Actorpool pipelining

How severe does this issue affect your experience of using Ray?

  • Low: It annoys or frustrates me for a moment.

With Ray tasks, we can easily pass object references around to build pipelines. However, it seems unnatural to do this with Ray actors.

After some research, it seems that actor pool is the only way to load balance actor resources (e.g., use 10 actors to read and process 1000 data). However, actor pool get methods returns ray.get() instead of future or promises, which denies potential downstream pipeline.

A possible solution is to just initiate 1000 actors and make use of the scheduler?

Is there something missing here, or is there other patterns to use in this type of scenario?

Could you elaborate the pipelining you want to do?

Also you don’t have to use the ray actor pool util, you can easily write your own actor pool that serves your purpose.

Sorry for late replies. Thank you for replying. I think my problem can be solved by modifying actorpool.