PostgreSQL backup and recovery - online backup & Point-In-Time-Recovery / iVAN

前面介绍的冷备份, 以及逻辑备份都是一个时间点的全量备份, 没有增量的概念.

如果数据库在运行过程中发生故障, 使用逻辑备份只能将数据库还原到备份时刻, 无法恢复到故障发生前的那个时刻.

又或者在使用过程中由于误操作修改或删除了重要数据, 需要还原到误操作前的那个时刻怎么办呢?

使用冷备份加上有效的归档文件可以实现任意时间点的恢复. 但是冷备份需要停库操作, 所以实用性不大.

本文要介绍的是在线的数据库文件备份, 弥补了冷备份的缺陷, 同时又支持基于时间点的恢复.

postgresql.conf参数配置 :

1. wal日志级别 :

wal_level=archive 或 hot_standby

查看当前数据库的wal level :

pg93@db-172-16-3-33-> pg_controldata |grep wal_level
Current wal_level setting:            hot_standby

如果结果不为hot_standby或者archive, 那么需要调整一下postgresql.conf. 并重启数据库.

2. 归档模式 :

    archive_mode = on

如果原来是off的, 调整成on, 那么需要重启数据库.

3. 归档命令 :

首先要创建用于存放wal归档文件目录, 数据库启动用户需要写权限.

[root@db-172-16-3-33 postgresql-9.3beta1]# mkdir -p /pgdata/digoal/1921/data04/pg93archdir
[root@db-172-16-3-33 postgresql-9.3beta1]# chown pg93:pg93 /pgdata/digoal/1921/data04/pg93archdir
[root@db-172-16-3-33 postgresql-9.3beta1]# chmod 700 /pgdata/digoal/1921/data04/pg93archdir

然后在$PGDATA中创建归档脚本

[root@db-172-16-3-33 postgresql-9.3beta1]# su - pg93
pg93@db-172-16-3-33-> cd $PGDATA
pg93@db-172-16-3-33-> cat archive.sh
#!/bin/bash

export LANG=en_US.utf8
export PGHOME=/opt/pgsql9.3
export LD_LIBRARY_PATH=$PGHOME/lib:/lib64:/usr/lib64:/usr/local/lib64:/lib:/usr/lib:/usr/local/lib:$LD_LIBRARY_PATH
export DATE=`date +"%Y%m%d"`
export PATH=$PGHOME/bin:$PATH:.

BASEDIR="/pgdata/digoal/1921/data04/pg93archdir"

if [ ! -d $BASEDIR/$DATE ]; then
  mkdir -p $BASEDIR/$DATE
  if [ ! -d $BASEDIR/$DATE ]; then
    echo "error mkdir -p $BASEDIR/$DATE"
    exit 1
  fi
fi

cp $1 $BASEDIR/$DATE/$2
if [ $? -eq 0 ]; then
  exit 0
else
  echo -e "cp $1 $BASEDIR/$DATE/$2 error"
  exit 1
fi

echo -e "backup failed"
exit 1

给archive.sh添加可执行权限 :

chmod 700 archive.sh

配置postgresql.conf中的归档命令 :

archive_command = 'archive.sh %p %f'

修改完后reload配置文件即可

pg_ctl reload

配置完后检查一下归档是否正常 :

pg93@db-172-16-3-33-> psql
psql (9.3devel)
Type "help" for help.
digoal=# checkpoint;
CHECKPOINT
digoal=# select pg_switch_xlog();
 pg_switch_xlog
----------------
 F/2095D8
(1 row)
digoal=# select pg_switch_xlog();
 pg_switch_xlog
----------------
 F/1000000
(1 row)
digoal=# checkpoint;
CHECKPOINT
digoal=# select pg_switch_xlog();
 pg_switch_xlog
----------------
 F/10000E8
(1 row)
digoal=# checkpoint;
^[[ACHECKPOINT
digoal=# select pg_switch_xlog();
 pg_switch_xlog
----------------
 F/20000E8
(1 row)

查看归档文件 :

pg93@db-172-16-3-33-> cd /pgdata/digoal/1921/data04/pg93archdir
pg93@db-172-16-3-33-> ll
total 4.0K
drwx------ 2 pg93 pg93 4.0K May 27 10:56 20130527
pg93@db-172-16-3-33-> cd 20130527/
pg93@db-172-16-3-33-> ll
total 48M
-rw------- 1 pg93 pg93 16M May 27 10:56 000000030000000F00000000
-rw------- 1 pg93 pg93 16M May 27 10:56 000000030000000F00000001
-rw------- 1 pg93 pg93 16M May 27 10:56 000000030000000F00000002

具备了这些前提条件之后, 就可以进行在线备份了.

在线备份可以使用拷贝, rsync, 或者pg_basebackup命令进行.

方法一, 拷贝

首先在数据库中以超级用户执行如下命令 :

pg93@db-172-16-3-33-> psql
psql (9.3devel)
Type "help" for help.
digoal=# select pg_start_backup(now()::text);
 pg_start_backup
-----------------
 F/3000028
(1 row)

这个命令执行的目的是在$PGDATA中记录一个标签文件, 包含标签名now()::text, 以及执行这条指令的启动时间和WAL位置.

标签文件信息如下 :

[root@db-172-16-3-33 postgresql-9.3beta1]# su - pg93
pg93@db-172-16-3-33-> cd $PGDATA
pg93@db-172-16-3-33-> cat backup_label
START WAL LOCATION: F/3000028 (file 000000030000000F00000003)
CHECKPOINT LOCATION: F/3000060
BACKUP METHOD: pg_start_backup
BACKUP FROM: master
START TIME: 2013-05-27 11:02:58 CST
LABEL: 2013-05-27 11:02:58.578713+08

接下来就可以拷贝数据文件了.

pg93@db-172-16-3-33-> cp -r -L $PGDATA /pgdata/digoal/1921/data04/pg93backup/20130527_onlinebackup

拷贝完后清除备份文件中pg_xlog和pg_log中的内容, 因为这些是不需要的.

cd /pgdata/digoal/1921/data04/pg93backup/20130527_onlinebackup/pg_xlog
rm -rf *
cd /pgdata/digoal/1921/data04/pg93backup/20130527_onlinebackup/pg_log
rm -rf *

接下来使用超级用户在数据库中执行如下命令 :

pg93@db-172-16-3-33-> psql
psql (9.3devel)
Type "help" for help.
digoal=# select pg_stop_backup();
NOTICE:  pg_stop_backup complete, all required WAL segments have been archived
 pg_stop_backup
----------------
 F/3001A20
(1 row)

然后要备份这段时间产生的归档文件 :

需要的归档文件是哪些呢?

备份开始是的XLOG文件在backup_label中已经指出了000000030000000F00000003, 结束时的XLOG文件也可以从pg_stop_backup()的输出中得到 :

digoal=#  select pg_xlogfile_name('F/3001A20');
     pg_xlogfile_name
--------------------------
 000000030000000F00000003
(1 row)

如果你的PostgreSQL版本没有这个函数, 那么就自己计算一下文件名 :

参考 :

http://blog.163.com/digoal@126/blog/static/1638770402012914112949546/

因此本例中需要的文件只有1个就是000000030000000F00000003

如果归档文件本来就不是放在本地文件系统中, 那么也不需要二次备份归档文件. 只要别删掉需要的归档文件就行了.

方法二, rsync

方法与拷贝类似, 也需要先执行pg_start_backup, 然后调用rsync, 最后调用pg_stop_backup.

命令举例 :

pg93@db-172-16-3-33-> rsync -acvz -L --exclude "pg_xlog" --exclude "pg_log" $PGDATA /pgdata/digoal/1921/data04/pg93backup/20130527_backup

方法三, pg_basebackup

pg_basebackup是PostgreSQL 9.1 引入的一个数据库备份命令, pg_basebackup利用流复制协议传输数据库文件.

使用pg_basebackup备份数据库需要用到超级用户或者replication角色用户.

另外就是pg_basebackup不需要执行pg_start_backup和pg_stop_backup, 因为命令中已经封装了这些操作.

并且支持多个pg_basebackup并行备份同一个数据库集群.

例如 :

pg93@db-172-16-3-33-> psql
psql (9.3devel)
Type "help" for help.
digoal=# create role replica login replication encrypted password 'replica';
CREATE ROLE

配置pg_hba.conf , 允许连接.

host    replication     replica        172.16.3.40/32            md5
pg_ctl reload

在另一台主机备份 :

pg93@db-172-16-3-40-> pg_basebackup -D ./basebackup_20130527 -F p -X stream -h 172.16.3.33 -p 1999 -U replica
Password:
WARNING:  skipping special file "./.s.PGSQL.1999"

需要注意, 以pg_basebackup备份, 如果表空间使用了软链接, 那么备份时在目标端也会创建软链接, 因此需要先保证目录存在并有写权限.

例如本例 :

pg93@db-172-16-3-40-> cd basebackup_20130527/
pg93@db-172-16-3-40-> cd pg_tblspc/
pg93@db-172-16-3-40-> ll
total 0
lrwxrwxrwx 1 pg93 pg93 47 May 27 12:28 26425 -> /pgdata/digoal/1921/data03/pg93/1999/tbs_digoal

那么目标端的/pgdata/digoal/1921/data03/pg93/1999 这个目录要先创建好, 并提供写权限.

PITR还原举例 :

PostgreSQL 支持指定还原点的位置, 即数据库恢复到什么位置停下来.

4个recovery.conf参数恢复停止在哪个位置.

1. recovery_target_name, 指pg_create_restore_point(text)创建的还原点, 如果有重名的还原点, 那么在recovery过程中第一个遇到的还原点即停止.

2. recovery_target_time, 指XLOG中记录的recordXtime(xl_xact_commit_compact->xact_time), 配合recovery_target_inclusive使用,

如果在同一个时间点有多个事务回滚或提交, 那么recovery_target_inclusive=false则恢复到这个时间点第一个回滚或提交的事务(含), recovery_target_inclusive=true则恢复到这个时间点最后一个回滚或提交的事务(含).

如果时间点上刚好只有1个事务回滚或提交, 那么recovery_target_inclusive=true和false一样, 恢复将处理到这个事务包含的xlog信息(含).

如果时间点没有匹配的事务提交或回滚信息, 那么recovery_target_inclusive=true和false一样, 恢复将处理到这个时间后的下一个事务回滚或提交的xlog信息(含).

3. recovery_target_xid, 指XLogRecord->xl_xid, 可以配合recovery_target_inclusive使用, 但是recovery_target_inclusive只影响日志的输出, 并不影响恢复进程截至点的选择, 截至都截止于这个xid的xlog位置. 也就是说无论如何都包含了这个事务的xlog信息的recovery.

这里需要特别注意xid的信息体现在结束时, 而不是分配xid时. 所以恢复到xid=100提交|回滚点, 可能xid=102已经先提交了. 那么包含xid=102的xlog信息会被recovery.

4. recovery_target_inclusive

还原点的详细介绍可参考 :

http://blog.163.com/digoal@126/blog/static/163877040201303082942271/

在还原前, 对数据库做一些DML操作, 配合这些DML操作来讲解一下这几个参数的含义.

begin1;query1;commit1; begin2; query2_1; pg_create_restore_point(text); query2_2; commit2;

按照以上描述, 使用recovery_target_name恢复数据库恢复后应该包含query1的变更, 但是不包含query2_1和query2_2的变更.

begin1;query1;commit1; -- SESSION A :

digoal=# create table pitr_test(id int, info text);
CREATE TABLE
digoal=# insert into pitr_test values (1,'test');
INSERT 0 1
begin2; query2_1; -- SESSION A :
digoal=# begin;
BEGIN
digoal=# insert into pitr_test values (2,'test');
INSERT 0 1

pg_create_restore_point(text); -- SESSION B :

digoal=# select pg_create_restore_point('pitr_test');
 pg_create_restore_point
-------------------------
 F/B01F8C8
(1 row)

query2_2; commit2; -- SESSION B :

digoal=# insert into pitr_test values (3,'test');
INSERT 0 1
digoal=# commit;
COMMIT

切换日志, 归档 :

digoal=# checkpoint;
CHECKPOINT
digoal=# select pg_xlogfile_name(pg_switch_xlog());
     pg_xlogfile_name
--------------------------
 000000030000000F0000000C
(1 row)
digoal=# checkpoint;
CHECKPOINT
digoal=# select pg_xlogfile_name(pg_switch_xlog());
     pg_xlogfile_name
--------------------------
 000000030000000F0000000D
(1 row)

使用拷贝的备份以及归档日志还原 :

关闭数据库

pg93@db-172-16-3-33-> pg_ctl stop -m fast
waiting for server to shut down.... done
server stopped

pg93@db-172-16-3-33-> cd $PGDATA
pg93@db-172-16-3-33-> cd pg_tblspc/
pg93@db-172-16-3-33-> ll
total 0
lrwxrwxrwx 1 pg93 pg93 47 May 26 18:04 26425 -> /pgdata/digoal/1921/data03/pg93/1999/tbs_digoal
pg93@db-172-16-3-33-> rm -rf /pgdata/digoal/1921/data03/pg93/1999/tbs_digoal/*
pg93@db-172-16-3-33-> cd $PGDATA
pg93@db-172-16-3-33-> rm -rf *

还原备份以及表空间软链接

pg93@db-172-16-3-33-> cp -r /pgdata/digoal/1921/data04/pg93backup/20130527_onlinebackup/* $PGDATA/
pg93@db-172-16-3-33-> cd $PGDATA
pg93@db-172-16-3-33-> cd pg_tblspc/
pg93@db-172-16-3-33-> ll
total 4.0K
drwx------ 3 pg93 pg93 4.0K May 27 14:32 26425
pg93@db-172-16-3-33-> mv 26425/* /pgdata/digoal/1921/data03/pg93/1999/tbs_digoal/
pg93@db-172-16-3-33-> rm -rf 26425
pg93@db-172-16-3-33-> ln -s /pgdata/digoal/1921/data03/pg93/1999/tbs_digoal ./26425
pg93@db-172-16-3-33-> ll
total 0
lrwxrwxrwx 1 pg93 pg93 47 May 27 14:34 26425 -> /pgdata/digoal/1921/data03/pg93/1999/tbs_digoal

创建pg_log目录, 存放日志 :

pg93@db-172-16-3-33-> cd $PGDATA
pg93@db-172-16-3-33-> mkdir -p pg_log

配置$PGDATA/recovery.conf

pg93@db-172-16-3-33-> cd $PGDATA
pg93@db-172-16-3-33-> cp $PGHOME/share/recovery.conf.sample ./recovery.conf
vi recovery.conf
restore_command = 'recovery.sh /pgdata/digoal/1921/data04/pg93archdir %f %p'            # e.g. 'cp /mnt/server/archivedir/%f %p'
recovery_target_name = 'pitr_test'      # e.g. 'daily backup 2011-01-26'
recovery_target_timeline = 'latest'

创建恢复脚本 :

pg93@db-172-16-3-33-> cat recovery.sh
#!/bin/bash

export LANG=en_US.utf8
export PGHOME=/opt/pgsql9.3
export LD_LIBRARY_PATH=$PGHOME/lib:/lib64:/usr/lib64:/usr/local/lib64:/lib:/usr/lib:/usr/local/lib:$LD_LIBRARY_PATH
export DATE=`date +"%Y%m%d"`
export PATH=$PGHOME/bin:$PATH:.

BASEDIR=$1

find $BASEDIR -name $2 -exec cp {} $3 \;

添加可执行权限 :

pg93@db-172-16-3-33-> chmod 700 recovery.sh

启动数据库

pg93@db-172-16-3-33-> pg_ctl start
server starting
pg93@db-172-16-3-33-> LOG:  00000: loaded library "pg_stat_statements"
LOCATION:  load_libraries, miscinit.c:1296

查看还原点是否与预计匹配 :

pg93@db-172-16-3-33-> psql
psql (9.3devel)
Type "help" for help.
digoal=# select ctid,* from pitr_test ;
 ctid  | id | info
-------+----+------
 (0,1) |  1 | test
(1 row)
digoal=# insert into pitr_test values(2,'new');
INSERT 0 1
digoal=# select ctid,* from pitr_test ;
 ctid  | id | info
-------+----+------
 (0,1) |  1 | test
 (0,3) |  2 | new
(2 rows)

新插入的数据ctid=3, 说明query2_1的xlog信息被恢复了, 但是回滚了. 如果没有执行info=new的ctid应该=2.

2.

commit_xact1; time1; rollback_xact2; time2; commit_xact3; commit_xact4;

按照以上描述, 使用recovery_target_time恢复,

如果recovery_target_time=time1, 不管recovery_target_inclusive=true|false, 恢复后应该都只能看到commit_xact1;的数据变更. 因为rollback_xact2回滚了, 不过使用ctid也能看出回滚的影响, 因为rollback_xact2;的xlog是需要处理的.

如果recovery_target_time=time2, 不管recovery_target_inclusive=true|false, 恢复后应该都只能看到commit_xact3;的数据变更. 但是绝对看不到commit_xact4的变更, 即使ctid也不会有影响, 因为commit_xact4;的xlog不会被处理.

commit_xact1; -- SESSION A :

digoal=# truncate pitr_test ;
TRUNCATE TABLE
digoal=# insert into pitr_test values (1,'test');
INSERT 0 1

time1; -- SESSION A :

digoal=# select now();
              now
-------------------------------
 2013-05-27 15:39:57.033134+08
(1 row)

rollback_xact2; -- SESSION A :

digoal=# begin;
BEGIN
digoal=# insert into pitr_test values (2,'rollback');
INSERT 0 1
digoal=# insert into pitr_test values (3,'rollback');
INSERT 0 1
digoal=# rollback;
ROLLBACK

time2; -- SESSION A :

digoal=# select now();
              now
-------------------------------
 2013-05-27 15:40:24.055161+08
(1 row)

commit_xact3; -- SESSION A :

digoal=# insert into pitr_test values (4,'new');
INSERT 0 1

commit_xact4; -- SESSION A :

digoal=# insert into pitr_test values (5,'new');
INSERT 0 1

切换日志, 归档 :

digoal=# checkpoint;
CHECKPOINT
digoal=# select pg_xlogfile_name(pg_switch_xlog());
     pg_xlogfile_name
--------------------------
 000000010000000000000003
(1 row)
digoal=# checkpoint;
CHECKPOINT
digoal=# select pg_xlogfile_name(pg_switch_xlog());
     pg_xlogfile_name
--------------------------
 000000010000000000000004
(1 row)

使用拷贝的备份以及归档日志还原 :

关闭数据库

pg93@db-172-16-3-33-> pg_ctl stop -m fast
waiting for server to shut down.... done
server stopped

删除数据库目录以及表空间目录(为简化例子, 本例无新建的表空间.)

pg93@db-172-16-3-33-> cd $PGDATA
pg93@db-172-16-3-33-> rm -rf *

还原备份以及表空间软链接

pg93@db-172-16-3-33-> cp -r /pgdata/digoal/1921/data04/pg93backup/20130527_onlinebackup/* $PGDATA

创建pg_log目录, 存放日志 :

pg93@db-172-16-3-33-> cd $PGDATA

配置$PGDATA/recovery.conf

pg93@db-172-16-3-33-> cd $PGDATA
pg93@db-172-16-3-33-> cp $PGHOME/share/recovery.conf.sample ./recovery.conf
vi recovery.conf
restore_command = 'recovery.sh /pgdata/digoal/1921/data04/pg93archdir %f %p'            # e.g. 'cp /mnt/server/archivedir/%f %p'
recovery_target_time = '2013-05-27 15:39:57.033134+08'
recovery_target_inclusive = false
recovery_target_timeline = 'latest'

创建恢复脚本 :

pg93@db-172-16-3-33-> cat recovery.sh
#!/bin/bash

export LANG=en_US.utf8
export PGHOME=/opt/pgsql9.3
export LD_LIBRARY_PATH=$PGHOME/lib:/lib64:/usr/lib64:/usr/local/lib64:/lib:/usr/lib:/usr/local/lib:$LD_LIBRARY_PATH
export DATE=`date +"%Y%m%d"`
export PATH=$PGHOME/bin:$PATH:.

BASEDIR=$1

find $BASEDIR -name $2 -exec cp {} $3 \;

添加可执行权限 :

pg93@db-172-16-3-33-> chmod 700 recovery.sh

启动数据库

pg93@db-172-16-3-33-> pg_ctl start
server starting
pg93@db-172-16-3-33-> LOG:  00000: loaded library "pg_stat_statements"
LOCATION:  load_libraries, miscinit.c:1296

查看还原点是否与预计匹配 :

pg93@db-172-16-3-33-> psql
psql (9.3devel)
Type "help" for help.
digoal=# select ctid,* from pitr_test ;
 ctid  | id | info
-------+----+------
 (0,1) |  1 | test
(1 row)
digoal=# insert into pitr_test values(10,'new');
INSERT 0 1
digoal=# select ctid,* from pitr_test ;
 ctid  | id | info
-------+----+------
 (0,1) |  1 | test
 (0,4) | 10 | new
(2 rows)

新插入的数据ctid=4, 说明rollback_xact2;的xlog信息被恢复了, 但是回滚了. 如果没有执行info=new的ctid应该=2.

其他几种搭配就不测试了, 原理已经介绍清楚了. 有兴趣的朋友可以自行测试.

commit_xact1;begin2;query2_1;commit_xact3;commit_xact4;query2_2;rollback2;commit_xact5;

recovery_target_xid=commit_xact1, 无论recovery_target_inclusive=true|false, 恢复后数据库将包含commit_xact1的数据变更;

recovery_target_xid=rollback2; 则包含xact1,xact3,xact4的变更以及xact2的回滚;

recovery_target_xid=xact5; 则包含所有xlog信息.

commit_xact1; -- SESSION A :

digoal=# truncate pitr_test ;
TRUNCATE TABLE
digoal=# begin;
BEGIN
digoal=# select txid_current();
 txid_current
--------------
         1691
(1 row)
digoal=# insert into pitr_test values (1,'test');
INSERT 0 1
digoal=# end;
COMMIT

begin2;query2_1; -- SESSION A :

digoal=# begin;
BEGIN
digoal=# select txid_current();
 txid_current
--------------
         1692
(1 row)
digoal=# insert into pitr_test values (1692,'test');
INSERT 0 1

commit_xact3;commit_xact4; -- SESSION B :

digoal=# begin;
BEGIN
digoal=# select txid_current();
 txid_current
--------------
         1693
(1 row)
digoal=# insert into pitr_test values (1693,'teset');
INSERT 0 1
digoal=# end;
COMMIT
digoal=# begin;
BEGIN
digoal=# select txid_current();
 txid_current
--------------
         1694
(1 row)
digoal=# insert into pitr_test values (1693,'teset');
INSERT 0 1
digoal=# insert into pitr_test values (1694,'teset');
INSERT 0 1
digoal=# end;
COMMIT

query2_2;rollback2; -- SESSION A :

digoal=# insert into pitr_test values (1692,'new_test');
INSERT 0 1
digoal=# rollback;
ROLLBACK

commit_xact5; -- SESSION A :

digoal=# begin;
BEGIN
digoal=# select txid_current();
 txid_current
--------------
         1695
(1 row)
digoal=# insert into pitr_test values (1695,'test');
INSERT 0 1
digoal=# commit;
COMMIT

切换日志, 归档 :

digoal=# checkpoint;
CHECKPOINT
digoal=# select pg_xlogfile_name(pg_switch_xlog());
     pg_xlogfile_name
--------------------------
 000000020000000000000003
(1 row)
digoal=# checkpoint;
CHECKPOINT
digoal=# select pg_xlogfile_name(pg_switch_xlog());
     pg_xlogfile_name
--------------------------
 000000020000000000000004
(1 row)

使用拷贝的备份以及归档日志还原 :

关闭数据库

pg93@db-172-16-3-33-> pg_ctl stop -m fast
waiting for server to shut down.... done
server stopped

删除数据库目录以及表空间目录(为简化例子, 本例无新建的表空间.)

pg93@db-172-16-3-33-> cd $PGDATA
pg93@db-172-16-3-33-> rm -rf *

还原备份以及表空间软链接

pg93@db-172-16-3-33-> cp -r /pgdata/digoal/1921/data04/pg93backup/20130527_onlinebackup/* $PGDATA

创建pg_log目录, 存放日志 :

pg93@db-172-16-3-33-> cd $PGDATA

配置$PGDATA/recovery.conf

pg93@db-172-16-3-33-> cd $PGDATA
pg93@db-172-16-3-33-> cp $PGHOME/share/recovery.conf.sample ./recovery.conf
vi recovery.conf
restore_command = 'recovery.sh /pgdata/digoal/1921/data04/pg93archdir %f %p'            # e.g. 'cp /mnt/server/archivedir/%f %p'
recovery_target_xid = '1692'
recovery_target_inclusive = false
recovery_target_timeline = 'latest'

创建恢复脚本 :

pg93@db-172-16-3-33-> cat recovery.sh
#!/bin/bash

export LANG=en_US.utf8
export PGHOME=/opt/pgsql9.3
export LD_LIBRARY_PATH=$PGHOME/lib:/lib64:/usr/lib64:/usr/local/lib64:/lib:/usr/lib:/usr/local/lib:$LD_LIBRARY_PATH
export DATE=`date +"%Y%m%d"`
export PATH=$PGHOME/bin:$PATH:.

BASEDIR=$1

find $BASEDIR -name $2 -exec cp {} $3 \;

添加可执行权限 :

pg93@db-172-16-3-33-> chmod 700 recovery.sh

启动数据库

pg93@db-172-16-3-33-> pg_ctl start
server starting
pg93@db-172-16-3-33-> LOG:  00000: loaded library "pg_stat_statements"
LOCATION:  load_libraries, miscinit.c:1296

查看还原点是否与预计匹配 :

pg93@db-172-16-3-33-> psql
psql (9.3devel)
Type "help" for help.
digoal=# select ctid,* from pitr_test ;
 ctid  |  id  | info
-------+------+-------
 (0,1) |    1 | test
 (0,3) | 1693 | teset
 (0,4) | 1693 | teset
 (0,5) | 1694 | teset
(4 rows)
digoal=# insert into pitr_test values(10,'new');
INSERT 0 1
digoal=# select ctid,* from pitr_test ;
 ctid  |  id  | info
-------+------+-------
 (0,1) |    1 | test
 (0,3) | 1693 | teset
 (0,4) | 1693 | teset
 (0,5) | 1694 | teset
 (0,7) |   10 | new
(5 rows)

因为只有当record_info等于XLOG_XACT_ABORT,XLOG_XACT_COMMIT或者XLOG_XACT_COMMIT_COMPACT, 或者自建还原点时才可以被作为截至点,

所以本例使用xid=1692作为截至点, xlog会应用到query2_2;rollback2; -- SESSION A : 这个位置(并且包含这个xlog信息).

因此在此之前应该插入了6条记录, 所以新插入的数据ctid=7, 再次说明query2_2;rollback2;的xlog信息被恢复了, 但是回滚了.

其他几种搭配就不测试了, 原理已经介绍清楚了. 有兴趣的朋友可以自行测试.

自动备份

这个可以结合操作系统的定时脚本来实现, 在Linux中可以使用crontab.

例如 :

1. 首先规划好存放数据库数据文件的目录.

mkdir -p /pgdata/digoal/1921/data04/pg93backup
chown -R pg93:pg93 /pgdata/digoal/1921/data04/pg93backup
chmod 700 /pgdata/digoal/1921/data04/pg93backup

2. 编写备份脚本

pg93@db-172-16-3-33-> cd $PGDATA
pg93@db-172-16-3-33-> vi backup.sh
#!/bin/bash

export LANG=en_US.utf8
export PGHOME=/opt/pgsql9.3
export LD_LIBRARY_PATH=$PGHOME/lib:/lib64:/usr/lib64:/usr/local/lib64:/lib:/usr/lib:/usr/local/lib:$LD_LIBRARY_PATH
export DATE=`date +"%Y%m%d"`
export PATH=$PGHOME/bin:$PATH:.
export PGDATA=/pgdata1999

BASEDIR="/pgdata/digoal/1921/data04/pg93backup"

date +%F%T

if [ ! -d $BASEDIR/$DATE ]; then
  mkdir -p $BASEDIR/$DATE
  if [ $? -eq 0 ]; then
    psql -h 127.0.0.1 -p 1999 -U postgres postgres -c "select pg_start_backup(now()::text)"
    if [ $? -eq 0 ]; then
      cp -r -L $PGDATA $BASEDIR/$DATE
    else
      echo -e "select pg_start_backup(now()::text) error"
      exit 1
    fi
    psql -h 127.0.0.1 -p 1999 -U postgres postgres -c "select pg_stop_backup()"
    date +%F%T
    echo -e "backup successed"
    exit 0
  else
    echo -e "mkdir -p $BASEDIR/$DATE error"
    exit 1
  fi
else
  echo -e "$DATE backuped, don't backup repeated"
  exit 1
fi

修改权限

pg93@db-172-16-3-33-> chmod 700 backup.sh

3. 测试

pg93@db-172-16-3-33-> ./backup.sh
 pg_start_backup
-----------------
 0/6000028
(1 row)

NOTICE:  pg_stop_backup complete, all required WAL segments have been archived
 pg_stop_backup
----------------
 0/60000F0
(1 row)
pg93@db-172-16-3-33-> 再次执行则退出, 因为今天已经备份了.
pg93@db-172-16-3-33-> ./backup.sh
20130527 backuped, don't backup repeated

4. 配置crontab

首先删除当天备份

rm -rf /pgdata/digoal/1921/data04/pg93backup/20130527

配置自动备份

su - pg93
crontab -e
40 16 * * * /pgdata1999/backup.sh >>/tmp/backup.log 2>&1

5. 检查是否自动备份.

pg93@db-172-16-3-33-> cat /tmp/backup.log
2013-05-2716:40:01
 pg_start_backup
-----------------
 0/8000028
(1 row)

NOTICE:  pg_stop_backup complete, all required WAL segments have been archived
 pg_stop_backup
----------------
 0/80000F0
(1 row)

2013-05-2716:40:02
backup successed

检查备份成功后把crontab调整为你需要的备份时间即可. 例如每天都凌晨2点开始备份.

0 2 * * * /pgdata1999/backup.sh >>/tmp/backup.log 2>&1

[参考]

1. http://blog.163.com/digoal@126/blog/static/1638770402012914112949546/

2. http://blog.163.com/digoal@126/blog/static/163877040201303082942271/

时间： 2024-09-20 09:34:46

PostgreSQL backup and recovery - online backup & Point-In-Time-Recovery

PostgreSQL backup and recovery - online backup & Point-In-Time-Recovery的相关文章

PostgreSQL backup and recovery - cold backup & recovery

PostgreSQL 9.3 Have pg_basebackup --write-recovery-conf output a minimal recovery.conf

小米3怎么Recovery模式？小米3进入Recovery图文教程

76.3. Bacula, the Open Source, Enterprise ready, Network Backup Tool for Linux, Unix, Mac and Windows.

【oracle】ORA-01580 error creating control backup file

RMAN backup validate database on databases in noarchivelog mode

Symantec Backup Exec 2012 Agent For Linux安装

[20160329]backup as copy问题.txt

rman backup遇到ORA-19811