RE: table shrink
- From: "Mark W. Farnham" <mwf@xxxxxxxx>
- To: <genegurevich@xxxxxxxxxxxx>
- Date: Thu, 10 Dec 2009 10:25:27 -0500
I use copy-keep as an umbrella for all of the methods having the same
resulting information as deleting what is to be removed. Moving the data
twice is usually suboptimal compared to moving the data once and doing the
appropriate dance with the object names and references. (And "temporary"
table has various distracting meanings in Oracle, so even if you choose to
do the data movement that way, I would use an "interim" permanent table.
Though I've had little success in de-ambigufying that in common conversation
since 1989.)
Notably, in your case, copying the data back into the original table would
not accomplish the apparent goal of releasing 40GB of extents allocated to
that table.
It is probably worth mentioning that if that table is the only table in that
tablespace, then copying the data to a diffent place and doing an offline
drop of the existing tablespace is probably the cheapest way to get your
storage back, and it will extend the reclamation all the way to your OS, or
in the case of ASM, to the disk group.
Regards,
mwf
-----Original Message-----
From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx]
On Behalf Of genegurevich@xxxxxxxxxxxx
Sent: Wednesday, December 09, 2009 6:28 PM
To: Mark W. Farnham
Cc: oracle-l@xxxxxxxxxxxxx
Subject: RE: table shrink
Mark,
When you refer to copy-keep are you referring to copying the data from my
table to a temporary table and loading them back into the original table?
Thank you
Eugene Gurevich
<snip>
No.
The undo required will be about:
1) the space management changes undo from all the other things
2) the overhead for the "block deltas" for the rows being moved
3) the whole size of the row from the block it is "deleted" from
4) the whole size of the row in the block it is placed in
5) all the index updates for any row that moves.
But without examining the table details, we have no idea what fraction of
the rows or total size might have to move to accomplish the shrink.
Now I don't see any size to keep info clause in the shrink clause.
What I see is:
shrink space [compact][cascade]
and a list of restrictions about when you can use it (which mostly appear
related to flows that would be really complex to code and really stupid to
try to do anyway.)
So I think this means you've purged (or are planning to purge) 40G of 41G
and then do the shrink. That's a really bad idea from the standpoint of how
much work the computer has to do to accomplish the goal and how much undo
you'll need. Copy-keep the 1G to somewhere else, even if you have to make a
union view to keep it continuously available if new inserts are continuing
during the move. Of course if you directly address the table right now,
then you'll have a slight bump when you rename the table and create the
view. If you already reference the table as an all inclusive view wrapper
or a synonyn, then create or replace view or synonym will handle this for
you. Sure, the database engine will have to reparse some queries, but the
data will be continuously available. I think I've written about this before
and it should be in the archives. Being able to do this thing, not the
convenience of not having to drop a view or synonym before creating a
different one of the same name, was the driving requirement of the Oracle
VLDB advising Oracle to implement this with "breakable parse locks" in the
early 1990's and it is a shame anyone purges by delete instead of using
copy-keep. Of course if you can stand an outage, you can still use
copy-keep and the much cheaper direct mode. Then you can unhook the old
table and dispose of it at your leisure in a mode that only has undo for
the dictionary and space management changes.
Obviously be careful and test.
-----Original Message-----
From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx]
On Behalf Of genegurevich@xxxxxxxxxxxx
Sent: Monday, December 07, 2009 5:51 PM
To: oracle-l@xxxxxxxxxxxxx
Subject: RE: table shrink
Thank you, Mark
Based on the oracle's recomendation, about 1 G will be kept in the table
and about 40G+ will be released. Does that this mean that the max
undo I will need is 1G?
Thank you
Eugene Gurevich
For a quick rough estimate, remember that the only undo you'll get from the
empty blocks is a tiny amount from the space management. It is also very
unlikely that any block lower in extent order than the first empty block
can
move. Finally, the total of blocks that still do have contents is a rough
ceiling on the undo, so if you're keeping less than you're shrinking that
is
a consideration.
If you can be out of action a bit, create table as select to keep the rows
you want to retain is probably the cheapest thing unless the amount kept is
greater than two times (and then measurements are required and the results
vary depending on indexes, etc.) the amount to be shuffled, and you might
have an opportunity to order the select so that in your destination your
most frequently used access path gets the best possible cluster score and
lowest cost for both the CBO and actual run time. Then of course you
probably do a rename and re-index.
-----Original Message-----
From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx]
On Behalf Of Adam Musch
Sent: Monday, December 07, 2009 10:58 AM
To: genegurevich@xxxxxxxxxxxx
Cc: oracle-l@xxxxxxxxxxxxx
Subject: Re: table shrink
Remember, if it's moving rows (and therefore, changing ROWIDs), it's
also got indexes to update. So it could require significantly more
UNDO than the size of the table. If you can afford the downtime, you
may be better off with impdp/expdp, alter table move, and/or online
table redefinition.
On Mon, Dec 7, 2009 at 9:46 AM, <genegurevich@xxxxxxxxxxxx> wrote:
> Hello everyone:
>
> I am researching alter table shrink space command in 10.2.0.3 before I
try
> it in production. According to oracle adviser, there is a table that can
be
> shrank
> to release over 40G of space and another which can release over 100G. I
> plan to execute the commands during a maintenance window to minimize the
> usage of these tables. I have run a small test in dev and saw that these
> command have been using undo tablespace. That tablespace is only 20G
> in my database. Does it need to match the size of the 'freeable' disk
> space, in other word do I need to increase it to about 100G? I also was
> told that
> i can cancel this command any time (in case there is a performance
> degradation) and whatever disk space has been released will stay released
> and
> I can run the command again to release the rest of the disk space. I have
> not been able to find this in any oracle documentation however. I don't
> have any table in dev to test a large shrink command.
>
> If anyone had any experience with this command, can you please address my
> questions on undo TS and cancelling the command please. I am
> alsowondering whether there is any downside to this operation.
>
> Thank you
>
> Eugene Gurevich
>
>
> Please consider the environment before printing this email.
>
>
> --
> //www.freelists.org/webpage/oracle-l
>
>
>
--
Adam Musch
ahmusch@xxxxxxxxx
--
//www.freelists.org/webpage/oracle-l
Please consider the environment before printing this email.
--
//www.freelists.org/webpage/oracle-l
Please consider the environment before printing this email.
i0zX+ n{+i^
--
//www.freelists.org/webpage/oracle-l
Other related posts:
