Re: Visibility regression test

On Thu, 2002-08-29 at 16:37, Tom Lane wrote:
> Manfred Koizar <mkoi-pg(at)aon(dot)at> writes:
> > On Thu, 29 Aug 2002 10:30:59 -0400, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>
> > wrote:
> >> Manfred Koizar <mkoi-pg(at)aon(dot)at> writes:
> > A new regression test trying to detect runaway INSERTs/UPDATEs.
> >>
> >> Why?
>
> > Because we do not want to run a database that gets hung in an endless
> > loop on INSERT or UPDATE. Better we find such bugs during regression
> > testing.
>
> If there is such a problem it will surely be found by the other
> regression tests. I don't see a need to insert a test that has an
> acknowledged system dependency in order to detect this.
>

I agree with this, but I think an earlier suggestion of Manfred's,
(namely tests that explicitly check concurrency issues) might be useful
to verify the integrity of MVCC.

How about the following as a possible approach:

We produce an application which opens two (or more?) database
connections and feeds appropriate SQL to them. ISTM that this need not
be a very complicated application. It takes one input file whose lines
begin with (say) '-' for a comment, '1' for connection 1, '2' for
connection 2 etc. followed by the SQL statement to send. (This is all
very sketchy, of course -there might be better ways to format it). The
output from each backend is sent to a separate file for comparison
against the expected results.

Does this sound feasible or useful? It would offer a means to test tuple
visibility, concurrent updates and deadlock detection in a controlled
way without too much difficulty.

Regards

John
--
John Gray
Azuli IT
www.azuli.co.uk