Date: Tue, 29 Mar 2022 18:08:46 +0800
From: Julien Rouhaud <rjuju123@gmail.com>
To: Magnus Hagander <magnus@hagander.net>
Cc: Andres Freund <andres@anarazel.de>,
	Justin Pryzby <pryzby@telsasoft.com>,
	pgsql-hackers@lists.postgresql.org
Subject: Re: Add parameter jit_warn_above_fraction
Message-ID: <20220329100846.3iedk5ear3ewdi5j@jrouhaud>
References: 
 <CABUevExR_9ZmkYj-aBvDreDKUinWLBBpORcmTbuPdNb5vGOLtA@mail.gmail.com>
 <20220225162304.GD9008@telsasoft.com>
 <CABUevExe1fV0RzjA7vQRp_WkjtTVF1O9E5STKodWa2CmpFoqHA@mail.gmail.com>
 <20220321235048.wb3ymun6lbi33b2u@alap3.anarazel.de>
 <CABUevEy9H+HPAuM1ofC=OuY2+bESOsA4hHSAY6HbLYyTG1oaPw@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: 
 <CABUevEy9H+HPAuM1ofC=OuY2+bESOsA4hHSAY6HbLYyTG1oaPw@mail.gmail.com>
Archived-At: 
 <https://www.postgresql.org/message-id/20220329100846.3iedk5ear3ewdi5j%40jrouhaud>
Precedence: bulk

Hi,

On Mon, Mar 28, 2022 at 10:11:16PM +0200, Magnus Hagander wrote:
> On Tue, Mar 22, 2022 at 12:50 AM Andres Freund <andres@anarazel.de> wrote:
> >
> > This fails on cfbot, due to compiler warnings:
> > https://cirrus-ci.com/task/5127667648299008?logs=mingw_cross_warning#L390
>
>
> Huh. That's annoying. I forgot int64 is %d on linux and %lld on Windows :/
>
> PFA a fix for that.

The v3 is now stable on all platforms according to the cfbot, and apart from
the log level the patch looks good to me.

The last remaining thing is whether logging at WARNING level is the correct
choice.  I'm personally fine with it, because the people who are going to use
it will probably use the same approach as for log_min_duration_statements:
enable it first with a high value, and check if that just lead to a massive log
spam.  If not, see if there's anything that needs attention and fix it,
otherwise keep lowering it and keep going.

Also, jitting isn't particularly fast in general, so it's probably somewhat
unlikely to have a massive amount of queries that both have a cost higher than
the jit threshold cost and still run fast enough to spam the log without jit
slowing everything down.

I'm not sure what status the patch should be.  I think it's the committer's
decision to chose which log level to use, and that won't affect the patch
enough to justify another review, so I'm switching the patch to ready for
committer.  This doesn't prevent any further discussions for the log level.