Skip to content

Fix the companion function changes step behavior #13

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

Merged
merged 1 commit into from
May 21, 2025
Merged

Fix the companion function changes step behavior #13

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
42 changes: 38 additions & 4 deletions velox/experimental/cudf/exec/CudfHashAggregation.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -392,8 +392,6 @@ std::unique_ptr<cudf_velox::CudfHashAggregation::Aggregator> createAggregator(
uint32_t inputIndex,
VectorPtr constant,
bool isGlobal) {
// Companion function may be count_merge_extract or count_partial or others,
// so use this to map
if (kind.rfind("sum", 0) == 0) {
return std::make_unique<SumAggregator>(
step, inputIndex, constant, isGlobal);
Expand All @@ -414,6 +412,39 @@ std::unique_ptr<cudf_velox::CudfHashAggregation::Aggregator> createAggregator(
}
}

static const std::unordered_map<std::string, core::AggregationNode::Step>
companionStep = {
{"_partial", core::AggregationNode::Step::kPartial},
{"_merge", core::AggregationNode::Step::kIntermediate},
{"_merge_extract", core::AggregationNode::Step::kFinal}};

/// \brief Convert companion function to step for the aggregation function
///
/// Companion functions are functions that are registered in velox along with
/// their main aggregation functions. These are designed to always function
/// with a fixed `step`. This is to allow spark style planNodes where `step` is
/// the property of the aggregation function rather than the planNode.
/// Companion functions allow us to override the planNode's step and use
/// aggregations of different steps in the same planNode
core::AggregationNode::Step getCompanionStep(
std::string const& kind,
core::AggregationNode::Step step) {
for (const auto& [k, v] : companionStep) {
if (folly::StringPiece(kind).endsWith(k)) {
step = v;
Copy link
Author

Choose a reason for hiding this comment

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

break;

break;
}
}
return step;
}

bool hasFinalAggs(
std::vector<core::AggregationNode::Aggregate> const& aggregates) {
return std::any_of(aggregates.begin(), aggregates.end(), [](auto const& agg) {
return folly::StringPiece(agg.call->name()).endsWith("_merge_extract");
});
}

auto toAggregators(
core::AggregationNode const& aggregationNode,
exec::OperatorCtx const& operatorCtx) {
Expand Down Expand Up @@ -453,8 +484,9 @@ auto toAggregators(
auto const kind = aggregate.call->name();
auto const inputIndex = aggInputs[0];
auto const constant = aggConstants.empty() ? nullptr : aggConstants[0];
auto const companionStep = getCompanionStep(kind, step);
aggregators.push_back(
createAggregator(step, kind, inputIndex, constant, isGlobal));
createAggregator(companionStep, kind, inputIndex, constant, isGlobal));
}
return aggregators;
}
Expand Down Expand Up @@ -505,7 +537,9 @@ CudfHashAggregation::CudfHashAggregation(
operatorId,
fmt::format("[{}]", aggregationNode->id())),
aggregationNode_(aggregationNode),
isPartialOutput_(exec::isPartialOutput(aggregationNode->step())),
isPartialOutput_(
exec::isPartialOutput(aggregationNode->step()) &&
!hasFinalAggs(aggregationNode->aggregates())),
isGlobal_(aggregationNode->groupingKeys().empty()),
isDistinct_(!isGlobal_ && aggregationNode->aggregates().empty()),
maxPartialAggregationMemoryUsage_(
Expand Down
27 changes: 27 additions & 0 deletions velox/experimental/cudf/tests/AggregationTest.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -479,6 +479,33 @@ TEST_F(AggregationTest, countPartialFinalGlobal) {
assertQuery(op, "SELECT count(*) FROM tmp");
}

/// Tests the spark scenario of having different types of aggs in the same
/// planNode Specific example being tested is
/// https://github.com/facebookincubator/velox/issues/12830#issuecomment-2783340233
TEST_F(AggregationTest, CompanionAggs) {
std::vector<int64_t> keys0{1, 1, 1, 2, 1, 1, 2, 2};
std::vector<int64_t> keys1{1, 2, 1, 2, 1, 2, 1, 2};
std::vector<int64_t> values{1, 2, 3, 4, 5, 6, 7, 8};
auto rowVector = makeRowVector(
{makeFlatVector<int64_t>(keys0),
makeFlatVector<int64_t>(keys1),
makeFlatVector<int64_t>(values)});

createDuckDbTable({rowVector});

auto op =
PlanBuilder()
.values({rowVector})
.singleAggregation({"c2", "c0"}, {"count_partial(c1)"})
.localPartition({"c2", "c0"})
.singleAggregation({"c0"}, {"count_merge(a0)", "count_partial(c2)"})
.localPartition({"c0"})
.singleAggregation({"c0"}, {"count_merge(a0)", "count_merge(a1)"})
.planNode();
assertQuery(
op, "SELECT c0, count(c1), count(distinct c2) FROM tmp GROUP BY c0");
}

TEST_F(AggregationTest, partialAggregationMemoryLimit) {
auto vectors = {
makeRowVector({makeFlatVector<int32_t>(
Expand Down