On Thu, Jun 01, 2006 at 01:20:28PM +0300, Daniel-Constantin Mierla wrote:
On 06/01/06 12:53, Alexandre Snarskii wrote:
On Wed, May 31, 2006 at 12:04:41AM +0300, Daniel-Constantin Mierla wrote:
Hello,
On 05/30/06 19:47, Rafael J. Risco G.V. wrote:
hi I would like to read some opinions to know which method its better to implement Call Hunting, using serial forking with 'avpops' or 'lcr' module 'load_contacts()/next_contact()' functions, does someone has an example of any of these methods?
if it is a global scope you can use any of them, Which is better? Depends on your environment and how you organize your data.
If is per user, then avpops is more suitable. An example of using avpops you can find at:
http://www.voice-system.ro/docs/avpops/ar01s08.html#ex_serial_forking
It is suitable, but for some scenarios it require much more complicated config, which is not easy to understand.
Scenario: User A (PSTN side) calls user B (SIP). With avpops we select some numbers for this user, trying original number (returns 486/BUSY), hunting to next number (creating new transaction on router).
it does not create a new transaction, just adds a new branch to it.
Phone B2 starts with 100/Trying 180/Ringing, then user A hangs. CANCEL message first visits original number, which returns 481/Call leg/transaction does not exists [1],
you mean that sends CANCEL to B1? I do not think so, CANCEL is not sent to completed branches, just to ones that haven't received a final response.
then hunts to second number and successfully cancel's call. But, at point [1], first transaction _does not store_ new reply code (modules/tm/t_reply.c, line 762 just logs message LOG(L_ERR, "ERROR: t_should_relay_response: status rewrite by UAS: " "stored: %d, received: %d\n", Trans->uac[branch].last_received, new_code ); goto discard; ), and again hunts, so, third number of user B starts ringing...
What version of openser do you use?
ser 0.9.6, freebsd 6.1, database backend is postgresql. Did not tried that setup with openser yet. Initial question was in ser-users mailing list..
The only solution for this which I found yet is:
#The only way to really check return code is to check it in onreply_route, #then set flag which will permit failure_route to hunt this call again. #You can't check status with t_check_status within failure_route, #because t_check_status in failure_route returns the lowest value #of all branches.. onreply_route[1] { if(status=~"40[48]" || status=~"486") { #not-so-fatal response, should hunt setflag(2); }; } failure_route[1] { if(isflagset(2)) { resetflag(2); if(avp_pushto("$ruri","$serial_fork")) { #and then everytingh should work as in example.
You can use t_was_cancelled() in failure route: http://openser.org/docs/modules/1.1.x/tm.html#AEN479 . It will ensure that you get the proper status of canceled transactions.
But, this config is not yet complete - with this configuration CANCEL's does not hunt after 481 Call leg does not exists, so, in situation when we need to cancel call on second or third number, we're not able to do that :( Just adding another code (481) to our list of not-so-fatal response does nothing but returns us to initial problem, and the only solution found is to complicate configuration again to:
#main post-routing code route[1] { if(method=="CANCEL") { t_on_reply("2"); } else { t_on_reply("1"); }; t_on_failure("1"); t_relay(); };
onreply_route[1] { #onreply for everyting but CANCEL if(status=~"40[48]" || status=~"486") { #not-so-fatal response, should hunt setflag(2); }; } onreply_route[2] { #onreply for CANCEL's if(status=~"481") { #not-so-fatal response, should hunt setflag(2); }; }
failure_route[1] { if(isflagset(2)) { resetflag(2); if(avp_pushto("$ruri","$serial_fork")) {
Please note, that config is in pre-production stage, tested, but without really massive tests. Also, note that this is not complete config, just parts to allow serial hunting without noted problem.
If anyone knows more elegant solution for my problem - please, let me know. Also, if anyone can see a problem with my config - let me know how to test it...
Could you try the latest version of OpenSER CVS, it is now in frozen/testing phase, just to be release in a few weeks. Any feedback will help to fix eventual issues. The part with reply code selections and CANCEL processing was refurbished and should have fixed some old issues.
Cheers, Daniel
PS: another problem may arise with maximal number of branches. With default configuration of ser (config.h, #define MAX_BRANCHES 12) maximum number of hunts is limited to 11. If you need more hunts than this - you need to recompile ser.