Kubernetes resource limits: predictability vs. efficiency

Milan Plžík, a Senior Software Engineer at Grafana Labs, argues for the case of Kubernetes resource limits by examining the trade-off between predictability and efficiency in workload performance. While some advocate against using resource limits to maximize compute efficiency, Plžík suggests that setting limits can prevent performance unpredictability and help manage resource allocation more effectively, especially during high-traffic periods like Black Friday and Cyber Monday. He discusses strategies such as configuring workload limits to be slightly above requests, known as fixed-fraction headroom, or setting requests equal to limits for predictable performance. These approaches ensure a balance between resource utilization and stability, although they may lead to some resource wastage. Plžík emphasizes that while configuring limits is challenging and can lead to inefficiencies if done incorrectly, it provides valuable insights into workload behavior under stress and can help prevent catastrophic failures. He highlights that the decision to set limits should consider various factors, including workload type and resource availability, to achieve an optimal balance between resource efficiency and performance predictability.