对临时段文件生成UNDO有什么意义呢？

Yong Huang

原帖由 alantany 于 2009-9-9 18:59 发表
这应该是一些维护字典表的undo，并非数据产生的Undo，比如extent扩展啊之类的东西。

That makes sense. But if the initial temp segment is already created and you keep inserting one row at a time into the simplest GTT (e.g. one that only has one single column of type number), "undo change vector size" still goes up. V$sort_usage doesn't say the temporary segment is using more and more blocks (mine stays at 256). But at this time, my v$transaction.used_urec increments by 1 every time I insert a row into my GTT.

So I don't think it's extent being extended. Got to be something. There're some undo related events (10221 for exampel). When I have time, I'll take a look at it.

Yong Huang

dhhb

原帖由 Yong Huang 于 2009-9-11 02:42 发表


That makes sense. But if the initial temp segment is already created and you keep inserting one row at a time into the simplest GTT (e.g. one that only has one single column of type number), "undo change vector size" still goes up. V$sort_usage doesn't say the temporary segment is using more and more blocks (mine stays at 256). But at this time, my v$transaction.used_urec increments by 1 every time I insert a row into my GTT.

So I don't think it's extent being extended. Got to be something. There're some undo related events (10221 for exampel). When I have time, I'll take a look at it.

Yong Huang

I have some procedures will use temp table and return refcursor:


        OPEN O_REF FOR  SELECT * FROM TMP_table
        DELETE FROM tmp_table  ;
        COMMIT;

I can fetch data from O_REF, I believe it will read the data from undo.

What I think is undo for temp segment should be necessary because  an transaction need to rollback if it fails.

[ 本帖最后由 dhhb 于 2009-9-11 16:30 编辑 ]

蓝魔之恋

回滚、数据恢复、读一致性，还有闪回。