Re: Which indexes to drop

On Tue, 2002-06-18 at 10:36, Josh Berkus wrote:
> John,
>
> > I'm optimising my data load by dropping indexes.
> >
> > My question is, what do I do with the indexes created by postgres for
> > primary keys ?
> >
> > Is it OK to drop them ?
>
> No.

I've done it many times...

alter table t_lane_tx
    add constraint c_lane_tx_pk primary key (lane_tx_id);

> > How do I create them again afterwards, so that they are used
> > correctly for constraint checking ?

In this instance, you can have speed or you can have correctness.
Pick one.  Of course, if you know that your input data is good,
you have both...

> I'm not sure that it's possible to add a primary key to an existing
> populated table.

See above.

> I'd reccommend instead dropping all other indexes but the primary key.
>  It's just not a good idea.   How much is this slowing down the data
> load?  Can you do a comparison test on a keyed vs. keyless table?

Ack!  Have you tried loading millions of rows to an indexed table?
Imagine, if you will, a 1MHz 8080 writing to a floppy drive...

> FYI, in 7.4 or 8.0 we will have DEFERRABLE UNIQUE constraints, which
> means that it may be possible for you to hold the PK checking until the
> data load is finished, adding some speed to the process.

Still, I bet that creating an (implicit or explicit) index would
be faster if you are loading the data in bite-sized chunks.

--
+-------------------------------------------------------------+
| Ron Johnson, Jr.        Home: ron.l.johnson@cox.net         |
| Jefferson, LA  USA      http://ronandheather.dhs.org:81     |
|                                                             |
| "Object-oriented programming is an exceptionally bad idea   |
|  which could only have originated in California."           |
|  --Edsger Dijkstra                                          |
+-------------------------------------------------------------+