allisonwang-db commented on code in PR #42420:
URL: https://github.com/apache/spark/pull/42420#discussion_r1297730442
##########
python/pyspark/worker.py:
##########
@@ -573,9 +574,73 @@ def read_udtf(pickleSer, infile, eval_type):
f"The return type of a UDTF must be a struct type, but got
{type(return_type)}."
)
+ class UDTFWithPartitions:
+ """
+ This implements the logic of a UDTF that accepts an input TABLE
argument with one or more
+ PARTITION BY expressions.
Review Comment:
It would be extremely helpful if we can add an example here. Maybe something
like
-- t: [c1, c2]
SELECT * FROM my_udtf(TABLE (t) PARTITION BY .. ORDER BY ..)
partition_child_indexes in this case will be ...
##########
python/pyspark/worker.py:
##########
@@ -573,9 +574,73 @@ def read_udtf(pickleSer, infile, eval_type):
f"The return type of a UDTF must be a struct type, but got
{type(return_type)}."
)
+ class UDTFWithPartitions:
+ """
+ This implements the logic of a UDTF that accepts an input TABLE
argument with one or more
+ PARTITION BY expressions.
+
+ Parameters
+ ----------
+ create_udtf: function
+ Function to create a new instance of the UDTF to be invoked.
+ partition_child_indexes: list
+ List of integers identifying zero-based indexes of the columns of
the input table that
+ contain projected partitioning expressions. This class will
inspect these values for
+ each pair of consecutive input rows. When they change, this
indicates the boundary
+ between two partitions, and we will invoke the 'terminate' method
on the UDTF class
+ instance and then destroy it and create a new one to implement the
desired partitioning
+ semantics.
+ """
+ def __init__(self, create_udtf, partition_child_indexes):
Review Comment:
nit: can we add type hint here?
##########
python/pyspark/worker.py:
##########
@@ -573,9 +574,73 @@ def read_udtf(pickleSer, infile, eval_type):
f"The return type of a UDTF must be a struct type, but got
{type(return_type)}."
)
+ class UDTFWithPartitions:
+ """
+ This implements the logic of a UDTF that accepts an input TABLE
argument with one or more
+ PARTITION BY expressions.
+
+ Parameters
+ ----------
+ create_udtf: function
+ Function to create a new instance of the UDTF to be invoked.
+ partition_child_indexes: list
+ List of integers identifying zero-based indexes of the columns of
the input table that
+ contain projected partitioning expressions. This class will
inspect these values for
+ each pair of consecutive input rows. When they change, this
indicates the boundary
+ between two partitions, and we will invoke the 'terminate' method
on the UDTF class
+ instance and then destroy it and create a new one to implement the
desired partitioning
+ semantics.
+ """
+ def __init__(self, create_udtf, partition_child_indexes):
+ self._create_udtf = create_udtf
+ self._udtf = create_udtf()
+ self._prev_arguments = None
+ self._partition_child_indexes = partition_child_indexes
+
+ def eval(self, *args, **kwargs):
+ changed_partitions = self._check_partition_boundaries(
+ list(args) + list(kwargs.values()))
+ if changed_partitions:
+ if self._udtf.terminate is not None:
+ result = self._udtf.terminate()
+ if result is not None:
+ for row in result:
+ yield row
+ self._udtf = self._create_udtf()
+ if self._udtf.eval is not None:
+ result = self._udtf.eval(*args, **kwargs)
+ if result is not None:
+ for row in result:
+ yield row
+
+ def terminate(self):
+ if self._udtf.terminate is not None:
+ return self._udtf.terminate()
+
+ def _check_partition_boundaries(self, arguments):
+ result = False
Review Comment:
nit: can we add type hinting for the function input and output, and
docstring comments? And also for _get_table_arg?
##########
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala:
##########
@@ -2118,11 +2121,32 @@ class Analyzer(override val catalogManager:
CatalogManager) extends RuleExecutor
tableArgs.size)
}
val alias =
SubqueryAlias.generateSubqueryName(s"_${tableArgs.size}")
+ // Propagate the column indexes for TABLE arguments to the
PythonUDTF instance.
+ val tvfWithTableColumnIndexes: LogicalPlan = tvf match {
+ case g @ Generate(p: PythonUDTF, _, _, _, _, _) =>
+ functionTableSubqueryArgs.headOption.map { tableArg =>
+ val indexes = PythonUDTFPartitionColumnIndexes(
+ tableArg.partitioningExpressionIndexes)
+ g.copy(generator = p.copy(pythonUDTFPartitionColumnIndexes
= Some(indexes)))
+ }.getOrElse {
+ g
+ }
+ case g @ Generate(p: UnresolvedPolymorphicPythonUDTF, _, _, _,
_, _) =>
+ functionTableSubqueryArgs.headOption.map { tableArg =>
+ val indexes = PythonUDTFPartitionColumnIndexes(
+ tableArg.partitioningExpressionIndexes)
+ g.copy(generator = p.copy(pythonUDTFPartitionColumnIndexes
= Some(indexes)))
+ }.getOrElse {
+ g
Review Comment:
We can combine this with the one above:
```
p : (PythonUDTF | UnresolvedPolymorphicPythonUDTF)
```
##########
sql/core/src/main/scala/org/apache/spark/sql/execution/python/BatchEvalPythonUDTFExec.scala:
##########
@@ -134,6 +134,13 @@ object PythonUDTFRunner {
dataOut.writeBoolean(false)
}
}
+ udtf.pythonUDTFPartitionColumnIndexes match {
+ case Some(partitionColumnIndexes) =>
+ dataOut.writeInt(partitionColumnIndexes.partitionChildIndexes.length)
+ partitionColumnIndexes.partitionChildIndexes.foreach(dataOut.writeInt)
Review Comment:
can we assert partitionColumnIndexes.partitionChildIndexes.length > 0 here?
##########
python/pyspark/worker.py:
##########
@@ -573,9 +574,73 @@ def read_udtf(pickleSer, infile, eval_type):
f"The return type of a UDTF must be a struct type, but got
{type(return_type)}."
)
+ class UDTFWithPartitions:
+ """
+ This implements the logic of a UDTF that accepts an input TABLE
argument with one or more
+ PARTITION BY expressions.
+
+ Parameters
+ ----------
+ create_udtf: function
+ Function to create a new instance of the UDTF to be invoked.
+ partition_child_indexes: list
+ List of integers identifying zero-based indexes of the columns of
the input table that
+ contain projected partitioning expressions. This class will
inspect these values for
+ each pair of consecutive input rows. When they change, this
indicates the boundary
+ between two partitions, and we will invoke the 'terminate' method
on the UDTF class
+ instance and then destroy it and create a new one to implement the
desired partitioning
+ semantics.
+ """
+ def __init__(self, create_udtf, partition_child_indexes):
+ self._create_udtf = create_udtf
+ self._udtf = create_udtf()
+ self._prev_arguments = None
+ self._partition_child_indexes = partition_child_indexes
+
+ def eval(self, *args, **kwargs):
+ changed_partitions = self._check_partition_boundaries(
+ list(args) + list(kwargs.values()))
+ if changed_partitions:
+ if self._udtf.terminate is not None:
+ result = self._udtf.terminate()
+ if result is not None:
+ for row in result:
+ yield row
Review Comment:
I wonder if this will cause issues when we join the output rows from the
UDTF with the input rows. Here is how we join the rows:
```
input_row_1 -> [eval_output_row_1, eval_output_row_2, .. ]
input_row_2 - >[eval_output_row_1, ..]
(now assume we consumed all input rows in a partition, input_row_2 is the
last one)
(we use the last row in a partition to join with the rows generated by
terminate)
input_row_2 -> [terminate_output_row_1, terminate_output_row_2, ...]
```
If we directly yield rows from `terminate` here, will they be joined with
the input row from the next partition?
##########
python/pyspark/worker.py:
##########
@@ -573,9 +574,73 @@ def read_udtf(pickleSer, infile, eval_type):
f"The return type of a UDTF must be a struct type, but got
{type(return_type)}."
)
+ class UDTFWithPartitions:
+ """
+ This implements the logic of a UDTF that accepts an input TABLE
argument with one or more
+ PARTITION BY expressions.
+
+ Parameters
+ ----------
+ create_udtf: function
+ Function to create a new instance of the UDTF to be invoked.
+ partition_child_indexes: list
+ List of integers identifying zero-based indexes of the columns of
the input table that
+ contain projected partitioning expressions. This class will
inspect these values for
+ each pair of consecutive input rows. When they change, this
indicates the boundary
+ between two partitions, and we will invoke the 'terminate' method
on the UDTF class
+ instance and then destroy it and create a new one to implement the
desired partitioning
+ semantics.
Review Comment:
I think these should be under `__init__`?
```
def __init__(self, ..):
"""
docstring here
"""
```
##########
python/pyspark/worker.py:
##########
@@ -573,9 +574,73 @@ def read_udtf(pickleSer, infile, eval_type):
f"The return type of a UDTF must be a struct type, but got
{type(return_type)}."
)
+ class UDTFWithPartitions:
+ """
+ This implements the logic of a UDTF that accepts an input TABLE
argument with one or more
+ PARTITION BY expressions.
+
+ Parameters
+ ----------
+ create_udtf: function
+ Function to create a new instance of the UDTF to be invoked.
+ partition_child_indexes: list
+ List of integers identifying zero-based indexes of the columns of
the input table that
+ contain projected partitioning expressions. This class will
inspect these values for
+ each pair of consecutive input rows. When they change, this
indicates the boundary
+ between two partitions, and we will invoke the 'terminate' method
on the UDTF class
+ instance and then destroy it and create a new one to implement the
desired partitioning
+ semantics.
+ """
+ def __init__(self, create_udtf, partition_child_indexes):
+ self._create_udtf = create_udtf
+ self._udtf = create_udtf()
+ self._prev_arguments = None
+ self._partition_child_indexes = partition_child_indexes
+
+ def eval(self, *args, **kwargs):
+ changed_partitions = self._check_partition_boundaries(
+ list(args) + list(kwargs.values()))
+ if changed_partitions:
+ if self._udtf.terminate is not None:
+ result = self._udtf.terminate()
+ if result is not None:
+ for row in result:
+ yield row
+ self._udtf = self._create_udtf()
Review Comment:
note: we should probably add some metrics here on how many times this udtf
instance is created.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
