Implementation:Triton inference server Server CommonUtils

Knowledge Sources	Triton Inference Server
Domains	Utilities, Error_Handling
Last Updated	2026-02-13 17:00 GMT

Overview

Foundational utility library providing shared constants, error-handling macros, and helper functions used across all Triton server components.

Description

common.h and common.cc form the foundational utility layer of the Triton server. The header defines HTTP header constants, server limits (max gRPC message size, JSON nesting depth, default max input size), and the critical error-handling macro family (RETURN_IF_ERR, FAIL_IF_ERR, THROW_IF_ERR, IGNORE_ERR). The implementation provides model version string parsing, environment variable retrieval, tensor shape formatting, element count calculation with overflow protection, Base64 decoding using libb64, and shared memory key validation.

Usage

Included by nearly every server source file. The macros and utility functions provide the common vocabulary for error handling and data manipulation throughout the codebase.

Code Reference

Source Location

• Repository: Triton Inference Server
• File: src/common.h
• Lines: 1-286
• File: src/common.cc
• Lines: 1-201

Signature

namespace triton { namespace server {

// Error-handling macros
#define RETURN_IF_ERR(X)     /* returns TRITONSERVER_Error* if X fails */
#define FAIL_IF_ERR(X, MSG)  /* logs and exits if X fails */
#define THROW_IF_ERR(X)      /* throws exception if X fails */
#define IGNORE_ERR(X)        /* deletes error without action */

// Utility functions
TRITONSERVER_Error* GetModelVersionFromString(
    const std::string& version_string, int64_t* version);

const char* GetEnvironmentVariableOrDefault(
    const char* variable_name, const char* default_value);

std::string ShapeToString(
    const int64_t* dims, size_t dim_count);

int64_t GetElementCount(
    const int64_t* dims, size_t dim_count);

TRITONSERVER_Error* Base64Decode(
    const char* encoded, size_t encoded_len,
    std::vector<char>* decoded);

TRITONSERVER_Error* ValidateSharedMemoryKey(
    const std::string& shm_key);

// Constants
constexpr int64_t kMaxGRPCMessageSize = INT32_MAX;
constexpr int kMaxJsonNestingDepth = 64;

}} // namespace triton::server

Import

#include "common.h"

I/O Contract

Inputs

Name	Type	Required	Description
version_string	string	Varies	Model version as string (e.g., "1", "latest")
dims	int64_t*	Varies	Tensor shape dimensions
encoded	const char*	Varies	Base64-encoded data

Outputs

Name	Type	Description
version	int64_t	Parsed model version number
shape_string	string	Human-readable shape (e.g., "[3,224,224]")
element_count	int64_t	Total elements in tensor (-1 on overflow)
decoded	vector<char>	Base64-decoded bytes

Usage Examples

Error Handling with Macros

#include "common.h"

TRITONSERVER_Error* ProcessRequest() {
  // Return early if any step fails
  RETURN_IF_ERR(ValidateInput());
  RETURN_IF_ERR(RunInference());
  RETURN_IF_ERR(FormatOutput());
  return nullptr;  // success
}

void SafeCleanup() {
  // Log error but don't propagate
  FAIL_IF_ERR(
      TRITONSERVER_ServerDelete(server),
      "Failed to delete server");
}

Related Pages

• Environment:Triton_inference_server_Server_GPU_CUDA_Runtime