A more traditional route before HA probe was around is the following and if you want to ensure there is absolutely no lag or software failure with HA probe.
You can deploy 2 Hubs if your using Linux make them VM(s) you can get by with 2xCPU(s) and 4 Gigs of ram if your doing minimal QoS and just worried about alarm data ( I can push 3k Robots easy using those specs ) . Remember Hub sizing isn't a clean cut thing as several variables go into affect i.e how many alarms are being pushed and QoS metrics etc.
On the customers robot configuration you need to set the following inside the <controller> section :
hub = <name of hub i.e hub01>
hubip = <hub ip>
hubport = <hub port>
hubrobotname = <hub robot name>
secondary_domain = <UIM Domain>
secondary_hub = <hub name i.e hub02>
secondary_hubip = <hub ip>
secondary_hubport = <hub port>
Now you can also just do the following keys as the controller will restart itself if it finds the hub name different and or hub robot name different and write that to the configuration file before the restart.
hubip = <hub ip>
hubport = <hub port>
secondary_hubip = <hub ip>
secondary_hubport = <hub port>
After you have both your hubs up and running you can do post or get/attach queues from your primary hub to both hubs (hub01 and hub02) if your using post queues and HA probe there always can be a potential delay on when it brings up queues which you have to account for or you may lose the alarm / qos data. . There is limitations though on the amount of queues and or the amount of tunnels to take into consideration and both Linux and Windows Hubs have different limitations and not exactly the same and depending the amount of hubs and the size of the infrastructure some deployments need a relay hub to consolidate everything.
I hope this helps!