半个PostgreSQL DBA,热衷于数据库相关的技术。我的ppt分享https://pan.baidu.com/s/1eRQsdAa https://github.com/chenhuajun https://chenhuajun.github.io
分类: Mysql/postgreSQL
2015-07-17 01:42:36
MySQL的复制存在一个回放延迟的问题,即备机回放的速度赶不上主机SQL执行的速度,导致延迟越来越厉害。原因在于MySQL的binlog回放是单线程,主机的SQL执行是多线程,单线程的回放速度跟不上。这个问题,BAT的工程师们在公开场合讲过N次了,印象很深刻。而对于这个问题,各路高手也对MySQL做了各种优化。比如备机回放也使用多线程,但仍然治标不治本,因为多线程仍然只能是多个datebase每个database一个线程,感觉没啥子用啊。还有一种方法就是让主机慢下来,等等备机,但这样主机的性能就会被备机生生给拖住,动用这样的方法也是不得已啊。
PostgreSQL流复制的备机也是单线程(进程)的,那么有没有类似的问题呢?
实际上应该不会,因为PostgreSQL的流复制是基于块的物理复制,回放的代价比执行SQL小得多,一般不会出现备机的回放速度跟不上主机的问题。下面看看实际的测试结果。
主机和备机都是以下配置的虚拟机
CPU:2 core
MEM: 2G
OS:CentOS release 6.5 (Final)
PostgreSQL:9.4.2(shared_buffers = 512MB)
chenhj=# create table tb1(id serial,c1 int); CREATE TABLE [chenhj@node1 ~]$ cat test.sql insert into tb1(c1) values(1);
在主机上执行pgbench。
[chenhj@node1 ~]$ pgbench -n -f test.sql -T 50 -c 20 transaction type: Custom query scaling factor: 1 query mode: simple number of clients: 20 number of threads: 1 duration: 50 s number of transactions actually processed: 95324 latency average: 10.491 ms tps = 1905.733296 (including connections establishing) tps = 1910.579339 (excluding connections establishing)
pgbench执行过程中,在备机不断执行下面的语句,发现偶尔会看到100个字节左右的回放延迟,但延迟没有积累,即没有出现备机追不上主机的情况。
SELECT pg_last_xlog_receive_location() - pg_last_xlog_replay_location();
根据下面的top资源占用情况可知,主机上WAL发送进程(18374)占用CPU最高,为21.6%。备机上WAL接受进程(8181)最高为17.3%,其次是Startup进程,也即是恢复进程(5012),占用CPU 8.0%。这说明即使由于主机的处理能力提升导致复制遇到了瓶颈,也是主机上的WAL发送进程先遇到瓶颈,而后是备机上的WAL接受进程,最后才轮到备机上的恢复进程。
主机
[root@node1 ~]# top top - 12:37:28 up 3:18, 3 users, load average: 0.02, 0.01, 0.01 Tasks: 156 total, 5 running, 151 sleeping, 0 stopped, 0 zombie Cpu(s): 24.9%us, 43.9%sy, 0.0%ni, 21.4%id, 5.4%wa, 3.1%hi, 1.3%si, 0.0%st Mem: 1922268k total, 1718824k used, 203444k free, 10072k buffers Swap: 2064376k total, 394032k used, 1670344k free, 463376k cached PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND 18374 chenhj 20 0 663m 2280 1432 R 21.6 0.1 0:10.99 postgres 18965 chenhj 20 0 13848 1176 812 R 10.0 0.1 0:00.74 pgbench 18967 chenhj 20 0 663m 5688 4688 S 8.3 0.3 0:00.59 postgres 18969 chenhj 20 0 663m 5660 4660 S 7.3 0.3 0:00.50 postgres 18968 chenhj 20 0 663m 5660 4660 S 7.0 0.3 0:00.50 postgres 18970 chenhj 20 0 663m 5664 4664 S 7.0 0.3 0:00.47 postgres 18971 chenhj 20 0 663m 5676 4676 S 6.6 0.3 0:00.47 postgres 18972 chenhj 20 0 663m 5648 4648 S 6.3 0.3 0:00.43 postgres 18975 chenhj 20 0 663m 5628 4636 S 6.3 0.3 0:00.37 postgres 18973 chenhj 20 0 663m 5636 4636 S 6.0 0.3 0:00.39 postgres 18982 chenhj 20 0 663m 5632 4636 S 6.0 0.3 0:00.38 postgres 18984 chenhj 20 0 663m 5652 4656 S 6.0 0.3 0:00.38 postgres 18974 chenhj 20 0 663m 5648 4648 R 5.6 0.3 0:00.40 postgres 18976 chenhj 20 0 663m 5632 4640 S 5.6 0.3 0:00.38 postgres 18978 chenhj 20 0 663m 5656 4664 S 5.6 0.3 0:00.37 postgres 18979 chenhj 20 0 663m 5644 4648 S 5.6 0.3 0:00.37 postgres 18983 chenhj 20 0 663m 5636 4640 S 5.6 0.3 0:00.37 postgres 18986 chenhj 20 0 663m 5624 4628 S 5.6 0.3 0:00.38 postgres 18977 chenhj 20 0 663m 5656 4664 S 5.3 0.3 0:00.36 postgres 18980 chenhj 20 0 663m 5636 4640 S 5.3 0.3 0:00.35 postgres 18981 chenhj 20 0 663m 5636 4640 S 5.3 0.3 0:00.36 postgres 18985 chenhj 20 0 663m 5656 4660 S 5.0 0.3 0:00.34 postgres [root@node1 ~]# ps -ef|grep postgres chenhj 18367 1 0 12:31 pts/1 00:00:00 /usr/local/pgsql/bin/postgres -D data chenhj 18369 18367 0 12:31 ? 00:00:00 postgres: checkpointer process chenhj 18370 18367 0 12:31 ? 00:00:00 postgres: writer process chenhj 18371 18367 0 12:31 ? 00:00:00 postgres: wal writer process chenhj 18372 18367 0 12:31 ? 00:00:00 postgres: autovacuum launcher process chenhj 18373 18367 0 12:31 ? 00:00:00 postgres: stats collector process chenhj 18374 18367 0 12:31 ? 00:00:00 postgres: wal sender process chenhj 192.168.150.201(34998) streaming 0/A5734050
备机
[root@node1 ~]# top top - 12:37:36 up 3:11, 4 users, load average: 0.19, 0.06, 0.02 Tasks: 138 total, 1 running, 137 sleeping, 0 stopped, 0 zombie Cpu(s): 0.7%us, 8.6%sy, 0.0%ni, 58.4%id, 32.0%wa, 0.2%hi, 0.2%si, 0.0%st Mem: 1922268k total, 1851180k used, 71088k free, 4816k buffers Swap: 2064376k total, 660616k used, 1403760k free, 459252k cached PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND 8181 chenhj 20 0 667m 2148 1332 D 17.3 0.1 0:15.65 postgres 5012 chenhj 20 0 662m 172m 171m S 8.0 9.2 0:30.94 postgres 23 root 20 0 0 0 0 S 1.0 0.0 0:06.64 kblockd/1 11 root 20 0 0 0 0 S 0.3 0.0 0:08.91 events/0 12 root 20 0 0 0 0 S 0.3 0.0 0:10.60 events/1 1263 root -2 0 2574m 1.2g 1540 S 0.3 65.5 0:52.81 corosync 2195 root -2 0 4900 452 420 S 0.3 0.0 0:00.56 iscsid 8188 root 20 0 15036 1216 908 R 0.3 0.1 0:00.32 top [chenhj@node2 ~]$ ps -ef|grep postgres chenhj 5011 1 0 11:08 ? 00:00:00 /usr/local/pgsql/bin/postgres -D data chenhj 5012 5011 0 11:08 ? 00:00:35 postgres: startup process recovering 0000000100000000000000A6 chenhj 5014 5011 0 11:08 ? 00:00:04 postgres: checkpointer process chenhj 5015 5011 0 11:08 ? 00:00:00 postgres: writer process chenhj 5016 5011 0 11:08 ? 00:00:02 postgres: stats collector process chenhj 8124 5011 0 12:13 ? 00:00:00 postgres: chenhj chenhj [local] idle chenhj 8181 5011 4 12:31 ? 00:00:24 postgres: wal receiver process streaming 0/A644EE20 chenhj 9477 8239 0 12:40 pts/3 00:00:00 grep postgres