[WIP][CI] Add multi node ci

[Potabk]

What this PR does / why we need it?

This pr introduces multi-node ci for vllm-ascend on different backend(mp, ray)

Does this PR introduce any user-facing change?

How was this patch tested?

• vLLM version: v0.10.0
• vLLM main: vllm-project/vllm@6807af8

[gemini-code-assist]

Code Review

This pull request introduces two example scripts for multi-node serving: multi_node_dp.sh for data parallelism and multi_node_ray.sh for Ray-based deployment. My review focuses on improving the robustness, correctness, and reusability of these scripts. For multi_node_dp.sh, I've identified several critical and high-severity issues including an undefined variable, hardcoded paths, and missing error checks, and I've provided suggestions to address them. For multi_node_ray.sh, I've pointed out that error messages should be redirected to stderr, which is a best practice for shell scripts. Applying these suggestions will make the examples more reliable and user-friendly.

[gemini-code-assist]

The MASTER_ADDR variable is used on line 52 but it is not defined. This will cause an error when running as a worker. Please add a check to ensure this variable is set.

Suggested change

	echo "====> Running worker node"
	if [ -z "${MASTER_ADDR-}" ]; then
	echo "Error: MASTER_ADDR environment variable must be set for worker node." >&2
	exit 1
	fi
	echo "====> Running worker node"

[gemini-code-assist]

The model path is hardcoded on lines 23 and 45, which limits the script's reusability. It's better to pass the model path as a command-line argument and validate it. You can then use the variable (e.g., $MODEL_PATH) in the vllm serve commands on lines 23 and 45.

Suggested change

	NODE_TYPE=$1
	NODE_TYPE=$1
	MODEL_PATH=$2
	if [ -z "$MODEL_PATH" ]; then
	echo "Error: model path must be provided as the second argument" >&2
	exit 1
	fi

[gemini-code-assist]

The hostname -I command can fail and return an empty string, which would cause issues in subsequent commands. It's safer to check if an IP was actually found.

Suggested change

	local_ip=$(hostname -I | awk '{print $1}')
	local_ip=$(hostname -I | awk '{print $1}')
	if [ -z "$local_ip" ]; then
	echo "Error: Could not determine local IP address." >&2
	exit 1
	fi

[gemini-code-assist]

Error messages should be printed to standard error (stderr) instead of standard output (stdout). This is a standard practice in shell scripting to allow for proper output redirection and error handling. This applies to other error and timeout messages in this script as well (e.g., lines 68, 93, 111, 116).

Suggested change

	echo "Error: Missing argument --ray_address"
	echo "Error: Missing argument --ray_address" >&2

[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.