Next Message ID selection issue

[darkopetrovic]

Hello, first thanks for the update concerning #5. Here a new proposal for another issue :)

I let my device working during this week-end and after ~20h the communication stops working. The log showed me a lot of Duplicate NON, ACK or RST received preventing the CoAP client to enter processResponse and print the final result.

I've found that this issue has occurred because the client has generated the same message ID for an exchange than a previous received message from the server. And since this generation has occurred within EXCHANGE_LIFETIME, the message ID being still present in the recent_messages dictionary, the client found a duplicate NON message while treating the response of the added exchange.

To prevent this to append I modified the function nextMessageID as follow:

def nextMessageID(self):
        """Reserve and return a new message ID."""
        message_id = self.message_id
        self.message_id = 0xFFFF & (1 + self.message_id)

        listOfMsgIdInRecentMsg = [key[0] for key in self.recent_messages]
        while self.message_id in listOfMsgIdInRecentMsg:
            self.message_id = random.randint(0, 65535)

        return message_id

Otherwise we could reduce EXCHANGE_LIFETIME but it is not guaranteed that even with a smaller lifetime the client doesn't generate the same message ID within this delay.

From the standard perspective (RFC 7252 document) it is stated in chapter 4.4 :

The same Message ID MUST NOT be reused (in communicating with the
   same endpoint) within the EXCHANGE_LIFETIME (Section 4.8.2).

I guess the application must take care of the incoming message ID too before generating the next one by itself. As it is implemented in your code though since you store every incoming message in recent_messages.

[mwasilak]

I am aware of this issue. It is described in: https://tools.ietf.org/html/draft-ietf-lwig-coap-03#section-2.3.2 I think it will be better to solve the problem by introducing separate namespaces (dictionaries) for local and remote IDs.

[mwasilak]

I uploaded new version of the library (0.1.3) to pypi.

[darkopetrovic]

You're right. Thanks for the speed!