From this it looks like the bottleneck happens when Postgres does an Index Scan using emotions_moment_id_idx on emotions before filtering on moments.inserted so I thought I'd try filtering on emotions.inserted instead but that only made it worse. At the same time, I noticed that "FROM pg_class, moments WHERE moments.tableoid = pg_class.oid" tends to run a bit faster than "FROM pg_class JOIN moments ON moments.tableoid = pg_class.oid". So I tried:
SELECT relname, emotion, COUNT(feedback_id)
FROM pg_class, moments, emotions
WHERE moments.tableoid = pg_class.oid
AND emotions.inserted > 'yesterday'
AND moments.inserted BETWEEN 'yesterday' AND 'today'
AND emotions.moment_id = moments.moment_id
GROUP BY relname, emotion
ORDER BY relname, emotion;
That was a bit faster, but still very slow. Here's the EXPLAIN: http://explain.depesz.com/s/ZdF
On Tue, Jan 31, 2012 at 2:53 PM, Alessandro Gagliardi <alessandro@xxxxxxxx> wrote:
I changed the query a bit so the results would not change over the
course of the day to:
WHERE moments.inserted BETWEEN 'yesterday' AND 'today' AND
SELECT relname, emotion, COUNT(feedback_id) FROM pg_class, moments
JOIN emotions USING (moment_id)
moments.tableoid = pg_class.oid
GROUP BY relname, emotion ORDER BY relname, emotion;
