Cloud Run job delay up to 12 hours: schedule Sume bulk runs

Use a delayed Cloud Run job when you want a one-off batch of Sume work to start later, such as an overnight render of up to 100 Format items, without standing up a scheduler. Google's Cloud Run release notes for September 8, 2026 describe job execution delay of up to 12 hours, marked Preview. For a repeating schedule, use Sume Actions or a scheduler instead.

What is new and what is not

Cloud Run also reached GA on September 29 for scaling controls with a custom target CPU or concurrency, which matters for services, not jobs.

Shape of the job

Keep the container small. It reads a work list, calls POST /v1/formats/{handle}/{slug}/bulk-runs with a stable Idempotency-Key, stores the queue id, and exits. Bulk runs accept up to 100 items and a concurrency value, return 202, and are read at GET /v1/format-run-queues/{id}.

The queue does the waiting, not your container. That is what makes a delayed job cheap: it runs for seconds, then Sume works through the items within your plan's concurrency.

Rules that avoid double spend

A replayed key on bulk runs returns the old queue. That is your protection if Cloud Run retries the job task. Build the key from the work list identity, not a timestamp.

• Do not poll for hours from the job; collect results with webhooks.
• The queue has no webhook of its own; each item can carry one, as format.run.terminal.
• Terminal run webhooks carry status OK or ERROR, an outcome of ok, degraded or error, and a receipt payload.
• Canceled and skipped runs deliver no webhook, so reconcile against the queue read.

A note on Preview

Preview features can change. Treat the delay setting as a convenience and keep a fallback: if the delay is withdrawn, the same job can be started by Cloud Scheduler. Because the Sume side is idempotent, switching the trigger does not change the work.