Add config flag for VLLM_DISABLE_COMPILE_CACHE

[elizabetht]

Purpose

Add config flag for VLLM_DISABLE_COMPILE_CACHE which should be overridden if env var is set

Test Plan

Scenario	Config Value	Env Var	Test
Default (nothing set)	False	Not set	✅ test_disable_compile_cache_config_without_env_var
User sets config to True	True	Not set	✅ test_disable_compile_cache_config + test_disable_compile_cache_config_without_env_var
User sets env var	False	1	✅ test_disable_compile_cache_env_var_override + test_VLLM_DISABLE_COMPILE_CACHE
Both set	True	1	✅ test_disable_compile_cache_both_set (newly added)
Config True, env var 0	True	0	Implicitly covered - env var 0 is treated as "not set" by the env parsing logic, so it falls back to "Config True, env var not set"

Test Result

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
• (Optional) Release notes update. If your change is user facing, please update the release notes draft in the Google Doc.

[elizabetht]

@ProExpertProg @zou3519

[gemini-code-assist]

Code Review

This pull request introduces a new configuration flag, disable_compile_cache, to control the vLLM compile cache, with the environment variable VLLM_DISABLE_COMPILE_CACHE taking precedence. The changes are well-implemented across the configuration, backend, and interface files. The documentation has also been updated accordingly. My main feedback is to refactor the newly added tests for better readability and maintainability.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

ℹ️ About Codex in GitHub

Codex has been enabled to automatically 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 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[mergify]

Documentation preview: https://vllm--30108.org.readthedocs.build/en/30108/

[mergify]

Hi @elizabetht, the pre-commit checks have failed. Please run:

uv pip install pre-commit
pre-commit install
pre-commit run --all-files

Then, commit the changes and push to your branch.

For future commits, pre-commit will run automatically on changed files before each commit.

[mergify]

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @elizabetht.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

[yewentao256]

Thanks for the work! But I don't think this flag is needed as we already have the env variable which increased the complexity.

[elizabetht]

@yewentao256 this issue #29917 talks about needing the config flag for this env variable? @ProExpertProg @zou3519 could you weigh in on this issue?

[ProExpertProg]

Yeah this flag is needed so that we can see whether cache is enabled or disabled when printing config. Env var should stay around so that cache can be disabled without modifying code.

[ProExpertProg]

Please just address the comments

[ProExpertProg]

Instead of adding the key to inductor config, let's just add a parameter

[ProExpertProg]

Let's just add a separate param. Could also pass the whole CompilationConfig around

[zou3519]

Btw, positive flags are generally better than negative flags. Something like enable_compile_cache. Because double negation gets confusing.

I know the envvar is already negative, but we should (in the future) add a positive envvar and then deprecate the negative envvar. In that sense, I'd prefer that we have the config flag be positive.