registry.py 2.9 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596
  1. # Copyright (c) Facebook, Inc. and its affiliates.
  2. # All rights reserved.
  3. #
  4. # This source code is licensed under the BSD-style license found in the
  5. # LICENSE file in the root directory of this source tree.
  6. import logging
  7. from importlib.metadata import entry_points
  8. from .api import (
  9. rendezvous_handler_registry as handler_registry,
  10. RendezvousHandler,
  11. RendezvousParameters,
  12. )
  13. from .dynamic_rendezvous import create_handler
  14. log = logging.getLogger(__name__)
  15. __all__ = ["get_rendezvous_handler"]
  16. def _create_static_handler(params: RendezvousParameters) -> RendezvousHandler:
  17. from . import static_tcp_rendezvous
  18. return static_tcp_rendezvous.create_rdzv_handler(params)
  19. def _create_etcd_handler(params: RendezvousParameters) -> RendezvousHandler:
  20. from . import etcd_rendezvous
  21. return etcd_rendezvous.create_rdzv_handler(params)
  22. def _create_etcd_v2_handler(params: RendezvousParameters) -> RendezvousHandler:
  23. from .etcd_rendezvous_backend import create_backend
  24. backend, store = create_backend(params)
  25. return create_handler(store, backend, params)
  26. def _create_c10d_handler(params: RendezvousParameters) -> RendezvousHandler:
  27. from .c10d_rendezvous_backend import create_backend
  28. backend, store = create_backend(params)
  29. return create_handler(store, backend, params)
  30. def _register_default_handlers() -> None:
  31. handler_registry.register("etcd", _create_etcd_handler)
  32. handler_registry.register("etcd-v2", _create_etcd_v2_handler)
  33. handler_registry.register("c10d", _create_c10d_handler)
  34. handler_registry.register("static", _create_static_handler)
  35. def _register_out_of_tree_handlers() -> None:
  36. discovered_handler_generators = entry_points(group="torchrun.handlers")
  37. for handler_generator in discovered_handler_generators:
  38. try:
  39. # pyrefly: ignore [bad-index]
  40. get_handler = discovered_handler_generators[handler_generator.name].load()
  41. handler_registry.register(handler_generator.name, get_handler())
  42. except Exception:
  43. log.warning(
  44. "Exception while registering out of tree plugin %s: ",
  45. handler_generator.name,
  46. exc_info=True,
  47. )
  48. def get_rendezvous_handler(params: RendezvousParameters) -> RendezvousHandler:
  49. """
  50. Obtain a reference to a :py:class`RendezvousHandler`.
  51. Custom rendezvous handlers can be registered by
  52. ::
  53. from torch.distributed.elastic.rendezvous import rendezvous_handler_registry
  54. from torch.distributed.elastic.rendezvous.registry import get_rendezvous_handler
  55. def create_my_rdzv(params: RendezvousParameters):
  56. return MyCustomRdzv(params)
  57. rendezvous_handler_registry.register("my_rdzv_backend_name", create_my_rdzv)
  58. my_rdzv_handler = get_rendezvous_handler(
  59. "my_rdzv_backend_name", RendezvousParameters
  60. )
  61. """
  62. return handler_registry.create_handler(params)