Observability In ArgoCD/Rollouts Using Streaming ML For Reducing MTTR
summary
At Intuit one third of P1/P2 outages trace to change across ~2500 Kubernetes services, so a K8s-native DAG streaming platform is built that runs real-time anomaly scoring and automated canary rollback. The pipeline uses Numaflow for streaming, Numalogic for multivariate time-series inference and online training, and Prometheus remote_write to feed metrics; it computes per-deployment anomaly scores and triggers Argo Rollouts auto-rollback, lowering MTTD/MTTR in production.
AI-written summary · about
chapters (13)
- 0:00ArgoCD observability with AI
- 1:02Platform modernization at Intuit
- 4:10Change-induced incidents
- 5:55AI-powered ArgoCD/Rollouts observability
- 6:16Demo: Developer bug introduction
- 8:45Metrics in ArgoCD UI
- 11:12AI Ops anomaly scoring
- 12:14Automated rollback with Argo Rollouts
- 13:42Streaming ML observability architecture
- 16:09Real-time streaming challenges
- 17:53Pneuma Flow for streaming
- 21:43ML for time-series analytics
- 23:29Future roadmap and OSS plans
related talks
30mArgo CD Observability with AIOps - Detect Incident Fast
07mArgo in an AI-Native World: Effective In-Cluster Machine Learning and A...
22mAnalysis in Action: Real-World Metrics Driven Argo Roll...
23mStop Deploying Blind! Using Observability and Argo Rollouts to Light the Way
22mReal Time Argo Rollouts Analytics Powered by Notification Engine
26mProgressive Delivery Made Easy with Argo Rollouts
25mPrometheus Data analysis and Event Notifications for Progressive Delivery
28mPanel: From Signals To Safety Nets: Self-Healing Progressive...