CI: a pull request head pushed by the rebase bot gets no workflow run at all (0 statuses, permanently) #2082

Closed
opened 2026-11-19 03:52:16 +00:00 by jaylouisw · 0 comments
Owner

Symptom. A pull request reaches 0 check statuses and stays there. Re-pushing does not cure it; a force-push that the rebase bot then rebuilds makes it worse.

Evidence (2026-11-19, three PRs, one symptom).

  • #2074 and #2075 opened at 02:54/02:56 with 0 statuses 20 min later. I amended and force-pushed both (7159f06c, d0db1f7f); the Rebase onto main and arm auto-merge job immediately rebuilt them onto then-main (882f98ec, 740a4235) and both then reported no runs at all for the new head, 8 min and then 40 min later.
  • #2049's last runs are for head eeeccb21 (eight ci.yml jobs, 2026-11-18 23:10-23:11Z). Its head then became d5e15d71 (committer packetforge-bot) and no run has been created for it since - 4 h+.
  • GET /commits/<sha>/status returns state: "", statuses: [] for those heads; /actions/tasks has no row for that SHA at all. So the run is never created, not created-and-cancelled.

The dividing line is who created the head. Every stuck head has packetforge-bot <packetforge-bot@users.noreply.git.jayw.uk> as committer - i.e. it was produced by the rebase bot's force-push. Heads pushed by the authoring loop (a90ea6d7, 6414f879, e641b356, and the amend I pushed at 7159f06c) get runs normally.

The cure, confirmed twice today. Rebase the branch onto current main so it is 0 commits behind, push it yourself (committer != packetforge-bot), and a full 8-check run appears within ~100 s:

  • #2074: 882f98ec -> 460c9281, all 8 green in 100 s, merged as 16a7ded7.
  • #2075: 740a4235 -> 18838f99, all 8 green, merged as ceb55abc.

Pushing 0-behind matters: a stale head re-triggers the rebase bot, whose push is exactly the head class that gets no run, so the PR bounces back into the dead state (#2074/#2075 did this at 03:13).

What I cannot see from here. Whether the skip is act_runner's synchronize handling, a duplicate-run dedupe keyed on the pusher, or the workflow event fan-out dropping the second pull_request event of the same SHAs. ssh odin (for runner logs and disk headroom) has refused for 9 consecutive runs, so runner-side logs are unreachable from this container.

Ask. Look at act_runner / Forgejo's pull_request event fan-out for pushes whose pusher is packetforge-bot. Until that is fixed, every rebase the bot performs silently removes the PR from CI - and a PR with no status can never be auto-merged, so the queue fills with green-looking branches nobody can verify.

**Symptom.** A pull request reaches 0 check statuses and stays there. Re-pushing does not cure it; a force-push that the rebase bot then rebuilds makes it worse. **Evidence (2026-11-19, three PRs, one symptom).** - #2074 and #2075 opened at 02:54/02:56 with 0 statuses 20 min later. I amended and force-pushed both (7159f06c, d0db1f7f); the *Rebase onto main and arm auto-merge* job immediately rebuilt them onto then-main (`882f98ec`, `740a4235`) and both then reported **no runs at all** for the new head, 8 min and then 40 min later. - #2049's last runs are for head `eeeccb21` (eight `ci.yml` jobs, 2026-11-18 23:10-23:11Z). Its head then became `d5e15d71` (committer `packetforge-bot`) and no run has been created for it since - 4 h+. - `GET /commits/<sha>/status` returns `state: ""`, `statuses: []` for those heads; `/actions/tasks` has no row for that SHA at all. So the run is never created, not created-and-cancelled. **The dividing line is who created the head.** Every stuck head has `packetforge-bot <packetforge-bot@users.noreply.git.jayw.uk>` as *committer* - i.e. it was produced by the rebase bot's force-push. Heads pushed by the authoring loop (a90ea6d7, 6414f879, e641b356, and the amend I pushed at 7159f06c) get runs normally. **The cure, confirmed twice today.** Rebase the branch onto *current* `main` so it is 0 commits behind, push it yourself (committer != packetforge-bot), and a full 8-check run appears within ~100 s: - #2074: 882f98ec -> `460c9281`, all 8 green in 100 s, merged as `16a7ded7`. - #2075: 740a4235 -> `18838f99`, all 8 green, merged as `ceb55abc`. Pushing 0-behind matters: a stale head re-triggers the rebase bot, whose push is exactly the head class that gets no run, so the PR bounces back into the dead state (#2074/#2075 did this at 03:13). **What I cannot see from here.** Whether the skip is act_runner's `synchronize` handling, a duplicate-run dedupe keyed on the pusher, or the workflow event fan-out dropping the second `pull_request` event of the same SHAs. `ssh odin` (for runner logs and disk headroom) has refused for 9 consecutive runs, so runner-side logs are unreachable from this container. **Ask.** Look at act_runner / Forgejo's pull_request event fan-out for pushes whose pusher is `packetforge-bot`. Until that is fixed, every rebase the bot performs silently removes the PR from CI - and a PR with no status can never be auto-merged, so the queue fills with green-looking branches nobody can verify.
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
jaylouisw/packetforge#2082
No description provided.