有一张点击表和曝光表 ad_click_b,ad_exposure_b,注意:这俩张表都是以小时为单位的增量的分区表,都拥有字段:rtime,req_id,data_date。(rtime 是精确到秒的时间戳类型;req_id 是这俩张表的主键,可以用于ad_click_b表与ad_exposure_b表的关联操作;data_date 是分区字段,并且分区单位是小时,格式为yyMMddHH)。以点击表为主表,求点击表在 data_date=2020042406中,在曝光表中往前延3个小时(2020042404,2020042405,2020042406), 统计曝光事件与点击事件之间的时间间隔分布情况,按30分钟为粒度。本需求的目的是求:在 data_date=2020042406这一小时的点击量中, 有多少的比例是广告曝光后在30分钟内点击的;有多少的比例是广告曝光后在60分钟内点击的;有多少的比例是广告曝光后在90分钟内点击的…
结果展示示例:
时间间隔(单位分钟) 占比情况(%)
分钟 百分比
30 80
60 95
90 96
120 98
150 99
180 99.6
210 100
30 80 这一行表示:在2020042406这一个小时的点击量中,80%以上是30分钟内(点击时间-曝光时间)的点击的。
60 95 这一行表示:在2020042406这一个小时的点击量中,95%以上是60分钟内(点击时间-曝光时间)的点击的。
思路:
1.用req_id 关联点击事件left join 曝光事件,且曝光时间和点击事件都存在的情况下,click[20200422],exposure[20200420,20200421,20200422]
2.获取点击时间和曝光时间
3.利用2步骤中的 点击时间-曝光时间差 得到 时间差
4.计算各个时间差的占比
function etldata() {
echo "start etldata"
beeline -e "
select
time
,count
,sum
,sum(rate) over(rows between UNBOUNDED PRECEDING and CURRENT ROW)
from(
select
time
,count
,sum
,(count*100.00)/sum as rate
from
(
select
time
,count
,sum(count) over() as sum
from(
select
(time+1)*30 as time
,count(*) as count
from
(
select
click.req_id
,click.rtime as ctime
,expo.rtime as etime
,floor((click.rtime-expo.rtime)/1000/60/30) as time
from (
select rtime,req_id
from tmp.ad_click_b
where data_date=2020042406
) click
left join
(
select rtime,req_id
from tmp.ad_exposure_b
where data_date>=2020042404 and data_date<=2020042406
) expo on click.req_id = expo.req_id
)cli_epo group by time
)t2
)t1
)t order by time asc
"
echo "end etldata"
}
function main() {
etldata
}
main
with v_cli_epo as (
select click.req_id,
click.rtime as ctime,
expo.rtime as etime,
floor((click.rtime - expo.rtime) / 1000 / 60 / 30) as time from(
select rtime,
req_id
from tmp.ad_click_b where data_date = 2020042406
)click
left join(
select rtime,
req_id
from tmp.ad_exposure_b where data_date >= 2020042404 and data_date <= 2020042406
)expo on click.req_id = expo.req_id
)
, v_cli_epo_sum as (
select time,
count,
sum(count) over() as sum
from(
select (time + 1) * 30 as time,
count(*) as count
from v_cli_epo
group by time
)v_cli_epo_rn
)
, v_cli_epo_rate as (
select time,
count,
sum,
(count * 100.00) / sum as rate
from v_cli_epo_sum
)
select
time,
count,
sum,
sum(rate) over(rows between UNBOUNDED PRECEDING and CURRENT ROW)
from v_cli_epo_rate order by time asc