Taming uncertainty in Observability
summary
Claim: model observability as a spatio-temporal uncertainty problem and ingest normal plus fault‑injection traces into a spatio-temporal database. The pipeline applies time series analysis, association rules, clustering, Markov models and Bayesian networks, plus a knowledge graph built from logs, to surface likely causal pathways, hierarchical alarms, and time‑to‑failure recommendations. The system returns ranked, human‑actionable recommendations rather than fully automated fixes and is meant to augment human diagnosis during chaos experiments.
AI-written summary · about
related talks
13mObservability Where are We Headed?
35mObservability: Building Confidence & Answering the What, Where, How
42mObservability can help you understand change
10mFueling system discovery with observability
42mObservability Is Not Analytics!
01h 21mBuilding a ML Model Observability pipeline
32mRules of Observability: The Data Edition
01h 00mWhy observability requires the marriage of AI, metrics, and logs