Re: Possible to improve optimisation / index usage based on domain properties of a function

On 20 Feb 2014, at 7:21, Sameer Kumar <sameer(dot)kumar(at)ashnik(dot)com> wrote:

> If however, I was to provide the below query, it uses a sequential scan based plan. The planner is unable to utilise any indexes because it can’t know what the function is going to return – thus unable to constrain the range at the time of planning the execution.
>
> select count(*) from streams where date(stream_date) = ‘2013-01-08’;

The inverse of that expression, if it’s possible to formulate one, would most likely use the index though:

select count(*) from streams where stream_date = inv_date(‘2013-01-08’);

>
>> I’m wondering if we could build into postgres some level of metadata regarding the properties of a function, such that the optimiser could filter against the range of values that the function is expected to return.
>>
>> In this case, it could deduce that the date function will only ever return a value for stream_date to within a certain maximum and minimum range.
>> Thus the planner could scan the index for all values of stream_date falling within +/- 24 hours of the right operand, and then check/re-check the results.
>>
> If you can't go for the smarter query, go for more optimum index by "expression based index"
>
> http://www.postgresql.org/docs/9.1/static/indexes-expressional.html

AFAIK, you can’t use expression based indexes to partition a table, so that won’t help the OP much.

Alban Hertroys
--
If you can't see the forest for the trees,
cut the trees and you'll find there is no forest.