如何在 pyspark 中使用外部（自定义）包？

Question

我正在尝试复制此处给出的灵魂 https://www.cloudera.com/documentation/enterprise/5-7-x/topics/spark_python.html 在 pypspark 中导入外部包。但它失败了。

我的代码：

spark_distro.py

from pyspark import SparkContext, SparkConf

def import_my_special_package(x):
    from external_package import external
    return external.fun(x)

conf = SparkConf()
sc = SparkContext()
int_rdd = sc.parallelize([1, 2, 3, 4])
int_rdd.map(lambda x: import_my_special_package(x)).collect()

external_package.py

class external:

    def __init__(self,in):
        self.in = in

    def fun(self,in):
        return self.in*3

spark 提交命令：

spark-submit \
   --master yarn \
  /path to script/spark_distro.py  \
  --py-files /path to script/external_package.py \
  1000

实际错误：

Actual:
  vs = list(itertools.islice(iterator, batch))
  File "/home/gsurapur/pyspark_examples/spark_distro.py", line 13, in <lambda>
  File "/home/gsurapur/pyspark_examples/spark_distro.py", line 6, in import_my_special_package
ImportError: No module named external_package

预期输出：

[3,6,9,12]

我也尝试了 sc.addPyFile 选项，但它因同样的问题而失败。

Answer 1

我知道，事后看来，这听起来很愚蠢，但 spark-submit 的参数顺序通常不可互换：所有与 Spark 相关的参数，包括 --py-file，必须是之前要执行的脚本：

# your case:
spark-submit --master yarn-client /home/ctsats/scripts/SO/spark_distro.py --py-files /home/ctsats/scripts/SO/external_package.py
[...]
ImportError: No module named external_package

# correct usage:
spark-submit --master yarn-client --py-files /home/ctsats/scripts/SO/external_package.py /home/ctsats/scripts/SO/spark_distro.py
[...]
[3, 6, 9, 12]

测试你的脚本修改如下：

spark_distro.py

from pyspark import SparkContext, SparkConf

def import_my_special_package(x):
    from external_package import external
    return external(x)

conf = SparkConf()
sc = SparkContext()
int_rdd = sc.parallelize([1, 2, 3, 4])
print int_rdd.map(lambda x: import_my_special_package(x)).collect()

external_package.py

def external(x):
     return x*3

修改可以说没有改变问题的本质...

Answer 2

这是关于 addPyFile 的情况：

spark_distro2.py

from pyspark import SparkContext, SparkConf

def import_my_special_package(x):
    from external_package import external
    return external(x)

conf = SparkConf()
sc = SparkContext()
sc.addPyFile("/home/ctsats/scripts/SO/external_package.py") # added
int_rdd = sc.parallelize([1, 2, 3, 4])
print int_rdd.map(lambda x: import_my_special_package(x)).collect()

测试：

spark-submit --master yarn-client /home/ctsats/scripts/SO/spark_distro2.py
[...]
[3, 6, 9, 12]

如何在 pyspark 中使用外部（自定义）包？

How to use external (custom) package in pyspark?

hadoop-yarn

apache-spark

pyspark