Unlocking Kubernetes Observability - Secure, Tenant-Centric Metrics for GPU Workloads

Learn to implement secure, tenant-centric observability for multi-tenant Kubernetes clusters with GPU workloads through this 35-minute conference talk from KubeCon + CloudNativeCon. Discover how Adobe's Tenant Exporter enhances monitoring by delivering curated namespace metrics built on Prometheus, exposing critical data including ingress requests, container CPU/memory usage, GPU utilization, and resource quotas. Explore the comprehensive architecture featuring Prometheus for metric collection, Nginx proxy for load balancing, and secure authentication through prom-label-proxy with kube-rbac-proxy integration. Watch a live demonstration of configuring self-service metrics for GPU namespaces and see how users can select specific metrics via ConfigMap while managing system load through quotas. Master deployment strategies, quota management techniques, and scaling approaches for metric delivery across multiple clusters, drawing from Adobe's real-world experience managing thousands of namespaces. Gain practical insights into best practices for Kubernetes observability using CNCF tools to reduce operational overhead while improving system visibility and GPU workload optimization.