diff options
| author | Milan Straka <fox@ucw.cz> | 2015-04-10 13:50:32 -0700 |
|---|---|---|
| committer | Josh Rosen <joshrosen@databricks.com> | 2015-04-10 13:50:32 -0700 |
| commit | 0375134f42197f2e29aa865a513cda381f0a1445 (patch) | |
| tree | ce4bfd1a4f9168b11bb183c6361503f9e9eafac6 /python | |
| parent | b9baa4cd9f6e9fc58161f79744b6b7729894d920 (diff) | |
| download | spark-0375134f42197f2e29aa865a513cda381f0a1445.tar.gz spark-0375134f42197f2e29aa865a513cda381f0a1445.tar.bz2 spark-0375134f42197f2e29aa865a513cda381f0a1445.zip | |
[SPARK-5969][PySpark] Fix descending pyspark.rdd.sortByKey.
The samples should always be sorted in ascending order, because bisect.bisect_left is used on it. The reverse order of the result is already achieved in rangePartitioner by reversing the found index.
The current implementation also work, but always uses only two partitions -- the first one and the last one (because the bisect_left return returns either "beginning" or "end" for a descending sequence).
Author: Milan Straka <fox@ucw.cz>
This patch had conflicts when merged, resolved by
Committer: Josh Rosen <joshrosen@databricks.com>
Closes #4761 from foxik/fix-descending-sort and squashes the following commits:
95896b5 [Milan Straka] Add regression test for SPARK-5969.
5757490 [Milan Straka] Fix descending pyspark.rdd.sortByKey.
Diffstat (limited to 'python')
| -rw-r--r-- | python/pyspark/rdd.py | 2 | ||||
| -rw-r--r-- | python/pyspark/tests.py | 11 |
2 files changed, 12 insertions, 1 deletions
diff --git a/python/pyspark/rdd.py b/python/pyspark/rdd.py index 1b18789040..c8e54ed5c6 100644 --- a/python/pyspark/rdd.py +++ b/python/pyspark/rdd.py @@ -595,7 +595,7 @@ class RDD(object): maxSampleSize = numPartitions * 20.0 # constant from Spark's RangePartitioner fraction = min(maxSampleSize / max(rddSize, 1), 1.0) samples = self.sample(False, fraction, 1).map(lambda (k, v): k).collect() - samples = sorted(samples, reverse=(not ascending), key=keyfunc) + samples = sorted(samples, key=keyfunc) # we have numPartitions many parts but one of the them has # an implicit boundary diff --git a/python/pyspark/tests.py b/python/pyspark/tests.py index 0bd5d20f78..0e3721b55a 100644 --- a/python/pyspark/tests.py +++ b/python/pyspark/tests.py @@ -829,6 +829,17 @@ class RDDTests(ReusedPySparkTestCase): rdd = self.sc.parallelize(range(1 << 20)).map(lambda x: str(x)) rdd._jrdd.first() + def test_sortByKey_uses_all_partitions_not_only_first_and_last(self): + # Regression test for SPARK-5969 + seq = [(i * 59 % 101, i) for i in range(101)] # unsorted sequence + rdd = self.sc.parallelize(seq) + for ascending in [True, False]: + sort = rdd.sortByKey(ascending=ascending, numPartitions=5) + self.assertEqual(sort.collect(), sorted(seq, reverse=not ascending)) + sizes = sort.glom().map(len).collect() + for size in sizes: + self.assertGreater(size, 0) + class ProfilerTests(PySparkTestCase): |