Kafka Failure Stories

Kafka incidents rarely start with Kafka being broken. More often, the failure is in consumer group design, retry behavior, database writes, payload size, or the assumptions around idempotency.

This hub groups the Bytz Echo Kafka articles by the kind of production signal engineers usually see first: duplicate work, stuck consumers, lag, and confusing monitoring.

Duplicate work and consumer group mistakes

Lag, writes, and throughput

Local infrastructure and schema

How to read these stories

Start with the incident article that matches your symptom. If the symptom is duplicate work, read the consumer group and duplicate report articles first. If the symptom is lag, start with the PostgreSQL write benchmark before changing Kafka broker settings.

Suggested next articles

  • Kafka Retry Strategy That Prevented Duplicate Jobs
  • Why Consumer Lag Increased After We Added an Index
  • Spring Boot Kafka Dead Letter Topics: What We Got Wrong First