Mistake: random-sampling logs and losing rare-but-critical bot events

Mistake: random-sampling logs and losing rare-but-critical bot events

Uniform sampling is fine for users but erases the rare 5xx Googlebot hit you most need to see. Sources:

→ Honeycomb on dynamic sampling — Sample common events hard, keep all the errors and outliers.
Takeaway: never sample 5xx, 429, or verified-bot lines — keep them whole.

→ Google Search Central (5xx and crawling) — A burst of 5xx to Googlebot cuts crawl rate fast.
Takeaway: those are the exact rows uniform sampling throws away.

★ Pick of the week — a tail-based sampling sketch — Keep 100% of errors, 1% of 200s.
Takeaway: size your sample by event importance, not a flat percentage.