一文详解Hive知识体系

2021-08-19 14:54

lateral view用于和split、explode等UDTF一起使用的,能将一行数据拆分成多行数据,在此基础上可以对拆分的数据进行聚合,lateral view首先为原始表的每行调用UDTF,UDTF会把一行拆分成一行或者多行,lateral view在把结果组合,产生一个支持别名表的虚拟表。

其中explode还可以用于将hive一列中复杂的array或者map结构拆分成多行

需求:现在有数据格式如下

zhangsan child1,child2,child3,child4 k1:v1,k2:v2
lisi child5,child6,child7,child8 k3:v3,k4:v4

字段之间使用分割,需求将所有的child进行拆开成为一列

将map的key和value也进行拆开,成为如下结果

+-----------+-------------+--+
| mymapkey | mymapvalue |
+-----------+-------------+--+
| k1 | v1 |
| k2 | v2 |
| k3 | v3 |
| k4 | v4 |
+-----------+-------------+--+
创建hive数据库创建hive数据库
hive (default)> create database hive_explode;
hive (default)> use hive_explode;
创建hive表,然后使用explode拆分map和arrayhive (hive_explode)> create table t3(name string,children array

需求: 需求:现在有一些数据格式如下:

a:shandong,b:beijing,c:hebei|1,2,3,4,5,6,7,8,9|[{"source":"7fresh","monthSales":4900,"userCount":1900,"score":"9．9"},{"source":"jd","monthSales":2090,"userCount":78981,"score":"9．8"},{"source":"jdmart","monthSales":6987,"userCount":1600,"score":"9．0"}]

其中字段与字段之间的分隔符是 |

我们要解析得到所有的monthSales对应的值为以下这一列(行转列)

4900

2090

6987

创建hive表hive (hive_explode)> create table explode_lateral_view
> (`area` string,
> `goods_id` string,
> `sale_info` string)
> ROW FORMAT DELIMITED
> FIELDS TERMINATED BY '|'
> STORED AS textfile;
准备数据并加载数据准备数据如下
cd /export/servers/hivedatas
vim explode_json
a:shandong,b:beijing,c:hebei|1,2,3,4,5,6,7,8,9|[{"source":"7fresh","monthSales":4900,"userCount":1900,"score":"9．9"},{"source":"jd","monthSales":2090,"userCount":78981,"score":"9．8"},{"source":"jdmart","monthSales":6987,"userCount":1600,"score":"9．0"}]
加载数据到hive表当中去
hive (hive_explode)> load data local inpath '/export/servers/hivedatas/explode_json' overwrite into table explode_lateral_view;
使用explode拆分Arrayhive (hive_explode)> select explode(split(goods_id,',')) as goods_id from explode_lateral_view;
使用explode拆解Maphive (hive_explode)> select explode(split(area,',')) as area from explode_lateral_view;
拆解json字段hive (hive_explode)> select explode(split(regexp_replace(regexp_replace(sale_info,'\[\{',''),'}]',''),'},\{')) as sale_info from explode_lateral_view;
然后我们想用get_json_object来获取key为monthSales的数据:
hive (hive_explode)> select get_json_object(explode(split(regexp_replace(regexp_replace(sale_info,'\[\{',''),'}]',''),'},\{')),'$．monthSales') as sale_info from explode_lateral_view;
然后挂了FAILED: SemanticException [Error 10081]: UDTF's are not supported outside the SELECT clause, nor nested in expressions
UDTF explode不能写在别的函数内
如果你这么写,想查两个字段,select explode(split(area,',')) as area,good_id from explode_lateral_view;
会报错FAILED: SemanticException 1:40 Only a single expression in the SELECT clause is supported with UDTF's． Error encountered near token 'good_id'
使用UDTF的时候,只支持一个字段,这时候就需要LATERAL VIEW出场了
配合LATERAL VIEW使用

配合lateral view查询多个字段

hive (hive_explode)> select goods_id2,sale_info from explode_lateral_view LATERAL VIEW explode(split(goods_id,','))goods as goods_id2;
其中LATERAL VIEW explode(split(goods_id,','))goods相当于一个虚拟表,与原表explode_lateral_view笛卡尔积关联

也可以多重使用

hive (hive_explode)> select goods_id2,sale_info,area2
from explode_lateral_view
LATERAL VIEW explode(split(goods_id,','))goods as goods_id2
LATERAL VIEW explode(split(area,','))area as area2;也是三个表笛卡尔积的结果

最终,我们可以通过下面的句子,把这个json格式的一行数据,完全转换成二维表的方式展现

hive (hive_explode)> select get_json_object(concat('{',sale_info_1,'}'),'$．source') as source,get_json_object(concat('{',sale_info_1,'}'),'$．monthSales') as monthSales,get_json_object(concat('{',sale_info_1,'}'),'$．userCount') as monthSales,get_json_object(concat('{',sale_info_1,'}'),'$．score') as monthSales from explode_lateral_view LATERAL VIEW explode(split(regexp_replace(regexp_replace(sale_info,'\[\{',''),'}]',''),'},\{'))sale_info as sale_info_1;

总结:

Lateral View通常和UDTF一起出现,为了解决UDTF不允许在select字段的问题。Multiple Lateral View可以实现类似笛卡尔乘积。Outer关键字可以把不输出的UDTF的空结果,输出成NULL,防止丢失数据。

行转列

相关参数说明:

CONCAT(string A/col, string B/col…):返回输入字符串连接后的结果,支持任意个输入字符串;

CONCAT_WS(separator, str1, str2,．．．):它是一个特殊形式的 CONCAT()。第一个参数剩余参数间的分隔符。分隔符可以是与剩余参数一样的字符串。如果分隔符是 NULL,返回值也将为 NULL。这个函数会跳过分隔符参数后的任何 NULL 和空字符串。分隔符将被加到被连接的字符串之间;

COLLECT_SET(col):函数只接受基本数据类型,它的主要作用是将某字段的值进行去重汇总,产生array类型字段。

数据准备:

nameconstellationblood_type孙悟空白羊座A老王射手座A宋宋白羊座B猪八戒白羊座A凤姐射手座A

需求: 把星座和血型一样的人归类到一起。结果如下:

射手座,A 老王|凤姐
白羊座,A 孙悟空|猪八戒
白羊座,B 宋宋

实现步骤:

创建本地constellation．txt,导入数据node03服务器执行以下命令创建文件,注意数据使用进行分割
cd /export/servers/hivedatas
vim constellation．txt
数据如下:
孙悟空白羊座 A
老王射手座 A
宋宋白羊座 B
猪八戒白羊座 A
凤姐射手座 A
创建hive表并导入数据创建hive表并加载数据
hive (hive_explode)> create table person_info(
name string,
constellation string,
blood_type string)
row format delimited fields terminated by " ";

加载数据
hive (hive_explode)> load data local inpath '/export/servers/hivedatas/constellation．txt' into table person_info;
按需求查询数据hive (hive_explode)> select
t1．base,
concat_ws('|', collect_set(t1．name)) name
from
(select
name,
concat(constellation, "," , blood_type) base
from
person_info) t1
group by
t1．base;
列转行

所需函数:

EXPLODE(col):将hive一列中复杂的array或者map结构拆分成多行。

LATERAL VIEW

用法:LATERAL VIEW udtf(expression) tableAlias AS columnAlias

解释:用于和split, explode等UDTF一起使用,它能够将一列数据拆成多行数据,在此基础上可以对拆分后的数据进行聚合。

数据准备:

cd /export/servers/hivedatas
vim movie．txt
文件内容如下: 数据字段之间使用进行分割
《疑犯追踪》悬疑,动作,科幻,剧情
《Lie to me》悬疑,警匪,动作,心理,剧情
《战狼2》战争,动作,灾难

需求: 将电影分类中的数组数据展开。结果如下:

《疑犯追踪》悬疑
《疑犯追踪》动作
《疑犯追踪》科幻
《疑犯追踪》剧情
《Lie to me》悬疑
《Lie to me》警匪
《Lie to me》动作
《Lie to me》心理
《Lie to me》剧情
《战狼2》战争
《战狼2》动作
《战狼2》灾难

实现步骤:

创建hive表create table movie_info(
movie string,
category array

reflect函数可以支持在sql中调用java中的自带函数,秒杀一切udf函数。

需求1: 使用java．lang．Math当中的Max求两列中最大值

实现步骤:

创建hive表create table test_udf(col1 int,col2 int) row format delimited fields terminated by ',';
准备数据并加载数据cd /export/servers/hivedatas
vim test_udf
文件内容如下:
1,2
4,3
6,4
7,5
5,6
加载数据hive (hive_explode)> load data local inpath '/export/servers/hivedatas/test_udf' overwrite into table test_udf;
使用java．lang．Math当中的Max求两列当中的最大值hive (hive_explode)> select reflect("java．lang．Math","max",col1,col2) from test_udf;

需求2: 文件中不同的记录来执行不同的java的内置函数

实现步骤:

创建hive表hive (hive_explode)> create table test_udf2(class_name string,method_name string,col1 int , col2 int) row format delimited fields terminated by ',';
准备数据cd /export/servers/hivedatas
vim test_udf2
文件内容如下:
java．lang．Math,min,1,2
java．lang．Math,max,2,3
加载数据hive (hive_explode)> load data local inpath '/export/servers/hivedatas/test_udf2' overwrite into table test_udf2;
执行查询hive (hive_explode)> select reflect(class_name,method_name,col1,col2) from test_udf2;

需求3: 判断是否为数字

实现方式:

使用apache commons中的函数,commons下的jar已经包含在hadoop的classpath中,所以可以直接使用。

select reflect("org．apache．commons．lang．math．NumberUtils","isNumber","123")
窗口函数与分析函数

在sql中有一类函数叫做聚合函数,例如sum()、avg()、max()等等,这类函数可以将多行数据按照规则聚集为一行,一般来讲聚集后的行数是要少于聚集前的行数的。但是有时我们想要既显示聚集前的数据,又要显示聚集后的数据,这时我们便引入了窗口函数。窗口函数又叫OLAP函数/分析函数,窗口函数兼具分组和排序功能。

窗口函数最重要的关键字是 partition by 和 order by。

具体语法如下:over (partition by xxx order by xxx)

sum、avg、min、max

准备数据

建表语句:
create table test_t1(
cookieid string,
createtime string, --day
pv int
) row format delimited
fields terminated by ',';
加载数据:
load data local inpath '/root/hivedata/test_t1．dat' into table test_t1;
cookie1,2020-04-10,1
cookie1,2020-04-11,5
cookie1,2020-04-12,7
cookie1,2020-04-13,3
cookie1,2020-04-14,2
cookie1,2020-04-15,4
cookie1,2020-04-16,4
开启智能本地模式
SET hive．exec．mode．local．auto=true;

SUM函数和窗口函数的配合使用:结果和ORDER BY相关,默认为升序。

select cookieid,createtime,pv,
sum(pv) over(partition by cookieid order by createtime) as pv1
from test_t1;
select cookieid,createtime,pv,
sum(pv) over(partition by cookieid order by createtime rows between unbounded preceding and current row) as pv2
from test_t1;
select cookieid,createtime,pv,
sum(pv) over(partition by cookieid) as pv3
from test_t1;
select cookieid,createtime,pv,
sum(pv) over(partition by cookieid order by createtime rows between 3 preceding and current row) as pv4
from test_t1;
select cookieid,createtime,pv,
sum(pv) over(partition by cookieid order by createtime rows between 3 preceding and 1 following) as pv5
from test_t1;
select cookieid,createtime,pv,
sum(pv) over(partition by cookieid order by createtime rows between current row and unbounded following) as pv6
from test_t1;
pv1: 分组内从起点到当前行的pv累积,如,11号的pv1=10号的pv+11号的pv, 12号=10号+11号+12号
pv2: 同pv1
pv3: 分组内(cookie1)所有的pv累加
pv4: 分组内当前行+往前3行,如,11号=10号+11号, 12号=10号+11号+12号,
13号=10号+11号+12号+13号, 14号=11号+12号+13号+14号
pv5: 分组内当前行+往前3行+往后1行,如,14号=11号+12号+13号+14号+15号=5+7+3+2+4=21
pv6: 分组内当前行+往后所有行,如,13号=13号+14号+15号+16号=3+2+4+4=13,
14号=14号+15号+16号=2+4+4=10

如果不指定rows between,默认为从起点到当前行;

如果不指定order by,则将分组内所有值累加;

关键是理解rows between含义,也叫做window子句:

preceding:往前

following:往后

current row:当前行

unbounded:起点

unbounded preceding 表示从前面的起点

unbounded following:表示到后面的终点

AVG,MIN,MAX,和SUM用法一样。

row_number、rank、dense_rank、ntile

准备数据

cookie1,2020-04-10,1
cookie1,2020-04-11,5
cookie1,2020-04-12,7
cookie1,2020-04-13,3
cookie1,2020-04-14,2
cookie1,2020-04-15,4
cookie1,2020-04-16,4
cookie2,2020-04-10,2
cookie2,2020-04-11,3
cookie2,2020-04-12,5
cookie2,2020-04-13,6
cookie2,2020-04-14,3
cookie2,2020-04-15,9
cookie2,2020-04-16,7

CREATE TABLE test_t2 (
cookieid string,
createtime string, --day
pv INT
) ROW FORMAT DELIMITED
FIELDS TERMINATED BY ','
stored as textfile;

加载数据:
load data local inpath '/root/hivedata/test_t2．dat' into table test_t2;

ROW_NUMBER()使用

ROW_NUMBER()从1开始,按照顺序,生成分组内记录的序列。

SELECT
cookieid,
createtime,
pv,
ROW_NUMBER() OVER(PARTITION BY cookieid ORDER BY pv desc) AS rn
FROM test_t2;

RANK 和 DENSE_RANK使用

RANK() 生成数据项在分组中的排名,排名相等会在名次中留下空位。

DENSE_RANK()生成数据项在分组中的排名,排名相等会在名次中不会留下空位。

SELECT
cookieid,
createtime,
pv,
RANK() OVER(PARTITION BY cookieid ORDER BY pv desc) AS rn1,
DENSE_RANK() OVER(PARTITION BY cookieid ORDER BY pv desc) AS rn2,
ROW_NUMBER() OVER(PARTITION BY cookieid ORDER BY pv DESC) AS rn3
FROM test_t2
WHERE cookieid = 'cookie1';

NTILE

有时会有这样的需求:如果数据排序后分为三部分,业务人员只关心其中的一部分,如何将这中间的三分之一数据拿出来呢?NTILE函数即可以满足。

ntile可以看成是:把有序的数据集合平均分配到指定的数量(num)个桶中, 将桶号分配给每一行。如果不能平均分配,则优先分配较小编号的桶,并且各个桶中能放的行数最多相差1。

然后可以根据桶号,选取前或后 n分之几的数据。数据会完整展示出来,只是给相应的数据打标签;具体要取几分之几的数据,需要再嵌套一层根据标签取出。

SELECT
cookieid,
createtime,
pv,
NTILE(2) OVER(PARTITION BY cookieid ORDER BY createtime) AS rn1,
NTILE(3) OVER(PARTITION BY cookieid ORDER BY createtime) AS rn2,
NTILE(4) OVER(ORDER BY createtime) AS rn3
FROM test_t2
ORDER BY cookieid,createtime;
其他一些窗口函数lag,lead,first_value,last_valueLAG
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)SELECT cookieid,
createtime,
url,
ROW_NUMBER() OVER(PARTITION BY cookieid ORDER BY createtime) AS rn,
LAG(createtime,1,'1970-01-01 00:00:00') OVER(PARTITION BY cookieid ORDER BY createtime) AS last_1_time,
LAG(createtime,2) OVER(PARTITION BY cookieid ORDER BY createtime) AS last_2_time
FROM test_t4;
last_1_time: 指定了往上第1行的值,default为'1970-01-01 00:00:00'
cookie1第一行,往上1行为NULL,因此取默认值 1970-01-01 00:00:00
cookie1第三行,往上1行值为第二行值,2015-04-10 10:00:02
cookie1第六行,往上1行值为第五行值,2015-04-10 10:50:01
last_2_time: 指定了往上第2行的值,为指定默认值
cookie1第一行,往上2行为NULL
cookie1第二行,往上2行为NULL
cookie1第四行,往上2行为第二行值,2015-04-10 10:00:02
cookie1第七行,往上2行为第五行值,2015-04-10 10:50:01
LEAD

与LAG相反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)

SELECT cookieid,
createtime,
url,
ROW_NUMBER() OVER(PARTITION BY cookieid ORDER BY createtime) AS rn,
LEAD(createtime,1,'1970-01-01 00:00:00') OVER(PARTITION BY cookieid ORDER BY createtime) AS next_1_time,
LEAD(createtime,2) OVER(PARTITION BY cookieid ORDER BY createtime) AS next_2_time
FROM test_t4;

FIRST_VALUE

取分组内排序后,截止到当前行,第一个值

SELECT cookieid,
createtime,
url,
ROW_NUMBER() OVER(PARTITION BY cookieid ORDER BY createtime) AS rn,
FIRST_VALUE(url) OVER(PARTITION BY cookieid ORDER BY createtime) AS first1
FROM test_t4;
LAST_VALUE

取分组内排序后,截止到当前行,最后一个值

SELECT cookieid,
createtime,
url,
ROW_NUMBER() OVER(PARTITION BY cookieid ORDER BY createtime) AS rn,
LAST_VALUE(url) OVER(PARTITION BY cookieid ORDER BY createtime) AS last1
FROM test_t4;

如果想要取分组内排序后最后一个值,则需要变通一下:

SELECT cookieid,
createtime,
url,
ROW_NUMBER() OVER(PARTITION BY cookieid ORDER BY createtime) AS rn,
LAST_VALUE(url) OVER(PARTITION BY cookieid ORDER BY createtime) AS last1,
FIRST_VALUE(url) OVER(PARTITION BY cookieid ORDER BY createtime DESC) AS last2
FROM test_t4
ORDER BY cookieid,createtime;

特别注意order by

如果不指定ORDER BY,则进行排序混乱,会出现错误的结果

SELECT cookieid,
createtime,
url,
FIRST_VALUE(url) OVER(PARTITION BY cookieid) AS first2
FROM test_t4;
cume_dist,percent_rank

这两个序列分析函数不是很常用,注意:序列函数不支持WINDOW子句

数据准备d1,user1,1000
d1,user2,2000
d1,user3,3000
d2,user4,4000
d2,user5,5000

CREATE EXTERNAL TABLE test_t3 (
dept STRING,
userid string,
sal INT
) ROW FORMAT DELIMITED
FIELDS TERMINATED BY ','
stored as textfile;
加载数据:
load data local inpath '/root/hivedata/test_t3．dat' into table test_t3;

CUME_DIST 和order byd的排序顺序有关系

CUME_DIST 小于等于当前值的行数/分组内总行数 order 默认顺序正序升序比如,统计小于等于当前薪水的人数,所占总人数的比例

SELECT
dept,
userid,
sal,
CUME_DIST() OVER(ORDER BY sal) AS rn1,
CUME_DIST() OVER(PARTITION BY dept ORDER BY sal) AS rn2
FROM test_t3;

rn1: 没有partition,所有数据均为1组,总行数为5,
第一行:小于等于1000的行数为1,因此,1/5=0．2
第三行:小于等于3000的行数为3,因此,3/5=0．6
rn2: 按照部门分组,dpet=d1的行数为3,
第二行:小于等于2000的行数为2,因此,2/3=0．6666666666666666

PERCENT_RANK

PERCENT_RANK 分组内当前行的RANK值-1/分组内总行数-1

经调研该函数显示现实意义不明朗有待于继续考证

SELECT
dept,
userid,
sal,
PERCENT_RANK() OVER(ORDER BY sal) AS rn1, --分组内
RANK() OVER(ORDER BY sal) AS rn11, --分组内RANK值
SUM(1) OVER(PARTITION BY NULL) AS rn12, --分组内总行数
PERCENT_RANK() OVER(PARTITION BY dept ORDER BY sal) AS rn2
FROM test_t3;

rn1: rn1 = (rn11-1) / (rn12-1)
第一行,(1-1)/(5-1)=0/4=0
第二行,(2-1)/(5-1)=1/4=0．25
第四行,(4-1)/(5-1)=3/4=0．75
rn2: 按照dept分组,
dept=d1的总行数为3
第一行,(1-1)/(3-1)=0
第三行,(3-1)/(3-1)=1
grouping sets,grouping__id,cube,rollup

这几个分析函数通常用于OLAP中,不能累加,而且需要根据不同维度上钻和下钻的指标统计,比如,分小时、天、月的UV数。

数据准备2020-03,2020-03-10,cookie1
2020-03,2020-03-10,cookie5
2020-03,2020-03-12,cookie7
2020-04,2020-04-12,cookie3
2020-04,2020-04-13,cookie2
2020-04,2020-04-13,cookie4
2020-04,2020-04-16,cookie4
2020-03,2020-03-10,cookie2
2020-03,2020-03-10,cookie3
2020-04,2020-04-12,cookie5
2020-04,2020-04-13,cookie6
2020-04,2020-04-15,cookie3
2020-04,2020-04-15,cookie2
2020-04,2020-04-16,cookie1

CREATE TABLE test_t5 (
month STRING,
day STRING,
cookieid STRING
) ROW FORMAT DELIMITED
FIELDS TERMINATED BY ','
stored as textfile;
加载数据:
load data local inpath '/root/hivedata/test_t5．dat' into table test_t5;
GROUPING SETS

grouping sets是一种将多个group by 逻辑写在一个sql语句中的便利写法。

等价于将不同维度的GROUP BY结果集进行UNION ALL。

GROUPING__ID,表示结果属于哪一个分组集合。

SELECT
month,
day,
COUNT(DISTINCT cookieid) AS uv,
GROUPING__ID
FROM test_t5
GROUP BY month,day
GROUPING SETS (month,day)
ORDER BY GROUPING__ID;
grouping_id表示这一组结果属于哪个分组集合,
根据grouping sets中的分组条件month,day,1是代表month,2是代表day
等价于
SELECT month,NULL,COUNT(DISTINCT cookieid) AS uv,1 AS GROUPING__ID FROM test_t5 GROUP BY month UNION ALL
SELECT NULL as month,day,COUNT(DISTINCT cookieid) AS uv,2 AS GROUPING__ID FROM test_t5 GROUP BY day;

再如:

SELECT
month,
day,
COUNT(DISTINCT cookieid) AS uv,
GROUPING__ID
FROM test_t5
GROUP BY month,day
GROUPING SETS (month,day,(month,day))
ORDER BY GROUPING__ID;
等价于
SELECT month,NULL,COUNT(DISTINCT cookieid) AS uv,1 AS GROUPING__ID FROM test_t5 GROUP BY month
UNION ALL
SELECT NULL,day,COUNT(DISTINCT cookieid) AS uv,2 AS GROUPING__ID FROM test_t5 GROUP BY day
UNION ALL
SELECT month,day,COUNT(DISTINCT cookieid) AS uv,3 AS GROUPING__ID FROM test_t5 GROUP BY month,day;
CUBE

根据GROUP BY的维度的所有组合进行聚合。

SELECT
month,
day,
COUNT(DISTINCT cookieid) AS uv,
GROUPING__ID
FROM test_t5
GROUP BY month,day
WITH CUBE
ORDER BY GROUPING__ID;
等价于
SELECT NULL,NULL,COUNT(DISTINCT cookieid) AS uv,0 AS GROUPING__ID FROM test_t5
UNION ALL
SELECT month,NULL,COUNT(DISTINCT cookieid) AS uv,1 AS GROUPING__ID FROM test_t5 GROUP BY month
UNION ALL
SELECT NULL,day,COUNT(DISTINCT cookieid) AS uv,2 AS GROUPING__ID FROM test_t5 GROUP BY day
UNION ALL
SELECT month,day,COUNT(DISTINCT cookieid) AS uv,3 AS GROUPING__ID FROM test_t5 GROUP BY month,day;
ROLLUP

是CUBE的子集,以最左侧的维度为主,从该维度进行层级聚合。

比如,以month维度进行层级聚合:
SELECT
month,
day,
COUNT(DISTINCT cookieid) AS uv,
GROUPING__ID
FROM test_t5
GROUP BY month,day
WITH ROLLUP
ORDER BY GROUPING__ID;
--把month和day调换顺序,则以day维度进行层级聚合:

SELECT
day,
month,
COUNT(DISTINCT cookieid) AS uv,
GROUPING__ID
FROM test_t5
GROUP BY day,month
WITH ROLLUP
ORDER BY GROUPING__ID;
(这里,根据天和月进行聚合,和根据天聚合结果一样,因为有父子关系,如果是其他维度组合的话,就会不一样)
七、Hive执行计划

Hive SQL的执行计划描述SQL实际执行的整体轮廓,通过执行计划能了解SQL程序在转换成相应计算引擎的执行逻辑,掌握了执行逻辑也就能更好地把握程序出现的瓶颈点,从而能够实现更有针对性的优化。此外还能帮助开发者识别看似等价的SQL其实是不等价的,看似不等价的SQL其实是等价的SQL。可以说执行计划是打开SQL优化大门的一把钥匙。

要想学SQL执行计划,就需要学习查看执行计划的命令:explain,在查询语句的SQL前面加上关键字explain是查看执行计划的基本方法。

学会explain,能够给我们工作中使用hive带来极大的便利!

查看SQL的执行计划

Hive提供的执行计划目前可以查看的信息有以下几种:

explain:查看执行计划的基本信息;

explain dependency:dependency在explain语句中使用会产生有关计划中输入的额外信息。它显示了输入的各种属性;

explain authorization:查看SQL操作相关权限的信息;

explain vectorization:查看SQL的向量化描述信息,显示为什么未对Map和Reduce进行矢量化。从 Hive 2．3．0 开始支持;

explain analyze:用实际的行数注释计划。从 Hive 2．2．0 开始支持;

explain cbo:输出由Calcite优化器生成的计划。CBO 从 Hive 4．0．0 版本开始支持;

explain locks:这对于了解系统将获得哪些锁以运行指定的查询很有用。LOCKS 从 Hive 3．2．0 开始支持;

explain ast:输出查询的抽象语法树。AST 在 Hive 2．1．0 版本删除了,存在bug,转储AST可能会导致OOM错误,将在4．0．0版本修复;

explain extended:加上 extended 可以输出有关计划的额外信息。这通常是物理信息,例如文件名,这些额外信息对我们用处不大;

1． explain 的用法

Hive提供了explain命令来展示一个查询的执行计划,这个执行计划对于我们了解底层原理,Hive 调优,排查数据倾斜等很有帮助。

使用语法如下:

explain query;

在 hive cli 中输入以下命令(hive 2．3．7):

explain select sum(id) from test1;

得到结果:

STAGE DEPENDENCIES:
Stage-1 is a root stage
Stage-0 depends on stages: Stage-1
STAGE PLANS:
Stage: Stage-1
Map Reduce
Map Operator Tree:
TableScan
alias: test1
Statistics: Num rows: 6 Data size: 75 Basic stats: COMPLETE Column stats: NONE
Select Operator
expressions: id (type: int)
outputColumnNames: id
Statistics: Num rows: 6 Data size: 75 Basic stats: COMPLETE Column stats: NONE
Group By Operator
aggregations: sum(id)
mode: hash
outputColumnNames: _col0
Statistics: Num rows: 1 Data size: 8 Basic stats: COMPLETE Column stats: NONE
Reduce Output Operator
sort order:
Statistics: Num rows: 1 Data size: 8 Basic stats: COMPLETE Column stats: NONE
value expressions: _col0 (type: bigint)
Reduce Operator Tree:
Group By Operator
aggregations: sum(VALUE．_col0)
mode: mergepartial
outputColumnNames: _col0
Statistics: Num rows: 1 Data size: 8 Basic stats: COMPLETE Column stats: NONE
File Output Operator
compressed: false
Statistics: Num rows: 1 Data size: 8 Basic stats: COMPLETE Column stats: NONE
table:
input format: org．apache．hadoop．mapred．SequenceFileInputFormat
output format: org．apache．hadoop．hive．ql．io．HiveSequenceFileOutputFormat
serde: org．apache．hadoop．hive．serde2．lazy．LazySimpleSerDe
Stage: Stage-0
Fetch Operator
limit: -1
Processor Tree:
ListSink

看完以上内容有什么感受,是不是感觉都看不懂,不要着急,下面将会详细讲解每个参数,相信你学完下面的内容之后再看 explain 的查询结果将游刃有余。

一个HIVE查询被转换为一个由一个或多个stage组成的序列(有向无环图DAG)。这些stage可以是MapReduce stage,也可以是负责元数据存储的stage,也可以是负责文件系统的操作(比如移动和重命名)的stage。

我们将上述结果拆分看,先从最外层开始,包含两个大的部分:

stage dependencies:各个stage之间的依赖性stage plan:各个stage的执行计划

先看第一部分 stage dependencies ,包含两个 stage,Stage-1 是根stage,说明这是开始的stage,Stage-0 依赖 Stage-1,Stage-1执行完成后执行Stage-0。

再看第二部分 stage plan,里面有一个 Map Reduce,一个MR的执行计划分为两个部分:

Map Operator Tree:MAP端的执行计划树Reduce Operator Tree:Reduce端的执行计划树

这两个执行计划树里面包含这条sql语句的 operator:

TableScan:表扫描操作,map端第一个操作肯定是加载表,所以就是表扫描操作,常见的属性:alias:表名称Statistics:表统计信息,包含表中数据条数,数据大小等Select Operator:选取操作,常见的属性 :expressions:需要的字段名称及字段类型outputColumnNames:输出的列名称Statistics:表统计信息,包含表中数据条数,数据大小等Group By Operator:分组聚合操作,常见的属性:aggregations:显示聚合函数信息mode:聚合模式,值有 hash:随机聚合,就是hash partition;partial:局部聚合;final:最终聚合keys:分组的字段,如果没有分组,则没有此字段outputColumnNames:聚合之后输出列名Statistics:表统计信息,包含分组聚合之后的数据条数,数据大小等Reduce Output Operator:输出到reduce操作,常见属性:sort order:值为空不排序;值为 + 正序排序,值为 - 倒序排序;值为 +- 排序的列为两列,第一列为正序,第二列为倒序Filter Operator:过滤操作,常见的属性:predicate:过滤条件,如sql语句中的where id>=1,则此处显示(id >= 1)Map Join Operator:join 操作,常见的属性:condition map:join方式 ,如Inner Join 0 to 1 Left Outer Join0 to 2keys: join 的条件字段outputColumnNames:join 完成之后输出的字段Statistics:join 完成之后生成的数据条数,大小等File Output Operator:文件输出操作,常见的属性compressed:是否压缩table:表的信息,包含输入输出文件格式化方式,序列化方式等Fetch Operator 客户端获取数据操作,常见的属性:limit,值为 -1 表示不限制条数,其他值为限制的条数2． explain 的使用场景

本节介绍 explain 能够为我们在生产实践中带来哪些便利及解决我们哪些迷惑

案例一:join 语句会过滤 null 的值吗?

现在,我们在hive cli 输入以下查询计划语句

select a．id,b．user_name from test1 a join test2 b on a．id=b．id;

问:上面这条 join 语句会过滤 id 为 null 的值吗

执行下面语句:

explain select a．id,b．user_name from test1 a join test2 b on a．id=b．id;

我们来看结果 (为了适应页面展示,仅截取了部分输出信息):

TableScan
alias: a
Statistics: Num rows: 6 Data size: 75 Basic stats: COMPLETE Column stats: NONE
Filter Operator
predicate: id is not null (type: boolean)
Statistics: Num rows: 6 Data size: 75 Basic stats: COMPLETE Column stats: NONE
Select Operator
expressions: id (type: int)
outputColumnNames: _col0
Statistics: Num rows: 6 Data size: 75 Basic stats: COMPLETE Column stats: NONE
HashTable Sink Operator
keys:
0 _col0 (type: int)
1 _col0 (type: int)
．．．

从上述结果可以看到 predicate: id is not null 这样一行,说明 join 时会自动过滤掉关联字段为 null值的情况,但 left join 或 full join 是不会自动过滤null值的,大家可以自行尝试下。

案例二:group by 分组语句会进行排序吗?

看下面这条sql

select id,max(user_name) from test1 group by id;

问:group by 分组语句会进行排序吗

直接来看 explain 之后结果 (为了适应页面展示,仅截取了部分输出信息)

TableScan
alias: test1
Statistics: Num rows: 9 Data size: 108 Basic stats: COMPLETE Column stats: NONE
Select Operator
expressions: id (type: int), user_name (type: string)
outputColumnNames: id, user_name
Statistics: Num rows: 9 Data size: 108 Basic stats: COMPLETE Column stats: NONE
Group By Operator
aggregations: max(user_name)
keys: id (type: int)
mode: hash
outputColumnNames: _col0, _col1
Statistics: Num rows: 9 Data size: 108 Basic stats: COMPLETE Column stats: NONE
Reduce Output Operator
key expressions: _col0 (type: int)
sort order: +
Map-reduce partition columns: _col0 (type: int)
Statistics: Num rows: 9 Data size: 108 Basic stats: COMPLETE Column stats: NONE
value expressions: _col1 (type: string)
．．．

我们看 Group By Operator,里面有 keys: id (type: int) 说明按照 id 进行分组的,再往下看还有 sort order: + ,说明是按照 id 字段进行正序排序的。

案例三:哪条sql执行效率高呢?

观察两条sql语句

SELECT
a．id,
b．user_name
FROM
test1 a
JOIN test2 b ON a．id = b．id
WHERE
a．id > 2;
SELECT
a．id,
b．user_name
FROM
(SELECT * FROM test1 WHERE id > 2) a
JOIN test2 b ON a．id = b．id;

这两条sql语句输出的结果是一样的,但是哪条sql执行效率高呢?

有人说第一条sql执行效率高,因为第二条sql有子查询,子查询会影响性能;

有人说第二条sql执行效率高,因为先过滤之后,在进行join时的条数减少了,所以执行效率就高了。

到底哪条sql效率高呢,我们直接在sql语句前面加上 explain,看下执行计划不就知道了嘛!

在第一条sql语句前加上 explain,得到如下结果

hive (default)> explain select a．id,b．user_name from test1 a join test2 b on a．id=b．id where a．id >2;
OK
Explain
STAGE DEPENDENCIES:
Stage-4 is a root stage
Stage-3 depends on stages: Stage-4
Stage-0 depends on stages: Stage-3
STAGE PLANS:
Stage: Stage-4
Map Reduce Local Work
Alias -> Map Local Tables:
$hdt$_0:a
Fetch Operator
limit: -1
Alias -> Map Local Operator Tree:
$hdt$_0:a
TableScan
alias: a
Statistics: Num rows: 6 Data size: 75 Basic stats: COMPLETE Column stats: NONE
Filter Operator
predicate: (id > 2) (type: boolean)
Statistics: Num rows: 2 Data size: 25 Basic stats: COMPLETE Column stats: NONE
Select Operator
expressions: id (type: int)
outputColumnNames: _col0
Statistics: Num rows: 2 Data size: 25 Basic stats: COMPLETE Column stats: NONE
HashTable Sink Operator
keys:
0 _col0 (type: int)
1 _col0 (type: int)
Stage: Stage-3
Map Reduce
Map Operator Tree:
TableScan
alias: b
Statistics: Num rows: 6 Data size: 75 Basic stats: COMPLETE Column stats: NONE
Filter Operator
predicate: (id > 2) (type: boolean)
Statistics: Num rows: 2 Data size: 25 Basic stats: COMPLETE Column stats: NONE
Select Operator
expressions: id (type: int), user_name (type: string)
outputColumnNames: _col0, _col1
Statistics: Num rows: 2 Data size: 25 Basic stats: COMPLETE Column stats: NONE
Map Join Operator
condition map:
Inner Join 0 to 1
keys:
0 _col0 (type: int)
1 _col0 (type: int)
outputColumnNames: _col0, _col2
Statistics: Num rows: 2 Data size: 27 Basic stats: COMPLETE Column stats: NONE
Select Operator
expressions: _col0 (type: int), _col2 (type: string)
outputColumnNames: _col0, _col1
Statistics: Num rows: 2 Data size: 27 Basic stats: COMPLETE Column stats: NONE
File Output Operator
compressed: false
Statistics: Num rows: 2 Data size: 27 Basic stats: COMPLETE Column stats: NONE
table:
input format: org．apache．hadoop．mapred．SequenceFileInputFormat
output format: org．apache．hadoop．hive．ql．io．HiveSequenceFileOutputFormat
serde: org．apache．hadoop．hive．serde2．lazy．LazySimpleSerDe
Local Work:
Map Reduce Local Work
Stage: Stage-0
Fetch Operator
limit: -1
Processor Tree:
ListSink

在第二条sql语句前加上 explain,得到如下结果

hive (default)> explain select a．id,b．user_name from(select * from test1 where id>2 ) a join test2 b on a．id=b．id;
OK
Explain
STAGE DEPENDENCIES:
Stage-4 is a root stage
Stage-3 depends on stages: Stage-4
Stage-0 depends on stages: Stage-3
STAGE PLANS:
Stage: Stage-4
Map Reduce Local Work
Alias -> Map Local Tables:
$hdt$_0:test1
Fetch Operator
limit: -1
Alias -> Map Local Operator Tree:
$hdt$_0:test1
TableScan
alias: test1
Statistics: Num rows: 6 Data size: 75 Basic stats: COMPLETE Column stats: NONE
Filter Operator
predicate: (id > 2) (type: boolean)
Statistics: Num rows: 2 Data size: 25 Basic stats: COMPLETE Column stats: NONE
Select Operator
expressions: id (type: int)
outputColumnNames: _col0
Statistics: Num rows: 2 Data size: 25 Basic stats: COMPLETE Column stats: NONE
HashTable Sink Operator
keys:
0 _col0 (type: int)
1 _col0 (type: int)
Stage: Stage-3
Map Reduce
Map Operator Tree:
TableScan
alias: b
Statistics: Num rows: 6 Data size: 75 Basic stats: COMPLETE Column stats: NONE
Filter Operator
predicate: (id > 2) (type: boolean)
Statistics: Num rows: 2 Data size: 25 Basic stats: COMPLETE Column stats: NONE
Select Operator
expressions: id (type: int), user_name (type: string)
outputColumnNames: _col0, _col1
Statistics: Num rows: 2 Data size: 25 Basic stats: COMPLETE Column stats: NONE
Map Join Operator
condition map:
Inner Join 0 to 1
keys:
0 _col0 (type: int)
1 _col0 (type: int)
outputColumnNames: _col0, _col2
Statistics: Num rows: 2 Data size: 27 Basic stats: COMPLETE Column stats: NONE
Select Operator
expressions: _col0 (type: int), _col2 (type: string)
outputColumnNames: _col0, _col1
Statistics: Num rows: 2 Data size: 27 Basic stats: COMPLETE Column stats: NONE
File Output Operator
compressed: false
Statistics: Num rows: 2 Data size: 27 Basic stats: COMPLETE Column stats: NONE
table:
input format: org．apache．hadoop．mapred．SequenceFileInputFormat
output format: org．apache．hadoop．hive．ql．io．HiveSequenceFileOutputFormat
serde: org．apache．hadoop．hive．serde2．lazy．LazySimpleSerDe
Local Work:
Map Reduce Local Work
Stage: Stage-0
Fetch Operator
limit: -1
Processor Tree:
ListSink