background-job-orchestrator（后台作业协调器）

Background Job Orchestrator

Expert in designing and implementing production-grade background job systems that handle long-running tasks without blocking API responses.

When to Use

✅ Use for:

• Long-running tasks (email sends, report generation, image processing)
• Batch operations (bulk imports, exports, data migrations)
• Scheduled tasks (daily digests, cleanup jobs, recurring reports)
• Tasks requiring retry logic (external API calls, flaky operations)
• Priority-based processing (premium users first, critical alerts)
• Rate-limited operations (API quotas, third-party service limits)

❌ NOT for:

• Real-time bidirectional communication (use WebSockets)
• Sub-second latency requirements (use in-memory caching)
• Simple delays (setTimeout is fine for <5 seconds)
• Synchronous API responses (keep logic in request handler)

Quick Decision Tree

Does this task:
├── Take >5 seconds? → Background job
├── Need to retry on failure? → Background job
├── Run on a schedule? → Background job (cron pattern)
├── Block user interaction? → Background job
├── Process in batches? → Background job
└── Return immediately? → Keep synchronous

Technology Selection

Node.js: BullMQ (Recommended 2024+)

When to use:

• TypeScript project
• Redis already in stack
• Need advanced features (rate limiting, priorities, repeatable jobs)

Why BullMQ over Bull:

• Bull (v3) → BullMQ (v4+): Complete rewrite in TypeScript
• Better Redis connection handling
• Improved concurrency and performance
• Active maintenance (Bull is in maintenance mode)

Python: Celery

When to use:

• Python/Django project
• Need distributed task execution
• Complex workflows (chains, groups, chords)

Alternatives:

• RQ (Redis Queue): Simpler, fewer features
• Dramatiq: Modern, less ecosystem
• Huey: Lightweight, good for small projects

Cloud-Native: AWS SQS, Google Cloud Tasks

When to use:

• Serverless architecture
• Don't want to manage Redis/RabbitMQ
• Need guaranteed delivery and dead-letter queues

Common Anti-Patterns

Anti-Pattern 1: No Dead Letter Queue

Novice thinking: "Retry 3 times, then fail silently"

Problem: Failed jobs disappear with no visibility or recovery path.

Correct approach:

// BullMQ with dead letter queue
const queue = new Queue('email-queue', {
  connection: redis,
  defaultJobOptions: {
    attempts: 3,
    backoff: {
      type: 'exponential',
      delay: 2000
    },
    removeOnComplete: 100, // Keep last 100 successful
    removeOnFail: false     // Keep all failed for inspection
  }
});

// Monitor failed jobs
const failedJobs = await queue.getFailed();

Timeline:

• Pre-2020: Retry and forget
• 2020+: Dead letter queues standard
• 2024+: Observability for job failures required

Anti-Pattern 2: Synchronous Job Processing

Symptom: API endpoint waits for job completion

Problem:

// ❌ WRONG - Blocks API response
app.post('/send-email', async (req, res) => {
  await sendEmail(req.body.to, req.body.subject);
  res.json({ success: true });
});

Why wrong: Timeout, poor UX, wastes server resources

Correct approach:

// ✅ RIGHT - Queue and return immediately
app.post('/send-email', async (req, res) => {
  const job = await emailQueue.add('send', {
    to: req.body.to,
    subject: req.body.subject
  });

  res.json({
    success: true,
    jobId: job.id,
    status: 'queued'
  });
});

// Separate worker processes the job
worker.process('send', async (job) => {
  await sendEmail(job.data.to, job.data.subject);
});

Anti-Pattern 3: No Idempotency

Problem: Job runs twice → duplicate charges, double emails

Why it happens:

• Redis connection drops mid-processing
• Worker crashes before job completion
• Job timeout triggers retry while still running

Correct approach:

// ✅ Idempotent job with deduplication key
await queue.add('charge-payment', {
  userId: 123,
  amount: 50.00
}, {
  jobId: `payment-${orderId}`, // Prevents duplicates
  attempts: 3
});

// In worker: Check if already processed
worker.process('charge-payment', async (job) => {
  const { userId, amount } = job.data;

  // Check idempotency
  const existing = await db.payments.findOne({
    jobId: job.id
  });
  if (existing) {
    return existing; // Already processed
  }

  // Process payment
  const result = await stripe.charges.create({...});

  // Store idempotency record
  await db.payments.create({
    jobId: job.id,
    result
  });

  return result;
});

Anti-Pattern 4: No Rate Limiting

Problem: Overwhelm third-party APIs or exhaust quotas

Symptom: "Rate limit exceeded" errors from Sendgrid, Stripe, etc.

Correct approach:

// BullMQ rate limiting
const queue = new Queue('api-calls', {
  limiter: {
    max: 100,        // Max 100 jobs
    duration: 60000  // Per 60 seconds
  }
});

// Or: Priority-based rate limits
await queue.add('send-email', data, {
  priority: user.isPremium ? 1 : 10,
  rateLimiter: {
    max: user.isPremium ? 1000 : 100,
    duration: 3600000 // Per hour
  }
});

Anti-Pattern 5: Forgetting Worker Scaling

Problem: Single worker can't keep up with queue depth

Symptom: Queue backs up, jobs delayed hours/days

Correct approach:

// Horizontal scaling with multiple workers
const worker = new Worker('email-queue', async (job) => {
  await processEmail(job.data);
}, {
  connection: redis,
  concurrency: 5  // Process 5 jobs concurrently per worker
});

// Run multiple worker processes (PM2, Kubernetes, etc.)
// Each worker processes concurrency * num_workers jobs

Monitoring:

// Set up alerts for queue depth
setInterval(async () => {
  const waiting = await queue.getWaitingCount();
  if (waiting > 1000) {
    alert('Queue depth exceeds 1000, scale workers!');
  }
}, 60000);

Implementation Patterns

Pattern 1: Email Campaigns

// Queue setup
const emailQueue = new Queue('email-campaign', { connection: redis });

// Enqueue batch
async function sendCampaign(userIds: number[], template: string) {
  const jobs = userIds.map(userId => ({
    name: 'send',
    data: { userId, template },
    opts: {
      attempts: 3,
      backoff: { type: 'exponential', delay: 5000 }
    }
  }));

  await emailQueue.addBulk(jobs);
}

// Worker with retry logic
const worker = new Worker('email-campaign', async (job) => {
  const { userId, template } = job.data;

  const user = await db.users.findById(userId);
  const email = renderTemplate(template, user);

  try {
    await sendgrid.send({
      to: user.email,
      subject: email.subject,
      html: email.body
    });
  } catch (error) {
    if (error.code === 'ECONNREFUSED') {
      throw error; // Retry
    }
    // Invalid email, don't retry
    console.error(`Invalid email for user ${userId}`);
  }
}, {
  connection: redis,
  concurrency: 10
});

Pattern 2: Scheduled Reports

// Daily report at 9 AM
await queue.add('daily-report', {
  type: 'sales',
  recipients: ['admin@company.com']
}, {
  repeat: {
    pattern: '0 9 * * *', // Cron syntax
    tz: 'America/New_York'
  }
});

// Worker generates and emails report
worker.process('daily-report', async (job) => {
  const { type, recipients } = job.data;

  const data = await generateReport(type);
  const pdf = await createPDF(data);

  await emailQueue.add('send', {
    to: recipients,
    subject: `Daily ${type} Report`,
    attachments: [{ filename: 'report.pdf', content: pdf }]
  });
});

Pattern 3: Video Transcoding Pipeline

// Multi-stage job with progress tracking
await videoQueue.add('transcode', {
  videoId: 123,
  formats: ['720p', '1080p', '4k']
}, {
  attempts: 2,
  timeout: 3600000 // 1 hour timeout
});

worker.process('transcode', async (job) => {
  const { videoId, formats } = job.data;

  for (let i = 0; i < formats.length; i++) {
    const format = formats[i];

    // Update progress
    await job.updateProgress((i / formats.length) * 100);

    // Transcode
    await ffmpeg.transcode(videoId, format);
  }

  await job.updateProgress(100);
});

// Client polls for progress
app.get('/videos/:id/status', async (req, res) => {
  const job = await queue.getJob(req.params.jobId);
  res.json({
    state: await job.getState(),
    progress: job.progress
  });
});

Monitoring & Observability

Essential Metrics

// Queue health dashboard
async function getQueueMetrics() {
  const [waiting, active, completed, failed, delayed] = await Promise.all([
    queue.getWaitingCount(),
    queue.getActiveCount(),
    queue.getCompletedCount(),
    queue.getFailedCount(),
    queue.getDelayedCount()
  ]);

  return {
    waiting,    // Jobs waiting to be processed
    active,     // Jobs currently processing
    completed,  // Successfully completed
    failed,     // Failed after retries
    delayed,    // Scheduled for future
    health: waiting < 1000 && failed < 100 ? 'healthy' : 'degraded'
  };
}

BullMQ Board (UI)

// Development: Monitor jobs visually
import { createBullBoard } from '@bull-board/api';
import { BullMQAdapter } from '@bull-board/api/bullMQAdapter';
import { ExpressAdapter } from '@bull-board/express';

const serverAdapter = new ExpressAdapter();

createBullBoard({
  queues: [
    new BullMQAdapter(emailQueue),
    new BullMQAdapter(videoQueue)
  ],
  serverAdapter
});

app.use('/admin/queues', serverAdapter.getRouter());
// Visit http://localhost:3000/admin/queues

Production Checklist

□ Dead letter queue configured
□ Retry strategy with exponential backoff
□ Job timeout limits set
□ Rate limiting for third-party APIs
□ Idempotency keys for critical operations
□ Worker concurrency tuned (CPU cores * 2)
□ Horizontal scaling configured (multiple workers)
□ Queue depth monitoring with alerts
□ Failed job inspection workflow
□ Job data doesn't contain PII in logs
□ Redis persistence enabled (AOF or RDB)
□ Graceful shutdown handling (SIGTERM)

When to Use vs Avoid

Technology Comparison

References

• /references/bullmq-patterns.md - Advanced BullMQ patterns and examples
• /references/celery-workflows.md - Celery chains, groups, and chords
• /references/job-observability.md - Monitoring, alerting, and debugging

Scripts

• scripts/setup_bullmq.sh - Initialize BullMQ with Redis
• scripts/queue_health_check.ts - Queue metrics dashboard
• scripts/retry_failed_jobs.ts - Bulk retry failed jobs

This skill guides: Background job implementation | Queue architecture | Retry strategies | Worker scaling | Job observability