[英]offline limited multi-master in Postgres
Site A will be generating a set of records. 站点A将生成一组记录。 Nightly they will backup their database and ftp it to Site B. Site B will not be modifying those records at all, but will be adding more records and other tables will be creating FK's to Site A's records.
每晚他们将备份他们的数据库并将其ftp到站点B.站点B根本不会修改这些记录,但会添加更多记录,其他表格将创建FK到站点A的记录。
So, essentially, I need to setup a system to take all the incremental changes from Site A's dump (mostly inserts and updates, but some deletes possible) and apply them at Site B. 因此,基本上,我需要设置一个系统来从站点A的转储中进行所有增量更改(主要是插入和更新,但有些删除可能)并在站点B应用它们。
At this point, we're using Postgres 8.3, but could upgrade if valuable. 此时,我们正在使用Postgres 8.3,但如果有价值可以升级。
I believe I can do this relatively straight-forwardly with Bucardo but I'd love to hear alternatives (or confirmation of Bucardo) before I setup a linux box to test it out. 我相信我可以用Bucardo相对直接地做到这一点,但在设置一个linux盒子来测试它之前,我很想听听替代品(或Bucardo的确认)。
Most every replication solution would do your trick. 大多数复制解决方案都可以解决问题。 The Postgres Wiki has a chapter on the topic .
Postgres Wiki有一个关于该主题的章节 。 But your case is simple enough.
但你的情况很简单。 I would just use dblink .
我只想使用dblink 。
This is generalized from a working implementation of mine: 这是从我的工作实现推广的:
Create a view in the master db that returns updated rows. 在主数据库中创建一个返回更新行的视图。
Let's call it myview
. 我们称之为
myview
。
Create one function per table in the slave db that fetches rows via dblink: 在从属数据库中为每个表创建一个函数,该函数通过dblink获取行:
CREATE OR REPLACE FUNCTION f_lnk_mytbl()
RETURNS TABLE(col_a integer, col_b text, col_c text) AS
$func$
SELECT *
FROM public.dblink('SELECT col_a, col_b, col_c FROM myview')
AS b(col_a integer, col_b text, col_c text);
$func$ LANGUAGE sql SECURITY DEFINER;
REVOKE ALL ON FUNCTION f_lnk_mytbl() FROM public;
GRANT EXECUTE ON FUNCTION f_lnk_mytbl() TO my_user;
CREATE OR REPLACE FUNCTION f_mysync()
RETURNS void AS
$func$
BEGIN
PERFORM dblink_connect(
'hostaddr=123.45.67.89 port=5432 dbname=mydb user=postgres password=secret');
-- Fetch data into local temporary table for fast processing.
CREATE TEMP TABLE tmp_i ON COMMIT DROP AS
SELECT * FROM f_lnk_mytbl();
-- *Or* read local files into temp tables with COPY so you don't need dblink.
-- UPDATE what's already there (instead of DELETE, to keep integrity).
UPDATE mytbl m
SET ( col_a, col_b, col_c) =
(i.col_a, i.col_b, i.col_c)
FROM tmp_i i
WHERE m.id = i.id
AND (m.col_a, m.col_b, m.col_c) IS DISTINCT FROM
(i.col_a, i.col_b, i.col_c);
-- INSERT new rows
INSERT INTO mytbl
SELECT * FROM tmp_i i
WHERE NOT EXISTS (SELECT 1 FROM mytbl m WHERE m.id = i.id);
-- DELETE anything? More tables?
PERFORM dblink_disconnect();
END
$func$ LANGUAGE plpgsql SECURITY DEFINER;
REVOKE ALL ON FUNCTION f_mysync() FROM public;
GRANT EXECUTE ON FUNCTION f_mysync() TO my_user;
my_user
. my_user
身份进行my_user
。 Schedule a cronjob or something. SELECT f_sync_mytbl();
In PostgreSQL 9.1 or later there is also the new CREATE FOREIGN TABLE . 在PostgreSQL 9.1或更高版本中,还有新的CREATE FOREIGN TABLE 。 Might be more elegant.
可能会更优雅。
声明:本站的技术帖子网页,遵循CC BY-SA 4.0协议,如果您需要转载,请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.