We have exposed more parameters to the cluster-autoscaler, allowing for more fine-grained control. initially, only the scale_down_utilization_threshold could be configured. Now this is extended with the following parameters:
balance_similar_node_groups
expander
max_graceful_termination_sec
max_node_provisioning_time
ok_total_unready_count
max_total_unready_percentage
new_pod_scale_up_delay
scale_down_delay_after_add
scale_down_delay_after_delete
scale_down_delay_after_failure
scan_interval
scale_down_unneeded_time
scale_down_unready_time
scale_down_utilization_threshold
skip_nodes_with_local_storage
skip_nodes_with_system_pods
max_empty_bulk_delete
You can check what each of these parameters control in the upstream documentation.
These are available on both Azure AKS and AWS EKS platforms.