HPA supports implicit maintenance mode deactivation which means that setting target replicas to 0 will deactivate HPA [1].

At this moment we mention in the documentation to delete HPA for upgrades but if properly managed by Operator the upgrades could work with HPA not requiring the users to remove the HPA temporarily.

Currently there seems to be issue when users updates cluster version while the HPA exists. The original cluster is removed but the new one isn't created. The Operator doesn't create new StatefulSet till next reconciliation is triggered.

[1] https://kubernetes.io/docs/tasks/run-application/horizontal-pod-autoscale/#implicit-maintenance-mode-deactivation