I have a table of messages with paths and inserted dates (among other
things), like so:

CREATE TABLE Messages (
   msgkey BIGSERIAL PRIMARY KEY,
   path TEXT NOT NULL,
   inserted TIMESTAMP WITHOUT TIMEZONE DEFAULT NOW()
);

I run a query to determine which days actually saw emails come in, like so:

SELECT DATE(inserted) FROM Messages GROUP BY DATE(inserted);

That's obviously not very efficient, so I made an index:

CREATE INDEX messages_date_inserted_ind ON Messages(DATE(inserted));

However, GROUP BY does not use this index:

=# explain analyze select date(inserted) from messages group by
date(inserted);
                                                         QUERY
PLAN
------------------------------------------------------------------------------------------------------------------------------
HashAggregate  (cost=104773.10..104789.51 rows=1313 width=8) (actual time=
31269.476..31269.557 rows=44 loops=1)
  ->  Seq Scan on messages  (cost=0.00..101107.25 rows=1466340 width=8)
(actual time=23.923..25248.400 rows=1467036 loops=1)
Total runtime: 31269.735 ms
(3 rows)


Is it possible to get pg to use an index in a group by?  I don't see why it
wouldn't be possible, but maybe I'm missing something.

Using pg 8.1.4...

Reply via email to