Kafka consumer lag — downstream processing 3 hours behind

Ситуация: Stream processing: Kafka → Flink → ClickHouse. Today consumer lag 3 hours (was 30 seconds). Dashboards stale, customers complain.

Stream: 100K events/sec peak. Flink consumer behind. Cause unclear.

Доступные данные

• kafka_metrics: topic, partition, consumer_group, lag, ts
• flink_metrics: job, processing_rate, backpressure
• clickhouse_metrics: query_time, writes_per_sec

Задачи

• Localize bottleneck.
• Immediate fix.
• Root cause.
• Post-mortem prevention.

Все кейсы для подготовки → ← Все кейсы