Users running workloads in a bin-packed topology seeing some pods not schedulable
Article ID: 440486
Updated On:
Products
Issue/Introduction
When deployments with resource requests are created, users will see that some or all the pods are not schedulable because nodes don’t have capacity to accommodate them.
This only applies to deployments which specify resource requests which exhaust the total allocatable resources of the worker node. For eg, deployments with 2 replicases requesting 1 CPU each when the node has 2 CPU capacity.
Environment
VKr Releases >=1.34.1
Cause
In VKr releases older than 1.34.1, VKS system components didn’t have resource requests defined. To improve resiliency of workloads and make sure correct resources are accounted for every component, resource requests have been added in VKr releases 1.34.1 and higher.
Resolution
Reduce resource requests to fit the "allocatable" minus "already reserved by system components" resources of the nodes or increase the node pool size.
Additional Information
Feedback
