Re: consistent gets

Mentre cercavo di capire la differenza tra una select e un drop database, Jonathan Lewis ha scritto:

>
> Immediately after you deleted the huge amount of
> data, there were lots of empty leaf blocks at the
> left-hand edge of the index.
>
> Your query for min() would have to start at the
> left hand edge, and walk through all those empty
> blocks (building read-consistent versions if necessary)
> to find the first row that still existed. This is why you
> get so many CR gets.

[CUT]
> The empty blocks will eventually be re-used or
> taken out of the index structure, but for very
> large deletes, the time-delay for a proper cleanout
> can be significant.

[CUT] Thank you very much Jonathan a very clear explanation. But this means that the index *is not* always balanced, because in such situations the leaf blocks at the left hand are empty.
The dml I've used was "delete from TABLE where cdate < data parallel ...." so I think that *all* the blocks starting from the left side up to a block in the middle (which contains data > data_used_on_dml) should be deleted. If this is right....than I don't understand how can the index stay balanced in such kind of situations.

regards Received on Tue Mar 22 2005 - 06:50:50 CST