Revert "[SPARK-38563][PYTHON] Upgrade to Py4J 0.10.9.4"

bin/pyspark

@@ -50,7 +50,7 @@ export PYSPARK_DRIVER_PYTHON_OPTS
 
 # Add the PySpark classes to the Python path:
 export PYTHONPATH="${SPARK_HOME}/python/:$PYTHONPATH"
-export PYTHONPATH="${SPARK_HOME}/python/lib/py4j-0.10.9.4-src.zip:$PYTHONPATH"
+export PYTHONPATH="${SPARK_HOME}/python/lib/py4j-0.10.9.3-src.zip:$PYTHONPATH"
 
 # Load the PySpark shell.py script when ./pyspark is used interactively:
 export OLD_PYTHONSTARTUP="$PYTHONSTARTUP"

bin/pyspark2.cmd

@@ -30,7 +30,7 @@ if "x%PYSPARK_DRIVER_PYTHON%"=="x" (
 )
 
 set PYTHONPATH=%SPARK_HOME%\python;%PYTHONPATH%
-set PYTHONPATH=%SPARK_HOME%\python\lib\py4j-0.10.9.4-src.zip;%PYTHONPATH%
+set PYTHONPATH=%SPARK_HOME%\python\lib\py4j-0.10.9.3-src.zip;%PYTHONPATH%
 
 set OLD_PYTHONSTARTUP=%PYTHONSTARTUP%
 set PYTHONSTARTUP=%SPARK_HOME%\python\pyspark\shell.py

core/pom.xml

@@ -423,7 +423,7 @@
     <dependency>
       <groupId>net.sf.py4j</groupId>
       <artifactId>py4j</artifactId>
-      <version>0.10.9.4</version>
+      <version>0.10.9.3</version>
     </dependency>
     <dependency>
       <groupId>org.apache.spark</groupId>

core/src/main/scala/org/apache/spark/api/python/PythonUtils.scala

@@ -27,7 +27,7 @@ import org.apache.spark.SparkContext
 import org.apache.spark.api.java.{JavaRDD, JavaSparkContext}
 
 private[spark] object PythonUtils {
-  val PY4J_ZIP_NAME = "py4j-0.10.9.4-src.zip"
+  val PY4J_ZIP_NAME = "py4j-0.10.9.3-src.zip"
 
   /** Get the PYTHONPATH for PySpark, either from SPARK_HOME, if it is set, or from our JAR */
   def sparkPythonPath: String = {

dev/deps/spark-deps-hadoop-2-hive-2.3

@@ -233,7 +233,7 @@ parquet-hadoop/1.12.2//parquet-hadoop-1.12.2.jar
 parquet-jackson/1.12.2//parquet-jackson-1.12.2.jar
 pickle/1.2//pickle-1.2.jar
 protobuf-java/2.5.0//protobuf-java-2.5.0.jar
-py4j/0.10.9.4//py4j-0.10.9.4.jar
+py4j/0.10.9.3//py4j-0.10.9.3.jar
 remotetea-oncrpc/1.1.2//remotetea-oncrpc-1.1.2.jar
 rocksdbjni/6.20.3//rocksdbjni-6.20.3.jar
 scala-collection-compat_2.12/2.1.1//scala-collection-compat_2.12-2.1.1.jar

dev/deps/spark-deps-hadoop-3-hive-2.3

@@ -221,7 +221,7 @@ parquet-hadoop/1.12.2//parquet-hadoop-1.12.2.jar
 parquet-jackson/1.12.2//parquet-jackson-1.12.2.jar
 pickle/1.2//pickle-1.2.jar
 protobuf-java/2.5.0//protobuf-java-2.5.0.jar
-py4j/0.10.9.4//py4j-0.10.9.4.jar
+py4j/0.10.9.3//py4j-0.10.9.3.jar
 remotetea-oncrpc/1.1.2//remotetea-oncrpc-1.1.2.jar
 rocksdbjni/6.20.3//rocksdbjni-6.20.3.jar
 scala-collection-compat_2.12/2.1.1//scala-collection-compat_2.12-2.1.1.jar

docs/job-scheduling.md

@@ -304,5 +304,5 @@ via `sc.setJobGroup` in a separate PVM thread, which also disallows to cancel th
 later.
 
 `pyspark.InheritableThread` is recommended to use together for a PVM thread to inherit the inheritable attributes
- such as local properties in a JVM thread.
+ such as local properties in a JVM thread, and to avoid resource leak.
 

python/docs/Makefile

@@ -21,7 +21,7 @@ SPHINXBUILD   ?= sphinx-build
 SOURCEDIR     ?= source
 BUILDDIR      ?= build
 
-export PYTHONPATH=$(realpath ..):$(realpath ../lib/py4j-0.10.9.4-src.zip)
+export PYTHONPATH=$(realpath ..):$(realpath ../lib/py4j-0.10.9.3-src.zip)
 
 # Put it first so that "make" without argument is like "make help".
 help:

python/docs/make2.bat

@@ -25,7 +25,7 @@ if "%SPHINXBUILD%" == "" (
 set SOURCEDIR=source
 set BUILDDIR=build
 
-set PYTHONPATH=..;..\lib\py4j-0.10.9.4-src.zip
+set PYTHONPATH=..;..\lib\py4j-0.10.9.3-src.zip
 
 if "%1" == "" goto help
 

python/docs/source/getting_started/install.rst

@@ -157,7 +157,7 @@ Package       Minimum supported version Note
 `pandas`      1.0.5                     Optional for Spark SQL
 `NumPy`       1.7                       Required for MLlib DataFrame-based API
 `pyarrow`     1.0.0                     Optional for Spark SQL
-`Py4J`        0.10.9.4                  Required
+`Py4J`        0.10.9.3                  Required
 `pandas`      1.0.5                     Required for pandas API on Spark
 `pyarrow`     1.0.0                     Required for pandas API on Spark
 `Numpy`       1.14                      Required for pandas API on Spark

python/pyspark/context.py

@@ -1365,7 +1365,7 @@ def setJobGroup(self, groupId: str, description: str, interruptOnCancel: bool =
         to HDFS-1208, where HDFS may respond to Thread.interrupt() by marking nodes as dead.
 
         If you run jobs in parallel, use :class:`pyspark.InheritableThread` for thread
-        local inheritance.
+        local inheritance, and preventing resource leak.
 
         Examples
         --------
@@ -1405,7 +1405,7 @@ def setLocalProperty(self, key: str, value: str) -> None:
         Notes
         -----
         If you run jobs in parallel, use :class:`pyspark.InheritableThread` for thread
-        local inheritance.
+        local inheritance, and preventing resource leak.
         """
         self._jsc.setLocalProperty(key, value)
 
@@ -1423,7 +1423,7 @@ def setJobDescription(self, value: str) -> None:
         Notes
         -----
         If you run jobs in parallel, use :class:`pyspark.InheritableThread` for thread
-        local inheritance.
+        local inheritance, and preventing resource leak.
         """
         self._jsc.setJobDescription(value)
 

python/pyspark/util.py

@@ -331,10 +331,13 @@ def inheritable_thread_target(f: Callable) -> Callable:
 
         @functools.wraps(f)
         def wrapped(*args: Any, **kwargs: Any) -> Any:
-            # Set local properties in child thread.
-            assert SparkContext._active_spark_context is not None
-            SparkContext._active_spark_context._jsc.sc().setLocalProperties(properties)
-            return f(*args, **kwargs)
+            try:
+                # Set local properties in child thread.
+                assert SparkContext._active_spark_context is not None
+                SparkContext._active_spark_context._jsc.sc().setLocalProperties(properties)
+                return f(*args, **kwargs)
+            finally:
+                InheritableThread._clean_py4j_conn_for_current_thread()
 
         return wrapped
     else:
@@ -374,7 +377,10 @@ def copy_local_properties(*a: Any, **k: Any) -> Any:
                 assert hasattr(self, "_props")
                 assert SparkContext._active_spark_context is not None
                 SparkContext._active_spark_context._jsc.sc().setLocalProperties(self._props)
-                return target(*a, **k)
+                try:
+                    return target(*a, **k)
+                finally:
+                    InheritableThread._clean_py4j_conn_for_current_thread()
 
             super(InheritableThread, self).__init__(
                 target=copy_local_properties, *args, **kwargs  # type: ignore[misc]
@@ -395,6 +401,25 @@ def start(self) -> None:
             self._props = SparkContext._active_spark_context._jsc.sc().getLocalProperties().clone()
         return super(InheritableThread, self).start()
 
+    @staticmethod
+    def _clean_py4j_conn_for_current_thread() -> None:
+        from pyspark import SparkContext
+
+        jvm = SparkContext._jvm
+        assert jvm is not None
+        thread_connection = jvm._gateway_client.get_thread_connection()
+        if thread_connection is not None:
+            try:
+                # Dequeue is shared across other threads but it's thread-safe.
+                # If this function has to be invoked one more time in the same thead
+                # Py4J will create a new connection automatically.
+                jvm._gateway_client.deque.remove(thread_connection)
+            except ValueError:
+                # Should never reach this point
+                return
+            finally:
+                thread_connection.close()
+
 
 if __name__ == "__main__":
     if "pypy" not in platform.python_implementation().lower() and sys.version_info[:2] >= (3, 7):

python/setup.py

@@ -258,7 +258,7 @@ def run(self):
         license='http://www.apache.org/licenses/LICENSE-2.0',
         # Don't forget to update python/docs/source/getting_started/install.rst
         # if you're updating the versions or dependencies.
-        install_requires=['py4j==0.10.9.4'],
+        install_requires=['py4j==0.10.9.3'],
         extras_require={
             'ml': ['numpy>=1.15'],
             'mllib': ['numpy>=1.15'],

sbin/spark-config.sh

@@ -28,6 +28,6 @@ export SPARK_CONF_DIR="${SPARK_CONF_DIR:-"${SPARK_HOME}/conf"}"
 # Add the PySpark classes to the PYTHONPATH:
 if [ -z "${PYSPARK_PYTHONPATH_SET}" ]; then
   export PYTHONPATH="${SPARK_HOME}/python:${PYTHONPATH}"
-  export PYTHONPATH="${SPARK_HOME}/python/lib/py4j-0.10.9.4-src.zip:${PYTHONPATH}"
+  export PYTHONPATH="${SPARK_HOME}/python/lib/py4j-0.10.9.3-src.zip:${PYTHONPATH}"
   export PYSPARK_PYTHONPATH_SET=1
 fi