fix: if the capability cpu or memory is not specified in the hierarch…

…ical queue, it will be set to the corresponding value of the parent queue Signed-off-by: JesseStutler <[email protected]>
volcano-sh · Dec 25, 2024 · 9754ae3 · 9754ae3
1 parent b0c1a56
commit 9754ae3
Show file tree

Hide file tree

Showing 2 changed files with 27 additions and 13 deletions.
diff --git a/pkg/scheduler/plugins/capacity/capacity.go b/pkg/scheduler/plugins/capacity/capacity.go
@@ -419,9 +419,7 @@ func (cp *capacityPlugin) buildHierarchicalQueueAttrs(ssn *framework.Session) bo
 			continue
 		}
 
-		attr := cp.newQueueAttr(queue)
-		cp.queueOpts[queue.UID] = attr
-		err := cp.updateParents(queue, ssn)
+		err := cp.initializeHierarchicalQueueAttrs(queue, ssn)
 		if err != nil {
 			klog.Errorf("Failed to update Queue <%s> attributes, error: %v", queue.Name, err)
 			return false
@@ -579,7 +577,7 @@ func (cp *capacityPlugin) buildHierarchicalQueueAttrs(ssn *framework.Session) bo
 	return true
 }
 
-func (cp *capacityPlugin) newQueueAttr(queue *api.QueueInfo) *queueAttr {
+func (cp *capacityPlugin) newQueueAttr(queue *api.QueueInfo, parent *api.QueueInfo) *queueAttr {
 	attr := &queueAttr{
 		queueID:  queue.UID,
 		name:     queue.Name,
@@ -593,13 +591,29 @@ func (cp *capacityPlugin) newQueueAttr(queue *api.QueueInfo) *queueAttr {
 		inqueue:   api.EmptyResource(),
 		guarantee: api.EmptyResource(),
 	}
+
+	if queue.Name == cp.rootQueue {
+		attr.realCapability = cp.totalResource
+		attr.capability = attr.realCapability
+		attr.deserved = cp.totalResource
+		return attr
+	}
+
 	if len(queue.Queue.Spec.Capability) != 0 {
 		attr.capability = api.NewResource(queue.Queue.Spec.Capability)
+		parentCapability := api.EmptyResource()
+		if parent.Queue.Name == cp.rootQueue {
+			// if current queue's parent is root, set parentCapability to totalResource (because root's realCapability is totalResource, and the attr of root queue may not have been initialized yet)
+			parentCapability = cp.totalResource
+		} else {
+			parentCapability = api.NewResource(parent.Queue.Spec.Capability)
+		}
+		// If the user does not set CPU or memory in capability, we set the value to be the same as parent (we do not consider the situation where the user sets CPU or memory<=0)
 		if attr.capability.MilliCPU <= 0 {
-			attr.capability.MilliCPU = math.MaxFloat64
+			attr.capability.MilliCPU = parentCapability.MilliCPU
 		}
 		if attr.capability.Memory <= 0 {
-			attr.capability.Memory = math.MaxFloat64
+			attr.capability.Memory = parentCapability.Memory
 		}
 	}
 
@@ -610,8 +624,9 @@ func (cp *capacityPlugin) newQueueAttr(queue *api.QueueInfo) *queueAttr {
 	return attr
 }
 
-func (cp *capacityPlugin) updateParents(queue *api.QueueInfo, ssn *framework.Session) error {
+func (cp *capacityPlugin) initializeHierarchicalQueueAttrs(queue *api.QueueInfo, ssn *framework.Session) error {
 	if queue.Name == cp.rootQueue {
+		cp.queueOpts[queue.UID] = cp.newQueueAttr(queue, nil)
 		return nil
 	}
 
@@ -624,10 +639,11 @@ func (cp *capacityPlugin) updateParents(queue *api.QueueInfo, ssn *framework.Ses
 	}
 
 	parentInfo := ssn.Queues[api.QueueID(parent)]
+	curQueueAttr := cp.newQueueAttr(queue, parentInfo)
+	cp.queueOpts[curQueueAttr.queueID] = curQueueAttr
+
 	if _, found := cp.queueOpts[parentInfo.UID]; !found {
-		parentAttr := cp.newQueueAttr(parentInfo)
-		cp.queueOpts[parentAttr.queueID] = parentAttr
-		err := cp.updateParents(parentInfo, ssn)
+		err := cp.initializeHierarchicalQueueAttrs(parentInfo, ssn)
 		if err != nil {
 			return err
 		}
@@ -653,8 +669,6 @@ func (cp *capacityPlugin) checkHierarchicalQueue(attr *queueAttr) error {
 	if attr.name == cp.rootQueue {
 		attr.guarantee = totalGuarantee
 		cp.totalGuarantee = totalGuarantee
-		attr.realCapability = cp.totalResource
-		attr.deserved = cp.totalResource
 	}
 
 	for _, childAttr := range attr.children {

diff --git a/pkg/scheduler/plugins/capacity/capacity_test.go b/pkg/scheduler/plugins/capacity/capacity_test.go
@@ -348,7 +348,7 @@ func Test_capacityPlugin_OnSessionOpenWithHierarchy(t *testing.T) {
 	// podgroup
 	pg1 := util.BuildPodGroup("pg1", "ns1", "q11", 1, nil, schedulingv1beta1.PodGroupInqueue)
 	// queue
-	root := buildQueueWithParents("root", "", api.BuildResourceList("8", "8Gi", []api.ScalarResource{{Name: "pods", Value: "10"}}...), api.BuildResourceList("8", "8Gi", []api.ScalarResource{{Name: "pods", Value: "10"}}...))
+	root := buildQueueWithParents("root", "", nil, nil)
 	queue1 := buildQueueWithParents("q1", "root", nil, api.BuildResourceList("4", "4Gi"))
 	queue2 := buildQueueWithParents("q2", "root", nil, api.BuildResourceList("4", "4Gi"))
 	queue11 := buildQueueWithParents("q11", "q1", nil, api.BuildResourceList("1", "1Gi"))