Writing
Writing on data architecture.
Notes from building, running, and improving data platforms.
The Kafka-to-Iceberg Pipe Is Disappearing. The Table Contract Is Not.
How Kafka topics became Iceberg tables before the convergence wave, how the new solutions actually differ, and the questions that decide which one fits your platform.
Retries Are Not Free
How to compare retry controls, retry budgets, idempotency, circuit breakers, queues, and replay design without pretending they solve the same problem.
When Iceberg Says 321 GB and S3 Says 46 TB
Iceberg metadata can be correct while S3 is still billing you for tens of terabytes under the same table path.
Why Chargeback Fails in Shared Data Platforms
Infrastructure tags tell you who owns the box, not who caused the bill.
Who Owns This Table?
Most data lake cost problems are ownership problems wearing a storage bill costume.
The Full Rewrite Anti-Pattern in Data Lakes
Full rewrites are often the most expensive way to make a small logical change.