Ideally yes, it's running on a different machine. But most certainly a different process.

For the queue, I've wrapped redis and rabbitmq clients with a simple retry logic that re-enqueues the job on failure to send. It's about choosing what's right for your use case.