* Use EmitterProcessor in the FluxMessageChannel
The `EmitterProcessor` has a good logic to block upstream producer
when its downstream subscriber cannot keep up with overproducing.
* Rework `FluxMessageChannel` logic to rely on the `EmitterProcessor`
instead of `Flux.create()`
* Cancel `FluxMessageChannel` internal subscriptions in the `destroy()`
* Fix `ReactiveStreamsTests.testFluxTransform()` for the splitter's
delimiter
* Ensure in the `FluxMessageChannelTests.testFluxMessageChannel`
that we can have several concurrent subscribers to the
`FluxMessageChannel`
* * Use `flux.onComplete()` instead of iteration over subscribers
* Change `subscribers` list into just `AtomicInteger` count marker
* fix `DefaultSplitterTests` according a new logic in the `FluxMessageChannel`
* GH-3107: Add errorOnTimeout for TcpInboundGateway
Fixes https://github.com/spring-projects/spring-integration/issues/3107
The `MessagingGatewaySupport` has an `errorOnTimeout` option to throw
a `MessageTimeoutException` when downstream reply doesn't come back in
time for configured reply timeout
* Expose an `errorOnTimeout` option as a `TcpInboundGateway` ctor
property
* Add new factory methods into a `Tcp` factory for Java DSL
* Ensure a property works as expected in the `IpIntegrationTests`
* Document a new option
* * Use `delaySubscription()` for subscribing publishers in the `FluxMessageChannel`
to wait until this one subscribed.
* Use an `EmitterProcessor` to catch subscriptions and pass them as a
signal to delayed upstream publishers
* Fix `FluxMessageChannelTests.testFluxMessageChannelCleanUp` to
verify an actual property instead of removed.
* Fix `RSocketOutboundGatewayIntegrationTests` for the proper subscription
into a `FluxMessageChannel` before actual interaction with an RSocket
gateway.
This should help us also to avoid some race conditions in the future
* Revert "GH-3107: Add errorOnTimeout for TcpInboundGateway"
This reverts commit fa6119ddc4e423e8b15ca3fdc9fc4c79ea9d56af.
* * Refactor `FluxMessageChannel` to use `ReplayProcessor` for `subscribedSignal`.
This one is used `delaySubscription` for the upstream publishers
* Use a `AtomicBoolean` for subscription state since `doOnSubscribe()`
is called before `EmitterProcessor` adds subscribers for its `downstreams`
* Use `publishOn(Schedulers.boundedElastic())` for upstream publishers
to avoid blocking over there when our `EmitterProcessor` doesn't have
enough demand
* Refactor reactive tests to have a subscription into the `FluxMessageChannel`
earlier than emission happens for it
* * Use `Flux.subscribe(Consumer)` instead of `doOnNext(Consumer).subscribe()`
* * Emit `subscribedSignal` value after `.subscribe(subscriber)`
instead of `doOnSubscribe`
* Check for `this.processor.hasDownstreams()` before emitting such an event
* * Use `this.processor.hasDownstreams()` as a value to emit for `subscribedSignal`.
This way we are less vulnerable race conditions when subscribers are changed
actively
94e08169a5