Re: Schema variables - new implementation for Postgres 15 (typo)

Hi

čt 22. 12. 2022 v 22:23 odesílatel Dmitry Dolgov <9erthalion6(at)gmail(dot)com>
napsal:

> > On Thu, Dec 22, 2022 at 08:45:57PM +0100, Pavel Stehule wrote:
> > > From the first look it seems some major topics the discussion is
> evolving
> > > are about:
> > >
> > > * Validity of the use case. Seems to be quite convincingly addressed in
> > > [1] and
> > > [2].
> > >
> > > * Complicated logic around invalidation, concurrent create/drop etc. (I
> > > guess
> > > the issue above is falling into the same category).
> > >
> > > * Concerns that session variables could repeat some problems of
> temporary
> > > tables.
> > >
>

I am sending an updated patch, fixing the mentioned issue. Big thanks for
testing, and checking.

> >
> > Why do you think so? The variable has no mvcc support - it is just stored
> > value with local visibility without mvcc support. There can be little bit
> > similar issues like with global temporary tables.
>
> Yeah, sorry for not being precise, I mean global temporary tables. This
> is not my analysis, I've simply picked up it was mentioned a couple of
> times here. The points above are not meant to serve as an objection
> against the patch, but rather to figure out if there are any gaps left
> to address and come up with some sort of plan with "committed" as a
> final destination.
>

There are some similarities, but there are a lot of differences too.
Handling of metadata is partially similar, but session variable is almost
the value cached in session memory. It has no statistics, it is not stored
in a file. Because there is different storage, I don't think there is some
intersection on implementation level.

Regards

Pavel