PostgreSQL JSON类型常用操作

PostgreSQL JSON类型常用操作

说明

根据RFC 7159[1]中的说明,JSON 数据类型是用来存储 JSON(JavaScript Object Notation) 数据的。这种数据也可以被存储为text,但是 JSON 数据类型的 优势在于能强制要求每个被存储的值符合 JSON 规则。也有很多 JSON 相关的函数和操作符可以用于存储在这些数据类型中的数据

PostgreSQL 提供存储JSON数据的两种类型:json 和 jsonb。

一个JSON数值可以是一个简单值(数字、字符串、true/null/false),数组,对象。下列都是合法的JSON表达式:

-- 简单标量/简单值
-- 简单值可以是数字、带引号的字符串、true、false或者null
SELECT '5'::json;

-- 零个或者更多个元素的数组(元素类型可以不同)
SELECT '[1, 2, "foo", null]'::json;

-- 含有键/值对的对象
-- 注意对象的键必须总是带引号的字符串
SELECT '{"bar": "baz", "balance": 7.77, "active": false}'::json;

-- 数组和对象可以任意嵌套
SELECT '{"foo": [true, "bar"], "tags": {"a": 1, "b": null}}'::json;

以上的JSON类型都可以写成JSONB类型的表达式,例如:

-- 简单标量/简单值,转化为jsonb类型
SELECT '5'::jsonb;

当一个 JSON 值被输入并且接着不做任何附加处理就输出时, json会输出和输入完全相同的文本,而jsonb 则不会保留语义上没有意义的细节(例如空格)。例如,注意下面的不同:

SELECT '{"bar": "baz", "balance": 7.77, "active":false}'::json;
                      json                       
-------------------------------------------------
 {"bar": "baz", "balance": 7.77, "active":false}
(1 row)

SELECT '{"bar": "baz", "balance": 7.77, "active":false}'::jsonb;
                      jsonb                       
--------------------------------------------------
 {"bar": "baz", "active": false, "balance": 7.77}
(1 row)

值得一提的一种语义上无意义的细节是,在jsonb中数据会被按照底层 numeric类型的行为来打印。实际上,这意味着用E记号 输入的数字被打印出来时就不会有该记号,例如:

SELECT '{"reading": 1.230e-5}'::json, '{"reading": 1.230e-5}'::jsonb;
         json          |          jsonb          
-----------------------+-------------------------
 {"reading": 1.230e-5} | {"reading": 0.00001230}
(1 row)

JSON 基本类型和相应的PostgreSQL类型

JSON 基本类型PostgreSQL类型注释
stringtext不允许\u0000,如果数据库编码不是 UTF8,非 ASCII Unicode 转义也是这样
numbernumeric不允许NaN 和 infinity值
booleanboolean只接受小写true和false拼写
null(无)SQL NULL是一个不同的概念

json 和 jsonb区别

json 和 jsonb数据类型接受几乎完全相同的值集合作为输入。

对比项jsonjsonb
特点json数据类型存储输入文本的精准拷贝,处理函数必须在每 次执行时必须重新解析该数据。jsonb数据被存储在一种分解好的二进制格式中,因为需要做附加的转换,它在输入时要稍慢一些。但是 jsonb在处理时要快很多,因为不需要重新解析。
字符处理json类型存储的是输入文本的准确拷贝,存储时会空格和JSON 对象内部的键的顺序。如果一个值中的 JSON 对象包含同一个键超过一次,所有的键/值对都会被保留(处理函数会把最后的值当作有效值)。jsonb不保留空格、不保留对象键的顺序并且不保留重复的对象键。如果在输入中指定了重复的键,只有最后一个值会被保留。
效率json类型存储快,使用慢(写入快,读取慢)jsonb类型存储稍慢,使用较快(写入慢,读取快)
索引支持不支持索引支持索引

示例:

postgres=# select '{"name":"zhangsan","age":"12","name":"lisi"}'::json as json;
                     json
----------------------------------------------
 {"name":"zhangsan","age":"12","name":"lisi"}
(1 row)

postgres=# select '{"name":"zhangsan","age":"12","name":"lisi"}'::jsonb as json;
             json
-------------------------------
 {"age": "12", "name": "lisi"}
(1 row)

创建测试数据

drop table if exists orders;
CREATE TABLE orders (
   ID serial NOT NULL PRIMARY KEY,
   info json NOT NULL
);
INSERT INTO orders (info) VALUES
    ('{ "customer": "John Doe", "items": {"product": "Beer","qty": 6}}'),
    ('{ "customer": "Lily Bush", "items": {"product": "Diaper","qty": 24}}'),
    ('{ "customer": "Josh William", "items": {"product": "Toy Car","qty": 1}}'),
    ('{ "customer": "Mary Clark", "items": {"product": "Toy Train","qty": 2}}'),
	('[{ "customer": "John Steven", "items": {"product": "Toy Car","qty": 5}},{ "customer": "Tom Hark", "items": {"product": "Diaper","qty": 7}}]'),
	('{"customer":[{"name":"Jose Manuel","country":"Germany"},{"name":"John Seo","country":"Australia"}],"items":{"product":"Beer","qty":"13"}}');

内容层级显示如下

postgres=# select id,jsonb_pretty(info::jsonb) from orders;
 id |            jsonb_pretty            
----+------------------------------------
  1 | {                                 +
    |     "items": {                    +
    |         "qty": 6,                 +
    |         "product": "Beer"         +
    |     },                            +
    |     "customer": "John Doe"        +
    | }
  2 | {                                 +
    |     "items": {                    +
    |         "qty": 24,                +
    |         "product": "Diaper"       +
    |     },                            +
    |     "customer": "Lily Bush"       +
    | }
  3 | {                                 +
    |     "items": {                    +
    |         "qty": 1,                 +
    |         "product": "Toy Car"      +
    |     },                            +
    |     "customer": "Josh William"    +
    | }
  4 | {                                 +
    |     "items": {                    +
    |         "qty": 2,                 +
    |         "product": "Toy Train"    +
    |     },                            +
    |     "customer": "Mary Clark"      +
    | }
  5 | [                                 +
    |     {                             +
    |         "items": {                +
    |             "qty": 5,             +
    |             "product": "Toy Car"  +
    |         },                        +
    |         "customer": "John Steven" +
    |     },                            +
    |     {                             +
    |         "items": {                +
    |             "qty": 7,             +
    |             "product": "Diaper"   +
    |         },                        +
    |         "customer": "Tom Hark"    +
    |     }                             +
    | ]
  6 | {                                 +
    |     "items": {                    +
    |         "qty": "13",              +
    |         "product": "Beer"         +
    |     },                            +
    |     "customer": [                 +
    |         {                         +
    |             "name": "Jose Manuel",+
    |             "country": "Germany"  +
    |         },                        +
    |         {                         +
    |             "name": "John Seo",   +
    |             "country": "Australia"+
    |         }                         +
    |     ]                             +
    | }
(6 rows)

操作符

操作符右操作数类型返回类型描述例子例子结果
->intjson or jsonb获得 JSON 数组元素(索引从 0 开始,负整数从末尾开始计)'[{"a":"foo"},{"b":"bar"},{"c":"baz"}]'::json->2{"c":"baz"}
->textjson or jsonb通过键获得 JSON 对象域'{"a": {"b":"foo"}}'::json->'a'{"b":"foo"}
->>inttexttext形式获得 JSON 数组元素'[1,2,3]'::json->>23
->>texttexttext形式获得 JSON 对象域'{"a":1,"b":2}'::json->>'b'2
#>text[]json or jsonb获取在指定路径的 JSON 对象'{"a": {"b":{"c": "foo"}}}'::json#>'{a,b}'{"c": "foo"}
#>>text[]texttext形式获取在指定路径的 JSON 对象'{"a":[1,2,3],"b":[4,5,6]}'::json#>>'{a,2}'3

额外的jsonb操作符

操作符右操作数类型描述例子
@>jsonb左边的 JSON 值是否在顶层包含右边的 JSON 路径/值项?'{"a":1, "b":2}'::jsonb @> '{"b":2}'::jsonb
<@jsonb左边的 JSON 路径/值项是否被包含在右边的 JSON 值的顶层?'{"b":2}'::jsonb <@ '{"a":1, "b":2}'::jsonb
?text键/元素字符串是否存在于 JSON 值的顶层?'{"a":1, "b":2}'::jsonb ? 'b'
`?`text[]这些数组字符串中的任何一个是否做为顶层键存在?
?&text[]是否所有这些数组字符串都作为顶层键存在?'["a", "b"]'::jsonb ?& array['a', 'b']
``jsonb
-text从左操作数删除键/值对或者string元素。键/值对基于它们的键值来匹配。'{"a": "b"}'::jsonb - 'a'
-text[]从左操作数中删除多个键/值对或者string元素。键/值对基于它们的键值来匹配。'{"a": "b", "c": "d"}'::jsonb - '{a,c}'::text[]
-integer删除具有指定索引(负值表示倒数)的数组元素。如果 顶层容器不是数组则抛出一个错误。'["a", "b"]'::jsonb - 1
#-text[]删除具有指定路径的域或者元素(对于 JSON 数组,负值 表示倒数)'["a", {"b":1}]'::jsonb #- '{1,b}'
@?jsonpathJSON路径是否返回指定的JSON值的任何项目?'{"a":[1,2,3,4,5]}'::jsonb @? '$.a[*] ? (@ > 2)'
@@jsonpath返回指定的JSON路径谓词检查结果。只考虑结果的第一项。 如果结果不是布尔值,那么返回 null'{"a":[1,2,3,4,5]}'::jsonb @@ '$.a[*] > 2'

注意:

  • ||操作符将其每一个操作数的顶层的元素串接起来。它不会递归 操作。例如,如果两个操作数都是具有公共域名称的对象,结果中的域值将只是来自右手操作数的值。
  • @?@@@操作符会抑制以下错误:缺乏对象字段或数组元素、意外的JSON项类型和数字错误。当搜索不同结构的JSON文档集合时,这种行为可能会有帮助。

增删改查

查询

下面查询使用->操作符,查询json中所有顾客作为键:

postgres=# SELECT info -> 'customer' AS customer FROM orders where id <=4;
    customer
----------------
 "John Doe"
 "Lily Bush"
 "Josh William"
 "Mary Clark"
(4 rows)

下面使用->>操作获取所有顾客作为文本:

postgres=# SELECT info ->> 'customer' AS customer FROM orders where id <=4;
   customer
--------------
 John Doe
 Lily Bush
 Josh William
 Mary Clark
(4 rows)

->操作返回json对象,我们可以链式方式继续使用->>返回特定节点。举例,下面语句返回所有购买的商品:

postgres=# SELECT info -> 'items' ->> 'product' as product FROM orders where id <=4;
  product
-----------
 Beer
 Diaper
 Toy Car
 Toy Train
(4 rows)

首先使用info->'item’返回json对象。然后使用info->‘item’->>'product’返回所有产品文本值。

where子句

我们能在where子句中使用json操作符过滤数据行。举例,查找买了Diaper的记录:

SELECT info ->> 'customer' AS customer FROM orders
WHERE info -> 'items' ->> 'product' = 'Diaper';
 customer
-----------
 Lily Bush
(1 row)

下面查询谁一次买了2个商品,语句如下:

SELECT
   info ->> 'customer' AS customer,
   info -> 'items' ->> 'product' AS product
FROM
   orders
WHERE
   CAST (info -> 'items' ->> 'qty' AS INTEGER) = 2;			-- 使用cast转换qty字段值为integer类型,然后和2进行比较

  customer  |  product
------------+-----------
 Mary Clark | Toy Train
(1 row)

以上两个案例都是where子句中的精确匹配,也可以模糊匹配

-- 查询customer的值中包含john的记录
postgres=# select * from orders where info #>> '{customer}' ~* 'john';
 id |                               info
----+------------------------------------------------------------------
  1 | { "customer": "John Doe", "items": {"product": "Beer","qty": 6}}
(1 row)

postgres=# select * from orders where info ->> 'customer' ~* 'john';
 id |                               info
----+------------------------------------------------------------------
  1 | { "customer": "John Doe", "items": {"product": "Beer","qty": 6}}
(1 row)

根据键名查询(函数实现)

-- 查询表某个键对应的值
postgres=# select json_object_field_text(info,'customer') from orders where id <=4;
 json_object_field_text
------------------------
 John Doe
 Lily Bush
 Josh William
 Mary Clark
(5 rows)

-- 查询customer中包含 lily的记录
postgres=# select * from orders where json_object_field_text(info,'customer') ~* 'lily';
 id |                                 info
----+----------------------------------------------------------------------
  2 | { "customer": "Lily Bush", "items": {"product": "Diaper","qty": 24}}
(1 row)

-- 嵌套里面的查询
select * from orders where json_object_field_text(info,'items')::json ->> 'product' ~* 'per';
 id |                                 info
----+----------------------------------------------------------------------
  2 | { "customer": "Lily Bush", "items": {"product": "Diaper","qty": 24}}
(1 row)

根据键名(属性)查询

-- 查询有 items 属性的记录
postgres=# select * from orders where info::jsonb ? 'items' and id <=4;
 id |                                  info
----+-------------------------------------------------------------------------
  1 | { "customer": "John Doe", "items": {"product": "Beer","qty": 6}}
  2 | { "customer": "Lily Bush", "items": {"product": "Diaper","qty": 24}}
  3 | { "customer": "Josh William", "items": {"product": "Toy Car","qty": 1}}
  4 | { "customer": "Mary Clark", "items": {"product": "Toy Train","qty": 2}}
(4 rows)

-- 查询有 items 属性的记录,并且items数据不为空的数据
select * from orders where info ->> 'items' is not null;

-- 查询有 customer属性或items的记录
select * from orders where info::jsonb ?| array['customer','items'];
-- 查询既有 customer属性又有items的记录
select * from orders where info::jsonb ?& array['customer','items'];

根据键值查询记录

-- 查询customer为'John Doe'购买的产品名和具体数量( @> 操作符表示:左侧顶层是否包含右侧)
select info -> 'items' ->> 'product' as product,info -> 'items' ->> 'qty' as qty from orders
where info::jsonb @> '{"customer":"John Doe"}';
 product | qty
---------+-----
 Beer    | 6
(1 row)

-- 查询 customer中包含'John Doe'的记录数
select * from orders where info::jsonb -> 'customer' ? 'John Doe';
 id |                               info
----+------------------------------------------------------------------
  1 | { "customer": "John Doe", "items": {"product": "Beer","qty": 6}}
(1 row)

-- 查询 product中包含'Toy Car'的记录
select * from orders where info::jsonb -> 'items' -> 'product'  ? 'Toy Car';
 id |                                  info
----+-------------------------------------------------------------------------
  3 | { "customer": "Josh William", "items": {"product": "Toy Car","qty": 1}}
(1 row)
order by
-- 按照商品购买数量(qty)的值降序
select * from orders where id <=4
order by (info -> 'items' ->> 'qty')::int desc;

 id |                                  info
----+-------------------------------------------------------------------------
  2 | { "customer": "Lily Bush", "items": {"product": "Diaper","qty": 24}}
  1 | { "customer": "John Doe", "items": {"product": "Beer","qty": 6}}
  4 | { "customer": "Mary Clark", "items": {"product": "Toy Train","qty": 2}}
  3 | { "customer": "Josh William", "items": {"product": "Toy Car","qty": 1}}
(4 rows)

-- 排序+分页
select * from orders where id <=4
order by (info -> 'items' ->> 'qty')::int desc limit 2 offset 1;
 id |                                  info
----+-------------------------------------------------------------------------
  1 | { "customer": "John Doe", "items": {"product": "Beer","qty": 6}}
  4 | { "customer": "Mary Clark", "items": {"product": "Toy Train","qty": 2}}
(2 rows)

-- 按照顾客(customer)的值排序
select * from orders where id <=4
order by info ->> 'customer';
 id |                                  info
----+-------------------------------------------------------------------------
  1 | { "customer": "John Doe", "items": {"product": "Beer","qty": 6}}
  3 | { "customer": "Josh William", "items": {"product": "Toy Car","qty": 1}}
  2 | { "customer": "Lily Bush", "items": {"product": "Diaper","qty": 24}}
  4 | { "customer": "Mary Clark", "items": {"product": "Toy Train","qty": 2}}
(4 rows)

-- 按照顾客(customer)的值排序降序,再按照商品购买数量(qty)的值升序
select * from orders where id <=4
order by info ->> 'customer' desc,(info -> 'items' ->> 'qty')::int asc;
 id |                                  info
----+-------------------------------------------------------------------------
  4 | { "customer": "Mary Clark", "items": {"product": "Toy Train","qty": 2}}
  2 | { "customer": "Lily Bush", "items": {"product": "Diaper","qty": 24}}
  3 | { "customer": "Josh William", "items": {"product": "Toy Car","qty": 1}}
  1 | { "customer": "John Doe", "items": {"product": "Beer","qty": 6}}
(4 rows)

json 数据的聚集函数

我们能对json数据使用聚集函数,如min,max,average,sum等。举例,下面语句返回最小数量,最大数量、平均数量以及总数量。

SELECT
    MIN (CAST (info -> 'items' ->> 'qty' AS INTEGER)),
    MAX (CAST (info -> 'items' ->> 'qty' AS INTEGER)),
    SUM (CAST (info -> 'items' ->> 'qty' AS INTEGER)),
    AVG (CAST (info -> 'items' ->> 'qty' AS INTEGER))
FROM orders where id <=4;

 min | max | sum |        avg
-----+-----+-----+--------------------
   1 |  24 |  33 | 8.2500000000000000
(1 row)
多表Join

创建表插入数据

create table employees(info json);
insert into employees values
('{"employee_id":101,"name":"Steven","hire_date":"1990-01-01","department_id":"10"}'),
('{"employee_id":102,"name":"Bruce","hire_date":"1993-01-01","department_id":"20"}'),
('{"employee_id":103,"name":"Nancy","hire_date":"1989-01-01","department_id":"30"}');

create table departments(info json);
insert into departments values
('{"department_id":10,"department_name":"Administration"}'),
('{"department_id":20,"department_name":"Marketing"}'),
('{"department_id":30,"department_name":"Purchasing"}');

关联employees和departments查询员工的姓名和部门

-- 查询employee_id 为101的员工的姓名和部门
select t1.info ->> 'name' as emp_name,t2.info ->> 'department_name' as dep_name
from employees t1 join departments t2
on t1.info ->> 'department_id' = t2.info ->> 'department_id'
where (t1.info ->> 'employee_id')::int = 101;

 emp_name |    dep_name
----------+----------------
 Steven   | Administration
(1 row)

增加JSON字段

1、jsonb键/值追加可通过||操作符

-- jsonb类型可以直接使用 || 操作符
update orders set info = info::jsonb||'{"remark":"The most important customer!"}'::jsonb
where (info -> 'items' ->> 'qty')::int = (select max((info -> 'items' ->> 'qty')::int) from orders);

2、jsonb_set函数

update orders set info = jsonb_set(info::jsonb,'{remark}':'"The most important customer!"')
where (info -> 'items' ->> 'qty')::int = (select max((info -> 'items' ->> 'qty')::int) from orders);

-- 更新前表信息
postgres=# select * from orders where id <=4;
 id |                                  info
----+-------------------------------------------------------------------------
  1 | { "customer": "John Doe", "items": {"product": "Beer","qty": 6}}
  2 | { "customer": "Lily Bush", "items": {"product": "Diaper","qty": 24}}
  3 | { "customer": "Josh William", "items": {"product": "Toy Car","qty": 1}}
  4 | { "customer": "Mary Clark", "items": {"product": "Toy Train","qty": 2}}
(4 rows)

-- 更新后表信息
postgres=# select * from orders where id <=4;
 id |                                                      info
----+----------------------------------------------------------------------------------------------------------------
  1 | { "customer": "John Doe", "items": {"product": "Beer","qty": 6}}
  3 | { "customer": "Josh William", "items": {"product": "Toy Car","qty": 1}}
  4 | { "customer": "Mary Clark", "items": {"product": "Toy Train","qty": 2}}
  2 | {"items": {"qty": 24, "product": "Diaper"}, "remark": "The most important customer!", "customer": "Lily Bush"}
(4 rows)

删除JSON字段

1、操作符-

update orders set info = (info::jsonb - 'remark')::json
where (info -> 'items' ->> 'qty')::int = (select max((info -> 'items' ->> 'qty')::int) from orders);

postgres=# select * from orders where id <=4;
 id |                                  info
----+-------------------------------------------------------------------------
  1 | { "customer": "John Doe", "items": {"product": "Beer","qty": 6}}
  3 | { "customer": "Josh William", "items": {"product": "Toy Car","qty": 1}}
  4 | { "customer": "Mary Clark", "items": {"product": "Toy Train","qty": 2}}
  2 | {"items": {"qty": 24, "product": "Diaper"}, "customer": "Lily Bush"}
(4 rows)

注意:操作符 - 用于从jsonb对象中删除键,如果类型为json需要转成jsonb才能使用,然后处理完后再转回json格式。

2、操作符#-

-- 根据键名删除
postgres=# select * from orders where id=1;
 id |                               info
----+------------------------------------------------------------------
  1 | { "customer": "John Doe", "items": {"product": "Beer","qty": 6}}
(1 row)
-- 删除id为1记录的customer键
postgres=# update orders set info=info::jsonb #- '{customer}' where id=1;
UPDATE 1
postgres=# select * from orders where id=1;
 id |                   info
----+------------------------------------------
  1 | {"items": {"qty": 6, "product": "Beer"}}
(1 row)


-- 根据键名删除嵌套的内层
postgres=# select * from orders where id=2;
 id |                                 info
----+----------------------------------------------------------------------
  2 | { "customer": "Lily Bush", "items": {"product": "Diaper","qty": 24}}
(1 row)
postgres=# update orders set info=info::jsonb #- '{items,product}' where id=2;
UPDATE 1
postgres=# select * from orders where id=2;
 id |                      info
----+-------------------------------------------------
  2 | {"items": {"qty": 24}, "customer": "Lily Bush"}
(1 row)


-- 根据元素位置删除
postgres=# SELECT '["red","green","blue"]'::jsonb - 0;
     ?column?
-------------------
 ["green", "blue"]
(1 row)
-- 删除嵌套aliases中的位置为1的键/值
postgres=# SELECT '{"name": "James", "aliases": ["Jamie","The Jamester","J Man"]}'::jsonb #- '{aliases,1}'::text[];
                     ?column?
--------------------------------------------------
 {"name": "James", "aliases": ["Jamie", "J Man"]}
(1 row)

修改json字段值

1、||操作符,||操作符可以连接json键,也可覆盖重复的键值

-- 修改外层值
update orders set info = info::jsonb || '{"customer":"Tom"}'::jsonb where (info ->> 'customer') = 'Josh William';

2、jsonb_set函数

-- 修改外层值
UPDATE orders
SET info = jsonb_set(info::jsonb,'{customer}','"Tom"')
where id = 1;

-- 修改内层值
UPDATE orders
SET info = jsonb_set(
    info::jsonb,
    '{items,qty}',
    '10')
WHERE (info -> 'items' ->> 'qty')::int = 1;

JSON类型数组常见操作

本次测试使用的JSON数组数据

postgres=# select id,jsonb_pretty(info::jsonb) from orders where id>=5;
 id |            jsonb_pretty            
----+------------------------------------
  5 | [                                 +
    |     {                             +
    |         "items": {                +
    |             "qty": 5,             +
    |             "product": "Toy Car"  +
    |         },                        +
    |         "customer": "John Steven" +
    |     },                            +
    |     {                             +
    |         "items": {                +
    |             "qty": 7,             +
    |             "product": "Diaper"   +
    |         },                        +
    |         "customer": "Tom Hark"    +
    |     }                             +
    | ]
  6 | {                                 +
    |     "items": {                    +
    |         "qty": "13",              +
    |         "product": "Beer"         +
    |     },                            +
    |     "customer": [                 +
    |         {                         +
    |             "name": "Jose Manuel",+
    |             "country": "Germany"  +
    |         },                        +
    |         {                         +
    |             "name": "John Seo",   +
    |             "country": "Australia"+
    |         }                         +
    |     ]                             +
    | }
(2 rows)

查询数组的长度

postgres=# select jsonb_array_length(info::jsonb) from orders where id=5;
 jsonb_array_length 
--------------------
                  2
(1 row)

postgres=# select jsonb_array_length(info::jsonb -> 'customer') from orders where id=6;
 jsonb_array_length 
--------------------
                  2
(1 row)

查询数组的指定元素(按照位置编号)

-- 查询顶层数组中的第二个元素
postgres=# select info::jsonb ->> 1 from orders where id=5;
                              ?column?                              
--------------------------------------------------------------------
 {"items": {"qty": 7, "product": "Diaper"}, "customer": "Tom Hark"}
(1 row)

-- 查询内层数组中的第一个元素
postgres=# select info::jsonb -> 'customer' ->> 0 from orders where id=6;
                   ?column?                    
-----------------------------------------------
 {"name": "Jose Manuel", "country": "Germany"}
(1 row)

将数组拆分为JSON对象

postgres=# select json_array_elements(info) from orders where id=5;
                          json_array_elements                           
------------------------------------------------------------------------
 { "customer": "John Steven", "items": {"product": "Toy Car","qty": 5}}
 { "customer": "Tom Hark", "items": {"product": "Diaper","qty": 7}}
(2 rows)

postgres=# select json_array_elements(info -> 'customer') from orders where id=6;
            json_array_elements             
--------------------------------------------
 {"name":"Jose Manuel","country":"Germany"}
 {"name":"John Seo","country":"Australia"}
(2 rows)

在以上的基础上,再根据键名获取值

-- 顶层
postgres=# select json_array_elements(info) #> '{customer}' as customer from orders where id=5;
   customer    
---------------
 "John Steven"
 "Tom Hark"
(2 rows)

postgres=# select json_array_elements(info) #>> '{customer}' as customer from orders where id=5;
  customer   
-------------
 John Steven
 Tom Hark
(2 rows)

-- 去重
select distinct customer||'' from (select json_array_elements(info) #>> '{customer}' as customer from orders where id=5) tmp;
                                   

-- 内层
postgres=# select json_array_elements(info -> 'customer') #> '{name}' as customer from orders where id=6;
   customer    
---------------
 "Jose Manuel"
 "John Seo"
(2 rows)

postgres=# select json_array_elements(info -> 'customer') #>> '{name}' as customer from orders where id=6;
  customer   
-------------
 Jose Manuel
 John Seo
(2 rows)

PostgreSQL 常用JSON 函数

JSON创建函数

函数描述例子结果
to_json (anyelement)返回该值作为一个合法的JSON对象。数组和组合会被递归处理并且转换成数组和对象。如果输入包含一个从该类型到JSON的造型,会使用该cast函数来执行转换,否则将会产生一个JSON标量值。对于任何非数字、布尔值或空值的标量类型,会使用其文本表示,并且加上适当的引号和转义让它变成一个合法的JSON字符串。to_json ('Fred said "Hi."'::text)"Fred said \"Hi.\""
array_to_json (anyarray [, pretty_bool])返回该数组为一个JSON数组。一个多维数组会变成一个JSON数组的数组。说明如果pretty_bool为true,在第一维元素之间会增加换行。array_to_json ('{{1,5},{99,100}}'::int[])[[1,5],[99,100]]
row_to_json (record [, pretty_bool])返回该行为一个JSON对象。说明如果pretty_bool为true,在第一级别元素之间会增加换行。row_to_json (row(1,'foo')){"f1":1,"f2":"foo"}
row_to_json(record [, pretty_bool])

把行作为一个 JSON 对象返回。如果*pretty_bool*为真,将在第1层元素之间增加换行。

此函数常用来生成json测试数据,比如将一个普通表转换成json类型表:

create table test_t(id int,name varchar(100),age int,set varchar(10));
insert into test_t values
(1,'zhangsan',18,'male'),
(2,'lisi',19,'female'),
(3,'wangwu',18,'male');

postgres=# select * from test_t;
 id |   name   | age |  set
----+----------+-----+--------
  1 | zhangsan |  18 | male
  2 | lisi     |  19 | female
  3 | wangwu   |  18 | male
(3 rows)

postgres=# select row_to_json(test_t) from test_t;
                   row_to_json
--------------------------------------------------
 {"id":1,"name":"zhangsan","age":18,"set":"male"}
 {"id":2,"name":"lisi","age":19,"set":"female"}
 {"id":3,"name":"wangwu","age":18,"set":"male"}
(3 rows)

JSON处理函数

函数返回类型描述例子例子结果
json_each(json)set of key text, value json set of key text, value jsonb把最外层的JSON对象展开成键/值对的集合。select * from json_each('{"a":"foo", "b":"bar"}')` key
json_each_text(json)set of key text, value text把最外层的JSON对象展开成键/值对的集合。返回值的类型是text。select * from json_each_text('{"a":"foo", "b":"bar"}')` key
json_extract_path(from_json json, VARIADIC path_elems text[])json返回path_elems指定的JSON值。等效于#>操作符。json_extract_path('{"f2":{"f3":1},"f4":{"f5":99,"f6":"foo"}}','f4'){"f5":99,"f6":"foo"}
json_extract_path_text(from_json json, VARIADIC path_elems text[])text返回path_elems指定的JSON值为文本。等效于#>>操作符。json_extract_path_text('{"f2":{"f3":1},"f4":{"f5":99,"f6":"foo"}}','f4', 'f6')foo
json_object_keys(json)setof text返回最外层JSON对象中的键集合。json_object_keys('{"f1":"abc","f2":{"f3":"a", "f4":"b"}}') json_object_keys ------------------ f1 f2
json_populate_record(base anyelement, from_json json)anyelement把Expands the object in from_json中的对象展开成一行,其中的列匹配由base定义的记录类型。select * from json_populate_record(null::myrowtype, '{"a":1,"b":2}')` a
json_populate_recordset(base anyelement, from_json json)set of anyelement将from_json中最外层的对象数组展开成一个行集合,其中的列匹配由base定义的记录类型。select * from json_populate_recordset(null::myrowtype, '[{"a":1,"b":2},{"a":3,"b":4}]')` a
json_array_elements(json)set of json将一个JSON数组展开成JSON值的一个集合。select * from json_array_elements('[1,true, [2,false]]') value ----------- 1 true [2,false]
json_object_field_text(json_obj json, key text)text提取JSON对象中指定字段的文本值,但它只能用于直接提取顶层字段,若需要过滤嵌套字段,可以利用->->>运算符
jsonb_pretty(from_json jsonb)text把*from_json*返回成一段 缩进后的 JSON 文本。jsonb_pretty(‘[{“f1”:1,“f2”:null},2,null,3]’)
jsonb_set(target jsonb, path text[], new_value jsonb[,create_missing boolean])jsonb返回*target*,其中由 *path*指定的节用 *new_value*替换,如果 *path*指定的项不存在并且 *create_missing为真(默认为 true)则加上 new_value。正如面向路径的 操作符一样,出现在path*中的 负整数表示从 JSON 数组的末尾开始数。jsonb_set(‘[{“f1”:1,“f2”:null},2,null,3]’, ‘{0,f1}’,‘[2,3,4]’, false)
jsonb_set(‘[{“f1”:1,“f2”:null},2]’, ‘{0,f3}’,‘[2,3,4]’)
[{“f1”:[2,3,4],“f2”:null},2,null,3] [{“f1”: 1, “f2”: null, “f3”: [2, 3, 4]}, 2]
json_each()

json_each()函数的作用是:将最外层的JSON对象展开为一组键值对。举例:

postgres=# SELECT json_each(info) FROM orders where id <=4;
                     json_each
---------------------------------------------------
 (customer,"""John Doe""")
 (items,"{""product"": ""Beer"",""qty"": 6}")
 (customer,"""Lily Bush""")
 (items,"{""product"": ""Diaper"",""qty"": 24}")
 (customer,"""Josh William""")
 (items,"{""product"": ""Toy Car"",""qty"": 1}")
 (customer,"""Mary Clark""")
 (items,"{""product"": ""Toy Train"",""qty"": 2}")
(8 rows)

如果想得到一组key-value对作为文本,可以使用json_each_text()函数。

postgres=# SELECT json_each_text(info) FROM orders where id <=4;
                  json_each_text
---------------------------------------------------
 (customer,"John Doe")
 (items,"{""product"": ""Beer"",""qty"": 6}")
 (customer,"Lily Bush")
 (items,"{""product"": ""Diaper"",""qty"": 24}")
 (customer,"Josh William")
 (items,"{""product"": ""Toy Car"",""qty"": 1}")
 (customer,"Mary Clark")
 (items,"{""product"": ""Toy Train"",""qty"": 2}")
(8 rows)

其他示例

postgres=# SELECT * FROM json_each_text('{"a":"foo", "b":"bar"}');
 key | value
-----+-------
 a   | foo
 b   | bar
(2 rows)

postgres=# SELECT json_each_text('{"a":"foo", "b":"bar"}');
 json_each_text
----------------
 (a,foo)
 (b,bar)
(2 rows)
json_object_keys()

json_object_keys()函数可以获得json对象最外层的一组键。举例:

postgres=# SELECT json_object_keys (info->'items') FROM orders where id <=4;
 json_object_keys
------------------
 product
 qty
 product
 qty
 product
 qty
 product
 qty
(8 rows)
json_typeof()

json_typeof函数返回json最外层key的数据类型作为字符串。可能是number, boolean, null, object, array, string。

-- 下面语句查询所有item的数据类型:
postgres=# SELECT json_typeof (info->'items') FROM orders where id <=4;
 json_typeof
-------------
 object
 object
 object
 object
(4 rows)

-- 下面语句返回嵌套类型中qty字段的数据类型:
postgres=# SELECT json_typeof (info->'items'->'qty') FROM orders where id <=4;
 json_typeof
-------------
 number
 number
 number
 number
(4 rows)
json_object_field_text(json_obj json, key text)

在 PostgreSQL 中,json_object_field_text 函数用于从 JSON 对象中提取指定键的文本值。以下是其详细用法:

语法

json_object_field_text(json_obj json, key text)text
  • json_obj: 目标 JSON 对象。
  • key: 需要提取的字段名(键)。

功能

json_object_field_text 函数从 JSON 对象中提取指定键的值,并返回该值的文本表示。如果键不存在,则返回 NULL

使用示例

假设有以下 JSON 数据:

{ "customer": "Lily Bush", "items": {"product": "Diaper", "qty": 24} }

1. 提取顶层字段值

info 字段中提取 customer 的值:

SELECT json_object_field_text(info, 'customer') AS customer_name
FROM orders;

2. 提取嵌套字段值

对于嵌套字段,首先需要提取包含目标字段的对象,然后从中提取字段值。例如,提取 items 对象中的 product 值:

SELECT json_object_field_text(info, 'items') AS items_json
FROM orders;

要进一步从 items JSON 对象中提取 product 字段,可以结合 jsonb 运算符:

SELECT (json_object_field_text(info, 'items')::jsonb->>'product') AS product_name
FROM orders;

注意事项

  • json_object_field_text 函数专用于提取 JSON 对象的顶层字段。对于嵌套字段,可能需要结合其他 JSON 操作函数和运算符。
  • 如果使用 jsonb 类型,可以直接利用 ->->> 运算符来处理嵌套字段,而不必转换为 text

示例表

假设 orders 表的 info 列为 JSON 类型,以下是如何使用这些函数的示例:

CREATE TABLE orders (
    id SERIAL PRIMARY KEY,
    info JSON
);

INSERT INTO orders (info) VALUES
('{"customer": "Lily Bush", "items": {"product": "Diaper", "qty": 24}}'),
('{"customer": "John Doe", "items": {"product": "Wipes", "qty": 50}}');

查询 customer 字段的值:

SELECT json_object_field_text(info, 'customer') AS customer_name
FROM orders;

提取 items 对象中的 product 值:

SELECT (json_object_field_text(info, 'items')::jsonb->>'product') AS product_name
FROM orders;
json_extract_path_text(from_json json, VARIADIC path_elems text[])

以text返回由path_elems指向的 JSON 值(等效于#>>操作符)。

select json_extract_path_text(info,'customer') from orders where id<=4;
-- 等价于
select info #>> '{customer}' from orders where id<=4;

   ?column?
--------------
 John Doe
 Lily Bush
 Josh William
 Mary Clark

(5 rows)
jsonb_pretty(from_json jsonb)

把*from_json*返回成一段 缩进后的 JSON 文本。

postgres=# select id,jsonb_pretty(info::jsonb) from orders where id=1;
 id |        jsonb_pretty
----+----------------------------
  1 | {                         +
    |     "items": {            +
    |         "qty": 6,         +
    |         "product": "Beer" +
    |     },                    +
    |     "customer": "John Doe"+
    | }
(1 row)
jsonb_set

jsonb_set() 函数参数如下:

jsonb_set(target         jsonb,  // 需要修改的数据
          path           text[], // 数据路径
          new_value      jsonb,  // 新数据
          create_missing boolean default true)

如果create_missing 是true (缺省是true),并且path指定的路径在target 中不存在,那么target将包含path指定部分, new_value替换部分, 或者new_value添加部分。

-- 更新 target 第0 个元素 key 为 f1 的值,如果f1 不存在 忽略
select jsonb_set('[{"f1":1,"f2":null},2,null,3]', '{0,f1}','[2,3,4]', false);
                  jsonb_set
---------------------------------------------
 [{"f1": [2, 3, 4], "f2": null}, 2, null, 3]

-- 更新 target 第0 个元素 key 为 f3 的值,如果f3 不存在 创建 
select jsonb_set('[{"f1":1,"f2":null},2]', '{0,f3}','[2,3,4]');
                  jsonb_set
---------------------------------------------
 [{"f1": 1, "f2": null, "f3": [2, 3, 4]}, 2]
 
 -- 更新 target 第0 个元素 key 为 f3 的值,如果f3 不存在 忽略
select jsonb_set('[{"f1":1,"f2":null},2]', '{0,f3}','[2,3,4]', false);
                  jsonb_set
---------------------------------------------
 [{"f1": 1, "f2": null}, 2]
json_array_elements*

把顶层 JSON 数组扩展成一个text值集合。

postgres=# select * from json_array_elements('[1,true,[2,false]]');
   value   
-----------
 1
 true
 [2,false]
(3 rows)

postgres=# select * from json_array_elements_text('["foo", "bar"]');
 value 
-------
 foo
 bar
(2 rows)
json_array_length

返回顶层json数组中的元素数量。

postgres=# select json_array_length('[1,2,3,{"f1":1,"f2":[5,6]},4]');
 json_array_length 
-------------------
                 5
(1 row)

postgres=# select json_array_length('["foo", "bar"]');
 json_array_length 
-------------------
                 2
(1 row)

JSON与JSONB读写性能测试

构建JSON、JSONB测试表

下面通过一个简单的例子测试下json、jsonb的读写性能差异,计划创建以下三张表:

  • user_ini:基础数据表,并插入200万测试数据;
  • tbl_user_json:: json 数据类型表,200万数据;
  • tbl_user_jsonb: jsonb 数据类型表,200万数据;
CREATE TABLE user_ini(
    id int4,
    user_id int8,
    user_name varchar(64),
    create_time timestamp(6) with time zone default clock_timestamp());

INSERT INTO user_ini(id,user_id,user_name)
SELECT r,round(random()*2000000), r || '_francs' FROM generate_series(1,2000000) as r;

计划使用user_ini表数据生成json、jsonb数据,创建user_ini_jsonuser_ini_jsonb表,如下所示:

CREATE TABLE tbl_user_json(id serial, user_info json);
CREATE TABLE tbl_user_jsonb(id serial, user_info jsonb);

JSON与JSONB表写性能测试

根据user_ini数据通过row_to_json函数向表user_ini_json插入200万json数据,如下:

postgres=# \timing
Timing is on.
postgres=# INSERT INTO tbl_user_json(user_info) SELECT row_to_json(user_ini) FROM user_ini;
INSERT 0 2000000
Time: 15093.043 ms (00:15.093)

从以上结果看出tbl_user_json插入200万数据花了15秒左右;接着根据user_ini表数据生成200万jsonb数据并插入表tbl_user_jsonb,如下:

postgres=# INSERT INTO tbl_user_jsonb(user_info) SELECT row_to_json(user_ini)::jsonb FROM user_ini;
INSERT 0 2000000
Time: 19801.533 ms (00:19.802)

从以上看出tbl_user_jsonb表插入200万jsonb数据花了19秒左右,正好验证了json数据写入比jsonb快。

比较两表占用空间大小,如下所示

postgres=# \dt+ tbl_user_json
                        List of relations
 Schema |     Name      | Type  |  Owner   |  Size  | Description 
--------+---------------+-------+----------+--------+-------------
 public | tbl_user_json | table | postgres | 281 MB | 
(1 row)

postgres=# \dt+ tbl_user_jsonb
                         List of relations
 Schema |      Name      | Type  |  Owner   |  Size  | Description 
--------+----------------+-------+----------+--------+-------------
 public | tbl_user_jsonb | table | postgres | 333 MB | 
(1 row)

从占用空间来看,同样的数据量jsonb数据类型占用空间比json稍大。

随机查一条数据对比

postgres=# select * from tbl_user_json limit 1;
 id |                                             user_info                                              
----+----------------------------------------------------------------------------------------------------
  1 | {"id":1,"user_id":1141402,"user_name":"1_francs","create_time":"2024-08-11T01:33:57.532707+08:00"}
(1 row)

Time: 0.600 ms
postgres=# select * from tbl_user_jsonb limit 1;
 id |                                                 user_info                                                 
----+-----------------------------------------------------------------------------------------------------------
  1 | {"id": 1, "user_id": 1141402, "user_name": "1_francs", "create_time": "2024-08-11T01:33:57.532707+08:00"}
(1 row)

Time: 0.710 ms

JSON与JSONB表读性能测试

对于json、jsonb读性能测试我们选择基于json、jsonb键值查询的场景,例如,根据user_info字段的user_name键的值查询,如下所示:

postgres=# EXPLAIN ANALYZE SELECT * FROM tbl_user_jsonb WHERE user_info->>'user_name'='1_francs';
                                                             QUERY PLAN                                                             
------------------------------------------------------------------------------------------------------------------------------------
 Gather  (cost=1000.00..57053.22 rows=10000 width=143) (actual time=0.576..213.625 rows=1 loops=1)
   Workers Planned: 2
   Workers Launched: 2
   ->  Parallel Seq Scan on tbl_user_jsonb  (cost=0.00..55053.22 rows=4167 width=143) (actual time=128.923..198.702 rows=0 loops=3)
         Filter: ((user_info ->> 'user_name'::text) = '1_francs'::text)
         Rows Removed by Filter: 666666
 Planning Time: 0.098 ms
 Execution Time: 213.656 ms
(8 rows)


postgres=# EXPLAIN ANALYZE SELECT * FROM tbl_user_json WHERE user_info->>'user_name'='1_francs';
                                                            QUERY PLAN                                                             
-----------------------------------------------------------------------------------------------------------------------------------
 Gather  (cost=1000.00..50401.43 rows=10000 width=113) (actual time=0.353..746.473 rows=1 loops=1)
   Workers Planned: 2
   Workers Launched: 2
   ->  Parallel Seq Scan on tbl_user_json  (cost=0.00..48401.43 rows=4167 width=113) (actual time=481.034..726.493 rows=0 loops=3)
         Filter: ((user_info ->> 'user_name'::text) = '1_francs'::text)
         Rows Removed by Filter: 666666
 Planning Time: 0.066 ms
 Execution Time: 746.494 ms
(8 rows)

如上普通查询没有走索引的情况可以看出jsonb的读更快一点。

为更好的对比tbl_user_json、tbl_user_jsonb表基于键值查询的效率,计划根据user_info字段id键进行范围扫描对比性能,创建索引如下:

CREATE INDEX idx_gin_user_info_id ON tbl_user_json USING btree (((user_info ->> 'id')::integer));
CREATE INDEX idx_gin_user_infob_id ON tbl_user_jsonb USING btree (((user_info ->> 'id')::integer));

创建索引后对比查询性能

EXPLAIN ANALYZE SELECT id,user_info->'id',user_info->'user_name' FROM tbl_user_json
WHERE (user_info->>'id')::int4>1 AND (user_info->>'id')::int4<10000;
                                                              QUERY PLAN                                                              
--------------------------------------------------------------------------------------------------------------------------------------
 Bitmap Heap Scan on tbl_user_json  (cost=214.93..22655.42 rows=10000 width=68) (actual time=1.538..24.549 rows=9998 loops=1)
   Recheck Cond: ((((user_info ->> 'id'::text))::integer > 1) AND (((user_info ->> 'id'::text))::integer < 10000))
   Heap Blocks: exact=173
   ->  Bitmap Index Scan on idx_gin_user_info_id  (cost=0.00..212.43 rows=10000 width=0) (actual time=1.495..1.495 rows=9998 loops=1)
         Index Cond: ((((user_info ->> 'id'::text))::integer > 1) AND (((user_info ->> 'id'::text))::integer < 10000))
 Planning Time: 0.546 ms
 Execution Time: 25.101 ms
(7 rows)


EXPLAIN ANALYZE SELECT id,user_info->'id',user_info->'user_name' FROM tbl_user_jsonb
WHERE (user_info->>'id')::int4>1 AND (user_info->>'id')::int4<10000;
                                                              QUERY PLAN                                                               
---------------------------------------------------------------------------------------------------------------------------------------
 Bitmap Heap Scan on tbl_user_jsonb  (cost=214.93..24049.23 rows=10000 width=68) (actual time=0.712..7.030 rows=9998 loops=1)
   Recheck Cond: ((((user_info ->> 'id'::text))::integer > 1) AND (((user_info ->> 'id'::text))::integer < 10000))
   Heap Blocks: exact=212
   ->  Bitmap Index Scan on idx_gin_user_infob_id  (cost=0.00..212.43 rows=10000 width=0) (actual time=0.672..0.673 rows=9998 loops=1)
         Index Cond: ((((user_info ->> 'id'::text))::integer > 1) AND (((user_info ->> 'id'::text))::integer < 10000))
 Planning Time: 0.276 ms
 Execution Time: 7.413 ms
(7 rows)

如上user_info字段id键值在1到10000范围内的记录走了索引,而且jsonb检索效率比json快了四倍以上。

从以上两个测试看出,正好验证了 “json写入比jsonb快,但检索时比jsonb慢” 的观点,值得一提的是如果需要通过key/value进行检索,例如以下。

EXPLAIN ANALYZE SELECT * FROM tbl_user_json WHERE user_info::jsonb @> '{"user_name": "2_francs"}';
                                                            QUERY PLAN                                                             
-----------------------------------------------------------------------------------------------------------------------------------
 Gather  (cost=1000.00..51684.33 rows=2000 width=113) (actual time=0.355..1144.857 rows=1 loops=1)
   Workers Planned: 2
   Workers Launched: 2
   ->  Parallel Seq Scan on tbl_user_json  (cost=0.00..50484.33 rows=833 width=113) (actual time=759.142..1140.358 rows=0 loops=3)
         Filter: ((user_info)::jsonb @> '{"user_name": "2_francs"}'::jsonb)
         Rows Removed by Filter: 666666
 Planning Time: 0.197 ms
 Execution Time: 1144.885 ms
(8 rows)



EXPLAIN ANALYZE SELECT * FROM tbl_user_jsonb WHERE user_info @> '{"user_name": "2_francs"}';
                                                            QUERY PLAN                                                             
-----------------------------------------------------------------------------------------------------------------------------------
 Gather  (cost=1000.00..54169.67 rows=2000 width=143) (actual time=0.473..233.123 rows=1 loops=1)
   Workers Planned: 2
   Workers Launched: 2
   ->  Parallel Seq Scan on tbl_user_jsonb  (cost=0.00..52969.67 rows=833 width=143) (actual time=150.780..227.646 rows=0 loops=3)
         Filter: (user_info @> '{"user_name": "2_francs"}'::jsonb)
         Rows Removed by Filter: 666666
 Planning Time: 0.114 ms
 Execution Time: 233.147 ms
(8 rows)

Time: 233.684 ms

tbl_user_jsonb字段user_info上创建gin索引,如下所示:

CREATE INDEX idx_tbl_user_jsonb_user_Info ON tbl_user_jsonb USING gin(user_Info);

索引创建后,再次执行一下,如下所示:

EXPLAIN ANALYZE SELECT * FROM tbl_user_jsonb WHERE user_info @> '{"user_name": "2_francs"}';
                                                               QUERY PLAN                                                                
-----------------------------------------------------------------------------------------------------------------------------------------
 Bitmap Heap Scan on tbl_user_jsonb  (cost=59.50..6647.38 rows=2000 width=143) (actual time=0.191..0.193 rows=1 loops=1)
   Recheck Cond: (user_info @> '{"user_name": "2_francs"}'::jsonb)
   Heap Blocks: exact=1
   ->  Bitmap Index Scan on idx_tbl_user_jsonb_user_info  (cost=0.00..59.00 rows=2000 width=0) (actual time=0.173..0.174 rows=1 loops=1)
         Index Cond: (user_info @> '{"user_name": "2_francs"}'::jsonb)
 Planning Time: 0.632 ms
 Execution Time: 0.228 ms
(7 rows)

PostgreSQ全文检索支持JSON和JSONB

PostgreSQL全文检索简介

对于大多数应用全文检索很少放到数据库中实现,一般使用单独的全文检索引擎,例如基于SQL全文检索引擎Sphinx。PostgreSQL支持全文检索,对于规模不大的应用如果不想搭建专门的搜索引擎,PostgreSQL的全文检索也可以满足需求。

如果没有使用专门的搜索引擎,大部检索需要通过数据库like操作匹配,这种检索方式主要缺点在于:

  • 不能很好的支持索引,通常需全表扫描检索数据,数据量大时检索性能很低。
  • 不提供检索结果排序,当输出结果数据量非常大时表现更加明显。

PostgreSQL全文检索能有效地解决这个问题,PostgreSQL全文检索通过以下两种数据类型来实现。

1、tsvector

tsvector全文检索数据类型代表一个被优化的可以基于搜索的文档,将一串字符串转换成tsvector全文检索数据类型,如下:

postgres=# SELECT 'Hello,cat,how are u? cat is smiling! '::tsvector;
                     tsvector                     
--------------------------------------------------
 'Hello,cat,how' 'are' 'cat' 'is' 'smiling!' 'u?'
(1 row)

可以看到,字符串的内容被分隔成好几段,但通过::tsvector只是做类型转换,没有进行数据标准化处理,对于英文全文检索可通过函数to_tsvector进行数据标准化,如下所示:

postgres=# SELECT to_tsvector('english','Hello cat,');
    to_tsvector    
-------------------
 'cat':2 'hello':1
(1 row)

2、tsquery

tsquery表示一个文本查询,存储用于搜索的词,并且支持布尔操作&|!,将字符串转换成tsquery,如下所示:

postgres=# SELECT 'hello&cat'::tsquery;
     tsquery     
-----------------
 'hello' & 'cat'
(1 row)

上述只是转换成tsquery类型,而并没有做标准化,使用to_tsquery函数可以执行标准化,如下所示:

postgres=# SELECT to_tsquery('hello&cat');
   to_tsquery    
-----------------
 'hello' & 'cat'
(1 row)

一个全文检索示例如下,检索字符串是否包括hello和cat字符,本例中返回真。

postgres=# SELECT to_tsvector('english','Hello cat,how are u') @@ to_tsquery('hello&cat');
 ?column? 
----------
 t
(1 row)

检索字符串是否包含字符hello和dog,本例中返回假。

postgres=# SELECT to_tsvector('english','Hello cat,how are u') @@ to_tsquery('hello&dog');
 ?column? 
----------
 f
(1 row)

有兴趣可以测试一下tsquery的其他操作符,例如|、!等。

注意:这里使用了带双参数的to_tsvector函数,函数to_tsvector双参数的格式如下:

to_tsvector([ config regconfig , ] document text)
#本节to_tsvector函数指定了config参数为english,如果不指定config参数,则默认使用default_text_search_config参数的配置。


postgres=# select name,setting,unit,context from pg_settings where name ~* 'default_text_search_config';
            name            |      setting       | unit | context 
----------------------------+--------------------+------+---------
 default_text_search_config | pg_catalog.english |      | user
(1 row)

英文全文检索例子

下面演示一个英文全文检索示例,创建一张测试表并插入200万测试数据,如下所示:

CREATE TABLE test_search(id int4,name text);
INSERT INTO test_search(id,name) SELECT n, n||'_francs' FROM generate_series(1,2000000) n;

执行以下SQL,查询test_search表name字段包含字符1_francs的记录。

postgres=# SELECT * FROM test_search WHERE name LIKE '1_francs';
 id |   name   
----+----------
  1 | 1_francs
(1 row)

-- 执行计划
EXPLAIN ANALYZE SELECT * FROM test_search WHERE name LIKE '1_francs';
                                                         QUERY PLAN                                                         
----------------------------------------------------------------------------------------------------------------------------
 Gather  (cost=1000.00..24166.67 rows=200 width=18) (actual time=0.423..84.066 rows=1 loops=1)
   Workers Planned: 2
   Workers Launched: 2
   ->  Parallel Seq Scan on test_search  (cost=0.00..23146.67 rows=83 width=18) (actual time=50.502..77.706 rows=0 loops=3)
         Filter: (name ~~ '1_francs'::text)
         Rows Removed by Filter: 666666
 Planning Time: 0.262 ms
 Execution Time: 84.090 ms
(8 rows)

以上执行计划走了全表扫描,执行时间为84毫秒左右,性能很低,接着创建索引,如下所示

CREATE INDEX idx_gin_search ON test_search USING gin(to_tsvector('english',name));

执行以下SQL,查询test_search表name字段包含字符1_francs的记录。

postgres=# SELECT * FROM test_search WHERE to_tsvector('english',name) @@ to_tsquery('english','1_francs');
 id |   name   
----+----------
  1 | 1_francs
(1 row)


-- 执行计划
EXPLAIN ANALYZE SELECT * FROM test_search WHERE to_tsvector('english',name) @@ to_tsquery('english','1_francs');
                                                       QUERY PLAN                                                        
-------------------------------------------------------------------------------------------------------------------------
 Bitmap Heap Scan on test_search  (cost=36.39..240.11 rows=50 width=18) (actual time=0.129..0.131 rows=1 loops=1)
   Recheck Cond: (to_tsvector('english'::regconfig, name) @@ '''1'' & ''franc'''::tsquery)
   Heap Blocks: exact=1
   ->  Bitmap Index Scan on idx_gin_search  (cost=0.00..36.38 rows=50 width=0) (actual time=0.111..0.112 rows=1 loops=1)
         Index Cond: (to_tsvector('english'::regconfig, name) @@ '''1'' & ''franc'''::tsquery)
 Planning Time: 0.289 ms
 Execution Time: 0.157 ms
(7 rows)

创建索引后,以上查询走了索引并且执行时间下降到0.157毫秒,性能提升很大,值得一提的是如果SQL改成以下,则不走索引,如下所示:

EXPLAIN ANALYZE SELECT * FROM test_search WHERE to_tsvector(name) @@ to_tsquery('1_francs');
                                                           QUERY PLAN                                                            
---------------------------------------------------------------------------------------------------------------------------------
 Gather  (cost=1000.00..440818.33 rows=50 width=18) (actual time=0.339..2869.023 rows=1 loops=1)
   Workers Planned: 2
   Workers Launched: 2
   ->  Parallel Seq Scan on test_search  (cost=0.00..439813.33 rows=21 width=18) (actual time=1908.625..2864.505 rows=0 loops=3)
         Filter: (to_tsvector(name) @@ to_tsquery('1_francs'::text))
         Rows Removed by Filter: 666666
 Planning Time: 0.160 ms
 Execution Time: 2869.147 ms
(8 rows)

由于创建索引时使用的是to_tsvector('english',name)函数索引,带了两个参数,因此where条件中的to_tsvector函数带两个参数才能走索引,而to_tsvector(name)不走索引。

JSON、JSONB全文检索实践

在PostgreSQL10版本之前全文检索不支持json和jsonb数据类型,10版本的一个重要特性是全文检索支持json和jsonb数据类型,这一小节演示下10版本以后的这个新特性。

当前测试版本为12版本

postgres=# select version();
                                                 version                                                  
----------------------------------------------------------------------------------------------------------
 PostgreSQL 12.17 on x86_64-pc-linux-gnu, compiled by gcc (GCC) 4.8.5 20150623 (Red Hat 4.8.5-36), 64-bit
(1 row)

PostgreSQL12版本与9.6版本to_tsvector函数的差异

先来看下9.6版本to_tsvector函数,如下:

[pg96@pghost1 ~]$ psql
psql (9.6.21)
Type "help" for help.

postgres=# \df *to_tsvector*
                                List of functions
   Schema   |       Name        | Result data type | Argument data types |  Type  
------------+-------------------+------------------+---------------------+--------
 pg_catalog | array_to_tsvector | tsvector         | text[]              | normal
 pg_catalog | to_tsvector       | tsvector         | regconfig, text     | normal
 pg_catalog | to_tsvector       | tsvector         | text                | normal
(3 rows)

从以上看出9.6版本to_tsvector函数的输入参数仅支持text、text[]数据类型,接着看下12版本的to_tsvector函数,如下所示:

[postgres@centos7 ~]$ psql
psql (12.17)
Type "help" for help.

postgres=# \df *to_tsvector*
                                 List of functions
   Schema   |       Name        | Result data type |   Argument data types   | Type 
------------+-------------------+------------------+-------------------------+------
 pg_catalog | array_to_tsvector | tsvector         | text[]                  | func
 pg_catalog | json_to_tsvector  | tsvector         | json, jsonb             | func
 pg_catalog | json_to_tsvector  | tsvector         | regconfig, json, jsonb  | func
 pg_catalog | jsonb_to_tsvector | tsvector         | jsonb, jsonb            | func
 pg_catalog | jsonb_to_tsvector | tsvector         | regconfig, jsonb, jsonb | func
 pg_catalog | to_tsvector       | tsvector         | json                    | func
 pg_catalog | to_tsvector       | tsvector         | jsonb                   | func
 pg_catalog | to_tsvector       | tsvector         | regconfig, json         | func
 pg_catalog | to_tsvector       | tsvector         | regconfig, jsonb        | func
 pg_catalog | to_tsvector       | tsvector         | regconfig, text         | func
 pg_catalog | to_tsvector       | tsvector         | text                    | func
(11 rows)

从以上看出,12版本的to_tsvector函数支持的数据类型增加了json和jsonb。

JSON数据全文检索测试

创建数据生成函数

为了便于生成测试数据,创建以下两个函数用来随机生成指定长度的字符串,创建random_range(int4, int4)函数如下:

CREATE OR REPLACE FUNCTION random_range(int4, int4)
RETURNS int4
LANGUAGE SQL
AS $$
    SELECT ($1 + FLOOR(($2 - $1 + 1) * random() ))::int4;
$$;

-- 接着创建random_text_simple(length int4)函数,此函数会调用random_range(int4, int4)函数。
CREATE OR REPLACE FUNCTION random_text_simple(length int4)
RETURNS text
LANGUAGE PLPGSQL
AS $$
DECLARE
    possible_chars text := '0123456789abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ';
    output text := '';
    i int4;
    pos int4;
BEGIN
    FOR i IN 1..length LOOP
        pos := random_range(1, length(possible_chars));
        output := output || substr(possible_chars, pos, 1);
    END LOOP;
    RETURN output;
END;
$$;

random_text_simple(length int4)函数可以随机生成指定长度字符串,如下示例:

postgres=# SELECT random_text_simple(3);
 random_text_simple 
--------------------
 4dI
(1 row)

postgres=# SELECT random_text_simple(6);
 random_text_simple 
--------------------
 3uLknl
(1 row)

创建JSON测试表

创建user_ini测试表,并通过random_text_simple(length int4)函数插入100万随机生成六位字符的字符串测试数据,如下所示:

drop table if exists user_ini;
CREATE TABLE user_ini(
    id int4,
    user_id int8,
    user_name varchar(64),
    create_time timestamp(6) with time zone default clock_timestamp());

INSERT INTO user_ini(id,user_id,user_name)
SELECT r,round(random()*1000000), random_text_simple(6) FROM generate_series(1,1000000) as r;

创建tbl_user_search_json表,并通过row_to_json函数将表user_ini行数据转换成json数据,如下所示:

CREATE TABLE tbl_user_search_json(id serial, user_info json);
INSERT INTO tbl_user_search_json(user_info) SELECT row_to_json(user_ini) FROM user_ini;

-- 生成的数据如下:
postgres=# SELECT * FROM tbl_user_search_json LIMIT 1;
 id |                                           user_info                                            
----+------------------------------------------------------------------------------------------------
  1 | {"id":1,"user_id":185716,"user_name":"mOuXBE","create_time":"2024-08-11T03:27:46.69352+08:00"}
(1 row)

使用全文检索查询表tbl_user_search_jsonuser_info字段中包含KTU89H字符的记录,如下所示:

postgres=# SELECT * FROM tbl_user_search_json WHERE to_tsvector('english',user_info) @@ to_tsquery('ENGLISH','bb7tQk');
 id  |                                             user_info                                             
-----+---------------------------------------------------------------------------------------------------
 100 | {"id":100,"user_id":640314,"user_name":"bb7tQk","create_time":"2024-08-11T03:27:46.694603+08:00"}
(1 row)

-- 执行计划
EXPLAIN ANALYZE SELECT * FROM tbl_user_search_json WHERE to_tsvector('english',user_info) @@ to_tsquery('ENGLISH','bb7tQk');
                                                                 QUERY PLAN                                                                  
---------------------------------------------------------------------------------------------------------------------------------------------
 Gather  (cost=1000.00..127886.00 rows=5000 width=104) (actual time=1.415..2299.886 rows=1 loops=1)
   Workers Planned: 2
   Workers Launched: 2
   ->  Parallel Seq Scan on tbl_user_search_json  (cost=0.00..126386.00 rows=2083 width=104) (actual time=1522.099..2287.579 rows=0 loops=3)
         Filter: (to_tsvector('english'::regconfig, user_info) @@ '''bb7tqk'''::tsquery)
         Rows Removed by Filter: 333333
 Planning Time: 0.091 ms
 Execution Time: 2299.910 ms
(8 rows)

以上SQL能正常执行说明全文检索支持json数据类型,只是上述SQL走了全表扫描性能低,执行时间为2299毫秒

创建索引

CREATE INDEX idx_gin_search_json ON tbl_user_search_json USING gin(to_tsvector('english',user_info));

索引创建后,再次执行以下SQL,如下所示:

EXPLAIN ANALYZE SELECT * FROM tbl_user_search_json WHERE to_tsvector('english',user_info) @@ to_tsquery('ENGLISH','bb7tQk');
                                                           QUERY PLAN                                                           
--------------------------------------------------------------------------------------------------------------------------------
 Bitmap Heap Scan on tbl_user_search_json  (cost=62.75..12193.30 rows=5000 width=104) (actual time=0.046..0.047 rows=1 loops=1)
   Recheck Cond: (to_tsvector('english'::regconfig, user_info) @@ '''bb7tqk'''::tsquery)
   Heap Blocks: exact=1
   ->  Bitmap Index Scan on idx_gin_search_json  (cost=0.00..61.50 rows=5000 width=0) (actual time=0.024..0.024 rows=1 loops=1)
         Index Cond: (to_tsvector('english'::regconfig, user_info) @@ '''bb7tqk'''::tsquery)
 Planning Time: 0.439 ms
 Execution Time: 0.079 ms
(7 rows)

从上述执行计划看出走了索引,并且执行时间降为0.079毫秒,性能非常不错。

索引支持

jsonb缺省的GIN操作符类支持使用@>??&?|操作符查询

JSONB类型支持GIN, BTree索引。一般情况下,我们会在JSONB类型字段上建GIN索引,语法如下:

CREATE INDEX idx_name ON table_name USING gin (idx_col);
CREATE INDEX idx_name ON table_name USING gin (idx_col jsonb_path_ops);

说明:在JSONB上创建GIN索引的方式有两种:使用默认的jsonb_ops操作符创建和使用jsonb_path_ops操作符创建。两者的区别在jsonb_ops的GIN索引中,JSONB数据中的每个key和value都是作为一个单独的索引项的,而jsonb_path_ops则只为每个value创建一个索引项。

JSON 函数索引

CREATE TABLE test_json (json_type text,obj json);
insert into test_json values
('aa', '{"f2":{"f3":1},"f4":{"f5":99,"f6":"foo"}}'),
('cc', '{"f7":{"f3":1},"f8":{"f5":99,"f6":"foo"}}');


select obj->'f2' from test_json where json_type = 'aa';
 ?column?
----------
 {"f3":1}
(1 row)

-- 创建函数索引
create index i on test_json (json_extract_path_text(obj, 'f4'));


select * from test_json where json_extract_path_text(obj, 'f4') = '{"f5":99,"f6":"foo"}';
 json_type |                 obj
-----------+-------------------------------------------
 aa        | {"f2":{"f3":1},"f4":{"f5":99,"f6":"foo"}}
(1 row)

JSONB创建索引

-- 创建测试表并生成数据
CREATE TABLE test_t (id int,info json);

-- 创建随机生成字符串函数
create or replace function random_string(integer)
returns text as
$body$
    select array_to_string(array(select substring('0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz' FROM (ceil(random()*62))::int FOR 1) FROM generate_series(1, $1)), '');
$body$
language sql volatile;

-- 插入数据
insert into test_t
select t.seq, ('{"a":{"a1":"a1a1", "a2":"a2a2"}, "name":"'||random_string(10)||'","b":"bbbbb"}')::json
from generate_series(1, 10000000) as t(seq);

CREATE TABLE test_t2 (id int,info jsonb);
CREATE TABLE test_t3 (id int,info jsonb);
insert into test_t2 select id, info::jsonb from test_t;
insert into test_t3 select id, info::jsonb from test_t;

-- 建立索引
CREATE INDEX idx_test2 ON test_t2 USING gin(info);
CREATE INDEX idx_test3 ON test_t3 USING gin(info jsonb_path_ops);

-- 未建索引执行
EXPLAIN ANALYZE SELECT * FROM test_t where info::jsonb @> '{"name":"FMYYFF6I5O"}';
                                                          QUERY PLAN
-------------------------------------------------------------------------------------------------------------------------------
 Gather  (cost=1000.00..207373.79 rows=100000 width=70) (actual time=2.398..6762.787 rows=1 loops=1)
   Workers Planned: 2
   Workers Launched: 2
   ->  Parallel Seq Scan on test_t  (cost=0.00..196373.79 rows=41667 width=70) (actual time=4496.908..6749.185 rows=0 loops=3)
         Filter: ((info)::jsonb @> '{"name": "FMYYFF6I5O"}'::jsonb)
         Rows Removed by Filter: 3333333
 Planning Time: 0.069 ms
 Execution Time: 6762.834 ms
(8 rows)

-- 使用jsonb_ops操作符创建索引执行
EXPLAIN ANALYZE SELECT * FROM test_t2 where info @> '{"name":"FMYYFF6I5O"}';
                                                      QUERY PLAN
----------------------------------------------------------------------------------------------------------------------
 Bitmap Heap Scan on test_t2  (cost=51.75..3802.38 rows=1000 width=88) (actual time=0.597..0.598 rows=1 loops=1)
   Recheck Cond: (info @> '{"name": "FMYYFF6I5O"}'::jsonb)
   Heap Blocks: exact=1
   ->  Bitmap Index Scan on idx_test2  (cost=0.00..51.50 rows=1000 width=0) (actual time=0.432..0.432 rows=1 loops=1)
         Index Cond: (info @> '{"name": "FMYYFF6I5O"}'::jsonb)
 Planning Time: 3.766 ms
 Execution Time: 0.635 ms
(7 rows)

-- 使用jsonb_path_ops操作符创建索引执行
postgres=# EXPLAIN ANALYZE SELECT * FROM test_t3 where info @> '{"name":"FMYYFF6I5O"}';
                                                      QUERY PLAN
----------------------------------------------------------------------------------------------------------------------
 Bitmap Heap Scan on test_t3  (cost=31.75..3782.38 rows=1000 width=88) (actual time=0.401..0.402 rows=1 loops=1)
   Recheck Cond: (info @> '{"name": "FMYYFF6I5O"}'::jsonb)
   Heap Blocks: exact=1
   ->  Bitmap Index Scan on idx_test3  (cost=0.00..31.50 rows=1000 width=0) (actual time=0.019..0.019 rows=1 loops=1)
         Index Cond: (info @> '{"name": "FMYYFF6I5O"}'::jsonb)
 Planning Time: 0.366 ms
 Execution Time: 0.437 ms
(7 rows)

JSONB性能分析

数据准备

-- account 表 id 使用uuid 类型,需要先添加uuid-ossp模块。
CREATE EXTENSION IF NOT EXISTS "uuid-ossp";

-- create table
create table account (id UUID NOT NULL PRIMARY KEY default uuid_generate_v1(), content jsonb, created_at timestamptz DEFAULT CURRENT_TIMESTAMP, updated_at timestamptz DEFAULT CURRENT_TIMESTAMP);
postgres=> \d account
                               Table "public.account"
    Column    |           Type           | Collation | Nullable |      Default
--------------+--------------------------+-----------+----------+--------------------
 id           | uuid                     |           | not null |uuid_generate_v1()
 content      | jsonb                    |           |          |
 created_at   | timestamp with time zone |           |          | CURRENT_TIMESTAMP
 updated_at   | timestamp with time zone |           |          | CURRENT_TIMESTAMP
Indexes:
    "account_pkey" PRIMARY KEY, btree (id)

content 数据结构为

content = {
    "nickname": {"type": "string"},
    "avatar": {"type": "string"},
    "weixin": {"type": "string"},
    "tags": {"type": "array", "items": {"type": "string"}},
}

批量插入数据

-- 插入100w条有 nickname avatar tags 为["python", "golang", "c"]的数据
insert into account select uuid_generate_v1(), ('{"nickname": "nn-' || round(random()*20000000) || '", "avatar": "avatar_url", "tags": ["python", "golang", "c"]}')::jsonb from (select * from generate_series(1,100000)) as tmp;

-- 插入100w条有 nickname tags 为["python", "golang"]的数据
insert into account select uuid_generate_v1(), ('{"nickname": "nn-' || round(random()*2000000) || '", "tags": ["python", "golang"]}')::jsonb from (select * from generate_series(1,1000000)) as tmp;

-- 插入100w条有 nickname tags 为["python"]的数据
insert into account select uuid_generate_v1(), ('{"nickname": "nn-' || round(random()*2000000) || '", "tags": ["python"]}')::jsonb from (select * from generate_series(1,1000000)) as tmp;

测试查询

content 中有avatar key 的数据条数 count(*) 查询不是一个好的测试语句,就算是有索引,也只能起到过滤的作用,如果结果集比较大,查询速度还是会很慢

explain analyze select count(*) from account where content::jsonb ? 'avatar';
                                                                QUERY PLAN
------------------------------------------------------------------------------------------------------------------------------------------
 Finalize Aggregate  (cost=45210.65..45210.66 rows=1 width=8) (actual time=148.609..159.876 rows=1 loops=1)
   ->  Gather  (cost=45210.43..45210.64 rows=2 width=8) (actual time=148.390..159.864 rows=3 loops=1)
         Workers Planned: 2
         Workers Launched: 2
         ->  Partial Aggregate  (cost=44210.43..44210.44 rows=1 width=8) (actual time=136.871..136.872 rows=1 loops=3)
               ->  Parallel Seq Scan on account  (cost=0.00..44111.50 rows=39573 width=0) (actual time=0.038..134.372 rows=33333 loops=3)
                     Filter: (content ? 'avatar'::text)
                     Rows Removed by Filter: 666667
 Planning Time: 1.671 ms
 Execution Time: 159.937 ms
(10 rows)

content 中没有avatar key 的数据条数

explain analyze select count(*) from account where content::jsonb ? 'avatar' = false;
                                                                 QUERY PLAN
--------------------------------------------------------------------------------------------------------------------------------------------
 Finalize Aggregate  (cost=47200.28..47200.29 rows=1 width=8) (actual time=206.111..217.322 rows=1 loops=1)
   ->  Gather  (cost=47200.07..47200.28 rows=2 width=8) (actual time=205.930..217.311 rows=3 loops=1)
         Workers Planned: 2
         Workers Launched: 2
         ->  Partial Aggregate  (cost=46200.07..46200.08 rows=1 width=8) (actual time=201.030..201.031 rows=1 loops=3)
               ->  Parallel Seq Scan on account  (cost=0.00..44111.50 rows=835427 width=0) (actual time=0.030..157.985 rows=666667 loops=3)
                     Filter: (NOT (content ? 'avatar'::text))
                     Rows Removed by Filter: 33333
 Planning Time: 0.112 ms
 Execution Time: 217.376 ms
(10 rows)

查询content 中nickname 为nn-194318的数据

explain analyze select * from account where content @> '{"nickname": "nn-194318"}';
                                                        QUERY PLAN
--------------------------------------------------------------------------------------------------------------------------
 Gather  (cost=1000.00..45132.50 rows=210 width=95) (actual time=197.300..208.413 rows=1 loops=1)
   Workers Planned: 2
   Workers Launched: 2
   ->  Parallel Seq Scan on account  (cost=0.00..44111.50 rows=88 width=95) (actual time=133.899..192.681 rows=0 loops=3)
         Filter: (content @> '{"nickname": "nn-194318"}'::jsonb)
         Rows Removed by Filter: 700000
 Planning Time: 0.206 ms
 Execution Time: 208.447 ms
(8 rows)

对应的查询id 为 ‘a830f154-5711-11ef-aac7-000c29d4de9c’ 的数据

explain analyze select * from account where id='a830f154-5711-11ef-aac7-000c29d4de9c';
                                                      QUERY PLAN
-----------------------------------------------------------------------------------------------------------------------
 Index Scan using account_pkey on account  (cost=0.43..8.45 rows=1 width=95) (actual time=0.029..0.031 rows=1 loops=1)
   Index Cond: (id = 'a830f154-5711-11ef-aac7-000c29d4de9c'::uuid)
 Planning Time: 0.089 ms
 Execution Time: 0.059 ms
(4 rows)

通过结果可以看到 使用 jsonb 查询和使用主键查询速度差异巨大,通过看查询分析记录可以看到,这两个语句最大的差别在于使用主键的查询用到了索引,而content nickname 的查询没有索引可以使用。 接下来测试一下使用索引时的查询速度。

索引

JSONB 最常用的是GIN 索引,GIN 索引可以被用来有效地搜索在大量jsonb文档(数据)中出现 的键或者键值对。

jsonb的默认 GIN 操作符类支持使用顶层键存在运算符??&以及?| 操作符和路径/值存在运算符@>的查询。

创建默认索引

CREATE INDEX idxgin ON account USING GIN (content);

create table account (id UUID NOT NULL PRIMARY KEY default uuid_generate_v1(), content jsonb, created_at timestamptz DEFAULT CURRENT_TIMESTAMP, updated_at timestamptz DEFAULT CURRENT_TIMESTAMP);

非默认的 GIN 操作符类jsonb_path_ops只支持索引@>操作符。

-- 创建指定路径的索引
CREATE INDEX idxginp ON account USING GIN (content jsonb_path_ops);

查询优化

创建默认索引

-- 创建简单索引
create index ix_account_content on account USING GIN (content);

现在下面这样的查询就能使用该索引:

-- content 中有avatar key 的数据条数
explain analyze select count(*) from account where content::jsonb ? 'avatar';
                                                                 QUERY PLAN
--------------------------------------------------------------------------------------------------------------------------------------------
 Aggregate  (cost=35490.68..35490.69 rows=1 width=8) (actual time=42.619..42.621 rows=1 loops=1)
   ->  Bitmap Heap Scan on account  (cost=892.06..35253.24 rows=94975 width=0) (actual time=8.532..36.016 rows=100000 loops=1)
         Recheck Cond: (content ? 'avatar'::text)
         Heap Blocks: exact=2041
         ->  Bitmap Index Scan on ix_account_content  (cost=0.00..868.31 rows=94975 width=0) (actual time=8.217..8.218 rows=100000 loops=1)
               Index Cond: (content ? 'avatar'::text)
 Planning Time: 0.495 ms
 Execution Time: 42.752 ms
(8 rows)

和之前没有添加索引时速度提升了3倍。

-- 查询content 中nickname 为nn-194318的数据
explain analyze select * from account where content@>'{"nickname": "nn-194318"}';
                                                          QUERY PLAN
------------------------------------------------------------------------------------------------------------------------------
 Bitmap Heap Scan on account  (cost=37.63..830.13 rows=210 width=95) (actual time=0.109..0.110 rows=1 loops=1)
   Recheck Cond: (content @> '{"nickname": "nn-194318"}'::jsonb)
   Heap Blocks: exact=1
   ->  Bitmap Index Scan on ix_account_content  (cost=0.00..37.58 rows=210 width=0) (actual time=0.098..0.099 rows=1 loops=1)
         Index Cond: (content @> '{"nickname": "nn-194318"}'::jsonb)
 Planning Time: 0.140 ms
 Execution Time: 0.147 ms
(7 rows)

这个查询效率提升更明显,竟然比使用主键还要高效。

但是下面这种查询并不能使用索引:

-- 查询content 中不存在 avatar key 的数据条数
explain analyze select count(*) from account where content::jsonb ? 'avatar' = false;
                                                                 QUERY PLAN
--------------------------------------------------------------------------------------------------------------------------------------------
 Finalize Aggregate  (cost=47200.28..47200.29 rows=1 width=8) (actual time=221.257..232.061 rows=1 loops=1)
   ->  Gather  (cost=47200.07..47200.28 rows=2 width=8) (actual time=221.052..232.049 rows=3 loops=1)
         Workers Planned: 2
         Workers Launched: 2
         ->  Partial Aggregate  (cost=46200.07..46200.08 rows=1 width=8) (actual time=215.582..215.583 rows=1 loops=3)
               ->  Parallel Seq Scan on account  (cost=0.00..44111.50 rows=835427 width=0) (actual time=0.033..170.202 rows=666667 loops=3)
                     Filter: (NOT (content ? 'avatar'::text))
                     Rows Removed by Filter: 33333
 Planning Time: 0.120 ms
 Execution Time: 232.136 ms
(10 rows)

该索引也不能被用于下面这样的查询,因为尽管操作符? 是可索引的,但它不能直接被应用于被索引列content:

explain analyze select count(1) from account where content -> 'tags' ? 'c';
                                                               QUERY PLAN
-----------------------------------------------------------------------------------------------------------------------------------------
 Finalize Aggregate  (cost=47321.09..47321.10 rows=1 width=8) (actual time=185.699..196.680 rows=1 loops=1)
   ->  Gather  (cost=47320.88..47321.08 rows=2 width=8) (actual time=185.500..196.670 rows=3 loops=1)
         Workers Planned: 2
         Workers Launched: 2
         ->  Partial Aggregate  (cost=46320.88..46320.89 rows=1 width=8) (actual time=180.745..180.747 rows=1 loops=3)
               ->  Parallel Seq Scan on account  (cost=0.00..46299.00 rows=8750 width=0) (actual time=0.030..178.474 rows=33333 loops=3)
                     Filter: ((content -> 'tags'::text) ? 'c'::text)
                     Rows Removed by Filter: 666667
 Planning Time: 0.108 ms
 Execution Time: 196.751 ms
(10 rows)

使用表达式索引

-- 创建路径索引
create index ix_account_content_tags on account USING GIN ((content->'tags'));

-- 测试查询性能
explain analyze select count(1) from account where content -> 'tags' ? 'c';
                                                                   QUERY PLAN
-------------------------------------------------------------------------------------------------------------------------------------------------
 Aggregate  (cost=31745.00..31745.01 rows=1 width=8) (actual time=49.878..49.880 rows=1 loops=1)
   ->  Bitmap Heap Scan on account  (cost=754.75..31692.50 rows=21000 width=0) (actual time=8.473..43.123 rows=100000 loops=1)
         Recheck Cond: ((content -> 'tags'::text) ? 'c'::text)
         Heap Blocks: exact=2041
         ->  Bitmap Index Scan on ix_account_content_tags  (cost=0.00..749.50 rows=21000 width=0) (actual time=8.138..8.138 rows=100000 loops=1)
               Index Cond: ((content -> 'tags'::text) ? 'c'::text)
 Planning Time: 0.251 ms
 Execution Time: 49.934 ms
(8 rows)

现在,WHERE 子句content -> 'tags' ? 'c' 将被识别为可索引操作符?在索引表达式content -> 'tags' 上的应用。

也可以利用包含查询的方式,例如:

-- 查寻 "tags" 包含数组元素 "c" 的数据的个数
explain analyze select count(1) from account where content @> '{"tags": ["c"]}';
                                                                  QUERY PLAN
----------------------------------------------------------------------------------------------------------------------------------------------
 Aggregate  (cost=35506.68..35506.69 rows=1 width=8) (actual time=68.712..68.714 rows=1 loops=1)
   ->  Bitmap Heap Scan on account  (cost=908.06..35269.24 rows=94975 width=0) (actual time=11.153..61.405 rows=100000 loops=1)
         Recheck Cond: (content @> '{"tags": ["c"]}'::jsonb)
         Heap Blocks: exact=2041
         ->  Bitmap Index Scan on ix_account_content  (cost=0.00..884.31 rows=94975 width=0) (actual time=10.838..10.838 rows=100000 loops=1)
               Index Cond: (content @> '{"tags": ["c"]}'::jsonb)
 Planning Time: 0.257 ms
 Execution Time: 68.777 ms
(8 rows)

content 列上的简单 GIN 索引(默认索引)就能支持索引查询。 但是索引将会存储content列中每一个键 和值的拷贝表达式索引只存储tags 键下找到的数据。

虽然简单索引的方法更加灵活(因为它支持有关任意键的查询),但定向的表达式索引更小并且搜索速度比简单索引更快。 尽管jsonb_path_ops操作符类只支持用 @>操作符的查询,但它比起默认的操作符类 jsonb_ops有更客观的性能优势。一个 jsonb_path_ops索引通常也比一个相同数据上的 jsonb_ops要小得多,并且搜索的专一性更好,特 别是当查询包含频繁出现在该数据中的键时。因此,其上的搜索操作 通常比使用默认操作符类的搜索表现更好。

总结

  • PG 有两种 JSON 数据类型:jsonjsonb,jsonb 性能优于json,且jsonb 支持索引。
  • jsonb 写入时会处理写入数据,写入相对较慢,json会保留原始数据(包括无用的空格)
  • jsonb 查询优化时一个好的方式是添加GIN 索引
    • 简单索引和路径索引相比更灵活,但是占用空间多
    • 路径索引比简单索引更高效,占用空间更小

PostgreSQL json 索引实践 - 检索(存在、包含、等值、范围等)加速

背景

用户在使用JSON类型时,常见的一些JSON搜索包括:

1、存在,JSON中是否存在某个KEY,某些KEY,某些KEY的任意一个

-- 存在某个KEY(TOP LEVEL)
'{"a":1, "b":2}'::jsonb ? 'b'

-- 存在所有KEY
'{"a":1, "b":2, "c":3}'::jsonb ?& array['b', 'c'] 

-- 存在任意key、元素
'["a", "b"]'::jsonb ?| array['a', 'b']  

2、等值,JSON中是否存在指定的key:value对(支持嵌套JSON)

'{"a":1, "b":2}'::jsonb @> '{"b":2}'::jsonb 

3、包含,JSON中某个路径下的VALUE(数组)中,是否包含指定的所有元素。

postgres=# select jsonb '{"a":1, "b": {"c":[1,2,3], "d":["k","y","z"]}, "d":"kbc"}' @> '{"b":{"c":[2,3]}}';
 ?column? 
----------
 t
(1 row)

4、相交,JSON中某个路径下的VALUE(数组)中,是否包含指定的任意元素。

postgres=# select jsonb '{"a":1, "b": {"c":[1,2,3], "d":["k","y","z"]}, "d":"kbc"}' @> '{"b":{"c":[2]}}' or jsonb '{"a":1, "b": {"c":[1,2,3], "d":["k","y","z"]}, "d":"kbc"}' @> '{"b":{"c":[3]}}'; 
 ?column? 
----------
 t
(1 row)

或(注意1,2,3需要双引号,作为text类型存储,因为操作符?| ?&暂时只支持了text[],如果是numeric匹配不上)

postgres=# select jsonb '{"a":1, "b": {"c":["1","2","3"], "d":["k","y","z"]}, "d":"kbc"}' -> 'b' -> 'c' ?& array['2','3','4'];
 ?column? 
----------
 f
(1 row)

postgres=# select jsonb '{"a":1, "b": {"c":["1","2","3"], "d":["k","y","z"]}, "d":"kbc"}' -> 'b' -> 'c' ?| array['2','3','4'];
 ?column? 
----------
 t
(1 row)

5、范围查找,JSON中某个路径下的VALUE,是否落在某个范围内。

(js ->> 'key1' )::numeric between xx and xx
  
(js ->> 'key2' )::numeric between xx and xx

这些操作如何加速,或者如何使用索引加速?

一、json 索引支持

GIN的两个OPS,分别支持JSON:

The default GIN operator class for jsonb supports queries with top-level key-exists operators ?, ?& and ?|operators and path/value-exists operator @>.

The non-default GIN operator class jsonb_path_ops supports indexing the @> operator only.

1、支持 @> 操作符的索引如下(jsonb_path_ops只支持 @> 操作符,但是效率高)

postgres=# create table tbl(id int, js jsonb);
CREATE TABLE  
postgres=# create index idx_tbl_1 on tbl using gin (js jsonb_path_ops);
CREATE INDEX 

2、支持除范围查询以外的所有查询的索引如下

postgres=# create table tbl(id int, js jsonb);
CREATE TABLE  
postgres=# create index idx_tbl_1 on tbl using gin (js);  -- 使用默认ops即可  
CREATE INDEX

二、JSON KEY VALUE值范围查询加速

某些使用,需要对VALUE使用范围查询,比如时间(如果要建索引,请使用numeric表示,否则需要自定义immutable函数),数值都有这些需求。

通常的做法,把范围查询的类型提取出来,创建btree表达式索引,如果有任意组合的范围查询,使用gin或rum表达式索引。

例子

create index idx1 on tbl ( ((js->>'k1')::float8) );  
create index idx2 on tbl ( ((js->>'k2')::numeric) );  
...  
create index idxn on tbl ( ((js->>'kn')::float8) );

create extension btree_gin;
create index idx1 on tbl using gin( ((js->>'k1')::float8), ((js->>'k2')::numeric), ... ((js->>'kn')::float8) );

create extension rum;
create index idx1 on tbl using rum( ((js->>'k1')::float8), ((js->>'k2')::numeric), ... ((js->>'kn')::float8) );

create or replace function to_timestamp(text) returns timestamp as $$  
  select $1::timestamp;  
$$ language sql strict immutable;  
  
  
create index idx1 on tbl using gin( ((js->>'k1')::float8), to_timestamp(js->>'k2'), ... ((js->>'kn')::float8) );  
-- 或  
create index idx1 on tbl using rum( ((js->>'k1')::float8), to_timestamp(js->>'k2'), ... ((js->>'kn')::float8) ); 

三、索引使用例子

rum插件说明

#代码仓
https://github.com/postgrespro/rum/releases

$ git clone https://github.com/postgrespro/rum
$ cd rum
$ make USE_PGXS=1
$ make USE_PGXS=1 install
$ make USE_PGXS=1 installcheck
$ psql DB -c "CREATE EXTENSION rum;"

创建测试表和索引

create or replace function to_timestamp(text) returns timestamp as $$
  select $1::timestamp;
$$ language sql strict immutable;

create table tbl(id int, js jsonb);
create index idx_tbl_1 on tbl using gin (js jsonb_path_ops);
create index idx_tbl_2 on tbl using gin (js);
create index idx_tbl_3 on tbl using rum( ((js->>'k1')::float8), to_timestamp(js->>'k2'), ((js->>'k3')::numeric) );

? 操作符,匹配JSON对象中的键走索引:

postgres=# explain analyze select * from tbl where js ? 'a';
                                                    QUERY PLAN                                                    
------------------------------------------------------------------------------------------------------------------
 Bitmap Heap Scan on tbl  (cost=8.01..12.02 rows=1 width=36) (actual time=0.031..0.033 rows=0 loops=1)
   Recheck Cond: (js ? 'a'::text)
   ->  Bitmap Index Scan on idx_tbl_2  (cost=0.00..8.01 rows=1 width=0) (actual time=0.023..0.025 rows=0 loops=1)
         Index Cond: (js ? 'a'::text)
 Planning Time: 0.581 ms
 Execution Time: 0.133 ms
(6 rows)

@> 操作符,匹配JSON对象中的键值对走索引:

postgres=# explain analyze select * from tbl where js @> '{"a":"b"}';
                                                    QUERY PLAN                                                    
------------------------------------------------------------------------------------------------------------------
 Bitmap Heap Scan on tbl  (cost=8.01..12.02 rows=1 width=36) (actual time=0.017..0.017 rows=0 loops=1)
   Recheck Cond: (js @> '{"a": "b"}'::jsonb)
   ->  Bitmap Index Scan on idx_tbl_1  (cost=0.00..8.01 rows=1 width=0) (actual time=0.014..0.014 rows=0 loops=1)
         Index Cond: (js @> '{"a": "b"}'::jsonb)
 Planning Time: 0.132 ms
 Execution Time: 0.046 ms
(6 rows)

@> 操作符,嵌套数组使用索引

postgres=# select * from tbl where js @> '{"a": {"b":"c"}}';
 id | js 
----+----
(0 rows)


SELECT doc->'site_name' FROM websites WHERE doc->'tags' @> '[{"term":"paris"}, {"term":"food"}]';  


postgres=# select jsonb '{"a":1, "b": {"c":[1,2,3], "d":["k","y","z"]}, "d":"kbc"}';
                               jsonb                               
-------------------------------------------------------------------
 {"a": 1, "b": {"c": [1, 2, 3], "d": ["k", "y", "z"]}, "d": "kbc"}
(1 row)

postgres=# select jsonb '{"a":1, "b": {"c":[1,2,3], "d":["k","y","z"]}, "d":"kbc"}' @> '{"b":{"c":[2,3]}}';
 ?column? 
----------
 t
(1 row)

postgres=# select jsonb '{"a":1, "b": {"c":[1,2,3], "d":["k","y","z"]}, "d":"kbc"}' @> '{"b":{"c":[2,4]}}';
 ?column? 
----------
 f
(1 row)

postgres=# explain analyze select * from tbl where js @> '{"b":{"c":[2,4]}}';
                                                    QUERY PLAN                                                     
-------------------------------------------------------------------------------------------------------------------
 Bitmap Heap Scan on tbl  (cost=12.01..16.02 rows=1 width=36) (actual time=0.012..0.013 rows=0 loops=1)
   Recheck Cond: (js @> '{"b": {"c": [2, 4]}}'::jsonb)
   ->  Bitmap Index Scan on idx_tbl_1  (cost=0.00..12.01 rows=1 width=0) (actual time=0.010..0.010 rows=0 loops=1)
         Index Cond: (js @> '{"b": {"c": [2, 4]}}'::jsonb)
 Planning Time: 0.111 ms
 Execution Time: 0.041 ms
(6 rows)

rum扩展将键的文本值转为timestamp类型,范围查询走索引:

explain analyze select * from tbl where to_timestamp(js->>'k2') between '2018-01-01' and '2018-01-02';
                                                                                                  QUERY PLAN                                    
                                                               
------------------------------------------------------------------------------------------------------------------------------------------------
---------------------------------------------------------------
 Bitmap Heap Scan on tbl  (cost=24.07..36.58 rows=6 width=36) (actual time=0.040..0.041 rows=0 loops=1)
   Recheck Cond: ((to_timestamp((js ->> 'k2'::text)) >= '2018-01-01 00:00:00'::timestamp without time zone) AND (to_timestamp((js ->> 'k2'::text
)) <= '2018-01-02 00:00:00'::timestamp without time zone))
   ->  Bitmap Index Scan on idx_tbl_3  (cost=0.00..24.06 rows=6 width=0) (actual time=0.037..0.037 rows=0 loops=1)
         Index Cond: ((to_timestamp((js ->> 'k2'::text)) >= '2018-01-01 00:00:00'::timestamp without time zone) AND (to_timestamp((js ->> 'k2'::
text)) <= '2018-01-02 00:00:00'::timestamp without time zone))
 Planning Time: 0.569 ms
 Execution Time: 0.090 ms
(6 rows)


explain analyze select * from tbl where to_timestamp(js->>'k2') between '2018-01-01' and '2018-01-02' and ((js->>'k3')::numeric) between 1 and 200;
                                                                                                                                                
     QUERY PLAN                                                                                                                                 
                     
------------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------------------
---------------------
 Index Scan using idx_tbl_3 on tbl  (cost=36.00..44.01 rows=1 width=36) (actual time=0.033..0.034 rows=0 loops=1)
   Index Cond: ((to_timestamp((js ->> 'k2'::text)) >= '2018-01-01 00:00:00'::timestamp without time zone) AND (to_timestamp((js ->> 'k2'::text))
 <= '2018-01-02 00:00:00'::timestamp without time zone) AND (((js ->> 'k3'::text))::numeric >= '1'::numeric) AND (((js ->> 'k3'::text))::numeric
 <= '200'::numeric))
 Planning Time: 0.299 ms
 Execution Time: 0.062 ms
(4 rows)

参考资料

官方文档:

https://www.postgresql.org/docs/current/functions-json.html
https://www.postgresql.org/docs/current/datatype-json.html

PostgreSQL JSON数据类型
https://blog.csdn.net/neweastsun/article/details/93345799

PostgreSQL JSON类型字段常用操作
https://blog.csdn.net/wilsonpeng3/article/details/128677263

JSON & JSONB 数据类型操作

https://help.aliyun.com/zh/analyticdb-for-postgresql/developer-reference/operations-of-json-data

PostgreSQL JSONB 使用入门

https://cloud.tencent.com/developer/article/1763846

PostgreSQL高级数据类型JSON和JSONB

https://bbs.huaweicloud.com/blogs/363682

https://emacsist.github.io/2016/10/09/postgresql中的json与jsonb/

PostgreSQL JSON函数和操作符

https://www.w3cschool.cn/postgresql13_1/postgresql13_1-n3ha3jbh.html

PostgreSQL操作JSON数据

https://blog.51cto.com/u_14441472/10771220

PostgreSQL json 索引实践 - 检索(存在、包含、等值、范围等)加速

https://billtian.github.io/digoal.blog/2018/07/31/01.html

  • 6
    点赞
  • 9
    收藏
    觉得还不错? 一键收藏
  • 0
    评论
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值