Writing
Practical thinking on distributed systems, observability, and production engineering. No hype, just things I've learned the hard way.
Collecting every trace in a high-volume system is wasteful and expensive. Learn why sampling is statistically sound, how head and tail sampling differ, and how the OpenTelemetry Collector gives you the control to do it right.
Traces, Metrics, Logs, Baggage, and Profiles: the definitive guide to understanding OpenTelemetry's five signal types and how they fit into a real observability strategy.
Running LLMs in production is not like running a REST API. Here's how to instrument, trace, and monitor large language models, AI agents, and retrieval-augmented generation pipelines so you can actually debug them when they fail.
Unstructured logs are a liability in distributed systems. Here's how to design structured, JSON-based logging that correlates with traces, survives high cardinality, and actually helps you debug at 3am.
The circuit breaker prevents cascading failures, but without proper instrumentation you're blind to why it trips. Here's how to make it observable end-to-end.
Discover Goku, a blazing-fast HTTP load testing tool written in Rust, scalable, and with built-in MCP (Model Context Protocol) support. Complete guide and practical examples.
Most teams instrument everything and understand nothing. A practical guide to designing spans that actually help you debug production issues.
Trace propagation through message queues and async workers looks straightforward until it isn't. Here's what actually breaks and how to fix it.
The difference between a reliability target that drives engineering decisions and one that collects dust in a wiki page.
Container memory limits and JVM ergonomics don't always agree. A deep dive into diagnosing and fixing OOM kills, GC storms, and off-heap leaks.
