Jayadevan M wrote:
> Hi,
> > The primary question that needs to be asked is what do you want to
> do with them?
> > It is not so much a performance issue as an admin issue. OIDs where
> created for
> > Postgres internal system use and leaked out to user space. As a
> result they
> > have some shortcomings as detailed in the above article. Given that
> sequences
> > are available as number generators, it was decided to
> encourage/force OIDs to
> > be for internal system use only. That decision is set and using OIDs
> on user
> > tables is setting yourself for future problems.
>
> I am an Oracle guy who is learning PostgreSQL. oid sounded a lot like
> rowid in Oracle. In Oracle, access by rowid is expected to be the
> fastest way of accessing a record, faster than even an index access
> followed by table access using the primary key. That was why I have
> this doubt about usage of oid being deprecated. Even if we use a
> sequence as PK (which is there in Oracle too), it is not as fast as
> access by rowid (I don't know if this applies to PostgreSQL's oid
> too). This is important when we use a cursors in an Oracle procedure
> (function in PostgreSQL) and loop through it and update specific
> records, when some conditions are met. Of course, that approach has
> its drawbacks -as in the case when row movement is enabled some
> maintenance activity moves the row to another location. Another
> scenario is when we want to delete duplicate records in a table.
well, postgres' OID's were never a direct row address of any sort. as
the previous poster said, OID's were an internal identifier, and were
never really meant for general use but their use was tolerated in
earlier versions of postgres when there were things you couldn't do
without them. Even in Oracle, I don't believe rowid bypasses
indexes, its more like an implicit SERIAL PRIMARY KEY field.