PostgreSQL10支持内置的逻辑复制(Logical Replication)。
逻辑复制
简单说一下什么是逻辑复制以及有什么好处。
大多数人都知道Streaming Replication已经成为PostgreSQL的一部分,并且通常用于高可用性和读写分离,流复制是基于WAL日志的物理复制,适用于整个数据库实例的复制,并且备库是只读的。
Logical Replication属于逻辑复制,适用于数据库实例的部分(单个数据库或者某些表)的复制,目前只支持表复制。
最大的不同就是可以向下游节点写入数据,也可以将多个数据库实例的数据,同步到一个目标数据库等等。
使用
Logical Replication使用piblish/subcribe概念,在上游节点创建发布者,下游节点创建订阅者。
配置postgresql.conf
wal_level = logical
创建发布者
yangjie=# create table users(id int, name name); CREATE TABLE yangjie=# create publication pub1 for table users; CREATE PUBLICATION
创建发布者pub1,并添加表users.
另一种用法是添加数据库中所有用户表到发布者alltables:
create publication alltables for all tables;
创建订阅者:
yangjie=# create subscription sub1 connection 'host=192.168.102.30 port=5432 dbname=yangjie' publication pub1; NOTICE: created replication slot "sub1" on publisher CREATE SUBSCRIPTION
订阅者sub1将会从发布者pub1复制表users.
这些都需要基础的复制工作,订阅者会拷贝所有数据到表中,创建订阅者时,表不会被复制,我们需要先自己创建这些表,如果没有发现本地表复制将会失败。
当发布者添加新表时,订阅者不能自动的获知,我们需要更新订阅者:
alter subscription sub refresh publication;
这会从新表中拷贝所有存在的数据。
示例:
先创建发布者、订阅者:
# publication yangjie=# create publication mypub for all tables; CREATE PUBLICATION yangjie=# create subscription mysub connection 'host=192.168.102.30 port=5432 dbname=yangjie' publication mypub; CREATE SUBSCRIPTION
上游节点添加一张表并插入数据:
yangjie=# create table a(id int); CREATE TABLE yangjie=# insert into a values (1); INSERT 0 1 yangjie=# select * from a ; id ---- 1 (1 row)
下游节点添加上表并查询:
yangjie# create table a(id int); CREATE TABLE yangjie=# select * from a; id ---- (0 rows)
更新订阅者:
yangjie=# alter subscription mysub refresh publication ; ALTER SUBSCRIPTION yangjie=# select * from a; id ---- 1 (1 row)
监控
现在我们已经配置好了逻辑复制,为了能清楚地知道它如何运行等,提供了两张视图,pg_stat_replication显示当前服务的所有复制连接,pg_stat_subscription下游节点显示订阅者的状态信息。
# publication yangjie=# select * from pg_stat_replication ; -[ RECORD 1 ]----+------------------------------ pid | 5743 usesysid | 10 usename | yangjie application_name | sub1 client_addr | 192.168.102.34 client_hostname | client_port | 34094 backend_start | 2017-11-23 17:22:08.460961+08 backend_xmin | state | streaming sent_lsn | 0/308DCB8 write_lsn | 0/308DCB8 flush_lsn | 0/308DCB8 replay_lsn | 0/308DCB8 write_lag | flush_lag | replay_lag | sync_priority | 0 sync_state | async
# subscription yangjie=# select * from pg_stat_subscription ; -[ RECORD 1 ]---------+------------------------------ subid | 16388 subname | sub1 pid | 10810 relid | received_lsn | 0/308DCB8 last_msg_send_time | 2017-11-23 17:22:08.484458+08 last_msg_receipt_time | 2017-11-23 17:22:08.476762+08 latest_end_lsn | 0/308DCB8 latest_end_time | 2017-11-23 17:22:08.484458+08
进程信息
# publication [yangjie@young-1 ~]$ ps -ef | grep postgres yangjie 4266 31571 0 Nov23 ? 00:00:00 postgres: yangjie yangjie [local] idle yangjie 5743 31571 0 Nov23 ? 00:00:00 postgres: wal sender process yangjie 192.168.102.34(34094) idle yangjie 14395 14347 0 09:22 pts/2 00:00:00 grep --color=auto postgres yangjie 31571 1 0 Nov23 ? 00:00:01 /opt/pgsql/pg101/bin/postgres -D ../data yangjie 31573 31571 0 Nov23 ? 00:00:00 postgres: checkpointer process yangjie 31574 31571 0 Nov23 ? 00:00:00 postgres: writer process yangjie 31575 31571 0 Nov23 ? 00:00:00 postgres: wal writer process yangjie 31576 31571 0 Nov23 ? 00:00:01 postgres: autovacuum launcher process yangjie 31577 31571 0 Nov23 ? 00:00:01 postgres: stats collector process yangjie 31578 31571 0 Nov23 ? 00:00:00 postgres: bgworker: logical replication launcher
# subscription [yangjie@young-2 ~]$ ps -ef | grep postgres yangjie 9222 1 0 Nov23 pts/1 00:00:01 /opt/pgsql/pg101-2/bin/postgres -D ../data yangjie 9224 9222 0 Nov23 ? 00:00:00 postgres: checkpointer process yangjie 9225 9222 0 Nov23 ? 00:00:00 postgres: writer process yangjie 9226 9222 0 Nov23 ? 00:00:00 postgres: wal writer process yangjie 9227 9222 0 Nov23 ? 00:00:01 postgres: autovacuum launcher process yangjie 9228 9222 0 Nov23 ? 00:00:02 postgres: stats collector process yangjie 9229 9222 0 Nov23 ? 00:00:00 postgres: bgworker: logical replication launcher yangjie 9287 9222 0 Nov23 ? 00:00:00 postgres: yangjie yangjie [local] idle yangjie 10810 9222 0 Nov23 ? 00:00:04 postgres: bgworker: logical replication worker for subscription 16388 yangjie 26627 26570 0 09:22 pts/0 00:00:00 grep --color=auto postgres
示例
在上游节点创建发布者:
yangjie=# create table users(id int, name name); CREATE TABLE yangjie=# insert into users values (1, 'Jie Yang'); INSERT 0 1 yangjie=# create publication pub1 for table users; CREATE PUBLICATION
设置订阅者:
yangjie=# create table users (id int, name name); CREATE TABLE yangjie=# create subscription sub1 connection 'host=192.168.102.30 port=5432 dbname=yangjie' publication pub1; NOTICE: created replication slot "sub1" on publisher CREATE SUBSCRIPTION
这里将会将会同步表信息并在上游节点创建一个复制槽sub1:
# subscription yangjie=# select * from users ; id | name ----+---------- 1 | Jie Yang (1 row) yangjie=# select * from pg_replication_slots ; -[ RECORD 1 ]-------+---------- slot_name | sub1 plugin | pgoutput slot_type | logical datoid | 16384 database | yangjie temporary | f active | t active_pid | 15386 xmin | catalog_xmin | 593 restart_lsn | 0/308DF28 confirmed_flush_lsn | 0/308DF60
查看订阅者状态:
# subscription yangjie=# select * from pg_stat_subscription ; -[ RECORD 1 ]---------+------------------------------ subid | 16393 subname | sub1 pid | 27705 relid | received_lsn | 0/308DF60 last_msg_send_time | 2017-11-24 09:29:41.820483+08 last_msg_receipt_time | 2017-11-24 09:29:41.818227+08 latest_end_lsn | 0/308DF60 latest_end_time | 2017-11-24 09:29:41.820483+08
查看复制状态:
# publication yangjie=# select * from pg_stat_replication ; -[ RECORD 1 ]----+------------------------------ pid | 15386 usesysid | 10 usename | yangjie application_name | sub1 client_addr | 192.168.102.34 client_hostname | client_port | 41152 backend_start | 2017-11-24 09:29:41.796288+08 backend_xmin | state | streaming sent_lsn | 0/308DF60 write_lsn | 0/308DF60 flush_lsn | 0/308DF60 replay_lsn | 0/308DF60 write_lag | flush_lag | replay_lag | sync_priority | 0 sync_state | async
注意这里的application_name与我们创建订阅者的名字相同。
在插入一行,看是否复制:
# publication yangjie=# insert into users values (2, 'Joe Yang'); INSERT 0 1
查看订阅者:
# subscription yangjie=# select * from users ; id | name ----+---------- 1 | Jie Yang 2 | Joe Yang (2 rows)
复制标识(replica identity):
为了逻辑复制能在下游节点正确执行UPDATE和DELETE,我们需要定义如何找到唯一行,这就是复制标识,默认情况下,复制标识将是表的主键,如果已经定义了主键,将不需要做任何动作,
yangjie=# update users set name = 'Jee Yang' where id = 2; ERROR: cannot update table "users" because it does not have a replica identity and publishes updates HINT: To enable updating the table, set REPLICA IDENTITY using ALTER TABLE.
配置replica identity:
yangjie=# alter table users add primary key(id); ALTER TABLE # 明确定义复制标识 yangjie=# alter table users replica identity using index users_pkey; ALTER TABLE yangjie=# \d+ users Table "public.users" Column | Type | Collation | Nullable | Default | Storage | Stats target | Description --------+---------+-----------+----------+---------+---------+--------------+------------- id | integer | | not null | | plain | | name | name | | | | plain | | Indexes: "users_pkey" PRIMARY KEY, btree (id) REPLICA IDENTITY Publications: "pub1" yangjie=# update users set name = 'Jee Yang' where id = 2; UPDATE 1
查询:
# subscription yangjie=# alter table users add primary key(id); ALTER TABLE yangjie=# select * from users ; id | name ----+---------- 1 | Jie Yang 2 | Jee Yang (2 rows)
灵活性
订阅者添加额外字段:
# subscription yangjie=# alter table users add age int; ALTER TABLE # publication yangjie=# insert INTO users values (3, 'Joe Yang'); INSERT 0 1 # subscription yangjie=# select * from users ; id | name | age ----+----------+----- 1 | Jie Yang | 2 | Jee Yang | 3 | Joe Yang | (3 rows)
多个数据库实例的数据,同步到一个目标数据库
# publication host=192.168.102.30 port=5431 yangjie=# create table users (id int primary key, name name, age int); CREATE TABLE yangjie=# insert into users values (11, 'Jre Yang', 24); INSERT 0 1 yangjie=# create publication pub1 for table users ; CREATE PUBLICATION # subscription yangjie=# create subscription sub2 connection 'host=localhost port=5431 dbname=yangjie' publication pub1; NOTICE: created replication slot "sub2" on publisher CREATE SUBSCRIPTION yangjie=# select * from users ; id | name | age ----+----------+----- 1 | Jie Yang | 2 | Jee Yang | 3 | Joe Yang | 11 | Jre Yang | 24 (4 rows) # publication 2 yangjie=# update users set age = 23 where id = 11; UPDATE 1 yangjie=# select * from users ; id | name | age ----+----------+----- 11 | Jre Yang | 23 (1 row) # NOTICE:创建 # subscription yangjie=# select * from users ; id | name | age ----+----------+----- 1 | Jie Yang | 2 | Jee Yang | 3 | Joe Yang | 11 | Jre Yang | 23 (4 rows)
同理:也可以将一个数据库实例的不同数据,复制到不同的目标库,或者多个数据库实例之间,共享部分数据等等。
总结
publication - 发布者
- 逻辑复制的前提是将数据库 wal_level 参数设置成 logical;
- 源库上逻辑复制的用户必须具有 replicatoin 或 superuser 角色;
- 逻辑复制目前仅支持数据库表逻辑复制,其它对象例如函数、视图不支持;
- 逻辑复制支持DML(UPDATE、INSERT、DELETE)操作,TRUNCATE 和 DDL 操作不支持;
- 需要发布逻辑复制的表,须配置表的 REPLICA IDENTITY 特性;
- 一个数据库中可以有多个publication,通过 pg_publication 查看;
- 允许一次发布所有表,语法: CREATE PUBLICATION alltables FOR ALL TABLES;
subscription - 订阅者
- 订阅节点需要指定发布者的连接信息;
- 一个数据库中可以有多个订阅者;
- 可以使用enable/disable启用/暂停该订阅;
- 发布节点和订阅节点表的模式名、表名必须一致,订阅节点允许表有额外字段;
- 发布节点增加表名,订阅节点需要执行: ALTER SUBSCRIPTION sub1 REFRESH PUBLICATION
参考:
Logical Replication in PostgreSQL 10