@absolutejs/queue-postgres
v0.1.8
Published
Postgres (Drizzle) storage adapter for @absolutejs/queue — postgres.js + Neon serverless
Downloads
2,333
Maintainers
Readme
@absolutejs/queue-postgres
Postgres storage adapter for @absolutejs/queue, built on Drizzle. The
production JobStore: durable, with atomic multi-worker claiming via
FOR UPDATE SKIP LOCKED. Ships convenience factories for both postgres.js
and Neon's WebSocket driver (@neondatabase/serverless); the underlying
buildPostgresJobStore accepts any Drizzle Postgres database, so other
drivers (including Bun SQL and node-postgres) work too. Its portable JSONB
encoding preserves payload objects across all supported drivers.
Install
# pick your driver
bun add @absolutejs/queue @absolutejs/queue-postgres drizzle-orm postgres
# …or
bun add @absolutejs/queue @absolutejs/queue-postgres drizzle-orm @neondatabase/serverlesspostgres and @neondatabase/serverless are optional peer deps —
install whichever one you'll use.
Usage with postgres.js
import { createPostgresJobStore } from '@absolutejs/queue-postgres/postgres';
import { createJobRegistry, defineJobs, queue, t } from '@absolutejs/queue';
import postgres from 'postgres';
const jobs = defineJobs({
'email.send': t.Object({ to: t.String(), subject: t.String() })
});
const registry = createJobRegistry(jobs).on(
'email.send',
async ({ to, subject }) => {}
);
// Share your app's existing postgres.js client (one pool)…
const client = postgres(process.env.DATABASE_URL, { prepare: false });
const store = createPostgresJobStore({ client, jobs });
// …or let the adapter open its own connection:
// const store = createPostgresJobStore({ connectionString: url, jobs });
app.use(queue({ registry, store }));Usage with Neon (@neondatabase/serverless)
createNeonJobStore mirrors createPostgresJobStore but uses Neon's
WebSocket Pool. Important: the queue's claimDue opens a transaction
and selects with FOR UPDATE SKIP LOCKED. Neon's HTTP driver
(drizzle-orm/neon-http) is single-statement and can't do row-level
locks — use the WebSocket Pool driver here. Your app's other code can
keep using the HTTP driver; they're independent.
import {
createNeonJobStore,
neonConfig
} from '@absolutejs/queue-postgres/neon';
import { createJobRegistry, defineJobs, queue, t } from '@absolutejs/queue';
import { Pool } from '@neondatabase/serverless';
// Bun ships a global WebSocket; for node, polyfill once:
// import ws from 'ws'; neonConfig.webSocketConstructor = ws;
const jobs = defineJobs({
'email.send': t.Object({ to: t.String(), subject: t.String() })
});
// Share an existing Neon Pool…
const pool = new Pool({ connectionString: process.env.DATABASE_URL });
const store = createNeonJobStore({ jobs, pool });
// …or let the adapter open its own:
// const store = createNeonJobStore({ connectionString: url, jobs });Usage with any other Drizzle Postgres adapter
Both factories are thin wrappers around buildPostgresJobStore(db, jobs),
which accepts any PgDatabase<any, any, any> from Drizzle. If you use
drizzle-orm/node-postgres, build the db yourself and pass it in:
import { buildPostgresJobStore, queueSchema } from '@absolutejs/queue-postgres/drizzle';
import { drizzle } from 'drizzle-orm/node-postgres';
import { Pool } from 'pg';
const pool = new Pool({ connectionString: process.env.DATABASE_URL });
const db = drizzle(pool, { schema: queueSchema });
const store = buildPostgresJobStore(db, jobs);Migrations
Add the table to your Drizzle schema so it's included in migrations:
export { queueJobsTable } from '@absolutejs/queue-postgres/drizzle';Then drizzle-kit generate / push as usual.
How claiming works
claimDue runs inside a transaction:
SELECT … FROM queue_jobs
WHERE status = 'pending' AND run_at <= $now
ORDER BY run_at LIMIT $n
FOR UPDATE SKIP LOCKED; -- concurrent workers skip locked rows
-- then UPDATE those ids → status='claimed'This guarantees a job is handed to exactly one worker even with many workers polling.
Crashed workers are recovered by reapStuck (lease expiry → back to pending).
License
Apache-2.0 — see LICENSE.
Atomic claim fencing
The adapter now assigns claim_token with each claim and implements
completeClaim / failClaim as conditional updates on job ID, token and
status = 'claimed'. An expired, canceled or superseded attempt returns false
without changing the current job. Updated queue workers use this capability
automatically; requireClaimFencing: true refuses an incompatible store.
Apply this additive migration before starting the updated adapter (or generate
it from the exported queueJobsTable with your application's migration tool):
ALTER TABLE queue_jobs ADD COLUMN IF NOT EXISTS claim_token text;Drain older workers during rollout. Older binaries still perform unfenced updates and cannot provide this guarantee. Claim fencing protects the queue row; handlers must separately fence application writes and external side effects.
Lease expiry consumes one attempt. Reaping moves an exhausted job to dead
with a lease-expiry error, so repeated worker death cannot bypass maxAttempts.
The /drizzle entry point (0.1.8+) does not load the optional postgres or Neon connection factories. Use it with Bun SQL or an existing Drizzle connection, including schema generation. Existing root exports remain compatible.
Mixed worker versions
With queue 0.8.4, enable requireKindFiltering: true on workers. Claims are
restricted to both the adapter's job definition and the worker's registered
handlers before the database applies its claim limit. Empty handler sets claim
nothing. Upgrade every worker attached to a shared table before producing new
job kinds; an older, unfiltered worker can still claim and reject those jobs.
