问题
I am creating a summary table that sums up all events in a given day.
INSERT INTO graph_6(
day,
event_type,
(SELECT COUNT(*) FROM event e
WHERE event_type = e.event_type
AND creation_time::DATE = sq.day)
FROM event_type
CROSS JOIN
(SELECT generate_series(
(SELECT '2014-01-01'::DATE),
(SELECT '2014-01-02'::DATE),
'1 day') as day) sq;
The creation_time
column is indexed:
CREATE INDEX event_creation_time_date_idx ON event USING BTREE(creation_time);
However, the query runs a pretty long time even when only querying two days of data with a handful of events (January 1-2 2014).
The EXPLAIN
on the query is pretty grim - it runs a sequential scan on the event
table, not utilizing the index at all:
-> Seq Scan on event e_1 (cost=0.00..12557.39 rows=531 width=38)
Filter: ... AND ((creation_time)::date = (generate_series(($12)::timestamp with time zone, ($13)::timestamp with time zone, '1 day'::interval))))
I assume this is because we compare a casted value - creation_time::DATE
, not creation_time
. I have tried indexing the cast:
CREATE INDEX event_creation_time_date_idx ON event USING BTREE(creation_time::DATE);
But got an error:
ERROR: syntax error at or near "::"
Is there a way to utilize PostgreSQL indices on a timezone column casted to DATE?
回答1:
An expression in an index declaration should be enclosed in additional brackets, try:
CREATE INDEX event_creation_time_date_idx ON event ((creation_time::DATE));
来源:https://stackoverflow.com/questions/34724403/postgresql-create-an-index-on-timestampdate