Performance testing Kubernetes workloads, with Stephan Schwarz
May 27, 2025
Stephan Schwarz, a DevOps engineer at iits-consulting specializing in Kubernetes, shares invaluable insights on performance testing workloads. He discusses defining performance metrics and the methodology of testing individual pods to uncover their limitations. The conversation delves into the impact of shared Kubernetes components on results and the complexities of configuring Horizontal Pod Autoscaling. Stephan also highlights the importance of tools like OpenTelemetry for monitoring performance in production, emphasizing a holistic approach to testing and continuous learning in the DevOps landscape.
AI Snips
Chapters
Transcript
Episode notes
Define Performance In Business Terms
- 'Performance' must be defined with business context and measurable targets before testing.
- Different stakeholders may prefer lower latency or faster error feedback; record expectations explicitly.
Isolate And Document Single-Pod Limits
- Disable autoscaling and other dynamic features when measuring a single pod's capacity.
- Change only one resource or setting at a time and document every step and result.
You Test The Whole Chain, Not Just The Pod
- Load tests exercise the entire request chain, not just the application pod.
- Shared components like ingress, CNIs, service meshes and external DBs often become the true bottlenecks.