Connections are not disposed when server is closed

[mostroverkhov]

Expected behavior

Connections are disposed on DisposableServer.disposeNow(), and Connection.onDispose() Publisher is terminated

Actual behavior

Connection.onDispose() Publisher is not terminated

Steps to reproduce

Code below is timing out

@Test
 public void nettyServerDisposalIssue() {
 MonoProcessor<Void> serverConnDisposed = MonoProcessor.create();

DisposableServer boundServer =
    TcpServer.create()
        .port(8082)
        .doOnConnection(
            serverConnection -> serverConnection.onDispose()
                .subscribe(serverConnDisposed))
        .bind()
        .block();

Connection clientConnection =
    TcpClient.create()
        .addressSupplier(boundServer::address)
        .connect()
        .block();

boundServer.disposeNow();
serverConnDisposed.block(Duration.ofSeconds(5));

Reactor Netty version

0.8.2.RELEASE

JVM version (e.g. java -version)

Java(TM) SE Runtime Environment (build 1.8.0_191-b12)
Java HotSpot(TM) 64-Bit Server VM (build 25.191-b12, mixed mode)

OS version (e.g. uname -a)

Linux desktop 4.4.0-138-generic #164~14.04.1-Ubuntu SMP Fri Oct 5 08:56:16 UTC 2018 x86_64 x86_64 x86_64 GNU/Linux

[smaldini]

Interestingly this is a costly thing to do and thats why Netty doesnt do it by default.
We would need to attach a callback for all connections (and remove them when they are disconnected). I'm not sure this should be a default behavior but something like an option could enable unless we find another algorithm. Would be worth asking Netty team too.

[smaldini]

Looking a bit further, Netty will mark the channels as closing if we shutdown the worker loops from the server (which we don't because they can be shared). I wonder if we should force such shutdown when a server is disposed (and let LoopResources rehydrate new loops for following use).

[dconnelly]

Does this mean we expect outstanding connections to be abruptly terminated when a server is disposed? If so, should we consider adding support for graceful shutdown with connection draining?

[violetagg]

@dconnelly We plan to shutdown the EventLoopGroup invoking EventExecutorGroup#shutdownGracefully(). Do you need a specific configuration for the quiet period and the timeout?

[dconnelly]

@violetagg This is what I was hoping would happen thanks for the confirmation. Default values (2 second quiet period, 15 second timeout) are fine for my use case at least.

[violetagg]

Approach with channel group will be implemented for TCP/HTTP server (as described here for the client):
#581 (comment)

As a workaround for now - doOnConnection and channelGroup.add(connection.channel()) can be used.

This issue will depend on #586

152476f is a test case using channelGroup approach

[violetagg]

Feature is made available as part of #1022