Anaconda pyspark——错误只出现在IPython中，而不会出现在vanila python中

如果我通过在控制台中键入

/usr/bin/pyspark

来启动pyspark，那么下面的示例代码运行时不会出现任何错误。但是，如果我将它与IPython一起使用，则

$IPYTHON_OPTS="notebook" /usr/bin/pyspark # notebook

或者

$IPYTHON=1 /usr/bin/pyspark

然后引发一个异常

代码如下：

from pyspark import SparkContext,SparkConf
from pyspark import SQLContext
from pyspark.sql.types import *
# sc is a SparkContex object created when pyspark is invoked
sqc = SQLContext(sc)

这是错误消息：

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-1-f0bbbc9cdb50> in <module>()
      3 from pyspark.sql.types import *
      4 # sc is a SparkContex object created when pyspark is invoked
----> 5 sqc = SQLContext(sc)

/opt/cloudera/parcels/CDH-5.4.2-1.cdh5.4.2.p0.2/lib/spark/python/pyspark/sql/context.py in __init__(self, sparkContext, sqlContext)
     91         """
     92         self._sc = sparkContext
---> 93         self._jsc = self._sc._jsc
     94         self._jvm = self._sc._jvm
     95         self._scala_SQLContext = sqlContext

AttributeError: 'module' object has no attribute '_jsc'

但是，如果我禁用anaconda发行版并使用系统附带的Python，那么一切都会正常工作

    $ ipython --version
    4.0.0
    $ python --version
    Python 2.7.3
    $ cat /etc/issue
    Debian GNU/Linux 7 \n \l

---

因此，问题出在Anaconda上，但仍然不知道问题出在哪里，不确定具体的错误，因为香草和Anaconda spark应该有相同的问题，但是，您可以检查以下几点：

确保驱动程序和工作程序上安装了相同的python版本。不同版本可能会导致序列化问题

IPYTHON_OPTS通常不推荐使用。相反，我定义了以下环境变量：

# tells pyspark to use notebook
export PYSPARK_DRIVER_PYTHON_OPS="notebook"
# tells pyspark to use the jupyter executable instead of python. In your case you might want this to be ipython instead
export PYSPARK_DRIVER_PYTHON=/opt/anaconda2/bin/jupyter
# tells pyspark where the python executable is on the executors. It MUST be the same version of python (preferably with the same packages if you are using them in a UDF or similar
export PYSPARK_PYTHON=/opt/anaconda2/bin/python

---

当然，我看到您没有将主控添加到命令行中，因此如果您没有更改spark默认值（即没有工作程序），这将在本地运行spark。

另一个软件包也有同样的问题。很烦人。你有没有在某处发布过一个问题？

# tells pyspark to use notebook
export PYSPARK_DRIVER_PYTHON_OPS="notebook"
# tells pyspark to use the jupyter executable instead of python. In your case you might want this to be ipython instead
export PYSPARK_DRIVER_PYTHON=/opt/anaconda2/bin/jupyter
# tells pyspark where the python executable is on the executors. It MUST be the same version of python (preferably with the same packages if you are using them in a UDF or similar
export PYSPARK_PYTHON=/opt/anaconda2/bin/python