mirror of
https://github.com/pytorch/pytorch.git
synced 2025-10-21 05:34:18 +08:00
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/25499 See #23110 for model parallel design details, and #26759 for the RRef protocol. This commit add support for using RRef as Python UDF arguments and return value. RRefs can now be shared from owner to user, from user to owner, or from user to user. Limitations: 1. No implicit type conversion yet. (#27099) 2. No failure handling and retry. (#26116) 3. UDF is not yet blocked until all RRefs are confirmed. (#27098) 4. Internal RRef control messages are not idempotent yet. (#26116) 5. Cannot delete RRefs correctly when there are circular dependencies. (#27096) Main changes: 1. Added `SCRIPT_REMOTE_CALL` and `PYTHON_REMOTE_CALL` to `Message.h` to represent `dist.remote` invocations. 2. Added `SCRIPT_RREF_FETCH_CALL`, `PYTHON_RREF_FETCH_CALL`, `RREF_USER_ACCEPT`, `RREF_USER_DELETE`, `RREF_CHILD_ACCEPT`, and `RREF_FORK_REQUEST` to `Message.h` as internal RRef control messages. 3. New message request handling code is added to `functions.cpp`, and message format is added in `script_remote_call.h`, `python_remote_call.h`, and `rref_proto.h`. 4. Added a `PyRRef` type in `py_rref.h` and `py_rref.cpp` which holds a shared pointer to C++ `RRef` type. `PyRRef` wraps the C++ API and also implements RRef pickling and unpickling. RRef fork related control messages will be sent during RRef pickling/unpickling procedure. 5. Update `RRef.h` and `RRef.cpp` accordingly to support `py::object` RRefs. 6. RRef context (reference count, etc.) are tracked in `rref_context.h` and `rref_context.cpp`. Test Plan: Imported from OSS buck test mode/dev-nosan //caffe2/test:rpc_fork Differential Revision: D17184146 Pulled By: mrshenli fbshipit-source-id: a3a268efc087ac1ef489136ab957080382629265
54 lines
1.9 KiB
C++
54 lines
1.9 KiB
C++
#pragma once
|
|
|
|
#include <torch/csrc/distributed/rpc/message.h>
|
|
#include <torch/csrc/distributed/rpc/types.h>
|
|
#include <torch/csrc/utils/pybind.h>
|
|
|
|
namespace torch {
|
|
namespace distributed {
|
|
namespace rpc {
|
|
|
|
// Singleton class provides interface to execute python UDF remote call
|
|
// and deserialize the returned results by running python function
|
|
// in internal_rpc_utilities.
|
|
// The singleton object is constructed at first when RPC agent is
|
|
// constructed, where the python function in
|
|
// torch/distributed/internal_rpc_utils.py are imported only once.
|
|
class PYBIND11_EXPORT PythonRpcHandler {
|
|
public:
|
|
static PythonRpcHandler& getInstance();
|
|
// Execute python UDF, result is pickled to binary string
|
|
std::vector<char> generatePythonUDFResult(
|
|
const std::vector<char>& pickledPayload,
|
|
const std::vector<torch::Tensor>& requestTensorTable,
|
|
std::vector<torch::Tensor>& responseTensorTable);
|
|
// Returned python UDF result is pickled binary string, so run python
|
|
// function to unpickle the python UDF result and return py::object to user
|
|
py::object loadPythonUDFResult(
|
|
const std::vector<char>& pickledPayload,
|
|
const std::vector<torch::Tensor>& tensorTable);
|
|
// Run a pickled Python UDF and return the result py::object
|
|
py::object runPythonUDF(const SerializedPyObj& serializedObj);
|
|
// Serialized a py::object into a string
|
|
SerializedPyObj serialize(const py::object& obj);
|
|
// Deserialize a string into a py::object
|
|
py::object deserialize(const SerializedPyObj& serializedObj);
|
|
|
|
private:
|
|
PythonRpcHandler();
|
|
~PythonRpcHandler() = default;
|
|
|
|
PythonRpcHandler(const PythonRpcHandler&) = delete;
|
|
PythonRpcHandler& operator=(const PythonRpcHandler&) = delete;
|
|
PythonRpcHandler(PythonRpcHandler&&) = delete;
|
|
PythonRpcHandler& operator=(PythonRpcHandler&&) = delete;
|
|
|
|
py::object runUDFFunction_;
|
|
py::object loadResultFunction_;
|
|
py::object serializeFunction_;
|
|
};
|
|
|
|
} // namespace rpc
|
|
} // namespace distributed
|
|
} // namespace torch
|