The documentation doesn't indicate that this is not needed for read only end points.
As read only instances can come and go it seems like a good thing for it to monitor the connections to make sure they are functional as well as pick up new instances if the cluster is scaled up.
However when using the :aurora: option on readonly endpoints, the connector seems to hunt for a write endpoint and when it cannot find one, it gives up.
We want to use the readonly endpoint because we frequently load the write instance so heavily that it would not be responsive for our readers. We need the readers to avoid the write instance.