Avoid initialize SparkSession but only JVM

Author: HyukjinKwon

sql/create-docs.sh

@@ -42,15 +42,11 @@ fi
 
 # Now create markdown file
 rm -fr docs
-rm -rf "$WAREHOUSE_DIR"
 mkdir docs
 echo "Generating markdown files for SQL documentation."
-"$SPARK_HOME/bin/spark-submit" \
-  --driver-java-options "-Dlog4j.configuration=file:$FWDIR/log4j.properties" \
-  --conf spark.sql.warehouse.dir="$WAREHOUSE_DIR" \
-  gen-sql-markdown.py
-rm -rf "$WAREHOUSE_DIR"
+"$SPARK_HOME/bin/spark-submit" gen-sql-markdown.py
 
 # Now create HTML files
 echo "Generating HTML files for SQL documentation."
 mkdocs build --clean
+rm -fr docs

sql/gen-sql-markdown.py

@@ -19,19 +19,16 @@
 import os
 from collections import namedtuple
 
-from pyspark.sql import SparkSession
-
 ExpressionInfo = namedtuple("ExpressionInfo", "className usage name extended")
 
 
-def _list_function_infos(spark):
+def _list_function_infos(jvm):
     """
     Returns a list of function information via JVM. Sorts wrapped expression infos by name
     and returns them.
     """
 
-    jinfos = spark.sparkContext._jvm \
-        .org.apache.spark.sql.api.python.PythonSQLUtils.listBuiltinFunctionInfos()
+    jinfos = jvm.org.apache.spark.sql.api.python.PythonSQLUtils.listBuiltinFunctionInfos()
     infos = []
     for jinfo in jinfos:
         name = jinfo.getName()
@@ -69,14 +66,14 @@ def _make_pretty_extended(extended):
         return "```%s```\n\n" % extended
 
 
-def generate_sql_markdown(spark, path):
+def generate_sql_markdown(jvm, path):
     """
     Generates a markdown file after listing the function information. The output file
     is created in `path`.
     """
 
     with open(path, 'w') as mdfile:
-        for info in _list_function_infos(spark):
+        for info in _list_function_infos(jvm):
             mdfile.write("### %s\n\n" % info.name)
             usage = _make_pretty_usage(info.usage)
             extended = _make_pretty_extended(info.extended)
@@ -87,10 +84,8 @@ def generate_sql_markdown(spark, path):
 
 
 if __name__ == "__main__":
-    spark = SparkSession \
-        .builder \
-        .appName("GenSQLDocs") \
-        .getOrCreate()
+    from pyspark.java_gateway import launch_gateway
+
+    jvm = launch_gateway().jvm
     markdown_file_path = "%s/docs/index.md" % os.path.dirname(sys.argv[0])
-    generate_sql_markdown(spark, markdown_file_path)
-    spark.stop()
+    generate_sql_markdown(jvm, markdown_file_path)