@groved what use case would require scoping PD max limits on a per node level vs application level or global max?
All nodes in k8s are generally supposed to be interchangable, and identical so they can be ready to fill in for a failed node at any time. Mixed node types normally merit separate clusters.
@lrvick node selectors exist for this reason. You may want to have nodes that cost much more to run for specific resources. IE. I need a graphics card for this. But I don't want to pay $7/hr for all my resources that don't need a graphics card. Each instance type, especially in AWS has different limits on the number of Volumes and ENIs that can be bound to it. The default scheduler does not receive this data so it has no real idea how to manage that. This is why the recommendation is to move to (1/2)
@lrvick the latest version of EKS along with using the new AWS CNI for volumes so that it _can_ manage this better. But if you're current cluster is in contention and resources are not able to scale your SOL (2/2)
This Mastodon instance is for people interested in technology. Discussions aren't limited to technology, because tech folks shouldn't be limited to technology either!