--1 环境信息
Postgresql 版本: Postgresql 9.1beta3
OS : Red Hat Enterprise Linux Server release 5.5
硬件环境 : 笔记本上的两台虚拟机
主库 IP : 192.168.1.25
主库 PORT : 1921
备库 IP : 192.168.1.26
备库 PORT : 1921
--2 如何区别主库和备库
有时在论坛上会有人问,如何区分主库和备库,这里提供两种方法。
--2.1 方法一,根据主机进程判断
[postgres@pg1 pg_root]$ ps -ef | grep "wal"
postgres 17715 17684 0 20:41 ? 00:00:00 postgres: wal writer process
postgres 17746 17684 0 20:43 ? 00:00:00 postgres: wal sender process repuser 192.168.1.26(43246) streaming 0/700178A8
postgres 17819 17590 0 21:00 pts/2 00:00:00 grep wal
备注:这里显示了wal 日志发送进程"wal sender process",说明是主库。
[postgres@pgb pg_xlog]$ ps -ef | grep wal
postgres 29436 29386 0 20:43 ? 00:00:00 postgres: wal receiver process streaming 0/700178A8
postgres 29460 29289 0 21:00 pts/3 00:00:00 grep wal
备注:这里显示了 wal 日志接收进程 "wal receiver process" ,说明是备库;
--2.2 方法二,根据 pg_controldata 输出
pg_controldata 输出数据库服务的当前状态,可以根据 "Database cluster state: " 的信息来判断,
如果值为 "in production" 说明是主库,如果值为 "in archive recovery" 说明是备库。
主库的 pg_controldata 输出
[postgres@pg1 pg_root]$ pg_controldata
pg_control version number: 903
Catalog version number: 201105231
Database system identifier: 5640897481082175487
Database cluster state: in production
...
备库的 pg_controldata 输出
[postgres@pgb pg_xlog]$ pg_controldata
pg_control version number: 903
Catalog version number: 201105231
Database system identifier: 5640897481082175487
Database cluster state: in archive recovery
...
--3 recovery.conf 文件介绍
recovery.conf 是一个配置文件,用于主库,备库切换时的参数配置,可以从 $PGHOME/share 目录下
复制一份 recovery.conf.sample 到备库 $PGDATA 目录,里面有众多参数,这里只介绍用于切换时的
关键参数
standby_mode = '' --标记PG为STANDBY SERVER
primary_conninfo = '' --标识主库信息
trigger_file = '' --标识触发器文件
主备切换
--4.1 创建备库recovery.conf 文件( On Slave )
cp $PGHOME/share/recovery.conf.sample $PGDATA/recovery.conf
配置以下参数
standby_mode = 'on' --标记PG为STANDBY SERVER
primary_conninfo = 'host=192.168.1.25 port=1921 user=repuser ' --标识主库信息
trigger_file = '/opt/pgdata/pg_root/postgresql.trigger.1921' --标识触发器文件
--4.2 关闭主库(on Primary)
[postgres@pg1 pg_root]$ pg_ctl stop -m fast -D $PGDATA
waiting for server to shut down....... done
server stopped
--4.3 激活备库到主库状态 ( on slave )
激活备库只要创建一个文件即可,根据备库 recovery.conf 配置文件的参数 trigger_file 值,
创建这个 trigger 文件即可。 例如 "touch /opt/pgdata/pg_root/postgresql.trigger.1921 "
[postgres@pgb pg_root]$ touch /opt/pgdata/pg_root/postgresql.trigger.1921
过一会儿发现 recovery.conf 文件变成 recovery.done ,说明备库已经激活。
[postgres@pgb pg_root]$ ll
total 176K
-rw------- 1 postgres postgres 168 Aug 24 10:24 backup_label.old
drwx------ 5 postgres postgres 4.0K Aug 15 10:03 base
drwx------ 2 postgres postgres 4.0K Aug 24 20:50 global
drwx------ 2 postgres postgres 4.0K Aug 15 10:03 pg_clog
-rw------- 1 postgres postgres 4.5K Aug 24 10:39 pg_hba.conf
-rw------- 1 postgres postgres 1.6K Aug 15 10:03 pg_ident.conf
drwx------ 4 postgres postgres 4.0K Aug 15 10:03 pg_multixact
drwx------ 2 postgres postgres 4.0K Aug 24 20:42 pg_notify
drwx------ 2 postgres postgres 4.0K Aug 15 10:03 pg_serial
drwx------ 2 postgres postgres 4.0K Aug 15 10:03 pg_stat_tmp
drwx------ 2 postgres postgres 4.0K Aug 15 10:03 pg_subtrans
drwx------ 2 postgres postgres 4.0K Aug 21 20:21 pg_tblspc
drwx------ 2 postgres postgres 4.0K Aug 15 10:03 pg_twophase
-rw------- 1 postgres postgres 4 Aug 15 10:03 PG_VERSION
drwx------ 3 postgres postgres 4.0K Aug 24 21:20 pg_xlog
-rw------- 1 postgres postgres 19K Aug 24 10:24 postgresql.conf
-rw------- 1 postgres postgres 51 Aug 24 20:42 postmaster.opts
-rw------- 1 postgres postgres 69 Aug 24 20:42 postmaster.pid
-rw-r--r-- 1 postgres postgres 4.7K Aug 24 20:42 recovery.conf
--查看从库CSV日志(正在激活成主库)
2011-08-24 21:20:55.130 CST,29388,4e54f1c5.72cc,11,2011-08-24 20:42:45 CST,1/0,LOG,00000,"selected new timeline ID: 6",""
2011-08-24 21:20:58.119 CST,12,"archive recovery complete",""
2011-08-24 21:20:58.495 CST,29386,4e54f1c3.72ca,5,2011-08-24 20:42:43 CST,"database system is ready to accept connections",""
说明从库已经为OPEN状态,可以进行读写操作。
--4.4 激活原来的主库,让其转变成从库 (在原来的主库上执行)
--创建 $PGDATA/recovery.conf 文件,配置以下参数
recovery_target_timeline = 'latest'
standby_mode = 'on' --标记PG为STANDBY SERVER
primary_conninfo = 'host=192.168.1.26 port=1921 user=repuser ' --标识主库信息
trigger_file = '/opt/pgdata/pg_root/postgresql.trigger.1921' --标识触发器文件
--创建密码文件 /home/postgres/.pgpass 密码文件,输入以下内容
192.168.1.26:1921:replication:repuser:rep123us345er
--修改 pg_hba.conf (现在的主库上增加),添加以下配置
host replication repuser 192.168.1.25/32 md5
--将原来的主库(现在的从库)启动
[postgres@pg1 pg_root]$ pg_ctl start -D $PGDATA
server starting
--查看从库日志, 发现大量 FATAL 错误信息
2011-08-24 21:31:59.178 CST,17889,4e54fd4f.45e1,1,2011-08-24 21:31:59 CST,FATAL,XX000,"timeline 6 of the primary does not match recovery target timeline 5",""
2011-08-24 21:32:04.208 CST,17891,4e54fd54.45e3,2011-08-24 21:32:04 CST,""
2011-08-24 21:32:09.135 CST,17892,4e54fd59.45e4,2011-08-24 21:32:09 CST,""
2011-08-24 21:32:14.136 CST,17895,4e54fd5e.45e7,2011-08-24 21:32:14 CST,""
备注:出现了大量 FATAL,"timeline 6 of the primary does not match recovery target timeline 5”
估计是时间线有问题,网上查了下资料也没啥结果,后来咨询了下德哥,只要将从库 $PGDATA/pg_xlog
一个文件考过来就行。
--将主库文件 00000006.history 复制到从库
[postgres@pgb pg_xlog]$ scp 00000006.history postgres@192.168.1.25:/opt/pgdata/pg_root/pg_xlog
postgres@192.168.1.25's password:
00000006.history
--再次查看从库日志
2011-08-24 21:36:04.819 CST,17948,4e54fe44.461c,2011-08-24 21:36:04 CST,""
2011-08-24 21:36:09.742 CST,17885,4e54fd44.45dd,2011-08-24 21:31:48 CST,"new target timeline is 6",""
2011-08-24 21:36:09.824 CST,17977,4e54fe49.4639,2011-08-24 21:36:09 CST,"streaming replication successfully connected to primary",""
备注:根据日志信息,说明从库已经恢复正常;
--4.5 测试
主库上创建一张表
postgres=# \c skytf skytf
You are now connected to database "skytf" as user "skytf".
skytf=> \d
List of relations
Schema | Name | Type | Owner
--------+--------------------+-------+----------
public | pg_stat_statements | view | postgres
skytf | pgbench_accounts | table | skytf
skytf | pgbench_branches | table | skytf
skytf | pgbench_history | table | skytf
skytf | pgbench_tellers | table | skytf
skytf | test_stadnby | table | skytf
(16 rows)
skytf=> create table test_11 (id integer,name varchar(32));
CREATE TABLE
skytf=> \d
List of relations
Schema | Name | Type | Owner
--------+--------------------+-------+----------
public | pg_stat_statements | view | postgres
skytf | pgbench_accounts | table | skytf
skytf | pgbench_branches | table | skytf
skytf | pgbench_history | table | skytf
skytf | pgbench_tellers | table | skytf
skytf | test_11 | table | skytf
skytf | test_stadnby | table | skytf
(17 rows)
从库上查询
[postgres@pgb pg_root]$ psql
psql (9.1beta3)
Type "help" for help.
postgres=# \c skytf skytf
skytf=> \d
List of relations
Schema | Name | Type | Owner
--------+--------------------+-------+----------
public | pg_stat_statements | view | postgres
skytf | pgbench_accounts | table | skytf
skytf | pgbench_branches | table | skytf
skytf | pgbench_history | table | skytf
skytf | pgbench_tellers | table | skytf
skytf | test_11 | table | skytf
skytf | test_stadnby | table | skytf
备注:可见表 test_11 迅速从主库上同步过来了,到此为止,库切换完成。
--5 总结: 1 Hot-Standby 切换步骤比较多,有些配置可以提前做好的,例如 .pgpass,pg_hba.conf 等; 2 主,备切换时,务必先将主库关闭,否则一旦从库被激活时,而主库尚未关闭,会有问题; 3 主,备切换可作为生产库迁移的一种方式,因为这最大限度减少了业务停机时间。