Re: On-demand running query plans using auto_explain and signals

Hello,

On 15/10/2015 16:04, Pavel Stehule wrote:
>
>
> 2015-10-15 15:42 GMT+02:00 Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us
> <mailto:tgl(at)sss(dot)pgh(dot)pa(dot)us>>:
>
> "Shulgin, Oleksandr" <oleksandr(dot)shulgin(at)zalando(dot)de
> <mailto:oleksandr(dot)shulgin(at)zalando(dot)de>> writes:
> > I was thinking about this and what seems to be the biggest problem is when
> > to actually turn the feature on. It seems unlikely that someone will want
> > to enable it unconditionally. Enabling per-backend also doesn't seem to be
> > a good approach because you don't know if the next query you'd like to look
> > at is going to run in this exact backend.
>
> Check.
>
> > What might be actually usable is poking pg_stat_statements for queryid to
> > decide if we need to do explain (and possibly analyze).
>
> Hm, interesting thought.
>
> > Does this make sense to you? Does this make a good argument for merging
> > pg_stat_statements and auto_explain into core?
>
> I'd say more that it's a good argument for moving this feature out to
> one of those extensions, or perhaps building a third extension that
> depends on both of those. TBH, none of this stuff smells to me like
> something that ought to be in core.
>
>
> There are few features, that I would to see in core:
>
> 1. possibility to get full SQL string
> 2. possibility to get state string
>
> We can speak how to do it well.
>

I registered as reviewer on this, but after reading the whole thread for
the second time, it's still not clear to me if the last two submitted
patches (0001-Add-auto_explain.publish_plans.patch and
0002-Add-SHM-table-of-contents-to-the-explain-DSM.patch) are still
needed reviews, since multiple refactoring ideas and objections have
been raised since.

Regards.

--
Julien Rouhaud
http://dalibo.com - http://dalibo.org