Skip to content

ci: stabilize readiness retry test - #242

Open
StevenWang-CY wants to merge 1 commit into
vllm-project:mainfrom
StevenWang-CY:fix-python-readiness-test-race
Open

ci: stabilize readiness retry test#242
StevenWang-CY wants to merge 1 commit into
vllm-project:mainfrom
StevenWang-CY:fix-python-readiness-test-race

Conversation

@StevenWang-CY

@StevenWang-CY StevenWang-CY commented Sep 4, 2026

Copy link
Copy Markdown
Contributor

Summary

  • replace the retry test delayed listener and wake-up cleanup with two deterministic URLError failures followed by a real /v1/models success through the shared server fixture
  • assert exactly three attempts and one actual server request, preserving the retry contract without port-reservation or listener-shutdown races
  • leave production code and behavior unchanged

On unmodified main at 661d8db, the Python 3.13 job completed the readiness request and then failed in test cleanup when its wake-up connection raced the listener closing: ConnectionRefusedError with 113 tests already passing. Python 3.10-3.12 passed in the same matrix. A 200-run local stress control on the untouched test did not reproduce the scheduling-dependent failure.

Test Plan

  • focused Python 3.13 regression repeated 50 times: 50 passed
  • CI-style Python 3.13 wheel build, install, full suite, and wheel validation: 114 passed; wheel validation passed
  • full editable Python 3.13 suite: 111 passed, 3 skipped
  • cargo fmt --all -- --check
  • cargo check --workspace --all-targets --all-features
  • cargo clippy --workspace --all-targets --all-features -- -D warnings
  • cargo test --workspace --all-features (passed; 5 PostgreSQL tests ignored locally without TEST_POSTGRES_URL)
  • launcher contracts and CLI end-to-end tests
  • cassette validation: 108/108 cassettes, 238 turns
  • strict MkDocs build

Local repository-wide pre-commit hooks passed except that the platform-hanging check-added-large-files all-files invocation was skipped; that hook passed directly against the only changed file.

Rebased the unchanged patch onto current main at c04ac196. All 12 hosted checks pass on head 6455cae, including Python 3.10-3.13, Rust, PostgreSQL, both CLI harnesses, pre-commit, and DCO.

Signed-off-by: Chuyue Wang <stevenwang0805@outlook.com>
@StevenWang-CY
StevenWang-CY force-pushed the fix-python-readiness-test-race branch from 3ee31b4 to 6455cae Compare September 5, 2026 15:19
@StevenWang-CY

Copy link
Copy Markdown
Contributor Author

@franciscojavierarceo could you review this? All 12 checks pass after rebasing onto main. This one-file test fix removes the cleanup race while retaining a real HTTP success check; production code is unchanged.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant