Exponential backoff is misinformation here. During the server startup, it reads a critical data, intentionally no retry, hence no backoff.

A better fix is to quickly spread the load to backup databases that already exist. There are other options too.