Re: Huge Dataset (perhaps?) made for Range Partitioning (Long)

[JinNo Kim]

Mikael Ronstrom wrote:
> Hi,

[...Snipped for brevity...]

> ALTER TABLE LARGE_TABLE ADD PARTITION (PARTITION
> new_of_new_partition VALUES LESS THAN
> (constant_expression))
>
> In the current version this will be slow since it
> recreates the entire table, but the optimised
> version runs on my desktop and is soon ready for
> intro into the 5.1 tree. The optimised version
> will simply create a new partition and not move
> any of the existing data.
>

I'll be able to start playing with this on Monday if all goes well - having 5.1 on the second server will hopefully prevent too many of the ALTER table queries that for now will force a recreate of the table :).

>
> There is also commands to drop partitions
> ALTER TABLE LARGE_TABLE DROP PARTITION (list of
> partition names to drop)
>
> There is also a command to reorganise the
> partitions (e.g. if you want the first 50 days of
> data to be stored in weekly partitions instead you
> can do that by using
> ALTER TABLE LARGE_TABLE REORGANISE PARTITION (list
> of partition names to reorganise)
> INTO (list of partition definitions)
>
> All lists are comma separated.
>
> In the code on my desktop there are also commands
> to
> REBUILD PARTITION (list of partition names) that
> will copy the partition to a new partition with
> the same name and characteristics to get data less
> fragmented e.g.
> OPTIMIZE PARTITION (list of partition names) more
> or similar as REBUILD but done by the
> storage engine internally, e.g. only rebuild
> indexes.
> also ANALYZE, CHECK and REPAIR PARTITION will be
> there.
>
> Documentation is also in the writings so I'll
> ensure that you get early access to that if you give
> me an email address to send it to.

That would be great, any additional references would be welcome. I'd love to get a better feel for how best to implement partitioning and see what I may have missed in my scans of this forum. Any documentation (understanding that it's still a work in progress) would be useful :)

Please use:

jinno_kim at yahoo dot com dot au

(Sorry, for the obfuscation...)


>
> > This is one of those times where I'm really motivated to help and excited about the
> > prospect of contributing anything (even just a testbed) to
> > something great.
> >
>
> Sounds great, it's also of immense value to us to
> check that we have developed the right tools
> for partitioning since obviously in an early
> version you can't do everything at once.
>
> Rgrds Mikael
>

Thank you,

-JNK