Postgres的日志实用功能

不得不说，Postgres的日志(pg_log，类似oracle的alter文件，非pg_xlog)确实是很灵活，功能也很丰富的，下面是借用postgres的日志来实现一些管理功能，下面涉及的参数都在文件$PGDATA/postgresql.conf里面。 

OS:CentOS 6.2 
DB:Postgres 9.2.3 

1.日志审计 
审计是值记录用户的登陆退出以及登陆后在数据库里的行为操作，可以根据安全等级不一样设置不一样级别的审计， 
此处涉及的参数文件有：

logging_collector      --是否开启日志收集开关，默认off，开启要重启DB
log_destination    --日志记录类型，默认是stderr，只记录错误输出
log_directory      --日志路径，默认是$PGDATA/pg_log
log_filename       --日志名称，默认是postgresql-%Y-%m-%d_%H%M%S.log
log_connections    --用户session登陆时是否写入日志，默认off
log_disconnections --用户session退出时是否写入日志，默认off
log_rotation_age   --保留单个文件的最大时长,默认是1d,也有1h,1min,1s,个人觉得不实用
log_rotation_size  --保留单个文件的最大尺寸，默认是10MB

配置值：

logging_collector = on
log_destination = 'csvlog'
log_directory = '/home/postgres/pg_log'
log_filename = 'postgresql-%Y-%m-%d_%H%M%S.log'
log_connections = on
log_disconnections = on
log_rotation_age = 1d
log_rotation_size = 20MB

配置完重启DB，检查日志情况

[postgres@localhost pg_log]$ ls -l
total 4
-rw-------. 1 postgres postgres  672 Mar 29 08:25 postgresql-2013-03-29_000000.csv
-rw-------. 1 postgres postgres    0 Mar 29 00:00 postgresql-2013-03-29_000000.log
[postgres@localhost pg_log]$ 

--登陆并退出，日志内容有访问的IP(local),访问用户，登陆和退出时间等信息，对检查超级用户的登陆退出是很有效的
[postgres@localhost pg_log]$ psql
psql (9.2.3)
Type "help" for help.

postgres=# \q
[postgres@localhost pg_log]$ tail -f postgresql-2013-03-29_000000.csv 
2013-03-29 10:38:36.934 PDT,,,2236,"",5155d19c.8bc,1,"",2013-03-29 10:38:36 PDT,,0,LOG,00000,"connection received: host=[local]",,,,,,,,,""
2013-03-29 10:38:36.938 PDT,"postgres","postgres",2236,"[local]",5155d19c.8bc,2,"authentication",2013-03-29 10:38:36 PDT,2/11858,0,LOG,00000,"connection authorized: user=postgres database=postgres",,,,,,,,,""
2013-03-29 10:38:42.365 PDT,"postgres","postgres",2236,"[local]",5155d19c.8bc,3,"idle",2013-03-29 10:38:36 PDT,,0,LOG,00000,"disconnection: session time: 0:00:05.431 user=postgres database=postgres host=[local]",,,,,,,,,"psql"

记录用户登陆数据库后的各种操作，postgres日志里分成了3类，通过参数pg_statement来控制，默认的pg_statement参数值是none，即不记录，可以设置ddl(记录create,drop和alter)、mod(记录ddl+insert,delete,update和truncate)和all(mod+select)。 

示例： 
[postgres@localhost ~]$ vi $PGDATA/postgresql.conf 
log_statement = ddl

postgres=# show log_statement;
 log_statement 
---------------
 ddl
(1 row)

postgres=# create table t_ken_yon(id int);
CREATE TABLE
postgres=# drop table t_ken_yon ;
DROP TABLE
postgres=# 

[postgres@localhost pg_log]$ tail -f postgresql-2013-03-29_000000.csv 
2013-03-29 11:01:29.048 PDT,"postgres","postgres",2324,"[local]",5155d681.914,3,"idle",2013-03-29 10:59:29 PDT,2/11945,0,LOG,00000,"statement: create table t_ken_yon(id int);",,,,,,,,,"psql"
2013-03-29 11:01:36.087 PDT,"postgres","postgres",2324,"[local]",5155d681.914,4,"idle",2013-03-29 10:59:29 PDT,2/11948,0,LOG,00000,"statement: drop table t_ken_yon ;",,,,,,,,,"psql"

--修改为mod级别，并reload
postgres=# show log_statement;
 log_statement 
---------------
 mod
(1 row)
postgres=# insert into t_ken_yon values(1),(2);
INSERT 0 2
postgres=# delete from t_ken_yon where id =1;
DELETE 1

[postgres@localhost pg_log]$ tail -f postgresql-2013-03-29_000000.csv
2013-03-29 11:04:08.148 PDT,,,5554,,514933a6.15b2,42,,2013-03-19 20:57:26 PDT,,0,LOG,00000,"received SIGHUP, reloading configuration files",,,,,,,,,""
2013-03-29 11:04:08.151 PDT,,,5554,,514933a6.15b2,43,,2013-03-19 20:57:26 PDT,,0,LOG,00000,"parameter ""log_statement"" changed to ""mod""",,,,,,,,,""
2013-03-29 11:05:33.346 PDT,"postgres","postgres",2324,"[local]",5155d681.914,6,"idle",2013-03-29 10:59:29 PDT,2/11952,0,LOG,00000,"statement: insert into t_ken_yon values(1),(2);",,,,,,,,,"psql"
2013-03-29 11:05:52.033 PDT,"postgres","postgres",2324,"[local]",5155d681.914,7,"idle",2013-03-29 10:59:29 PDT,2/11953,0,LOG,00000,"statement: delete from t_ken_yon where id =1;",,,,,,,,,"psql"

--修改为all级别，并reload
postgres=# show log_statement;
 log_statement 
---------------
 all
(1 row)

postgres=# select * from t_ken_yon;
 id 
----
  2
(1 row)

[postgres@localhost pg_log]$ tail -f postgresql-2013-03-29_000000.csv 
2013-03-29 11:07:14.820 PDT,,,5554,,514933a6.15b2,44,,2013-03-19 20:57:26 PDT,,0,LOG,00000,"received SIGHUP, reloading configuration files",,,,,,,,,""
2013-03-29 11:07:14.821 PDT,,,5554,,514933a6.15b2,45,,2013-03-19 20:57:26 PDT,,0,LOG,00000,"parameter ""log_statement"" changed to ""all""",,,,,,,,,""
2013-03-29 11:07:19.784 PDT,"postgres","postgres",2324,"[local]",5155d681.914,8,"idle",2013-03-29 10:59:29 PDT,2/11954,0,LOG,00000,"statement: show log_statement;",,,,,,,,,"psql"
2013-03-29 11:07:28.631 PDT,"postgres","postgres",2324,"[local]",5155d681.914,9,"idle",2013-03-29 10:59:29 PDT,2/11955,0,LOG,00000,"statement: select * from t_ken_yon;",,,,,,,,,"psql"

一般的OLTP系统审计级别设置为ddl就够了，因为记录输出各种SQL对性能的影响还是蛮大的，安全级别高一点的也可以设置mod模式，有条件也可以不在数据库层面做，而是购买设备放在网络层监控解析。 

   2.定位慢查询SQL  
可以设置一定时长的参数(log_min_duration_statement)，来记录超过该时长的所有SQL，对找出当前数据库的慢查询很有效。 比如log_min_duration_statement = 2s,记录超过2秒的SQL，改完需要reload 

示例:

postgres=# show log_min_duration_statement ;
 log_min_duration_statement 
----------------------------
 2s
(1 row)

postgres=# \timing 
Timing is on.
postgres=# select now(),pg_sleep(1);
             now              | pg_sleep 
------------------------------+----------
 2013-03-29 12:36:48.13353-07 | 
(1 row)

Time: 1001.844 ms
postgres=# select now(),pg_sleep(4);
              now              | pg_sleep 
-------------------------------+----------
 2013-03-29 12:36:28.309595-07 | 
(1 row)

Time: 4002.273 ms

[postgres@localhost pg_log]$ tail -f postgresql-2013-03-29_000000.csv 
2013-03-29 12:36:19.265 PDT,"postgres","postgres",2324,"[local]",5155d681.914,10,"SELECT",2013-03-29 10:59:29 PDT,2/0,0,LOG,00000,"duration: 4027.183 ms  statement: select now(),pg_sleep(4);",,,,,,,,,"psql"

可以看到只记录了4秒的那个SQL，而没有记录1秒的SQL。 

3.监控数据库的checkpoint 
当数据库进行一项大更新操作时，如果参数设置不当，会在日志里留下大量的告警信息，频繁的做checkpoint会导致系统变慢，如:

2013-03-28 17:01:39.523 CST,,,10350,,50bd676b.286e,1,,2012-12-04 11:00:59 CST,,0,LOG,00000,"checkpoints are occurring too frequently (8 seconds apart)",,"Consider increasing the configuration parameter ""checkpoint_segments"".",,,,,,,""
2013-03-28 17:01:50.427 CST,,,10350,,50bd676b.286e,2,,2012-12-04 11:00:59 CST,,0,LOG,00000,"checkpoints are occurring too frequently (11 seconds apart)",,"Consider increasing the configuration parameter ""checkpoint_segments"".",,,,,,,""

但是不会记录系统正常的checkpoint，如果你想看系统一天之类发生了多少次checkpoint，以及每次checkpoint的一些详细信息，比如buffer,sync等，就可以通过设置log_checkpoints，该参数默认值是off，修改log_checkpoints = on 示例：

postgres=# show log_checkpoints ;
 log_checkpoints 
-----------------
 on
(1 row)
postgres=# checkpoint;
CHECKPOINT
postgres=#

[postgres@localhost pg_log]$ tail -f postgresql-2013-03-29_000000.csv 
2013-03-29 12:43:38.900 PDT,,,5557,,514933a7.15b5,45,,2013-03-19 20:57:27 PDT,,0,LOG,00000,"checkpoint starting: immediate force wait",,,,,,,,,""
2013-03-29 12:43:38.941 PDT,,,5557,,514933a7.15b5,46,,2013-03-19 20:57:27 PDT,,0,LOG,00000,"checkpoint complete: wrote 0 buffers (0.0%); 0 transaction log file(s) added, 0 removed, 0 recycled; write=0.009 s, sync=0.000 s, total=0.040 s; sync files=0, longest=0.000 s, average=0.000 s",,,,,,,,,""

 4.监控数据库的锁 
数据库的锁通常可以在pg_locks这个系统表里找，但这只是当前的锁表/行信息，如果你想看一天内有多少个超过死锁时间的锁发生，可以在日志里设置并查看，log_lock_waits 默认是off，可以设置开启。这个可以区分SQL慢是资源紧张还是锁等待的问题。 示例:

postgres=# show log_lock_waits ;
 log_lock_waits 
----------------
 on
(1 row)

postgres=# show deadlock_timeout ;
 deadlock_timeout 
------------------
 1s
(1 row)

--模拟锁
postgres=# begin;
BEGIN
postgres=# SELECT * FROM t_ken_yon ;
 id 
----
 11
(1 row)

postgres=# delete from t_ken_yon ;
DELETE 1

--另一个session
postgres=# begin;
BEGIN
postgres=# delete from t_ken_yon;

--查看日志
[postgres@localhost pg_log]$ tail -f postgresql-2013-03-29_000000.csv
2013-03-29 14:01:02.673 PDT,"postgres","postgres",3056,"[local]",5155f4d9.bf0,6,"DELETE waiting",2013-03-29 13:08:57 PDT,5/12502,2659,LOG,00000,"process 3056 still waiting for ShareLock on transaction 2658 after 1000.398 ms",,,,,,"delete from t_ken_yon;",,,"psql"
2013-03-29 14:02:06.208 PDT,"postgres","postgres",3056,"[local]",5155f4d9.bf0,7,"DELETE waiting",2013-03-29 13:08:57 PDT,5/12502,2659,LOG,00000,"process 3056 acquired ShareLock on transaction 2658 after 64535.339 ms",,,,,,"delete from t_ken_yon;",,,"psql"
2013-03-29 14:02:06.209 PDT,"postgres","postgres",3056,"[local]",5155f4d9.bf0,8,"DELETE",2013-03-29 13:08:57 PDT,5/12502,2659,LOG,00000,"duration: 64536.118 ms  statement: delete from t_ken_yon;",,,,,,,,,"psql"

还有一些Debug功能，适合修改源码调试，一般的的系统上并不需要，暂时比较关注的就这些。