oracle学习----统计信息

浏览数：24 / 时间：2015年06月12日

1.收集统计信息的方式

for all columns size skewonly

BEGIN
  DBMS_STATS.GATHER_TABLE_STATS(ownname          => ‘SCOTT‘,
                                tabname          => ‘TEST‘,
                                estimate_percent => 100,
                                method_opt       => ‘for all columns size skewonly‘,
                                no_invalidate    => FALSE,
                                degree           => 8,
                                cascade          => TRUE);
END;
/

查看统计信息

select a.column_name,
b.num_rows,
a.num_distinct Cardinality,
round(a.num_distinct / b.num_rows * 100, 2) selectivity,
a.histogram,
a.num_buckets
from dba_tab_col_statistics a, dba_tables b
where a.owner = b.owner
and a.table_name = b.table_name
and a.owner = ‘SCOTT‘
and a.table_name = ‘TEST‘;

method_opt设置成for all columns size skewnoly时，让oracle自动判断是否需要收集直方图，除了主键列和唯一值列之外的列都收集直方图。这不符合我们收集直方图的规则，只有在where关键字之后出现的列才收集直方图。

for all columns size auto

创建一个新表，使用auto的方式对表test收集直方图。

BEGIN
  DBMS_STATS.GATHER_TABLE_STATS(ownname          => ‘SCOTT‘,
                                tabname          => ‘TEST‘,
                                estimate_percent => 100,
                                method_opt       => ‘for all columns size auto‘,
                                no_invalidate    => FALSE,
                                degree           => 8,
                                cascade          => TRUE);
END;
/

查看统计信息

可以看到，当对一个新表使用auto方式收集直方图的时候，num_buckets数为1，说明没有收集直方图

接下来运行sql，在收集直方图

SQL> select count(*) from test where owner=‘SB‘;

COUNT(*)
----------
0

exec dbms_stats.flush_database_monitoring_info; --刷新内存

查看where关键字后列值进行过滤操作的信息

select
r.name owner,
o.name table_name ,
c.name column_name,
equality_preds, ---等值过滤
equijoin_preds, ---等值JOIN过滤比如where a.id=b.id
nonequijoin_preds, ----不等JOIN过滤
range_preds, ----范围过滤 > >= < <= between and
like_preds, ----LIKE过滤
null_preds, ----NULL 过滤
timestamp
from
sys.col_usage$ u,
sys.obj$ o,
sys.col$ c,
sys.user$ r
where
o.obj# = u.obj#
and c.obj# = u.obj#
and c.col# = u.intcol#
and r.name=‘SCOTT‘ and o.name=‘TEST‘;

通过显示结果可以知道，用户scott下test表中的owner列进行了一次等值连接

再次执行auto模式收集统计信息

查询统计信息

可以看出对owner列收集了直方图，如果有sql运行，有where列，就收集直方图。

使用auto方式收集统计信息也是不可用的，在oracle11g里用没有问题，因为解决了绑定变量窥探的问题，在11g之前使用for all columns size 1方式，遇到需要收集的列单独收集。

for all columns size repeat

BEGIN
  DBMS_STATS.GATHER_TABLE_STATS(ownname          => ‘SCOTT‘,
                                tabname          => ‘TEST‘,
                                estimate_percent => 100,
                                method_opt       => ‘for all columns size repeat‘,
                                no_invalidate    => FALSE,
                                degree           => 8,
                                cascade          => TRUE);
END;
/

这种方式是推荐的方式，对以前收集过的列，进行收集，以前没收集过的，就不收集。

当系统运行稳定的时候采用这种方式收集统计信息。第一次上线的时候先使用for all columns size 1方式收集，在发现运行慢的sql对单独列进行收集统计信息，之后等系统运行稳定，使用repeat方式收集统计信息。

for columns <column name> size skewonly

BEGIN
  DBMS_STATS.GATHER_TABLE_STATS(ownname          => ‘SCOTT‘,
                                tabname          => ‘TEST‘,
                                estimate_percent => 100,
                                method_opt       => ‘for columns object_name size skewonly‘,
                                no_invalidate    => FALSE,
                                degree           => 8,
                                cascade          => TRUE);
END;
/

查看统计信息

可以看到对object_name列单独收集了直方图。

常用的统计信息相关的脚本

1.查看收集统计信息的采样率

SELECT owner,
table_name,
num_rows,
sample_size,
trunc(sample_size / num_rows * 100) estimate_percent
FROM DBA_TAB_STATISTICS
WHERE owner=‘SCOTT‘ AND table_name=‘TEST‘;

采样率是否合适，根据表的大小而定，根究dba_segments查看表的大小，制定采样率

2G以内采样率为100%

2G-5G采样率为50%

5G-10G采样率为30%

10G-20G采样率为10%

20G-30G采样率为5%

30G以上采样率为1%

2.查看统计信息是否过期

exec dbms_stats.flush_database_monitoring_info;

select owner, table_name name, object_type, stale_stats, last_analyzed
  from dba_tab_statistics
where table_name in (‘TEST‘)
   and owner = ‘SCOTT‘
   and (stale_stats = ‘YES‘ or last_analyzed is null);

统计信息过期的标准是，变化率超过10%

如果一个表一年都没有DML操作，是否对他重新收集统计信息呢？不需要，因为统计信息根本没有过期。

3.定制收集统计信息的脚本

DECLARE
CURSOR STALE_TABLE IS
    SELECT OWNER,
           SEGMENT_NAME,
           CASE
             WHEN SIZE_GB < 2 THEN
              100
             WHEN SIZE_GB >= 2 AND SIZE_GB < 5 THEN
              50
             WHEN SIZE_GB >= 5 AND SIZE_GB < 10 THEN
              30
             WHEN SIZE_GB >= 10 AND SIZE_GB < 30 THEN
              10
             WHEN SIZE_GB >= 30 THEN
              1
           END AS PERCENT,
           8 AS DEGREE
      FROM (SELECT OWNER,
                   SEGMENT_NAME,
                   SUM(BYTES / 1024 / 1024 / 1024) SIZE_GB
              FROM DBA_SEGMENTS
             WHERE OWNER = ‘ADWU_OPTIMA_AP11‘
               AND SEGMENT_NAME IN
                   (SELECT /*+ UNNEST */ DISTINCT TABLE_NAME
                      FROM DBA_TAB_STATISTICS
                     WHERE (LAST_ANALYZED IS NULL OR STALE_STATS = ‘YES‘)
                       AND OWNER = ‘ADWU_OPTIMA_AP11‘)
             GROUP BY OWNER, SEGMENT_NAME);

BEGIN
DBMS_STATS.FLUSH_DATABASE_MONITORING_INFO;
FOR STALE IN STALE_TABLE LOOP
    DBMS_STATS.GATHER_TABLE_STATS(OWNNAME          => STALE.OWNER,
                                  TABNAME          => STALE.SEGMENT_NAME,
                                  ESTIMATE_PERCENT => STALE.PERCENT,
                                  METHOD_OPT       => ‘for all columns size repeat‘,
                                  DEGREE           => 8,
                                  GRANULARITY      => ‘ALL‘,--这个是针对分区表，非分区表要进行改写

NO_INVALIDATE => FALSE,
CASCADE => TRUE);
END LOOP;
END;
/

4.DBA_TAB_STATS_HISTORY 可以查看表历史收集统计信息的时间，利用下面脚本分析某个表统计信息收集时间间隔

select owner,
      table_name,
      partition_name,
      subpartition_name,
      stats_update_time,
      stats_update_time - lag(stats_update_time, 1, null) over(partition by owner, table_name order by stats_update_time) interval
from DBA_TAB_STATS_HISTORY
where owner = ‘SCOTT‘
  and table_name = ‘TEST‘
order by owner, table_name, stats_update_time desc;

5.下面SQL是查询 TOP 50 INSERTS,UPDATES,DELETES 的表

select * from
(
select * from
(
select * from
(
select u.name owner, o.name table_name, null partition_name, null subpartition_name,
       m.inserts, m.updates, m.deletes, m.timestamp,
       decode(bitand(m.flags,1),1,‘YES‘,‘NO‘) truncated,
       m.drop_segments
from sys.mon_mods_all$ m, sys.obj$ o, sys.tab$ t, sys.user$ u
where o.obj# = m.obj# and o.obj# = t.obj# and o.owner# = u.user#
union all
select u.name, o.name, o.subname, null,
       m.inserts, m.updates, m.deletes, m.timestamp,
       decode(bitand(m.flags,1),1,‘YES‘,‘NO‘),
       m.drop_segments
from sys.mon_mods_all$ m, sys.obj$ o, sys.user$ u
where o.owner# = u.user# and o.obj# = m.obj# and o.type#=19
union all
select u.name, o.name, o2.subname, o.subname,
       m.inserts, m.updates, m.deletes, m.timestamp,
       decode(bitand(m.flags,1),1,‘YES‘,‘NO‘),
       m.drop_segments
from sys.mon_mods_all$ m, sys.obj$ o, sys.tabsubpart$ tsp, sys.obj$ o2,
     sys.user$ u
where o.obj# = m.obj# and o.owner# = u.user# and
      o.obj# = tsp.obj# and o2.obj# = tsp.pobj#
) where owner not like ‘%SYS%‘ and owner not like ‘XDB‘
union all
select * from
(
select u.name owner, o.name table_name, null partition_name, null subpartition_name,
       m.inserts, m.updates, m.deletes, m.timestamp,
       decode(bitand(m.flags,1),1,‘YES‘,‘NO‘) truncated,
       m.drop_segments
from sys.mon_mods$ m, sys.obj$ o, sys.tab$ t, sys.user$ u
where o.obj# = m.obj# and o.obj# = t.obj# and o.owner# = u.user#
union all
select u.name, o.name, o.subname, null,
       m.inserts, m.updates, m.deletes, m.timestamp,
       decode(bitand(m.flags,1),1,‘YES‘,‘NO‘),
       m.drop_segments
from sys.mon_mods$ m, sys.obj$ o, sys.user$ u
where o.owner# = u.user# and o.obj# = m.obj# and o.type#=19
union all
select u.name, o.name, o2.subname, o.subname,
       m.inserts, m.updates, m.deletes, m.timestamp,
       decode(bitand(m.flags,1),1,‘YES‘,‘NO‘),
       m.drop_segments
from sys.mon_mods$ m, sys.obj$ o, sys.tabsubpart$ tsp, sys.obj$ o2,
     sys.user$ u
where o.obj# = m.obj# and o.owner# = u.user# and
      o.obj# = tsp.obj# and o2.obj# = tsp.pobj#
) where owner not like ‘%SYS%‘ and owner not like ‘%XDB%‘
) order by inserts desc
) where rownum<=50;