Skip to content

Decouple Cordoning from Draining to Avoid Uncordoning User-Cordoned Nodes in Preservation Flow #1119

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
thiyyakat wants to merge 9 commits into
base: master
Choose a base branch
from
Open

Decouple Cordoning from Draining to Avoid Uncordoning User-Cordoned Nodes in Preservation Flow #1119

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
44e82f9
De-couple cordoning and drain
thiyyakat Jul 9, 2026
ea9f84f
Address reviewer comments
thiyyakat Jul 16, 2026
07849d2
Address reviewer comments - part 2
thiyyakat Jul 17, 2026
fd46abb
Address reviewer comments - part 3
thiyyakat Jul 17, 2026
d2f925a
Address reviewer comments - part 4
thiyyakat Jul 17, 2026
117fb4b
Address review comments (5) and correct rebase
thiyyakat Jul 27, 2026
6850d21
Fix bug in `preserveMachine` that returns early when a machine preser…
thiyyakat Jul 27, 2026
fa15010
Fix comment for clarity.
thiyyakat Jul 28, 2026
0baf17e
Improve logs, and revert to original pattern of returning errors.
thiyyakat Jul 29, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 0 additions & 27 deletions pkg/util/provider/drain/drain.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -235,10 +235,6 @@ func (o *Options) RunDrain(ctx context.Context) error {
)
}()

if err := o.RunCordonOrUncordon(drainContext, true); err != nil {
klog.Errorf("Drain Error: Cordoning of node failed with error: %v", err)
return err
}
if !cache.WaitForCacheSync(drainContext.Done(), o.podSynced) {
err := fmt.Errorf("timed out waiting for pod cache to sync")
return err
Expand Down Expand Up @@ -1164,29 +1160,6 @@ func SupportEviction(clientset kubernetes.Interface) (string, error) {
return "", nil
}

// RunCordonOrUncordon runs either Cordon or Uncordon. The desired value for
// "Unschedulable" is passed as the first arg.
func (o *Options) RunCordonOrUncordon(ctx context.Context, desired bool) error {
node, err := o.client.CoreV1().Nodes().Get(ctx, o.nodeName, metav1.GetOptions{})
if err != nil {
// Deletion could be triggered when machine is just being created, no node present then
return nil
}
unsched := node.Spec.Unschedulable
if unsched == desired {
klog.V(3).Infof("Scheduling state for node %q is already in desired state", node.Name)
} else {
clone := node.DeepCopy()
clone.Spec.Unschedulable = desired

_, err = o.client.CoreV1().Nodes().Update(ctx, clone, metav1.UpdateOptions{})
if err != nil {
return err
}
}
return nil
}

func getPdbForPod(pdbLister policyv1listers.PodDisruptionBudgetLister, pod *corev1.Pod) *policyv1.PodDisruptionBudget {
// GetPodPodDisruptionBudgets returns an error only if no PodDisruptionBudgets are found.
// We don't return that as an error to the caller.
Expand Down
16 changes: 9 additions & 7 deletions pkg/util/provider/machinecontroller/machine.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,6 +13,7 @@ import (
"strings"
"time"

"github.com/gardener/machine-controller-manager/pkg/util/nodeops"
corev1 "k8s.io/api/core/v1"
apierrors "k8s.io/apimachinery/pkg/api/errors"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
Expand Down Expand Up @@ -806,6 +807,8 @@ func (c *controller) manageMachinePreservation(ctx context.Context, machine *v1a
var removeAnnotations bool
clone := machine.DeepCopy()
switch effectivePreserveValue {
// effectivePreserveValue == "" implies the preservation annotation was deleted to indicate that
// preservation must be stopped
case "", machineutils.PreserveMachineAnnotationValueFalse:
clone, err = c.stopPreservationIfActive(ctx, clone, removeAnnotations)
case machineutils.PreserveMachineAnnotationValueWhenFailed:
Expand Down Expand Up @@ -840,19 +843,18 @@ func (c *controller) manageMachinePreservation(ctx context.Context, machine *v1a
if err != nil {
return
}
// For the preserve=now path (preservation still active), uncordon the node if the machine
// has recovered to Running. For all other cases, stopPreservationIfActive path handles uncordon internally.
// For the preserve=now path (preservation still active), untaint the node if the machine
// has recovered to Running. For all other cases, stopPreservationIfActive path handles untainting internally.
if clone.Status.CurrentStatus.PreserveExpiryTime != nil && clone.Status.CurrentStatus.Phase == v1alpha1.MachineRunning && nodeName != "" {
var node *corev1.Node
node, err = c.nodeLister.Get(nodeName)
if err != nil {
if apierrors.IsNotFound(err) {
err = nil
return
}
return
}
err = c.uncordonNodeIfCordoned(ctx, node)
err = nodeops.RemoveTaintOffNode(ctx, c.targetCoreClient, node.Name, node, &corev1.Taint{
Key: machineutils.NodePreservedTaintKey,
Effect: corev1.TaintEffectNoSchedule,
})
if err != nil {
return
}
Expand Down
60 changes: 34 additions & 26 deletions pkg/util/provider/machinecontroller/machine_test.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,6 +10,7 @@ import (
"math"
"time"

taintutils "github.com/gardener/machine-controller-manager/pkg/util/taints"
k8stesting "k8s.io/client-go/testing"

. "github.com/onsi/ginkgo/v2"
Expand Down Expand Up @@ -4105,7 +4106,7 @@ var _ = Describe("machine", func() {
nodeName string
machinePhase v1alpha1.MachinePhase
preserveExpiryTime *metav1.Time
nodeUnschedulable bool
nodeTainted bool
}
type expect struct {
retry machineutils.RetryPeriod
Expand All @@ -4114,7 +4115,7 @@ var _ = Describe("machine", func() {
machineAnnotationValue string
laNodePreserveValue string
err error
nodeUnschedulable *bool
nodeTainted bool
}
type testCase struct {
setup setup
Expand Down Expand Up @@ -4165,12 +4166,20 @@ var _ = Describe("machine", func() {
Annotations: nodeAnnotations,
},
Spec: corev1.NodeSpec{
Unschedulable: tc.setup.nodeUnschedulable,
Taints: []corev1.Taint{},
},
Status: corev1.NodeStatus{
Conditions: []corev1.NodeCondition{},
},
}
if tc.setup.nodeTainted {
node.Spec.Taints = []corev1.Taint{
{
Key: machineutils.NodePreservedTaintKey,
Effect: corev1.TaintEffectNoSchedule,
},
}
}
targetCoreObjects = append(targetCoreObjects, node)
}
c, trackers := createController(stop, testNamespace, controlMachineObjects, nil, targetCoreObjects, nil, false)
Expand Down Expand Up @@ -4211,9 +4220,7 @@ var _ = Describe("machine", func() {
} else {
Expect(found).To(BeFalse())
}
if tc.expect.nodeUnschedulable != nil {
Expect(updatedNode.Spec.Unschedulable).To(Equal(*tc.expect.nodeUnschedulable))
}
Expect(taintutils.TaintExists(updatedNode.Spec.Taints, &corev1.Taint{Key: machineutils.NodePreservedTaintKey, Effect: corev1.TaintEffectNoSchedule})).To(Equal(tc.expect.nodeTainted))
}
Expect(updatedMachine.Annotations[machineutils.PreserveMachineAnnotationKey]).To(Equal(tc.expect.machineAnnotationValue))
Expect(updatedMachine.Annotations[machineutils.LastAppliedNodePreserveValueAnnotationKey]).To(Equal(tc.expect.laNodePreserveValue))
Expand Down Expand Up @@ -4271,7 +4278,8 @@ var _ = Describe("machine", func() {
nodeCondition: &corev1.NodeCondition{
Type: v1alpha1.NodePreserved,
Status: corev1.ConditionTrue},
retry: machineutils.LongRetry,
retry: machineutils.LongRetry,
nodeTainted: true,
},
}),
Entry("when node of preserved machine is annotated with preserve value 'false', should stop preservation", testCase{
Expand Down Expand Up @@ -4303,6 +4311,7 @@ var _ = Describe("machine", func() {
Status: corev1.ConditionTrue},
retry: machineutils.LongRetry,
machineAnnotationValue: machineutils.PreserveMachineAnnotationValueAutoPreserved,
nodeTainted: true,
},
}),
Entry("when machine is annotated with preserve=now and preservation times out, should stop preservation, and remove annotation", testCase{
Expand All @@ -4326,6 +4335,7 @@ var _ = Describe("machine", func() {
nodeName: "node-1",
machinePhase: v1alpha1.MachineFailed,
preserveExpiryTime: &metav1.Time{Time: metav1.Now().Add(-1 * time.Minute)},
nodeTainted: true,
},
expect: expect{
preserveExpiryTimeIsSet: false,
Expand Down Expand Up @@ -4414,11 +4424,13 @@ var _ = Describe("machine", func() {
nodeName: "node-1",
machinePhase: v1alpha1.MachineRunning,
preserveExpiryTime: &metav1.Time{Time: metav1.Now().Add(1 * time.Hour)},
nodeTainted: true,
},
expect: expect{
preserveExpiryTimeIsSet: false,
nodeCondition: &corev1.NodeCondition{Type: v1alpha1.NodePreserved, Status: corev1.ConditionFalse},
machineAnnotationValue: "",
nodeTainted: false,
retry: machineutils.LongRetry,
err: nil,
},
Expand Down Expand Up @@ -4462,81 +4474,77 @@ var _ = Describe("machine", func() {
err: nil,
},
}),
Entry("when preserve=when-failed machine with cordoned node recovers to Running, should stop preservation and uncordon node", func() testCase {
uncordoned := false
Entry("when preserved machine with preserve=when-failed recovers to Running, should stop preservation and untaint node", func() testCase {
return testCase{
setup: setup{
machineAnnotationValue: machineutils.PreserveMachineAnnotationValueWhenFailed,
machineAnnotated: true,
nodeName: "node-1",
machinePhase: v1alpha1.MachineRunning,
preserveExpiryTime: &metav1.Time{Time: metav1.Now().Add(1 * time.Hour)},
nodeUnschedulable: true,
nodeTainted: true,
},
expect: expect{
preserveExpiryTimeIsSet: false,
nodeCondition: &corev1.NodeCondition{Type: v1alpha1.NodePreserved, Status: corev1.ConditionFalse},
machineAnnotationValue: machineutils.PreserveMachineAnnotationValueWhenFailed,
retry: machineutils.LongRetry,
nodeUnschedulable: &uncordoned,
nodeTainted: false,
},
}
}()),
Entry("when preserve=when-failed machine with cordoned node has expired preservation while Running, should stop preservation and uncordon node", func() testCase {
uncordoned := false
Entry("when preserved machine with preserve=when-failed has expired preservation while in Failed phase, should stop preservation and untaint node", func() testCase {
return testCase{
setup: setup{
machineAnnotationValue: machineutils.PreserveMachineAnnotationValueWhenFailed,
machineAnnotated: true,
nodeName: "node-1",
machinePhase: v1alpha1.MachineRunning,
machinePhase: v1alpha1.MachineFailed,
preserveExpiryTime: &metav1.Time{Time: metav1.Now().Add(-1 * time.Minute)},
nodeUnschedulable: true,
nodeTainted: true,
},
expect: expect{
preserveExpiryTimeIsSet: false,
nodeCondition: &corev1.NodeCondition{Type: v1alpha1.NodePreserved, Status: corev1.ConditionFalse},
machineAnnotationValue: "",
retry: machineutils.LongRetry,
nodeUnschedulable: &uncordoned,
nodeTainted: false,
},
}
}()),
Entry("when preserve=when-failed machine is Failed with active preservation and cordoned node, node should stay cordoned", func() testCase {
cordoned := true
Entry("when preserve=when-failed machine is in Failed phase with active preservation and tainted node, node should stay tainted", func() testCase {
return testCase{
setup: setup{
nodeAnnotationValue: machineutils.PreserveMachineAnnotationValueWhenFailed,
nodeAnnotated: true,
nodeName: "node-1",
machinePhase: v1alpha1.MachineFailed,
nodeUnschedulable: true,
nodeTainted: true,
},
expect: expect{
laNodePreserveValue: machineutils.PreserveMachineAnnotationValueWhenFailed,
preserveExpiryTimeIsSet: true,
nodeCondition: &corev1.NodeCondition{Type: v1alpha1.NodePreserved, Status: corev1.ConditionTrue},
retry: machineutils.LongRetry,
nodeUnschedulable: &cordoned,
nodeTainted: true,
},
}
}()),
Entry("when preserve=now machine is Failed with active preservation and cordoned node, node should stay cordoned", func() testCase {
cordoned := true
Entry("when preserve=now machine is Failed with active preservation and tainted node, node should stay tainted", func() testCase {
return testCase{
setup: setup{
machineAnnotationValue: machineutils.PreserveMachineAnnotationValueNow,
machineAnnotated: true,
nodeName: "node-1",
machinePhase: v1alpha1.MachineFailed,
nodeUnschedulable: true,
nodeTainted: true,
},
expect: expect{
preserveExpiryTimeIsSet: true,
nodeCondition: &corev1.NodeCondition{Type: v1alpha1.NodePreserved, Status: corev1.ConditionTrue},
machineAnnotationValue: machineutils.PreserveMachineAnnotationValueNow,
retry: machineutils.LongRetry,
nodeUnschedulable: &cordoned,
nodeTainted: true,
},
}
}()),
Expand Down Expand Up @@ -4588,8 +4596,8 @@ var _ = Describe("machine", func() {
Expect(updatedNode.Spec.Unschedulable).To(Equal(nodeUnschedulable))
}
},
Entry("node is cordoned", true, true),
Entry("node is uncordoned", false, true),
Entry("node is tainted", true, true),
Entry("node is untainted", false, true),
Entry("node retrieval fails (node not found)", false, false),
)
})
Expand Down
Loading
Loading