StackGres offers an integration that allow to configure automatic scaling of instances both horizontally (number of replicas) and verticlly (CPU and memory).
Horizontal automatic scaling is based on KEDA that extends the HorizontalPodAutoscaler in order to scale the number of read only instances based on active connections statistics from the database itself.
Vertical automatic scaling is implemeted though the VerticalPodAutoscaler that depending on usage on each instance will allow to increase or decrease the CPU and memory resource requests and limits and thus re-create an instance in a node with more or less resources.
IMPORTANT: Note that using vertical autoscaling will disrupt the database service when the scaling is performed on the primary. High availability will minimize this disruption but your application must be prepared if you decide to use such functionality.
You can configure the automatic scaling of the cluter in the SGCluster CRD autoscaling section.