Taming the Beast: Managing the day 2 operational complexity of Kubeflow
summary
Practical notes for running Kubeflow: the manifests repository is reorganized into apps / common / contrib with component manifests moved upstream so distributions can build from a single tested baseline; installers shift from kfctl/kfdef toward kustomize-based installs; an available Kubeflow Operator on Operator Hub watches and repairs child resources; kfserving keeps its manifests in its repo with a Kubeflow overlay for integration; expect Istio 1.9 in Kubeflow 1.3, watch for stale webhooks, prefer external databases for persistent state, and use Prometheus/Grafana/Alertmanager for monitoring.
AI-written summary · about
chapters (21)
- 0:00Intro
- 0:58Overview
- 1:49Challenges
- 2:31Taming the Beast
- 2:57Kubeflow Deployment
- 5:28Kubeflow Manifests Restructure
- 6:48How This Can Help
- 9:06Another Deployment Option
- 10:18Kubeflow Operator
- 11:35Day 2
- 12:52Kubeflow Component Update
- 14:25Stale Webhooks
- 14:50Securing Your Deployment
- 16:26Integration with Different Istio
- 16:49Updates / Upgrades
- 17:50External DB
- 18:32Kubernetes Version Update
- 21:30Troubleshoot Platform
- 22:15Monitoring
- 22:50The Future
- 23:52References
related talks
05mCI/CD Pipelines and Machine Learning
06mMLOps, Kubeflow, and Tekton
33mCentralizing Kubernetes and Container Operations
38mAutomated & Observable Deployments with Flux + Keptn
27mUnder the Hood with Fluent Bit Operator: Kubernetes-native Log Processor
34mManaging Machine Learning in Production with Kubeflow and DevOps
37mBuilding a Managed Platform While Maintaining a Good Developer Experience
22mDevOps in Data Science: What Works and What Doesn’t