Is gracefull shutdown possible?

[wmww]

Is there a way to gracefully shut down the rtc_server.recv() loop without waiting for the next packet?

EDIT: I'm currently using futures::future::Abortable to kill it, seems like a better way would be optimal but this is working for now.

[kyren]

Right now (with one small commit that I'm adding) it's possible to implement a graceful shutdown on top of the current Server I think. The graceful shutdown paths are a very poorly tested, but I think it would go like this:

1. Hook up some kind of signal to the futures driving SessionEndpoint and Server::recv and select! on it in both cases.

2. Once the shutdown signal is received by the future driving the SessionEndpoint, stop creating new sessions (you can drop the SessionEndpoint at this point).

3. Once the shutdown signal is received by the future driving the Server itself, start a shutdown loop:
   a. For every connected client (via the Server::connected_clients method), call Server::disconnect on it.
   b. Call Server::recv with a short timeout.
   c. Check Server::active_clients to see if any clients are left in any state where they require more processing.
   d. Repeat until some global timeout is reached.

The last step would allow for any new incoming clients to finish establishing and then be completely closed, and I think there are no weird race conditions there.

This requires a new method Server::active_clients to get the count of clients in any state other than completely shut down, which I am currently adding.

[kyren]

I would accept a PR to make this a method on Server as well, if anyone is interested, but I actually think that before this happens the entire graceful shutdown procedure needs to be looked at much more closely and improved. I don't think it currently works in the presence of packet loss and I'm not entirely sure how to make it work properly at all using openssl's API.

[wmww]

Thanks! I will give implementing this a shot at some point, when I get back to the project that needs it.