一、目的

当Hive的计算引擎是spark或mr时，发现海豚调度HQL任务的脚本并不同，mr更简洁

二、Hive的计算引擎是Spark时

（一）海豚调度脚本

#! /bin/bash
source /etc/profile

nowdate=`date --date='0 days ago' "+%Y%m%d"`
yesdate=`date -d yesterday +%Y-%m-%d`

hive -e "
use hurys_dc_dwd;

set hive.vectorized.execution.enabled=false;
set hive.auto.convert.join=false;
set mapreduce.map.memory.mb=10150;
set mapreduce.map.java.opts=-Xmx6144m;
set mapreduce.reduce.memory.mb=10150;
set mapreduce.reduce.java.opts=-Xmx8120m;
set hive.exec.dynamic.partition.mode=nonstrict;
set hive.exec.dynamic.partition=true;
set hive.exec.parallel=true;
set hive.support.concurrency=false;
set mapreduce.map.memory.mb=4128;
set hive.vectorized.execution.enabled=false;

set hive.exec.dynamic.partition=true;
set hive.exec.dynamic.partition.mode=nonstrict;
set hive.exec.max.dynamic.partitions.pernode=1000;
set hive.exec.max.dynamic.partitions=1500;

insert  overwrite  table dwd_evaluation partition(day='$yesdate')
select device_no,
       cycle,
       lane_num,
       create_time,
       lane_no,
       volume,
       queue_len_max,
       sample_num,
       stop_avg,
       delay_avg,
       stop_rate,
       travel_dist,
       travel_time_avg
from hurys_dc_ods.ods_evaluation
where volume is not null  and   date(create_time)= '$yesdate'
group by device_no, cycle, lane_num, create_time, lane_no,
         volume, queue_len_max, sample_num, stop_avg, delay_avg, stop_rate, travel_dist, travel_time_avg
"

（二）任务流执行结果

调度执行成功，时间需要1m29s

三、Hive的计算引擎是MR时

（一）海豚调度脚本

#! /bin/bash
source /etc/profile

nowdate=`date --date='0 days ago' "+%Y%m%d"`
yesdate=`date -d yesterday +%Y-%m-%d`

hive -e "
use hurys_dc_dwd;

set hive.exec.dynamic.partition=true;
set hive.exec.dynamic.partition.mode=nonstrict;
set hive.exec.max.dynamic.partitions.pernode=1000;
set hive.exec.max.dynamic.partitions=1500;

insert  overwrite  table dwd_evaluation partition(day='$yesdate')
select device_no,
       cycle,
       lane_num,
       create_time,
       lane_no,
       volume,
       queue_len_max,
       sample_num,
       stop_avg,
       delay_avg,
       stop_rate,
       travel_dist,
       travel_time_avg
from hurys_dc_ods.ods_evaluation
where volume is not null  and   date(create_time)= '$yesdate'
group by device_no, cycle, lane_num, create_time, lane_no,
         volume, queue_len_max, sample_num, stop_avg, delay_avg, stop_rate, travel_dist, travel_time_avg
"

（二）任务流执行结果

调度执行成功，时间需要1m3s

四、脚本区别

计算引擎为spark时，脚本比计算引擎为mr多，而且spark运行速度比mr慢

set hive.vectorized.execution.enabled=false;
set hive.auto.convert.join=false;
set mapreduce.map.memory.mb=10150;
set mapreduce.map.java.opts=-Xmx6144m;
set mapreduce.reduce.memory.mb=10150;
set mapreduce.reduce.java.opts=-Xmx8120m;
set hive.exec.dynamic.partition.mode=nonstrict;
set hive.exec.dynamic.partition=true;
set hive.exec.parallel=true;
set hive.support.concurrency=false;
set mapreduce.map.memory.mb=4128;
set hive.vectorized.execution.enabled=false;

mr为计算引擎时任务流脚本不能添加上面这些优化语句，不然会报错

在海豚调度HiveSQL任务流，推荐使用mr作为Hive的计算引擎。

不仅不需要安装spark，而且脚本简洁、任务执行速度快！