[Freeswitch-users] Asynchronous communication with FreeSWITCH's mod_event_socket

[Luke Graybill]

I've been doing a lot of work recently with FreeSWITCH's mod_event_socket,
and I wanted to comment a bit about the syntax used for commands through the
socket while using asynchronous mode. I haven't tried the synchronous mode
yet, as I always want to be free to be able to execute commands without
waiting for other commands to finish. For instance, I need to be able to
collect DTMF events while I'm playing a sound file, so that the user can do
things like select menu items without listening to the entire menu first.

Asterisk's AMI protocol allows you to specify an ActionID along with every
command that you send. Asterisk then includes this ActionID with every event
that is related to that command, making it cake to coordinate an
asynchronous client.

However, even in async mode, FreeSWITCH's mod_event_socket doesn't
communicate any identifying information for command responses, or for events
triggered by a previous command, unless one uses the bgapi command set. This
command set is not applicable to every situation, though. It only applies to
commands which manipulate the call; if one needs to manipulate the channel,
then messages must be used through the sendmsg command set, which doesn't
provide any specific identifying information.

Now, to complicate things, with all commands (even bgapi) the protocol works
something like this: you send a command, and wait for a response from
mod_event_socket. This response is assumed to be immediate before anything
else the client might receive from mod_event_socket, and in the case of
bgapi, this response will contain a job-id to use for comparing job-related
events later.

For example, for the following command:

sendmsg
call-command: execute
execute-app-name: answer\n\n

The response is this:

Content-Type: command/reply
Reply-Text: +OK

That response is generic to nearly every single command sent, and is only
really saying "The last transmission was a valid command, and didn't
immediately fail". The command may actually fail later, and command specific
feedback is generally contained in later events (which have no unique
identifying information).

My issue here is that this seems to be forcing an asynchronous client to
rely upon a synchronous ordering of response directly following command,
thus violating the very concepts of an asynchronous protocol in which there
should be no assumed order. Not only that, but this method increases the
complexity of a client, which must be aware of limitations that wouldn't
ordinarily be required by a true asynchronous protocol. An asynchronous
client should be unconcerned with listening for a synchronous response to
every command.

My suggested solution is to apply the job-id concept from bgapi to messages
as well, and to go a step further; borrow the Asterisk idea of transmitting
an identifier along with each command. Every response and event related to
that command should then contain the very same identifier in the header.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.freeswitch.org/pipermail/freeswitch-users/attachments/20080918/f54dda8c/attachment.html