Background Jobs Checklist: Field Kit

Schedule and dispatch

List every job that must run without a user request: reports, reminders, syncsWork nobody triggers is the work that gets forgotten until a customer notices it never happened.
Declare each schedule as a crons entry in vercel.json pointing at a routeWithout a native schedule, recurring work depends on someone or something remembering to start it.
Reject any cron request that lacks the CRON_SECRET bearer tokenThe cron path is a public URL, so anyone who finds it can trigger the job at will.
Keep the cron handler a dispatcher that finds work and hands each piece offA handler that does the heavy work inline hits the function time limit and dies halfway.
Break any long job into many small units, each well inside the time limitOne ten-minute run on a short-lived function gets killed partway, leaving work half done.

Run safely

Move heavy or must-not-fail work to a queue with automatic retriesInline work that fails is simply lost, while a queued message is redelivered until it succeeds.
Take a lock with a TTL before each run, and skip the tick if it is heldOverlapping runs process the same records twice, sending duplicate emails or double charges.
Release the lock in a finally block and rely on the TTL if a run diesA lock that never releases blocks every future run, and the job stops without a single error.
Make writes idempotent with stable ids and upsertsRetries are normal, and without idempotency each retry duplicates what the last attempt did.
Process large jobs in batches and save a checkpoint after each oneWithout a saved cursor, an interrupted job restarts from zero or doubles up on finished work.

Observe

Log each run's start, end, duration and outcomeBackground work is invisible, and without a log you are guessing when something goes wrong.
Alert when a job fails repeatedly or its backlog keeps growingSilent failures accumulate until the damage shows up in customer complaints or missing revenue.
Add a heartbeat check, such as no nightly success in 25 hoursA job that stops running throws no error, so only a missing success can reveal it.
Route permanently failing work to a dead-letter store for reviewWork that retries forever wastes capacity, and work that is dropped is lost for good.

How to use this

Use it once per job, top to bottom. A good result: every job has a guarded schedule, runs in short idempotent units behind a lock, and would alert you within a day if it failed or simply stopped running.

The rest of the kit is one step away.

Leave your email and the full kit opens right here, ready to work through or save as a PDF.

Kept private. Never shared or sold.