Beyond Root Cause Analysis for K8s: What Your Logs Should Be Telling You
Talkat Observability Day 2025 · on Tuesday, 2025-04-01· at 10:45· Level 1 | Hall Entrance S10 | Room C·
summary
Claims that better RCA requires better data and shows concrete fixes: convert unstructured logs to structured fields, aggregate similar messages and sample repetitive health checks, convert logs to metrics for numeric values, and summarize transactions via trace IDs by grouping per transaction and using an LLM to produce one structured summary log per path. Demonstrated on large services where ~80% of volume comes from ~20% of paths, replacing multiple lines with a single summary reduces volume and improves root cause visibility.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 1:06Logs and RCA challenges
- 3:31Log data problems
- 7:45Improving log hygiene
- 10:37AI log narratives
- 14:11Summary
related talks
32mGenerative AI for Observability: Automating Root Cause Analysis in Modern IT Systems
30mThere is No Root Cause...
07mAccelerating RCA with AI
41mImprove observability for your container workloads
21mWhy Do My Logs Suck?! A Practical Guide to Improving Your Logs
27mAI-Powered Root Cause Analysis at Scale: From Theory To Production...
28mAsk your logs
22mHidden Signals in K8s Clusters