register controller.

Author: zetxqx

config/charts/inferencepool/templates/rbac.yaml

@@ -46,7 +46,7 @@ metadata:
     {{- include "gateway-api-inference-extension.labels" . | nindent 4 }}
 rules:
 - apiGroups: ["inference.networking.x-k8s.io"]
-  resources: ["inferenceobjectives"]
+  resources: ["inferenceobjectives", "inferencemodelrewrites"]
   verbs: ["get", "watch", "list"]
 - apiGroups: ["{{ (split "/" .Values.inferencePool.apiVersion)._0 }}"]
   resources: ["inferencepools"]

pkg/epp/controller/inferencemodelrewrite_reconciler.go

@@ -54,7 +54,9 @@ func (c *InferenceModelRewriteReconciler) Reconcile(ctx context.Context, req ctr
 		notFound = true
 	}
 
-	if notFound || !infModelRewrite.DeletionTimestamp.IsZero() || infModelRewrite.Spec.PoolRef.Name != v1alpha2.ObjectName(c.PoolGKNN.Name) {
+	if notFound || !infModelRewrite.DeletionTimestamp.IsZero() || infModelRewrite.Spec.PoolRef == nil ||
+		infModelRewrite.Spec.PoolRef.Name != v1alpha2.ObjectName(c.PoolGKNN.Name) ||
+		(infModelRewrite.Spec.PoolRef.Group != v1alpha2.Group(c.PoolGKNN.Group) && infModelRewrite.Spec.PoolRef.Group != "inference.networking.x-k8s.io") {
 		// InferenceModelRewrite object got deleted or changed the referenced pool.
 		c.Datastore.RewriteDelete(req.NamespacedName)
 		return ctrl.Result{}, nil

pkg/epp/controller/inferencemodelrewrite_reconciler_test.go

@@ -36,6 +36,7 @@ import (
 	"sigs.k8s.io/gateway-api-inference-extension/pkg/common"
 	backendmetrics "sigs.k8s.io/gateway-api-inference-extension/pkg/epp/backend/metrics"
 	"sigs.k8s.io/gateway-api-inference-extension/pkg/epp/datastore"
+	poolutil "sigs.k8s.io/gateway-api-inference-extension/pkg/epp/util/pool"
 	utiltest "sigs.k8s.io/gateway-api-inference-extension/pkg/epp/util/testing"
 )
 
@@ -48,7 +49,10 @@ var (
 			CreationTimestamp: metav1.Unix(1000, 0),
 		},
 		Spec: v1alpha2.InferenceModelRewriteSpec{
-			PoolRef: &v1alpha2.PoolObjectReference{Name: v1alpha2.ObjectName(poolForRewrite.Name)},
+			PoolRef: &v1alpha2.PoolObjectReference{
+				Name:  v1alpha2.ObjectName(poolForRewrite.Name),
+				Group: v1alpha2.Group(poolForRewrite.GroupVersionKind().Group),
+			},
 		},
 	}
 	rewrite1Pool2 = &v1alpha2.InferenceModelRewrite{
@@ -58,7 +62,10 @@ var (
 			CreationTimestamp: metav1.Unix(1001, 0),
 		},
 		Spec: v1alpha2.InferenceModelRewriteSpec{
-			PoolRef: &v1alpha2.PoolObjectReference{Name: "test-pool2"},
+			PoolRef: &v1alpha2.PoolObjectReference{
+				Name:  "test-pool2",
+				Group: v1alpha2.Group(poolForRewrite.GroupVersionKind().Group),
+			},
 		},
 	}
 	rewrite1Updated = &v1alpha2.InferenceModelRewrite{
@@ -68,8 +75,11 @@ var (
 			CreationTimestamp: metav1.Unix(1003, 0),
 		},
 		Spec: v1alpha2.InferenceModelRewriteSpec{
-			PoolRef: &v1alpha2.PoolObjectReference{Name: v1alpha2.ObjectName(poolForRewrite.Name)},
-			Rules:   []v1alpha2.InferenceModelRewriteRule{{}},
+			PoolRef: &v1alpha2.PoolObjectReference{
+				Name:  v1alpha2.ObjectName(poolForRewrite.Name),
+				Group: v1alpha2.Group(poolForRewrite.GroupVersionKind().Group),
+			},
+			Rules: []v1alpha2.InferenceModelRewriteRule{{}},
 		},
 	}
 	rewrite1Deleted = &v1alpha2.InferenceModelRewrite{
@@ -78,9 +88,13 @@ var (
 			Namespace:         rewrite1.Namespace,
 			CreationTimestamp: metav1.Unix(1004, 0),
 			DeletionTimestamp: &metav1.Time{Time: time.Now()},
+			Finalizers:        []string{"deleted"},
 		},
 		Spec: v1alpha2.InferenceModelRewriteSpec{
-			PoolRef: &v1alpha2.PoolObjectReference{Name: v1alpha2.ObjectName(poolForRewrite.Name)},
+			PoolRef: &v1alpha2.PoolObjectReference{
+				Name:  v1alpha2.ObjectName(poolForRewrite.Name),
+				Group: v1alpha2.Group(poolForRewrite.GroupVersionKind().Group),
+			},
 		},
 	}
 	rewrite2 = &v1alpha2.InferenceModelRewrite{
@@ -90,7 +104,10 @@ var (
 			CreationTimestamp: metav1.Unix(1001, 0),
 		},
 		Spec: v1alpha2.InferenceModelRewriteSpec{
-			PoolRef: &v1alpha2.PoolObjectReference{Name: v1alpha2.ObjectName(poolForRewrite.Name)},
+			PoolRef: &v1alpha2.PoolObjectReference{
+				Name:  v1alpha2.ObjectName(poolForRewrite.Name),
+				Group: v1alpha2.Group(poolForRewrite.GroupVersionKind().Group),
+			},
 		},
 	}
 )
@@ -155,7 +172,7 @@ func TestInferenceModelRewriteReconciler(t *testing.T) {
 			_ = v1alpha2.Install(scheme)
 			_ = v1.Install(scheme)
 			initObjs := []client.Object{}
-			if test.rewrite != nil && test.rewrite.DeletionTimestamp.IsZero() {
+			if test.rewrite != nil {
 				initObjs = append(initObjs, test.rewrite)
 			}
 			for _, r := range test.rewritesInAPIServer {
@@ -170,7 +187,8 @@ func TestInferenceModelRewriteReconciler(t *testing.T) {
 			for _, r := range test.rewritesInStore {
 				ds.RewriteSet(r)
 			}
-			_ = ds.PoolSet(context.Background(), fakeClient, poolForRewrite)
+			endpointPool := poolutil.InferencePoolToEndpointPool(poolForRewrite)
+			_ = ds.PoolSet(context.Background(), fakeClient, endpointPool)
 			reconciler := &InferenceModelRewriteReconciler{
 				Reader:    fakeClient,
 				Datastore: ds,

pkg/epp/datastore/datastore.go

@@ -31,7 +31,6 @@ import (
 
 	"sigs.k8s.io/controller-runtime/pkg/client"
 	"sigs.k8s.io/controller-runtime/pkg/log"
-	v1 "sigs.k8s.io/gateway-api-inference-extension/api/v1"
 	"sigs.k8s.io/gateway-api-inference-extension/apix/v1alpha2"
 	backendmetrics "sigs.k8s.io/gateway-api-inference-extension/pkg/epp/backend/metrics"
 	"sigs.k8s.io/gateway-api-inference-extension/pkg/epp/datalayer"
@@ -101,7 +100,7 @@ type datastore struct {
 	parentCtx context.Context
 	// mu is used to synchronize access to pool, objectives, and rewrites.
 	mu   sync.RWMutex
-	pool *v1.InferencePool
+	pool *datalayer.EndpointPool
 	// key: InferenceObjective name, value: *InferenceObjective
 	objectives map[string]*v1alpha2.InferenceObjective
 	// rewrites store for InferenceModelRewrite objects.
@@ -156,7 +155,7 @@ func (ds *datastore) PoolSet(ctx context.Context, reader client.Reader, endpoint
 	return nil
 }
 
-func (ds *datastore) PoolGet() (*v1.InferencePool, error) {
+func (ds *datastore) PoolGet() (*datalayer.EndpointPool, error) {
 	ds.mu.RLock()
 	defer ds.mu.RUnlock()
 	if !ds.PoolHasSynced() {

pkg/epp/server/controller_manager.go

@@ -55,6 +55,16 @@ func defaultManagerOptions(disableK8sCrdReconcile bool, gknn common.GKNN, metric
 						gknn.Namespace: {},
 					},
 				},
+				&v1alpha2.InferenceObjective{}: {
+					Namespaces: map[string]cache.Config{
+						gknn.Namespace: {},
+					},
+				},
+				&v1alpha2.InferenceModelRewrite{}: {
+					Namespaces: map[string]cache.Config{
+						gknn.Namespace: {},
+					},
+				},
 			},
 		},
 		Metrics: metricsServerOptions,

pkg/epp/server/runserver.go

@@ -131,6 +131,14 @@ func (r *ExtProcServerRunner) SetupWithManager(ctx context.Context, mgr ctrl.Man
 		}
 	}
 
+	if err := (&controller.InferenceModelRewriteReconciler{
+		Datastore: r.Datastore,
+		Reader:    mgr.GetClient(),
+		PoolGKNN:  r.GKNN,
+	}).SetupWithManager(ctx, mgr); err != nil {
+		return fmt.Errorf("failed setting up InferenceModelRewriteReconciler: %w", err)
+	}
+
 	if err := (&controller.PodReconciler{
 		Datastore: r.Datastore,
 		Reader:    mgr.GetClient(),

test/testdata/inferencepool-e2e.yaml

@@ -44,7 +44,7 @@ metadata:
   namespace: $E2E_NS
 rules:
 - apiGroups: [ "inference.networking.x-k8s.io" ]
-  resources: [ "inferenceobjectives", "inferencepools" ]
+  resources: [ "inferenceobjectives", "inferencepools", "inferencemodelrewrites" ]
   verbs: [ "get", "watch", "list" ]
 - apiGroups: [ "inference.networking.k8s.io" ]
   resources: [ "inferencepools" ]