max-pods now taking into account CNI custom networking?

[jwelch649]

Hello - We've been trying to solve an issue with the max-pods value on bottlerocket nodes not taking into account the use of CNI custom networking (which reduces the available ENIs by 1). The solution we've leveraged is calculating max-pods in a bootstrap container.

We just noticed that an EKS cluster (K8s 1.27 and ami-0b312c9ed12d199b0) that is using CNI custom networking appears to be calculating the correct max pods by default (without the bootstrap container).

Has there been a change that allows for bottlerocket to know that CNI customer networking is in use and to calculate max-pods appropriately?

Thanks!

[jwelch649]

Jumped the gun a bit. Found out EKS managed node groups do the correct max-pods calculation based on CNI config during creation.

https://aws.amazon.com/blogs/containers/amazon-vpc-cni-increases-pods-per-node-limits/

Our cluster that 'magically' calculated max-pods correctly had a managed node group, so all was well. For our self-managed node groups, it appears we still need to leverage the bootstrap container method to dynamically calculate max-pods.