See #8329 (comment):

We do need to update the nodes here. It looks like we calculate write nodes based on pods/replicas, and then read nodes based on write nodes. This is assuming a constant ratio between read/write nodes, but that ratio will change in basic vs super query configurations.