I'm using K8s Deployment API. When applying update I use RollingUpdate strategy and works perfectly.
However when I do "rollout undo", k8s kills all of the pods and create new ones, i.e. the rollback is disruptive.
Is there a way to configure "rollout undo" to use a RollingUpdate (to previous version) as well?
Rollback should use the same strategy you specified (i.e. RollingUpdate). The behavior you saw isn't expected. Can you provide the configuration (yaml/json) of your Deployment?