opt: add cost penalty for scans with large cardinality

This commit adds a new cost function, largeCardinalityRowCountPenalty,
which calculates a penalty that should be added to the row count of scans.
It is non-zero for expressions with unbounded maximum cardinality or with
maximum cardinality exceeding the row count estimate. Adding a few rows
worth of cost helps prevent surprising plans for very small tables or for
when stats are stale.

Fixes cockroachdb#64570

Release note (performance improvement): When choosing between index
scans that are estimated to have the same number of rows, the optimizer
now prefers indexes for which it has higher certainty about the maximum
number of rows over indexes for which there is more uncertainty in the
estimated row count. This helps to avoid choosing suboptimal plans for
small tables or if the statistics are stale.

pkg/sql/logictest/testdata/logic_test/alter_primary_key

@@ -118,7 +118,7 @@ SELECT * FROM parent
 4 5
 
 query T
-SELECT * FROM [EXPLAIN SELECT * FROM child WHERE x >= 1 AND x < 5 AND y >= 2 AND y <= 6] OFFSET 2
+SELECT * FROM [EXPLAIN SELECT * FROM child@primary WHERE x >= 1 AND x < 5 AND y >= 2 AND y <= 6] OFFSET 2
 ----
 ·
 • filter
@@ -130,7 +130,7 @@ SELECT * FROM [EXPLAIN SELECT * FROM child WHERE x >= 1 AND x < 5 AND y >= 2 AND
       spans: [/1/2 - /4/6]
 
 query III rowsort
-SELECT * FROM child WHERE x >= 1 AND x < 5 AND y >= 2 AND y <= 6
+SELECT * FROM child@primary WHERE x >= 1 AND x < 5 AND y >= 2 AND y <= 6
 ----
 1 2 3
 4 5 6

pkg/sql/logictest/testdata/logic_test/prepare

@@ -1171,14 +1171,14 @@ select
  ├── columns: k:1 str:2
  ├── immutable
  ├── stats: [rows=333.333333]
- ├── cost: 1064.43
+ ├── cost: 1074.83
  ├── key: (1)
  ├── fd: (1)-->(2)
  ├── prune: (2)
  ├── scan t2
  │    ├── columns: k:1 str:2
  │    ├── stats: [rows=1000]
- │    ├── cost: 1054.41
+ │    ├── cost: 1064.81
  │    ├── key: (1)
  │    ├── fd: (1)-->(2)
  │    ├── prune: (1,2)

pkg/sql/logictest/testdata/logic_test/vectorize_local

@@ -104,35 +104,35 @@ EXPLAIN (OPT, VERBOSE) SELECT c.a FROM c INNER MERGE JOIN d ON c.a = d.b
 project
  ├── columns: a:1
  ├── stats: [rows=10]
- ├── cost: 1100.77
+ ├── cost: 1122.17
  ├── prune: (1)
  └── inner-join (merge)
       ├── columns: c.a:1 d.b:8
       ├── flags: force merge join
       ├── left ordering: +1
       ├── right ordering: +8
       ├── stats: [rows=10, distinct(1)=1, null(1)=0, distinct(8)=1, null(8)=0]
-      ├── cost: 1100.66
+      ├── cost: 1122.06
       ├── fd: (1)==(8), (8)==(1)
       ├── sort
       │    ├── columns: c.a:1
       │    ├── stats: [rows=1, distinct(1)=1, null(1)=0]
-      │    ├── cost: 15.93
+      │    ├── cost: 26.73
       │    ├── ordering: +1
       │    ├── prune: (1)
       │    ├── interesting orderings: (+1)
       │    ├── unfiltered-cols: (1-6)
       │    └── scan c@sec
       │         ├── columns: c.a:1
       │         ├── stats: [rows=1, distinct(1)=1, null(1)=0]
-      │         ├── cost: 15.89
+      │         ├── cost: 26.69
       │         ├── prune: (1)
       │         ├── interesting orderings: (+1)
       │         └── unfiltered-cols: (1-6)
       ├── scan d
       │    ├── columns: d.b:8
       │    ├── stats: [rows=1000, distinct(8)=100, null(8)=0]
-      │    ├── cost: 1074.61
+      │    ├── cost: 1085.21
       │    ├── ordering: +8
       │    ├── prune: (8)
       │    ├── interesting orderings: (+8)

pkg/sql/opt/exec/execbuilder/testdata/aggregate

@@ -670,7 +670,7 @@ vectorized: true
 • group (scalar)
 │ columns: (min int)
 │ estimated row count: 1 (missing stats)
-│ aggregate 0: min(x)
+│ aggregate 0: any_not_null(x)
 │
 └── • project
     │ columns: (x int)
@@ -680,6 +680,7 @@ vectorized: true
           estimated row count: 1 (missing stats)
           table: xyz@zyx
           spans: /3/2-/3/3
+          limit: 1
 
 statement ok
 SET tracing = on,kv,results; SELECT min(x) FROM xyz WHERE (y, z) = (2, 3.0); SET tracing = off
@@ -701,16 +702,17 @@ vectorized: true
 • group (scalar)
 │ columns: (max int)
 │ estimated row count: 1 (missing stats)
-│ aggregate 0: max(x)
+│ aggregate 0: any_not_null(x)
 │
 └── • project
     │ columns: (x int)
     │
-    └── • scan
+    └── • revscan
           columns: (x int, y int, z float)
           estimated row count: 1 (missing stats)
           table: xyz@zyx
           spans: /3/2-/3/3
+          limit: 1
 
 # VARIANCE/STDDEV
 

pkg/sql/opt/exec/execbuilder/testdata/distsql_agg

@@ -963,15 +963,15 @@ group-by
  ├── grouping columns: b:2
  ├── internal-ordering: +2 opt(1)
  ├── stats: [rows=9.5617925, distinct(2)=9.5617925, null(2)=0]
- ├── cost: 15.1256179
+ ├── cost: 25.9256179
  ├── key: (2)
  ├── fd: (2)-->(5)
  ├── prune: (5)
  ├── scan data2
  │    ├── columns: a:1 b:2
  │    ├── constraint: /1/2: [/1 - /1]
  │    ├── stats: [rows=10, distinct(1)=1, null(1)=0, distinct(2)=9.5617925, null(2)=0]
- │    ├── cost: 14.81
+ │    ├── cost: 25.61
  │    ├── key: (2)
  │    ├── fd: ()-->(1)
  │    ├── ordering: +2 opt(1) [actual: +2]

pkg/sql/opt/exec/execbuilder/testdata/explain

@@ -1097,21 +1097,21 @@ EXPLAIN (OPT,VERBOSE) SELECT * FROM tc WHERE a = 10 ORDER BY b
 sort
  ├── columns: a:1 b:2
  ├── stats: [rows=10, distinct(1)=1, null(1)=0]
- ├── cost: 76.7943856
+ ├── cost: 87.5943856
  ├── fd: ()-->(1)
  ├── ordering: +2 opt(1) [actual: +2]
  ├── prune: (2)
  └── index-join tc
       ├── columns: a:1 b:2
       ├── stats: [rows=10, distinct(1)=1, null(1)=0]
-      ├── cost: 75.72
+      ├── cost: 86.52
       ├── fd: ()-->(1)
       ├── prune: (2)
       └── scan tc@c
            ├── columns: a:1 rowid:3
            ├── constraint: /1/3: [/10 - /10]
            ├── stats: [rows=10, distinct(1)=1, null(1)=0]
-           ├── cost: 14.81
+           ├── cost: 25.61
            ├── key: (3)
            └── fd: ()-->(1)
 
@@ -1121,21 +1121,21 @@ EXPLAIN (OPT,TYPES) SELECT * FROM tc WHERE a = 10 ORDER BY b
 sort
  ├── columns: a:1(int!null) b:2(int)
  ├── stats: [rows=10, distinct(1)=1, null(1)=0]
- ├── cost: 76.7943856
+ ├── cost: 87.5943856
  ├── fd: ()-->(1)
  ├── ordering: +2 opt(1) [actual: +2]
  ├── prune: (2)
  └── index-join tc
       ├── columns: a:1(int!null) b:2(int)
       ├── stats: [rows=10, distinct(1)=1, null(1)=0]
-      ├── cost: 75.72
+      ├── cost: 86.52
       ├── fd: ()-->(1)
       ├── prune: (2)
       └── scan tc@c
            ├── columns: a:1(int!null) rowid:3(int!null)
            ├── constraint: /1/3: [/10 - /10]
            ├── stats: [rows=10, distinct(1)=1, null(1)=0]
-           ├── cost: 14.81
+           ├── cost: 25.61
            ├── key: (3)
            └── fd: ()-->(1)
 
@@ -1183,20 +1183,20 @@ inner-join (hash)
  ├── columns: a:1 b:2 k:6 v:7
  ├── multiplicity: left-rows(zero-or-one), right-rows(zero-or-more)
  ├── stats: [rows=990, distinct(1)=99, null(1)=0, distinct(6)=99, null(6)=0]
- ├── cost: 2249.87625
+ ├── cost: 2271.67625
  ├── fd: (6)-->(7), (1)==(6), (6)==(1)
  ├── prune: (2,7)
  ├── scan tc
  │    ├── columns: a:1 b:2
  │    ├── stats: [rows=1000, distinct(1)=100, null(1)=10]
- │    ├── cost: 1115.01
+ │    ├── cost: 1126.01
  │    ├── prune: (1,2)
  │    ├── interesting orderings: (+1)
  │    └── unfiltered-cols: (1-5)
  ├── scan t
  │    ├── columns: k:6 v:7
  │    ├── stats: [rows=1000, distinct(6)=1000, null(6)=0]
- │    ├── cost: 1094.81
+ │    ├── cost: 1105.61
  │    ├── key: (6)
  │    ├── fd: (6)-->(7)
  │    ├── prune: (6,7)
@@ -1224,7 +1224,7 @@ sort
  ├── columns: a:1 b:2  [hidden: column6:6]
  ├── immutable
  ├── stats: [rows=333.333333]
- ├── cost: 1204.26951
+ ├── cost: 1215.26951
  ├── fd: (1,2)-->(6)
  ├── ordering: +6
  ├── prune: (1,2,6)
@@ -1233,20 +1233,20 @@ sort
       ├── columns: column6:6 a:1 b:2
       ├── immutable
       ├── stats: [rows=333.333333]
-      ├── cost: 1131.70667
+      ├── cost: 1142.70667
       ├── fd: (1,2)-->(6)
       ├── prune: (1,2,6)
       ├── interesting orderings: (+1)
       ├── select
       │    ├── columns: a:1 b:2
       │    ├── immutable
       │    ├── stats: [rows=333.333333]
-      │    ├── cost: 1125.03
+      │    ├── cost: 1136.03
       │    ├── interesting orderings: (+1)
       │    ├── scan tc
       │    │    ├── columns: a:1 b:2
       │    │    ├── stats: [rows=1000]
-      │    │    ├── cost: 1115.01
+      │    │    ├── cost: 1126.01
       │    │    ├── prune: (1,2)
       │    │    └── interesting orderings: (+1)
       │    └── filters
@@ -1261,7 +1261,7 @@ sort
  ├── columns: a:1(int) b:2(int)  [hidden: column6:6(int)]
  ├── immutable
  ├── stats: [rows=333.333333]
- ├── cost: 1204.26951
+ ├── cost: 1215.26951
  ├── fd: (1,2)-->(6)
  ├── ordering: +6
  ├── prune: (1,2,6)
@@ -1270,20 +1270,20 @@ sort
       ├── columns: column6:6(int) a:1(int) b:2(int)
       ├── immutable
       ├── stats: [rows=333.333333]
-      ├── cost: 1131.70667
+      ├── cost: 1142.70667
       ├── fd: (1,2)-->(6)
       ├── prune: (1,2,6)
       ├── interesting orderings: (+1)
       ├── select
       │    ├── columns: a:1(int) b:2(int)
       │    ├── immutable
       │    ├── stats: [rows=333.333333]
-      │    ├── cost: 1125.03
+      │    ├── cost: 1136.03
       │    ├── interesting orderings: (+1)
       │    ├── scan tc
       │    │    ├── columns: a:1(int) b:2(int)
       │    │    ├── stats: [rows=1000]
-      │    │    ├── cost: 1115.01
+      │    │    ├── cost: 1126.01
       │    │    ├── prune: (1,2)
       │    │    └── interesting orderings: (+1)
       │    └── filters

pkg/sql/opt/exec/execbuilder/testdata/inverted_index

@@ -1468,7 +1468,7 @@ inner-join (lookup geo_table)
  ├── lookup columns are key
  ├── immutable
  ├── stats: [rows=9801]
- ├── cost: 112694.84
+ ├── cost: 112705.64
  ├── key: (1,5)
  ├── fd: (1)-->(2), (5)-->(6)
  ├── prune: (1,5)
@@ -1477,13 +1477,13 @@ inner-join (lookup geo_table)
  │    ├── inverted-expr
  │    │    └── st_intersects(geo_table2.geom:2, geo_table.geom:12)
  │    ├── stats: [rows=10000, distinct(1)=999.956829, null(1)=0, distinct(11)=999.956829, null(11)=0]
- │    ├── cost: 41794.82
+ │    ├── cost: 41805.62
  │    ├── key: (1,11)
  │    ├── fd: (1)-->(2)
  │    ├── scan geo_table2
  │    │    ├── columns: geo_table2.k:1 geo_table2.geom:2
  │    │    ├── stats: [rows=1000, distinct(1)=1000, null(1)=0, distinct(2)=100, null(2)=10]
- │    │    ├── cost: 1094.81
+ │    │    ├── cost: 1105.61
  │    │    ├── key: (1)
  │    │    ├── fd: (1)-->(2)
  │    │    ├── prune: (1,2)
@@ -1558,7 +1558,7 @@ left-join (lookup geo_table)
  ├── second join in paired joiner
  ├── immutable
  ├── stats: [rows=10000]
- ├── cost: 112894.84
+ ├── cost: 112905.64
  ├── key: (1,5)
  ├── fd: (1)-->(2), (5)-->(6)
  ├── prune: (1,5)
@@ -1568,13 +1568,13 @@ left-join (lookup geo_table)
  │    ├── inverted-expr
  │    │    └── st_intersects(geo_table2.geom:2, geo_table.geom:12)
  │    ├── stats: [rows=10000, distinct(1)=1000, null(1)=0, distinct(11)=999.956829, null(11)=0]
- │    ├── cost: 41994.82
+ │    ├── cost: 42005.62
  │    ├── key: (1,11)
  │    ├── fd: (1)-->(2), (11)-->(16)
  │    ├── scan geo_table2
  │    │    ├── columns: geo_table2.k:1 geo_table2.geom:2
  │    │    ├── stats: [rows=1000, distinct(1)=1000, null(1)=0]
- │    │    ├── cost: 1094.81
+ │    │    ├── cost: 1105.61
  │    │    ├── key: (1)
  │    │    ├── fd: (1)-->(2)
  │    │    ├── prune: (1,2)
@@ -1594,7 +1594,7 @@ semi-join (lookup geo_table)
  ├── second join in paired joiner
  ├── immutable
  ├── stats: [rows=10]
- ├── cost: 112694.84
+ ├── cost: 112705.64
  ├── key: (1)
  ├── fd: (1)-->(2)
  ├── prune: (1)
@@ -1604,13 +1604,13 @@ semi-join (lookup geo_table)
  │    ├── inverted-expr
  │    │    └── st_intersects(geo_table2.geom:2, geo_table.geom:12)
  │    ├── stats: [rows=10000, distinct(1)=999.956829, null(1)=0, distinct(11)=999.956829, null(11)=0]
- │    ├── cost: 41994.82
+ │    ├── cost: 42005.62
  │    ├── key: (1,11)
  │    ├── fd: (1)-->(2), (11)-->(16)
  │    ├── scan geo_table2
  │    │    ├── columns: geo_table2.k:1 geo_table2.geom:2
  │    │    ├── stats: [rows=1000, distinct(1)=1000, null(1)=0, distinct(2)=100, null(2)=10]
- │    │    ├── cost: 1094.81
+ │    │    ├── cost: 1105.61
  │    │    ├── key: (1)
  │    │    ├── fd: (1)-->(2)
  │    │    ├── prune: (1,2)
@@ -1631,7 +1631,7 @@ anti-join (lookup geo_table)
  ├── second join in paired joiner
  ├── immutable
  ├── stats: [rows=990]
- ├── cost: 112694.84
+ ├── cost: 112705.64
  ├── key: (1)
  ├── fd: (1)-->(2)
  ├── prune: (1)
@@ -1641,13 +1641,13 @@ anti-join (lookup geo_table)
  │    ├── inverted-expr
  │    │    └── st_intersects(geo_table2.geom:2, geo_table.geom:12)
  │    ├── stats: [rows=10000, distinct(1)=1000, null(1)=0, distinct(11)=999.956829, null(11)=0]
- │    ├── cost: 41994.82
+ │    ├── cost: 42005.62
  │    ├── key: (1,11)
  │    ├── fd: (1)-->(2), (11)-->(16)
  │    ├── scan geo_table2
  │    │    ├── columns: geo_table2.k:1 geo_table2.geom:2
  │    │    ├── stats: [rows=1000, distinct(1)=1000, null(1)=0]
- │    │    ├── cost: 1094.81
+ │    │    ├── cost: 1105.61
  │    │    ├── key: (1)
  │    │    ├── fd: (1)-->(2)
  │    │    ├── prune: (1,2)