重庆分公司,新征程启航
为企业提供网站建设、域名注册、服务器等服务
本篇内容介绍了“PostgreSQL HA环境分析”的有关知识,在实际案例的操作过程中,不少人都会遇到这样的困境,接下来就让小编带领大家学习一下如何处理这些情况吧!希望大家仔细阅读,能够学有所成!
我们提供的服务有:成都网站设计、成都网站制作、微信公众号开发、网站优化、网站认证、乌苏ssl等。为上千企事业单位解决了网站和推广的问题。提供周到的售前咨询和贴心的售后服务,是有科学管理、有技术的乌苏网站制作公司
基于streaming replication搭建的PostgreSQL HA环境,Old Standby节点升级为New Master节点后,时间线会切换为新的时间线,比如从n变为n + 1.而Old Master节点的时间线仍然为原来的时间线,比如仍为n,通过使用pg_rewind工具,可使原来其实”完好”的Old Master成为New Standby节点.
图解
如下图所示:
在执行主备切换后,New Master节点的时间线切换为n + 1,通过pg_rewind可使Old Master在分叉点开始与New Master同步,成为New Standby节点.
实测
New Master
切换后为New Master后,执行以下SQL:
testdb=# create table t_new(id int,flag varchar(40)); CREATE TABLE testdb=# insert into t_new select c,'flag'||c from generate_series(1,1000000) as c; INSERT 0 1000000 testdb=#
Old Master
执行pg_rewind前,重启主库,执行以下SQL
testdb=# create table t_fork(id int,flag varchar(40)); CREATE TABLE testdb=# insert into t_fork select c,'flag'||c from generate_series(1,1000000) as c; INSERT 0 1000000 testdb=#
执行pg_rewind
[xdb@localhost testdb]$ cp /data/archivelog/* ./pg_wal [xdb@localhost testdb]$ pg_rewind --target-pgdata=$PGDATA --source-server="host=192.168.26.25 port=5432 dbname=testdb" --progress connected to server servers diverged at WAL location 0/B41F12B8 on timeline 23 rewinding from last common checkpoint at 0/AFCF99E0 on timeline 23 reading source file list reading target file list reading WAL in target need to copy 360 MB (total source directory size is 501 MB) 369312/369312 kB (100%) copied creating backup label and updating control file syncing target data directory Done!
配置recovery.conf文件
[xdb@localhost testdb]$ mv recovery.done recovery.conf [xdb@localhost testdb]$ vim recovery.conf [xdb@localhost testdb]$ cat recovery.conf standby_mode = 'on' primary_conninfo = 'user=replicator password=replicator host=192.168.26.26 port=5432 sslmode=prefer sslcompression=0 krbsrvname=postgres target_session_attrs=any' restore_command = 'cp /data/archivelog/%f %p'
重启数据库
[xdb@localhost testdb]$ pg_ctl start waiting for server to start....2019-03-28 12:39:31.918 CST [1961] LOG: listening on IPv4 address "0.0.0.0", port 5432 2019-03-28 12:39:31.918 CST [1961] LOG: listening on IPv6 address "::", port 5432 2019-03-28 12:39:31.920 CST [1961] LOG: listening on Unix socket "/tmp/.s.PGSQL.5432" 2019-03-28 12:39:31.970 CST [1961] LOG: redirecting log output to logging collector process 2019-03-28 12:39:31.970 CST [1961] HINT: Future log output will appear in directory "pg_log". ... done server started
与New Master数据同步了,而在原TL上的t_fork数据表消失了.
testdb=# select count(*) from t_new; count --------- 1000000 (1 row) testdb=# select count(*) from t_old; count --------- 1000000 (1 row) testdb=# select count(*) from t_fork; ERROR: relation "t_fork" does not exist LINE 1: select count(*) from t_fork; ^ testdb=#
“PostgreSQL HA环境分析”的内容就介绍到这里了,感谢大家的阅读。如果想了解更多行业相关的知识可以关注创新互联网站,小编将为大家输出更多高质量的实用文章!