Re: Tbs READ ONLY and Snapshot too old
- From: "Binley Lim" <Binley.Lim@xxxxxxxxxx>
- To: <oracle-l@xxxxxxxxxxxxx>
- Date: Sat, 19 Jun 2004 00:58:51 +1200
MessageActually, you are both right.
[BL] They cannot both be right -- either you get ORA-1555 on a READ ONLY
tablespace or you do not.
Delayed Block Cleanout can cause an ORA-1555, even in a read only tablespace.
[BL] Agreed.
However, if the tablespace has been read only "for a long enough period of
time",
every query against it will realize that all updates to objects in that
tablespace have
been committed, and will never try to reconstruct the table.
[BL] Disagree. What is "long enough" and how can a query "realise" something?
Problem is the ITL entries which tell the query to visit the rollback segments
cannot be cleaned out because they are in a READ ONLY tablespace. Every time
the block(s) are queried, they repeat the same exercise. This problem is
magnified when you do selective restores of tablespaces without the rollback
segments, and find you cannot read from those tablespaces without the rollback
tablespaces. Also, you don't reconstruct the table, but just reconstruct the
individual blocks.
As it happens, I have never had an ORA-1555 on a table which is (was?) being
updated during the query; I've had lots of them due to delayed block cleanout.
This was my first thought when I read the original post, however, as I read the
OP,
I think that the errors are appearing after the tablespace has been read only
for several days.
[BL] A table being updated is guaranteed to have its undo entries intact.Since
this is the only way the rollback can happen if the update does not commit.
Your query will find what it needs in the rollback segments, and not get an
ORA-1555, unless the update has already committed, or rolled-back.
As a test, (if the OP is still reading, and my time sequence is spot on), try
computing all statistics on all objects in the tablespace (not estimate, full
compute). This will visit every block on all tables and indexes (if any are
in the tablespace) and should clean out all the blocks.
[BL] Much easier to simply do "select max(column)..." on an un-indexed column
to cause FTS to cause cleanouts. Statistics is a bit of an overkill.
Other related posts:
