vendor update for CSI 0.3.0

This commit is contained in:
gman
2018-07-18 16:47:22 +02:00
parent 6f484f92fc
commit 8ea659f0d5
6810 changed files with 438061 additions and 193861 deletions

View File

@ -11,7 +11,7 @@ go_library(
"autoscaling_timer.go",
"cluster_autoscaler_scalability.go",
"cluster_size_autoscaling.go",
"custom_metrics_autoscaling.go",
"custom_metrics_stackdriver_autoscaling.go",
"dns_autoscaling.go",
"framework.go",
"horizontal_pod_autoscaling.go",
@ -24,6 +24,7 @@ go_library(
"//test/e2e/instrumentation/monitoring:go_default_library",
"//test/e2e/scheduling:go_default_library",
"//test/utils:go_default_library",
"//test/utils/image:go_default_library",
"//vendor/github.com/golang/glog:go_default_library",
"//vendor/github.com/onsi/ginkgo:go_default_library",
"//vendor/github.com/onsi/gomega:go_default_library",
@ -33,7 +34,7 @@ go_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/scheduling/v1alpha1:go_default_library",
"//vendor/k8s.io/api/scheduling/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",

View File

@ -93,7 +93,7 @@ var _ = SIGDescribe("[Feature:ClusterSizeAutoscalingScaleUp] [Slow] Autoscaling"
nodeMemoryMB := (&nodeMemoryBytes).Value() / 1024 / 1024
memRequestMB := nodeMemoryMB / 10 // Ensure each pod takes not more than 10% of node's allocatable memory.
replicas := 1
resourceConsumer := common.NewDynamicResourceConsumer("resource-consumer", f.Namespace.Name, common.KindDeployment, replicas, 0, 0, 0, cpuRequestMillis, memRequestMB, f.ClientSet, f.InternalClientset)
resourceConsumer := common.NewDynamicResourceConsumer("resource-consumer", f.Namespace.Name, common.KindDeployment, replicas, 0, 0, 0, cpuRequestMillis, memRequestMB, f.ClientSet, f.InternalClientset, f.ScalesGetter)
defer resourceConsumer.CleanUp()
resourceConsumer.WaitForReplicas(replicas, 1*time.Minute) // Should finish ~immediately, so 1 minute is more than enough.

View File

@ -31,6 +31,7 @@ import (
clientset "k8s.io/client-go/kubernetes"
"k8s.io/kubernetes/test/e2e/framework"
testutils "k8s.io/kubernetes/test/utils"
imageutils "k8s.io/kubernetes/test/utils/image"
"github.com/golang/glog"
. "github.com/onsi/ginkgo"
@ -347,7 +348,7 @@ var _ = framework.KubeDescribe("Cluster size autoscaler scalability [Slow]", fun
timeToWait := 5 * time.Minute
podsConfig := reserveMemoryRCConfig(f, "unschedulable-pod", unschedulablePodReplicas, totalMemReservation, timeToWait)
framework.RunRC(*podsConfig) // Ignore error (it will occur because pods are unschedulable)
defer framework.DeleteRCAndPods(f.ClientSet, f.InternalClientset, f.Namespace.Name, podsConfig.Name)
defer framework.DeleteRCAndWaitForGC(f.ClientSet, f.Namespace.Name, podsConfig.Name)
// Ensure that no new nodes have been added so far.
Expect(framework.NumberOfReadyNodes(f.ClientSet)).To(Equal(nodeCount))
@ -417,7 +418,7 @@ func simpleScaleUpTestWithTolerance(f *framework.Framework, config *scaleUpTestC
}
timeTrack(start, fmt.Sprintf("Scale up to %v", config.expectedResult.nodes))
return func() error {
return framework.DeleteRCAndPods(f.ClientSet, f.InternalClientset, f.Namespace.Name, config.extraPods.Name)
return framework.DeleteRCAndWaitForGC(f.ClientSet, f.Namespace.Name, config.extraPods.Name)
}
}
@ -432,7 +433,7 @@ func reserveMemoryRCConfig(f *framework.Framework, id string, replicas, megabyte
Name: id,
Namespace: f.Namespace.Name,
Timeout: timeout,
Image: framework.GetPauseImageName(f.ClientSet),
Image: imageutils.GetPauseImageName(),
Replicas: replicas,
MemRequest: int64(1024 * 1024 * megabytes / replicas),
}
@ -492,7 +493,7 @@ func createHostPortPodsWithMemory(f *framework.Framework, id string, replicas, p
Name: id,
Namespace: f.Namespace.Name,
Timeout: timeout,
Image: framework.GetPauseImageName(f.ClientSet),
Image: imageutils.GetPauseImageName(),
Replicas: replicas,
HostPorts: map[string]int{"port1": port},
MemRequest: request,
@ -500,7 +501,7 @@ func createHostPortPodsWithMemory(f *framework.Framework, id string, replicas, p
err := framework.RunRC(*config)
framework.ExpectNoError(err)
return func() error {
return framework.DeleteRCAndPods(f.ClientSet, f.InternalClientset, f.Namespace.Name, id)
return framework.DeleteRCAndWaitForGC(f.ClientSet, f.Namespace.Name, id)
}
}
@ -540,7 +541,7 @@ func distributeLoad(f *framework.Framework, namespace string, id string, podDist
framework.ExpectNoError(framework.RunRC(*rcConfig))
framework.ExpectNoError(waitForAllCaPodsReadyInNamespace(f, f.ClientSet))
return func() error {
return framework.DeleteRCAndPods(f.ClientSet, f.InternalClientset, f.Namespace.Name, id)
return framework.DeleteRCAndWaitForGC(f.ClientSet, f.Namespace.Name, id)
}
}

View File

@ -30,7 +30,7 @@ import (
"k8s.io/api/core/v1"
policy "k8s.io/api/policy/v1beta1"
"k8s.io/api/scheduling/v1alpha1"
schedulerapi "k8s.io/api/scheduling/v1beta1"
"k8s.io/apimachinery/pkg/api/errors"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/fields"
@ -45,6 +45,7 @@ import (
"k8s.io/kubernetes/test/e2e/framework"
"k8s.io/kubernetes/test/e2e/scheduling"
testutils "k8s.io/kubernetes/test/utils"
imageutils "k8s.io/kubernetes/test/utils/image"
"github.com/golang/glog"
. "github.com/onsi/ginkgo"
@ -168,7 +169,7 @@ var _ = SIGDescribe("Cluster size autoscaling [Slow]", func() {
It("shouldn't increase cluster size if pending pod is too large [Feature:ClusterSizeAutoscalingScaleUp]", func() {
By("Creating unschedulable pod")
ReserveMemory(f, "memory-reservation", 1, int(1.1*float64(memAllocatableMb)), false, defaultTimeout)
defer framework.DeleteRCAndPods(f.ClientSet, f.InternalClientset, f.Namespace.Name, "memory-reservation")
defer framework.DeleteRCAndWaitForGC(f.ClientSet, f.Namespace.Name, "memory-reservation")
By("Waiting for scale up hoping it won't happen")
// Verify that the appropriate event was generated
@ -195,7 +196,7 @@ var _ = SIGDescribe("Cluster size autoscaling [Slow]", func() {
simpleScaleUpTest := func(unready int) {
ReserveMemory(f, "memory-reservation", 100, nodeCount*memAllocatableMb, false, 1*time.Second)
defer framework.DeleteRCAndPods(f.ClientSet, f.InternalClientset, f.Namespace.Name, "memory-reservation")
defer framework.DeleteRCAndWaitForGC(f.ClientSet, f.Namespace.Name, "memory-reservation")
// Verify that cluster size is increased
framework.ExpectNoError(WaitForClusterSizeFuncWithUnready(f.ClientSet,
@ -206,6 +207,108 @@ var _ = SIGDescribe("Cluster size autoscaling [Slow]", func() {
It("should increase cluster size if pending pods are small [Feature:ClusterSizeAutoscalingScaleUp]",
func() { simpleScaleUpTest(0) })
supportedGpuTypes := []string{"nvidia-tesla-k80", "nvidia-tesla-v100", "nvidia-tesla-p100"}
for _, gpuType := range supportedGpuTypes {
gpuType := gpuType // create new variable for each iteration step
It(fmt.Sprintf("Should scale up GPU pool from 0 [GpuType:%s] [Feature:ClusterSizeAutoscalingGpu]", gpuType), func() {
framework.SkipUnlessProviderIs("gke")
const gpuPoolName = "gpu-pool"
addGpuNodePool(gpuPoolName, gpuType, 1, 0)
defer deleteNodePool(gpuPoolName)
installNvidiaDriversDaemonSet()
By("Enable autoscaler")
framework.ExpectNoError(enableAutoscaler(gpuPoolName, 0, 1))
defer disableAutoscaler(gpuPoolName, 0, 1)
Expect(len(getPoolNodes(f, gpuPoolName))).Should(Equal(0))
By("Schedule a pod which requires GPU")
framework.ExpectNoError(scheduleGpuPod(f, "gpu-pod-rc"))
framework.ExpectNoError(WaitForClusterSizeFunc(f.ClientSet,
func(size int) bool { return size == nodeCount+1 }, scaleUpTimeout))
Expect(len(getPoolNodes(f, gpuPoolName))).Should(Equal(1))
})
It(fmt.Sprintf("Should scale up GPU pool from 1 [GpuType:%s] [Feature:ClusterSizeAutoscalingGpu]", gpuType), func() {
framework.SkipUnlessProviderIs("gke")
const gpuPoolName = "gpu-pool"
addGpuNodePool(gpuPoolName, gpuType, 1, 1)
defer deleteNodePool(gpuPoolName)
installNvidiaDriversDaemonSet()
By("Schedule a single pod which requires GPU")
framework.ExpectNoError(scheduleGpuPod(f, "gpu-pod-rc"))
By("Enable autoscaler")
framework.ExpectNoError(enableAutoscaler(gpuPoolName, 0, 2))
defer disableAutoscaler(gpuPoolName, 0, 2)
Expect(len(getPoolNodes(f, gpuPoolName))).Should(Equal(1))
By("Scale GPU deployment")
framework.ScaleRC(f.ClientSet, f.ScalesGetter, f.Namespace.Name, "gpu-pod-rc", 2, true)
framework.ExpectNoError(WaitForClusterSizeFunc(f.ClientSet,
func(size int) bool { return size == nodeCount+2 }, scaleUpTimeout))
Expect(len(getPoolNodes(f, gpuPoolName))).Should(Equal(2))
})
It(fmt.Sprintf("Should not scale GPU pool up if pod does not require GPUs [GpuType:%s] [Feature:ClusterSizeAutoscalingGpu]", gpuType), func() {
framework.SkipUnlessProviderIs("gke")
const gpuPoolName = "gpu-pool"
addGpuNodePool(gpuPoolName, gpuType, 1, 0)
defer deleteNodePool(gpuPoolName)
installNvidiaDriversDaemonSet()
By("Enable autoscaler")
framework.ExpectNoError(enableAutoscaler(gpuPoolName, 0, 1))
defer disableAutoscaler(gpuPoolName, 0, 1)
Expect(len(getPoolNodes(f, gpuPoolName))).Should(Equal(0))
By("Schedule bunch of pods beyond point of filling default pool but do not request any GPUs")
ReserveMemory(f, "memory-reservation", 100, nodeCount*memAllocatableMb, false, 1*time.Second)
defer framework.DeleteRCAndWaitForGC(f.ClientSet, f.Namespace.Name, "memory-reservation")
// Verify that cluster size is increased
framework.ExpectNoError(WaitForClusterSizeFunc(f.ClientSet,
func(size int) bool { return size >= nodeCount+1 }, scaleUpTimeout))
// Expect gpu pool to stay intact
Expect(len(getPoolNodes(f, gpuPoolName))).Should(Equal(0))
})
It(fmt.Sprintf("Should scale down GPU pool from 1 [GpuType:%s] [Feature:ClusterSizeAutoscalingGpu]", gpuType), func() {
framework.SkipUnlessProviderIs("gke")
const gpuPoolName = "gpu-pool"
addGpuNodePool(gpuPoolName, gpuType, 1, 1)
defer deleteNodePool(gpuPoolName)
installNvidiaDriversDaemonSet()
By("Schedule a single pod which requires GPU")
framework.ExpectNoError(scheduleGpuPod(f, "gpu-pod-rc"))
By("Enable autoscaler")
framework.ExpectNoError(enableAutoscaler(gpuPoolName, 0, 1))
defer disableAutoscaler(gpuPoolName, 0, 1)
Expect(len(getPoolNodes(f, gpuPoolName))).Should(Equal(1))
By("Remove the only POD requiring GPU")
framework.DeleteRCAndWaitForGC(f.ClientSet, f.Namespace.Name, "gpu-pod-rc")
framework.ExpectNoError(WaitForClusterSizeFunc(f.ClientSet,
func(size int) bool { return size == nodeCount }, scaleDownTimeout))
Expect(len(getPoolNodes(f, gpuPoolName))).Should(Equal(0))
})
}
It("should increase cluster size if pending pods are small and one node is broken [Feature:ClusterSizeAutoscalingScaleUp]",
func() {
framework.TestUnderTemporaryNetworkFailure(c, "default", getAnyNode(c), func() { simpleScaleUpTest(1) })
@ -222,7 +325,7 @@ var _ = SIGDescribe("Cluster size autoscaling [Slow]", func() {
By("Schedule more pods than can fit and wait for cluster to scale-up")
ReserveMemory(f, "memory-reservation", 100, nodeCount*memAllocatableMb, false, 1*time.Second)
defer framework.DeleteRCAndPods(f.ClientSet, f.InternalClientset, f.Namespace.Name, "memory-reservation")
defer framework.DeleteRCAndWaitForGC(f.ClientSet, f.Namespace.Name, "memory-reservation")
status, err = waitForScaleUpStatus(c, func(s *scaleUpStatus) bool {
return s.status == caOngoingScaleUpStatus
@ -265,8 +368,8 @@ var _ = SIGDescribe("Cluster size autoscaling [Slow]", func() {
By("Reserving 0.1x more memory than the cluster holds to trigger scale up")
totalMemoryReservation := int(1.1 * float64(nodeCount*memAllocatableMb+extraMemMb))
defer framework.DeleteRCAndWaitForGC(f.ClientSet, f.Namespace.Name, "memory-reservation")
ReserveMemory(f, "memory-reservation", 100, totalMemoryReservation, false, defaultTimeout)
defer framework.DeleteRCAndPods(f.ClientSet, f.InternalClientset, f.Namespace.Name, "memory-reservation")
// Verify, that cluster size is increased
framework.ExpectNoError(WaitForClusterSizeFunc(f.ClientSet,
@ -289,7 +392,7 @@ var _ = SIGDescribe("Cluster size autoscaling [Slow]", func() {
It("should increase cluster size if pods are pending due to host port conflict [Feature:ClusterSizeAutoscalingScaleUp]", func() {
scheduling.CreateHostPortPods(f, "host-port", nodeCount+2, false)
defer framework.DeleteRCAndPods(f.ClientSet, f.InternalClientset, f.Namespace.Name, "host-port")
defer framework.DeleteRCAndWaitForGC(f.ClientSet, f.Namespace.Name, "host-port")
framework.ExpectNoError(WaitForClusterSizeFunc(f.ClientSet,
func(size int) bool { return size >= nodeCount+2 }, scaleUpTimeout))
@ -304,12 +407,12 @@ var _ = SIGDescribe("Cluster size autoscaling [Slow]", func() {
}
By("starting a pod with anti-affinity on each node")
framework.ExpectNoError(runAntiAffinityPods(f, f.Namespace.Name, pods, "some-pod", labels, labels))
defer framework.DeleteRCAndPods(f.ClientSet, f.InternalClientset, f.Namespace.Name, "some-pod")
defer framework.DeleteRCAndWaitForGC(f.ClientSet, f.Namespace.Name, "some-pod")
framework.ExpectNoError(waitForAllCaPodsReadyInNamespace(f, c))
By("scheduling extra pods with anti-affinity to existing ones")
framework.ExpectNoError(runAntiAffinityPods(f, f.Namespace.Name, newPods, "extra-pod", labels, labels))
defer framework.DeleteRCAndPods(f.ClientSet, f.InternalClientset, f.Namespace.Name, "extra-pod")
defer framework.DeleteRCAndWaitForGC(f.ClientSet, f.Namespace.Name, "extra-pod")
framework.ExpectNoError(waitForAllCaPodsReadyInNamespace(f, c))
framework.ExpectNoError(framework.WaitForReadyNodes(c, nodeCount+newPods, scaleUpTimeout))
@ -323,14 +426,14 @@ var _ = SIGDescribe("Cluster size autoscaling [Slow]", func() {
"anti-affinity": "yes",
}
framework.ExpectNoError(runAntiAffinityPods(f, f.Namespace.Name, pods, "some-pod", labels, labels))
defer framework.DeleteRCAndPods(f.ClientSet, f.InternalClientset, f.Namespace.Name, "some-pod")
defer framework.DeleteRCAndWaitForGC(f.ClientSet, f.Namespace.Name, "some-pod")
By("waiting for all pods before triggering scale up")
framework.ExpectNoError(waitForAllCaPodsReadyInNamespace(f, c))
By("creating a pod requesting EmptyDir")
framework.ExpectNoError(runVolumeAntiAffinityPods(f, f.Namespace.Name, newPods, "extra-pod", labels, labels, emptyDirVolumes))
defer framework.DeleteRCAndPods(f.ClientSet, f.InternalClientset, f.Namespace.Name, "extra-pod")
defer framework.DeleteRCAndWaitForGC(f.ClientSet, f.Namespace.Name, "extra-pod")
framework.ExpectNoError(waitForAllCaPodsReadyInNamespace(f, c))
framework.ExpectNoError(framework.WaitForReadyNodes(c, nodeCount+newPods, scaleUpTimeout))
@ -359,11 +462,10 @@ var _ = SIGDescribe("Cluster size autoscaling [Slow]", func() {
},
Prebind: nil,
}
emptyStorageClass := ""
pvcConfig := framework.PersistentVolumeClaimConfig{
Annotations: map[string]string{
v1.BetaStorageClassAnnotation: "",
},
Selector: selector,
Selector: selector,
StorageClassName: &emptyStorageClass,
}
pv, pvc, err := framework.CreatePVPVC(c, pvConfig, pvcConfig, f.Namespace.Name, false)
@ -388,7 +490,7 @@ var _ = SIGDescribe("Cluster size autoscaling [Slow]", func() {
}
framework.ExpectNoError(runAntiAffinityPods(f, f.Namespace.Name, pods, "some-pod", labels, labels))
defer func() {
framework.DeleteRCAndPods(f.ClientSet, f.InternalClientset, f.Namespace.Name, "some-pod")
framework.DeleteRCAndWaitForGC(f.ClientSet, f.Namespace.Name, "some-pod")
glog.Infof("RC and pods not using volume deleted")
}()
@ -401,7 +503,7 @@ var _ = SIGDescribe("Cluster size autoscaling [Slow]", func() {
volumes := buildVolumes(pv, pvc)
framework.ExpectNoError(runVolumeAntiAffinityPods(f, f.Namespace.Name, newPods, pvcPodName, labels, labels, volumes))
defer func() {
framework.DeleteRCAndPods(f.ClientSet, f.InternalClientset, f.Namespace.Name, pvcPodName)
framework.DeleteRCAndWaitForGC(f.ClientSet, f.Namespace.Name, pvcPodName)
framework.ExpectNoError(waitForAllCaPodsReadyInNamespace(f, c))
}()
@ -506,7 +608,7 @@ var _ = SIGDescribe("Cluster size autoscaling [Slow]", func() {
defer removeLabels(registeredNodes)
framework.ExpectNoError(waitForAllCaPodsReadyInNamespace(f, c))
framework.ExpectNoError(framework.DeleteRCAndPods(f.ClientSet, f.InternalClientset, f.Namespace.Name, "node-selector"))
framework.ExpectNoError(framework.DeleteRCAndWaitForGC(f.ClientSet, f.Namespace.Name, "node-selector"))
})
It("should scale up correct target pool [Feature:ClusterSizeAutoscalingScaleUp]", func() {
@ -524,8 +626,8 @@ var _ = SIGDescribe("Cluster size autoscaling [Slow]", func() {
extraPods := extraNodes + 1
totalMemoryReservation := int(float64(extraPods) * 1.5 * float64(memAllocatableMb))
By(fmt.Sprintf("Creating rc with %v pods too big to fit default-pool but fitting extra-pool", extraPods))
defer framework.DeleteRCAndWaitForGC(f.ClientSet, f.Namespace.Name, "memory-reservation")
ReserveMemory(f, "memory-reservation", extraPods, totalMemoryReservation, false, defaultTimeout)
defer framework.DeleteRCAndPods(f.ClientSet, f.InternalClientset, f.Namespace.Name, "memory-reservation")
// Apparently GKE master is restarted couple minutes after the node pool is added
// reseting all the timers in scale down code. Adding 5 extra minutes to workaround
@ -663,7 +765,7 @@ var _ = SIGDescribe("Cluster size autoscaling [Slow]", func() {
By("Run a scale-up test")
ReserveMemory(f, "memory-reservation", 1, 100, false, 1*time.Second)
defer framework.DeleteRCAndPods(f.ClientSet, f.InternalClientset, f.Namespace.Name, "memory-reservation")
defer framework.DeleteRCAndWaitForGC(f.ClientSet, f.Namespace.Name, "memory-reservation")
// Verify that cluster size is increased
framework.ExpectNoError(WaitForClusterSizeFunc(f.ClientSet,
@ -776,7 +878,7 @@ var _ = SIGDescribe("Cluster size autoscaling [Slow]", func() {
framework.TestUnderTemporaryNetworkFailure(c, "default", ntb, testFunction)
} else {
ReserveMemory(f, "memory-reservation", 100, nodeCount*memAllocatableMb, false, defaultTimeout)
defer framework.DeleteRCAndPods(f.ClientSet, f.InternalClientset, f.Namespace.Name, "memory-reservation")
defer framework.DeleteRCAndWaitForGC(f.ClientSet, f.Namespace.Name, "memory-reservation")
time.Sleep(scaleUpTimeout)
currentNodes := framework.GetReadySchedulableNodesOrDie(f.ClientSet)
framework.Logf("Currently available nodes: %v, nodes available at the start of test: %v, disabled nodes: %v", len(currentNodes.Items), len(nodes.Items), nodesToBreakCount)
@ -957,6 +1059,12 @@ var _ = SIGDescribe("Cluster size autoscaling [Slow]", func() {
})
})
func installNvidiaDriversDaemonSet() {
By("Add daemonset which installs nvidia drivers")
// the link differs from one in GKE documentation; discussed with @mindprince this one should be used
framework.RunKubectlOrDie("apply", "-f", "https://raw.githubusercontent.com/GoogleCloudPlatform/container-engine-accelerators/master/daemonset.yaml")
}
func execCmd(args ...string) *exec.Cmd {
glog.Infof("Executing: %s", strings.Join(args, " "))
return exec.Command(args[0], args[1:]...)
@ -974,7 +1082,7 @@ func runDrainTest(f *framework.Framework, migSizes map[string]int, namespace str
labelMap := map[string]string{"test_id": testID}
framework.ExpectNoError(runReplicatedPodOnEachNode(f, nodes.Items, namespace, podsPerNode, "reschedulable-pods", labelMap, 0))
defer framework.DeleteRCAndPods(f.ClientSet, f.InternalClientset, namespace, "reschedulable-pods")
defer framework.DeleteRCAndWaitForGC(f.ClientSet, namespace, "reschedulable-pods")
By("Create a PodDisruptionBudget")
minAvailable := intstr.FromInt(numPods - pdbSize)
@ -1300,20 +1408,37 @@ func addNodePool(name string, machineType string, numNodes int) {
framework.ExpectNoError(err, string(output))
}
func addGpuNodePool(name string, gpuType string, gpuCount int, numNodes int) {
args := []string{"beta", "container", "node-pools", "create", name, "--quiet",
"--accelerator", "type=" + gpuType + ",count=" + strconv.Itoa(gpuCount),
"--num-nodes=" + strconv.Itoa(numNodes),
"--cluster=" + framework.TestContext.CloudConfig.Cluster}
output, err := execCmd(getGcloudCommand(args)...).CombinedOutput()
glog.Infof("Creating node-pool %s: %s", name, output)
framework.ExpectNoError(err, string(output))
}
func deleteNodePool(name string) {
glog.Infof("Deleting node pool %s", name)
args := []string{"container", "node-pools", "delete", name, "--quiet",
"--cluster=" + framework.TestContext.CloudConfig.Cluster}
output, err := execCmd(getGcloudCommand(args)...).CombinedOutput()
if err != nil {
glog.Infof("Error: %v", err)
}
glog.Infof("Node-pool deletion output: %s", output)
err := wait.ExponentialBackoff(
wait.Backoff{Duration: 1 * time.Minute, Factor: float64(3), Steps: 3},
func() (bool, error) {
output, err := execCmd(getGcloudCommand(args)...).CombinedOutput()
if err != nil {
glog.Warningf("Error deleting nodegroup - error:%v, output: %s", err, output)
return false, nil
}
glog.Infof("Node-pool deletion output: %s", output)
return true, nil
})
framework.ExpectNoError(err)
}
func getPoolNodes(f *framework.Framework, poolName string) []*v1.Node {
nodes := make([]*v1.Node, 0, 1)
nodeList := framework.GetReadySchedulableNodesOrDie(f.ClientSet)
nodeList := framework.GetReadyNodesIncludingTaintedOrDie(f.ClientSet)
for _, node := range nodeList.Items {
if node.Labels[gkeNodepoolNameKey] == poolName {
nodes = append(nodes, &node)
@ -1388,7 +1513,7 @@ func reserveMemory(f *framework.Framework, id string, replicas, megabytes int, e
Name: id,
Namespace: f.Namespace.Name,
Timeout: timeout,
Image: framework.GetPauseImageName(f.ClientSet),
Image: imageutils.GetPauseImageName(),
Replicas: replicas,
MemRequest: request,
NodeSelector: selector,
@ -1404,7 +1529,7 @@ func reserveMemory(f *framework.Framework, id string, replicas, megabytes int, e
framework.ExpectNoError(err)
}
return func() error {
return framework.DeleteRCAndPods(f.ClientSet, f.InternalClientset, f.Namespace.Name, id)
return framework.DeleteRCAndWaitForGC(f.ClientSet, f.Namespace.Name, id)
}
}
framework.Failf("Failed to reserve memory within timeout")
@ -1617,6 +1742,26 @@ func makeNodeSchedulable(c clientset.Interface, node *v1.Node, failOnCriticalAdd
return fmt.Errorf("Failed to remove taint from node in allowed number of retries")
}
func scheduleGpuPod(f *framework.Framework, id string) error {
config := &testutils.RCConfig{
Client: f.ClientSet,
InternalClient: f.InternalClientset,
Name: id,
Namespace: f.Namespace.Name,
Timeout: 3 * scaleUpTimeout, // spinning up GPU node is slow
Image: imageutils.GetPauseImageName(),
Replicas: 1,
GpuLimit: 1,
Labels: map[string]string{"requires-gpu": "yes"},
}
err := framework.RunRC(*config)
if err != nil {
return err
}
return nil
}
// Create an RC running a given number of pods with anti-affinity
func runAntiAffinityPods(f *framework.Framework, namespace string, pods int, id string, podLabels, antiAffinityLabels map[string]string) error {
config := &testutils.RCConfig{
@ -1626,7 +1771,7 @@ func runAntiAffinityPods(f *framework.Framework, namespace string, pods int, id
Name: id,
Namespace: namespace,
Timeout: scaleUpTimeout,
Image: framework.GetPauseImageName(f.ClientSet),
Image: imageutils.GetPauseImageName(),
Replicas: pods,
Labels: podLabels,
}
@ -1650,7 +1795,7 @@ func runVolumeAntiAffinityPods(f *framework.Framework, namespace string, pods in
Name: id,
Namespace: namespace,
Timeout: scaleUpTimeout,
Image: framework.GetPauseImageName(f.ClientSet),
Image: imageutils.GetPauseImageName(),
Replicas: pods,
Labels: podLabels,
}
@ -1731,7 +1876,7 @@ func runReplicatedPodOnEachNode(f *framework.Framework, nodes []v1.Node, namespa
Name: id,
Namespace: namespace,
Timeout: defaultTimeout,
Image: framework.GetPauseImageName(f.ClientSet),
Image: imageutils.GetPauseImageName(),
Replicas: 0,
Labels: labels,
MemRequest: memRequest,
@ -1790,7 +1935,7 @@ func runReplicatedPodOnEachNode(f *framework.Framework, nodes []v1.Node, namespa
func runReplicatedPodOnEachNodeWithCleanup(f *framework.Framework, nodes []v1.Node, namespace string, podsPerNode int, id string, labels map[string]string, memRequest int64) (func(), error) {
err := runReplicatedPodOnEachNode(f, nodes, namespace, podsPerNode, id, labels, memRequest)
return func() {
framework.DeleteRCAndPods(f.ClientSet, f.InternalClientset, namespace, id)
framework.DeleteRCAndWaitForGC(f.ClientSet, namespace, id)
}, err
}
@ -1853,7 +1998,7 @@ type scaleUpStatus struct {
// Try to get timestamp from status.
// Status configmap is not parsing-friendly, so evil regexpery follows.
func getStatusTimestamp(status string) (time.Time, error) {
timestampMatcher, err := regexp.Compile("Cluster-autoscaler status at \\s*([0-9\\-]+ [0-9]+:[0-9]+:[0-9]+\\.[0-9]+ \\+[0-9]+ [A-Za-z]+):")
timestampMatcher, err := regexp.Compile("Cluster-autoscaler status at \\s*([0-9\\-]+ [0-9]+:[0-9]+:[0-9]+\\.[0-9]+ \\+[0-9]+ [A-Za-z]+)")
if err != nil {
return time.Time{}, err
}
@ -2006,13 +2151,13 @@ func createPriorityClasses(f *framework.Framework) func() {
highPriorityClassName: 1000,
}
for className, priority := range priorityClasses {
_, err := f.ClientSet.SchedulingV1alpha1().PriorityClasses().Create(&v1alpha1.PriorityClass{ObjectMeta: metav1.ObjectMeta{Name: className}, Value: priority})
_, err := f.ClientSet.SchedulingV1beta1().PriorityClasses().Create(&schedulerapi.PriorityClass{ObjectMeta: metav1.ObjectMeta{Name: className}, Value: priority})
Expect(err == nil || errors.IsAlreadyExists(err)).To(Equal(true))
}
return func() {
for className := range priorityClasses {
f.ClientSet.SchedulingV1alpha1().PriorityClasses().Delete(className, nil)
f.ClientSet.SchedulingV1beta1().PriorityClasses().Delete(className, nil)
}
}
}

View File

@ -1,271 +0,0 @@
/*
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 autoscaling
import (
"context"
"time"
gcm "google.golang.org/api/monitoring/v3"
as "k8s.io/api/autoscaling/v2beta1"
corev1 "k8s.io/api/core/v1"
extensions "k8s.io/api/extensions/v1beta1"
"k8s.io/apimachinery/pkg/api/resource"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/util/wait"
clientset "k8s.io/client-go/kubernetes"
"k8s.io/kubernetes/test/e2e/framework"
"k8s.io/kubernetes/test/e2e/instrumentation/monitoring"
. "github.com/onsi/ginkgo"
"golang.org/x/oauth2/google"
)
const (
stackdriverExporterDeployment = "stackdriver-exporter-deployment"
dummyDeploymentName = "dummy-deployment"
stackdriverExporterPod = "stackdriver-exporter-pod"
)
var _ = SIGDescribe("[HPA] Horizontal pod autoscaling (scale resource: Custom Metrics from Stackdriver)", func() {
BeforeEach(func() {
framework.SkipUnlessProviderIs("gce", "gke")
})
f := framework.NewDefaultFramework("horizontal-pod-autoscaling")
It("should scale down with Custom Metric of type Pod from Stackdriver [Feature:CustomMetricsAutoscaling]", func() {
initialReplicas := 2
scaledReplicas := 1
// metric should cause scale down
metricValue := int64(100)
metricTarget := 2 * metricValue
deployment := monitoring.SimpleStackdriverExporterDeployment(stackdriverExporterDeployment, f.Namespace.ObjectMeta.Name, int32(initialReplicas), metricValue)
customMetricTest(f, f.ClientSet, simplePodsHPA(f.Namespace.ObjectMeta.Name, metricTarget), deployment, nil, initialReplicas, scaledReplicas)
})
It("should scale down with Custom Metric of type Object from Stackdriver [Feature:CustomMetricsAutoscaling]", func() {
initialReplicas := 2
scaledReplicas := 1
// metric should cause scale down
metricValue := int64(100)
metricTarget := 2 * metricValue
deployment := monitoring.SimpleStackdriverExporterDeployment(dummyDeploymentName, f.Namespace.ObjectMeta.Name, int32(initialReplicas), metricValue)
pod := monitoring.StackdriverExporterPod(stackdriverExporterPod, f.Namespace.Name, stackdriverExporterPod, monitoring.CustomMetricName, metricValue)
customMetricTest(f, f.ClientSet, objectHPA(f.Namespace.ObjectMeta.Name, metricTarget), deployment, pod, initialReplicas, scaledReplicas)
})
It("should scale down with Custom Metric of type Pod from Stackdriver with Prometheus [Feature:CustomMetricsAutoscaling]", func() {
initialReplicas := 2
scaledReplicas := 1
// metric should cause scale down
metricValue := int64(100)
metricTarget := 2 * metricValue
deployment := monitoring.PrometheusExporterDeployment(stackdriverExporterDeployment, f.Namespace.ObjectMeta.Name, int32(initialReplicas), metricValue)
customMetricTest(f, f.ClientSet, simplePodsHPA(f.Namespace.ObjectMeta.Name, metricTarget), deployment, nil, initialReplicas, scaledReplicas)
})
It("should scale up with two metrics of type Pod from Stackdriver [Feature:CustomMetricsAutoscaling]", func() {
initialReplicas := 1
scaledReplicas := 3
// metric 1 would cause a scale down, if not for metric 2
metric1Value := int64(100)
metric1Target := 2 * metric1Value
// metric2 should cause a scale up
metric2Value := int64(200)
metric2Target := int64(0.5 * float64(metric2Value))
containers := []monitoring.CustomMetricContainerSpec{
{
Name: "stackdriver-exporter-metric1",
MetricName: "metric1",
MetricValue: metric1Value,
},
{
Name: "stackdriver-exporter-metric2",
MetricName: "metric2",
MetricValue: metric2Value,
},
}
metricTargets := map[string]int64{"metric1": metric1Target, "metric2": metric2Target}
deployment := monitoring.StackdriverExporterDeployment(stackdriverExporterDeployment, f.Namespace.ObjectMeta.Name, int32(initialReplicas), containers)
customMetricTest(f, f.ClientSet, podsHPA(f.Namespace.ObjectMeta.Name, stackdriverExporterDeployment, metricTargets), deployment, nil, initialReplicas, scaledReplicas)
})
})
func customMetricTest(f *framework.Framework, kubeClient clientset.Interface, hpa *as.HorizontalPodAutoscaler,
deployment *extensions.Deployment, pod *corev1.Pod, initialReplicas, scaledReplicas int) {
projectId := framework.TestContext.CloudConfig.ProjectID
ctx := context.Background()
client, err := google.DefaultClient(ctx, gcm.CloudPlatformScope)
// Hack for running tests locally, needed to authenticate in Stackdriver
// If this is your use case, create application default credentials:
// $ gcloud auth application-default login
// and uncomment following lines:
/*
ts, err := google.DefaultTokenSource(oauth2.NoContext)
framework.Logf("Couldn't get application default credentials, %v", err)
if err != nil {
framework.Failf("Error accessing application default credentials, %v", err)
}
client := oauth2.NewClient(oauth2.NoContext, ts)
*/
gcmService, err := gcm.New(client)
if err != nil {
framework.Failf("Failed to create gcm service, %v", err)
}
// Set up a cluster: create a custom metric and set up k8s-sd adapter
err = monitoring.CreateDescriptors(gcmService, projectId)
if err != nil {
framework.Failf("Failed to create metric descriptor: %v", err)
}
defer monitoring.CleanupDescriptors(gcmService, projectId)
err = monitoring.CreateAdapter()
if err != nil {
framework.Failf("Failed to set up: %v", err)
}
defer monitoring.CleanupAdapter()
// Run application that exports the metric
err = createDeploymentToScale(f, kubeClient, deployment, pod)
if err != nil {
framework.Failf("Failed to create stackdriver-exporter pod: %v", err)
}
defer cleanupDeploymentsToScale(f, kubeClient, deployment, pod)
// Wait for the deployment to run
waitForReplicas(deployment.ObjectMeta.Name, f.Namespace.ObjectMeta.Name, kubeClient, 15*time.Minute, initialReplicas)
// Autoscale the deployment
_, err = kubeClient.AutoscalingV2beta1().HorizontalPodAutoscalers(f.Namespace.ObjectMeta.Name).Create(hpa)
if err != nil {
framework.Failf("Failed to create HPA: %v", err)
}
waitForReplicas(deployment.ObjectMeta.Name, f.Namespace.ObjectMeta.Name, kubeClient, 15*time.Minute, scaledReplicas)
}
func createDeploymentToScale(f *framework.Framework, cs clientset.Interface, deployment *extensions.Deployment, pod *corev1.Pod) error {
if deployment != nil {
_, err := cs.Extensions().Deployments(f.Namespace.ObjectMeta.Name).Create(deployment)
if err != nil {
return err
}
}
if pod != nil {
_, err := cs.CoreV1().Pods(f.Namespace.ObjectMeta.Name).Create(pod)
if err != nil {
return err
}
}
return nil
}
func cleanupDeploymentsToScale(f *framework.Framework, cs clientset.Interface, deployment *extensions.Deployment, pod *corev1.Pod) {
if deployment != nil {
_ = cs.Extensions().Deployments(f.Namespace.ObjectMeta.Name).Delete(deployment.ObjectMeta.Name, &metav1.DeleteOptions{})
}
if pod != nil {
_ = cs.CoreV1().Pods(f.Namespace.ObjectMeta.Name).Delete(pod.ObjectMeta.Name, &metav1.DeleteOptions{})
}
}
func simplePodsHPA(namespace string, metricTarget int64) *as.HorizontalPodAutoscaler {
return podsHPA(namespace, stackdriverExporterDeployment, map[string]int64{monitoring.CustomMetricName: metricTarget})
}
func podsHPA(namespace string, deploymentName string, metricTargets map[string]int64) *as.HorizontalPodAutoscaler {
var minReplicas int32 = 1
metrics := []as.MetricSpec{}
for metric, target := range metricTargets {
metrics = append(metrics, as.MetricSpec{
Type: as.PodsMetricSourceType,
Pods: &as.PodsMetricSource{
MetricName: metric,
TargetAverageValue: *resource.NewQuantity(target, resource.DecimalSI),
},
})
}
return &as.HorizontalPodAutoscaler{
ObjectMeta: metav1.ObjectMeta{
Name: "custom-metrics-pods-hpa",
Namespace: namespace,
},
Spec: as.HorizontalPodAutoscalerSpec{
Metrics: metrics,
MaxReplicas: 3,
MinReplicas: &minReplicas,
ScaleTargetRef: as.CrossVersionObjectReference{
APIVersion: "extensions/v1beta1",
Kind: "Deployment",
Name: deploymentName,
},
},
}
}
func objectHPA(namespace string, metricTarget int64) *as.HorizontalPodAutoscaler {
var minReplicas int32 = 1
return &as.HorizontalPodAutoscaler{
ObjectMeta: metav1.ObjectMeta{
Name: "custom-metrics-objects-hpa",
Namespace: namespace,
},
Spec: as.HorizontalPodAutoscalerSpec{
Metrics: []as.MetricSpec{
{
Type: as.ObjectMetricSourceType,
Object: &as.ObjectMetricSource{
MetricName: monitoring.CustomMetricName,
Target: as.CrossVersionObjectReference{
Kind: "Pod",
Name: stackdriverExporterPod,
},
TargetValue: *resource.NewQuantity(metricTarget, resource.DecimalSI),
},
},
},
MaxReplicas: 3,
MinReplicas: &minReplicas,
ScaleTargetRef: as.CrossVersionObjectReference{
APIVersion: "extensions/v1beta1",
Kind: "Deployment",
Name: dummyDeploymentName,
},
},
}
}
func waitForReplicas(deploymentName, namespace string, cs clientset.Interface, timeout time.Duration, desiredReplicas int) {
interval := 20 * time.Second
err := wait.PollImmediate(interval, timeout, func() (bool, error) {
deployment, err := cs.ExtensionsV1beta1().Deployments(namespace).Get(deploymentName, metav1.GetOptions{})
if err != nil {
framework.Failf("Failed to get replication controller %s: %v", deployment, err)
}
replicas := int(deployment.Status.ReadyReplicas)
framework.Logf("waiting for %d replicas (current: %d)", desiredReplicas, replicas)
return replicas == desiredReplicas, nil // Expected number of replicas found. Exit.
})
if err != nil {
framework.Failf("Timeout waiting %v for %v replicas", timeout, desiredReplicas)
}
}

View File

@ -0,0 +1,451 @@
/*
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 autoscaling
import (
"context"
"math"
"time"
gcm "google.golang.org/api/monitoring/v3"
as "k8s.io/api/autoscaling/v2beta1"
corev1 "k8s.io/api/core/v1"
extensions "k8s.io/api/extensions/v1beta1"
"k8s.io/apimachinery/pkg/api/resource"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/util/wait"
clientset "k8s.io/client-go/kubernetes"
"k8s.io/kubernetes/test/e2e/framework"
"k8s.io/kubernetes/test/e2e/instrumentation/monitoring"
. "github.com/onsi/ginkgo"
"golang.org/x/oauth2/google"
)
const (
stackdriverExporterDeployment = "stackdriver-exporter-deployment"
dummyDeploymentName = "dummy-deployment"
stackdriverExporterPod = "stackdriver-exporter-pod"
externalMetricValue = int64(85)
)
var _ = SIGDescribe("[HPA] Horizontal pod autoscaling (scale resource: Custom Metrics from Stackdriver)", func() {
BeforeEach(func() {
framework.SkipUnlessProviderIs("gce", "gke")
})
f := framework.NewDefaultFramework("horizontal-pod-autoscaling")
It("should scale down with Custom Metric of type Pod from Stackdriver [Feature:CustomMetricsAutoscaling]", func() {
initialReplicas := 2
// metric should cause scale down
metricValue := int64(100)
metricTarget := 2 * metricValue
tc := CustomMetricTestCase{
framework: f,
kubeClient: f.ClientSet,
initialReplicas: initialReplicas,
scaledReplicas: 1,
deployment: monitoring.SimpleStackdriverExporterDeployment(stackdriverExporterDeployment, f.Namespace.ObjectMeta.Name, int32(initialReplicas), metricValue),
hpa: simplePodsHPA(f.Namespace.ObjectMeta.Name, metricTarget)}
tc.Run()
})
It("should scale down with Custom Metric of type Object from Stackdriver [Feature:CustomMetricsAutoscaling]", func() {
initialReplicas := 2
// metric should cause scale down
metricValue := int64(100)
metricTarget := 2 * metricValue
tc := CustomMetricTestCase{
framework: f,
kubeClient: f.ClientSet,
initialReplicas: initialReplicas,
scaledReplicas: 1,
// Metric exported by deployment is ignored
deployment: monitoring.SimpleStackdriverExporterDeployment(dummyDeploymentName, f.Namespace.ObjectMeta.Name, int32(initialReplicas), 0 /* ignored */),
pod: monitoring.StackdriverExporterPod(stackdriverExporterPod, f.Namespace.Name, stackdriverExporterPod, monitoring.CustomMetricName, metricValue),
hpa: objectHPA(f.Namespace.ObjectMeta.Name, metricTarget)}
tc.Run()
})
It("should scale down with External Metric with target value from Stackdriver [Feature:CustomMetricsAutoscaling]", func() {
initialReplicas := 2
// metric should cause scale down
metricValue := externalMetricValue
metricTarget := 2 * metricValue
metricTargets := map[string]externalMetricTarget{
"target": {
value: metricTarget,
isAverage: false,
},
}
tc := CustomMetricTestCase{
framework: f,
kubeClient: f.ClientSet,
initialReplicas: initialReplicas,
scaledReplicas: 1,
// Metric exported by deployment is ignored
deployment: monitoring.SimpleStackdriverExporterDeployment(dummyDeploymentName, f.Namespace.ObjectMeta.Name, int32(initialReplicas), 0 /* ignored */),
pod: monitoring.StackdriverExporterPod(stackdriverExporterPod, f.Namespace.Name, stackdriverExporterPod, "target", metricValue),
hpa: externalHPA(f.Namespace.ObjectMeta.Name, metricTargets)}
tc.Run()
})
It("should scale down with External Metric with target average value from Stackdriver [Feature:CustomMetricsAutoscaling]", func() {
initialReplicas := 2
// metric should cause scale down
metricValue := externalMetricValue
metricAverageTarget := 2 * metricValue
metricTargets := map[string]externalMetricTarget{
"target_average": {
value: metricAverageTarget,
isAverage: true,
},
}
tc := CustomMetricTestCase{
framework: f,
kubeClient: f.ClientSet,
initialReplicas: initialReplicas,
scaledReplicas: 1,
// Metric exported by deployment is ignored
deployment: monitoring.SimpleStackdriverExporterDeployment(dummyDeploymentName, f.Namespace.ObjectMeta.Name, int32(initialReplicas), 0 /* ignored */),
pod: monitoring.StackdriverExporterPod(stackdriverExporterPod, f.Namespace.Name, stackdriverExporterPod, "target_average", externalMetricValue),
hpa: externalHPA(f.Namespace.ObjectMeta.Name, metricTargets)}
tc.Run()
})
It("should scale down with Custom Metric of type Pod from Stackdriver with Prometheus [Feature:CustomMetricsAutoscaling]", func() {
initialReplicas := 2
// metric should cause scale down
metricValue := int64(100)
metricTarget := 2 * metricValue
tc := CustomMetricTestCase{
framework: f,
kubeClient: f.ClientSet,
initialReplicas: initialReplicas,
scaledReplicas: 1,
deployment: monitoring.PrometheusExporterDeployment(stackdriverExporterDeployment, f.Namespace.ObjectMeta.Name, int32(initialReplicas), metricValue),
hpa: simplePodsHPA(f.Namespace.ObjectMeta.Name, metricTarget)}
tc.Run()
})
It("should scale up with two metrics of type Pod from Stackdriver [Feature:CustomMetricsAutoscaling]", func() {
initialReplicas := 1
// metric 1 would cause a scale down, if not for metric 2
metric1Value := int64(100)
metric1Target := 2 * metric1Value
// metric2 should cause a scale up
metric2Value := int64(200)
metric2Target := int64(0.5 * float64(metric2Value))
containers := []monitoring.CustomMetricContainerSpec{
{
Name: "stackdriver-exporter-metric1",
MetricName: "metric1",
MetricValue: metric1Value,
},
{
Name: "stackdriver-exporter-metric2",
MetricName: "metric2",
MetricValue: metric2Value,
},
}
metricTargets := map[string]int64{"metric1": metric1Target, "metric2": metric2Target}
tc := CustomMetricTestCase{
framework: f,
kubeClient: f.ClientSet,
initialReplicas: initialReplicas,
scaledReplicas: 3,
deployment: monitoring.StackdriverExporterDeployment(stackdriverExporterDeployment, f.Namespace.ObjectMeta.Name, int32(initialReplicas), containers),
hpa: podsHPA(f.Namespace.ObjectMeta.Name, stackdriverExporterDeployment, metricTargets)}
tc.Run()
})
It("should scale up with two External metrics from Stackdriver [Feature:CustomMetricsAutoscaling]", func() {
initialReplicas := 1
// metric 1 would cause a scale down, if not for metric 2
metric1Value := externalMetricValue
metric1Target := 2 * metric1Value
// metric2 should cause a scale up
metric2Value := externalMetricValue
metric2Target := int64(math.Ceil(0.5 * float64(metric2Value)))
metricTargets := map[string]externalMetricTarget{
"external_metric_1": {
value: metric1Target,
isAverage: false,
},
"external_metric_2": {
value: metric2Target,
isAverage: false,
},
}
containers := []monitoring.CustomMetricContainerSpec{
{
Name: "stackdriver-exporter-metric1",
MetricName: "external_metric_1",
MetricValue: metric1Value,
},
{
Name: "stackdriver-exporter-metric2",
MetricName: "external_metric_2",
MetricValue: metric2Value,
},
}
tc := CustomMetricTestCase{
framework: f,
kubeClient: f.ClientSet,
initialReplicas: initialReplicas,
scaledReplicas: 3,
deployment: monitoring.StackdriverExporterDeployment(dummyDeploymentName, f.Namespace.ObjectMeta.Name, int32(initialReplicas), containers),
hpa: externalHPA(f.Namespace.ObjectMeta.Name, metricTargets)}
tc.Run()
})
})
type CustomMetricTestCase struct {
framework *framework.Framework
hpa *as.HorizontalPodAutoscaler
kubeClient clientset.Interface
deployment *extensions.Deployment
pod *corev1.Pod
initialReplicas int
scaledReplicas int
}
func (tc *CustomMetricTestCase) Run() {
projectId := framework.TestContext.CloudConfig.ProjectID
ctx := context.Background()
client, err := google.DefaultClient(ctx, gcm.CloudPlatformScope)
// Hack for running tests locally, needed to authenticate in Stackdriver
// If this is your use case, create application default credentials:
// $ gcloud auth application-default login
// and uncomment following lines:
/*
ts, err := google.DefaultTokenSource(oauth2.NoContext)
framework.Logf("Couldn't get application default credentials, %v", err)
if err != nil {
framework.Failf("Error accessing application default credentials, %v", err)
}
client := oauth2.NewClient(oauth2.NoContext, ts)
*/
gcmService, err := gcm.New(client)
if err != nil {
framework.Failf("Failed to create gcm service, %v", err)
}
// Set up a cluster: create a custom metric and set up k8s-sd adapter
err = monitoring.CreateDescriptors(gcmService, projectId)
if err != nil {
framework.Failf("Failed to create metric descriptor: %v", err)
}
defer monitoring.CleanupDescriptors(gcmService, projectId)
err = monitoring.CreateAdapter(monitoring.AdapterDefault)
if err != nil {
framework.Failf("Failed to set up: %v", err)
}
defer monitoring.CleanupAdapter(monitoring.AdapterDefault)
// Run application that exports the metric
err = createDeploymentToScale(tc.framework, tc.kubeClient, tc.deployment, tc.pod)
if err != nil {
framework.Failf("Failed to create stackdriver-exporter pod: %v", err)
}
defer cleanupDeploymentsToScale(tc.framework, tc.kubeClient, tc.deployment, tc.pod)
// Wait for the deployment to run
waitForReplicas(tc.deployment.ObjectMeta.Name, tc.framework.Namespace.ObjectMeta.Name, tc.kubeClient, 15*time.Minute, tc.initialReplicas)
// Autoscale the deployment
_, err = tc.kubeClient.AutoscalingV2beta1().HorizontalPodAutoscalers(tc.framework.Namespace.ObjectMeta.Name).Create(tc.hpa)
if err != nil {
framework.Failf("Failed to create HPA: %v", err)
}
defer tc.kubeClient.AutoscalingV2beta1().HorizontalPodAutoscalers(tc.framework.Namespace.ObjectMeta.Name).Delete(tc.hpa.ObjectMeta.Name, &metav1.DeleteOptions{})
waitForReplicas(tc.deployment.ObjectMeta.Name, tc.framework.Namespace.ObjectMeta.Name, tc.kubeClient, 15*time.Minute, tc.scaledReplicas)
}
func createDeploymentToScale(f *framework.Framework, cs clientset.Interface, deployment *extensions.Deployment, pod *corev1.Pod) error {
if deployment != nil {
_, err := cs.Extensions().Deployments(f.Namespace.ObjectMeta.Name).Create(deployment)
if err != nil {
return err
}
}
if pod != nil {
_, err := cs.CoreV1().Pods(f.Namespace.ObjectMeta.Name).Create(pod)
if err != nil {
return err
}
}
return nil
}
func cleanupDeploymentsToScale(f *framework.Framework, cs clientset.Interface, deployment *extensions.Deployment, pod *corev1.Pod) {
if deployment != nil {
_ = cs.Extensions().Deployments(f.Namespace.ObjectMeta.Name).Delete(deployment.ObjectMeta.Name, &metav1.DeleteOptions{})
}
if pod != nil {
_ = cs.CoreV1().Pods(f.Namespace.ObjectMeta.Name).Delete(pod.ObjectMeta.Name, &metav1.DeleteOptions{})
}
}
func simplePodsHPA(namespace string, metricTarget int64) *as.HorizontalPodAutoscaler {
return podsHPA(namespace, stackdriverExporterDeployment, map[string]int64{monitoring.CustomMetricName: metricTarget})
}
func podsHPA(namespace string, deploymentName string, metricTargets map[string]int64) *as.HorizontalPodAutoscaler {
var minReplicas int32 = 1
metrics := []as.MetricSpec{}
for metric, target := range metricTargets {
metrics = append(metrics, as.MetricSpec{
Type: as.PodsMetricSourceType,
Pods: &as.PodsMetricSource{
MetricName: metric,
TargetAverageValue: *resource.NewQuantity(target, resource.DecimalSI),
},
})
}
return &as.HorizontalPodAutoscaler{
ObjectMeta: metav1.ObjectMeta{
Name: "custom-metrics-pods-hpa",
Namespace: namespace,
},
Spec: as.HorizontalPodAutoscalerSpec{
Metrics: metrics,
MaxReplicas: 3,
MinReplicas: &minReplicas,
ScaleTargetRef: as.CrossVersionObjectReference{
APIVersion: "extensions/v1beta1",
Kind: "Deployment",
Name: deploymentName,
},
},
}
}
func objectHPA(namespace string, metricTarget int64) *as.HorizontalPodAutoscaler {
var minReplicas int32 = 1
return &as.HorizontalPodAutoscaler{
ObjectMeta: metav1.ObjectMeta{
Name: "custom-metrics-objects-hpa",
Namespace: namespace,
},
Spec: as.HorizontalPodAutoscalerSpec{
Metrics: []as.MetricSpec{
{
Type: as.ObjectMetricSourceType,
Object: &as.ObjectMetricSource{
MetricName: monitoring.CustomMetricName,
Target: as.CrossVersionObjectReference{
Kind: "Pod",
Name: stackdriverExporterPod,
},
TargetValue: *resource.NewQuantity(metricTarget, resource.DecimalSI),
},
},
},
MaxReplicas: 3,
MinReplicas: &minReplicas,
ScaleTargetRef: as.CrossVersionObjectReference{
APIVersion: "extensions/v1beta1",
Kind: "Deployment",
Name: dummyDeploymentName,
},
},
}
}
type externalMetricTarget struct {
value int64
isAverage bool
}
func externalHPA(namespace string, metricTargets map[string]externalMetricTarget) *as.HorizontalPodAutoscaler {
var minReplicas int32 = 1
metricSpecs := []as.MetricSpec{}
selector := &metav1.LabelSelector{
MatchLabels: map[string]string{"resource.type": "gke_container"},
MatchExpressions: []metav1.LabelSelectorRequirement{
{
Key: "resource.labels.namespace_id",
Operator: metav1.LabelSelectorOpIn,
// TODO(bskiba): change default to real namespace name once it is available
// from Stackdriver.
Values: []string{"default", "dummy"},
},
{
Key: "resource.labels.pod_id",
Operator: metav1.LabelSelectorOpExists,
Values: []string{},
},
},
}
for metric, target := range metricTargets {
var metricSpec as.MetricSpec
metricSpec = as.MetricSpec{
Type: as.ExternalMetricSourceType,
External: &as.ExternalMetricSource{
MetricName: "custom.googleapis.com|" + metric,
MetricSelector: selector,
},
}
if target.isAverage {
metricSpec.External.TargetAverageValue = resource.NewQuantity(target.value, resource.DecimalSI)
} else {
metricSpec.External.TargetValue = resource.NewQuantity(target.value, resource.DecimalSI)
}
metricSpecs = append(metricSpecs, metricSpec)
}
hpa := &as.HorizontalPodAutoscaler{
ObjectMeta: metav1.ObjectMeta{
Name: "custom-metrics-external-hpa",
Namespace: namespace,
},
Spec: as.HorizontalPodAutoscalerSpec{
Metrics: metricSpecs,
MaxReplicas: 3,
MinReplicas: &minReplicas,
ScaleTargetRef: as.CrossVersionObjectReference{
APIVersion: "extensions/v1beta1",
Kind: "Deployment",
Name: dummyDeploymentName,
},
},
}
return hpa
}
func waitForReplicas(deploymentName, namespace string, cs clientset.Interface, timeout time.Duration, desiredReplicas int) {
interval := 20 * time.Second
err := wait.PollImmediate(interval, timeout, func() (bool, error) {
deployment, err := cs.ExtensionsV1beta1().Deployments(namespace).Get(deploymentName, metav1.GetOptions{})
if err != nil {
framework.Failf("Failed to get replication controller %s: %v", deployment, err)
}
replicas := int(deployment.Status.ReadyReplicas)
framework.Logf("waiting for %d replicas (current: %d)", desiredReplicas, replicas)
return replicas == desiredReplicas, nil // Expected number of replicas found. Exit.
})
if err != nil {
framework.Failf("Timeout waiting %v for %v replicas", timeout, desiredReplicas)
}
}

View File

@ -99,9 +99,11 @@ var _ = SIGDescribe("DNS horizontal autoscaling", func() {
// This test is separated because it is slow and need to run serially.
// Will take around 5 minutes to run on a 4 nodes cluster.
It("[Serial] [Slow] kube-dns-autoscaler should scale kube-dns pods when cluster size changed", func() {
numNodes, err := framework.NumberOfRegisteredNodes(c)
Expect(err).NotTo(HaveOccurred())
By("Replace the dns autoscaling parameters with testing parameters")
err := updateDNSScalingConfigMap(c, packDNSScalingConfigMap(packLinearParams(&DNSParams_1)))
err = updateDNSScalingConfigMap(c, packDNSScalingConfigMap(packLinearParams(&DNSParams_1)))
Expect(err).NotTo(HaveOccurred())
defer func() {
By("Restoring initial dns autoscaling parameters")
@ -117,25 +119,21 @@ var _ = SIGDescribe("DNS horizontal autoscaling", func() {
Expect(waitForDNSReplicasSatisfied(c, getExpectReplicasLinear, DNSdefaultTimeout)).NotTo(HaveOccurred())
originalSizes := make(map[string]int)
sum := 0
for _, mig := range strings.Split(framework.TestContext.CloudConfig.NodeInstanceGroup, ",") {
size, err := framework.GroupSize(mig)
Expect(err).NotTo(HaveOccurred())
By(fmt.Sprintf("Initial size of %s: %d", mig, size))
originalSizes[mig] = size
sum += size
}
By("Manually increase cluster size")
increasedSize := 0
increasedSizes := make(map[string]int)
for key, val := range originalSizes {
increasedSizes[key] = val + 1
increasedSize += increasedSizes[key]
}
setMigSizes(increasedSizes)
Expect(WaitForClusterSizeFunc(c,
func(size int) bool { return size == increasedSize }, scaleUpTimeout)).NotTo(HaveOccurred())
func(size int) bool { return size == numNodes+len(originalSizes) }, scaleUpTimeout)).NotTo(HaveOccurred())
By("Wait for kube-dns scaled to expected number")
getExpectReplicasLinear = getExpectReplicasFuncLinear(c, &DNSParams_1)
@ -151,7 +149,7 @@ var _ = SIGDescribe("DNS horizontal autoscaling", func() {
By("Restoring cluster size")
setMigSizes(originalSizes)
Expect(framework.WaitForReadyNodes(c, sum, scaleDownTimeout)).NotTo(HaveOccurred())
Expect(framework.WaitForReadyNodes(c, numNodes, scaleDownTimeout)).NotTo(HaveOccurred())
By("Wait for kube-dns scaled to expected number")
Expect(waitForDNSReplicasSatisfied(c, getExpectReplicasLinear, DNSdefaultTimeout)).NotTo(HaveOccurred())

View File

@ -116,7 +116,7 @@ type HPAScaleTest struct {
// TODO The use of 3 states is arbitrary, we could eventually make this test handle "n" states once this test stabilizes.
func (scaleTest *HPAScaleTest) run(name string, kind schema.GroupVersionKind, rc *common.ResourceConsumer, f *framework.Framework) {
const timeToWait = 15 * time.Minute
rc = common.NewDynamicResourceConsumer(name, f.Namespace.Name, kind, int(scaleTest.initPods), int(scaleTest.totalInitialCPUUsage), 0, 0, scaleTest.perPodCPURequest, 200, f.ClientSet, f.InternalClientset)
rc = common.NewDynamicResourceConsumer(name, f.Namespace.Name, kind, int(scaleTest.initPods), int(scaleTest.totalInitialCPUUsage), 0, 0, scaleTest.perPodCPURequest, 200, f.ClientSet, f.InternalClientset, f.ScalesGetter)
defer rc.CleanUp()
hpa := common.CreateCPUHorizontalPodAutoscaler(rc, scaleTest.targetCPUUtilizationPercent, scaleTest.minPods, scaleTest.maxPods)
defer common.DeleteHorizontalPodAutoscaler(rc, hpa.Name)