ray-project / ray

Ray is a unified framework for scaling AI and Python applications. Ray consists of a core distributed runtime and a set of AI Libraries for accelerating ML workloads.
https://ray.io
Apache License 2.0
33.25k stars 5.63k forks source link

[core][experimental] Remove separate asyncio path for accelerated DAGs #46307

Open stephanie-wang opened 3 months ago

stephanie-wang commented 3 months ago

Description

Right now asyncio support in accelerated DAGs uses a slightly different codepath from normal DAGs:

This is because asyncio execution currently starts a background thread to read/write channels, so we need these extra APIs for coordinating with the background thread. With timeout support for channel reads and writes (#46259), we might be able to remove some of these APIs and support asyncio with single-threaded execution, using mostly the same codepath as non-asyncio execution.

Use case

No response

rkooo567 commented 1 month ago

downgrade for now. let's revisit after resolving p1s.