Initial commit (#1307)

Author: kfswain

apix/v1alpha2/inferenceobjective_types.go

@@ -63,20 +63,6 @@ type InferenceObjectiveList struct {
 // creation timestamp, will be selected to remain valid. In the event of a race
 // condition, one will be selected at random.
 type InferenceObjectiveSpec struct {
-	// ModelName is the name of the model as it will be set in the "model" parameter for an incoming request.
-	// ModelNames must be unique for a referencing InferencePool
-	// (names can be reused for a different pool in the same cluster).
-	// The modelName with the oldest creation timestamp is retained, and the incoming
-	// InferenceObjective's Ready status is set to false with a corresponding reason.
-	// In the rare case of a race condition, one Model will be selected randomly to be considered valid, and the other rejected.
-	// Names can be reserved without an underlying model configured in the pool.
-	// This can be done by specifying a target model and setting the weight to zero,
-	// an error will be returned specifying that no valid target model is found.
-	//
-	// +kubebuilder:validation:MaxLength=256
-	// +kubebuilder:validation:Required
-	// +kubebuilder:validation:XValidation:rule="self == oldSelf",message="modelName is immutable"
-	ModelName string `json:"modelName"`
 
 	// Criticality defines how important it is to serve the model compared to other models referencing the same pool.
 	// Criticality impacts how traffic is handled in resource constrained situations. It handles this by

client-go/applyconfiguration/apix/v1alpha2/inferenceobjectivespec.go

@@ -83,22 +83,6 @@ spec:
                 - Standard
                 - Sheddable
                 type: string
-              modelName:
-                description: |-
-                  ModelName is the name of the model as it will be set in the "model" parameter for an incoming request.
-                  ModelNames must be unique for a referencing InferencePool
-                  (names can be reused for a different pool in the same cluster).
-                  The modelName with the oldest creation timestamp is retained, and the incoming
-                  InferenceObjective's Ready status is set to false with a corresponding reason.
-                  In the rare case of a race condition, one Model will be selected randomly to be considered valid, and the other rejected.
-                  Names can be reserved without an underlying model configured in the pool.
-                  This can be done by specifying a target model and setting the weight to zero,
-                  an error will be returned specifying that no valid target model is found.
-                maxLength: 256
-                type: string
-                x-kubernetes-validations:
-                - message: modelName is immutable
-                  rule: self == oldSelf
               poolRef:
                 description: PoolRef is a reference to the inference pool, the pool
                   must exist in the same namespace.
@@ -171,7 +155,6 @@ spec:
                 - message: Weights should be set for all models, or none of the models.
                   rule: self.all(model, has(model.weight)) || self.all(model, !has(model.weight))
             required:
-            - modelName
             - poolRef
             type: object
           status:

config/crd/bases/inference.networking.x-k8s.io_inferenceobjectives.yaml

@@ -3,7 +3,6 @@ kind: InferenceObjective
 metadata:
   name: food-review
 spec:
-  modelName: food-review
   criticality: Standard
   poolRef:
     name: vllm-llama3-8b-instruct
@@ -16,7 +15,6 @@ kind: InferenceObjective
 metadata:
   name: base-model
 spec:
-  modelName: meta-llama/Llama-3.1-8B-Instruct
   criticality: Critical
   poolRef:
     name: vllm-llama3-8b-instruct
@@ -26,7 +24,6 @@ kind: InferenceObjective
 metadata:
   name: base-model-cpu
 spec:
-  modelName: Qwen/Qwen2.5-1.5B-Instruct
   criticality: Critical
   poolRef:
     name: vllm-llama3-8b-instruct

config/manifests/inferenceobjective.yaml

@@ -3,7 +3,6 @@ kind: InferenceObjective
 metadata:
   name: adapter-0
 spec:
-  modelName: adapter-0
   criticality: Critical
   poolRef:
     name: vllm-llama3-8b-instruct
@@ -18,7 +17,6 @@ kind: InferenceObjective
 metadata:
   name: adapter-1
 spec:
-  modelName: adapter-1
   criticality: Critical
   poolRef:
     name: vllm-llama3-8b-instruct
@@ -33,7 +31,6 @@ kind: InferenceObjective
 metadata:
   name: adapter-2
 spec:
-  modelName: adapter-2
   criticality: Critical
   poolRef:
     name: vllm-llama3-8b-instruct
@@ -48,7 +45,6 @@ kind: InferenceObjective
 metadata:
   name: adapter-3
 spec:
-  modelName: adapter-3
   criticality: Critical
   poolRef:
     name: vllm-llama3-8b-instruct
@@ -63,7 +59,6 @@ kind: InferenceObjective
 metadata:
   name: adapter-4
 spec:
-  modelName: adapter-4
   criticality: Critical
   poolRef:
     name: vllm-llama3-8b-instruct
@@ -78,7 +73,6 @@ kind: InferenceObjective
 metadata:
   name: adapter-5
 spec:
-  modelName: adapter-5
   criticality: Critical
   poolRef:
     name: vllm-llama3-8b-instruct
@@ -93,7 +87,6 @@ kind: InferenceObjective
 metadata:
   name: adapter-6
 spec:
-  modelName: adapter-6
   criticality: Critical
   poolRef:
     name: vllm-llama3-8b-instruct
@@ -108,7 +101,6 @@ kind: InferenceObjective
 metadata:
   name: adapter-7
 spec:
-  modelName: adapter-7
   criticality: Critical
   poolRef:
     name: vllm-llama3-8b-instruct
@@ -123,7 +115,6 @@ kind: InferenceObjective
 metadata:
   name: adapter-8
 spec:
-  modelName: adapter-8
   criticality: Critical
   poolRef:
     name: vllm-llama3-8b-instruct
@@ -138,7 +129,6 @@ kind: InferenceObjective
 metadata:
   name: adapter-9
 spec:
-  modelName: adapter-9
   criticality: Critical
   poolRef:
     name: vllm-llama3-8b-instruct
@@ -153,7 +143,6 @@ kind: InferenceObjective
 metadata:
   name: adapter-10
 spec:
-  modelName: adapter-10
   criticality: Critical
   poolRef:
     name: vllm-llama3-8b-instruct
@@ -168,7 +157,6 @@ kind: InferenceObjective
 metadata:
   name: adapter-11
 spec:
-  modelName: adapter-11
   criticality: Critical
   poolRef:
     name: vllm-llama3-8b-instruct
@@ -183,7 +171,6 @@ kind: InferenceObjective
 metadata:
   name: adapter-12
 spec:
-  modelName: adapter-12
   criticality: Critical
   poolRef:
     name: vllm-llama3-8b-instruct
@@ -199,7 +186,6 @@ kind: InferenceObjective
 metadata:
   name: adapter-13
 spec:
-  modelName: adapter-13
   criticality: Critical
   poolRef:
     name: vllm-llama3-8b-instruct
@@ -215,7 +201,6 @@ kind: InferenceObjective
 metadata:
   name: adapter-14
 spec:
-  modelName: adapter-14
   criticality: Critical
   poolRef:
     name: vllm-llama3-8b-instruct
@@ -231,7 +216,6 @@ kind: InferenceObjective
 metadata:
   name: base-model
 spec:
-  modelName: meta-llama/Llama-3.1-8B-Instruct
   criticality: Critical
   poolRef:
     name: vllm-llama3-8b-instruct

config/manifests/regression-testing/inferenceobjective.yaml

@@ -6,7 +6,6 @@ metadata:
   name: conformance-fake-model-server
   namespace: gateway-conformance-app-backend
 spec:
-  modelName: conformance-fake-model
   criticality: Critical # Mark it as critical to bypass the saturation check since the model server is fake and don't have such metrics. 
   poolRef:
     name: secondary-inference-pool

conformance/tests/epp_unavailable_fail_open.yaml

@@ -6,7 +6,6 @@ metadata:
   name: conformance-fake-model-server
   namespace: gateway-conformance-app-backend
 spec:
-  modelName: conformance-fake-model
   criticality: Critical # Mark it as critical to bypass the saturation check since the model server is fake and don't have such metrics. 
   poolRef:
     name: primary-inference-pool

conformance/tests/gateway_following_epp_routing.yaml

@@ -341,7 +341,6 @@ kind: InferenceModel
 metadata:
   name: sql-code-assist
 spec:
-  modelName: sql-code-assist
   poolRef:
     name: base-model-pool
 ---
@@ -350,7 +349,6 @@ kind: InferenceModel
 metadata:
   name: npc-bot
 spec:
-  modelName: npc-bot
   criticality: Critical
   targetModels:
     - name: npc-bot-v1

docs/proposals/002-api-proposal/README.md

@@ -18,10 +18,8 @@ package controller
 
 import (
 	"context"
-	"fmt"
 
 	"k8s.io/apimachinery/pkg/api/errors"
-	"k8s.io/apimachinery/pkg/types"
 	ctrl "sigs.k8s.io/controller-runtime"
 	"sigs.k8s.io/controller-runtime/pkg/client"
 	"sigs.k8s.io/controller-runtime/pkg/event"
@@ -58,60 +56,19 @@ func (c *InferenceObjectiveReconciler) Reconcile(ctx context.Context, req ctrl.R
 
 	if notFound || !infObjective.DeletionTimestamp.IsZero() || infObjective.Spec.PoolRef.Name != v1alpha2.ObjectName(c.PoolGKNN.Name) {
 		// InferenceObjective object got deleted or changed the referenced pool.
-		err := c.handleObjectiveDeleted(ctx, req.NamespacedName)
-		return ctrl.Result{}, err
+		c.Datastore.ObjectiveDelete(req.NamespacedName)
+		return ctrl.Result{}, nil
 	}
 
 	// Add or update if the InferenceObjective instance has a creation timestamp older than the existing entry of the model.
-	logger = logger.WithValues("poolRef", infObjective.Spec.PoolRef).WithValues("modelName", infObjective.Spec.ModelName)
-	if !c.Datastore.ObjectiveSetIfOlder(infObjective) {
-		logger.Info("Skipping InferenceObjective, existing instance has older creation timestamp")
-	} else {
-		logger.Info("Added/Updated InferenceObjective")
-	}
+	logger = logger.WithValues("poolRef", infObjective.Spec.PoolRef)
+	c.Datastore.ObjectiveSet(infObjective)
+	logger.Info("Added/Updated InferenceObjective")
 
 	return ctrl.Result{}, nil
 }
 
-func (c *InferenceObjectiveReconciler) handleObjectiveDeleted(ctx context.Context, req types.NamespacedName) error {
-	logger := log.FromContext(ctx)
-
-	// We will lookup and delete the modelName associated with this object, and search for
-	// other instances referencing the same modelName if exist, and store the oldest in
-	// its place. This ensures that the InferenceObjective with the oldest creation
-	// timestamp is active.
-	existing := c.Datastore.ObjectiveDelete(req)
-	if existing == nil {
-		// No entry exists in the first place, nothing to do.
-		return nil
-	}
-	logger.Info("InferenceObjective removed from datastore", "poolRef", existing.Spec.PoolRef, "modelName", existing.Spec.ModelName)
-
-	// TODO(#409): replace this backfill logic with one that is based on InferenceObjective Ready conditions once those are set by an external controller.
-	updated, err := c.Datastore.ObjectiveResync(ctx, c.Reader, existing.Spec.ModelName)
-	if err != nil {
-		return err
-	}
-	if updated {
-		logger.Info("Model replaced.", "modelName", existing.Spec.ModelName)
-	}
-	return nil
-}
-
-func indexInferenceObjectivesByModelName(obj client.Object) []string {
-	m, ok := obj.(*v1alpha2.InferenceObjective)
-	if !ok {
-		return nil
-	}
-	return []string{m.Spec.ModelName}
-}
-
 func (c *InferenceObjectiveReconciler) SetupWithManager(ctx context.Context, mgr ctrl.Manager) error {
-	// Create an index on ModelName for InferenceObjective objects.
-	indexer := mgr.GetFieldIndexer()
-	if err := indexer.IndexField(ctx, &v1alpha2.InferenceObjective{}, datastore.ModelNameIndexKey, indexInferenceObjectivesByModelName); err != nil {
-		return fmt.Errorf("setting index on ModelName for InferenceObjective: %w", err)
-	}
 	return ctrl.NewControllerManagedBy(mgr).
 		For(&v1alpha2.InferenceObjective{}).
 		WithEventFilter(predicate.Funcs{