[Samba 4] smbtorture BENCH-NBENCH returns a (seemingly) false failure

[Jeremy Allison]

On Thu, Jun 09, 2005 at 11:30:11AM +1000, Andrew Tridgell wrote:
> Marc,
> 
>  > I'm found that when I run smbtorture --num-progs>1 (i.e. --num-progs 20)
>  > smbtorture gets a false "TEST BENCH-NBENCH FAILED!" message. At least it
>  > seems like this is a false failure, since the test completes
>  > successfully. 
> 
> yes, I noticed that yesterday as well. For me it happens only when
> talking to a Samba3 smbd and not a Samba4 smbd. Does it happen to you
> for both?
> 
> It is a strange one, as it seems the child is existing cleanly, but
> the parent is getting a -1/ECHILD from the waitpid() call, even though
> the child has not been reaped in any other way. An strace shows the
> fork() but no way that the child could have disappeared. Unfortunately
> strace -f hangs (ptrace on linux is just so darn unreliable ...).
> 
> Jeremy, any ideas?

Not immediately, sorry. I'm tracking down a lost wakeup in the
asyncio code currently and that's consuming all my energy. I'm
not 100% convinced it's a glibc bug on Linux... yet :-).

Jeremy.