Horton

Re: [HACKERS] AdvanceXLInsertBuffer vs. WAL segment compressibility

Tom Lane <tgl@sss.pgh.pa.us>

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Chapman Flack <chap@anastigmatix.net>
Cc: Tomas Vondra <tomas.vondra@2ndquadrant.com>, Michael Paquier <michael@paquier.xyz>, Stephen Frost <sfrost@snowman.net>, Daniel Gustafsson <daniel@yesql.se>, Michael Paquier <michael.paquier@gmail.com>, Heikki Linnakangas <hlinnaka@iki.fi>, Robert Haas <robertmhaas@gmail.com>, Bruce Momjian <bruce@momjian.us>, PostgreSQL Hackers <pgsql-hackers@postgresql.org>
Date: 2018-03-30T18:18:46Z
Lists: pgsql-hackers 
Chapman Flack <chap@anastigmatix.net> writes:
> On 03/27/18 22:10, Michael Paquier wrote:
>> Here you go for one example:
>> https://sourceforge.net/projects/pglesslog/

> In any case, from my study of the commit, it is hard for me to see an issue.
> The code comment says: "mark the header to indicate that WAL records
> beginning in this page have removable backup blocks."

Yeah, that commit just moved a flag from individual WAL records to page
headers, arguing that it was okay to assume that the same flag value
applies to all records on a page.  If there are no records in the page,
it doesn't matter what you think the flag value is.

A potentially stronger complaint is that WAL-reading tools might fail
outright on a page with an invalid header, but I'd say that's a robustness
issue that they'd need to address anyway.  There's never been any
guarantee that the trailing pages of a WAL segment are valid.

			regards, tom lane

Commits

  1. Ensure that WAL pages skipped by a forced WAL switch are zero-filled.

  2. Improve scalability of WAL insertions.

  3. Move BKP_REMOVABLE bit from individual WAL records to WAL page headers.