# CNCF End User TAB curates KubeCon NA 2026 sessions on Kubernetes at scale

> **Key takeaway:** The CNCF End User Technical Advisory Board has published its curated session recommendations for KubeCon + CloudNativeCon NA 2026, highlighting production war stories from Anthropic, Walmart, Salesforce, NVIDIA, and others running Kubernetes at massive scale.

**Published:** 2026-10-30T00:43:04+00:00 
**Source:** Cloud Native Computing Foundation (CNCF) 
**Category:** cloud-native 
**Canonical URL:** https://fosswire.org/news/cncf-end-user-tab-curates-kubecon-na-2026-sessions-on-kubern.html 

## Summary
The Cloud Native Computing Foundation's End User Technical Advisory Board has released its guide to KubeCon + CloudNativeCon North America 2026, taking place in Salt Lake City. The post highlights Birds of Feather sessions including newly formed end user groups for Platform Engineering, Public Sector, Telecom and Research, plus a live attempt to build a reference architecture for Agentic AI deployments. The Town Hall will remain in the project pavilion after positive feedback from Amsterdam.

The TAB members recommend several technically deep sessions drawn from production environments. Joseph Sandoval highlights talks on apiserver churn at Anthropic's tens-of-thousands-node clusters, a Salesforce-NVIDIA multi-cluster proxy that preserves kubectl semantics across fleet deployments, Isomorphic Labs and Google's work stabilizing bursty ML workloads against the Kubernetes scheduler with Kueue, and Verkada's three-year FedRAMP Moderate authorization journey. Juliano Martinez points to Walmart's zero-downtime migration from Canal to Cilium across hundreds of clusters, Adobe's eBPF-based network cost attribution across 400,000 pods, and sessions on saturated on-premises cluster allocation and secure agent execution at scale.

## Production Kubernetes war stories in the schedule
- **Apiserver scalability:** Anthropic's session covers read replicas and controllers for busy clusters, specifically addressing how LIST operations on tens of thousands of nodes triggered cascading 410 Gone relist storms that crashed clusters twice in two weeks.
- **Multi-cluster federation:** Salesforce and NVIDIA built a proxy layer atop OCM-style multi-cluster that makes fleet resources feel like a single cluster to kubectl, solving the problem where hub-template source-of-truth causes kubectl scale and edit operations to be silently reverted.
- **ML workload scheduling:** Isomorphic Labs and Google stabilized bursty ML workloads that timed out webhooks and took down control planes, using decoupled queues and Kueue tuning rather than the original scheduler configuration.
- **CNI migration at scale:** Walmart migrated from Canal to Cilium across hundreds of production clusters spanning cloud and on-premises, addressing Flannel route-drop failures beyond 300 nodes, iptables rule explosion, and etcd v2 incompatibility blocking upgrades.
- **eBPF cost attribution:** Adobe uses Retina eBPF agents to capture pod-level network flows across 400,000 pods in 31,000 namespaces across 85+ clusters, enriching with Prometheus node topology and classifying destinations into cost categories matching cloud provider billing structures.

## Why it matters
These curated sessions represent some of the most detailed public documentation of Kubernetes operating at extreme scale, with concrete failure modes and remediation strategies that platform engineering teams can adapt. The inclusion of FedRAMP authorization timelines, CNI migration playbooks, and GPU observability tooling signals maturation in areas where many enterprises remain stuck in early adoption phases.