拉链表流程
1.从ODS层获取增量数据(上一天新增和更新的数据) 2.拿着DWD原始拉链表数据 left join 增量数据 ,修改原始拉链中历史数据的结束时间 3.拿着left join 的结果集 union all 增量数据 4.把最新的拉链数据优先保存到DWD对应的临时表中 5.使用insert+select 方式把临时表中数据灌入DWD拉链表中
拉链表实现流程:
DWD层开发
DWD层: 数仓明细层(清洗转换、降维操作) 此层核心目标: 基于数据探查情况, 对相关表数据进行合并
会员基础信息表:
-
需要对此表进行历史数据拉链
说明: 将每日会员基础信息表中新增及更新的数据, 通过拉链的方式记录下来
建表操作:
CREATE TABLE IF NOT EXISTS dwd.dwd_mem_member_union_i(
zt_id BIGINT COMMENT '中台会员ID',
member_id BIGINT COMMENT '会员ID',
user_id BIGINT COMMENT '用户ID',
card_no STRING COMMENT '卡号',
member_name STRING COMMENT '会员名称',
mobile STRING COMMENT '手机号',
user_email STRING COMMENT '邮箱',
sex BIGINT COMMENT '用户的性别，1男性，2女性，0未知',
birthday_date STRING COMMENT '生日',
address STRING COMMENT '地址',
reg_time TIMESTAMP COMMENT '注册时间',
reg_md STRING COMMENT '注册门店',
bind_md STRING COMMENT '绑定门店',
flag BIGINT COMMENT '0正常,1删除',
is_black BIGINT COMMENT '是否被拉黑 1被拉黑,0正常用户',
user_state BIGINT COMMENT '会员状态，0停用/注销,1正常,2冻结',
user_type STRING COMMENT '用户类型（-1:传智鲜用户;0:普通用户;1:企业用户 2:内部员工 3:黑马门店 4:商铺会员 5:大买家 6:中间商 7:军区员工）',
member_type BIGINT COMMENT '会员状态 10：未付费会员 20：付费会员',
member_status BIGINT COMMENT '付费会员状态 -1:未付费会员 1：正常 2：试用 3：过期 4:试用已过期',
expired_time TIMESTAMP COMMENT '过期时间',
user_source BIGINT COMMENT '用户来源 ',
member_level BIGINT COMMENT '会员等级',
growth BIGINT COMMENT '成长值',
invite_member_id BIGINT COMMENT '邀请人标识',
invite_type BIGINT COMMENT '邀请类型，0为内部',
register_store_leader_id BIGINT COMMENT '注册归属团长 ID',
last_update_time TIMESTAMP COMMENT '更新日期',
end_date STRING COMMENT '生效结束日期'
)
comment '会员基础信息表'
partitioned by (start_date STRING COMMENT '生效开始日期')
row format delimited fields terminated by ','
stored as orc
tblproperties ('orc.compress'='SNAPPY');需要对此表进行历史数据拉链
说明: 将每日会员基础信息表中新增及更新的数据, 通过拉链的方式记录下来
首次导入
-- 会员基础信息表首次导入:
insert overwrite table dwd.dwd_mem_member_union_i partition (start_date)
select
zt_id,
member_id,
user_id,
card_no,
member_name,
mobile,
user_email,
sex,
birthday_date,
address,
reg_time,
reg_md,
bind_md,
flag,
is_black,
user_state,
user_type,
member_type,
member_status,
expired_time,
user_source,
member_level,
growth,
invite_member_id,
invite_type,
register_store_leader_id,
last_update_time,
'9999-99-99' as end_date,
'2023-11-29' as start_date
from ods.ods_mem_member_union_i;
后续导入
1.mysql业务数据变化
注意: 实际开发没有自己准备业务数据这一步!!!
-- 步骤一: 在MySQL中, 添加增量的测试数据, 包含 新增 和 更新的数据 (测试)
-- 模拟新增数据
insert into member.member_union (zt_id, member_id, user_id, card_no, member_name, mobile, user_email, sex, birthday_date, address,reg_time, reg_md, bind_md, flag, is_black, user_state, user_type, member_type, member_status, expired_time, user_source, member_level, growth, invite_member_id, invite_type, register_store_leader_id,last_update_time)
values ('32015926',2160344,NULL,'','32015925',114,163,0,'','不详','2023-11-30 17:09:28','W121','W121',0,0,1,-1,10,-1,NULL,-1,0,0,NULL,NULL,NULL,'2023-11-30 17:09:28');
-- 模拟更新数据
UPDATE member.member_union SET SEX = 1, last_update_time = '2023-11-30 17:10:20' WHERE zt_id = '32015925';
-- 验证数据
select *
from member.member_union
where date_format(reg_time,'%Y-%m-%d') = date_format(date_sub(Now(),INTERVAL 1 DAY),'%Y-%m-%d')
OR date_format(last_update_time,'%Y-%m-%d') = date_format(date_sub(Now(),INTERVAL 1 DAY),'%Y-%m-%d')2.datax-web导数据到ods
-- 步骤二: 执行DataX, 将新增数据和增量数据导入到ODS层 (应该在数据采集中执行)
-- 说明: 此步骤详细过程参考day02实施
-- 注意: mysqlreader中记得补充条件
date_format(reg_time,'%Y-%m-%d') = date_format(date_sub(Now(),INTERVAL 1 DAY),'%Y-%m-%d')
OR date_format(last_update_time,'%Y-%m-%d') = date_format(date_sub(Now(),INTERVAL 1 DAY),'%Y-%m-%d')
-- 注意: hdfswriter中记得补一个后置sql语句,内容如下
"postSql":[
"msck repair table ods.ods_mem_member_union_i"
],3.hive增量导数据到dwd
-- 步骤三: 执行增量数据导入
-- 先创建一张目标表的临时表, 用于放置计算后的结果
CREATE TABLE IF NOT EXISTS dwd.dwd_mem_member_union_i_temp(
zt_id BIGINT COMMENT '中台会员ID',
member_id BIGINT COMMENT '会员ID',
user_id BIGINT COMMENT '用户ID',
card_no STRING COMMENT '卡号',
member_name STRING COMMENT '会员名称',
mobile STRING COMMENT '手机号',
user_email STRING COMMENT '邮箱',
sex BIGINT COMMENT '用户的性别，1男性，2女性，0未知',
birthday_date STRING COMMENT '生日',
address STRING COMMENT '地址',
reg_time TIMESTAMP COMMENT '注册时间',
reg_md STRING COMMENT '注册门店',
bind_md STRING COMMENT '绑定门店',
flag BIGINT COMMENT '0正常,1删除',
is_black BIGINT COMMENT '是否被拉黑 1被拉黑,0正常用户',
user_state BIGINT COMMENT '会员状态，0停用/注销,1正常,2冻结',
user_type STRING COMMENT '用户类型（-1:传智鲜用户;0:普通用户;1:企业用户 2:内部员工 3:黑马门店 4:商铺会员 5:大买家 6:中间商 7:军区员工）',
member_type BIGINT COMMENT '会员状态 10：未付费会员 20：付费会员',
member_status BIGINT COMMENT '付费会员状态 -1:未付费会员 1：正常 2：试用 3：过期 4:试用已过期',
expired_time TIMESTAMP COMMENT '过期时间',
user_source BIGINT COMMENT '用户来源 ',
member_level BIGINT COMMENT '会员等级',
growth BIGINT COMMENT '成长值',
invite_member_id BIGINT COMMENT '邀请人标识',
invite_type BIGINT COMMENT '邀请类型，0为内部',
register_store_leader_id BIGINT COMMENT '注册归属团长 ID',
last_update_time TIMESTAMP COMMENT '更新日期',
end_date STRING COMMENT '生效结束日期'
)
comment '会员基础信息表'
partitioned by (start_date STRING COMMENT '生效开始日期')
row format delimited fields terminated by ','
stored as orc
tblproperties ('orc.compress'='SNAPPY');
-- 循环导入数据
with t2 as (
select
t1.zt_id,
t1.member_id,
t1.user_id,
t1.card_no,
t1.member_name,
t1.mobile,
t1.user_email,
t1.sex,
t1.birthday_date,
t1.address,
t1.reg_time,
t1.reg_md,
t1.bind_md,
t1.flag,
t1.is_black,
t1.user_state,
t1.user_type,
t1.member_type,
t1.member_status,
t1.expired_time,
t1.user_source,
t1.member_level,
t1.growth,
t1.invite_member_id,
t1.invite_type,
t1.register_store_leader_id,
t1.last_update_time,
if(
t2.zt_id is null OR t1.end_date != '9999-99-99',
t1.end_date,
t2.dt
) as end_date,
t1.start_date
from dwd.dwd_mem_member_union_i t1
left join (select * from ods.ods_mem_member_union_i
where dt = date_format(date_sub(current_date(),1),'yyyy-MM-dd')
) as t2 on t1.zt_id = t2.zt_id
union all
select
zt_id,
member_id,
user_id,
card_no,
member_name,
mobile,
user_email,
sex,
birthday_date,
address,
reg_time,
reg_md,
bind_md,
flag,
is_black,
user_state,
user_type,
member_type,
member_status,
expired_time,
user_source,
member_level,
growth,
invite_member_id,
invite_type,
register_store_leader_id,
last_update_time,
'9999-99-99' as end_date,
date_format(date_sub(current_date(),1),'yyyy-MM-dd') as start_date
from ods.ods_mem_member_union_i
where dt = date_format(date_sub(current_date(),1),'yyyy-MM-dd')
)
insert overwrite table dwd.dwd_mem_member_union_i_temp partition (start_date)
select
*
from t2 ;
-- 将临时表数据覆盖回目标表中
insert overwrite table dwd.dwd_mem_member_union_i partition (start_date)
select * from dwd.dwd_mem_member_union_i_temp;
-- 将临时表删除
drop table dwd.dwd_mem_member_union_i_temp;情况说明:
1- 目前所做的拉链表是针对历史所有数据, 哪怕这个数据是五年前创建后, 然后五年后发生修改, 我们依然会进行维护
2- 目前所有的拉链表是针对表中所有的字段, 只要表中任何字段发生变更, 都会进行维护
但是:
在实际开发中,我们一般不需要维护历史所有数据, 也不需要维护表中所有的字段
一般维护最近一段周期的数据(一个月、一个季度、一年(最常用))
一般维护的核心与后续指标计算相关的字段: 用哪些一般维护哪些
会员积分变动表:
-
用于记录每个会员每天积分的变动情况(需统计)
建表操作:
因为占用主体ID，分为两部分，一部分occupy_subject_id 为0，即全部，另一部分是各种主体，所以这里计算时，分为两部分计算，然后将结果进行合并。
CREATE TABLE IF NOT EXISTS dwd.dwd_mem_member_point_change_i(
trade_date STRING COMMENT '快照时间',
zt_id BIGINT COMMENT '中台ID',
occupy_subject_id BIGINT COMMENT '占用主体ID，0为全部，101优选,102传智鲜,103传智商城',
point_add BIGINT COMMENT '增加积分，没有则为0',
point_reduce BIGINT COMMENT '减少积分，没有则为0',
point_change BIGINT COMMENT '变动积分，没有则为0'
)
comment '会员积分变动表'
partitioned by (dt STRING COMMENT '统计日期')
row format delimited fields terminated by ','
stored as orc
tblproperties ('orc.compress'='SNAPPY');数据导入:
-- 会员主题 DWD层开发 会员积分变动表
-- 需求: 统计每天各个会员积分变动情况
-- 注意: 主体分为两部分 , 一部分是全部 一部分为各个主体
insert overwrite table dwd.dwd_mem_member_point_change_i partition(dt)
select
dt as trade_date,
zt_id,
occupy_subject_id,
sum( if( change_type = 1,point_c,0) ) as point_add,
sum( if( change_type = 0,-point_c,0) ) as point_reduce,
sum(if( change_type = 1,point_c,-point_c)) as point_change,
dt
from ods.ods_mem_user_point_log_detailed_i
group by
dt,
zt_id,
occupy_subject_id
union all
select
dt as trade_date,
zt_id,
0 as occupy_subject_id,
sum( if( change_type = 1,point_c,0) ) as point_add,
sum( if( change_type = 0,-point_c,0) ) as point_reduce,
sum(if( change_type = 1,point_c,-point_c)) as point_change,
dt
from ods.ods_mem_user_point_log_detailed_i
group by
dt,
zt_id;线上会员每日余额变动表:
用于记录每个会员每天余额的变动情况(需统计)
建表操作:
CREATE TABLE IF NOT EXISTS dwd.dwd_mem_balance_change_i(
trade_date STRING COMMENT '统计日期',
zt_id BIGINT COMMENT '中台ID',
member_id BIGINT COMMENT '会员ID',
record_type BIGINT COMMENT '记录类型，0全部,1消费,2充值,3退款,4.清退余额,5.转化,6.系统清除,7.礼品卡兑换,8.现付结余,9.结余退款,10.退卡',
times BIGINT COMMENT '次数',
change_amount DECIMAL(27, 2) COMMENT '变动金额'
)
comment '线上会员每日余额变动表'
partitioned by (dt STRING COMMENT '统计日期')
row format delimited fields terminated by ','
stored as orc
tblproperties ('orc.compress'='SNAPPY');数据导入:
-- DWD 会员余额变动表
-- 需求: 统计每天各个会员余额变动情况
-- 注意: 记录类型也分为二部分 一个是全部 一个是 各个记录类型 union all 将两部分结果进行合并
insert overwrite table dwd.dwd_mem_balance_change_i partition (dt)
select
dt as trade_date,
zt_id,
member_id,
record_type,
count(1) as times,
sum(amount) as change_amount,
dt
from ods.ods_mem_store_amount_record_i
group by
dt,
zt_id,
member_id,
record_type
union all
select
dt as trade_date,
zt_id,
member_id,
0 as record_type,
count(1) as times,
sum(amount) as change_amount,
dt
from ods.ods_mem_store_amount_record_i
group by
dt,
zt_id,
member_id;线上会员每日余额表:
用于记录每个会员每天余额的情况(需统计)
建表操作:
CREATE TABLE IF NOT EXISTS dwd.dwd_mem_balance_online_i(
trade_date STRING COMMENT '统计日期',
zt_id BIGINT COMMENT '中台ID',
member_id BIGINT COMMENT '会员ID',
member_type BIGINT COMMENT '会员类型 1:线下会员 2:线上会员',
member_type_name STRING COMMENT '会员类型名称',
store_no STRING COMMENT '门店编码',
city_id BIGINT COMMENT '城市ID',
balance_amount DECIMAL(27, 2) COMMENT '余额'
)
comment '线上会员每日余额表'
partitioned by (dt STRING COMMENT '统计日期')
row format delimited fields terminated by ','
stored as orc
tblproperties ('orc.compress'='SNAPPY');需求分析:
这个需求是将变动日志进行处理，基于日志明细构建余额快照表。
先取到每天最后一条，
select
date_format(trade_date,'yyyy-MM-dd') as trade_date,
max(id) as id
from ods.ods_mem_store_amount_record_i
group by date_format(trade_date,'yyyy-MM-dd'),member_id 然后使用 lead 函数按用户id进行分组，按日期进行排序，取到下一条对应的日期。
select
trade_date,zt_id,member_id,store_no,city_id,left_store_amount,
lead(trade_date,1,'9999-12-31') over(partition by member_id order by trade_date) as next_date
from
(select a.trade_date,b.zt_id,b.member_id,b.store_no,b.city_id,b.left_store_amount
from
(select date_format(trade_date,'yyyy-MM-dd') as trade_date,
max(id) as id
from ods.ods_mem_store_amount_record_i
group by date_format(trade_date,'yyyy-MM-dd'),member_id ) a
inner join ods.ods_mem_store_amount_record_i b on a.id=b.id
) t然后像使用拉链表一样用'${inputdate}'去卡日期，即可取到当天对应的余额。
注意: 因为这个需求只是把有余额的记录记录到表中，所以需要去除掉 left_store_amount 为0 的情况。在这里把 left_store_amount<>0 写到了where条件中，是先对结果进行了过滤，这样在匹配时就匹配不到对应的数据了，也就是不会把对应的记录插入到表中了。
where trade_date<='${inputdate}' and '${inputdate}'<next_date and left_store_amount<>0
-- 思考: 如何拿到最后一次余额变动数据呢?
-- 尝试先找到每天 每个用户 ID最大值
select
dt,
zt_id,
max(id) as last_id
from ods.ods_mem_store_amount_record_i
group by dt,zt_id;数据导入:
-- DWD 线上会员每日余额表
-- 说明: 此表是用于记录每个会员每天(某一天)会员余额是多少
-- 先找到最大id
-- 基于最大的ID, 找到对应的余额数据
-- 目前写的这条SQL 其实已经拿到了每天每个用户的余额,但是这个数据是来源于用户余额变动表, 如果用户在某一天没有变化, 在这一天就不会有这个用户余额
with t1 as (
select
dt as trade_date,
zt_id,
max(id) as last_id
from ods.ods_mem_store_amount_record_i
group by dt,zt_id
),
t2 as(
select
t1.trade_date as start_date,
t1.zt_id,
t2.member_id,
t2.store_no,
t2.city_id,
t2.left_store_amount,
lead(t1.trade_date,1,'9999-99-99') over(partition by t1.zt_id order by t1.trade_date) as end_date
from t1 inner join ods.ods_mem_store_amount_record_i t2 on t1.last_id = t2.id
)
insert overwrite table dwd.dwd_mem_balance_online_i partition (dt)
select
start_date as trade_date,
zt_id,
member_id,
2 as member_type,
'线上会员' as member_type_name,
store_no,
city_id,
left_store_amount as balance_amount,
'2023-12-01' as dt
from t2 where start_date <= '2023-12-01' and end_date > '2023-12-01' and left_store_amount <> 0;DWM层开发
各类会员数量统计: 指标：新增注册会员数、累计注册会员数、新增消费会员数、累计消费会员数、新增复购会员数、累计复购会员数、活跃会员数、沉睡会员数、会员消费金额 维度: 时间维度（天、周、月） 门店会员分析: 指标: 门店销售额、门店总订单量、当日注册人数、累计注册会员数、当日注册且充值会员数、当日注册且充值且消费会员数、当日注册且消费会员数、充值会员数、充值金额、累计会员充值金额、当日有余额的会员人数、当日会员余额、余额消费人数/单量、余额支付金额、余额消费金额、会员消费人数/单量、会员消费金额、会员首单人数/订单量/销售额、会员非首单人数/订单量/销售额 维度: 时间维度（天、周、月） 说明: 由于各类会员数据统计分析和门店会员分析中, 有大量的指标存在一定的依赖关系, 所以在此处我们合并在一起进行分析, 向上抽取出一些公共的DWM层的数据表, 便于后续两个DWS层表数据的聚合统计, 本次主要涉及有四张DWM层表:会员销售订单表、会员首次消费表、会员第二次消费表、会员行为天表
会员销售订单表
建表操作:
CREATE TABLE IF NOT EXISTS dwm.dwm_mem_sell_order_i(
create_time STRING COMMENT '订单创建时间',
trade_date STRING COMMENT '交易日期',
week_trade_date STRING COMMENT '周一日期',
month_trade_date STRING COMMENT '月一日期',
hourly BIGINT COMMENT '交易小时(0-23)',
quarter BIGINT COMMENT '刻钟:1.0-15,2.15-30,3.30-45,4.45-60',
quarters BIGINT COMMENT '刻钟数:hourly*4+quarters',
parent_order_no STRING COMMENT '父单订单号/源单号',
order_no STRING COMMENT '订单编号',
trade_type BIGINT COMMENT '结算类型(0.正常交易,1.赠品发放,2.退货,4.培训,5.取消交易)',
source_type BIGINT COMMENT '交易来源1:线下POS;2:三方平台;3:传智鲜商城;4:黑马优选团;5:传智大客户;6:传智其他;7:黑马优选;8:优选海淘;9:优选大客户;10:优选POS;11:优选APP;12:优选H5;13:店长工具线下;14:店长工具线上;15:黑马其他',
source_type_name STRING COMMENT '交易来源名称',
sale_type BIGINT COMMENT '销售类型 1.实物,2.代客,3.优选小程序,4.离店,5.传智鲜小程序,6.第三方平台,7.其他,8.大客户',
is_online_order BIGINT COMMENT '是否为线上单:0否,1是',
member_type BIGINT COMMENT '会员类型:0非会员,1线上会员,2实体卡会员',
is_balance_consume BIGINT COMMENT '是否有余额支付:0否,1是',
order_type BIGINT COMMENT '配送类型（真正的订单类型由业务类型来决定）：1-及时送；2-隔日送；3-自提单；4-线下单',
express_type BIGINT COMMENT '配送方式：0-三方平台配送；1-自配送；2-快递；3-自提；4-线下',
store_no STRING COMMENT '店铺编码',
store_name STRING COMMENT '店铺名称',
store_sale_type BIGINT COMMENT '店铺销售类型',
store_type_code BIGINT COMMENT '分店类型',
worker_num BIGINT COMMENT '员工人数',
store_area DECIMAL(27, 2) COMMENT '门店面积',
city_id BIGINT COMMENT '城市ID',
city_name STRING COMMENT '城市名称',
region_code STRING COMMENT '区域编码',
region_name STRING COMMENT '区域名称',
is_day_clear BIGINT COMMENT '是否日清:0否,1是',
is_cancel BIGINT COMMENT '是否取消',
cancel_time STRING COMMENT '取消时间',
cancel_reason STRING COMMENT '取消原因',
last_update_time TIMESTAMP COMMENT '最新更新时间',
cashier_no STRING COMMENT '收银员编码',
cashier_name STRING COMMENT '收银员名称',
zt_id BIGINT COMMENT '中台ID',
member_id BIGINT COMMENT '会员ID',
card_no STRING COMMENT '卡号',
r_name STRING COMMENT '收货人姓名',
r_province STRING COMMENT '收货人省份',
r_city STRING COMMENT '收货人城市',
r_district STRING COMMENT '收货人区域',
is_tuan_head BIGINT COMMENT '是否为团长订单',
store_leader_id BIGINT COMMENT '团长id',
order_group_no STRING COMMENT '团单号',
settle_amount DECIMAL(27, 2) COMMENT '结算金额',
share_user_id BIGINT COMMENT '分享人用户ID',
commission_amount DECIMAL(27, 2) COMMENT '佣金',
order_total_amount DECIMAL(27, 2) COMMENT '订单总金额',
product_total_amount DECIMAL(27, 2) COMMENT '商品总金额（原价）',
pack_amount DECIMAL(27, 2) COMMENT '餐盒费/打包费',
delivery_amount DECIMAL(27, 2) COMMENT '配送费',
discount_amount DECIMAL(27, 2) COMMENT '订单优惠金额=商家承担优惠金额+平台补贴金额',
seller_discount_amount DECIMAL(27, 2) COMMENT '商家承担优惠金额',
platform_allowance_amount DECIMAL(27, 2) COMMENT '平台补贴金额',
real_paid_amount DECIMAL(27, 2) COMMENT '实付金额',
product_discount DECIMAL(27, 2) COMMENT '商品优惠金额',
real_product_amount DECIMAL(27, 2) COMMENT '商品实际金额',
round_amount DECIMAL(27, 2) COMMENT '舍分金额',
wechat_amount DECIMAL(27, 4) COMMENT '微信支付',
ali_pay_amount DECIMAL(27, 4) COMMENT '支付宝支付',
cash_amount DECIMAL(27, 4) COMMENT '现金支付',
balance_amount DECIMAL(27, 4) COMMENT '余额支付',
point_amount DECIMAL(27, 4) COMMENT '积分支付',
unionpay_amount DECIMAL(27, 4) COMMENT '银行支付',
member_card_amount DECIMAL(27, 4) COMMENT '线下实体卡支付',
gift_amount DECIMAL(27, 4) COMMENT '礼品卡支付',
czapi_amount DECIMAL(27, 4) COMMENT '传智支付',
other_pay_amount DECIMAL(27, 4) COMMENT '其他支付'
)
comment '会员销售订单表'
partitioned by (dt STRING COMMENT '销售日期')
row format delimited fields terminated by ','
stored as orc
tblproperties ('orc.compress'='SNAPPY');
思路分析:
构建这张表的主要原因在于，后续的会员数据分析可以基于这张会员销售明细表来进行计算，相比全量的销售明细表，可以极大地减少数据量。
从dwm_sell_o2o_order_i表中获取，只取member_type 为1的。
准备数据:
在实际工作中, 部分表需要依赖于其他开发人员, 当天dwm_sell_o2o_order_i是属于售卖主题中的相关表
需要执行脚本目录中售卖主题准备工作脚本的<<售卖主题dwm_sell_o2o_order_i表>>
数据导入:
-- DWM层: 会员销售订单明细表
insert overwrite table dwm.dwm_mem_sell_order_i partition (dt)
select
create_time,
trade_date,
week_trade_date,
month_trade_date,
hourly,
quarter,
quarters,
parent_order_no,
order_no,
trade_type,
source_type,
source_type_name,
sale_type,
is_online_order,
member_type,
is_balance_consume,
order_type,
express_type,
store_no,
store_name,
store_sale_type,
store_type_code,
worker_num,
store_area,
city_id,
city_name,
region_code,
region_name,
is_day_clear,
is_cancel,
cancel_time,
cancel_reason,
last_update_time,
cashier_no,
cashier_name,
zt_id,
member_id,
card_no,
r_name,
r_province,
r_city,
r_district,
is_tuan_head,
store_leader_id,
order_group_no,
settle_amount,
share_user_id,
commission_amount,
order_total_amount,
product_total_amount,
pack_amount,
delivery_amount,
discount_amount,
seller_discount_amount,
platform_allowance_amount,
real_paid_amount,
product_discount,
real_product_amount,
round_amount,
wechat_amount,
ali_pay_amount,
cash_amount,
balance_amount,
point_amount,
unionpay_amount,
member_card_amount,
gift_amount,
czapi_amount,
other_pay_amount,
dt
from dwm.dwm_sell_o2o_order_i where member_type = 1;会员首次消费表:
建表操作:
CREATE TABLE IF NOT EXISTS dwm.dwm_mem_first_buy_i(
zt_id BIGINT COMMENT '中台 会员id',
trade_date_time STRING COMMENT '首次消费时间',
trade_date STRING COMMENT '首次消费日期',
week_trade_date STRING COMMENT '周一日期',
month_trade_date STRING COMMENT '月一日期',
store_no STRING COMMENT '消费门店',
sale_amount DECIMAL(27, 2) COMMENT '消费金额',
order_no STRING COMMENT '订单编号',
source_type BIGINT COMMENT '交易来源'
)
comment '会员首次消费表'
partitioned by (dt STRING COMMENT '消费日期')
row format delimited fields terminated by ','
stored as orc
tblproperties ('orc.compress'='SNAPPY');思路分析:
注意：这里不能直接对全量数据使用over窗口，然后取 row_number() 为1的数据，因为这样会极大的消耗没有用的IO资源。
思路：通过思考发现，每日新增首次消费会员一定是当天消费中的首次会员，并且不在历史首次消费的会员中。基于这个特性，可以先算中当天消费中的首次会员，然后再和 dwm_mem_first_buy_i 关联，使用左关联，取出 dwm_mem_first_buy_i 中没有的，即关联不上的，则是首次消费会员，然后存到对应分区即可。
数据导入:
-- DWM 会员首次消费表 dwm_mem_first_buy_i
-- 第一步: 计算出当天首次消费的用户 (此用户并不代表历史首次消费)
with t1 as (
select
zt_id,
create_time as trade_date_time,
trade_date,
week_trade_date,
month_trade_date,
store_no,
real_paid_amount as sale_amount,
order_no,
source_type,
row_number() over (partition by zt_id order by create_time) as rn
from dwm.dwm_mem_sell_order_i where dt = '2023-11-14' and zt_id is not null
),
t2 as(
select
zt_id,
trade_date_time,
trade_date,
week_trade_date,
month_trade_date,
store_no,
sale_amount,
order_no,
source_type
from t1 where rn = 1
)
-- 第二步: 用第一步的结果 和 截止当天之前的历史首次消费表进行关联 (left Join)
insert overwrite table dwm.dwm_mem_first_buy_i partition (dt)
select
t2.zt_id,
t2.trade_date_time,
t2.trade_date,
t2.week_trade_date,
t2.month_trade_date,
t2.store_no,
t2.sale_amount,
t2.order_no,
t2.source_type,
'2023-11-14' as dt
from t2 left join dwm.dwm_mem_first_buy_i t3 on t2.zt_id = t3.zt_id and t3.dt < '2023-11-14'
-- 第三步: 判断: 如果 没有关联上, 说明在历史首次消费中并未发现有消费, 我们就认为当天的消费就是历史首次
where t3.zt_id is null;
-- 注意:运行完后可以依次修改时间把dwm_mem_sell_order_i所有分区数据都导入
-- 说明: 在实施中 大家需要调整日期, 依次将14~20号的数据跑出来即可会员第二次消费表:
建表操作:
CREATE TABLE IF NOT EXISTS dwm.dwm_mem_second_buy_i(
zt_id BIGINT COMMENT '中台 会员id',
trade_date_time STRING COMMENT '第二次消费时间',
trade_date STRING COMMENT '第二次消费日期',
week_trade_date STRING COMMENT '周一日期',
month_trade_date STRING COMMENT '月一日期',
store_no STRING COMMENT '消费门店',
sale_amount DECIMAL(27, 2) COMMENT '消费金额',
order_no STRING COMMENT '订单编号',
source_type BIGINT COMMENT '交易来源'
)
comment '会员第二次消费表'
partitioned by (dt STRING COMMENT '消费日期')
row format delimited fields terminated by ','
stored as orc
tblproperties ('orc.compress'='SNAPPY');思路分析:
注意：同dwm_mem_first_buy_i不能直接对全量数据使用over窗口，然后取 row_number() 为2的数据。
思路：这种用户分为两种，一种是历史上有过首次购买的但没有二次购买的，这种用户如果当日有首次购买，则为第二次购买。另一种是历史上从没有购买过，这种用户如果当天首次购买，并且发生第二次购买则是第二次购买。
第一种会员：先求出有过首次购买但没有二次购买的会员
select f.zt_id
from dwm.dwm_mem_first_buy_i f
left join dwm.dwm_mem_second_buy_i s on f.zt_id=s.zt_id and s.dt < '${inputdate}'
where f.dt < '${inputdate}' and s.zt_id is null然后和当天首次消费的会员进行关联。
第二种：先求出当天购买两次的会员。
select
*
from
(select
zt_id,
create_time as trade_date_time,
trade_date,
week_trade_date,
month_trade_date,
store_no,
real_paid_amount as sale_amount,
order_no,
source_type,
row_number() over(partition by zt_id order by create_time) as rn
from dwm.dwm_mem_sell_order_i
where dt = '${inputdate}' ) t
where t.rn=2然后和当天是首单的会员进行关联：
inner join dwm.dwm_mem_first_buy_i tmp
on t.zt_id=tmp.zt_id and tmp.dt = '${inputdate}'数据导入:
-- DWM层: 会员二次消费表
--思路: 统计日期: 2023-11-14
-- 情况一: 历史上有过首次购买, 但没有二次购买 和 今日的首次购买用户进行 关联 得出二次购买用户
with t3 as (
-- 步骤一: 得到史上有过首次购买, 但没有二次购买用户有哪些
select
t1.zt_id,
t1.trade_date_time,
t1.trade_date,
t1.week_trade_date,
t1.month_trade_date,
t1.store_no,
t1.sale_amount,
t1.order_no,
t1.source_type
from (select * from dwm.dwm_mem_first_buy_i where dt < '2023-11-14') t1 -- 历史所有的首次购买
left join dwm.dwm_mem_second_buy_i t2 on t1.zt_id = t2.zt_id and t2.dt < '2023-11-14' -- 历史所有的二次购买用户
where t2.zt_id is null -- 判断 如果关联不上, 那就表示有过历史首次购买, 但没有二次购买用户
),
-- 步骤二: 基于这个结果 和 今日首次购买的用户进行关联
t5 as ( -- 历史上有过首次购买 但没有二次购买和当日首次购买用户结果 (情况一结果表)
select
t3.zt_id,
t3.trade_date_time,
t3.trade_date,
t3.week_trade_date,
t3.month_trade_date,
t3.store_no,
t3.sale_amount,
t3.order_no,
t3.source_type
from t3
inner join
(
select
zt_id,
create_time as trade_date_time,
trade_date,
week_trade_date,
month_trade_date,
store_no,
real_paid_amount as sale_amount,
order_no,
source_type,
row_number() over (partition by zt_id order by create_time) as rn
from dwm.dwm_mem_sell_order_i where dt = '2023-11-14'
) t4
on t3.zt_id = t4.zt_id and t4.rn = 1
),
-- 情况二: 历史上从没有购买过, 但是当天发生了多次购买, 获取其中第二次购买即可
t6 as (
select
zt_id,
create_time as trade_date_time,
trade_date,
week_trade_date,
month_trade_date,
store_no,
real_paid_amount as sale_amount,
order_no,
source_type,
row_number() over (partition by zt_id order by create_time) as rn
from dwm.dwm_mem_sell_order_i where dt = '2023-11-14'
),
t7 as (
select
zt_id,
trade_date_time,
trade_date,
week_trade_date,
month_trade_date,
store_no,
sale_amount,
order_no,
source_type
from t6 where rn = 2
),
-- 历史上没有购买过, 但是当天发生了二次购买的用户(情况二结果表)
t9 as (
select
t7.zt_id,
t7.trade_date_time,
t7.trade_date,
t7.week_trade_date,
t7.month_trade_date,
t7.store_no,
t7.sale_amount,
t7.order_no,
t7.source_type
from t7 inner join dwm.dwm_mem_first_buy_i t8 on t7.zt_id = t8.zt_id and t8.dt = '2023-11-14'
)
insert overwrite table dwm.dwm_mem_second_buy_i partition (dt)
select
zt_id,
trade_date_time,
trade_date,
week_trade_date,
month_trade_date,
store_no,
sale_amount,
order_no,
source_type,
'2023-11-14' as dt
from t5 where zt_id is not null
union all
select
zt_id,
trade_date_time,
trade_date,
week_trade_date,
month_trade_date,
store_no,
sale_amount,
order_no,
source_type,
'2023-11-14' as dt
from t9 where zt_id is not null
-- 说明: 在实施中 大家需要调整日期, 依次将14~20号的数据跑出来即可
/* 在生产环境中(工作中), 我们可以通过海豚调度器提供的补数方案, 指定需要补数的范围, 调度器会自动将过去的几天数据全部补回来(无需执行, 了解即可, 面试中按照这个说即可)*/会员行为天表
建表操作:
CREATE TABLE IF NOT EXISTS dwm.dwm_mem_member_behavior_day_i(
trade_date STRING COMMENT '时间',
week_trade_date STRING COMMENT '周一日期',
month_trade_date STRING COMMENT '月一日期',
zt_id BIGINT COMMENT '中台 会员id',
bind_md STRING COMMENT '归属门店(绑定门店)',
reg_md STRING COMMENT '注册门店',
reg_time TIMESTAMP COMMENT '中台 注册时间',
is_register BIGINT COMMENT '当日是否注册',
is_recharge BIGINT COMMENT '当日是否充值',
recharge_times BIGINT COMMENT '充值次数，没有充值则为0',
recharge_amount DECIMAL(27, 2) COMMENT '充值金额，没有充值则为0',
is_consume BIGINT COMMENT '当日是否消费',
consume_times BIGINT COMMENT '消费次数，没有消费则为0',
consume_amount DECIMAL(27, 2) COMMENT '消费金额，没有消费则为0',
is_first_consume BIGINT COMMENT '当日是否首次消费',
first_consume_store STRING COMMENT '首次消费门店，没有则为null',
first_consume_amount DECIMAL(27, 2) COMMENT '首次消费金额，没有消费则为0',
is_balance_consume BIGINT COMMENT '当日是否余额消费',
balance_consume_times BIGINT COMMENT '余额消费次数，没有消费则为0',
balance_pay_amount DECIMAL(27, 2) COMMENT '余额支付金额，没有消费则为0',
balance_consume_amount DECIMAL(27, 2) COMMENT '余额消费金额，没有消费则为0',
is_point_consume BIGINT COMMENT '当日是否积分消费',
point_consume_times BIGINT COMMENT '积分消费次数，没有消费则为0',
point_pay_amount DECIMAL(27, 2) COMMENT '积分支付金额，没有消费则为0',
point_consume_amount DECIMAL(27, 2) COMMENT '积分消费金额，没有消费则为0',
point_add BIGINT COMMENT '增加积分，没有则为0',
point_reduce BIGINT COMMENT '减少积分，没有则为0',
point_change BIGINT COMMENT '变动积分，没有则为0',
online_consume_times BIGINT COMMENT '线上订单量',
online_consume_amount DECIMAL(27, 2) COMMENT '线上消费金额',
offline_consume_times BIGINT COMMENT '线下订单量',
offline_consume_amount DECIMAL(27, 2) COMMENT '线下消费金额'
)
comment '会员行为天表'
partitioned by (dt STRING COMMENT '统计日期')
row format delimited fields terminated by ','
stored as orc
tblproperties ('orc.compress'='SNAPPY');思路分析:
从dwd.dwd_mem_member_union_i中获取注册信息，
dwd.dwd_mem_balance_change_i中获取充值信息，
dwm.dwm_mem_sell_order_i中获取销售信息，
dwm.dwm_mem_first_buy_i中获取首次消费信息，
dwd.dwd_mem_member_point_change_i中获取积分信息。
数据导入:
-- DWM: 会员行为数据表
with t1 as (
-- 注册信息数据
select
'2023-11-14' as trade_date,
zt_id,
if(
date_format(reg_time,'yyyy-MM-dd') = '2023-11-14',1,0
) as is_register,
0 as is_recharge,
0 as recharge_times,
0 as recharge_amount,
0 as is_consume,
0 as consume_times,
0 as consume_amount,
0 as is_first_consume,
'' as first_consume_store,
0 as first_consume_amount,
0 as is_balance_consume,
0 as balance_consume_times,
0 as balance_pay_amount,
0 as balance_consume_amount,
0 as is_point_consume,
0 as point_consume_times,
0 as point_pay_amount,
0 as point_consume_amount,
0 as point_add,
0 as point_reduce,
0 as point_change,
0 as online_consume_times,
0 as online_consume_amount,
0 as offline_consume_times,
0 as offline_consume_amount
from dwd.dwd_mem_member_union_i
-- 第一次导入: start_date 更改为 <= 但是第二次及其后续, 直接用 = 获取当天的日期注册数据
where date_format(reg_time,'yyyy-MM-dd') <= '2023-11-14' and end_date = '9999-99-99'
union all
-- 充值数据
select
trade_date,
zt_id,
0 as is_register,
1 as is_recharge,
times as recharge_times,
change_amount as recharge_amount,
0 as is_consume,
0 as consume_times,
0 as consume_amount,
0 as is_first_consume,
'' as first_consume_store,
0 as first_consume_amount,
0 as is_balance_consume,
0 as balance_consume_times,
0 as balance_pay_amount,
0 as balance_consume_amount,
0 as is_point_consume,
0 as point_consume_times,
0 as point_pay_amount,
0 as point_consume_amount,
0 as point_add,
0 as point_reduce,
0 as point_change,
0 as online_consume_times,
0 as online_consume_amount,
0 as offline_consume_times,
0 as offline_consume_amount
from dwd.dwd_mem_balance_change_i
where dt ='2023-11-14' and record_type = 2
union all
-- 消费情况
select
trade_date,
zt_id,
0 as is_register,
0 as is_recharge,
0 as recharge_times,
0 as recharge_amount,
1 as is_consume,
count( distinct if(trade_type = 0,parent_order_no,NULL)) - count( distinct if(trade_type = 5,parent_order_no,NULL)) as consume_times,
sum(real_paid_amount) as consume_amount,
0 as is_first_consume,
'' as first_consume_store,
0 as first_consume_amount,
max(is_balance_consume) as is_balance_consume,
count( distinct if(trade_type = 0 and is_balance_consume = 1,parent_order_no,NULL)) - count( distinct if(trade_type = 5 and is_balance_consume = 1,parent_order_no,NULL)) as balance_consume_times,
sum(
if(is_balance_consume = 1,balance_amount,0)
) as balance_pay_amount,
sum(
if(is_balance_consume = 1,real_paid_amount,0)
) as balance_consume_amount,
max(
if(point_amount > 0,1,0)
) as is_point_consume,
count( DISTINCT if(trade_type = 0 and point_amount > 0,parent_order_no,NULL) ) - count( DISTINCT if(trade_type = 5 and point_amount > 0,parent_order_no,NULL) ) as point_consume_times,
sum(
if(point_amount > 0,point_amount,0)
) as point_pay_amount,
sum(
if(point_amount > 0,real_paid_amount,0)
) as point_consume_amount,
0 as point_add,
0 as point_reduce,
0 as point_change,
count( DISTINCT if(trade_type = 0 and is_online_order = 1,parent_order_no,NULL) ) - count( DISTINCT if(trade_type = 5 and is_online_order = 1,parent_order_no,NULL) ) as online_consume_times,
sum(
if(is_online_order = 1,real_paid_amount,0)
) as online_consume_amount,
count( DISTINCT if(trade_type = 0 and is_online_order = 0,parent_order_no,NULL) ) - count( DISTINCT if(trade_type = 5 and is_online_order = 0,parent_order_no,NULL) ) as offline_consume_times,
sum(
if(is_online_order = 0,real_paid_amount,0)
) as offline_consume_amount
from dwm.dwm_mem_sell_order_i where dt = '2023-11-14'
group by trade_date,zt_id
union all
-- 首次消费
select
trade_date,
zt_id,
0 as is_register,
0 as is_recharge,
0 as recharge_times,
0 as recharge_amount,
0 as is_consume,
0 as consume_times,
0 as consume_amount,
1 as is_first_consume,
store_no as first_consume_store,
sale_amount as first_consume_amount,
0 as is_balance_consume,
0 as balance_consume_times,
0 as balance_pay_amount,
0 as balance_consume_amount,
0 as is_point_consume,
0 as point_consume_times,
0 as point_pay_amount,
0 as point_consume_amount,
0 as point_add,
0 as point_reduce,
0 as point_change,
0 as online_consume_times,
0 as online_consume_amount,
0 as offline_consume_times,
0 as offline_consume_amount
from dwm.dwm_mem_first_buy_i where dt = '2023-11-14'
union all
-- 积分变动表
select
trade_date,
zt_id,
0 as is_register,
0 as is_recharge,
0 as recharge_times,
0 as recharge_amount,
0 as is_consume,
0 as consume_times,
0 as consume_amount,
0 as is_first_consume,
'' as first_consume_store,
0 as first_consume_amount,
0 as is_balance_consume,
0 as balance_consume_times,
0 as balance_pay_amount,
0 as balance_consume_amount,
0 as is_point_consume,
0 as point_consume_times,
0 as point_pay_amount,
0 as point_consume_amount,
point_add,
point_reduce,
point_change,
0 as online_consume_times,
0 as online_consume_amount,
0 as offline_consume_times,
0 as offline_consume_amount
from dwd.dwd_mem_member_point_change_i where dt = '2023-11-14'
),
t2 as (
select
trade_date,
zt_id,
max(is_register) as is_register,
max(is_recharge) as is_recharge,
sum(recharge_times) as recharge_times,
sum(recharge_amount) as recharge_amount,
max(is_consume) as is_consume,
sum(consume_times) as consume_times,
sum(consume_amount) as consume_amount,
max(is_first_consume) as is_first_consume,
max(first_consume_store) as first_consume_store,
sum(first_consume_amount) as first_consume_amount,
max(is_balance_consume) as is_balance_consume,
sum(balance_consume_times) as balance_consume_times,
sum(balance_pay_amount) as balance_pay_amount,
sum(balance_consume_amount) as balance_consume_amount,
max(is_point_consume) as is_point_consume,
sum(point_consume_times) as point_consume_times,
sum(point_pay_amount) as point_pay_amount,
sum(point_consume_amount) as point_consume_amount,
sum(point_add) as point_add,
sum(point_reduce) as point_reduce,
sum(point_change) as point_change,
sum(online_consume_times) as online_consume_times,
sum(online_consume_amount) as online_consume_amount,
sum(offline_consume_times) as offline_consume_times,
sum(offline_consume_amount) as offline_consume_amount,
trade_date as dt
from t1
group by trade_date,zt_id
)
insert overwrite table dwm.dwm_mem_member_behavior_day_i partition (dt)
select
t2.trade_date,
t3.week_trade_date,
t3.month_trade_date,
t2.zt_id,
t4.bind_md,
t4.reg_md,
t4.reg_time,
t2.is_register,
t2.is_recharge,
t2.recharge_times,
t2.recharge_amount,
t2.is_consume,
t2.consume_times,
t2.consume_amount,
t2.is_first_consume,
t2.first_consume_store,
t2.first_consume_amount,
t2.is_balance_consume,
t2.balance_consume_times,
t2.balance_pay_amount,
t2.balance_consume_amount,
t2.is_point_consume,
t2.point_consume_times,
t2.point_pay_amount,
t2.point_consume_amount,
t2.point_add,
t2.point_reduce,
t2.point_change,
t2.online_consume_times,
t2.online_consume_amount,
t2.offline_consume_times,
t2.offline_consume_amount,
t2.dt
from t2 left join dim.dwd_dim_date_f t3 on t2.trade_date = t3.trade_date
left join dwd.dwd_mem_member_union_i t4 on t2.zt_id = t4.zt_id and t4.end_date = '9999-99-99'
-- 说明: 在实施中 大家需要调整日期, 依次将14~20号的每天会员的余额数据跑出来即可
/* 在生产环境中(工作中), 我们可以通过海豚调度器提供的补数方案, 指定需要补数的范围, 调度器会自动将过去的几天数据全部补回来(无需执行, 了解即可, 面试中按照这个说即可)*/
![LeetCode [中等]全排列(回溯算法)](https://img-blog.csdnimg.cn/direct/851755e8324449e7825ef1d18c43c454.png)
