Hive SQL经典优化案例

SELECT dt as DATA_DATE,STRATEGY,AB_GROUP,SOURCE,
    count(distinct case when lower(event) not like '%push%' and event!='corner_mark_show' then udid else null end) as DAU,
    count(case when event='client_show' then 1 else null end) as TOTAL_VSHOW,
    count(distinct case when event='client_show' then vid else null end) as TOTAL_VIDEO_VSHOW,
    count(case when event='video_play' then 1 else null end) as TOTAL_VV_VP,
    count(distinct case when event='video_play' then udid else null end) as TOTAL_USERS_VP,
    count(case when event='effective_play' then 1 else null end) as TOTAL_VV_EP,
    count(distinct case when event='effective_play' then udid else null end) as TOTAL_USERS_EP,
    sum(case when event='video_over' then duration else 0 end) as TOTAL_DURATION,
    count(case when event='video_over' then 1 else null end) as TOTAL_VOVER,
    sum(case when event='video_over' then play_cnts else 0 end) as TOTAL_VOVER_PCNTS,
    count(case when event='push_video_clk' then 1 else null end) as TOTAL_PUSH_VC,
    count(distinct case when event='app_start' and body_source = 'push' then udid else null end) as TOTAL_PUSH_START,
    count(case when event='post_comment' then 1 else null end) as TOTAL_REPLY,
    count(distinct case when event='post_comment' then udid else null end) as TOTAL_USERS_REPLY
    FROM dwb_v8sp_tmp.base_report_bystrategy_byab_source_column_zkl
group by dt,strategy,ab_group,source;

jdbc:hive2://ks-hdp-master-01.dns.rightpad (default)> select count(*) from dwb_v8sp_tmp.base_report_bystrategy_byab_source_column_zkl;

jdbc:hive2://ks-hdp-master-01.dns.rightpad (default)> 
select count(distinct dt) as dis_dt, count(distinct strategy) as dis_strategy, 
count(distinct ab_group) as dis_ab_group, 
count(distinct source) as dis_source
from dwb_v8sp_tmp.base_report_bystrategy_byab_source_column_zkl;

[hue@ks-hdp-client-v02 10:55:08 /usr/local/hue]$ python
Python 2.7.12 (default, Dec  4 2017, 14:50:18)
[GCC 5.4.0 20160609] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> 2*14*72
2016
-- 2016 个分区还可以接受。

show create table dwb_v8sp_tmp.base_report_bystrategy_byab_source_column_zkl;

jdbc:hive2://ks-hdp-master-01.dns.rightpad (default)> show create table dwb_v8sp_tmp.base_report_bystrategy_byab_source_column_zkl;

create external table `dwb_v8sp_tmp.base_report_bystrategy_byab_source_column_lym`(
  event string,
  udid string,
  vid string,
  duration string,
  body_source string,
  play_cnts string
)
PARTITIONED BY (
  dt string,
  source string,
  strategy string,
  ab_group string
);

insert into `dwb_v8sp_tmp.base_report_bystrategy_byab_source_column_lym` partition(dt,source,strategy,ab_group)
select event,udid,vid,duration,body_source,play_cnts,dt,source,strategy,ab_group
from `dwb_v8sp_tmp.base_report_bystrategy_byab_source_column_zkl`;

create table bi_tmp.aloha_UserLoyalty_190301_190303 as 
    select aid, imei, idfa, udid, event, duration, dt, time_local, hour, source, 
        first_value(time_local) over(partition by udid, event order by time_local) as first_time,
        last_value(time_local) over(partition by udid, event order by time_local) as last_time,
        count(time_local) over(partition by udid, event, dt) as event_count_per_day,
        sum(duration) over(partition by udid, event, dt) as event_duration_each_day
    from dwb_v8sp.event_column_info_new_hour
    where event in ('app_start', 'app_exit', 'effective_play', 'share_succ', 'like', 'unlike', 'like_comment', 'unlike_comment', 
        'comment_success')
        and dt >= '2019-03-01' and dt <= '2019-03-03';

select count(*) 
from dwb_v8sp.event_column_info_new_hour
where event in ('app_start', 'app_exit', 'effective_play', 'share_succ', 'like', 'unlike', 'like_comment', 'unlike_comment', 'comment_success')
and dt >= '2019-03-01' and dt <= '2019-03-03';

select count(distinct event) as dis_event
from dwb_v8sp.event_column_info_new_hour
where event in ('app_start', 'app_exit', 'effective_play', 'share_succ', 'like', 'unlike', 'like_comment', 'unlike_comment', 'comment_success')
and dt >= '2019-03-01' and dt <= '2019-03-03';

-- drop table if exists bi_tmp.zyt1;
create table bi_tmp.zyt1 partitioned by(event)
as
select udid, 
       min(time_local) as first_time,
       max(time_local) as last_time,
       event
from dwb_v8sp.event_column_info_new_hour
where event in ('app_start', 'app_exit', 'effective_play', 'share_succ', 'like', 'unlike', 'like_comment', 'unlike_comment', 'comment_success')
and dt >= '2019-03-01' and dt <= '2019-03-03'
group by udid, event;

-- drop table if exists bi_tmp.zyt2 purge;
create table bi_tmp.zyt2 partitioned by(dt,event)
as
select udid, 
       count(time_local) as event_count_per_day,
       sum(duration) as event_duration_each_day,
       dt,
       event
from dwb_v8sp.event_column_info_new_hour
where event in ('app_start', 'app_exit', 'effective_play', 'share_succ', 'like', 'unlike', 'like_comment', 'unlike_comment', 'comment_success')
and dt >= '2019-03-01' and dt <= '2019-03-03'
group by udid, dt, event;

create table bi_tmp.zyt3 partitioned by(dt,event)
as select aid, imei, idfa, udid, duration, time_local, hour, source, dt, event
from dwb_v8sp.event_column_info_new_hour t3
    where event in ('app_start', 'app_exit', 'effective_play', 'share_succ', 'like', 'unlike', 'like_comment', 'unlike_comment', 
        'comment_success')
        and dt >= '2019-03-01' and dt <= '2019-03-03';

-- 插入目标表：
create table bi_tmp.aloha_UserLoyalty_190301_190303 as 
    select t3.aid, t3.imei, t3.idfa, t3.udid, t3.event, t3.duration, t3.dt, t3.time_local, t3.hour, t3.source, 
        t1.first_time,
        t1.last_time,
        t2.event_count_per_day,
        t2.event_duration_each_day
    from bi_tmp.zyt1 t1 join bi_tmp.zyt2 t2 on t1.event=t2.event and t1.udid=t2.udid
    join bi_tmp.zyt3 t3 on t2.dt=t3.dt and t2.event= t3.event and t2.udid=t3.udid;

-- 验证数据：（与上面的查询记录行数对的上)