The Shrinking ISRKafka · medium · ~12 minNo broker has crashed, yet UnderReplicatedPartitions keeps flapping between 0 and ~14 all afternoon, always involving replicas hosted on kafka-4. Producers are fine. It clears for ten minutes, then it's back.Something is making followers on kafka-4 fall in and out of sync. Find it.Evidence

Metrics

Broker Logs

Config

Checkpoints

What does the ISR shrinking (without any broker dying) mean?

What's driving the flapping?

The right fix?

Why is raising replica.lag.time.max.ms to 'make the alert stop' a dangerous fix? Answer interview-style.