File: request_callback.h

package info (click to toggle)
pytorch 1.7.1-7
  • links: PTS, VCS
  • area: main
  • in suites: bullseye
  • size: 80,340 kB
  • sloc: cpp: 670,830; python: 343,991; ansic: 67,845; asm: 5,503; sh: 2,924; java: 2,888; xml: 266; makefile: 244; ruby: 148; yacc: 144; objc: 51; lex: 44
file content (33 lines) | stat: -rw-r--r-- 1,183 bytes parent folder | download
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
#pragma once

#include <torch/csrc/distributed/rpc/message.h>

namespace torch {
namespace distributed {
namespace rpc {

// Functor which is invoked to process an RPC message. This is an abstract class
// with some common functionality across all request handlers. Users need to
// implement this interface to perform the actual business logic.
class TORCH_API RequestCallback {
 public:
  // Invoke the callback.
  std::shared_ptr<FutureMessage> operator()(Message& request) const;

  virtual ~RequestCallback() {}

 protected:
  // RpcAgent implementation should invoke ``RequestCallback`` to process
  // received requests. There is no restriction on the implementation's
  // threading model. This function takes an rvalue reference of the Message
  // object. It is expected to return the future to a response message or
  // message containing an exception. Different rpc agent implementations are
  // expected to ensure delivery of the response/exception based on their
  // implementation specific mechanisms.
  virtual std::shared_ptr<FutureMessage> processMessage(
      Message& request) const = 0;
};

} // namespace rpc
} // namespace distributed
} // namespace torch