Blog

What happens in production, with sources.

Short notes on cloud, legacy systems, payments and AI in production. Each one starts from a published fact and ends with what a team would do about it.

Page 1 of 3

A cooling failure in Virginia left Coinbase unable to operate for hours

On May 8, 2026, cooling failed at an AWS data center in us-east-1, in Northern Virginia, and Coinbase went offline for several hours. In October 2025, a DNS failure in the same region had affected 64 internal AWS services and hundreds of applications. If your whole system lives in a single availability zone, your availability depends on what happens in those buildings.

Staying on an old Kubernetes version multiplies the per-cluster charge on EKS by six

AWS charges $0.10 per hour per cluster under standard support and $0.60 under extended support, which starts 14 months after each version's release. It's the charge for the control plane, not for the nodes, but you pay it for every cluster, every month. Putting off the upgrade has a cost: you pay more to stand still.

SQL Server 2016 went out of support in July. Windows Server 2016 follows on January 12

Microsoft ended extended support for SQL Server 2016 on July 15, 2026, and extended support for Windows Server 2016 ends on January 12, 2027. After those dates, security fixes come only through the Extended Security Updates (ESU) program, which is paid outside Azure. Everything keeps working the next day; what changes is who pays to stay secure.