Highly available VIPs on OpenStack VMs with VRRP

Introduction

Virtual IPs (VIPs) are widely used to enable high availabilty for IT infrastructure services. Although the use case is very common, it is surprisingly tricky to set up in an OpenStack environment without a deeper knowledge about neutron and its port security mechanisms. In this blog post I will walk you through the creation of a pair of VMs that share a VIP via the Virtual Router Redundancy Protocol (VRRP) on an OpenStack cloud.

Neutron setup

My OpenStack project has an internal network (192.168.0.0/24) and a router connecting it to an external network (10.98.208.0/24). To setup a HA VIP I need 3 ports that I named vip-port, vm1-port and vm2-port. The VM ports will connect my virtual machines to the internal network, the VIP port actually is just a dummy port that allocates an internal IP address. Neutron will not instantiate this port, it s just a database entry that “blocks” an IP address that will later be used on the VM ports. I want both VMs to be able to communicate via their own IP and the VIP, but the neutron port security mechanisms will drop all traffic not coming from the IP and MAC address pair owned by the VM. Therefore I need to explicitly tell neutron to also allow the IP address of the VIP on the VM ports. This can be done with the --allowed-address-pair parameter on the neutron port-create command:

To allow VRRP traffic between the VMs I will now create a security group and assign it to the VM ports we created. Just for convenience reasons I will also add rules for allowing incoming SSH (for login) and ICMP (for ping) traffic.

As I want to access the VMs and the VIP from outside the cloud I create three floating IPs from my external network called floating and assign them to the three ports.

That is all that is needed within neutron.

Create the VMs

Now I am able to create two Ubuntu VMs and conenct them to the VM ports:

I wait until the VMs are running and log in to install keepalived:

To configure keepalived I lookup the name of the network interface on the VM:

With the interface name and the IP of the vip-port I can now continue to configure keepalived by editing /etc/keepalived/keepalived.conf:

Now I restart the keepalived service:

And repeat those steps on the second VM.

Checking the output of ip a on the VMs I see that one of my VMs assigned the VIP IP address 192.168.0.14 to its ens3 interface.

Testing failover

I open a terminal and ping the VIP:

In a second terminal I log in to the VM that currently has the VIP listed under the ens3 and stop the keepalived:

If everything worked I should now see that the ping in my first terminal is still working and checking ip a on both VMs shows that the VIP failed over to the other VM.