Fix query planning for complex queries with impossible conditions.

Signed-off-by: Arthur Schreiber <[email protected]>

Author: arthurschreiber

go/vt/vtgate/planbuilder/operators/subquery_builder.go

@@ -171,7 +171,9 @@ func createSubquery(
 	sqc := &SubQueryBuilder{totalID: totalID, subqID: subqID, outerID: outerID}
 
 	predicates, joinCols := sqc.inspectStatement(ctx, subq.Select)
-	correlated := !ctx.SemTable.RecursiveDeps(subq).IsEmpty()
+
+	subqDependencies := ctx.SemTable.RecursiveDeps(subq)
+	correlated := subqDependencies.KeepOnly(outerID).NotEmpty()
 
 	opInner := translateQueryToOp(ctx, subq.Select)
 
@@ -209,7 +211,9 @@ func createSubqueryFromPath(
 	sqc := &SubQueryBuilder{totalID: totalID, subqID: subqID, outerID: outerID}
 
 	predicates, joinCols := sqc.inspectStatement(ctx, subq.Select)
-	correlated := !ctx.SemTable.RecursiveDeps(subq).IsEmpty()
+
+	subqDependencies := ctx.SemTable.RecursiveDeps(subq)
+	correlated := subqDependencies.KeepOnly(outerID).NotEmpty()
 
 	opInner := translateQueryToOp(ctx, subq.Select)
 

go/vt/vtgate/planbuilder/testdata/select_cases.json

@@ -4411,7 +4411,7 @@
       ]
     }
   },
-  {
+    {
     "comment": "Subquery with `IN` condition using columns with matching lookup vindexes, with inner scatter query",
     "query": "SELECT music.id FROM music WHERE music.id IN (SELECT music.id FROM music WHERE music.foo = 'bar') AND music.user_id IN (3, 4, 5)",
     "plan": {
@@ -4438,6 +4438,85 @@
     },
     "skip_e2e": true
   },
+  {
+    "comment": "Subquery with `IN` condition using columns with matching lookup vindexes, impossible conditions and limit clause",
+    "query": "SELECT music.id FROM music WHERE music.id IN (SELECT * FROM (SELECT music.id FROM music WHERE music.user_id IN (1, 2, 3) AND 1 = 0 AND music.foo = 'bar' LIMIT 0, 100) _inner)",
+    "plan": {
+      "Type": "Complex",
+      "QueryType": "SELECT",
+      "Original": "SELECT music.id FROM music WHERE music.id IN (SELECT * FROM (SELECT music.id FROM music WHERE music.user_id IN (1, 2, 3) AND 1 = 0 AND music.foo = 'bar' LIMIT 0, 100) _inner)",
+      "Instructions": {
+        "OperatorType": "UncorrelatedSubquery",
+        "Variant": "PulloutIn",
+        "PulloutVars": [
+          "__sq_has_values",
+          "__sq1"
+        ],
+        "Inputs": [
+          {
+            "InputName": "SubQuery",
+            "OperatorType": "Limit",
+            "Count": "100",
+            "Offset": "0",
+            "Inputs": [
+              {
+                "OperatorType": "Route",
+                "Variant": "None",
+                "Keyspace": {
+                  "Name": "user",
+                  "Sharded": true
+                },
+                "FieldQuery": "select id from (select music.id from music where 1 != 1) as _inner where 1 != 1",
+                "Query": "select id from (select music.id from music where 0) as _inner limit 100"
+              }
+            ]
+          },
+          {
+            "InputName": "Outer",
+            "OperatorType": "VindexLookup",
+            "Variant": "IN",
+            "Keyspace": {
+              "Name": "user",
+              "Sharded": true
+            },
+            "Values": [
+              "::__sq1"
+            ],
+            "Vindex": "music_user_map",
+            "Inputs": [
+              {
+                "OperatorType": "Route",
+                "Variant": "IN",
+                "Keyspace": {
+                  "Name": "user",
+                  "Sharded": true
+                },
+                "FieldQuery": "select `name`, keyspace_id from name_user_vdx where 1 != 1",
+                "Query": "select `name`, keyspace_id from name_user_vdx where `name` in ::__vals",
+                "Values": [
+                  "::name"
+                ],
+                "Vindex": "user_index"
+              },
+              {
+                "OperatorType": "Route",
+                "Variant": "ByDestination",
+                "Keyspace": {
+                  "Name": "user",
+                  "Sharded": true
+                },
+                "FieldQuery": "select music.id from music where 1 != 1",
+                "Query": "select music.id from music where :__sq_has_values and music.id in ::__vals"
+              }
+            ]
+          }
+        ]
+      },
+      "TablesUsed": [
+        "user.music"
+      ]
+    }
+  },
   {
     "comment": "Subquery with `IN` condition using columns with matching lookup vindexes",
     "query": "SELECT music.id FROM music WHERE music.id IN (SELECT music.id FROM music WHERE music.user_id IN (1, 2, 3)) and music.user_id = 5",
@@ -6372,10 +6451,10 @@
         },
         "FieldQuery": "select r from (select rank() over (partition by col) as r from `user` where 1 != 1) as t where 1 != 1",
         "Query": "select r from (select rank() over (partition by col) as r from `user` where id = 1) as t",
-        "Vindex": "user_index",
         "Values": [
           "1"
-        ]
+        ],
+        "Vindex": "user_index"
       },
       "TablesUsed": [
         "user.user"

go/vt/vtgate/semantics/semantic_table.go

@@ -544,7 +544,20 @@ func (st *SemTable) CopySemanticInfo(from, to sqlparser.SQLNode) {
 		if !ok {
 			return
 		}
-		st.CopyDependencies(f, t)
+
+		// Not all expressions are valid map keys
+		if !ValidAsMapKey(t) || !ValidAsMapKey(f) {
+			return
+		}
+
+		if _, ok := t.(*sqlparser.ColName); ok {
+			// If this is introducing a new column, we should copy all dependencies over
+			// as we can't recalculate them later
+			st.CopyDependencies(f, t)
+		} else {
+			// Otherwise, we only copy over the type information
+			st.ExprTypes[t] = st.ExprTypes[f]
+		}
 	case *sqlparser.Union:
 		t, ok := to.(*sqlparser.Union)
 		if !ok {