Merge pull request #318 from Tal-or/best_effort_pod_fix

NodeResourceTopology: handle best-effort pods in score plugin

pkg/noderesourcetopology/score.go

@@ -24,6 +24,7 @@ import (
 
 	v1 "k8s.io/api/core/v1"
 	"k8s.io/klog/v2"
+	v1qos "k8s.io/kubernetes/pkg/apis/core/v1/helper/qos"
 	"k8s.io/kubernetes/pkg/scheduler/framework"
 	apiconfig "sigs.k8s.io/scheduler-plugins/pkg/apis/config"
 
@@ -55,6 +56,11 @@ func (rw resourceToWeightMap) weight(r v1.ResourceName) int64 {
 
 func (tm *TopologyMatch) Score(ctx context.Context, state *framework.CycleState, pod *v1.Pod, nodeName string) (int64, *framework.Status) {
 	klog.V(5).InfoS("Scoring node", "nodeName", nodeName)
+	// if it's a non-guaranteed pod, every node is considered to be a good fit
+	if v1qos.GetPodQOS(pod) != v1.PodQOSGuaranteed {
+		return framework.MaxNodeScore, nil
+	}
+
 	nodeTopology := findNodeTopology(nodeName, tm.lister)
 
 	if nodeTopology == nil {

test/integration/noderesourcetopology_test.go

@@ -553,6 +553,32 @@ func TestTopologyMatchPlugin(t *testing.T) {
 			},
 			expectedNodes: []string{"fake-node-1"},
 		},
+		// in the following "Scheduling Best-Effort pod with least-allocatable/balanced-allocation/most-allocatable strategy" tests,
+		// both nodes are expected to get the maximum score.
+		// in case of multiple nodes with the same highest score, the chosen node selection might vary.
+		// thus, we can't tell exactly which node will get selected, so the only option is to accept both of them,
+		// and it's still good enough for making sure that pod not stuck in pending state and/or the program isn't crashing
+		{
+			name: "Scheduling Best-Effort pod with least-allocatable strategy scheduler",
+			pods: []*v1.Pod{
+				st.MakePod().Namespace(ns.Name).Name("topology-aware-scheduler-pod").SchedulerName(leastAllocatableScheduler).Container(pause).Obj(),
+			},
+			expectedNodes: []string{"fake-node-1", "fake-node-2"},
+		},
+		{
+			name: "Scheduling Best-Effort pod with balanced-allocation strategy scheduler",
+			pods: []*v1.Pod{
+				st.MakePod().Namespace(ns.Name).Name("topology-aware-scheduler-pod").SchedulerName(balancedAllocationScheduler).Container(pause).Obj(),
+			},
+			expectedNodes: []string{"fake-node-1", "fake-node-2"},
+		},
+		{
+			name: "Scheduling Best-Effort pod with most-allocatable strategy scheduler",
+			pods: []*v1.Pod{
+				st.MakePod().Namespace(ns.Name).Name("topology-aware-scheduler-pod").SchedulerName(mostAllocatableScheduler).Container(pause).Obj(),
+			},
+			expectedNodes: []string{"fake-node-1", "fake-node-2"},
+		},
 	} {
 		t.Run(tt.name, func(t *testing.T) {
 			t.Logf("Start-topology-match-test %v", tt.name)