Oracle query execution time

空扰寡人 提交于 2019-11-30 16:53:48
Adam Musch

One can issue the SQL*Plus command SET TIMING ON to get wall-clock times, but one can't take, for example, fetch time out of that trivially.

The AUTOTRACE setting, when used as SET AUTOTRACE TRACEONLY will suppress output, but still perform all of the work to satisfy the query and send the results back to SQL*Plus, which will suppress it.

Lastly, one can trace the SQL*Plus session, and manually calculate the time spent waiting on events which are client waits, such as "SQL*Net message to client", "SQL*Net message from client".

Use:

set serveroutput on
variable n number
exec :n := dbms_utility.get_time;
select ......
exec dbms_output.put_line( (dbms_utility.get_time-:n)/100) || ' seconds....' );

Or possibly:

SET TIMING ON;

-- do stuff

SET TIMING OFF;

...to get the hundredths of seconds that elapsed.

In either case, time elapsed can be impacted by server load/etc.

Reference:

select LAST_LOAD_TIME, ELAPSED_TIME, MODULE, SQL_TEXT elapsed from v$sql
  order by LAST_LOAD_TIME desc

More complicated example (don't forget to delete or to substitute PATTERN):

select * from (
  select LAST_LOAD_TIME, to_char(ELAPSED_TIME/1000, '999,999,999.000') || ' ms' as TIME,
         MODULE, SQL_TEXT from SYS."V_\$SQL"
    where SQL_TEXT like '%PATTERN%'
    order by LAST_LOAD_TIME desc
  ) where ROWNUM <= 5;

I'd recommend looking at consistent gets/logical reads as a better proxy for 'work' than run time. The run time can be skewed by what else is happening on the database server, how much stuff is in the cache etc.

But if you REALLY want SQL executing time, the V$SQL view has both CPU_TIME and ELAPSED_TIME.

易学教程内所有资源均来自网络或用户发布的内容,如有违反法律规定的内容欢迎反馈
该文章没有解决你所遇到的问题?点击提问,说说你的问题,让更多的人一起探讨吧!