I’ve come across comments online and in older akka docs re: Fusing.aggressive() and how that may help reduce materialization costs in, for example, a web service gateway application that materializes a new stream for every request. However, it appears that class is no longer in akka 2.5.+.
Have the auto-fusing costs during materialization been improved since then to make that pre-materialization fusing no longer of any benefit?
thanks
Having a separate fusing operation was very much an artifact of how the old materializer was designed. The redesigned materializer (PhasedFusingActorMaterializer, completed spring 2017), also encompass how the streams are laid out, tries to make as much of the work already being done when constructing the graph.
I don’t have any numbers on hand for the improvements, I remember them as substantial though. They could be checked by running the materialization benchmarks in the akka-jmh-bench on 2.5 and 2.4 respectively if you want to dig into it.