Distributed Systems
Notification Fanout Architecture
Notification fanout comes down to fan-out-on-write vs on-read, and the celebrity problem that breaks the naive choice. The hybrid design that scales, explained.
All writing
Archive
45 articles and counting.
System Design
WebSocket Capacity Planning for Social Products
WebSocket capacity planning for social products: budget memory, file descriptors, and fan-out per connection so you scale before connections, not CPU, break first.
Distributed Systems
Backpressure Design for Real-Time Systems
Backpressure keeps real-time systems alive under load by making producers slow down instead of drowning consumers. Strategies, tradeoffs, and a checklist.
Distributed Systems
Timeout Budgets Across Service Chains
A timeout budget is one deadline split across a service chain so the whole request fails fast instead of piling up doomed work. How to set and propagate it.
Distributed Systems
Idempotency Keys for Distributed Systems
Idempotency keys make retried requests safe so a timed-out payment or duplicate POST applies exactly once. The design, storage, and TTL decisions that matter.
Distributed Systems
Sharding Strategies Before You Need Them
Pick database sharding strategies before traffic forces your hand. Shard keys, hash vs range vs directory, online resharding, and the traps that bite early.
Engineering Strategy
Grafana Dashboards for Operators, Not Executives
Most Grafana dashboards are decoration. An operator dashboard answers one question fast during an incident. How to design dashboards that speed up debugging.
Distributed Systems
Jaeger Tracing for Cross-Service Debugging
Jaeger turns a slow request across many services into one visual trace. How distributed tracing works, what to instrument, and the sampling tradeoff that bites.
Kubernetes
K3s on Hetzner: What to Trust and Isolate
Running K3s on Hetzner is the cheapest path to real Kubernetes, but bare-metal means you own the security boundaries. What to trust, what to isolate, and how.
Engineering Strategy
Bazel Monorepo Builds for Large Systems
Bazel makes large monorepo builds fast through hermetic, cached, incremental builds, but the cost is up-front rigor. When it pays off, and how to adopt it.
Distributed Systems
Kafka Replay Strategy Without Duplicate Events
Replaying a Kafka topic re-delivers events, so duplicates are guaranteed unless consumers are idempotent. The safe replay playbook: dedup, offsets, and isolation.
Kubernetes
Readiness Probes That Don't Lie
Most Kubernetes readiness probes lie: they return 200 because the process started, not because the service can serve. How to write probes that tell the truth.