BIND 9.5.0-P1 exiting due to assertion failure in rbtdb.c

[JINMEI Tatuya / 神明達哉]

At Thu, 7 Aug 2008 18:21:42 -0700,
Kai Lanz <lanz at stanford.edu> wrote:

> > Thanks, some additional questions:
> >
> > - how often have you seen this?  Have you seen it multiple times?
> > - especially if it repeats, does it happen if you build named
> >   --disable-atomic?
> > - also especially if it repeats, does it happen for 9.4 or only for
> >   9.5?
> 
> It has happened 3 times since we switched to 9.5.0-P1 yesterday.

Okay, this sounds like not very rare events.  It may not be possible
to identify the cause of bug just from the back trace, but I'll think
about additional assertion checks to catch it at an earlier stage.

> I haven't tried building with --disable-atomic; if you think that's
> a possible fix, I'll give it a try.

I'm not confident, but it's possible.  The crash happened in a code
path which is only effective 'with-atomic', and since the code
with-atomic tries to do tricky locking strategy to improve
performance, disabling it may help avoid the trap.

> We haven't tried 9.4, so I can't say if the same problem occurs in
> that version. We specifically upgraded to 9.5 because of the
> security advisory regarding the DNS protocol. (Previously we
> were running Bind-8.)

Okay, then you may also want to try 9.4.2-P1 or 9.4.2-P2 (if you don't
have other performance with 9.5.0-P1, the former should be fine, but
the latter would also work).

---
JINMEI, Tatuya
Internet Systems Consortium, Inc.