/writing

Engineering write-ups.

Debugging stories and MLOps patterns from systems I run in production. No listicles — every post comes from a real repo with the commits to prove it.

2026-07

My Kafka producer said it delivered 500 messages. It delivered zero.

A Kubernetes smoke test exposed a component failing while reporting success — confluent-kafka's flush() returns undelivered counts instead of raising. In distributed systems, silent success is worse than loud failure.

read →