indexrec: include != and IS DISTINCT FROM and IS NOT NULL columns in …

…index recommendations Fixes #102206 As for the tests, since !=, IS DISTINT FROM and IS NOT are categorized as range index, so I copied similar tests with "R": R, EQ + R, J + R, EQ + J + R. Release note: None
cockroachdb · Jun 1, 2023 · 5f50b7a · 5f50b7a
1 parent 78d177b
commit 5f50b7a
Show file tree

Hide file tree

Showing 2 changed files with 218 additions and 1 deletion.
diff --git a/pkg/sql/opt/indexrec/candidate.go b/pkg/sql/opt/indexrec/candidate.go
@@ -29,7 +29,7 @@ import (
 //     indexes by table. For Order By, the index column ordering and column
 //     directions are the same as how it is in the Order By.
 //  2. Add a single-column index on any Range expression, comparison
-//     expression (=, <, >, <=, >=), and IS expression.
+//     expression (=, !=, <, >, <=, >=), IS and IS NOT expression.
 //  3. Add a single-column index on any column that appears in a JOIN predicate.
 //  4. If there exist multiple columns from the same table in a JOIN predicate,
 //     create a single index on all such columns.
@@ -123,8 +123,13 @@ func (ics *indexCandidateSet) categorizeIndexCandidates(expr opt.Expr) {
 	case *memo.EqExpr:
 		ics.addVariableExprIndex(expr.Left, ics.equalCandidates)
 		ics.addVariableExprIndex(expr.Right, ics.equalCandidates)
+	case *memo.NeExpr:
+		ics.addVariableExprIndex(expr.Left, ics.rangeCandidates)
+		ics.addVariableExprIndex(expr.Right, ics.rangeCandidates)
 	case *memo.IsExpr:
 		ics.addVariableExprIndex(expr.Left, ics.equalCandidates)
+	case *memo.IsNotExpr:
+		ics.addVariableExprIndex(expr.Left, ics.rangeCandidates)
 	case *memo.LtExpr:
 		ics.addVariableExprIndex(expr.Left, ics.rangeCandidates)
 		ics.addVariableExprIndex(expr.Right, ics.rangeCandidates)

diff --git a/pkg/sql/opt/indexrec/testdata/index b/pkg/sql/opt/indexrec/testdata/index
@@ -275,6 +275,42 @@ scan t1@_hyp_1
  ├── cost: 28.6200001
  └── fd: ()-->(4)
 
+index-candidates
+SELECT i FROM t1 WHERE i != 5
+----
+t1:
+ (i)
+
+index-recommendations
+SELECT i FROM t1 WHERE i != 5
+----
+creation: CREATE INDEX ON t.public.t1 (i);
+--
+optimal plan:
+scan t1@_hyp_1
+ ├── columns: i:2!null
+ ├── constraint: /2/5
+ │    ├── (/NULL - /4]
+ │    └── [/6 - ]
+ └── cost: 375.353333
+
+index-candidates
+SELECT s FROM t1 WHERE s IS NOT NULL
+----
+t1:
+ (s)
+
+index-recommendations
+SELECT s FROM t1 WHERE s IS NOT NULL
+----
+creation: CREATE INDEX ON t.public.t1 (s);
+--
+optimal plan:
+scan t1@_hyp_1
+ ├── columns: s:4!null
+ ├── constraint: /4/5: (/NULL - ]
+ └── cost: 1067.42
+
 index-candidates
 SELECT t1.k FROM t1 JOIN t2 ON t1.k = t2.i
 ----
@@ -624,6 +660,32 @@ scan t1@_hyp_3
  ├── cost: 28.1933333
  └── fd: ()-->(1)
 
+index-candidates
+SELECT * FROM t1 WHERE k = 1 AND f != 0
+----
+t1:
+ (f)
+ (k)
+ (k, f)
+
+index-recommendations
+SELECT * FROM t1 WHERE k = 1 AND f != 0
+----
+creation: CREATE INDEX ON t.public.t1 (k) STORING (i, f, s);
+--
+optimal plan:
+select
+ ├── columns: k:1!null i:2 f:3!null s:4
+ ├── cost: 29.0500001
+ ├── fd: ()-->(1)
+ ├── scan t1@_hyp_1
+ │    ├── columns: k:1!null i:2 f:3 s:4
+ │    ├── constraint: /1/5: [/1 - /1]
+ │    ├── cost: 28.9200001
+ │    └── fd: ()-->(1)
+ └── filters
+      └── f:3 != 0.0 [outer=(3), constraints=(/3: (/NULL - /-5e-324] [/5e-324 - ]; tight)]
+
 # Multi-column combinations used: EQ, EQ + R.
 index-candidates
 SELECT  k, i, f FROM t1 WHERE k = 1 AND i = 2 AND f > 0
@@ -654,6 +716,28 @@ inner-join (zigzag t1@_hyp_1 t1@_hyp_2)
       ├── i:2 = 2 [outer=(2), constraints=(/2: [/2 - /2]; tight), fd=()-->(2)]
       └── f:3 > 0.0 [outer=(3), constraints=(/3: [/5e-324 - ]; tight)]
 
+index-candidates
+SELECT  k, i, s FROM t1 WHERE k = 1 AND i = 2 AND s IS NOT NULL
+----
+t1:
+ (i)
+ (k)
+ (k, i)
+ (k, i, s)
+ (s)
+
+index-recommendations
+SELECT k, i, s FROM t1 WHERE k = 1 AND i = 2 AND s IS NOT NULL
+----
+creation: CREATE INDEX ON t.public.t1 (k, i, s);
+--
+optimal plan:
+scan t1@_hyp_5
+ ├── columns: k:1!null i:2!null s:4!null
+ ├── constraint: /1/2/4/5: (/1/2/NULL - /1/2]
+ ├── cost: 18.907147
+ └── fd: ()-->(1,2)
+
 # Multi-column combinations used: J + R.
 index-candidates
 SELECT t1.k, t1.f FROM t1 JOIN t2 ON t1.k != t2.k WHERE t1.f > 0
@@ -687,6 +771,39 @@ project
       └── filters
            └── t1.k:1 != t2.k:8 [outer=(1,8), constraints=(/1: (/NULL - ]; /8: (/NULL - ])]
 
+index-candidates
+SELECT t1.k, t1.s FROM t1 JOIN t2 ON t1.k = t2.k WHERE t1.s IS NOT NULL
+----
+t1:
+ (k)
+ (k, s)
+ (s)
+t2:
+ (k)
+
+index-recommendations
+SELECT t1.k, t1.s FROM t1 JOIN t2 ON t1.k = t2.k WHERE t1.s IS NOT NULL
+----
+creation: CREATE INDEX ON t.public.t1 (s) STORING (k);
+--
+optimal plan:
+project
+ ├── columns: k:1!null s:4!null
+ ├── cost: 2379.90804
+ └── inner-join (hash)
+      ├── columns: t1.k:1!null t1.s:4!null t2.k:8!null
+      ├── cost: 2282.85814
+      ├── fd: (1)==(8), (8)==(1)
+      ├── scan t2
+      │    ├── columns: t2.k:8
+      │    └── cost: 1078.52
+      ├── scan t1@_hyp_1
+      │    ├── columns: t1.k:1 t1.s:4!null
+      │    ├── constraint: /4/5: (/NULL - ]
+      │    └── cost: 1077.32
+      └── filters
+           └── t1.k:1 = t2.k:8 [outer=(1,8), constraints=(/1: (/NULL - ]; /8: (/NULL - ]), fd=(1)==(8), (8)==(1)]
+
 # Multi-column combinations used: EQ, EQ + J.
 index-candidates
 SELECT t1.i, t1.s FROM t1 JOIN t2 ON t1.k != t2.k WHERE t1.i = 2 AND t1.s = 'NG'
@@ -956,6 +1073,101 @@ union-all
       └── aggregations
            └── count-rows [as=count_rows:22]
 
+index-candidates
+SELECT count(*)
+FROM t1 LEFT JOIN t2
+ON t1.k = t2.k
+GROUP BY t2.s, t2.i
+UNION ALL
+SELECT count(*)
+FROM (
+  SELECT *
+  FROM t1
+  WHERE t1.f != 0
+  AND t1.s = 'NG'
+)
+----
+t1:
+ (f)
+ (k)
+ (k, f)
+ (s)
+ (s, f)
+ (s, k)
+ (s, k, f)
+t2:
+ (i, s)
+ (k)
+
+index-recommendations
+SELECT count(*)
+FROM t1 LEFT JOIN t2
+ON t1.k = t2.k
+GROUP BY t2.s, t2.i
+UNION ALL
+SELECT count(*)
+FROM (
+  SELECT *
+  FROM t1
+  WHERE t1.f != 0
+  AND t1.s = 'NG'
+)
+----
+creation: CREATE INDEX ON t.public.t1 (k);
+creation: CREATE INDEX ON t.public.t1 (s) STORING (f);
+creation: CREATE INDEX ON t.public.t2 (k) STORING (i, s);
+--
+optimal plan:
+union-all
+ ├── columns: count:23!null
+ ├── left columns: count_rows:14
+ ├── right columns: count_rows:22
+ ├── cardinality: [1 - ]
+ ├── cost: 2766.55828
+ ├── project
+ │    ├── columns: count_rows:14!null
+ │    ├── cost: 2727.55537
+ │    └── group-by (hash)
+ │         ├── columns: t2.i:9 t2.s:10 count_rows:14!null
+ │         ├── grouping columns: t2.i:9 t2.s:10
+ │         ├── cost: 2717.53581
+ │         ├── key: (9,10)
+ │         ├── fd: (9,10)-->(14)
+ │         ├── left-join (merge)
+ │         │    ├── columns: t1.k:1 t2.k:8 t2.i:9 t2.s:10
+ │         │    ├── left ordering: +1
+ │         │    ├── right ordering: +8
+ │         │    ├── cost: 2307.36
+ │         │    ├── scan t1@_hyp_3
+ │         │    │    ├── columns: t1.k:1
+ │         │    │    ├── cost: 1088.62
+ │         │    │    └── ordering: +1
+ │         │    ├── scan t2@_hyp_2
+ │         │    │    ├── columns: t2.k:8 t2.i:9 t2.s:10
+ │         │    │    ├── cost: 1098.72
+ │         │    │    └── ordering: +8
+ │         │    └── filters (true)
+ │         └── aggregations
+ │              └── count-rows [as=count_rows:14]
+ └── scalar-group-by
+      ├── columns: count_rows:22!null
+      ├── cardinality: [1 - 1]
+      ├── cost: 28.9733334
+      ├── key: ()
+      ├── fd: ()-->(22)
+      ├── select
+      │    ├── columns: f:17!null t1.s:18!null
+      │    ├── cost: 28.8500001
+      │    ├── fd: ()-->(18)
+      │    ├── scan t1@_hyp_1
+      │    │    ├── columns: f:17 t1.s:18!null
+      │    │    ├── constraint: /18/19: [/'NG' - /'NG']
+      │    │    ├── cost: 28.7200001
+      │    │    └── fd: ()-->(18)
+      │    └── filters
+      │         └── f:17 != 0.0 [outer=(17), constraints=(/17: (/NULL - /-5e-324] [/5e-324 - ]; tight)]
+      └── aggregations
+           └── count-rows [as=count_rows:22]
 
 # No rule 5 multi-column index combinations.
 index-candidates