[Feature]: Toleration configuration for device-plugin daemonset

[c-fteixeira]

Suggestion Description

Currently it is not possible to add tolerations to the daemonset created based on the DeviceConfig.

Operating System

No response

GPU

No response

ROCm Component

No response

[sajmera-pensando]

Thanks @c-fteixeira for filing this feature request. Could you please provide a bit more detail about the use case for which you need to add Tolerations to the DevicePlugin daemonset?

[ESlam-Yousry]

@sajmera-pensando there's a use case when you have some taints for your worker nodes,
For example { "effect": "NoSchedule", "key": "amd.com/gpu", "value": "present" }
so we need to allow daemonsets to tolerate with these taints

[taddeusb90]

@sajmera-pensando could you please point me to the repo that contains the internal reconciliation logic for tolerations? If it's not supported i would like to add support for it

[sajmera-pensando]

Hi @ESlam-Yousry @taddeusb90 , we will add support for it and will update this issue once its available. Thanks.

[ESlam-Yousry]

@sajmera-pensando Thanks! ,Super excited to have this ^^