I have a use case where one of our apps is very memory heavy and need a big c5a.xlarge instance to run, but the other ones don’t need as much memory to run.
So, in my cluster, is it possible to define different instance types for each application? For example, keep the c5a.xlarge for the memory heavy app, and use t3.medium for the other less demanding apps.
As of today, that’s not something Qovery supports. But we are working on it and it should be available at some point in couple weeks / months using Karpenter on AWS.
Just for curiosity, why do you want to run
c5a.xlarge instances for all? It will avoid wasting resources and will need less nodes than using
What’s your actual node usage on the cluster?
Thanks for your answer.
I guess you mean why don’t I want to run
c5a.xlarge for all?
Well I’m having another problem with the auto-scaling: Auto-scaling with multiple apps - #7 by Pierre_Gerbelot and because of that, it’s costing us a lot of money.
So as we are working on trying to fix the auto-scaling issue, I thought that having an issue with auto-scaling on
t3.medium instances would cost less than an issue with
c5a.xlarge nodes ^^
Side note: Our Managed GCP offer supports GKE Autopilot, where you can have multiple types of instances for your Kubernetes cluster. So what you expect will work on GCP - but as mentioned @bchastanier, we’re working on it to provide this feature to AWS EKS with Karpenter.