Re: [PATCH] nptl: Avoid expected SIGALRM in most tests [BZ #20432]

[Florian Weimer <fweimer at redhat dot com>]

On 08/26/2016 07:01 PM, Carlos O'Donell wrote:
> > > +  pthread_t thr;
> > > +  int ret = pthread_create (&thr, attr, thread_func, closure);
> > > +  if (ret != 0)
> Could we make this more robust by looping on EAGAIN and trying
> to create the thread until it succeeds or we timeout the test?
>
> This way embedded systems with less memory might operate correctly,
> and we might avoid the slow reaping problems in the linux kernel?

This would be undesirable because it would paper over glibc or kernel 
bugs.  Such bugs typically involve races in RLIMIT_NPROC accouting, for 
example:

  <https://sourceware.org/ml/libc-alpha/2012-09/msg00538.html>
  <https://sourceware.org/ml/libc-alpha/2013-04/msg00294.html>

I believe this has subsequently been fixed in the kernel, but it appears 
to have resurfaced in Linux 4.6+ in some configurations (far fewer than 
before).

Florian