Using HA Cluster to Manage Fabric
Contrail Networking Release 1911 supports High Availability (HA) cluster to manage fabrics.
In earlier releases, the All-in-One (AIO) cluster that contains the key components to run Contrail Networking on a single server, was used. After the fabric device discovery process begins, the AIO server becomes the DHCP server.
With the introduction of this high availability scenario, the DHCP server (dnsmasq) runs only during the zero-touch-provisioning (greenfield onboarding) process. After the fabric onboarding process is complete, the config files that are generated by the device manager and applied to dnsmasq, are deleted. This ensures that dnsmasq and device manager are active only on one node. After the files are deleted, the dnsmasq does not serve any more clients on the ZTP network.
In earlier releases, lease file records are maintained in
/var/lib/dnsmasq/dnsmasq.leases. Starting in Contrail Networking Release 1911, lease
file records are maintained in an external storage called Cassandra
You cannot use both AIO cluster and HA cluster at the same time to manage the same fabric.
Figure 1 shows an HA topology for ZTP. All spine and leaf switches, controllers, and nodes that are part of the deployment, are on the same ZTP subnet. All controllers are connected to a single ToR switch. The HA cluster is connected to a leaf switch.