Disposing Connection after closing it with timeout causes deadlock

[JanEggers]

Describe the bug

[TestMethod]
public async Task DisposeWhileCatchingTimeoutDeadlocksRepro()
{
    var factory = new ConnectionFactory
    {
        UserName = "rabbit",
        Password = "rabbit",
        HostName = "localhost",
        VirtualHost = "/",
        AutomaticRecoveryEnabled = true
    };

    var receiveConnection = await factory.CreateConnectionAsync();
    try
    {
        await receiveConnection.CloseAsync(TimeSpan.Zero);
    }
    catch (Exception e)
    {
    }

    // this blocks forever
    await receiveConnection.DisposeAsync();
}

Reproduction steps

see code

Expected behavior

Dispose should never block it must either throw or work

Additional context

Im currently migrating to version from 6.8.1 to 7.0 but this causes my tests to be stuck.

[michaelklishin]

@JanEggers Dispose for most objects, perhaps, should. In this client, Dispose has always been an equivalent of Close (or rather, Abort IIRC, which is the same thing but with all exceptions ignored).

Close can block, that's just the reality of distributed systems. It can also block then succeed.

There aren't any details as to what specifically blocks forever, do you have evidence of a deadlock or is the term just used generally to say "blocks forever"?

[michaelklishin]

Dispose in main seems to be doing roughly what I'd expect it to do.

So most likely the answer lies in

receiveConnection.CloseAsync(TimeSpan.Zero)

and how it handles timeouts. Unfortunately, CloseAsync is anything but trivial and it tries to handle half a dozen of different failures along the way.

@JanEggers do you have a thread state dump from when the method in question blocks?

[lukebakken]

Please continue discussion here, thanks: #1759