Add gpu taints toleration

aws-samples / aws-efa-eks

Deploying EFA in EKS utilizing GPUDirectRDMA where supported

MIT No Attribution

37 stars 19 forks source link

Add gpu taints toleration #3

Closed VoVAllen closed 3 years ago

VoVAllen commented 3 years ago

P4d instances are also frequently labeled with gpu taints. Should also tolerate such scenario

amrragab8080 commented 3 years ago

The k8s-eks plugin is not exclusive to GPU instances it can work for CPU based instances with EFA enabled.

VoVAllen commented 3 years ago

The toleration means thr pod can be scheduled to the node with nvidia.com/gpu taint, doesn't mean it can only be scheduled to gpu node. Pod with such toleration can still be scheduled to cpu nodes. Also I believe since there's already node affinity, here toleration should tolerate all the taints for the corresponding node