Thread: Regression test failure.

Regression test failure.

From
Robert Hentosh
Date:
I am receiving a regression test failure in about 1 out of 12 test runs.
I searched the bugs mailling list and have not found a similar problem.

The system is a RedHat 7.3 system running the UP kernel 2.4.18-10 on an
800MHz Athlon. I didn't do any kernel tuning.

I am compiling postgresql-7.2.3 with the following options:

 ./configure  --with-tcl --enable-syslog --enable-multibyte

greping FAILED on regression.out gives:
     create_index         ... FAILED
test sanity_check         ... FAILED
test select               ... FAILED

I have attached a bzip2 tar file that contains

  log/postmaster.log
  regression.diffs



Thanks,
Robert

Re: Regression test failure.

From
Tom Lane
Date:
Robert Hentosh <hentosh@io.com> writes:
> I am receiving a regression test failure in about 1 out of 12 test runs.

I am amazed no one has reported this before :-(

Upon looking, I see that the parallel test schedule tries to run
create_misc and create_index in parallel, but the latter uses tables
that are created by the former.  AFAICT the diffs you show can all
be explained by create_misc not having executed completely when
create_index looks at those tables.

By rights this failure should have been noticed long ago.  Perhaps there
is some scheduler peculiarity on your machine that makes it more prone
than others to execute the different jobs with just the right timing to
manifest the error.

We really gotta do some work to validate the parallel test schedule
a little better... this sort of thing has popped up before.

            regards, tom lane

Re: Regression test failure.

From
Peter Eisentraut
Date:
Tom Lane writes:

> Robert Hentosh <hentosh@io.com> writes:
> > I am receiving a regression test failure in about 1 out of 12 test runs.
>
> I am amazed no one has reported this before :-(

Someone did. :-)

--
Peter Eisentraut   peter_e@gmx.net

Re: Regression test failure.

From
Tom Lane
Date:
Peter Eisentraut <peter_e@gmx.net> writes:
> Tom Lane writes:
>> Robert Hentosh <hentosh@io.com> writes:
> I am receiving a regression test failure in about 1 out of 12 test runs.
>>
>> I am amazed no one has reported this before :-(

> Someone did. :-)

Yeah, I found your message awhile later.  I was surprised you hadn't
fixed the schedule file...

            regards, tom lane

Re: Regression test failure.

From
Robert Hentosh
Date:
On Tue, 22 Oct 2002, Tom Lane wrote:

> Robert Hentosh <hentosh@io.com> writes:
> > I am receiving a regression test failure in about 1 out of 12 test runs.
>
> I am amazed no one has reported this before :-(
>
> Upon looking, I see that the parallel test schedule tries to run
> create_misc and create_index in parallel, but the latter uses tables
> that are created by the former.  AFAICT the diffs you show can all
> be explained by create_misc not having executed completely when
> create_index looks at those tables.

Thanks for the prompt diagnosis.

> By rights this failure should have been noticed long ago.  Perhaps there
> is some scheduler peculiarity on your machine that makes it more prone
> than others to execute the different jobs with just the right timing to
> manifest the error.

It occured on my first test after I compiled it.  I then repeated the
regression test and couldn't get it to happen again and was thinking maybe
it only occured on a fresh install.  I was about to do a fresh install,
when I got it to fail again on the 24th "make check" according to my
history file.

I probably should have scripted it...

Regards,
Robert

Re: Regression test failure.

From
Robert Hentosh
Date:
On Tue, 22 Oct 2002, Tom Lane wrote:

> Peter Eisentraut <peter_e@gmx.net> writes:
> > Tom Lane writes:
> >> Robert Hentosh <hentosh@io.com> writes:
> > I am receiving a regression test failure in about 1 out of 12 test runs.
> >>
> >> I am amazed no one has reported this before :-(
>
> > Someone did. :-)
>
> Yeah, I found your message awhile later.  I was surprised you hadn't
> fixed the schedule file...
>
>             regards, tom lane


Was this a post in the bugs mailling list?  I am just wondering if I
didn't search for the problem correctly.

Regards,
Robert