2007/2/16, Colin Whittaker <[EMAIL PROTECTED]>:
> Well, that didn't really work... The for loop doesn't traverse when there
> is one server.
Thanks for the patches. I tried to figure out what the surrounding
code does, and I think the logic calling sres_next_server() and
reacting to permanent/transient errors was pretty demented. So, in
addition to your fixes I tried to clarify the logic there, too. I have
not given the patch a real try yet, however.
--Pekka
> Here is another attempt:
> sres.c
> 2688a2689,2690
> > #define DNS_ICMP_TIMEOUT 60 // Retry after 1 min and
> ICMP error
> > #define DNS_ERROR_TIMEOUT 3600 // Retry after 1 hour and "perm"
> error
> 2697a2700
> > time_t now = time(NULL);
> 2705a2709,2720
> >
> > for ( j=0; j<N; j++ ) {
> > if ( servers[j] )
> > {
> > if (servers[j]->dns_icmp + DNS_ICMP_TIMEOUT < now ) {
> > servers[j]->dns_icmp = 0;
> > }
> > if (servers[j]->dns_error + DNS_ERROR_TIMEOUT < now ) {
> > servers[j]->dns_error = 0;
> > }
> > }
> > }
>
> Colin Whittaker wrote:
> Ok, I took a stab a this, but I think you may have a better or more
> complete solution.
> I basically just timed out the dns_icmp and dns_error on the servers.
> Here is the diff:
> sres.c
> 2688a2689,2690
> > #define DNS_ICMP_TIMEOUT 60 // Retry after 1 min and
> ICMP error
> > #define DNS_ERROR_TIMEOUT 3600 // Retry after 1 hour and "perm"
> error
> 2697a2700
> > time_t now = time(NULL);
> 2708a2712,2714
> > if (servers[j]->dns_icmp + DNS_ICMP_TIMEOUT > now ) {
> > servers[j]->dns_icmp = 0;
> > }
> 2714a2721,2723
> > if (servers[j]->dns_error + DNS_ERROR_TIMEOUT > now ) {
> > servers[j]->dns_error = 0;
> > }
>
>
>
> Pekka Pessi wrote:
> Sorry for the way too long delay, Iäve been away for the holidays.
>
> On 12/12/06, Colin Whittaker <[EMAIL PROTECTED]> wrote:
>
>
>
> You are right, it does not use /etc/hosts or nsswitch.conf (or
> /etc/host.conf)
>
>
>
> Hmm, seems like folks setting these things up would expect an application
> that uses DNS to use them. At least a lookup in /etc/hosts if DNS fails.
>
> You are absolutely correct, the folks expect them to be used, but it
> is a bit too hairy for my taste. If you like to add host name
> resolving to the nta or sresolv, please go ahead.
>
>
>
>
> That is correct. The default domain or search domains are only used
> with sres_search(). The SIP/SIPS URI are resolved using three-phase
> algorithm explained in RFC 3263 and with sres_query() which never uses
> search domains.
>
>
>
> It doesn't appear sres_search() is used by the NUA stack.
>
> Nope, only sres_query(). sres_search() (for initial A/AAAA-record
> query?) should be a drop-in replacement.
>
>
>
> There are some problems with sres_query(), too (it abandons the search
> after the initial absolute query fails).
>
> This seems like a bad bug. If the DNS server ever looses contact, you
> basically loose all functionality until you bounce the stack.
>
>
>
> I think I have seen this. If the DNS is not up when I start a REGISTER, I
> get a 503 error. Then when DNS is up, I try to REGISTER again, the stack
> never sends another DNS query and I get a 503 error. Forever. seem like a
> blocker to me. :-)
>
> Hm. It looks like once a permanent error occurs, the server is not
> tried again. I think this is not the intention.
>
>
>
> Do you think this something easy to fix ? Should I start digging into it ?
>
> Please go ahead.
>
> I think the algorithm selecting the server in sres_send_dns_query()
> and sres_next_server() should work so that
> 1) a query is always sent once unless if all servers have a permanent
> error (send
> returns immediately -1 for every server)
> 2) a query is retried only when there has been "enough time" from last
> dns_icmp
>
>
>
>
> There are a few open issues with using /etc/hosts or search domains.
> At which point we should use /etc/hosts (or gethostbyname())? Should
> nua or nta expand a host name in URI to a FQDN before sending it to a
> network?
>
>
>
> I don't think the URI should be modified. If the URI is entered as just a
> host, I think it is reasonable to expect that all servers will be able to
> resolve that host as entered in the URI.
>
> OK.
>
>
>
> I am using NUTAG_PROXY() and NUTAG_REGISTRAR(), which take URIs, but just
> specify hosts right ?
> Here is where I really expected search domains to be used.
>
> I guess NUTAG_PROXY() is a special case, and there we should try to
> resolve host names, too.
>
>
> ________________________________
>
> -------------------------------------------------------------------------
> Take Surveys. Earn Cash. Influence the Future of IT
> Join SourceForge.net's Techsay panel and you'll get the chance to share your
> opinions on IT & business topics through brief surveys-and earn cash
> http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
> ________________________________
>
> _______________________________________________
> Sofia-sip-devel mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/sofia-sip-devel
>
>
--
Pekka.Pessi mail at nokia.com
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
Sofia-sip-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/sofia-sip-devel
