Skip to content

build(deps): require ray[default]>=2.60.0,<3 for the ray extra (ITL-1210) - #427

Open
eywalker wants to merge 2 commits into
mainfrom
eywalker/itl-1210-orcapod-python-triage-the-open-dependabot-ray
Open

eywalker wants to merge 2 commits into
mainfrom
eywalker/itl-1210-orcapod-python-triage-the-open-dependabot-ray

Conversation

@eywalker

@eywalker eywalker commented Nov 17, 2026 •

Copy link
Copy Markdown
Contributor

Note

Safe to merge after 0.4.0rc1, not for rc1. Merge #426 first, then this branch takes main back in and relocks.

With Ray 2.54 or later, a driver started with uv run hangs on its first RayExecutor call when the project's version comes from version control. No task starts and no error is raised. Ray rebuilds the project on its workers from an upload of the working directory, and since 2.54 that upload leaves .git out. Affected: a developer running RayExecutor from an Orcapod source checkout with uv run, and, by the same mechanism, a user whose own project takes its version from VCS (hatch-vcs or setuptools-scm) and who starts the driver with uv run. Not affected: a driver started with python, a project with a fixed version (Orcapod installed from PyPI), and a remote cluster with its own environment.

Verified locally with one function pod on a local cluster. ray 2.48.0 with uv run: works. ray 2.60.0 with uv run: hangs. ray 2.60.0 started with .venv/bin/python: works. ray 2.60.0 with uv run and RAY_ENABLE_UV_RUN_RUNTIME_ENV=0: works. This PR documents that workaround in docs/troubleshooting.md, the RayExecutor docstring and the changelog fragment (Known limitations). ITL-1213 follows up on RayExecutor raising instead of hanging.

Raise the Ray requirement (ITL-1210). The ray extra and the dev group pinned ray[default]==2.48.0, the version that happened to be installed when the pin was written. They now take ray[default]>=2.60.0,<3, and uv.lock holds ray 2.60.0, the latest release (wheels for CPython 3.11 to 3.13). The lower bound is the version tested here.

Related: ITL-1210

Changes

  • pyproject.toml: ray[default]==2.48.0 becomes ray[default]>=2.60.0,<3 in the ray extra and in the dev group.
  • uv.lock: ray 2.48.0 → 2.60.0, plus aiohttp 3.12.15 → 3.14.5 and yarl 1.20.1 → 1.25.1, which ray 2.60 requires.
  • docs/troubleshooting.md and the RayExecutor docstring: the uv run hang and its workaround.
  • changelog.d/ITL-1210.md.

What changes for a user

Ray 2.59 turned on token authentication by default for local clusters. RayExecutor calls ray.init() with no address when it is given no ray_address. That call now enables authentication and writes a token to ~/.ray/auth_token if none exists. Set RAY_AUTH_MODE=disabled to opt out. Connecting to a remote or multi-node cluster (ray_address, including ray://) is unchanged.

Release notes for Ray 2.49 to 2.60, checked against what Orcapod calls (ray.init(address=, runtime_env=), ray.remote(...), ray.get, ObjectRef.future() including Ray Client, ray.nodes, ray.cluster_resources, ray.available_resources, ray.is_initialized, ray.shutdown): none of these is removed or changed. Besides the authentication default, the relevant changes are: Python 3.9 dropped (2.54); Pydantic v1 support removed (2.56; Orcapod requires Pydantic 2); cloudpickle 3.1.2 (2.55); hardened runtime_env package extraction and pickle deserialization (2.56).

Merge order

This PR touches uv.lock and changelog.d/ITL-1210.md, as #426 does. Merge #426 first. This branch then takes main back in, relocks and combines the two fragments.

Checked locally (Python 3.12, ray 2.60.0)

  • The tests that use Ray (tests/test_core/test_regression_fixes.py, tests/test_pipeline/test_logging_capture.py, tests/test_core/function_pod/test_function_pod_decorator.py): 80 passed, no warnings. These mock Ray or run the capture wrapper in-process: no test in the suite starts a Ray cluster.
  • ruff check, ruff format --check, basedpyright (0 errors), and the changelog fragment check.

🤖 Generated with Claude Code

Posted by a Claude Code agent on behalf of @eywalker.

…210)

The ray extra and the dev group pinned ray[default]==2.48.0. They now take
ray[default]>=2.60.0,<3, and uv.lock holds ray 2.60.0 (with aiohttp 3.14.5
and yarl 1.25.1, which it requires).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@eywalker
eywalker requested a balanced review from Copilot November 17, 2026 19:09
@coderabbitai

coderabbitai Bot commented Nov 17, 2026 •

Copy link
Copy Markdown

Warning

Review limit reached

  • Run on-demand review

This review includes 4 billable files and costs up to $1.00.

  • Ask an admin to make reviews automatic

Open in CodeRabbit

Reviews can continue after your included limit without a manual trigger. An admin must approve usage-based billing.

Or wait 22 minutes for your next included review.

Check out review usage here.

View limit details

Limit details: You’ve used the included review currently available. Your 108 included PR review attempts over the past 7 days set your current allowance at 1 review per hour.

Learn how review limits work.

Review configuration:

Run configuration

Commits

Reviewing files that changed from the base of the PR and between 51be5e4 and 1662988.

Files ignored due to path filters: 1

  • uv.lock is excluded by !**/*.lock
Files selected for processing: 4

  • changelog.d/ITL-1210.md
  • docs/troubleshooting.md
  • pyproject.toml
  • src/orcapod/core/executors/ray.py
  • Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🟡 Changes recommended

The unconditional Ray 2.60.0 development dependency makes normal uv sync fail on Intel macOS because no compatible wheels or source distribution are available.

1 open issue
What changed in this PR

Updates Orcapod’s optional Ray execution dependency from an exact pin to a bounded version range, alongside the broader dependency refresh in #426.

Changes:

  • Requires ray[default]>=2.60.0,<3 in the Ray extra and development group.
  • Locks Ray 2.60.0 and updates aiohttp and yarl.
  • Documents local-cluster token authentication and its opt-out setting.
File Description
uv.lock Updates Ray, aiohttp, yarl, and their locked artifacts.
pyproject.toml Replaces both exact Ray pins with the new version range.
changelog.d/​ITL-1210.md Records the dependency change and local authentication behavior.

Review effort: Balanced

Comment thread pyproject.toml
"pytest-xdist>=3.6",
"psycopg[binary]>=3.0",
"ray[default]==2.48.0",
"ray[default]>=2.60.0,<3",
@codecov

codecov Bot commented Nov 17, 2026 •

Copy link
Copy Markdown

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines Coverage Δ
src/orcapod/core/executors/ray.py 66.41% <ø> (ø)

... and 2 files with indirect coverage changes

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

…-1210)

A driver started with uv run from a project whose version comes from
version control hangs on its first RayExecutor call: Ray rebuilds the
project on its workers without .git. The troubleshooting page, the
RayExecutor docstring and the changelog fragment give the workaround,
RAY_ENABLE_UV_RUN_RUNTIME_ENV=0 or starting the driver with python.
ITL-1213 follows up on RayExecutor itself.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🔵 Needs a closer look

Ray 2.60.0 lacks Intel macOS wheels or a source distribution, breaking development installation on those machines because Ray is an unconditional development dependency.

1 open issue

Review effort: Balanced

This branch has not been deployed

No deployments
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.

2 participants