QUESTION: do workloads drain automatically if running in EKS and using SPOT instances?

[jobcespedes]

Do workloads drain automatically if running in EKS and using SPOT instances?
I found settings.ecs.enable-spot-instance-draining ( from #1100 by @mkulke) in the documentation but I am not sure it should be set for EKS or just ECS. In addition it made me wonder if draining is already done in EKS for spot instances. Thanks

[arnaldo2792]

Hello there! Sorry for the late response.

I found settings.ecs.enable-spot-instance-draining

Configurations within the ecs key will only work for ECS images. If you try to set it through user data your k8s hosts won't be able to boot since the configuration doesn't exist in the API model for k8s images. On the other hand, if you try to use apiclient you will get an error saying that the key ecs is an invalid key.

In addition it made me wonder if draining is already done in EKS for spot instances

It seems like there is already some guides to use EC2 spot instances in EKS:

https://aws.amazon.com/premiumsupport/knowledge-center/eks-spot-instance-best-practices/

[jobcespedes]

Thanks for your reply. It is now clear for me the option is only for ECS.

It was my understanding that draining is done out of the box with EKS + managed groups, since managed node groups doc says so. I just wanted to make sure it remains the same when using Bottlerocket (EKS + managed groups + Bottlerocket). Seeing settings.ecs.enable-spot-instance-draining made me hesitate and considered asking here.

[arnaldo2792]

Yes! Bottlerocket does support the features offered by managed node groups 👍 (we try to keep feature parity with what the EKS optimized AMI)