[EKS/Fargate] [request]: Support for pod anti-affinity on fargate

[lennartt]

Community Note

• Please vote on this issue by adding a 👍 reaction to the original issue to help the community and maintainers prioritize this request
• Please do not leave "+1" or "me too" comments, they generate extra noise for issue followers and do not help prioritize the request
• If you are interested in working on this issue or have submitted a pull request, please leave a comment

Tell us about your request EKS fargate used in combination with the pod anti-affinity feature does not seem to work as expected. Potentially this is because the fargate provisioned nodes are only added to the cluster at the time of scheduling?

Which service(s) is this request for? EKS on fargate

Tell us about the problem you're trying to solve. What are you trying to do, and why is it hard? I am trying to deploy an application with high availability over 2 AZ's on Fargate. I tried using the PodTopologySpread feature, but it seems this is currently not supported either: https://github.com/aws/containers-roadmap/issues/824 Am I right to conclude that you currently cannot deploy an application highly available over multiple AZ's on fargate and keep it balanced after scale out/in events?

Are you currently working around this issue? I currently have not found a workaround for Fargate.

[witspirit]

For our use-case it is not that important that the load ends up balanced (like with PodTopologySpread), but it is important that when at least 2 pods exist, at least each of 2 AZs hosts one of the pods. Such that we have at least 1 warm pod available in an AZ when one of the AZs become unavailable.

[rsmets]

This is a required feature for me to adopt Fargate with EKS. Curious if there has been any progress on this or a workaround via Fargate profiles?

Thank you.