Re: [HACKERS] having and union in v7beta

Jose Soares <jose@sferacarta.com> writes:
>>>> SELECT ... UNION (is 3 / 4 times slow)
>> 
>> Can't help you on that without more details, either.  What is the
>> query exactly, what plan does EXPLAIN show, and what plan did you
>> get from 6.5?

> psql7=> EXPLAIN select distretto from comuni union select codice_fiscale from comuni;
> NOTICE:  QUERY PLAN:

> Unique  (cost=1767.19..1808.90 rows=1668 width=12)
-> Sort  (cost=1767.19..1767.19 rows=16684 width=12)
-> Append  (cost=0.00..464.84 rows=16684 width=12)
-> Seq Scan on comuni  (cost=0.00..232.42 rows=8342 width=12)
-> Seq Scan on comuni  (cost=0.00..232.42 rows=8342 width=12)

> [ and exactly the same plan for 6.5 ]

OK, so much for my first thought that the 7.0 planner was choosing a
bad plan.

One relevant change is that Unique nodes now invoke the proper
type-specific equality function(s) to decide whether tuples are distinct
or not, instead of doing a bitwise comparison (memcmp()) like they did
before.  But it's tough to believe that that accounts for a 3-to-4x
slowdown of this query; certainly I don't see much performance
difference on the datatypes I tried.  What datatypes are your fields,
anyway?

The other possibility is that the Sort step is a lot slower in 7.0,
although I don't think it should be.  Are you running both versions
with the same -S setting, and if so what is it?  Could it be that
the query is right on the edge of needing to switch from memory-based
to disk-based sort?  Perhaps 7.0 is deciding that it needs to go to
disk a little sooner than 6.5 did.
        regards, tom lane