github/kubernetes
1
0
Fork 0
mirror of https://github.com/k3s-io/kubernetes.git synced 2025-08-06 10:43:56 +00:00
Code Issues Projects Releases Wiki Activity

Merge pull request #106544 from ehashman/fix-flake-restart

Browse Source 
Deflake "Kubelet should correctly account for terminated pods after restart"
This commit is contained in:
Kubernetes Prow Robot 2021-11-20 00:04:59 -08:00 committed by GitHub
commit 21d3acc787
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 11 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
test/e2e_node/restart_test.go
View File

@ -73,7 +73,7 @@ var _ = SIGDescribe("Restart [Serial] [Slow] [Disruptive]", func() {
podCount = 100 podCount = 100
podCreationInterval = 100 * time.Millisecond podCreationInterval = 100 * time.Millisecond
recoverTimeout = 5 * time.Minute recoverTimeout = 5 * time.Minute
startTimeout = 5 * time.Minute startTimeout = 3 * time.Minute
// restartCount is chosen so even with minPods we exhaust the default // restartCount is chosen so even with minPods we exhaust the default
// allocation of a /24. // allocation of a /24.
minPods = 50 minPods = 50
@ -150,6 +150,10 @@ var _ = SIGDescribe("Restart [Serial] [Slow] [Disruptive]", func() {
if numCpus < 1 { if numCpus < 1 {
e2eskipper.Skipf("insufficient CPU available for kubelet restart test") e2eskipper.Skipf("insufficient CPU available for kubelet restart test")
} }
if numCpus > 18 {
// 950m * 19 = 1805 CPUs -> not enough to block the scheduling of another 950m pod
e2eskipper.Skipf("test will return false positives on a machine with >18 cores")
}
// create as many restartNever pods as there are allocatable CPU // create as many restartNever pods as there are allocatable CPU
// nodes; if they are not correctly accounted for as terminated // nodes; if they are not correctly accounted for as terminated
@ -161,7 +165,7 @@ var _ = SIGDescribe("Restart [Serial] [Slow] [Disruptive]", func() {
pod.Spec.RestartPolicy = "Never" pod.Spec.RestartPolicy = "Never"
pod.Spec.Containers[0].Command = []string{"echo", "hi"} pod.Spec.Containers[0].Command = []string{"echo", "hi"}
pod.Spec.Containers[0].Resources.Limits = v1.ResourceList{ pod.Spec.Containers[0].Resources.Limits = v1.ResourceList{
v1.ResourceCPU: resource.MustParse("1"), v1.ResourceCPU: resource.MustParse("950m"), // leave a little room for other workloads
} }
} }
createBatchPodWithRateControl(f, restartNeverPods, podCreationInterval) createBatchPodWithRateControl(f, restartNeverPods, podCreationInterval)
@ -199,10 +203,12 @@ var _ = SIGDescribe("Restart [Serial] [Slow] [Disruptive]", func() {
// restart may think these old pods are consuming CPU and we // restart may think these old pods are consuming CPU and we
// will get an OutOfCpu error. // will get an OutOfCpu error.
ginkgo.By("verifying restartNever pods succeed and restartAlways pods stay running") ginkgo.By("verifying restartNever pods succeed and restartAlways pods stay running")
for start := time.Now(); time.Since(start) < startTimeout; time.Sleep(10 * time.Second) {
postRestartRunningPods := waitForPods(f, numAllPods, recoverTimeout) postRestartRunningPods := waitForPods(f, numAllPods, recoverTimeout)
if len(postRestartRunningPods) < numAllPods { if len(postRestartRunningPods) < numAllPods {
framework.Failf("less pods are running after node restart, got %d but expected %d", len(postRestartRunningPods), numAllPods) framework.Failf("less pods are running after node restart, got %d but expected %d", len(postRestartRunningPods), numAllPods)
} }
}
}) })
}) })
}) })