[runtime env] API refactor

Search before asking

• I had searched in the issues and found no similar feature requirement.

Description

1、Can we make the runtime env API strong-typed instead of a json or dict?

In python now, we provide runtime env API which is a dict. We need to do a lot validation logics here https://github.com/ray-project/ray/blob/master/python/ray/_private/runtime_env/validation.py. And If we support runtime env cross-language API, the validation code will be more complex and it couldn’t be reused in Java and C++. Can we make the runtime env API strong-typed instead of a json or dict? If we do, I think runtime env will be clearer and easy to maintain. Maybe the strong-typed API is like:

runtime_env = RuntimeEnvBuilder.set_env_vars({"a": "b"}).set_pip_packages(["requests", "pendulum==2.1.2"]).build()
ray.init(runtime_env=runtime_env)

The cross-language API is like:

java_runtime_env = RuntimeEnvBuilder.set_env_vars({"a": "b"}).set_jars(["s3://abc.jar"]).build()
ray.init(runtime_env=java_runtime_env)

2、Can we remove the code_search_path in Python, Java and C++? We can support this logic in runtime env:

• In python, we can reuse the py_modules field. https://github.com/ray-project/ray/pull/21504/files#diff-ff7ace6af03d75dc9d8e031310da8f3994c5f71c8f3b29b587c0d841b63e1a1fR73
• In Java, we can add a new field jars like https://github.com/ray-project/ray/pull/21504/files#diff-ff7ace6af03d75dc9d8e031310da8f3994c5f71c8f3b29b587c0d841b63e1a1fR95.
• In C++, we can add a new field dynamic_library like https://github.com/ray-project/ray/pull/21504/files#diff-ff7ace6af03d75dc9d8e031310da8f3994c5f71c8f3b29b587c0d841b63e1a1fR110.

Use case

No response

Related issues

No response

Are you willing to submit a PR?

• Yes I am willing to submit a PR!

[architkulkarni]

Adding a prefix for language-specific options makes sense to me!

[edoakes]

I don’t see a strong reason to veer from what we currently have. Making it more structured makes sense though. e.g., support both of the following:

runtime_env={"pip": ["requests"]}
runtime_env = ray.runtime_env.RuntimeEnv(pip=["requests"])