Re: Temporary tables under hot standby

On Thu, Apr 26, 2012 at 5:57 AM, Robert Haas <robertmhaas@gmail.com> wrote:
> On Wed, Apr 25, 2012 at 10:04 PM, Noah Misch <noah@leadboat.com> wrote:
>> Based on the range of assessments spanning your "almost useless" to Merlin's
>> "killer feature", I gather that its utility is exceptionally site-specific.
>
> Well said, sir.

Not well said. This is spurious. This is not a balanced point.

100% of all temp table use cases currently define them dynamically. If
we implement a feature that *requires* them to all suddenly be defined
statically, then we're requiring those applications to
Either 1)a rewrite *all* their existing custom code
1)b Don't work with external applications that don't use GTTs (who the
hell does?)
2) have problems if any of their code doesn't allow static definition

Implementing a feature that *requires* those things is madness and
obscuring those crucial points is not balanced or fair.

Merlin is right to say that having GTTs would be a step up from what
we have now, but then so would a different and actually useful
implementation. So GTTs alone are not a killer feature. The killer
feature comes from solving the problems of running temp tables on HS.
I accept GTTs are a feature, but certainly not a killer one.

If we are interested in well-principled software then we must address
*both* the utility and the quality of our implementation. We cannot
just shrug our way through the boring bit of feature design and then
agonise over a well principled implementation. We must recognise that
our wish to get on with the interesting bit gets in the way of
spending sensible time balancing what we do to get the most out of our
work. Noah will enjoy writing this; I just want to make sure he is
enjoying something that will be useful, not just a tick box on the
release notes that will cause further frustration when users find out
the details.

--
 Simon Riggs                   http://www.2ndQuadrant.com/
 PostgreSQL Development, 24x7 Support, Training & Services