Tips and Tricks for Using Vertical Pod Autoscaler in Kubernetes

It’s essential to monitor and alert for any containers that get VPA recommendations exceeding the capacity of your worker nodes. I’ve set up a script that flags any VPAs suggesting over 88 CPUs so I can address potential issues before they become real problems!

I find it ridiculous that the only option for installation is via a bash script—it really needs to change. Also, I struggled with Grafana Mimir because it had way too many requirements compared to what I needed. Each component felt like a separate deployment, which added complexity. In my experience, EKS pod creation slowed down after I added it, so I eventually removed it. I’d love to try Goldilocks for better resource management.

Yeah, I hear you on Goldilocks! Just a heads up: watch out for the millibytes issue. People using Kubernetes metrics API sometimes report memory in millibytes instead of the usual bytes, which can mess up your scaling decisions. It’s tricky because it might appear that your container is using way more resources than it actually is. Just avoid using the "humanize-memory" option to keep things straightforward.

One crucial thing to remember is that you shouldn't deploy VPA and Horizontal Pod Autoscaler (HPA) at the same time. They can conflict with each other like dueling keyboards! Use VPA in recommendation mode first—it analyzes and helps you scale vertically without immediate deployment. Then, let HPA handle the regular autoscaling. If you're looking for something a bit more advanced, KEDA can be a solid choice as it extends HPA, but keep it simple if you're new to these tools.