标签：
这篇文章是记录一次ORACLE数据库UNDO表空间爆满的分析过程，主要整理、梳理了同事分析的思路。具体过程如下所示：
早上收到一数据库服务器的UNDO表空间的告警邮件，最早一封是7：55发出的（监控作业是15分钟一次），从告警邮件分析，好像是UNDO表空间突然一下子被耗尽了。
| DB | Tablespace | Allocated | Free | Used | % Free | % Used |
| 192.168.xxx.xxx:1521 | UNDOTBS1 | 16384 | 190.25 | 16193.75 | 1.16 | 99 |
使用一些SQL分析了undo表空间使用情况，以及undo segment状态等等，非常想定位到是哪个或那些SQL耗尽了UNDO表空间，但是没有一个SQL能实现我的想法，抑或是我不了解。
SELECT UPPER(F.TABLESPACE_NAME) AS "TABLESPACE_NAME",
ROUND(D.MAX_BYTES,2) AS "TBS_TOTAL_SIZE" , ROUND(D.AVAILB_BYTES ,2) AS "TABLESPACE_SIZE", ROUND(D.MAX_BYTES - D.AVAILB_BYTES +USED_BYTES,2) AS "TBS_AVABLE_SIZE", ROUND((D.AVAILB_BYTES - F.USED_BYTES),2) AS "TBS_USED_SIZE",TO_CHAR(ROUND((D.AVAILB_BYTES - F.USED_BYTES) / D.AVAILB_BYTES * 100,
2),
‘999.99‘) AS "USED_RATE(%)",
ROUND(F.USED_BYTES, 6) AS "FREE_SIZE(G)"FROM (SELECT TABLESPACE_NAME,
ROUND(SUM(BYTES) / (1024 * 1024 * 1024), 6) USED_BYTES, ROUND(MAX(BYTES) / (1024 * 1024 * 1024), 6) MAX_BYTES FROM SYS.DBA_FREE_SPACEGROUP BY TABLESPACE_NAME) F,
(SELECT DD.TABLESPACE_NAME, ROUND(SUM(DD.BYTES) / (1024 * 1024 * 1024), 6) AVAILB_BYTES, ROUND(SUM(DECODE(DD.MAXBYTES, 0, DD.BYTES, DD.MAXBYTES))/(1024*1024*1024),6) MAX_BYTES FROM SYS.DBA_DATA_FILES DDGROUP BY DD.TABLESPACE_NAME) D
HERE D.TABLESPACE_NAME = F.TABLESPACE_NAME
AND D.TABLESPACE_NAME=&UNDO_TABLESPACE_NAMERDER BY 5 DESC;
select usn,xacts,rssize/1024/1024/1024,hwmsize/1024/1024/1024,shrinksfrom v$rollstat order by rssize;
既然直接入手，无法定位，那就曲线分析，首先检查、分析了一下redo log，发现在7点这段时间，日志切换了83次之多，横向、纵向对比，明显异常，如下截图所示：
SELECT TO_CHAR(FIRST_TIME,‘YYYY-MM-DD‘) DAY,
TO_CHAR(SUM(DECODE(TO_CHAR(FIRST_TIME,‘HH24‘),‘00‘,1,0)),‘99‘) "00",
TO_CHAR(SUM(DECODE(TO_CHAR(FIRST_TIME,‘HH24‘),‘01‘,1,0)),‘99‘) "01",
TO_CHAR(SUM(DECODE(TO_CHAR(FIRST_TIME,‘HH24‘),‘02‘,1,0)),‘99‘) "02",
TO_CHAR(SUM(DECODE(TO_CHAR(FIRST_TIME,‘HH24‘),‘03‘,1,0)),‘99‘) "03",
TO_CHAR(SUM(DECODE(TO_CHAR(FIRST_TIME,‘HH24‘),‘04‘,1,0)),‘99‘) "04",
TO_CHAR(SUM(DECODE(TO_CHAR(FIRST_TIME,‘HH24‘),‘05‘,1,0)),‘99‘) "05",
TO_CHAR(SUM(DECODE(TO_CHAR(FIRST_TIME,‘HH24‘),‘06‘,1,0)),‘99‘) "06",
TO_CHAR(SUM(DECODE(TO_CHAR(FIRST_TIME,‘HH24‘),‘07‘,1,0)),‘99‘) "07",
TO_CHAR(SUM(DECODE(TO_CHAR(FIRST_TIME,‘HH24‘),‘08‘,1,0)),‘99‘) "0",
TO_CHAR(SUM(DECODE(TO_CHAR(FIRST_TIME,‘HH24‘),‘09‘,1,0)),‘99‘) "09",
TO_CHAR(SUM(DECODE(TO_CHAR(FIRST_TIME,‘HH24‘),‘10‘,1,0)),‘99‘) "10",
TO_CHAR(SUM(DECODE(TO_CHAR(FIRST_TIME,‘HH24‘),‘11‘,1,0)),‘99‘) "11",
TO_CHAR(SUM(DECODE(TO_CHAR(FIRST_TIME,‘HH24‘),‘12‘,1,0)),‘99‘) "12",
TO_CHAR(SUM(DECODE(TO_CHAR(FIRST_TIME,‘HH24‘),‘13‘,1,0)),‘99‘) "13",
TO_CHAR(SUM(DECODE(TO_CHAR(FIRST_TIME,‘HH24‘),‘14‘,1,0)),‘99‘) "14",
TO_CHAR(SUM(DECODE(TO_CHAR(FIRST_TIME,‘HH24‘),‘15‘,1,0)),‘99‘) "15",
TO_CHAR(SUM(DECODE(TO_CHAR(FIRST_TIME,‘HH24‘),‘16‘,1,0)),‘99‘) "16",
TO_CHAR(SUM(DECODE(TO_CHAR(FIRST_TIME,‘HH24‘),‘17‘,1,0)),‘99‘) "17",
TO_CHAR(SUM(DECODE(TO_CHAR(FIRST_TIME,‘HH24‘),‘18‘,1,0)),‘99‘) "18",
TO_CHAR(SUM(DECODE(TO_CHAR(FIRST_TIME,‘HH24‘),‘19‘,1,0)),‘99‘) "19",
TO_CHAR(SUM(DECODE(TO_CHAR(FIRST_TIME,‘HH24‘),‘20‘,1,0)),‘99‘) "20",
TO_CHAR(SUM(DECODE(TO_CHAR(FIRST_TIME,‘HH24‘),‘21‘,1,0)),‘99‘) "21",
TO_CHAR(SUM(DECODE(TO_CHAR(FIRST_TIME,‘HH24‘),‘22‘,1,0)),‘99‘) "22",
TO_CHAR(SUM(DECODE(TO_CHAR(FIRST_TIME,‘HH24‘),‘23‘,1,0)),‘99‘) "23"
FROMV$LOG_HISTORY
GROUP BY
TO_CHAR(FIRST_TIME,‘YYYY-MM-DD‘) ORDER BY 1 DESC;
生成了实例在7:00~8:00时间段的AWR报告，从下面指标我们可以看出，数据库实例在这段时间呢，其实是非常空闲的，因为DB Time为9.74（mins)
另外，从Time Model Statistics部分来看，主要时间花在background elapsed time，而不是DB Time,我们可以判断时间主要耗费在后台进程，而不是前台进程。另外sql execute elapsed time耗用了DB Time的70.36的时间。
然后我们来看SQL order by Gets部分信息, 第一个SQL是删除WRH$_SQL_PLAN的记录，当然也有删除wrh$_sqltext、WRH$_SEG_STAT_OBJ表记录的SQL，如下所示
DELETEFROM WRH$_SQL_PLAN tabWHERE (:beg_snap <= tab.snap_idAND tab.snap_id <= :end_snapAND dbid = :dbid)AND NOT EXISTS
(SELECT 1 FROM WRM$_BASELINE b WHERE (tab.dbid = b.dbid) AND (tab.snap_id >= b.start_snap_id) AND (tab.snap_id <= b.end_snap_id))
DELETEFROM wrh$_sqltext tabWHERE (tab.dbid = :dbidAND :beg_snap <= tab.snap_idAND tab.snap_id <= :end_snapAND tab.ref_count = 0)AND NOT EXISTS
(SELECT 1 FROM WRM$_BASELINE b WHERE (b.dbid = :dbid2 AND tab.snap_id >= b.start_snap_id AND tab.snap_id <= b.end_snap_id));
DELETEFROM WRH$_SEG_STAT_OBJ tabWHERE (:beg_snap <= tab.snap_idAND tab.snap_id <= :end_snapAND dbid = :dbid)AND NOT EXISTS
(SELECT 1 FROM WRM$_BASELINE b WHERE (tab.dbid = b.dbid) AND (tab.snap_id >= b.start_snap_id) AND (tab.snap_id <= b.end_snap_id));
查看SQL ordered by Reads部分信息，发现主要也是删除系统表WRH$_SQL_PLAN记录 （这个表是非常大的）
DELETEFROM WRH$_SQL_PLAN tabWHERE (:beg_snap <= tab.snap_idAND tab.snap_id <= :end_snapAND dbid = :dbid)AND NOT EXISTS
(SELECT 1 FROM WRM$_BASELINE b WHERE (tab.dbid = b.dbid) AND (tab.snap_id >= b.start_snap_id) AND (tab.snap_id <= b.end_snap_id))
然后我们查看AWR报告的Tablespace IO Stats部分，IO主要集中在SYSAUX,UNDOTBS1这两个表空间，然后你会发现那个表WRH$_SQL_PLAN就是在SYSAUX下
所以，上面种种证据显示，让我们几乎可以断定主要是下面这个SQL导致了UNDO表空间使用的暴增。当然分析过程中，还有一些旁听佐证。在此感觉没有必要一一列举了。
DELETEFROM WRH$_SQL_PLAN tabWHERE (:beg_snap <= tab.snap_idAND tab.snap_id <= :end_snapAND dbid = :dbid)AND NOT EXISTS
(SELECT 1 FROM WRM$_BASELINE b WHERE (tab.dbid = b.dbid) AND (tab.snap_id >= b.start_snap_id) AND (tab.snap_id <= b.end_snap_id))
标签：
原文地址：http://www.cnblogs.com/kerrycode/p/5693629.html
