> Hi,
>
> I have several authoritative and recursive combined servers running bind
> 9.4.3b2 that all crash in the same manor. They will normally run for a
> day or two between crashes, however. The syslog messages only show:
>
> Jul 21 12:08:32 named[17892]: socket.c:1710: fatal error:
> Jul 21 12:08:32 named[17892]:
> RUNTIME_CHECK(((pthread_mutex_destroy(((&sock->lock))) == 0) ? 0 : 34)
> == 0) failed
> Jul 21 12:08:32 named[17892]: exiting (due to fatal error in library)
>
> but elsewhere in the log it has these messages (don't result in named
> exiting):
>
> Jul 21 10:53:39 named[17892]: socket.c:525: unexpected error:
> Jul 21 10:53:39 named[17892]: epoll_ctl(DEL), 73: Bad file descriptor
> Jul 21 10:53:39 named[17892]: socket.c:525: unexpected error:
> Jul 21 10:53:39 named[17892]: epoll_ctl(DEL), 69: Bad file descriptor
>
> 9.4.2 was quite a bit more stable (although I won't say it never exited,
> different errors though IIRC) on the same hardware / OS patch level.


The b2 in BIND 9.4.3b2 stands for BETA 2. It's not a final
release and has lots of new code in it to support many more
concurrent file descriptors as well as getting back some
of the performance loss in the BIND 9.4.2-P1 compared with
BIND 9.4.2. If we had our druthers, we would have only
introduced a change like this in a .0 release.

The changes for BIND 9.4.2-P1 are much more conservative
than those for BIND 9.4.3b2.

> Both servers average 50-100 recursive clients, although I've seen spikes
> to 2000+ (which I imagine is malicious traffic or bots trying to exploit
> the recent vulnerability that I upgraded to fix).
>
> $ uname -a
> Linux 2.6.17-ARCH #1 SMP PREEMPT Sun Jul 16 09:29:38 CEST 2006 i686
> Intel(R) Pentium(R) D CPU 3.00GHz GenuineIntel GNU/Linux
>
> $ gcc -v
> Using built-in specs.
> Target: i686-pc-linux-gnu
> Configured with: ../gcc-4.1.1/configure --prefix=/usr --enable-shared
> --enable-languages=c,c++,objc --enable-threads=posix
> --enable-__cxa_atexit --disable-multilib --libdir=/usr/lib
> --enable-clocale=gnu
> Thread model: posix
> gcc version 4.1.1
>
> Anything I can try to stop these crashes?


Stack backtraces are useful to isolate the problem code.
Backing down to BIND 9.4.2-P1 is a option if the performance
hit is not to big.

Mark

> Thanks,
>
> Dale
>
>

--
Mark Andrews, ISC
1 Seymour St., Dundas Valley, NSW 2117, Australia
PHONE: +61 2 9871 4742 INTERNET: Mark_Andrews@isc.org