[ray,trainer] feat: add master port range configuration for port range

[RobotGF]

…cation

What does this PR do?

support train master port range configuration

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, veomni, sglang, vllm, rollout, trainer, ci, training_utils, recipe, hardware, deployment, ray, worker, single_controller, misc, perf, model, algo, env, tool, ckpt, doc, data, cfg, reward
  • 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).)
• If your PR is related to the recipe submodule, please also update the reference to the submodule commit via git submodule update --remote or cd recipe && git pull origin main.

[gemini-code-assist]

Code Review

This pull request introduces a feature to configure a range of ports for the master process. While the intent is good, the implementation has a few critical issues that prevent it from working correctly. There's a logic error in port selection that could lead to using an invalid port. Additionally, the new configuration parameter is not correctly passed through all the necessary components, rendering the feature ineffective. I've provided detailed comments on these points.

[gemini-code-assist]

If all ports within the specified master_port_range are in use, the current implementation will fall through the loop and return addr, str(port) will be called with port equal to master_port_range[1]. This port is outside the specified range [start, end) and is not guaranteed to be free. This can lead to unexpected behavior or failures. The function should instead raise an exception if no free port can be found within the given range.

Suggested change

	else:
	port = master_port_range[0]
	while port < master_port_range[1]:
	try:
	with socket.socket() as s:
	s.bind(("", port))
	break
	except OSError:
	port += 1 # Increment port number if already in use
	logger.info("Port %d is already in use, trying port %d", port - 1, port)
	else:
	port = master_port_range[0]
	while port < master_port_range[1]:
	try:
	with socket.socket() as s:
	s.bind(('', port))
	break
	except OSError:
	port += 1 # Increment port number if already in use
	logger.info("Port %d is already in use, trying port %d", port - 1, port)
	else:
	raise RuntimeError(f"Could not find a free port in range {master_port_range}")

[gemini-code-assist]

While master_port_range is correctly stored as an instance attribute here, it is not used later in this method. The call to self._get_master_addr_port at line 555 needs to be updated to pass self.master_port_range to make this feature functional. This is one of several places where the new parameter needs to be plumbed through.

[gemini-code-assist]

The master_port_range is being added to wg_kwargs, but the RayWorkerGroup constructor does not handle this argument. It will be ignored, and the port range configuration will have no effect. You need to update RayWorkerGroup.__init__ to accept master_port_range from **kwargs and then pass it down to _init_with_resource_pool.

[vermouth1992]

Remove this, we can't setup it in the global config

[vermouth1992]

We don't need to pass the config to the trainer, just provide the API so that it is configurable based on actual use case.