feat(cubesql): Add zero_members_wrapper cost component

Author: mcheshkov

rust/cubesql/cubesql/src/compile/mod.rs

@@ -2274,7 +2274,7 @@ from
             logical_plan.find_cube_scan().request,
             V1LoadRequestQuery {
                 measures: Some(vec![]),
-                dimensions: Some(vec![]),
+                dimensions: Some(vec!["KibanaSampleDataEcommerce.order_date".to_string()]),
                 segments: Some(vec![]),
                 order: Some(vec![]),
                 ungrouped: Some(true),
@@ -7362,7 +7362,10 @@ ORDER BY "source"."str0" ASC
             query_plan.as_logical_plan().find_cube_scan().request,
             V1LoadRequestQuery {
                 measures: Some(vec![]),
-                dimensions: Some(vec![]),
+                dimensions: Some(vec![
+                    "WideCube.dim1".to_string(),
+                    "WideCube.dim2".to_string(),
+                ]),
                 segments: Some(vec![]),
                 order: Some(vec![]),
                 ungrouped: Some(true),

rust/cubesql/cubesql/src/compile/rewrite/cost.rs

@@ -78,6 +78,11 @@ impl BestCubePlan {
             _ => 0,
         };
 
+        let zero_members_wrapper = match enode {
+            LogicalPlanLanguage::WrappedSelect(_) => 1,
+            _ => 0,
+        };
+
         let cube_members = match enode {
             LogicalPlanLanguage::Measure(_) => 1,
             LogicalPlanLanguage::Dimension(_) => 1,
@@ -196,6 +201,7 @@ impl BestCubePlan {
             non_pushed_down_window,
             non_pushed_down_grouping_sets,
             non_pushed_down_limit_sort,
+            zero_members_wrapper,
             cube_members,
             errors: this_errors,
             time_dimensions_used_as_dimensions,
@@ -247,6 +253,11 @@ pub struct CubePlanCost {
     filters: i64,
     structure_points: i64,
     filter_members: i64,
+    // This is separate from both non_detected_cube_scans and cube_members
+    // Because it's ok to use all members inside wrapper (so non_detected_cube_scans would be zero)
+    // And we want to select representation with less members
+    // But only when members are present!
+    zero_members_wrapper: i64,
     cube_members: i64,
     errors: i64,
     time_dimensions_used_as_dimensions: i64,
@@ -350,6 +361,11 @@ impl CubePlanCost {
             non_pushed_down_limit_sort: self.non_pushed_down_limit_sort
                 + other.non_pushed_down_limit_sort,
             member_errors: self.member_errors + other.member_errors,
+            zero_members_wrapper: (if other.cube_members == 0 {
+                self.zero_members_wrapper
+            } else {
+                0
+            }) + other.zero_members_wrapper,
             cube_members: self.cube_members + other.cube_members,
             errors: self.errors + other.errors,
             structure_points: self.structure_points + other.structure_points,
@@ -403,6 +419,8 @@ impl CubePlanCost {
                 SortState::Current if top_down => self.non_pushed_down_limit_sort,
                 _ => 0,
             },
+            // Don't track state here: we want representation that have fewer wrappers with zero members _in total_
+            zero_members_wrapper: self.zero_members_wrapper,
             cube_members: self.cube_members,
             errors: self.errors,
             structure_points: self.structure_points,

rust/cubesql/cubesql/src/compile/test/test_wrapper.rs

@@ -1261,6 +1261,7 @@ async fn test_wrapper_filter_flatten() {
 }
 
 /// Regular aggregation over CubeScan(limit=n, ungrouped=true) is NOT pushed to CubeScan
+/// and inner ungrouped CubeScan should have both proper members and limit
 #[tokio::test]
 async fn wrapper_agg_over_limit() {
     if !Rewriter::sql_push_down_enabled() {
@@ -1299,7 +1300,9 @@ async fn wrapper_agg_over_limit() {
         logical_plan.find_cube_scan().request,
         V1LoadRequestQuery {
             measures: Some(vec![]),
-            dimensions: Some(vec![]),
+            dimensions: Some(vec![
+                "KibanaSampleDataEcommerce.customer_gender".to_string(),
+            ]),
             segments: Some(vec![]),
             order: Some(vec![]),
             limit: Some(5),
@@ -1324,6 +1327,7 @@ async fn wrapper_agg_over_limit() {
 }
 
 /// Aggregation(dimension) over CubeScan(limit=n, ungrouped=true) is NOT pushed to CubeScan
+/// and inner ungrouped CubeScan should have both proper members and limit
 #[tokio::test]
 async fn wrapper_agg_dimension_over_limit() {
     if !Rewriter::sql_push_down_enabled() {
@@ -1360,7 +1364,9 @@ async fn wrapper_agg_dimension_over_limit() {
         logical_plan.find_cube_scan().request,
         V1LoadRequestQuery {
             measures: Some(vec![]),
-            dimensions: Some(vec![]),
+            dimensions: Some(vec![
+                "KibanaSampleDataEcommerce.customer_gender".to_string(),
+            ]),
             segments: Some(vec![]),
             order: Some(vec![]),
             limit: Some(5),