[asterisk-dev] [Code Review] 3343: res_pjsip: Enable DNS support.

[Olle E. Johansson]

On 14 Mar 2014, at 14:51, Sean Bright <sean.bright at gmail.com> wrote:

> On 3/14/2014 2:41 AM, Olle E. Johansson wrote:
>> 
>> On 13 Mar 2014, at 22:13, Sean Bright <sean.bright at gmail.com> wrote:
>> 
>>> On 3/13/2014 4:42 PM, Paul Belanger wrote:
>>>> +1 with Dan.  Comments aside on DNS functionality (I have opinions but
>>>> sitting this one out). Any functionality should be channel agnostic.
>>>> I too am a little concern'd that statement seems to have changed.
>>> 
>>> In order to make this "channel agnostic" you have three (equally bad) options:
>>> Replace Asterisk's internal DNS facilities with PJLIB's, creating a mandatory dependency on PJSIP.
>>> Roll a shiny new DNS API into Asterisk that supports all address types (multiple results, weighting, etc.).  Bear in mind that PJSIP would not use this new API at all, you would still need to create a PJLIB DNS resolver and feed it the nameservers to use.
>>> Use PJLIB's DNS interface if it is available, otherwise fall back to Asterisk's current DNS interface.  This means that you are now maintaining two separate interfaces and have to throw a layer of abstraction in while you're at it.  In fact, by adding an abstraction layer you would force res_pjsip to then unwrap and then re-wrap the abstraction just to get at the necessary PJLIB data structures.
>>> Frankly, I don't see what all the hubbub is about.  99.9% of users will never touch the nameservers configuration               option and it will behave exactly as if the system resolver was being used.
>>> 
>>> 
>> If there is a configuration people will teach it and people will use it. Later on, the sysadmin change /etc/resolv.conf since the DNS servers used change and PJsip stops working. This is not a good solution. There's no reason for that configuration option at all. No one has stepped forward to explain a situation where it would be needed, right?
> 
> Even if the 'nameservers' configuration option is removed and Asterisk defaults to using the results of res_[n]init, an     administrator changing the name servers in /etc/resolv.conf will not automatically be picked up by PJLIB's resolver.  A reload of some kind would still be required to pick up the changes.
That was my next question. Yes, that's bad.

> 
>> Regarding the resolver issue, I have no clear indication on where to go. I only know I don't want to support a product with multiple resolvers in it. I'm currently working on adding proper SRV support to the old SIP driver and have been digging through a lot of the DNS code. If you ask me today, anything will be better, even a core dependency on PJSIP. ;-)
> 
> That's a bit like rearranging the deck chairs on the Titanic.  Why would anyone continue to use chan_sip when chan_pjsip is available?
I am doing this for 1.8, there's no PJSIP available there. And until PJSIP is on par or ahead of chan_sip, it will be used in large production sites. Moving to PJSIP will require as much testing as moving to another platform.

> 
>> There are other options for asynch DNS too - like C-ARES. It's used in a lot of products and embedded in Resiprocate.
>> 
>> Regarding changing PJSIP - it's just code, right? Why can't you change PJSIP to use another API? That's a very strange statement.
> 
> You certainly could do that, but it's a lot of work for very little gain. 
I disagree on "little gain".

> It would mean continuing to maintain Asterisk's pjproject fork until those changes were (hopefully) accepted upstream, released, and then waiting for the rpm/deb packages to catch up.  Not to mention that someone would actually have to _do_ all of this work.  Could all volunteers please raise their hands? ;-)

If this is how we are going to manage our product then I'm getting really worried. Are we controlling our own software?

/O

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.digium.com/pipermail/asterisk-dev/attachments/20140314/3e055828/attachment-0001.html>