Re: pgsql: Use slots in trigger infrastructure, except for the actual invoc

Hi,

On 2019-02-27 10:16:21 -0800, Andres Freund wrote:
> On 2019-02-27 12:59:16 -0500, Andrew Dunstan wrote:
> > Backtraces should be enabled in the animal - I will investigate why
> > we're not getting them here. Meanwhile, here's one for this problem:
>
> Thanks, that helps! My first theories as to what's going on fell flat,
> unfortunately. I guess I'll have to figure out how to run that locally
> :(.

The issue was just that we didn't allow materializing virtual tuples
stored in a buffer tuple table slot. Locally allowing that fixed the
problem for me, I assume that'll also fix the crake. It was already on a
list of patches for pluggable storage...

Btw, shouldn't redisExecForeignDelete return NULL when there's no row
matched for the deletion? That's possible, I'd assume, if there's a
concurrent deletion.

Greetings,

Andres Freund