ARTEMIS-6142 FollowUp to fix tests

[clebertsuconic]

The purpose I'm sending this PR is basically to open a discussion.

Certain failover tests are now failing with InVM. This is a major concern for Wildfly integratoin as they have a lot of tests doing that.

For reference run LockManagerBackupSyncJournalTest, and you wlil see the faliure without this change on this PR.

@MrEasy / @jbertram

[MrEasy]

Can confirm the LockManagerBackupSyncJournalTest showing this issue.
Did ran only mvn -Pfast-tests install so far, strictly following the guide ;)

Can also confirm test succeeds with your commit.
Have to wrap my head around the change again, but would agree, that in VMConnector the disconnect should always call connection.disconnect.

edit:
see below
or should it not better be like this? LockManagerBackupSyncJournalTest succeeds with that. Think in case of failed, connection should get closed and otherwise disconnected (was the other (wrong) way around in my commit):

   public void disconnect(final String connectionID, final boolean failed) {
      if (!started) {
         return;
      }

      Connection conn = connections.get(connectionID);

      if (conn != null) {
         if (failed) {
            conn.close();
         } else {
            conn.disconnect();
         }
      }
   }

That wrong logic would then however also be present in InVMConnector

edit 2:
nah, disconnect raises failed=true, so passing this along seems correct to me

[clebertsuconic]

@MrEasy so, you think my fix is good with your changes?

[clebertsuconic]

I'm keeping the best of both worlds here. for a case where this is used on serverLocator.connect() the acceptor will close connections like it used to. while keeping your changes for regular clients.

[clebertsuconic]

we are doing a release very soon... and I reverted ARTEMIS-6142. Reapplied in this PR with some follow up that will need take place.

[MrEasy]

Sounds good to me - have to say not much experience with the mechanics in a clustering scenario.
Definitely still avoids the InVMConnections from leaking after close, which is the main point for me.