Date: Fri, 15 Nov 2019 18:41:02 -0300
From: Alvaro Herrera <alvherre@2ndquadrant.com>
To: Antonin Houska <ah@cybertec.at>
Cc: Michael Paquier <michael@paquier.xyz>,
	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: <20191115214102.GA15616@alvherre.pgsql>
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <44237.1573556793@antos>
User-Agent: Mutt/1.10.1 (2018-07-13)
Precedence: bulk

On 2019-Nov-12, Antonin Houska wrote:

> ok, the next version uses explicit lseek(). Maybe the fact that XLOG is mostly
> read sequentially (i.e. without frequent seeks) is the reason pread() has't
> been adopted so far.

I don't quite understand why you backed off from switching to pread.  It
seemed a good change to me.

Here's a few edits on top of your latest.

The new routine WALRead() is not at all the same as the previous
XLogRead, so I don't see why we would keep the name.  Hence renamed.

I see no reason for the openSegment callback to return the FD in an out
param instead of straight return value.  Changed that way.

Having seek/open be a boolean "xlr_seek" seems a bit weird.  Changed to
an "operation" enum.  (Maybe if we go back to pg_pread we can get rid of
this.)  Accordingly, change WALReadRaiseError and WALDumpReadPage.

Change xlr_seg to be a struct rather than pointer to struct.  It seems a
bit dangerous to me to return a pointer that we don't know is going to
be valid at raise-error time.  Struct assignment works fine for the
purpose.

Renamed XLogDumpReadPage to WALDumpReadPage, because what the heck is
XLogDump anyway?  That doesn't exist.

I would only like to switch this back to pg_pread() (from seek/read) and
I'd be happy to commit this.

What is logical_read_local_xlog_page all about?  Seems useless.  Let's
get rid of it.

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