long delay in Connection creation on cancellation

[JanEggers]

Describe the bug

I discovered this while testing. when a tests fails due to some assert I dispose a service that manages my rabbitmq clients. first thing in there is triggering a cancellation token so all initialization attempts are aborted. the rabbitmq client is stuck in this callstack for more than 10 seconds for each client so my tests run into a timeout inside of dispose:

Reproduction steps

whenever the time is less or more it seems to work properly on my machine so you may need to adjust the delay to hit the stack mentioned above.

[TestClass]
public class RabbitRepro
{
	private static readonly RabbitMqContainer RabbitMqContainer = new RabbitMqBuilder("rabbitmq:4-management")
		.Build();

	[TestMethod]
	public async Task TestCreateClientWithCancellation()
	{
		await RabbitMqContainer.StartAsync();

		var factory = new ConnectionFactory
		{
			Uri = new Uri(RabbitMqContainer.GetConnectionString()),
			AutomaticRecoveryEnabled = true
		};

		using var cts = new CancellationTokenSource();
		cts.CancelAfter(TimeSpan.FromMilliseconds(20));

		await factory.CreateConnectionAsync(cts.Token);
	}
}

Expected behavior

CreateConnectionAsync and its internals should respect the cancellation request and throw as soon as cancellation is requested and not be stuck somewhere for a long time.

what I have seen while debugging is that Channel.DisposeAsync calls CloseAsync which is stuck because its waiting for a response from the server which is not comming because the underlying connection is already closed / not opened yet

Additional context

rabbitmq client 7.2

[JanEggers]

I just added factory.CreateConnectionAsync(cts.Token).WaitAsync(cts.Token) to our code so we are not stuck in the timeouts. but there are still tests failing with timeouts and other tests failing because of resource in use exceptions because of lingering connections. so i guess there are more cancellation related issues. sadly dotnet no longer has a simple way to capture callstacks of different threads. that makes it quite hard to diagnose. if you have suggestions let me know. I will keep you posted if i have more findings.

a little more context: we have over 1500 tests and up until lately we mocked rabbitmq clients. im now trying to switch to testcontainers to have a more realistic coverage. I create vhosts for testisolation and use pooling so i dont have to create 1500 vhosts. is this the right approach or should i get rid of the pooling and just create 1500 vhosts? that would at least get rid of the resource in use issue.