You are right patch does not work for Async commands. I will try to fix it per your guidelines. Once again many thanks for your insight into the matter.

Thank you.




On Sat, Mar 29, 2014 at 3:53 AM, Muhammad Shahzad <shaheryarkh@gmail.com> wrote:
humm,

When datagram server is initialized in method "mi_init_datagram_server", it set tx_sock equal to rx_sock, instead of creating a new reply socket. So that when an async command is received in method "mi_datagram_server" with tx_sock reference, i pass it to "build_async_handler" method which is then stored to a newly added parameter to "my_address_socket" structure (see the last 5 lines of this method). This same structure is passed and used to send reply in "datagram_close_async" method, so i just used the existing socket reference stored in structure to send the reply, instead of creating new one.

So basically original socket reference is preserved for reply, instead of creating new. Shouldn't this work for async process or i am missing something here?

Thank you.




On Fri, Mar 28, 2014 at 11:28 PM, Daniel-Constantin Mierla <miconda@gmail.com> wrote:
Have you tested with async mi commands? The function build_async_handler() is cloning the information in shared memory, expecting that the response to the mi command to be done from another process. In that case, the socket is no longer valid, because it was created only in the mi process.

I haven't analyzed deeply, but it is very likely to be what I said above.

Options coming in my mind:

- for async case keep the old way of creating a new socket. This can be propagated by setting field tx_sock=0
- try to bind to same socket in async handler, provided the initial socket is created with resuse addr/port

Cheers,
Daniel


On 28/03/14 21:49, Muhammad Shahzad wrote:
Please find attached updated patch as requested.


On Fri, Mar 28, 2014 at 8:49 PM, Daniel-Constantin Mierla <miconda@gmail.com> wrote:
Thanks for the patch. Can you resend it without commented code? Just remove the lines that are no longer used, to have clean code.

Cheers,
Daniel


On 28/03/14 01:29, Muhammad Shahzad wrote:
Hi,

After wasting most of the day trying to make mi_datagram over UDP socket work. I eventually realize that it does asymmetric UDP communication, which creates a lot of trouble for writing a useful MI script using PERL or Python etc.

Anyhow, i go through the module code and was able to write a patch for symmetric UDP communication using mi_datagram.

I have tested the patch using both Kamailio 4.1 and trunk branches and it seems to compile and run flawlessly.

I am attaching here with this patch for Kamailio community. It would be great if Kamailio developers would review and commit this patch to source trunk.

Thank you.



--
Daniel-Constantin Mierla - http://www.asipto.com
http://twitter.com/#!/miconda - http://www.linkedin.com/in/miconda
Kamailio World Conference - April 2-4, 2014, Berlin, Germany
http://www.kamailioworld.com



-- 
Daniel-Constantin Mierla - http://www.asipto.com
http://twitter.com/#!/miconda - http://www.linkedin.com/in/miconda
Kamailio World Conference - April 2-4, 2014, Berlin, Germany
http://www.kamailioworld.com