On 01/02/2012 11:01 PM, Henrik Størner wrote:
> I am trying to use C-ARES for an application that will do DNS lookups in
> parallel with other network I/O (it is a network test-tool, meant to
> check the availability of various network services). And either I do not
> understand how the timeout mechanism is supposed to work in C-ARES, or
> there is a bug somewhere in the timeout handling.
>
> To demonstrate this, I have written a small test application based
> mostly on the "adig" utility that comes with C-ARES. This application
> does the following:
>
> - use ares_library_init() to initialize c-ares library;
> - create a channel with the options including ARES_OPT_TIMEOUT with a
> value of 10;
> - use ares_set_servers() to point the channel to an IP that does not
> respond;
> - use ares_query() to send a single query to the non-responding DNS server.
> It then loops doing ares_fds(), ares_timeout(), select() and
> ares_process(). I pass a "maxtv" value to ares_timeout(), because I only
> want to wait 1 second between each select() call (to do other I/O
> between the ARES checks); as I understand the man-pages, this should be
> perfectly fine.
>
> The callback for the channel simply prints out a message that it was
> invoked.
>
> I then thought that after 10 seconds, my callback would be invoked with
> a status of ARES_ETIMEOUT, but that does not happen. Instead,
> ares_timeout() begins to return a timeout value of 0.0 seconds;
> ares_fds() still indicates that the channel is active. The result is
> that the application starts busy-looping in the
> ares_fds()/ares_timeout()/select()/ares_process() loop and stays there
> forever.
>
> I have tested this with version 1.7.5 of C-ARES.
>
> If you want to check the details, the test application can be found at
> http://henrik.hswn.dk/test-arestimeout.c - it isn't completely polished
> for portability, but it does compile on Linux with "gcc -I./c-ares-1.7.5
> test-arestimeout.c c-ares-1.7.5/.libs/libcares.a"
>
>
> A workaround is to let my application keep track of how long the lookup
> has been underway, and then invoke ares_destroy() on the channel after
> 10 seconds - in that case the callback *is* invoked with the
> ARES_EDESTRUCTION status, as expected. That may be the solution for now,
> but then I don't see what the point is in having the timeout option in
> C-ARES at all...
>
> I hope someone can shed some light on this.
Your test program is flawed. It should call ares_process() *also* when
select() returns time out.
-Timo
Received on 2012-01-03