Practical JSON in MySQL 5.7 and Beyond

How people build software
!
"
Practical JSON in MySQL 5.7 and Beyond
IkeWalker
GitHub
Percona Live
April 27, 2017

How people build software!
What this talk is about
2
!

!
What this talk is about
3
• Using JSON in MySQL

What this talk is not about
4
!

!
What this talk is not about
5
• Whether you should use JSON in MySQL
YES! NO!

A brief history of JSON in
MySQL
6
!

!
Timeline: MySQL and JSON
7
1995: MYSQL
2002: JSON
2011: COMMON_SCHEMA
2013: MYSQL JSON UDFS
2015: NATIVE JSON

2002 - 2011
“There’s an app for that”
8
!

!
2002-2011: “There’s an app for that”
9
• Store JSON as text
• Rewrite full string every time
• Parsing happens exclusively in the application layer
• Or write your own stored procedures/functions
BUILD JSON
WRITE TO DB
STORE AS TEXT
READ FROM DB
PARSE JSON

2011 - 2013
“Standard procedures”
10
!

!
2011-2013: “Standard procedures”
11
• Rewrite full string every time
• Some simple parsing can be done with common_schema:
• get_option()
• extract_json_value()

2013 - 2015
“Lab experiments”
13
!

!
2013-2015: “Lab experiments”
14
• Some updates can be done with UDFs
• Much faster parsing supported by UDFs

!
2013-2015: “Lab experiments”
15
mysql> select json_extract(f.event_data,'state') as state,
-> json_extract(f.event_data,'age') as age,
-> json_extract(f.event_data,'gender') as gender,
-> sum(f.event_count) as event_count
-> from json_event_fact f
-> group by state, age, gender;
+-------+----------+--------+-------------+
| state | age | gender | event_count |
+-------+----------+--------+-------------+
| CA | 18-24 | female | 5384 |
| CA | 18-24 | male | 791 |
| CA | 18-24 | null | 25 |
| CA | 25-34 | female | 6731 |
| CA | 25-34 | male | 1292 |
| CA | 25-34 | null | 38 |
| CA | 35-44 | female | 10638 |
| CA | 35-44 | male | 1822 |
| CA | 35-44 | null | 20 |
+-------+----------+--------+-------------+

2015 - present
“Going native”
16
!

!
2015-present: “Going native”
17
• Store JSON as text or JSON datatype
• JSON datatype is binary, with keys sorted
• Fast access to embedded data
• Updates executed via native functions
• Extensive parsing supported by native functions

JSON function examples
18
!

!
JSON_EXTRACT
19
mysql> select json_unquote(json_extract(event_data,'$.country’)) as country,
-> sum(event_count) as events
-> from json_event_fact
-> where d = current_date() - interval 1 day
-> and ad_id = 2
-> group by country;
+---------+--------+
| country | events |
+---------+--------+
| nl | 107954 |
| us | 27373 |
+---------+--------+
2 rows in set (0.00 sec)

!
JSON_SEARCH
20
mysql> select json_search(event_data,'one','android') as json_path
-> having json_path is not null
-> limit 1;
+-----------+
| json_path |
+-----------+
| "$.os" |
+-----------+
1 row in set (0.01 sec)
mysql> select json_search('{"fa":"la","la":["la","la"]}','all','la')G
*************************** 1. row ***************************
json_search('{"fa":"la","la":["la","la"]}','all','la'): ["$.fa", "$.la[0]",
"$.la[1]"]

!
JSON_REPLACE
21
mysql> set @json = cast('{"foo":"bar"}' as json);
Query OK, 0 rows affected (0.00 sec)
mysql> set @json = json_replace(@json, '$.foo', 'UPDATED');
mysql> select @json;
+--------------------+
| @json |
+--------------------+
| {"foo": "UPDATED"} |
+--------------------+

!
JSON_ARRAY
22
mysql> set @json = cast('{"foo":"bar"}' as json);
mysql> set @json = json_replace(@json, '$.foo', json_array('bar', 'car',
'far'));
mysql> select @json;
+--------------------------------+
| @json |
+--------------------------------+
| {"foo": ["bar", "car", "far"]} |
+--------------------------------+

!
JSON_OBJECT
23
mysql> set @json = json_object(
-> 'id','007',
-> 'name','James Bond',
-> 'cars',json_array('Alfa Romeo','Aston Martin','BMW')
-> );
mysql> select @jsonG
*************************** 1. row ***************************
@json: {"id": "007", "cars": ["Alfa Romeo", "Aston Martin", "BMW"], "name":
"James Bond"}

!
JSON COMPARATOR
24
mysql> select cast('"hello"' as json) = 'hello';
1
mysql> select cast(42 as json) = 42;
1
mysql> select cast(false as json) = false;
1
mysql> select cast('{"foo":"bar"}' as json) = cast('{"foo":"bar"}' as json);
1

!
But wait! There’s more!
25
https://dev.mysql.com/doc/refman/5.7/en/json-function-reference.html

Use case #1
Flexible rollups
26
!

!
Flexible rollups
27
• GOAL: Support diﬀerent ﬁelds for multiple customers within a single
dimension

!
Flexible rollups
28
For example, some ads track age and gender and others track country and os:
{"age":"Over 30","gender":"female"}
{"country":"us","os":"android"}

!
Flexible rollups
29
-- try to create rollup with JSON datatype
mysql> create table json_event_fact (
-> d date not null,
-> ad_id int not null,
-> event_data json not null,
-> event_count int not null,
-> primary key (d,ad_id,event_data)
-> );
ERROR 3152 (42000): JSON column 'event_data' cannot be used in key
specification.

!
Flexible rollups
30
-- use text instead
-> d date not null,
-> event_data varchar(750) not null,
-> primary key (d,ad_id,event_data)
-> );

!
Flexible rollups
31
-- generated column hack
-> d date not null,
-> event_data json not null,
-> event_data_text varchar(750) as (cast(event_data as char)) stored,
-> primary key (d,ad_id,event_data_text)
-> );

!
Flexible rollups
32
mysql> select event_data->'$.age' as age,
-> and ad_id = 1
-> group by age;
+------------+---------+
| age | events |
+------------+---------+
| "Over 30" | 810424 |
| "Under 30" | 1205544 |
+------------+---------+

!
Flexible rollups
33
mysql> select json_unquote(event_data->'$.country') as country,
-> and ad_id = 2
-> group by country;
+---------+--------+
| country | events |
+---------+--------+
| nl | 107954 |
| us | 27373 |
+---------+--------+

Use case #2
Conﬁguration data
34
!

!
Conﬁguration data
35
• GOAL: Store extensible conﬁguration data of arbitrary depth for each ad

!
Conﬁguration data
36
create table ad_config_data (
ad_config_data_id int not null primary key,
ad_id int not null,
name varchar(50) not null,
config_data json not null
);

!
Conﬁguration data
37
{
"a56a81eb": {
"type": "AD",
"uuid": "d9e9d8ae-8a33-11e6-97e0-22000b93579c",
"subConfig": {},
"dataSupport": true
},
"a6529578": {
"type": "VIDEO",
"uuid": "e09b40af-8a33-11e6-97e0-22000b93579c",
"subConfig": {
"video_url": "https://www.youtube.com/watch?v=dQw4w9WgXcQ",
"hd": "true”
}
},
"a6caab6e": {
"type": "AD",
"uuid": "e89a3877-8a33-11e6-97e0-22000b93579c",
"subConfig": {}
}
}

!
Conﬁguration data
38
{
"paths": {
"path_3007ea93": ["action_312c40f4"],
"path_30972a80": ["action_3158f2a7", "action_3185a6da",
"action_31aedd9b"],
…
},
"actions": {
"action_312c40f4": {
…
}
}
}

!
Conﬁguration data
39
mysql> select json_extract(config_data,'$.paths.path_30c06190') as sub_paths from
ad_config_data where ad_id = 1 and name = 'actions'G
*************************** 1. row ***************************
sub_paths: ["action_5b5343af", "action_5b8b6c39", "action_5bbb05d6"]
mysql> update ad_config_data
-> set config_data =
json_replace(config_data,'$.paths.path_30c06190',json_array('action_5b5343af',
'action_5bbb05d6'))
-> where ad_id = 1 and name = 'actions';
Query OK, 1 row affected (0.00 sec)
Rows matched: 1 Changed: 1 Warnings: 0
mysql> select json_extract(config_data,'$.paths.path_30c06190') as sub_paths from
ad_config_data where ad_id = 1 and name = 'actions'G
*************************** 1. row ***************************
sub_paths: ["action_5b5343af", "action_5bbb05d6"]

!
Conﬁguration data
40
mysql> select json_extract(config_data,'$.*.type')
-> from ad_config_data
-> where ad_id = 1
-> and name = 'layers';
+--------------------------------------+
| json_extract(config_data,'$.*.type') |
+--------------------------------------+
| ["AD", "VIDEO", "AD"] |
+--------------------------------------+

!
Conﬁguration data
41
mysql> select json_search(config_data,'one','action_312c40f4')
-> from ad_config_data
-> where ad_id = 1
-> and name = 'actions';
+--------------------------------------------------+
| json_search(config_data,'one','action_312c40f4') |
+--------------------------------------------------+
| "$.paths.path_3007ea93[0]" |
+--------------------------------------------------+

Use case #3
EAV antidote
42
!

!
EAV antidote
43
• GOAL: Store Entity-Attribute-Value style data while avoiding the EAV
antipattern

!
EAV antidote
44
• The EAV antipattern is described well by Bill Karwin in his book: “SQL
Antipatterns”
• In Bill’s example the EAV anti-pattern is used to store two types of issues
(bugs and features) in a single shared table.

!
EAV antidote
45
CREATE TABLE Issues (
issue_id SERIAL PRIMARY KEY
);
CREATE TABLE IssueAttributes (
issue_id BIGINT UNSIGNED NOT NULL,
attr_name VARCHAR(100) NOT NULL,
attr_value VARCHAR(100),
PRIMARY KEY (issue_id, attr_name),
FOREIGN KEY (issue_id) REFERENCES Issues(issue_id)
);
ATTRIBUTES STORED AS KEY-VALUE PAIRS

!
EAV antidote
46
issue_id SERIAL PRIMARY KEY,
reported_by BIGINT UNSIGNED NOT NULL,
product_id BIGINT UNSIGNED,
priority VARCHAR(20),
version_resolved VARCHAR(20),
status VARCHAR(20),
issue_type VARCHAR(10), -- BUG or FEATURE
attributes TEXT NOT NULL, -- all dynamic attributes for the row
FOREIGN KEY (reported_by) REFERENCES Accounts(account_id),
FOREIGN KEY (product_id) REFERENCES Products(product_id)
);

!
EAV antidote
47
issue_id SERIAL PRIMARY KEY,
reported_by BIGINT UNSIGNED NOT NULL,
product_id BIGINT UNSIGNED,
priority VARCHAR(20),
version_resolved VARCHAR(20),
status VARCHAR(20),
issue_type VARCHAR(10), -- BUG or FEATURE
attributes JSON NOT NULL, -- all dynamic attributes for the row
severity VARCHAR(20) AS (attributes->"$.severity"), -- only for bugs
version_affected VARCHAR(20) AS (attributes->"$.version_affected"), -- only for bugs
sponsor VARCHAR(50) AS (attributes->"$.sponsor"), -- only for feature requests
FOREIGN KEY (reported_by) REFERENCES Accounts(account_id),
FOREIGN KEY (product_id) REFERENCES Products(product_id)
);

JSON + Generated Columns
Two great techs that tech great together.
48
!

!
JSON + Generated Columns
49
• Allows you to expose one or more JSON ﬁelds as table columns
• Supports indexes
• Choose virtual (not stored) unless the index is Primary Key, FULLTEXT, or
GIS

!
Example #1: Add age column to rollup table
50
mysql> alter table json_event_fact
-> add column age varchar(20) as (event_data->'$.age');
Records: 0 Duplicates: 0 Warnings: 0
mysql> select age,
-> and ad_id = 1
-> group by age;
+------------+---------+
| age | events |
+------------+---------+
| "Over 30" | 810424 |
| "Under 30" | 1205544 |
+------------+---------+

!
Example #2: Add gender column/index to rollup
51
mysql> alter table json_event_fact
-> add column gender varchar(20) as (event_data->'$.gender'),
-> add index (gender);
Records: 0 Duplicates: 0 Warnings: 0
mysql> select sum(event_count) as events
-> where gender = '"female"';
+---------+
| events |
+---------+
| 1081286 |
+---------+

JSON vs TEXT
JSON as text versus JSON datatype
52
!

!
TEXT vs. JSON datatype
53
• Use the JSON datatype in general
• There are a few exceptions. Use text types if you need to:
• Include the column in a primary key
• Store heterogeneous data (some rows are strings, some are JSON)
• Store JSON with depth greater than 100

!
54
JSON data type sorts by keys:
mysql> select cast('{"c":"3","b":
{"2":"2","1":"1"},"a":"1"}' as json) as key_sort_test;
+-------------------------------------------------+
| key_sort_test |
+-------------------------------------------------+
| {"a": "1", "b": {"1": "1", "2": "2"}, "c": "3"} |
+-------------------------------------------------+

!
55
• Use the JSON_VALID() function to test validity of existing column values
before you run ALTER TABLE
• Be aware that the JSON type automatically uses the utf8mb4 character set
• Morgan Tocker has a good blog post on this topic: http://
mysqlserverteam.com/upgrading-json-data-stored-in-text-columns/

Reads versus writes
Read/write balance considerations
56
!

!
Read/write balance considerations
57
• New JSON type is read-optimized
• Every update requires rewriting the entire object
• Performance improvements are in progress for writes:
• http://dev.mysql.com/worklog/task/?id=9141
• http://dev.mysql.com/worklog/task/?id=8985

Disk storage implications
58
!

!
Disk storage implications
59
• The JSON data type’s binary format uses about the same amount of space
as text types.
• Given the nature of JSON data (keys are repeated in every row), JSON data
tends to compress well.

Gotchas
60
!

!
Gotchas
61
Namespace collisions between JSON UDFs and native functions:
JSON_APPEND()
JSON_DEPTH()
JSON_EXTRACT()
JSON_MERGE()
JSON_REMOVE()
JSON_REPLACE()
JSON_SEARCH()
JSON_SET()
JSON_VALID()

!
Gotchas
62
Stricter behavior of native functions:
-- MySQL 5.6 UDF
mysql> select json_extract('','foo');
+------------------------+
| json_extract('','foo') |
+------------------------+
| NULL |
+------------------------+
-- MySQL 5.7 native function
mysql> select json_extract('','$.foo');
ERROR 3141 (22032): Invalid JSON text in argument 1 to function json_extract: "The
document is empty." at position 0.

!
Gotchas
63
Native functions output JSON, not text:
-- MySQL 5.6 UDF
mysql> select json_extract('{"foo":"bar"}','foo');
+-------------------------------------+
| json_extract('{"foo":"bar"}','foo') |
+-------------------------------------+
| bar |
+-------------------------------------+
mysql> select json_extract('{"foo":"bar"}','$.foo');
+---------------------------------------+
| json_extract('{"foo":"bar"}','$.foo') |
+---------------------------------------+
| "bar" |
+---------------------------------------+

!
Gotchas
64
Path diﬀerences between JSON UDFs and native functions:
-- MySQL 5.6 UDF
mysql> select json_extract('{"parent":{"child":"hello"}}','parent','child');
+---------------------------------------------------------------+
| json_extract('{"parent":{"child":"hello"}}','parent','child') |
+---------------------------------------------------------------+
| hello |
+---------------------------------------------------------------+
mysql> select json_extract('{"parent":{"child":"hello"}}','$.parent.child');
+---------------------------------------------------------------+
| json_extract('{"parent":{"child":"hello"}}','$.parent.child') |
+---------------------------------------------------------------+
| "hello" |
+---------------------------------------------------------------+

!
Gotchas
65
• Cannot index JSON columns directly: use generated columns for indexing

!
Gotchas
66
Validating existing JSON values can be diﬃcult if some rows are not valid
JSON and other rows have a depth higher than 100:
mysql> select json_text from old_table where json_valid(json_text) = 0;
ERROR 3157 (22032): The JSON document exceeds the maximum depth.
mysql> select id, json_depth(json_text) from old_table;
ERROR 3141 (22032): Invalid JSON text in argument 1 to function json_depth:
"Invalid value." at position 0.

MySQL 8.0
67
!

!
JSON changes in MySQL 8.0
68
• JSON aggregate functions: http://mysqlserverteam.com/mysql-8-0-labs-json-
aggregation-functions/
• JSON_ARRAYAGG()
• JSON_OBJECTAGG()
• New function to prettify JSON values: https://dev.mysql.com/doc/refman/8.0/
en/json-utility-functions.html
• JSON_PRETTY()
• Storage used/free functions for inspecting native type
• JSON_STORAGE_FREE()
• JSON_STORAGE_SIZE()
• “JSON in place” work underway
• Optimizer piece complete
• InnoDB BLOB refactoring complete

!
Help shape the future!!!
69
• Proposal to change the behavior of JSON_MERGE: http://
mysqlserverteam.com/proposal-to-change-the-behavior-of-json_merge/
• How should JSON_MERGE() deal with overlapping values?
• Current behavior = combine all values in an array
• How should duplicate keys be handled?
• Current behavior = ﬁrst key wins
• Common behavior = last key wins

Thanks!
Questions?
Please rate my talk :)
@iowalker
My team is hiring! http://bit.ly/platform-data
70
!

Practical JSON in MySQL 5.7 and Beyond

Recommended

More Related Content

What's hot (20)

Similar to Practical JSON in MySQL 5.7 and Beyond (20)

Recently uploaded (20)

Practical JSON in MySQL 5.7 and Beyond