MySQL: Select N rows, but with only unique values in one column

BlaM picture BlaM · Oct 10, 2008 · Viewed 74.3k times · Source

Given this data set:

ID  Name            City            Birthyear
1   Egon Spengler   New York        1957
2   Mac Taylor      New York        1955
3   Sarah Connor    Los Angeles     1959
4   Jean-Luc Picard La Barre        2305
5   Ellen Ripley    Nostromo        2092
6   James T. Kirk   Riverside       2233
7   Henry Jones     Chicago         1899

I need to find the 3 oldest persons, but only one of every city.

If it would just be the three oldest, it would be...

  • Henry Jones / Chicago
  • Mac Taylor / New York
  • Egon Spengler / New York

However since both Egon Spengler and Mac Taylor are located in New York, Egon Spengler would drop out and the next one (Sarah Connor / Los Angeles) would come in instead.

Any elegant solutions?

Update:

Currently a variation of PConroy is the best/fastest solution:

SELECT P.*, COUNT(*) AS ct
   FROM people P
   JOIN (SELECT MIN(Birthyear) AS Birthyear
              FROM people 
              GROUP by City) P2 ON P2.Birthyear = P.Birthyear
   GROUP BY P.City
   ORDER BY P.Birthyear ASC 
   LIMIT 10;

His original query with "IN" is extremly slow with big datasets (aborted after 5 minutes), but moving the subquery to a JOIN will speed it up a lot. It took about 0.15 seconds for approx. 1 mio rows in my test environment. I have an index on "City, Birthyear" and a second one just on "Birthyear".

Note: This is related to...

Answer

ConroyP picture ConroyP · Oct 10, 2008

Probably not the most elegant of solutions, and the performance of IN may suffer on larger tables.

The nested query gets the minimum Birthyear for each city. Only records who have this Birthyear are matched in the outer query. Ordering by age then limiting to 3 results gets you the 3 oldest people who are also the oldest in their city (Egon Spengler drops out..)

SELECT Name, City, Birthyear, COUNT(*) AS ct
FROM table
WHERE Birthyear IN (SELECT MIN(Birthyear)
               FROM table
               GROUP by City)
GROUP BY City
ORDER BY Birthyear DESC LIMIT 3;

+-----------------+-------------+------+----+
| name            | city        | year | ct |
+-----------------+-------------+------+----+
| Henry Jones     | Chicago     | 1899 | 1  |
| Mac Taylor      | New York    | 1955 | 1  |
| Sarah Connor    | Los Angeles | 1959 | 1  |
+-----------------+-------------+------+----+

Edit - added GROUP BY City to outer query, as people with same birth years would return multiple values. Grouping on the outer query ensures that only one result will be returned per city, if more than one person has that minimum Birthyear. The ct column will show if more than one person exists in the city with that Birthyear