Fix flaky cudf-polars Quent integration test

[TomAugspurger]

Description

On main and release/26.08, the following will error reliably:

pytest -p no:randomly -n0 \
	python/cudf_polars/tests/streaming/test_ray.py::test_yields_engine \
	python/cudf_polars/tests/quent/test_quent_integration.py

In CI, it will fail if a worker that previously happened to run one of the ray tests with 0 GPUs happened to worksteal the quent integration tests.

This fixes the issue by passing through ray_init_options and ray_num_ranks when creating the RayEngine for the quent tests, similar to the rest of the tests

Closes #23441

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 75a00f90-f90a-4dbd-813b-4582e543c3b9

📥 Commits

Reviewing files that changed from the base of the PR and between 4a902a2 and 35074e5.

📒 Files selected for processing (1)

• python/cudf_polars/tests/quent/test_quent_integration.py

---

📝 Walkthrough

Summary by CodeRabbit

• Tests
  • Enhanced Ray-backed integration test setup with configurable rank counts and initialization options.
  • Enabled GPU sharing during Ray test execution.
  • Improved support for externally controlled Ray engine configurations.

Walkthrough

The Quent integration fixture now accepts Ray rank and initialization options, passes them to RayEngine, and enables GPU sharing for the Ray backend.

Changes

Ray Quent fixture

Layer / File(s)	Summary
Configure RayEngine fixture setup python/cudf_polars/tests/quent/test_quent_integration.py	The fixture adds typed Ray configuration inputs and constructs RayEngine with explicit ranks, initialization options, and GPU sharing enabled.

Estimated code review effort: 2 (Simple) | ~10 minutes

Possibly related issues

• #23292 — Both changes update RayEngine test initialization with rank configuration and GPU-sharing behavior.

Suggested reviewers: mroeschke

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly matches the main change: fixing a flaky Quent integration test.
Description check	✅ Passed	The description directly explains the RayEngine setup fix and the flaky test failure it addresses.
Linked Issues check	✅ Passed	The changes address issue #23441 by passing Ray init options and num_ranks to avoid the no-GPU startup error.
Out of Scope Changes check	✅ Passed	The PR stays focused on the Quent Ray test fixture and related Ray initialization handling.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands.}

[vyasr]

Approving for now to avoid the issues, although it would be nice if there was a context manager or similar approach that we could do to avoid global state mutations causing problems like this.

[TomAugspurger]

Agreed, though I'm not sure how easy that is to do... Our tests really do need to share the underlying Ray cluster resources between tests, since they're so expensive to create. The current API of RayEngine accepting ray_init_options and optionally create those resources makes things awkward. If RayEngine only accepted a connection to a running ray cluster things might be less awkward in our tests.

[TomAugspurger]

/merge

[mroeschke]

/merge