postgres--流复制

配置

master  192.168.2.21    5432
slave   192.168.2.22    5432
目录      /var/lib/pgsql/10/data/

配置主库

配置用户

psql
create role rep login replication encrypted password '1qaz2wsx';

配置访问权限 pg_hba.conf

在/var/lib/pgsql/10/data/pg_hba.conf增加

host    replication rep     192.168.2.22/32     md5

配置同步文件

cd /var/lib/pgsql/10/data/

vi recovery.done

recovery_target_timeline = 'latest'
standby_mode = on
primary_conninfo = 'host=192.168.2.21 port=5432 user=rep password=1qaz2wsx'
# recovery_min_apply_delay = 0   #延迟多少分钟应用，用户可以配置延迟的备库，例如给一点误操作的缓冲时间。在备库不会这么早被应用。

启动主机

配置从库

创建目录

安装从机之后，不要初始化和启动数据库

mkdir /var/lib/pgsql/10/data/
chown postgres:postgres /var/lib/pgsql/10/data/
chmod 700 /var/lib/pgsql/10/data/

同步主库数据

su - postgres
pg_basebackup -D /var/lib/pgsql/10/data/ -F p -X stream -h 192.168.2.21 -p 5432 -U rep

修改配置文件

根据从机配置适当修改/var/lib/pgsql/10/data/postgresql.conf

配置recovery.conf

cd /var/lib/pgsql/10/data/
mv recovery.done recovery.conf

启动从库

service postgresql-10 restart

流复制节点的状态监控

主库

su - postgres
psql
postgres=# \x
    Expanded display is on.  

postgres=# select pg_size_pretty(pg_wal_lsn_diff(pg_current_wal_lsn(), sent_lsn)) as sent_delay,
pg_size_pretty(pg_wal_lsn_diff(pg_current_wal_lsn(), write_lsn)) as write_delay,
pg_size_pretty(pg_wal_lsn_diff(pg_current_wal_lsn(), flush_lsn)) as flush_delay,
pg_size_pretty(pg_wal_lsn_diff(pg_current_wal_lsn(), replay_lsn)) as replay_delay,
*
from pg_stat_replication;  

    -[ RECORD 1 ]----+------------------------------
    sent_delay       | 0 bytes
    write_delay      | 0 bytes
    flush_delay      | 0 bytes
    replay_delay     | 0 bytes
    pid              | 25628
    usesysid         | 16399
    usename          | rep
    application_name | walreceiver
    client_addr      | 192.168.2.22
    client_hostname  |
    client_port      | 59674
    backend_start    | 2018-01-03 17:57:00.690789+08
    backend_xmin     |
    state            | streaming
    sent_lsn         | 0/F787EA30
    write_lsn        | 0/F787EA30
    flush_lsn        | 0/F787EA30
    replay_lsn       | 0/F787EA30
    write_lag        | 00:00:00.000259
    flush_lag        | 00:00:00.013562
    replay_lag       | 00:00:00.013563
    sync_priority    | 0
    sync_state       | async

从库

进入数据库

su - postgres
psql

查看当前WAL应用是否暂停

postgres=# select pg_is_wal_replay_paused();
    -[ RECORD 1 ]-----------+--
    pg_is_wal_replay_paused | f

f表示没有暂停，t表示暂停

查看WAL接收LSN点

postgres=# select pg_last_wal_receive_lsn();
    -[ RECORD 1 ]-----------+-----------
    pg_last_wal_receive_lsn | 0/FFA88FB8

查看WAL应用LSN点

postgres=# select pg_last_wal_replay_lsn();
    -[ RECORD 1 ]----------+---------
    pg_last_wal_replay_lsn | 1/1EC488

查看wal receiver的统计信息

postgres=# \x
    Expanded display is on.  

postgres=# select * from pg_stat_get_wal_receiver();
    -[ RECORD 1 ]---------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
    pid                   | 11648
    status                | streaming
    receive_start_lsn     | 0/3A000000
    receive_start_tli     | 1
    received_lsn          | 1/2DE1E28
    received_tli          | 1
    last_msg_send_time    | 2018-01-04 17:20:38.645492+08
    last_msg_receipt_time | 2018-01-04 17:20:38.640939+08
    latest_end_lsn        | 1/2DE1E28
    latest_end_time       | 2018-01-04 17:20:38.645492+08
    slot_name             |
    conninfo              | user=rep password=******** dbname=replication host=192.168.2.21 port=5432 fallback_application_name=walreceiver sslmode=prefer sslcompression=1 krbsrvname=postgres target_session_attrs=any

暂停WAL的应用

select pg_wal_replay_pause();
    -[ RECORD 1 ]-------+-
    pg_wal_replay_pause | 

postgres=# select pg_is_wal_replay_paused();
    -[ RECORD 1 ]-----------+--
    pg_is_wal_replay_paused | t

postgres=# select * from pg_stat_get_wal_receiver();
    -[ RECORD 1 ]---------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
    pid                   | 11648
    status                | streaming
    receive_start_lsn     | 0/3A000000
    receive_start_tli     | 1
    received_lsn          | 1/71D3E88
    received_tli          | 1
    last_msg_send_time    | 2018-01-04 17:23:52.302411+08
    last_msg_receipt_time | 2018-01-04 17:23:52.297611+08
    latest_end_lsn        | 1/71D3E88
    latest_end_time       | 2018-01-04 17:23:52.302411+08
    slot_name             |
    conninfo              | user=rep password=******** dbname=replication host=192.168.2.21 port=5432 fallback_application_name=walreceiver sslmode=prefer sslcompression=1 krbsrvname=postgres target_session_attrs=any

恢复WAL的应用

postgres=# select pg_wal_replay_resume();
    -[ RECORD 1 ]--------+-
    pg_wal_replay_resume |

postgres=# select pg_is_wal_replay_paused();
    -[ RECORD 1 ]-----------+--
    pg_is_wal_replay_paused | f

注意事项

防止主库wal文件被删除，从库没有接收到

方法有：使用slot、配置足够大的wal keep、wal归档。

slot

配置方法

主 postgresql.conf
max_replication_slots = 10   

备 recovery.conf
primary_slot_name = ''

当备库挂了或者不再使用，而忘记删除对应的SLOT时，可能导致主库WAL无限膨胀。

wal keep

配置方法

主 postgresql.conf
wal_keep_segments = 1024

会导致主库的WAL预留足够的个数，占用一定空间（walkeepsegments*16M
= 16G）。

wal归档

配置主库的归档，同时备库需要能获取到已归档的WAL文件。

配置方式

主 postgresql.conf
    archive_mode = on
    archive_command = '......' 

备 recovery.conf
    restore_command = '....'

同步账号密码的安全性

由于recovery.conf是明文，需要保护好recovery.conf文件的访问权限
或者限制只允许特定的从机访问