kube-batch
kube-batch copied to clipboard
Preemption/Reclaim enhancement
Is this a BUG REPORT or FEATURE REQUEST?:
/kind feature
Description:
Currently, the preemption/reclaim in kube-batch are not good enough; it's better to enhance it for elastic workload, e.g. elastic training of pytorch.
The idea is to leverage ROI (Return of Investment)
to enhance preemption/reclaim.