Reputation: 4774
I can't believe what I see with a simple select with an ORDER BY clause. Here is my query and the faulty result:
SELECT date_valid, id_variable FROM myTable
WHERE id_stn='78224' AND date_valid BETWEEN '2014-07-03 09:00:00'
AND '2014-07-03 21:00:00' AND id_variable IN (11012,12004)
ORDER BY date_valid ;
date_valid | id_variable
---------------------+-------------
2014-07-03 09:00:00 | 11012
2014-07-03 15:00:00 | 11012
2014-07-03 21:00:00 | 11012
2014-07-03 09:00:00 | 12004
2014-07-03 15:00:00 | 12004
2014-07-03 21:00:00 | 12004
As you see, the sorting seems to be done on id_variable instead of date_valid. To get the expected result, I have to create a new field that Postgresql can't optimize or to give a timestamp range of more than 1 day:
SELECT date_valid,id_variable FROM myTable
WHERE id_stn='78224' AND date_valid BETWEEN '2014-07-03 09:00:00'
AND '2014-07-03 21:00:00' AND id_variable IN (11012,12004)
ORDER BY date_valid + '0 hours'::INTERVAL;
date_valid | id_variable
---------------------+-------------
2014-07-03 09:00:00 | 11012
2014-07-03 09:00:00 | 12004
2014-07-03 15:00:00 | 11012
2014-07-03 15:00:00 | 12004
2014-07-03 21:00:00 | 11012
2014-07-03 21:00:00 | 12004
Here is a partial table definition, it's partitionned on date_valid for each month:
Column | Type
---------------+-----------------------------
id_obs | bigint
date_valid | timestamp without time zone
id_variable | integer
id_stn | character varying(50)
Indexes:
"myTable_pkey" PRIMARY KEY, btree (id_obs)
"myTable_ukey" UNIQUE CONSTRAINT, btree (date_valid, id_variable, lat, lon)
Check constraints:
"myTable_date_valid_check" CHECK (date_valid >= '2014-07-01 00:00:00'::timestamp without time zone AND date_valid < '2014-08-01 00:00:00'::timestamp without time zone)
Triggers:
myTable_before_update BEFORE UPDATE ON myTable_201407 FOR EACH ROW EXECUTE PROCEDURE obs_update()
Inherits: myTable_parent
Has OIDs: no
It seems like a bug that Postgresql doesn't sort on hours if the result is on the same day. It must be an optimizer issue as I don't have this problem if I sort on another timestamp field which is not indexed. The result is the same (unsorted) if I specify ::TIMESTAMP after each date string, or if I enclose the select in another on with: SELECT * FROM (SELECT ...) x ORDER BY DATE_VALID. I have the same problem with other tables with similar structure.
This is the EXPLAIN result with Postgresql 9.2.8:
Result (cost=0.02..62864.86 rows=10 width=87)
-> Merge Append (cost=0.02..62864.86 rows=10 width=87)
Sort Key: myTable.date_valid
-> Sort (cost=0.01..0.02 rows=1 width=220)
Sort Key: myTable.date_valid
-> Seq Scan on myTable (cost=0.00..0.00 rows=1 width=220)
Filter: ((date_valid >= '2014-07-03 09:00:00'::timestamp without time zone) AND (date_valid <= '2014-07-03 21:00:00'::timestamp without time zone) AND (id_variable = ANY ('{11012,12004}'::integer[])) AND ((id
_stn)::text = '78224'::text))
-> Index Scan using myTable_201407_ukey on myTable_201407 myTable (cost=0.00..62864.71 rows=9 width=72)
Index Cond: ((date_valid >= '2014-07-03 09:00:00'::timestamp without time zone) AND (date_valid <= '2014-07-03 21:00:00'::timestamp without time zone) AND (id_variable = ANY ('{11012,12004}'::integer[])))
Filter: ((id_stn)::text = '78224'::text)
Upvotes: 1
Views: 204
Reputation: 125214
Probably this error fixed in 9.2.1
Fix possible incorrect sorting of output from queries involving WHERE indexed_column IN (list_of_values)
http://www.postgresql.org/docs/9.2/static/release-9-2-1.html
9.2 is already at 9.2.8
Upvotes: 1