Cluster-autoscaler by default evict all those pods -including those coming from daemon sets-
In the case of EFS-CSI drivers, which are mounted as nfs volumes, this is causing nfs stale and that application worloads are not terminated gracefully. 

4.11

- While scaling down a node from the cluster-autoscaler-operator, the DS pods are beeing evicted.

1.
2.
3.

CSI pods might not be evicted by the cluster autoscaler (at least prior to workloads termination) as it might produce data corruption

Is possible to disable csi pods eviction adding the following annotation label on the csi driver pod
cluster-autoscaler.kubernetes.io/enable-ds-eviction: "false"