resource less than the guaranteed value.") + // update config + cm, err := kClient.GetConfigMap(constants.ConfigMapName, configmanager.YuniKornTestConfig.YkNamespace) + Ω(err).NotTo(HaveOccurred()) + cm.Data[configmanager.DefaultPolicyGroup] = fmt.Sprintf(preemptionQueueYamlFormat, sleepPodMemLimit, "default", sleepPodMemLimit) + _, err = kClient.UpdateConfigMap(cm, configmanager.YuniKornTestConfig.YkNamespace) + Ω(err).NotTo(HaveOccurred()) + + // Define sleepPod + sleepPod1Configs := k8s.SleepPodConfig{Name: "sleepjob1", NS: dev, Mem: sleepPodMemLimit, Time: 600, Optedout: true, Labels: map[string]string{"queue": "root.sandbox1"}} + sleepPod2Configs := k8s.SleepPodConfig{Name: "sleepjob2", NS: dev, Mem: sleepPodMemLimit, Time: 600, Optedout: true, Labels: map[string]string{"queue": "root.sandbox1"}} + sleepPod3Configs := k8s.SleepPodConfig{Name: "sleepjob3", NS: dev, Mem: sleepPodMemLimit, Time: 600, Optedout: true, Labels: map[string]string{"queue": "root.sandbox1"}} + sleepPod4Configs := k8s.SleepPodConfig{Name: "sleepjob4", NS: dev, Mem: sleepPodMemLimit, Time: 600, Optedout: true, Labels: map[string]string{"queue": "root.sandbox2"}} + + for _, config := range []k8s.SleepPodConfig{sleepPod1Configs, sleepPod2Configs, sleepPod3Configs, sleepPod4Configs} { + ginkgo.By("Deploy the sleep pod " + config.Name + " to the development namespace") + sleepObj, podErr := k8s.InitSleepPod(config) + Ω(podErr).NotTo(gomega.HaveOccurred()) + sleepRespPod, podErr := kClient.CreatePod(sleepObj, dev) + gomega.Ω(podErr).NotTo(gomega.HaveOccurred()) + + // Wait for pod to move to running state + podErr = kClient.WaitForPodBySelectorRunning(dev, + fmt.Sprintf("app=%s", sleepRespPod.ObjectMeta.Labels["app"]), + 60) + gomega.Ω(podErr).NotTo(gomega.HaveOccurred()) + } + + // assert one of the pods in root.sandbox1 is preempted + ginkgo.By("One of the pods in root.sanbox1 is preempted") + sandbox1RunningPodsCnt := 0 + pods, err := kClient.ListPodsByLabelSelector(dev, "queue=root.sandbox1") + gomega.Ω(err).NotTo(gomega.HaveOccurred()) + for _, pod := range pods.Items { + if pod.DeletionTimestamp != nil { + continue + } + if pod.Status.Phase == v1.PodRunning { + sandbox1RunningPodsCnt++ + } + } + Ω(sandbox1RunningPodsCnt).To(gomega.Equal(2), "One of the pods in root.sandbox1 should be preempted") + }) + + ginkgo.It("Verify_no_preemption_on_resources_less_than_guaranteed_value", func() { + ginkgo.By("A queue uses resource less than the guaranteed value can't be preempted.") + // update config + cm, err := kClient.GetConfigMap(constants.ConfigMapName, configmanager.YuniKornTestConfig.YkNamespace) + Ω(err).NotTo(HaveOccurred()) + cm.Data[configmanager.DefaultPolicyGroup] = fmt.Sprintf(preemptionQueueYamlFormat, WorkerMemRes, "default", WorkerMemRes) + _, err = kClient.UpdateConfigMap(cm, configmanager.YuniKornTestConfig.YkNamespace) + Ω(err).NotTo(HaveOccurred()) + + // Define sleepPod + sleepPod1Configs := k8s.SleepPodConfig{Name: "sleepjob1", NS: dev, Mem: sleepPodMemLimit, Time: 30, Optedout: true, Labels: map[string]string{"queue": "root.sandbox1"}} + sleepPod2Configs := k8s.SleepPodConfig{Name: "sleepjob2", NS: dev, Mem: sleepPodMemLimit, Time: 30, Optedout: true, Labels: map[string]string{"queue": "root.sandbox1"}} + sleepPod3Configs := k8s.SleepPodConfig{Name: "sleepjob3", NS: dev, Mem: sleepPodMemLimit, Time: 30, Optedout: true, Labels: map[string]string{"queue": "root.sandbox1"}} + sleepPod4Configs := k8s.SleepPodConfig{Name: "sleepjob4", NS: dev, Mem: sleepPodMemLimit, Time: 30, Optedout: true, Labels: map[string]string{"queue": "root.sandbox2"}} + + // Deploy pods in root.sandbox1 + for _, config := range []k8s.SleepPodConfig{sleepPod1Configs, sleepPod2Configs, sleepPod3Configs} { + ginkgo.By("Deploy the sleep pod " + config.Name + " to the development namespace") + sleepObj, podErr := k8s.InitSleepPod(config) + Ω(podErr).NotTo(gomega.HaveOccurred()) + sleepRespPod, podErr := kClient.CreatePod(sleepObj, dev) + gomega.Ω(podErr).NotTo(gomega.HaveOccurred()) + + // Wait for pod to move to running state + podErr = kClient.WaitForPodBySelectorRunning(dev, + fmt.Sprintf("app=%s", sleepRespPod.ObjectMeta.Labels["app"]), + 30) + gomega.Ω(podErr).NotTo(gomega.HaveOccurred()) + } + + // Deploy sleepjob4 pod in root.sandbox2 + ginkgo.By("Deploy the sleep pod " + sleepPod4Configs.Name + " to the development namespace") + sleepObj, podErr := k8s.InitSleepPod(sleepPod4Configs) + Ω(podErr).NotTo(gomega.HaveOccurred()) + sleepRespPod4, err := kClient.CreatePod(sleepObj, dev) + gomega.Ω(err).NotTo(gomega.HaveOccurred()) + + // sleepjob4 pod can't be scheduled before pods in root.sandbox1 are succeeded + ginkgo.By("The sleep pod " + sleepPod4Configs.Name + " can't be scheduled") + err = kClient.WaitForPodUnschedulable(sleepRespPod4, time.Duration(60)*time.Second) + gomega.Ω(err).NotTo(gomega.HaveOccurred()) + + // pods in root.sandbox1 can be succeeded + ginkgo.By("The pods in root.sandbox1 can be succeeded") + for _, config := range []k8s.SleepPodConfig{sleepPod1Configs, sleepPod2Configs, sleepPod3Configs} { + err = kClient.WaitForPodScheduled(dev, config.Name, time.Duration(30)*time.Second) + gomega.Ω(err).NotTo(gomega.HaveOccurred()) + } + }) + + ginkgo.It("Verify_no_preemption_outside_fence", func() { + ginkgo.By("The preemption can't go outside the fence.") + // update config + cm, err := kClient.GetConfigMap(constants.ConfigMapName, configmanager.YuniKornTestConfig.YkNamespace) + Ω(err).NotTo(HaveOccurred()) + cm.Data[configmanager.DefaultPolicyGroup] = fmt.Sprintf(preemptionQueueYamlFormat, sleepPodMemLimit, "fence", sleepPodMemLimit) + _, err = kClient.UpdateConfigMap(cm, configmanager.YuniKornTestConfig.YkNamespace) + Ω(err).NotTo(HaveOccurred()) + + // Define sleepPod + sleepPod1Configs := k8s.SleepPodConfig{Name: "sleepjob1", NS: dev, Mem: sleepPodMemLimit, Time: 30, Optedout: true, Labels: map[string]string{"queue": "root.sandbox1"}} + sleepPod2Configs := k8s.SleepPodConfig{Name: "sleepjob2", NS: dev, Mem: sleepPodMemLimit, Time: 30, Optedout: true, Labels: map[string]string{"queue": "root.sandbox1"}} + sleepPod3Configs := k8s.SleepPodConfig{Name: "sleepjob3", NS: dev, Mem: sleepPodMemLimit, Time: 30, Optedout: true, Labels: map[string]string{"queue": "root.sandbox1"}} + sleepPod4Configs := k8s.SleepPodConfig{Name: "sleepjob4", NS: dev, Mem: sleepPodMemLimit, Time: 30, Optedout: true, Labels: map[string]string{"queue": "root.sandbox2"}} + + // Deploy pods in root.sandbox1 + for _, config := range []k8s.SleepPodConfig{sleepPod1Configs, sleepPod2Configs, sleepPod3Configs} { + ginkgo.By("Deploy the sleep pod " + config.Name + " to the development namespace") + sleepObj, podErr := k8s.InitSleepPod(config) + Ω(podErr).NotTo(gomega.HaveOccurred()) + sleepRespPod, podErr := kClient.CreatePod(sleepObj, dev) + gomega.Ω(podErr).NotTo(gomega.HaveOccurred()) + + // Wait for pod to move to running state + podErr = kClient.WaitForPodBySelectorRunning(dev, + fmt.Sprintf("app=%s", sleepRespPod.ObjectMeta.Labels["app"]), + 30) + gomega.Ω(podErr).NotTo(gomega.HaveOccurred()) + } + + // Deploy sleepjob4 pod in root.sandbox2 + ginkgo.By("Deploy the sleep pod " + sleepPod4Configs.Name + " to the development namespace") + sleepObj, podErr := k8s.InitSleepPod(sleepPod4Configs) + Ω(podErr).NotTo(gomega.HaveOccurred()) + sleepRespPod4, err := kClient.CreatePod(sleepObj, dev) + gomega.Ω(err).NotTo(gomega.HaveOccurred()) + + // sleepjob4 pod can't be scheduled before pods in root.sandbox1 are succeeded + ginkgo.By("The sleep pod " + sleepPod4Configs.Name + " can't be scheduled") + err = kClient.WaitForPodUnschedulable(sleepRespPod4, time.Duration(60)*time.Second) Review Comment: Nit: `60 * time.Second` should work too -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
