Abstract
The LoadBalancer is built with two network interfaces in mind. This has been made in order to provide a better network architecture: administrations services are on an internal (usually private) network interface, and all user-defined services along with Virtual IPs are on a dedicated public interface.
The drawback of this approach is that you have to define two different network mask and some route definition. When this is not possible due to business rules, it’s possible to configure the load balancer to act as if it was with only one interface.
Before jumping to the configuration, here are some things to keep in mind to understand the changes:
- cluster replica is done through SSH connection, using the IP assigned on the internal interface;
- public IPs aren’t a strict requirement, if not useful for publishing services;
- Virtual IPs are bound to the public interface, and are managed through ARP messages.
How to configure
The first thing to do, is to make sure that both the interfaces of the cluster nodes are connected to the same “physical” (i.e. of the Hypervisor) network. Since we are trying to melt the two interfaces, the physical communication must be shared.
The configuration for the nodes could be the following (e.g. network 192.168.1.0/24 with gateway 192.168.1.254/24):
| Configuration | Master node | Backup node |
|---|---|---|
| Private IP | 192.168.1.12/24 | 192.168.1.34/24 |
| Private GW | default 192.168.1.254/24 | default 192.168.1.254/24 |
| Public IP | 127.1.0.1/8 | 127.1.0.2/8 |
| Public GW | not set | not set |
| Virtual IP | 192.168.1.56 | |
And that’s all. Note that local IPs are assigned to public interfaces and that internal interfaces are “published” just like the virtual IPs.
22 and 1443 must be limited to administrator IPs or network