Let’s unroll my tweet (yes, I’m not over saying Tweet even with X), where I stressed Karpenter with EKS and Qovery.
⚠️ This is a stress test where I spent 1:30 hour; it’s not a benchmark. I’ll probably run some benchmarking to compare with Kubernetes node-autoscaler at some point, but not today :)
Setup
- 30 environments
- Each environment: 3 containerized apps + 2 container databases (Redis and Postgres) with EBS storage
- 1 EKS cluster with Karpenter deployed with Qovery
If you’re not familiar with Karpenter — watch my quick intro. But in a nutshell, Karpenter is a better node autoscaler for Kubernetes (say goodbye to wasted compute resources). It is open-source and built by the AWS team. Qovery is an Internal Developer Platform I’m a co-founder) that we’ll use to spin up our EKS cluster with Karpenter.
Let’s see how things behaved.
30 Environments Deployed
I created and deployed my 30 environments — and we can see that my apps are scheduled on my cluster. The EC2 nodes are dynamically allocated to the cluster (we can see t3.smallinstances). I don’t even have to choose the instance type. Karpenter is smart to choose the right instance depending on the following criteria:
- The workload of my apps
- The instance types available
- The best choice in terms of performance vs. costs
It’s truly a game-changer!
Note also that I’ve requested to allocate spot instances. (configurable via Qovery cluster interface)
Dynamic Node Allocation
I changed the CPU and RAM of one app from my env-1 environment to see how Karpenter will reallocate resources.
- From: 250m CPU and 256 MB RAM
- To: 2000m CPU and 8GB RAM
It just works nicely, and we can see that Karpenter scheduled a t3a.xlarge instance to handle my app requirement.
Cluster Scaled Down
If you are familiar with the default Kubernetes node autoscaler — you know how hard it is to downscale your cluster. For many reasons that I’ll not enumerate here… But in the case of Karpenter. It just works once again!
I stopped all my environments, and as expected, my cluster is down-scaled in less than 10 minutes as well. So I don’t have wasted resources!
Cluster Scaled Up.. Again
I wanted to see the cluster scaling up from 3 nodes to x nodes and see how fast it was... In less than 15 minutes, I got all my nodes provisioned and my 30 environments up and running.
Use Cases
At Qovery, we have hundreds of customers running testing environments with our Preview Environment product. Karpenter will solve many issues that they were facing previously. Just to mention a few...
- No need to allocate a specific node pool size now
- Cluster Scale-down is well handled, and no more resources are wasted (reducing AWS costs)
Interested in giving it a shot? Sign up to Qovery and create your EKS cluster with Karpenter :)
Top comments (0)