Short notes from production
Not every lesson deserves two thousand words. These are the short ones: decisions, defaults and details our engineers apply on real infrastructure.
The collection
Choosing NVMe for Blockchain Databases
Capacity is the easy number. Sustained write endurance and latency under compaction are the ones that decide node health.
Why Latency Matters for RPC Infrastructure
Median latency is a vanity metric. Products fail at the tail.
What We Monitor on Production Nodes
The short list we put on every node dashboard before anything else.
How We Think About Multi Region Architecture
Two regions solve different problems depending on why you added the second one.
Cloud Egress, the Hidden Infrastructure Cost
Peer to peer traffic and public API responses make egress the line item nobody forecasts.
Why Infrastructure Isolation Matters
Shared capacity is fine until one consumer decides what everyone else experiences.
When Kubernetes Adds Too Much Complexity
Orchestration earns its keep on stateless services. On multi terabyte nodes it mostly adds a scheduler that can hurt you.
Monitoring ClickHouse at Scale
Most ClickHouse incidents are visible in part counts long before they are visible to users.
Terraform Modules for Repeatable Infrastructure
A module per role beats a module per resource, every time.
Designing Better Failover
Automatic failover that has never been exercised is a hypothesis, not a control.