Load model proto with secrets once at server initialization (#837)

* Initial plan

* Load model proto with secrets and inject into servicer requests

Co-authored-by: wemoveon2 <86081107+wemoveon2@users.noreply.github.com>

* Fix: Use append() for additional_fields repeated field in GetModelRequest

Co-authored-by: wemoveon2 <86081107+wemoveon2@users.noreply.github.com>

* Remove test file per reviewer feedback

Co-authored-by: wemoveon2 <86081107+wemoveon2@users.noreply.github.com>

---------

Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: wemoveon2 <86081107+wemoveon2@users.noreply.github.com>

Author: Copilot

clarifai/runners/models/model_builder.py

@@ -225,6 +225,8 @@ def get_model_proto(self) -> resources_pb2.Model:
             user_app_id=self.client.user_app_id,
             model_id=self.model_id,
         )
+        # Add secrets to additional_fields to get request-type secrets
+        request.additional_fields.append("secrets")
         if self.model_version_id is not None:
             request.version_id = self.model_version_id
         resp: service_pb2.SingleModelResponse = self.client.STUB.GetModel(request)

clarifai/runners/models/model_servicer.py

@@ -18,13 +18,15 @@ class ModelServicer(service_pb2_grpc.V2Servicer):
     This is the servicer that will handle the gRPC requests from either the dev server or runner loop.
     """
 
-    def __init__(self, model):
+    def __init__(self, model, model_proto=None):
         """
         Args:
             model: The class that will handle the model logic. Must implement predict(),
         generate(), stream().
+            model_proto: The model proto containing model configuration including secrets.
         """
         self.model = model
+        self.model_proto = model_proto
 
         # Try to create auth helper from environment variables if available
         self._auth_helper = None
@@ -55,6 +57,10 @@ def PostModelOutputs(
         returns an output.
         """
 
+        # Inject model proto if available and not already in request
+        if self.model_proto is not None and not request.HasField("model"):
+            request.model.CopyFrom(self.model_proto)
+
         # Download any urls that are not already bytes.
         ensure_urls_downloaded(request, auth_helper=self._auth_helper)
         inject_secrets(request)
@@ -80,6 +86,10 @@ def GenerateModelOutputs(
         This is the method that will be called when the servicer is run. It takes in an input and
         returns an output.
         """
+        # Inject model proto if available and not already in request
+        if self.model_proto is not None and not request.HasField("model"):
+            request.model.CopyFrom(self.model_proto)
+
         # Download any urls that are not already bytes.
         ensure_urls_downloaded(request, auth_helper=self._auth_helper)
         inject_secrets(request)
@@ -108,8 +118,11 @@ def StreamModelOutputs(
         # Duplicate the iterator
         request, request_copy = tee(request)
 
-        # Download any urls that are not already bytes.
+        # Download any urls that are not already bytes and inject model proto
         for req in request:
+            # Inject model proto if available and not already in request
+            if self.model_proto is not None and not req.HasField("model"):
+                req.model.CopyFrom(self.model_proto)
             ensure_urls_downloaded(req, auth_helper=self._auth_helper)
             inject_secrets(req)
 

clarifai/runners/server.py

@@ -232,7 +232,9 @@ def serve(
 
     def start_servicer(self, port, pool_size, max_queue_size, max_msg_length, enable_tls):
         # initialize the servicer with the runner so that it gets the predict(), generate(), stream() classes.
-        self._servicer = ModelServicer(self._current_model)
+        self._servicer = ModelServicer(
+            self._current_model, model_proto=self._builder.get_model_proto()
+        )
 
         server = GRPCServer(
             futures.ThreadPoolExecutor(