The problem is: writer.close() is a regular method which calls transport.close().

But transport.close() actually closes the connection only on next loop iteration by self._loop.call_soon(self._call_connection_lost, None) call`.

Now a safe way for stream closing is

writer.close()
await sleep(0)

but it looks ugly.

I see two possible solutions:

1. Introducing writer.wait_closed() coroutine.
2. Changing writer.close() signature to return a future object. The future will be resumed on protocol's connection_lost() callback.

Honestly I vote on point 2: it looks more native and fully backward compatible with existing code.