env: add isolated CUDA 13 / vLLM 0.25.1 / LMCache 0.5.2 profile

[rickisba]

Summary

Add an isolated, additive development-image profile for the target serving baseline in #148:

• CUDA 13.0.0 on Ubuntu 22.04;
• Python 3.12 in /opt/venv;
• PyTorch 2.11.0+cu130;
• torchvision / torchaudio 0.26.0 / 2.11.0;
• vLLM 0.25.1;
• LMCache 0.5.2;
• compatible CacheRoute application/development dependency ranges.

Changes

• Add env/docker/cu130/Dockerfile.
• Add env/docker/cu130/constraints.txt for the exact serving stack and shared compatibility ranges.
• Add env/docker/cu130/requirements-dev.txt for the new profile without modifying the legacy root requirements.txt.
• Add env/docker/cu130/README.md with build, container, editable-install, runtime, and validation instructions.

Compatibility and non-regression

This PR is intentionally additive:

• it does not modify env/docker/Dockerfile;
• it does not modify the legacy root requirements.txt;
• it does not modify pyproject.toml;
• it does not change Scheduler, KDN, Proxy, Instance, routing, queueing, or storage behavior;
• the Docker build uses env/docker/cu130 as its context, avoiding large model/KDN/log directories without adding a repository-wide .dockerignore that could affect existing builds.

The new profile uses /opt/venv to isolate the serving stack from Ubuntu system Python packages. It installs FFmpeg for TorchCodec and excludes the Python-2-era Booktype==1.5, which can overwrite the modern redis module with invalid Python 2 source.

Validation performed

The image was built successfully and validated with:

• Python 3.12.13;
• PyTorch 2.11.0+cu130 and CUDA runtime 13.0;
• vLLM 0.25.1;
• LMCache 0.5.2;
• Transformers 5.12.1;
• Redis Python client 8.0.1;
• successful python3 -m pip check;
• successful imports of Torch, TorchCodec, vLLM, LMCache, and Redis;
• successful startup of the existing CacheRoute Scheduler, KDN, Proxy, and Instance workflow.

Scope boundary

This PR provides the new installation and image-build interface, but does not claim that #148 is complete. The KDN raw Redis dump/inject contract still requires runtime migration and validation against LMCache MP RESP L2 behavior. The README records that limitation explicitly.

Partially addresses #148.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 66a9a26e27

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Keep Ubuntu's system Python on 3.10

When this development image later runs Ubuntu Python-based administration tools such as add-apt-repository, replacing /usr/bin/python3 with Python 3.12 makes those tools load the wrong interpreter; Ubuntu 22.04 installs extensions such as apt_pkg for its system Python 3.10, so the commands can fail with import errors. The checked update-alternatives --help confirms that --install <link> <name> <path> turns the supplied link into an alternatives-managed symlink. Create the virtual environment with /usr/bin/python${PYTHON_VERSION} -m venv directly and leave /usr/bin/python3 unchanged; the existing PATH already selects /opt/venv/bin/python3 for application commands.

Useful? React with 👍 / 👎.

[rickisba]

The isolated CUDA 13 / vLLM 0.25.1 / LMCache 0.5.2 environment profile from this PR has now been incorporated into draft PR #149 (v1/runtime-compat).

#149 contains the same additive environment boundary plus the runtime compatibility layer and KDN migration work. The imported profile is under env/docker/cu130/, and the dedicated image now defaults to CACHEROUTE_RUNTIME_PROFILE=v1.

No changes from this PR were applied to the legacy Dockerfile, root requirements.txt, or pyproject.toml. After verifying #149, this PR can be closed as superseded.

[rickisba]

This environment profile has been superseded. Its CUDA 13 / vLLM 0.25.1 / LMCache 0.5.2 files were consolidated and merged through PR #149, the homepage/quick-start entry was merged through PR #151, and the migration closeout plus reproducible validation pipeline is now tracked in PR #152. Closing #150 avoids accidentally merging the older standalone branch a second time.