Re: Simplifying replication

From: Josh Berkus <josh(at)agliodbs(dot)com>
To: Fujii Masao <masao(dot)fujii(at)gmail(dot)com>
Cc: Robert Haas <robertmhaas(at)gmail(dot)com>, Bruce Momjian <bruce(at)momjian(dot)us>, Greg Stark <gsstark(at)mit(dot)edu>, pgsql-hackers(at)postgresql(dot)org
Subject: Re: Simplifying replication
Date: 2010-10-22 17:18:06
Message-ID: 4CC1C74E.8080409@agliodbs.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers


> Please see
> http://archives.postgresql.org/pgsql-docs/2010-10/msg00038.php

Ye gods and little fishes!

You really want to talk arcane formulas. I've re-read that
three times, and am still not sure that I could tell someone
definitively how much disk space WAL needs for a given group of
settings. I'll also point out that that formula is not in our docs --
what's an appropriate location?

I think this needs to be corrected in 9.1, *even if it means breaking
backwards compatibility*.

What would be sensible for DBAs is to have two settings:

max_wal_size
min_wal_size

These would be expresses in MB or GB and would be simple direct
quantities, which our formulas would work backwards from. max_wal_size
would be a hard limit (i.e. Postgres would stop accepting writes if we
hit it), and Admins would not be allowed to set min_wal_size to more
than max_wal_size - 2.

Even better would be to replace min_wal_size with min_wal_time, which
would set a time span for the oldest WAL segment to be kept (up to
max_wal_size - 2). Hmmm. That doesn't seem that hard to implement.
Is it?

(BTW, Robert, that e-mail is what I meant by "relationship")

--
-- Josh Berkus
PostgreSQL Experts Inc.
http://www.pgexperts.com

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Marios Vodas 2010-10-22 17:19:49 Re: gist DatumGetPointer returns pointer to corrupted data
Previous Message Dimitri Fontaine 2010-10-22 16:53:03 Re: Extensions, this time with a patch