Kube Intelligence - A Metric Based Insightful Remediation Recommender
Talkat KubeCon + CloudNativeCon China 2025 · on Wednesday, 2025-06-11· at 11:45· Level 19 | Crystal Court I·
summary
An internal recommender ingests metrics via Prometheus remote_write, traces via OpenTelemetry tracing, and runs modular recommendation engines (rightsizing, serverless candidacy, maintenance windows, priority classes, node defragmentation). It uses seasonal ARMA, moving‑average spike detection and exponential smoothing plus graph grouping to propose concrete actions. Piloted across ~20 teams (900 nodes, ~$1M spend, 1k+ pods, ~100 services, 10 metrics), it flags 19 spike‑y services, recommends 5 serverless moves, yields one $3,200/month win, a $230/month 5% saving, 15% fewer restarts and measurable throughput/cost gains.
AI-written summary · about
chapters (9)
- 0:00Kube Intelligence intro
- 1:05Kubernetes optimization challenges
- 3:32Intelligent remediation platform
- 8:10Recommendation engines design
- 9:34Maintenance windows detection
- 12:48Latency and replication costs
- 16:42Real-world results and benefits
- 20:54LLMs in recommendations
- 21:35Q&A
related talks
27mMastering Cloud Cost Optimization with Kruize
22mHidden Signals in K8s Clusters
11mK8s Resources Ensuring Performance and Optimization
11mLightning Talk: Grasp the Real Worth of Your Microservices by Brewing a Truth Poti...
32mOptimizing Application Performance on Kubernetes
34mOptimizing operations & container management
40mOptimizing AI/ML Workloads
29mRight-Size Your Kubernetes Deployments