我们经常困惑在数据挖掘和报表分析场景中sql不会写,或者因为sql太长以至于可读性降低; 今天我为大家总结了一些Spark SQL中的高阶函数,它们将会对你的业务形成助力,百倍提升你的工作效率

GROUPING__ID,CUBE,ROLLUP

可快速实现多维度自由组合分析查询,主要应用于OLAP钻取分析场景,比如,分小时、天、月的UV数。

  • cube
    cube函数 多用来实现钻取查询
    将一个group by中单一维度分组后进行聚合,等价于将不同维度的GROUP BY结果集进行UNION ALL
select * from eqs_1234;

month      day        GROUPING__ID
------------------------------------
2015-03 2015-03-10      cookie1
2015-03 2015-03-10      cookie5
2015-03 2015-03-12      cookie7
2015-04 2015-04-12      cookie3
2015-04 2015-04-13      cookie2
2015-04 2015-04-13      cookie4
2015-04 2015-04-16      cookie4
2015-03 2015-03-10      cookie2
2015-03 2015-03-10      cookie3
2015-04 2015-04-12      cookie5
2015-04 2015-04-13      cookie6
2015-04 2015-04-15      cookie3
2015-04 2015-04-15      cookie2
2015-04 2015-04-16      cookie1

SELECT 
month,
day,
COUNT(DISTINCT cookieid) AS uv 
FROM eqs_1234
GROUP BY  cube (month,day,(month,day))  

month         day             uv      GROUPING__ID
------------------------------------------------
2015-03       NULL            5       1
2015-04       NULL            6       1
NULL          2015-03-10      4       2
NULL          2015-03-12      1       2
NULL          2015-04-12      2       2
NULL          2015-04-13      3       2
NULL          2015-04-15      2       2
NULL          2015-04-16      2       2
2015-03       2015-03-10      4       3
2015-03       2015-03-12      1       3
2015-04       2015-04-12      2       3
2015-04       2015-04-13      3       3
2015-04       2015-04-15      2       3
2015-04       2015-04-16      2       3

## 如果不知道cube函数,那么可能会用下面的方式来实现,SQL的可读性和性能大大降低
SELECT month,NULL,COUNT(DISTINCT cookieid) AS uv,1 AS GROUPING__ID FROM lxw1234 GROUP BY month 
UNION ALL 
SELECT NULL,day,COUNT(DISTINCT cookieid) AS uv,2 AS GROUPING__ID FROM lxw1234 GROUP BY day
UNION ALL 
SELECT month,day,COUNT(DISTINCT cookieid) AS uv,3 AS GROUPING__ID FROM lxw1234 GROUP BY month,day
  • grouping_id
    标记出属于哪一类维度组合,相同的组合方式grouping_id的结果一样
SELECT name, grouping_id(), sum(age), avg(height) FROM VALUES (2, 'Alice', 165), (5, 'Bob', 180) people(age, name, height) GROUP BY cube(name, height);
  NULL    2       2       165.0
  Alice   0       2       165.0
  NULL    2       5       180.0
  NULL    3       7       172.5
  Bob     0       5       180.0
  Bob     1       5       180.0
  Alice   1       2       165.0
  • rollup
    以左侧维度为主聚合维度进行层级聚合,所有维度都为NULL时代表全部数据,rollup是cube的子集;可以快速实现由左及右的下钻分析。
SELECT name, age, count(*) FROM VALUES (2, 'Alice'), (5, 'Bob') people(age, name) GROUP BY rollup(name, age);
  NULL    NULL    2
  Alice   2       1
  Bob     5       1
  Bob     NULL    1
  Alice   NULL    1

LAG,LEAD,FIRST_VALUE,LAST_VALUE

快速获取窗口内往上或往下第几行的数据

  • lag
    向上取数;lag(col,n,DEFAULT) 用于统计窗口内往上第n行值
    第一个参数为列名,第二个参数为往上第n行(可选,默认为1),第三个参数为默认值(当往上第n行为NULL时候,取默认值,如不指定,则为NULL)
sql("SELECT name, age, lag(age,1) over(partition by name order by age) lag FROM VALUES (2, 'Alice'), (5, 'Bob'), (12, 'Alice')  people(age, name)").show
+-----+---+----+                                                                
| name|age| lag|
+-----+---+----+
|  Bob|  5|null|
|Alice|  2|null|
|Alice| 12|   2|
+-----+---+----+
  • lead
    向下取数;lead(col,n,DEFAULT) 用于统计窗口内往下第n行值
    第一个参数为列名,第二个参数为往下第n行(可选,默认为1),第三个参数为默认值(当往下第n行为NULL时候,取默认值,如不指定,则为NULL)
sql("SELECT name, age, lead(age,1,age) over(partition by name order by age) leadFROM VALUES (2, 'Alice') 'Bob'),, (5, 'Bob'), (12, 'Alice')  people(age, name)").show
+-----+---+---+
| name|age|lead|
+-----+---+---+
|  Bob|  5|  5|
|Alice|  2| 12|
|Alice| 12| 12|
+-----+---+---+
  • first_value/first
    first_value(expr[, isIgnoreNull]) 用来获取分组内第一个满足expr表达式的值
    第一个参数是列名 支持表达式,第二个参数可选,当为true时 只返回非NULL数值
sql("SELECT name, age, first(age,true) over(partition by name order by age) first FROM VALUES (2, 'Alice'), (5, 'Bob'), (12, 'Alice')  people(age, name)").show
+-----+---+---+
| name|age|first|
+-----+---+---+
|  Bob|  5|  5|
|Alice|  2|  2|
|Alice| 12|  2|
+-----+---+---+
  • last_value/last
    last_value(expr[, isIgnoreNull]) 用来获取分组内最后一个满足expr表达式的值
    第一个参数是列名 支持表达式,第二个参数可选,当为true时 只返回非NULL数值
sql("SELECT name, age, last(age,true) over(partition by name) last FROM VALUES (2, 'Alice'), (5, 'Bob'), (12, 'Alice'), (1, 'Alice')   people(age, name)").show
+-----+---+-----+
| name|age|last|
+-----+---+-----+
|  Bob|  5|    5|
|Alice| 12|    2|
|Alice|  1|    2|
|Alice|  2|    2|
+-----+---+-----+

## 如果要获取窗口排序后的末尾值,需要使用first函数实现
sql("SELECT name, age, last(age,true) over(partition by name) last1, first(age) over(partition by name order by age desc) last2  FROM VALUES (2, 'Alice'), (5, 'Bob'), (12, 'Alice'), (1, 'Alice')   people(age, name)").show
+-----+---+-----+-----+
| name|age|last1|last2|
+-----+---+-----+-----+
|  Bob|  5|    5|    5|
|Alice| 12|    1|   12|
|Alice|  2|    1|   12|
|Alice|  1|    1|   12|
+-----+---+-----+-----+

NTILE,ROW_NUMBER,DENSE_RANK

常用窗口函数

  • ntile
    对数据按照某一维度进行等比切片,如果数据不均匀,会优先补充上面分片的数据量
sql("select name,age,ntile(2) over(partition by name order by age) ntile from VALUES (2, 'Alice'), (5, 'Bob'), (12, 'Alice'), (1, 'Alice')   people(age, name)").show
+-----+---+-----+
| name|age|ntile|
+-----+---+-----+
|  Bob|  5|    1|
|Alice|  1|    1|
|Alice|  2|    1|
|Alice| 12|    2|
+-----+---+-----+

sql("select name,age,ntile(3) over(partition by name order by age) ntile from VALUES (2, 'Alice'), (5, 'Bob'), (12, 'Alice'), (1, 'Alice')   people(age, name)").show
+-----+---+-----+
| name|age|ntile|
+-----+---+-----+
|  Bob|  5|    1|
|Alice|  1|    1|
|Alice|  2|    2|
|Alice| 12|    3|
+-----+---+-----+

## 如统计某个用户一天内pv最多的前1/3是那几天
## 那么将数据3等分后 取rn=1就是我们需要的记录

cookieid day           pv       rn
----------------------------------
cookie1 2015-04-12      7       1
cookie1 2015-04-11      5       1
cookie1 2015-04-15      4       1
cookie1 2015-04-16      4       2
cookie1 2015-04-13      3       2
cookie1 2015-04-14      2       3
cookie1 2015-04-10      1       3
cookie2 2015-04-15      9       1
cookie2 2015-04-16      7       1
cookie2 2015-04-13      6       1
cookie2 2015-04-12      5       2
cookie2 2015-04-14      3       2
cookie2 2015-04-11      3       3
cookie2 2015-04-10      2       3
  • row_number
    row_number从1开始,按照指定字段排序后生成递增序列号
    如 按照日期分组后对作品pv进行降序排列,生成组内pv排序的名次
sql("select day, sid, pv ,row_number() over(partition by day order by pv desc)rn from VALUES('2020-04-04','a1',11), ('2020-04-01','b1',51), ('2020-04-04','b1',11), ('2020-04-01','c1',21), ('2020-04-01','a1',1) log(day, sid, pv)").show
+----------+---+---+---+
|       day|sid| pv| rn|
+----------+---+---+---+
|2020-04-01| b1| 51|  1|
|2020-04-01| c1| 21|  2|
|2020-04-01| a1|  1|  3|
|2020-04-04| a1| 11|  1|
|2020-04-04| b1| 11|  2|
+----------+---+---+---+
  • dense_rank
    与row_number不同的是相同 pv 的序列号,dense_rank返回值是相同的
sql("select day, sid, pv, dense_rank() over(partition by day order by pv desc) dense_rank,row_number() over(partition by day order by pv desc)rn from VALUES('2020-04-04','a1',11), ('2020-04-01','b1',51), ('2020-04-04','b1',11), ('2020-04-01','c1',21), ('2020-04-01','a1',1) log(day, sid, pv)").show
+----------+---+---+----------+---+
|       day|sid| pv|dense_rank| rn|
+----------+---+---+----------+---+
|2020-04-01| b1| 51|         1|  1|
|2020-04-01| c1| 21|         2|  2|
|2020-04-01| a1|  1|         3|  3|
|2020-04-04| a1| 11|         1|  1|
|2020-04-04| b1| 11|         1|  2|
+----------+---+---+----------+---+

最后补充SUM,AVG,MIN,MAX聚合函数的窗口化支持

  • 统计某个作品随时间增长的累计pv
sql("select day, sid, pv, sum(pv) over(partition by sid order by day) pv1 from VALUES('2020-04-04','a1',11), ('2020-04-03','d1',51), ('2020-04-02','d1',11), ('2020-04-01','d1',21), ('2020-04-04','d1',1) log(day, sid, pv)").show
+----------+---+---+---+
|       day|sid| pv|pv1|
+----------+---+---+---+
|2020-04-01| d1| 21| 21|
|2020-04-02| d1| 11| 32|
|2020-04-03| d1| 51| 83|
|2020-04-04| d1|  1| 84|
|2020-04-04| a1| 11| 11|
+----------+---+---+---+
  • ROWS BETWEEN是窗口子函数,借助该函数可限定累计的范围
## ROWS BETWEEN 2 PRECEDING AND CURRENT ROW 意思是当前行pv + 往前2行pv值
sql("select day, sid, pv, sum(pv) over(partition by sid order by day) pv1, sum(pv) over(partition by sid order by day  ROWS BETWEEN 2 PRECEDING AND CURRENT ROW) pv2  from VALUES('2020-04-04','a1',11), ('2020-04-03','d1',51), ('2020-04-02','d1',11), ('2020-04-01','d1',21), ('2020-04-04','d1',1) log(day, sid, pv)").show
+----------+---+---+---+---+
|       day|sid| pv|pv1|pv2|
+----------+---+---+---+---+
|2020-04-01| d1| 21| 21| 21|
|2020-04-02| d1| 11| 32| 32|
|2020-04-03| d1| 51| 83| 83|
|2020-04-04| d1|  1| 84| 63|
|2020-04-04| a1| 11| 11| 11|
+----------+---+---+---+---+

作者:易企秀工程师 Yarn ->个人主页


易企秀工程师
91 声望10 粉丝