Introduction to MySQL Query Tuning for Dev[Op]s

Introduction to MySQL Query Tuning
for Dev[Op]s
October 4, 2019
Sveta Smirnova

•Basics
•When MySQL Uses Indexes
•Diagnostics
EXPLAIN: estimation on how Optimizer works
Real Numbers: Inside Storage Engine
Real Numbers: Inside the Server
•How to Aﬀect Query Plans
Table of Contents
2

• MySQL Support engineer
• Author of
• MySQL Troubleshooting
• JSON UDF functions
• FILTER clause for MySQL
• Speaker
• Percona Live, OOW, Fosdem,
DevConf, HighLoad...
Sveta Smirnova
3

Query Tuning
Concurrency
Environment
Troubleshooting Workﬂow
5

Query Tuning
Concurrency
Environment
Troubleshooting Workﬂow: This Session
6

$system = System::factory()
->setName($this->form->get(Field::NAME))
->setDescription(
$this->form->get(Field::DESCRIPTION)
);
DAO::system()->take($system);
The Query
7

cursor = conn.cursor()
q = ’’’UPDATE ‘foo‘ SET my_date=NOW(),
subject = %s,
msg = %s,
address = %s,
updated_at = NOW()
WHERE id=%s
’’’
cursor.execute(q, [
remote_resp.get(’subject’),
remote_resp.get(’msg’),
remote_resp.get(’address’),
my_id
])
The Query
7

SELECT dept_name, title, gender,
min(salary) AS mins, max(salary) AS maxs
FROM employees
JOIN salaries USING(emp_no)
JOIN titles USING(emp_no)
JOIN dept_emp USING(emp_no)
JOIN departments USING(dept_no)
JOIN dept_manager USING(dept_no)
WHERE dept_manager.to_date = ’9999-01-01’
GROUP BY dept_name, title, gender
ORDER BY gender, maxs DESC;
The Query
7

• PMM QAN
Allways Tune Raw Query
8

• PMM QAN
• Slow Query Log
8

• PMM QAN
• Slow Query Log
• Application log
8

• PMM QAN
• Slow Query Log
• Application log
• ...
8

• Mind you data!
•
75,000,000 rows
• (INT, INT)
75,000,000 * (4 + 4) = 600,000,000 bytes = 572 MB
• (INT, INT, DATETIME, VARCHAR(255),
VARCHAR(255))
75,000,000 * (4 + 4 + 8 + 256 + 256) = 39,600,000,000 bytes = 37 G
• 39,600,000,000 / 600,000,000 = 66
Slow is relative
9

• Mind you data!
• Mind use case
• Popular website
•
Admin interface
• Weekly cron job
Slow is relative
9

• Mind you data!
• Mind use case
• Mind location
• Server, used by multiple connections
•
Dedicated for OLAP queries
Slow is relative
9

• MySQL performs a job to execute a query
Why Query can be Slow
10

• In worst case scenario it will do a full table scan
•
CREATE INDEX
• ANALYZE TABLE ... UPDATE HISTOGRAM ON
10

• In worst case scenario it will do a full table scan
•
CREATE INDEX
• ANALYZE TABLE ... UPDATE HISTOGRAM ON
• Incorrect index can be used
10

Query sent
Connection Pool: Authentication, Caches; SQL interface; Parser
Optimizer
Storage engines
Hardware
Query Execution Workﬂow
11

SELECT name FROM users SELECT name FROM users WHERE id=12
1 2 5 6 7 9 12 16 18 21 22 23 24 25
Full Table Scan
12

SELECT name FROM users SELECT name FROM users WHERE id=12
1 2 5 6 7 9 12 16 18 21 22 23 24 25
After Index Added
13

d001
d003
d008
d009
d003******
d009******
d008******
d009******
d001******
d003******
d009******
d008******
d009******
d001******
d008******
d008******
d001******
• B-Tree (Mostly)
• Fractal Tree
• LSM Tree
• R-Tree (Spatial)
• Hash (Memory SE)
• Engine-dependent
MySQL Indexes
14

• Single column
CREATE INDEX index name ON
the table(the column)
• Multiple columns
CREATE INDEX index name ON
the table(column1, column2)
How to Create an Index
15

• Single column
ALTER TABLE table name ADD INDEX
[index name] (the column)
• Multiple columns
ALTER TABLE table name ADD INDEX
[index name] (column1, column2)
How to Create an Index
15

•
WHERE the column = a value
• WHERE the column IN(value1, value2,
value3)
• WHERE the column LIKE ’value%’
• WHERE the column LIKE ’%value’
Conditions
17

•
WHERE left part = value1 AND
right part = value2
• WHERE left part = value1 OR
right part = value2
• WHERE right part = value1 AND
left part = value2
•
WHERE right part = value1 OR left part =
value2
Conditions
17

•
table1 JOIN table2 ON table1.column1
= table2.column2
Joins
18

•
table1 JOIN table2 ON table1.column1
= table2.column2
• Same as FROM table1, table2 WHERE
table1.column1 = table2.column2
Joins
18

•
GROUP BY the column
• GROUP BY left part, right part
• GROUP BY right part, left part
• GROUP BY the index, another index
GROUP BY
19

•
ORDER BY the column
• ORDER BY left part, right part
• ORDER BY right part, left part
• ORDER BY the index, another index
ORDER BY
20

5.7 ORDER BY left part DESC, right part ASC
8.0 ORDER BY left part DESC, right part
ASC
• left part must be descending
• right part must be ascending
• the index(left part DESC, right part ASC)
ORDER BY
20

• Deterministic, built-in
•
Return same value for the same argument
• WHERE the column = FLOOR(123.45)
Expressions
21

•
• Non-deterministic
•
Return diﬀerent values for diﬀerent invocations
• WHERE the column = RAND() ∗ 100
Expressions
21

•
• Non-deterministic
•
Return diﬀerent values for diﬀerent invocations
• WHERE the column = RAND() ∗ 100
•
Stored functions and UDFs
• Indexes are not used
Use generated column indexes
Expressions
21

Diagnostics
EXPLAIN: estimation on how Optimizer works

• EXPLAIN
•
Estimates what happens during query execution
5.6- EXTENDED
5.6- PARTITIONS
5.6+ FORMAT=JSON
8.0+ FORMAT=TREE
How to Find how MySQL Uses Indexes
24

• EXPLAIN
•
Estimates what happens during query execution
5.6- EXTENDED
5.6- PARTITIONS
5.6+ FORMAT=JSON
8.0+ FORMAT=TREE
• INFORMATION SCHEMA.OPTIMIZER TRACE
•
Real data, collected after query was executed
• Advanced topic
How to Find how MySQL Uses Indexes
24

mysql> explain select * from t1G
*************************** 1. row ***************************
...
rows: 12
Extra: NULL
mysql> explain select * from t1 where f2=12G
*************************** 1. row ***************************
...
key: NULL
...
rows: 12
Extra: Using where
Same number of examined rows for both queries
Eﬀect of Indexes: Before
25

mysql> alter table t1 add index(f2);
Query OK, 12 rows affected (0.07 sec)
Records: 12 Duplicates: 0 Warnings: 0
mysql> explain select * from t1 where f2=12G
*************************** 1. row ***************************
...
key: f2
key_len: 5
ref: const
rows: 1
Extra: NULL
1 row in set (0.00 sec)
Much more eﬀective!
Only 1 row examined
Eﬀect of Indexes: After
26

mysql> explain extended select * from t1 join t2 where t1.int_key=1;
+----+-------------+-------+-------+---------------+---------+---------+-------+------+------+---------
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | f... | Extra
+----+-------------+-------+-------+---------------+---------+---------+-------+------+------+---------
| 1 | SIMPLE | t1 | ref | int_key,ik | int_key | 5 | const | 4 | 100. | NULL
| 1 | SIMPLE | t2 | index | NULL | pk | 9 | NULL | 6 | 100. | Using in
Using join buffer |
(Block Nested Loop) |
+----+-------------+-------+-------+---------------+---------+---------+-------+------+------+---------
2 rows in set, 1 warning (0.00 sec)
Note (Code 1003): /* select#1 */ select ‘test‘.‘t1‘.‘pk‘ AS ‘pk‘,‘test‘.‘t1‘.‘int_key‘ AS ‘int_key‘,‘te
AS ‘pk‘,‘test‘.‘t2‘.‘int_key‘ AS ‘int_key‘ from ‘test‘.‘t1‘ join ‘test‘.‘t2‘ where (‘test‘.‘t1‘.‘int_ke
Number of select
Select type
Tables, for which information is printed
How data is accessed
Possible keys
Key, which was actually used
Length of the key
Which columns were compared with the index
Number of examined rows
% of ﬁltered rows
rows x ﬁltered / 100 — number of row
which will be joined with another table
Additional informatio
Table, for which information is printed
Product of rows here: how many rows in all tables will be accessed
For this example estimated value is 4*6 = 24
Actual (optimized) query as executed by MySQL Server
EXPLAIN: overview
27

mysql> explain extended select * from t1 join t2 where...
+----+-------------+-------+-------+***
| id | select_type | table | type |***
+----+-------------+-------+-------+***
| 1 | SIMPLE | t1 | ref |***
| 1 | SIMPLE | t2 | index |***
+----+-------------+-------+-------+***
SIMPLE;PRIMARY;UNION;DEPENDENT UNION;UNION RESULT;
SUBQUERY;DEPENDENT SUBQUERY;DERIVED;MATERIALIZED
system
const
eq ref
ref
fulltext
ref or null
index merge
unique subquery
index subquery
range
index
ALL
EXPLAIN in Details
28

***+---------------+---------+---------+-------+***
***| possible_keys | key | key_len | ref |***
***+---------------+---------+---------+-------+***
***| int_key,ik | int_key | 5 | const |***
***| NULL | pk | 9 | NULL |***
***+---------------+---------+---------+-------+***
Keys, which can be used for resolving the query
Only one key was actually used
Actual length of the key (Important for multiple-column keys)
Which columns were compared with the index
Constant
Numeric in our case
Index used
to resolve rows
EXPLAIN in Details: keys
29

***+-------+----------+----------------------------------------------------+
***| rows | filtered | Extra |
***+-------+----------+----------------------------------------------------+
***| 4 | 100.00 | NULL |
***| 6 | 100.00 | Using index; Using join buffer (Block Nested Loop) |
***+-------+----------+----------------------------------------------------+
Number of rows accessed
% of rows ﬁltered
Additional information:
how query is resolved
Using ﬁlesort
Using temporary
etc.
4X6
=
24
All rows used
EXPLAIN in Details: rows
30

mysql> explain select count(*) from employees where hire_date > ’1995-01-01’
********************** 1. row **********************
id: 1
select_type: SIMPLE
table: employees
type: ALL
possible_keys: NULL
key: NULL
key_len: NULL
ref: NULL
rows: 300157
Extra: Using where
All rows in the table examined
Worst plan ever!
EXPLAIN Type by Example: ALL
31

• We need to add index to table employees ﬁrst
mysql> alter table employees add index(hire_date);
Query OK, 0 rows affected (3.48 sec)
Records: 0 Duplicates: 0 Warnings: 0
EXPLAIN Type by Example: range
32

mysql> explain select count(*) from employees where hire_date>’1995-01-01’G
********************** 1. row **********************
id: 1
select_type: SIMPLE
table: employees
type: range
possible_keys: hire_date
key: hire_date
key_len: 3
ref: NULL
rows: 68654
Extra: Using where; Using index
Only rows from given range used
Compare with ALL:
300157/68654 = 4.3720
4 times less rows examined!
EXPLAIN Type by Example: range
33

• Consists of two or more columns
Combined Indexes
34

• Consists of two or more columns
• Only leftmost part used
mysql> alter table City add key
-> comb(CountryCode, District, Population),
-> drop key CountryCode;
Combined Indexes
34

mysql> explain select * from City where CountryCode = ’USA’G
********************** 1. row ******************
table: City
type: ref
possible_keys: comb
key: comb
key_len: 3
ref: const
rows: 273
Uses ﬁrst ﬁeld from the comb key
Combined Indexes: example 1
35

mysql> explain select * from City where
-> District = ’California’ and population > 10000G
********************** 1. row ******************
table: City
type: ALL
possible_keys: NULL
key: NULL
key_len: NULL
ref: NULL
rows: 3868
Can’t use combined index:
not a leftmost part
Does not have the CountryCode
in the where clause
= can’t use comb index
Combined Indexes: example 2
36

• Key len = total size (in bytes)
• Index
• comb(CountryCode, District, Population)
Explain: Fields:
key: comb CountryCode char(3)
key len: 3 District char(20)
Population int(11)
3 ->Char(3) ->First ﬁeld is used
Combined Indexes: key len
37

mysql> explain select count(*) from titles where title=’Senior Engineer’G
********************** 1. row **********************
id: 1
select_type: SIMPLE
table: titles
type: index
possible_keys: NULL
key: emp_no
key_len: 4
ref: NULL
rows: 444033
No row in the table was accessed to resolve the query!
Only index used
Still all records in the index were scanned
EXPLAIN Type by Example: index
38

•
Covered index = cover all ﬁelds in the query
select name from City
where CountryCode = ’USA’ and District = ’Alaska’ and population > 10000
mysql> alter table City add key
-> cov1(CountryCode, District, population, name);
Uses all ﬁelds in the query in particular order
1. Where part 2. Group By/Order (not used now) 3. Select part
Covered Indexes
39

mysql> explain select name from City where CountryCode = ’USA’
-> and District = ’Alaska’ and population > 10000G
*************************** 1. row ***********
table: City
type: range
possible_keys: cov1
key: cov1
key_len: 27
ref: NULL
rows: 1
Covered index is used
MySQL will only use index
Will not go to the data ﬁle
EXPLAIN by Example: Covered Indexes
40

Diagnostics
Real Numbers: Inside Storage Engine

• EXPLAIN is optimistic
mysql> explain select * from ol
-> where thread_id=10432 and site_id != 9939 order by id limit 3G
*************************** 1. row ***************************
id: 1 | ref: NULL
select_type: SIMPLE | rows: 33
table: ol | filtered: 8.07
partitions: NULL | Extra: Using where
type: index
possible_keys: thread_id
key: PRIMARY
key_len: 4
1 row in set, 1 warning (0,00 sec)
Handler * Status Variables
42

Diagnostics
Real Numbers: Inside the Server

•
SHOW [FULL] PROCESSLIST
•
INFORMATION SCHEMA.PROCESSLIST
•
performance schema.THREADS
PROCESSLIST
44

•
•
•
• Your ﬁrst alert in case of performance issue
PROCESSLIST
44

•
•
•
• Your ﬁrst alert in case of performance issue
• Shows all queries, running at the moment
PROCESSLIST
44

• Can be seen in PROCESSLIST
mysql> show processlistG
************************ 1. row ************************
Id: 7
User: root
Host: localhost:48799
db: employees
Command: Query
Time: 2
State: Sending data
Info: select count(*) from employees join titles using(emp_no)
where title=’Senior Engineer’
...
Execution Stages
45

• Can be seen in PROCESSLIST
• Very useful when you need to answer on question:
”What is my server doing now?”
Execution Stages
45

•
PERFORMANCE SCHEMA.EVENTS STAGES *
mysql> select eshl.event_name, substr(sql_text, 1, 15) as ‘sql‘,
-> eshl.timer_wait/1000000000000 w_s from events_stages_history_long
-> eshl join events_statements_history_long esthl on
-> (eshl.nesting_event_id = esthl.event_id) where
-> esthl.current_schema=’employees’ and sql_text like
-> ’select count(*) from employees%’ order by eshl.timer_start asc;
+--------------------------------+-----------------+--------+
| event_name | sql | w_s |
+--------------------------------+-----------------+--------+
| stage/sql/starting | select count(*) | 0.0002 |
| stage/sql/checking permissions | select count(*) | 0.0000 |
...
Execution Stages
45

• Status variables
mysql> flush status;
Query OK, 0 rows affected (0,01 sec)
mysql> select count(*) from employees join titles using(emp_no)
-> where title=’Senior Engineer’;
+----------+
| count(*) |
+----------+
| 97750 |
+----------+
1 row in set (5,44 sec)
Temporary tables and other job
46

• Status variables
mysql> select * from performance_schema.session_status
-> where variable_name in (’Created_tmp_tables’,
-> ’Created_tmp_disk_tables’, ’Select_full_join’,
-> ’Select_full_range_join’, ’Select_range’,
-> ’Select_range_check’, ’Select_scan’, ’Sort_merge_passes’,
-> ’Sort_range’, ’Sort_rows’, ’Sort_scan’) and variable_value > 0;
+------------------------+----------------+
| VARIABLE_NAME | VARIABLE_VALUE |
+------------------------+----------------+
| Select_scan | 2 |
+------------------------+----------------+
46

• PERFORMANCE SCHEMA.EVENTS STATEMENTS *
mysql> select * from performance_schema.events_statements_history_long
-> where sql_text like ’select count(*) from employees join %’G
*************************** 1. row ****************************
...
ROWS_SENT: 1 SELECT_RANGE_CHECK: 0
ROWS_EXAMINED: 541058 SELECT_SCAN: 1
CREATED_TMP_DISK_TABLES: 0 SORT_MERGE_PASSES: 0
CREATED_TMP_TABLES: 0 SORT_RANGE: 0
SELECT_FULL_JOIN: 0 SORT_ROWS: 0
SELECT_FULL_RANGE_JOIN: 0 SORT_SCAN: 0
SELECT_RANGE: 0 NO_INDEX_USED: 0
46

•
sys.statement analysis
mysql> select * from statement_analysis where query like ’SELECT COUNT
-> ( * ) FROM ‘emplo%’ and db=’employees’G
*************************** 1. row ***************************
query: SELECT COUNT ( * ) FROM ‘emplo ... ‘emp_no‘ ) WHE...
db: employees max_latency: 5.59 s
full_scan: avg_latency: 5.41 s
exec_count: 7 lock_latency: 2.24 ms
err_count: 0 rows_sent: 7
warn_count: 0 rows_sent_avg: 1
total_latency: 37.89 s rows_examined: 3787406
46

•
sys.statement analysis
rows_examined_avg: 541058
rows_affected: 0
rows_affected_avg: 0
tmp_tables: 0
tmp_disk_tables: 0
rows_sorted: 0
sort_merge_passes: 0
digest: 4086bc3dc6510a1d9c8f2fe1f59f0943
first_seen: 2016-04-14 15:19:19
last_seen: 2016-04-14 16:13:14
46

• Index statistics
• Histogram statistics
• Optimizer switches
• Bugs in optimizer
What has Eﬀect on Query Optimizer Plans?
48

• Collected by storage engine
Index Statistics
49

• Collected by storage engine
• Used by Optimizer
Index Statistics
49

• Can be examined by SHOW INDEX command
mysql> show index from sbtest1;
+---------+----------+-------------+-------------+
| Table | Key_name | Column_name | Cardinality |
+---------+----------+-------------+-------------+
| sbtest1 | k_1 | k | 49142 |
+---------+----------+-------------+-------------+
mysql> select count(distinct id), count(distinct k) from sbtest1;
+--------------------+-------------------+
| count(distinct id) | count(distinct k) |
+--------------------+-------------------+
| 100000 | 17598 |
+--------------------+-------------------+
Index Statistics
49

• Can be updated
• ANALYZE TABLE
•
If does not help: rebuild table
OPTIMIZE TABLE
ALTER TABLE ENGINE=INNODB; ANALYZE TABLE
Index Statistics
49

• Since version 8.0
Histogram Statistics
50

• Collected and used by the Optimizer
50

• Can be examined in Information Schema
mysql> select HISTOGRAM from information_schema.column_statistics
-> where table_name=’example’G
*************************** 1. row ***************************
HISTOGRAM: {"buckets": [[1, 0.6], [2, 0.8], [3, 1.0]],
"data-type": "int", "null-values": 0.0, "collation-id": 8,
"last-updated": "2018-11-07 09:07:19.791470",
"sampling-rate": 1.0, "histogram-type": "singleton",
"number-of-buckets-specified": 3}
50

• Can be examined in Information Schema
More details
50

mysql> select @@optimizer_switchG
*************************** 1. row ***************************
@@optimizer_switch: index_merge=on,index_merge_union=on,
index_merge_sort_union=on,index_merge_intersection=on,
engine_condition_pushdown=on,index_condition_pushdown=on,
mrr=on,mrr_cost_based=on,
block_nested_loop=on,batched_key_access=off,
materialization=on,semijoin=on,loosescan=on,firstmatch=on,
duplicateweedout=on,subquery_materialization_cost_based=on,
use_index_extensions=on,condition_fanout_filter=on,derived_merge=on
Optimizer Switches
51

• Turn ON and OFF particular optimization
Optimizer Switches
51

• Can be not helpful
• Especially for queries, tuned for previous versions
Optimizer Switches
51

• Work with them as with any other option
• Turn OFF and try
SET optimizer_switch = ’use_index_extensions=off’;
SELECT ...
EXPLAIN SELECT ...
Optimizer Switches
51

• Work with them as with any other option
•
If helps implement in queries
SELECT /*+ SEMIJOIN(FIRSTMATCH, LOOSESCAN) */ * FROM t1 ...;
SELECT /*+ BKA(t1) NO_BKA(t2) */ * FROM t1 INNER JOIN t2 WHERE ...;
Optimizer Switches
51

• Optimizer choses wrong index for no reason
Bugs in Optimizer
52

• Statistics is up to date
• Histograms are not usable
Bugs in Optimizer
52

• Solution
•
Use index hints
FORCE INDEX
IGNORE INDEX
Bugs in Optimizer
52

• Solution
• On every upgrade
• Remove index hints
• Test if query improved
•
You must do it even for minor version upgrades!
Bugs in Optimizer
52

• EXPLAIN is essential for query tuning
• Real job is done by storage engine
• Index statistics aﬀect query execution plan
• All index hints, optimizer hints and other
workarounds must be validated on each
upgrade
Summary
53

EXPLAIN Syntax
EXPLAIN FORMAT=JSON is Cool! series
Troubleshooting Performance add-ons
Optimizer Satistics aka Histograms
Optimizer Hints
Tracing the Optimizer
More information
54

Alexander Rubin for combined and covered index examples
Special thanks
55

www.slideshare.net/SvetaSmirnova
twitter.com/svetsmirnova
github.com/svetasmirnova
Thank you!
56

Percona’s open source database experts are true
superheroes, improving database performance for
customers across the globe.
Percona’s open source database experts are true
superheroes, improving database performance for
customers across the globe.
Discover what it means to have a Percona career
with the smartest people in the database
performance industries, solving the most
challenging problems our customers come across.
We’re Hiring!
57

Introduction to MySQL Query Tuning for Dev[Op]s

Recommended

More Related Content

What's hot (20)

Similar to Introduction to MySQL Query Tuning for Dev[Op]s (20)

More from Sveta Smirnova (14)

Recently uploaded (20)

Introduction to MySQL Query Tuning for Dev[Op]s