github/kubernetes
1
0
Fork 0
mirror of https://github.com/k3s-io/kubernetes.git synced 2025-07-24 20:24:09 +00:00
Code Issues Projects Releases Wiki Activity

Merge pull request #43453 from MaciekPytel/ca_more_drain_e2e

Browse Source 
Automatic merge from submit-queue (batch tested with PRs 43453, 42992)

Cluster-autoscaler multistep node drain e2e

**What this PR does / why we need it**:
This is another Cluster-Autoscaler node drain test. It tests draining a node, when there is pdb allowing for moving some pods, but not all the required pods in one go. Basically it's a more complex scenario of other CA drain e2e, testing the retry logic in CA scale down.

**Which issue this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close that issue when PR gets merged)*: fixes #

**Special notes for your reviewer**:

**Release note**:

```release-note
```
This commit is contained in:
Kubernetes Submit Queue 2017-04-04 04:27:17 -07:00 committed by GitHub
commit 826069e666
1 changed files with 62 additions and 36 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

98
test/e2e/autoscaling/cluster_size_autoscaling.go
View File

@ -330,7 +330,7 @@ var _ = framework.KubeDescribe("Cluster size autoscaling [Slow]", func() {
})
It("should be able to scale down when rescheduling a pod is required and pdb allows for it[Feature:ClusterSizeAutoscalingScaleDown]", func() {
runDrainTest(f, originalSizes, 1, func(increasedSize int) {
runDrainTest(f, originalSizes, 1, 1, func(increasedSize int) {
By("Some node should be removed")
framework.ExpectNoError(WaitForClusterSizeFunc(f.ClientSet,
func(size int) bool { return size < increasedSize }, scaleDownTimeout))
@ -338,7 +338,7 @@ var _ = framework.KubeDescribe("Cluster size autoscaling [Slow]", func() {
})
It("shouldn't be able to scale down when rescheduling a pod is required, but pdb doesn't allow drain[Feature:ClusterSizeAutoscalingScaleDown]", func() {
runDrainTest(f, originalSizes, 0, func(increasedSize int) {
runDrainTest(f, originalSizes, 1, 0, func(increasedSize int) {
By("No nodes should be removed")
time.Sleep(scaleDownTimeout)
nodes := framework.GetReadySchedulableNodesOrDie(f.ClientSet)
@ -346,9 +346,17 @@ var _ = framework.KubeDescribe("Cluster size autoscaling [Slow]", func() {
})
})
It("should be able to scale down by draining multiple pods one by one as dictated by pdb[Feature:ClusterSizeAutoscalingScaleDown]", func() {
runDrainTest(f, originalSizes, 2, 1, func(increasedSize int) {
By("Some node should be removed")
framework.ExpectNoError(WaitForClusterSizeFunc(f.ClientSet,
func(size int) bool { return size < increasedSize }, scaleDownTimeout))
})
})
})
func runDrainTest(f *framework.Framework, migSizes map[string]int, pdbSize int, verifyFunction func(int)) {
func runDrainTest(f *framework.Framework, migSizes map[string]int, podsPerNode, pdbSize int, verifyFunction func(int)) {
increasedSize := manuallyIncreaseClusterSize(f, migSizes)
nodes, err := f.ClientSet.Core().Nodes().List(metav1.ListOptions{FieldSelector: fields.Set{
@ -356,10 +364,10 @@ func runDrainTest(f *framework.Framework, migSizes map[string]int, pdbSize int,
}.AsSelector().String()})
framework.ExpectNoError(err)
namespace := f.Namespace.Name
numPods := len(nodes.Items)
numPods := len(nodes.Items) * podsPerNode
testId := string(uuid.NewUUID()) // So that we can label and find pods
labelMap := map[string]string{"test_id": testId}
framework.ExpectNoError(runReplicatedPodOnEachNode(f, nodes.Items, "reschedulable-pods", labelMap))
framework.ExpectNoError(runReplicatedPodOnEachNode(f, nodes.Items, podsPerNode, "reschedulable-pods", labelMap))
defer framework.DeleteRCAndPods(f.ClientSet, f.InternalClientset, f.Namespace.Name, "reschedulable-pods")
@ -666,55 +674,73 @@ func setMigSizes(sizes map[string]int) {
func makeNodeUnschedulable(c clientset.Interface, node *v1.Node) error {
By(fmt.Sprintf("Taint node %s", node.Name))
freshNode, err := c.Core().Nodes().Get(node.Name, metav1.GetOptions{})
if err != nil {
return err
}
for _, taint := range freshNode.Spec.Taints {
if taint.Key == disabledTaint {
for j := 0; j < 3; j++ {
freshNode, err := c.Core().Nodes().Get(node.Name, metav1.GetOptions{})
if err != nil {
return err
}
for _, taint := range freshNode.Spec.Taints {
if taint.Key == disabledTaint {
return nil
}
}
freshNode.Spec.Taints = append(freshNode.Spec.Taints, v1.Taint{
Key: disabledTaint,
Value: "DisabledForTest",
Effect: v1.TaintEffectNoSchedule,
})
_, err = c.Core().Nodes().Update(freshNode)
if err == nil {
return nil
}
if !errors.IsConflict(err) {
return err
}
glog.Warningf("Got 409 conflict when trying to taint node, retries left: %v", 3-j)
}
freshNode.Spec.Taints = append(freshNode.Spec.Taints, v1.Taint{
Key: disabledTaint,
Value: "DisabledForTest",
Effect: v1.TaintEffectNoSchedule,
})
_, err = c.Core().Nodes().Update(freshNode)
return err
return fmt.Errorf("Failed to taint node in allowed number of retries")
}
func makeNodeSchedulable(c clientset.Interface, node *v1.Node) error {
By(fmt.Sprintf("Remove taint from node %s", node.Name))
freshNode, err := c.Core().Nodes().Get(node.Name, metav1.GetOptions{})
if err != nil {
return err
}
newTaints := make([]v1.Taint, 0)
for _, taint := range freshNode.Spec.Taints {
if taint.Key != disabledTaint {
newTaints = append(newTaints, taint)
for j := 0; j < 3; j++ {
freshNode, err := c.Core().Nodes().Get(node.Name, metav1.GetOptions{})
if err != nil {
return err
}
newTaints := make([]v1.Taint, 0)
for _, taint := range freshNode.Spec.Taints {
if taint.Key != disabledTaint {
newTaints = append(newTaints, taint)
}
}
}
if len(newTaints) != len(freshNode.Spec.Taints) {
if len(newTaints) == len(freshNode.Spec.Taints) {
return nil
}
freshNode.Spec.Taints = newTaints
_, err = c.Core().Nodes().Update(freshNode)
return err
if err == nil {
return nil
}
if !errors.IsConflict(err) {
return err
}
glog.Warningf("Got 409 conflict when trying to taint node, retries left: %v", 3-j)
}
return nil
return fmt.Errorf("Failed to remove taint from node in allowed number of retries")
}
// Creat an RC running a single pod on each node without adding any constraint forcing such
// pod distribution. This is meant to create a bunch of underutilized (but not unused) nodes
// Creat an RC running a given number of pods on each node without adding any constraint forcing
// such pod distribution. This is meant to create a bunch of underutilized (but not unused) nodes
// with pods that can be rescheduled on different nodes.
// This is achieved using the following method:
// 1. disable scheduling on each node
// 2. create an empty RC
// 3. for each node:
// 3a. enable scheduling on that node
// 3b. increase number of replicas in RC by 1
func runReplicatedPodOnEachNode(f *framework.Framework, nodes []v1.Node, id string, labels map[string]string) error {
// 3b. increase number of replicas in RC by podsPerNode
func runReplicatedPodOnEachNode(f *framework.Framework, nodes []v1.Node, podsPerNode int, id string, labels map[string]string) error {
By("Run a pod on each node")
for _, node := range nodes {
err := makeNodeUnschedulable(f.ClientSet, &node)
@ -754,7 +780,7 @@ func runReplicatedPodOnEachNode(f *framework.Framework, nodes []v1.Node, id stri
// Update replicas count, to create new pods that will be allocated on node
// (we retry 409 errors in case rc reference got out of sync)
for j := 0; j < 3; j++ {
*rc.Spec.Replicas = int32(i + 1)
*rc.Spec.Replicas = int32((i + 1) * podsPerNode)
rc, err = f.ClientSet.Core().ReplicationControllers(f.Namespace.Name).Update(rc)
if err == nil {
break
@ -771,7 +797,7 @@ func runReplicatedPodOnEachNode(f *framework.Framework, nodes []v1.Node, id stri
err = wait.PollImmediate(5*time.Second, podTimeout, func() (bool, error) {
rc, err = f.ClientSet.Core().ReplicationControllers(f.Namespace.Name).Get(id, metav1.GetOptions{})
if err != nil || rc.Status.ReadyReplicas < int32(i+1) {
if err != nil || rc.Status.ReadyReplicas < int32((i+1)*podsPerNode) {
return false, nil
}
return true, nil