This is a combination of a few things I've noticed, and resulted in me pulling the reliable shutdown patch - for now.
My attempt at adding some sort of automatic teardown is fraught with interesting corner cases, but I think also potentially highlighting some issues in our test cases.
[ ] Test cases hang on shutdown as we often take down the server before the client. Client tries to cleanup nicely and ends up stuck and timing out as there's nowhere to send the appropriate goodbye message to.
[ ] Proxied traffic is getting torn down incorrectly. When the proxy tests are torn down, order of shutdown matters too. But if the proxy server is torn down first, it, thinking its incorrectly responsible for a dialog, will send an independent shutdown message without the client's knowledge.
This is a combination of a few things I've noticed, and resulted in me pulling the reliable shutdown patch - for now.
My attempt at adding some sort of automatic teardown is fraught with interesting corner cases, but I think also potentially highlighting some issues in our test cases.