[Bizgres-general] Re: Any changes to partitioning strategy?

[Simon Riggs]

On Sun, 2005-06-19 at 10:59 -0700, Josh Berkus wrote:
> Simon,
> 
> Have the discussions on -hackers changed any aspects of your original table 
> partitioning specification?  I need to include it in my presentation for 
> OSCON.

I'm currently implementing a generalised form of partitioning that will
work for 
i) direct restriction of columns in the WHERE clause (not joins)
ii) where a CHECK constraints on inherited tables will be used to
exclude tables.
iii) only for non-prepared queries (since elimination will only be done
at planning time to begin with).

That forms the basis for all further work.

> Also, when will we be ready to talk about syntax?

With this generalised form, there is no need for syntax at all, apart
from the need to specify the check constraints.

However, without some hint to the planner, generalised partition
elimination *could* make planning take too long. If this is the case, it
would be my suggestion to introduce optional syntax that would direct
the planner and reduce time taken. My suggested syntax would look much
like the usual RANGE PARTITIONED BY (col) ideas discussed before.

What I'm thinking now is we only have syntax *if* we need it, rather
than smothering the code and manuals with lots of additional stuff. 

There are some additional things like mutually exclusive constraints
that will require syntax...

There is likely to be a new GUC called something like
enable_constraint_based_partition_elimination but I'll bet that a
shorter name will be the name that wins. 

Best Regards, Simon Riggs