@edoakes There is a new issue I want to discuss:
Currently, ray.get_runtime_context().runtime_env return a new dict instance so we can modify it directly for inheritance. Like:

Actor.options(runtime_env=ray.get_runtime_context().runtime_env.update(
            {"env_vars": {"A": "a", "B": "b"}}))

This use case is a little weird because runtime_env is a property API which often be seen as a reference. So, do we need to add a new API in runtime_context? Like:

Actor.options(runtime_env=ray.get_runtime_context().get_runtime_env().update(
            {"env_vars": {"A": "a", "B": "b"}}))

@SongGuyang it's not safe to return a reference to the worker's runtime_env anyways (don't want users messing with it), so I think let's just document that it returns a copy

linter failing on docs

@SongGuyang ping to fix tests and get this merged

@edoakes do we still need this PR? The previous PR has been reverted.

Yes we should still merge this; it updates tests and removes the currently unused code

Oh, sorry, now I realize that this makes no sense after the revert.... I was just looking at the diff but didn't realize it doesn't apply to master. My bad.

OK.