Filter inference objectives based on inference pool group (#1306)

* filter inference objective based on inference pool group

* update eventPredicate

* Fix e2e test by explicitly specifying inference pool group

* fix unit test

* minor changes to unit test

Author: nicolexin

pkg/epp/controller/inferenceobjective_reconciler.go

@@ -54,7 +54,7 @@ func (c *InferenceObjectiveReconciler) Reconcile(ctx context.Context, req ctrl.R
 		notFound = true
 	}
 
-	if notFound || !infObjective.DeletionTimestamp.IsZero() || infObjective.Spec.PoolRef.Name != v1alpha2.ObjectName(c.PoolGKNN.Name) {
+	if notFound || !infObjective.DeletionTimestamp.IsZero() || infObjective.Spec.PoolRef.Name != v1alpha2.ObjectName(c.PoolGKNN.Name) || infObjective.Spec.PoolRef.Group != v1alpha2.Group(c.PoolGKNN.Group) {
 		// InferenceObjective object got deleted or changed the referenced pool.
 		c.Datastore.ObjectiveDelete(req.NamespacedName)
 		return ctrl.Result{}, nil
@@ -83,5 +83,5 @@ func (c *InferenceObjectiveReconciler) SetupWithManager(ctx context.Context, mgr
 }
 
 func (c *InferenceObjectiveReconciler) eventPredicate(infObjective *v1alpha2.InferenceObjective) bool {
-	return string(infObjective.Spec.PoolRef.Name) == c.PoolGKNN.Name
+	return string(infObjective.Spec.PoolRef.Name) == c.PoolGKNN.Name && string(infObjective.Spec.PoolRef.Group) == c.PoolGKNN.Group
 }

pkg/epp/controller/inferenceobjective_reconciler_test.go

@@ -45,27 +45,37 @@ var (
 			Namespace(pool.Namespace).
 			Criticality(v1alpha2.Standard).
 			CreationTimestamp(metav1.Unix(1000, 0)).
-			PoolName(pool.Name).ObjRef()
+			PoolName(pool.Name).
+			PoolGroup("inference.networking.k8s.io").ObjRef()
 	infObjective1Pool2 = utiltest.MakeInferenceObjective(infObjective1.Name).
 				Namespace(infObjective1.Namespace).
 				Criticality(*infObjective1.Spec.Criticality).
 				CreationTimestamp(metav1.Unix(1001, 0)).
-				PoolName("test-pool2").ObjRef()
+				PoolName("test-pool2").
+				PoolGroup("inference.networking.k8s.io").ObjRef()
 	infObjective1Critical = utiltest.MakeInferenceObjective(infObjective1.Name).
 				Namespace(infObjective1.Namespace).
 				Criticality(v1alpha2.Critical).
 				CreationTimestamp(metav1.Unix(1003, 0)).
-				PoolName(pool.Name).ObjRef()
+				PoolName(pool.Name).
+				PoolGroup("inference.networking.k8s.io").ObjRef()
 	infObjective1Deleted = utiltest.MakeInferenceObjective(infObjective1.Name).
 				Namespace(infObjective1.Namespace).
 				CreationTimestamp(metav1.Unix(1004, 0)).
 				DeletionTimestamp().
-				PoolName(pool.Name).ObjRef()
-
+				PoolName(pool.Name).
+				PoolGroup("inference.networking.k8s.io").ObjRef()
+	infObjective1DiffGroup = utiltest.MakeInferenceObjective(infObjective1.Name).
+				Namespace(pool.Namespace).
+				Criticality(v1alpha2.Standard).
+				CreationTimestamp(metav1.Unix(1005, 0)).
+				PoolName(pool.Name).
+				PoolGroup("inference.networking.x-k8s.io").ObjRef()
 	infObjective2 = utiltest.MakeInferenceObjective("model2").
 			Namespace(pool.Namespace).
 			CreationTimestamp(metav1.Unix(1000, 0)).
-			PoolName(pool.Name).ObjRef()
+			PoolName(pool.Name).
+			PoolGroup("inference.networking.k8s.io").ObjRef()
 )
 
 func TestInferenceObjectiveReconciler(t *testing.T) {
@@ -119,6 +129,17 @@ func TestInferenceObjectiveReconciler(t *testing.T) {
 			objective:          infObjective2,
 			wantObjectives:     []*v1alpha2.InferenceObjective{infObjective1, infObjective2},
 		},
+		{
+			name:               "Objective deleted due to group mismatch for the inference pool",
+			objectivessInStore: []*v1alpha2.InferenceObjective{infObjective1},
+			objective:          infObjective1DiffGroup,
+			wantObjectives:     []*v1alpha2.InferenceObjective{},
+		},
+		{
+			name:           "Objective ignored due to group mismatch for the inference pool",
+			objective:      infObjective1DiffGroup,
+			wantObjectives: []*v1alpha2.InferenceObjective{},
+		},
 	}
 	for _, test := range tests {
 		t.Run(test.name, func(t *testing.T) {
@@ -134,7 +155,6 @@ func TestInferenceObjectiveReconciler(t *testing.T) {
 			for _, m := range test.objectivesInAPIServer {
 				initObjs = append(initObjs, m)
 			}
-
 			fakeClient := fake.NewClientBuilder().
 				WithScheme(scheme).
 				WithObjects(initObjs...).

pkg/epp/util/testing/wrappers.go

@@ -142,7 +142,12 @@ func (m *InferenceObjectiveWrapper) TargetModel(modelName string) *InferenceObje
 }
 
 func (m *InferenceObjectiveWrapper) PoolName(poolName string) *InferenceObjectiveWrapper {
-	m.Spec.PoolRef = v1alpha2.PoolObjectReference{Name: v1alpha2.ObjectName(poolName)}
+	m.Spec.PoolRef.Name = v1alpha2.ObjectName(poolName)
+	return m
+}
+
+func (m *InferenceObjectiveWrapper) PoolGroup(poolGroup string) *InferenceObjectiveWrapper {
+	m.Spec.PoolRef.Group = v1alpha2.Group(poolGroup)
 	return m
 }
 
@@ -175,6 +180,10 @@ func MakeInferencePool(name string) *InferencePoolWrapper {
 			ObjectMeta: metav1.ObjectMeta{
 				Name: name,
 			},
+			TypeMeta: metav1.TypeMeta{
+				APIVersion: "inference.networking.k8s.io/v1",
+				Kind:       "InferencePool",
+			},
 			Spec: v1.InferencePoolSpec{},
 		},
 	}

test/testdata/inferencepool-with-model-hermetic.yaml

@@ -19,6 +19,7 @@ spec:
   criticality: Critical
   poolRef:
     name: vllm-llama3-8b-instruct-pool
+    group: inference.networking.k8s.io
   targetModels:
   - name: sql-lora-1fdg2
     weight: 100
@@ -31,6 +32,7 @@ metadata:
 spec:
   poolRef:
     name: vllm-llama3-8b-instruct-pool
+    group: inference.networking.k8s.io
   targetModels:
   - name: sql-lora-1fdg3
     weight: 100
@@ -44,6 +46,7 @@ spec:
   criticality: Critical
   poolRef:
     name: vllm-llama3-8b-instruct-pool
+    group: inference.networking.k8s.io
   targetModels:
   - name: my-model-12345
     weight: 100    
@@ -56,4 +59,5 @@ metadata:
 spec:
   criticality: Critical
   poolRef:
-    name: vllm-llama3-8b-instruct-pool
+    name: vllm-llama3-8b-instruct-pool
+    group: inference.networking.k8s.io