Suppose a browser is forcefully terminated, and no close message is sent to the Tornado server. How does Tornado know (or does it even know?) that this connection has been terminated on the client side? Looking through the Tornado websocket code doesn't make this obvious to me.
I ask because I've had times where the server-side on_close method is not called (presumably because the client failed to send its closing message) for a couple of minutes, but then Tornado seems to notice that the client is gone and closes the socket itself.
Any insight?
Tornado monitors if underlying TCP connection is closed. It does not matter if connection was closed due to closing handshake or because network error occurred. In either case, on_close will be called.
If client loses network connection, TCP connection might be considered open for few minutes more before TCP stack will decide that connection is timed out. After that, it will close socket, Tornado will be notified, etc.