Hello, I haven't looked at the issue reported by Hugh, so by now just comments on srutils/sruid. The idea was to have an unique id generator without linking to an external library -- my first purpose was to use it for temporary GRUU ids (RFC5627), I am just about to push to master the gruu support in registrar/usrloc. Then I thought it might be useful in other places, such as dialog unique id. I added it as part of lib, since its target usage was for modules so far, but if needed for some core processing, the two files (rather small by now) can be moved in the core. Cheers, Daniel On 4/13/12 2:50 PM, Henning Westerholt wrote: -- Daniel-Constantin Mierla Kamailio Advanced Training, April 23-26, 2012, Berlin, Germany http://www.asipto.com/index.php/kamailio-advanced-training/

I now realise that rls was deliberately using the same sequence of random numbers to satisfy RFC4662 §5.5 5.5. Instance Attributes Each resource element contains zero or more instance elements. These instance elements are used to represent a single notifier for the resource. For event packages that allow forking, multiple virtual subscriptions may exist for a given resource. Multiple virtual subscriptions are represented as multiple instance elements in the corresponding resource element. For subscriptions in which forking does not occur, at most one instance will be present for a given resource. The "id" attribute contains an opaque string used to uniquely identify the instance of the resource. The "id" attribute is unique only within the context of a resource. Construction of this string is an implementation decision. Any mechanism for generating this string is valid, as long as uniqueness within the resource is assured. Seeding srand() with the values 0, 1, 2... always gives the same sequence of ID strings. Maybe someone has an opinion on whether this should be implemented differently, or whether a value should be stored for situations where multiple resource instances are removed/reordered. Maybe kamailio should use a unique ID (from srutils) in the Via header for uac requests (in modules/tm/h_table.h) anyway, to prevent a module from affecting Via branch params by using srand(). Regards, Hugh On 13/04/2012 15:26, Hugh Waite wrote: -- Hugh Waite Senior Design Engineer Crocodile RCS Ltd.

Hi Daniel, Calling srand() to re-seed frequently in RLS meant that in full-state NOTIFY requests the instance IDs within the resource nodes in the NOTIFYs were repeatable. This is required because, when you have forking of back-end SUBSCRIBEs, you need to be able to uniquely (but consistently) identify presentities for different instances of the same resource. Unfortunately, using srand() doesn't work here because you need to store the instance IDs so that they can be used in non-full-state NOTIFY requests (which may only contain updates for specific instances of resources rather than the full set). If you had two instances of a resource A and B in a full-state NOTIFY A would get ID 1 and B would get ID 2. If you then have a non-full-state NOTIFY updates for A must still have ID 1 and updates for B must still have ID 2. However, using srand() to re-seed means that the first instance (which might be B is only B has been updated) gets ID 1. So using srand() here fundamentally doesn't work. The bad side-effect of re-seeding srand() frequently was that it broke other things that, quite rightly, relied on rand() giving random numbers. This point is academic anyway. PUA doesn't support multiple final responses/parallel-forking of back-end SUBSCRIBEs, and the structure and use of the rls_presentity table means that only a single presentity instance can be stored for a resource anyway. Until these things change all that is needed for the instance ID is that it is unique within a resource node and consistent between NOTIFYs for that resource within a dialog. A simple fixed string meets this requirement. So I removed the srand() from PUA and fixed the branch-tag related problem it was causing us. There is still a more general problem in Kamailio, srand() is called during running (not just startup) in quite a lot of cases. This is not ideal and could cause problems similar to the one we saw here at Crocodile in other use cases. I think srand() should be called just once per process during init(), and probably with some seed value involving time and the process ID number. It might also be worth having some Kamailio specific random number library functions in the core and simply telling developers that they should always use them and never srand(), rand() and so on. Regards, Peter On Fri, 2012-04-20 at 12:39 +0200, Daniel-Constantin Mierla wrote: -- Peter Dunkley Technical Director Crocodile RCS Ltd