Rules of Observability: The Data Edition
summary
Claims that observability is a data problem: keep metrics, traces, logs at full fidelity because sampling and aggregation introduce blind spots. Demonstrates head‑ vs tail‑based sampling and how a 5% tail sample gives ~95% chance of missing an outlier, shows a 29–40s trace that costs conversions, and a 30 rps threshold hidden by 10s aggregation. Emphasizes precision vs accuracy, drift and skew in elastic infrastructure, prefer median for historical baselines, and head-based vs tail-based sampling, 30 rps alert missed by aggregation, keep full-fidelity traces metrics logs.
AI-written summary · about
chapters (8)
- 0:00Observability and data challenges
- 3:35Cloud native observability complexity
- 6:43Beyond traditional monitoring
- 9:12Rule one: Use all data
- 13:07Pitfalls of trace sampling
- 19:52Rule two: App and infra speed
- 27:50Drift and skew in clouds
- 30:59Predictive alerts and reliability
related talks
36mObservability? It's a Data Analytics Problem, You Fool!
35mBeyond Logs, Metrics and Traces. The Data Analytics Paradigm Shift in Observability — Dotan Horovits
38mThe Murphy's Laws of Observability
26mYou Can’t Fix What You Can’t See: Tales from the Observability Frontier
30mMurphy’s Laws for observability
30mObservability is More than Logs, Metrics & Traces
52mObservability: it's not just an ops thing
42mObservability can help you understand change