Re: Query palns and tug-of-war with enable_sort

Glyn Astill <glynastill(at)yahoo(dot)co(dot)uk> writes:
> With enable_sort on this is the plan it chooses:

> HashAggregate (cost=14.72..14.73 rows=1 width=9)
> -> Nested Loop (cost=0.00..14.72 rows=1 width=9)
> -> Index Scan using credit_index02 on credit (cost=0.00..7.04 rows=1 width=9)
> Index Cond: ((date >= '2009-02-16'::date) AND (date <= '2009-02-16'::date))
> Filter: (((cancelled)::text = ' '::text) AND ((show = 450000::numeric) OR (show = 450
> 001::numeric)))
> -> Index Scan using mult_ord_index02 on mult_ord (cost=0.00..7.67 rows=1 width=17)
> Index Cond: (mult_ord.transno = credit.transno)

> That's what I want, good. Now with enable_sort off this is the plan it chooses:

> Group (cost=0.00..11149194.48 rows=1 width=9)

That's just bizarre. Can you put together a self-contained test case
for this? Also, what version is it exactly? ("8.3" is the wrong
answer.)

regards, tom lane