accelerator.go 3.9 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134
  1. /*
  2. Copyright 2017 The Kubernetes Authors.
  3. Licensed under the Apache License, Version 2.0 (the "License");
  4. you may not use this file except in compliance with the License.
  5. You may obtain a copy of the License at
  6. http://www.apache.org/licenses/LICENSE-2.0
  7. Unless required by applicable law or agreed to in writing, software
  8. distributed under the License is distributed on an "AS IS" BASIS,
  9. WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  10. See the License for the specific language governing permissions and
  11. limitations under the License.
  12. */
  13. package monitoring
  14. import (
  15. "context"
  16. "os"
  17. "time"
  18. "github.com/onsi/ginkgo"
  19. "golang.org/x/oauth2/google"
  20. gcm "google.golang.org/api/monitoring/v3"
  21. v1 "k8s.io/api/core/v1"
  22. "k8s.io/apimachinery/pkg/api/resource"
  23. metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
  24. "k8s.io/apimachinery/pkg/util/wait"
  25. "k8s.io/kubernetes/test/e2e/framework"
  26. "k8s.io/kubernetes/test/e2e/framework/gpu"
  27. e2elog "k8s.io/kubernetes/test/e2e/framework/log"
  28. instrumentation "k8s.io/kubernetes/test/e2e/instrumentation/common"
  29. "k8s.io/kubernetes/test/e2e/scheduling"
  30. "k8s.io/kubernetes/test/utils/image"
  31. )
  32. // Stackdriver container accelerator metrics, as described here:
  33. // https://cloud.google.com/monitoring/api/metrics_gcp#gcp-container
  34. var acceleratorMetrics = []string{
  35. "accelerator/duty_cycle",
  36. "accelerator/memory_total",
  37. "accelerator/memory_used",
  38. }
  39. var _ = instrumentation.SIGDescribe("Stackdriver Monitoring", func() {
  40. ginkgo.BeforeEach(func() {
  41. framework.SkipUnlessProviderIs("gce", "gke")
  42. })
  43. f := framework.NewDefaultFramework("stackdriver-monitoring")
  44. ginkgo.It("should have accelerator metrics [Feature:StackdriverAcceleratorMonitoring]", func() {
  45. testStackdriverAcceleratorMonitoring(f)
  46. })
  47. })
  48. func testStackdriverAcceleratorMonitoring(f *framework.Framework) {
  49. projectID := framework.TestContext.CloudConfig.ProjectID
  50. ctx := context.Background()
  51. client, err := google.DefaultClient(ctx, gcm.CloudPlatformScope)
  52. gcmService, err := gcm.New(client)
  53. framework.ExpectNoError(err)
  54. // set this env var if accessing Stackdriver test endpoint (default is prod):
  55. // $ export STACKDRIVER_API_ENDPOINT_OVERRIDE=https://test-monitoring.sandbox.googleapis.com/
  56. basePathOverride := os.Getenv("STACKDRIVER_API_ENDPOINT_OVERRIDE")
  57. if basePathOverride != "" {
  58. gcmService.BasePath = basePathOverride
  59. }
  60. scheduling.SetupNVIDIAGPUNode(f, false)
  61. f.PodClient().Create(&v1.Pod{
  62. ObjectMeta: metav1.ObjectMeta{
  63. Name: rcName,
  64. },
  65. Spec: v1.PodSpec{
  66. RestartPolicy: v1.RestartPolicyNever,
  67. Containers: []v1.Container{
  68. {
  69. Name: rcName,
  70. Image: image.GetE2EImage(image.CudaVectorAdd),
  71. Command: []string{"/bin/sh", "-c"},
  72. Args: []string{"nvidia-smi && sleep infinity"},
  73. Resources: v1.ResourceRequirements{
  74. Limits: v1.ResourceList{
  75. gpu.NVIDIAGPUResourceName: *resource.NewQuantity(1, resource.DecimalSI),
  76. },
  77. },
  78. },
  79. },
  80. },
  81. })
  82. metricsMap := map[string]bool{}
  83. pollingFunction := checkForAcceleratorMetrics(projectID, gcmService, time.Now(), metricsMap)
  84. err = wait.Poll(pollFrequency, pollTimeout, pollingFunction)
  85. if err != nil {
  86. e2elog.Logf("Missing metrics: %+v", metricsMap)
  87. }
  88. framework.ExpectNoError(err)
  89. }
  90. func checkForAcceleratorMetrics(projectID string, gcmService *gcm.Service, start time.Time, metricsMap map[string]bool) func() (bool, error) {
  91. return func() (bool, error) {
  92. counter := 0
  93. for _, metric := range acceleratorMetrics {
  94. metricsMap[metric] = false
  95. }
  96. for _, metric := range acceleratorMetrics {
  97. // TODO: check only for metrics from this cluster
  98. ts, err := fetchTimeSeries(projectID, gcmService, metric, start, time.Now())
  99. framework.ExpectNoError(err)
  100. if len(ts) > 0 {
  101. counter = counter + 1
  102. metricsMap[metric] = true
  103. e2elog.Logf("Received %v timeseries for metric %v", len(ts), metric)
  104. } else {
  105. e2elog.Logf("No timeseries for metric %v", metric)
  106. }
  107. }
  108. if counter < 3 {
  109. return false, nil
  110. }
  111. return true, nil
  112. }
  113. }