15 Pro-tips for MySQL Users 4Developers - 12 april 2013 Warsaw - Poland 16 

Joshua Thijssen Freelance consultant, developer and trainer @ NoxLogic Founder of the Dutch Web Alliance Development in PHP, Python, Perl, C, Java. Lead developer of Saffire. Blog: http://adayinthelifeof.nl Email: [email protected] Twitter: @jaytaph 

“Pro”-tips for MySQL. Rough estimation: 147 seconds per “tip”. So let’s get going. 

http://distrowatch.com/table.php?distribution=redhat http://distrowatch.com/table.php?distribution=debian 0: Use an up-to-date version RHEL 5.7: 5.0.77 RHEL 6.1: 5.1.52 RHEL 6.4: 5.1.67 Debian (Lenny): 5.0.51a Debian (Squeeze): 5.1.49 Debian (Wheezy) 5.5.30 

0: Use an up-to-date version MariaDB is a very good (maybe better?) option too! 

1: Know how to use explain Not enough time to talk about EXPLAIN today 

mysql> desc varchartest; +----------+--------------+------+-----+---------+-------+ | Field | Type | Null | Key | Default | Extra | +----------+--------------+------+-----+---------+-------+ | id | int(11) | NO | PRI | 0 | | | name | varchar(255) | NO | MUL | NULL | | | utf8name | varchar(255) | NO | MUL | NULL | | +----------+--------------+------+-----+---------+-------+ 3 rows in set (0.01 sec) 1: Know how to use explain 

mysql> EXPLAIN EXTENDED -> SELECT * FROM varchartest WHERE name LIKE 'joshua'; +----+-------------+-------------+-------+---------------+----------+---------+------+------+----------+-------------+ | id | select_type | table | type | possible_keys | key | key_len | ref | rows | filtered | Extra | +----+-------------+-------------+-------+---------------+----------+---------+------+------+----------+-------------+ | 1 | SIMPLE | varchartest | range | idx_name | idx_name | 257 | NULL | 1 | 100.00 | Using where | +----+-------------+-------------+-------+---------------+----------+---------+------+------+----------+-------------+ 1 row in set, 1 warning (0.01 sec) 1: Know how to use explain 

mysql> SHOW WARNINGS\g +-------+------ +--------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------+ | Level | Code | Message | +-------+------ +--------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------+ | Note | 1003 | select `zend`.`varchartest`.`id` AS `id`,`zend`.`varchartest`.`name` AS `name`,`zend`.`varchartest`.`utf8name` AS `utf8name` from `zend`.`varchartest` where (`zend`.`varchartest`.`name` like 'joshua') | +-------+------ +--------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------+ 1: Know how to use explain 

mysql> SET profiling=1; mysql> SELECT * FROM table; mysql> SHOW PROFILE CPU, BLOCK IO; +--------------------------------+----------+----------+------------+--------------+---------------+ | Status | Duration | CPU_user | CPU_system | Block_ops_in | Block_ops_out | +--------------------------------+----------+----------+------------+--------------+---------------+ | starting | 0.000930 | 0.000000 | 0.000000 | 0 | 8 | | checking query cache for query | 0.000547 | 0.000000 | 0.000000 | 0 | 0 | | checking permissions | 0.000045 | 0.000000 | 0.000000 | 0 | 0 | | Opening tables | 0.000142 | 0.000000 | 0.000000 | 0 | 0 | | System lock | 0.000030 | 0.000000 | 0.000000 | 0 | 0 | | Table lock | 0.000045 | 0.000000 | 0.000000 | 0 | 0 | | init | 0.000207 | 0.000000 | 0.000000 | 0 | 0 | | optimizing | 0.000007 | 0.000000 | 0.000000 | 0 | 0 | | statistics | 0.000068 | 0.000000 | 0.000000 | 0 | 0 | | preparing | 0.001393 | 0.004000 | 0.000000 | 0 | 0 | | Creating tmp table | 0.001658 | 0.000000 | 0.000000 | 0 | 8 | | executing | 0.000005 | 0.000000 | 0.000000 | 0 | 0 | | Copying to tmp table | 0.000834 | 0.000000 | 0.004001 | 0 | 8 | | Sorting result | 0.000179 | 0.000000 | 0.000000 | 0 | 0 | | Sending data | 0.000089 | 0.000000 | 0.000000 | 0 | 0 | | end | 0.000022 | 0.000000 | 0.000000 | 0 | 0 | | removing tmp table | 0.000089 | 0.000000 | 0.000000 | 0 | 0 | | end | 0.000005 | 0.000000 | 0.000000 | 0 | 0 | | query end | 0.000004 | 0.000000 | 0.000000 | 0 | 0 | | freeing items | 0.000844 | 0.000000 | 0.000000 | 0 | 0 | | logging slow query | 0.000023 | 0.000000 | 0.000000 | 0 | 0 | | logging slow query | 0.000060 | 0.000000 | 0.000000 | 0 | 8 | | cleaning up | 0.000007 | 0.000000 | 0.000000 | 0 | 0 | +--------------------------------+----------+----------+------------+--------------+---------------+ 1: Know how to use explain 

2: Know the most basic my.cnf settings 

Know the most important ones: key_buffer_size, innodb_buffer_pool_size, sort_buffer_size, max_connections 2: Know the most basic my.cnf settings 

2: Know the most basic my.cnf settings Some settings work on global level, some per connection! Know some quirks: (max_heap_table_size vs tmp_table_size, binlog-do-db, replicate- ignore-db etc) 

http://www.omh.cc/mycnf/ https://github.com/rackerhacker/ MySQLTuner-perl http://www.day32.com/MySQL/ PHPMyAdmin 2: Know the most basic my.cnf settings 

3: Backup on table level mysqldump can dump per database OR by table. Simple scripts to scan/dump tables. Easy restore for single table (or part of table) 

Don’t use “SELECT *” when you only need one or two fields. 4: Don’t select what you don’t need Much more data to be read from disk Much more data will be send over, thus slower (blobs/texts) Cannot use covering indices 

mysql> SHOW FULL COLUMNS FROM `covering`; +--------------+------------------+------------------+------+-----+---------+----------------+ | Field | Type | Collation | Null | Key | Default | Extra | +--------------+------------------+------------------+------+-----+---------+----------------+ | id | int(10) unsigned | NULL | NO | PRI | NULL | auto_increment | | email | varchar(255) | ascii_general_ci | NO | MUL | NULL | | | want_mailing | tinyint(1) | NULL | NO | MUL | NULL | | | extra_info | varchar(255) | ascii_general_ci | NO | MUL | NULL | | +--------------+------------------+------------------+------+-----+---------+----------------+ 3 rows in set (0.00 sec) mysql> SHOW INDEXES FROM `covering`; +----------+------------+------------+--------------+--------------+-----------+-------------+ | Table | Non_unique | Key_name | Seq_in_index | Column_name | Collation | Cardinality | +----------+------------+------------+--------------+--------------+-----------+-------------+ | covering | 0 | PRIMARY | 1 | id | A | 3 | | covering | 1 | idx_email | 1 | email | A | 3 | | covering | 1 | idx_email | 2 | want_mailing | A | 3 | | covering | 1 | idx_email2 | 1 | want_mailing | A | 1 | | covering | 1 | idx_email2 | 2 | email | A | 3 | +----------+------------+------------+--------------+--------------+-----------+-------------+ 5 rows in set (0.01 sec) 4: Don’t select what you don’t need 

mysql> EXPLAIN SELECT want_mailing FROM `covering` WHERE email LIKE '[email protected]'; +----+-------------+----------+-------+---------------+-----------+---------+------+------+--------------------------+ | id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra | +----+-------------+----------+-------+---------------+-----------+---------+------+------+--------------------------+ | 1 | SIMPLE | covering | range | idx_email | idx_email | 257 | NULL | 1 | Using where; Using index | +----+-------------+----------+-------+---------------+-----------+---------+------+------+--------------------------+ 1 row in set (0.00 sec) 4: Don’t select what you don’t need 

5: Use triggers and stored procedures for DB logic 6 triggers per table (insert, update, delete, before and after the mutation) 3rd party tools (phpmyadmin etc) can also use the database without loosing data consistency. Watch out with (phpmyadmin) table dumps! 

6: Don’t use FULLTEXT searches They only work for MyISAM tables. Not compatible with other DB’s. Slow (especially compared to Solr, Sphinx). No extra features (faceted search, spell checking etc). You want to move to Lucene/Solr/ElasticSearch 

Some wildcard searches (%item%) are bad for performance 7: Be careful with %wildcards% MySQL cannot use indexing! Revert your data: search for ‘ln.cigolxon@%’ instead of ‘%@noxlogic.nl’. Use a better solution (solr, sphinx). You probably want it (no, really) 

8: Shard your volatile and non-volatile data 

mysql> SHOW FULL COLUMNS FROM `pages`; +------------+------------------+-------------------+------+-----+-------------------+-------+ | Field | Type | Collation | Null | Key | Default | Extra | +------------+------------------+-------------------+------+-----+-------------------+-------+ | page_id | int(10) unsigned | NULL | NO | PRI | NULL | | | created_dt | timestamp | NULL | NO | | CURRENT_TIMESTAMP | | | creator_id | int(11) | NULL | NO | | NULL | | | title | varchar(100) | latin1_swedish_ci | NO | | NULL | | | contents | text | latin1_swedish_ci | NO | | NULL | | | hit_count | int(11) | NULL | NO | | 0 | | +------------+------------------+-------------------+------+-----+-------------------+-------+ 6 rows in set (0.01 sec) 8: Shard your volatile and non-volatile data 

Remember: an update on a table will invalidate ALL caches referring to that table. UPDATE pages SET hit_count = hit_count + 1; Thus: page table will NEVER be cached. 8: Shard your volatile and non-volatile data 

Define hot data (volatile, changes often) and cold data (static, changes never or infrequently) move to different tables UPDATE page_stats SET hit_count = hit_count + 1; Query cache is happy again Bulk update (hitcount += 100) 8: Shard your volatile and non-volatile data 

MySQL 5.1 and higher has got partitioning (but you should shard anyway). 8: Shard your volatile and non-volatile data 

9: Don’t use large PKs for InnoDB tables InnoDB adds the primary key to EACH index. No primary key given? It uses an internal 6 byte key. 

10: Don’t select count(*) on InnoDB 

InnoDB implements MVCC (multi-version concurrency control). COUNT(*) must be counted and is not fetched from metadata. 10: Don’t select count(*) on InnoDB 

What do you want to COUNT(*)? Just for displaying purposes (there are X amount of pages): do you need the EXACT amount? (guess &| cache) 10: Don’t select count(*) on InnoDB 

11: Don’t rely on the VARCHAR() VARCHAR() is a storage solution, not a processing solution. 

11: Don’t rely on the VARCHAR() 

mysql> SHOW FULL COLUMNS FROM `varchartest`; +----------+--------------+-------------------+------+-----+---------+-------+ | Field | Type | Collation | Null | Key | Default | Extra | +----------+--------------+-------------------+------+-----+---------+-------+ | id | int(11) | NULL | NO | PRI | 0 | | | name | varchar(255) | latin1_swedish_ci | NO | MUL | NULL | | | utf8name | varchar(255) | utf8_general_ci | NO | MUL | NULL | | +----------+--------------+-------------------+------+-----+---------+-------+ 3 rows in set (0.01 sec) mysql> SHOW INDEXES FROM `varchartest`; +-------------+------------+--------------+--------------+-------------+-----------+-------------+ | Table | Non_unique | Key_name | Seq_in_index | Column_name | Collation | Cardinality | +-------------+------------+--------------+--------------+-------------+-----------+-------------+ | varchartest | 0 | PRIMARY | 1 | id | A | 3 | | varchartest | 1 | idx_name | 1 | name | A | 3 | | varchartest | 1 | idx_utf8name | 1 | utf8name | A | 3 | +-------------+------------+--------------+--------------+-------------+-----------+-------------+ 3 rows in set (0.00 sec) 11: Don’t rely on the VARCHAR() 

mysql> EXPLAIN SELECT * FROM `varchartest` WHERE name LIKE 'jthijssen'; +----+-------------+-------------+-------+---------------+----------+---------+------+------+-------------+ | id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra | +----+-------------+-------------+-------+---------------+----------+---------+------+------+-------------+ | 1 | SIMPLE | varchartest | range | idx_name | idx_name | 257 | NULL | 1 | Using where | +----+-------------+-------------+-------+---------------+----------+---------+------+------+-------------+ 1 row in set (0.00 sec) 11: Don’t rely on the VARCHAR() 

UTF-8 isn’t the enemy, but it’s certainly not your friend. Dr Jeckyl and Mr Hyde Solves all your multi-language problems! But gives back performance issues. 12: Don’t trust UTF-8 

mysql> SHOW FULL COLUMNS FROM `varchartest`; +----------+--------------+-------------------+------+-----+---------+-------+ | Field | Type | Collation | Null | Key | Default | Extra | +----------+--------------+-------------------+------+-----+---------+-------+ | id | int(11) | NULL | NO | PRI | 0 | | | name | varchar(255) | latin1_swedish_ci | NO | MUL | NULL | | | utf8name | varchar(255) | utf8_general_ci | NO | MUL | NULL | | +----------+--------------+-------------------+------+-----+---------+-------+ 3 rows in set (0.01 sec) mysql> SHOW INDEXES FROM `varchartest`; +-------------+------------+--------------+--------------+-------------+-----------+-------------+ | Table | Non_unique | Key_name | Seq_in_index | Column_name | Collation | Cardinality | +-------------+------------+--------------+--------------+-------------+-----------+-------------+ | varchartest | 0 | PRIMARY | 1 | id | A | 3 | | varchartest | 1 | idx_name | 1 | name | A | 3 | | varchartest | 1 | idx_utf8name | 1 | utf8name | A | 3 | +-------------+------------+--------------+--------------+-------------+-----------+-------------+ 3 rows in set (0.00 sec) 12: Don’t trust UTF-8 

mysql> EXPLAIN SELECT * FROM `varchartest` WHERE name LIKE 'jthijssen'; +----+-------------+-------------+-------+---------------+----------+---------+------+------+-------------+ | id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra | +----+-------------+-------------+-------+---------------+----------+---------+------+------+-------------+ | 1 | SIMPLE | varchartest | range | idx_name | idx_name | 257 | NULL | 1 | Using where | +----+-------------+-------------+-------+---------------+----------+---------+------+------+-------------+ 1 row in set (0.00 sec) mysql> EXPLAIN SELECT * FROM `varchartest` WHERE utf8name LIKE 'jthijssen'; +----+-------------+-------------+-------+---------------+--------------+---------+------+------+-------------+ | id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra | +----+-------------+-------------+-------+---------------+--------------+---------+------+------+-------------+ | 1 | SIMPLE | varchartest | range | idx_utf8name | idx_utf8name | 767 | NULL | 1 | Using where | +----+-------------+-------------+-------+---------------+--------------+---------+------+------+-------------+ 1 row in set (0.01 sec) 12: Don’t trust UTF-8 

ALL temporary buffers are allocated for worst-case scenario’s. This means a varchar(255) in UTF-8 uses 255*3 + 2 = 767 bytes PER row, even if you have only 1 single char inside. 12: Don’t trust UTF-8 

13: Know your cardinality and selectivity Cardinality: the number of unique entries inside the index. Selectivity: percentage of unique entries. S(I) = cardinality / count * 100% 

13: Know your cardinality and selectivity 

with 10 records: 5/10 * 100% = 50% with 1000 records: 75/1000 * 100 = 7.5% with 10.000 records: 200/10000 * 100% = 2% country_id (max +-200, but effectively +- 50, maybe less) 13: Know your cardinality and selectivity 

A selectivity < 30% ? Full table scan! ANALYZE TABLE frequently. Adding records changes your cardinality and thus selectivity. Develop against a “real” dataset. 13: Know your cardinality and selectivity 

14: Non-deterministic functions do not go well with query caching SELECT .. FROM table ORDER BY RAND() 

SELECT * FROM table WHERE YEAR(created_dt) < YEAR(NOW()); vs SELECT * FROM table WHERE YEAR(created_dt) < ‘2010’; 14: Non-deterministic functions do not go well with query caching 

15: Certify yourself as a DBA and/or DBE. Oracle Certified MySQL Associate Oracle Certified Professional MySQL 5.0 Developer Oracle Certified Professional MySQL 5.0 Database Administrator Oracle Certified Expert, MySQL 5.1 Cluster Database Administrator “Old”, but still, get them all! 

(1) Know how to use explain. (2) Know the most basic my.cnf settings. (3) Backup on table level. (4) Don’t use “SELECT *” when you only need 1 or 2 fields. (5) Use triggers and stored procedures. (6) Don’t use FULLTEXT searches. (7) Wildcard searches (%item%) are bad for performance. (8) Shard your volatile and non-volatile data. (9) Don’t use a large Primary Key for InnoDB tables. (10) Don’t “Select COUNT(*)” on InnoDB. (11) Don’t rely on the VARCHAR(). (12) UTF-8 is not the enemy, but it certainly isn’t your friend. (13) Know your cardinality & selectivity. (14) Non-deterministic functions do not go well with query caching. (15) Certify yourself as a DBA and/or DBE. 

Questions? 

Find me on twitter: @jaytaph Find me for development or training: www.noxlogic.nl Find me on email: [email protected] Find me for blogs: www.adayinthelifeof.nl THANK YOU!