when pleg channel is full, discard events and record how many events discard

This commit is contained in:
changyaowei 2019-01-30 20:43:54 +08:00
parent af31a202ea
commit b52afc350f
3 changed files with 116 additions and 8 deletions

View File

@ -36,6 +36,7 @@ const (
CgroupManagerOperationsKey = "cgroup_manager_latency_microseconds"
PodWorkerStartLatencyKey = "pod_worker_start_latency_microseconds"
PLEGRelistLatencyKey = "pleg_relist_latency_microseconds"
PLEGDiscardEventsKey = "pleg_discard_events"
PLEGRelistIntervalKey = "pleg_relist_interval_microseconds"
EvictionStatsAgeKey = "eviction_stats_age_microseconds"
VolumeStatsCapacityBytesKey = "volume_stats_capacity_bytes"
@ -109,6 +110,14 @@ var (
Help: "Latency in microseconds for relisting pods in PLEG.",
},
)
PLEGDiscardEvents = prometheus.NewCounterVec(
prometheus.CounterOpts{
Subsystem: KubeletSubsystem,
Name: PLEGDiscardEventsKey,
Help: "The number of discard events in PLEG.",
},
[]string{},
)
PLEGRelistInterval = prometheus.NewSummary(
prometheus.SummaryOpts{
Subsystem: KubeletSubsystem,
@ -214,6 +223,7 @@ func Register(containerCache kubecontainer.RuntimeCache, collectors ...prometheu
prometheus.MustRegister(ContainersPerPodCount)
prometheus.MustRegister(newPodAndContainerCollector(containerCache))
prometheus.MustRegister(PLEGRelistLatency)
prometheus.MustRegister(PLEGDiscardEvents)
prometheus.MustRegister(PLEGRelistInterval)
prometheus.MustRegister(RuntimeOperations)
prometheus.MustRegister(RuntimeOperationsLatency)

View File

@ -265,7 +265,12 @@ func (g *GenericPLEG) relist() {
if events[i].Type == ContainerChanged {
continue
}
g.eventChannel <- events[i]
select {
case g.eventChannel <- events[i]:
default:
metrics.PLEGDiscardEvents.WithLabelValues().Inc()
klog.Error("event channel is full, discard this relist() cycle event")
}
}
}

View File

@ -30,10 +30,16 @@ import (
"k8s.io/apimachinery/pkg/util/diff"
kubecontainer "k8s.io/kubernetes/pkg/kubelet/container"
containertest "k8s.io/kubernetes/pkg/kubelet/container/testing"
"github.com/prometheus/client_golang/prometheus"
"net"
"net/http"
"k8s.io/kubernetes/pkg/kubelet/metrics"
)
const (
testContainerRuntimeType = "fooRuntime"
// largeChannelCap is a large enough capacity to hold all events in a single test.
largeChannelCap = 100
)
type TestGenericPLEG struct {
@ -42,7 +48,11 @@ type TestGenericPLEG struct {
clock *clock.FakeClock
}
func newTestGenericPLEG() *TestGenericPLEG {
func newTestGenericPLEGWithLargeChannel() *TestGenericPLEG {
return newTestGenericPLEG(largeChannelCap)
}
func newTestGenericPLEG(eventChannelCap int) *TestGenericPLEG {
fakeRuntime := &containertest.FakeRuntime{}
clock := clock.NewFakeClock(time.Time{})
// The channel capacity should be large enough to hold all events in a
@ -50,7 +60,7 @@ func newTestGenericPLEG() *TestGenericPLEG {
pleg := &GenericPLEG{
relistPeriod: time.Hour,
runtime: fakeRuntime,
eventChannel: make(chan *PodLifecycleEvent, 100),
eventChannel: make(chan *PodLifecycleEvent, eventChannelCap),
podRecords: make(podRecords),
clock: clock,
}
@ -93,7 +103,7 @@ func verifyEvents(t *testing.T, expected, actual []*PodLifecycleEvent) {
}
func TestRelisting(t *testing.T) {
testPleg := newTestGenericPLEG()
testPleg := newTestGenericPLEGWithLargeChannel()
pleg, runtime := testPleg.pleg, testPleg.runtime
ch := pleg.Watch()
// The first relist should send a PodSync event to each pod.
@ -157,6 +167,89 @@ func TestRelisting(t *testing.T) {
verifyEvents(t, expected, actual)
}
// TestEventChannelFull test when channel is full, the events will be discard.
func TestEventChannelFull(t *testing.T) {
prometheus.MustRegister(metrics.PLEGDiscardEvents)
temporalServer := "127.0.0.1:1234"
l, err := net.Listen("tcp", temporalServer)
assert.NoError(t, err)
defer l.Close()
prometheusUrl := "http://" + temporalServer + "/metrics"
mux := http.NewServeMux()
mux.Handle("/metrics", prometheus.Handler())
server := &http.Server{
Addr: temporalServer,
Handler: mux,
}
go func() {
server.Serve(l)
}()
testPleg := newTestGenericPLEG(4)
pleg, runtime := testPleg.pleg, testPleg.runtime
ch := pleg.Watch()
// The first relist should send a PodSync event to each pod.
runtime.AllPodList = []*containertest.FakePod{
{Pod: &kubecontainer.Pod{
ID: "1234",
Containers: []*kubecontainer.Container{
createTestContainer("c1", kubecontainer.ContainerStateExited),
createTestContainer("c2", kubecontainer.ContainerStateRunning),
createTestContainer("c3", kubecontainer.ContainerStateUnknown),
},
}},
{Pod: &kubecontainer.Pod{
ID: "4567",
Containers: []*kubecontainer.Container{
createTestContainer("c1", kubecontainer.ContainerStateExited),
},
}},
}
pleg.relist()
// Report every running/exited container if we see them for the first time.
expected := []*PodLifecycleEvent{
{ID: "1234", Type: ContainerStarted, Data: "c2"},
{ID: "4567", Type: ContainerDied, Data: "c1"},
{ID: "1234", Type: ContainerDied, Data: "c1"},
}
actual := getEventsFromChannel(ch)
verifyEvents(t, expected, actual)
runtime.AllPodList = []*containertest.FakePod{
{Pod: &kubecontainer.Pod{
ID: "1234",
Containers: []*kubecontainer.Container{
createTestContainer("c2", kubecontainer.ContainerStateExited),
createTestContainer("c3", kubecontainer.ContainerStateRunning),
},
}},
{Pod: &kubecontainer.Pod{
ID: "4567",
Containers: []*kubecontainer.Container{
createTestContainer("c4", kubecontainer.ContainerStateRunning),
},
}},
}
pleg.relist()
// event channel is full, discard events
expected = []*PodLifecycleEvent{
{ID: "1234", Type: ContainerRemoved, Data: "c1"},
{ID: "1234", Type: ContainerDied, Data: "c2"},
{ID: "1234", Type: ContainerStarted, Data: "c3"},
{ID: "4567", Type: ContainerRemoved, Data: "c1"},
}
actual = getEventsFromChannel(ch)
verifyEvents(t, expected, actual)
plegEventsDiscardCounterExpected := "pleg_discard_events 1"
assert.HTTPBodyContains(t, http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
mux.ServeHTTP(w, r)
}), "GET", prometheusUrl, nil, plegEventsDiscardCounterExpected)
}
func TestDetectingContainerDeaths(t *testing.T) {
// Vary the number of relists after the container started and before the
// container died to account for the changes in pleg's internal states.
@ -168,7 +261,7 @@ func TestDetectingContainerDeaths(t *testing.T) {
}
func testReportMissingContainers(t *testing.T, numRelists int) {
testPleg := newTestGenericPLEG()
testPleg := newTestGenericPLEGWithLargeChannel()
pleg, runtime := testPleg.pleg, testPleg.runtime
ch := pleg.Watch()
runtime.AllPodList = []*containertest.FakePod{
@ -209,7 +302,7 @@ func testReportMissingContainers(t *testing.T, numRelists int) {
}
func testReportMissingPods(t *testing.T, numRelists int) {
testPleg := newTestGenericPLEG()
testPleg := newTestGenericPLEGWithLargeChannel()
pleg, runtime := testPleg.pleg, testPleg.runtime
ch := pleg.Watch()
runtime.AllPodList = []*containertest.FakePod{
@ -345,7 +438,7 @@ func TestRemoveCacheEntry(t *testing.T) {
}
func TestHealthy(t *testing.T) {
testPleg := newTestGenericPLEG()
testPleg := newTestGenericPLEGWithLargeChannel()
// pleg should initially be unhealthy
pleg, _, clock := testPleg.pleg, testPleg.runtime, testPleg.clock
@ -441,7 +534,7 @@ func TestRelistWithReinspection(t *testing.T) {
// Test detecting sandbox state changes.
func TestRelistingWithSandboxes(t *testing.T) {
testPleg := newTestGenericPLEG()
testPleg := newTestGenericPLEGWithLargeChannel()
pleg, runtime := testPleg.pleg, testPleg.runtime
ch := pleg.Watch()
// The first relist should send a PodSync event to each pod.