您还可以按照您的方法避免流式处理和python以及所有这些，但是使用hivesubstr（）

CREATE TABLE IF NOT EXISTS input_raw(line STRING);

LOAD DATA LOCAL INPATH '${hiveconf:input}' OVERWRITE INTO TABLE input_raw;

CREATE TABLE IF NOT EXISTS processed_data(
field1 STRING,
field2 STRING
field3 STRING);

INSERT INTO TABLE processed_data
SELECT
substr(line,1,4) as field1,
substr(line,5,4) as field2,
substr(line,9,4) as field3
FROM input_raw;

DROP TABLE input_raw;

我用这种方法解决了类似的问题：
给定换行分隔行、固定偏移列的输入，
首先将数据输入到一个包含一列的表中，一个字符串
然后通过流式hadoop传递java类或python模块，该模块接收一行并返回多个字段：

import sys

for line in sys.stdin:
    # line will have a newline on the end you don't want
    line = line.strip()
    output = []
    output.append(line[:4])
    output.append(line[4:8])
    output.append(line[8:12])
    print '\t'.join(output)

您的配置单元脚本将如下所示：

CREATE TABLE IF NOT EXISTS input_raw(line STRING);

LOAD DATA LOCAL INPATH '${hiveconf:input}' OVERWRITE INTO TABLE input_raw;

CREATE TABLE IF NOT EXISTS processed_data(
field1 STRING,
field2 STRING
field3 STRING);

delete FILE processing.py;

add FILE processing.py;

INSERT INTO TABLE processed_data
SELECT
TRANSFORM (line)
USING 'python processing.py'
AS(field1, field2, field3)
FROM input_raw;

DROP TABLE input_raw;