Date: Thu, 4 Apr 2019 14:45:20 -0700
From: Andres Freund <andres@anarazel.de>
To: Alvaro Herrera <alvherre@2ndquadrant.com>
Cc: Pavan Deolasee <pavan.deolasee@gmail.com>,
	Masahiko Sawada <sawada.mshk@gmail.com>,
	pgsql-hackers <pgsql-hackers@postgresql.org>,
	Jeff Janes <jeff.janes@gmail.com>,
	Simon Riggs <simon@2ndquadrant.com>,
	Tomas Vondra <tomas.vondra@2ndquadrant.com>
Subject: Re: COPY FREEZE and setting PD_ALL_VISIBLE/visibility map bits
Message-ID: <20190404214520.7rfufsubxekodypr@alap3.anarazel.de>
References: <20190404172417.gua2xmn754uoxzto@alap3.anarazel.de>
 <20190404191554.GA7161@alvherre.pgsql>
 <20190404192308.mumsbtmccqyuebdh@alap3.anarazel.de>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20190404192308.mumsbtmccqyuebdh@alap3.anarazel.de>
Precedence: bulk

Hi,

On 2019-04-04 12:23:08 -0700, Andres Freund wrote:
> Also, how is this code even close to correct?
> CheckAndSetPageAllVisible() modifies the buffer in a crucial way, and
> there's no WAL logging? Without even a comment arguing why that's OK (I
> don't think it is)?

Peter Geoghegan just reminded me over IM that there's actually logging
inside log_heap_visible(), called from visibilitymap_set(). Still lacks
a critical section though.

I still think this is the wrong architecture.

Greetings,

Andres Freund

PS: We're going to have to revamp visibilitymap_set() soon-ish - the
fact that it directly calls heap routines inside is bad, means that
additional AMs e.g. zheap has to reimplement that routine.