Add improved test for the SQL optimization level in the request

metricflow/dataflow/optimizer/source_scan/cm_branch_combiner.py

@@ -133,7 +133,7 @@ def __init__(self, left_branch_node: DataflowPlanNode) -> None:  # noqa: D107
         self._current_left_node: DataflowPlanNode = left_branch_node
 
     def _log_visit_node_type(self, node: DataflowPlanNode) -> None:
-        logger.debug(lambda: f"Visiting {node.node_id}")
+        logger.debug(LazyFormat(lambda: f"Visiting {node.node_id}"))
 
     def _log_combine_failure(
         self,

metricflow/dataflow/optimizer/source_scan/source_scan_optimizer.py

@@ -280,7 +280,14 @@ def visit_combine_aggregated_outputs_node(  # noqa: D102
                     for branch_combination_result in combination_results
                 ]
 
-        logger.debug(lambda: f"Got {len(combined_parent_branches)} branches after combination")
+        logger.debug(
+            LazyFormat(
+                "Possible branches combined.",
+                count_of_branches_before_combination=len(optimized_parent_branches),
+                count_of_branches_after_combination=len(combined_parent_branches),
+            )
+        )
+
         assert len(combined_parent_branches) > 0
 
         # If we were able to reduce the parent branches of the CombineAggregatedOutputsNode into a single one, there's

metricflow/sql/optimizer/optimization_levels.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import functools
 from dataclasses import dataclass
 from enum import Enum
 from typing import Tuple
@@ -13,6 +14,7 @@
 from metricflow.sql.optimizer.table_alias_simplifier import SqlTableAliasSimplifier
 
 
+@functools.total_ordering
 class SqlQueryOptimizationLevel(Enum):
     """Defines the level of query optimization and the associated optimizers to apply."""
 
@@ -27,6 +29,12 @@ class SqlQueryOptimizationLevel(Enum):
     def default_level() -> SqlQueryOptimizationLevel:  # noqa: D102
         return SqlQueryOptimizationLevel.O5
 
+    def __lt__(self, other: SqlQueryOptimizationLevel) -> bool:  # noqa: D105
+        if not isinstance(other, SqlQueryOptimizationLevel):
+            return NotImplemented
+
+        return self.name < other.name
+
 
 @dataclass(frozen=True)
 class SqlGenerationOptionSet:

tests_metricflow/engine/test_explain.py

@@ -1,10 +1,20 @@
 from __future__ import annotations
 
+import logging
 from concurrent.futures import Future, ThreadPoolExecutor
 from typing import Mapping, Sequence
 
+import pytest
+from _pytest.fixtures import FixtureRequest
+from metricflow_semantics.mf_logging.pretty_print import mf_pformat_dict
+from metricflow_semantics.test_helpers.config_helpers import MetricFlowTestConfiguration
+
 from metricflow.engine.metricflow_engine import MetricFlowEngine, MetricFlowExplainResult, MetricFlowQueryRequest
+from metricflow.sql.optimizer.optimization_levels import SqlQueryOptimizationLevel
 from tests_metricflow.fixtures.manifest_fixtures import MetricFlowEngineTestFixture, SemanticManifestSetup
+from tests_metricflow.snapshot_utils import assert_str_snapshot_equal
+
+logger = logging.getLogger(__name__)
 
 
 def _explain_one_query(mf_engine: MetricFlowEngine) -> str:
@@ -29,3 +39,39 @@ def test_concurrent_explain_consistency(
         results = [future.result() for future in futures]
         for result in results:
             assert result == results[0], "Expected only one unique result / results to be the same"
+
+
+@pytest.mark.sql_engine_snapshot
+@pytest.mark.duckdb_only
+def test_optimization_level(
+    request: FixtureRequest,
+    mf_test_configuration: MetricFlowTestConfiguration,
+    mf_engine_test_fixture_mapping: Mapping[SemanticManifestSetup, MetricFlowEngineTestFixture],
+) -> None:
+    """Tests that the results of explain reflect the SQL optimization level in the request."""
+    mf_engine = mf_engine_test_fixture_mapping[SemanticManifestSetup.SIMPLE_MANIFEST].metricflow_engine
+
+    results = {}
+    for optimization_level in SqlQueryOptimizationLevel:
+        # Skip lower optimization levels as they are generally not used.
+        if optimization_level <= SqlQueryOptimizationLevel.O3:
+            continue
+
+        explain_result: MetricFlowExplainResult = mf_engine.explain(
+            MetricFlowQueryRequest.create_with_random_request_id(
+                metric_names=("bookings", "views"),
+                group_by_names=("metric_time", "listing__country_latest"),
+                sql_optimization_level=optimization_level,
+            )
+        )
+        results[optimization_level.value] = explain_result.rendered_sql_without_descriptions.sql_query
+
+    assert_str_snapshot_equal(
+        request=request,
+        mf_test_configuration=mf_test_configuration,
+        snapshot_id="result",
+        snapshot_str=mf_pformat_dict(
+            description=None, obj_dict=results, preserve_raw_strings=True, pad_items_with_newlines=True
+        ),
+        expectation_description=f"The result for {SqlQueryOptimizationLevel.O5} should be SQL uses a CTE.",
+    )

tests_metricflow/integration/test_mf_engine.py

@@ -3,8 +3,6 @@
 from _pytest.fixtures import FixtureRequest
 from metricflow_semantics.test_helpers.config_helpers import MetricFlowTestConfiguration
 
-from metricflow.engine.metricflow_engine import MetricFlowExplainResult, MetricFlowQueryRequest
-from metricflow.sql.optimizer.optimization_levels import SqlQueryOptimizationLevel
 from tests_metricflow.integration.conftest import IntegrationTestHelpers
 from tests_metricflow.snapshot_utils import assert_object_snapshot_equal
 
@@ -18,26 +16,3 @@ def test_list_dimensions(  # noqa: D103
         obj_id="result0",
         obj=sorted([dim.qualified_name for dim in it_helpers.mf_engine.list_dimensions()]),
     )
-
-
-def test_sql_optimization_level(it_helpers: IntegrationTestHelpers) -> None:
-    """Check that different SQL optimization levels produce different SQL."""
-    assert (
-        SqlQueryOptimizationLevel.default_level() != SqlQueryOptimizationLevel.O0
-    ), "The default optimization level should be different from the lowest level."
-    explain_result_at_default_level: MetricFlowExplainResult = it_helpers.mf_engine.explain(
-        MetricFlowQueryRequest.create_with_random_request_id(
-            metric_names=("bookings",),
-            group_by_names=("metric_time",),
-            sql_optimization_level=SqlQueryOptimizationLevel.default_level(),
-        )
-    )
-    explain_result_at_level_0: MetricFlowExplainResult = it_helpers.mf_engine.explain(
-        MetricFlowQueryRequest.create_with_random_request_id(
-            metric_names=("bookings",),
-            group_by_names=("metric_time",),
-            sql_optimization_level=SqlQueryOptimizationLevel.O0,
-        )
-    )
-
-    assert explain_result_at_default_level.rendered_sql.sql_query != explain_result_at_level_0.rendered_sql.sql_query

tests_metricflow/snapshot_utils.py

@@ -141,4 +141,8 @@ def assert_str_snapshot_equal(  # type: ignore[misc]
         snapshot_file_extension=".txt",
         additional_sub_directories_for_snapshots=(sql_engine.value,) if sql_engine is not None else (),
         expectation_description=expectation_description,
+        incomparable_strings_replacement_function=make_schema_replacement_function(
+            system_schema=mf_test_configuration.mf_system_schema,
+            source_schema=mf_test_configuration.mf_source_schema,
+        ),
     )

tests_metricflow/snapshots/test_explain.py/str/test_optimization_level__result.txt

@@ -0,0 +1,125 @@
+test_name: test_optimization_level
+test_filename: test_explain.py
+docstring:
+  Tests that the results of explain reflect the SQL optimization level in the request.
+expectation_description:
+  The result for SqlQueryOptimizationLevel.O5 should be SQL uses a CTE.
+---
+O4:
+  SELECT
+    COALESCE(subq_8.metric_time__day, subq_17.metric_time__day) AS metric_time__day
+    , COALESCE(subq_8.listing__country_latest, subq_17.listing__country_latest) AS listing__country_latest
+    , MAX(subq_8.bookings) AS bookings
+    , MAX(subq_17.views) AS views
+  FROM (
+    SELECT
+      subq_1.metric_time__day AS metric_time__day
+      , listings_latest_src_10000.country AS listing__country_latest
+      , SUM(subq_1.bookings) AS bookings
+    FROM (
+      SELECT
+        DATE_TRUNC('day', ds) AS metric_time__day
+        , listing_id AS listing
+        , 1 AS bookings
+      FROM ***************************.fct_bookings bookings_source_src_10000
+    ) subq_1
+    LEFT OUTER JOIN
+      ***************************.dim_listings_latest listings_latest_src_10000
+    ON
+      subq_1.listing = listings_latest_src_10000.listing_id
+    GROUP BY
+      subq_1.metric_time__day
+      , listings_latest_src_10000.country
+  ) subq_8
+  FULL OUTER JOIN (
+    SELECT
+      subq_10.metric_time__day AS metric_time__day
+      , listings_latest_src_10000.country AS listing__country_latest
+      , SUM(subq_10.views) AS views
+    FROM (
+      SELECT
+        DATE_TRUNC('day', ds) AS metric_time__day
+        , listing_id AS listing
+        , 1 AS views
+      FROM ***************************.fct_views views_source_src_10000
+    ) subq_10
+    LEFT OUTER JOIN
+      ***************************.dim_listings_latest listings_latest_src_10000
+    ON
+      subq_10.listing = listings_latest_src_10000.listing_id
+    GROUP BY
+      subq_10.metric_time__day
+      , listings_latest_src_10000.country
+  ) subq_17
+  ON
+    (
+      subq_8.listing__country_latest = subq_17.listing__country_latest
+    ) AND (
+      subq_8.metric_time__day = subq_17.metric_time__day
+    )
+  GROUP BY
+    COALESCE(subq_8.metric_time__day, subq_17.metric_time__day)
+    , COALESCE(subq_8.listing__country_latest, subq_17.listing__country_latest)
+
+O5:
+  WITH sma_10014_cte AS (
+    SELECT
+      listing_id AS listing
+      , country AS country_latest
+    FROM ***************************.dim_listings_latest listings_latest_src_10000
+  )
+
+  SELECT
+    COALESCE(subq_8.metric_time__day, subq_16.metric_time__day) AS metric_time__day
+    , COALESCE(subq_8.listing__country_latest, subq_16.listing__country_latest) AS listing__country_latest
+    , MAX(subq_8.bookings) AS bookings
+    , MAX(subq_16.views) AS views
+  FROM (
+    SELECT
+      subq_1.metric_time__day AS metric_time__day
+      , sma_10014_cte.country_latest AS listing__country_latest
+      , SUM(subq_1.bookings) AS bookings
+    FROM (
+      SELECT
+        DATE_TRUNC('day', ds) AS metric_time__day
+        , listing_id AS listing
+        , 1 AS bookings
+      FROM ***************************.fct_bookings bookings_source_src_10000
+    ) subq_1
+    LEFT OUTER JOIN
+      sma_10014_cte sma_10014_cte
+    ON
+      subq_1.listing = sma_10014_cte.listing
+    GROUP BY
+      subq_1.metric_time__day
+      , sma_10014_cte.country_latest
+  ) subq_8
+  FULL OUTER JOIN (
+    SELECT
+      subq_10.metric_time__day AS metric_time__day
+      , sma_10014_cte.country_latest AS listing__country_latest
+      , SUM(subq_10.views) AS views
+    FROM (
+      SELECT
+        DATE_TRUNC('day', ds) AS metric_time__day
+        , listing_id AS listing
+        , 1 AS views
+      FROM ***************************.fct_views views_source_src_10000
+    ) subq_10
+    LEFT OUTER JOIN
+      sma_10014_cte sma_10014_cte
+    ON
+      subq_10.listing = sma_10014_cte.listing
+    GROUP BY
+      subq_10.metric_time__day
+      , sma_10014_cte.country_latest
+  ) subq_16
+  ON
+    (
+      subq_8.listing__country_latest = subq_16.listing__country_latest
+    ) AND (
+      subq_8.metric_time__day = subq_16.metric_time__day
+    )
+  GROUP BY
+    COALESCE(subq_8.metric_time__day, subq_16.metric_time__day)
+    , COALESCE(subq_8.listing__country_latest, subq_16.listing__country_latest)