The deploy is the cache
2026.197Presigned URLs cap at seven days -- a protocol ceiling. A static site mints URLs at build and serves them until the next deploy. The clock starts at mint.
Infrastructure
tag:
infrastructure
all tags →2026
The personal agent trap
2026.059A week in the agent ecosystem. Routing layers with an LLM attached, not development partners.
2021
Evidence
2021.253The debate ended when the PM pulled up a graph.
PostgreSQL HA
2021.074Streaming replication, automated failover, and the operational lessons that aren't in the PostgreSQL docs -- split-brain, silent lag, and why DR drills matter more than DR plans.
2020
When the aquarium went dark
2020.228The checklist ports to Slack. The hazard lights are still wired. Nobody filed the ticket.
2019
The board stayed green
2019.295The board put us ahead of every customer ticket - until it stayed green while the data underneath it went wrong.
Prescaling for a known spike
2019.074Autoscale never helped in the time of crisis. What helped was prescaling -- capacity ready before the traffic arrives.
2018
Consul in practice
2018.253Service discovery, health checks, and dynamic config across a growing fleet of microservices -- and the moment KV store got too popular.
The week pgbouncer stopped being news
2018.193Transaction pooling, one SET that didn't stick, and a standup that moved on.
A week before the sale
2018.154The checklist is the same as last year. That's the point.
The GraphQL buffer
2018.110A GraphQL gateway that started as a unified API layer for clients became the place where we could intervene without touching the backends.
2017
Hazard lights
2017.161We wired physical lights to 5xx alerts. It changed how the team reacts to production issues.
Nginx load balancing decisions
2017.138Algorithm selection, passive health checking, per-endpoint rate limiting, and upstream timing -- the decisions that matter when Nginx sits in front of production traffic.
2014
Node.js on a Raspberry Pi
2014.005Running a custom Node.js blog engine on a Raspberry Pi at home -- from a 45-day wait and a dead board to DuckDNS, ISP port blocking, and the first external request.
2013
Running Node.js in production
2013.149Upstart, clustering, zero-downtime deploys, heartbeat monitoring, and an embedded REPL -- what it takes to keep Node.js running in production in 2013.
MySQL on XFS
2013.101Using bind mounts to move MySQL data, config, and logs to an XFS volume without changing a single MySQL setting.