Daily scheduled workflows missing or delayed by more than 7 hours; manual and push runs succeed #213265
Replies: 3 comments
|
Hi @yrasema-sa — the most informative result in your report is the smoke test: a date-specific cron a few minutes ahead that never appears rules out "the daily expression is wrong" and "the first run needs to register" — a brand-new workflow file with a fresh cron should have fired. And since disable/re-enable didn't help either, the problem is very likely not in the workflow file at all. There is one documented failure mode that matches your symptoms, and it's about the actor, not the schedule: A scheduled workflow can silently stop when its actor's account is no longer active. From GitHub's docs ("Events that trigger workflows"): certain repository events change the How to test it — two checks, no changes needed:
One honest caveat: you mentioned changing the cron in the owner's authenticated web session, producing a verified web-flow commit. If that commit actually changed the What I ruled out: I checked the GitHub status history for your dates. The Nov 13 incident was pull-request merges timing out in the afternoon (first posted 15:47 UTC) — it doesn't cover a 06:17 UTC schedule slot — and Nov 14 had no incidents at all. So no platform incident explains the missing Nov 13 and 14 runs. The Nov 15 run arriving 7h37m late is within the documented best-effort behaviour (the docs say schedule events can be delayed under high load and some queued jobs may be dropped), but load alone doesn't explain consecutive missing days plus smoke tests never appearing. If the actor checks out: the fix is a single commit by an active user with write access that changes the cron value — that reactivates the schedule and makes them the actor. No credential changes, no permission changes, nothing weakened. I'd change the minute by one, let it run once, and only then decide whether to move it back. On getting GitHub to look at it: there is no public API or UI to inspect scheduler registration — GitHub doesn't expose one, so there's no self-serve check beyond the actor and the runs API above. If the actor fix doesn't bring the runs back, the remaining route is GitHub Support; since your account has no technical-support entitlement, this discussion is the channel GitHub points free accounts to, so keeping the evidence here (actor, dates, API results) is the right way to get it in front of staff. Your guarded 3-hour retries are a sensible mitigation in the meantime — they don't depend on any single slot firing. But if the actor is the cause, no retry schedule will help until the actor is refreshed, so I'd check that first. |
|
One way to sharpen the evidence you already collected is to keep three UTC timestamps for each real scheduled run: expected cron time, REST The guarded three-hour retries help with occasional misses, but they still use the same scheduler. If the copy has a hard deadline, a separate heartbeat that alerts when the verification receipt is missing can tell you the job was missed even when Actions has no run or log for that tick. |
|
Use |
Uh oh!
There was an error while loading. Please reload this page.
🏷️ Discussion Type
Question
💬 Feature/Topic Area
Schedule & Cron Jobs
Discussion Details
I am troubleshooting a GitHub.com private repository where scheduled workflows are missing or delayed by hours, while workflow_dispatch and push-triggered workflows work successfully.
The workflow exists on the default branch (main) and is enabled. The daily UTC schedule when these delays were observed was:
Observed timeline:
What we have checked:
I understand GitHub documents that high load may delay or drop schedule events. We have now added guarded retries with
17 */3 * * *UTC; they skip an existing daily copy after checking its verification receipt and metadata, but still depend on the same event delivery. A manual read-only check of that gate succeeds; that does not validate schedule delivery. The internal cause remains unconfirmed.How can we diagnose scheduler registration, the actor associated with a schedule, or an account-level restriction that is not shown in the UI? Is there a supported way to request review of missing schedule events without changing credentials or weakening repository permissions?
We have not established the internal cause and are not treating a successful manual execution as proof that scheduling works.
All reactions