spark-submit python egg 解决三方件依赖问题

假设spark里用到了purl这个三方件，https://github.com/ultrabluewolf/p.url，他还额外依赖futures这个三方件（six的话，anaconda2自带）。

pyspark 代码如下：

from pyspark import SparkConf, SparkContext
conf = SparkConf().setMaster("local").setAppName("My test App")
sc = SparkContext(conf=conf)
#from purl import Purl

def get_purl(x):
    from purl import Purl
    url = Purl('https://github.com/search?q={}'.format(x))
    return str(url.add_query('name', 'dog'))

int_rdd = sc.parallelize([1, 2, 3, 4])
r =int_rdd.map(lambda x: get_purl(x))
print(r.collect())

下面说明如何编译打包egg。

通过https://pypi.org/project/p.url/#files 下载源码。然后解压：

python setup.py  bdist_egg

在dist目录下可以看到有egg文件生成。

同理，下载https://pypi.org/project/future/#files futures的源码，然后解压生成egg文件。

最终运行：

spark-submit --py-files p.url-0.1.0a4-py2.7.egg,future-0.17.1-py2.7.egg main_dep.py

结果输出：

['https://github.com/search?q=1&name=dog', 'https://github.com/search?q=2&name=dog', 'https://github.com/search?q=3&name=dog', 'https://github.com/search?q=4&name=dog']

补充官方文档，比较蛋疼，没有说具体操作：

Complex Dependencies

Some operations rely on complex packages that also have many dependencies. For example, the following code snippet imports the Python pandas data analysis library:

def import_pandas(x):
 import pandas
 return x

int_rdd = sc.parallelize([1, 2, 3, 4])
int_rdd.map(lambda x: import_pandas(x))
int_rdd.collect()

pandas depends on NumPy, SciPy, and many other packages. Although pandas is too complex to distribute as a *.py file, you can create an egg for it and its dependencies and send that to executors.

Limitations of Distributing Egg Files

In both self-contained and complex dependency scenarios, sending egg files is problematic because packages that contain native code must be compiled for the specific host on which it will run. When doing distributed computing with industry-standard hardware, you must assume is that the hardware is heterogeneous. However, because of the required C compilation, a Python egg built on a client host is specific to the client CPU architecture. Therefore, distributing an egg for complex, compiled packages like NumPy, SciPy, and pandas often fails. Instead of distributing egg files you should install the required Python packages on each host of the cluster and specify the path to the Python binaries for the worker hosts to use.