Multiple Ranks in one table
Solution 1:
This is a bit tricky. You may want to use variables, such as in the following example:
SELECT (
CASE cust_type
WHEN @curType
THEN @curRow := @curRow + 1
ELSE @curRow := 1 AND @curType := cust_type END
) + 1 AS rank,
cust_type,
cust_name,
revenue
FROM sales,
(SELECT @curRow := 0, @curType := '') r
ORDER BY cust_type DESC, revenue DESC;
The (SELECT @curRow := 0, @curType := '') r
part allows the variable initialization without requiring a separate SET
command.
Test case:
CREATE TABLE sales (cust_type varchar(10), cust_name varchar(10), revenue int);
INSERT INTO sales VALUES ('Top', 'A', 10000);
INSERT INTO sales VALUES ('Top', 'B', 9000);
INSERT INTO sales VALUES ('Top', 'C', 8000);
INSERT INTO sales VALUES ('Bottom', 'X', 5000);
INSERT INTO sales VALUES ('Bottom', 'Y', 6000);
INSERT INTO sales VALUES ('Bottom', 'Z', 7000);
Result:
+------+-----------+-----------+---------+
| rank | cust_type | cust_name | revenue |
+------+-----------+-----------+---------+
| 1 | Top | A | 10000 |
| 2 | Top | B | 9000 |
| 3 | Top | C | 8000 |
| 1 | Bottom | Z | 7000 |
| 2 | Bottom | Y | 6000 |
| 3 | Bottom | X | 5000 |
+------+-----------+-----------+---------+
6 rows in set (0.00 sec)
Another test case:
CREATE TABLE sales (cust_type varchar(10), cust_name varchar(10), revenue int);
INSERT INTO sales VALUES ('Type X', 'A', 7000);
INSERT INTO sales VALUES ('Type X', 'B', 8000);
INSERT INTO sales VALUES ('Type Y', 'C', 5000);
INSERT INTO sales VALUES ('Type Y', 'D', 6000);
INSERT INTO sales VALUES ('Type Y', 'E', 4000);
INSERT INTO sales VALUES ('Type Z', 'F', 4000);
INSERT INTO sales VALUES ('Type Z', 'G', 3000);
Result:
+------+-----------+-----------+---------+
| rank | cust_type | cust_name | revenue |
+------+-----------+-----------+---------+
| 1 | Type Z | F | 4000 |
| 2 | Type Z | G | 3000 |
| 1 | Type Y | D | 6000 |
| 2 | Type Y | C | 5000 |
| 3 | Type Y | E | 4000 |
| 1 | Type X | B | 8000 |
| 2 | Type X | A | 7000 |
+------+-----------+-----------+---------+
7 rows in set (0.00 sec)
You can obviously order the cust_type
in ascending order instead of descending. I used descending just to have Top
before Bottom
in the original test case.
Solution 2:
I found a problem with the solution using CASE
, @curRow
, and @curType
. It depends on the execution plan MySQL uses to process the query. For example, it shows up if you add a join to the query. Then there is no guarantee that the rank is going to be computed correctly.
Making a slight change to the answer:
CREATE TABLE sales (cust_type_id int, cust_name varchar(10), revenue int);
CREATE TABLE cust_type (cust_type_id int, type_name varchar(10));
INSERT INTO cust_type VALUES (1, 'Bottom');
INSERT INTO cust_type VALUES (2, 'Top');
INSERT INTO sales VALUES (2, 'A', 10000);
INSERT INTO sales VALUES (2, 'B', 9000);
INSERT INTO sales VALUES (2, 'C', 8000);
INSERT INTO sales VALUES (1, 'X', 5000);
INSERT INTO sales VALUES (1, 'Y', 6000);
INSERT INTO sales VALUES (1, 'Z', 7000);
If I query only the sales
table I get the rank in the correct order, but if I join to the cust_type
table the rank values are no longer correct
SELECT (
CASE s.cust_type_id
WHEN @curType
THEN @curRow := @curRow + 1
ELSE @curRow := 1 AND @curType := s.cust_type_id END
) AS rank,
t.type_name,
s.cust_name,
s.revenue
FROM sales s,
cust_type t,
(SELECT @curRow := 0, @curType := 0) r
WHERE s.cust_type_id = t.cust_type_id
ORDER BY t.type_name DESC, s.revenue DESC;
Result:
+------+-----------+-----------+---------+
| rank | type_name | cust_name | revenue |
+------+-----------+-----------+---------+
| 1 | Top | A | 10000 |
| 2 | Top | B | 9000 |
| 3 | Top | C | 8000 |
| 3 | Bottom | Z | 7000 |
| 2 | Bottom | Y | 6000 |
| 1 | Bottom | X | 5000 |
+------+-----------+-----------+---------+
MySQL is running the initial query into a temp table and then the ORDER BY
is executing against the temp table after a rank was already computed.
Solution 3:
This is similar to Thomas's answer, but a little simpler:
SELECT (SELECT COUNT(Cust_Type) FROM sales
WHERE Cust_Type = S.Cust_Type AND Revenue >= S.Revenue) AS Rank,
Cust_Type,
Cust_Name,
Revenue
FROM sales AS S
ORDER BY Cust_Type DESC, Rank;
I'm not sure how the performance compares to the Daniel's solution, particularly on very large data sets, or if you have to use complex joins.
Solution 4:
What is not exactly clear is how the items should be ranked (I assumed by Revenue) or whether you are only pulling a certain number of values (e.g. the top 3 and the bottom 3) so I assumed you wanted all values. Given those assumptions,
Select Cust_Name, Cust_Type
, (Select Count(*)
From Table As T1
Where T1.Revenue > T.Revenue ) + 1 As Rank
From Table As T
Where Cust_Type = 'Top'
Union All
Select Cust_Name, Cust_Type
, (Select Count(*)
From Table As T1
Where T1.Revenue < T.Revenue ) + 1 As Rank
From Table As T
Where Cust_Type = 'Bottom'
If you were trying to do this in a single non-union query you could do:
Select Cust_Name, Cust_Type
, Case Z.Cust_Type
When 'Top' Then Z.TopRank
Else Z.BottomRank
End As Rank
From (
Select Cust_Name, Cust_Type
, (Select Count(*)
From Table As T1
Where T1.Revenue > T.Revenue ) + 1 As TopRank
, (Select Count(*)
From Table As T1
Where T1.Revenue < T.Revenue ) + 1 As BottomRank
From Table As T
) As Z