Reputation: 2014
I have a table with a multi-column primary key (city/state/date) and many more columns of data. I'm looking to get the latest data for each city/state. How do I do that cleanly/efficiently? Right now I can do this by doing a first query to get the list of all the rows I'm trying to fetch, followed by a second query with a massive WHERE clause:
SELECT state, city, max(date) from data GROUP BY city, state;
+-------+---------------------+------------+
| state | city | MAX(date) |
+-------+---------------------+------------+
| CA | San Francisco | 2013-09-01 |
| CA | Los Angeles | 2013-08-01 |
| NY | New York | 2013-10-01 |
| ... | ... (many rows) ... | ... |
+-------+---------------------+------------+
SELECT * FROM data WHERE
(state = "CA" AND city = "San Francisco" AND date='2013-09-01') OR
(state = "CA" AND city = "Los Angeles" AND date='2013-08-01') OR
(state = "NY" AND city = "New York" AND date='2013-10-01') OR
...
This is really ugly and inefficient, and if the first query returns a lot of rows my second query might be too long. Clearly if I have a single-column primary key I could use a subselect with IN(), but that's not really possible here. Any suggestions?
UPDATE: I tried Bill's suggestion with a subselect, but it's not using any keys and is taking forever. If I restrict the subselect to only return 5 rows it returns in 0.64s. If I let it return all 73 city/state combinations, it takes a very long time (query still running).
EXPLAIN SELECT * FROM data WHERE (city, state, date) IN (SELECT state, city, MAX(date) FROM data GROUP BY city, state)
+----+--------------------+-------+-------+---------------+---------+---------+------+-------+-------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+--------------------+-------+-------+---------------+---------+---------+------+-------+-------------+
| 1 | PRIMARY | data | ALL | NULL | NULL | NULL | NULL | 13342 | Using where |
| 2 | DEPENDENT SUBQUERY | data | index | NULL | PRIMARY | 57 | NULL | 8058 | Using index |
+----+--------------------+-------+-------+---------------+---------+---------+------+-------+-------------+
Upvotes: 10
Views: 8926
Reputation: 1531
I think this should do the trick for you:
select
*
from
data t1
natural join
(
select
city,
state,
max(date) as date
from
data
group by
city,
state
) t2;
Upvotes: 6
Reputation: 562368
MySQL supports tuple comparisons:
SELECT * FROM data WHERE
(state, city, date) IN (
('CA', 'San Francisco', '2013-09-01'),
('CA', 'Los Angeles', '2013-08-01'),
('NY', 'New York', '2013-10-01'));
Upvotes: 5