Skip to content
Closed
Show file tree
Hide file tree
Changes from 1 commit
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 10 additions & 2 deletions pkg/scheduler/scheduler.go
Original file line number Diff line number Diff line change
Expand Up @@ -1038,6 +1038,14 @@ func (s *Scheduler) Filter(args extenderv1.ExtenderArgs) (*extenderv1.ExtenderFi
if args.Nodes != nil {
return s.filterSimulation(args, resourceReqs)
}
if args.NodeNames == nil {
Comment thread
nishantbkl3345-ship-it marked this conversation as resolved.
Resolved
// The live filter path has no candidate list to work with, and running it
// anyway would drop the pod from the scheduling caches for nothing.
err := fmt.Errorf("extender args contain neither Nodes nor NodeNames")

@moezdil moezdil Aug 30, 2026 •

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

getNodesUsage never reads its NodeNames parameter, and calcScore scores every node. so is rejecting right, or is nil safe enough?

klog.ErrorS(err, "Rejecting filter request", "pod", klog.KObj(args.Pod))

@moezdil moezdil Aug 30, 2026 •

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

which caller sends args with neither field set?

s.recordScheduleFilterResultEvent(args.Pod, EventReasonFilteringFailed, "", err)
return &extenderv1.ExtenderFilterResult{Error: err.Error()}, nil
}

if pi, ok := s.podManager.TakeAndDeletePod(args.Pod); ok {
s.quotaManager.RmUsage(args.Pod, pi.Devices)
Expand All @@ -1058,7 +1066,7 @@ func (s *Scheduler) Filter(args extenderv1.ExtenderArgs) (*extenderv1.ExtenderFi
}
if len((*nodeScores).NodeList) == 0 {
klog.V(4).InfoS("No available nodes meet the required scores", "pod", args.Pod.Name)
s.recordScheduleFilterResultEvent(args.Pod, EventReasonFilteringFailed, "", fmt.Errorf("no available node, %d nodes do not meet", len(*args.NodeNames)))
s.recordScheduleFilterResultEvent(args.Pod, EventReasonFilteringFailed, "", fmt.Errorf("no available node, %d nodes do not meet", nodeNamesLen(args.NodeNames)))
return &extenderv1.ExtenderFilterResult{
FailedNodes: failedNodes,
}, nil
Expand Down Expand Up @@ -1097,7 +1105,7 @@ func (s *Scheduler) Filter(args extenderv1.ExtenderArgs) (*extenderv1.ExtenderFi
}
}

successMsg := genSuccessMsg(len(*args.NodeNames), m.NodeID, nodeScores.NodeList)
successMsg := genSuccessMsg(nodeNamesLen(args.NodeNames), m.NodeID, nodeScores.NodeList)
s.recordScheduleFilterResultEvent(args.Pod, EventReasonFilteringSucceed, successMsg, nil)
res := extenderv1.ExtenderFilterResult{NodeNames: &[]string{m.NodeID}}
return &res, nil
Expand Down
53 changes: 53 additions & 0 deletions pkg/scheduler/scheduler_test.go
Original file line number Diff line number Diff line change
Expand Up @@ -2285,3 +2285,56 @@ func Test_Bind_PodGroupPodNonContentionErrorDoesNotRetry(t *testing.T) {
require.Equal(t, int32(1), mock.lockCalls.Load(),
"non-contention error must not trigger retry")
}

// TestFilterWithoutCandidateNodes covers extender args that carry a
// device-requesting pod but neither Nodes nor NodeNames. The live filter path
// used to dereference the nil *[]string while building the failure message and
// took the extender process down with it.
func TestFilterWithoutCandidateNodes(t *testing.T) {
s := NewScheduler()

sConfig := &config.Config{
NvidiaConfig: nvidia.NvidiaConfig{
ResourceCountName: "hami.io/gpu",
ResourceMemoryName: "hami.io/gpumem",
ResourceMemoryPercentageName: "hami.io/gpumem-percentage",
ResourceCoreName: "hami.io/gpucores",
DefaultGPUNum: 1,
},
}
require.NoError(t, config.InitDevicesWithConfig(sConfig))

pod := &corev1.Pod{
ObjectMeta: metav1.ObjectMeta{
Name: "no-candidate-nodes",
Namespace: "default",
UID: types.UID("no-candidate-nodes-uid"),
},
Spec: corev1.PodSpec{
Containers: []corev1.Container{{
Name: "worker",
Resources: corev1.ResourceRequirements{
Limits: corev1.ResourceList{
"hami.io/gpu": *resource.NewQuantity(1, resource.BinarySI),
"hami.io/gpucores": *resource.NewQuantity(20, resource.BinarySI),
"hami.io/gpumem": *resource.NewQuantity(1024, resource.BinarySI),
},
},
}},
},
}
require.True(t, s.podManager.AddPod(pod, "node-a", device.PodDevices{}))

res, err := s.Filter(extenderv1.ExtenderArgs{Pod: pod})
require.NoError(t, err)
require.NotNil(t, res)
require.Equal(t, "extender args contain neither Nodes nor NodeNames", res.Error)
require.Nil(t, res.NodeNames)
require.Nil(t, res.Nodes)

// The request is rejected before the live path evicts the pod from the
// scheduling caches, so an already-scheduled pod keeps its accounting.
cached, ok := s.podManager.GetPod(pod)
require.True(t, ok)
require.Equal(t, "node-a", cached.NodeID)
}
Loading