Bump required Py4J version and add test for large broadcast variables.

pyspark/README

@@ -36,7 +36,7 @@ examples.
 PySpark requires a development version of Py4J, a Python library for
 interacting with Java processes.  It can be installed from
 https://github.com/bartdag/py4j; make sure to install a version that
-contains at least the commits through 3dbf380d3d.
+contains at least the commits through b7924aabe9.
 
 PySpark uses the `PYTHONPATH` environment variable to search for Python
 classes; Py4J should be on this path, along with any libraries used by

pyspark/pyspark/broadcast.py

@@ -13,6 +13,8 @@
 
 >>> sc.parallelize([0, 0]).flatMap(lambda x: b.value).collect()
 [1, 2, 3, 4, 5, 1, 2, 3, 4, 5]
+
+>>> large_broadcast = sc.broadcast(list(range(10000)))
 """
 # Holds broadcasted data received from Java, keyed by its id.
 _broadcastRegistry = {}

pyspark/requirements.txt

@@ -3,4 +3,4 @@
 # package is not at the root of the git repository.  It may be possible to
 # install Py4J from git once https://github.com/pypa/pip/pull/526 is merged.
 
-# git+git://github.com/bartdag/py4j.git@3dbf380d3d2cdeb9aab394454ea74d80c4aba1ea
+# git+git://github.com/bartdag/py4j.git@b7924aabe9c5e63f0a4d8bbd17019534c7ec014e