[SPARK-39057][SQL] Offset could work without Limit

### What changes were proposed in this pull request?
Currently, `Offset` must work with `Limit`. The behavior not allow to use offset alone and add offset API into `DataFrame`.

If we use `Offset` alone, there are two situations:
1. If `Offset` is the last operator, collect the result to the driver and then drop/skip the first n (offset value) rows. Users can test or debug `Offset` in the way.
2. If `Offset` is the intermediate operator, shuffle all the result to one task and drop/skip the first n (offset value) rows and the result will be passed to the downstream operator.

For example, `SELECT * FROM a offset 10; ` parsed to the logic plan as below:
```
Offset (offset = 10) // Only offset clause
|--Relation
```

and then the physical plan as below:
```
CollectLimitExec(limit = -1, offset = 10) // Collect the result to the driver and skip the first 10 rows
|--JDBCRelation
```
or
```
GlobalLimitAndOffsetExec(limit = -1, offset = 10) // Collect the result and skip the first 10 rows
|--JDBCRelation
```

After this PR merged, users could input the SQL show below:
```
SELECT '' AS ten, unique1, unique2, stringu1
 		FROM onek
 		ORDER BY unique1 OFFSET 990;
```

Note: apache#35975 supports offset clause, it create a logical node named
`GlobalLimitAndOffset`. In fact, we can avoid use this node and use `Offset` instead and the latter is good with unify name.

### Why are the changes needed?
Improve the implement of offset clause.

### Does this PR introduce _any_ user-facing change?
'No'.
New feature.

### How was this patch tested?
Exists test cases.

Closes apache#36417 from beliefer/SPARK-28330_followup2.

Authored-by: Jiaan Geng <[email protected]>
Signed-off-by: Wenchen Fan <[email protected]>

Author: beliefer

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/CheckAnalysis.scala

@@ -404,6 +404,13 @@ trait CheckAnalysis extends PredicateHelper with LookupCatalog {
 
           case Offset(offsetExpr, _) => checkLimitLikeClause("offset", offsetExpr)
 
+          case o if !o.isInstanceOf[GlobalLimit] && !o.isInstanceOf[LocalLimit]
+            && o.children.exists(_.isInstanceOf[Offset]) =>
+            failAnalysis(
+              s"""
+                 |The OFFSET clause is allowed in the LIMIT clause or be the outermost node,
+                 |but the OFFSET clause found in: ${o.nodeName}.""".stripMargin.replace("\n", " "))
+
           case Tail(limitExpr, _) => checkLimitLikeClause("tail", limitExpr)
 
           case _: Union | _: SetOperation if operator.children.length > 1 =>
@@ -564,7 +571,6 @@ trait CheckAnalysis extends PredicateHelper with LookupCatalog {
         }
     }
     checkCollectedMetrics(plan)
-    checkOffsetOperator(plan)
     extendedCheckRules.foreach(_(plan))
     plan.foreachUp {
       case o if !o.resolved =>
@@ -712,30 +718,6 @@ trait CheckAnalysis extends PredicateHelper with LookupCatalog {
     check(plan)
   }
 
-  /**
-   * Validate whether the [[Offset]] is valid.
-   */
-  private def checkOffsetOperator(plan: LogicalPlan): Unit = {
-    plan.foreachUp {
-      case o if !o.isInstanceOf[GlobalLimit] && !o.isInstanceOf[LocalLimit]
-        && o.children.exists(_.isInstanceOf[Offset]) =>
-        failAnalysis(
-          s"""
-             |The OFFSET clause is only allowed in the LIMIT clause, but the OFFSET
-             |clause found in: ${o.nodeName}.""".stripMargin.replace("\n", " "))
-      case _ =>
-    }
-    plan match {
-      case Offset(offsetExpr, _) =>
-        checkLimitLikeClause("offset", offsetExpr)
-        failAnalysis(
-          s"""
-             |The OFFSET clause is only allowed in the LIMIT clause, but the OFFSET
-             |clause is found to be the outermost node.""".stripMargin.replace("\n", " "))
-      case _ =>
-    }
-  }
-
   /**
    * Validates to make sure the outer references appearing inside the subquery
    * are allowed.

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala

@@ -93,9 +93,8 @@ abstract class Optimizer(catalogManager: CatalogManager)
         OptimizeWindowFunctions,
         CollapseWindow,
         CombineFilters,
-        RewriteOffsets,
+        EliminateOffsets,
         EliminateLimits,
-        RewriteOffsets,
         CombineUnions,
         // Constant folding and strength reduction
         OptimizeRepartition,
@@ -640,7 +639,7 @@ object RemoveNoopUnion extends Rule[LogicalPlan] {
 }
 
 /**
- * Pushes down [[LocalLimit]] beneath UNION ALL and joins.
+ * Pushes down [[LocalLimit]] beneath UNION ALL, OFFSET and joins.
  */
 object LimitPushDown extends Rule[LogicalPlan] {
 
@@ -710,6 +709,14 @@ object LimitPushDown extends Rule[LogicalPlan] {
     // There is a Project between LocalLimit and Join if they do not have the same output.
     case LocalLimit(exp, project @ Project(_, join: Join)) =>
       LocalLimit(exp, project.copy(child = pushLocalLimitThroughJoin(exp, join)))
+    // Push down limit 1 through Aggregate and turn Aggregate into Project if it is group only.
+    case Limit(le @ IntegerLiteral(1), a: Aggregate) if a.groupOnly =>
+      Limit(le, Project(a.aggregateExpressions, LocalLimit(le, a.child)))
+    case Limit(le @ IntegerLiteral(1), p @ Project(_, a: Aggregate)) if a.groupOnly =>
+      Limit(le, p.copy(child = Project(a.aggregateExpressions, LocalLimit(le, a.child))))
+    // Merge offset value and limit value into LocalLimit and pushes down LocalLimit through Offset.
+    case LocalLimit(le, Offset(oe, grandChild)) =>
+      Offset(oe, LocalLimit(Add(le, oe), grandChild))
   }
 }
 
@@ -1785,34 +1792,22 @@ object EliminateLimits extends Rule[LogicalPlan] {
 }
 
 /**
- * Rewrite [[Offset]] as [[Limit]] or combines two adjacent [[Offset]] operators into one,
- *  merging the expressions into one single expression.
- */
-object RewriteOffsets extends Rule[LogicalPlan] {
-  def apply(plan: LogicalPlan): LogicalPlan = plan transform {
-    case GlobalLimit(le, Offset(oe, grandChild)) =>
-      GlobalLimitAndOffset(le, oe, grandChild)
-    case LocalLimit(le, Offset(oe, grandChild)) =>
-      Offset(oe, LocalLimit(Add(le, oe), grandChild))
-  }
-}
-
-/**
- * Rewrite [[Offset]] as [[GlobalLimitAndOffset]] or [[LocalLimit]],
- * merging the expressions into one single expression. See [[Limit]] for more information
- * about the difference between [[LocalLimit]] and [[GlobalLimit]].
+ * This rule optimizes Offset operators by:
+ * 1. Eliminate [[Offset]] operators if offset == 0.
+ * 2. Replace [[Offset]] operators to empty [[LocalRelation]]
+ *    if [[Offset]]'s child max row <= offset.
+ * 3. Combines two adjacent [[Offset]] operators into one, merging the
+ *    expressions into one single expression.
  */
-object RewriteOffsets extends Rule[LogicalPlan] {
+object EliminateOffsets extends Rule[LogicalPlan] {
   def apply(plan: LogicalPlan): LogicalPlan = plan transform {
-    case GlobalLimit(le, Offset(oe, grandChild)) =>
-      GlobalLimitAndOffset(le, oe, grandChild)
-    case localLimit @ LocalLimit(le, Offset(oe, grandChild)) =>
-      val offset = oe.eval().asInstanceOf[Int]
-      if (offset == 0) {
-        localLimit.withNewChildren(Seq(grandChild))
-      } else {
-        Offset(oe, LocalLimit(Add(le, oe), grandChild))
-      }
+    case Offset(oe, child) if oe.foldable && oe.eval().asInstanceOf[Int] == 0 =>
+      child
+    case Offset(oe, child)
+      if oe.foldable && child.maxRows.exists(_ <= oe.eval().asInstanceOf[Int]) =>
+      LocalRelation(child.output, data = Seq.empty, isStreaming = child.isStreaming)
+    case Offset(oe1, Offset(oe2, child)) =>
+      Offset(Add(oe1, oe2), child)
   }
 }
 

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/basicLogicalOperators.scala

@@ -1245,36 +1245,16 @@ case class LocalLimit(limitExpr: Expression, child: LogicalPlan) extends OrderPr
 }
 
 object LimitAndOffset {
-  def unapply(p: GlobalLimitAndOffset): Option[(Expression, Expression, LogicalPlan)] = {
+  def unapply(p: GlobalLimit): Option[(Expression, Expression, LogicalPlan)] = {
     p match {
-      case GlobalLimitAndOffset(le1, le2, LocalLimit(le3, child)) if le1.eval().asInstanceOf[Int]
+      case GlobalLimit(le1, Offset(le2, LocalLimit(le3, child))) if le1.eval().asInstanceOf[Int]
         + le2.eval().asInstanceOf[Int] == le3.eval().asInstanceOf[Int] =>
         Some((le1, le2, child))
       case _ => None
     }
   }
 }
 
-/**
- * A global (coordinated) limit with offset. This operator can skip at most `offsetExpr` number and
- * emit at most `limitExpr` number in total. For example, if we have LIMIT 10 OFFSET 5, we impose a
- * total limit of 10 + 5 = 15 rows and then discard the first 5, leaving 10 rows remaining.
- */
-case class GlobalLimitAndOffset(
-    limitExpr: Expression,
-    offsetExpr: Expression,
-    child: LogicalPlan) extends OrderPreservingUnaryNode {
-  override def output: Seq[Attribute] = child.output
-  override def maxRows: Option[Long] = {
-    limitExpr match {
-      case IntegerLiteral(limit) => Some(limit)
-      case _ => None
-    }
-  }
-  override protected def withNewChildInternal(newChild: LogicalPlan): GlobalLimitAndOffset =
-    copy(child = newChild)
-}
-
 /**
  * This is similar with [[Limit]] except:
  *

sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/analysis/AnalysisErrorSuite.scala

@@ -574,18 +574,11 @@ class AnalysisErrorSuite extends AnalysisTest {
     "The offset expression must be equal to or greater than 0, but got -1" :: Nil
   )
 
-  errorTest(
-    "OFFSET clause is outermost node",
-    testRelation.offset(Literal(10, IntegerType)),
-    "The OFFSET clause is only allowed in the LIMIT clause, but the OFFSET" +
-      " clause is found to be the outermost node." :: Nil
-  )
-
   errorTest(
     "OFFSET clause in other node",
     testRelation2.offset(Literal(10, IntegerType)).where('b > 1),
-    "The OFFSET clause is only allowed in the LIMIT clause, but the OFFSET" +
-      " clause found in: Filter." :: Nil
+    "The OFFSET clause is allowed in the LIMIT clause or be the outermost node," +
+      " but the OFFSET clause found in: Filter." :: Nil
   )
 
   errorTest(

sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/optimizer/EliminateOffsetsSuite.scala

@@ -0,0 +1,100 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.sql.catalyst.optimizer
+
+import org.apache.spark.sql.Row
+import org.apache.spark.sql.catalyst.dsl.expressions._
+import org.apache.spark.sql.catalyst.dsl.plans._
+import org.apache.spark.sql.catalyst.expressions.{Add, Literal}
+import org.apache.spark.sql.catalyst.plans.PlanTest
+import org.apache.spark.sql.catalyst.plans.logical.{LocalRelation, LogicalPlan}
+import org.apache.spark.sql.catalyst.rules.RuleExecutor
+
+class EliminateOffsetsSuite extends PlanTest {
+  object Optimize extends RuleExecutor[LogicalPlan] {
+    val batches =
+      Batch("Eliminate Offset", FixedPoint(10), EliminateOffsets) :: Nil
+  }
+
+  val testRelation = LocalRelation.fromExternalRows(
+    Seq("a".attr.int, "b".attr.int, "c".attr.int),
+    1.to(10).map(_ => Row(1, 2, 3))
+  )
+
+  test("Offsets: eliminate Offset operators if offset == 0") {
+    val originalQuery =
+      testRelation
+        .select($"a")
+        .offset(0)
+
+    val optimized = Optimize.execute(originalQuery.analyze)
+    val correctAnswer =
+      testRelation
+        .select($"a")
+        .analyze
+
+    comparePlans(optimized, correctAnswer)
+  }
+
+  test("Offsets: cannot eliminate Offset operators if offset > 0") {
+    val originalQuery =
+      testRelation
+        .select($"a")
+        .offset(2)
+
+    val optimized = Optimize.execute(originalQuery.analyze)
+    val correctAnswer =
+      testRelation
+        .select($"a")
+        .offset(2)
+        .analyze
+
+    comparePlans(optimized, correctAnswer)
+  }
+
+  test("Replace Offset operators to empty LocalRelation if child max row <= offset") {
+    val child = testRelation.select($"a").analyze
+    val originalQuery = child.offset(10)
+
+    val optimized = Optimize.execute(originalQuery.analyze)
+    val correctAnswer =
+      LocalRelation(child.output, data = Seq.empty, isStreaming = child.isStreaming).analyze
+
+    comparePlans(optimized, correctAnswer)
+  }
+
+  test("Cannot replace Offset operators to empty LocalRelation if child max row > offset") {
+    val child = testRelation.select($"a").analyze
+    val originalQuery = child.offset(3)
+
+    val optimized = Optimize.execute(originalQuery.analyze)
+    val correctAnswer = originalQuery.analyze
+
+    comparePlans(optimized, correctAnswer)
+  }
+
+  test("Combines Offset operators") {
+    val child = testRelation.select($"a").analyze
+    val originalQuery = child.offset(2).offset(3)
+
+    val optimized = Optimize.execute(originalQuery.analyze)
+    val correctAnswer = child.offset(Add(Literal(3), Literal(2))).analyze
+
+    comparePlans(optimized, correctAnswer)
+  }
+}

sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/optimizer/LimitPushdownSuite.scala

@@ -239,4 +239,10 @@ class LimitPushdownSuite extends PlanTest {
       Limit(5, LocalLimit(5, x).join(y, LeftOuter, joinCondition).select("x.a".attr)).analyze
     comparePlans(optimized, correctAnswer)
   }
+
+  test("Push down limit 1 through Offset") {
+    comparePlans(
+      Optimize.execute(testRelation.offset(2).limit(1).analyze),
+      GlobalLimit(1, Offset(2, LocalLimit(3, testRelation))).analyze)
+  }
 }

sql/core/src/main/scala/org/apache/spark/sql/execution/SparkStrategies.scala

@@ -92,14 +92,16 @@ abstract class SparkStrategies extends QueryPlanner[SparkPlan] {
         case LimitAndOffset(IntegerLiteral(limit), IntegerLiteral(offset),
           Sort(order, true, child)) if limit + offset < conf.topKSortFallbackThreshold =>
           TakeOrderedAndProjectExec(
-            limit, order, child.output, planLater(child), Some(offset)) :: Nil
+            limit, order, child.output, planLater(child), offset) :: Nil
         case LimitAndOffset(IntegerLiteral(limit), IntegerLiteral(offset),
           Project(projectList, Sort(order, true, child)))
             if limit + offset < conf.topKSortFallbackThreshold =>
           TakeOrderedAndProjectExec(
-            limit, order, projectList, planLater(child), Some(offset)) :: Nil
+            limit, order, projectList, planLater(child), offset) :: Nil
         case LimitAndOffset(IntegerLiteral(limit), IntegerLiteral(offset), child) =>
-          CollectLimitExec(limit, planLater(child), Some(offset)) :: Nil
+          CollectLimitExec(limit, planLater(child), offset) :: Nil
+        case logical.Offset(IntegerLiteral(offset), child) =>
+          CollectLimitExec(child = planLater(child), offset = offset) :: Nil
         case Tail(IntegerLiteral(limit), child) =>
           CollectTailExec(limit, planLater(child)) :: Nil
         case other => planLater(other) :: Nil
@@ -115,20 +117,18 @@ abstract class SparkStrategies extends QueryPlanner[SparkPlan] {
       // In this case we generate a physical top-K sorting operator, passing down
       // the limit and offset values to be evaluated inline during the physical
       // sorting operation for greater efficiency.
-      case LimitAndOffset(
-          IntegerLiteral(limit),
-          IntegerLiteral(offset),
-          Sort(order, true, child))
-          if limit + offset < conf.topKSortFallbackThreshold =>
-        TakeOrderedAndProjectExec(
-          limit, order, child.output, planLater(child), Some(offset)) :: Nil
-      case LimitAndOffset(
-          IntegerLiteral(limit),
-          IntegerLiteral(offset),
-          Project(projectList, Sort(order, true, child)))
+      case LimitAndOffset(IntegerLiteral(limit), IntegerLiteral(offset),
+        Sort(order, true, child)) if limit + offset < conf.topKSortFallbackThreshold =>
+          TakeOrderedAndProjectExec(
+            limit, order, child.output, planLater(child), offset) :: Nil
+      case LimitAndOffset(IntegerLiteral(limit), IntegerLiteral(offset),
+        Project(projectList, Sort(order, true, child)))
           if limit + offset < conf.topKSortFallbackThreshold =>
-        TakeOrderedAndProjectExec(limit, order, projectList, planLater(child), Some(offset)) :: Nil
-      case _ => Nil
+        TakeOrderedAndProjectExec(limit, order, projectList, planLater(child), offset) :: Nil
+      case LimitAndOffset(IntegerLiteral(limit), IntegerLiteral(offset), child) =>
+        GlobalLimitAndOffsetExec(limit, offset, planLater(child)) :: Nil
+      case _ =>
+        Nil
     }
   }
 
@@ -790,8 +790,8 @@ abstract class SparkStrategies extends QueryPlanner[SparkPlan] {
         execution.LocalLimitExec(limit, planLater(child)) :: Nil
       case logical.GlobalLimit(IntegerLiteral(limit), child) =>
         execution.GlobalLimitExec(limit, planLater(child)) :: Nil
-      case logical.GlobalLimitAndOffset(IntegerLiteral(limit), IntegerLiteral(offset), child) =>
-        execution.GlobalLimitAndOffsetExec(limit, offset, planLater(child)) :: Nil
+      case logical.Offset(IntegerLiteral(offset), child) =>
+        GlobalLimitAndOffsetExec(offset = offset, child = planLater(child)) :: Nil
       case union: logical.Union =>
         execution.UnionExec(union.children.map(planLater)) :: Nil
       case g @ logical.Generate(generator, _, outer, _, _, child) =>