Implementation:Triton inference server Server DataCompressor

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

Overview

Concrete tool for compressing and decompressing HTTP request/response data using gzip and deflate algorithms via zlib.

Description

The DataCompressor class provides static methods for compressing and decompressing data that flows through Triton's HTTP endpoints. It operates directly on libevent evbuffer objects, supporting GZIP, DEFLATE, and IDENTITY (no-op) compression types. The implementation manages streaming compression with configurable maximum decompressed size limits to prevent memory exhaustion from malicious payloads.

Usage

Used internally by the HTTP server to compress responses and decompress incoming requests based on Content-Encoding and Accept-Encoding headers. Not directly imported by external users.

Code Reference

Source Location

• Repository: Triton Inference Server
• File: src/data_compressor.h
• Lines: 1-382

Signature

class DataCompressor {
 public:
  enum Type { IDENTITY, DEFLATE, GZIP };

  static TRITONSERVER_Error* CompressData(
      Type type, evbuffer* source, evbuffer* compressed);

  static TRITONSERVER_Error* DecompressData(
      Type type, evbuffer* source, evbuffer* decompressed,
      size_t max_decompressed_size = 0);

  static Type GetType(const std::string& encoding);
  static const char* GetName(Type type);
};

Import

#include "data_compressor.h"

I/O Contract

Inputs

Name	Type	Required	Description
type	DataCompressor::Type	Yes	Compression algorithm (IDENTITY, DEFLATE, GZIP)
source	evbuffer*	Yes	Input data buffer from libevent
max_decompressed_size	size_t	No	Maximum allowed decompressed size (0 = unlimited)

Outputs

Name	Type	Description
compressed/decompressed	evbuffer*	Output buffer containing processed data
error	TRITONSERVER_Error*	nullptr on success, error object on failure

Usage Examples

Compressing HTTP Response

#include "data_compressor.h"

// Compress response data before sending
evbuffer* raw_response = evbuffer_new();
evbuffer* compressed_response = evbuffer_new();

// Add data to raw_response...

TRITONSERVER_Error* err = DataCompressor::CompressData(
    DataCompressor::GZIP, raw_response, compressed_response);
if (err == nullptr) {
  // Send compressed_response with Content-Encoding: gzip
}

Related Pages

• Environment:Triton_inference_server_Server_GPU_CUDA_Runtime