Thread

Re: [PATCH] refint: Avoid reusing cascade UPDATE plans.

Ayush Tiwari <ayushtiwari.slg01@gmail.com> — 2026-05-15T15:04:55Z

Hi,


On Fri, 15 May 2026 at 20:16, Nathan Bossart <nathandbossart@gmail.com>
wrote:

> On Fri, May 15, 2026 at 04:42:45PM +0200, Álvaro Herrera wrote:
> > On 2026-May-15, Nathan Bossart wrote:
> >> Yeah, this may be the best option.  I'm curious what others think.
> >
> > I think this bug makes it clear that this code has few enough users, and
> > a sufficient number of bugs, that we should just nuke it.  Do you really
> > want to spend so much time trying to fix it?  The module's documentation
> > says
> >
> >    (This functionality is long since superseded by the built-in foreign
> >    key mechanism, of course, but the module is still useful as an
> example.)
> >
> > but I question whether the usefulness value is really there.
> >
> > Would anyone oppose that?
>
> Big +1 for removing it in v20.  Or maybe even v19.  I do think it is worth
> trying to fix the more egregious bugs, though, as the code will still be
> supported for a few years.
>
>
I agree on the removal part.

In the meantime, since the module will still be
supported for a while, this seems like a focused fix for the more
egregious cache behavior.

Attached v3 removes the private SPI plan cache from refint entirely.
Both check_primary_key() and check_foreign_key() now prepare their SPI
plans per trigger invocation and let SPI_finish() release them.

Regards,
Ayush