Implementation:Tensorflow Serving Predict Util Test

Knowledge Sources	Tensorflow_Serving
Domains	Testing, Prediction
Last Updated	2026-02-13 00:00 GMT

Overview

Test suite validating the predict utility functions including RunPredict which executes prediction directly on a SavedModelBundle.

Description

This test file validates the lower-level predict utility functions that operate directly on SavedModelBundle instances rather than going through ServerCore. The PredictImplTest fixture creates multiple ServerCore instances (including one with a "bad" model for error testing) and obtains ServableHandle<SavedModelBundle> to call RunPredict directly. A FakeSession is also provided for isolated unit testing that copies input tensors to output.

Key areas tested include:

• Model spec validation (missing, empty, not found)
• Input tensor validation (empty, mismatched, wrong type, invalid tensor)
• Output filter validation
• Model with missing signatures
• Successful prediction with various signature types
• Customized signature handling
• Thread pool options propagation
• Method name checking feature

Usage

Run these tests to validate changes to the RunPredict utility function and related prediction helpers.

Code Reference

Source Location

• Repository: Tensorflow_Serving
• File: tensorflow_serving/servables/tensorflow/predict_util_test.cc
• Lines: 1-614

Test Fixture

class FakeSession : public tensorflow::Session {
 public:
  // Copies input tensors to output for testing
  absl::Status Run(const RunOptions& run_options,
                   const std::vector<std::pair<string, Tensor>>& inputs,
                   const std::vector<string>& output_names,
                   const std::vector<string>& target_nodes,
                   std::vector<Tensor>* outputs,
                   RunMetadata* run_metadata,
                   const thread::ThreadPoolOptions& thread_pool_options) override;
};

class PredictImplTest : public ::testing::Test {
 protected:
  absl::Status CallPredict(ServerCore* server_core,
                           const PredictRequest& request,
                           PredictResponse* response,
                           const thread::ThreadPoolOptions& opts = {});
};

Build Target

bazel test //tensorflow_serving/servables/tensorflow:predict_util_test

Test Coverage

Key Test Cases

Test Name	Category	Description
MissingOrEmptyModelSpec	Validation	Tests error on missing or empty model spec
EmptyInputList	Validation	Tests error on empty input list
InputTensorsDontMatchModelSpecInputs	Validation	Tests error on input tensor mismatch
PredictionInvalidTensor	Validation	Tests error on invalid tensor data
OutputFiltersDontMatchModelSpecOutputs	Validation	Tests error on invalid output filters
InputTensorsHaveWrongType	Validation	Tests error on wrong tensor dtype
ModelMissingSignatures	Error Handling	Tests error when model has no signatures
PredictionSuccess	Integration	Tests successful prediction execution
PredictionWithNamedRegressionSignature	Integration	Tests prediction with named regression signature
PredictionWithNamedClassificationSignature	Integration	Tests prediction with named classification signature
PredictionWithCustomizedSignatures	Integration	Tests prediction with custom signatures
ThreadPoolOptions	Integration	Tests thread pool options are passed through
MethodNameCheck	Feature Toggle	Tests method name checking feature

Usage Examples

Test Pattern

TEST_F(PredictImplTest, PredictionSuccess) {
  PredictRequest request;
  PredictResponse response;

  ModelSpec* model_spec = request.mutable_model_spec();
  model_spec->set_name(kTestModelName);
  model_spec->mutable_version()->set_value(kTestModelVersion);

  TensorProto tensor_proto;
  tensor_proto.add_float_val(2.0);
  tensor_proto.set_dtype(tensorflow::DT_FLOAT);
  (*request.mutable_inputs())[kInputTensorKey] = tensor_proto;

  TF_EXPECT_OK(CallPredict(GetServerCore(), request, &response));
  // Validates model spec and output tensor in response
}

Related Pages

• Implementation:Tensorflow_Serving_Predict_Impl_Test