[OpenSIPS-Users] Failed INVITE tcp_send

[opensipslist at encambio.com]

Hello,

An ven., déc 18, 2009, Bogdan-Andrei Iancu schrieb:
>opensipslist at encambio.com wrote:
>>>> You may try to increase the number of tries to something higher -
>>>> 3200, just to see if that is the problem:
>>>>
>>>>    see tls/tls_server.c , line 689
>>>>       #define MAX_SSL_RETRIES 320
>>>>
>>> This doesn't make sense to me. If 320 retries are attempted with no
>>> write op success, then trying 3200 can't be the solution. Rather, it
>>> must be that I have an error in the config script or some permission
>>> problem no?
>>>
>> Sadly, the only thing I could do to solve the problem was to
>> increase 320 as Bogdan suggested. I raised it to 3200, but
>> surely there is a lower 'cieling' that would work. It leaves
>> me with a somewhat sick feeling however, because this seems
>> very hacky and probably only masks the symptom and leaves
>> the real problem intact (which could surface again in some
>> other form.)
>>
>It is not really a hack :) .... i tend to think this number vary from OS 
>to OS, from server to server .....like how slow the write ops take place 
>- on some system is faster, on other is not.....
>
By the way, even when MAX_SSL_RETRIES is set to almost infinity, the
same errors reappear when lowering the number of child processes.
This could be a clue for whoever decides to look into this bug.

In the current state of development, it seems the workaround is:

  Increase MAX_SSL_RETRIES to almost infinity
  Increase tcp_children to the number of UACs (not scalable)
  Log how many SSL retries are really needed until success
  Lower MAX_SSL_RETRIES from almost infinity to what is needed

Does that sound right? Is this acceptable as far as scalability goes?