Pods from deployment pin on same node - rancher 2.5.7, k8s 1.20.4

Rancher 2.x
1

After upgrade to 2.5.7, and RKE managed k8s 1.20.4 cluster. I am seeing many deployments has pods always pin to same node.

When pods from the same deployment/statefulset pin on the same node, they are prone to outage when the node went down. (I found this out when rebooting a host to test failover…)

This behavior was not observed in the past.

Any one running into this issue?

2

If you want that behavior, you have to put in an [anti]affinity rule on the workload.

Without that, the the scheduler can and will place them wherever it feels like, according to many non-obvious criteria that can easily change over time.

3

I realized that too, but it is bizarre that this problem started over update…