Repository: spark Updated Branches: refs/heads/branch-2.3 65600bfdb -> 3b6fc286d
[SPARK-23233][PYTHON] Reset the cache in asNondeterministic to set deterministic properly ## What changes were proposed in this pull request? Reproducer: ```python from pyspark.sql.functions import udf f = udf(lambda x: x) spark.range(1).select(f("id")) # cache JVM UDF instance. f = f.asNondeterministic() spark.range(1).select(f("id"))._jdf.logicalPlan().projectList().head().deterministic() ``` It should return `False` but the current master returns `True`. Seems it's because we cache the JVM UDF instance and then we reuse it even after setting `deterministic` disabled once it's called. ## How was this patch tested? Manually tested. I am not sure if I should add the test with a lot of JVM accesses with the intetnal stuff .. Let me know if anyone feels so. I will add. Author: hyukjinkwon <gurwls...@gmail.com> Closes #20409 from HyukjinKwon/SPARK-23233. (cherry picked from commit 3227d14feb1a65e95a2bf326cff6ac95615cc5ac) Signed-off-by: gatorsmile <gatorsm...@gmail.com> Project: http://git-wip-us.apache.org/repos/asf/spark/repo Commit: http://git-wip-us.apache.org/repos/asf/spark/commit/3b6fc286 Tree: http://git-wip-us.apache.org/repos/asf/spark/tree/3b6fc286 Diff: http://git-wip-us.apache.org/repos/asf/spark/diff/3b6fc286 Branch: refs/heads/branch-2.3 Commit: 3b6fc286d105ae7de737c46e50cf941e6831ab98 Parents: 65600bf Author: hyukjinkwon <gurwls...@gmail.com> Authored: Sat Jan 27 11:26:09 2018 -0800 Committer: gatorsmile <gatorsm...@gmail.com> Committed: Sat Jan 27 11:26:28 2018 -0800 ---------------------------------------------------------------------- python/pyspark/sql/tests.py | 13 +++++++++++++ python/pyspark/sql/udf.py | 3 +++ 2 files changed, 16 insertions(+) ---------------------------------------------------------------------- http://git-wip-us.apache.org/repos/asf/spark/blob/3b6fc286/python/pyspark/sql/tests.py ---------------------------------------------------------------------- diff --git a/python/pyspark/sql/tests.py b/python/pyspark/sql/tests.py index fbb18c4..a8e337d 100644 --- a/python/pyspark/sql/tests.py +++ b/python/pyspark/sql/tests.py @@ -435,6 +435,19 @@ class SQLTests(ReusedSQLTestCase): pydoc.render_doc(random_udf1) pydoc.render_doc(udf(lambda x: x).asNondeterministic) + def test_nondeterministic_udf3(self): + # regression test for SPARK-23233 + from pyspark.sql.functions import udf + f = udf(lambda x: x) + # Here we cache the JVM UDF instance. + self.spark.range(1).select(f("id")) + # This should reset the cache to set the deterministic status correctly. + f = f.asNondeterministic() + # Check the deterministic status of udf. + df = self.spark.range(1).select(f("id")) + deterministic = df._jdf.logicalPlan().projectList().head().deterministic() + self.assertFalse(deterministic) + def test_nondeterministic_udf_in_aggregate(self): from pyspark.sql.functions import udf, sum import random http://git-wip-us.apache.org/repos/asf/spark/blob/3b6fc286/python/pyspark/sql/udf.py ---------------------------------------------------------------------- diff --git a/python/pyspark/sql/udf.py b/python/pyspark/sql/udf.py index 134badb..0795255 100644 --- a/python/pyspark/sql/udf.py +++ b/python/pyspark/sql/udf.py @@ -181,6 +181,9 @@ class UserDefinedFunction(object): .. versionadded:: 2.3 """ + # Here, we explicitly clean the cache to create a JVM UDF instance + # with 'deterministic' updated. See SPARK-23233. + self._judf_placeholder = None self.deterministic = False return self --------------------------------------------------------------------- To unsubscribe, e-mail: commits-unsubscr...@spark.apache.org For additional commands, e-mail: commits-h...@spark.apache.org