Logs Are Letters to Future-You
The log line you skipped is the outage you cannot explain.
[ essay ]
A log line is a letter to whoever is debugging this at 3am — and the signal-to-noise ratio decides whether that letter helps or becomes more weather to scroll past.
I learned this running vision batches on the mystic-bytes cover pipeline. Throughput dropped forty percent over a weekend. No stack trace, no failed health check — just thousands of processing image lines with no batch ID, no slug, no elapsed time. The regression was real; the narrative was missing. I had written logs for a dashboard, not for a stranger under pressure who needed to know which image and when the slowdown started.1
Good logging is not more logging. It is the right event, at the right level, with correlation IDs that survive across preprocess, upload, and manifest write. Write as if someone who does not have your context must reconstruct the story from lines alone.
If you would not thank yourself for reading it during an incident, delete it or fix it.
— JV · Dark Heart Labs.
-
Charity Majors, “Observability — A 3-Year Retrospective” (Honeycomb blog, 2022) — on structured events and high-cardinality context as the unit of debugging, not strings sprinkled after the fact. ↩