Support multiple time spines using new time spine configs

.changes/unreleased/Features-20240725-160038.yaml

@@ -0,0 +1,6 @@
+kind: Features
+body: Support multiple time spines with different granularities.
+time: 2024-07-25T16:00:38.07984-07:00
+custom:
+  Author: courtneyholcomb
+  Issue: "1348"

extra-hatch-configuration/requirements.txt

@@ -1,5 +1,5 @@
 Jinja2>=3.1.3
-dbt-semantic-interfaces==0.6.9
+dbt-semantic-interfaces==0.6.10
 more-itertools>=8.10.0, <10.2.0
 pydantic>=1.10.0, <3.0
 tabulate>=0.8.9

metricflow-semantics/extra-hatch-configuration/requirements.txt

@@ -1,5 +1,5 @@
 # dbt Cloud depends on metricflow-semantics (dependency set in dbt-mantle), so DSI must always point to a production version here.
-dbt-semantic-interfaces>=0.6.9, <2.0.0
+dbt-semantic-interfaces>=0.6.10, <2.0.0
 graphviz>=0.18.2, <0.21
 python-dateutil>=2.9.0, <2.10.0
 rapidfuzz>=3.0, <4.0

metricflow/dataflow/builder/dataflow_plan_builder.py

@@ -996,7 +996,7 @@ def _find_dataflow_recipe(
             desired_linkable_specs=linkable_specs,
             nodes=candidate_nodes_for_right_side_of_join,
             metric_time_dimension_reference=self._metric_time_dimension_reference,
-            time_spine_node=self._source_node_set.time_spine_node,
+            time_spine_nodes=self._source_node_set.time_spine_nodes_tuple,
         )
         logger.info(
             f"After removing unnecessary nodes, there are {len(candidate_nodes_for_right_side_of_join)} candidate "
@@ -1034,7 +1034,7 @@ def _find_dataflow_recipe(
             semantic_model_lookup=self._semantic_model_lookup,
             nodes_available_for_joins=self._sort_by_suitability(candidate_nodes_for_right_side_of_join),
             node_data_set_resolver=self._node_data_set_resolver,
-            time_spine_node=self._source_node_set.time_spine_node,
+            time_spine_nodes=self._source_node_set.time_spine_nodes_tuple,
         )
 
         # Dict from the node that contains the source node to the evaluation results.

metricflow/dataflow/builder/node_evaluator.py

@@ -168,7 +168,7 @@ def __init__(
         semantic_model_lookup: SemanticModelLookup,
         nodes_available_for_joins: Sequence[DataflowPlanNode],
         node_data_set_resolver: DataflowPlanNodeOutputDataSetResolver,
-        time_spine_node: MetricTimeDimensionTransformNode,
+        time_spine_nodes: Sequence[MetricTimeDimensionTransformNode],
     ) -> None:
         """Initializer.
 
@@ -185,7 +185,7 @@ def __init__(
         self._node_data_set_resolver = node_data_set_resolver
         self._partition_resolver = PartitionJoinResolver(self._semantic_model_lookup)
         self._join_evaluator = SemanticModelJoinEvaluator(self._semantic_model_lookup)
-        self._time_spine_node = time_spine_node
+        self._time_spine_nodes = time_spine_nodes
 
     def _find_joinable_candidate_nodes_that_can_satisfy_linkable_specs(
         self,
@@ -201,7 +201,7 @@ def _find_joinable_candidate_nodes_that_can_satisfy_linkable_specs(
         left_node_spec_set = left_node_instance_set.spec_set
         for right_node in self._nodes_available_for_joins:
             # If right node is time spine source node, use cross join.
-            if right_node == self._time_spine_node:
+            if right_node in self._time_spine_nodes:
                 needed_metric_time_specs = group_specs_by_type(needed_linkable_specs).metric_time_specs
                 candidates_for_join.append(
                     JoinLinkableInstancesRecipe(

metricflow/dataflow/builder/source_node.py

@@ -1,9 +1,10 @@
 from __future__ import annotations
 
 from dataclasses import dataclass
-from typing import List, Sequence, Tuple
+from typing import List, Mapping, Sequence, Tuple
 
 from dbt_semantic_interfaces.references import TimeDimensionReference
+from dbt_semantic_interfaces.type_enums import TimeGranularity
 from metricflow_semantics.model.semantic_manifest_lookup import SemanticManifestLookup
 from metricflow_semantics.query.query_parser import MetricFlowQueryParser
 from metricflow_semantics.specs.column_assoc import ColumnAssociationResolver
@@ -36,14 +37,16 @@ class SourceNodeSet:
     # below. See usage in `DataflowPlanBuilder`.
     source_nodes_for_group_by_item_queries: Tuple[DataflowPlanNode, ...]
 
-    # Provides the time spine.
-    time_spine_node: MetricTimeDimensionTransformNode
+    # Provides the time spines.
+    time_spine_nodes: Mapping[TimeGranularity, MetricTimeDimensionTransformNode]
 
     @property
     def all_nodes(self) -> Sequence[DataflowPlanNode]:  # noqa: D102
-        return (
-            self.source_nodes_for_metric_queries + self.source_nodes_for_group_by_item_queries + (self.time_spine_node,)
-        )
+        return self.source_nodes_for_metric_queries + self.source_nodes_for_group_by_item_queries
+
+    @property
+    def time_spine_nodes_tuple(self) -> Tuple[MetricTimeDimensionTransformNode, ...]:  # noqa: D102
+        return tuple(self.time_spine_nodes.values())
 
 
 class SourceNodeBuilder:
@@ -56,13 +59,16 @@ def __init__(  # noqa: D107
     ) -> None:
         self._semantic_manifest_lookup = semantic_manifest_lookup
         data_set_converter = SemanticModelToDataSetConverter(column_association_resolver)
-        time_spine_source = TimeSpineSource.create_from_manifest(semantic_manifest_lookup.semantic_manifest)
-        time_spine_data_set = data_set_converter.build_time_spine_source_data_set(time_spine_source)
-        time_dim_reference = TimeDimensionReference(element_name=time_spine_source.time_column_name)
-        self._time_spine_source_node = MetricTimeDimensionTransformNode.create(
-            parent_node=ReadSqlSourceNode.create(data_set=time_spine_data_set),
-            aggregation_time_dimension_reference=time_dim_reference,
-        )
+        self._time_spine_source_nodes = {}
+        for granularity, time_spine_source in TimeSpineSource.create_from_manifest(
+            semantic_manifest_lookup.semantic_manifest
+        ).items():
+            data_set = data_set_converter.build_time_spine_source_data_set(time_spine_source)
+            self._time_spine_source_nodes[granularity] = MetricTimeDimensionTransformNode.create(
+                parent_node=ReadSqlSourceNode.create(data_set),
+                aggregation_time_dimension_reference=TimeDimensionReference(time_spine_source.time_column_name),
+            )
+
         self._query_parser = MetricFlowQueryParser(semantic_manifest_lookup)
 
     def create_from_data_sets(self, data_sets: Sequence[SemanticModelDataSet]) -> SourceNodeSet:
@@ -93,8 +99,9 @@ def create_from_data_sets(self, data_sets: Sequence[SemanticModelDataSet]) -> So
                     source_nodes_for_metric_queries.append(metric_time_transform_node)
 
         return SourceNodeSet(
-            time_spine_node=self._time_spine_source_node,
-            source_nodes_for_group_by_item_queries=tuple(group_by_item_source_nodes) + (self._time_spine_source_node,),
+            time_spine_nodes=self._time_spine_source_nodes,
+            source_nodes_for_group_by_item_queries=tuple(group_by_item_source_nodes)
+            + tuple(self._time_spine_source_nodes.values()),
             source_nodes_for_metric_queries=tuple(source_nodes_for_metric_queries),
         )
 

metricflow/engine/metricflow_engine.py

@@ -368,7 +368,7 @@ def __init__(
             DunderColumnAssociationResolver(semantic_manifest_lookup)
         )
         self._time_source = time_source
-        self._time_spine_source = TimeSpineSource.create_from_manifest(semantic_manifest_lookup.semantic_manifest)
+        self._time_spine_sources = TimeSpineSource.create_from_manifest(semantic_manifest_lookup.semantic_manifest)
         self._source_data_sets: List[SemanticModelDataSet] = []
         converter = SemanticModelToDataSetConverter(column_association_resolver=self._column_association_resolver)
         for semantic_model in sorted(

metricflow/plan_conversion/dataflow_to_sql.py

@@ -183,7 +183,7 @@ def __init__(
         self._semantic_manifest_lookup = semantic_manifest_lookup
         self._metric_lookup = semantic_manifest_lookup.metric_lookup
         self._semantic_model_lookup = semantic_manifest_lookup.semantic_model_lookup
-        self._time_spine_source = TimeSpineSource.create_from_manifest(semantic_manifest_lookup.semantic_manifest)
+        self._time_spine_sources = TimeSpineSource.create_from_manifest(semantic_manifest_lookup.semantic_manifest)
 
     @property
     def column_association_resolver(self) -> ColumnAssociationResolver:  # noqa: D102
@@ -222,24 +222,49 @@ def _next_unique_table_alias(self) -> str:
         """Return the next unique table alias to use in generating queries."""
         return SequentialIdGenerator.create_next_id(StaticIdPrefix.SUB_QUERY).str_value
 
+    def _choose_time_spine_source(
+        self, agg_time_dimension_instances: Tuple[TimeDimensionInstance, ...]
+    ) -> TimeSpineSource:
+        """Determine which time spine source to use when building time spine dataset.
+
+        Will choose the time spine with the largest granularity that can be used to get the smallest granularity requested
+        in the agg time dimension instances. Example:
+        - Time spines available: SECOND, MINUTE, DAY
+        - Agg time dimension granularity needed for request: HOUR, DAY
+        --> Selected time spine: MINUTE
+        """
+        assert (
+            agg_time_dimension_instances
+        ), "Building time spine dataset requires agg_time_dimension_instances, but none were found."
+        smallest_agg_time_grain = min(dim.spec.time_granularity for dim in agg_time_dimension_instances)
+        compatible_time_spine_grains = [
+            grain for grain in self._time_spine_sources.keys() if grain.to_int() <= smallest_agg_time_grain.to_int()
+        ]
+        if not compatible_time_spine_grains:
+            raise RuntimeError(
+                # TODO: update docs link when new docs are available
+                f"This query requires a time spine with granularity {smallest_agg_time_grain.name} or smaller, which is not configured. "
+                f"The smallest available time spine granularity is {min(self._time_spine_sources.keys()).name}, which is too large."
+                "See documentation for how to configure a new time spine: https://docs.getdbt.com/docs/build/metricflow-time-spine"
+            )
+        return self._time_spine_sources[max(compatible_time_spine_grains)]
+
     def _make_time_spine_data_set(
         self,
         agg_time_dimension_instances: Tuple[TimeDimensionInstance, ...],
-        time_spine_source: TimeSpineSource,
         time_range_constraint: Optional[TimeRangeConstraint] = None,
     ) -> SqlDataSet:
-        """Make a time spine data set, which contains all date/time values like '2020-01-01', '2020-01-02'...
+        """Returns a dataset with a datetime column for each agg_time_dimension granularity requested.
 
-        Returns a dataset with a column selected for each agg_time_dimension requested.
         Column alias will use 'metric_time' or the agg_time_dimension name depending on which the user requested.
         """
         time_spine_instance_set = InstanceSet(time_dimension_instances=agg_time_dimension_instances)
         time_spine_table_alias = self._next_unique_table_alias()
 
+        time_spine_source = self._choose_time_spine_source(agg_time_dimension_instances)
         column_expr = SqlColumnReferenceExpression.from_table_and_column_names(
             table_alias=time_spine_table_alias, column_name=time_spine_source.time_column_name
         )
-
         select_columns: Tuple[SqlSelectColumn, ...] = ()
         apply_group_by = False
         for agg_time_dimension_instance in agg_time_dimension_instances:
@@ -314,7 +339,6 @@ def visit_join_over_time_range_node(self, node: JoinOverTimeRangeNode) -> SqlDat
         # 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_spine_source=self._time_spine_source,
             time_range_constraint=node.time_range_constraint,
         )
         table_alias_to_instance_set[time_spine_data_set_alias] = time_spine_data_set.instance_set
@@ -1234,7 +1258,6 @@ def visit_join_to_time_spine_node(self, node: JoinToTimeSpineNode) -> SqlDataSet
         time_spine_alias = self._next_unique_table_alias()
         time_spine_dataset = self._make_time_spine_data_set(
             agg_time_dimension_instances=(agg_time_dimension_instance_for_join,),
-            time_spine_source=self._time_spine_source,
             time_range_constraint=node.time_range_constraint,
         )
 

metricflow/plan_conversion/node_processor.py

@@ -619,7 +619,7 @@ def remove_unnecessary_nodes(
         desired_linkable_specs: Sequence[LinkableInstanceSpec],
         nodes: Sequence[DataflowPlanNode],
         metric_time_dimension_reference: TimeDimensionReference,
-        time_spine_node: MetricTimeDimensionTransformNode,
+        time_spine_nodes: Sequence[MetricTimeDimensionTransformNode],
     ) -> List[DataflowPlanNode]:
         """Filters out many of the nodes that can't possibly be useful for joins to obtain the desired linkable specs.
 
@@ -661,7 +661,7 @@ def remove_unnecessary_nodes(
                 continue
 
             # Used for group-by-item-values queries.
-            if node == time_spine_node:
+            if node in time_spine_nodes:
                 logger.debug(f"Including {node} since it matches `time_spine_node`")
                 relevant_nodes.append(node)
                 continue

metricflow/plan_conversion/time_spine.py

@@ -2,11 +2,10 @@
 
 import logging
 from dataclasses import dataclass
-from typing import Optional
+from typing import Dict, Optional
 
 from dbt_semantic_interfaces.protocols import SemanticManifest
 from dbt_semantic_interfaces.type_enums.time_granularity import TimeGranularity
-from metricflow_semantics.mf_logging.pretty_print import mf_pformat
 from metricflow_semantics.specs.time_dimension_spec import DEFAULT_TIME_GRANULARITY
 
 from metricflow.sql.sql_table import SqlTable
@@ -34,26 +33,37 @@ def spine_table(self) -> SqlTable:
         return SqlTable(schema_name=self.schema_name, table_name=self.table_name, db_name=self.db_name)
 
     @staticmethod
-    def create_from_manifest(semantic_manifest: SemanticManifest) -> TimeSpineSource:
+    def create_from_manifest(semantic_manifest: SemanticManifest) -> Dict[TimeGranularity, TimeSpineSource]:
         """Creates a time spine source based on what's in the manifest."""
-        time_spine_table_configurations = semantic_manifest.project_configuration.time_spine_table_configurations
-
-        if not (
-            len(time_spine_table_configurations) == 1
-            and time_spine_table_configurations[0].grain == DEFAULT_TIME_GRANULARITY
-        ):
-            raise NotImplementedError(
-                f"Only a single time spine table configuration with {DEFAULT_TIME_GRANULARITY} is currently "
-                f"supported. Got:\n"
-                f"{mf_pformat(time_spine_table_configurations)}"
+        time_spine_sources = {
+            time_spine.primary_column.time_granularity: TimeSpineSource(
+                schema_name=time_spine.node_relation.schema_name,
+                table_name=time_spine.node_relation.relation_name,
+                db_name=time_spine.node_relation.database,
+                time_column_name=time_spine.primary_column.name,
+                time_column_granularity=time_spine.primary_column.time_granularity,
             )
+            for time_spine in semantic_manifest.project_configuration.time_spines
+        }
 
-        time_spine_table_configuration = time_spine_table_configurations[0]
-        time_spine_table = SqlTable.from_string(time_spine_table_configuration.location)
-        return TimeSpineSource(
-            schema_name=time_spine_table.schema_name,
-            table_name=time_spine_table.table_name,
-            db_name=time_spine_table.db_name,
-            time_column_name=time_spine_table_configuration.column_name,
-            time_column_granularity=time_spine_table_configuration.grain,
-        )
+        # For backward compatibility: if legacy time spine config exists in the manifest, add that time spine here for
+        # backward compatibility. Ignore it if there is a new time spine config with the same granularity.
+        legacy_time_spines = semantic_manifest.project_configuration.time_spine_table_configurations
+        for legacy_time_spine in legacy_time_spines:
+            if legacy_time_spine.grain not in time_spine_sources:
+                time_spine_table = SqlTable.from_string(legacy_time_spine.location)
+                time_spine_sources[legacy_time_spine.grain] = TimeSpineSource(
+                    schema_name=time_spine_table.schema_name,
+                    table_name=time_spine_table.table_name,
+                    db_name=time_spine_table.db_name,
+                    time_column_name=legacy_time_spine.column_name,
+                    time_column_granularity=legacy_time_spine.grain,
+                )
+
+        # Sanity check: this should have been validated during manifest parsing.
+        if not time_spine_sources:
+            raise RuntimeError(
+                "At least one time spine must be configured to use the semantic layer, but none were found."
+            )
+
+        return time_spine_sources

metricflow/validation/data_warehouse_model_validator.py

@@ -44,7 +44,6 @@
 from metricflow.dataset.dataset_classes import DataSet
 from metricflow.engine.metricflow_engine import MetricFlowEngine, MetricFlowExplainResult, MetricFlowQueryRequest
 from metricflow.plan_conversion.dataflow_to_sql import DataflowToSqlQueryPlanConverter
-from metricflow.plan_conversion.time_spine import TimeSpineSource
 from metricflow.protocols.sql_client import SqlClient
 
 
@@ -59,7 +58,6 @@ class QueryRenderingTools:
     semantic_manifest_lookup: SemanticManifestLookup
     source_node_builder: SourceNodeBuilder
     converter: SemanticModelToDataSetConverter
-    time_spine_source: TimeSpineSource
     plan_converter: DataflowToSqlQueryPlanConverter
 
     def __init__(self, manifest: SemanticManifest) -> None:  # noqa: D107
@@ -68,7 +66,6 @@ def __init__(self, manifest: SemanticManifest) -> None:  # noqa: D107
             column_association_resolver=DunderColumnAssociationResolver(self.semantic_manifest_lookup),
             semantic_manifest_lookup=self.semantic_manifest_lookup,
         )
-        self.time_spine_source = TimeSpineSource.create_from_manifest(manifest)
         self.converter = SemanticModelToDataSetConverter(
             column_association_resolver=DunderColumnAssociationResolver(
                 semantic_manifest_lookup=self.semantic_manifest_lookup

tests_metricflow/dataflow/builder/test_node_evaluator.py

@@ -48,7 +48,7 @@ def node_evaluator(
         ].semantic_manifest_lookup.semantic_model_lookup,
         nodes_available_for_joins=tuple(mf_engine_fixture.read_node_mapping.values()),
         node_data_set_resolver=node_data_set_resolver,
-        time_spine_node=mf_engine_fixture.source_node_set.time_spine_node,
+        time_spine_nodes=mf_engine_fixture.source_node_set.time_spine_nodes_tuple,
     )
 
 
@@ -72,7 +72,7 @@ def make_multihop_node_evaluator(
         desired_linkable_specs=desired_linkable_specs,
         nodes=source_node_set.source_nodes_for_metric_queries,
         metric_time_dimension_reference=DataSet.metric_time_dimension_reference(),
-        time_spine_node=source_node_set.time_spine_node,
+        time_spine_nodes=source_node_set.time_spine_nodes_tuple,
     )
 
     nodes_available_for_joins = list(
@@ -87,7 +87,7 @@ def make_multihop_node_evaluator(
         semantic_model_lookup=semantic_manifest_lookup_with_multihop_links.semantic_model_lookup,
         nodes_available_for_joins=nodes_available_for_joins,
         node_data_set_resolver=node_data_set_resolver,
-        time_spine_node=source_node_set.time_spine_node,
+        time_spine_nodes=source_node_set.time_spine_nodes_tuple,
     )
 
 
@@ -518,7 +518,7 @@ def test_node_evaluator_with_scd_target(
         # Use all nodes in the simple model as candidates for joins.
         nodes_available_for_joins=tuple(mf_engine_fixture.read_node_mapping.values()),
         node_data_set_resolver=node_data_set_resolver,
-        time_spine_node=mf_engine_fixture.source_node_set.time_spine_node,
+        time_spine_nodes=mf_engine_fixture.source_node_set.time_spine_nodes_tuple,
     )
 
     evaluation = node_evaluator.evaluate_node(