[cluster-autoscaler-1.32]: prevent panic in SimulateNodeRemoval by handling missing node info

Author: jackfrancis

cluster-autoscaler/simulator/cluster.go

@@ -17,10 +17,12 @@ limitations under the License.
 package simulator
 
 import (
+	"errors"
 	"fmt"
 	"time"
 
 	apiv1 "k8s.io/api/core/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/autoscaler/cluster-autoscaler/core/scaledown/pdb"
 	"k8s.io/autoscaler/cluster-autoscaler/simulator/clustersnapshot"
 	"k8s.io/autoscaler/cluster-autoscaler/simulator/drainability/rules"
@@ -91,6 +93,8 @@ const (
 	BlockedByPod
 	// UnexpectedError - node can't be removed because of an unexpected error.
 	UnexpectedError
+	// NoNodeInfo - node can't be removed because it doesn't have any node info in the cluster snapshot.
+	NoNodeInfo
 )
 
 // RemovalSimulator is a helper object for simulating node removal scenarios.
@@ -151,6 +155,12 @@ func (r *RemovalSimulator) SimulateNodeRemoval(
 	nodeInfo, err := r.clusterSnapshot.GetNodeInfo(nodeName)
 	if err != nil {
 		klog.Errorf("Can't retrieve node %s from snapshot, err: %v", nodeName, err)
+		unremovableReason := UnexpectedError
+		if errors.Is(err, clustersnapshot.ErrNodeNotFound) {
+			unremovableReason = NoNodeInfo
+		}
+		unremovableNode := &UnremovableNode{Node: &apiv1.Node{ObjectMeta: metav1.ObjectMeta{Name: nodeName}}, Reason: unremovableReason}
+		return nil, unremovableNode
 	}
 	klog.V(2).Infof("Simulating node %s removal", nodeName)
 

cluster-autoscaler/simulator/cluster_test.go

@@ -258,6 +258,201 @@ func TestFindNodesToRemove(t *testing.T) {
 	}
 }
 
+type simulateNodeRemovalTestConfig struct {
+	name        string
+	pods        []*apiv1.Pod
+	allNodes    []*apiv1.Node
+	nodeName    string
+	toRemove    *NodeToBeRemoved
+	unremovable *UnremovableNode
+}
+
+func TestSimulateNodeRemoval(t *testing.T) {
+	emptyNode := BuildTestNode("n1", 1000, 2000000)
+
+	// two small pods backed by ReplicaSet
+	drainableNode := BuildTestNode("n2", 1000, 2000000)
+	drainableNodeInfo := framework.NewTestNodeInfo(drainableNode)
+
+	// one small pod, not backed by anything
+	nonDrainableNode := BuildTestNode("n3", 1000, 2000000)
+	nonDrainableNodeInfo := framework.NewTestNodeInfo(nonDrainableNode)
+
+	// one very large pod
+	fullNode := BuildTestNode("n4", 1000, 2000000)
+	fullNodeInfo := framework.NewTestNodeInfo(fullNode)
+
+	// noExistNode it doesn't have any node info in the cluster snapshot.
+	noExistNode := &apiv1.Node{ObjectMeta: metav1.ObjectMeta{Name: "n5"}}
+
+	SetNodeReadyState(emptyNode, true, time.Time{})
+	SetNodeReadyState(drainableNode, true, time.Time{})
+	SetNodeReadyState(nonDrainableNode, true, time.Time{})
+	SetNodeReadyState(fullNode, true, time.Time{})
+
+	replicas := int32(5)
+	replicaSets := []*appsv1.ReplicaSet{
+		{
+			ObjectMeta: metav1.ObjectMeta{
+				Name:      "rs",
+				Namespace: "default",
+			},
+			Spec: appsv1.ReplicaSetSpec{
+				Replicas: &replicas,
+			},
+		},
+	}
+	rsLister, err := kube_util.NewTestReplicaSetLister(replicaSets)
+	assert.NoError(t, err)
+	registry := kube_util.NewListerRegistry(nil, nil, nil, nil, nil, nil, nil, rsLister, nil)
+
+	ownerRefs := GenerateOwnerReferences("rs", "ReplicaSet", "extensions/v1beta1", "")
+
+	pod1 := BuildTestPod("p1", 100, 100000)
+	pod1.OwnerReferences = ownerRefs
+	pod1.Spec.NodeName = "n2"
+	drainableNodeInfo.AddPod(&framework.PodInfo{Pod: pod1})
+
+	pod2 := BuildTestPod("p2", 100, 100000)
+	pod2.OwnerReferences = ownerRefs
+	pod2.Spec.NodeName = "n2"
+	drainableNodeInfo.AddPod(&framework.PodInfo{Pod: pod2})
+
+	pod3 := BuildTestPod("p3", 100, 100000)
+	pod3.Spec.NodeName = "n3"
+	nonDrainableNodeInfo.AddPod(&framework.PodInfo{Pod: pod3})
+
+	pod4 := BuildTestPod("p4", 1000, 100000)
+	pod4.Spec.NodeName = "n4"
+	fullNodeInfo.AddPod(&framework.PodInfo{Pod: pod4})
+
+	clusterSnapshot := testsnapshot.NewTestSnapshotOrDie(t)
+
+	topoNode1 := BuildTestNode("topo-n1", 1000, 2000000)
+	topoNode2 := BuildTestNode("topo-n2", 1000, 2000000)
+	topoNode3 := BuildTestNode("topo-n3", 1000, 2000000)
+	topoNode1.Labels = map[string]string{"kubernetes.io/hostname": "topo-n1"}
+	topoNode2.Labels = map[string]string{"kubernetes.io/hostname": "topo-n2"}
+	topoNode3.Labels = map[string]string{"kubernetes.io/hostname": "topo-n3"}
+
+	SetNodeReadyState(topoNode1, true, time.Time{})
+	SetNodeReadyState(topoNode2, true, time.Time{})
+	SetNodeReadyState(topoNode3, true, time.Time{})
+
+	minDomains := int32(2)
+	maxSkew := int32(1)
+	topoConstraint := apiv1.TopologySpreadConstraint{
+		MaxSkew:           maxSkew,
+		TopologyKey:       "kubernetes.io/hostname",
+		WhenUnsatisfiable: apiv1.DoNotSchedule,
+		MinDomains:        &minDomains,
+		LabelSelector: &metav1.LabelSelector{
+			MatchLabels: map[string]string{
+				"app": "topo-app",
+			},
+		},
+	}
+
+	pod5 := BuildTestPod("p5", 100, 100000)
+	pod5.Labels = map[string]string{"app": "topo-app"}
+	pod5.OwnerReferences = ownerRefs
+	pod5.Spec.NodeName = "topo-n1"
+	pod5.Spec.TopologySpreadConstraints = []apiv1.TopologySpreadConstraint{topoConstraint}
+
+	pod6 := BuildTestPod("p6", 100, 100000)
+	pod6.Labels = map[string]string{"app": "topo-app"}
+	pod6.OwnerReferences = ownerRefs
+	pod6.Spec.NodeName = "topo-n2"
+	pod6.Spec.TopologySpreadConstraints = []apiv1.TopologySpreadConstraint{topoConstraint}
+
+	pod7 := BuildTestPod("p7", 100, 100000)
+	pod7.Labels = map[string]string{"app": "topo-app"}
+	pod7.OwnerReferences = ownerRefs
+	pod7.Spec.NodeName = "topo-n3"
+	pod7.Spec.TopologySpreadConstraints = []apiv1.TopologySpreadConstraint{topoConstraint}
+
+	blocker1 := BuildTestPod("blocker1", 100, 100000)
+	blocker1.Spec.NodeName = "topo-n2"
+	blocker2 := BuildTestPod("blocker2", 100, 100000)
+	blocker2.Spec.NodeName = "topo-n3"
+
+	tests := []simulateNodeRemovalTestConfig{
+		{
+			name:        "just an empty node, should be removed",
+			nodeName:    emptyNode.Name,
+			allNodes:    []*apiv1.Node{emptyNode},
+			toRemove:    &NodeToBeRemoved{Node: emptyNode},
+			unremovable: nil,
+		},
+		{
+			name:     "just a drainable node, but nowhere for pods to go to",
+			pods:     []*apiv1.Pod{pod1, pod2},
+			nodeName: drainableNode.Name,
+			allNodes: []*apiv1.Node{drainableNode},
+			toRemove: nil,
+			unremovable: &UnremovableNode{
+				Node:   drainableNode,
+				Reason: NoPlaceToMovePods,
+			},
+		},
+		{
+			name:        "drainable node, and a mostly empty node that can take its pods",
+			pods:        []*apiv1.Pod{pod1, pod2, pod3},
+			nodeName:    drainableNode.Name,
+			allNodes:    []*apiv1.Node{drainableNode, nonDrainableNode},
+			toRemove:    &NodeToBeRemoved{Node: drainableNode, PodsToReschedule: []*apiv1.Pod{pod1, pod2}},
+			unremovable: nil,
+		},
+		{
+			name:        "drainable node, and a full node that cannot fit anymore pods",
+			pods:        []*apiv1.Pod{pod1, pod2, pod4},
+			nodeName:    drainableNode.Name,
+			allNodes:    []*apiv1.Node{drainableNode, fullNode},
+			toRemove:    nil,
+			unremovable: &UnremovableNode{Node: drainableNode, Reason: NoPlaceToMovePods},
+		},
+		{
+			name:        "4 nodes, 1 empty, 1 drainable",
+			pods:        []*apiv1.Pod{pod1, pod2, pod3, pod4},
+			nodeName:    emptyNode.Name,
+			allNodes:    []*apiv1.Node{emptyNode, drainableNode, fullNode, nonDrainableNode},
+			toRemove:    &NodeToBeRemoved{Node: emptyNode},
+			unremovable: nil,
+		},
+		{
+			name:        "topology spread constraint test - one node should be removable",
+			pods:        []*apiv1.Pod{pod5, pod6, pod7, blocker1, blocker2},
+			allNodes:    []*apiv1.Node{topoNode1, topoNode2, topoNode3},
+			nodeName:    topoNode1.Name,
+			toRemove:    &NodeToBeRemoved{Node: topoNode1, PodsToReschedule: []*apiv1.Pod{pod5}},
+			unremovable: nil,
+		},
+		{
+			name:        "candidate not in clusterSnapshot should be marked unremovable",
+			nodeName:    noExistNode.Name,
+			allNodes:    []*apiv1.Node{},
+			pods:        []*apiv1.Pod{},
+			toRemove:    nil,
+			unremovable: &UnremovableNode{Node: noExistNode, Reason: NoNodeInfo},
+		},
+	}
+
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			destinations := make(map[string]bool)
+			for _, node := range test.allNodes {
+				destinations[node.Name] = true
+			}
+			clustersnapshot.InitializeClusterSnapshotOrDie(t, clusterSnapshot, test.allNodes, test.pods)
+			r := NewRemovalSimulator(registry, clusterSnapshot, testDeleteOptions(), nil, false)
+			toRemove, unremovable := r.SimulateNodeRemoval(test.nodeName, destinations, time.Now(), nil)
+			fmt.Printf("Test scenario: %s, toRemove=%v, unremovable=%v\n", test.name, toRemove, unremovable)
+			assert.Equal(t, test.toRemove, toRemove)
+			assert.Equal(t, test.unremovable, unremovable)
+		})
+	}
+}
+
 func testDeleteOptions() options.NodeDeleteOptions {
 	return options.NodeDeleteOptions{
 		SkipNodesWithSystemPods:           true,