fix(translator): let a manual run lift the debounce it is waiting on

[SearheiParkhamchuk]

Closes #127.

What was wrong

The panel's Run now button did nothing for a job queued with an auto-translate debounce, and then said the job was already in progress.

run() hands the job to Payload's picker, which declines a job that is processing, carries hasError, or has a waitUntil that has not elapsed. Blockers were cleared only for a job that was processing or had failed — a queued one is neither, so its waitUntil survived, the picker took nothing, and the single failure reason the runner has (already_running) went back to the editor.

Two things were wrong with that. The button did not do what it is named. And the explanation was confidently false: the job was not running, it was waiting.

The fix

One line in PayloadJobsTaskRunner.run(): clear the blockers unconditionally. A manual run means now, so it lifts the debounce along with the lock and the spent retry budget.

After this, the case where the picker still declines is one where already_running is true — another job for the same document holds the concurrency key, when the host has enabled Payload's concurrency control.

PayloadJob also gained the waitUntil field. The row always had it; the type did not, which is part of why this was invisible from the runner.

How it was verified

Test first. Against the unchanged code the new case fails with Number of calls: 0 — the blockers were never cleared. With the fix it passes, and putting the old condition back reddens it again.

One existing test asserted expect(mockPayload.update).not.toHaveBeenCalled() for a queued job — it pinned exactly the behaviour this change reverses. Rather than delete the assertion, it now checks that the clear happens before the picker is called, which is what makes lifting the debounce meaningful.

• Unit: 1493
• Integration: three adapters × two queue modes. SQLite and MongoDB 130/130 in each; Postgres 127/130 in each — the three failures are the known auto-translate cases from translator: auto-translate does nothing on Postgres — the hook's work runs outside the triggering transaction #124, red on main as well.
• check-types clean in both packages; lint 0 errors on the changed files.

Scope note

#127 also asked for a distinct RunResult member so the admin could name the reason. That is not added here: after this change the one remaining decline genuinely is "already running", so a new member would have no case to describe. If a future decline reason appears that is neither, it should come with its own member then.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
ideal-cms	Ready	Preview	Sep 13, 2026 11:37am UTC