06-wide-events-scuba-burmistrov

All you need is Wide Events, not “Metrics, Logs and Traces”

• Author: Ivan Burmistrov (ex-Meta)
• Link: https://isburmistrov.substack.com/p/all-you-need-is-wide-events-not-metrics
• Status: Read

Takeaway

• Wide Event = one fat JSON per event, cram in every field you might need later
• Traces, logs, metrics are all just special cases of Wide Events (add traceId = trace, add log_message = log, periodic system snapshot = metric)
• Meta Scuba: columnar store, no pre-aggregation, keep raw events, query-time aggregation, brute-force scan
• Core workflow: pick a metric, group by any field, drill down — slice and dice until you find the problem
• Dynamic sampling controls cost: normal traffic 1/1000, errors 1/1, UI auto-upscales so users dont notice
• Storage cost is the main tradeoff: raw events vs pre-aggregated metrics differ by orders of magnitude, sampling is the bridge
• External alternatives: Honeycomb (closest to Scuba philosophy), Axiom, ClickHouse