workaround spark context config issue

Author: HyukjinKwon

python/pyspark/sql/tests.py

@@ -80,7 +80,7 @@
 _have_pyarrow = _pyarrow_requirement_message is None
 _test_compiled = _test_not_compiled_message is None
 
-from pyspark import SparkContext
+from pyspark import SparkContext, SparkConf
 from pyspark.sql import SparkSession, SQLContext, HiveContext, Column, Row
 from pyspark.sql.types import *
 from pyspark.sql.types import UserDefinedType, _infer_type, _make_type_verifier
@@ -4559,11 +4559,8 @@ class MaxResultArrowTests(unittest.TestCase):
 
     @classmethod
     def setUpClass(cls):
-        cls.spark = SparkSession.builder \
-            .master("local[4]") \
-            .appName(cls.__name__) \
-            .config("spark.driver.maxResultSize", "10k") \
-            .getOrCreate()
+        cls.spark = SparkSession(SparkContext(
+            'local[4]', cls.__name__, conf=SparkConf().set("spark.driver.maxResultSize", "10k")))
 
         # Explicitly enable Arrow and disable fallback.
         cls.spark.conf.set("spark.sql.execution.arrow.enabled", "true")