diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala
index 1e934d0aa0e51..dd868f3e89329 100644
--- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala
+++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala
@@ -1115,6 +1115,8 @@ class Analyzer(
           g.copy(join = true, child = addMissingAttr(g.child, missing))
         case d: Distinct =>
           throw new AnalysisException(s"Can't add $missingAttrs to $d")
+        case u: Union =>
+          u.withNewChildren(u.children.map(addMissingAttr(_, missingAttrs)))
         case u: UnaryNode =>
           u.withNewChildren(addMissingAttr(u.child, missingAttrs) :: Nil)
         case other =>
@@ -1133,6 +1135,8 @@ class Analyzer(
         resolved
       } else {
         plan match {
+          case u: Union if !u.children.head.isInstanceOf[SubqueryAlias] =>
+            resolveExpressionRecursively(resolved, u.children.head)
           case u: UnaryNode if !u.isInstanceOf[SubqueryAlias] =>
             resolveExpressionRecursively(resolved, u.child)
           case other => resolved
diff --git a/sql/core/src/test/scala/org/apache/spark/sql/DataFrameSuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/DataFrameSuite.scala
index 13341645e8ff8..045b3cfbb14e5 100644
--- a/sql/core/src/test/scala/org/apache/spark/sql/DataFrameSuite.scala
+++ b/sql/core/src/test/scala/org/apache/spark/sql/DataFrameSuite.scala
@@ -2039,4 +2039,13 @@ class DataFrameSuite extends QueryTest with SharedSQLContext {
       testData2.select(lit(7), 'a, 'b).orderBy(lit(1), lit(2), lit(3)),
       Seq(Row(7, 1, 1), Row(7, 1, 2), Row(7, 2, 1), Row(7, 2, 2), Row(7, 3, 1), Row(7, 3, 2)))
   }
+
+  test("SPARK-21966: ResolveMissingReference rule should not ignore Union") {
+    val df1 = Seq((1, 1), (2, 1), (2, 2)).toDF("a", "b")
+    val df2 = Seq((1, 1), (1, 2), (2, 3)).toDF("a", "b")
+    val df3 = df1.cube("a").sum("b")
+    val df4 = df2.cube("a").sum("b")
+    val df = df3.union(df4).filter("grouping_id() = 0")
+    checkAnswer(df, Seq(Row(1, 1), Row(2, 3), Row(1, 3), Row(2, 3)))
+  }
 }