I would like to determine if a data storage scenario described below is possible with AWS EKS/Kubernetes. Here is the spec:
The goal here is to use the multi-attach EBS drive as a stand-in for a database. We want data to be accessible by several EC2 nodes, see note 1.
As far as I can tell, the EBS driver for EKS does not support multi-attach drives (stackoverflow, github issue).
So then my other option is to directly configure the nodes in the auto-scaling group to mount the EBS volume as part of their initial start-up. Then, use a local
PV type in the K8s manifest.
Is this second option technically viable?
1 Why are we not using an actual database, or an EFS? Because they are too slow. We have a bit of a special use case, where we are trying to establish a multi-gigabyte-per-second read rate between compute and storage, and on AWS that can only (?) be achieved with an IO optimized EBS drive directly attached to an EC2 node.
Version 1.25.0 of AWS EBS CSI Driver just released includes this functionality. Perhaps this is now a possibility for you?
https://github.com/kubernetes-sigs/aws-ebs-csi-driver/blob/master/CHANGELOG.md