Implementation:Tensorflow Serving Json Tensor Test

Knowledge Sources	Tensorflow_Serving
Domains	Testing, JSON, REST API
Last Updated	2026-02-13 00:00 GMT

Overview

Test suite validating JSON tensor conversion utilities for the REST API, covering both request parsing and response serialization.

Description

This test file provides extensive coverage of the JSON tensor conversion functions in tensorflow_serving/util/json_tensor.h. It tests FillPredictRequestFromJson for parsing JSON predict requests into TensorFlow protobuf format, and various FromJson* functions for reverse conversion. The tests use RapidJSON for JSON parsing and protobuf text format for expected value comparisons.

Key areas tested include:

• Single unnamed tensors (row and columnar format)
• Multiple named tensors
• Base64 encoded bytes (scalars and lists)
• Mixed integer/float inputs for float and double tensor types
• Precision loss and overflow handling for float tensors
• Non-null-terminated buffer handling
• Deeply nested JSON (well-formed and malformed)
• Signature name support
• Comprehensive error cases for both row and columnar formats
• JSON-to-tensor conversion for single and multiple tensors
• Zero-batch tensor handling
• Classification and regression result serialization to JSON
• Non-finite float values (NaN, Infinity)

Usage

Run these tests to validate changes to the REST API JSON parsing/serialization, tensor format conversion, or error handling for malformed JSON inputs.

Code Reference

Source Location

• Repository: Tensorflow_Serving
• File: tensorflow_serving/util/json_tensor_test.cc
• Lines: 1-1619

Test Fixture

using TensorInfoMap = ::google::protobuf::Map<string, TensorInfo>;
using TensorMap = ::google::protobuf::Map<string, TensorProto>;

std::function<absl::Status(const string&, TensorInfoMap*)> getmap(
    const TensorInfoMap& map) {
  return [&map](const string&, TensorInfoMap* m) {
    *m = map;
    return absl::OkStatus();
  };
}

// Tests use getmap() to provide tensor type info for parsing
TEST(JsontensorTest, SingleUnnamedTensor) {
  TensorInfoMap infomap;
  ASSERT_TRUE(
      TextFormat::ParseFromString("dtype: DT_INT32", &infomap["default"]));
  PredictRequest req;
  JsonPredictRequestFormat format;
  TF_EXPECT_OK(FillPredictRequestFromJson(..., getmap(infomap), &req, &format));
}

Build Target

bazel test //tensorflow_serving/util:json_tensor_test

Test Coverage

Key Test Cases

Test Name	Category	Description
SingleUnnamedTensor	Parsing	Tests parsing single unnamed tensor in row format
DeeplyNestedWellFormed	Parsing	Tests parsing deeply nested but valid JSON
DeeplyNestedMalformed	Error Handling	Tests rejection of deeply nested malformed JSON
MixedInputForFloatTensor	Type Handling	Tests mixed int/float input for float tensors
MixedInputForDoubleTensor	Type Handling	Tests mixed int/float input for double tensors
FloatTensorWithPrecisionLoss	Precision	Tests float precision loss handling
FloatTensorThatExceedsMaxReturnsInf	Overflow	Tests float overflow returns infinity
FloatTensorThatExceedsMinReturnsZero	Underflow	Tests float underflow returns zero
SingleUnnamedTensorWithSignature	Parsing	Tests parsing with signature name
TensorFromNonNullTerminatedBuffer	Edge Case	Tests parsing from non-null-terminated buffer
SingleUnnamedTensorBase64Scalars	Base64	Tests base64 scalar byte value parsing
SingleUnnamedTensorBase64Lists	Base64	Tests base64 list byte value parsing
SingleNamedTensorBase64	Base64	Tests base64 named tensor parsing
MultipleNamedTensor	Parsing	Tests parsing multiple named tensors
SingleUnnamedTensorColumnarFormat	Format	Tests columnar (inputs) format parsing
MultipleNamedTensorColumnarFormat	Format	Tests columnar format with multiple named tensors
SingleUnnamedTensorErrors	Error Handling	Tests various error cases for row format
MultipleNamedTensorErrors	Error Handling	Tests various error cases for named tensors
FromJsonSingleTensor	Serialization	Tests single tensor JSON serialization
FromJsonSingleScalarTensor	Serialization	Tests scalar tensor JSON serialization
FromJsonSingleBytesTensor	Serialization	Tests bytes tensor base64 serialization
FromJsonSingleFloatTensorSixDigitPrecision	Precision	Tests float serialization precision
FromJsonSingleFloatTensorNonFinite	Serialization	Tests NaN/Infinity serialization
FromJsonMultipleNamedTensors	Serialization	Tests multiple tensor JSON serialization
FromJsonSingleZeroBatchTensor	Edge Case	Tests zero-batch tensor handling
FromJsonMultipleZeroBatchTensors	Edge Case	Tests multiple zero-batch tensors
JsonFromClassificationResult	Serialization	Tests classification result to JSON
JsonFromRegressionResult	Serialization	Tests regression result to JSON
JsonFromRegressionResultWithNonFinite	Serialization	Tests regression with NaN/Infinity
JsonFromResultErrors	Error Handling	Tests error cases in result serialization
MakeJsonFromTensors.StatusOK	Serialization	Tests successful tensor to JSON conversion
MakeJsonFromTensors.StatusError	Error Handling	Tests error in tensor to JSON conversion

Usage Examples

Test Pattern

TEST(JsontensorTest, SingleUnnamedTensor) {
  TensorInfoMap infomap;
  ASSERT_TRUE(
      TextFormat::ParseFromString("dtype: DT_INT32", &infomap["default"]));

  PredictRequest req;
  JsonPredictRequestFormat format;
  TF_EXPECT_OK(FillPredictRequestFromJson(R"(
    {
      "instances": [[1,2],[3,4],[5,6]]
    })",
    getmap(infomap), &req, &format));
  auto tmap = req.inputs();
  EXPECT_EQ(tmap.size(), 1);
  EXPECT_EQ(format, JsonPredictRequestFormat::kRow);
}