Remove assertion for connection close instead throwing StreamClosedError

[jc-fireball]

this is fix for issue: #289

[jc-fireball]

#290 this is same patch but compare to master.

[abhinav]

Why don't you just call close right here before making the call instead of using the hook?

[jc-fireball]

If I call close, then we make a call, we will choose peer and rebuilt a connection.

[abhinav]

Sorry, I meant what if you stop the server here?

[jc-fireball]

the way iostream calls the callback method is

    def _run_callback(self, callback, *args):
       ....
        with stack_context.NullContext():
            self._pending_callbacks += 1
            self.io_loop.add_callback(wrapper)

from the last line, we see the callback won't be called immediately, there is some delay. That's why even if I stop the server, the call back is triggered later.

for testing purpose: I manually set peer.connections[0].closed = True

[abhinav]

That's not the problem I'm trying to solve here. I'm trying to solve the hook being used to close the connection. Why not,

mock_server.tchannel._dep_tchannel.close()

That will tell the server to immediately stop listening for new incoming connections, so when the call happens two lines below, it'll be trying to hit a hostport that doesn't exist.

[abhinav]

Also, just to confirm: Is this test just trying to verify that we raise a NetworkError when the destination doesn't exist?

[abhinav]

I'm a little confused here. Are you trying to re-raise the StreamClosedError you caught or do you want the new NetworkError to be raised?

[jc-fireball]

It is will be raised again since we have a try catch from the caller to tell whether we should retry based on the exception or not.

This part will be refined in this task: #285