[sr-dev] Dialog2 and proxy initiated early dialog termination

[Timo Reimann]

Hey Jason,


On 29.09.2011 16:54, Jason Penton wrote:
> Awesome, thanks Timo. However, The example you give here is to store
> dlg_cell in transaction. Actually, we are using the reverse, se pseudo
> code below:
> 
> when INVITE req_forwarded callback is called, create dialog_in
> structure, link and store pointer to transaction in dlg_cell
> if we get a request to terminate dialog that is unconfirmed we get the
> transaction ptr from dlg_cell and traverse through branches, sending
> fake_reply (480/408/x).

Dialogs aren't stored in transactions, they are stored (hackishly) in
transaction callbacks (better: attached to them). But anyways, I think
the uses-relationships are similar in both cases: The dialog module is
called and needs to refer to its transaction. The current module's
pseudo-code:

"When INVITE on_create callback is called: create dialog structure, link
and store pointer to transaction in tm callback.
If we a response, fetch the transaction ptr from the tm callback to
allow access to dialog variables."

AFAICS, all that differs is the location of the transaction pointer
which is currently stored in a tm callback (bad) while you use the
dialog structure (good). Let me know if I get things wrong.


Cheers,

--Timo



> On Thu, Sep 29, 2011 at 4:34 PM, Timo Reimann <timo.reimann at 1und1.de
> <mailto:timo.reimann at 1und1.de>> wrote:
> 
>     Hi Jason,
> 
> 
>     On 29.09.2011 16:06, Jason Penton wrote:
>     > Ok Dialog2 progressing nicely. We now have dialogs and their
>     associated
>     > out dialogs (branches / forking) stored and managed within the dialog2
>     > module. For the moment, we have excluded DB support but will add
>     once we
>     > check in to git. One thing we need a little assistance with:
>     >
>     > We have just finished the prototype for proxy initiated early dialog
>     > termination, but we are concerned with the way it has been done.
>     > Basically as mentioned in the wiki
>     >
>     (http://www.kamailio.org/dokuwiki/doku.php/modules-new-design:dialog-module-design),
>     >
>     >
>     >     *
>     >       It should be possible to terminate dialogs in the “early” state,
>     >       i.e., sending out BYE/CANCEL requests in order to terminate all
>     >       branches appropriately.
>     >           o
>     >             ibc: IMHO it would be easier just to cancel the
>     transaction
>     >             as when fr_inv_timer expires, this is, by sending a CANCEL
>     >             to all the pending branches and a 408 to the UAC
>     (perhaps in
>     >             this case a 480 would be more appropriate).
>     >
>     > The only way we could think of doing this was to send a fake reply via
>     > the TM module. We have therefore exposed the fake_reply function from
>     > the TM module and using that to terminate early dialogs. It works
>     in the
>     > test scenarios we have performed, but the main drawback we can see
>     here
>     > is that the dialog module needs to hold a pointer to the
>     transaction for
>     > each dialog (not sure how bad this is as my experience with tm is not
>     > expert yet ;) )
>     >
>     > So any thoughts/ideas. Is this the correct way to do it? Would it be
>     > okay to expose a fake_reply function through TM API?
> 
>     I cannot comment on how good or bad it is to expose the fake_reply
>     function.
> 
>     Regarding pointing each dialog to its associated transaction at a given
>     time: This is already implemented in the current dialog(1) module. It
>     was needed for several reasons, one of them being to allow access to
>     dialog variables in responses. The way the link between dialogs and
>     transactions is done is by attaching a transaction pointer to the
>     TMCB_MAX callback which is fetched when required. Look at
>     store_dlg_in_tm() in dlg_handlers.c and get_dialog_from_tm() in
>     dlg_profile.c.
> 
>     Yes, abusing TMCB_MAX to store additional data is kinda hackish. The
>     point here is that the dialog/transaction linkage is definitely needed,
>     so IMHO you may go ahead and re-establish it in dialog2. Finding a
>     cleaner implementation approach that doesn't involve TMCB_MAX is highly
>     appreciated though.
> 
> 
>     HTH,
> 
>     --Timo