Re: Dynamic Partitioning using Segment Visibility Maps

On Thu, Jan 10, 2008 at 04:51:04PM +0000, Simon Riggs wrote:
> On Thu, 2008-01-10 at 03:06 +0100, Gavin Sherry wrote:
>
> > > If people with large tables like partitioning why is Oracle moving
> > > towards automated partitioning in 11g? Automated partitioning was one of
> >
> > Have you used Oracle's partitioning?
>
> Since you ask, yep, certified on it, plus DB2, Teradata and Greenwich.

It was just a joke.

>
> > What about consistency over time? High levels of control for users if
> > they want it? Even simple things like tablespace support? High
> > performance unload and load performance much better than now?
> > Reliable and consistent query plans? Support for datatypes other than
> > those relating to time.
>
> Most of these comments seem like emotional appeals, so refuting them one
> by one is just going to look and smell like an argument, and a fruitless
> one as well since we agree on so many things.

Sigh. You can call those questions what you like but your approach
doesn't deal with them and the current partitioning approach, for it's
problems, does deal with them.

> So, I get the message that you really want the DDL approach and agree
> that you've demonstrated there are use cases that need it that you are
> interested in. That's fine by me as long as we can each work on parts of
> it to get it done. Will it be possible for you to do that?

I assured you offlist that it was. This is something Greenplum needs
right now and I'm being paid to deliver it.

> I feel I can say that because AFAICS we can in principle have both
> dynamic and declarative techniques, even on the same table. Around half
> of the mechanics would be identical anyway.
>
> So from here I say lets work together to get the basic mechanics right.
> My experience with PostgreSQL is that the team/list always comes up with
> a better idea in the end and I'm sure we will this time too.

I agree. Scrutiny of ideas almost always leads to better ideas.

> I have one main technical concern which is the storage model, which is
> the source of the difference between the two types of partitioning we've
> been discussing. I see difficulties originating there, so I will start
> another thread to examine just those items in detail.

I look forward to it. As promised, I'll soon post some syntax of what we
have been working on.

Thanks,

Gavin