Argo CD Observability with AIOps - Detect Incident Fast
summary
Intuit runs 2,500+ services on Kubernetes and finds one third of incidents come from rollouts so Argo CD gets an integrated metrics tab in Argo CD showing golden signals, plus an AI pipeline using Numaflow pipeline and Prometheus remote_write that produces 16M+ predictions per day from 2.5B events daily. An autoencoder model delivers a anomaly score 0–10 (above 7 = bad) with ~0.5s inference latency, online retraining via Argo Workflows, and scores written back to Prometheus for faster detection and automated rollback integration.
AI-written summary · about
chapters (8)
- 0:00Start
- 5:14Demo Plan
- 12:27Architecture - Overall
- 13:32Architecture - Real-time Streaming AlOps
- 16:40Real-time Streaming Barriers for AlOps
- 19:58Ideal Streaming AlOps Features
- 22:46Numaflow - Streaming AlOps Platform
- 26:07The Road Ahead
related talks
25mObservability In ArgoCD/Rollouts Using Streaming ML For Reducing MTTR
22mReal Time Argo Rollouts Analytics Powered by Notification Engine
23mArgo at Scale: Navigating Complex Multi-Dimensional Deployments Across H...
22mAnalysis in Action: Real-World Metrics Driven Argo Roll...
26mBeyond Argo Rollouts: Boosting Developer Experience With Intelligen...
31mAIOps: (near) Zero-Touch Production Rollout Fixes
23mStop Deploying Blind! Using Observability and Argo Rollouts to Light the Way
24mFix Production Rollouts on the Fly With Agentic AIOps