diff options
Diffstat (limited to 'python')
-rw-r--r-- | python/pyspark/rdd.py | 13 | ||||
-rw-r--r-- | python/pyspark/tests.py | 10 |
2 files changed, 20 insertions, 3 deletions
diff --git a/python/pyspark/rdd.py b/python/pyspark/rdd.py index ed81eb16df..0e2ae19ca3 100644 --- a/python/pyspark/rdd.py +++ b/python/pyspark/rdd.py @@ -2017,8 +2017,7 @@ class RDD(object): >>> len(rdd.repartition(10).glom().collect()) 10 """ - jrdd = self._jrdd.repartition(numPartitions) - return RDD(jrdd, self.ctx, self._jrdd_deserializer) + return self.coalesce(numPartitions, shuffle=True) def coalesce(self, numPartitions, shuffle=False): """ @@ -2029,7 +2028,15 @@ class RDD(object): >>> sc.parallelize([1, 2, 3, 4, 5], 3).coalesce(1).glom().collect() [[1, 2, 3, 4, 5]] """ - jrdd = self._jrdd.coalesce(numPartitions, shuffle) + if shuffle: + # In Scala's repartition code, we will distribute elements evenly across output + # partitions. However, the RDD from Python is serialized as a single binary data, + # so the distribution fails and produces highly skewed partitions. We need to + # convert it to a RDD of java object before repartitioning. + data_java_rdd = self._to_java_object_rdd().coalesce(numPartitions, shuffle) + jrdd = self.ctx._jvm.SerDeUtil.javaToPython(data_java_rdd) + else: + jrdd = self._jrdd.coalesce(numPartitions, shuffle) return RDD(jrdd, self.ctx, self._jrdd_deserializer) def zip(self, other): diff --git a/python/pyspark/tests.py b/python/pyspark/tests.py index b0756911bf..3e0bd16d85 100644 --- a/python/pyspark/tests.py +++ b/python/pyspark/tests.py @@ -914,6 +914,16 @@ class RDDTests(ReusedPySparkTestCase): self.assertEqual(partitions[0], [(0, 5), (0, 8), (2, 6)]) self.assertEqual(partitions[1], [(1, 3), (3, 8), (3, 8)]) + def test_repartition_no_skewed(self): + num_partitions = 20 + a = self.sc.parallelize(range(int(1000)), 2) + l = a.repartition(num_partitions).glom().map(len).collect() + zeros = len([x for x in l if x == 0]) + self.assertTrue(zeros == 0) + l = a.coalesce(num_partitions, True).glom().map(len).collect() + zeros = len([x for x in l if x == 0]) + self.assertTrue(zeros == 0) + def test_distinct(self): rdd = self.sc.parallelize((1, 2, 3)*10, 10) self.assertEqual(rdd.getNumPartitions(), 10) |