AWR报告里和sql语句有关的section有SQL ordered by Elapsed Time、SQL ordered by CPU Time、SQL ordered by User I/O Wait Time

AWR报告里和sql语句有关的section有SQL ordered by Elapsed Time、SQL ordered by CPU Time、SQL ordered by User I/O Wait Time、SQL ordered by Gets等。一次在分析一个负荷较高的数据库时为了能在上述section中看到更多的SQL语句，特意通过DBMS_WORKLOAD_REPOSITORY.MODIFY_SNAPSHOT_SETTINGS(topnsql=>50)将AWR捕捉的SQL语句数量上限调整为50，，可以从dba_hist_wr_control里查到调整后的结果：
col SNAP_INTERVAL format a30
col RETENTION format a30
set linesize 150
select * from dba_hist_wr_control;
DBID SNAP_INTERVAL RETENTION TOPNSQL
---------- ------------------------------ ------------------------------ ----------
617151977 +00000 01:00:00.0 +00035 00:00:00.0 50

但后来在生成的AWR报告里，在"SQL ordered by Elapsed Time"、"SQL ordered by CPU Time"等section里sql语句的数量仍然是10条，并没有变成50条。后来发现在11gR2里有DBMS_WORKLOAD_REPOSITORY.AWR_SET_REPORT_THRESHOLDS可以用来控制AWR报告里的SQL语句的数量，用法如下
DBMS_WORKLOAD_REPOSITORY.AWR_SET_REPORT_THRESHOLDS(
top_n_events IN NUMBER DEFAULT NULL,
top_n_files IN NUMBER DEFAULT NULL,
top_n_segments IN NUMBER DEFAULT NULL,
top_n_services IN NUMBER DEFAULT NULL,
top_n_sql IN NUMBER DEFAULT NULL,
top_n_sql_max IN NUMBER DEFAULT NULL,
top_sql_pct IN NUMBER DEFAULT NULL,
shmem_threshold IN NUMBER DEFAULT NULL,
versions_threshold IN NUMBER DEFAULT NULL);

其中于到sql语句数量有关的参数是top_n_sql、top_n_sql_max、top_sql_pct，如果我们要在生成的AWR报告里包含50条语句，那么可以先执行
exec DBMS_WORKLOAD_REPOSITORY.AWR_SET_REPORT_THRESHOLDS(top_n_sql=>50)，然后再使用@?/rdbms/admin/awrrpt生成报告，如此报告里便能看到top 50的SQL了，记住DBMS_WORKLOAD_REPOSITORY.AWR_SET_REPORT_THRESHOLDS的执行结果仅在session级别生效。
对@?/rdbms/admin/awrrpt的执行过程进行了10046 trace，从trace结果里摘录了一段生成"SQL ordered by Elapsed Time"信息的sql：

with sqt as (select elap, cput, exec, iowt, norm_val, sql_id, module,rnum from (select sql_id, module, elap, norm_val, cput, exec, iowt,rownum rnum from (select sql_id,max(module) module,sum(elapsed_time_delta) elap,(100 *(sum(elapsed_time_delta) / nullif(:dbtime,0))) norm_val ,sum(cpu_time_delta) cput,sum(executions_delta) exec,sum(iowait_delta) iowt from dba_hist_sqlstat where dbid = :dbid and instance_number = :inst_num and :bid < snap_id and snap_id <= :eid group by sql_id order by nvl(sum(elapsed_time_delta), -1) desc, sql_id)) where rnum < :tsql_max and (rnum <= :tsql_min or norm_val > :top_pct_sql)) select /*+ NO_MERGE(sqt) */ nvl((sqt.elap/1000000),to_number(null)),sqt.exec,
decode(sqt.exec, 0,to_number(null),(sqt.elap / sqt.exec / 1000000)),sqt.norm_val,decode(sqt.elap, 0, to_number(null), (100 * (sqt.cput / sqt.elap))),decode(sqt.elap, 0, to_number(null), (100 * (sqt.iowt / sqt.elap))),sqt.sql_id,to_clob(decode(sqt.module, null,null,'Module: ' || sqt.module)),nvl(st.sql_text,to_clob('** SQL Text Not Available **')) from sqt, dba_hist_sqltext st where st.sql_id(+) = sqt.sql_id and st.dbid(+) = :dbid order by sqt.rnum

可以看出其中有一段where rnum < :tsql_max and (rnum <= :tsql_min or norm_val > :top_pct_sql)这便是AWR_SET_REPORT_THRESHOLDS的设置值对AWR结果起到的过滤作用，也可以看出dba_hist_sqlstat和dba_hist_sqltext是sql统计结果的主要来源。解释一下AWR_SET_REPORT_THRESHOLDS里三个参数top_n_sql、top_n_sql_max、top_sql_pct的用途。抽象一点，top_sql_pct表示某个sql_id对应的sql语句所消耗的资源占整个DB资源的百分比；
具体一点，
拿sql ordered by elapsed time里列出的sql来说，top_sql_pct表示每条sql语句消耗的时间占db time的百分比：top_sql_pct_for_elapsed_time%=(sql_elapse_time/db_time)*100%；
如果拿sql ordered by cpu time里列出的sql来说，top_sql_pct表示每条sql语句消耗的cpu时间占db cpu的百分比：top_sql_pct_for_cpu_time%=（sql_cpu_time/db_cpu)*100%;
以此类推，AWR里用%Total表示了这个百分比值。

top_n_sql、top_n_sql_max都是表示按照Elapsed Time、CPU Time、Gets等指标数值从高到低排序后返回的sql语句的数量，结合top_sql_pct通过下面几个场景解释一下top_n_sql、top_n_sql_max所起的作用(以sql ordered by elapsed time为例)

(1)如果top_n_sql=top_n_sql_max=N
返回elapsed time最长的N-1条sql，top_sql_pct值被忽略

(2)如果top_n_sql=N，top_n_sql_max=M，且满足N+1=M
返回elapsed time最长的N条sql，top_sql_pct值被忽略