opt: fix statistics estimation for semi and anti joins

Prior to this commit, the statisticsBuilder always estimated that
the number of output rows for a semi or anti join was equal to the
number of rows on the left side. It ignored any ON conditions.
This commit improves the estimate by taking into account the ON
conditions.

Release note: None

pkg/sql/opt/exec/execbuilder/testdata/lookup_join

@@ -244,8 +244,6 @@ SELECT start_key, end_key, replicas, lease_holder from [SHOW RANGES FROM TABLE b
 start_key  end_key  replicas  lease_holder
 NULL       NULL     {5}       5
 
-# TODO(radu): this doesn't seem to be a lookup join, but it should be.
-
 query T
 SELECT url FROM [EXPLAIN (DISTSQL) SELECT DISTINCT authors.name FROM books AS b1, books2 AS b2, authors WHERE b1.title = b2.title AND authors.book = b1.title AND b1.shelf <> b2.shelf]
 ----
@@ -675,18 +673,18 @@ render            ·         ·            (a, b, c)  ·
 query TTTTT
 EXPLAIN (VERBOSE) SELECT * from abc WHERE EXISTS (SELECT * FROM def WHERE a=f AND c=e)
 ----
-render            ·                      ·                (a, b, c)        ·
- │                render 0               a                ·                ·
- │                render 1               b                ·                ·
- │                render 2               c                ·                ·
- └── lookup-join  ·                      ·                (a, b, c, e, f)  ·
-      │           table                  def@primary      ·                ·
-      │           type                   inner            ·                ·
-      │           equality               (a, c) = (f, e)  ·                ·
-      │           equality cols are key  ·                ·                ·
-      └── scan    ·                      ·                (a, b, c)        ·
-·                 table                  abc@primary      ·                ·
-·                 spans                  ALL              ·                ·
+render            ·                      ·                (a, b, c)  ·
+ │                render 0               a                ·          ·
+ │                render 1               b                ·          ·
+ │                render 2               c                ·          ·
+ └── lookup-join  ·                      ·                (a, b, c)  ·
+      │           table                  def@primary      ·          ·
+      │           type                   semi             ·          ·
+      │           equality               (a, c) = (f, e)  ·          ·
+      │           equality cols are key  ·                ·          ·
+      └── scan    ·                      ·                (a, b, c)  ·
+·                 table                  abc@primary      ·          ·
+·                 spans                  ALL              ·          ·
 
 query TTTTT
 EXPLAIN (VERBOSE) SELECT * from abc WHERE NOT EXISTS (SELECT * FROM def WHERE a=f AND c=e)
@@ -741,7 +739,7 @@ SELECT url FROM [ EXPLAIN (DISTSQL)
   SELECT a,b from small WHERE EXISTS (SELECT a FROM data WHERE small.a=data.a) ORDER BY a
 ]
 ----
-https://cockroachdb.github.io/distsqlplan/decode.html#eJzMlEFr2zAUx-_7FOJdJ-NIdtLUp1w6SOmS0eY2ctCiR-fN0TOSDCsl3304HqtdWlngQ3q0lD__X35PvGcwpHGjjuig-A4COEjgkAGHHDjMYc-htnRA58i2P-kCa_0HihmH0tSNb4_3HA5kEYpn8KWvEArYqR8V3qPSaNMZcNDoVVmda2pbHpV9Wrmjqirg8FAr4wqWpIIpo5lg5H-iBQ7bxhdsJfhKwv7EgRr_0ue8ekQoxInHM91Saf4hzYdIu6caC3Z382XHHm6-rtntdr0B_p9UK6-Awx3R76Zmv6g0jEyLFgcp34V8YSOr0aIeYq3EZ9if3vgnG0qoTsVQ7Hv12aBexM9NRM0tFUkqp05uhKo3ucXlJifj1ck4dTJJs6nqRqh66q4upy6LV5fFqcuSNJ-qboSqp255OXV5vLo8Tl2epPOp6kaoeuquP8aqfQPyHl1NxmHUFp21exj1I3Z721FjD_jN0uFc031uz7nzgUbnu1vRfaxNd9UC9sMiGJaDsHgdluHmkeosmM7D4XwK9zwYXoSbF1Oar4LhZbh5OaX5Ojyr2cgzCT-y193706e_AQAA__8Shyy4
+https://cockroachdb.github.io/distsqlplan/decode.html#eJzEll9v2jAUxd_3KdB92lSjxE7Cn0iTeNhLJ62dur1NPLjEg0g0RraRVlV89wkyKSKBe-0F6Fsp-XGPfc65yhtUulAP8kVZyH8BBwYCGCTAIAUGGcwZbIxeKGu12T9SA_fFH8hjBmW12br9v-cMFtooyN_AlW6tIIef8nmtnpQslIliYFAoJ8v1YczGlC_SvM7si1yvgcGPjaxsPhhGfCCrYsAH2q2UAQaPW5cPZpzNBMx3DPTW_ZvXjHl-HaykXR0PmHGY7-YMrJNLBTnfsf_TnZ3WXUgnu7JFLds2uluiG0EiRNCX0rqyWriIx51TMng0hTKqQMclZ8c1U3T9O-0Jd_VFnteUBWkK8y3tp5s1T22rU8-dPNo3ZZbqqy4rZSI-Ov7dtfrtPs743afPplyu6j-xmDZHyS5uwYMe6k0keOtEp8ePjsZz_wZwr-ZGfBiJW3Q3QPnIp7uIcM_yEoqaovCLlJe_QwnG1ymB8PdS-KVQDKPkFikMUD72SuF54Z4pJBQ1KRQXSaF4hxROrpPCxN_LxC-FyTBKb5HCAOUTrxSeF-6ZQkJRk8LkIilM3iGF0-ukMPX3MvVLYTqMslukMED51CuFe-HBySNUNMlLL5K89PbJE_H1X0VPSHhSdqMrq7zeMuP9IVSxVPW9WL01C_Xd6MVhTP3x8cAdXmQKZV39bVp_uK_qr_YC_eFpH5gnvehxH1pwnOZtOj6ij-C4DQsUJk6dBLglwuBpH7jlVig97kO33OrQKXrhGX7hGW414fWoT7lwmCgXDlPlImiiXDhNlWuM3vgEv_BJn3JN-wSF91qkBE2YTdCU2xRO2E3g5DLFtyknKsZ77VPeWS1BnndWS5DnOE15jtOk5wROeY7jpOf4WuUjwvPOhgnyHN8wlOedFRPkOU5TnuM06TmBU57jOOk5vl5FTLw2dbZMiOcCXzJtz-e7D38DAAD__xe8Uow=
 
 query T
 SELECT url FROM [ EXPLAIN (DISTSQL)

pkg/sql/opt/memo/extract.go

@@ -178,6 +178,32 @@ func ExtractJoinEqualityColumns(
 	return leftEq, rightEq
 }
 
+// ExtractJoinEqualityFilters returns the filters containing pairs of columns
+// (one from the left side, one from the right side) which are constrained to
+// be equal in a join (and have equivalent types).
+func ExtractJoinEqualityFilters(leftCols, rightCols opt.ColSet, on FiltersExpr) FiltersExpr {
+	// We want to avoid allocating a new slice unless strictly necessary.
+	var newFilters FiltersExpr
+	for i := range on {
+		condition := on[i].Condition
+		ok, _, _ := isJoinEquality(leftCols, rightCols, condition)
+		if ok {
+			if newFilters != nil {
+				newFilters = append(newFilters, on[i])
+			}
+		} else {
+			if newFilters == nil {
+				newFilters = make(FiltersExpr, i, len(on)-1)
+				copy(newFilters, on[:i])
+			}
+		}
+	}
+	if newFilters != nil {
+		return newFilters
+	}
+	return on
+}
+
 func isVarEquality(condition opt.ScalarExpr) (leftVar, rightVar *VariableExpr, ok bool) {
 	if eq, ok := condition.(*EqExpr); ok {
 		if leftVar, ok := eq.Left.(*VariableExpr); ok {

pkg/sql/opt/memo/memo.go

@@ -359,8 +359,7 @@ func (m *Memo) RequestColStat(
 }
 
 // RowsProcessed calculates and returns the number of rows processed by the
-// relational expression. It is currently only supported for lookup joins and
-// merge joins.
+// relational expression. It is currently only supported for joins.
 func (m *Memo) RowsProcessed(expr RelExpr) (_ float64, ok bool) {
 	// When SetRoot is called, the statistics builder may have been cleared.
 	// If this happens, we can't serve the request anymore.