fix panic when get job's elastic resource

pkg/scheduler/api/job_info.go

@@ -572,12 +572,15 @@ func (ji *JobInfo) DeductSchGatedResources(res *Resource) *Resource {
 	return result
 }
 
+// GetElasticResources returns those partly resources in allocated which are more than its minResource
 func (ji *JobInfo) GetElasticResources() *Resource {
-	minResource := ji.GetMinResources()
-	if ji.Allocated.LessEqualPartly(minResource, Zero) {
+	if ji.Allocated == nil {
 		return EmptyResource()
 	}
-	return ji.Allocated.Clone().Sub(minResource)
+	minResource := ji.GetMinResources()
+	elastic := ExceededPart(ji.Allocated, minResource)
+
+	return elastic
 }
 
 func (ji *JobInfo) addTaskIndex(ti *TaskInfo) {

pkg/scheduler/api/job_info_test.go

@@ -385,3 +385,58 @@ func TestJobInfo(t *testing.T) {
 		}
 	}
 }
+
+func TestGetElasticResources(t *testing.T) {
+	resNoGPU := BuildResourceList("1", "1G")
+	resWithGPU := BuildResourceListWithGPU("1", "1G", "1")
+	wantNoGPU := BuildResourceList("1", "1G", []ScalarResource{{Name: "pods", Value: "1"}}...)
+	wantWithGPU := BuildResourceListWithGPU("1", "1G", "1", []ScalarResource{{Name: "pods", Value: "1"}}...)
+	tests := []struct {
+		pods     []*v1.Pod
+		podgroup scheduling.PodGroup
+		want     *Resource
+	}{
+		{
+			pods: []*v1.Pod{
+				buildPod("ns1", "task-1", "node1", v1.PodRunning, resWithGPU, nil, make(map[string]string)),
+			},
+			podgroup: BuildPodgroup("pg1", "ns1", 1, wantWithGPU),
+			want:     EmptyResource(),
+		},
+		{
+			pods: []*v1.Pod{
+				buildPod("ns1", "task-1", "node1", v1.PodRunning, resWithGPU, nil, make(map[string]string)),
+				buildPod("ns1", "task-2", "node2", v1.PodRunning, resNoGPU, nil, make(map[string]string)),
+			},
+			podgroup: BuildPodgroup("pg1", "ns1", 1, wantWithGPU),
+			want:     NewResource(wantNoGPU),
+		},
+		{
+			pods: []*v1.Pod{
+				buildPod("ns1", "task-1", "node1", v1.PodRunning, resNoGPU, nil, make(map[string]string)),
+				buildPod("ns1", "task-2", "node2", v1.PodRunning, resNoGPU, nil, make(map[string]string)),
+			},
+			podgroup: BuildPodgroup("pg1", "ns1", 1, wantWithGPU),
+			want:     NewResource(wantNoGPU),
+		},
+		{
+			pods: []*v1.Pod{
+				buildPod("ns1", "task-1", "node1", v1.PodRunning, resWithGPU, nil, make(map[string]string)),
+				buildPod("ns1", "task-2", "node2", v1.PodRunning, resWithGPU, nil, make(map[string]string)),
+			},
+			podgroup: BuildPodgroup("pg1", "ns1", 1, wantWithGPU),
+			want:     NewResource(wantWithGPU),
+		},
+	}
+
+	for i, test := range tests {
+		job := NewJobInfo("job")
+		for _, pod := range test.pods {
+			job.AddTaskInfo(NewTaskInfo(pod))
+		}
+		job.SetPodGroup(&PodGroup{PodGroup: test.podgroup})
+		if elastic := job.GetElasticResources(); !elastic.Equal(test.want, Zero) {
+			t.Fatalf("case %d:expected %+v, got %+v", i, test.want, elastic)
+		}
+	}
+}

pkg/scheduler/api/resource_info.go

@@ -771,3 +771,17 @@ func (r ResourceNameList) Contains(rr ResourceNameList) bool {
 func IsCountQuota(name v1.ResourceName) bool {
 	return strings.HasPrefix(string(name), "count/")
 }
+
+// ExceededPart returns the partly resource in left which exceed right
+func ExceededPart(left, right *Resource) *Resource {
+	if right == nil {
+		return left
+	}
+
+	if left == nil {
+		return EmptyResource()
+	}
+
+	diff, _ := left.Diff(right, Zero)
+	return diff
+}

pkg/scheduler/api/test_utils.go

@@ -23,6 +23,8 @@ import (
 	"k8s.io/apimachinery/pkg/api/resource"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/types"
+
+	"volcano.sh/apis/pkg/apis/scheduling"
 )
 
 func buildNode(name string, alloc v1.ResourceList) *v1.Node {
@@ -90,7 +92,7 @@ type ScalarResource struct {
 	Value string
 }
 
-// BuildResourceList builts resource list object
+// BuildResourceList builds resource list object
 func BuildResourceList(cpu string, memory string, scalarResources ...ScalarResource) v1.ResourceList {
 	resourceList := v1.ResourceList{
 		v1.ResourceCPU:    resource.MustParse(cpu),
@@ -103,7 +105,7 @@ func BuildResourceList(cpu string, memory string, scalarResources ...ScalarResou
 	return resourceList
 }
 
-// BuildResourceListWithGPU builts resource list with GPU
+// BuildResourceListWithGPU builds resource list with GPU
 func BuildResourceListWithGPU(cpu string, memory string, GPU string, scalarResources ...ScalarResource) v1.ResourceList {
 	resourceList := v1.ResourceList{
 		v1.ResourceCPU:    resource.MustParse(cpu),
@@ -116,3 +118,17 @@ func BuildResourceListWithGPU(cpu string, memory string, GPU string, scalarResou
 
 	return resourceList
 }
+
+// BuildPodgroup builds podgroup
+func BuildPodgroup(name, ns string, minMember int32, minResource v1.ResourceList) scheduling.PodGroup {
+	return scheduling.PodGroup{
+		ObjectMeta: metav1.ObjectMeta{
+			Namespace: ns,
+			Name:      name,
+		},
+		Spec: scheduling.PodGroupSpec{
+			MinMember:    minMember,
+			MinResources: &minResource,
+		},
+	}
+}

pkg/scheduler/plugins/capacity/capacity.go

@@ -116,7 +116,7 @@ func (cp *capacityPlugin) OnSessionOpen(ssn *framework.Session) {
 			if len(queue.Queue.Spec.Guarantee.Resource) != 0 {
 				attr.guarantee = api.NewResource(queue.Queue.Spec.Guarantee.Resource)
 			}
-			realCapability := cp.totalResource.Clone().Sub(cp.totalGuarantee).Add(attr.guarantee)
+			realCapability := api.ExceededPart(cp.totalResource, cp.totalGuarantee).Add(attr.guarantee)
 			if attr.capability == nil {
 				attr.realCapability = realCapability
 			} else {

pkg/scheduler/plugins/proportion/proportion.go

@@ -117,7 +117,7 @@ func (pp *proportionPlugin) OnSessionOpen(ssn *framework.Session) {
 			if len(queue.Queue.Spec.Guarantee.Resource) != 0 {
 				attr.guarantee = api.NewResource(queue.Queue.Spec.Guarantee.Resource)
 			}
-			realCapability := pp.totalResource.Clone().Sub(pp.totalGuarantee).Add(attr.guarantee)
+			realCapability := api.ExceededPart(pp.totalResource, pp.totalGuarantee).Add(attr.guarantee)
 			if attr.capability == nil {
 				attr.realCapability = realCapability
 			} else {