Skip to content

ARROW-10008: [C++][Dataset] Fix filtering/row group statistics of dict columns #8311

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.

Sign up for GitHub

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

Closed
bkietz wants to merge 2 commits into from
Closed

ARROW-10008: [C++][Dataset] Fix filtering/row group statistics of dict columns #8311

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
2ab5a8f
ARROW-10008: [C++][Dataset] Fix filtering/row group statistics of dic…
bkietz Sep 30, 2020
5f20975
add test
bkietz Oct 5, 2020
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
10 changes: 9 additions & 1 deletion cpp/src/arrow/dataset/file_parquet.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -131,7 +131,6 @@ static Result<SchemaManifest> GetSchemaManifest(

static std::shared_ptr<StructScalar> MakeMinMaxScalar(std::shared_ptr<Scalar> min,
std::shared_ptr<Scalar> max) {
DCHECK(min->type->Equals(max->type));
return std::make_shared<StructScalar>(ScalarVector{min, max},
struct_({
field("min", min->type),
Expand Down Expand Up @@ -171,6 +170,15 @@ static std::shared_ptr<StructScalar> ColumnChunkStatisticsAsStructScalar(
return nullptr;
}

auto maybe_min = min->CastTo(field->type());
auto maybe_max = max->CastTo(field->type());
Copy link
Member

@jorisvandenbossche jorisvandenbossche Oct 2, 2020

Choose a reason for hiding this comment

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

Does this change behaviour? For a dictionary with string values, is field->type() string or dictionary?

Copy link
Member Author

@bkietz bkietz Oct 2, 2020

Choose a reason for hiding this comment

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

StatisticsAsScalars returns scalars whose types are the correct physical type, so even if the column was dictionary(string) min and max would be just string before this cast

Copy link
Member Author

@bkietz bkietz Oct 2, 2020

Choose a reason for hiding this comment

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

(IE, it only changes behavior in cases where the physical type wasn't appropriate)

if (maybe_min.ok() && maybe_max.ok()) {
min = maybe_min.MoveValueUnsafe();
max = maybe_max.MoveValueUnsafe();
} else {
return nullptr;
}

return MakeMinMaxScalar(std::move(min), std::move(max));
}

Expand Down
17 changes: 8 additions & 9 deletions cpp/src/arrow/dataset/file_parquet_test.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -480,21 +480,20 @@ TEST_F(TestParquetFileFormat, PredicatePushdownRowGroupFragments) {
}

TEST_F(TestParquetFileFormat, PredicatePushdownRowGroupFragmentsUsingStringColumn) {
auto table =
TableFromJSON(schema({field("x", utf8())}), {
R"([{"x": "a"}, {"x": "a"}])",
R"([{"x": "b"}, {"x": "b"}])",
R"([{"x": "c"}, {"x": "c"}])",
R"([{"x": "a"}, {"x": "b"}])",
});
auto table = TableFromJSON(schema({field("x", utf8())}),
{
R"([{"x": "a"}])",
R"([{"x": "b"}, {"x": "b"}])",
R"([{"x": "c"}, {"x": "c"}, {"x": "c"}])",
R"([{"x": "a"}, {"x": "b"}, {"x": "c"}, {"x": "d"}])",
});
TableBatchReader reader(*table);
auto source = GetFileSource(&reader);

opts_ = ScanOptions::Make(reader.schema());
ASSERT_OK_AND_ASSIGN(auto fragment, format_->MakeFragment(*source));

// TODO(bkietz): support strings in StatisticsAsScalars
// CountRowGroupsInFragment(fragment, {0, 3}, "x"_ == "a");
CountRowGroupsInFragment(fragment, {0, 3}, "x"_ == "a");
}

TEST_F(TestParquetFileFormat, ExplicitRowGroupSelection) {
Expand Down
44 changes: 36 additions & 8 deletions cpp/src/arrow/dataset/filter.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1179,8 +1179,8 @@ struct TreeEvaluator::Impl {
Result<Datum> kernel(const Datum& left,
const Datum& right,
ExecContext* ctx)) const {
ARROW_ASSIGN_OR_RAISE(auto lhs, Evaluate(*expr.left_operand()));
ARROW_ASSIGN_OR_RAISE(auto rhs, Evaluate(*expr.right_operand()));
ARROW_ASSIGN_OR_RAISE(Datum lhs, Evaluate(*expr.left_operand()));
ARROW_ASSIGN_OR_RAISE(Datum rhs, Evaluate(*expr.right_operand()));

if (lhs.is_scalar()) {
ARROW_ASSIGN_OR_RAISE(
Expand All @@ -1200,7 +1200,7 @@ struct TreeEvaluator::Impl {
}

Result<Datum> operator()(const NotExpression& expr) const {
ARROW_ASSIGN_OR_RAISE(auto to_invert, Evaluate(*expr.operand()));
ARROW_ASSIGN_OR_RAISE(Datum to_invert, Evaluate(*expr.operand()));
if (IsNullDatum(to_invert)) {
return NullDatum();
}
Expand All @@ -1214,7 +1214,7 @@ struct TreeEvaluator::Impl {
}

Result<Datum> operator()(const InExpression& expr) const {
ARROW_ASSIGN_OR_RAISE(auto operand_values, Evaluate(*expr.operand()));
ARROW_ASSIGN_OR_RAISE(Datum operand_values, Evaluate(*expr.operand()));
if (IsNullDatum(operand_values)) {
return Datum(expr.set()->null_count() != 0);
}
Expand All @@ -1224,7 +1224,7 @@ struct TreeEvaluator::Impl {
}

Result<Datum> operator()(const IsValidExpression& expr) const {
ARROW_ASSIGN_OR_RAISE(auto operand_values, Evaluate(*expr.operand()));
ARROW_ASSIGN_OR_RAISE(Datum operand_values, Evaluate(*expr.operand()));
if (IsNullDatum(operand_values)) {
return Datum(false);
}
Expand Down Expand Up @@ -1255,14 +1255,42 @@ struct TreeEvaluator::Impl {
}

Result<Datum> operator()(const ComparisonExpression& expr) const {
ARROW_ASSIGN_OR_RAISE(auto lhs, Evaluate(*expr.left_operand()));
ARROW_ASSIGN_OR_RAISE(auto rhs, Evaluate(*expr.right_operand()));
ARROW_ASSIGN_OR_RAISE(Datum lhs, Evaluate(*expr.left_operand()));
ARROW_ASSIGN_OR_RAISE(Datum rhs, Evaluate(*expr.right_operand()));

if (IsNullDatum(lhs) || IsNullDatum(rhs)) {
return Datum(std::make_shared<BooleanScalar>());
}

DCHECK(lhs.is_array());
if (lhs.type()->id() == Type::DICTIONARY && rhs.type()->id() == Type::DICTIONARY) {
Copy link
Member Author

@bkietz bkietz Sep 30, 2020

Choose a reason for hiding this comment

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

@wesm What do you think about adding kernels to scalar_compare.cc which do this inside compute:: ?

Copy link
Member

@wesm wesm Oct 22, 2020

Choose a reason for hiding this comment

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

Yes, this sounds fine, can you open a JIRA issue about it?

if (lhs.is_array() && rhs.is_array()) {
// decode dictionary arrays
for (Datum* arg : {&lhs, &rhs}) {
auto dict = checked_pointer_cast<DictionaryArray>(arg->make_array());
ARROW_ASSIGN_OR_RAISE(*arg, compute::Take(dict->dictionary(), dict->indices(),
compute::TakeOptions::Defaults()));
}
} else if (lhs.is_array() || rhs.is_array()) {
auto dict = checked_pointer_cast<DictionaryArray>(
(lhs.is_array() ? lhs : rhs).make_array());

ARROW_ASSIGN_OR_RAISE(auto scalar, checked_cast<const DictionaryScalar&>(
*(lhs.is_scalar() ? lhs : rhs).scalar())
.GetEncodedValue());
if (lhs.is_array()) {
lhs = dict->dictionary();
rhs = std::move(scalar);
} else {
lhs = std::move(scalar);
rhs = dict->dictionary();
}
ARROW_ASSIGN_OR_RAISE(
Datum out_dict,
compute::Compare(lhs, rhs, compute::CompareOptions(expr.op()), &ctx_));

return compute::Take(out_dict, dict->indices(), compute::TakeOptions::Defaults());
}
}

return compute::Compare(lhs, rhs, compute::CompareOptions(expr.op()), &ctx_);
}
Expand Down
18 changes: 18 additions & 0 deletions python/pyarrow/tests/test_dataset.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -826,6 +826,24 @@ def test_fragments_parquet_row_groups(tempdir):
assert len(result) == 1


@pytest.mark.pandas
@pytest.mark.parquet
def test_fragments_parquet_row_groups_dictionary(tempdir):
import pandas as pd

df = pd.DataFrame(dict(col1=['a', 'b'], col2=[1, 2]))
df['col1'] = df['col1'].astype("category")

import pyarrow.parquet as pq
pq.write_table(pa.table(df), tempdir / "test_filter_dictionary.parquet")

import pyarrow.dataset as ds
dataset = ds.dataset(tempdir / 'test_filter_dictionary.parquet')
result = dataset.to_table(filter=ds.field("col1") == "a")

assert (df.iloc[0] == result.to_pandas()).all().all()


@pytest.mark.pandas
@pytest.mark.parquet
def test_fragments_parquet_ensure_metadata(tempdir, open_logging_fs):
Expand Down