I don't understand the approach, really. They manage a 25+ node cluster of RethinkDB, but are reluctant to introduce a message broker? In my experience, a mess

I don't understand the approach, really. They manage a 25+ node cluster of RethinkDB, but are reluctant to introduce a message broker? In my experience, a message broker isn't such a big operational burden, even more so if it doesn't have to persist messages. And a message broker with pub/sub doesn't make the architecture necessarily more complicated. Somehow that seems like optimizing along the wrong axis.