Re: Read uncommitted ever possible?

On Thu, Mar 10, 2011 at 12:31 PM, hans wulf <lotu1(at)gmx(dot)net> wrote:
> I don't know the postgres code, but I don't thing it is a big deal, not to care about consistancy. The code for executing such a query should be quite basic, because no MVCC-Stuff has to be done.

I remember I used to think this would be simple -- just return all
rows regardless of visibility.

However I later learned things were more complex than that. You
probably want to return the latest version of any row regardless of
whether it's committed but *not* return two or more copies of the same
row which would really make the results entirely meaningless. That
alone would make it prohibitively hard to do.

I think I remember issues with potentially running into old rows that
don't even match the current definition of the table. That would
potentially cause you to crash or output garbage. However offhand I
don't see how that would be possible so perhaps I'm misremembering
this issue.

--
greg