20 Oct
2008
20 Oct
'08
5:53 a.m.
Juha Heinanen schrieb:
Klaus Darilion writes:
Ok. I missed that next_contact have different return value on no AVPs
depending on request_route or failure_route. This is fine, as long as
the first usage is in a request route. If the load_contacts happens in a
failure route, this can cause problems.
But I wonder what if all the branches after
enum_query() have the same q
value. According to the README:
load_contact: If all contacts in the destination set have the same
qvalue, load_contacts() does not do anything thus minimizing performance
impact of sequential forking capability when it is not needed. Returns 1
if loading of contacts succeeded or there was nothing to do.
Thus, next_contacts() will be called although there are no AVPs to
fetch. Do I miss something here?
klaus.
i don't think you miss anything. if all contacts have same q value,
load_contacts() does not do anything and does not add anything to the
avp. when next_contacts() is then called, it, in turn. does not do
anything because there is no avp values. if t_relay fails, for example, due to tcp connect
failure to the first
contact. your failure route will never get called and no other contacts
even if they exist, will never be tried.
i personally think this is a major problem with current t_relay
implementation. i my opinion, failure route should be called also in
case of internally generated negative replies.
Yes. I read your other emails. I always thought this was solved with the
t_relay("2") flag, but obviously not.
regards
klaus