Running Percona Operators in Production: Lessons from 40+ Databases per Cluster

Abstract

At b’nerd, we operate Kubernetes platforms for multiple clients and run Percona Operators in production environments with more than 40 databases per cluster. What started as a customer requirement for self-service backups and restores quickly evolved into our standard approach for Kubernetes-native database operations.

As adoption grew, so did the operational complexity. Exposing database operations through Kubernetes CRDs changed how these systems were used: backups became more frequent, restores became self-service, and concurrent operations became part of everyday platform usage.

In this session, we share practical lessons learned from operating Percona Operators at scale. We will discuss how reconciliation behavior, state transitions, and resource pressure become increasingly important as database density grows, and why observability needs to go beyond Kubernetes resource status alone.

Join us for a field report from production and learn about the operational patterns, scaling considerations, and unexpected behaviors that emerge when running Percona Operators in heavily utilized Kubernetes environments.

Slides

Work in progress :)

Presented at

• Upcoming: Percona Live, Amsterdam, September 2026