How Should We Effectively Manage Resources in Kubernetes?

It's essential to conduct performance tests and benchmarks as this helps identify appropriate resource settings. If your application sees huge spikes in memory and CPU during heavy loads, that indicates a potential software engineering issue. In a shared resource environment, chaotic scaling isn't ideal and could be seen as a bug.

Auto scaling and proper metrics are key here. Most services start small, and as usage grows, you gather data from tools like the metrics server to understand patterns. You can set initial resource limits to minimize blast radius and ensure services can auto scale based on demand. The vertical pod autoscaler (VPA) has also been introduced to help with this recently.

You really need to embrace a DevOps approach with this. It's not enough to have developers code things and pass the runtime decisions off to someone else. Teams should take responsibility for tuning their services and managing costs. If you're breaking things into microservices, you also have to handle the complexity that comes with it. Ideally, those managing services should know them inside and out to avoid random adjustments that could lead to downtime. It can be tricky, but with well-tuned alerts and a solid runbook, it can work out well.

Using observability tools during development and in production can be a game changer. They help you measure how resources are being used, giving you the insights needed to make informed decisions about resource management.