Re: WIP Join Removal

Simon Riggs <simon(at)2ndQuadrant(dot)com> writes:

> Same answer, just slower. Removing the join makes the access to a into a
> SeqScan, whereas it was a two-table index plan when both tables present.
> The two table plan is added by the immediately preceding call add_... -
> i.e. that plan is only added during join time not during planning of
> base relations.

Perhaps it would clearer to discuss a non-outer join here:

select invoices.*
from customer join invoices using (company_id,customer_id)
where customer_id = ?

where there's a foreign key relation guaranteeing that every invoice has a
matching <company_id, customer_id>.

If there's an index on customer(customer_id) but not on invoices(customer_id)
then conceivably it would be faster to use that than scan all of the invoices.

I wonder if it would be more worthwhile to remove them and have a subsequent
phase where we look for possible joins to *add*. So even if the user writes
"select * from invoices where customer_id=?" the planner might be able to
discover that it can find those records quicker by scanning customer, finding
the matching <company_id,customer_id> and then using an index to look them up
in invoices.

--
Gregory Stark
EnterpriseDB http://www.enterprisedb.com
Ask me about EnterpriseDB's 24x7 Postgres support!