Articles
- Head-of-line blocking: how one dead output stalled a Logstash pipeline
A production incident write-up: one dead Elasticsearch output wedged a whole Logstash pipeline with a persisted queue, and how it was detected, fixed and drained.
- Metrics vs Logs: where it lives and what it costs
How I decide whether a field belongs in a time-series database or a log store, measured on a day of real e-commerce edge traffic.
- Non-Determinism Doesn't Matter If the Tests Are Strict Enough
Why LLM non-determinism stops mattering once the spec and the tests pin down the required behavior, and where the remaining variance comes from.
- Keeping an LLM honest about what it doesn't know
A three-layer setup (a SessionStart hook, verification rules and claude.ai preferences) that makes Claude check current facts instead of answering from stale training data.
- epm: watching an Elasticsearch cluster from the terminal
Why I wrote epm, a terminal dashboard for Elasticsearch, and how it turns cumulative counters into rates without an Elasticsearch client library.