should filter nodes that do not have enough cpu when cross numa nodes

pkg/plugins/noderesourcetopology/filter.go

@@ -76,7 +76,10 @@ func (tm *TopologyMatch) Filter(
 			return status
 		}
 	}
-	assignTopologyResult(nw, s.targetContainerResource.Clone())
+
+	if status := assignTopologyResult(nw, s.targetContainerResource.Clone()); status != nil {
+		return status
+	}
 
 	s.Lock()
 	defer s.Unlock()

pkg/plugins/noderesourcetopology/filter_test.go

@@ -294,7 +294,7 @@ func TestTopologyMatch_Filter(t *testing.T) {
 			want: framework.NewStatus(framework.Unschedulable, ErrReasonNUMAResourceNotEnough),
 		},
 		{
-			name: "enough cpu resource in node with default none topology manager policy",
+			name: "not enough cpu resource in node with default none topology manager policy",
 			args: args{
 				pod: newResourcePod(false, nil, framework.Resource{MilliCPU: 2 * CPUTestUnit, Memory: MemTestUnit}),
 				nodeInfo: framework.NewNodeInfo(
@@ -310,7 +310,7 @@ func TestTopologyMatch_Filter(t *testing.T) {
 				}(),
 				topologyAwareResources: sets.NewString(string(corev1.ResourceCPU)),
 			},
-			want: nil,
+			want: framework.NewStatus(framework.Unschedulable, ErrReasonNUMAResourceNotEnough),
 		},
 		{
 			name: "no enough cpu resource in one NUMA node with default single numa topology manager policy",
@@ -357,6 +357,23 @@ func TestTopologyMatch_Filter(t *testing.T) {
 			},
 			want: framework.NewStatus(framework.Unschedulable, ErrReasonNUMAResourceNotEnough),
 		},
+		{
+			name: "no enough cpu resource in one NUMA node, but enough with cross numa pods",
+			args: args{
+				pod: newResourcePod(false, nil, framework.Resource{MilliCPU: 4 * CPUTestUnit}),
+				nodeInfo: framework.NewNodeInfo(
+					newResourcePod(true, newZoneList([]zone{{name: "node1", cpu: 1 * CPUTestUnit}}),
+						framework.Resource{MilliCPU: 1 * CPUTestUnit, Memory: 2 * MemTestUnit}),
+				),
+				nrt: func() *topologyv1alpha1.NodeResourceTopology {
+					nrtCopy := nrt.DeepCopy()
+					nrtCopy.CraneManagerPolicy.TopologyManagerPolicy = topologyv1alpha1.TopologyManagerPolicyNone
+					return nrtCopy
+				}(),
+				topologyAwareResources: sets.NewString(string(corev1.ResourceCPU)),
+			},
+			want: nil,
+		},
 	}
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {

pkg/plugins/noderesourcetopology/helper.go

@@ -170,7 +170,7 @@ func (nw *nodeWrapper) addNUMAResources(numaNodeResult topologyv1alpha1.ZoneList
 	}
 }
 
-func assignTopologyResult(nw *nodeWrapper, request *framework.Resource) {
+func assignTopologyResult(nw *nodeWrapper, request *framework.Resource) *framework.Status {
 	// sort by free CPU resource
 	sort.Slice(nw.numaNodes, func(i, j int) bool {
 		nodeI, nodeJ := nw.numaNodes[i], nw.numaNodes[j]
@@ -187,10 +187,10 @@ func assignTopologyResult(nw *nodeWrapper, request *framework.Resource) {
 				},
 			},
 		}
-		return
+		return nil
 	}
 
-	for _, node := range nw.numaNodes {
+	for i, node := range nw.numaNodes {
 		node.allocatable.MilliCPU = node.allocatable.MilliCPU / 1000 * 1000
 		res, finished := assignRequestForNUMANode(request, node)
 		if capacity := ResourceListIgnoreZeroResources(res); len(capacity) != 0 {
@@ -205,10 +205,16 @@ func assignTopologyResult(nw *nodeWrapper, request *framework.Resource) {
 		if finished {
 			break
 		}
+		// if we reach the last numa node and still not finished,
+		// then we are done for this node.
+		if i == len(nw.numaNodes)-1 {
+			return framework.NewStatus(framework.Unschedulable, ErrReasonNUMAResourceNotEnough)
+		}
 	}
 	sort.Slice(nw.result, func(i, j int) bool {
 		return nw.result[i].Name < nw.result[j].Name
 	})
+	return nil
 }
 
 func computeContainerSpecifiedResourceRequest(pod *corev1.Pod, indices []int, names sets.String) *framework.Resource {