From: Zardosht Kasheff
Date: July 16 2009 6:05pm
Subject: Re: help with index_merge and clustering keys
List-Archive: http://lists.mysql.com/internals/37199
Message-Id: <2f9663ba0907161105l5b85dd7r48b7781f1d4b395d@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: quoted-printable

Hello,

In TokuDB, one can cluster on multiple keys
(http://blogs.tokutek.com/tokuview/introducing_multiple_clustering_indexes)=
.
With this feature, the index_merge intersect code seems to need some
tweaking. Sergey and Rick have been kind enough to help me out in
solving this problem on this thread.

In the example below, I think Rick's implicit assumption was that keys
were either non-covering or clustered.

Thanks
-Zardosht

On Thu, Jul 16, 2009 at 1:34 PM, Michael Widenius<monty@stripped> wrote=
:
>
> This is true for the given query, but not in general.
>
> Assume you have in an InnoDB table and an index on (b,c) and 'a' is
> the primary key. In this case:
>
> select c from foo where a=3D1 and b =3D 1
>
> I't better to scan the index (b,c) than scan the primary key as you
> get all key parts from this index, but it's much less to scan as you
> don't have to access all the other columns from the table.
>
>
> Rick> * Both a and b are clustered: =A0Again, the other key is useless; s=
imply use the clustered key.
>
> How can you have something clustered on to different key?
> (Normally you can only cluster on one key at a time)
>
> Regards,
> Monty
>