Taming the Wild West of Research Computing: How Policies Saved Us a Thousand...
summary
Describes deploying OpenShift GitOps/Argo CD with ApplicationSet plus Kyverno and Kueue to stop GPU waste and noisy CPU jobs: automatic Argo Application creation with self-heal and pruning, Kyverno policies that add GPU-avoid anti-affinity for CPU-only jobs, block exec into GPU pods with an exemption config map and allowlist of rsync/tar for file copies, annotate creators for fast triage, and Kueue resource-flavored queues that yield fair GPU access; result: near-zero Slack support and predictable scheduling.
AI-written summary · about
chapters (12)
- 0:00Accelerated discovery research
- 1:49Research computing challenges
- 3:34Automating cluster goals
- 6:01Argo CD automation foundation
- 8:32Customizing user projects
- 11:02Namespace lifecycle automation
- 11:44CPU/GPU workload policies
- 14:14Blocking interactive pods
- 17:04Policy feedback refinement
- 18:55Resource tracking annotations
- 19:16Fair GPU access queues
- 22:03Benefits and key takeaways
related talks
25mPolicy-Based GitOps: How Policies Can Help Secure and Automate GitOps...
28mGPU sharing done right: Secrets, security, and scaling with Vault and vCluster
29mTowards Sustainable Multi-Cluster Management in Real-World Academic Infrastructure
30mGitOps: Keeping It Light, Bright, and Deployed Right with OpenShift!
26mGitOps Me Some of That! Managing Hundreds of Clusters with Argo CD
32mSmart GPU Management: Dynamic Pooling, Sharing, and Scheduling for AI Workloads in Kubernetes
19mPolicy-as-Code for LLM Inference: Cost & Security Guardrails
06mLightning Talk: Secure GitOps for Regulated Workloads: Argo CD Meets Confidential...