Re: pgsql: Include information on buffer usage during planning phase, in EX

Fujii Masao <masao(dot)fujii(at)oss(dot)nttdata(dot)com> writes:
> On 2020/04/03 12:30, Tom Lane wrote:
>> so probably the thing to do is adapt
>> the filter functions in explain.sql so that they suppress Buffers lines
>> altogether in text output. Kind of annoying, but ...

> I'm thinking to suppress only Buffers line just after Planning Time line,
> by applying something like the following changes to explain_filter().
> Thought?

I think this is a bad idea. It's overcomplicated, and more to the
point: now that we've seen the problem we should realize that we're
eventually going to have failures for *any* Buffers line in text-mode
output. We're already filtering them so hard as to be nearly useless
(see a couple lines further down). I think we should just drop them
in text mode and be content with checking for them in non-text modes.

Robert's muttered about possibly introducing some EXPLAIN option
that would make the output a bit less variable. Perhaps we should
do that. I'm not quite sure I like the idea on theoretical grounds:
testing a different behavior from what you do in production seems
like the wrong thing. But it'd allow undoing decisions like this one.
It's too late to consider such a thing for v13, though --- four days
before feature freeze is no time to be designing new features.
So for now, we have to make the output stable with the tools we have,
and text-mode Buffers output is unstable by design.

regards, tom lane