本地数据仓库项目(一) —— 本地数仓搭建详细流程

1 准备

1.1 本地虚拟机器准备

使用本地搭建三台机器，搭建数仓，模拟实际平台数据仓库的搭建。具体的搭建框架如下

安装软件	角色	主机1	主机2	主机3
Hadoop	NameNode	√
	SecondaryNameNode		√
	DataNode	√	√	√
	NodeManager	√	√	√
	ResourceManager	√	√
Zookeeper	zk server	√	√	√
Flume	flume			√
Kafka	kafka	√	√	√
Hive	hive	√
Mysql	mysql	√	√

1.2 项目技术选型

类型	技术
数据采集	Flum,Kafka,Sqoop,Logstash,DataX
数据存储	Mysql,HDFS,Hbase,Redis,MongoDB
数据计算	HIve,Tez,Spark,Flink,Storm
数据查询	Presto,Druid,Impala,Kylin

1.3 数据来源

数据来自本地模拟创建的数据，数据格式内容如下：

{
"ap":"xxxxx",//项目数据来源 app pc
"cm": {  //公共字段
		"mid": "",  // (String) 设备唯一标识
        "uid": "",  // (String) 用户标识
        "vc": "1",  // (String) versionCode，程序版本号
        "vn": "1.0",  // (String) versionName，程序版本名
        "l": "zh",  // (String) language系统语言
        "sr": "",  // (String) 渠道号，应用从哪个渠道来的。
        "os": "7.1.1",  // (String) Android系统版本
        "ar": "CN",  // (String) area区域
        "md": "BBB100-1",  // (String) model手机型号
        "ba": "blackberry",  // (String) brand手机品牌
        "sv": "V2.2.1",  // (String) sdkVersion
        "g": "",  // (String) gmail
        "hw": "1620x1080",  // (String) heightXwidth，屏幕宽高
        "t": "1506047606608",  // (String) 客户端日志产生时的时间
        "nw": "WIFI",  // (String) 网络模式
        "ln": 0,  // (double) lng经度
        "la": 0  // (double) lat 纬度
    },
"et":  [  //事件
            {
                "ett": "1506047605364",  //客户端事件产生时间
                "en": "display",  //事件名称
                "kv": {  //事件结果，以key-value形式自行定义
                    "goodsid": "236",
                    "action": "1",
                    "extend1": "1",
"place": "2",
"category": "75"
                }
            }
        ]
}

样例：

1540934156385|{
    "ap": "gmall", 
    "cm": {
        "uid": "1234", 
        "vc": "2", 
        "vn": "1.0", 
        "la": "EN", 
        "sr": "", 
        "os": "7.1.1", 
        "ar": "CN", 
        "md": "BBB100-1", 
        "ba": "blackberry", 
        "sv": "V2.2.1", 
        "g": "abc@gmail.com", 
        "hw": "1620x1080", 
        "t": "1506047606608", 
        "nw": "WIFI", 
        "ln": 0
    }, 
        "et": [
            {
                "ett": "1506047605364",  //客户端事件产生时间
                "en": "display",  //事件名称
                "kv": {  //事件结果，以key-value形式自行定义
                    "goodsid": "236",
                    "action": "1",
                    "extend1": "1",
"place": "2",
"category": "75"
                }
            },{
		        "ett": "1552352626835",
		        "en": "active_background",
		        "kv": {
			         "active_source": "1"
		        }
	        }
        ]
    }
}

2 软件安装

2.1 安装hadoop

2.1.1 配置支持LZO压缩

安装过程不再赘述，注意这里的hadoop需要支持LZO压缩格式。配置core-site.yml文件内容如下，支持LZO

<property>
<name>io.compression.codecs</name>
<value>
org.apache.hadoop.io.compress.GzipCodec,
org.apache.hadoop.io.compress.DefaultCodec,
org.apache.hadoop.io.compress.BZip2Codec,
org.apache.hadoop.io.compress.SnappyCodec,
com.hadoop.compression.lzo.LzoCodec,
com.hadoop.compression.lzo.LzopCodec
</value>
</property>
<property>
    <name>io.compression.codec.lzo.class</name>
    <value>com.hadoop.compression.lzo.LzoCodec</value>
</property>

2.1.2 进行HDFS读写性能测试

向HDFS集群写10个128M的文件

hadoop jar /opt/module/hadoop-2.7.2/share/hadoop/mapreduce/hadoop-mapreduce-client-jobclient-2.7.2-tests.jar TestDFSIO -write -nrFiles 10 -fileSize 128MB

读取HDFS集群10个128M的文件

hadoop jar /opt/module/hadoop-2.7.2/share/hadoop/mapreduce/hadoop-mapreduce-client-jobclient-2.7.2-tests.jar TestDFSIO -read -nrFiles 10 -fileSize 128MB

2.1.3 hadoop调优

① HDFS参数调优hdfs-site.xm
dfs.namenode.handler.count=20 * log2(Cluster Size)，比如集群规模为8台时，此参数设置为60

The number of Namenode RPC server threads that listen to requests from clients. If dfs.namenode.servicerpc-address is not configured then Namenode RPC server threads listen to requests from all nodes.
NameNode有一个工作线程池，用来处理不同DataNode的并发心跳以及客户端并发的元数据操作。对于大集群或者有大量客户端的集群来说，通常需要增大参数dfs.namenode.handler.count的默认值10。设置该值的一般原则是将其设置为集群大小的自然对数乘以20，即20logN，N为集群大小。

② 编辑日志存储路径dfs.namenode.edits.dir设置与镜像文件存储路径dfs.namenode.name.dir尽量分开，达到最低写入延迟
③ YARN参数调优yarn-site.xml

情景描述：总共7台机器，每天几亿条数据，数据源->Flume->Kafka->HDFS->Hive
面临问题：数据统计主要用HiveSQL，没有数据倾斜，小文件已经做了合并处理，开启的JVM重用，而且IO没有阻塞，内存用了不到50%。但是还是跑的非常慢，而且数据量洪峰过来时，整个集群都会宕掉。基于这种情况有没有优化方案。

解决办法：

内存利用率不够。这个一般是Yarn的2个配置造成的，单个任务可以申请的最大内存大小，
和Hadoop单个节点可用内存大小。调节这两个参数能提高系统内存的利用率。

（a）yarn.nodemanager.resource.memory-mb
表示该节点上YARN可使用的物理内存总量，默认是8192（MB），注意，如果你的节点内存资源不够8GB，则需要调减小这个值，而YARN不会智能的探测节点的物理内存总量。
（b）yarn.scheduler.maximum-allocation-mb
单个任务可申请的最多物理内存量，默认是8192（MB）。

2.2 安装zookeeper

三节点安装三个，安装的大致流程可以看这篇文章，这里具体安装流程不再赘述

2.3 安装Flume

具体安装流程可以看这篇文章

2.4 安装Kafka

安装具体可以看这篇文章

2.4.1 压测

用Kafka官方自带的脚本，对Kafka进行压测。Kafka压测时，可以查看到哪个地方出现了瓶颈（CPU，内存，网络IO）。一般都是网络IO达到瓶颈。
kafka-consumer-perf-test.sh
kafka-producer-perf-test.sh

bin/kafka-producer-perf-test.sh  --topic test --record-size 100 --num-records 100000 
--throughput 1000 --producer-props bootstrap.servers=wavehouse-1:9092,wavehouse-2:9092
,wavehouse-3:9092
#record-size是一条信息有多大，单位是字节。num-records是总共发送多少条信息。
#throughput 是每秒多少条信息。

2.4.2 kafka数量计算

Kafka机器数量（经验公式）=2*（峰值生产速度副本数/100）+1
先拿到峰值生产速度，再根据设定的副本数，就能预估出需要部署Kafka的数量。 比如我们的峰值生产速度是50M/s。副本数为2。
Kafka机器数量=2（50*2/100）+ 1=3台

3 dwd数据明细层

3.1 安装Mysql

3.1.1 Keepalived介绍

这里使用keepalived实现Mysql的高可用，Keepalived是基于VRRP（Virtual Router Redundancy Protocol，虚拟路由器冗余协议）协议的一款高可用软件。Keepailived有一台主服务器（master）和多台备份服务器（backup），在主服务器和备份服务器上面部署相同的服务配置，使用一个虚拟IP地址对外提供服务，当主服务器出现故障时，虚拟IP地址会自动漂移到备份服务器。

3.1.2 配置双主模式

wavehouse-1	Mysql(master,slave)
wavehouse-2	Mysql(slave,master)

wavehouse-1节点中MySQL的/usr/my.cnf配置文件

[mysqld]
#开启binlog
log_bin = mysql-bin
#binlog日志类型
binlog_format = row
#MySQL服务器唯一id
server_id = 2
#开启slave中继日志
relay_log=mysql-relay

重启wavehouse-1的mysql服务

service mysql restart

配置wavehouse-2节点

[mysqld]
#MySQL服务器唯一id
server_id = 1

#开启binlog
log_bin = mysql-bin
#binlog日志类型
binlog_format = row
#开启slave中继日志
relay_log=mysql-relay

重启Mysql服务

service mysql restart

配置wavehouse-1节点keepalived配置文件/etc/keepalived/keepalived.conf

global_defs {
    router_id MySQL-ha
}
vrrp_instance VI_1 {
    state master #初始状态
    interface eth0 #网卡
    virtual_router_id 51 #虚拟路由id
    priority 100 #优先级
    advert_int 1 #Keepalived心跳间隔
    nopreempt #只在高优先级配置，原master恢复之后不重新上位
    authentication {
        auth_type PASS #认证相关
        auth_pass 1111
    }
    virtual_ipaddress {
        192.168.1.100 #虚拟ip
    }
} 

#声明虚拟服务器
virtual_server 192.168.1.100 3306 {
    delay_loop 6
    persistence_timeout 30
    protocol TCP
    #声明真实服务器
    real_server 192.168.1.103 3306 {
        notify_down /var/lib/mysql/killkeepalived.sh #真实服务故障后调用脚本
        TCP_CHECK {
            connect_timeout 3 #超时时间
            nb_get_retry 1 #重试次数
            delay_before_retry 1 #重试时间间隔
        }
    }
}

编辑脚本文件/var/lib/mysql/killkeepalived.sh

#! /bin/bash
service keepalived stop

启动keepalived服务

service keepalived start

配置wavehouse-2节点/etc/keepalived/keepalived.conf

! Configuration File for keepalived
global_defs {
    router_id MySQL-ha
}
vrrp_instance VI_1 {
    state master #初始状态
    interface eth0 #网卡
    virtual_router_id 51 #虚拟路由id
    priority 100 #优先级
    advert_int 1 #Keepalived心跳间隔
    authentication {
        auth_type PASS #认证相关
        auth_pass 1111
    }
    virtual_ipaddress {
        192.168.1.100 #虚拟ip
    }
} 

#声明虚拟服务器
virtual_server 192.168.1.100 3306 {
    delay_loop 6
    persistence_timeout 30
    protocol TCP
    #声明真实服务器
    real_server 192.168.1.104 3306 {
        notify_down /var/lib/mysql/killkeepalived.sh #真实服务故障后调用脚本
        TCP_CHECK {
            connect_timeout 3 #超时时间
            nb_get_retry 1 #重试次数
            delay_before_retry 1 #重试时间间隔
        }
    }
}

其他修改内容同wavehouse-1节点

3.2 安装Hive

3.2.1 准备条件

① 保证环境已安装jdk，且JAVA_HOME变量存在
② hadoop已完成安装，HADOOP_HOME已配置
③ 配置HIVE_HOME
④ 拷贝Mysql驱动jar包到Hive的lib目录下

3.2.2 配置文件修改

hive-site.xml

<?xml version="1.0"?>
<?xml-stylesheet type="text/xsl" href="configuration.xsl"?>
<configuration>
	<property>
	  <name>javax.jdo.option.ConnectionURL</name>
	  <value>jdbc:mysql://192.168.2.100:3306/metastore?createDatabaseIfNotExist=true</value>
	  <description>JDBC connect string for a JDBC metastore</description>
	</property>
	<property>
	  <name>javax.jdo.option.ConnectionDriverName</name>
	  <value>com.mysql.jdbc.Driver</value>
	  <description>Driver class name for a JDBC metastore</description>
	</property>

	<property>
	  <name>javax.jdo.option.ConnectionUserName</name>
	  <value>root</value>
	  <description>username to use against metastore database</description>
	</property>
	<property>
	  <name>javax.jdo.option.ConnectionPassword</name>
	  <value>123456</value>
	  <description>password to use against metastore database</description>
	</property>
</configuration>

注意这里jdbc的链接Ip要配置为虚拟ip，这里的链接密码根据前面mysql设置的密码填写。

3.2.3 在mysql中创建metastore表

metastore表，并指定字符集为latin1

以上配置完成后，启动hive成功后会在mysql的metastore数据库生成很多hive的原始数据表

3.3 安装Tez

Tez是一个Hive的运行引擎，性能优于MR。为什么优于MR呢？看下图。

用Hive直接编写MR程序，假设有四个有依赖关系的MR作业，上图中，绿色是Reduce Task，云状表示写屏蔽，需要将中间结果持久化写到HDFS。
Tez可以将多个有依赖的作业转换为一个作业，这样只需写一次HDFS，且中间节点较少，从而大大提升作业的计算性能。

3.3.1 解压Tez包并上传到hdfs

由于Hive只安装在wavehouse-1节点，当执行任务时，需要到其他集群执行任务时，其他节点没有tez的包，因此上传到hdfs当需要使用时，从hdfs去获取。

hadoop fs -mkdir /tez
hadoop fs -put /root/soft/apache-tez-0.9.1-bin.tar.gz /tez

3.3.2 配置hive中的tez

在hive的conf目录下新建tez-site.xml文件，并加入如下内容。这里配置tez在hdfs中的路径

<?xml version="1.0" encoding="UTF-8"?>
<?xml-stylesheet type="text/xsl" href="configuration.xsl"?>
<configuration>
<property>
	<name>tez.lib.uris</name>
    <value>${fs.defaultFS}/tez/apache-tez-0.9.1-bin.tar.gz</value>
</property>
<property>
     <name>tez.use.cluster.hadoop-libs</name>
     <value>true</value>
</property>
<property>
     <name>tez.history.logging.service.class</name> 
     <value>org.apache.tez.dag.history.logging.ats.ATSHistoryLoggingService</value>
</property>
</configuration>

3.3.3 修改hive-env.sh

添加如下内容，让hive启动时加载Tez引擎的jar包

export TEZ_HOME=/opt/module/tez-0.9.1    #是你的tez的解压目录
export TEZ_JARS=""
for jar in `ls $TEZ_HOME |grep jar`; do
    export TEZ_JARS=$TEZ_JARS:$TEZ_HOME/$jar
done
for jar in `ls $TEZ_HOME/lib`; do
    export TEZ_JARS=$TEZ_JARS:$TEZ_HOME/lib/$jar
done

export HIVE_AUX_JARS_PATH=/opt/module/hadoop-2.7.2/share/hadoop/common/hadoop-lzo-0.4.20.jar$TEZ_JARS

3.3.4 hive-site.xml修改为tez引擎

<property>
    <name>hive.execution.engine</name>
    <value>tez</value>
</property>

3.3.5 修改hive-site.xml元数据校验

0.9版本的Tez有对应的hive版本，为了不触发校验，设置配置

<property>
    <name>hive.metastore.schema.verification</name>
    <value>false</value>
</property>

3.3.6 测试

启动hive，并执行建表语句

create table student(
id int,
name string);

插入数据

insert into student values(1,"zhangsan");

看到如下页面则说明，Tez已部署好

3.3.7 可能遇到的问题

以Tez为引擎执行时报错：

Caused by: org.apache.tez.dag.api.SessionNotRunning: TezSession has already shutdown. Application application_1546781144082_0005 failed 2 times due to AM Container for appattempt_1546781144082_0005_000002 exited with exitCode: -103
For more detailed output, check application tracking page:http://wavehouse-1:8088/cluster/app/application_1546781144082_0005Then, click on links to logs of each attempt.
Diagnostics: Container [pid=11116,containerID=container_1546781144082_0005_02_000001] is running beyond virtual memory limits. Current usage: 216.3 MB of 1 GB physical memory used; 2.6 GB of 2.1 GB virtual memory used. Killing container.

解决方案：
这种问题是从机上运行的Container试图使用过多的内存，而被NodeManager kill掉了。
关掉虚拟内存检查，修改yarn-site.xml，修改后一定要分发，并重新启动hadoop集群。

<property>
<name>yarn.nodemanager.vmem-check-enabled</name>
<value>false</value>
</property>

3 ods原始数据层导入数据

3.1 创建数据库

create database gmall;

use gmall;

3.2 创建启动日志表

CREATE EXTERNAL TABLE ods_start_log (`line` string) PARTITIONED BY (`dt` string)
stored as INPUTFORMAT 'com.hadoop.mapred.DeprecatedLzoTextInputFormat' 
OUTPUTFORMAT 'org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat'
LOCATION '/wavehouse/gmall/ods/ods_start_log'

3.3 加载数据

load data inpath '/origin_data/gmall/log/topic_start/2022-12-05' into table gmall.ods_start_log partition(dt='2022-12-05');

查看是否加载成功

select * from ods_start_log limit 2;

3.4 创建索引

为lzo压缩文件创建索引,lzo压缩文件的可切片特性依赖于其索引，故我们需要手动为lzo压缩文件创建索引。若无索引，则lzo文件的切片只有一个。

hadoop jar /root/soft/hadoop-2.7.2/share/hadoop/common/hadoop-lzo-0.4.20.jar 
com.hadoop.compression.lzo.DistributedLzoIndexer
/user/hive/warehouse/gmall.db/ods_start_log/dt=2022-12-05

4 DWD数据明细层

4.1 创建启动表和事件表

根据ods层中json格式数据字段，在dwd层创建表。

drop table if exists dwd_start_log;
CREATE EXTERNAL TABLE dwd_start_log(
`mid_id` string,
`user_id` string, 
`version_code` string, 
`version_name` string, 
`lang` string, 
`source` string, 
`os` string, 
`area` string, 
`model` string,
`brand` string, 
`sdk_version` string, 
`gmail` string, 
`height_width` string,  
`app_time` string,
`network` string, 
`lng` string, 
`lat` string, 
`entry` string, 
`open_ad_type` string, 
`action` string, 
`loading_time` string, 
`detail` string, 
`extend1` string
)
PARTITIONED BY (dt string)
stored as parquet
location '/warehouse/gmall/dwd/dwd_start_log/'
TBLPROPERTIES('parquet.compression'='lzo');

创建事件表

drop table if exists ods_event_log;
CREATE EXTERNAL TABLE ods_event_log(`line` string)
PARTITIONED BY (`dt` string)
STORED AS
  INPUTFORMAT 'com.hadoop.mapred.DeprecatedLzoTextInputFormat'
  OUTPUTFORMAT 'org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat'
LOCATION '/wavehouse/gmall/ods/ods_event_log';

4.2 导入数据

导入数据到原始层启动表

load data inpath '/origin_data/gmall/log/topic_start/2022-12-15' into table gmall.ods_start_log partition(dt='2022-12-15');
load data inpath '/origin_data/gmall/log/topic_start/2022-12-16' into table gmall.ods_start_log partition(dt='2022-12-16');

导入数据到数据原始层事件表

load data inpath '/origin_data/gmall/log/topic_event/2022-12-16' into table gmall.ods_event_log partition(dt='2022-12-16');
load data inpath '/origin_data/gmall/log/topic_event/2022-12-15' into table gmall.ods_event_log partition(dt='2022-12-15');

4.3 创建索引

hadoop jar /root/soft/hadoop-2.7.2/share/hadoop/common/hadoop-lzo-0.4.20.jar com.hadoop.compression.lzo.DistributedLzoIndexer /wavehouse/gmall/ods/ods_start_log/dt=2022-12-15
hadoop jar /root/soft/hadoop-2.7.2/share/hadoop/common/hadoop-lzo-0.4.20.jar com.hadoop.compression.lzo.DistributedLzoIndexer /wavehouse/gmall/ods/ods_start_log/dt=2022-12-16

hadoop jar /root/soft/hadoop-2.7.2/share/hadoop/common/hadoop-lzo-0.4.20.jar  com.hadoop.compression.lzo.DistributedLzoIndexer /wavehouse/gmall/ods/ods_event_log/dt=2022-12-15
hadoop jar /root/soft/hadoop-2.7.2/share/hadoop/common/hadoop-lzo-0.4.20.jar  com.hadoop.compression.lzo.DistributedLzoIndexer /wavehouse/gmall/ods/ods_event_log/dt=2022-12-16

添加完索引，可以在HDFS浏览中查看到相关索引

注： 此时原始数据层，数据格式如下：

接下来将会在数据明细层DWD进行处理

5 数据明细层DWD

5.1 预备知识

5.1.1 get_json_object函数

A limited version of JSONPath supported: $ : Root object
代表整个JSON对象 . : Child operator 代表获取JSON对象中子元素(属性)的操作符 [] :
Subscript operator for array 获取JSONArray中的某个元素

如下json字符串：

{"name":"jack","age":18,"parents":[{"name":"oldjack","age":48},{"name":"jackmom","age":49}]}

获取18

select get_json_object('{"name":"jack","age":18,"parents":
[{"name":"oldjack","age":48},{"name":"jackmom","age":48}]}','$.age')

获取49

select get_json_object('{"name":"jack","age":18,"parents":
[{"name":"oldjack","age":48},{"name":"jackmom","age":49}]}','$.parents[1].age')

5.2 创建dwd明细表

这里创建表的字段按照ods层中的json数据格式进行，先创建启动明细表

drop table if exists dwd_start_log;
CREATE EXTERNAL TABLE dwd_start_log(
`mid_id` string,
`user_id` string, 
`version_code` string, 
`version_name` string, 
`lang` string, 
`source` string, 
`os` string, 
`area` string, 
`model` string,
`brand` string, 
`sdk_version` string, 
`gmail` string, 
`height_width` string,  
`app_time` string,
`network` string, 
`lng` string, 
`lat` string, 
`entry` string, 
`open_ad_type` string, 
`action` string, 
`loading_time` string, 
`detail` string, 
`extend1` string
)
PARTITIONED BY (dt string)
stored as parquet
location '/wavehouse/gmall/dwd/dwd_start_log/'
TBLPROPERTIES('parquet.compression'='lzo');

创建事件明细表

drop table if exists dwd_base_event_log;
CREATE EXTERNAL TABLE dwd_base_event_log(
`mid_id` string,
`user_id` string, 
`version_code` string, 
`version_name` string, 
`lang` string, 
`source` string, 
`os` string, 
`area` string, 
`model` string,
`brand` string, 
`sdk_version` string, 
`gmail` string, 
`height_width` string, 
`app_time` string, 
`network` string, 
`lng` string, 
`lat` string, 
`event_name` string, 
`event_json` string, 
`server_time` string)
PARTITIONED BY (`dt` string)
stored as parquet
location '/wavehouse/gmall/dwd/dwd_base_event_log/'
TBLPROPERTIES('parquet.compression'='lzo');

5.3 导入数据

按照不同的分区进行导入，其他日期的数据修改相应的分区和查询条件即可。

insert overwrite table dwd_start_log
PARTITION (dt='2022-12-16')
select 
    get_json_object(line,'$.mid') mid_id,
    get_json_object(line,'$.uid') user_id,
    get_json_object(line,'$.vc') version_code,
    get_json_object(line,'$.vn') version_name,
    get_json_object(line,'$.l') lang,
    get_json_object(line,'$.sr') source,
    get_json_object(line,'$.os') os,
    get_json_object(line,'$.ar') area,
    get_json_object(line,'$.md') model,
    get_json_object(line,'$.ba') brand,
    get_json_object(line,'$.sv') sdk_version,
    get_json_object(line,'$.g') gmail,
    get_json_object(line,'$.hw') height_width,
    get_json_object(line,'$.t') app_time,
    get_json_object(line,'$.nw') network,
    get_json_object(line,'$.ln') lng,
    get_json_object(line,'$.la') lat,
    get_json_object(line,'$.entry') entry,
    get_json_object(line,'$.open_ad_type') open_ad_type,
    get_json_object(line,'$.action') action,
    get_json_object(line,'$.loading_time') loading_time,
    get_json_object(line,'$.detail') detail,
    get_json_object(line,'$.extend1') extend1
from ods_start_log 
where dt='2022-12-16';

在导入事件数据时需要使用自定义的UDF和UDTF函数，具体实现逻辑如下：

package com.chen.hive.function;

import org.apache.hadoop.hive.ql.exec.UDF;
import org.json.JSONException;
import org.json.JSONObject;

/**
 * @author:cc
 * @create: 2022-12-16 15:47
 * @Description: Never pleased by external gains, and never saddened by personal losses.
 * 1671184702500|{"cm":{"ln":"-89.8","sv":"V2.4.7","os":"8.0.6","g":"C5KV50A1@gmail.com","mid":"1","nw":"4G","l":"en",
 * "vc":"11","hw":"640*1136","ar":"MX","uid":"1","t":"1671119837180","la":"-32.4","md":"HTC      -6","vn":"1.2.2",
 * "ba":"HTC","sr":"E"},"ap":"app","et":[{"ett":"1671126951625","en":"newsdetail","kv":{"entry":"2","goodsid":"0",
 * "news_staytime":"0","loading_time":"12","action":"4","showtype":"3","catego      ry":"73","type1":""}},
 * {"ett":"1671093609828","en":"ad","kv":{"entry":"3","show_style":"3","action":"5","detail":"","source":"3",
 * "behavior":"1","content":"1","newstype":"0"}},{"ett":"1671172497391","en":"      active_foreground",
 * "kv":{"access":"","push_id":"3"}},{"ett":"1671139666780","en":"active_background","kv":{"active_source":"1"}}]}
 */

public class BaseFeildUDF extends UDF {

    public String evaluate(String line,String param) throws JSONException {

        //1 切割line数据
        String[] strings = line.split("\\|");
        //2 校验strings
        if(strings.length != 2||strings[1] == null){
            return "";
        }
        //3 根据传入参数返回相应的值
        String result = "";
        JSONObject jsonObject = new JSONObject(strings[1]);
        if(param.equalsIgnoreCase("et")){
            if(jsonObject.has("et")){
                result = jsonObject.getString("et");
            }
        }else if(param.equalsIgnoreCase("st")){
            result = strings[0].trim();
        }else {
            JSONObject cm = jsonObject.getJSONObject("cm");
            if(cm.has(param)){
                result =  cm.getString(param);
            }

        }
        return result;
    }

}

package com.chen.hive.function;

import org.apache.hadoop.hive.ql.exec.UDFArgumentException;
import org.apache.hadoop.hive.ql.metadata.HiveException;
import org.apache.hadoop.hive.ql.udf.generic.GenericUDTF;
import org.apache.hadoop.hive.serde2.objectinspector.ObjectInspector;
import org.apache.hadoop.hive.serde2.objectinspector.ObjectInspectorFactory;
import org.apache.hadoop.hive.serde2.objectinspector.StructObjectInspector;
import org.apache.hadoop.hive.serde2.objectinspector.primitive.PrimitiveObjectInspectorFactory;
import org.json.JSONArray;
import org.json.JSONException;
import org.json.JSONObject;

import java.util.ArrayList;
import java.util.List;
import java.util.concurrent.Semaphore;

/**
 * @author:cc
 * @create: 2022-12-16 16:12
 * @Description: Never pleased by external gains, and never saddened by personal losses.
 * 数据格式[{},{},{}]
 */
public class EventJsonUDTF extends GenericUDTF {

    //在函数运行之前被调用一次，作用是告诉MapTask,当前函数返回的结果类型和个数，以便MapTask在运行时，函数的返回值进行检查
    @Override
    public StructObjectInspector initialize(StructObjectInspector argOIs) throws UDFArgumentException {
        //当前返回的两列的字段别名
        List<String> fieldNames = new ArrayList<String>();
        fieldNames.add("event_name");
        fieldNames.add("event_json");
        //　当前返回的两列的类型检查器
        List<ObjectInspector> fieldOIs = new ArrayList<ObjectInspector>();
        fieldOIs.add(PrimitiveObjectInspectorFactory.javaStringObjectInspector);
        fieldOIs.add(PrimitiveObjectInspectorFactory.javaStringObjectInspector);
        return ObjectInspectorFactory.getStandardStructObjectInspector(fieldNames, fieldOIs);

    }

    //执行函数的功能，处理数据后调用forward()返回结果，args是传参传过来的参数
    // 返回的应该是2列Ｎ行的数据，
    @Override
    public void process(Object[] args) throws HiveException {
        //校验传参是否合理
        if (args.length == 0 || args[0] == null) return;
        //构建json对象
        try {
            JSONArray jsonArray = new JSONArray(args[0].toString());
            if (jsonArray == null || jsonArray.length() == 0) return;
            //遍历json数组
            for (int i = 0; i < jsonArray.length(); i++) {
                try {
                    //每遍历一次，需要写出去一行两列的数据，构建一个数组，存储一行两列的数据
                    String [] result = new String[2];
                    JSONObject jsonObject = jsonArray.getJSONObject(i);
                    //取事件类型
                    result[0] = jsonObject.getString("en");
                    result[1] = jsonObject.toString();
                    //将result写出
                    forward(result);
                }catch (Exception e){
                    //过滤有问题的数据，并继续开始遍历下一个{}
                    continue;
                }
            }
        } catch (JSONException e) {
            e.printStackTrace();
        }
    }
    @Override
    public void close() throws HiveException {

    }
}

打成jar包上传到hive目录下的auxlib目录，如果该目录不存在则新建该目录。注：这里打jar包的时候，如果hive版本比较老可能会遇到不能成功编译的问题，具体可看这篇文章。

5.4 重启hive

测试数据，并保存为/root/hivetest文件

1670356265677|{"cm":{"ln":"-66.0","sv":"V2.4.5","os":"8.1.8","g":"8S19MC59@gmail.com","mid":"994","nw":"4G","l":"pt","vc":"1","hw":"640*1136","ar":"MX","uid":"994","t":"1670342099717","la":"15.3","md":"HTC-0","vn":"1.3.8","ba":"HTC","sr":"Y"},"ap":"app","et":[{"ett":"1670335111586","en":"loading","kv":{"extend2":"","loading_time":"54","action":"1","extend1":"","type":"2","type1":"","loading_way":"2"}},{"ett":"1670337626239","en":"comment","kv":{"p_comment_id":1,"addtime":"1670276780061","praise_count":891,"other_id":0,"comment_id":6,"reply_count":103,"userid":3,"content":"瞅非衅鸯垂棠应"}},{"ett":"1670278354353","en":"praise","kv":{"target_id":9,"id":1,"type":3,"add_time":"1670320899047","userid":0}}]}

启动hive时将样例指定为特殊变量chen,方便调用。

hive -d chen=$(cat /root/hivetest)

5.5 创建函数

create function base_analizer as 'com.chen.hive.function.BaseFeildUDF'
create function flat_analizer as 'com.chen.hive.function.EventJsonUDTF'

测试：

5.6 利用自定义函数插入dwd层数据

利用自定义的函数，将数据经过处理之后插入dwd_base_event_log表

insert overwrite TABLE gmall.dwd_base_event_log PARTITION(dt='2022-12-16')
SELECT 
base_analizer(line,'mid') as mid_id, 
base_analizer(line,'uid') as user_id, 
base_analizer(line,'vc') as version_code, 
base_analizer(line,'vn') as version_name, 
base_analizer(line,'l') as lang, 
base_analizer(line,'sr') as source, 
base_analizer(line,'os') as os, 
base_analizer(line,'ar') as area, 
base_analizer(line,'md') as model, 
base_analizer(line,'ba') as brand, 
base_analizer(line,'sv') as sdk_version, 
base_analizer(line,'g') as gmail, 
base_analizer(line,'hw') as height_width, 
base_analizer(line,'t') as app_time, 
base_analizer(line,'nw') as network, 
base_analizer(line,'ln') as lng, 
base_analizer(line,'la') as lat, 
en event_name, 
ej event_json, 
base_analizer(line,'st') as server_time
FROM
gmall.ods_event_log LATERAL VIEW flat_analizer(base_analizer(line,'et')) tmp as en,ej
WHERE dt='2022-12-16';

这里的flat_analizer函数需要配合lateral view使用，该功能是一个侧写视图，根据返回的视图插入新表中。其他日期的数据修改分区时间和ods_event_log表查询时间即可插入。

5.7 创建明细表

根据dwd_base_event_log中不同的even_name创建不同类型的表。

5.7.1 商品点击表

drop table if exists dwd_display_log;
CREATE EXTERNAL TABLE dwd_display_log(
`mid_id` string,
`user_id` string,
`version_code` string,
`version_name` string,
`lang` string,
`source` string,
`os` string,
`area` string,
`model` string,
`brand` string,
`sdk_version` string,
`gmail` string,
`height_width` string,
`app_time` string,
`network` string,
`lng` string,
`lat` string,
`action` string,
`goodsid` string,
`place` string,
`extend1` string,
`category` string,
`server_time` string
)
PARTITIONED BY (dt string)
stored as parquet
location '/wavehouse/gmall/dwd/dwd_display_log/'
TBLPROPERTIES('parquet.compression'='lzo');

插入数据

insert overwrite table dwd_display_log
PARTITION (dt='2022-12-17')
select 
mid_id,
user_id,
version_code,
version_name,
lang,
source,
os,
area,
model,
brand,
sdk_version,
gmail,
height_width,
app_time,
network,
lng,
lat,
get_json_object(event_json,'$.kv.action') action,
get_json_object(event_json,'$.kv.goodsid') goodsid,
get_json_object(event_json,'$.kv.place') place,
get_json_object(event_json,'$.kv.extend1') extend1,
get_json_object(event_json,'$.kv.category') category,
server_time
from dwd_base_event_log 
where dt='2022-12-17' and event_name='display';

5.7.2 商品详情页表

drop table if exists dwd_newsdetail_log;
CREATE EXTERNAL TABLE dwd_newsdetail_log(
`mid_id` string,
`user_id` string, 
`version_code` string, 
`version_name` string, 
`lang` string, 
`source` string, 
`os` string, 
`area` string, 
`model` string,
`brand` string, 
`sdk_version` string, 
`gmail` string, 
`height_width` string, 
`app_time` string,  
`network` string, 
`lng` string, 
`lat` string, 
`entry` string,
`action` string,
`goodsid` string,
`showtype` string,
`news_staytime` string,
`loading_time` string,
`type1` string,
`category` string,
`server_time` string)
PARTITIONED BY (dt string)
stored as parquet
location '/wavehouse/gmall/dwd/dwd_newsdetail_log/'
TBLPROPERTIES('parquet.compression'='lzo');

插入数据

insert overwrite table dwd_newsdetail_log
PARTITION (dt='2022-12-17')
select 
mid_id,
user_id,
version_code,
version_name,
lang,
source,
os,
area,
model,
brand,
sdk_version,
gmail,
height_width,
app_time,
network,
lng,
lat,
get_json_object(event_json,'$.kv.entry') entry,
get_json_object(event_json,'$.kv.action') action,
get_json_object(event_json,'$.kv.goodsid') goodsid,
get_json_object(event_json,'$.kv.showtype') showtype,
get_json_object(event_json,'$.kv.news_staytime') news_staytime,
get_json_object(event_json,'$.kv.loading_time') loading_time,
get_json_object(event_json,'$.kv.type1') type1,
get_json_object(event_json,'$.kv.category') category,
server_time
from dwd_base_event_log
where dt='2022-12-17' and event_name='newsdetail';

5.7.3 商品列表页表

drop table if exists dwd_loading_log;
CREATE EXTERNAL TABLE dwd_loading_log(
`mid_id` string,
`user_id` string, 
`version_code` string, 
`version_name` string, 
`lang` string, 
`source` string, 
`os` string, 
`area` string, 
`model` string,
`brand` string, 
`sdk_version` string, 
`gmail` string,
`height_width` string,  
`app_time` string,
`network` string, 
`lng` string, 
`lat` string, 
`action` string,
`loading_time` string,
`loading_way` string,
`extend1` string,
`extend2` string,
`type` string,
`type1` string,
`server_time` string)
PARTITIONED BY (dt string)
stored as parquet
location '/wavehouse/gmall/dwd/dwd_loading_log/'
TBLPROPERTIES('parquet.compression'='lzo');

插入数据

insert overwrite table dwd_loading_log
PARTITION (dt='2022-12-17')
select 
mid_id,
user_id,
version_code,
version_name,
lang,
source,
os,
area,
model,
brand,
sdk_version,
gmail,
height_width,
app_time,
network,
lng,
lat,
get_json_object(event_json,'$.kv.action') action,
get_json_object(event_json,'$.kv.loading_time') loading_time,
get_json_object(event_json,'$.kv.loading_way') loading_way,
get_json_object(event_json,'$.kv.extend1') extend1,
get_json_object(event_json,'$.kv.extend2') extend2,
get_json_object(event_json,'$.kv.type') type,
get_json_object(event_json,'$.kv.type1') type1,
server_time
from dwd_base_event_log
where dt='2022-12-17' and event_name='loading';

5.7.4 广告表

drop table if exists dwd_ad_log;
CREATE EXTERNAL TABLE dwd_ad_log(
`mid_id` string,
`user_id` string, 
`version_code` string, 
`version_name` string, 
`lang` string, 
`source` string, 
`os` string, 
`area` string, 
`model` string,
`brand` string, 
`sdk_version` string, 
`gmail` string, 
`height_width` string,  
`app_time` string,
`network` string, 
`lng` string, 
`lat` string, 
`entry` string,
`action` string,
`content` string,
`detail` string,
`ad_source` string,
`behavior` string,
`newstype` string,
`show_style` string,
`server_time` string)
PARTITIONED BY (dt string)
stored as parquet
location '/wavehouse/gmall/dwd/dwd_ad_log/'
TBLPROPERTIES('parquet.compression'='lzo');

插入数据

insert overwrite table dwd_ad_log
PARTITION (dt='2022-12-17')
select 
mid_id,
user_id,
version_code,
version_name,
lang,
source,
os,
area,
model,
brand,
sdk_version,
gmail,
height_width,
app_time,
network,
lng,
lat,
get_json_object(event_json,'$.kv.entry') entry,
get_json_object(event_json,'$.kv.action') action,
get_json_object(event_json,'$.kv.content') content,
get_json_object(event_json,'$.kv.detail') detail,
get_json_object(event_json,'$.kv.source') ad_source,
get_json_object(event_json,'$.kv.behavior') behavior,
get_json_object(event_json,'$.kv.newstype') newstype,
get_json_object(event_json,'$.kv.show_style') show_style,
server_time
from dwd_base_event_log 
where dt='2022-12-17' and event_name='ad';

5.7.5 消息通知表

drop table if exists dwd_notification_log;
CREATE EXTERNAL TABLE dwd_notification_log(
`mid_id` string,
`user_id` string, 
`version_code` string, 
`version_name` string, 
`lang` string,
`source` string, 
`os` string, 
`area` string, 
`model` string,
`brand` string, 
`sdk_version` string, 
`gmail` string, 
`height_width` string,  
`app_time` string,
`network` string, 
`lng` string, 
`lat` string, 
`action` string,
`noti_type` string,
`ap_time` string,
`content` string,
`server_time` string
)
PARTITIONED BY (dt string)
stored as parquet
location '/wavehouse/gmall/dwd/dwd_notification_log/'
TBLPROPERTIES('parquet.compression'='lzo');

插入数据

insert overwrite table dwd_notification_log
PARTITION (dt='2022-12-17')
select 
mid_id,
user_id,
version_code,
version_name,
lang,
source,
os,
area,
model,
brand,
sdk_version,
gmail,
height_width,
app_time,
network,
lng,
lat,
get_json_object(event_json,'$.kv.action') action,
get_json_object(event_json,'$.kv.noti_type') noti_type,
get_json_object(event_json,'$.kv.ap_time') ap_time,
get_json_object(event_json,'$.kv.content') content,
server_time
from dwd_base_event_log
where dt='2022-12-17' and event_name='notification';

5.7.6 用户前台活跃表

drop table if exists dwd_active_foreground_log;
CREATE EXTERNAL TABLE dwd_active_foreground_log(
`mid_id` string,
`user_id` string,
`version_code` string,
`version_name` string,
`lang` string,
`source` string,
`os` string,
`area` string,
`model` string,
`brand` string,
`sdk_version` string,
`gmail` string,
`height_width` string,
`app_time` string,
`network` string,
`lng` string,
`lat` string,
`push_id` string,
`access` string,
`server_time` string)
PARTITIONED BY (dt string)
stored as parquet
location '/wavehouse/gmall/dwd/dwd_foreground_log/'
TBLPROPERTIES('parquet.compression'='lzo');

插入数据

insert overwrite table dwd_active_foreground_log
PARTITION (dt='2022-12-17')
select 
mid_id,
user_id,
version_code,
version_name,
lang,
source,
os,
area,
model,
brand,
sdk_version,
gmail,
height_width,
app_time,
network,
lng,
lat,
get_json_object(event_json,'$.kv.push_id') push_id,
get_json_object(event_json,'$.kv.access') access,
server_time
from dwd_base_event_log
where dt='2022-12-17' and event_name='active_foreground';

5.7.7 用户后台活跃表

drop table if exists dwd_active_background_log;
CREATE EXTERNAL TABLE dwd_active_background_log(
`mid_id` string,
`user_id` string,
`version_code` string,
`version_name` string,
`lang` string,
`source` string,
`os` string,
`area` string,
`model` string,
`brand` string,
`sdk_version` string,
`gmail` string,
 `height_width` string,
`app_time` string,
`network` string,
`lng` string,
`lat` string,
`active_source` string,
`server_time` string
)
PARTITIONED BY (dt string)
stored as parquet
location '/wavehouse/gmall/dwd/dwd_background_log/'
TBLPROPERTIES('parquet.compression'='lzo');

插入数据

insert overwrite table dwd_active_background_log
PARTITION (dt='2022-12-17')
select 
mid_id,
user_id,
version_code,
version_name,
lang,
source,
os,
area,
model,
brand,
sdk_version,
gmail,
height_width,
app_time,
network,
lng,
lat,
get_json_object(event_json,'$.kv.active_source') active_source,
server_time
from dwd_base_event_log
where dt='2022-12-17' and event_name='active_background';

5.7.8 评论表

drop table if exists dwd_comment_log;
CREATE EXTERNAL TABLE dwd_comment_log(
`mid_id` string,
`user_id` string,
`version_code` string,
`version_name` string,
`lang` string,
`source` string,
`os` string,
`area` string,
`model` string,
`brand` string,
`sdk_version` string,
`gmail` string,
`height_width` string,
`app_time` string,
`network` string,
`lng` string,
`lat` string,
`comment_id` int,
`userid` int,
`p_comment_id` int, 
`content` string,
`addtime` string,
`other_id` int,
`praise_count` int,
`reply_count` int,
`server_time` string
)
PARTITIONED BY (dt string)
stored as parquet
location '/wavehouse/gmall/dwd/dwd_comment_log/'
TBLPROPERTIES('parquet.compression'='lzo');

插入数据

insert overwrite table dwd_comment_log
PARTITION (dt='2022-12-17')
select 
mid_id,
user_id,
version_code,
version_name,
lang,
source,
os,
area,
model,
brand,
sdk_version,
gmail,
height_width,
app_time,
network,
lng,
lat,
get_json_object(event_json,'$.kv.comment_id') comment_id,
get_json_object(event_json,'$.kv.userid') userid,
get_json_object(event_json,'$.kv.p_comment_id') p_comment_id,
get_json_object(event_json,'$.kv.content') content,
get_json_object(event_json,'$.kv.addtime') addtime,
get_json_object(event_json,'$.kv.other_id') other_id,
get_json_object(event_json,'$.kv.praise_count') praise_count,
get_json_object(event_json,'$.kv.reply_count') reply_count,
server_time
from dwd_base_event_log
where dt='2022-12-17' and event_name='comment';

5.7.9 收藏表

drop table if exists dwd_favorites_log;
CREATE EXTERNAL TABLE dwd_favorites_log(
`mid_id` string,
`user_id` string, 
`version_code` string, 
`version_name` string, 
`lang` string, 
`source` string, 
`os` string, 
`area` string, 
`model` string,
`brand` string, 
`sdk_version` string, 
`gmail` string, 
`height_width` string,  
`app_time` string,
`network` string, 
`lng` string, 
`lat` string, 
`id` int, 
`course_id` int, 
`userid` int,
`add_time` string,
`server_time` string
)
PARTITIONED BY (dt string)
stored as parquet
location '/wavehouse/gmall/dwd/dwd_favorites_log/'
TBLPROPERTIES('parquet.compression'='lzo');

插入数据

insert overwrite table dwd_favorites_log
PARTITION (dt='2022-12-17')
select 
mid_id,
user_id,
version_code,
version_name,
lang,
source,
os,
area,
model,
brand,
sdk_version,
gmail,
height_width,
app_time,
network,
lng,
lat,
get_json_object(event_json,'$.kv.id') id,
get_json_object(event_json,'$.kv.course_id') course_id,
get_json_object(event_json,'$.kv.userid') userid,
get_json_object(event_json,'$.kv.add_time') add_time,
server_time
from dwd_base_event_log 
where dt='2022-12-17' and event_name='favorites';

5.7.10 点赞表

drop table if exists dwd_praise_log;
CREATE EXTERNAL TABLE dwd_praise_log(
`mid_id` string,
`user_id` string, 
`version_code` string, 
`version_name` string, 
`lang` string, 
`source` string, 
`os` string, 
`area` string, 
`model` string,
`brand` string, 
`sdk_version` string, 
`gmail` string, 
`height_width` string,  
`app_time` string,
`network` string, 
`lng` string, 
`lat` string, 
`id` string, 
`userid` string, 
`target_id` string,
`type` string,
`add_time` string,
`server_time` string
)
PARTITIONED BY (dt string)
stored as parquet
location '/wavehouse/gmall/dwd/dwd_praise_log/'
TBLPROPERTIES('parquet.compression'='lzo');

插入数据

insert overwrite table dwd_praise_log
PARTITION (dt='2022-12-17')
select 
mid_id,
user_id,
version_code,
version_name,
lang,
source,
os,
area,
model,
brand,
sdk_version,
gmail,
height_width,
app_time,
network,
lng,
lat,
get_json_object(event_json,'$.kv.id') id,
get_json_object(event_json,'$.kv.userid') userid,
get_json_object(event_json,'$.kv.target_id') target_id,
get_json_object(event_json,'$.kv.type') type,
get_json_object(event_json,'$.kv.add_time') add_time,
server_time
from dwd_base_event_log
where dt='2022-12-17' and event_name='praise';

5.7.11 错误日志表

drop table if exists dwd_error_log;
CREATE EXTERNAL TABLE dwd_error_log(
`mid_id` string,
`user_id` string, 
`version_code` string, 
`version_name` string, 
`lang` string, 
`source` string, 
`os` string, 
`area` string, 
`model` string,
`brand` string, 
`sdk_version` string, 
`gmail` string, 
`height_width` string,  
`app_time` string,
`network` string, 
`lng` string, 
`lat` string, 
`errorBrief` string, 
`errorDetail` string, 
`server_time` string)
PARTITIONED BY (dt string)
stored as parquet
location '/wavehouse/gmall/dwd/dwd_error_log/'
TBLPROPERTIES('parquet.compression'='lzo');

插入数据

insert overwrite table dwd_error_log
PARTITION (dt='2022-12-17')
select 
mid_id,
user_id,
version_code,
version_name,
lang,
source,
os,
area,
model,
brand,
sdk_version,
gmail,
height_width,
app_time,
network,
lng,
lat,
get_json_object(event_json,'$.kv.errorBrief') errorBrief,
get_json_object(event_json,'$.kv.errorDetail') errorDetail,
server_time
from dwd_base_event_log 
where dt='2022-12-17' and event_name='error';

至此DWD层的数据已经全部导入

6 DWS层

6.1 需求1

求每日、每周、每月活跃设备明细。

6.1.1 每日设备活跃明细

建表

drop table if exists dws_uv_detail_day;
create external table dws_uv_detail_day
(
    `mid_id` string COMMENT '设备唯一标识',
    `user_id` string COMMENT '用户标识', 
    `version_code` string COMMENT '程序版本号', 
    `version_name` string COMMENT '程序版本名', 
    `lang` string COMMENT '系统语言', 
    `source` string COMMENT '渠道号', 
    `os` string COMMENT '安卓系统版本', 
    `area` string COMMENT '区域', 
    `model` string COMMENT '手机型号', 
    `brand` string COMMENT '手机品牌', 
    `sdk_version` string COMMENT 'sdkVersion', 
    `gmail` string COMMENT 'gmail', 
    `height_width` string COMMENT '屏幕宽高',
    `app_time` string COMMENT '客户端日志产生时的时间',
    `network` string COMMENT '网络模式',
    `lng` string COMMENT '经度',
    `lat` string COMMENT '纬度'
)
partitioned by(dt string)
stored as parquet
location '/wavehouse/gmall/dws/dws_uv_detail_day';

插入数据

insert overwrite TABLE gmall.dws_uv_detail_day PARTITION(dt='2022-12-17')
SELECT 
mid_id, 
concat_ws('|',collect_set(user_id)) user_id, 
concat_ws('|',collect_set(version_code)) version_code, 
concat_ws('|',collect_set(version_name)) version_name, 
concat_ws('|',collect_set(lang)) lang, 
concat_ws('|',collect_set(source)) source, 
concat_ws('|',collect_set(user_id)) os, 
concat_ws('|',collect_set(area)) area, 
concat_ws('|',collect_set(model)) model, 
concat_ws('|',collect_set(brand)) brand, 
concat_ws('|',collect_set(sdk_version)) sdk_version, 
concat_ws('|',collect_set(gmail)) gmail, 
concat_ws('|',collect_set(height_width)) height_width, 
concat_ws('|',collect_set(app_time)) app_time, 
concat_ws('|',collect_set(network)) network, 
concat_ws('|',collect_set(lng)) lng, 
concat_ws('|',collect_set(lat)) lat
FROM
gmall.dwd_start_log
WHERE dt='2022-12-17'
GROUP BY mid_id;

6.1.2 每周设备活跃明细

drop table if exists dws_uv_detail_wk;
create external table dws_uv_detail_wk( 
    `mid_id` string COMMENT '设备唯一标识',
    `user_id` string COMMENT '用户标识', 
    `version_code` string COMMENT '程序版本号', 
    `version_name` string COMMENT '程序版本名', 
    `lang` string COMMENT '系统语言', 
    `source` string COMMENT '渠道号', 
    `os` string COMMENT '安卓系统版本', 
    `area` string COMMENT '区域', 
    `model` string COMMENT '手机型号', 
    `brand` string COMMENT '手机品牌', 
    `sdk_version` string COMMENT 'sdkVersion', 
    `gmail` string COMMENT 'gmail', 
    `height_width` string COMMENT '屏幕宽高',
    `app_time` string COMMENT '客户端日志产生时的时间',
    `network` string COMMENT '网络模式',
    `lng` string COMMENT '经度',
    `lat` string COMMENT '纬度',
    `monday_date` string COMMENT '周一日期',
    `sunday_date` string COMMENT  '周日日期' 
) COMMENT '活跃用户按周明细'
PARTITIONED BY (`wk_dt` string)
stored as parquet
location '/wavehouse/gmall/dws/dws_uv_detail_wk/';

插入数据

set hive.exec.dynamic.partition.mode=nonstrict;
insert overwrite TABLE gmall.dws_uv_detail_wk PARTITION(wk_dt)
SELECT 
mid_id, 
concat_ws('|',collect_set(user_id)) user_id, 
concat_ws('|',collect_set(version_code)) version_code, 
concat_ws('|',collect_set(version_name)) version_name, 
concat_ws('|',collect_set(lang)) lang, 
concat_ws('|',collect_set(source)) source, 
concat_ws('|',collect_set(user_id)) os, 
concat_ws('|',collect_set(area)) area, 
concat_ws('|',collect_set(model)) model, 
concat_ws('|',collect_set(brand)) brand, 
concat_ws('|',collect_set(sdk_version)) sdk_version, 
concat_ws('|',collect_set(gmail)) gmail, 
concat_ws('|',collect_set(height_width)) height_width, 
concat_ws('|',collect_set(app_time)) app_time, 
concat_ws('|',collect_set(network)) network, 
concat_ws('|',collect_set(lng)) lng, 
concat_ws('|',collect_set(lat)) lat,
date_add(next_day('2022-12-17','mo'),-7) monday_date,
date_add(next_day('2022-12-17','mo'),-1) sunday_date,
concat(date_add(next_day('2022-12-17','mo'),-7),'-',date_add(next_day('2022-12-17','mo'),-1)) wk_dt
FROM
gmall.dwd_start_log
WHERE dt BETWEEN date_add(next_day('2022-12-17','mo'),-7) AND date_add(next_day('2022-12-17','mo'),-1)
GROUP BY mid_id;

6.1.3 每月设备活跃明细

建表

drop table if exists dws_uv_detail_mn;
create external table dws_uv_detail_mn( 
    `mid_id` string COMMENT '设备唯一标识',
    `user_id` string COMMENT '用户标识', 
    `version_code` string COMMENT '程序版本号', 
    `version_name` string COMMENT '程序版本名', 
    `lang` string COMMENT '系统语言', 
    `source` string COMMENT '渠道号', 
    `os` string COMMENT '安卓系统版本', 
    `area` string COMMENT '区域', 
    `model` string COMMENT '手机型号', 
    `brand` string COMMENT '手机品牌', 
    `sdk_version` string COMMENT 'sdkVersion', 
    `gmail` string COMMENT 'gmail', 
    `height_width` string COMMENT '屏幕宽高',
    `app_time` string COMMENT '客户端日志产生时的时间',
    `network` string COMMENT '网络模式',
    `lng` string COMMENT '经度',
    `lat` string COMMENT '纬度'
) COMMENT '活跃用户按月明细'
PARTITIONED BY (`mn` string)
stored as parquet
location '/wavehouse/gmall/dws/dws_uv_detail_mn/';

插入数据

set hive.exec.dynamic.partition.mode=nonstrict;
insert overwrite TABLE gmall.dws_uv_detail_mn PARTITION(mn)
SELECT 
mid_id, 
concat_ws('|',collect_set(user_id)) user_id, 
concat_ws('|',collect_set(version_code)) version_code, 
concat_ws('|',collect_set(version_name)) version_name, 
concat_ws('|',collect_set(lang)) lang, 
concat_ws('|',collect_set(source)) source, 
concat_ws('|',collect_set(user_id)) os, 
concat_ws('|',collect_set(area)) area, 
concat_ws('|',collect_set(model)) model, 
concat_ws('|',collect_set(brand)) brand, 
concat_ws('|',collect_set(sdk_version)) sdk_version, 
concat_ws('|',collect_set(gmail)) gmail, 
concat_ws('|',collect_set(height_width)) height_width, 
concat_ws('|',collect_set(app_time)) app_time, 
concat_ws('|',collect_set(network)) network, 
concat_ws('|',collect_set(lng)) lng, 
concat_ws('|',collect_set(lat)) lat,
date_format('2022-12-17','yyyy-MM') mn
FROM
gmall.dwd_start_log
WHERE date_format(dt,'yyyy-MM') = date_format('2022-12-17','yyyy-MM') 
GROUP BY mid_id;

注意：周活/月活需要使用动态分区，因此需要开启非严格模式。动态分区会根据sql函数字段进行动态分区，由于日活是的日期是可以指定且固定的某一天，因此对于日活统计不需要动态分区。

set hive.exec.dynamic.partition.mode=nonstrict;

6.2 需求2

求每日新增用户数
首次联网使用应用的用户。如果一个用户首次打开某APP，那这个用户定义为新增用户；卸载再安装的设备，不会被算作一次新增。新增用户包括日新增用户、周新增用户、月新增用户。

6.2.1 建每日新增用户表

新增注册时间字段，用于判断是否为新用户

drop table if exists dws_new_mid_day;
create external table dws_new_mid_day
(
    `mid_id` string COMMENT '设备唯一标识',
    `user_id` string COMMENT '用户标识', 
    `version_code` string COMMENT '程序版本号', 
    `version_name` string COMMENT '程序版本名', 
    `lang` string COMMENT '系统语言', 
    `source` string COMMENT '渠道号', 
    `os` string COMMENT '安卓系统版本', 
    `area` string COMMENT '区域', 
    `model` string COMMENT '手机型号', 
    `brand` string COMMENT '手机品牌', 
    `sdk_version` string COMMENT 'sdkVersion', 
    `gmail` string COMMENT 'gmail', 
    `height_width` string COMMENT '屏幕宽高',
    `app_time` string COMMENT '客户端日志产生时的时间',
    `network` string COMMENT '网络模式',
    `lng` string COMMENT '经度',
    `lat` string COMMENT '纬度',
    `create_date`  string  comment '创建时间' 
)  COMMENT '每日新增设备信息'
stored as parquet
location '/wavehouse/gmall/dws/dws_new_mid_day/';

导入数据

INSERT INTO TABLE dws_new_mid_day
SELECT 
ud.mid_id, 
ud.user_id, 
ud.version_code, 
ud.version_name, 
ud.lang, 
ud.source, 
ud.os, 
ud.area, 
ud.model, 
ud.brand, 
ud.sdk_version, 
ud.gmail, 
ud.height_width, 
ud.app_time, 
ud.network, 
ud.lng, 
ud.lat, 
'2022-12-05'
FROM
dws_uv_detail_day ud LEFT JOIN dws_new_mid_day nm ON ud.mid_id = nm.mid_id
where ud.dt='2022-12-05' and nm.mid_id is NULL;

6.3 需求3

统计每日用户留存用户和留存率。

留存用户：某段时间内新增用户，经过一段时间后，又继续使用的用户。
留存率：留存用户占当时新增用户的比例
如：2月10日新增用户100人，2月11日这100人中有30人继续登录，留存率则为30%；2月12日有25人登录，则留存率25%

建表

drop table if exists dws_user_retention_day;
create external table dws_user_retention_day 
(
    `mid_id` string COMMENT '设备唯一标识',
    `user_id` string COMMENT '用户标识', 
    `version_code` string COMMENT '程序版本号', 
    `version_name` string COMMENT '程序版本名', 
    `lang` string COMMENT '系统语言', 
    `source` string COMMENT '渠道号', 
    `os` string COMMENT '安卓系统版本', 
    `area` string COMMENT '区域', 
    `model` string COMMENT '手机型号', 
    `brand` string COMMENT '手机品牌', 
    `sdk_version` string COMMENT 'sdkVersion', 
    `gmail` string COMMENT 'gmail', 
    `height_width` string COMMENT '屏幕宽高',
    `app_time` string COMMENT '客户端日志产生时的时间',
    `network` string COMMENT '网络模式',
    `lng` string COMMENT '经度',
    `lat` string COMMENT '纬度',
   `create_date`    string  comment '设备新增时间',
   `retention_day`  int comment '截止当前日期留存天数'
)  COMMENT '每日用户留存情况'
PARTITIONED BY (`dt` string)
stored as parquet
location '/wavehouse/gmall/dws/dws_user_retention_day/';

插入数据

INSERT overwrite TABLE dws_user_retention_day PARTITION(dt='2022-12-20')
SELECT 
t2.mid_id, 
t2.user_id, 
t2.version_code, 
t2.version_name, 
t2.lang, 
t2.source, 
t2.os, 
t2.area, 
t2.model, 
t2.brand, 
t2.sdk_version, 
t2.gmail, 
t2.height_width, 
t2.app_time, 
t2.network, 
t2.lng, 
t2.lat, 
t2.create_date, 
1 retention_day
FROM
dws_uv_detail_day t1 JOIN dws_new_mid_day t2 on t1.mid_id=t2.mid_id
WHERE t1.dt='2022-12-20' AND t2.create_date=date_add('2022-12-20',-1);

6.4 需求4

求沉默用户，沉默用户：指的是只在安装当天启动过，且启动时间是在一周前。
这里可以使用dws层的dws_uv_detail_day表进行查询沉默用户操作

6.5 本周回流用户数

本周回流=本周活跃-本周新增-上周活跃。使用日活明细表dws_uv_detail_day作为DWS层数据。
###6.6 最近连续三周活跃用户数
最近3周连续活跃的用户：通常是周一对前3周的数据做统计，该数据一周计算一次。

6.7 最近7天内连续3天活跃用户数

使用日活明细表dws_uv_detail_day作为DWS层数据

6.8 新收藏用户数

考虑到后面的多个需求会同时用到多张表中的数据, 如果每次都join操作, 则影响查询的效率. 可以先提前做一张宽表, 提高其他查询的执行效率。每个用户对每个商品的点击次数, 点赞次数, 收藏次数
建表

drop table if exists dws_user_action_wide_log;
CREATE EXTERNAL TABLE dws_user_action_wide_log(
    `mid_id` string COMMENT '设备id',
    `goodsid` string COMMENT '商品id',
    `display_count` string COMMENT '点击次数',
    `praise_count` string COMMENT '点赞次数',
    `favorite_count` string COMMENT '收藏次数')
PARTITIONED BY (`dt` string)
stored as parquet
location '/wavehouse/gmall/dws/dws_user_action_wide_log/'
TBLPROPERTIES('parquet.compression'='lzo');

导入数据

INSERT overwrite TABLE dws_user_action_wide_log PARTITION(dt='2022-12-20')
SELECT 
mid_id, 
goodsid, 
sum(display_count) display_count, 
sum(praise_count) praise_count, 
sum(favorite_count) favorite_count
FROM 
(
	SELECT
	mid_id,
	goodsid,
	COUNT(*) display_count,
	0 praise_count,
	0 favorite_count
	FROM
	dwd_display_log
	WHERE dt<='2022-12-20'
	GROUP BY mid_id,goodsid
UNION ALL
	SELECT
	mid_id,
	target_id as goodsid,
	0 display_count,
	COUNT(mid_id) praise_count,
	0 favorite_count
	FROM
	dwd_praise_log
	WHERE dt<='2022-12-20'
	GROUP BY mid_id,target_id
UNION ALL
	SELECT
	mid_id,
	course_id as goodsid,
	0 display_count,
	0 praise_count,
	COUNT(mid_id) favorite_count
	FROM
	dwd_favorites_log
	WHERE dt<='2022-12-20'
	GROUP BY mid_id,course_id
) tmp
GROUP BY mid_id,goodsid

6.9 需求9 各个商品点击次数top3的用户

使用日志数据用户行为宽表作为DWS层表

7 ADS 应用数据层

7.1 需求1

对应6.1中DWS层需求1，在ADS数据应用层创建表

drop table if exists ads_uv_count;
create external table ads_uv_count( 
    `dt` string COMMENT '统计日期',
    `day_count` bigint COMMENT '当日用户数量',
    `wk_count`  bigint COMMENT '当周用户数量',
    `mn_count`  bigint COMMENT '当月用户数量',
    `is_weekend` string COMMENT 'Y,N是否是周末,用于得到本周最终结果',
    `is_monthend` string COMMENT 'Y,N是否是月末,用于得到本月最终结果' 
) COMMENT '活跃设备数'
row format delimited fields terminated by '\t'
location '/wavehouse/gmall/ads/ads_uv_count/';

插入数据

INSERT into TABLE ads_uv_count
SELECT
'2022-12-18' dt,
day_count, 
wk_count, 
mn_count, 
if(date_add(next_day('2022-12-18','mo'),-1)='2022-12-18','Y','N') is_weekend, 
if(last_day('2022-12-18')='2022-12-18','Y','N') is_monthend
FROM
(SELECT count(mid_id) day_count FROM dws_uv_detail_day WHERE dt='2022-12-18') t1
JOIN
(SELECT count(mid_id) wk_count FROM dws_uv_detail_wk WHERE  wk_dt=concat(date_add(next_day('2022-12-18','mo'),-7),'-',date_add(next_day('2022-12-18','mo'),-1))) t2 
JOIN
(SELECT COUNT(mid_id) mn_count FROM dws_uv_detail_mn WHERE mn=date_format('2022-12-18','yyyy-MM')) t3;

7.2 需求2

根据dws层dws_new_mid_day表求聚合操作可以求出每天的新增用户，创建ads层表

drop table if exists ads_new_mid_count;
create external table ads_new_mid_count
(
    `create_date`     string comment '创建时间' ,
    `new_mid_count`   BIGINT comment '新增设备数量' 
)  COMMENT '每日新增设备信息数量'
row format delimited fields terminated by '\t'
location '/wavehouse/gmall/ads/ads_new_mid_count/';

导入数据

INSERT into TABLE ads_new_mid_count
SELECT 
create_date, 
count(*)
FROM
dws_new_mid_day
WHERE create_date='2022-12-05'
GROUP BY create_date;

7.3 需求3

求留存用户和留存率，根据dws层表数据进行计算。
建表

drop table if exists ads_user_retention_day_count;
create external table ads_user_retention_day_count 
(
   `create_date`       string  comment '设备新增日期',
   `retention_day`     int comment '截止当前日期留存天数',
   `retention_count`    bigint comment  '留存数量'
)  COMMENT '每日用户留存情况'
row format delimited fields terminated by '\t'
location '/wavehouse/gmall/ads/ads_user_retention_day_count/';

导入数据

INSERT overwrite TABLE ads_user_retention_day_count
SELECT 
create_date, 
retention_day, 
count(*) retention_count
FROM 
dws_user_retention_day
WHERE dt='2022-12-20'
GROUP BY create_date,retention_day;

留存率计算
建表

drop table if exists ads_user_retention_day_rate;
create external table ads_user_retention_day_rate 
(
     `stat_date`          string comment '统计日期',
     `create_date`       string  comment '设备新增日期',
     `retention_day`     int comment '截止当前日期留存天数',
     `retention_count`    bigint comment  '留存数量',
     `new_mid_count`     bigint comment '当日设备新增数量',
     `retention_ratio`   decimal(10,2) comment '留存率'
)  COMMENT '每日用户留存情况'
row format delimited fields terminated by '\t'
location '/wavehouse/gmall/ads/ads_user_retention_day_rate/';

插入数据

INSERT INTO TABLE ads_user_retention_day_rate
SELECT 
'2022-12-20' stat_date,
nm.create_date,
ur.retention_day, 
ur.retention_count, 
nm.new_mid_count, 
CAST(ur.retention_count / nm.new_mid_count as decimal(10,2)) retention_ratio
FROM 
ads_new_mid_count nm JOIN ads_user_retention_day_count ur on nm.create_date=ur.create_date
-- 为了求出留存率大于1的情况，需要进去筛选和过率
where date_add(ur.create_date,ur.retention_day)='2022-12-20';

7.4 需求4

统计沉默用户，创建ads数据应用层表。

drop table if exists ads_silent_count;
create external table ads_silent_count( 
    `dt` string COMMENT '统计日期',
    `silent_count` bigint COMMENT '沉默设备数'
) 
row format delimited fields terminated by '\t'
location '/wavehouse/gmall/ads/ads_silent_count';

插入数据

INSERT into TABLE ads_silent_count
SELECT
'2022-12-19' dt,
COUNT(*) silent_count
FROM
	(SELECT 
	mid_id
	FROM 
	dws_uv_detail_day
	where dt <= '2022-12-19'
	GROUP BY mid_id
	HAVING COUNT(mid_id)=1 AND MIN(dt) < date_sub('2022-12-19',7)) tmp;

7.5 需求5

建表

drop table if exists ads_back_count;
create external table ads_back_count( 
    `dt` string COMMENT '统计日期',
    `wk_dt` string COMMENT '统计日期所在周',
    `wastage_count` bigint COMMENT '回流设备数'
) 
row format delimited fields terminated by '\t'
location '/wavehouse/gmall/ads/ads_back_count';

导入数据

INSERT INTO TABLE ads_back_count
SELECT 
'2022-12-20' dt,
concat(date_add(next_day('2022-12-20','mo'),-7),'-',date_add(next_day('2022-12-20','mo'),-1)) wk_dt,
COUNT(*) wastage_count
FROM
(	
	SELECT t1.mid_id
	FROM
	(	SELECT mid_id
		FROM dws_uv_detail_wk
		WHERE wk_dt=concat(date_add(next_day('2022-12-20','mo'),-7),'-',date_add(next_day('2022-12-20','mo'),-1))
	) t1
	LEFT JOIN
	(	SELECT mid_id
		FROM dws_new_mid_day
		WHERE create_date <=date_add('2022-12-20',-1) AND create_date >=date_add('2022-12-20',-7)
	) t2
	ON t1.mid_id=t2.mid_id
	LEFT JOIN
	(	SELECT mid_id
		FROM dws_uv_detail_wk
		WHERE wk_dt=concat(date_add(next_day('2022-12-20','mo'),-7*2),'-',date_add(next_day('2022-12-20','mo'),-1-7))
	) t3
	ON t1.mid_id=t3.mid_id
	WHERE t2.mid_id is null and t3.mid_id is null
) t4;

7.6 需求6

drop table if exists ads_continuity_wk_count;
create external table ads_continuity_wk_count( 
    `dt` string COMMENT '统计日期,一般用结束周周日日期,如果每天计算一次,可用当天日期',
    `wk_dt` string COMMENT '持续时间',
    `continuity_count` bigint
) 
row format delimited fields terminated by '\t'
location '/wavehouse/gmall/ads/ads_continuity_wk_count';

插入数据逻辑

INSERT INTO TABLE ads_continuity_wk_count
SELECT 
'2022-12-20' dt, 
concat(date_sub(next_day('2022-12-20','MO'),7*3),'-',date_sub(next_day('2022-12-20','MO'),1)) wk_dt, 
count(*) continuity_count
FROM 
dws_uv_detail_wk
WHERE wk_dt BETWEEN concat(date_sub(next_day('2022-12-20','MO'),7*3),'-',date_sub(next_day('2022-12-20','MO'),7*2+1)) AND
concat(date_sub(next_day('2022-12-20','MO'),7),'-',date_sub(next_day('2022-12-20','MO'),1))
GROUP BY mid_id
HAVING COUNT(mid_id)>=3;

7.7 需求7

建表

drop table if exists ads_continuity_uv_count;
create external table ads_continuity_uv_count( 
    `dt` string COMMENT '统计日期',
    `wk_dt` string COMMENT '最近7天日期',
    `continuity_count` bigint
) COMMENT '连续活跃设备数'
row format delimited fields terminated by '\t'
location '/wavehouse/gmall/ads/ads_continuity_uv_count';

插入数据

SELECT
'2022-12-20' dt,
concat(date_sub(next_day('2022-12-20','MO'),7),'-',date_sub(next_day('2022-12-20','MO'),1)) wk_dt,
count(DISTINCT mid_id) continuity_count
FROM
(	SELECT
	mid_id
	FROM
(		SELECT
		mid_id,
		dt,
		ROW_NUMBER() OVER(PARTITION BY mid_id ORDER BY dt asc) rn,
		date_sub(dt,ROW_NUMBER() OVER(PARTITION BY mid_id ORDER BY dt asc)) diff_dt
		FROM
		dws_uv_detail_day
		WHERE dt BETWEEN date_sub(next_day('2022-12-20','MO'),7) AND date_sub(next_day('2022-12-20','MO'),1)
) t1 
	GROUP BY mid_id,diff_dt
	HAVING COUNT(mid_id)>=3
) t2

7.8 需求8

ads层建表

drop table if exists ads_new_favorites_mid_day;
create external table ads_new_favorites_mid_day( 
    `dt` string COMMENT '日期',
    `favorites_users` bigint COMMENT '新收藏用户数'
) 
row format delimited fields terminated by '\t'
location '/wavehouse/gmall/ads/ads_new_favorites_mid_day';

插入数据

INSERT INTO TABLE ads_new_favorites_mid_day
SELECT
'2022-12-20' dt,
COUNT(*) 
FROM
( SELECT 
mid_id
FROM 
dws_user_action_wide_log
WHERE favorite_count > 0
GROUP BY mid_id
HAVING MIN(dt)='2022-12-20'
) t1

7.9 需求9

建表

drop table if exists ads_goods_count;
create external table ads_goods_count( 
    `dt` string COMMENT '统计日期',
    `goodsid` string COMMENT '商品',
    `user_id` string COMMENT '用户',
    `goodsid_user_count` bigint COMMENT '商品用户点击次数'
) 
row format delimited fields terminated by '\t'
location '/wavehouse/gmall/ads/ads_goods_count';

导入数据

SELECT
'2022-12-20' dt,
goodsid,
mid_id,
total_count
FROM
(
SELECT
mid_id,
goodsid,
total_count,
RANK() OVER(PARTITION BY goodsid ORDER BY total_count DESC) rn
FROM
(
SELECT 
mid_id,
goodsid,
sum(display_count) total_count
FROM
dws_user_action_wide_log
WHERE dt<='2022-12-20' AND display_count > 0
GROUP BY mid_id,goodsid
) t1) t2
WHERE rn<=3;

接下来是本地业务数仓的详细搭建流程，详见《本地数仓项目(二)——搭建系统业务数仓详细流程》