[Router][Bugfix] propagate client cancellation

[unsorted-otter]

When a client disconnects from a non-streaming request, the router just keeps waiting for the backend's response headers, it never checks whether the client is still around. So the engine has no way to know it should stop, runs the request to completion, and the router throws the answer away.

This races waiting for the backend headers against the client disconnecting, same idea as vLLM's own entrypoints.utils.with_cancellation. Whichever happens first wins: if the client leaves, the pending call gets cancelled, which closes the backend connection and lets the engine abort. A normal response works exactly as before.

#867 already tried to fix this and never got merged. Two differences here, both directly answering what that review pointed out:

• It polled request.is_disconnected(). vLLM's own code avoids that on purpose, because it breaks once there's middleware in the stack. This uses a background task racing request.receive() for http.disconnect instead, same as with_cancellation.
• Its review caught a log-injection issue from logging the raw X-Request-Id header. The one new log line here strips that before logging it.

Also moved on_request_complete into a finally block, since it used to only run after a successful response. A cancelled or failed request stayed marked as in-flight forever, and the routing logic kept treating that engine as busier than it really was.

Test plan

• src/tests/test_client_disconnect.py: covers a client disconnecting before the backend responds (checks the backend call actually gets torn down, not just that we return 499), a normal connected client still getting its response, and the cleanup helper correctly telling apart its own cancellation from an unrelated one hitting the caller at the same time (there's a test that fails against the naive version of that check, so it doesn't creep back in).
• Ran pytest src/tests/test_client_disconnect.py locally, all 6 pass.
• black, isort, and ruff all pass on the changed files.

Fixes #634

---

• Make sure the code changes pass the pre-commit checks.
• Sign-off your commit by using -s when doing git commit
• Try to classify PRs for easy understanding of the type of changes, such as [Bugfix], [Feat], and [CI].

Detailed Checklist (Click to Expand)

Thank you for your contribution to production-stack! Before submitting the pull request, please ensure the PR meets the following criteria. This helps us maintain the code quality and improve the efficiency of the review process.

PR Title and Classification

Please try to classify PRs for easy understanding of the type of changes. The PR title is prefixed appropriately to indicate the type of change. Please use one of the following:

• [Bugfix] for bug fixes.
• [CI/Build] for build or continuous integration improvements.
• [Doc] for documentation fixes and improvements.
• [Feat] for new features in the cluster (e.g., autoscaling, disaggregated prefill, etc.).
• [Router] for changes to the vllm_router (e.g., routing algorithm, router observability, etc.).
• [Misc] for PRs that do not fit the above categories. Please use this sparingly.

Note: If the PR spans more than one category, please include all relevant prefixes.

Code Quality

The PR need to meet the following code quality standards:

• Pass all linter checks. Please use pre-commit to format your code. See README.md for installation.
• The code need to be well-documented to ensure future contributors can easily understand the code.
• Please include sufficient tests to ensure the change is stay correct and robust. This includes both unit tests and integration tests.

DCO and Signed-off-by

When contributing changes to this project, you must agree to the DCO. Commits must include a Signed-off-by: header which certifies agreement with the terms of the DCO.

Using -s with git commit will automatically add this header.

What to Expect for the Reviews

We aim to address all PRs in a timely manner. If no one reviews your PR within 5 days, please @-mention one of YuhanLiu11
, Shaoting-Feng or ApostaC.

[gemini-code-assist]

Code Review

This pull request implements client disconnect handling in the router, aborting backend requests and returning a 499 status code if a client disconnects before a response is sent. It also ensures request statistics are properly cleaned up on all exit paths. The review feedback highlights a potential resource leak in the disconnect handling logic, Python compatibility issues with asyncio.Task.cancelling() on Python versions older than 3.11, and a recommendation to make the log sanitization helper safer against non-string inputs.