Python client for lightstreamer

[gianluca.finocchiaro]

Hi epep,

you can always test your code against a Lightstreamer version available from our site (for example, you can install a free Moderato version or a Vivace demo). After installation, please follow the instructions on how to deploy the StockList Demo Adapter.

That said, your modifications look very good, just a couple of observations:

1. Replacing the "raise IOError()" in the _handle_stream method is not a good idea, because a server error response while connecting is a symptom that something is not working properly, as you can read from Paragraph 4.1 of the Network Protocol Tutorial
2. It is necessary to remove the following line:
   self._stream_connection = None
   before invoking reconnect(), otherwise you'll run against a runtime error: it is a bug that I'm going to fix in our example

Please let us know about your progresses.

Thanks and Regards,
Gianluca

[epep]

Replacing the "raise IOError()" in the _handle_stream method is not a good idea, because a server error response while connecting is a symptom that something is not working properly, as you can read from Paragraph 4.1 of the Network Protocol Tutorial

In fact the only error I encountered for now with IG's stream server happen here, about 3 times a day, while trying to bind. I receive a SYNC ERROR about half a second after a LOOP message, before the binding is done. In this situation the reconnection work at first try.

I understand that errors in this point should be used with care for a clean client, but for my project I believe that the only errors that can occur here are the temporary ones, which would be resolved with a later reconnection.

It is necessary to remove the following line:self._stream_connection = None before invoking reconnect(), otherwise you'll run against a runtime error: it is a bug that I'm going to fix in our example

Thanks a lot ! I see the point : None has no close() method.

This line is also worthless before calling the bind() method ?

[gianluca.finocchiaro]

Hi epep,

...
This line is also worthless before calling the bind() method ?

yes because the bind() method replaces self._stream_connection with a newly created object: see fixed example here.

Let me know.

[epep]

Ok, it works fine without the "self._stream_connection = None".

I notice that my logging was badly formated in reconnect(), I should have written : def reconnect(self):<br /> fail = 0<br /> while True :<br /> try :<br /> log.info("Trying to reconnect")<br /> self._stream_connection.close()<br /> self._session.clear()<br /> self.connect()<br /> self.resubscribe()<br /> break<br /> except :<br /> log.info("fail : {0}".format(sys.exc_info()[0]))<br /> fail += 1<br /> wait = min( 300, 0.125 * ( 2 ** fail ))<br /> log.info("reconnect in {0} seconds".format(wait))<br /> time.sleep(wait)

About errors documented in 4.1 of the network protocol tutorial, do you agree that they should not happen after the connection is established ? I've done this at the end of the _handle_stream method : if self._bind_counter == 0 :<br /> raise IOError()<br /> else :<br /> self.reconnect() Like this, if an error occurred at the first connection, it is raised. If it encounters the bind problem from IG, it reconnects. Is it clean ?

[gianluca.finocchiaro]

Hi epep,

...

About errors documented in 4.1 of the network protocol tutorial, do you agree that they should not happen after the connection is established ? I've done this at the end of the _handle_stream method : if self._bind_counter == 0 :<br /> raise IOError()<br /> else :<br /> self.reconnect() Like this, if an error occurred at the first connection, it is raised. If it encounters the bind problem from IG, it reconnects. Is it clean ?

In case of binding an existing session (hence _bind_counter is greater then 0), you could run against similar issues, as specified in the Paragraph 4.2 of the Network Protocol tutorial.

Obviously you are free to implement the reconnection logic as per your requirements, but in my opinion you should trigger it only if you can be sure that a real network problem has been verified.

Let me know,
Gianluca