SELECT r.name rr,
nvl(s.username,'no transaction') us,
s.osuser os,
s.terminal te
FROM v$lock l, v$session s,v$rollname r
WHERE l.sid = s.sid(+) AND
trunc(l.id1/65536) = r.usn AND
l.type = 'TX' AND
l.lmode = 6
ORDER BY r.name
RR US OS TE
---------- ---------- ---------- ------------------------------
_SYSSMU10$ APPLE Oracle pts/2
_SYSSMU4$ SYS oracle pts/0
觀察UNDO使用狀況
select b.NAME, a.USN seg#, GETS, WAITS, round(((GETS-WAITS)*100)/GETS,2) hit_ratio, XACTS active_transactions, WRITES
from v$rollstat a, v$rollname b where a.USN = b.USN
NAME SEG# GETS WAITS HIT_RATIO ACTIVE_TRANSACTIONS WRITES
------------------------------ ---------- ---------- ---------- ---------- ------------------- ----------
SYSTEM 0 385 0 100 0 7016
_SYSSMU1$ 1 36196 0 100 0 2397118
_SYSSMU2$ 2 36196 0 100 0 2397218
_SYSSMU3$ 3 32197 0 100 1 2136840
_SYSSMU4$ 4 33015 0 100 0 2183608
_SYSSMU5$ 5 36199 0 100 0 2395922
_SYSSMU6$ 6 36212 0 100 0 2398430
_SYSSMU7$ 7 36210 0 100 0 2398762
_SYSSMU8$ 8 36238 0 100 0 2400166
_SYSSMU9$ 9 36236 1 100 0 2440250
_SYSSMU10$ 10 36209 0 100 0 2396632
11 rows selected.
若HIT_RATIO < 99，表示你可能要考慮增加rollback segment了。
不管如何，在一個大量進(jìn)行交易的batch job里，適當(dāng)?shù)腸ommit可以大大減輕undo的負(fù)擔(dān)，或避免snapshoot too old的錯誤發(fā)生，又或者你是使用9i以上的版本，利用AUM可能比你自己去設(shè)定個別的RBS來的不易出狀況吧。
