Add request-evict-only annotation to virt-launcher pods

Add the `descheduler.alpha.kubernetes.io/request-evict-only` annotation to virt-launcher pods. This annotation should be added to already running pods too. ref: kubevirt/community#258 Signed-off-by: fossedihelm <[email protected]>
fossedihelm · Jun 6, 2024 · 9f109a3 · 9f109a3
1 parent a1c1c67
commit 9f109a3
Show file tree

Hide file tree

Showing 5 changed files with 18 additions and 8 deletions.
diff --git a/pkg/virt-controller/services/template.go b/pkg/virt-controller/services/template.go
@@ -1359,6 +1359,8 @@ func generatePodAnnotations(vmi *v1.VirtualMachineInstance, config *virtconfig.C
 	// Set this annotation now to indicate that the newly created virt-launchers will use
 	// unix sockets as a transport for migration
 	annotationsSet[v1.MigrationTransportUnixAnnotation] = "true"
+	annotationsSet[v1.DeschedulerRequestEvictOnlyAnnotation] = ""
+
 	return annotationsSet, nil
 }
 

diff --git a/pkg/virt-controller/services/template_test.go b/pkg/virt-controller/services/template_test.go
@@ -427,6 +427,7 @@ var _ = Describe("Template", func() {
 					"post.hook.backup.velero.io/command":      "[\"/usr/bin/virt-freezer\", \"--unfreeze\", \"--name\", \"testvmi\", \"--namespace\", \"testns\"]",
 					"kubevirt.io/migrationTransportUnix":      "true",
 					"kubectl.kubernetes.io/default-container": "compute",
+					v1.DeschedulerRequestEvictOnlyAnnotation:  "",
 				}))
 				Expect(pod.ObjectMeta.OwnerReferences).To(Equal([]metav1.OwnerReference{{
 					APIVersion:         v1.VirtualMachineInstanceGroupVersionKind.GroupVersion().String(),

diff --git a/pkg/virt-controller/watch/vmi.go b/pkg/virt-controller/watch/vmi.go
@@ -23,6 +23,7 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
+	"maps"
 	"sort"
 	"strings"
 	"time"
@@ -1271,14 +1272,14 @@ func (c *VMIController) sync(vmi *virtv1.VirtualMachineInstance, pod *k8sv1.Pod,
 	}
 
 	if !isTempPod(pod) && isPodReady(pod) {
-		newAnnotations := network.GeneratePodAnnotations(vmi.Spec.Networks, vmi.Spec.Domain.Devices.Interfaces, pod.Annotations[networkv1.NetworkStatusAnnot], c.clusterConfig.GetNetworkBindings())
-		if len(newAnnotations) != 0 {
-			patchedPod, err := c.syncPodAnnotations(pod, newAnnotations)
-			if err != nil {
-				return &syncErrorImpl{err, FailedPodPatchReason}
-			}
-			*pod = *patchedPod
+		newAnnotations := map[string]string{virtv1.DeschedulerRequestEvictOnlyAnnotation: ""}
+		maps.Copy(newAnnotations, network.GeneratePodAnnotations(vmi.Spec.Networks, vmi.Spec.Domain.Devices.Interfaces, pod.Annotations[networkv1.NetworkStatusAnnot], c.clusterConfig.GetNetworkBindings()))
+
+		patchedPod, err := c.syncPodAnnotations(pod, newAnnotations)
+		if err != nil {
+			return &syncErrorImpl{err, FailedPodPatchReason}
 		}
+		*pod = *patchedPod
 
 		hotplugVolumes := getHotplugVolumes(vmi, pod)
 		hotplugAttachmentPods, err := controller.AttachmentPods(pod, c.podIndexer)

diff --git a/pkg/virt-controller/watch/vmi_test.go b/pkg/virt-controller/watch/vmi_test.go
@@ -3942,7 +3942,8 @@ func setDataVolumeCondition(dv *cdiv1.DataVolume, cond cdiv1.DataVolumeCondition
 
 func NewPodForVirtualMachine(vmi *virtv1.VirtualMachineInstance, phase k8sv1.PodPhase) *k8sv1.Pod {
 	podAnnotations := map[string]string{
-		virtv1.DomainAnnotation: vmi.Name,
+		virtv1.DomainAnnotation:                      vmi.Name,
+		virtv1.DeschedulerRequestEvictOnlyAnnotation: "",
 	}
 	return &k8sv1.Pod{
 		ObjectMeta: metav1.ObjectMeta{

diff --git a/staging/src/kubevirt.io/api/core/v1/types.go b/staging/src/kubevirt.io/api/core/v1/types.go
@@ -1055,6 +1055,11 @@ const (
 	// in which freePageReporting is always disabled.
 	FreePageReportingDisabledAnnotation string = "kubevirt.io/free-page-reporting-disabled"
 
+	// DeschedulerRequestEvictOnlyAnnotation indicates pods whose eviction is not expected to be completed right away.
+	// Instead, an eviction request is expected to be intercepted by an external component which will initiate the
+	// eviction process for the pod.
+	DeschedulerRequestEvictOnlyAnnotation string = "descheduler.alpha.kubernetes.io/request-evict-only"
+
 	// VirtualMachinePodCPULimitsLabel indicates VMI pod CPU resource limits
 	VirtualMachinePodCPULimitsLabel string = "kubevirt.io/vmi-pod-cpu-resource-limits"
 	// VirtualMachinePodMemoryRequestsLabel indicates VMI pod Memory resource requests