Trigger the rebase on merge, because the schedule never reaches a runner #1083

Merged
jaylouisw merged 1 commit from ci/auto-rebase-on-merge into main 2026-11-08 15:48:58 +00:00
Owner

#1071 was wrong, and the data now says so.

It found a real defect — a scheduled run entering a cancel-in-progress: false group and blocking every later tick — and fixing that was correct. But the scheduled runs still don't start:

692 | schedule          | waiting | created 15:30:41 | started 1970-01-01 00:00:00
693 | workflow_dispatch | waiting | created 15:32:04 | started 1970-01-01 00:00:00

Both have their action_run_job rows. Neither has ever been handed to a runner, while pull-request and push runs are dispatched normally on the same instance, same runner, same labels. That is a dispatcher problem on this Forgejo build, not a concurrency one, and it is not fixable from this repository — which means auto-rebase has never run on schedule and never will until that is fixed upstream or on the host.

push is a trigger that demonstrably works, and it is the better trigger regardless: when main changes, every open pull request is stale by definition. That is exactly the moment the queue needs rebasing, and it needs no cron to notice.

The schedule stays as a safety net for a quiet main, and workflow_dispatch keeps manual runs. Neither needs work: the script already defaults batch to 5 and settle to 120 when the inputs are absent, and validates them numeric.

Combined with cancel-in-progress: true (#1071), a burst of merges leaves one rebase run standing rather than a queue of them.

Once this lands, the queue rebases itself on every merge — which is what this workflow was supposed to be doing since #409.

#1071 was wrong, and the data now says so. It found a real defect — a scheduled run entering a `cancel-in-progress: false` group and blocking every later tick — and fixing that was correct. But the scheduled runs still don't start: ``` 692 | schedule | waiting | created 15:30:41 | started 1970-01-01 00:00:00 693 | workflow_dispatch | waiting | created 15:32:04 | started 1970-01-01 00:00:00 ``` Both have their `action_run_job` rows. Neither has ever been handed to a runner, while pull-request and push runs are dispatched normally on the same instance, same runner, same labels. That is a dispatcher problem on this Forgejo build, not a concurrency one, and it is not fixable from this repository — which means `auto-rebase` has never run on schedule and never will until that is fixed upstream or on the host. `push` is a trigger that demonstrably works, and it is the better trigger regardless: **when `main` changes, every open pull request is stale by definition.** That is exactly the moment the queue needs rebasing, and it needs no cron to notice. The schedule stays as a safety net for a quiet `main`, and `workflow_dispatch` keeps manual runs. Neither needs work: the script already defaults `batch` to 5 and `settle` to 120 when the inputs are absent, and validates them numeric. Combined with `cancel-in-progress: true` (#1071), a burst of merges leaves one rebase run standing rather than a queue of them. Once this lands, the queue rebases itself on every merge — which is what this workflow was supposed to be doing since #409.
Trigger the rebase on merge, because the schedule never reaches a runner
All checks were successful
CI / Test (Node 18) (pull_request) Successful in 1m43s
CI / Test (Node 20) (pull_request) Successful in 1m35s
CI / Server smoke test (pull_request) Successful in 28s
CI / Docker build (pull_request) Successful in 10s
CI / Test (Node 22) (pull_request) Successful in 1m26s
b96523f84f
#1071 was wrong. It found a real deadlock - a scheduled run entering a
cancel-in-progress:false group and blocking every later tick - and fixed it, but the
scheduled runs still do not start. The database after that merge:

  692 | schedule          | waiting | created 15:30:41 | started 1970-01-01
  693 | workflow_dispatch | waiting | created 15:32:04 | started 1970-01-01

Both have their jobs and neither has ever been handed to a runner, while
pull-request and push runs are dispatched normally on the same instance. That is a
dispatcher problem, not a concurrency one, and it is not fixable from this
repository.

Push is a trigger that works, and it is the better trigger anyway: when main
changes, every open pull request is stale by definition. That is precisely the
moment the queue needs rebasing, and it needs no cron to notice.

The schedule stays as a safety net for a quiet main; the dispatch input keeps
manual runs working, and both already default correctly when absent (batch=5,
settle=120), so a push run needs no inputs.
jaylouisw scheduled this pull request to auto merge when all checks succeed 2026-11-08 15:34:57 +00:00
Sign in to join this conversation.
No reviewers
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!1083
No description provided.