Scaling notes — horizontal vs vertical, CAP and PACELC, sharding strategies, idempotency, back-pressure, and the cost curves that decide whether 'just throw money at it' is the right answer.
5 articles · updated regularly
Under a load test our service started throwing 'cannot assign requested address' even though every machine had spare CPU, memory, and bandwidth. We were not out of capacity in any way a dashboard tracks. We were out of source ports, because every outbound request opened a fresh connection and left a pile of sockets sitting in TIME_WAIT.
A frank account of moving from one big shared system to many small isolated cells. The migration sucked. The blast-radius math afterward was a religious experience.
Networks drop packets. Clients retry. Without idempotency keys you charge the card twice. It's a 20-line fix for a 2 AM class of bug.
Consistency, Availability, Partition-tolerance — pick two. That summary is almost useless. Here's the real version.
Every team debates this eventually. Here's the math, the trade-offs, and why 'just scale out' is often the wrong first answer.