Microservices Outbox Pattern

Microservices: Outbox Pattern

Guarantee at-least-once event delivery using a transactional outbox and polling publisher.

When to Use

• You publish events to Kafka/RabbitMQ/SNS as part of database transactions and can't afford to lose events
• Your service has experienced dual-write inconsistencies (DB wrote but event wasn't published, or vice versa)
• You need guaranteed event delivery without distributed transactions (no 2PC)
• You're building event-sourced or CQRS systems where event publication is critical

Instructions

The dual-write problem:

WITHOUT OUTBOX:
1. BEGIN TRANSACTION
2. INSERT INTO orders → success
3. COMMIT TRANSACTION → success
4. kafka.produce('order.created') → CRASH → event never sent
   ↑ DB and message broker are now inconsistent

WITH OUTBOX:
1. BEGIN TRANSACTION
2. INSERT INTO orders → success
3. INSERT INTO outbox (same transaction) → success
4. COMMIT TRANSACTION → both writes are atomic
5. Separate publisher process reads outbox → publishes to Kafka → marks as published

Full implementation with Prisma and Kafka:

import { PrismaClient } from '@prisma/client';

const prisma = new PrismaClient();

// Writing to DB + outbox in ONE transaction
async function createOrder(input: CreateOrderInput): Promise<Order> {
  return prisma.$transaction(async (tx) => {
    const order = await tx.order.create({
      data: {
        userId: input.userId,
        status: 'PENDING',
        total: calculateTotal(input.items),
        items: { create: input.items },
      },
    });

    // Outbox record — same transaction, same commit
    await tx.outboxEvent.create({
      data: {
        aggregateId: order.id,
        aggregateType: 'Order',
        eventType: 'order.created',
        payload: {
          eventId: crypto.randomUUID(),
          orderId: order.id,
          userId: order.userId,
          total: order.total,
          items: input.items,
          occurredAt: new Date().toISOString(),
        },
      },
    });

    return order;
  });
}

// For every domain operation, add an outbox record in the same transaction
async function cancelOrder(orderId: string, reason: string): Promise<void> {
  await prisma.$transaction(async (tx) => {
    await tx.order.update({
      where: { id: orderId },
      data: { status: 'CANCELLED' },
    });

    await tx.outboxEvent.create({
      data: {
        aggregateId: orderId,
        aggregateType: 'Order',
        eventType: 'order.cancelled',
        payload: {
          eventId: crypto.randomUUID(),
          orderId,
          reason,
          occurredAt: new Date().toISOString(),
        },
      },
    });
  });
}

Polling publisher (runs as a separate process or scheduled job):

class OutboxPublisher {
  private timer: NodeJS.Timeout | null = null;

  constructor(
    private readonly db: PrismaClient,
    private readonly kafka: KafkaProducer,
    private readonly options = {
      pollIntervalMs: 1_000,
      batchSize: 50,
      maxAttempts: 5,
    }
  ) {}

  start(): void {
    this.poll();
  }

  stop(): void {
    if (this.timer) {
      clearTimeout(this.timer);
      this.timer = null;
    }
  }

  private poll(): void {
    this.processBatch()
      .catch((err) => console.error('Outbox publisher error:', err))
      .finally(() => {
        this.timer = setTimeout(() => this.poll(), this.options.pollIntervalMs);
      });
  }

  private async processBatch(): Promise<void> {
    const events = await this.db.outboxEvent.findMany({
      where: {
        publishedAt: null,
        attempts: { lt: this.options.maxAttempts },
      },
      orderBy: { createdAt: 'asc' },
      take: this.options.batchSize,
    });

    for (const event of events) {
      try {
        await this.kafka.send({
          topic: event.eventType,
          messages: [
            {
              key: event.aggregateId,
              value: JSON.stringify(event.payload),
              headers: {
                'outbox-event-id': event.id,
                'aggregate-type': event.aggregateType,
              },
            },
          ],
          acks: -1, // wait for all replicas
        });

        await this.db.outboxEvent.update({
          where: { id: event.id },
          data: { publishedAt: new Date() },
        });
      } catch (err) {
        await this.db.outboxEvent.update({
          where: { id: event.id },
          data: {
            attempts: { increment: 1 },
            lastError: (err as Error).message,
          },
        });
        console.error(`Failed to publish event ${event.id}:`, err);
      }
    }
  }
}

Outbox schema (Prisma):

model OutboxEvent {
  id            String    @id @default(uuid())
  aggregateId   String
  aggregateType String
  eventType     String
  payload       Json
  publishedAt   DateTime?
  attempts      Int       @default(0)
  lastError     String?
  createdAt     DateTime  @default(now())

  @@index([publishedAt, attempts]) // index for the polling query
}

Pruning published events:

// Run as a scheduled job (e.g., daily cron)
async function pruneOutbox(retentionDays = 7): Promise<number> {
  const cutoff = new Date(Date.now() - retentionDays * 86_400_000);
  const { count } = await prisma.outboxEvent.deleteMany({
    where: {
      publishedAt: { lt: cutoff },
    },
  });
  console.log(`Pruned ${count} outbox events`);
  return count;
}

Details

CDC alternative: Instead of polling, use Debezium (Change Data Capture) to stream the outbox table changes directly to Kafka via Kafka Connect. Near-zero latency, no polling overhead.

Consumer idempotency required: The outbox guarantees at-least-once delivery. Consumers must be idempotent (use the outbox-event-id header as the deduplication key). See events-idempotency skill.

Multiple publishers: If running multiple service instances, each polls the outbox. Use SELECT ... FOR UPDATE SKIP LOCKED to prevent duplicate publishing:

SELECT * FROM outbox_events
WHERE published_at IS NULL AND attempts < 5
ORDER BY created_at
LIMIT 50
FOR UPDATE SKIP LOCKED;

Anti-patterns:

• Publishing before committing — the transaction may roll back but the event is already sent
• Outbox table with no index on published_at — polling becomes a full table scan
• Unbounded retention — the outbox table grows forever; schedule pruning

Source

microservices.io/patterns/data/transactional-outbox.html

Process

1. Read the instructions and examples in this document.
2. Apply the patterns to your implementation, adapting to your specific context.
3. Verify your implementation against the details and edge cases listed above.

Harness Integration

• Type: knowledge — this skill is a reference document, not a procedural workflow.
• No tools or state — consumed as context by other skills and agents.

Success Criteria

• The patterns described in this document are applied correctly in the implementation.
• Edge cases and anti-patterns listed in this document are avoided.