Re: Fast object checkpoint
- From: "Alex Gorbachev" <gorbyx@xxxxxxxxx>
- To: mgogala@xxxxxxxxxxx
- Date: Sat, 14 Oct 2006 22:50:29 -0400
So you assume it's just buggy. Fair enough. :-) I guess you already though of ALTER SYSTEM CHECKPOINT before your operation with direct IO. Not exactly old behavior (flushing all dirty blocks instead of just one tablespace).
On 9/23/06, Mladen Gogala <mgogala@xxxxxxxxxxx> wrote:
On 09/23/2006 02:21:21 PM, Jonathan Lewis wrote: > > > I don't understand why you think that writing the > dirty blocks from one object is likely to cause > more problems than writing the dirty blocks from > every object in the tablespace that holds the one > object that needs to be written. >
Jonathan, the session that motivated me to start investigating spent significant time waiting for the "KO" enqueue. V$LOCK_TYPE tells me that this is "Multiple Object Checkpoint" and that it "Coordinates checkpointing of multiple objects". I haven't had such problems with the usual checkpoints, which were well described in the documentation. The mere fact that I noticed waiting for the checkpoint means that there is a problem. Principle sounds right, on paper, but whoever has written anything more complex then printf("%s\n","Hello World"); knows that the devil lies in the details. I believe that the problem is in the sloppy coding and Elbonian project management. As Oracle 10.2 is the first version having such a nice feature that I had discovered yesterday, I'm inclined to turn it off until it's ready for the prime time.
-- Mladen Gogala http://www.mgogala.com
-- Best regards, Alex Gorbachev
The Pythian Group Sr. Oracle DBA
http://www.pythian.com/blogs/author/alex/ http://blog.oracloid.com -- http://www.freelists.org/webpage/oracle-l