Sometimes I find that when DNS name resolution is not immediately 
successful, gethostbyname() may block for unacceptable time, causing my 
client program to become unresponsive. My question is, how should I 
implement either DNS timeouts, or totally asynchronous DNS name 
resolution? Do I need to write a proper DNS client from scratch from the 
RFC? This is worse on some platforms than others.
Also gethostbyname() is poor because it is returning a pointer to static 
data structure, and I currently must use mutexes to protect this call in 
the multithreaded code. My code must be portable across HP/UX, Irix, 
AIX, Linux, Solaris, Tru64, and Windows, so I have minimal choice about 
what functions to use (gethostbyname_r() not supported on everything)
Thanks for any feedback on this.

root wrote:
> Sometimes I find that when DNS name resolution is not immediately
> successful, gethostbyname() may block for unacceptable time, causing my
> client program to become unresponsive. My question is, how should I
> implement either DNS timeouts, or totally asynchronous DNS name
> resolution? Do I need to write a proper DNS client from scratch from the
> RFC? This is worse on some platforms than others.
> Also gethostbyname() is poor because it is returning a pointer to static
> data structure, and I currently must use mutexes to protect this call in
> the multithreaded code. My code must be portable across HP/UX, Irix,
> AIX, Linux, Solaris, Tru64, and Windows, so I have minimal choice about
> what functions to use (gethostbyname_r() not supported on everything)

Try calling alarm() before the call to gethostbyname(), so that it is
interrupted when the alarm timeout reaches and SIGALRM is sent to your
application.

On Fri, 31 Mar 2006 15:46:11 +0000, root wrote:

> Sometimes I find that when DNS name resolution is not immediately 
> successful, gethostbyname() may block for unacceptable time, causing my 
> client program to become unresponsive. My question is, how should I 
> implement either DNS timeouts

For how long can you block? Regardless of how low you can set your
system resolver options, anything less than 5 seconds or so is asking for
a ton of false failures.

If 5 seconds is just fine, than the SIGALRM approach as suggested
elsewhere in this thread is fine.

> or totally asynchronous DNS name 
> resolution? Do I need to write a proper DNS client from scratch from the 
> RFC? This is worse on some platforms than others.

I know of three: ADNS, C-Ares and UDNS. I've used the first two, and have
stuck w/ C-Ares because I've written a library suite around it. UDNS
looks interesting but I haven't found occasion to try it out (it's a late
comer).

> Also gethostbyname() is poor because it is returning a pointer to static 
> data structure, and I currently must use mutexes to protect this call in 
> the multithreaded code. My code must be portable across HP/UX, Irix, 
> AIX, Linux, Solaris, Tru64, and Windows, so I have minimal choice about 
> what functions to use (gethostbyname_r() not supported on everything)
> Thanks for any feedback on this.

gethostbyname(3) is an old, ugly interface. You should be using
getnameinfo(3)--which is thread-safe by design--and if you want a
portable implementation you can pull it out of the portable-OpenSSH
source. AIX, Linux, Solaris and I think HP/UX should definitely have it.
I'm suspect that Irix or Tru64 will since I dunno how well they've been
maintained.

- Bill