Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

GH-41692: [Python] Improve substrait extended expressions support #41693

Merged
merged 10 commits into from
Oct 16, 2024
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion cpp/src/arrow/engine/substrait/serde.cc
Original file line number Diff line number Diff line change
Expand Up @@ -56,7 +56,7 @@ Status ParseFromBufferImpl(const Buffer& buf, const std::string& full_name,
if (message->ParseFromZeroCopyStream(&buf_stream)) {
return Status::OK();
}
return Status::IOError("ParseFromZeroCopyStream failed for ", full_name);
return Status::Invalid("ParseFromZeroCopyStream failed for ", full_name);
Copy link
Member Author

@amol- amol- Jul 3, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It seemed odd that it failed with an IOError given that the ArrayInputStream was built from already in-memory Buffer, so no IO was involved. If it fails, it actually means that the Substrait data is invalid

}

template <typename Message>
Expand Down
3 changes: 3 additions & 0 deletions docs/source/python/api/substrait.rst
Original file line number Diff line number Diff line change
Expand Up @@ -43,6 +43,9 @@ compute expressions.
BoundExpressions
deserialize_expressions
serialize_expressions
serialize_schema
deserialize_schema
SubstraitSchema

Utility
-------
Expand Down
1 change: 1 addition & 0 deletions docs/source/python/integration.rst
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,7 @@ This allows to easily integrate PyArrow with other languages and technologies.
.. toctree::
:maxdepth: 2

integration/substrait
integration/python_r
integration/python_java
integration/extending
Expand Down
249 changes: 249 additions & 0 deletions docs/source/python/integration/substrait.rst
Original file line number Diff line number Diff line change
@@ -0,0 +1,249 @@
.. Licensed to the Apache Software Foundation (ASF) under one
.. or more contributor license agreements. See the NOTICE file
.. distributed with this work for additional information
.. regarding copyright ownership. The ASF licenses this file
.. to you under the Apache License, Version 2.0 (the
.. "License"); you may not use this file except in compliance
.. with the License. You may obtain a copy of the License at

.. http://www.apache.org/licenses/LICENSE-2.0

.. Unless required by applicable law or agreed to in writing,
.. software distributed under the License is distributed on an
.. "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
.. KIND, either express or implied. See the License for the
.. specific language governing permissions and limitations
.. under the License.

=========
Substrait
=========

The ``arrow-substrait`` module implements support for the Substrait_ format,
enabling conversion to and from Arrow objects.

The ``arrow-dataset`` module can execute Substrait_ plans via the
:doc:`Acero <../cpp/streaming_execution>` query engine.

.. contents::

Working with Schemas
====================

Arrow schemas can be encoded and decoded using the :meth:`pyarrow.substrait.serialize_schema` and
:meth:`pyarrow.substrait.deserialize_schema` functions.

.. code-block:: python

import pyarrow as pa
import pyarrow.substrait as pa_substrait

arrow_schema = pa.schema([
pa.field("x", pa.int32()),
pa.field("y", pa.string())
])
substrait_schema = pa_substrait.serialize_schema(arrow_schema)

The schema marshalled as a Substrait ``NamedStruct`` is directly
available as ``substrait_schema.schema``::

>>> print(substrait_schema.schema)
b'\n\x01x\n\x01y\x12\x0c\n\x04*\x02\x10\x01\n\x04b\x02\x10\x01'

In case arrow custom types were used, the schema will require
extensions for those types to be actually usable, for this reason
the schema is also available as an `Extended Expression`_ including
all the extensions types::

>>> print(substrait_schema.expression)
b'"\x14\n\x01x\n\x01y\x12\x0c\n\x04*\x02\x10\x01\n\x04b\x02\x10\x01:\x19\x10,*\x15Acero 17.0.0'

If ``Substrait Python`` is installed, the schema can also be converted to
a ``substrait-python`` object::

>>> print(substrait_schema.to_pysubstrait())
version {
minor_number: 44
producer: "Acero 17.0.0"
}
base_schema {
names: "x"
names: "y"
struct {
types {
i32 {
nullability: NULLABILITY_NULLABLE
}
}
types {
string {
nullability: NULLABILITY_NULLABLE
}
}
}
}

Working with Expressions
========================

Arrow compute expressions can be encoded and decoded using the
:meth:`pyarrow.substrait.serialize_expressions` and
:meth:`pyarrow.substrait.deserialize_expressions` functions.

.. code-block:: python

import pyarrow as pa
import pyarrow.compute as pa
import pyarrow.substrait as pa_substrait

arrow_schema = pa.schema([
pa.field("x", pa.int32()),
pa.field("y", pa.int32())
])

substrait_expr = pa_substrait.serialize_expressions(
exprs=[pc.field("x") + pc.field("y")],
names=["total"],
schema=arrow_schema
)

The result of encoding to substrait an expression will be the
protobuf ``ExtendedExpression`` message data itself::

>>> print(bytes(substrait_expr))
b'\nZ\x12Xhttps://github.com/substrait-io/substrait/blob/main/extensions/functions_arithmetic.yaml\x12\x07\x1a\x05\x1a\x03add\x1a>\n5\x1a3\x1a\x04*\x02\x10\x01"\n\x1a\x08\x12\x06\n\x02\x12\x00"\x00"\x0c\x1a\n\x12\x08\n\x04\x12\x02\x08\x01"\x00*\x11\n\x08overflow\x12\x05ERROR\x1a\x05total"\x14\n\x01x\n\x01y\x12\x0c\n\x04*\x02\x10\x01\n\x04*\x02\x10\x01:\x19\x10,*\x15Acero 17.0.0'

So in case a ``Substrait Python`` object is required, the expression
has to be decoded from ``substrait-python`` itself::

>>> import substrait
>>> pysubstrait_expr = substrait.proto.ExtendedExpression.FromString(substrait_expr)
>>> print(pysubstrait_expr)
version {
minor_number: 44
producer: "Acero 17.0.0"
}
extension_uris {
uri: "https://github.com/substrait-io/substrait/blob/main/extensions/functions_arithmetic.yaml"
}
extensions {
extension_function {
name: "add"
}
}
referred_expr {
expression {
scalar_function {
arguments {
value {
selection {
direct_reference {
struct_field {
}
}
root_reference {
}
}
}
}
arguments {
value {
selection {
direct_reference {
struct_field {
field: 1
}
}
root_reference {
}
}
}
}
options {
name: "overflow"
preference: "ERROR"
}
output_type {
i32 {
nullability: NULLABILITY_NULLABLE
}
}
}
}
output_names: "total"
}
base_schema {
names: "x"
names: "y"
struct {
types {
i32 {
nullability: NULLABILITY_NULLABLE
}
}
types {
i32 {
nullability: NULLABILITY_NULLABLE
}
}
}
}

Executing Queries Using Substrait Extended Expressions
======================================================

Dataset supports executing queries using Substrait's `Extended Expression`_,
the expressions can be passed to the dataset scanner in the form of
:class:`pyarrow.substrait.BoundExpressions`

.. code-block:: python

import pyarrow.dataset as ds
import pyarrow.substrait as pa_substrait

# Use substrait-python to create the queries
from substrait import proto

dataset = ds.dataset("./data/index-0.parquet")
substrait_schema = pa_substrait.serialize_schema(dataset.schema).to_pysubstrait()

# SELECT project_name FROM dataset WHERE project_name = 'pyarrow'

projection = proto.ExtendedExpression(referred_expr=[
{"expression": {"selection": {"direct_reference": {"struct_field": {"field": 0}}}},
"output_names": ["project_name"]}
])
projection.MergeFrom(substrait_schema)

filtering = proto.ExtendedExpression(
extension_uris=[{"extension_uri_anchor": 99, "uri": "/functions_comparison.yaml"}],
extensions=[{"extension_function": {"extension_uri_reference": 99, "function_anchor": 199, "name": "equal:any1_any1"}}],
referred_expr=[
{"expression": {"scalar_function": {"function_reference": 199, "arguments": [
{"value": {"selection": {"direct_reference": {"struct_field": {"field": 0}}}}},
{"value": {"literal": {"string": "pyarrow"}}}
], "output_type": {"bool": {"nullability": False}}}}}
]
)
filtering.MergeFrom(substrait_schema)

results = dataset.scanner(
columns=pa.substrait.BoundExpressions.from_substrait(projection),
filter=pa.substrait.BoundExpressions.from_substrait(filtering)
).head(5)


.. code-block:: text

project_name
0 pyarrow
1 pyarrow
2 pyarrow
3 pyarrow
4 pyarrow


.. _`Substrait`: https://substrait.io/
.. _`Substrait Python`: https://github.com/substrait-io/substrait-python
.. _`Acero`: https://arrow.apache.org/docs/cpp/streaming_execution.html
.. _`Extended Expression`: https://github.com/substrait-io/substrait/blob/main/site/docs/expressions/extended_expression.md
6 changes: 3 additions & 3 deletions python/pyarrow/_compute.pyx
Original file line number Diff line number Diff line change
Expand Up @@ -2441,7 +2441,7 @@ cdef class Expression(_Weakrefable):
)

@staticmethod
def from_substrait(object buffer not None):
def from_substrait(object message not None):
"""
Deserialize an expression from Substrait

Expand All @@ -2453,15 +2453,15 @@ cdef class Expression(_Weakrefable):

Parameters
----------
buffer : bytes or Buffer
message : bytes or Buffer or a protobuf Message
The Substrait message to deserialize

Returns
-------
Expression
The deserialized expression
"""
expressions = _pas().deserialize_expressions(buffer).expressions
expressions = _pas().BoundExpressions.from_substrait(message).expressions
if len(expressions) == 0:
raise ValueError("Substrait message did not contain any expressions")
if len(expressions) > 1:
Expand Down
19 changes: 17 additions & 2 deletions python/pyarrow/_dataset.pyx
Original file line number Diff line number Diff line change
Expand Up @@ -39,6 +39,11 @@ from pyarrow.util import _is_iterable, _is_path_like, _stringify_path
from pyarrow._json cimport ParseOptions as JsonParseOptions
from pyarrow._json cimport ReadOptions as JsonReadOptions

try:
import pyarrow.substrait as pa_substrait
except ImportError:
pa_substrait = None


_DEFAULT_BATCH_SIZE = 2**17
_DEFAULT_BATCH_READAHEAD = 16
Expand Down Expand Up @@ -272,6 +277,13 @@ cdef class Dataset(_Weakrefable):

# at the moment only support filter
requested_filter = options.get("filter")
if pa_substrait and isinstance(requested_filter, pa_substrait.BoundExpressions):
expressions = list(requested_filter.expressions.values())
if len(expressions) != 1:
raise ValueError(
"Only one BoundExpressions with a single expression are supported")
new_options["filter"] = requested_filter = expressions[0]

current_filter = self._scan_options.get("filter")
if requested_filter is not None and current_filter is not None:
new_options["filter"] = current_filter & requested_filter
Expand All @@ -282,7 +294,7 @@ cdef class Dataset(_Weakrefable):

def scanner(self,
object columns=None,
Expression filter=None,
object filter=None,
int batch_size=_DEFAULT_BATCH_SIZE,
int batch_readahead=_DEFAULT_BATCH_READAHEAD,
int fragment_readahead=_DEFAULT_FRAGMENT_READAHEAD,
Expand Down Expand Up @@ -3410,6 +3422,9 @@ cdef void _populate_builder(const shared_ptr[CScannerBuilder]& ptr,
filter, pyarrow_wrap_schema(builder.schema()))))

if columns is not None:
if pa_substrait and isinstance(columns, pa_substrait.BoundExpressions):
columns = columns.expressions

if isinstance(columns, dict):
for expr in columns.values():
if not isinstance(expr, Expression):
Expand Down Expand Up @@ -3490,7 +3505,7 @@ cdef class Scanner(_Weakrefable):
@staticmethod
def from_dataset(Dataset dataset not None, *,
object columns=None,
Expression filter=None,
object filter=None,
int batch_size=_DEFAULT_BATCH_SIZE,
int batch_readahead=_DEFAULT_BATCH_READAHEAD,
int fragment_readahead=_DEFAULT_FRAGMENT_READAHEAD,
Expand Down
Loading
Loading