feat(optimizer): index accelerating TopN (#7726)

y-wei · web-flow · commit 2d7989469450 · 2023-02-09T17:11:50.000Z
Optimizing applicable `LogicalTopN-LogicalScan` to `LogicalLimit-LogicalScan` by a new rule named `top_n_on_index` that arbitrarily match the above structure.

Please explain **IN DETAIL** what the changes are in this PR and why they are needed:

- Summarize your change (**mandatory**)
`LogicalTopN` will be replaced with `LogicalLimit` if there is applicable index or primary key.

On index:
```SQL
dev=&gt; explain select v1 from t order by v1 limit 1;
QUERY PLAN
--------------------------------------------------------
BatchLimit { limit: 1, offset: 0 }
└─BatchExchange { order: [idx1.v1 ASC], dist: Single }
└─BatchLimit { limit: 1, offset: 0 }
└─BatchScan { table: idx1, columns: [v1] }
(4 rows)
```

On primary key:
```SQL
dev=&gt; explain select * from t order by k limit 1;
QUERY PLAN
----------------------------------------------------
BatchLimit { limit: 1, offset: 0 }
└─BatchExchange { order: [t.k ASC], dist: Single }
└─BatchLimit { limit: 1, offset: 0 }
└─BatchScan { table: t, columns: [k, cnt] }
(4 rows)
```

Approved-By: chenzl25
Approved-By: BugenZhao

Co-Authored-By: Eurekaaw &lt;yifei.c.wei@gmail.com&gt;
Co-Authored-By: Clearlove &lt;52417396+Eurekaaw@users.noreply.github.com&gt;
diff --git a/dashboard/proto/gen/batch_plan.ts b/dashboard/proto/gen/batch_plan.ts
diff --git a/e2e_test/batch/top_n/top_n_on_index.slt b/e2e_test/batch/top_n/top_n_on_index.slt
@@ -0,0 +1,47 @@
+statement ok
+SET RW_IMPLICIT_FLUSH TO true;
+
+statement ok
+create table t(x int, y int);
+
+statement ok
+create index idx on t(x);
+
+statement ok
+insert into t values (100, 3), (1, 0), (2, 3), (3, 4), (5, 4);
+
+query II
+select * from t order by x limit 1;
+----
+1 0
+
+query II
+select * from t order by x limit 3;
+----
+1 0
+2 3
+3 4
+
+statement ok
+create table t1(x int primary key, y int);
+
+statement ok
+insert into t1 values (100, 3), (1, 0), (2, 3), (3, 4), (5, 4);
+
+query II
+select * from t1 order by x limit 1;
+----
+1 0
+
+query II
+select * from t1 order by x limit 3;
+----
+1 0
+2 3
+3 4
+
+statement ok
+drop table t;
+
+statement ok
+drop table t1;
diff --git a/proto/batch_plan.proto b/proto/batch_plan.proto
@@ -24,6 +24,12 @@ message RowSeqScanNode {
   common.Buffer vnode_bitmap = 4;
   // Whether the order on output columns should be preserved.
   bool ordered = 5;
+
+  message ChunkSize {
+    uint32 chunk_size = 1;
+  }
+  // If along with `batch_limit`, `chunk_size` will be set.
+  ChunkSize chunk_size = 6;
 }
 
 message SysRowSeqScanNode {
diff --git a/src/batch/src/executor/join/local_lookup_join.rs b/src/batch/src/executor/join/local_lookup_join.rs
@@ -115,6 +115,7 @@ impl<C: BatchTaskContext> InnerSideExecutorBuilder<C> {
             scan_ranges,
             ordered: false,
             vnode_bitmap: Some(vnode_bitmap.finish().to_protobuf()),
+            chunk_size: None,
         });
 
         Ok(row_seq_scan_node)
diff --git a/src/batch/src/executor/row_seq_scan.rs b/src/batch/src/executor/row_seq_scan.rs
@@ -241,7 +241,13 @@ impl BoxedExecutorBuilder for RowSeqScanExecutorBuilder {
         let ordered = seq_scan_node.ordered;
 
         let epoch = source.epoch.clone();
-        let chunk_size = source.context.get_config().developer.batch_chunk_size;
+        let chunk_size = if let Some(chunk_size_) = &seq_scan_node.chunk_size {
+            chunk_size_
+                .get_chunk_size()
+                .min(source.context.get_config().developer.batch_chunk_size as u32)
+        } else {
+            source.context.get_config().developer.batch_chunk_size as u32
+        };
         let metrics = source.context().task_metrics();
 
         dispatch_state_store!(source.context().state_store(), state_store, {
@@ -262,7 +268,7 @@ impl BoxedExecutorBuilder for RowSeqScanExecutorBuilder {
                 scan_ranges,
                 ordered,
                 epoch,
-                chunk_size,
+                chunk_size as usize,
                 source.plan_node().get_identity().clone(),
                 metrics,
             )))
diff --git a/src/frontend/planner_test/tests/testdata/explain.yaml b/src/frontend/planner_test/tests/testdata/explain.yaml
@@ -51,7 +51,7 @@
       "stages": {
         "0": {
           "root": {
-            "plan_node_id": 28,
+            "plan_node_id": 29,
             "plan_node_type": "BatchProject",
             "schema": [
               {
@@ -64,7 +64,7 @@
             ],
             "children": [
               {
-                "plan_node_id": 26,
+                "plan_node_id": 27,
                 "plan_node_type": "BatchValues",
                 "schema": [],
                 "children": [],
diff --git a/src/frontend/planner_test/tests/testdata/index_selection.yaml b/src/frontend/planner_test/tests/testdata/index_selection.yaml
@@ -578,3 +578,23 @@
   batch_plan: |
     BatchExchange { order: [], dist: Single }
     └─BatchScan { table: idx1, columns: [idx1.a, idx1.b, idx1.c], scan_ranges: [idx1.a = Int32(1)], distribution: UpstreamHashShard(idx1.a) }
+- name: topn on index
+  sql: |
+    create table t1 (a int, b int);
+    create index idx1 on t1(a);
+    select * from t1 order by a limit 1
+  batch_plan: |
+    BatchLimit { limit: 1, offset: 0 }
+    └─BatchExchange { order: [idx1.a ASC], dist: Single }
+      └─BatchLimit { limit: 1, offset: 0 }
+        └─BatchScan { table: idx1, columns: [idx1.a, idx1.b], distribution: UpstreamHashShard(idx1.a) }
+- name: topn on primary key
+  sql: |
+    create table t1 (a int primary key, b int);
+    create index idx1 on t1(a);
+    select * from t1 order by a limit 1
+  batch_plan: |
+    BatchLimit { limit: 1, offset: 0 }
+    └─BatchExchange { order: [t1.a ASC], dist: Single }
+      └─BatchLimit { limit: 1, offset: 0 }
+        └─BatchScan { table: t1, columns: [t1.a, t1.b], distribution: UpstreamHashShard(t1.a) }
diff --git a/src/frontend/planner_test/tests/testdata/nexmark_source.yaml b/src/frontend/planner_test/tests/testdata/nexmark_source.yaml
@@ -350,7 +350,7 @@
           |     └─StreamExchange { dist: HashShard(auction, window_start) }
           |       └─StreamHopWindow { time_col: date_time, slide: 00:00:02, size: 00:00:10, output: [auction, window_start, _row_id] }
           |         └─StreamProject { exprs: [auction, date_time, _row_id] }
-          |           └─StreamShare { id = 744 }
+          |           └─StreamShare { id = 764 }
           |             └─StreamProject { exprs: [auction, date_time, _row_id] }
           |               └─StreamRowIdGen { row_id_index: 7 }
           |                 └─StreamSource { source: "bid", columns: ["auction", "bidder", "price", "channel", "url", "date_time", "extra", "_row_id"] }
@@ -362,7 +362,7 @@
                     └─StreamExchange { dist: HashShard(auction, window_start) }
                       └─StreamHopWindow { time_col: date_time, slide: 00:00:02, size: 00:00:10, output: [auction, window_start, _row_id] }
                         └─StreamProject { exprs: [auction, date_time, _row_id] }
-                          └─StreamShare { id = 744 }
+                          └─StreamShare { id = 764 }
                             └─StreamProject { exprs: [auction, date_time, _row_id] }
                               └─StreamRowIdGen { row_id_index: 7 }
                                 └─StreamSource { source: "bid", columns: ["auction", "bidder", "price", "channel", "url", "date_time", "extra", "_row_id"] }
@@ -480,7 +480,7 @@
         └─StreamHashJoin { type: Inner, predicate: price = max(price), output: all }
           ├─StreamExchange { dist: HashShard(price) }
           | └─StreamProject { exprs: [auction, bidder, price, date_time, _row_id] }
-          |   └─StreamShare { id = 434 }
+          |   └─StreamShare { id = 446 }
           |     └─StreamProject { exprs: [auction, bidder, price, date_time, _row_id] }
           |       └─StreamRowIdGen { row_id_index: 7 }
           |         └─StreamSource { source: "bid", columns: ["auction", "bidder", "price", "channel", "url", "date_time", "extra", "_row_id"] }
@@ -489,7 +489,7 @@
               └─StreamAppendOnlyHashAgg { group_key: [(TumbleStart(date_time, '00:00:10':Interval) + '00:00:10':Interval)], aggs: [count, max(price)] }
                 └─StreamExchange { dist: HashShard((TumbleStart(date_time, '00:00:10':Interval) + '00:00:10':Interval)) }
                   └─StreamProject { exprs: [(TumbleStart(date_time, '00:00:10':Interval) + '00:00:10':Interval), price, _row_id] }
-                    └─StreamShare { id = 434 }
+                    └─StreamShare { id = 446 }
                       └─StreamProject { exprs: [auction, bidder, price, date_time, _row_id] }
                         └─StreamRowIdGen { row_id_index: 7 }
                           └─StreamSource { source: "bid", columns: ["auction", "bidder", "price", "channel", "url", "date_time", "extra", "_row_id"] }
@@ -1331,7 +1331,7 @@
         |     |     └─StreamSource { source: "auction", columns: ["id", "item_name", "description", "initial_bid", "reserve", "date_time", "expires", "seller", "category", "_row_id"] }
         |     └─StreamExchange { dist: HashShard(auction) }
         |       └─StreamProject { exprs: [auction, _row_id] }
-        |         └─StreamShare { id = 553 }
+        |         └─StreamShare { id = 571 }
         |           └─StreamProject { exprs: [auction, _row_id] }
         |             └─StreamRowIdGen { row_id_index: 7 }
         |               └─StreamSource { source: "bid", columns: ["auction", "bidder", "price", "channel", "url", "date_time", "extra", "_row_id"] }
@@ -1343,7 +1343,7 @@
                   └─StreamAppendOnlyHashAgg { group_key: [auction], aggs: [count, count] }
                     └─StreamExchange { dist: HashShard(auction) }
                       └─StreamProject { exprs: [auction, _row_id] }
-                        └─StreamShare { id = 553 }
+                        └─StreamShare { id = 571 }
                           └─StreamProject { exprs: [auction, _row_id] }
                             └─StreamRowIdGen { row_id_index: 7 }
                               └─StreamSource { source: "bid", columns: ["auction", "bidder", "price", "channel", "url", "date_time", "extra", "_row_id"] }
diff --git a/src/frontend/planner_test/tests/testdata/share.yaml b/src/frontend/planner_test/tests/testdata/share.yaml
@@ -48,7 +48,7 @@
               | └─StreamProject { exprs: [id, _row_id] }
               |   └─StreamFilter { predicate: (initial_bid = 1:Int32) }
               |     └─StreamProject { exprs: [id, initial_bid, _row_id] }
-              |       └─StreamShare { id = 519 }
+              |       └─StreamShare { id = 539 }
               |         └─StreamProject { exprs: [id, initial_bid, _row_id] }
               |           └─StreamFilter { predicate: ((initial_bid = 1:Int32) OR (initial_bid = 2:Int32)) }
               |             └─StreamRowIdGen { row_id_index: 9 }
@@ -57,7 +57,7 @@
                 └─StreamProject { exprs: [id, _row_id] }
                   └─StreamFilter { predicate: (initial_bid = 2:Int32) }
                     └─StreamProject { exprs: [id, initial_bid, _row_id] }
-                      └─StreamShare { id = 519 }
+                      └─StreamShare { id = 539 }
                         └─StreamProject { exprs: [id, initial_bid, _row_id] }
                           └─StreamFilter { predicate: ((initial_bid = 1:Int32) OR (initial_bid = 2:Int32)) }
                             └─StreamRowIdGen { row_id_index: 9 }
@@ -125,7 +125,7 @@
           |     └─StreamExchange { dist: HashShard(auction, window_start) }
           |       └─StreamHopWindow { time_col: date_time, slide: 00:00:02, size: 00:00:10, output: [auction, window_start, _row_id] }
           |         └─StreamProject { exprs: [auction, date_time, _row_id] }
-          |           └─StreamShare { id = 744 }
+          |           └─StreamShare { id = 764 }
           |             └─StreamProject { exprs: [auction, date_time, _row_id] }
           |               └─StreamRowIdGen { row_id_index: 4 }
           |                 └─StreamSource { source: "bid", columns: ["auction", "bidder", "price", "date_time", "_row_id"] }
@@ -137,7 +137,7 @@
                     └─StreamExchange { dist: HashShard(auction, window_start) }
                       └─StreamHopWindow { time_col: date_time, slide: 00:00:02, size: 00:00:10, output: [auction, window_start, _row_id] }
                         └─StreamProject { exprs: [auction, date_time, _row_id] }
-                          └─StreamShare { id = 744 }
+                          └─StreamShare { id = 764 }
                             └─StreamProject { exprs: [auction, date_time, _row_id] }
                               └─StreamRowIdGen { row_id_index: 4 }
                                 └─StreamSource { source: "bid", columns: ["auction", "bidder", "price", "date_time", "_row_id"] }
diff --git a/src/frontend/planner_test/tests/testdata/shared_views.yaml b/src/frontend/planner_test/tests/testdata/shared_views.yaml
@@ -26,7 +26,7 @@
     └─StreamHashJoin { type: Inner, predicate: (t1.x + t1.y) = (t1.x * (t1.x + t1.y)), output: [(t1.x + t1.y), (t1.x * (t1.x + t1.y)), (t1.y * (t1.x + t1.y)), t1._row_id, t1._row_id, t1._row_id, t1.x, (t1.x + t1.y)] }
       ├─StreamExchange { dist: HashShard((t1.x + t1.y)) }
       | └─StreamProject { exprs: [(t1.x + t1.y), t1._row_id] }
-      |   └─StreamShare { id = 207 }
+      |   └─StreamShare { id = 212 }
       |     └─StreamProject { exprs: [(t1.x + t1.y), t1._row_id] }
       |       └─StreamFilter { predicate: (t1.y > 0:Int32) }
       |         └─StreamTableScan { table: t1, columns: [t1.x, t1.y, t1._row_id], pk: [t1._row_id], dist: UpstreamHashShard(t1._row_id) }
@@ -37,7 +37,7 @@
             | └─StreamTableScan { table: t1, columns: [t1.x, t1.y, t1._row_id], pk: [t1._row_id], dist: UpstreamHashShard(t1._row_id) }
             └─StreamExchange { dist: HashShard((t1.x + t1.y)) }
               └─StreamProject { exprs: [(t1.x + t1.y), t1._row_id] }
-                └─StreamShare { id = 207 }
+                └─StreamShare { id = 212 }
                   └─StreamProject { exprs: [(t1.x + t1.y), t1._row_id] }
                     └─StreamFilter { predicate: (t1.y > 0:Int32) }
                       └─StreamTableScan { table: t1, columns: [t1.x, t1.y, t1._row_id], pk: [t1._row_id], dist: UpstreamHashShard(t1._row_id) }
diff --git a/src/frontend/planner_test/tests/testdata/tpch.yaml b/src/frontend/planner_test/tests/testdata/tpch.yaml
@@ -2573,7 +2573,7 @@
       |   ├─StreamExchange { dist: HashShard(supplier.s_suppkey) }
       |   | └─StreamTableScan { table: supplier, columns: [supplier.s_suppkey, supplier.s_name, supplier.s_address, supplier.s_phone], pk: [supplier.s_suppkey], dist: UpstreamHashShard(supplier.s_suppkey) }
       |   └─StreamProject { exprs: [lineitem.l_suppkey, sum((lineitem.l_extendedprice * (1:Int32 - lineitem.l_discount)))] }
-      |     └─StreamShare { id = 900 }
+      |     └─StreamShare { id = 921 }
       |       └─StreamProject { exprs: [lineitem.l_suppkey, sum((lineitem.l_extendedprice * (1:Int32 - lineitem.l_discount)))] }
       |         └─StreamHashAgg { group_key: [lineitem.l_suppkey], aggs: [count, sum((lineitem.l_extendedprice * (1:Int32 - lineitem.l_discount)))] }
       |           └─StreamExchange { dist: HashShard(lineitem.l_suppkey) }
@@ -2587,7 +2587,7 @@
               └─StreamHashAgg { group_key: [Vnode(lineitem.l_suppkey)], aggs: [count, max(sum((lineitem.l_extendedprice * (1:Int32 - lineitem.l_discount))))] }
                 └─StreamProject { exprs: [lineitem.l_suppkey, sum((lineitem.l_extendedprice * (1:Int32 - lineitem.l_discount))), Vnode(lineitem.l_suppkey)] }
                   └─StreamProject { exprs: [lineitem.l_suppkey, sum((lineitem.l_extendedprice * (1:Int32 - lineitem.l_discount)))] }
-                    └─StreamShare { id = 900 }
+                    └─StreamShare { id = 921 }
                       └─StreamProject { exprs: [lineitem.l_suppkey, sum((lineitem.l_extendedprice * (1:Int32 - lineitem.l_discount)))] }
                         └─StreamHashAgg { group_key: [lineitem.l_suppkey], aggs: [count, sum((lineitem.l_extendedprice * (1:Int32 - lineitem.l_discount)))] }
                           └─StreamExchange { dist: HashShard(lineitem.l_suppkey) }
diff --git a/src/frontend/src/optimizer/mod.rs b/src/frontend/src/optimizer/mod.rs
@@ -429,6 +429,13 @@ impl PlanRoot {
             ApplyOrder::TopDown,
         );
 
+        plan = self.optimize_by_rules(
+            plan,
+            "Agg on Index".to_string(),
+            vec![TopNOnIndexRule::create()],
+            ApplyOrder::TopDown,
+        );
+
         #[cfg(debug_assertions)]
         InputRefValidator.validate(plan.clone());
 
diff --git a/src/frontend/src/optimizer/plan_node/batch_limit.rs b/src/frontend/src/optimizer/plan_node/batch_limit.rs
@@ -49,8 +49,15 @@ impl BatchLimit {
         let new_offset = 0;
         let logical_partial_limit = LogicalLimit::new(input, new_limit, new_offset);
         let batch_partial_limit = Self::new(logical_partial_limit);
-        let ensure_single_dist = RequiredDist::single()
-            .enforce_if_not_satisfies(batch_partial_limit.into(), &Order::any())?;
+        let any_order = Order::any();
+        let ensure_single_dist = RequiredDist::single().enforce_if_not_satisfies(
+            batch_partial_limit.into(),
+            if self.order().field_order.is_empty() {
+                &any_order
+            } else {
+                self.order()
+            },
+        )?;
         let batch_global_limit = self.clone_with_input(ensure_single_dist);
         Ok(batch_global_limit.into())
     }
diff --git a/src/frontend/src/optimizer/plan_node/batch_seq_scan.rs b/src/frontend/src/optimizer/plan_node/batch_seq_scan.rs
@@ -20,6 +20,7 @@ use risingwave_common::error::Result;
 use risingwave_common::types::ScalarImpl;
 use risingwave_common::util::scan_range::{is_full_range, ScanRange};
 use risingwave_pb::batch_plan::plan_node::NodeBody;
+use risingwave_pb::batch_plan::row_seq_scan_node::ChunkSize;
 use risingwave_pb::batch_plan::{RowSeqScanNode, SysRowSeqScanNode};
 use risingwave_pb::plan_common::ColumnDesc as ProstColumnDesc;
 
@@ -240,6 +241,10 @@ impl ToBatchProst for BatchSeqScan {
                 // To be filled by the scheduler.
                 vnode_bitmap: None,
                 ordered: !self.order().is_any(),
+                chunk_size: self
+                    .logical
+                    .chunk_size()
+                    .map(|chunk_size| ChunkSize { chunk_size }),
             })
         }
     }
diff --git a/src/frontend/src/optimizer/plan_node/generic/scan.rs b/src/frontend/src/optimizer/plan_node/generic/scan.rs
@@ -37,6 +37,8 @@ pub struct Scan {
     pub indexes: Vec<Rc<IndexCatalog>>,
     /// The pushed down predicates. It refers to column indexes of the table.
     pub predicate: Condition,
+    /// Help RowSeqScan executor use a better chunk size
+    pub chunk_size: Option<u32>,
 }
 
 impl Scan {
diff --git a/src/frontend/src/optimizer/plan_node/logical_scan.rs b/src/frontend/src/optimizer/plan_node/logical_scan.rs
diff --git a/src/frontend/src/optimizer/rule/mod.rs b/src/frontend/src/optimizer/rule/mod.rs
diff --git a/src/frontend/src/optimizer/rule/top_n_on_index_rule.rs b/src/frontend/src/optimizer/rule/top_n_on_index_rule.rs

Original file line number	Diff line number	Diff line change
`@@ -37,6 +37,8 @@ pub struct Scan {`
`37`	`37`	`pub indexes: Vec<Rc<IndexCatalog>>,`
`38`	`38`	`/// The pushed down predicates. It refers to column indexes of the table.`
`39`	`39`	`pub predicate: Condition,`
	`40`	`+ /// Help RowSeqScan executor use a better chunk size`
	`41`	`+ pub chunk_size: Option<u32>,`
`40`	`42`	`}`
`41`	`43`
`42`	`44`	`impl Scan {`