im-server | IM application , including server | Websocket library

Based on this

I can see from other testing I'm running that this isn't just going slowly and that the messages just aren't being taken from the channel.

I think you have a live lock in addRedisSender()

The select statement will pseudo randomly select one of the cases, either the killSwitch case, or the <-messageChannel. Except there's also another case, the default. This will always be true, meaning the for { loop will spin for ever consuming all the resources and causing a live lock as the go runtime tries to schedule more and more competing go routines.

If you remove the default: continue case then the for loop will block on the select until there's a message to read, or a kill signal.

Unfortunately vendor support was unable to provide an answer as to why there was a conflict between JAX-WS RI and saaj but we received information on what breaks if we remove that jar.

We ended up electing to remove that saaj jar and overhaul some legacy, existing SOAP code to use JAX-WS RI. Doing so cleaned up a considerable amount of request message scaffolding work that was being done manually with strings (because of no domain visibility).

I don't particularly like this answer but will mark this question as answered since we've made a decision and will move forward with it. We'll just have to ensure that future product updates do not reintroduce this jar file back into the environment. It's possible one of the other implementations like Apache's CXF will have better luck.