github/kubernetes
4
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

daemonset: Implement MaxSurge on daemonset update

Browse Source 
If MaxSurge is set, the controller will attempt to double up nodes
up to the allowed limit with a new pod, and then when the most recent
(by hash) pod is ready, trigger deletion on the old pod. If the old
pod goes unready before the new pod is ready, the old pod is immediately
deleted. If an old pod goes unready before a new pod is placed on that
node, a new pod is immediately added for that node even past the MaxSurge
limit.

The backoff clock is used consistently throughout the daemonset controller
as an injectable clock for the purposes of testing.
This commit is contained in:
Clayton Coleman
2021-01-27 00:20:56 -05:00
parent 6bac5019aa
commit 18f43e4120
6 changed files with 1190 additions and 68 deletions
Download Patch File Download Diff File Expand all files Collapse all files

351
pkg/controller/daemon/update_test.go
View File

@@ -18,12 +18,20 @@ package daemon
import (
"testing"
"time"
apps "k8s.io/api/apps/v1"
v1 "k8s.io/api/core/v1"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/labels"
"k8s.io/apimachinery/pkg/util/clock"
"k8s.io/apimachinery/pkg/util/intstr"
utilfeature "k8s.io/apiserver/pkg/util/feature"
featuregatetesting "k8s.io/component-base/featuregate/testing"
"k8s.io/klog/v2"
podutil "k8s.io/kubernetes/pkg/api/v1/pod"
"k8s.io/kubernetes/pkg/controller/daemon/util"
"k8s.io/kubernetes/pkg/features"
)
func TestDaemonSetUpdatesPods(t *testing.T) {
@@ -67,6 +75,48 @@ func TestDaemonSetUpdatesPods(t *testing.T) {
clearExpectations(t, manager, ds, podControl)
}
func TestDaemonSetUpdatesPodsWithMaxSurge(t *testing.T) {
defer featuregatetesting.SetFeatureGateDuringTest(t, utilfeature.DefaultFeatureGate, features.DaemonSetUpdateSurge, true)()
ds := newDaemonSet("foo")
manager, podControl, _, err := newTestController(ds)
if err != nil {
t.Fatalf("error creating DaemonSets controller: %v", err)
}
addNodes(manager.nodeStore, 0, 5, nil)
manager.dsStore.Add(ds)
expectSyncDaemonSets(t, manager, ds, podControl, 5, 0, 0)
markPodsReady(podControl.podStore)
// surge is thhe controlling amount
maxSurge := 2
ds.Spec.Template.Spec.Containers[0].Image = "foo2/bar2"
ds.Spec.UpdateStrategy = newUpdateSurge(intstr.FromInt(maxSurge))
manager.dsStore.Update(ds)
clearExpectations(t, manager, ds, podControl)
expectSyncDaemonSets(t, manager, ds, podControl, maxSurge, 0, 0)
clearExpectations(t, manager, ds, podControl)
expectSyncDaemonSets(t, manager, ds, podControl, 0, 0, 0)
markPodsReady(podControl.podStore)
clearExpectations(t, manager, ds, podControl)
expectSyncDaemonSets(t, manager, ds, podControl, maxSurge, maxSurge, 0)
clearExpectations(t, manager, ds, podControl)
expectSyncDaemonSets(t, manager, ds, podControl, 0, 0, 0)
markPodsReady(podControl.podStore)
clearExpectations(t, manager, ds, podControl)
expectSyncDaemonSets(t, manager, ds, podControl, 5%maxSurge, maxSurge, 0)
clearExpectations(t, manager, ds, podControl)
expectSyncDaemonSets(t, manager, ds, podControl, 0, 0, 0)
markPodsReady(podControl.podStore)
clearExpectations(t, manager, ds, podControl)
expectSyncDaemonSets(t, manager, ds, podControl, 0, 5%maxSurge, 0)
clearExpectations(t, manager, ds, podControl)
expectSyncDaemonSets(t, manager, ds, podControl, 0, 0, 0)
}
func TestDaemonSetUpdatesWhenNewPosIsNotReady(t *testing.T) {
ds := newDaemonSet("foo")
manager, podControl, _, err := newTestController(ds)
@@ -138,6 +188,149 @@ func TestDaemonSetUpdatesAllOldPodsNotReady(t *testing.T) {
clearExpectations(t, manager, ds, podControl)
}
func TestDaemonSetUpdatesAllOldPodsNotReadyMaxSurge(t *testing.T) {
defer featuregatetesting.SetFeatureGateDuringTest(t, utilfeature.DefaultFeatureGate, features.DaemonSetUpdateSurge, true)()
ds := newDaemonSet("foo")
manager, podControl, _, err := newTestController(ds)
if err != nil {
t.Fatalf("error creating DaemonSets controller: %v", err)
}
addNodes(manager.nodeStore, 0, 5, nil)
manager.dsStore.Add(ds)
expectSyncDaemonSets(t, manager, ds, podControl, 5, 0, 0)
maxSurge := 3
ds.Spec.Template.Spec.Containers[0].Image = "foo2/bar2"
ds.Spec.UpdateStrategy = newUpdateSurge(intstr.FromInt(maxSurge))
manager.dsStore.Update(ds)
// all old pods are unavailable so should be surged
manager.failedPodsBackoff.Clock = clock.NewFakeClock(time.Unix(100, 0))
clearExpectations(t, manager, ds, podControl)
expectSyncDaemonSets(t, manager, ds, podControl, 5, 0, 0)
// waiting for pods to go ready, old pods are deleted
manager.failedPodsBackoff.Clock = clock.NewFakeClock(time.Unix(200, 0))
clearExpectations(t, manager, ds, podControl)
expectSyncDaemonSets(t, manager, ds, podControl, 0, 5, 0)
setPodReadiness(t, manager, true, 5, func(_ *v1.Pod) bool { return true })
ds.Spec.MinReadySeconds = 15
ds.Spec.Template.Spec.Containers[0].Image = "foo3/bar3"
manager.dsStore.Update(ds)
manager.failedPodsBackoff.Clock = clock.NewFakeClock(time.Unix(300, 0))
clearExpectations(t, manager, ds, podControl)
expectSyncDaemonSets(t, manager, ds, podControl, 3, 0, 0)
hash, err := currentDSHash(manager, ds)
if err != nil {
t.Fatal(err)
}
currentPods := podsByNodeMatchingHash(manager, hash)
// mark two updated pods as ready at time 300
setPodReadiness(t, manager, true, 2, func(pod *v1.Pod) bool {
return pod.Labels[apps.ControllerRevisionHashLabelKey] == hash
})
// mark one of the old pods that is on a node without an updated pod as unready
setPodReadiness(t, manager, false, 1, func(pod *v1.Pod) bool {
nodeName, err := util.GetTargetNodeName(pod)
if err != nil {
t.Fatal(err)
}
return pod.Labels[apps.ControllerRevisionHashLabelKey] != hash && len(currentPods[nodeName]) == 0
})
// the new pods should still be considered waiting to hit min readiness, so one pod should be created to replace
// the deleted old pod
manager.failedPodsBackoff.Clock = clock.NewFakeClock(time.Unix(310, 0))
clearExpectations(t, manager, ds, podControl)
expectSyncDaemonSets(t, manager, ds, podControl, 1, 0, 0)
// the new pods are now considered available, so delete the old pods
manager.failedPodsBackoff.Clock = clock.NewFakeClock(time.Unix(320, 0))
clearExpectations(t, manager, ds, podControl)
expectSyncDaemonSets(t, manager, ds, podControl, 1, 3, 0)
// mark all updated pods as ready at time 320
currentPods = podsByNodeMatchingHash(manager, hash)
setPodReadiness(t, manager, true, 3, func(pod *v1.Pod) bool {
return pod.Labels[apps.ControllerRevisionHashLabelKey] == hash
})
// the new pods are now considered available, so delete the old pods
manager.failedPodsBackoff.Clock = clock.NewFakeClock(time.Unix(340, 0))
clearExpectations(t, manager, ds, podControl)
expectSyncDaemonSets(t, manager, ds, podControl, 0, 2, 0)
// controller has completed upgrade
manager.failedPodsBackoff.Clock = clock.NewFakeClock(time.Unix(350, 0))
clearExpectations(t, manager, ds, podControl)
expectSyncDaemonSets(t, manager, ds, podControl, 0, 0, 0)
}
func podsByNodeMatchingHash(dsc *daemonSetsController, hash string) map[string][]string {
byNode := make(map[string][]string)
for _, obj := range dsc.podStore.List() {
pod := obj.(*v1.Pod)
if pod.Labels[apps.ControllerRevisionHashLabelKey] != hash {
continue
}
nodeName, err := util.GetTargetNodeName(pod)
if err != nil {
panic(err)
}
byNode[nodeName] = append(byNode[nodeName], pod.Name)
}
return byNode
}
func setPodReadiness(t *testing.T, dsc *daemonSetsController, ready bool, count int, fn func(*v1.Pod) bool) {
t.Helper()
for _, obj := range dsc.podStore.List() {
if count <= 0 {
break
}
pod := obj.(*v1.Pod)
if pod.DeletionTimestamp != nil {
continue
}
if podutil.IsPodReady(pod) == ready {
continue
}
if !fn(pod) {
continue
}
condition := v1.PodCondition{Type: v1.PodReady}
if ready {
condition.Status = v1.ConditionTrue
} else {
condition.Status = v1.ConditionFalse
}
if !podutil.UpdatePodCondition(&pod.Status, &condition) {
t.Fatal("failed to update pod")
}
// TODO: workaround UpdatePodCondition calling time.Now() directly
setCondition := podutil.GetPodReadyCondition(pod.Status)
setCondition.LastTransitionTime.Time = dsc.failedPodsBackoff.Clock.Now()
klog.Infof("marked pod %s ready=%t", pod.Name, ready)
count--
}
if count > 0 {
t.Fatalf("could not mark %d pods ready=%t", count, ready)
}
}
func currentDSHash(dsc *daemonSetsController, ds *apps.DaemonSet) (string, error) {
// Construct histories of the DaemonSet, and get the hash of current history
cur, _, err := dsc.constructHistory(ds)
if err != nil {
return "", err
}
return cur.Labels[apps.DefaultDaemonSetUniqueLabelKey], nil
}
func TestDaemonSetUpdatesNoTemplateChanged(t *testing.T) {
ds := newDaemonSet("foo")
manager, podControl, _, err := newTestController(ds)
@@ -163,12 +356,34 @@ func TestDaemonSetUpdatesNoTemplateChanged(t *testing.T) {
clearExpectations(t, manager, ds, podControl)
}
func newUpdateSurge(value intstr.IntOrString) apps.DaemonSetUpdateStrategy {
zero := intstr.FromInt(0)
return apps.DaemonSetUpdateStrategy{
Type: apps.RollingUpdateDaemonSetStrategyType,
RollingUpdate: &apps.RollingUpdateDaemonSet{
MaxUnavailable: &zero,
MaxSurge: &value,
},
}
}
func newUpdateUnavailable(value intstr.IntOrString) apps.DaemonSetUpdateStrategy {
return apps.DaemonSetUpdateStrategy{
Type: apps.RollingUpdateDaemonSetStrategyType,
RollingUpdate: &apps.RollingUpdateDaemonSet{
MaxUnavailable: &value,
},
}
}
func TestGetUnavailableNumbers(t *testing.T) {
cases := []struct {
name string
Manager *daemonSetsController
ds *apps.DaemonSet
nodeToPods map[string][]*v1.Pod
enableSurge bool
maxSurge int
maxUnavailable int
numUnavailable int
Err error
@@ -184,8 +399,7 @@ func TestGetUnavailableNumbers(t *testing.T) {
}(),
ds: func() *apps.DaemonSet {
ds := newDaemonSet("x")
intStr := intstr.FromInt(0)
ds.Spec.UpdateStrategy.RollingUpdate = &apps.RollingUpdateDaemonSet{MaxUnavailable: &intStr}
ds.Spec.UpdateStrategy = newUpdateUnavailable(intstr.FromInt(0))
return ds
}(),
nodeToPods: make(map[string][]*v1.Pod),
@@ -204,8 +418,7 @@ func TestGetUnavailableNumbers(t *testing.T) {
}(),
ds: func() *apps.DaemonSet {
ds := newDaemonSet("x")
intStr := intstr.FromInt(1)
ds.Spec.UpdateStrategy.RollingUpdate = &apps.RollingUpdateDaemonSet{MaxUnavailable: &intStr}
ds.Spec.UpdateStrategy = newUpdateUnavailable(intstr.FromInt(1))
return ds
}(),
nodeToPods: func() map[string][]*v1.Pod {
@@ -233,8 +446,7 @@ func TestGetUnavailableNumbers(t *testing.T) {
}(),
ds: func() *apps.DaemonSet {
ds := newDaemonSet("x")
intStr := intstr.FromInt(0)
ds.Spec.UpdateStrategy.RollingUpdate = &apps.RollingUpdateDaemonSet{MaxUnavailable: &intStr}
ds.Spec.UpdateStrategy = newUpdateUnavailable(intstr.FromInt(0))
return ds
}(),
nodeToPods: func() map[string][]*v1.Pod {
@@ -244,7 +456,32 @@ func TestGetUnavailableNumbers(t *testing.T) {
mapping["node-0"] = []*v1.Pod{pod0}
return mapping
}(),
maxUnavailable: 0,
maxUnavailable: 1,
numUnavailable: 1,
},
{
name: "Two nodes, one node without pods, surge",
Manager: func() *daemonSetsController {
manager, _, _, err := newTestController()
if err != nil {
t.Fatalf("error creating DaemonSets controller: %v", err)
}
addNodes(manager.nodeStore, 0, 2, nil)
return manager
}(),
ds: func() *apps.DaemonSet {
ds := newDaemonSet("x")
ds.Spec.UpdateStrategy = newUpdateSurge(intstr.FromInt(0))
return ds
}(),
nodeToPods: func() map[string][]*v1.Pod {
mapping := make(map[string][]*v1.Pod)
pod0 := newPod("pod-0", "node-0", simpleDaemonSetLabel, nil)
markPodReady(pod0)
mapping["node-0"] = []*v1.Pod{pod0}
return mapping
}(),
maxUnavailable: 1,
numUnavailable: 1,
},
{
@@ -259,8 +496,7 @@ func TestGetUnavailableNumbers(t *testing.T) {
}(),
ds: func() *apps.DaemonSet {
ds := newDaemonSet("x")
intStr := intstr.FromString("50%")
ds.Spec.UpdateStrategy.RollingUpdate = &apps.RollingUpdateDaemonSet{MaxUnavailable: &intStr}
ds.Spec.UpdateStrategy = newUpdateUnavailable(intstr.FromString("50%"))
return ds
}(),
nodeToPods: func() map[string][]*v1.Pod {
@@ -276,6 +512,66 @@ func TestGetUnavailableNumbers(t *testing.T) {
maxUnavailable: 1,
numUnavailable: 0,
},
{
name: "Two nodes with pods, MaxUnavailable in percents, surge",
Manager: func() *daemonSetsController {
manager, _, _, err := newTestController()
if err != nil {
t.Fatalf("error creating DaemonSets controller: %v", err)
}
addNodes(manager.nodeStore, 0, 2, nil)
return manager
}(),
ds: func() *apps.DaemonSet {
ds := newDaemonSet("x")
ds.Spec.UpdateStrategy = newUpdateSurge(intstr.FromString("50%"))
return ds
}(),
nodeToPods: func() map[string][]*v1.Pod {
mapping := make(map[string][]*v1.Pod)
pod0 := newPod("pod-0", "node-0", simpleDaemonSetLabel, nil)
pod1 := newPod("pod-1", "node-1", simpleDaemonSetLabel, nil)
markPodReady(pod0)
markPodReady(pod1)
mapping["node-0"] = []*v1.Pod{pod0}
mapping["node-1"] = []*v1.Pod{pod1}
return mapping
}(),
enableSurge: true,
maxSurge: 1,
maxUnavailable: 0,
numUnavailable: 0,
},
{
name: "Two nodes with pods, MaxUnavailable is 100%, surge",
Manager: func() *daemonSetsController {
manager, _, _, err := newTestController()
if err != nil {
t.Fatalf("error creating DaemonSets controller: %v", err)
}
addNodes(manager.nodeStore, 0, 2, nil)
return manager
}(),
ds: func() *apps.DaemonSet {
ds := newDaemonSet("x")
ds.Spec.UpdateStrategy = newUpdateSurge(intstr.FromString("100%"))
return ds
}(),
nodeToPods: func() map[string][]*v1.Pod {
mapping := make(map[string][]*v1.Pod)
pod0 := newPod("pod-0", "node-0", simpleDaemonSetLabel, nil)
pod1 := newPod("pod-1", "node-1", simpleDaemonSetLabel, nil)
markPodReady(pod0)
markPodReady(pod1)
mapping["node-0"] = []*v1.Pod{pod0}
mapping["node-1"] = []*v1.Pod{pod1}
return mapping
}(),
enableSurge: true,
maxSurge: 2,
maxUnavailable: 0,
numUnavailable: 0,
},
{
name: "Two nodes with pods, MaxUnavailable in percents, pod terminating",
Manager: func() *daemonSetsController {
@@ -288,8 +584,7 @@ func TestGetUnavailableNumbers(t *testing.T) {
}(),
ds: func() *apps.DaemonSet {
ds := newDaemonSet("x")
intStr := intstr.FromString("50%")
ds.Spec.UpdateStrategy.RollingUpdate = &apps.RollingUpdateDaemonSet{MaxUnavailable: &intStr}
ds.Spec.UpdateStrategy = newUpdateUnavailable(intstr.FromString("50%"))
return ds
}(),
nodeToPods: func() map[string][]*v1.Pod {
@@ -310,20 +605,26 @@ func TestGetUnavailableNumbers(t *testing.T) {
}
for _, c := range cases {
c.Manager.dsStore.Add(c.ds)
nodeList, err := c.Manager.nodeLister.List(labels.Everything())
if err != nil {
t.Fatalf("error listing nodes: %v", err)
}
maxUnavailable, numUnavailable, err := c.Manager.getUnavailableNumbers(c.ds, nodeList, c.nodeToPods)
if err != nil && c.Err != nil {
if c.Err != err {
t.Errorf("Test case: %s. Expected error: %v but got: %v", c.name, c.Err, err)
t.Run(c.name, func(t *testing.T) {
defer featuregatetesting.SetFeatureGateDuringTest(t, utilfeature.DefaultFeatureGate, features.DaemonSetUpdateSurge, c.enableSurge)()
c.Manager.dsStore.Add(c.ds)
nodeList, err := c.Manager.nodeLister.List(labels.Everything())
if err != nil {
t.Fatalf("error listing nodes: %v", err)
}
} else if err != nil {
t.Errorf("Test case: %s. Unexpected error: %v", c.name, err)
} else if maxUnavailable != c.maxUnavailable || numUnavailable != c.numUnavailable {
t.Errorf("Test case: %s. Wrong values. maxUnavailable: %d, expected: %d, numUnavailable: %d. expected: %d", c.name, maxUnavailable, c.maxUnavailable, numUnavailable, c.numUnavailable)
}
maxSurge, maxUnavailable, numUnavailable, err := c.Manager.getUnavailableNumbers(c.ds, nodeList, c.nodeToPods)
if err != nil && c.Err != nil {
if c.Err != err {
t.Fatalf("Expected error: %v but got: %v", c.Err, err)
}
}
if err != nil {
t.Fatalf("Unexpected error: %v", err)
}
if maxSurge != c.maxSurge || maxUnavailable != c.maxUnavailable || numUnavailable != c.numUnavailable {
t.Fatalf("Wrong values. maxSurge: %d, expected %d, maxUnavailable: %d, expected: %d, numUnavailable: %d. expected: %d", maxSurge, c.maxSurge, maxUnavailable, c.maxUnavailable, numUnavailable, c.numUnavailable)
}
})
}
}