Few adjustments on the Artemis Test

[franz1981]

I've filing the 2 things I believe should be fixed on the ActiveMQ Artemis test:

• netty-connector and netty-acceptor are the same and clients are connecting to netty-acceptor: I suggest to keep netty-acceptor as the acceptor responsible to deal with clients connections and define a new acceptor (let's call it cluster-acceptor, configured with protocols=CORE and remotingThreads=1) that would handle the connectivity with the other broker (could be either master or slave), and netty-connector should be use this cluster-acceptor address instead
• as said on [Artemis] Migrate to 3 masters topology #47 (comment) the test shouldn't use the other live brokers to distribute the traffic, given that Artemis won't guarantee automatic balanced resource partitions

The former is quite important to guarantee good results given that replicating is mandatory to return backup responses to clients and having a separate connection with its own thread guarantee to be able to response promptly.

[franz1981]

I see that the former proposal cannot be achieved with the current broker behavior while updating client topologies, so I'm going to fix at least the latter.

The former won't be a issue when apache/artemis#3479 will be merged ;)

[franz1981]

#57 (comment) isn't correct on:

I see that the former proposal cannot be achieved with the current broker behavior while updating client topologies

This could be achieved by disabling the topology updates of clients and just putting the list of brokers (ie live, backup urls) statically into it.