func createRC(wg *sync.WaitGroup, config *testutils.RCConfig, creatingTime time.Duration) {
defer GinkgoRecover()
defer wg.Done()
sleepUpTo(creatingTime)
framework.ExpectNoError(framework.RunRC(*config), fmt.Sprintf("creating rc %s", config.Name))
}
func runServiceLatencies(f *framework.Framework, inParallel, total int) (output []time.Duration, err error) {
cfg := testutils.RCConfig{
Client: f.Client,
Image: framework.GetPauseImageName(f.Client),
Name: "svc-latency-rc",
Namespace: f.Namespace.Name,
Replicas: 1,
PollInterval: time.Second,
}
if err := framework.RunRC(cfg); err != nil {
return nil, err
}
// Run a single watcher, to reduce the number of API calls we have to
// make; this is to minimize the timing error. It's how kube-proxy
// consumes the endpoints data, so it seems like the right thing to
// test.
endpointQueries := newQuerier()
startEndpointWatcher(f, endpointQueries)
defer close(endpointQueries.stop)
// run one test and throw it away-- this is to make sure that the pod's
// ready status has propagated.
singleServiceLatency(f, cfg.Name, endpointQueries)
// These channels are never closed, and each attempt sends on exactly
// one of these channels, so the sum of the things sent over them will
// be exactly total.
errs := make(chan error, total)
durations := make(chan time.Duration, total)
blocker := make(chan struct{}, inParallel)
for i := 0; i < total; i++ {
go func() {
defer GinkgoRecover()
blocker <- struct{}{}
defer func() { <-blocker }()
if d, err := singleServiceLatency(f, cfg.Name, endpointQueries); err != nil {
errs <- err
} else {
durations <- d
}
}()
}
errCount := 0
for i := 0; i < total; i++ {
select {
case e := <-errs:
framework.Logf("Got error: %v", e)
errCount += 1
case d := <-durations:
output = append(output, d)
}
}
if errCount != 0 {
return output, fmt.Errorf("got %v errors", errCount)
}
return output, nil
}
func runResourceTrackingTest(f *framework.Framework, podsPerNode int, nodeNames sets.String, rm *framework.ResourceMonitor,
expectedCPU map[string]map[float64]float64, expectedMemory framework.ResourceUsagePerContainer) {
numNodes := nodeNames.Len()
totalPods := podsPerNode * numNodes
By(fmt.Sprintf("Creating a RC of %d pods and wait until all pods of this RC are running", totalPods))
rcName := fmt.Sprintf("resource%d-%s", totalPods, string(uuid.NewUUID()))
// TODO: Use a more realistic workload
Expect(framework.RunRC(testutils.RCConfig{
Client: f.ClientSet,
InternalClient: f.InternalClientset,
Name: rcName,
Namespace: f.Namespace.Name,
Image: framework.GetPauseImageName(f.ClientSet),
Replicas: totalPods,
})).NotTo(HaveOccurred())
// Log once and flush the stats.
rm.LogLatest()
rm.Reset()
By("Start monitoring resource usage")
// Periodically dump the cpu summary until the deadline is met.
// Note that without calling framework.ResourceMonitor.Reset(), the stats
// would occupy increasingly more memory. This should be fine
// for the current test duration, but we should reclaim the
// entries if we plan to monitor longer (e.g., 8 hours).
deadline := time.Now().Add(monitoringTime)
for time.Now().Before(deadline) {
timeLeft := deadline.Sub(time.Now())
framework.Logf("Still running...%v left", timeLeft)
if timeLeft < reportingPeriod {
time.Sleep(timeLeft)
} else {
time.Sleep(reportingPeriod)
}
logPodsOnNodes(f.ClientSet, nodeNames.List())
}
By("Reporting overall resource usage")
logPodsOnNodes(f.ClientSet, nodeNames.List())
usageSummary, err := rm.GetLatest()
Expect(err).NotTo(HaveOccurred())
// TODO(random-liu): Remove the original log when we migrate to new perfdash
framework.Logf("%s", rm.FormatResourceUsage(usageSummary))
// Log perf result
framework.PrintPerfData(framework.ResourceUsageToPerfData(rm.GetMasterNodeLatest(usageSummary)))
verifyMemoryLimits(f.ClientSet, expectedMemory, usageSummary)
cpuSummary := rm.GetCPUSummary()
framework.Logf("%s", rm.FormatCPUSummary(cpuSummary))
// Log perf result
framework.PrintPerfData(framework.CPUUsageToPerfData(rm.GetMasterNodeCPUSummary(cpuSummary)))
verifyCPULimits(expectedCPU, cpuSummary)
By("Deleting the RC")
framework.DeleteRCAndPods(f.ClientSet, f.InternalClientset, f.Namespace.Name, rcName)
}
func runServiceAndWorkloadForResourceConsumer(c *client.Client, ns, name, kind string, replicas int, cpuLimitMillis, memLimitMb int64) {
By(fmt.Sprintf("Running consuming RC %s via %s with %v replicas", name, kind, replicas))
_, err := c.Services(ns).Create(&api.Service{
ObjectMeta: api.ObjectMeta{
Name: name,
},
Spec: api.ServiceSpec{
Ports: []api.ServicePort{{
Port: port,
TargetPort: intstr.FromInt(targetPort),
}},
Selector: map[string]string{
"name": name,
},
},
})
framework.ExpectNoError(err)
rcConfig := framework.RCConfig{
Client: c,
Image: resourceConsumerImage,
Name: name,
Namespace: ns,
Timeout: timeoutRC,
Replicas: replicas,
CpuRequest: cpuLimitMillis,
CpuLimit: cpuLimitMillis,
MemRequest: memLimitMb * 1024 * 1024, // MemLimit is in bytes
MemLimit: memLimitMb * 1024 * 1024,
}
switch kind {
case kindRC:
framework.ExpectNoError(framework.RunRC(rcConfig))
break
case kindDeployment:
dpConfig := framework.DeploymentConfig{
RCConfig: rcConfig,
}
framework.ExpectNoError(framework.RunDeployment(dpConfig))
break
case kindReplicaSet:
rsConfig := framework.ReplicaSetConfig{
RCConfig: rcConfig,
}
framework.ExpectNoError(framework.RunReplicaSet(rsConfig))
break
default:
framework.Failf(invalidKind)
}
// Make sure endpoints are propagated.
// TODO(piosz): replace sleep with endpoints watch.
time.Sleep(10 * time.Second)
}
func ReserveMemory(f *framework.Framework, id string, megabytes int) {
By(fmt.Sprintf("Running RC which reserves %v MB of memory", megabytes))
config := &framework.RCConfig{
Client: f.Client,
Name: id,
Namespace: f.Namespace.Name,
Timeout: 10 * time.Minute,
Image: "gcr.io/google_containers/pause:2.0",
Replicas: megabytes / 500,
MemRequest: 500 * 1024 * 1024,
}
framework.ExpectNoError(framework.RunRC(*config))
}
func ReserveCpu(f *framework.Framework, id string, millicores int) {
By(fmt.Sprintf("Running RC which reserves %v millicores", millicores))
config := &framework.RCConfig{
Client: f.Client,
Name: id,
Namespace: f.Namespace.Name,
Timeout: 10 * time.Minute,
Image: "gcr.io/google_containers/pause:2.0",
Replicas: millicores / 100,
CpuRequest: 100,
}
framework.ExpectNoError(framework.RunRC(*config))
}
func ReserveCpu(f *framework.Framework, id string, replicas, millicores int) {
By(fmt.Sprintf("Running RC which reserves %v millicores", millicores))
request := int64(millicores / replicas)
config := &testutils.RCConfig{
Client: f.Client,
Name: id,
Namespace: f.Namespace.Name,
Timeout: defaultTimeout,
Image: framework.GetPauseImageName(f.Client),
Replicas: replicas,
CpuRequest: request,
}
framework.ExpectNoError(framework.RunRC(*config))
}
func ReserveCpu(f *framework.Framework, id string, replicas, millicores int) {
By(fmt.Sprintf("Running RC which reserves %v millicores", millicores))
request := int64(millicores / replicas)
config := &framework.RCConfig{
Client: f.Client,
Name: id,
Namespace: f.Namespace.Name,
Timeout: scaleTimeout,
Image: "gcr.io/google_containers/pause-amd64:3.0",
Replicas: replicas,
CpuRequest: request,
}
framework.ExpectNoError(framework.RunRC(*config))
}
// runDensityTest will perform a density test and return the time it took for
// all pods to start
func runDensityTest(dtc DensityTestConfig) time.Duration {
defer GinkgoRecover()
// Start all replication controllers.
startTime := time.Now()
wg := sync.WaitGroup{}
wg.Add(len(dtc.Configs))
for i := range dtc.Configs {
rcConfig := dtc.Configs[i]
go func() {
defer GinkgoRecover()
// Call wg.Done() in defer to avoid blocking whole test
// in case of error from RunRC.
defer wg.Done()
framework.ExpectNoError(framework.RunRC(rcConfig))
}()
}
logStopCh := make(chan struct{})
go logPodStartupStatus(dtc.ClientSet, dtc.PodCount, map[string]string{"type": "densityPod"}, dtc.PollInterval, logStopCh)
wg.Wait()
startupTime := time.Now().Sub(startTime)
close(logStopCh)
framework.Logf("E2E startup time for %d pods: %v", dtc.PodCount, startupTime)
framework.Logf("Throughput (pods/s) during cluster saturation phase: %v", float32(dtc.PodCount)/float32(startupTime/time.Second))
// Print some data about Pod to Node allocation
By("Printing Pod to Node allocation data")
podList, err := dtc.ClientSet.Core().Pods(api.NamespaceAll).List(api.ListOptions{})
framework.ExpectNoError(err)
pausePodAllocation := make(map[string]int)
systemPodAllocation := make(map[string][]string)
for _, pod := range podList.Items {
if pod.Namespace == api.NamespaceSystem {
systemPodAllocation[pod.Spec.NodeName] = append(systemPodAllocation[pod.Spec.NodeName], pod.Name)
} else {
pausePodAllocation[pod.Spec.NodeName]++
}
}
nodeNames := make([]string, 0)
for k := range pausePodAllocation {
nodeNames = append(nodeNames, k)
}
sort.Strings(nodeNames)
for _, node := range nodeNames {
framework.Logf("%v: %v pause pods, system pods: %v", node, pausePodAllocation[node], systemPodAllocation[node])
}
return startupTime
}
func CreateHostPortPods(f *framework.Framework, id string, replicas int, expectRunning bool) {
By(fmt.Sprintf("Running RC which reserves host port"))
config := &testutils.RCConfig{
Client: f.Client,
Name: id,
Namespace: f.Namespace.Name,
Timeout: defaultTimeout,
Image: framework.GetPauseImageName(f.Client),
Replicas: replicas,
HostPorts: map[string]int{"port1": 4321},
}
err := framework.RunRC(*config)
if expectRunning {
framework.ExpectNoError(err)
}
}
func ReserveMemory(f *framework.Framework, id string, replicas, megabytes int, expectRunning bool) {
By(fmt.Sprintf("Running RC which reserves %v MB of memory", megabytes))
request := int64(1024 * 1024 * megabytes / replicas)
config := &testutils.RCConfig{
Client: f.Client,
Name: id,
Namespace: f.Namespace.Name,
Timeout: defaultTimeout,
Image: framework.GetPauseImageName(f.Client),
Replicas: replicas,
MemRequest: request,
}
err := framework.RunRC(*config)
if expectRunning {
framework.ExpectNoError(err)
}
}
func CreateHostPortPods(f *framework.Framework, id string, replicas int, expectRunning bool) {
By(fmt.Sprintf("Running RC which reserves host port"))
config := &framework.RCConfig{
Client: f.Client,
Name: id,
Namespace: f.Namespace.Name,
Timeout: scaleTimeout,
Image: "gcr.io/google_containers/pause-amd64:3.0",
Replicas: replicas,
HostPorts: map[string]int{"port1": 4321},
}
err := framework.RunRC(*config)
if expectRunning {
framework.ExpectNoError(err)
}
}
func ReserveMemory(f *framework.Framework, id string, replicas, megabytes int, expectRunning bool) {
By(fmt.Sprintf("Running RC which reserves %v MB of memory", megabytes))
request := int64(1024 * 1024 * megabytes / replicas)
config := &framework.RCConfig{
Client: f.Client,
Name: id,
Namespace: f.Namespace.Name,
Timeout: scaleTimeout,
Image: "gcr.io/google_containers/pause-amd64:3.0",
Replicas: replicas,
MemRequest: request,
}
err := framework.RunRC(*config)
if expectRunning {
framework.ExpectNoError(err)
}
}
func CreateNodeSelectorPods(f *framework.Framework, id string, replicas int, nodeSelector map[string]string, expectRunning bool) {
By(fmt.Sprintf("Running RC which reserves host port and defines node selector"))
config := &testutils.RCConfig{
Client: f.Client,
Name: "node-selector",
Namespace: f.Namespace.Name,
Timeout: defaultTimeout,
Image: framework.GetPauseImageName(f.Client),
Replicas: replicas,
HostPorts: map[string]int{"port1": 4321},
NodeSelector: map[string]string{"cluster-autoscaling-test.special-node": "true"},
}
err := framework.RunRC(*config)
if expectRunning {
framework.ExpectNoError(err)
}
}
func proxyContext(version string) {
options := framework.FrameworkOptions{
ClientQPS: -1.0,
}
f := framework.NewFramework("proxy", options, nil)
prefix := "/api/" + version
// Port here has to be kept in sync with default kubelet port.
It("should proxy logs on node with explicit kubelet port [Conformance]", func() { nodeProxyTest(f, prefix+"/proxy/nodes/", ":10250/logs/") })
It("should proxy logs on node [Conformance]", func() { nodeProxyTest(f, prefix+"/proxy/nodes/", "/logs/") })
It("should proxy to cadvisor [Conformance]", func() { nodeProxyTest(f, prefix+"/proxy/nodes/", ":4194/containers/") })
It("should proxy logs on node with explicit kubelet port using proxy subresource [Conformance]", func() { nodeProxyTest(f, prefix+"/nodes/", ":10250/proxy/logs/") })
It("should proxy logs on node using proxy subresource [Conformance]", func() { nodeProxyTest(f, prefix+"/nodes/", "/proxy/logs/") })
It("should proxy to cadvisor using proxy subresource [Conformance]", func() { nodeProxyTest(f, prefix+"/nodes/", ":4194/proxy/containers/") })
// using the porter image to serve content, access the content
// (of multiple pods?) from multiple (endpoints/services?)
It("should proxy through a service and a pod [Conformance]", func() {
start := time.Now()
labels := map[string]string{"proxy-service-target": "true"}
service, err := f.Client.Services(f.Namespace.Name).Create(&api.Service{
ObjectMeta: api.ObjectMeta{
GenerateName: "proxy-service-",
},
Spec: api.ServiceSpec{
Selector: labels,
Ports: []api.ServicePort{
{
Name: "portname1",
Port: 80,
TargetPort: intstr.FromString("dest1"),
},
{
Name: "portname2",
Port: 81,
TargetPort: intstr.FromInt(162),
},
{
Name: "tlsportname1",
Port: 443,
TargetPort: intstr.FromString("tlsdest1"),
},
{
Name: "tlsportname2",
Port: 444,
TargetPort: intstr.FromInt(462),
},
},
},
})
Expect(err).NotTo(HaveOccurred())
defer func(name string) {
err := f.Client.Services(f.Namespace.Name).Delete(name)
if err != nil {
framework.Logf("Failed deleting service %v: %v", name, err)
}
}(service.Name)
// Make an RC with a single pod. The 'porter' image is
// a simple server which serves the values of the
// environmental variables below.
By("starting an echo server on multiple ports")
pods := []*api.Pod{}
cfg := framework.RCConfig{
Client: f.Client,
Image: "gcr.io/google_containers/porter:cd5cb5791ebaa8641955f0e8c2a9bed669b1eaab",
Name: service.Name,
Namespace: f.Namespace.Name,
Replicas: 1,
PollInterval: time.Second,
Env: map[string]string{
"SERVE_PORT_80": `<a href="/rewriteme">test</a>`,
"SERVE_PORT_1080": `<a href="/rewriteme">test</a>`,
"SERVE_PORT_160": "foo",
"SERVE_PORT_162": "bar",
"SERVE_TLS_PORT_443": `<a href="/tlsrewriteme">test</a>`,
"SERVE_TLS_PORT_460": `tls baz`,
"SERVE_TLS_PORT_462": `tls qux`,
},
Ports: map[string]int{
"dest1": 160,
"dest2": 162,
"tlsdest1": 460,
"tlsdest2": 462,
},
ReadinessProbe: &api.Probe{
Handler: api.Handler{
HTTPGet: &api.HTTPGetAction{
Port: intstr.FromInt(80),
},
},
InitialDelaySeconds: 1,
TimeoutSeconds: 5,
PeriodSeconds: 10,
},
Labels: labels,
CreatedPods: &pods,
}
Expect(framework.RunRC(cfg)).NotTo(HaveOccurred())
defer framework.DeleteRCAndPods(f.Client, f.Namespace.Name, cfg.Name)
Expect(f.WaitForAnEndpoint(service.Name)).NotTo(HaveOccurred())
// table constructors
// Try proxying through the service and directly to through the pod.
svcProxyURL := func(scheme, port string) string {
return prefix + "/proxy/namespaces/" + f.Namespace.Name + "/services/" + net.JoinSchemeNamePort(scheme, service.Name, port)
}
subresourceServiceProxyURL := func(scheme, port string) string {
return prefix + "/namespaces/" + f.Namespace.Name + "/services/" + net.JoinSchemeNamePort(scheme, service.Name, port) + "/proxy"
}
podProxyURL := func(scheme, port string) string {
return prefix + "/proxy/namespaces/" + f.Namespace.Name + "/pods/" + net.JoinSchemeNamePort(scheme, pods[0].Name, port)
}
subresourcePodProxyURL := func(scheme, port string) string {
return prefix + "/namespaces/" + f.Namespace.Name + "/pods/" + net.JoinSchemeNamePort(scheme, pods[0].Name, port) + "/proxy"
}
// construct the table
expectations := map[string]string{
svcProxyURL("", "portname1") + "/": "foo",
svcProxyURL("", "80") + "/": "foo",
svcProxyURL("", "portname2") + "/": "bar",
svcProxyURL("", "81") + "/": "bar",
svcProxyURL("http", "portname1") + "/": "foo",
svcProxyURL("http", "80") + "/": "foo",
svcProxyURL("http", "portname2") + "/": "bar",
svcProxyURL("http", "81") + "/": "bar",
svcProxyURL("https", "tlsportname1") + "/": "tls baz",
svcProxyURL("https", "443") + "/": "tls baz",
svcProxyURL("https", "tlsportname2") + "/": "tls qux",
svcProxyURL("https", "444") + "/": "tls qux",
subresourceServiceProxyURL("", "portname1") + "/": "foo",
subresourceServiceProxyURL("http", "portname1") + "/": "foo",
subresourceServiceProxyURL("", "portname2") + "/": "bar",
subresourceServiceProxyURL("http", "portname2") + "/": "bar",
subresourceServiceProxyURL("https", "tlsportname1") + "/": "tls baz",
subresourceServiceProxyURL("https", "tlsportname2") + "/": "tls qux",
podProxyURL("", "1080") + "/": `<a href="` + podProxyURL("", "1080") + `/rewriteme">test</a>`,
podProxyURL("", "160") + "/": "foo",
podProxyURL("", "162") + "/": "bar",
podProxyURL("http", "1080") + "/": `<a href="` + podProxyURL("http", "1080") + `/rewriteme">test</a>`,
podProxyURL("http", "160") + "/": "foo",
podProxyURL("http", "162") + "/": "bar",
subresourcePodProxyURL("", "") + "/": `<a href="` + subresourcePodProxyURL("", "") + `/rewriteme">test</a>`,
subresourcePodProxyURL("", "1080") + "/": `<a href="` + subresourcePodProxyURL("", "1080") + `/rewriteme">test</a>`,
subresourcePodProxyURL("http", "1080") + "/": `<a href="` + subresourcePodProxyURL("http", "1080") + `/rewriteme">test</a>`,
subresourcePodProxyURL("", "160") + "/": "foo",
subresourcePodProxyURL("http", "160") + "/": "foo",
subresourcePodProxyURL("", "162") + "/": "bar",
subresourcePodProxyURL("http", "162") + "/": "bar",
subresourcePodProxyURL("https", "443") + "/": `<a href="` + subresourcePodProxyURL("https", "443") + `/tlsrewriteme">test</a>`,
subresourcePodProxyURL("https", "460") + "/": "tls baz",
subresourcePodProxyURL("https", "462") + "/": "tls qux",
// TODO: below entries don't work, but I believe we should make them work.
// podPrefix + ":dest1": "foo",
// podPrefix + ":dest2": "bar",
}
wg := sync.WaitGroup{}
errs := []string{}
errLock := sync.Mutex{}
recordError := func(s string) {
errLock.Lock()
defer errLock.Unlock()
errs = append(errs, s)
}
d := time.Since(start)
framework.Logf("setup took %v, starting test cases", d)
numberTestCases := len(expectations)
totalAttempts := numberTestCases * proxyAttempts
By(fmt.Sprintf("running %v cases, %v attempts per case, %v total attempts", numberTestCases, proxyAttempts, totalAttempts))
for i := 0; i < proxyAttempts; i++ {
wg.Add(numberTestCases)
for path, val := range expectations {
go func(i int, path, val string) {
defer wg.Done()
// this runs the test case
body, status, d, err := doProxy(f, path, i)
if err != nil {
if serr, ok := err.(*errors.StatusError); ok {
recordError(fmt.Sprintf("%v (%v; %v): path %v gave status error: %+v",
i, status, d, path, serr.Status()))
} else {
recordError(fmt.Sprintf("%v: path %v gave error: %v", i, path, err))
}
return
}
if status != http.StatusOK {
recordError(fmt.Sprintf("%v: path %v gave status: %v", i, path, status))
}
if e, a := val, string(body); e != a {
recordError(fmt.Sprintf("%v: path %v: wanted %v, got %v", i, path, e, a))
}
if d > proxyHTTPCallTimeout {
recordError(fmt.Sprintf("%v: path %v took %v > %v", i, path, d, proxyHTTPCallTimeout))
}
}(i, path, val)
}
wg.Wait()
}
if len(errs) != 0 {
body, err := f.Client.Pods(f.Namespace.Name).GetLogs(pods[0].Name, &api.PodLogOptions{}).Do().Raw()
if err != nil {
framework.Logf("Error getting logs for pod %s: %v", pods[0].Name, err)
} else {
framework.Logf("Pod %s has the following error logs: %s", pods[0].Name, body)
}
Fail(strings.Join(errs, "\n"))
}
})
}
BeforeEach(func() {
// These tests require SSH
framework.SkipUnlessProviderIs(framework.ProvidersWithSSH...)
ns = f.Namespace.Name
// All the restart tests need an rc and a watch on pods of the rc.
// Additionally some of them might scale the rc during the test.
config = framework.RCConfig{
Client: f.Client,
Name: rcName,
Namespace: ns,
Image: framework.GetPauseImageName(f.Client),
Replicas: numPods,
CreatedPods: &[]*api.Pod{},
}
Expect(framework.RunRC(config)).NotTo(HaveOccurred())
replacePods(*config.CreatedPods, existingPods)
stopCh = make(chan struct{})
tracker = newPodTracker()
newPods, controller = cache.NewInformer(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
options.LabelSelector = labelSelector
return f.Client.Pods(ns).List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
options.LabelSelector = labelSelector
return f.Client.Pods(ns).Watch(options)
},
},
func runServiceAndWorkloadForResourceConsumer(c clientset.Interface, internalClient internalclientset.Interface, ns, name, kind string, replicas int, cpuLimitMillis, memLimitMb int64) {
By(fmt.Sprintf("Running consuming RC %s via %s with %v replicas", name, kind, replicas))
_, err := c.Core().Services(ns).Create(&v1.Service{
ObjectMeta: v1.ObjectMeta{
Name: name,
},
Spec: v1.ServiceSpec{
Ports: []v1.ServicePort{{
Port: port,
TargetPort: intstr.FromInt(targetPort),
}},
Selector: map[string]string{
"name": name,
},
},
})
framework.ExpectNoError(err)
rcConfig := testutils.RCConfig{
Client: c,
InternalClient: internalClient,
Image: resourceConsumerImage,
Name: name,
Namespace: ns,
Timeout: timeoutRC,
Replicas: replicas,
CpuRequest: cpuLimitMillis,
CpuLimit: cpuLimitMillis,
MemRequest: memLimitMb * 1024 * 1024, // MemLimit is in bytes
MemLimit: memLimitMb * 1024 * 1024,
}
switch kind {
case kindRC:
framework.ExpectNoError(framework.RunRC(rcConfig))
break
case kindDeployment:
dpConfig := testutils.DeploymentConfig{
RCConfig: rcConfig,
}
framework.ExpectNoError(framework.RunDeployment(dpConfig))
break
case kindReplicaSet:
rsConfig := testutils.ReplicaSetConfig{
RCConfig: rcConfig,
}
By(fmt.Sprintf("creating replicaset %s in namespace %s", rsConfig.Name, rsConfig.Namespace))
framework.ExpectNoError(framework.RunReplicaSet(rsConfig))
break
default:
framework.Failf(invalidKind)
}
By(fmt.Sprintf("Running controller"))
controllerName := name + "-ctrl"
_, err = c.Core().Services(ns).Create(&v1.Service{
ObjectMeta: v1.ObjectMeta{
Name: controllerName,
},
Spec: v1.ServiceSpec{
Ports: []v1.ServicePort{{
Port: port,
TargetPort: intstr.FromInt(targetPort),
}},
Selector: map[string]string{
"name": controllerName,
},
},
})
framework.ExpectNoError(err)
dnsClusterFirst := v1.DNSClusterFirst
controllerRcConfig := testutils.RCConfig{
Client: c,
Image: resourceConsumerControllerImage,
Name: controllerName,
Namespace: ns,
Timeout: timeoutRC,
Replicas: 1,
Command: []string{"/controller", "--consumer-service-name=" + name, "--consumer-service-namespace=" + ns, "--consumer-port=80"},
DNSPolicy: &dnsClusterFirst,
}
framework.ExpectNoError(framework.RunRC(controllerRcConfig))
// Wait for endpoints to propagate for the controller service.
framework.ExpectNoError(framework.WaitForServiceEndpointsNum(
c, ns, controllerName, 1, startServiceInterval, startServiceTimeout))
}
uLock.Lock()
defer uLock.Unlock()
updateCount++
},
},
)
go updateController.Run(stop)
// Start all replication controllers.
startTime := time.Now()
wg := sync.WaitGroup{}
wg.Add(len(RCConfigs))
for i := range RCConfigs {
rcConfig := RCConfigs[i]
go func() {
framework.ExpectNoError(framework.RunRC(rcConfig))
wg.Done()
}()
}
logStopCh := make(chan struct{})
go logPodStartupStatus(c, totalPods, ns, map[string]string{"type": "densityPod"}, itArg.interval, logStopCh)
wg.Wait()
e2eStartupTime = time.Now().Sub(startTime)
close(logStopCh)
framework.Logf("E2E startup time for %d pods: %v", totalPods, e2eStartupTime)
framework.Logf("Throughput (pods/s) during cluster saturation phase: %v", float32(totalPods)/float32(e2eStartupTime/time.Second))
By("Waiting for all events to be recorded")
last := -1
current := len(events)
lastCount := -1
var _ = framework.KubeDescribe("Etcd failure [Disruptive]", func() {
f := framework.NewDefaultFramework("etcd-failure")
BeforeEach(func() {
// This test requires:
// - SSH
// - master access
// ... so the provider check should be identical to the intersection of
// providers that provide those capabilities.
framework.SkipUnlessProviderIs("gce")
Expect(framework.RunRC(framework.RCConfig{
Client: f.Client,
Name: "baz",
Namespace: f.Namespace.Name,
Image: "gcr.io/google_containers/pause:2.0",
Replicas: 1,
})).NotTo(HaveOccurred())
})
It("should recover from network partition with master", func() {
etcdFailTest(
f,
"sudo iptables -A INPUT -p tcp --destination-port 4001 -j DROP",
"sudo iptables -D INPUT -p tcp --destination-port 4001 -j DROP",
)
})
It("should recover from SIGKILL", func() {
etcdFailTest(
func runServiceAndWorkloadForResourceConsumer(c *client.Client, ns, name, kind string, replicas int, cpuLimitMillis, memLimitMb int64) {
By(fmt.Sprintf("Running consuming RC %s via %s with %v replicas", name, kind, replicas))
_, err := c.Services(ns).Create(&api.Service{
ObjectMeta: api.ObjectMeta{
Name: name,
},
Spec: api.ServiceSpec{
Ports: []api.ServicePort{{
Port: port,
TargetPort: intstr.FromInt(targetPort),
}},
Selector: map[string]string{
"name": name,
},
},
})
framework.ExpectNoError(err)
rcConfig := testutils.RCConfig{
Client: c,
Image: resourceConsumerImage,
Name: name,
Namespace: ns,
Timeout: timeoutRC,
Replicas: replicas,
CpuRequest: cpuLimitMillis,
CpuLimit: cpuLimitMillis,
MemRequest: memLimitMb * 1024 * 1024, // MemLimit is in bytes
MemLimit: memLimitMb * 1024 * 1024,
}
switch kind {
case kindRC:
framework.ExpectNoError(framework.RunRC(rcConfig))
break
case kindDeployment:
dpConfig := testutils.DeploymentConfig{
RCConfig: rcConfig,
}
framework.ExpectNoError(framework.RunDeployment(dpConfig))
break
case kindReplicaSet:
rsConfig := testutils.ReplicaSetConfig{
RCConfig: rcConfig,
}
By(fmt.Sprintf("creating replicaset %s in namespace %s", rsConfig.Name, rsConfig.Namespace))
framework.ExpectNoError(framework.RunReplicaSet(rsConfig))
break
default:
framework.Failf(invalidKind)
}
By(fmt.Sprintf("Running controller"))
controllerName := name + "-ctrl"
_, err = c.Services(ns).Create(&api.Service{
ObjectMeta: api.ObjectMeta{
Name: controllerName,
},
Spec: api.ServiceSpec{
Ports: []api.ServicePort{{
Port: port,
TargetPort: intstr.FromInt(targetPort),
}},
Selector: map[string]string{
"name": controllerName,
},
},
})
framework.ExpectNoError(err)
dnsClusterFirst := api.DNSClusterFirst
controllerRcConfig := testutils.RCConfig{
Client: c,
Image: resourceConsumerControllerImage,
Name: controllerName,
Namespace: ns,
Timeout: timeoutRC,
Replicas: 1,
Command: []string{"/controller", "--consumer-service-name=" + name, "--consumer-service-namespace=" + ns, "--consumer-port=80"},
DNSPolicy: &dnsClusterFirst,
}
framework.ExpectNoError(framework.RunRC(controllerRcConfig))
// Make sure endpoints are propagated.
// TODO(piosz): replace sleep with endpoints watch.
time.Sleep(10 * time.Second)
}
}
deleteTests := []DeleteTest{
{podsPerNode: 10, timeout: 1 * time.Minute},
}
for _, itArg := range deleteTests {
name := fmt.Sprintf(
"kubelet should be able to delete %d pods per node in %v.", itArg.podsPerNode, itArg.timeout)
It(name, func() {
totalPods := itArg.podsPerNode * numNodes
By(fmt.Sprintf("Creating a RC of %d pods and wait until all pods of this RC are running", totalPods))
rcName := fmt.Sprintf("cleanup%d-%s", totalPods, string(util.NewUUID()))
Expect(framework.RunRC(framework.RCConfig{
Client: f.Client,
Name: rcName,
Namespace: f.Namespace.Name,
Image: framework.GetPauseImageName(f.Client),
Replicas: totalPods,
})).NotTo(HaveOccurred())
// Perform a sanity check so that we know all desired pods are
// running on the nodes according to kubelet. The timeout is set to
// only 30 seconds here because framework.RunRC already waited for all pods to
// transition to the running status.
Expect(waitTillNPodsRunningOnNodes(f.Client, nodeNames, rcName, f.Namespace.Name, totalPods,
time.Second*30)).NotTo(HaveOccurred())
resourceMonitor.LogLatest()
By("Deleting the RC")
framework.DeleteRC(f.Client, f.Namespace.Name, rcName)
// Check that the pods really are gone by querying /runningpods on the
// node. The /runningpods handler checks the container runtime (or its
// runDensityTest will perform a density test and return the time it took for
// all pods to start
func runDensityTest(dtc DensityTestConfig) time.Duration {
defer GinkgoRecover()
// Create a listener for events.
// eLock is a lock protects the events
var eLock sync.Mutex
events := make([](*api.Event), 0)
_, controller := controllerframework.NewInformer(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
return dtc.Client.Events(dtc.Namespace).List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
return dtc.Client.Events(dtc.Namespace).Watch(options)
},
},
&api.Event{},
0,
controllerframework.ResourceEventHandlerFuncs{
AddFunc: func(obj interface{}) {
eLock.Lock()
defer eLock.Unlock()
events = append(events, obj.(*api.Event))
},
},
)
stop := make(chan struct{})
go controller.Run(stop)
// Create a listener for api updates
// uLock is a lock protects the updateCount
var uLock sync.Mutex
updateCount := 0
label := labels.SelectorFromSet(labels.Set(map[string]string{"type": "densityPod"}))
_, updateController := controllerframework.NewInformer(
&cache.ListWatch{
ListFunc: func(options api.ListOptions) (runtime.Object, error) {
options.LabelSelector = label
return dtc.Client.Pods(dtc.Namespace).List(options)
},
WatchFunc: func(options api.ListOptions) (watch.Interface, error) {
options.LabelSelector = label
return dtc.Client.Pods(dtc.Namespace).Watch(options)
},
},
&api.Pod{},
0,
controllerframework.ResourceEventHandlerFuncs{
UpdateFunc: func(_, _ interface{}) {
uLock.Lock()
defer uLock.Unlock()
updateCount++
},
},
)
go updateController.Run(stop)
// Start all replication controllers.
startTime := time.Now()
wg := sync.WaitGroup{}
wg.Add(len(dtc.Configs))
for i := range dtc.Configs {
rcConfig := dtc.Configs[i]
go func() {
framework.ExpectNoError(framework.RunRC(rcConfig))
wg.Done()
}()
}
logStopCh := make(chan struct{})
go logPodStartupStatus(dtc.Client, dtc.PodCount, dtc.Namespace, map[string]string{"type": "densityPod"}, dtc.PollInterval, logStopCh)
wg.Wait()
startupTime := time.Now().Sub(startTime)
close(logStopCh)
framework.Logf("E2E startup time for %d pods: %v", dtc.PodCount, startupTime)
framework.Logf("Throughput (pods/s) during cluster saturation phase: %v", float32(dtc.PodCount)/float32(startupTime/time.Second))
By("Waiting for all events to be recorded")
last := -1
current := len(events)
lastCount := -1
currentCount := updateCount
for start := time.Now(); (last < current || lastCount < currentCount) && time.Since(start) < dtc.Timeout; time.Sleep(10 * time.Second) {
func() {
eLock.Lock()
defer eLock.Unlock()
last = current
current = len(events)
}()
func() {
uLock.Lock()
defer uLock.Unlock()
lastCount = currentCount
currentCount = updateCount
}()
}
close(stop)
if current != last {
framework.Logf("Warning: Not all events were recorded after waiting %.2f minutes", dtc.Timeout.Minutes())
}
framework.Logf("Found %d events", current)
if currentCount != lastCount {
framework.Logf("Warning: Not all updates were recorded after waiting %.2f minutes", dtc.Timeout.Minutes())
}
framework.Logf("Found %d updates", currentCount)
// Tune the threshold for allowed failures.
badEvents := framework.BadEvents(events)
Expect(badEvents).NotTo(BeNumerically(">", int(math.Floor(0.01*float64(dtc.PodCount)))))
// Print some data about Pod to Node allocation
By("Printing Pod to Node allocation data")
podList, err := dtc.Client.Pods(api.NamespaceAll).List(api.ListOptions{})
framework.ExpectNoError(err)
pausePodAllocation := make(map[string]int)
systemPodAllocation := make(map[string][]string)
for _, pod := range podList.Items {
if pod.Namespace == api.NamespaceSystem {
systemPodAllocation[pod.Spec.NodeName] = append(systemPodAllocation[pod.Spec.NodeName], pod.Name)
} else {
pausePodAllocation[pod.Spec.NodeName]++
}
}
nodeNames := make([]string, 0)
for k := range pausePodAllocation {
nodeNames = append(nodeNames, k)
}
sort.Strings(nodeNames)
for _, node := range nodeNames {
framework.Logf("%v: %v pause pods, system pods: %v", node, pausePodAllocation[node], systemPodAllocation[node])
}
return startupTime
}
deleteTests := []DeleteTest{
{podsPerNode: 10, timeout: 1 * time.Minute},
}
for _, itArg := range deleteTests {
name := fmt.Sprintf(
"kubelet should be able to delete %d pods per node in %v.", itArg.podsPerNode, itArg.timeout)
It(name, func() {
totalPods := itArg.podsPerNode * numNodes
By(fmt.Sprintf("Creating a RC of %d pods and wait until all pods of this RC are running", totalPods))
rcName := fmt.Sprintf("cleanup%d-%s", totalPods, string(uuid.NewUUID()))
Expect(framework.RunRC(testutils.RCConfig{
Client: f.ClientSet,
InternalClient: f.InternalClientset,
Name: rcName,
Namespace: f.Namespace.Name,
Image: framework.GetPauseImageName(f.ClientSet),
Replicas: totalPods,
NodeSelector: nodeLabels,
})).NotTo(HaveOccurred())
// Perform a sanity check so that we know all desired pods are
// running on the nodes according to kubelet. The timeout is set to
// only 30 seconds here because framework.RunRC already waited for all pods to
// transition to the running status.
Expect(waitTillNPodsRunningOnNodes(f.ClientSet, nodeNames, rcName, f.Namespace.Name, totalPods,
time.Second*30)).NotTo(HaveOccurred())
if resourceMonitor != nil {
resourceMonitor.LogLatest()
}
By("Deleting the RC")
}
deleteTests := []DeleteTest{
{podsPerNode: 10, timeout: 1 * time.Minute},
}
for _, itArg := range deleteTests {
name := fmt.Sprintf(
"kubelet should be able to delete %d pods per node in %v.", itArg.podsPerNode, itArg.timeout)
It(name, func() {
totalPods := itArg.podsPerNode * numNodes
By(fmt.Sprintf("Creating a RC of %d pods and wait until all pods of this RC are running", totalPods))
rcName := fmt.Sprintf("cleanup%d-%s", totalPods, string(util.NewUUID()))
Expect(framework.RunRC(framework.RCConfig{
Client: f.Client,
Name: rcName,
Namespace: f.Namespace.Name,
Image: "gcr.io/google_containers/pause-amd64:3.0",
Replicas: totalPods,
})).NotTo(HaveOccurred())
// Perform a sanity check so that we know all desired pods are
// running on the nodes according to kubelet. The timeout is set to
// only 30 seconds here because framework.RunRC already waited for all pods to
// transition to the running status.
Expect(waitTillNPodsRunningOnNodes(f.Client, nodeNames, rcName, f.Namespace.Name, totalPods,
time.Second*30)).NotTo(HaveOccurred())
resourceMonitor.LogLatest()
By("Deleting the RC")
framework.DeleteRC(f.Client, f.Namespace.Name, rcName)
// Check that the pods really are gone by querying /runningpods on the
// node. The /runningpods handler checks the container runtime (or its
var _ = framework.KubeDescribe("Etcd failure [Disruptive]", func() {
f := framework.NewDefaultFramework("etcd-failure")
BeforeEach(func() {
// This test requires:
// - SSH
// - master access
// ... so the provider check should be identical to the intersection of
// providers that provide those capabilities.
framework.SkipUnlessProviderIs("gce")
Expect(framework.RunRC(testutils.RCConfig{
Client: f.ClientSet,
Name: "baz",
Namespace: f.Namespace.Name,
Image: framework.GetPauseImageName(f.ClientSet),
Replicas: 1,
})).NotTo(HaveOccurred())
})
It("should recover from network partition with master", func() {
etcdFailTest(
f,
"sudo iptables -A INPUT -p tcp --destination-port 2379 -j DROP",
"sudo iptables -D INPUT -p tcp --destination-port 2379 -j DROP",
)
})
It("should recover from SIGKILL", func() {
etcdFailTest(
func proxyContext(version string) {
f := framework.NewDefaultFramework("proxy")
prefix := "/api/" + version
// Port here has to be kept in sync with default kubelet port.
It("should proxy logs on node with explicit kubelet port [Conformance]", func() { nodeProxyTest(f, prefix+"/proxy/nodes/", ":10250/logs/") })
It("should proxy logs on node [Conformance]", func() { nodeProxyTest(f, prefix+"/proxy/nodes/", "/logs/") })
It("should proxy to cadvisor [Conformance]", func() { nodeProxyTest(f, prefix+"/proxy/nodes/", ":4194/containers/") })
It("should proxy logs on node with explicit kubelet port using proxy subresource [Conformance]", func() { nodeProxyTest(f, prefix+"/nodes/", ":10250/proxy/logs/") })
It("should proxy logs on node using proxy subresource [Conformance]", func() { nodeProxyTest(f, prefix+"/nodes/", "/proxy/logs/") })
It("should proxy to cadvisor using proxy subresource [Conformance]", func() { nodeProxyTest(f, prefix+"/nodes/", ":4194/proxy/containers/") })
It("should proxy through a service and a pod [Conformance]", func() {
labels := map[string]string{"proxy-service-target": "true"}
service, err := f.Client.Services(f.Namespace.Name).Create(&api.Service{
ObjectMeta: api.ObjectMeta{
GenerateName: "proxy-service-",
},
Spec: api.ServiceSpec{
Selector: labels,
Ports: []api.ServicePort{
{
Name: "portname1",
Port: 80,
TargetPort: intstr.FromString("dest1"),
},
{
Name: "portname2",
Port: 81,
TargetPort: intstr.FromInt(162),
},
{
Name: "tlsportname1",
Port: 443,
TargetPort: intstr.FromString("tlsdest1"),
},
{
Name: "tlsportname2",
Port: 444,
TargetPort: intstr.FromInt(462),
},
},
},
})
Expect(err).NotTo(HaveOccurred())
defer func(name string) {
err := f.Client.Services(f.Namespace.Name).Delete(name)
if err != nil {
framework.Logf("Failed deleting service %v: %v", name, err)
}
}(service.Name)
// Make an RC with a single pod.
pods := []*api.Pod{}
cfg := framework.RCConfig{
Client: f.Client,
Image: "gcr.io/google_containers/porter:cd5cb5791ebaa8641955f0e8c2a9bed669b1eaab",
Name: service.Name,
Namespace: f.Namespace.Name,
Replicas: 1,
PollInterval: time.Second,
Env: map[string]string{
"SERVE_PORT_80": `<a href="/rewriteme">test</a>`,
"SERVE_PORT_160": "foo",
"SERVE_PORT_162": "bar",
"SERVE_TLS_PORT_443": `<a href="/tlsrewriteme">test</a>`,
"SERVE_TLS_PORT_460": `tls baz`,
"SERVE_TLS_PORT_462": `tls qux`,
},
Ports: map[string]int{
"dest1": 160,
"dest2": 162,
"tlsdest1": 460,
"tlsdest2": 462,
},
ReadinessProbe: &api.Probe{
Handler: api.Handler{
HTTPGet: &api.HTTPGetAction{
Port: intstr.FromInt(80),
},
},
InitialDelaySeconds: 1,
TimeoutSeconds: 5,
PeriodSeconds: 10,
},
Labels: labels,
CreatedPods: &pods,
}
Expect(framework.RunRC(cfg)).NotTo(HaveOccurred())
defer framework.DeleteRC(f.Client, f.Namespace.Name, cfg.Name)
Expect(f.WaitForAnEndpoint(service.Name)).NotTo(HaveOccurred())
// Try proxying through the service and directly to through the pod.
svcProxyURL := func(scheme, port string) string {
return prefix + "/proxy/namespaces/" + f.Namespace.Name + "/services/" + net.JoinSchemeNamePort(scheme, service.Name, port)
}
subresourceServiceProxyURL := func(scheme, port string) string {
return prefix + "/namespaces/" + f.Namespace.Name + "/services/" + net.JoinSchemeNamePort(scheme, service.Name, port) + "/proxy"
}
podProxyURL := func(scheme, port string) string {
return prefix + "/proxy/namespaces/" + f.Namespace.Name + "/pods/" + net.JoinSchemeNamePort(scheme, pods[0].Name, port)
}
subresourcePodProxyURL := func(scheme, port string) string {
return prefix + "/namespaces/" + f.Namespace.Name + "/pods/" + net.JoinSchemeNamePort(scheme, pods[0].Name, port) + "/proxy"
}
expectations := map[string]string{
svcProxyURL("", "portname1") + "/": "foo",
svcProxyURL("", "80") + "/": "foo",
svcProxyURL("", "portname2") + "/": "bar",
svcProxyURL("", "81") + "/": "bar",
svcProxyURL("http", "portname1") + "/": "foo",
svcProxyURL("http", "80") + "/": "foo",
svcProxyURL("http", "portname2") + "/": "bar",
svcProxyURL("http", "81") + "/": "bar",
svcProxyURL("https", "tlsportname1") + "/": "tls baz",
svcProxyURL("https", "443") + "/": "tls baz",
svcProxyURL("https", "tlsportname2") + "/": "tls qux",
svcProxyURL("https", "444") + "/": "tls qux",
subresourceServiceProxyURL("", "portname1") + "/": "foo",
subresourceServiceProxyURL("http", "portname1") + "/": "foo",
subresourceServiceProxyURL("", "portname2") + "/": "bar",
subresourceServiceProxyURL("http", "portname2") + "/": "bar",
subresourceServiceProxyURL("https", "tlsportname1") + "/": "tls baz",
subresourceServiceProxyURL("https", "tlsportname2") + "/": "tls qux",
podProxyURL("", "80") + "/": `<a href="` + podProxyURL("", "80") + `/rewriteme">test</a>`,
podProxyURL("", "160") + "/": "foo",
podProxyURL("", "162") + "/": "bar",
podProxyURL("http", "80") + "/": `<a href="` + podProxyURL("http", "80") + `/rewriteme">test</a>`,
podProxyURL("http", "160") + "/": "foo",
podProxyURL("http", "162") + "/": "bar",
subresourcePodProxyURL("", "") + "/": `<a href="` + subresourcePodProxyURL("", "") + `/rewriteme">test</a>`,
subresourcePodProxyURL("", "80") + "/": `<a href="` + subresourcePodProxyURL("", "80") + `/rewriteme">test</a>`,
subresourcePodProxyURL("http", "80") + "/": `<a href="` + subresourcePodProxyURL("http", "80") + `/rewriteme">test</a>`,
subresourcePodProxyURL("", "160") + "/": "foo",
subresourcePodProxyURL("http", "160") + "/": "foo",
subresourcePodProxyURL("", "162") + "/": "bar",
subresourcePodProxyURL("http", "162") + "/": "bar",
subresourcePodProxyURL("https", "443") + "/": `<a href="` + subresourcePodProxyURL("https", "443") + `/tlsrewriteme">test</a>`,
subresourcePodProxyURL("https", "460") + "/": "tls baz",
subresourcePodProxyURL("https", "462") + "/": "tls qux",
// TODO: below entries don't work, but I believe we should make them work.
// podPrefix + ":dest1": "foo",
// podPrefix + ":dest2": "bar",
}
wg := sync.WaitGroup{}
errors := []string{}
errLock := sync.Mutex{}
recordError := func(s string) {
errLock.Lock()
defer errLock.Unlock()
errors = append(errors, s)
}
for i := 0; i < proxyAttempts; i++ {
for path, val := range expectations {
wg.Add(1)
go func(i int, path, val string) {
defer wg.Done()
body, status, d, err := doProxy(f, path)
if err != nil {
recordError(fmt.Sprintf("%v: path %v gave error: %v", i, path, err))
return
}
if status != http.StatusOK {
recordError(fmt.Sprintf("%v: path %v gave status: %v", i, path, status))
}
if e, a := val, string(body); e != a {
recordError(fmt.Sprintf("%v: path %v: wanted %v, got %v", i, path, e, a))
}
if d > proxyHTTPCallTimeout {
recordError(fmt.Sprintf("%v: path %v took %v > %v", i, path, d, proxyHTTPCallTimeout))
}
}(i, path, val)
// default QPS is 5
time.Sleep(200 * time.Millisecond)
}
}
wg.Wait()
if len(errors) != 0 {
Fail(strings.Join(errors, "\n"))
}
})
}