Mysql compound index cardinality is not same on all fields

Question

I have created an compound index on mysql, here is the command that i use: create index deliver_aid_sid_rcnt_idx on DELIVER_SM(AID,STATUSID,RETRY_CNT) USING BTREE; The weird is that i got different cardinality value for each field. is that normal? i have created other compound index as well and the cardinality value for each field is the same.

+--------------+-----------------+-------------+--------------------------+-------------+
| TABLE_SCHEMA | TABLE_NAME      | COLUMN_NAME | INDEX_NAME               | CARDINALITY |
+--------------+-----------------+-------------+--------------------------+-------------+
| prddb        | DELIVER_SM      | AID         | deliver_aid_sid_rcnt_idx |          28 |
| prddb        | DELIVER_SM      | STATUSID    | deliver_aid_sid_rcnt_idx |         286 |
| prddb        | DELIVER_SM      | RETRY_CNT   | deliver_aid_sid_rcnt_idx |         286 |
+--------------+-----------------+-------------+--------------------------+-------------+
 

Answer 1

Index cardinality is not a number of distinct values of columns, it is rather a number of nodes in the b-tree index.

Consider below example:

CREATE TABLE abc(   a int, b int, c int );

set @x = 0;
INSERT INTO abc( a, b, c ) 
SELECT (@x:=@x+1),
       round( @x / 10 ),
       round( @x / 100 )
FROM information_schema.columns
LIMIT 421;

CREATE INDEX ix1 ON abc( a, b, c );
CREATE INDEX ix2 ON abc( c, b, a );

ANALYZE TABLE abc;

and queries that show cardinality of indexes:

SELECT COUNT( distinct a) a,
       COUNT( distinct b) b,
       COUNT( distinct c) c,
       COUNT( * )
FROM abc;

SELECT table_name, index_name, column_name, cardinality
FROM INFORMATION_SCHEMA.STATISTICS
WHERE table_name = 'abc' AND index_name = 'ix1';

SELECT table_name, index_name, column_name, cardinality
FROM INFORMATION_SCHEMA.STATISTICS
WHERE table_name = 'abc' AND index_name = 'ix2';

Look at this demo to see results: http://www.sqlfiddle.com/#!2/b5987/1

The table has 421 rows.
Column a has 421 distinct values.
Column b has 43 distinct values.
Column c has 5 distinct values.

I am a poor drawer so I don't attach drawings of these b-tree indexes here :)
But I hope you can imagine a picture of a b-tree index in your head, like in this link: http://docs.oracle.com/cd/E11882_01/server.112/e25789/indexiot.htm
(BTW I recommend you to study this material, it's oracle related, not MySql, but it is great explanation how indexes work and how they are organized).

For ix1 ON abc( a, b, c ) index MySql shows following cardinalites:
a --> 407
b --> 407
c --> 407

Remember that cardinality numbers are not exact values, but rather estimates.

Here a is a leading column in the index (this is the column with the highest number of distinct values), and because of this it creates a high number of top level nodes in the index. Remaining columns (their values) are also stored in (or maybe "under) these top-level index nodes.

Hovever for ix2 ON abc( c, b, a ) estimate values of cardinality are different:
c --> 9
b --> 101
a --> 407

In this case, where c is the leading column of the index, MySql "thinks" (estimates) that the index has 9 "top level" nodes, values of b take 101 nodes "below c", and a takes 407 nodes in the index.