需求背景
需求
- 一张数据表
- 其中有sid字段,代表tag,每行数据可能有多个tag字段
- 统计全量数据中所有tag出现的次数(按tag分组,分别有多少数据)
source table demo
id |
sid |
1 |
a3,a4,a1,a2,a5 |
2 |
a5,a3,a4,a2 |
3 |
a5,a3,a4 |
target table demo
sid_tag |
sid_occurrence |
a1 |
1 |
a2 |
2 |
a3 |
3 |
a4 |
3 |
a5 |
3 |
SQL query in mysql
select substring_index( substring_index( sid , ',', id_table.help_topic_id + 1 ), ',',- 1 ) as sid_tag,count(sid) as sid_occurrence from src_t JOIN mysql.help_topic id_table on id_table.help_topic_id < (
length(src_t.sid) - length(replace(src_t.sid, ',', '')) + 1
) GROUP BY sid_tag
hive
select tag,count(*) from(
select tag from table_name
lateral view explode(split(taglist,',')) r1 AS tag) a group by a.tag;
解析
- 利用系统表mysql.help_topic进行了index的记录(可以使用其他表),更暴力的解法是直接新建一张只有id列的表,可以预防系统表id不够用(一般足够了)
- JOIN a.id < b.id,解决了复制行的操作;若b.id = 5,a.id 为(1,2,3,4,5),返回数据为 [(1,5),(2,5),(3,5),(4,5)]
- substring_index( substring_index( sid , ',', id_table.help_topic_id + 1 ), ',',- 1 ) 利用index表中的index,作为子串索引进行数据拆分。索引构建基于上一步的join。
- 随后正常groupby统计
实验
mysql
select * from (
SELECT concat('a',topic.help_topic_id) as sid,topic.help_topic_id AS id
FROM mysql.help_topic topic
WHERE help_topic_id IN ('1', '2', '3', '4', '5')
) as id_table JOIN (
SELECT concat('b',topic.help_topic_id) as sid,topic.help_topic_id AS id
FROM mysql.help_topic topic
WHERE help_topic_id IN ('1', '2', '3')
) as id_table2 on id_table2.id <= id_table.id
结果
sid |
id |
sid |
id |
a1 |
1 |
b1 |
1 |
a2 |
2 |
b1 |
1 |
a2 |
2 |
b2 |
2 |
a3 |
3 |
b1 |
1 |
a3 |
3 |
b2 |
2 |
a3 |
3 |
b3 |
3 |
a4 |
4 |
b1 |
1 |
a4 |
4 |
b2 |
2 |
a4 |
4 |
b3 |
3 |
a5 |
5 |
b1 |
1 |
a5 |
5 |
b2 |
2 |
a5 |
5 |
b3 |
3 |