Running a serverless platform on top of Kubernetes sounds straightforward in theory, until real-world scale exposes the limits of autoscaling, scheduling, and cluster design.

In this talk, I'd like to share our journey building Elastic Cloud Serverless on Kubernetes and the operational challenges we encountered while running highly dynamic, multi-tenant workloads at cloud scale. I’ll walk through the architectural decisions behind our Kubernetes-backed control plane, including why we adopted a multi-cluster, cell-based design, where standard HPA/VPA fell short, and how we approached custom autoscaling, provisioning, and lifecycle management.

Rather than focusing on product features, this session dives into practical lessons and trade-offs: what broke under load, what worked in production, and what we would do differently if we were starting today. We’ll also cover networking considerations, capacity planning, and the realities of operating large Kubernetes fleets reliably and cost-effectively.

This talk is aimed at platform engineers, SREs, and operators who are building or running Kubernetes platforms at scale and want concrete, experience-driven insights into designing resilient and scalable cloud-native systems.