All components are installed in all servers but they are not necessarily activated, because special configuration is done.

In order to have high availability on the services, it’s important to maintain the rule N/2+1 rule, which means that more than half of the servers must be up, in order to provide service.

In this case, considering N = 3 ABIS node servers, you HA will work as long as more than the half of the servers is up, applying the formula, it’s required a minimum of 2 servers to keep operations up.

Let's assume this is the cluster configuration:

So the HA scenarios, keeping minimum 2 servers alive, (and all the template partitions at least in two servers) could be:

1. Server 001 goes down

   Server name	AMQ	Manager	Extractor	Dispatcher	Matcher
   Node 1	GONE	GONE	GONE	GONE	GONE
   Node 2		ACTIVE	ACTIVE	ACTIVE	ACTIVE
   Node 3	ACTIVE				ACTIVE

2. Server 002 goes down

   Server name	AMQ	Manager	Extractor	Dispatcher	Matcher
   Node 1	ACTIVE	ACTIVE	ACTIVE	ACTIVE	ACTIVE
   Node 2		GONE	GONE	GONE	GONE
   Node 3	BACKUP				ACTIVE

3. Server 003 goes down
   
   

   Server name	AMQ	Manager	Extractor	Dispatcher	Matcher
   Node 1	ACTIVE	ACTIVE	ACTIVE	ACTIVE	ACTIVE
   Node 2		BACKUP	BACKUP	BACKUP	ACTIVE
   Node 3	GONE				GONE

   
   

As you can see, all the combinations following the N/2+1 rule, allows the system to continue operational, but if more than one server is down, then it would not be possible to maintain in memory not only the services but neither the three template partitions in memory; as the whole applicant database is to be split into three partitions, to have a mirror factor 2 (two entire copies of the applicant’s templates distributed in three partitions across the ABIS’ cluster nodes) as follows:

The ABIS database is partitioned, on this case, 3 partitions, as follows: