hive自定义mapred脚本运行机制分析

     1.  hive自定义mapred脚本运行机制

        1.1）HQL中使用自定义mapred脚本的语法格式及案例

语法格式：

FROM (
    FROM src
    MAP expression (',' expression)*
    (inRowFormat)?
    USING 'my_map_script'
    ( AS colName (',' colName)* )?
    (outRowFormat)? (outRecordReader)?
    ( clusterBy? | distributeBy? sortBy? ) src_alias
  )
  REDUCE expression (',' expression)*
    (inRowFormat)?
    USING 'my_reduce_script'
    ( AS colName (',' colName)* )?
    (outRowFormat)? (outRecordReader)?
 
  FROM (
    FROM src
    SELECT TRANSFORM '(' expression (',' expression)* ')'
    (inRowFormat)?
    USING 'my_map_script'
    ( AS colName (',' colName)* )?
    (outRowFormat)? (outRecordReader)?
    ( clusterBy? | distributeBy? sortBy? ) src_alias
  )
  SELECT TRANSFORM '(' expression (',' expression)* ')'
    (inRowFormat)? 
    USING 'my_reduce_script'
    ( AS colName (',' colName)* )?
    (outRowFormat)? (outRecordReader)?

HQL使用案例：

FROM (
  FROM pv_users
  MAP pv_users.userid, pv_users.date
  USING 'map_script'
  AS dt, uid
  CLUSTER BY dt) map_output
INSERT OVERWRITE TABLE pv_users_reduced
  REDUCE map_output.dt, map_output.uid
  USING 'reduce_script'
  AS date, count;
FROM (
  FROM pv_users
  SELECT TRANSFORM(pv_users.userid, pv_users.date)
  USING 'map_script'
  AS dt, uid
  CLUSTER BY dt) map_output
INSERT OVERWRITE TABLE pv_users_reduced
  SELECT TRANSFORM(map_output.dt, map_output.uid)
  USING 'reduce_script'
  AS date, count;

1.2）HIVE运行自定义mapred脚本的机制

MAP和REDUCE关键字都是SELECT TRANSFORM关键字的语法转换，也就是说：在使用MAP关键字时，自定义脚本不一定是在map阶段运行，使用REDUCE关键字时自定义脚本不一定是在red阶段运行；

可以参考如下案例：

#red.py 代码

#!/usr/bin/python
# coding=utf-8

import sys
import time
import traceback

sys.stderr.write("we are running at " + time.strftime("%Y-%m-%d %H:%M:%S") + "\n")

num = 10
i = 0

try:
    for l in sys.stdin:
        if i < num:
            print l[:-1]
            i += 1
except:
    traceback.print_exc(file=sys.stderr)

案例1：该HQL语句生成一个mapred Job，该Job只执行了map阶段，red数量为零；

FROM (
  FROM testhive
  MAP testhive.channel, testhive.pv
  USING '/bin/cat'
  AS dt, uid) map_output
REDUCE map_output.dt, map_output.uid
  USING '/bin/cat'
  AS date, count;

案例2：该HQL语句生成一个mapred Job，该Job只执行了map阶段，red数量为零；

FROM testhive
MAP testhive.channel, testhive.pv
USING 'red.py'
AS c, p;

FROM testhive
REDUCE testhive.channel, testhive.pv
USING 'red.py'
AS c, p;

案例3：该HQL语句生成一个mapred Job，该Job执行了map red阶段

FROM (
  FROM testhive
  MAP testhive.channel, testhive.pv
  USING 'red.py'
  AS dt, uid
  cluster By dt) map_output
REDUCE map_output.dt, map_output.uid
  USING 'red.py'
  AS date, count;

总结：HQL中嵌入自定义脚本时，自定义脚本的运行阶段取决于自定义脚本在HQL语句中的语法位置。也就是说，hive 在将hql语句翻译成mapred任务时，如果hql中存在group by、cluster by、sort by等需要reduce阶段的命令时，且MAP、REDUCE命令在上述命令之后，则自定义脚本运行在red端。

注意事项：如果是在windows上开发的mapred脚本，在提交到linux后最好先执行一下 dos2unix 命令，否则可能会出现错误：

hadoop mapred框架错误： 
java.io.IOException: Cannot run program "/data9/mapred/local/taskTracker/zhaoxiuxiang/jobcache/job_201311120949_47441/attempt_201311120949_47441_m_000003_2/work/././map.py": error=2, No such file or directory
本地错误：
-bash: ./map.py: /usr/bin/python^M: bad interpreter: No such file or directory

注意事项：使用下面的参数打开report和调试信息输出

stream.stderr.reporter.enabled=true
stream.stderr.reporter.prefix=reporter: