[alec-flowers]

Purpose

Expose the existing vllm-rs executable on PATH in the CUDA serving image. The wheel already bundles it inside the vllm package, but installing the wheel only creates the Python vllm console script. This prevents calling vllm-rs from a container shell or selecting it with --entrypoint vllm-rs.

After installing the wheel, vllm-base links the bundled binary into /usr/local/bin and runs vllm-rs --help to check the executable at image-build time. The package location comes from Python's sysconfig, so it follows the selected Python version. The serving images inherit the link. The default vllm serve entrypoint is preserved.

Checked the published linux/amd64 tags on 2026-09-18:

Tag	Build	Manifest digest
latest	v0.29.0 / 98dff2a81d74	sha256:082ca6f035279109041ffd3fe0695cb568b29bc580b35c4f297a66a08b216c1b
nightly	0.29.1rc1.dev347+gdee37d891	sha256:b88541ab10490f6ee43d2e352edcfcb4b4e0e6b1fdd26ed427c89d73e3fa6a7b

In both published containers, the bundled executable is /usr/local/lib/python3.12/dist-packages/vllm/vllm-rs and works when invoked by its full path. Bare vllm-rs --help and docker run --entrypoint vllm-rs ... --help both fail with exit status 127.

Duplicate-work checks: searched open PRs for "vllm-rs" "PATH", "vllm-rs" "symlink", and "vllm-rs" "entrypoint". No open PR addresses this change. #51503 adds a separate standalone renderer image; this fixes the existing CUDA serving image.

AI assistance was used to prepare this change.

Test Plan

For each pinned image above, build a small validation image with FROM <published-image> and the exact new RUN instruction extracted from this Dockerfile. Run these checks against the resulting image ($patched_image):

docker run --rm --network none --entrypoint /bin/sh "$patched_image" \
  -c 'set -e; command -v vllm-rs; readlink /usr/local/bin/vllm-rs; vllm-rs --version; vllm-rs --help'
docker run --rm --network none --entrypoint vllm-rs "$patched_image" --help
docker run --rm --network none --user 2000:0 --entrypoint vllm-rs "$patched_image" --help

Test Result

• Reproduced the missing command and Docker entrypoint failure on both unmodified published images: exit status 127.
• Built the validation layer on both images with docker build --network none --build-arg BASE_IMAGE=<pinned-image> ...: passed, including the new build-time vllm-rs --help check.
• All three runtime checks above passed for both patched images. command -v vllm-rs returns /usr/local/bin/vllm-rs, linked to the original bundled binary. Tested linux/amd64 without GPUs.
• .venv/bin/pre-commit run --files docker/Dockerfile docs/contributing/dockerfile/dockerfile.md docs/deployment/docker.md: passed, including the Dockerfile dependency-graph check.
• git diff --check origin/main...HEAD: passed.
• The full CUDA build from source and model inference were not run; validation applied the exact added instruction to both actual published images. Model evaluations are not applicable to this packaging-only change.

[mergify]

Documentation preview: https://vllm--57606.org.readthedocs.build/en/57606/

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[BugenZhao]

[AUTOMATED] @codex review

[BugenZhao]

Thanks @alec-flowers

[BugenZhao]

/ci run --allow-stale

[github-actions]

✅ @alec-flowers, CI is now available for this PR.

• /ci run starts upstream CI; /amd-ci run starts AMD CI only.
• Your branch must contain every commit currently on its upstream target branch. Merge or rebase onto the latest target branch, then rerun the command. Append --allow-stale to a run command to test an outdated branch at your own risk.
• /ci retry retries failed jobs in the CI build for the current PR head. If the current head has no CI build, it starts a new CI build for the current head containing only jobs that failed in the latest earlier CI build for this PR.
• /amd-ci retry retries failed jobs in AMD CI for the current PR head. Use /amd-ci run when the current head has no AMD CI build.
• /ci cancel cancels scheduled or running CI builds for this PR branch; /amd-ci cancel does the same for AMD CI only.

[github-actions]

✅ Triggered Buildkite CI #90191 for commit 1e69b7ebc5b6.

⚠️ This PR is 88 commits behind upstream main. Running CI at your own risk because --allow-stale was requested; outdated CI configuration may cause failures. Before merging, merge or rebase onto the latest main, then rerun /ci run on the latest PR commit.

[alec-flowers]

/ci run

[github-actions]

✅ Triggered Buildkite CI #90268 for commit 5f08aba989be.

[alec-flowers]

/ci retry

[github-actions]

✅ Queued 4 failed job(s) for retry in Buildkite CI #90268.

[alec-flowers]

/ci retry

[github-actions]

✅ Queued 1 failed job(s) for retry in Buildkite CI #90268.