Not exactly related to this ...
but I came across Note 4517470.8 which makes for interesting if not
depressing reading.
To mount a database exclusive, all shared instances would have had to be
shutdown first.
That would mean that all dirty buffers would have had to be written to disk.
However, this note seemingly implies that subsequent writes in the
exclusive mode
could corrupt the data -- actually that a select count(*) would return an
incorrect count.
How, I wonder ?
The text in the example
"This insert can put data in blocks above the hi HWM (hi water mark)
(Note: truncate is just an example of how this can occur and is not a
required action)"
isn't very clear.
Reading the bug, it would seem that the problem may not be with the datablocks but with the BMBs listing available blocks.
Is this bug on 10gR2 *ONLY* ?
I don't have a solution with the existing technology. This sounds like an excellent performance enhancement request to Oracle. They shouldn't need to write dirty buffers for the table being truncated, but they will need to be marked non-dirty or invalid, I think, so they don't get written out later.
Hemant K Chitale http://web.singnet.com.sg/~hkchital
-- //www.freelists.org/webpage/oracle-l
