bug#11956: misc/tty-eof: sometimes failure under heavy load

[Bernhard Voelker]

On 08/01/2012 01:41 PM, Jim Meyering wrote:
> Thanks for the report.
> I'll bet that the 1-second timeout is expiring.
> Does this solve it for you?
> 
> diff --git a/tests/misc/tty-eof b/tests/misc/tty-eof
> index e9eed2e..28b4fa3 100755
> --- a/tests/misc/tty-eof
> +++ b/tests/misc/tty-eof
> @@ -73,7 +73,7 @@ $@
>          or $exp->send("a b\n");
>        $exp->send("\cD");  # This is Control-D.  FIXME: what if that's not 
> EOF?
>        $exp->expect (0, '-re', "^a b\\r?\$");
> -      my $found = $exp->expect (1, '-re', "^.+\$");
> +      my $found = $exp->expect (10, '-re', "^.+\$");
>        $found and warn "F: $found: " . $exp->exp_match () . "\n";
>        $exp->expect(10, 'eof');
>        # Expect no output from cut, since we gave it no input.

Hi Jim,

thanks.

Unfortunately, I cannot reproduce that failure anymore in a constant way
to find out if the change has any effect.

I set my PC under disk load like this:

  while : ; do dd if=/dev/zero of=x1 bs=300M count=4; rm x1 ; done &
  while : ; do dd if=/dev/zero of=x2 bs=400M count=3; rm x2 ; done &
  while : ; do dd if=/dev/zero of=x3 bs=500M count=3; rm x3 ; done &
  while : ; do dd if=/dev/zero of=x4 bs=100M count=10; rm x4 ; done &

and started some tasks to make the CPU (quadcore) busy:

  dd if=/dev/zero of=/dev/null bs=100M &
  dd if=/dev/zero of=/dev/null bs=400M &

The load was at 8-10 with ~95% for sy or wa.

Then I copied tty-eof to tty-eof10 with you applied your patch to
the latter and added it to test/Makefile.

Finally, I ran "make check TESTS='misc/tty-eof misc/tty-eof10'
for about 50 times.

However, I cannot tell if the patch makes a significant change.
Mostly both versions succeded, and the very few times they failed
it happened that both failed.

Unless someone else can reproduce it, I'd tend to close the bug.

Have a nice day,
Berny