From: Tom Lane <tgl@sss.pgh.pa.us>
To: Stephen Frost <sfrost@snowman.net>
cc: Bruce Momjian <bruce@momjian.us>, Christoph Berg <myon@debian.org>,
        Julien Rouhaud <rjuju123@gmail.com>,
        Pavel Stehule <pavel.stehule@gmail.com>,
        Kyotaro Horiguchi <horikyota.ntt@gmail.com>,
        Magnus Hagander <magnus@hagander.net>,
        Fujii Masao <masao.fujii@oss.nttdata.com>,
        Michael Paquier <michael@paquier.xyz>,
        Andres Freund <andres@anarazel.de>,
        Alvaro Herrera <alvherre@alvh.no-ip.org>,
        Peter Eisentraut <peter.eisentraut@enterprisedb.com>,
        PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Subject: Re: compute_query_id and pg_stat_statements
In-reply-to: <20210513180706.GG20766@tamriel.snowman.net>
References: <20210513005236.aj7n6b76yjsxythc@nol>
 <20210513011325.GA11075@momjian.us> <20210513015700.fp33q5rfvwl2ahfx@nol>
 <20210513030652.GB11075@momjian.us> <20210513031613.j47baglf6xl3mgbs@nol>
 <20210513033332.GF11075@momjian.us> <20210513040342.fouh7ryglrfgxglu@nol>
 <20210513144143.GG11075@momjian.us> <YJ1kYeKvYBk0cxti@msg.df7cb.de>
 <20210513175947.GP11075@momjian.us>
 <20210513180706.GG20766@tamriel.snowman.net>
Comments: In-reply-to Stephen Frost <sfrost@snowman.net>
	message dated "Thu, 13 May 2021 14:07:06 -0400"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <1579534.1620930549.1@sss.pgh.pa.us>
Date: Thu, 13 May 2021 14:29:09 -0400
Message-ID: <1579535.1620930549@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/1579535.1620930549%40sss.pgh.pa.us>
Precedence: bulk

Stephen Frost <sfrost@snowman.net> writes:
> There's a ridiculously simple option here which is: drop the idea that
> we support an extension redefining the query id and then just make it
> on/off with the default to be 'on'.

I do not think that defaulting it to 'on' is acceptable unless you can
show that the added overhead is negligible.  IIUC the measurements that
have been done show the opposite.

Maybe we should revert this thing pending somebody doing the work to
make a version of queryid labeling that actually is negligibly cheap.
It certainly seems like that could be done; one more traversal of the
parse tree can't be that expensive in itself.  I suspect that the
performance problem is with the particular hashing mechanism that
was used, which looks mighty ad-hoc anyway.

			regards, tom lane