bind-9.4.0b2 exits unexpected...

[JINMEI Tatuya / 神明達哉]

>>>>> On Sun, 08 Oct 2006 21:03:23 +0200, 
>>>>> Marco Schumann <schumann at strato-rz.de> said:

> another mystic(?) bind-9.4.0b2 behaviour.
> Last two nights one of our resolvers stopped working:

> ...
> 07-Oct-2006 21:03:29.393 general: rbtdb.c:1158: REQUIRE(prev > 0) failed
> 07-Oct-2006 21:03:29.393 general: exiting (due to assertion failure)
> ...
> 08-Oct-2006 19:32:31.666 general: rbtdb.c:1158: REQUIRE(prev > 0) failed
> 08-Oct-2006 19:32:31.666 general: exiting (due to assertion failure)
> ...

> What happens exactly? isc_refcount_decrement returns NULL when... when?
> And why is it so fatal, that the whole process must die? Was this
> introduced in this version? In bind-9.3.2-P1 the macro was called only
> once in that file, bind-9.4.0b2 executes it 7 times. Does this behaviour
> correlate with the number of worker threads as it seems to be a locking
> issue? And if so, which way?

Did named dump core?  If so, showing its backtrace would be helpful.

> But most of all we are interested in how we can avoid that.

As you speculated, this is most likely a thread-related bug that is
specific to 9.4.  So, if you need a quick remedy, I'd suggest to
rebuild named with --disable-threads or --disable-atomic.  (The former
should be obvious, and the latter disables optimization newly
introduced in 9.4).

					JINMEI, Tatuya
					Communication Platform Lab.
					Corporate R&D Center, Toshiba Corp.
					jinmei at isl.rdc.toshiba.co.jp