Topic: cost-optimization
- Drowning in Logs, Starving for Answers
A production Kubernetes cluster emitted 25 million structured log lines and 62 GB a day, but no one could answer 'which service is failing right now?' with a query. Here is the pipeline that turned that firehose of JSON into columnar Parquet you can hit with plain SQL, and the design decisions behind it.
- Zero-Downtime Across AWS Regions: The Decisions Behind Moving 33 Production Microservices
A region-to-region migration is sold as a redeploy. The hard part is everything the old region quietly assumed: a machine image baked per release, a self-hosted artifact server, and one load balancer per service. Here is how 33 microservices moved with no outage, and the decisions that made it possible.