From 67a1d4d7b14baaba74c50dc642dd9022b774e98b Mon Sep 17 00:00:00 2001
From: Rebecca Taft <becca@cockroachlabs.com>
Date: Wed, 29 Apr 2020 10:42:06 -0500
Subject: [PATCH] opt: add rule to eliminate Exists when input has zero rows

This commit adds a new rule, EliminateExistsZeroRows, which
converts an Exists subquery to False when it's known
that the input produces zero rows.

Informs #47058

Release note (performance improvement): The optimizer can now
detect when an Exists subquery can be eliminated because the input
has zero rows. This leads to better plans in some cases.
---
 pkg/sql/opt/norm/rules/scalar.opt      |  5 ++++
 pkg/sql/opt/norm/testdata/rules/join   |  4 +--
 pkg/sql/opt/norm/testdata/rules/scalar | 34 ++++++++++++++------------
 3 files changed, 26 insertions(+), 17 deletions(-)

diff --git a/pkg/sql/opt/norm/rules/scalar.opt b/pkg/sql/opt/norm/rules/scalar.opt
index ad5e38003f0e..1d9f2edef4f3 100644
--- a/pkg/sql/opt/norm/rules/scalar.opt
+++ b/pkg/sql/opt/norm/rules/scalar.opt
@@ -90,6 +90,11 @@
     $result
 )
 
+# EliminateExistsZeroRows converts an Exists subquery to False when it's known
+# that the input produces zero rows.
+[EliminateExistsZeroRows, Normalize]
+(Exists $input:* & (HasZeroRows $input)) => (False)
+
 # EliminateExistsProject discards a Project input to the Exists operator. The
 # Project operator never changes the row cardinality of its input, and row
 # cardinality is the only thing that Exists cares about, so Project is a no-op.
diff --git a/pkg/sql/opt/norm/testdata/rules/join b/pkg/sql/opt/norm/testdata/rules/join
index 9e1685bbd37e..7749f0201d4f 100644
--- a/pkg/sql/opt/norm/testdata/rules/join
+++ b/pkg/sql/opt/norm/testdata/rules/join
@@ -2165,7 +2165,7 @@ scan a
 # SimplifyZeroCardinalitySemiJoin
 # --------------------------------------------------
 # TODO(justin): figure out if there's a good way to make this still apply.
-norm disable=SimplifyZeroCardinalityGroup expect=SimplifyZeroCardinalitySemiJoin
+norm disable=(SimplifyZeroCardinalityGroup,EliminateExistsZeroRows) expect=SimplifyZeroCardinalitySemiJoin
 SELECT * FROM a WHERE EXISTS(SELECT * FROM (VALUES (k)) OFFSET 1)
 ----
 values
@@ -2178,7 +2178,7 @@ values
 # EliminateAntiJoin
 # --------------------------------------------------
 # TODO(justin): figure out if there's a good way to make this still apply.
-norm disable=SimplifyZeroCardinalityGroup expect=EliminateAntiJoin
+norm disable=(SimplifyZeroCardinalityGroup,EliminateExistsZeroRows) expect=EliminateAntiJoin
 SELECT * FROM a WHERE NOT EXISTS(SELECT * FROM (VALUES (k)) OFFSET 1)
 ----
 scan a
diff --git a/pkg/sql/opt/norm/testdata/rules/scalar b/pkg/sql/opt/norm/testdata/rules/scalar
index 772d554e7247..7e729f43d138 100644
--- a/pkg/sql/opt/norm/testdata/rules/scalar
+++ b/pkg/sql/opt/norm/testdata/rules/scalar
@@ -249,6 +249,20 @@ values
  ├── fd: ()-->(1)
  └── (true IN (NULL, NULL, ('201.249.149.90/18' & '97a7:3650:3dd8:d4e9:35fe:6cfb:a714:1c17/61') << 'e22f:2067:2ed2:7b07:b167:206f:f17b:5b7d/82'),)
 
+# --------------------------------------------------
+# EliminateExistsZeroRows
+# --------------------------------------------------
+
+norm expect=EliminateExistsZeroRows
+SELECT EXISTS(SELECT * FROM (VALUES (1)) WHERE false)
+----
+values
+ ├── columns: exists:2!null
+ ├── cardinality: [1 - 1]
+ ├── key: ()
+ ├── fd: ()-->(2)
+ └── (false,)
+
 # --------------------------------------------------
 # EliminateExistsProject
 # --------------------------------------------------
@@ -524,21 +538,11 @@ anti-join (hash)
 norm expect-not=EliminateExistsLimit
 SELECT * FROM a a1 WHERE EXISTS(SELECT i FROM a a2 where a1.i = a2.i LIMIT 0)
 ----
-select
- ├── columns: k:1!null i:2 f:3 s:4 j:5 arr:6
- ├── key: (1)
- ├── fd: (1)-->(2-6)
- ├── scan a1
- │    ├── columns: a1.k:1!null a1.i:2 a1.f:3 a1.s:4 a1.j:5 a1.arr:6
- │    ├── key: (1)
- │    └── fd: (1)-->(2-6)
- └── filters
-      └── exists [subquery]
-           └── values
-                ├── columns: a2.i:8!null
-                ├── cardinality: [0 - 0]
-                ├── key: ()
-                └── fd: ()-->(8)
+values
+ ├── columns: k:1!null i:2!null f:3!null s:4!null j:5!null arr:6!null
+ ├── cardinality: [0 - 0]
+ ├── key: ()
+ └── fd: ()-->(1-6)
 
 # Don't eliminate a limit from a non-correlated subquery.
 norm expect-not=EliminateExistsLimit