Distributed RPC Framework

The distributed RPC framework provides mechanisms for multi-machine model training through a set of primitives to allow for remote communication, and a higher-level API to automatically differentiate models split across several machines.

Warning

APIs in the RPC package are stable. There are multiple ongoing work items to improve performance and error handling, which will ship in future releases.

Note

Please refer to PyTorch Distributed Overview for a brief introduction to all features related to distributed training.

Basics

The distributed RPC framework makes it easy to run functions remotely, supports referencing remote objects without copying the real data around, and provides autograd and optimizer APIs to transparently run backward and update parameters across RPC boundaries. These features can be categorized into four sets of APIs.

1. Remote Procedure Call (RPC) supports running a function on the specified destination worker with the given arguments and getting the return value back or creating a reference to the return value. There are three main RPC APIs: rpc_sync() (synchronous), rpc_async() (asynchronous), and remote() (asynchronous and returns a reference to the remote return value). Use the synchronous API if the user code cannot proceed without the return value. Otherwise, use the asynchronous API to get a future, and wait on the future when the return value is needed on the caller. The remote() API is useful when the requirement is to create something remotely but never need to fetch it to the caller. Imagine the case that a driver process is setting up a parameter server and a trainer. The driver can create an embedding table on the parameter server and then share the reference to the embedding table with the trainer, but itself will never use the embedding table locally. In this case, rpc_sync() and rpc_async() are no longer appropriate, as they always imply that the return value will be returned to the caller immediately or in the future.

2. Remote Reference (RRef) serves as a distributed shared pointer to a local or remote object. It can be shared with other workers and reference counting will be handled transparently. Each RRef only has one owner and the object only lives on that owner. Non-owner workers holding RRefs can get copies of the object from the owner by explicitly requesting it. This is useful when a worker needs to access some data object, but itself is neither the creator (the caller of remote()) or the owner of the object. The distributed optimizer, as we will discuss below, is one example of such use cases.

3. Distributed Autograd stitches together local autograd engines on all the workers involved in the forward pass, and automatically reach out to them during the backward pass to compute gradients. This is especially helpful if the forward pass needs to span multiple machines when conducting, e.g., distributed model parallel training, parameter-server training, etc. With this feature, user code no longer needs to worry about how to send gradients across RPC boundaries and in which order should the local autograd engines be launched, which can become quite complicated where there are nested and inter-dependent RPC calls in the forward pass.

4. Distributed Optimizer’s constructor takes a Optimizer() (e.g., SGD(), Adagrad(), etc.) and a list of parameter RRefs, creates an Optimizer() instance on each distinct RRef owner, and updates parameters accordingly when running step(). When you have distributed forward and backward passes, parameters and gradients will be scattered across multiple workers, and hence it requires an optimizer on each of the involved workers. Distributed Optimizer wraps all those local optimizers into one, and provides a concise constructor and step() API.

RPC

Before using RPC and distributed autograd primitives, initialization must take place. To initialize the RPC framework we need to use init_rpc() which would initialize the RPC framework, RRef framework and distributed autograd.

The following APIs allow users to remotely execute functions as well as create references (RRefs) to remote data objects. In these APIs, when passing a Tensor as an argument or a return value, the destination worker will try to create a Tensor with the same meta (i.e., shape, stride, etc.). We intentionally disallow transmitting CUDA tensors because it might crash if the device lists on source and destination workers do not match. In such cases, applications can always explicitly move the input tensors to CPU on the caller and move it to the desired devices on the callee if necessary.

Warning

TorchScript support in RPC is a prototype feature and subject to change. Since v1.5.0, torch.distributed.rpc supports calling TorchScript functions as RPC target functions, and this will help improve parallelism on the callee side as executing TorchScript functions does not require GIL.

The RPC package also provides decorators which allow applications to specify how a given function should be treated on the callee side.

torch.distributed.rpc.functions.async_execution(fn)

A decorator for a function indicating that the return value of the function is guaranteed to be a Future object and this function can run asynchronously on the RPC callee. More specifically, the callee extracts the Future returned by the wrapped function and installs subsequent processing steps as a callback to that Future. The installed callback will read the value from the Future when completed and send the value back as the RPC response. That also means the returned Future only exists on the callee side and is never sent through RPC. This decorator is useful when the wrapped function’s (fn) execution needs to pause and resume due to, e.g., containing rpc_async() or waiting for other signals.

Note

To enable asynchronous execution, applications must pass the function object returned by this decorator to RPC APIs. If RPC detected attributes installed by this decorator, it knows that this function returns a Future object and will handle that accordingly. However, this does not mean this decorator has to be outmost one when defining a function. For example, when combined with @staticmethod or @classmethod, @rpc.functions.async_execution needs to be the inner decorator to allow the target function be recognized as a static or class function. This target function can still execute asynchronously because, when accessed, the static or class method preserves attributes installed by @rpc.functions.async_execution.

Example::

The returned Future object can come from rpc_async(), then(), or Future constructor. The example below shows directly using the Future returned by then().

>>> from torch.distributed import rpc
>>>
>>> # omitting setup and shutdown RPC
>>>
>>> # On all workers
>>> @rpc.functions.async_execution
>>> def async_add_chained(to, x, y, z):
>>>     # This function runs on "worker1" and returns immediately when
>>>     # the callback is installed through the `then(cb)` API. In the
>>>     # mean time, the `rpc_async` to "worker2" can run concurrently.
>>>     # When the return value of that `rpc_async` arrives at
>>>     # "worker1", "worker1" will run the lambda function accordingly
>>>     # and set the value for the previously returned `Future`, which
>>>     # will then trigger RPC to send the result back to "worker0".
>>>     return rpc.rpc_async(to, torch.add, args=(x, y)).then(
>>>         lambda fut: fut.wait() + z
>>>     )
>>>
>>> # On worker0
>>> ret = rpc.rpc_sync(
>>>     "worker1",
>>>     async_add_chained,
>>>     args=("worker2", torch.ones(2), 1, 1)
>>> )
>>> print(ret)  # prints tensor([3., 3.])

When combined with TorchScript decorators, this decorator must be the outmost one.

>>> from torch import Tensor
>>> from torch.futures import Future
>>> from torch.distributed import rpc
>>>
>>> # omitting setup and shutdown RPC
>>>
>>> # On all workers
>>> @torch.jit.script
>>> def script_add(x: Tensor, y: Tensor) -> Tensor:
>>>     return x + y
>>>
>>> @rpc.functions.async_execution
>>> @torch.jit.script
>>> def async_add(to: str, x: Tensor, y: Tensor) -> Future[Tensor]:
>>>     return rpc.rpc_async(to, script_add, (x, y))
>>>
>>> # On worker0
>>> ret = rpc.rpc_sync(
>>>     "worker1",
>>>     async_add,
>>>     args=("worker2", torch.ones(2), 1)
>>> )
>>> print(ret)  # prints tensor([2., 2.])

When combined with static or class method, this decorator must be the inner one.

>>> from torch.distributed import rpc
>>>
>>> # omitting setup and shutdown RPC
>>>
>>> # On all workers
>>> class AsyncExecutionClass:
>>>
>>>     @staticmethod
>>>     @rpc.functions.async_execution
>>>     def static_async_add(to, x, y, z):
>>>         return rpc.rpc_async(to, torch.add, args=(x, y)).then(
>>>             lambda fut: fut.wait() + z
>>>         )
>>>
>>>     @classmethod
>>>     @rpc.functions.async_execution
>>>     def class_async_add(cls, to, x, y, z):
>>>         ret_fut = torch.futures.Future()
>>>         rpc.rpc_async(to, torch.add, args=(x, y)).then(
>>>             lambda fut: ret_fut.set_result(fut.wait() + z)
>>>         )
>>>         return ret_fut
>>>
>>>     @rpc.functions.async_execution
>>>     def bound_async_add(self, to, x, y, z):
>>>         return rpc.rpc_async(to, torch.add, args=(x, y)).then(
>>>             lambda fut: fut.wait() + z
>>>         )
>>>
>>> # On worker0
>>> ret = rpc.rpc_sync(
>>>     "worker1",
>>>     AsyncExecutionClass.static_async_add,
>>>     args=("worker2", torch.ones(2), 1, 2)
>>> )
>>> print(ret)  # prints tensor([4., 4.])
>>>
>>> ret = rpc.rpc_sync(
>>>     "worker1",
>>>     AsyncExecutionClass.class_async_add,
>>>     args=("worker2", torch.ones(2), 1, 2)
>>> )
>>> print(ret)  # prints tensor([4., 4.])

This decorator also works with RRef helpers, i.e., . torch.distributed.rpc.RRef.rpc_sync(), torch.distributed.rpc.RRef.rpc_async(), and torch.distributed.rpc.RRef.remote().

>>> from torch.distributed import rpc
>>>
>>> # reuse the AsyncExecutionClass class above
>>> rref = rpc.remote("worker1", AsyncExecutionClass)
>>> ret = rref.rpc_sync().static_async_add("worker2", torch.ones(2), 1, 2)
>>> print(ret)  # prints tensor([4., 4.])
>>>
>>> rref = rpc.remote("worker1", AsyncExecutionClass)
>>> ret = rref.rpc_async().static_async_add("worker2", torch.ones(2), 1, 2).wait()
>>> print(ret)  # prints tensor([4., 4.])
>>>
>>> rref = rpc.remote("worker1", AsyncExecutionClass)
>>> ret = rref.remote().static_async_add("worker2", torch.ones(2), 1, 2).to_here()
>>> print(ret)  # prints tensor([4., 4.])

Backends

The RPC module can leverage different backends to perform the communication between the nodes. The backend to be used can be specified in the init_rpc() function, by passing a certain value of the BackendType enum. Regardless of what backend is used, the rest of the RPC API won’t change. Each backend also defines its own subclass of the RpcBackendOptions class, an instance of which can also be passed to init_rpc() to configure the backend’s behavior.

TensorPipe Backend

The TensorPipe agent, which is the default, leverages the TensorPipe library, which provides a natively point-to-point communication primitive specifically suited for machine learning that fundamentally addresses some of the limitations of Gloo. Compared to Gloo, it has the advantage of being asynchronous, which allows a large number of transfers to occur simultaneously, each at their own speed, without blocking each other. It will only open pipes between pairs of nodes when needed, on demand, and when one node fails only its incident pipes will be closed, while all other ones will keep working as normal. In addition, it is able to support multiple different transports (TCP, of course, but also shared memory, NVLink, InfiniBand, …) and can automatically detect their availability and negotiate the best transport to use for each pipe.

The TensorPipe backend has been introduced in PyTorch v1.6 and is being actively developed. At the moment, it only supports CPU tensors, with GPU support coming soon. It comes with a TCP-based transport, just like Gloo. It is also able to automatically chunk and multiplex large tensors over multiple sockets and threads in order to achieve very high bandwidths. The agent will be able to pick the best transport on its own, with no intervention required.

Example:

>>> import os
>>> from torch.distributed import rpc
>>> os.environ['MASTER_ADDR'] = 'localhost'
>>> os.environ['MASTER_PORT'] = '29500'
>>>
>>> rpc.init_rpc(
>>>     "worker1",
>>>     rank=0,
>>>     world_size=2,
>>>     rpc_backend_options=rpc.TensorPipeRpcBackendOptions(
>>>         num_worker_threads=8,
>>>         rpc_timeout=20 # 20 second timeout
>>>     )
>>> )
>>>
>>> # omitting init_rpc invocation on worker2

Process Group Backend

Warning

The Process Group Backend will be deprecated soon, we recommend using the TensorPipe Backend instead.

The Process Group agent instantiates a process group from the distributed module and utilizes its point-to-point communication capabilities to send RPC messages. Internally, the process group uses the Gloo library.

Gloo has been hardened by years of extensive use in PyTorch and is thus very reliable. However, as it was designed to perform collective communication, it may not always be the best fit for RPC. For example, each networking operation is synchronous and blocking, which means that it cannot be run in parallel with others. Moreover, it opens a connection between all pairs of nodes, and brings down all of them when one fails, thus reducing the resiliency and the elasticity of the system.

Example:

>>> import os
>>> from torch.distributed import rpc
>>> os.environ['MASTER_ADDR'] = 'localhost'
>>> os.environ['MASTER_PORT'] = '29500'
>>>
>>> rpc.init_rpc(
>>>     "worker1",
>>>     rank=0,
>>>     world_size=2,
>>>     backend=rpc.BackendType.PROCESS_GROUP,
>>>     rpc_backend_options=rpc.ProcessGroupRpcBackendOptions(
>>>         num_send_recv_threads=16,
>>>         rpc_timeout=20 # 20 second timeout
>>>     )
>>> )
>>>
>>> # omitting init_rpc invocation on worker2

RRef

An RRef (Remote REFerence) is a reference to a value of some type T (e.g. Tensor) on a remote worker. This handle keeps the referenced remote value alive on the owner, but there is no implication that the value will be transferred to the local worker in the future. RRefs can be used in multi-machine training by holding references to nn.Modules that exist on other workers, and calling the appropriate functions to retrieve or modify their parameters during training. See Remote Reference Protocol for more details.

More Information about RRef

• Remote Reference Protocol
  • Background
  • Assumptions
  • RRef Lifetime
    • Design Reasoning
    • Implementation
  • Protocol Scenarios
    • User Share RRef with Owner as Return Value
    • User Share RRef with Owner as Argument
    • Owner Share RRef with User
    • User Share RRef with User

Distributed Autograd Framework

This module provides an RPC-based distributed autograd framework that can be used for applications such as model parallel training. In short, applications may send and receive gradient recording tensors over RPC. In the forward pass, we record when gradient recording tensors are sent over RPC and during the backward pass we use this information to perform a distributed backward pass using RPC. For more details see Distributed Autograd Design.

class torch.distributed.autograd.context

Context object to wrap forward and backward passes when using distributed autograd. The context_id generated in the with statement is required to uniquely identify a distributed backward pass on all workers. Each worker stores metadata associated with this context_id, which is required to correctly execute a distributed autograd pass.

Example::

>>> import torch.distributed.autograd as dist_autograd
>>> with dist_autograd.context() as context_id:
>>>   t1 = torch.rand((3, 3), requires_grad=True)
>>>   t2 = torch.rand((3, 3), requires_grad=True)
>>>   loss = rpc.rpc_sync("worker1", torch.add, args=(t1, t2)).sum()
>>>   dist_autograd.backward(context_id, [loss])

More Information about RPC Autograd

• Distributed Autograd Design
  • Background
  • Autograd recording during the forward pass
  • Distributed Autograd Context
  • Distributed Backward Pass
    • Computing dependencies
    • FAST mode algorithm
    • SMART mode algorithm
  • Distributed Optimizer
  • Simple end to end example

Distributed Optimizer

Design Notes

The distributed autograd design note covers the design of the RPC-based distributed autograd framework that is useful for applications such as model parallel training.

• Distributed Autograd Design

The RRef design note covers the design of the RRef (Remote REFerence) protocol used to refer to values on remote workers by the framework.

• Remote Reference Protocol

Tutorials

The RPC tutorials introduce users to the RPC framework, provide several example applications using torch.distributed.rpc APIs, and demonstrate how to use the profiler to profile RPC-based workloads.

• Getting started with Distributed RPC Framework

• Implementing a Parameter Server using Distributed RPC Framework

• Combining Distributed DataParallel with Distributed RPC Framework

• Profiling RPC-based Workloads

• Implementing batch RPC processing

• Distributed Pipeline Parallel