Date: Fri, 22 Nov 2019 11:25:28 -0300
From: Alvaro Herrera <alvherre@2ndquadrant.com>
To: Michael Paquier <michael@paquier.xyz>
Cc: Antonin Houska <ah@cybertec.at>, Thomas Munro <thomas.munro@gmail.com>,
	Robert Haas <robertmhaas@gmail.com>,
	"pgsql-hackers@postgresql.org" <pgsql-hackers@postgresql.org>
Subject: Re: Attempt to consolidate reading of XLOG page
Message-ID: <20191122142528.GA6465@alvherre.pgsql>
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <20191122134007.GA180646@paquier.xyz>
User-Agent: Mutt/1.10.1 (2018-07-13)
Precedence: bulk

On 2019-Nov-22, Michael Paquier wrote:

> On Fri, Nov 22, 2019 at 10:35:51AM -0300, Alvaro Herrera wrote:
> > FWIW I think the new code is buggy because it doesn't seem to be setting
> > ws_off, so I suppose the optimization in ReadPageInternal to skip
> > reading the page when it's already the page we have is not hit, except
> > for the first page in the segment.  I didn't verify this, just my
> > impression while reading the code.
> 
> FWIW, this matches with my impression here, third paragraph:
> https://www.postgresql.org/message-id/20191120083802.GB47145@paquier.xyz

Ah, right.

I was wondering if we shouldn't do away with the concept of "offset" as
such, since the offset there is always forcibly set to the start of a
page.  Why don't we count page numbers instead?  It seems like the
interface is confusingly generic (measure in bytes) yet not offer any
extra functionality that could not be obtained with a simpler struct
repr (measure in pages).

But then that's not something that we need to change in this patch.

-- 
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services