fix(scheduler): clear node stats when it is removed

pkg/scheduler/nodes.go

@@ -100,6 +100,12 @@ func (m *nodeManager) rmNodeDevices(nodeID string, deviceVendor string) {
 	klog.InfoS("Removing device from node", "nodeName", nodeID, "deviceVendor", deviceVendor, "remainingDevices", devices)
 }
 
+func (m *nodeManager) removeNode(nodeID string) {
+	m.mutex.Lock()
+	defer m.mutex.Unlock()
+	delete(m.nodes, nodeID)
+}
+
 func (m *nodeManager) GetNode(nodeID string) (*util.NodeInfo, error) {
 	m.mutex.RLock()
 	defer m.mutex.RUnlock()

pkg/scheduler/scheduler.go

@@ -300,6 +300,18 @@ func (s *Scheduler) RegisterFromNodeAnnotations() {
 				}
 			}
 		}
+		activeNodes := make(map[string]struct{}, len(nodeNames))
+		for _, name := range nodeNames {
+			activeNodes[name] = struct{}{}
+		}
+		allRegistered, _ := s.ListNodes()
+		for id := range allRegistered {
+			if _, exists := activeNodes[id]; !exists {
+				klog.InfoS("Removing stale node from scheduler", "nodeName", id)
+				s.removeNode(id)
+			}
+		}
+
 		_, _, err = s.getNodesUsage(&nodeNames, nil)
 		if err != nil {
 			klog.ErrorS(err, "Failed to get node usage", "nodeNames", nodeNames)