November 25, 2014

MySQL Optimizer and Innodb Primary Key

Innodb primary key is special in many senses and I was always wondering how well MySQL is integrated with Innodb to take advantage of these special features.

Lets see which things work and which things do not:

I used the following simple table for tests:

“myisam” is same table created with MyISAM storage engine used to show difference:

MySQL Optimizer correctly knows Innodb tables is clustered by primary key in the sense it would not be faster to do external filesort than to do lookups in primary key order:

MySQL Optimizer is also able to figure out every key also holds primary key value so primary key value can be read from index, making some queries index covered which previously was not: Notice “Using Index” difference

MySQL However is unable to benefit from the fact each index internally has primary key as last column, so key on (a) is effectively key on (a,id) which means MySQL could skip filesort if ordering is done by primary key:

Filesort should be avoided in this case which it is not. I now filed it as a
bug while I do not really think it would be fixed soon

About Peter Zaitsev

Peter managed the High Performance Group within MySQL until 2006, when he founded Percona. Peter has a Master's Degree in Computer Science and is an expert in database kernels, computer hardware, and application scaling.


  1. Roland Volkmann says:

    Hello Peter,

    as long as the bug is not fixed, you can define key a as “KEY a (a,id)” to avoid filesort. I did so in my projects.

    With best regards


  2. peter says:

    Roland yes. The solution you’re mentioning does work and it is a way to get such behavior for non-Innodb tables.
    My point is simply it could be done for Innodb with given index structure but it is not.

    Also quite curious if key (a,id) is used is Innodb is smart enough to notice it already had id in the end so it does not have to be added second time.

  3. grantsucceeded says:

    It is not true that primary key lookup is faster than filesort, or at least should not be true.

    If one is sorting a significant number of rows, sort is faster because it can do sequential IOs then O(logN) sort instead of random IOs.

  4. peter says:

    Why is that ?

    The choice is basically if you access data by primary key and send it or you access data by primary key to sort it and return back to the client ?

    Innodb full table scan will be reading via Innodb key and the data comes sorted already so it will be just overhead to sort it again.

  5. Raj says:

    Will Primary key improve performance ?

  6. Geo says:

    This is now fixed, at least my tests show so

    CREATE TABLE pricesbig_inno (
    id int(10) unsigned NOT NULL,
    url_id int(10) unsigned NOT NULL,
    product_id int(10) unsigned NOT NULL,
    PRIMARY KEY (product_id,id),
    KEY id (id)

    mysql> EXPLAIN SELECT id, product_id FROM pricesbig_inno WHERE id =1515;
    | id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
    | 1 | SIMPLE | pricesbig_inno | ref | id | id | 4 | const | 1 | Using index |
    1 row in set (0.00 sec)

  7. peter says:

    What is fixed ?

    This seems to be similar to my example:

    mysql> EXPLAIN SELECT id FROM innodb WHERE a=3 \G
    *************************** 1. row ***************************
    id: 1
    select_type: SIMPLE
    TABLE: innodb
    type: ref
    possible_keys: a
    KEY: a
    key_len: 5
    ref: const
    rows: 1
    1 row IN SET (0.00 sec)

    Which already had using index.

  8. Cyril Scetbon says:

    It seems to be not the case anymore in 5.1 :

    mysql> EXPLAIN SELECT id FROM innodb WHERE a=3 ORDER BY id \G
    *************************** 1. row ***************************
    id: 1
    select_type: SIMPLE
    table: innodb
    type: ref
    possible_keys: a
    key: a
    key_len: 5
    ref: const
    rows: 3
    Extra: Using where; Using index
    1 row in set (0.51 sec)

  9. peter says:


    Thank you for update. Indeed problems are getting fixed so you have to look at the old posts with care.
    In particular as developers often read this blog it is not a surprise :)

Speak Your Mind