Use cached TemplateNodeInfos in DRA processor

This change makes the DRACustomResourcesProcessor use TemplateNodeInfo
from the previous autoscaling loop iteration preserving custom node
options and devices. It falls back to NodeGroup.TemplateNodeInfo() if
there are no good candidates.

Fixes #8881

Author: Choraden

cluster-autoscaler/processors/customresources/dra_processor.go

@@ -19,6 +19,7 @@ package customresources
 import (
 	apiv1 "k8s.io/api/core/v1"
 	resourceapi "k8s.io/api/resource/v1"
+	"k8s.io/autoscaler/cluster-autoscaler/simulator/framework"
 
 	"k8s.io/apimachinery/pkg/util/sets"
 	"k8s.io/autoscaler/cluster-autoscaler/cloudprovider"
@@ -57,11 +58,19 @@ func (p *DraCustomResourcesProcessor) FilterOutNodesWithUnreadyResources(autosca
 			continue
 		}
 
-		nodeInfo, err := ng.TemplateNodeInfo()
-		if err != nil {
-			newReadyNodes = append(newReadyNodes, node)
-			klog.Warningf("Failed to get template node info for node group %s with error: %v", ng.Id(), err)
-			continue
+		var nodeInfo *framework.NodeInfo
+		if autoscalingCtx.TemplateNodeInfoRegistry != nil {
+			if ni, found := autoscalingCtx.TemplateNodeInfoRegistry.GetNodeInfo(ng.Id()); found {
+				nodeInfo = ni
+			}
+		}
+		if nodeInfo == nil {
+			nodeInfo, err = ng.TemplateNodeInfo()
+			if err != nil {
+				newReadyNodes = append(newReadyNodes, node)
+				klog.Warningf("Failed to get template node info for node group %s with error: %v", ng.Id(), err)
+				continue
+			}
 		}
 
 		nodeResourcesSlices, _ := draSnapshot.NodeResourceSlices(node.Name)

cluster-autoscaler/processors/customresources/dra_processor_test.go

@@ -35,12 +35,32 @@ import (
 	utils "k8s.io/autoscaler/cluster-autoscaler/utils/test"
 )
 
+type mockTemplateNodeInfoRegistry struct {
+	nodeInfos map[string]*framework.NodeInfo
+}
+
+func newMockTemplateNodeInfoRegistry(nodeInfos map[string]*framework.NodeInfo) *mockTemplateNodeInfoRegistry {
+	return &mockTemplateNodeInfoRegistry{
+		nodeInfos: nodeInfos,
+	}
+}
+
+func (m *mockTemplateNodeInfoRegistry) GetNodeInfo(id string) (*framework.NodeInfo, bool) {
+	nodeInfo, found := m.nodeInfos[id]
+	return nodeInfo, found
+}
+
+func (m *mockTemplateNodeInfoRegistry) GetNodeInfos() map[string]*framework.NodeInfo {
+	return m.nodeInfos
+}
+
 func TestFilterOutNodesWithUnreadyDRAResources(t *testing.T) {
 	testCases := map[string]struct {
 		nodeGroupsAllNodes        map[string][]*apiv1.Node
 		nodeGroupsTemplatesSlices map[string][]*resourceapi.ResourceSlice
 		nodesSlices               map[string][]*resourceapi.ResourceSlice
 		expectedNodesReadiness    map[string]bool
+		registryNodeInfos         map[string]*framework.NodeInfo
 	}{
 		"1 DRA node group all totally ready": {
 			nodeGroupsAllNodes: map[string][]*apiv1.Node{
@@ -306,6 +326,29 @@ func TestFilterOutNodesWithUnreadyDRAResources(t *testing.T) {
 				"node_7": true,
 			},
 		},
+		"Custom DRA driver retrieved via cached template node info": {
+			nodeGroupsAllNodes: map[string][]*apiv1.Node{
+				"ng1": {
+					buildTestNode("node_1", true),
+					buildTestNode("node_2", true),
+				},
+			},
+			nodeGroupsTemplatesSlices: map[string][]*resourceapi.ResourceSlice{},
+			registryNodeInfos: map[string]*framework.NodeInfo{
+				"ng1": framework.NewNodeInfo(
+					buildTestNode("ng1_template", true),
+					createNodeResourceSlices("ng1_template", []int{1}),
+				),
+			},
+			nodesSlices: map[string][]*resourceapi.ResourceSlice{
+				"node_1": createNodeResourceSlices("node_1", []int{1}),
+				"node_2": {},
+			},
+			expectedNodesReadiness: map[string]bool{
+				"node_1": true,
+				"node_2": false,
+			},
+		},
 	}
 
 	for tcName, tc := range testCases {
@@ -336,7 +379,11 @@ func TestFilterOutNodesWithUnreadyDRAResources(t *testing.T) {
 			clusterSnapshotStore.SetClusterState([]*apiv1.Node{}, []*apiv1.Pod{}, draSnapshot)
 			clusterSnapshot, _, _ := testsnapshot.NewCustomTestSnapshotAndHandle(clusterSnapshotStore)
 
-			autoscalingCtx := &ca_context.AutoscalingContext{CloudProvider: provider, ClusterSnapshot: clusterSnapshot}
+			autoscalingCtx := &ca_context.AutoscalingContext{
+				CloudProvider:            provider,
+				ClusterSnapshot:          clusterSnapshot,
+				TemplateNodeInfoRegistry: newMockTemplateNodeInfoRegistry(tc.registryNodeInfos),
+			}
 			processor := DraCustomResourcesProcessor{}
 			newAllNodes, newReadyNodes := processor.FilterOutNodesWithUnreadyResources(autoscalingCtx, initialAllNodes, initialReadyNodes, draSnapshot)