Field notes from running Kafka in production.
October 26, 2026 · MessagePilots
An operator removes most of the toil of running Kafka on Kubernetes, but not the decisions. Storage classes, rack awareness, KRaft node pools, and rolling-update behaviour still belong to you.
Read moreOctober 19, 2026
One unparseable record can stall a partition indefinitely. The three honest options — skip, retry, divert — what each costs, and why a DLQ quietly breaks per-key ordering.
Read moreOctober 12, 2026 · MessagePilots
Both will do your joins and aggregations. They differ in who owns state, how you recover, and what the on-call rotation looks like. A decision framework, plus the cases where neither is the answer.
Read moreOctober 5, 2026 · MessagePilots
Replication factor 3 does not mean your writes survive. The producer and broker settings that decide whether an acknowledged record is still there after a broker dies, and the availability you trade for them.
Read moreSeptember 28, 2026 · MessagePilots
One shared cluster, twenty teams, and no client quotas is a latency incident waiting for its trigger. How broker-side quotas, namespacing, and per-tenant visibility keep a shared Kafka estate boring.
Read more