Simplify dataflow to SQL logic for JoinOverTimeRangeNode

metricflow/dataflow/nodes/join_over_time.py

@@ -1,7 +1,7 @@
 from __future__ import annotations
 
 from dataclasses import dataclass
-from typing import Optional, Sequence
+from typing import Optional, Sequence, Tuple
 
 from dbt_semantic_interfaces.protocols import MetricTimeWindow
 from dbt_semantic_interfaces.type_enums import TimeGranularity
@@ -26,7 +26,7 @@ class JoinOverTimeRangeNode(DataflowPlanNode):
         time_range_constraint: Time range to aggregate over.
     """
 
-    queried_agg_time_dimension_specs: Sequence[TimeDimensionSpec]
+    queried_agg_time_dimension_specs: Tuple[TimeDimensionSpec, ...]
     window: Optional[MetricTimeWindow]
     grain_to_date: Optional[TimeGranularity]
     time_range_constraint: Optional[TimeRangeConstraint]
@@ -38,7 +38,7 @@ def __post_init__(self) -> None:  # noqa: D105
     @staticmethod
     def create(  # noqa: D102
         parent_node: DataflowPlanNode,
-        queried_agg_time_dimension_specs: Sequence[TimeDimensionSpec],
+        queried_agg_time_dimension_specs: Tuple[TimeDimensionSpec, ...],
         window: Optional[MetricTimeWindow] = None,
         grain_to_date: Optional[TimeGranularity] = None,
         time_range_constraint: Optional[TimeRangeConstraint] = None,

metricflow/dataset/sql_dataset.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+from dataclasses import dataclass
 from typing import List, Optional, Sequence, Tuple
 
 from dbt_semantic_interfaces.references import SemanticModelReference
@@ -160,3 +161,25 @@ def column_association_for_time_dimension(self, time_dimension_spec: TimeDimensi
     @override
     def semantic_model_reference(self) -> Optional[SemanticModelReference]:
         return None
+
+    def annotate(self, alias: str, metric_time_spec: TimeDimensionSpec) -> AnnotatedSqlDataSet:
+        """Convert to an AnnotatedSqlDataSet with specified metadata."""
+        metric_time_column_name = self.column_association_for_time_dimension(metric_time_spec).column_name
+        return AnnotatedSqlDataSet(data_set=self, alias=alias, _metric_time_column_name=metric_time_column_name)
+
+
+@dataclass(frozen=True)
+class AnnotatedSqlDataSet:
+    """Class to bind a DataSet to transient properties associated with it at a given point in the SqlQueryPlan."""
+
+    data_set: SqlDataSet
+    alias: str
+    _metric_time_column_name: Optional[str] = None
+
+    @property
+    def metric_time_column_name(self) -> str:
+        """Direct accessor for the optional metric time name, only safe to call when we know that value is set."""
+        assert (
+            self._metric_time_column_name
+        ), "Expected a valid metric time dimension name to be associated with this dataset, but did not get one!"
+        return self._metric_time_column_name

metricflow/plan_conversion/dataflow_to_sql.py

@@ -468,70 +468,41 @@ def visit_source_node(self, node: ReadSqlSourceNode) -> SqlDataSet:
     def visit_join_over_time_range_node(self, node: JoinOverTimeRangeNode) -> SqlDataSet:
         """Generate time range join SQL."""
         table_alias_to_instance_set: OrderedDict[str, InstanceSet] = OrderedDict()
-        input_data_set = node.parent_node.accept(self)
-        input_data_set_alias = self._next_unique_table_alias()
+        parent_data_set = node.parent_node.accept(self)
+        parent_data_set_alias = self._next_unique_table_alias()
 
-        # Find requested agg_time_dimensions in parent instance set.
-        # Will use instance with the smallest base granularity in time spine join.
-        agg_time_dimension_instance_for_join: Optional[TimeDimensionInstance] = None
-        requested_agg_time_dimension_instances: Tuple[TimeDimensionInstance, ...] = ()
-        for instance in input_data_set.instance_set.time_dimension_instances:
-            if instance.spec in node.queried_agg_time_dimension_specs:
-                requested_agg_time_dimension_instances += (instance,)
-                if not agg_time_dimension_instance_for_join or (
-                    instance.spec.time_granularity.base_granularity.to_int()
-                    < agg_time_dimension_instance_for_join.spec.time_granularity.base_granularity.to_int()
-                ):
-                    agg_time_dimension_instance_for_join = instance
-        assert (
-            agg_time_dimension_instance_for_join
-        ), "Specified metric time spec not found in parent data set. This should have been caught by validations."
+        # For the purposes of this node, use base grains. Custom grains will be joined later in the dataflow plan.
+        agg_time_dimension_specs = tuple({spec.with_base_grain() for spec in node.queried_agg_time_dimension_specs})
 
+        # Assemble time_spine dataset with a column for each agg_time_dimension requested.
+        agg_time_dimension_instances = parent_data_set.instances_for_time_dimensions(agg_time_dimension_specs)
         time_spine_data_set_alias = self._next_unique_table_alias()
-
-        # Assemble time_spine dataset with requested agg time dimension instances selected.
         time_spine_data_set = self._make_time_spine_data_set(
-            agg_time_dimension_instances=requested_agg_time_dimension_instances,
-            time_range_constraint=node.time_range_constraint,
+            agg_time_dimension_instances=agg_time_dimension_instances, time_range_constraint=node.time_range_constraint
         )
-        table_alias_to_instance_set[time_spine_data_set_alias] = time_spine_data_set.instance_set
 
+        # Build the join description.
+        join_spec = self._choose_instance_for_time_spine_join(agg_time_dimension_instances).spec
+        annotated_parent = parent_data_set.annotate(alias=parent_data_set_alias, metric_time_spec=join_spec)
+        annotated_time_spine = time_spine_data_set.annotate(alias=time_spine_data_set_alias, metric_time_spec=join_spec)
         join_desc = SqlQueryPlanJoinBuilder.make_cumulative_metric_time_range_join_description(
-            node=node,
-            metric_data_set=AnnotatedSqlDataSet(
-                data_set=input_data_set,
-                alias=input_data_set_alias,
-                _metric_time_column_name=input_data_set.column_association_for_time_dimension(
-                    agg_time_dimension_instance_for_join.spec
-                ).column_name,
-            ),
-            time_spine_data_set=AnnotatedSqlDataSet(
-                data_set=time_spine_data_set,
-                alias=time_spine_data_set_alias,
-                _metric_time_column_name=time_spine_data_set.column_association_for_time_dimension(
-                    agg_time_dimension_instance_for_join.spec
-                ).column_name,
-            ),
+            node=node, metric_data_set=annotated_parent, time_spine_data_set=annotated_time_spine
         )
 
-        # Remove instances of agg_time_dimension from input data set. They'll be replaced with time spine instances.
-        agg_time_dimension_specs = tuple(dim.spec for dim in requested_agg_time_dimension_instances)
-        modified_input_instance_set = input_data_set.instance_set.transform(
+        # Build select columns, replacing agg_time_dimensions from the parent node with columns from the time spine.
+        table_alias_to_instance_set[time_spine_data_set_alias] = time_spine_data_set.instance_set
+        table_alias_to_instance_set[parent_data_set_alias] = parent_data_set.instance_set.transform(
             FilterElements(exclude_specs=InstanceSpecSet(time_dimension_specs=agg_time_dimension_specs))
         )
-        table_alias_to_instance_set[input_data_set_alias] = modified_input_instance_set
-
-        # The output instances are the same as the input instances.
-        output_instance_set = ChangeAssociatedColumns(self._column_association_resolver).transform(
-            input_data_set.instance_set
+        select_columns = create_simple_select_columns_for_instance_sets(
+            column_resolver=self._column_association_resolver, table_alias_to_instance_set=table_alias_to_instance_set
         )
+
         return SqlDataSet(
-            instance_set=output_instance_set,
+            instance_set=parent_data_set.instance_set,  # The output instances are the same as the input instances.
             sql_select_node=SqlSelectStatementNode.create(
                 description=node.description,
-                select_columns=create_simple_select_columns_for_instance_sets(
-                    self._column_association_resolver, table_alias_to_instance_set
-                ),
+                select_columns=select_columns,
                 from_source=time_spine_data_set.checked_sql_select_node,
                 from_source_alias=time_spine_data_set_alias,
                 join_descs=(join_desc,),
@@ -1392,6 +1363,21 @@ def visit_semi_additive_join_node(self, node: SemiAdditiveJoinNode) -> SqlDataSe
             ),
         )
 
+    def _choose_instance_for_time_spine_join(
+        self, agg_time_dimension_instances: Sequence[TimeDimensionInstance]
+    ) -> TimeDimensionInstance:
+        """Find the agg_time_dimension instance with the smallest grain to use for the time spine join."""
+        # We can't use a date part spec to join to the time spine, so filter those out.
+        agg_time_dimension_instances = [
+            instance for instance in agg_time_dimension_instances if not instance.spec.date_part
+        ]
+        assert len(agg_time_dimension_instances) > 0, (
+            "No appropriate agg_time_dimension was found to join to the time spine. "
+            "This indicates that the dataflow plan was configured incorrectly."
+        )
+        agg_time_dimension_instances.sort(key=lambda instance: instance.spec.time_granularity.base_granularity.to_int())
+        return agg_time_dimension_instances[0]
+
     def visit_join_to_time_spine_node(self, node: JoinToTimeSpineNode) -> SqlDataSet:  # noqa: D102
         parent_data_set = node.parent_node.accept(self)
         parent_alias = self._next_unique_table_alias()

metricflow/plan_conversion/instance_converters.py

@@ -803,6 +803,7 @@ def transform(self, instance_set: InstanceSet) -> SelectColumnSet:  # noqa: D102
         )
 
 
+# TODO: delete this class & all uses. It doesn't do anything.
 class ChangeAssociatedColumns(InstanceSetTransform[InstanceSet]):
     """Change the columns associated with instances to the one specified by the resolver.
 

metricflow/plan_conversion/sql_join_builder.py

@@ -13,7 +13,7 @@
 from metricflow.dataflow.nodes.join_over_time import JoinOverTimeRangeNode
 from metricflow.dataflow.nodes.join_to_base import JoinDescription
 from metricflow.dataflow.nodes.join_to_time_spine import JoinToTimeSpineNode
-from metricflow.dataset.sql_dataset import SqlDataSet
+from metricflow.dataset.sql_dataset import AnnotatedSqlDataSet
 from metricflow.plan_conversion.sql_expression_builders import make_coalesced_expr
 from metricflow.sql.sql_exprs import (
     SqlColumnReference,
@@ -46,23 +46,6 @@ class ColumnEqualityDescription:
     treat_nulls_as_equal: bool = False
 
 
-@dataclass(frozen=True)
-class AnnotatedSqlDataSet:
-    """Class to bind a DataSet to transient properties associated with it at a given point in the SqlQueryPlan."""
-
-    data_set: SqlDataSet
-    alias: str
-    _metric_time_column_name: Optional[str] = None
-
-    @property
-    def metric_time_column_name(self) -> str:
-        """Direct accessor for the optional metric time name, only safe to call when we know that value is set."""
-        assert (
-            self._metric_time_column_name
-        ), "Expected a valid metric time dimension name to be associated with this dataset, but did not get one!"
-        return self._metric_time_column_name
-
-
 class SqlQueryPlanJoinBuilder:
     """Helper class for constructing various join components in a SqlQueryPlan."""