数据库SQL的监控和优化的核心,其实主要在于“行读”!
监控找出行读高的SQL,而优化的最直接手段就是降低行读数。
oracle的监控方法和监控工具可能有千百种,这里介绍我常用的几种简单而直接的方式:
注意:以下监控SQL需要用户有oracle的管理视图查询权限。
1、根据pid查SQL
假如你的oracle此时响应速度较慢,从服务器的TOP可以看到oracle进程占用了较多CPU,或者有较多的iowait,那么可以直接用下面的SQL语句来查询正在执行的较耗资源的SQL。
SELECT A.,B.SQLTEXT FROM
(SELECT z.SPID AS 操作系统PID
,x.SID
, x.SERIAL#
, x.USERNAME
, x.SQLID
FROM v$session x
, v$process z
WHERE x.PADDR = z.ADDR
AND Z.SPID = ‘10341‘)A
LEFT JOIN v$sql B ON A.SQLID = B.SQLID;
恭喜你,你可能已经直接找到有问题的SQL了。
2、正在执行中的SQL
如果不知道pid,那稍微改动下上面的SQL,可以直接查当前oracle库正在执行的所有SQL,按照时间倒序排序,一样能抓到最可疑的SQL。
SELECT z.SPID AS 操作系统PID
,x.SID
, x.SERIAL#
, x.USERNAME
, x.LOCKWAIT
, x.STATUS
, x.SCHEMANAME
, x.OSUSER
, x.PROCESS
, x.MACHINE
, x.PORT
, x.PROGRAM
, x.SQL_ID
, x.SQL_CHILD_NUMBER
, x.SQL_EXEC_START
, x.PREV_SQL_ID
, CASE x.ROW_WAIT_OBJ# WHEN - 1 THEN NULL ELSE (dbms_rowid.rowid_create (1, ROW_WAIT_OBJ#, ROW_WAIT_FILE#, ROW_WAIT_BLOCK#, ROW_WAIT_ROW#)) END AS rowinfo
, x.LAST_CALL_ET
, x.BLOCKING_SESSION_STATUS
, x.BLOCKING_SESSION
, x.FINAL_BLOCKING_SESSION_STATUS
, x.FINAL_BLOCKING_SESSION
, x.EVENT
, x.WAIT_CLASS
, x.STATE
, x.WAIT_TIME_MICRO
, y.SQL_TEXT
, y.EXECUTIONS
, CASE EXECUTIONS WHEN 0 THEN y.BUFFER_GETS ELSE y.BUFFER_GETS / EXECUTIONS END AS AVG_BUFFER_GETS
, CASE EXECUTIONS WHEN 0 THEN y.DISK_READS ELSE y.DISK_READS / EXECUTIONS END AS AVG_DISK_READS
, y.ELAPSED_TIME / 1000000 AS AVG_ELAPSED_TIME
, y.CPU_TIME / 1000000 AS AVG_CPU_TIME
FROM v$session x
, v$process z
, v$sql y
WHERE x.PADDR = z.ADDR AND x.SQL_ID = y.SQL_ID
--AND Z.SPID = ‘XXXX‘
--AND x.USERNAME=‘DEV‘
AND x.STATUS = ‘ACTIVE‘
ORDER BY x.WAIT_TIME_MICRO desc;
3、刚刚执行过的SQL
如果是要查刚刚(由v$sqlarea存活时间而定)执行过的SQL中,哪些是比较耗资源的,可以试试下面这个:
SELECT *
FROM (SELECT EXECUTIONS
, BUFFER_GETS / EXECUTIONS AS AVG_BUFFER_GETS
, DISK_READS / EXECUTIONS AS AVG_DISK_READS
, ELAPSED_TIME / (EXECUTIONS*1000000) AS AVG_ELAPSED_TIME
, CPU_TIME / (EXECUTIONS*1000000) AS AVG_CPU_TIME
, SQL_TEXT
, SQL_ID
, SORTS / EXECUTIONS AS AVG_SORTS
, SHARABLE_MEM
, APPLICATION_WAIT_TIME / EXECUTIONS AS AVG_APPLICATION_WAIT_TIME
, CONCURRENCY_WAIT_TIME / EXECUTIONS AS AVG_CONCURRENCY_WAIT_TIME
, USER_IO_WAIT_TIME / EXECUTIONS AS AVG_USER_IO_WAIT_TIME
, ROWS_PROCESSED
FROM v$sqlarea
WHERE EXECUTIONS > 0 AND PARSING_SCHEMA_NAME<>‘SYS‘
AND LAST_ACTIVE_TIME BETWEEN to_date(‘2016-04-25 08:00:00‘,‘yyyy-mm-dd hh24:mi:ss‘) AND to_date(‘2017-07-25 12:00:00‘,‘yyyy-mm-dd hh24:mi:ss‘)
ORDER BY AVG_ELAPSED_TIME DESC) x
WHERE ROWNUM < 100;
修改ORDER BY AVG_ELAPSED_TIME DESC 为其他字段,比如 AVG_BUFFER_GETS就能查出哪些SQL的缓冲池读比较高。
注意:v$sqlarea是在内存中的,此SQL的查询时间范围没法太久远。
4、过去七天内执行过的SQL
如果是要查最近7天内执行过的SQL中,哪些是比较耗资源的,那我们可以改用dba_hist_sqlstat进行查询,它默认是存储7/8天的。不过需要预先知道dbid、instance_number、snap_id这几个数据,熟悉oracle的朋友应该清楚,awr报告的数据正是来源于此。
SELECT y.*,round(y.elap/y.exec,2) AS AVG_ELAP,z.SQL_TEXT FROM
(SELECT x.sql_id
, max (module) module
, sum (elapsed_time_delta)/ 1000000 elap
, sum (cpu_time_delta) cput
, sum (executions_delta) exec
FROM dba_hist_sqlstat x
WHERE x.dbid = 901548675 AND x.instance_number = 1 AND 5604 < x.snap_id AND x.snap_id <= 5612
AND x.MODULE = ‘JDBC Thin Client‘
GROUP BY x.sql_id)y
LEFT JOIN dba_hist_sqltext z ON y.sql_id = z.sql_id
WHERE y.exec >0
ORDER BY y.elap/y.exec DESC
5、AWR报告
如果以上4种方式还不足以解君之忧的话,那么还是来份AWR报告吧。
当然AWR报告也不是万能的,比如报告中TOP SQL的排序是按照总资源消耗倒序排序的,假设某些SQL执行次数很少但是单次消耗资源很多的,那可能会被掩盖掉而没有在报告中展示出来。
至于AWR报告如何生成如何解读,本篇不再赘述。
6、总结
以上介绍了几种非专业DBA快速定位oracle TOP SQL的方式,希望于君有益。