nscd frequent assertion failure...

[Petr Baudis]

  Hi,

On Sat, Nov 22, 2008 at 12:54:18PM +0000, Daniel J Blueman wrote:
> glibc 2.8's name-service cache daemon (nscd) is seeing frequent
> assertion failure [1], at least in Ubuntu 8.10, due to internal
> consistency checks failing (memory object misalignment). Right now in
> the 8.10 Intrepid cycle, this is 100% of the crashes I've had over the
> last week's workflow.
> 
> Since nscd is not installed by default on many distros, this isn't
> getting the minimum exposure and attention to get addressed, sadly.
> I've been unable to find any fixed/unfixed bug reports, and I've
> tracked this with valgrind [2].
> 
> Is anyone else seeing this, and has any suggestions about the best way forward?

  I think the best solution is to just upgrade to glibc-2.9's nscd code,
which works much better - we will be doing that within a maintenance
update for OpenSUSE 11.0:

	https://bugzilla.novell.com/show_bug.cgi?id=387202

  Even in glibc-2.9, there is still an open race between gc() and adding
to the cache, as described in:

	http://sourceware.org/bugzilla/show_bug.cgi?id=5381

This still leads to occassional crashes, but much less frequent than
in glibc-2.8 and should be fixed soon.

  SUSE does enable nscd by default - I have been unable to look into
these issues before, but we are very careful to have rock-solid stable
nscd in our upcoming SLES11. In the longer term, we are considering to
switch to unscd for its more robust design.

-- 
				Petr "Pasky" Baudis
People who take cold baths never have rheumatism, but they have
cold baths.