espnet3.parallel.parallel.get_client

source

espnet3.parallel.parallel.get_client(config: DictConfig = None, setup_fn: Callable[[], dict] | None = None) → Generator[Client, None, None]

Yield a Dask client configured for shard-based runner execution.

• Parameters:
  • config (DictConfig , optional) – Cluster config.
  • setup_fn (Callable [ [ ] , dict ] , optional) – A setup function that runs on each worker and returns a dictionary of environment variables.
• Yields:Client – A Dask client instance tied to the global cluster.

Notes

• setup_fn is executed once per worker and should return a dict.
• Callers should prefer BaseRunner instead of interacting with the client directly for ESPnet3 workflows.

Example

>>> from omegaconf import OmegaConf
>>> cfg = OmegaConf.create({"env": "local", "n_workers": 2, "options": {}})
>>> set_parallel(cfg)
>>> with get_client() as client:
...     future = client.submit(lambda: 42)
...     result = future.result()