Hive 在多维统计分析中的应用 & 技巧总结
本文原地址：https://my.oschina.net/leejun2005/blog/121945
多维统计一般分两种，我们看看 Hive 中如何解决：
1、同属性的多维组合统计
（1）问题：
有如下数据，字段内容分别为：url, catePath0, catePath1, catePath2, unitparams
https://cwiki.apache.org/confluence 0 1 8 {"store":{"fruit":[{"weight":1,"type":"apple"},{"weight":9,"type":"pear"}],"bicycle":{"price":19.951,"color":"red1"}},"email":"amy@only_for_json_udf_test.net","owner":"amy1"}
http://my.oschina.net/leejun2005/blog/83058 0 1 23 {"store":{"fruit":[{"weight":1,"type":"apple"},{"weight":9,"type":"pear"}],"bicycle":{"price":19.951,"color":"red1"}},"email":"amy@only_for_json_udf_test.net","owner":"amy1"}
http://www.hao123.com/indexnt.html?sto 0 1 25 {"store":{"fruit":[{"weight":1,"type":"apple"},{"weight":9,"type":"pear"}],"bicycle":{"price":19.951,"color":"red1"}},"email":"amy@only_for_json_udf_test.net","owner":"amy1"}
https://cwiki.apache.org/confluence 0 5 18 {"store":{"fruit":[{"weight":5,"type":"apple"},{"weight":9,"type":"pear"}],"bicycle":{"price":19.951,"color":"red1"}},"email":"amy@only_for_json_udf_test.net","owner":"amy1"}
http://my.oschina.net/leejun2005/blog/83058 0 5 118 {"store":{"fruit":[{"weight":5,"type":"apple"},{"weight":9,"type":"pear"}],"bicycle":{"price":19.951,"color":"red1"}},"email":"amy@only_for_json_udf_test.net","owner":"amy1"}
http://www.hao123.com/indexnt.html?sto 0 3 98 {"store":{"fruit":[{"weight":3,"type":"apple"},{"weight":9,"type":"pear"}],"bicycle":{"price":19.951,"color":"red1"}},"email":"amy@only_for_json_udf_test.net","owner":"amy1"}
http://www.hao123.com/indexnt.html?sto 0 3 8 {"store":{"fruit":[{"weight":3,"type":"apple"},{"weight":9,"type":"pear"}],"bicycle":{"price":19.951,"color":"red1"}},"email":"amy@only_for_json_udf_test.net","owner":"amy1"}
http://my.oschina.net/leejun2005/blog/83058 0 5 81 {"store":{"fruit":[{"weight":5,"type":"apple"},{"weight":9,"type":"pear"}],"bicycle":{"price":19.951,"color":"red1"}},"email":"amy@only_for_json_udf_test.net","owner":"amy1"}
http://www.hao123.com/indexnt.html?sto 0 9 8 {"store":{"fruit":[{"weight":9,"type":"apple"},{"weight":9,"type":"pear"}],"bicycle":{"price":19.951,"color":"red1"}},"email":"amy@only_for_json_udf_test.net","owner":"amy1"}
（2）需求：
计算 catePath0, catePath1, catePath2 这三种维度组合下，各个 url 对应的 pv、uv，如：
0 1 23 1 1
0 1 25 1 1
0 1 8 1 1
0 1 ALL 3 3
0 3 8 1 1
0 3 98 1 1
0 3 ALL 2 1
0 5 118 1 1
0 5 18 1 1
0 5 81 1 1
0 5 ALL 3 2
0 ALL ALL 8 3
ALL ALL ALL 8 3
（3）解决思路：
hive 中同属性多维统计问题通常用 union all 组合出各种维度然后 group by 进行求解：
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
|
create
EXTERNAL
table
IF
NOT
EXISTS t_log ( url string, c0 string, c1 string, c2 string, unitparams string) ROW FORMAT DELIMITED FIELDS TERMINATED
BY
'\t'
location
'/tmp/decli/1'
;
select
*
from
(
select
host, c0, c1, c2
from
t_log t0
LATERAL
VIEW
parse_url_tuple(url,
'HOST'
) t1
as
host
where
get_json_object(t0.unitparams,
'$.store.fruit[0].weight'
) != 9
union
all
select
host, c0, c1,
'ALL'
c2
from
t_log t0
LATERAL
VIEW
parse_url_tuple(url,
'HOST'
) t1
as
host
where
get_json_object(t0.unitparams,
'$.store.fruit[0].weight'
) != 9
union
all
select
host, c0,
'ALL'
c1,
'ALL'
c2
from
t_log t0
LATERAL
VIEW
parse_url_tuple(url,
'HOST'
) t1
as
host
where
get_json_object(t0.unitparams,
'$.store.fruit[0].weight'
) != 9
union
all
select
host,
'ALL'
c0,
'ALL'
c1,
'ALL'
c2
from
t_log t0
LATERAL
VIEW
parse_url_tuple(url,
'HOST'
) t1
as
host
where
get_json_object(t0.unitparams,
'$.store.fruit[0].weight'
) != 9) test;
select
c0, c1, c2,
count
(host) PV,
count
(
distinct
(host)) UV
from
(
select
host, c0, c1, c2
from
t_log t0
LATERAL
VIEW
parse_url_tuple(url,
'HOST'
) t1
as
host
where
get_json_object(t0.unitparams,
'$.store.fruit[0].weight'
) != 9
union
all
select
host, c0, c1,
'ALL'
c2
from
t_log t0
LATERAL
VIEW
parse_url_tuple(url,
'HOST'
) t1
as
host
where
get_json_object(t0.unitparams,
'$.store.fruit[0].weight'
) != 9
union
all
select
host, c0,
'ALL'
c1,
'ALL'
c2
from
t_log t0
LATERAL
VIEW
parse_url_tuple(url,
'HOST'
) t1
as
host
where
get_json_object(t0.unitparams,
'$.store.fruit[0].weight'
) != 9
union
all
select
host,
'ALL'
c0,
'ALL'
c1,
'ALL'
c2
from
t_log t0
LATERAL
VIEW
parse_url_tuple(url,
'HOST'
) t1
as
host
where
get_json_object(t0.unitparams,
'$.store.fruit[0].weight'
) != 9) test
group
by
c0, c1, c2;
|
2、不同属性的多维组合统计
这种场景下我们一般选择 Multi Table/File Inserts，下面选自《programming hive》P124
Making Multiple Passes over the Same Data
Hive has a special syntax for producing multiple aggregations from a single pass
through a source of data, rather than rescanning it for each aggregation. This change
can save considerable processing time for large input data sets. We discussed the details
previously in Chapter 5.
For example, each of the following two queries creates a table from the same source
table, history:
hive> INSERT OVERWRITE TABLE sales
> SELECT * FROM history WHERE action='purchased';
hive> INSERT OVERWRITE TABLE credits
> SELECT * FROM history WHERE action='returned';
This syntax is correct, but inefficient. The following rewrite achieves the same thing,
but using a single pass through the source history table:
hive> FROM history
> INSERT OVERWRITE sales SELECT * WHERE action='purchased'
> INSERT OVERWRITE credits SELECT * WHERE action='returned';
|
1
2
|
FROM
pv_users
INSERT
OVERWRITE
TABLE
pv_gender_sum
SELECT
pv_users.gender, count_distinct(pv_users.userid)
GROUP
BY
pv_users.gender
INSERT
OVERWRITE DIRECTORY
'/user/data/tmp/pv_age_sum'
SELECT
pv_users.age, count_distinct(pv_users.userid)
GROUP
BY
pv_users.age;
|
https://cwiki.apache.org/confluence/display/Hive/Tutorial
注意事项以及一些小技巧：
1、hive union all 的用法：不支持 top level，以及各个select字段名称、属性必须严格一致
2、结果的顺序问题，可以自己加字符控制排序
3、多重insert和union all一样也只扫描一次，但因为要insert到多个分区，所以做了很多其他的事情，导致消耗的时间非常长，其会产生多个job，union all 本身只有一个job
关于 insert overwrite 产生多 job 并行执行的问题：
set hive.exec.parallel=true; //打开任务并行执行
set hive.exec.parallel.thread.number=16; //同一个sql允许最大并行度，默认为8。
http://superlxw1234.iteye.com/blog/1703713
4、当前HIVE 不支持 not in 中包含查询子句的语法，形如如下的HQ语句是不被支持的:
查询在key字段在a表中，但不在b表中的数据
select a.key from a where key not in(select key from b) 该语句在hive中不支持
可以通过left outer join进行查询,（假设B表中包含另外的一个字段 key1
select a.key from a left outer join b on a.key=b.key where b.key1 is null
5、left out join 不能连续3个以上使用，必须2个一组，2个一组包装起来使用。
|
1
2
3
4
5
6
7
8
9
10
|
select
p.ssi,p.pv,p.uv,p.nuv,p.visits,
'2012-06-19 17:00:00'
from
(
select
*
from
(
select
*
from
(
select
ssi,
count
(1) pv,
sum
(visits) visits
from
FactClickAnalysis
where
logTime <=
'2012-06-19 18:00:00'
and
logTime >=
'2012-06-19 17:00:00'
group
by
ssi ) p1
left
outer
join
(
select
ssi,
count
(1) uv
from
(
select
ssi,cookieid
from
FactClickAnalysis
where
logTime <=
'2012-06-19 18:00:00'
and
logTime >=
'2012-06-19 17:00:00'
group
by
ssi,cookieid ) t1
group
by
ssi
) p2
on
p1.ssi=p2.ssi
) p3
left
outer
join
(
select
ssi,
count
(1) nuv
from
FactClickAnalysis
where
logTime = insertTime
and
logTime <=
'2012-06-19 18:00:00'
and
logTime >=
'2012-06-19 17:00:00'
group
by
ssi
) p4
on
p3.ssi=p4.ssi
) p
|
6、hive本地执行mr
http://superlxw1234.iteye.com/blog/1703546
7、hive动态分区创建过多遇到的一个错误
http://superlxw1234.iteye.com/blog/1677938
8、hive中巧用正则表达式的贪婪匹配
http://superlxw1234.iteye.com/blog/1751216
9、hive匹配全中文字段
用java中匹配中文的正则即可：
name rlike '^[\\u4e00-\\u9fa5]+$'
判断一个字段是否全数字：
select mobile from woa_login_log_his where pt = '2012-01-10' and mobile rlike '^\\d+$' limit 50;
10、hive中使用sql window函数 LAG/LEAD/FIRST/LAST
http://superlxw1234.iteye.com/blog/1600323
http://www.shaoqun.com/a/18839.aspx
11、hive优化之------控制hive任务中的map数和reduce数
http://superlxw1234.iteye.com/blog/1582880
12、hive中转义$等特殊字符
http://superlxw1234.iteye.com/blog/1568739
13、日期处理：
查看N天前的日期：
select from_unixtime(unix_timestamp('20111102','yyyyMMdd') - N*86400,'yyyyMMdd') from t_lxw_test1 limit 1;
获取两个日期之间的天数/秒数/分钟数等等：
select ( unix_timestamp('2011-11-02','yyyy-MM-dd')-unix_timestamp('2011-11-01','yyyy-MM-dd') ) / 86400 from t_lxw_test limit 1;
14、删除 Hive 临时文件 hive.exec.scratchdir
http://hi.baidu.com/youziguo/item/1dd7e6315dcc0f28b2c0c576
REF：
http://superlxw1234.iteye.com/blog/1536440
http://liubingwwww.blog.163.com/blog/static/3048510720125201749323/
http://blog.csdn.net/azhao_dn/article/details/6921429
