[Misc] allow user to specify where to write gpu_p2p_access_cache through VLLM_CACHE_DIR env var

[sfc-gh-zhwang]

FILL IN THE PR DESCRIPTION HERE

Context: This is related to #4472. For security reasons of our company(might be same for other companies 😄 ), we do not allow dump a json to a directory where there are existing .so files.

In this PR, I added an environment var VLLM_CACHE_DIR(follow the same pattern in triton), if unset, it's a no-op as today. If set, the gpu_p2p_access_cache will be dumped under $VLLM_CACHE_DIR instead.

BEFORE SUBMITTING, PLEASE READ THE CHECKLIST BELOW AND FILL IN THE DESCRIPTION ABOVE

---

PR Checklist (Click to Expand)

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

PR Title and Classification

Only specific types of PRs will be reviewed. 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.
• [Model] for adding a new model or improving an existing model. Model name should appear in the title.
• [Frontend] For changes on the vLLM frontend (e.g., OpenAI API server, LLM class, etc.)
• [Kernel] for changes affecting CUDA kernels or other compute kernels.
• [Core] for changes in the core vLLM logic (e.g., LLMEngine, AsyncLLMEngine, Scheduler, etc.)
• [Hardware][Vendor] for hardware-specific changes. Vendor name should appear in the prefix (e.g., [Hardware][AMD]).
• [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:

• We adhere to Google Python style guide and Google C++ style guide.
• Pass all linter checks. Please use format.sh to format your code.
• The code need to be well-documented to ensure future contributors can easily understand the code.
• Include sufficient tests to ensure the project to stay correct and robust. This includes both unit tests and integration tests.
• Please add documentation to docs/source/ if the PR modifies the user-facing behaviors of vLLM. It helps vLLM user understand and utilize the new features or changes.

Notes for Large Changes

Please keep the changes as concise as possible. For major architectural changes (>500 LOC excluding kernel/data/config/test), we would expect a GitHub issue (RFC) discussing the technical design and justification. Otherwise, we will tag it with rfc-required and might not go through the PR.

What to Expect for the Reviews

The goal of the vLLM team is to be a transparent reviewing machine. We would like to make the review process transparent and efficient and make sure no contributor feel confused or frustrated. However, the vLLM team is small, so we need to prioritize some PRs over others. Here is what you can expect from the review process:

• After the PR is submitted, the PR will be assigned to a reviewer. Every reviewer will pick up the PRs based on their expertise and availability.
• After the PR is assigned, the reviewer will provide status update every 2-3 days. If the PR is not reviewed within 7 days, please feel free to ping the reviewer or the vLLM team.
• After the review, the reviewer will put an action-required label on the PR if there are changes required. The contributor should address the comments and ping the reviewer to re-review the PR.
• Please respond to all comments within a reasonable time frame. If a comment isn't clear or you disagree with a suggestion, feel free to ask for clarification or discuss the suggestion.

Thank You

Finally, thank you for taking the time to read these guidelines and for your interest in contributing to vLLM. Your contributions make vLLM a great tool for everyone!

[sfc-gh-zhwang]

now i see it, looks like the whole logic is not even needed, we can just populate _gpu_p2p_access_cache?

[youkaichao]

What do you mean by the whole logic is not even needed ?

[sfc-gh-zhwang]

I meant, we don't need to dump a json to disk, instead we just populate _gpu_p2p_access_cache if it's None, otherwise, we just read from _gpu_p2p_access_cache. Am i missing anything?

[youkaichao]

save to disk is useful to save memory, unless you want the first rank of the worker to have less gpu memory: #4021 .

[youkaichao]

For the nccl directory, it is much more complicated than it seems. It is not just a runtime directory, it is an installation time directory. The libnccl.so.2.18.1 is downloaded in installation time, and once it is downloaded, pip thinks the package is already installed, and will not trigger installation the next time.

All the trouble stems from pypi/support#3792 and NVIDIA/nccl#1234 . Until either one is resolved, we can do nothing :(

[sfc-gh-zhwang]

For the nccl directory, it is much more complicated than it seems. It is not just a runtime directory, it is an installation time directory. The libnccl.so.2.18.1 is downloaded in installation time, and once it is downloaded, pip thinks the package is already installed, and will not trigger installation the next time.

All the trouble stems from pypi/support#3792 and NVIDIA/nccl#1234 . Until either one is resolved, we can do nothing :(

Hi sorry for being confusing, i didn't intend to solve that issue. The whole idea of this PR is to allow us to specify VLLM_CACHE_DIR so that _gpu_p2p_access_cache json is dumped to that dir.

The reason i wanted to have this ability is to prevent _gpu_p2p_access_cache.json write to the same directory of libnccl.so.2.18.1 for security reason of our company. (and it might be the same case for other companies 😄 )

[youkaichao]

#4548 will introduce VLLM_CONFIG_ROOT used in many places, including the p2p access cache file. Please comment there.