With the client inactivity timeout increased LS shouldn't be disconnecting the clients. Back-pressure from ES shouldn't cause disconnection either.
If you run logstash with trace logging enabled this might help figure out why the client is disconnecting (--log.level trace).