Hive SQL 优化案例详解

HiveSQL 经典优化案例一：
1.1 将要执行的查询(执行了 1 个多小时才出结果)：

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;复制代码

1.2 查询语句涉及到的表有 7.7 亿 + 数据。（查询如下)

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

1.3 优化思路：既然将要执行的查询是按照 dt, strategy, ab_group, source 这 4 个字段分组， 那么在建表的时候，就按这四个字段中的 N 个（1 或 2 或 3 或 4）个字段组合分区，直接让 count(distinct xx) 之类的查询定位到“更少的数据子集”，其执行效率就应该更高了（不需要每个子任务均从 7.7 亿 + 的数据中(去重)统计)。

1.4 先看每个字段将会有多少分区（因为 Hive 表分区也不宜过多，一般一个查询语句涉及到的 hive 分区 应该控制在 2K 内)

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;复制代码

[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。复制代码

1.5 根据原表，新建分区表，并将原表数据插入新表：

show create table dwb_v8sp_tmp.base_report_bystrategy_byab_source_column_zkl;
复制代码

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

创建新表：按 dt,source,stragegy,ab_group 分区(注意先后顺序，一般习惯分区数越少的越靠前，根据 1.5 的查询可知：dt=1,source=2,strategy=14,ab_group=72)

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
);复制代码

将原表数据插入新表：

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`;复制代码

核对两表的数据是否一致：

1.6 基于新表执行查询(执行 5 分钟出结果)：

HiveSQL 经典优化案例二：

问题描述：一个复杂的 SQL，查询执行一段时间后报错：基本上是查不出来;

分析函数对于大表来说不是 hive 的强项，这个时候我们将其分解成很多子集，并且合理利用 hive 分区表的优势，然后去 join 。

2.1 将要执行的查询

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';
复制代码

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';复制代码

分解成三个子集，并保存到三张表： bi_tmp.zyt1, bi_tmp.zyt2, bi_tmp.zyt3

-- 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)复制代码

HiveSQL 经典优化案例三：

如下 SQL，用到了 PERCENTILE_APPROX 函数，问题描述：如下 SQL，用到了 PERCENTILE_APPROX 函数，个人初步分析认为：由于用到该函数的次数太多，导致性能严重下降。

我仔细查了一下该函数，发现：它是支持“数组传参”的，那么就不难找到优化该 SQL 的方法了。

3.1 原 SQL 性能测试：

3.2 优化后的 SQL，性能测试：

优化后的 SQL，性能提升了 4 倍多。