Run fork PR test suites in contributor forks by sdf-jkl · Pull Request #11788 · apache/arrow-rs

sdf-jkl · 2026-10-21T22:42:27Z

Which issue does this PR close?

No linked Arrow issue. Adapts the approach in apache/datafusion#25591.

Rationale for this change

Fork pull requests currently run their test suites upstream, consuming ASF's shared CI capacity before the merge queue validates the merged result. This change moves that initial test feedback to contributors' forks while retaining full upstream validation in the merge queue.

What changes are included in this PR?

Run all regular CI suites, excluding Miri, on fork branch pushes and manual runs. This includes documentation-only pushes. Superseded pushes to the same fork branch cancel earlier runs.
Keep lightweight checks upstream for fork PRs to main. Same-repository and release-branch PRs retain existing upstream testing; the merge queue still runs every suite, including Miri.
Show an informational CI (fork) check linking to the fork run for the current commit. A status bridge uses workflow_run and scheduled polling to discover results and reruns. It reads fork run metadata and writes upstream checks without checking out PR code or consuming PR artifacts.
Keep docs deployment upstream-only and add contributor setup instructions and CI configuration checks.

Fork results are informational because contributors control their workflows. They do not replace the required merge-queue checks.

Are these changes tested?

Routing and configuration tests cover fork pushes, PR targets, manual runs, and merge-queue validation. Bridge tests use mocked GitHub APIs to cover missing runs, reruns, stale commits, API failures, polling, and cancellation. All 8 routing/configuration tests and 29 bridge tests passed locally, along with workflow linting, spelling, and Markdown formatting checks.

Hosted end-to-end validation remains outstanding. The bridge must be installed on a test repository's default branch to exercise workflow_run and scheduled updates.

Are there any significant user-facing changes?

Contributors must enable Actions in their forks and update their branches to include these workflows. Results and reruns are available in the fork's Actions tab; the upstream status check may update later. No library API changes.

sdf-jkl · 2026-10-21T22:44:20Z

The CI runs on my fork https://github.com/sdf-jkl/arrow-rs/actions/runs/39688408876 🎉

sdf-jkl · 2026-10-21T22:45:59Z

@kumarUjjawal @Jefffrey @neilconway @comphead PTAL 😎

alamb · 2026-10-22T00:02:56Z

Are we sure this process is working well in DataFusion? I noticed a bunch fo the Ci today was broken / red. I haven't had time to investigate but maybe we can wait to see how it goes over there before merging this in

alamb · 2026-10-22T00:03:03Z

Thank you for pushing this along @sdf-jkl

neilconway · 2026-10-22T00:08:47Z

Are we sure this process is working well in DataFusion? I noticed a bunch fo the Ci today was broken / red. I haven't had time to investigate but maybe we can wait to see how it goes over there before merging this in

CI failures were due to an unrelated issue (security audit), I believe.

alamb · 2026-10-22T14:57:32Z

Are we sure this process is working well in DataFusion? I noticed a bunch fo the Ci today was broken / red. I haven't had time to investigate but maybe we can wait to see how it goes over there before merging this in

CI failures were due to an unrelated issue (security audit), I believe.

thanks -- I have been somewhat overwhelmed trying to get this arrow release out I haven't had much time for DataFusion recently or seen the effects of the merge queue improements

kumarUjjawal · 2026-10-22T15:37:48Z

+                for (const pr of page.data) {
+                  if (new Date(pr.updated_at).getTime() < cutoff) { stale = true; break; }
+                  seen.add(pr.number);
+                  await syncCheck(pr, sweepNumbers.has(pr.number));


We can catch errors per PR, continue the sweep, then fail the job after reporting the errors.

kumarUjjawal · 2026-10-22T15:38:46Z

+              pr.head.repo.full_name.toLowerCase() !== repository.toLowerCase() && pr.base.ref === 'main';
+            const externalId = (pr) => 'arrow-rs-fork-ci:' + pr.number;
+            const conclusions = new Set([
+              'action_required', 'cancelled', 'failure', 'neutral', 'skipped', 'stale', 'success', 'timed_out',


this allows stale through to check creation/update, although GitHub reserves that conclusion for itself. Map it to failure, as already done for startup_failure, so encountering it does not break the bridge.

comphead · 2026-10-22T17:12:20Z

I would prob let DF to run for 2-3 weeks to fix the issues and then we can port the changes to other repos.

sdf-jkl · 2026-10-22T18:57:21Z

I would prob let DF to run for 2-3 weeks to fix the issues and then we can port the changes to other repos.

Given the current state of CI I don't think we can wait 😢

alamb · 2026-10-22T19:45:14Z

I did some poking around in DataFusion I think that the CI is working nicely there. Let's try it here too

alamb

Thank you @sdf-jkl -- this makes sense to me and I think shifts the accounting to where it belongs (the submitter). I agree that we should give this a try

alamb · 2026-10-22T19:47:05Z

 -->

-CI runs a subset of checks on pull requests, selected by the changed files.
+CI runs lightweight checks upstream for fork pull requests to `main`. Fork


I think this documentation should go

In the contributor guide so it is clear what is happening

The description of what each job does (e.g. ``fork_ci_status.ymlmirrors fork results in an informationalCI (fork)` check,`) I think should be comments in the scripts themselves along with the commands to test

alamb · 2026-10-22T19:50:13Z

+3. Push your branch, or manually run `CI` from your fork's **Actions** tab.
+   View results and rerun failed jobs there; the PR's `CI (fork)` check may lag.
+
+Fork results are informational. The merge queue tests the proposed merge with


Maybe we could phrase this like "When your PR is approved and added to the merge queue, the final CI checks are run using the ASF runners"

Co-authored-by: Andrew Lamb <andrew@nerdnetworks.org>

kumarUjjawal

Looks good 🚀

comphead

it looks good to me, but 2 cents here is:
can we make the flow and scripts/yml be DF alike, ideally copy-paste, it would make much easier to port fixes.

In Apache Spark they follow the same rules/strategy for user forks across repos, and it is super easy to adopt user forks on a new repo whenever its created

sdf-jkl · 2026-10-23T15:43:46Z

can we make the flow and scripts/yml be DF alike, ideally copy-paste, it would makes much easier to port fixes.

I think it's a good idea. I'll do that now

sdf-jkl · 2026-10-23T19:18:30Z

added the @kumarUjjawal's and @alamb's suggestions above.

can we make the flow and scripts/yml be DF alike, ideally copy-paste, it would makes much easier to port fixes.

I think it's a good idea. I'll do that now

@comphead On a second thought, fully copy and pasting adds more churn and would require another round of review, so I'd defer the unification

comphead · 2026-10-23T21:08:40Z

I would still think we need to follow DF practices, especially that pull_request_target is restricted after Dec 2026, which makes this approach already obsolete.

we need to follow https://issues.apache.org/jira/browse/SPARK-60440

its fine to have this PR for now, but we later we need to adopt approach based on workflow_run

comphead · 2026-10-23T21:26:31Z

DF migration example apache/datafusion#26683

sdf-jkl · 2026-10-23T21:33:47Z

its fine to have this PR for now, but we later we need to adopt approach based on workflow_run

The changes in this PR already use workflow_run. I didn't touch the old usage of pull_request_target

DF migration example apache/datafusion#26683

🙏 I will file a follow up based on this later. I think the goal should is to make the CI usable again first.

alamb · 2026-10-23T23:21:20Z

oh boy -here we go!

Jefffrey · 2026-10-24T07:20:46Z

sorry i wasnt able to keep up with this, thanks for adding it in

one question, what impact did this have on the merge queue? assuming it was this PR, i noticed PRs are now speeding through the merge queue. are the checks not run in merge queue anymore?

fix(arrow-select): include column name in BatchCoalescer errors #11870 this was added to merge queue then merged to main in 1 minute
same for these chore(deps): bump black from 26.3.1 to 26.10.1 in /parquet/pytest #11847 chore(deps): bump the all-other-cargo-deps group across 1 directory with 4 updates #11863

Jefffrey · 2026-10-24T07:32:00Z

i clicked merge when ready for this PR #11745 and it went straight to the merge queue (and then merged) without waiting for all the checks to complete 🤔

fyi @Rich-T-kid @etseidl if you plan to merge PRs, things might be funky so probably need to babysit the PRs to ensure they actuallly run their CI first

sdf-jkl · 2026-10-24T11:54:39Z

I thought it was weird when the 60.1 release merged on the mq so fast. I'll triage

Run fork PR test suites in contributor forks

91e1d89

github-actions Bot added the development-process Internal changes; PRs with this label are excluded from changelog label Oct 21, 2026

sdf-jkl marked this pull request as ready for review October 21, 2026 22:48

Merge branch 'main' into codex/fork-pr-ci

e364baa

kumarUjjawal reviewed Oct 22, 2026

View reviewed changes

alamb approved these changes Oct 22, 2026

View reviewed changes

Update CONTRIBUTING.md

ee6f0b8

Co-authored-by: Andrew Lamb <andrew@nerdnetworks.org>

kumarUjjawal approved these changes Oct 23, 2026

View reviewed changes

sdf-jkl added 2 commits October 23, 2026 10:32

Merge upstream main into codex/fork-pr-ci

47eb26e

Address fork CI review feedback

db1c466

comphead reviewed Oct 23, 2026

View reviewed changes

sdf-jkl added 2 commits October 23, 2026 13:36

Fix fork CI rerun cancellation and redundant check updates

9f49725

Merge reviewed contributor documentation

dc8c716

sdf-jkl enabled auto-merge October 23, 2026 19:23

alamb mentioned this pull request Oct 23, 2026

CI takes hours to start apache/datafusion#25088

Open

sdf-jkl added this pull request to the merge queue Oct 23, 2026

github-merge-queue Bot removed this pull request from the merge queue due to a conflict with the base branch Oct 23, 2026

Merge branch 'main' into codex/fork-pr-ci

62db833

sdf-jkl enabled auto-merge October 23, 2026 23:14

sdf-jkl added this pull request to the merge queue Oct 23, 2026

Merged via the queue into apache:main with commit a1eb450 Oct 23, 2026
20 checks passed

sdf-jkl deleted the codex/fork-pr-ci branch October 24, 2026 04:21

Conversation

sdf-jkl commented Oct 21, 2026

Which issue does this PR close?

Rationale for this change

What changes are included in this PR?

Are these changes tested?

Are there any significant user-facing changes?

Uh oh!

sdf-jkl commented Oct 21, 2026

Uh oh!

sdf-jkl commented Oct 21, 2026

Uh oh!

alamb commented Oct 22, 2026

Uh oh!

alamb commented Oct 22, 2026

Uh oh!

neilconway commented Oct 22, 2026

Uh oh!

alamb commented Oct 22, 2026

Uh oh!

kumarUjjawal Oct 22, 2026

Choose a reason for hiding this comment

Uh oh!

kumarUjjawal Oct 22, 2026

Choose a reason for hiding this comment

Uh oh!

comphead commented Oct 22, 2026

Uh oh!

sdf-jkl commented Oct 22, 2026

Uh oh!

alamb commented Oct 22, 2026

Uh oh!

alamb left a comment

Choose a reason for hiding this comment

Uh oh!

alamb Oct 22, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

alamb Oct 22, 2026

Choose a reason for hiding this comment

Uh oh!

kumarUjjawal left a comment

Choose a reason for hiding this comment

Uh oh!

comphead left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

sdf-jkl commented Oct 23, 2026

Uh oh!

sdf-jkl commented Oct 23, 2026

Uh oh!

comphead commented Oct 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

comphead commented Oct 23, 2026

Uh oh!

sdf-jkl commented Oct 23, 2026

Uh oh!

Uh oh!

alamb commented Oct 23, 2026

Uh oh!

Jefffrey commented Oct 24, 2026

Uh oh!

Jefffrey commented Oct 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

sdf-jkl commented Oct 24, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

comphead left a comment •

edited

Loading

comphead commented Oct 23, 2026 •

edited

Loading

Jefffrey commented Oct 24, 2026 •

edited

Loading