Leveraging Argo Ecosystem for Scalable Cloud-Native Solutions at Databricks
summary
Databricks uses Argo Events, Argo Rollouts and Argo Workflows with Prometheus, Envoy and Kubernetes to run Gen Serving that handles ~5,000 active endpoints and up to 25,000 QPS per cloud region. Rollouts use staged steps (example 20% then 40%) and an Argo Rollouts dry-run path to avoid false-positive rollbacks; workflows power warm pools sized by a proprietary ML model across 70 regions, build-and-deploy pipelines with user-visible build logs, and periodic metric delivery.
AI-written summary · about
chapters (9)
- 0:00Argo ecosystem at Databricks
- 3:54Model serving lifecycle with Gen Serving
- 7:06Safe Ingress stack updates
- 10:18Zero-downtime deployments with Rollouts
- 12:04Safe checks with Rollouts dry run
- 15:35Capacity planning with Workflows
- 18:02Container builds for model serving
- 21:12Metrics pipeline with Workflows
- 23:17Q&A on Rollouts adoption
related talks
27mRevolutionizing Continuous Deployment: How Databricks Integrates ArgoRollouts to Achieve Zero-Downtime Releases
25mLeveraging Argo Workflows and Argo Events for Cloud Agnostic Enter...
07mArgo in an AI-Native World: Effective In-Cluster Machine Learning and A...
18mTailored Deployment Strategies: Argo Rollouts for Diverse Applications
25mHow to Multiply the Power of Argo Projects By Using Them Together
29mTuning Argo Rollouts for Thousands of Workloads
25mSailing Across the Ocean of Argo Ecosystem Like a Pro
22mReal Time Argo Rollouts Analytics Powered by Notification Engine