postgresql-JSON使用
json,jsonb区别
json,jsonb区别
json和jsonb，而两者唯一的区别在于效率,json是对输入的完整拷贝，使用时再去解析，所以它会保留输入的空格，重复键以及顺序等。而jsonb是解析输入后保存的二进制，它在解析时会删除不必要的空格和重复的键，顺序和输入可能也不相同。使用时不用再次解析。两者对重复键的处理都是保留最后一个键值对。效率的差别：json类型存储快，查询慢，jsonb类型存储稍慢，查询较快(支持许多额外的操作符)。
关于json和jsonb存储和解析效率连接
这里主要测试jsonb的增删改查
json和jsonb共同操作符
| 操作符 | 返回类型 | 数组[1,2,3] | {"a":{"b":{"c":1}},"d":[4,5,6]} | |
|---|---|---|---|---|
| -> | json | select '[1,2,3]'::jsonb ->2 = 3 | select '{"a":1,"b":2,"c":3}'::jsonb-> 'a'=1 | select '{"a":{"b":{"c":1}},"d":[4,5,6]}'::jsonb ->'a'={"b": {"c": 1}} |
| ->> | text | select '[1,2,3]'::jsonb ->>2 = 3 | select '{"a":1,"b":2,"c":3}'::jsonb->> 'a'=1 | select '{"a":{"b":{"c":1}},"d":[4,5,6]}'::jsonb ->>'a'={"b": {"c": 1}} |
| #> | json | -- | -- | select '{"a":{"b":{"c":1}},"d":[4,5,6]}'::jsonb #> '{a,b}' = |
| #>> | text | -- | -- | select '{"a":{"b":{"c":1}},"d":[4,5,6]}'::jsonb #> '{a,b}' = |
jsonb额外操作符
| 操作符 | 右操作数类型 | 描述 | 例子 |
|---|---|---|---|
| @> | jsonb | 左边的 JSON 值是否包含顶层右边JSON路径/值项? | '{"a":1, "b":2}'::jsonb @> '{"b":2}'::jsonb |
| <@ | jsonb | 左边的JSON路径/值是否包含在顶层右边JSON值中？ | '{"b":2}'::jsonb <@ '{"a":1, "b":2}'::jsonb |
| ? | text | 字符串是否作为顶层键值存在于JSON值中？ | '{"a":1, "b":2}'::jsonb ? 'b' |
| ?| | text[] | 这些数组字符串中的任何一个是否作为顶层键值存在？ | '{"a":1, "b":2, "c":3}'::jsonb ?|array['b',c'] |
| ?& | text[] | 这些数组字符串是否作为顶层键值存在？ | '["a", "b"]'::jsonb ?& array['a', 'b'] |
| || | jsonb | 连接两个jsonb值到新的jsonb值 | '["a", "b"]'::jsonb| '["c", "d"]'::jsonb |
| - | text | 从左操作数中删除键/值对或字符串元素。基于键值匹配键/值对。 | '{"a": "b"}'::jsonb - 'a' |
| - | integer | 删除指定索引的数组元素（负整数结尾）。如果顶层容器不是一个数组，那么抛出错误。 | '["a", "b"]'::jsonb - 1 |
| #- | text[] | 删除指定路径的域或元素（JSON数组，负整数结尾） | '["a", {"b":1}]'::jsonb #- '{1,b}' |
jsonb增删改
--1.1建表
abase=> create table test_jsonb(c_bh char(32),j_jsonb jsonb);
CREATE TABLE
--插入数据
insert into test_jsonb(c_bh,j_jsonb) values(replace(uuid_generate_v4()::text,'-',''),'{"c_xm":"张三","c_mx":{"c_ssdw":"一大队","c_dwbm":"11"}}');
INSERT 0 1
--查看数据
abase=# select * from test_jsonb where j_jsonb @> '{"c_xm":"张三","c_mx":{"c_ssdw":"一大队","c_dwbm":"11"}}';
c_bh | j_jsonb
----------------------------------+--------------------------------------------
c217c624152943ab93f502117514f432 | {"c_mx": {"c_dwbm": "11", "c_ssdw": "一大队"}, "c_xm": "张三"}
(1 row)
--1.2操作符||可用于添加元素，添加元素'{"c_id":"111"}'
abase=# update test_jsonb set j_jsonb = j_jsonb ||'{"c_id":"111"}'::jsonb where c_bh = 'c217c624152943ab93f502117514f432';
UPDATE 1
abase=# select j_jsonb from test_jsonb where c_bh = 'c217c624152943ab93f502117514f432';
j_jsonb
-------------------------------------------------------------------------------
{"c_id": "111", "c_mx": {"c_dwbm": "11", "c_ssdw": "一大队"}, "c_xm": "张三"}
(1 row)
--1.3更新元素（方法1），如果jsonb中有相同的元素则覆盖,使用'||'将'{"c_id":"111"}'更新为112
abase=# update test_jsonb set j_jsonb = j_jsonb ||'{"c_id":"112"}'::jsonb where c_bh = 'c217c624152943ab93f502117514f432';
UPDATE 1
abase=# select j_jsonb from test_jsonb where c_bh = 'c217c624152943ab93f502117514f432';
j_jsonb
-------------------------------------------------------------------------------
{"c_id": "112", "c_mx": {"c_dwbm": "11", "c_ssdw": "一大队"}, "c_xm": "张三"}
(1 row)
--更新元素（方法2），使用jsonb_set，将"c_id": "112"更新为123
abase=# update test_jsonb set j_jsonb= jsonb_set(j_jsonb,'{c_id}','"123"'::jsonb,false) where c_bh = 'c217c624152943ab93f502117514f432';
UPDATE 1
abase=# select j_jsonb from test_jsonb where c_bh = 'c217c624152943ab93f502117514f432';
j_jsonb
-------------------------------------------------------------------------------
{"c_id": "123", "c_mx": {"c_dwbm": "11", "c_ssdw": "一大队"}, "c_xm": "张三"}
(1 row)
--1.4更新嵌套元素，使用jsonb_set（pg9.5以上才支持），更新c_ssdw为二大队
abase=# update test_jsonb set j_jsonb= jsonb_set(j_jsonb,'{c_mx,c_ssdw}','"二大队"'::jsonb,false) where c_bh = 'c217c624152943ab93f502117514f432';
UPDATE 1
abase=# select j_jsonb from test_jsonb where c_bh = 'c217c624152943ab93f502117514f432';
j_jsonb
-------------------------------------------------------------------------------
{"c_id": "123", "c_mx": {"c_dwbm": "11", "c_ssdw": "二大队"}, "c_xm": "张三"}
(1 row)
--1.5删除元素，删除c_id元素
abase=# update test_jsonb set j_jsonb = j_jsonb-'c_id' where c_bh = 'c217c624152943ab93f502117514f432' ;
UPDATE 1
abase=# select j_jsonb from test_jsonb where c_bh = 'c217c624152943ab93f502117514f432';
j_jsonb
----------------------------------------------------------------
{"c_mx": {"c_dwbm": "11", "c_ssdw": "二大队"}, "c_xm": "张三"}
(1 row)
jsonb查询
--1.随机文本脚本
abase=> create or replace function random_string(INTEGER)
abase-> RETURNS TEXT AS
abase-> $BODY$
abase$> select array_to_string(
abase$> array(
abase$> select substring(
abase$> 'pg社区的作风非常严谨，一个补丁可能在邮件组中讨论几个月甚至几年，根据大家的意见反复的修正，补丁合并到主干已经非常成熟，所以pg的稳定性也是远近闻名的'
abase$> from (ceil(random()*73))::int FOR 2
abase$> )
abase$> from generate_series(1,$1)
abase$> ),''
abase$> ) $BODY$
abase-> LANGUAGE sql VOLATILE;
CREATE FUNCTION
--2.初始化数据：
abase=> insert into test_jsonb select replace(uuid_generate_v4()::text,'-',''),('{"a":'||random()*100||', "kxhbsl":"'|| random_string(10) ||'"}')::jsonb from generate_series(1,2000000);
INSERT 0 2000000
abase=> insert into test_jsonb select replace(uuid_generate_v4()::text,'-',''),('{"a":'||random()*100||', "kxhbsl":"索尼是大法官"}')::jsonb from generate_series(1,10000);
INSERT 0 10000
--3.第一种查询：获取包含'{"kxhbsl": "索尼是大法官"}'，全表扫描
abase=# explain analyze select j_jsonb->>'kxhbsl',j_jsonb from test_jsonb where j_jsonb @> '{"kxhbsl": "索尼是大法官"}';
QUERY PLAN
-------------------------------------------------------------------------------------------------------------
Gather (cost=1000.00..53379.78 rows=2010 width=134) (actual time=470.729..490.979 rows=10000 loops=1)
Workers Planned: 2
Workers Launched: 2
-> Parallel Seq Scan on test_jsonb (cost=0.00..52175.85 rows=838 width=134) (actual time=465.234..480.57
3 rows=3333 loops=3)
Filter: (j_jsonb @> '{"kxhbsl": "索尼是大法官"}'::jsonb)
Rows Removed by Filter: 666667
Planning time: 0.318 ms
Execution time: 506.204 ms
(8 rows)
--j_jsonb字段创建gin索引后，可走索引
abase=# create index i_t_test_jsonb_j_jsonb on test_jsonb using gin(j_jsonb);
CREATE INDEX
abase=# explain analyze select j_jsonb->>'kxhbsl',* from test_jsonb where j_jsonb @> '{"kxhbsl": "索尼是大法官"}';
QUERY PLAN
-------------------------------------------------------------------------------------------------------------
Bitmap Heap Scan on test_jsonb (cost=59.58..6664.09 rows=2010 width=167) (actual time=3.579..17.065 rows=10
000 loops=1)
Recheck Cond: (j_jsonb @> '{"kxhbsl": "索尼是大法官"}'::jsonb)
Heap Blocks: exact=481
-> Bitmap Index Scan on i_t_test_jsonb_j_jsonb (cost=0.00..59.08 rows=2010 width=0) (actual time=3.480..
3.480 rows=10000 loops=1)
Index Cond: (j_jsonb @> '{"kxhbsl": "索尼是大法官"}'::jsonb)
Planning time: 0.429 ms
Execution time: 17.964 ms
(7 rows)
--4.第二种查询，获取包含:'索尼是大法官'，全表扫描
abase=# explain analyze select j_jsonb->>'kxhbsl',j_jsonb from test_jsonb where j_jsonb -> 'kxhbsl' ? '索尼是大法官';
QUERY PLAN
-------------------------------------------------------------------------------------------------------------
Gather (cost=1000.00..55473.53 rows=2010 width=134) (actual time=1724.170..1769.543 rows=10000 loops=1)
Workers Planned: 2
Workers Launched: 0
-> Parallel Seq Scan on test_jsonb (cost=0.00..54269.60 rows=838 width=134) (actual time=1723.752..1767.
187 rows=10000 loops=1)
Filter: ((j_jsonb -> 'kxhbsl'::text) ? '索尼是大法官'::text)
Rows Removed by Filter: 2000000
Planning time: 0.267 ms
Execution time: 1770.422 ms
(8 rows)
--针对jsonb字段的kxhbsl元素创建gin索引。 可走索引
abase=# create index i_t_test_jsonb_j_jsonb_kxhbsl on test_jsonb using gin((j_jsonb->'kxhbsl'));
CREATE INDEX
abase=# explain analyze select j_jsonb->>'kxhbsl',j_jsonb from test_jsonb where j_jsonb -> 'kxhbsl' ? '索尼是大法官';
QUERY PLAN
-------------------------------------------------------------------------------------------------------------
Bitmap Heap Scan on test_jsonb (cost=39.58..6649.12 rows=2010 width=134) (actual time=2.166..13.999 rows=10
000 loops=1)
Recheck Cond: ((j_jsonb -> 'kxhbsl'::text) ? '索尼是大法官'::text)
Heap Blocks: exact=481
-> Bitmap Index Scan on i_t_test_jsonb_j_jsonb_kxhbsl (cost=0.00..39.08 rows=2010 width=0) (actual time=
2.045..2.045 rows=10000 loops=1)
Index Cond: ((j_jsonb -> 'kxhbsl'::text) ? '索尼是大法官'::text)
Planning time: 0.221 ms
Execution time: 14.715 ms
(7 rows)
--或者等价写法：
abase=# explain analyze select j_jsonb->>'kxhbsl',j_jsonb from test_jsonb where j_jsonb -> 'kxhbsl' @>'"索尼是大法官"';
QUERY PLAN
-------------------------------------------------------------------------------------------------------------
Bitmap Heap Scan on test_jsonb (cost=39.58..6649.12 rows=2010 width=134) (actual time=2.080..14.959 rows=10
000 loops=1)
Recheck Cond: ((j_jsonb -> 'kxhbsl'::text) @> '"索尼是大法官"'::jsonb)
Heap Blocks: exact=481
-> Bitmap Index Scan on i_t_test_jsonb_j_jsonb_kxhbsl (cost=0.00..39.08 rows=2010 width=0) (actual time=
1.980..1.980 rows=10000 loops=1)
Index Cond: ((j_jsonb -> 'kxhbsl'::text) @> '"索尼是大法官"'::jsonb)
Planning time: 0.199 ms
Execution time: 15.635 ms
(7 rows)
--5.第三种查询，获取'{"kxhbsl": "索尼是大法官"}'，全表扫描
abase=# explain analyze select * from test_jsonb where j_jsonb->>'kxhbsl' = '索尼是大法官';
QUERY PLAN
-------------------------------------------------------------------------------------------------------------
Gather (cost=1000.00..56272.50 rows=10050 width=135) (actual time=458.676..476.454 rows=10000 loops=1)
Workers Planned: 2
Workers Launched: 2
-> Parallel Seq Scan on test_jsonb (cost=0.00..54267.50 rows=4188 width=135) (actual time=453.472..466.5
44 rows=3333 loops=3)
Filter: ((j_jsonb ->> 'kxhbsl'::text) = '索尼是大法官'::text)
Rows Removed by Filter: 666667
Planning time: 0.821 ms
Execution time: 492.763 ms
(8 rows)
--针对这类查询，j_jsonb->>'kxhbsl'返回类型为text，那么可以考虑创建一个btree索引，也可以走索引
abase=# create index i_test_jsonb_j_jsonb_btree on test_jsonb using btree((j_jsonb ->> 'kxhbsl') );
CREATE INDEX
abase=# explain analyze select * from test_jsonb where j_jsonb->>'kxhbsl' = '索尼是大法官';
QUERY PLAN
-------------------------------------------------------------------------------------------------------------
Bitmap Heap Scan on test_jsonb (cost=498.44..24049.15 rows=10050 width=135) (actual time=4.150..8.168 rows=
10000 loops=1)
Recheck Cond: ((j_jsonb ->> 'kxhbsl'::text) = '索尼是大法官'::text)
Heap Blocks: exact=481
-> Bitmap Index Scan on i_test_jsonb_j_jsonb_btree (cost=0.00..495.93 rows=10050 width=0) (actual time=4
.042..4.042 rows=10000 loops=1)
Index Cond: ((j_jsonb ->> 'kxhbsl'::text) = '索尼是大法官'::text)
Planning time: 0.684 ms
Execution time: 8.991 ms
(7 rows)
--6.由于j_jsonb->>'kxhbsl'返回为text类型，所以可在其上面做许多操作，比如in，exists等
--查看执行计划，in查询：
abase=# explain analyze select * from test_jsonb where j_jsonb->>'kxhbsl' in ('索尼是大法官','3');
QUERY PLAN
-------------------------------------------------------------------------------------------------------------
Bitmap Heap Scan on test_jsonb (cost=992.88..35800.76 rows=20100 width=135) (actual time=2.666..5.992 rows=
10000 loops=1)
Recheck Cond: ((j_jsonb ->> 'kxhbsl'::text) = ANY ('{索尼是大法官,3}'::text[]))
Heap Blocks: exact=481
-> Bitmap Index Scan on i_test_jsonb_j_jsonb_btree (cost=0.00..987.86 rows=20100 width=0) (actual time=2
.576..2.576 rows=10000 loops=1)
Index Cond: ((j_jsonb ->> 'kxhbsl'::text) = ANY ('{索尼是大法官,3}'::text[]))
Planning time: 0.360 ms
Execution time: 6.856 ms
(7 rows)
三种查询都能得到相同的结果，可以看出第一种针对于jsonb字段的gin索引，适用于jsonb字段所有的元素，而第二种和第三种分别是对单个元素创建的gin和btree索引。
等值查询方面可能单个元素的btree索引占用空间小，且效率较高，如果单独某个元素的查询较为频繁可选择btree索引，而整个jsonb创建gin对所有元素有效。
第一种传入的是一个json，而第二种，第三种传入的是字符串
jsonb元素值模糊匹配
--1.有时候需要对jsonb的元素值进行模糊匹配
--在前面只有j_jsonb gin索引情况下，like全模糊匹配不能走索引
abase=# explain analyze select * from test_jsonb where j_jsonb->>'kxhbsl' like '%大法官%';
QUERY PLAN
-------------------------------------------------------------------------------------------------------------
Gather (cost=1000.00..55287.60 rows=201 width=135) (actual time=832.031..857.306 rows=10000 loops=1)
Workers Planned: 2
Workers Launched: 2
-> Parallel Seq Scan on test_jsonb (cost=0.00..54267.50 rows=84 width=135) (actual time=826.065..844.494
rows=3333 loops=3)
Filter: ((j_jsonb ->> 'kxhbsl'::text) ~~ '%大法官%'::text)
Rows Removed by Filter: 666667
Planning time: 0.314 ms
Execution time: 873.938 ms
(8 rows)
--由于(j_jsonb ->>'kxhbsl')返回的是text类型，所以考虑再其上面使用pg_trgm,创建gin索引。
abase=# create index i_test_jsonb_j_jsonb_gin on test_jsonb using gin((j_jsonb ->>'kxhbsl') gin_trgm_ops);
CREATE INDEX
--查看执行计划，模糊匹配可走索引。
abase=# explain analyze select * from test_jsonb where j_jsonb->>'kxhbsl' like '%大法官%';
QUERY PLAN
-------------------------------------------------------------------------------------------------------------
Bitmap Heap Scan on test_jsonb (cost=17.56..782.71 rows=201 width=135) (actual time=3.781..16.256 rows=1000
0 loops=1)
Recheck Cond: ((j_jsonb ->> 'kxhbsl'::text) ~~ '%大法官%'::text)
Heap Blocks: exact=481
-> Bitmap Index Scan on i_test_jsonb_j_jsonb_gin (cost=0.00..17.51 rows=201 width=0) (actual time=3.649.
.3.649 rows=10000 loops=1)
Index Cond: ((j_jsonb ->> 'kxhbsl'::text) ~~ '%大法官%'::text)
Planning time: 0.575 ms
Execution time: 17.514 ms
(7 rows)
--2.当然还有一种方式就是将该jsonb字段转为text，然后再创建gin索引
--创建gin索引
abase=#create index i_jsonb_ops on test_jsonb using gin ((j_jsonb::text) gin_trgm_ops);
CREATE INDEX
--但是这样的模糊匹配，可能匹配到其他元素中包含同样的值，所以需要加上辅助条件：j_jsonb->>'kxhbsl' like '%索尼是大法官%'，用来确保是该元素
abase=# explain analyze select * from test_jsonb where j_jsonb->>'kxhbsl' like '%大法官%' and j_jsonb ::text like '%大法官%';
QUERY PLAN
-------------------------------------------------------------------------------------------------------------
Bitmap Heap Scan on test_jsonb (cost=1297.51..2064.17 rows=1 width=135) (actual time=5.318..28.149 rows=100
00 loops=1)
Recheck Cond: ((j_jsonb)::text ~~ '%大法官%'::text)
Filter: ((j_jsonb ->> 'kxhbsl'::text) ~~ '%大法官%'::text)
Heap Blocks: exact=481
-> Bitmap Index Scan on i_jsonb_ops (cost=0.00..1297.51 rows=201 width=0) (actual time=5.198..5.198 rows
=10000 loops=1)
Index Cond: ((j_jsonb)::text ~~ '%大法官%'::text)
Planning time: 0.479 ms
Execution time: 29.147 ms
(8 rows)
第二种方法效率相对于第一种要低一点，但是所有元素都可使用
结语
1.在json和jsonb选择上，json更加适合用于存储，jsonb更加适用于检索。
2.可以对整个jsonb字段创建gin索引，同时也可以对jsonb中某个元素创建gin索引，或者btree。btree效率最高。
3.(j_jsonb ->> 'kxhbsl')返回的是一个text类型，所以可以在该属性上创建对应类型的索引，比如btree，gin索引。
4.对于元素值的模糊匹配可以创建单个元素的gin索引，也可以创建整个jsonb字段的gin索引，前者效率较高，后者适用所有元素。
