[SR-Users] Proposal to extend acc/dialog modules in order to log CDRs

[Timo Reimann]

Hey Carsten,


On 07.02.2011 12:04, Carsten Bock wrote:
> The functionality for storing additional data for a dialog is like this:
> 
> $dlg_var(ecscf_server) = ‘sip:ecscf.ng-voice.com’;
> xlog(“Emergency CSCF is $dlg_var(ecscf_server)\n”);
> 
> However, it might be worth noting, that i haven't added bindings to
> the dialog-API and that the Pseudo-Variable currently only supports
> Strings.

For CDR generation purposes, I wouldn't need any additional dialog API
bindings. The PV interface should suffice as Daniel and I have concluded
on the mailing-list and at FOSDEM.


> I think, my branch needs to exists a little longer, since the changes
> take quite a lot of time. I am currently doing a lot of cleanup where
> many methods get obsolete; so i do not want to add docs for later
> obsolete methods and parameters.

What aspects of your branch do you plan to change in the future? Would
that affect the PV interface as well?

If you can provide a rough plan in order for me to consider whether the
changes may affect my CDR approach I'd be very glad.

Ideally, if the dialog data storage functionality is somewhat complete,
we may consider putting it into the master branch prior to any
outstanding changes in your branch.

Let me know if there's anything I can do.


Cheers,

--Timo



> 2011/2/2 Timo Reimann <timo.reimann at 1und1.de>:
>> On 02.02.2011 14:03, Daniel-Constantin Mierla wrote:
>>>> On 01.02.2011 21:15, Daniel-Constantin Mierla wrote:
>>>>> On 2/1/11 8:18 PM, Timo Reimann wrote:
>>>>>> [...]
>>>>>>
>>>>>> Apart from this very minimum CDR content, however, one can think of a
>>>>>> number of CDR fields that cannot be filled easily. For instance, caller
>>>>>> identity isn't always contained in the From header (think of CLIR
>>>>>> calls)
>>>>>> but need to be determined from other headers depending on the type of
>>>>>> call, possibly even involving a database lookup.
>>>>>  From header URI is not accounted automatically, you can specify any
>>>>> variable that can contain caller id as you need (e.g., it can be an AVP
>>>>> that you previously set for in config) -- see *_extra parameters of acc
>>>>> module.
>>>> I initially thought about having a cdr_extra parameter similar to
>>>> log_extra. However, I see a problem with AVPs which live in single
>>>> transactions to work with dialogs that span multiple transactions.
>>>
>>> well, some of the avps are transaction persistent, but there are also
>>> global avps which are available as long as kamailio runs. None of these
>>> should be used, I gave avp just as generic PV example. The idea was to
>>> make new PV available with the data stored by dlg module.
>>
>> Ok, so instead of providing getter/setter functions, let's use PVs.
>> However, users will have varying needs regarding the to-be-defined CDR
>> fields next to the canonical ones (start time, end time, duration,
>> etc.). We, for example, use a rather large number of fields, many of
>> them being too specific for general integration into upstream Kamailio.
>> That's why I would rather see a way to dynamically specify CDR fields.
>> Carsten's new code would be the right approach to do this, right? His
>> dlg_var PV takes arbitrary key/value pairs which looks just like what is
>> needed.
>>
>>
>>>>>> That's why I'd like to
>>>>>> provide a way to pass additional, CDR-specific data from the
>>>>>> configuration file to the dialog structure associated with the call.
>>>>>> When the dialog is about to terminate, the extended acc module would
>>>>>> make sure that the added CDR data associated to the dialog is included
>>>>>> in the CDR.
>>>>> Not sure what you mean here with "acc module will make sure ...", but I
>>>>> hope is not going to be cross reference/dependency, so that acc has to
>>>>> walk through dlg structures.
>>>> If I get you right, by cross reference/dependency you mean that A
>>>> requires B and vice versa, i.e., cyclic dependency. I do not intend to
>>>> let that happen.
>>>>
>>>> Instead, the acc module would use getter functions attached to the
>>>> dialog interface (which is what I mean by "acc module will make sure").
>>>> dialog would never touch the acc module or even know about the fact that
>>>> the acc module is using it.
>>>
>>> But then acc has to know the dialog internals, what the getter function
>>> will return and how to access those structures. PV framework is exactly
>>> the same, but available for all components/modules, no need to develop
>>> specif ones each time.
>>
>> Agreed, let's use PVs. If you think of Carsten's new dlg_var PV, it's
>> basically just a way to get/set arbitrary values; so I suppose this is
>> what I should have expressed.
>>
>>
>>> With your solution, dialog module has to know acc api to call the
>>> recording function, and acc has to know dialog module to know its
>>> exported structures. This is cross dependency imo.
>>
>> I think this point shows the major difference between our solutions: My
>> approach doesn't make the dialog module call the recording function.
>> Instead, the recording function is called by the acc module as soon as
>> the dialog terminates where call termination can is observed by means of
>> using dialog callbacks. (Sorry, I should have possibly mentioned the
>> words "dialog callback" before.)
>>
>> Therefore, module dependency is unidirectional: acc registers for some
>> dialog callbacks and, upon execution of a callback, records the CDR.
>>
>>
>>>>>> Third, the CDR is persisted to either log file, database, or both.
>>>>> If this new thing is not going to support what acc module API has for
>>>>> backends (radius is missing), then will not make sense to tie the two.
>>>>> dialog module can do its accounting alone.
>>>> Of course, I would want to take advantage of acc's existing backend
>>>> connectors. There's no need to re-invent the wheel, acc will still be
>>>> responsible for writing out CDRs; the difference is that a few
>>>> additional calls to a well-defined dialog interface will be used in
>>>> order to collect the data that constitute the CDR.
>>>
>>> What will happen if a new call tracking module will be available, should
>>> acc be changed again?
>>
>> I am not completely convinced that call tracking is such a flexible
>> thing even though you gave a few examples below. A call is pretty much a
>> dialog, and proper tracking of a dialog should be dialog module's job.
>>
>> Moreover, I believe your approach has some dependency issues as well --
>> see my comment at the very end of this mail.
>>
>>
>>>>> - such functionality is independent of dlg module or any new call
>>>>> tracking extension in the future, also writing full CDR can be achieved
>>>>> from config file by tracking INVITE and BYE, calling
>>>>> acc_start()/acc_stop() from cfg
>>>> Tracking INVITE and BYE messages from the config file may not be enough
>>>> if you want to create CDRs for failed calls too, e.g., 408/487.
>>> What stops tracking these replies in the cfg?
>>
>> You can do it, but for a large config file with many different paths,
>> it's easy to miss a spot where you should have added a call to
>> acc_stop(). On the contrary, if recording happens automatically when the
>> call ends, you don't have to deal with any manual work.
>>
>>
>>> Anyhow, I think you divert from the initial goal. If you just break the
>>> message in pieces and reply on each word will lose the meaning of the
>>> entire content. I presented my plans above and below I added acc_cdr()
>>> that seems to suit your needs.
>>
>> Maybe I got you wrong but my understanding is that acc_start() records
>> one part of a CDR (start time, etc.) when the INVITE is routed while
>> acc_stop() finishes it off (end time, duration, etc.) on reception of
>> the BYE. I was trying to stress that a single call to the recording
>> function is likely preferred (at least by me), which acc_cdr() seems to
>> accomplish.
>>
>>>>> Therefore if I would do it:
>>>>> - enhance acc module to export via cfg exports and inter-module api
>>>>> three functions:
>>>>>     - acc_start() - write the initial call record at start
>>>>>     - acc_stop() - update the call record at stop, based on a matching
>>>>> condition specified as parameter
>>>>>     - acc_cdr() - write a full CDR
>>>>>
>>>>> Data to be written in db (or other backend) is going to be taken from
>>>>> PVs, independent of who (cfg, dialog, or other module) is calling the
>>>>> function, specified in a similar form like db_extra.
>>>>>
>>>>> First two functions will work for db only. Third can work also without
>>>>> dialog, e.g., I can store the start of a call, a.s.o. in hash table and
>>>>> get it at BYE time to build the full cdr.
>>>> Where would you maintain that hash table containing the call start time
>>>> and similar data if you do not seek to use the dialog module? Such data
>>>> seems to be naturally associated to a dialog, so my impression is that
>>>> it should be kept there too, i.e., in the dialog module.
>>> The hash table module (htable) keeps the data in shared memory. It is a
>>> generic container, so you can store anything there. dialog module is
>>> just an automatic call tracker, handling sip messages passing through
>>> kamailio. Most of that can be done from config file. Internally, dialog
>>> module has also a hash table. Also, I can track active calls using a db
>>> table (in memory or not for speed), with a timer process (rtimer module)
>>> to handle missing BYEs, and so there could be other options to track
>>> dialogs. My point here is that acc should be independent of all these
>>> potential options available now or added in the future.
>>>
>>> You additions to dialog module can export the new stored data inside
>>> dialog via some PV, e.g.,
>>>
>>> $dlgacc(caller), $dlgacc(callee), $dlgacc(duration), a.s.o.
>>>
>>> dialog module will call acc_cdr() when the call ended, making sure the
>>> PVs are properly available at that time.
>>
>> If you call an acc module-provided method from the dialog module, you
>> face a similar independence problem: Should accounting functionality
>> ever be changed or replaced of by another module, then you'd need to
>> change the dialog implementation as well.
>>
>> There's a conceptual issue related to the given usage relationship as
>> well: If component A requires B to implement a particular functionality,
>> A should use B. Having the dialog module trigger the recording function
>> puts this concept upside down -- CDR accounting isn't needed by dialog
>> tracking, it's the other way around.
>>
>>
>> Cheers,
>>
>> --Timo