Address comments

Author: liu-cong

pkg/epp/scheduling/plugins/filter.go

@@ -28,19 +28,19 @@ import (
 	logutil "sigs.k8s.io/gateway-api-inference-extension/pkg/epp/util/logging"
 )
 
-type BasicFilter struct {
+type Filter struct {
 	name   string
 	filter filterFunc
 }
 
-func (bf *BasicFilter) Name() string {
+func (bf *Filter) Name() string {
 	if bf == nil {
 		return "nil"
 	}
 	return bf.name
 }
 
-func (bf *BasicFilter) Filter(ctx *types.Context, pods []types.Pod) ([]types.Pod, error) {
+func (bf *Filter) Filter(ctx *types.Context, pods []types.Pod) ([]types.Pod, error) {
 	loggerTrace := ctx.Logger.V(logutil.TRACE)
 	loggerTrace.Info("Running a filter", "name", bf.Name(), "podCount", len(pods))
 
@@ -123,7 +123,7 @@ func toFilterFunc(pp podPredicate) filterFunc {
 	}
 }
 
-var LeastQueueFilter = &BasicFilter{
+var LeastQueueFilter = &Filter{
 	name:   "least queuing",
 	filter: leastQueuingFilterFunc,
 }
@@ -157,12 +157,12 @@ func leastQueuingFilterFunc(ctx *types.Context, pods []types.Pod) ([]types.Pod,
 	return filtered, nil
 }
 
-var LowQueueFilter = &BasicFilter{
+var LowQueueFilter = &Filter{
 	name:   "low queueing filter",
 	filter: toFilterFunc((queueThresholdPredicate(config.Conf.QueueingThresholdLoRA))),
 }
 
-var LeastKVCacheFilter = &BasicFilter{
+var LeastKVCacheFilter = &Filter{
 	name:   "least KV cache percent",
 	filter: leastKVCacheFilterFunc,
 }
@@ -195,7 +195,7 @@ func leastKVCacheFilterFunc(ctx *types.Context, pods []types.Pod) ([]types.Pod,
 	return filtered, nil
 }
 
-var LoRAAffinityFilter = &BasicFilter{
+var LoRAAffinityFilter = &Filter{
 	name:   "affinity LoRA",
 	filter: loRASoftAffinityFilterFunc,
 }
@@ -254,12 +254,12 @@ func loRASoftAffinityFilterFunc(ctx *types.Context, pods []types.Pod) ([]types.P
 	return filtered_available, nil
 }
 
-var HasCapacityFilter = &BasicFilter{
+var HasCapacityFilter = &Filter{
 	name:   "has capacity for sheddable requests",
 	filter: toFilterFunc(queueThresholdPredicate(config.Conf.QueueThresholdCritical).and(kvCacheThresholdPredicate(config.Conf.KVCacheThreshold))),
 }
 
-var DropRequestFilter = &BasicFilter{
+var DropRequestFilter = &Filter{
 	name: "drop request",
 	filter: func(ctx *types.Context, pods []types.Pod) ([]types.Pod, error) {
 		ctx.Logger.V(logutil.DEFAULT).Info("Request dropped", "request", ctx.Req)
@@ -269,7 +269,7 @@ var DropRequestFilter = &BasicFilter{
 	},
 }
 
-var NoopFilter = &BasicFilter{
+var NoopFilter = &Filter{
 	name: "noop",
 	filter: func(ctx *types.Context, pods []types.Pod) ([]types.Pod, error) {
 		ctx.Logger.V(logutil.DEBUG).Info("All pods pass")

pkg/epp/scheduling/plugins/filter_test.go

@@ -40,7 +40,7 @@ func TestFilter(t *testing.T) {
 		{
 			name: "simple filter without successor, failure",
 			filter: &DecisionTreeFilter{
-				Current: &BasicFilter{
+				Current: &Filter{
 					name: "error",
 					filter: func(ctx *types.Context, pods []types.Pod) ([]types.Pod, error) {
 						return nil, errors.New("filter error")

pkg/epp/scheduling/scheduler.go

@@ -70,9 +70,9 @@ func NewScheduler(datastore Datastore) *Scheduler {
 
 	return &Scheduler{
 		datastore:           datastore,
-		preSchedulePlugins:  []types.PreSchedule{defaultPlugin},
-		postSchedulePlugins: []types.PostSchedule{defaultPlugin},
-		scorers:             []types.Scorer{defaultPlugin},
+		preSchedulePlugins:  []types.PreSchedule{},
+		postSchedulePlugins: []types.PostSchedule{},
+		scorers:             []types.Scorer{},
 		filters:             []types.Filter{defaultPlugin},
 		picker:              defaultPlugin,
 	}
@@ -97,7 +97,7 @@ func (s *Scheduler) Schedule(ctx context.Context, req *types.LLMRequest) (*types
 	loggerDebug := logger.V(logutil.DEBUG)
 
 	// Snapshot pod metrics from the datastore to:
-	// 1. Reduce conCurrent access to the datastore.
+	// 1. Reduce concurrent access to the datastore.
 	// 2. Ensure consistent data during the scheduling operation of a request.
 	sCtx := types.NewContext(ctx, req, types.ToSchedulerPodMetrics(s.datastore.PodGetAll()))
 	loggerDebug.Info(fmt.Sprintf("Scheduling a request. Metrics: %+v", sCtx.PodsSnapshot))
@@ -197,7 +197,6 @@ func runScorersForPod(ctx *types.Context, scorers []types.Scorer, pod types.Pod)
 }
 
 type defaultPlugin struct {
-	plugins.NoopPlugin
 	plugins.RandomPicker
 }
 

pkg/epp/scheduling/types/types.go

@@ -37,7 +37,7 @@ type LLMRequest struct {
 }
 
 func (r *LLMRequest) String() string {
-	return fmt.Sprintf("Model: %s, TargetModels: %v, ResolvedTargetModel: %s, Critical: %t, Prompt Size: %v", r.Model, r.TargetModels, r.ResolvedTargetModel, r.Critical, len(r.Prompt))
+	return fmt.Sprintf("Model: %s, TargetModels: %v, ResolvedTargetModel: %s, Critical: %t, PromptLength: %v", r.Model, r.TargetModels, r.ResolvedTargetModel, r.Critical, len(r.Prompt))
 }
 
 // Context holds contextual information during a scheduling operation.