Pavan Deolasee, 22.04.2008 12:57:
On Tue, Apr 22, 2008 at 4:11 PM, Thomas Kellerer <spam_eater@xxxxxxx> wrote:
I am interested in the technical details on how PG determines that a block
needs to be read from from "some other place than the data block" because
another transaction has updated the data block.
Postgres uses MVCC for transaction consistency. When a row is updated,
a new copy of the row is created and the old version is marked for
deletion (though its not removed immediately). The old readers
continue to read from the old copy whereas new transactions see the
new copy. This is all managed by XMIN and XMAX which are transaction
ids of the inserting and deleting transactions respectively and
control the visibility of the different versions.
If you want to understand MVCC, please see this excellent
documentation by Tom Lane:
http://www.postgresql.org/files/developer/transactions.pdf
The xmin, xmax info is basically what I was looking for.
As far as I can tell (from the PDF and your quote) Postgres uses a very similar concept as Oracle. .
Each transaction has a uniqe number and each tuple contains the information for which transaction number it is visible.
Thanks for the quick answer.
Thomas
