Run queues across a cluster

Use a shared Postgres authority for queues on more than one server. The Postgres-backed pgliteCollections backend serializes conditional claims, so configured concurrency is cluster-wide rather than per process.

const collections = await pgliteCollections({
  pgUrl: process.env.PG_URL!, collections: app.collections ?? {}, tablePrefix: 'app_',
})
createSuperLineServer(app, {
  nodeName: process.env.NODE_NAME, nodeKey: process.env.NODE_KEY!,
  collections, plugins: [queueKit.plugin],
})

Every node runs the same queue definition. For concurrency two, the plugin creates two durable slots; all nodes compete for those slots. Keep nodeKey stable for a replica across restarts. A super-line adapter is optional for job correctness, but useful for Control Center, topology, and low-latency wake and cancellation hints. Durable polling and Postgres conditional batches remain the correctness path.

On server.close(), the plugin stops claiming and waits for shutdownGraceMs. It then aborts remaining workers. After a crash, an expired lease makes the job available again; a fencing runId prevents an old worker settling a reclaimed run. Make every external effect idempotent.

See queue-cluster for Docker and direct-host two-node operation with libp2p, Caddy, a browser dashboard, and Control Center.