[single_controller, ray] fix: shut ray down after initializes it

[lantian7]

What does this PR do?

Add concise overview of what this PR aims to achieve or accomplish. Reference related GitHub issues and PRs that help with the review.

To prevent Ascend NPU TBE errors caused by resource leakage, ensure that ray.shutdown()is explicitly called after initializing Ray with ray.init().

Address the first issue in #3316

Checklist Before Starting

• Search for similar PRs. Paste at least one query link here: ...
• Format the PR title as [{modules}] {type}: {description} (This will be checked by the CI)
  • {modules} include fsdp, megatron, sglang, vllm, rollout, trainer, ci, training_utils, recipe, hardware, deployment, ray, worker, single_controller, misc, perf, model, algo, env, tool, ckpt, doc, data
  • If this PR involves multiple modules, separate them with , like [megatron, fsdp, doc]
  • {type} is in feat, fix, refactor, chore, test
  • If this PR breaks any API (CLI arguments, config, function signature, etc.), add [BREAKING] to the beginning of the title.
  • Example: [BREAKING][fsdp, megatron] feat: dynamic batching

Test

For changes that can not be tested by CI (e.g., algorithm implementation, new model support), validate by experiment(s) and show results like training curve plots, evaluation results, etc.

API and Usage Example

Demonstrate how the API changes if any, and provide usage example(s) if possible.

# Add code snippet or script demonstrating how to use this

Design & Code Changes

Demonstrate the high-level design if this PR is complex, and list the specific changes.

Checklist Before Submitting

Important

Please check all the following items before requesting a review, otherwise the reviewer might deprioritize this PR for review.

• Read the Contribute Guide.
• Apply pre-commit checks: pre-commit install && pre-commit run --all-files --show-diff-on-failure --color=always
• Add / Update the documentation.
• Add unit or end-to-end test(s) to the CI workflow to cover all the code. If not feasible, explain why: ...
• Once your PR is ready for CI, send a message in the ci-request channel in the verl Slack workspace. (If not accessible, please try the Feishu group (飞书群).)

[gemini-code-assist]

Code Review

This pull request adds a call to ray.shutdown() to prevent resource leaks. While this is a good practice, the current implementation calls ray.shutdown() unconditionally. This could lead to issues if the run_ppo function is ever used as a library call within a larger Ray application, as it would tear down the entire Ray cluster. I've added a comment with a suggestion to only shut down Ray if it was initialized within the same function, which makes the code more robust and reusable.

[gemini-code-assist]

Calling ray.shutdown() unconditionally can cause issues if this function is used as part of a larger Ray application where Ray was initialized externally. It would shut down the entire Ray cluster, which is likely not the intended behavior for the caller.

A more robust approach is to only shut down Ray if it was initialized within this function. This can be achieved by tracking the initialization status.

Here is a suggested implementation for the run_ppo function:

def run_ppo(config) -> None:
    initialized_by_this_function = False
    if not ray.is_initialized():
        # this is for local ray cluster
        default_runtime_env = {
            "env_vars": {"TOKENIZERS_PARALLELISM": "true", "NCCL_DEBUG": "WARN", "VLLM_LOGGING_LEVEL": "WARN"}
        }
        ray_init_kwargs = config.ray_kwargs.get("ray_init", {})
        runtime_env_kwargs = ray_init_kwargs.get("runtime_env", {})
        runtime_env = OmegaConf.merge(default_runtime_env, runtime_env_kwargs)
        ray_init_kwargs = OmegaConf.create({**ray_init_kwargs, "runtime_env": runtime_env})
        print(f"ray init kwargs: {ray_init_kwargs}")
        ray.init(**OmegaConf.to_container(ray_init_kwargs))
        initialized_by_this_function = True

    try:
        if (
            is_cuda_available
            and config.global_profiler.tool == "nsys"
            and OmegaConf.select(config.global_profiler, "steps") is not None
            and len(OmegaConf.select(config.global_profiler, "steps")) > 0
        ):
            nsight_options = OmegaConf.to_container(
                config.global_profiler.global_tool_config.nsys.controller_nsight_options
            )
            runner = TaskRunner.options(runtime_env={"nsight": nsight_options}).remote()
        else:
            runner = TaskRunner.remote()
        ray.get(runner.run.remote(config))
    finally:
        if initialized_by_this_function:
            ray.shutdown()

This ensures that ray.shutdown() is only called when ray.init() was called in the same scope. I've also wrapped the Ray task execution in a try...finally block to ensure ray.shutdown() is called even if the task fails, but only when appropriate.

[lantian7]

done

[CLAassistant]

All committers have signed the CLA.