search

Azure / AKS

Azure Kubernetes Service
https://azure.github.io/AKS/
1.97k stars 311 forks source link

Mig Autoscaling not supported #4216

Open KeerthiPulle opened 7 months ago

KeerthiPulle commented 7 months ago

Cx has configured multi-instance GPU nodepool. Create a multi-instance GPU node pool in Azure Kubernetes Service (AKS) - Azure Kubernetes Service | Microsoft Learn

When workloads are scaled up and there is a need to scale up the nodepool, autoscaling the nodepool does not happen but manual scaling works for customer. Cx is looking for a feature request to get the autoscaling support in such scenarios.

KeerthiPulle commented 5 months ago

Hi @kevinkrp93

checking if there is any update on mig autoscaling feature

dsp05 commented 2 months ago

The doc does say "You can't use Cluster Autoscaler with multi-instance node pools.". But I have been doing some experiments and it seems the autoscaler works for MIG node pool on AKS. It can scale up the node pool for pending pods and scale it down when it's completed. Is the doc out dated and is there any official support for this? Thanks!