Hello,
being brought in discussions many times, including in the recent past days, I replaced the static define of max branches limit with a core parameter. It impacts the destination set stored in core and the uac fields stored in transaction.
It can be set now with global parameter: max_branches
The old limit was 12, now it can be set in the range of 1 to 31.
Few aspects that I would like to discuss to get this refactored to fit the best of common scenarios:
1) the 31 upper limit comes from tm cancelled branches bitmask, which now is stored in an integer. Should be easy to get it to 63 (planned after some testing that proves the concept used now is ok). Would it be a need for even more?
2) would it make sense to specify the max number of branches per transaction, in config, before creating the transaction? The upper limit will be the max_branches value from the core.
3) thinking of common cases of what can be forked a lot, I thought that we can a simplification of 2) by specifying two limits: one for initial requests which are very likely to have many branches (think of initial INVITE via LCR or location) and one for requests within dialog which are likely to have one or very few branches (e.g., replicating BYE to a peer server). Opinions?
The main benefit for 2) and 3) would be less memory usage.
Testing and feedback of the new max_branches parameter is very appreciated.
Cheers, Daniel
On 17 Oct 2014, at 11:39, Daniel-Constantin Mierla miconda@gmail.com wrote:
Yes! We will need this to implement the max-breadth limit at some point and manage it. It must be very very related to what you're up to now.
/O
Good stuff. I am upgrading a couple of kamailio dev servers now to test this. I will get back to you if i find any problem later today.
I think max 31 branches are fine, at least for my needs.
Max branches per transaction would be really great. Per my own requirements, I need higher number of max branches for only a few special cases (e.g. call over push notification, location based on-net call routing etc.), for other cases i would hardly need 3-5 max branches.
Regarding point 3, i am not sure if i have any use of it (besides the better performance as you mentioned), but i guess there is no harm in it and it may be useful for other users.
Thank you.
On Fri, Oct 17, 2014 at 11:39 AM, Daniel-Constantin Mierla < miconda@gmail.com> wrote:
group messages (if implemented using forking) would be one app that requires lost of branches for initial request. for this kind of app, number of branches would need to be configurable per message. perhaps there are better means to implement group messages.
-- juha
We did some preliminary work to make this 63 as well (some years ago). I'm glad to see this getting traction.
I'm generally opposed to hard limits like this. There are many cases where you might want to ring a number of phones at once, and the sizes of those ring groups are currently limited by this. I suspect that once we have > 31 in our production environment, we'll quickly have a case that wants > 63.
But right now, I can only say that we have use cases > 31.
-- Jorj
On Friday 17 October 2014, Daniel-Constantin Mierla wrote:
I suggest to skip option 3, as 2 is a superset of it. No need to introduce another limited interface.
I would prefer a different solution though: remove the maximum altogether and dynamically allocate branch/uac structures. A lot of memory is wasted now because memory is always allocated for the maximum number of branches even though they're rarely being used.
My team is very interested in this solution so I tried to size up the effort. My analysis revealed it would take a major redesign to support this since many static structures, like struct cell, depend on it. If I'm wrong, we would really like to see this changed to dynamic.
Bob
On Tue, Oct 21, 2014 at 11:20 AM, Alex Hermann alex@speakup.nl wrote:
-
Alex Hermann -
Daniel-Constantin Mierla -
Jorj Bauer -
Juha Heinanen -
Muhammad Shahzad -
Olle E. Johansson -
Robert Boisvert