Implementation:Tensorflow Serving Get Model Metadata Impl

Knowledge Sources	Tensorflow_Serving
Domains	Model Serving, Model Metadata
Last Updated	2026-02-13 00:00 GMT

Overview

Implements the GetModelMetadata API for retrieving signature definitions from loaded SavedModel bundles via ServerCore.

Description

GetModelMetadataImpl is a static utility class that provides the implementation for the GetModelMetadata RPC. It supports retrieving signature_def metadata from a loaded SavedModelBundle.

The implementation consists of:

• GetModelMetadata: Entry point that validates the presence of a ModelSpec in the request, then delegates to GetModelMetadataWithModelSpec.
• GetModelMetadataWithModelSpec: Validates that all requested metadata_fields are supported (only "signature_def" is supported), then iterates through requested fields. For the "signature_def" field, it calls SavedModelGetSignatureDef which obtains a ServableHandle to the SavedModelBundle from ServerCore, copies all signature definitions from the MetaGraphDef into a SignatureDefMap protobuf, sets the response's model_spec with the servable's name and version, and packs the SignatureDefMap into the response's metadata Any field.

The constant kSignatureDef ("signature_def") defines the only supported metadata field type.

Usage

Use this module for handling GetModelMetadata requests in the standard TensorFlow serving pipeline (non-TFRT). It is typically called by the PredictionServiceImpl gRPC handler. For TFRT models, metadata retrieval is handled directly by TfrtSavedModelServable::GetModelMetadata.

Code Reference

Source Location

• Repository: Tensorflow_Serving
• Files:
  • tensorflow_serving/servables/tensorflow/get_model_metadata_impl.h (lines 1-45)
  • tensorflow_serving/servables/tensorflow/get_model_metadata_impl.cc (lines 1-103)

Signature

class GetModelMetadataImpl {
 public:
  static constexpr const char kSignatureDef[] = "signature_def";

  static Status GetModelMetadata(ServerCore* core,
                                 const GetModelMetadataRequest& request,
                                 GetModelMetadataResponse* response);

  static Status GetModelMetadataWithModelSpec(
      ServerCore* core, const ModelSpec& model_spec,
      const GetModelMetadataRequest& request,
      GetModelMetadataResponse* response);
};

Import

#include "tensorflow_serving/servables/tensorflow/get_model_metadata_impl.h"

I/O Contract

Inputs

Name	Type	Required	Description
core	ServerCore*	Yes	ServerCore for accessing loaded servables
request	GetModelMetadataRequest	Yes	Request specifying model_spec and metadata_field(s)

Outputs

Name	Type	Description
response	GetModelMetadataResponse*	Contains model_spec and metadata map with SignatureDefMap packed as Any
return	Status	OK on success; InvalidArgument if model_spec is missing or unsupported metadata field is requested

Usage Examples

Retrieving Model Metadata

GetModelMetadataRequest request;
request.mutable_model_spec()->set_name("my_model");
request.add_metadata_field("signature_def");

GetModelMetadataResponse response;
Status status = GetModelMetadataImpl::GetModelMetadata(
    server_core, request, &response);

Related Pages

• Principle:Tensorflow_Serving_Model_Metadata_Query