NestJS BullMQ Implementation

Priority: P0 (Critical)

Guidelines

• Set idle polling: Add drainDelay + stalledInterval + maxStalledCount to every @Processor. Default drainDelay (5 ms) burns 570M Redis commands/day at idle. See patterns.md.
• Throttle worker error logs: BullMQ workers emit raw unhandled ReplyErrors on Redis failure (e.g. Upstash rate limits). Always extend BaseProcessor instead of WorkerHost to rate-limit these logs. See patterns.md.
• Set job retention: Add removeOnComplete, removeOnFail, attempts, backoff to every BullModule.registerQueue. See patterns.md.
• Use shared constants: All numeric options live in src/common/constants/bull-queue.constants.ts. Key constants: QUEUE_DRAIN_DELAY_MS (10 000 ms), QUEUE_STALLED_INTERVAL_MS (60 000 ms). Use getSharedBullQueueOptions helper for registerQueue. Queue/job names go in {feature}.constants.ts. Never inline magic numbers.
• Wrap every queue.add(): Persist DB record first, then enqueue inside try-catch. Redis errors must not surface as 500s. See patterns.md.
• Throttler fail-open: ThrottlerGuard is registered as global APP_GUARD — a Redis blip propagates errors to ALL HTTP routes. RedisThrottlerStorage.increment() must catch all Redis errors and return a fail-open pass-through record. A Redis blip must not kill all HTTP routes. See patterns.md.
• Guard new queues: Follow isRedisEnabled() conditional + mock token pattern in every module. NestJS DI throws on startup without mock.
• Keep processor and cron: Cron schedules; processor executes. Both always required — they are complementary. See patterns.md.
• Use local Redis in dev: Never point dev machines at Upstash — idle workers exhaust free tier (500K/day) in minutes.

Anti-Patterns

• No bare @Processor(NAME): Always pass worker options object with drainDelay and stalledInterval.
• No bare WorkerHost extension: Always extend BaseProcessor instead to intercept and rate-limit worker errors.
• No registerQueue without defaultJobOptions: Omitting causes unbounded Redis memory growth.
• No inline numbers: Use bull-queue.constants.ts — never write 10_000, 60_000, 50, 20, 3, or 5_000 directly.
• No unguarded queue.add(): Wrap in try-catch; persist DB state first.
• No throws in throttler increment: Catch Redis errors; return fail-open record.
• No missing mock token: Provide getQueueToken mock when redisEnabled = false.
• No removing processor because cron exists: They serve different roles.
• No cloud Redis in dev: Use local Docker Redis.

References

• All Code Patterns
• Evals