Events Kafka Patterns

Events: Kafka Patterns

Produce and consume Kafka messages with partitioning, consumer groups, and offset management.

When to Use

• You need high-throughput, durable event streaming (millions of events/day)
• You need event replay — re-consume historical events to rebuild state or debug
• You have multiple independent consumer groups that each need all events
• You need ordered processing within a partition (e.g., all events for one user in order)
• You're building event sourcing, CDC pipelines, or stream processing

Instructions

KafkaJS producer with partitioning:

import { Kafka, Partitioners, CompressionTypes } from 'kafkajs';

const kafka = new Kafka({
  clientId: 'order-service',
  brokers: ['kafka:9092'],
  retry: { initialRetryTime: 300, retries: 8 },
});

const producer = kafka.producer({
  createPartitioner: Partitioners.LegacyPartitioner,
  idempotent: true, // enable exactly-once for the producer
  transactionTimeout: 30_000,
});

await producer.connect();

// Send with a partition key — ensures ordering for the same key
await producer.send({
  topic: 'order.events',
  messages: [
    {
      key: `order:${orderId}`, // same key → same partition → ordered
      value: JSON.stringify({
        eventType: 'ORDER_CREATED',
        orderId,
        userId,
        amount,
        timestamp: new Date().toISOString(),
      }),
      headers: {
        'event-type': 'ORDER_CREATED',
        'source-service': 'order-service',
        'schema-version': '1',
      },
    },
  ],
  compression: CompressionTypes.GZIP,
  acks: -1, // wait for all ISR replicas to ack
});

Consumer with consumer group and manual offset commit:

const consumer = kafka.consumer({
  groupId: 'fulfillment-service', // all instances share this group — each partition assigned to one
  sessionTimeout: 30_000,
  heartbeatInterval: 3_000,
});

await consumer.connect();
await consumer.subscribe({ topic: 'order.events', fromBeginning: false });

await consumer.run({
  eachBatchAutoResolve: false, // manual offset commit for fine-grained control
  eachBatch: async ({ batch, resolveOffset, heartbeat, commitOffsetsIfNecessary }) => {
    for (const message of batch.messages) {
      const event = JSON.parse(message.value!.toString());

      try {
        await processOrderEvent(event);
        resolveOffset(message.offset); // mark this message as processed
      } catch (err) {
        console.error(`Failed to process ${message.offset}:`, err);
        // Do NOT resolve offset — message will be reprocessed
        break; // stop processing this batch
      }

      await heartbeat(); // prevent session timeout during slow processing
    }

    await commitOffsetsIfNecessary(); // commit resolved offsets
  },
});

Transactional producer (atomic produce + consume):

const transactionalProducer = kafka.producer({
  transactionalId: 'order-processor-1', // unique per producer instance
  idempotent: true,
  maxInFlightRequests: 1,
});

await transactionalProducer.connect();

// Atomic: consume from input topic, produce to output topic
const transaction = await transactionalProducer.transaction();
try {
  await transaction.send({
    topic: 'shipping.commands',
    messages: [{ key: `order:${orderId}`, value: JSON.stringify({ orderId, address }) }],
  });

  // Commit offsets as part of the transaction
  await transaction.sendOffsets({
    consumerGroupId: 'order-processor',
    topics: [{ topic: 'order.events', partitions: [{ partition: 0, offset: '42' }] }],
  });

  await transaction.commit();
} catch (err) {
  await transaction.abort();
  throw err;
}

Topic and partition strategy:

// Partition key selection strategy
function getPartitionKey(event: OrderEvent): string {
  // User-scoped events: partition by userId for ordering
  if ('userId' in event) return `user:${event.userId}`;
  // Order events: partition by orderId
  if ('orderId' in event) return `order:${event.orderId}`;
  // Global events: null key → round-robin
  return '';
}

// Topic naming convention
const TOPICS = {
  ORDER_EVENTS: 'order.events', // all order lifecycle events
  FULFILLMENT_COMMANDS: 'fulfillment.commands', // commands to fulfillment
  NOTIFICATION_EVENTS: 'notification.events', // fan-out notifications
  DLQ: 'order.events.dlq', // failed processing
} as const;

Details

Consumer group mechanics: All consumers in the same groupId share partitions — each partition assigned to one consumer. Add consumers to scale (up to partitionCount consumers can process in parallel). Different groupId values create independent consumers that each receive all messages.

Offset management:

• fromBeginning: true — start from the earliest retained offset (useful for new consumer groups)
• fromBeginning: false — start from the latest offset (only new messages)
• Manual offset commit gives exactly-once semantics when combined with idempotent processing

Retention and compaction:

• Default retention: time-based (7 days). Good for audit logs, replay.
• Log compaction: keeps only the latest message per key. Good for state materialization (like a distributed KV store).

Anti-patterns:

• Single partition for all messages — no parallelism, no ordering control
• Committing offsets before processing — risk of data loss on crash
• Consumer that blocks eachMessage without heartbeats — causes session timeout and partition reassignment
• Sharing transactionalId across multiple producer instances — causes transaction fencing errors

Schema evolution: Use Avro or Protobuf with a schema registry. Never break consumers by removing fields without versioning. See events-event-schema skill.

Source

kafka.apache.org/documentation/

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.