Observability 2.0 - Honeycomb / Charity Majors

Takeaway

  • O11y 1.0 = metrics + logs + traces as three separate pillars, each with its own tool
  • O11y 2.0 = unify everything into wide structured log events, one data source to rule them all
  • Shift from write-time aggregation (pre-compute metrics) to read-time aggregation (keep raw events, query on the fly)
  • Honeycomb claims logs are cheaper than metrics when you count total cost (no need to maintain 3 separate systems)
  • But dodges the hard question: at massive scale, raw events are way more data than pre-aggregated metrics
  • In practice, sampling + columnar storage + cheaper hardware makes it work for most companies
  • Industry trend is moving this direction, but metrics still win for ultra-fast pre-computed queries