AttributeError: 'NoneType' object has no attribute 'setCallSite'

Question

In PySpark, I want to calculate the correlation between two dataframe vectors, using the following code (I do not have any problem in importing pyspark or createDataFrame):

from pyspark.ml.linalg import Vectors
from pyspark.ml.stat import Correlation
import pyspark

spark = pyspark.sql.SparkSession.builder.master("local[*]").getOrCreate()

data = [(Vectors.sparse(4, [(0, 1.0), (3, -2.0)]),),
        (Vectors.dense([4.0, 5.0, 0.0, 3.0]),)]
df = spark.createDataFrame(data, ["features"])

r1 = Correlation.corr(df, "features").head()
print("Pearson correlation matrix:\n" + str(r1[0]))

But, I got the AttributeError (AttributeError: 'NoneType' object has no attribute 'setCallSite') as:

AttributeError                            Traceback (most recent call last)
<ipython-input-136-d553c1ade793> in <module>()
      6 df = spark.createDataFrame(data, ["features"])
      7 
----> 8 r1 = Correlation.corr(df, "features").head()
      9 print("Pearson correlation matrix:\n" + str(r1[0]))

/usr/local/lib/python3.6/dist-packages/pyspark/sql/dataframe.py in head(self, n)
   1130         """
   1131         if n is None:
-> 1132             rs = self.head(1)
   1133             return rs[0] if rs else None
   1134         return self.take(n)

/usr/local/lib/python3.6/dist-packages/pyspark/sql/dataframe.py in head(self, n)
   1132             rs = self.head(1)
   1133             return rs[0] if rs else None
-> 1134         return self.take(n)
   1135 
   1136     @ignore_unicode_prefix

/usr/local/lib/python3.6/dist-packages/pyspark/sql/dataframe.py in take(self, num)
    502         [Row(age=2, name=u'Alice'), Row(age=5, name=u'Bob')]
    503         """
--> 504         return self.limit(num).collect()
    505 
    506     @since(1.3)

/usr/local/lib/python3.6/dist-packages/pyspark/sql/dataframe.py in collect(self)
    463         [Row(age=2, name=u'Alice'), Row(age=5, name=u'Bob')]
    464         """
--> 465         with SCCallSiteSync(self._sc) as css:
    466             port = self._jdf.collectToPython()
    467         return list(_load_from_socket(port, BatchedSerializer(PickleSerializer())))

/usr/local/lib/python3.6/dist-packages/pyspark/traceback_utils.py in __enter__(self)
     70     def __enter__(self):
     71         if SCCallSiteSync._spark_stack_depth == 0:
---> 72             self._context._jsc.setCallSite(self._call_site)
     73         SCCallSiteSync._spark_stack_depth += 1
     74 

AttributeError: 'NoneType' object has no attribute 'setCallSite'

Any solution?

Answer 1

I got the same error not only with Correlation.corr(...) dataframe, but with ldaModel.describeTopics() as well.

Most probably it is the SPARK bug.

They forget to initialise DataFrame::_sc._jsc member when created resulting dataframe.

Each dataframe has normally this member initialised with proper JavaObject.

Answer 2

There's an open issue around this:

https://issues.apache.org/jira/browse/SPARK-27335?jql=text%20~%20%22setcallsite%22

The poster suggests forcing to sync your DF's backend with your Spark context:

df.sql_ctx.sparkSession._jsparkSession = spark._jsparkSession
df._sc = spark._sc

This worked for us, hopefully can work in other cases as well.