chore: don't emit disruption events for non-karpenter nodes (#1644)

kubernetes-sigs · Sep 5, 2024 · 140d35b · 140d35b
1 parent 75fcd2a
commit 140d35b
Show file tree

Hide file tree

Showing 3 changed files with 11 additions and 5 deletions.
diff --git a/pkg/controllers/disruption/suite_test.go b/pkg/controllers/disruption/suite_test.go
@@ -1492,7 +1492,7 @@ var _ = Describe("Candidate Filtering", func() {
 		Expect(cluster.Nodes()).To(HaveLen(1))
 		_, err := disruption.NewCandidate(ctx, env.Client, recorder, fakeClock, cluster.Nodes()[0], pdbLimits, nodePoolMap, nodePoolInstanceTypeMap, queue, disruption.GracefulDisruptionClass)
 		Expect(err).To(HaveOccurred())
-		Expect(err.Error()).To(Equal("state node doesn't contain both a node and a nodeclaim"))
+		Expect(err.Error()).To(Equal("state node does not have a nodeclaim representation"))
 	})
 	It("should not consider candidate that has just a NodeClaim representation", func() {
 		nodeClaim, _ := test.NodeClaimAndNode(v1.NodeClaim{
@@ -1511,7 +1511,7 @@ var _ = Describe("Candidate Filtering", func() {
 		Expect(cluster.Nodes()).To(HaveLen(1))
 		_, err := disruption.NewCandidate(ctx, env.Client, recorder, fakeClock, cluster.Nodes()[0], pdbLimits, nodePoolMap, nodePoolInstanceTypeMap, queue, disruption.GracefulDisruptionClass)
 		Expect(err).To(HaveOccurred())
-		Expect(err.Error()).To(Equal("state node doesn't contain both a node and a nodeclaim"))
+		Expect(err.Error()).To(Equal("state node does not have a node representation"))
 	})
 	It("should not consider candidates that are nominated", func() {
 		nodeClaim, node := test.NodeClaimAndNode(v1.NodeClaim{

diff --git a/pkg/controllers/disruption/types.go b/pkg/controllers/disruption/types.go
@@ -72,7 +72,10 @@ func NewCandidate(ctx context.Context, kubeClient client.Client, recorder events
 	var err error
 	var pods []*corev1.Pod
 	if err = node.ValidateNodeDisruptable(ctx, kubeClient); err != nil {
-		recorder.Publish(disruptionevents.Blocked(node.Node, node.NodeClaim, err.Error())...)
+		// Only emit an event if the NodeClaim is not nil, ensuring that we only emit events for Karpenter-managed nodes
+		if node.NodeClaim != nil {
+			recorder.Publish(disruptionevents.Blocked(node.Node, node.NodeClaim, err.Error())...)
+		}
 		return nil, err
 	}
 	// If the orchestration queue is already considering a candidate we want to disrupt, don't consider it a candidate.

diff --git a/pkg/controllers/state/statenode.go b/pkg/controllers/state/statenode.go
@@ -189,8 +189,11 @@ func (in *StateNode) Pods(ctx context.Context, kubeClient client.Client) ([]*cor
 //
 //nolint:gocyclo
 func (in *StateNode) ValidateNodeDisruptable(ctx context.Context, kubeClient client.Client) error {
-	if in.Node == nil || in.NodeClaim == nil {
-		return fmt.Errorf("state node doesn't contain both a node and a nodeclaim")
+	if in.NodeClaim == nil {
+		return fmt.Errorf("state node does not have a nodeclaim representation")
+	}
+	if in.Node == nil {
+		return fmt.Errorf("state node does not have a node representation")
 	}
 	if !in.Initialized() {
 		return fmt.Errorf("state node isn't initialized")