sdf-jkl opened a new issue, #11456: URL: https://github.com/apache/arrow-rs/issues/11456
### Is your feature request related to a problem or challenge? Make CI go :rocket: DataFusion has already moved several CI jobs to RunsOn in [apache/datafusion#20107](https://github.com/apache/datafusion/pull/20107) and [apache/datafusion#20274](https://github.com/apache/datafusion/pull/20274). A sample of recent arrow-rs CI runs gives us a baseline: | CI stage | Successful first-attempt runs sampled | Median elapsed time | Middle 50% of runs | | --- | ---: | ---: | ---: | | PR checks | 26 | 16.5 min | 11–29 min | | Merge-queue checks | 30 | 24.8 min | 24–36 min | The sample covers October 6–8, 2026. Times are measured from workflow creation to the final job's completion, including waiting. Failed, cancelled, and rerun workflows are excluded; PR coverage varies with the files changed. Example runs: [PR checks](https://github.com/apache/arrow-rs/actions/runs/37692867787) and [merge-queue checks](https://github.com/apache/arrow-rs/actions/runs/37556929578). The longest-running Miri shard in each merge-queue run took about 24 minutes at the median. Waiting can also dominate: [one recent PR run](https://github.com/apache/arrow-rs/actions/runs/37790159316) took about 80 minutes overall, while its longest job executed for 19 minutes. RunsOn could help with both runner availability and execution time. ### Describe the solution you'd like Start with one or two expensive Linux build/test jobs (Miri :smiley:), measure the results, and expand if the change is worthwhile. - Use Arrow's existing RunsOn infrastructure, choosing CPU, memory, and disk sizes appropriate for Rust builds. - Keep a fallback to GitHub-hosted runners for forks, following DataFusion's approach. - Compare queue time, build/test duration, cache effectiveness, and AWS resource usage before migrating more jobs. Repository permissions appear to be in place already. In [this comment on apache/arrow-nanoarrow#814](https://github.com/apache/arrow-nanoarrow/issues/814#issuecomment-3718550233), @raulcd confirmed on January 7, 2026: > Gavin has confirmed that `arrow` and `arrow-rs` had permissions initially. This gives us an existing setup to build on. We should coordinate runner sizing and expected usage with the maintainers of that infrastructure. cc @blaginin ### Additional context - [INFRA-27359: Enable use of RunsOn for ephemeral AWS runners](https://issues.apache.org/jira/browse/INFRA-27359) — completed setup for Arrow, including Infra's acceptance of the app's required permissions. - [ASF's approved GitHub Actions list](https://github.com/apache/infrastructure-actions/blob/main/actions.yml) — already includes `runs-on/action`. - [DataFusion's initial migration](https://github.com/apache/datafusion/pull/20107) and [follow-up migration](https://github.com/apache/datafusion/pull/20274). - [Arrow C++ large-memory tests](https://github.com/apache/arrow/blob/main/.github/workflows/cpp_extra.yml) — existing CPU use of RunsOn within Arrow. - [nanoarrow device tests](https://github.com/apache/arrow-nanoarrow/blob/main/.github/workflows/build-and-test-device.yaml) — existing CUDA use in another Arrow repository. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
