Metrics that matter Metric Why Consumer lag (per group/partition) the #1 health signal — growing lag = consumers can't keep up Under-replicated partitions broker or disk trouble; should be 0 Offline partitions / active controller count availability Request latency (produce/fetch p99) client experience Disk usage & retention brokers stop when disks fill Rebalance rate unstable consumers Tools: Kafka UI / AKHQ / Conduktor for browsing, Burrow or exporter + Prometheus/Grafana for lag, JMX metrics from brokers and clients. Everyday operations Reset offsets to replay (--reset-offsets) — the group must be stopped. Add partitions for throughput (mind key re-mapping). Rolling broker restarts one at a time; wait for ISR to recover. Since Kafka 4.0, clusters run in KRaft mode — ZooKeeper is gone.