During our migrations from KOPS to EKS clusters, some customer Pods had issues launching, due to hitting fs.inotify.max_user_instances and/or fs.inotify.max_user_watches limits. Turns out these sysctl have been raised from their defaults for the KOPS base images, but the EKS AMIs still use the OS defaults.
We have raised these limits accordingly in our own EKS images and will open a PR upstream.