Answer
Even with the expected Pod CIDR range set correctly in configuration, Kubernetes can still assign pod IPs from the wrong range (for example, overlapping the local/database network range) if that range wasn't actually applied during the Kubernetes installation step itself — running the installer's pod-IP-range-correction resource step after Kubernetes is up (but before finishing the rest of setup) can retroactively fix the assigned range. Separately, confirm the middle tier server's firewall (UFW on Ubuntu) isn't active and blocking pod-to-database traffic, since that's a second common contributor to this same class of database-connectivity failure during install.