AWS_VPC_K8S_CNI_CUSTOM_NETWORK_CFG and --use-max-pods

[cnelson]

I'm not sure if this issue belongs here or in awslabs/amazon-eks-ami but I'm running into an issue with the max pods being calculated incorrectly when running with AWS_VPC_K8S_CNI_CUSTOM_NETWORK_CFG

When using a custom network config where the pods are on a different subnet than the host then the number of the available address (if I understand this correctly) is not as described in eni-max-pods.txt:

number of ENI * (# of IPv4 per ENI - 1) + 2

but should actually be:

number of ENI -1 * (# of IPv4 per ENI - 1) + 2`

because the primary interface can't be used to allocate addresses on the custom subnet.

This leads to pods being scheduled onto nodes that then get stuck in ContainerCreating as this plugin can't assign an address once all the secondary interfaces have been used up.

What's the recommended way to handle this?

[cnelson]

Closing, this is a dupe of #331 - I didn't see it, apologies