工作中碰到个需求，需要我去SFTP服务器采集HIVE导出的数据，由于这个大数据平台设计问题，很多东西没考虑到，他们的导出方式是用HIVE提供的insert overwrite local directory这个语句来导出HIVE数据。
导出文件的大小和数量，以及每个文件的平均大小对于我的采集程序而言都是有影响的。由于他们平台设计的问题，导致只能嵌入sql，所以这面我需要通过hive sql的一些设置来实现这个功能。
HIVE设置reduce的数量
set mapred.reduce.tasks=5;
HIVE导出数据的语句
insert overwrite local directory '/opt/test/data' row format delimited fields terminated by '\t' select customer_id, customer_fname, customer_zipcode from customers cluster by customer_zipcode
hive> set mapred.reduce.tasks=5;
hive> insert overwrite local directory '/opt/test/data' row format delimited fields terminated by '\t' select customer_id, customer_fname, customer_zipcode from customers cluster by customer_zipcode;
Query ID = hdfs_20190111202929_76c074d9-c0ac-4073-a867-4c69aca85c7d
Total jobs = 1
Launching Job 1 out of 1
Number of reduce tasks not specified. Defaulting to jobconf value of: 5
In order to change the average load for a reducer (in bytes):
set hive.exec.reducers.bytes.per.reducer=
In order to limit the maximum number of reducers:
set hive.exec.reducers.max=
In order to set a constant number of reducers:
set mapreduce.job.reduces=
Starting Job = job_1544843719855_0021, Tracking URL = http://quickstart.cloudera:8088/proxy/application_1544843719855_0021/
Kill Command = /usr/lib/hadoop/bin/hadoop job -kill job_1544843719855_0021
Hadoop job information for Stage-1: number of mappers: 1; number of reducers: 5
2019-01-11 20:30:06,871 Stage-1 map = 0%, reduce = 0%
2019-01-11 20:30:18,011 Stage-1 map = 100%, reduce = 0%, Cumulative CPU 3.85 sec
2019-01-11 20:30:29,015 Stage-1 map = 100%, reduce = 20%, Cumulative CPU 6.36 sec
2019-01-11 20:30:30,091 Stage-1 map = 100%, reduce = 40%, Cumulative CPU 8.78 sec
2019-01-11 20:30:39,016 Stage-1 map = 100%, reduce = 60%, Cumulative CPU 8.78 sec
2019-01-11 20:30:40,077 Stage-1 map = 100%, reduce = 80%, Cumulative CPU 13.86 sec
2019-01-11 20:30:47,615 Stage-1 map = 100%, reduce = 100%, Cumulative CPU 17.0 sec
MapReduce Total cumulative CPU time: 17 seconds 0 msec
Ended Job = job_1544843719855_0021
Copying data to local directory /opt/test/data
MapReduce Jobs Launched:
Stage-Stage-1: Map: 1 Reduce: 5 Cumulative CPU: 17.0 sec HDFS Read: 110232 HDFS Write: 215580 SUCCESS
Total MapReduce CPU Time Spent: 17 seconds 0 msec
OK
Time taken: 56.395 seconds
进入/opt/test/data文件夹查看结果
[root@quickstart data]# ll -h
total 220K
-rw-r--r-- 1 hdfs hdfs 25K Jan 11 20:30 000000_0
-rw-r--r-- 1 hdfs hdfs 28K Jan 11 20:30 000001_0
-rw-r--r-- 1 hdfs hdfs 26K Jan 11 20:30 000002_0
-rw-r--r-- 1 hdfs hdfs 28K Jan 11 20:30 000003_0
-rw-r--r-- 1 hdfs hdfs 107K Jan 11 20:30 000004_0
上面的语句如果customer_zipcode不均匀的话，会造成数据倾斜,比如 000004_0这个文件就比其他的文件要大，为了保证数据的均匀，故使用distribute by rand()，同时调整reduce数量为10，验证是否生成了10个结果文件
hive> set mapred.reduce.tasks=10;
hive> insert overwrite local directory '/opt/test/data' row format delimited fields terminated by '\t' select customer_id, customer_fname, customer_zipcode from customers distribute by rand();
Query ID = hdfs_20190111205252_2d4b8cde-7d9b-4276-94fa-8fa83b77b4db
Total jobs = 1
Launching Job 1 out of 1
Number of reduce tasks not specified. Defaulting to jobconf value of: 10
In order to change the average load for a reducer (in bytes):
set hive.exec.reducers.bytes.per.reducer=
In order to limit the maximum number of reducers:
set hive.exec.reducers.max=
In order to set a constant number of reducers:
set mapreduce.job.reduces=
Starting Job = job_1544843719855_0022, Tracking URL = http://quickstart.cloudera:8088/proxy/application_1544843719855_0022/
Kill Command = /usr/lib/hadoop/bin/hadoop job -kill job_1544843719855_0022
Hadoop job information for Stage-1: number of mappers: 1; number of reducers: 10
2019-01-11 20:52:24,166 Stage-1 map = 0%, reduce = 0%
2019-01-11 20:52:33,763 Stage-1 map = 100%, reduce = 0%, Cumulative CPU 4.71 sec
2019-01-11 20:52:45,572 Stage-1 map = 100%, reduce = 20%, Cumulative CPU 9.66 sec
2019-01-11 20:52:55,322 Stage-1 map = 100%, reduce = 40%, Cumulative CPU 14.71 sec
2019-01-11 20:53:03,935 Stage-1 map = 100%, reduce = 50%, Cumulative CPU 17.42 sec
2019-01-11 20:53:06,107 Stage-1 map = 100%, reduce = 60%, Cumulative CPU 20.6 sec
2019-01-11 20:53:13,643 Stage-1 map = 100%, reduce = 70%, Cumulative CPU 22.89 sec
2019-01-11 20:53:14,718 Stage-1 map = 100%, reduce = 80%, Cumulative CPU 25.14 sec
2019-01-11 20:53:21,092 Stage-1 map = 100%, reduce = 90%, Cumulative CPU 27.3 sec
2019-01-11 20:53:23,198 Stage-1 map = 100%, reduce = 100%, Cumulative CPU 29.8 sec
MapReduce Total cumulative CPU time: 29 seconds 800 msec
Ended Job = job_1544843719855_0022
Copying data to local directory /opt/test/data
MapReduce Jobs Launched:
Stage-Stage-1: Map: 1 Reduce: 10 Cumulative CPU: 29.8 sec HDFS Read: 126360 HDFS Write: 215580 SUCCESS
Total MapReduce CPU Time Spent: 29 seconds 800 msec
OK
Time taken: 68.831 seconds
hive>
进入文件夹查看文件，可以看到一共10个文件，大小均匀
[root@quickstart data]# ll -h
total 240K
-rw-r--r-- 1 hdfs hdfs 21K Jan 11 20:53 000000_0
-rw-r--r-- 1 hdfs hdfs 22K Jan 11 20:53 000001_0
-rw-r--r-- 1 hdfs hdfs 21K Jan 11 20:53 000002_0
-rw-r--r-- 1 hdfs hdfs 22K Jan 11 20:53 000003_0
-rw-r--r-- 1 hdfs hdfs 22K Jan 11 20:53 000004_0
-rw-r--r-- 1 hdfs hdfs 22K Jan 11 20:53 000005_0
-rw-r--r-- 1 hdfs hdfs 22K Jan 11 20:53 000006_0
-rw-r--r-- 1 hdfs hdfs 21K Jan 11 20:53 000007_0
-rw-r--r-- 1 hdfs hdfs 21K Jan 11 20:53 000008_0
-rw-r--r-- 1 hdfs hdfs 22K Jan 11 20:53 000009_0
[root@quickstart data]#