Re: Any real known bugs about wrong selects?

On Tue, 13 Jan 2004, Harry Jackson wrote:

> David Teran wrote:
> >
> > The real question was: are there open known bugs where a select
> > statement does not return a correct result, meaning a wrong number of
> > rows? Were there a lot of errors like this?
> > We are asking this because this makes trouble with the database we are
> > currently using. Of course we know that there is no guarantee that pgsql
> > works 100% bugfree but we are only asking for user experience.
>
> An error of this nature in any database is a very serious problem and
> you should really take it up with the vendor. If you can reproduce it I
> am sure they would try and supply a patch as soon as feasibly possible.
>
> I have seen databases return the wrong number of rows to queries loads
> of times or at least that is what it appears to be. On further
> investigation these apparent bugs turn out to be user errors in
> complicated sql statements. To date I have never seen a database return
> the anything other than what I have asked it for and Postgres is in this
> list.

Just an FYI, MSSQL server, in an older flavor from several years ago, had
a bug where a statement like:

select * into newtable from oldtable

would return the right number of rows, as would

select * from oldtable order by somefield

however,

select * into newtable from oldtable order by somefield

would, under heavy load, return fewer rows than it should. With no error.

Seeing as how the third query is often viewed as a "poor man's cluster"
and wrapped in a begin;end; as well as rename / drop table script, that's
quite dangerous to your data. MS fixed that one pretty quick, even though
they hemmed and hawwed about how it wasn't very common, etc...

I've seen a few other isolated instances of this kind of thing happening
to other people, but seeing as how I use postgresql, I've never seen it
myself. :-)