I just encountered something like this in an execution plan:
-> Hash (cost=19865.48..19865.48 rows=489 width=12) (never executed)
Output: ly.total_count, ly.customer_id
-> Subquery Scan on ly (cost=19864.50..19865.48 rows=489 width=12) (never executed)
Output: ly.total_count, ly.customer_id
-> HashAggregate (cost=19864.50..19864.99 rows=489 width=4) (never executed)
Output: orders_1.customer_id, count(*)
-> Seq Scan on public.orders orders_1 (cost=0.00..19847.00 rows=3500 width=4) (never executed)
Output: orders_1.id, orders_1.customer_id, orders_1.order_date, orders_1.amount,
orders_1.sales_person_id
Filter: (date_part('year'::text, (orders_1.order_date)::timestamp without time zone) =
(date_part('year'::text,(('now'::cstring)::date)::timestamp without time zone) - 1::double precision))
The above is only a part of the execution plan and represents a derived table that is outer joined to the main table.
Postgres is correct to not execute this, because the condition in the sub-query will indeed not return any rows.
I can see why the Hash Aggregate and the Hash Join nodes can be marked as "(never executed").
But why does the Seq Scan node have the "(never executed)" as well?
I can't see how Postgres could tell that the condition won't return anything without actually doing the Seq Scan (there
isno index on the column order_date)
Thomas