[GitHub] [arrow] rtpsw commented on a diff in pull request #34311: GH-32884: [C++] Add ordered aggregation

["rtpsw (via GitHub)" <gi...@apache.org>]

rtpsw commented on code in PR #34311:
URL: https://github.com/apache/arrow/pull/34311#discussion_r1122260682


##########
cpp/src/arrow/compute/exec/aggregate_node.cc:
##########
@@ -169,35 +186,117 @@ void AggregatesToString(std::stringstream* ss, const Schema& input_schema,
   *ss << ']';
 }
 
+template <typename BatchHandler>
+Status HandleSegments(std::unique_ptr<GroupingSegmenter>& segmenter,
+                      const ExecBatch& batch, const std::vector<int>& ids,
+                      const BatchHandler& handle_batch) {
+  int64_t offset = 0;
+  ARROW_ASSIGN_OR_RAISE(auto segment_exec_batch, batch.SelectValues(ids));
+  ExecSpan segment_batch(segment_exec_batch);
+  while (true) {
+    ARROW_ASSIGN_OR_RAISE(auto segment, segmenter->GetNextSegment(segment_batch, offset));
+    if (segment.offset >= segment_batch.length) break;  // condition of no-next-segment
+    ARROW_RETURN_NOT_OK(handle_batch(batch, segment));
+    offset = segment.offset + segment.length;
+  }
+  return Status::OK();
+}
+
+Status GetScalarFields(std::vector<Datum>* values_ptr, const ExecBatch& input_batch,
+                       const std::vector<int>& field_ids) {
+  DCHECK_GT(input_batch.length, 0);
+  std::vector<Datum>& values = *values_ptr;
+  int64_t row = input_batch.length - 1;
+  values.clear();
+  values.resize(field_ids.size());
+  for (size_t i = 0; i < field_ids.size(); i++) {
+    const Datum& value = input_batch.values[field_ids[i]];
+    if (value.is_scalar()) {
+      values[i] = value;
+    } else if (value.is_array()) {
+      ARROW_ASSIGN_OR_RAISE(auto scalar, value.make_array()->GetScalar(row));
+      values[i] = scalar;
+    } else {
+      DCHECK(false);
+    }
+  }
+  return Status::OK();
+}
+
+void PlaceFields(ExecBatch& batch, size_t base, std::vector<Datum>& values) {
+  DCHECK_LE(base + values.size(), batch.values.size());
+  for (size_t i = 0; i < values.size(); i++) {
+    batch.values[base + i] = values[i];
+  }
+}
+
 class ScalarAggregateNode : public ExecNode, public TracedNode {
  public:
   ScalarAggregateNode(ExecPlan* plan, std::vector<ExecNode*> inputs,
                       std::shared_ptr<Schema> output_schema,
+                      std::unique_ptr<GroupingSegmenter> segmenter,
+                      std::vector<int> segment_field_ids,
                       std::vector<std::vector<int>> target_fieldsets,
                       std::vector<Aggregate> aggs,
                       std::vector<const ScalarAggregateKernel*> kernels,
                       std::vector<std::vector<std::unique_ptr<KernelState>>> states)
       : ExecNode(plan, std::move(inputs), {"target"},
                  /*output_schema=*/std::move(output_schema)),
         TracedNode(this),
+        segmenter_(std::move(segmenter)),
+        segment_field_ids_(std::move(segment_field_ids)),
         target_fieldsets_(std::move(target_fieldsets)),
         aggs_(std::move(aggs)),
         kernels_(std::move(kernels)),
-        states_(std::move(states)) {}
+        states_(std::move(states)) {
+    const auto& input_schema = *this->inputs()[0]->output_schema();
+    for (size_t i = 0; i < kernels_.size(); ++i) {
+      std::vector<TypeHolder> in_types;
+      for (const auto& target : target_fieldsets_[i]) {
+        in_types.emplace_back(input_schema.field(target)->type().get());
+      }
+      in_typesets_.push_back(std::move(in_types));
+    }
+  }
 
   static Result<ExecNode*> Make(ExecPlan* plan, std::vector<ExecNode*> inputs,
                                 const ExecNodeOptions& options) {
     RETURN_NOT_OK(ValidateExecNodeInputs(plan, inputs, 1, "ScalarAggregateNode"));
 
     const auto& aggregate_options = checked_cast<const AggregateNodeOptions&>(options);
     auto aggregates = aggregate_options.aggregates;
+    const auto& keys = aggregate_options.keys;
+    const auto& segment_keys = aggregate_options.segment_keys;
+
+    if (keys.size() > 0) {
+      return Status::Invalid("Scalar aggregation with some key");
+    }
+    if (plan->query_context()->exec_context()->executor()->GetCapacity() > 1 &&
+        segment_keys.size() > 0) {
+      return Status::NotImplemented("Segmented aggregation in a multi-threaded plan");
+    }
 
     const auto& input_schema = *inputs[0]->output_schema();
     auto exec_ctx = plan->query_context()->exec_context();
 
+    std::vector<int> segment_field_ids(segment_keys.size());
+    std::vector<TypeHolder> segment_key_types(segment_keys.size());
+    for (size_t i = 0; i < segment_keys.size(); i++) {
+      ARROW_ASSIGN_OR_RAISE(auto match, segment_keys[i].FindOne(input_schema));
+      if (match.indices().size() > 1) {
+        // ARROW-18369: Support nested references as segment ids

Review Comment:
   Nested references is a reference to a nested struct. It refers to `FieldRef` with a multi-component path, e.g. `[0, 2, 1]` selects the index-0 member, within it the index-2 field, and within it the index-1 field. The ARROW issue mentioned in the comment leads to [this post](https://github.com/apache/arrow/pull/14352#discussion_r1026945315).



-- 
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: github-unsubscribe@arrow.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org