Use new shutdown hooks in Akka 2.5 to prevent AbruptTerminationException in Http infrastructure

[jrudolph]

There are already some hook phases pre-allocated for exactly this purpose.

All active components should be shutdown in the hook:

• Http().shutdownAllConnectionPools
• Also introduce some infrastructure to track open server bindings and shut them down.

[jrudolph]

Also, see akka/akka#18747 tracking to fix the problem in the streaming infrastructure directly.

[jonas]

I think this is highly related with the shutdown improvements proposed in #54 and #188

[jrudolph]

Great, thanks for looking them up, @jonas.

[l15k4]

So the idea is that http.shutdownAllConnectionPools should be called internally when actor system is terminating but currently one has to do it explicitly ?

[mr-git]

We do not like this "harmless" error message either.
We are using Akka 2.5.6 and Akka-http 10.0.10
Currently for integration tests we have code like:

  override protected def afterAll(): Unit = {
    Http().shutdownAllConnectionPools().andThen { case _ =>
      materializer.shutdown()
      system.terminate()
    }.futureValue
  }

And still in some cases we manage to get that exception, sometimes several of them.
I added extra useless code to materialiaze responses, like response.entity.toStrict(3.seconds).futureValue.data.decodeString("UTF-8"), this allowed to remove few such errors, but still one is left for each test suite.

Are we doing something wrong? Is it possible, that we are leaving something streaming behind?
In https://doc.akka.io/docs/akka-http/current/scala/http/client-side/host-level.html I didn't find any other pointers.

According to #497 it seems that flow...via...map is preferred over Http().singleRequest(...) calls. Is that expected approach?