Just got into this exact issue when trying to connect with a non existent virtual host. Will try to work on a pull request now.

Please look my PR. There was some problem in there tests I never managed to fix....

I think I found the issue. Even though what you did was the right thing, it conflicted with other close and error events that were added to the socket. I added some more code based on your PR and now the connect method behaves properly and every test passes.

Processing leak after rejection

We had a similar problem with an integration test. In our case, the symptom: connect() produces a Promise (or, more precisely, a BlueBird) which is duly rejected a bit later. Our test waits for that rejection to happen. So far, so well. But then, the test framework complains that some background activity is still ongoing, even though the test has finished.

Our expectation is: At the time the promise rejects, all resources have been freed and all background processing pertaining the connection attempt has ceased. Our test framework (we use Jest) waits for another second, and in that second, not all background processing has terminated.

Details

We have a standard RabbitMQ running via

docker run --rm -ti -p 127.0.0.1:5672:5672 -p 127.0.0.1:15672:15672 rabbitmq:3-management

and try to connect to it, using the default existing user and password, but also using a vhost that does not exist.

The pertinent part of our (typescript) code is this:

  const connectionConfig: Options.Connect = {
    protocol: 'amqp',
    hostname: '127.0.0.1',
    port: 5672,
    username: 'guest',
    password: 'guest',
    heartbeat: 2,
    vhost: 'no_such_vhost',
  };
  const amqplibConnection = connect(connectionConfig);
  await expect(amqplibConnection).rejects.toThrowError();

amqplib v0.9.0 includes PR 647 which ends and destroys the socket.