sched: controller: set scheduler priority

So far the scheduler priority is set to default which is 0 this is risky especially when the preemtion of pods is needed to fit more important pods. The NRS is important enough to deserve the most critical priority class system-node-critical which is the same priority for the kube-scheduler. addresses openshift-kni#974 Signed-off-by: Shereen Haj <[email protected]>
shajmakh · Aug 12, 2024 · 506b82b · 506b82b
1 parent 6c6c29e
commit 506b82b
Show file tree

Hide file tree

Showing 2 changed files with 21 additions and 1 deletion.
diff --git a/controllers/numaresourcesscheduler_controller.go b/controllers/numaresourcesscheduler_controller.go
@@ -55,7 +55,8 @@ import (
 )
 
 const (
-	leaderElectionResourceName = "numa-scheduler-leader"
+	leaderElectionResourceName      = "numa-scheduler-leader"
+	SystemNodeCriticalPriorityClass = "system-node-critical"
 )
 
 const (
@@ -217,6 +218,9 @@ func (r *NUMAResourcesSchedulerReconciler) syncNUMASchedulerResources(ctx contex
 	// TODO: if replicas doesn't make sense (autodetect disabled and user set impossible value) then we
 	// should set a degraded state
 
+	// node-critical so the pod won't be preempted by pods having the most critical priority class
+	r.SchedulerManifests.Deployment.Spec.Template.Spec.PriorityClassName = SystemNodeCriticalPriorityClass
+
 	schedupdate.DeploymentImageSettings(r.SchedulerManifests.Deployment, schedSpec.SchedulerImage)
 	cmHash := hash.ConfigMapData(r.SchedulerManifests.ConfigMap)
 	schedupdate.DeploymentConfigMapSettings(r.SchedulerManifests.Deployment, r.SchedulerManifests.ConfigMap.Name, cmHash)

diff --git a/controllers/numaresourcesscheduler_controller_test.go b/controllers/numaresourcesscheduler_controller_test.go
@@ -208,6 +208,22 @@ var _ = ginkgo.Describe("Test NUMAResourcesScheduler Reconcile", func() {
 			gomega.Expect(nrs.Status.CacheResyncPeriod.Seconds()).To(gomega.Equal(resyncPeriod.Seconds()))
 		})
 
+		ginkgo.It("should have the correct priority class", func() {
+			key := client.ObjectKeyFromObject(nrs)
+			_, err := reconciler.Reconcile(context.TODO(), reconcile.Request{NamespacedName: key})
+			gomega.Expect(err).ToNot(gomega.HaveOccurred())
+
+			key = client.ObjectKey{
+				Name:      "secondary-scheduler",
+				Namespace: testNamespace,
+			}
+
+			dp := &appsv1.Deployment{}
+			gomega.Expect(reconciler.Client.Get(context.TODO(), key, dp)).ToNot(gomega.HaveOccurred())
+
+			gomega.Expect(dp.Spec.Template.Spec.PriorityClassName).To(gomega.BeEquivalentTo(SystemNodeCriticalPriorityClass))
+		})
+
 		ginkgo.It("should have a config hash annotation under deployment", func() {
 			key := client.ObjectKeyFromObject(nrs)
 			_, err := reconciler.Reconcile(context.TODO(), reconcile.Request{NamespacedName: key})