|
| 1 | +/* |
| 2 | +Copyright 2024 The Kubeflow Authors. |
| 3 | +
|
| 4 | +Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | +you may not use this file except in compliance with the License. |
| 6 | +You may obtain a copy of the License at |
| 7 | +
|
| 8 | + http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | +
|
| 10 | +Unless required by applicable law or agreed to in writing, software |
| 11 | +distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | +See the License for the specific language governing permissions and |
| 14 | +limitations under the License. |
| 15 | +*/ |
| 16 | + |
| 17 | +package controllerv2 |
| 18 | + |
| 19 | +import ( |
| 20 | + kubeflowv2 "github.com/kubeflow/training-operator/pkg/apis/kubeflow.org/v2alpha1" |
| 21 | + testingutil "github.com/kubeflow/training-operator/pkg/util.v2/testing" |
| 22 | + "github.com/kubeflow/training-operator/test/integration/framework" |
| 23 | + "github.com/kubeflow/training-operator/test/util" |
| 24 | + "github.com/onsi/ginkgo/v2" |
| 25 | + "github.com/onsi/gomega" |
| 26 | + corev1 "k8s.io/api/core/v1" |
| 27 | + metav1 "k8s.io/apimachinery/pkg/apis/meta/v1" |
| 28 | + "k8s.io/utils/ptr" |
| 29 | + "sigs.k8s.io/controller-runtime/pkg/client" |
| 30 | +) |
| 31 | + |
| 32 | +var _ = ginkgo.Describe("TrainingRuntime marker validations and defaulting", ginkgo.Ordered, func() { |
| 33 | + var ns *corev1.Namespace |
| 34 | + |
| 35 | + ginkgo.BeforeAll(func() { |
| 36 | + fwk = &framework.Framework{} |
| 37 | + cfg = fwk.Init() |
| 38 | + ctx, k8sClient = fwk.RunManager(cfg) |
| 39 | + }) |
| 40 | + ginkgo.AfterAll(func() { |
| 41 | + fwk.Teardown() |
| 42 | + }) |
| 43 | + |
| 44 | + ginkgo.BeforeEach(func() { |
| 45 | + ns = &corev1.Namespace{ |
| 46 | + TypeMeta: metav1.TypeMeta{ |
| 47 | + APIVersion: corev1.SchemeGroupVersion.String(), |
| 48 | + Kind: "Namespace", |
| 49 | + }, |
| 50 | + ObjectMeta: metav1.ObjectMeta{ |
| 51 | + GenerateName: "training-runtime-marker-", |
| 52 | + }, |
| 53 | + } |
| 54 | + gomega.Expect(k8sClient.Create(ctx, ns)).To(gomega.Succeed()) |
| 55 | + }) |
| 56 | + ginkgo.AfterEach(func() { |
| 57 | + gomega.Expect(k8sClient.DeleteAllOf(ctx, &kubeflowv2.TrainingRuntime{}, client.InNamespace(ns.Name))).Should(gomega.Succeed()) |
| 58 | + gomega.Expect(k8sClient.DeleteAllOf(ctx, &kubeflowv2.ClusterTrainingRuntime{})).Should(gomega.Succeed()) |
| 59 | + }) |
| 60 | + |
| 61 | + ginkgo.When("Creating TrainingRuntime", func() { |
| 62 | + ginkgo.DescribeTable("Validate TrainingRuntime on creation", func(trainingRuntime func() *kubeflowv2.TrainingRuntime, errorMatcher gomega.OmegaMatcher) { |
| 63 | + gomega.Expect(k8sClient.Create(ctx, trainingRuntime())).Should(errorMatcher) |
| 64 | + }, |
| 65 | + ginkgo.Entry("Should succeed to create trainingRuntime", |
| 66 | + func() *kubeflowv2.TrainingRuntime { |
| 67 | + return testingutil.MakeTrainingRuntimeWrapper(ns.Name, "runtime"). |
| 68 | + Obj() |
| 69 | + }, |
| 70 | + gomega.Succeed()), |
| 71 | + ginkgo.Entry("Should succeed to create clusterTrainingRuntime", |
| 72 | + func() *kubeflowv2.TrainingRuntime { |
| 73 | + return testingutil.MakeTrainingRuntimeWrapper(ns.Name, "runtime"). |
| 74 | + Obj() |
| 75 | + }, |
| 76 | + gomega.Succeed()), |
| 77 | + ginkgo.Entry("Should fail to create trainingRuntime with both MPI and Torch runtimes", |
| 78 | + func() *kubeflowv2.TrainingRuntime { |
| 79 | + runtime := testingutil.MakeTrainingRuntimeWrapper(ns.Name, "runtime").Obj() |
| 80 | + runtime.Spec.MLPolicy = &kubeflowv2.MLPolicy{ |
| 81 | + MLPolicySource: kubeflowv2.MLPolicySource{ |
| 82 | + Torch: &kubeflowv2.TorchMLPolicySource{}, |
| 83 | + MPI: &kubeflowv2.MPIMLPolicySource{}, |
| 84 | + }, |
| 85 | + } |
| 86 | + return runtime |
| 87 | + }, |
| 88 | + testingutil.BeInvalidError()), |
| 89 | + ginkgo.Entry("Should fail to create trainingRuntime with minNodes and torch.elasticPolicy", |
| 90 | + func() *kubeflowv2.TrainingRuntime { |
| 91 | + runtime := testingutil.MakeTrainingRuntimeWrapper(ns.Name, "runtime").Obj() |
| 92 | + runtime.Spec.MLPolicy = &kubeflowv2.MLPolicy{ |
| 93 | + NumNodes: ptr.To(int32(2)), |
| 94 | + MLPolicySource: kubeflowv2.MLPolicySource{ |
| 95 | + Torch: &kubeflowv2.TorchMLPolicySource{ |
| 96 | + ElasticPolicy: &kubeflowv2.TorchElasticPolicy{}, |
| 97 | + }, |
| 98 | + }, |
| 99 | + } |
| 100 | + return runtime |
| 101 | + }, |
| 102 | + testingutil.BeInvalidError()), |
| 103 | + ) |
| 104 | + ginkgo.DescribeTable("Defaulting TrainingRuntime on creation", func(trainingRuntime func() *kubeflowv2.TrainingRuntime, wantTrainingRuntime func() *kubeflowv2.TrainingRuntime) { |
| 105 | + created := trainingRuntime() |
| 106 | + gomega.Expect(k8sClient.Create(ctx, created)).Should(gomega.Succeed()) |
| 107 | + gomega.Expect(created).Should(gomega.BeComparableTo(wantTrainingRuntime(), util.IgnoreObjectMetadata)) |
| 108 | + }, |
| 109 | + ginkgo.Entry("Should succeed to default TorchMLPolicySource.NumProcPerNode=auto", |
| 110 | + func() *kubeflowv2.TrainingRuntime { |
| 111 | + runtime := testingutil.MakeTrainingRuntimeWrapper(ns.Name, "runtime").Obj() |
| 112 | + runtime.Spec.MLPolicy = &kubeflowv2.MLPolicy{ |
| 113 | + MLPolicySource: kubeflowv2.MLPolicySource{ |
| 114 | + Torch: &kubeflowv2.TorchMLPolicySource{}, |
| 115 | + }, |
| 116 | + } |
| 117 | + return runtime |
| 118 | + }, |
| 119 | + func() *kubeflowv2.TrainingRuntime { |
| 120 | + runtime := testingutil.MakeTrainingRuntimeWrapper(ns.Name, "runtime").Obj() |
| 121 | + runtime.Spec.MLPolicy = &kubeflowv2.MLPolicy{ |
| 122 | + MLPolicySource: kubeflowv2.MLPolicySource{ |
| 123 | + Torch: &kubeflowv2.TorchMLPolicySource{ |
| 124 | + NumProcPerNode: ptr.To("auto"), |
| 125 | + }, |
| 126 | + }, |
| 127 | + } |
| 128 | + return runtime |
| 129 | + }), |
| 130 | + ) |
| 131 | + }) |
| 132 | +}) |
0 commit comments