integrate model handler into HTTP handler

Author: ilopezluna

main.go

@@ -152,10 +152,8 @@ func main() {
 			mlx.Name:      mlxBackend,
 		},
 		llamaCppBackend,
-		modelHandler,
 		modelManager,
 		http.DefaultClient,
-		nil,
 		metrics.NewTracker(
 			http.DefaultClient,
 			log.WithField("component", "metrics"),
@@ -166,7 +164,7 @@ func main() {
 	)
 
 	// Create the HTTP handler for the scheduler
-	schedulerHTTP := scheduling.NewHTTPHandler(scheduler, nil)
+	schedulerHTTP := scheduling.NewHTTPHandler(scheduler, modelHandler, nil)
 
 	router := routing.NewNormalizedServeMux()
 

pkg/inference/scheduling/http_handler.go

@@ -12,6 +12,7 @@ import (
 
 	"github.com/docker/model-runner/pkg/distribution/distribution"
 	"github.com/docker/model-runner/pkg/inference"
+	"github.com/docker/model-runner/pkg/inference/models"
 	"github.com/docker/model-runner/pkg/metrics"
 	"github.com/docker/model-runner/pkg/middleware"
 )
@@ -23,15 +24,18 @@ type HTTPHandler struct {
 	scheduler   *Scheduler
 	router      *http.ServeMux
 	httpHandler http.Handler
-	lock        sync.RWMutex
+	// modelHandler is the shared model handler.
+	modelHandler *models.HTTPHandler
+	lock         sync.RWMutex
 }
 
 // NewHTTPHandler creates a new HTTP handler that wraps the scheduler.
 // This is the primary HTTP interface for the scheduling package.
-func NewHTTPHandler(s *Scheduler, allowedOrigins []string) *HTTPHandler {
+func NewHTTPHandler(s *Scheduler, modelHandler *models.HTTPHandler, allowedOrigins []string) *HTTPHandler {
 	h := &HTTPHandler{
-		scheduler: s,
-		router:    http.NewServeMux(),
+		scheduler:    s,
+		modelHandler: modelHandler,
+		router:       http.NewServeMux(),
 	}
 
 	// Register routes
@@ -217,7 +221,7 @@ func (h *HTTPHandler) handleOpenAIInference(w http.ResponseWriter, r *http.Reque
 // handleModels handles GET /engines/{backend}/v1/models* requests
 // by delegating to the model manager
 func (h *HTTPHandler) handleModels(w http.ResponseWriter, r *http.Request) {
-	h.scheduler.modelHandler.ServeHTTP(w, r)
+	h.modelHandler.ServeHTTP(w, r)
 }
 
 // GetBackendStatus returns the status of all backends.

pkg/inference/scheduling/scheduler.go

@@ -30,8 +30,6 @@ type Scheduler struct {
 	backends map[string]inference.Backend
 	// defaultBackend is the default inference backend. It may be nil.
 	defaultBackend inference.Backend
-	// modelHandler is the shared model handler.
-	modelHandler *models.HTTPHandler
 	// modelManager is the shared model manager.
 	modelManager *models.Manager
 	// installer is the backend installer.
@@ -49,10 +47,8 @@ func NewScheduler(
 	log logging.Logger,
 	backends map[string]inference.Backend,
 	defaultBackend inference.Backend,
-	handler *models.HTTPHandler,
 	modelManager *models.Manager,
 	httpClient *http.Client,
-	allowedOrigins []string,
 	tracker *metrics.Tracker,
 	sysMemInfo memory.SystemMemoryInfo,
 ) *Scheduler {
@@ -63,7 +59,6 @@ func NewScheduler(
 		log:            log,
 		backends:       backends,
 		defaultBackend: defaultBackend,
-		modelHandler:   handler,
 		modelManager:   modelManager,
 		installer:      newInstaller(log, backends, httpClient),
 		loader:         newLoader(log, backends, modelManager, openAIRecorder, sysMemInfo),

pkg/inference/scheduling/scheduler_test.go

@@ -44,8 +44,8 @@ func TestCors(t *testing.T) {
 			discard := logrus.New()
 			discard.SetOutput(io.Discard)
 			log := logrus.NewEntry(discard)
-			s := NewScheduler(log, nil, nil, nil, nil, nil, []string{"*"}, nil, systemMemoryInfo{})
-			httpHandler := NewHTTPHandler(s, []string{"*"})
+			s := NewScheduler(log, nil, nil, nil, nil, nil, systemMemoryInfo{})
+			httpHandler := NewHTTPHandler(s, nil, []string{"*"})
 			req := httptest.NewRequest(http.MethodOptions, "http://model-runner.docker.internal"+tt.path, http.NoBody)
 			req.Header.Set("Origin", "docker.com")
 			w := httptest.NewRecorder()