Implementation:Tensorflow Serving EvHTTP Request

Knowledge Sources	Tensorflow_Serving
Domains	HTTP, Networking
Last Updated	2026-02-13 00:00 GMT

Overview

A libevent-based implementation of the ServerRequestInterface that wraps evhttp_request to provide HTTP request parsing, response writing, gzip decompression, and reply dispatching.

Description

EvHTTPRequest implements the ServerRequestInterface and wraps a ParsedEvRequest which contains the decoded libevent request. ParsedEvRequest parses the incoming evhttp_request to extract the HTTP method (GET, POST, PUT, DELETE, HEAD, OPTIONS, TRACE, CONNECT, PATCH), URI, path, query, fragment, and headers. EvHTTPRequest provides: ReadRequestBytes() to read the request body (with automatic gzip decompression when Content-Encoding indicates gzip and the handler has auto_uncompress_input enabled), WriteResponseBytes()/WriteResponseString() to append data to the output buffer, GetRequestHeader() and request_headers() for header access, OverwriteResponseHeader()/AppendResponseHeader() for setting response headers, and ReplyWithStatus()/Reply()/Abort() for completing the response. Replies are scheduled on the libevent event loop via EventLoopSchedule to ensure thread safety with libevent's single-threaded model. The request object deletes itself after sending the reply via EvSendReply.

Usage

This is the internal request implementation used by EvHTTPServer. Application code interacts with it through the ServerRequestInterface abstraction when handling HTTP requests in the serving system.

Code Reference

Source Location

• Repository: Tensorflow_Serving
• File: tensorflow_serving/util/net_http/server/internal/evhttp_request.cc
• Lines: 1-389

Signature

struct ParsedEvRequest {
  explicit ParsedEvRequest(evhttp_request* request_in);
  ~ParsedEvRequest();
  bool decode();

  evhttp_request* request;
  evhttp_uri* decoded_uri;
  std::string method;
  const char* uri;
  const char* path;
  std::string path_and_query;
  evkeyvalq* headers;
};

class EvHTTPRequest : public ServerRequestInterface {
 public:
  EvHTTPRequest(std::unique_ptr<ParsedEvRequest> request, ServerSupport* server);
  bool Initialize();
  absl::string_view uri_path() const override;
  absl::string_view http_method() const override;
  void WriteResponseBytes(const char* data, int64_t size) override;
  void WriteResponseString(absl::string_view data) override;
  std::unique_ptr<char[], BlockDeleter> ReadRequestBytes(int64_t* size) override;
  void ReplyWithStatus(HTTPStatusCode status) override;
  void Reply() override;
  void Abort() override;
};

Import

#include "tensorflow_serving/util/net_http/server/internal/evhttp_request.h"

I/O Contract

Inputs

Name	Type	Required	Description
request	evhttp_request*	Yes	The raw libevent HTTP request
server	ServerSupport*	Yes	The server support interface for event loop scheduling

Outputs

Name	Type	Description
uri_path()	absl::string_view	The request path including query and fragment
http_method()	absl::string_view	The HTTP method string (e.g., "GET", "POST")
ReadRequestBytes()	std::unique_ptr<char[], BlockDeleter>	The request body bytes, optionally gzip-decompressed

Usage Examples

Handling a Request (Application Code)

// This is how application code sees the request through the interface
void MyHandler(ServerRequestInterface* request) {
  int64_t size;
  auto body = request->ReadRequestBytes(&size);
  if (body != nullptr) {
    ProcessBody(body.get(), size);
  }
  request->WriteResponseString("OK");
  request->Reply();
}

Related Pages

• Principle:Tensorflow_Serving_HTTP_Server_Implementation