DBA任务---确保统计信息准确性

本文转自：http://blog.csdn.net/robinson1988/article/details/6321537

最近忙得不可开交，项目进入了cut over阶段，压力之大，前所未有。我的任务就是，负责优化long running的SQL，让其可以在3小时以内完成。昨天就出现一个Long running 的SQL，它跑了16小时，经过2小时的奋斗，终于把它优化到了2小时10分钟。

    虽然那个Long running SQL 与统计信息无关，但是我还是提出要确保统计信息的准确性。作为DBA，我必须定制出收集统计信息的策略，以及相关脚本，下面就是一个关于确保统计信息准确性的脚本，拿出来分享一下。

DECLARE
  CURSOR STALE_TABLE IS
    SELECT OWNER,
           SEGMENT_NAME,
           CASE
             WHEN SIZE_GB < 0.5 THEN
              30
             WHEN SIZE_GB >= 0.5 AND SIZE_GB < 1 THEN
              20
             WHEN SIZE_GB >= 1 AND SIZE_GB < 5 THEN
              10
             WHEN SIZE_GB >= 5 AND SIZE_GB < 10 THEN
              5
             WHEN SIZE_GB >= 10 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',
                                  CASCADE          => TRUE);
  END LOOP;
END;
/

根据实际情况，可以选择每天晚上在数据库不繁忙的时候运行上述脚本。