Testing and local dev

Jobs code needs a real database: leases, sweeps, and leader election are all rows. For local development and integration tests, an in-memory MongoDB replica set gives you that in milliseconds, with no Docker and no installed Mongo. This is the same pattern Payload's own test suite uses.

The memory-Mongo pattern

Add mongodb-memory-server as a dev dependency and boot a single-node replica set:

import { MongoMemoryReplSet } from 'mongodb-memory-server'

let replSet: MongoMemoryReplSet | undefined

export const startMemoryMongo = async (): Promise<string> => {
  replSet = await MongoMemoryReplSet.create({ replSet: { count: 1 } })

  const stop = (): void => {
    void replSet?.stop()
    replSet = undefined
  }
  process.once('SIGINT', stop)
  process.once('SIGTERM', stop)
  process.once('exit', stop)

  return replSet.getUri()
}

Use it in your Payload config when no external database is configured:

import { mongooseAdapter } from '@payloadcms/db-mongodb'
import { startMemoryMongo } from './memoryDb'

const url = process.env.DATABASE_URI ?? (await startMemoryMongo())

export default buildConfig({
  db: mongooseAdapter({ url }),
  // ...
})

Two details matter. It must be a replica set, not a standalone: Payload's mongoose adapter enables transactions only when the connection URL carries a replicaSet parameter, which MongoMemoryReplSet.getUri() includes, so a standalone memory server would silently run without transactions and behave differently from production. And stop it on exit: the process.once handlers keep test runs and dev servers from leaking mongod processes.

Exercising the reliability layer in tests

The reliability internals are exported precisely so tests can drive them without a worker process:

import { createWorker, resolveReliabilityOptions, runSweep, getQueueHealth } from '@10x-media/jobs'

const reliability = resolveReliabilityOptions({ jobLeaseTtlMs: 1_000, sweepIntervalMs: 200 })

// A worker without signal handlers, safe to create per test:
const worker = createWorker({ payload, reliability: reliability!, installSignals: false })
worker.start()
// ...
await worker.drain()

Short TTLs make orphan-recovery paths testable in seconds: claim a job, skip the heartbeat, advance past jobLeaseTtlMs, and assert runSweep requeues it. getQueueHealth gives you the same counts as the dashboard for assertions.

Only one worker per process may install signal handlers, so always pass installSignals: false in tests that create several.