【main】patch sched_yield

[fems14]

What this PR does / why we need it?

On Arm systems, os.sched_yield() does not take effect, causing the GIL (Global Interpreter Lock) to remain unrelinquished and resulting in CPU bound issues. This PR applies a patch to sched_yield in vLLM, making the process execute time.sleep(0) instead to release the GIL.

Does this PR introduce any user-facing change?

How was this patch tested?

• vLLM version: v0.11.0rc3
• vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.0

[github-actions]

👋 Hi! Thank you for contributing to the vLLM Ascend project. The following points will speed up your PR merge:‌‌

• A PR should do only one thing, smaller PRs enable faster reviews.
• Every PR should include unit tests and end-to-end tests ‌to ensure it works and is not broken by other future PRs.
• Write the commit message by fulfilling the PR description to help reviewer and future developers understand.

If CI fails, you can run linting and testing checks locally according Contributing and Testing.

[gemini-code-assist]

Code Review

This pull request introduces a patch to disable sched_yield in vllm by monkey-patching the USE_SCHED_YIELD flag in vllm.distributed.utils. While this may be necessary for performance on Ascend hardware, the current implementation is fragile. It directly assigns a value to an attribute of a dependency, which will cause the application to crash with an AttributeError if that attribute is ever renamed or removed in a future version of vllm. My review includes a critical suggestion to make this patch more robust by checking for the attribute's existence before modification, thus preventing potential crashes and improving maintainability.

[gemini-code-assist]

This monkey patch is fragile and can cause the application to crash on startup if the USE_SCHED_YIELD attribute is removed or renamed in a future version of the vllm library. To make the patch more robust, you should defensively handle the attribute assignment. This prevents an AttributeError and ensures the application can still run, while also providing a warning that the patch was not applied.

Suggested change

	import vllm.distributed.utils
	vllm.distributed.utils.USE_SCHED_YIELD = False
	import logging
	import vllm.distributed.utils
	try:
	vllm.distributed.utils.USE_SCHED_YIELD = False
	except AttributeError:
	logging.warning(
	"Could not apply patch for sched_yield: 'vllm.distributed.utils.USE_SCHED_YIELD' not found. "
	"This may be due to a vLLM version update.")

[jianzs]

@fems14 Explain the reasoning behind your code changes in both comments and commit messages.

[wangxiyuan]

add os.arch check? I assume this patch should only work on aarch64?