[FLINK-34702][planner] When converting Rank to Deduplicate, it is nec…

flink-table/flink-table-planner/src/main/java/org/apache/flink/table/planner/plan/rules/physical/stream/StreamPhysicalDeduplicateRule.java

@@ -0,0 +1,108 @@
+/*
+ * 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.flink.table.planner.plan.rules.physical.stream;
+
+import org.apache.flink.table.planner.calcite.FlinkTypeFactory;
+import org.apache.flink.table.planner.plan.nodes.physical.stream.StreamPhysicalDeduplicate;
+import org.apache.flink.table.planner.plan.nodes.physical.stream.StreamPhysicalRank;
+import org.apache.flink.table.planner.plan.utils.ChangelogPlanUtils;
+import org.apache.flink.table.planner.plan.utils.RankUtil;
+
+import org.apache.calcite.plan.RelOptRuleCall;
+import org.apache.calcite.plan.RelRule;
+import org.apache.calcite.rel.RelFieldCollation;
+import org.apache.calcite.rel.type.RelDataType;
+import org.immutables.value.Value;
+
+/**
+ * Rule that matches {@link StreamPhysicalRank} which is sorted by time attribute and limits 1 and
+ * its rank type is ROW_NUMBER and input doesn't produce changelog, and converts it to {@link
+ * StreamPhysicalDeduplicate}.
+ *
+ * <p>NOTES: Queries that can be converted to {@link StreamPhysicalDeduplicate} could be converted
+ * to {@link StreamPhysicalRank} too. {@link StreamPhysicalDeduplicate} is more efficient than
+ * {@link StreamPhysicalRank} due to mini-batch and less state access.
+ *
+ * <p>e.g. 1. {@code SELECT a, b, c FROM ( SELECT a, b, c, proctime, ROW_NUMBER() OVER (PARTITION BY
+ * a ORDER BY proctime ASC) as row_num FROM MyTable ) WHERE row_num <= 1 } will be converted to
+ * StreamExecDeduplicate which keeps first row in proctime.
+ *
+ * <p>2. {@code SELECT a, b, c FROM ( SELECT a, b, c, rowtime, ROW_NUMBER() OVER (PARTITION BY a
+ * ORDER BY rowtime DESC) as row_num FROM MyTable ) WHERE row_num <= 1 } will be converted to
+ * StreamExecDeduplicate which keeps last row in rowtime.
+ */
+@Value.Enclosing
+public class StreamPhysicalDeduplicateRule
+        extends RelRule<StreamPhysicalDeduplicateRule.StreamPhysicalDeduplicateRuleConfig> {
+
+    public static final StreamPhysicalDeduplicateRule INSTANCE =
+            StreamPhysicalDeduplicateRule.StreamPhysicalDeduplicateRuleConfig.DEFAULT.toRule();
+
+    private StreamPhysicalDeduplicateRule(StreamPhysicalDeduplicateRuleConfig config) {
+        super(config);
+    }
+
+    @Override
+    public boolean matches(RelOptRuleCall call) {
+        StreamPhysicalRank rank = call.rel(0);
+        return ChangelogPlanUtils.inputInsertOnly(rank) && RankUtil.canConvertToDeduplicate(rank);
+    }
+
+    @Override
+    public void onMatch(RelOptRuleCall call) {
+        StreamPhysicalRank rank = call.rel(0);
+
+        // order by timeIndicator desc ==> lastRow, otherwise is firstRow
+        RelFieldCollation fieldCollation = rank.orderKey().getFieldCollations().get(0);
+        boolean isLastRow = fieldCollation.direction.isDescending();
+
+        RelDataType fieldType =
+                rank.getInput()
+                        .getRowType()
+                        .getFieldList()
+                        .get(fieldCollation.getFieldIndex())
+                        .getType();
+        boolean isRowtime = FlinkTypeFactory.isRowtimeIndicatorType(fieldType);
+
+        StreamPhysicalDeduplicate deduplicate =
+                new StreamPhysicalDeduplicate(
+                        rank.getCluster(),
+                        rank.getTraitSet(),
+                        rank.getInput(),
+                        rank.partitionKey().toArray(),
+                        isRowtime,
+                        isLastRow);
+        call.transformTo(deduplicate);
+    }
+
+    /** Rule configuration. */
+    @Value.Immutable(singleton = false)
+    public interface StreamPhysicalDeduplicateRuleConfig extends RelRule.Config {
+        StreamPhysicalDeduplicateRule.StreamPhysicalDeduplicateRuleConfig DEFAULT =
+                ImmutableStreamPhysicalDeduplicateRule.StreamPhysicalDeduplicateRuleConfig.builder()
+                        .build()
+                        .withOperandSupplier(b0 -> b0.operand(StreamPhysicalRank.class).anyInputs())
+                        .withDescription("StreamPhysicalDeduplicateRule");
+
+        @Override
+        default StreamPhysicalDeduplicateRule toRule() {
+            return new StreamPhysicalDeduplicateRule(this);
+        }
+    }
+}

flink-table/flink-table-planner/src/main/scala/org/apache/flink/table/planner/plan/optimize/program/FlinkStreamProgram.scala

@@ -296,6 +296,14 @@ object FlinkStreamProgram {
           "watermark transpose"
         )
         .addProgram(new FlinkChangelogModeInferenceProgram, "Changelog mode inference")
+        .addProgram(
+          FlinkHepRuleSetProgramBuilder.newBuilder
+            .setHepRulesExecutionType(HEP_RULES_EXECUTION_TYPE.RULE_COLLECTION)
+            .setHepMatchOrder(HepMatchOrder.BOTTOM_UP)
+            .add(FlinkStreamRuleSets.PHYSICAL_DEDUPLICATE_REWRITE)
+            .build(),
+          "physical rewrite rank with deduplicate"
+        )
         .addProgram(
           new FlinkMiniBatchIntervalTraitInitProgram,
           "Initialization for mini-batch interval inference")

flink-table/flink-table-planner/src/main/scala/org/apache/flink/table/planner/plan/rules/FlinkStreamRuleSets.scala

@@ -441,7 +441,6 @@ object FlinkStreamRuleSets {
     StreamPhysicalTemporalSortRule.INSTANCE,
     // rank
     StreamPhysicalRankRule.INSTANCE,
-    StreamPhysicalDeduplicateRule.INSTANCE,
     // expand
     StreamPhysicalExpandRule.INSTANCE,
     // group agg
@@ -487,6 +486,15 @@ object FlinkStreamRuleSets {
     WatermarkAssignerChangelogNormalizeTransposeRule.WITHOUT_CALC
   )
 
+  /**
+   * RuleSet to rewrite rank with deduplicate. And it should be before the rule of mini-batch
+   * interval and after changelog mode inference.
+   */
+  val PHYSICAL_DEDUPLICATE_REWRITE: RuleSet = RuleSets.ofList(
+    // optimize rank rule
+    StreamPhysicalDeduplicateRule.INSTANCE
+  )
+
   /** RuleSet related to mini-batch. */
   val MINI_BATCH_RULES: RuleSet = RuleSets.ofList(
     // mini-batch interval infer rule

flink-table/flink-table-planner/src/main/scala/org/apache/flink/table/planner/plan/rules/physical/stream/StreamPhysicalRankRule.scala

@@ -34,11 +34,6 @@ import org.apache.calcite.rel.convert.ConverterRule.Config
  */
 class StreamPhysicalRankRule(config: Config) extends ConverterRule(config) {
 
-  override def matches(call: RelOptRuleCall): Boolean = {
-    val rank: FlinkLogicalRank = call.rel(0)
-    !RankUtil.canConvertToDeduplicate(rank)
-  }
-
   override def convert(rel: RelNode): RelNode = {
     val rank = rel.asInstanceOf[FlinkLogicalRank]
     val input = rank.getInput

flink-table/flink-table-planner/src/main/scala/org/apache/flink/table/planner/plan/utils/ChangelogPlanUtils.scala

@@ -23,6 +23,7 @@ import org.apache.flink.table.planner.plan.nodes.physical.stream.StreamPhysicalR
 import org.apache.flink.table.planner.plan.optimize.program.FlinkChangelogModeInferenceProgram
 import org.apache.flink.types.RowKind
 
+import org.apache.calcite.plan.hep.HepRelVertex
 import org.apache.calcite.rel.RelNode
 
 import scala.collection.JavaConversions._
@@ -46,7 +47,11 @@ object ChangelogPlanUtils {
    * <p>Note: this method must be called after [[FlinkChangelogModeInferenceProgram]] is applied.
    */
   def inputInsertOnly(node: StreamPhysicalRel): Boolean = {
-    node.getInputs.forall { case input: StreamPhysicalRel => isInsertOnly(input) }
+    node.getInputs.forall {
+      case input: StreamPhysicalRel => isInsertOnly(input)
+      case hepRelVertex: HepRelVertex =>
+        isInsertOnly(hepRelVertex.getCurrentRel.asInstanceOf[StreamPhysicalRel])
+    }
   }
 
   /**

flink-table/flink-table-planner/src/main/scala/org/apache/flink/table/planner/plan/utils/RankUtil.scala

@@ -22,7 +22,7 @@ import org.apache.flink.table.planner.calcite.FlinkTypeFactory
 import org.apache.flink.table.planner.codegen.ExpressionReducer
 import org.apache.flink.table.planner.plan.nodes.calcite.Rank
 import org.apache.flink.table.planner.plan.nodes.logical.FlinkLogicalRank
-import org.apache.flink.table.planner.plan.nodes.physical.stream.StreamPhysicalDeduplicate
+import org.apache.flink.table.planner.plan.nodes.physical.stream.{StreamPhysicalDeduplicate, StreamPhysicalRank}
 import org.apache.flink.table.runtime.operators.rank.{ConstantRankRange, ConstantRankRangeWithoutEnd, RankRange, RankType, VariableRankRange}
 
 import org.apache.calcite.plan.RelOptUtil
@@ -357,6 +357,35 @@ object RankUtil {
     !rank.outputRankNumber && isLimit1 && isSortOnTimeAttribute && isRowNumberType
   }
 
+  /**
+   * Whether the given rank could be converted to [[StreamPhysicalDeduplicate]].
+   *
+   * Returns true if the given rank is sorted by time attribute and limits 1 and its RankFunction is
+   * ROW_NUMBER, else false.
+   *
+   * @param rank
+   *   The [[StreamPhysicalRank]] node
+   * @return
+   *   True if the input rank could be converted to [[StreamPhysicalDeduplicate]]
+   */
+  def canConvertToDeduplicate(rank: StreamPhysicalRank): Boolean = {
+    val sortCollation = rank.orderKey
+    val rankRange = rank.rankRange
+
+    val isRowNumberType = rank.rankType == RankType.ROW_NUMBER
+
+    val isLimit1 = rankRange match {
+      case rankRange: ConstantRankRange =>
+        rankRange.getRankStart == 1 && rankRange.getRankEnd == 1
+      case _ => false
+    }
+
+    val inputRowType = rank.getInput.getRowType
+    val isSortOnTimeAttribute = sortOnTimeAttribute(sortCollation, inputRowType)
+
+    !rank.outputRankNumber && isLimit1 && isSortOnTimeAttribute && isRowNumberType
+  }
+
   private def sortOnTimeAttribute(
       sortCollation: RelCollation,
       inputRowType: RelDataType): Boolean = {

flink-table/flink-table-planner/src/test/resources/org/apache/flink/table/planner/plan/stream/sql/DeduplicateTest.xml

@@ -360,6 +360,41 @@ Calc(select=[a, b, c, PROCTIME_MATERIALIZE(proctime) AS proctime, rowtime, 1 AS
 +- Deduplicate(keep=[LastRow], key=[a], order=[PROCTIME])
    +- Exchange(distribution=[hash[a]])
       +- DataStreamScan(table=[[default_catalog, default_database, MyTable]], fields=[a, b, c, proctime, rowtime])
+]]>
+    </Resource>
+  </TestCase>
+  <TestCase name="testRankConsumeChangelog">
+    <Resource name="sql">
+      <![CDATA[
+SELECT *
+FROM (
+  SELECT *,
+    ROW_NUMBER() OVER (PARTITION BY a ORDER BY PROCTIME() ASC) as rowNum
+  FROM (SELECT a, count(b) as b FROM MyTable GROUP BY a)
+)
+WHERE rowNum = 1
+      ]]>
+    </Resource>
+    <Resource name="ast">
+      <![CDATA[
+LogicalProject(a=[$0], b=[$1], rowNum=[$2])
++- LogicalFilter(condition=[=($2, 1)])
+   +- LogicalProject(a=[$0], b=[$1], rowNum=[ROW_NUMBER() OVER (PARTITION BY $0 ORDER BY PROCTIME() NULLS FIRST)])
+      +- LogicalAggregate(group=[{0}], b=[COUNT($1)])
+         +- LogicalProject(a=[$0], b=[$1])
+            +- LogicalTableScan(table=[[default_catalog, default_database, MyTable]])
+]]>
+    </Resource>
+    <Resource name="optimized exec plan">
+      <![CDATA[
+Calc(select=[a, b, 1 AS $2])
++- Rank(strategy=[RetractStrategy], rankType=[ROW_NUMBER], rankRange=[rankStart=1, rankEnd=1], partitionBy=[a], orderBy=[$2 ASC], select=[a, b, $2])
+   +- Exchange(distribution=[hash[a]])
+      +- Calc(select=[a, b, PROCTIME() AS $2])
+         +- GroupAggregate(groupBy=[a], select=[a, COUNT(b) AS b])
+            +- Exchange(distribution=[hash[a]])
+               +- Calc(select=[a, b])
+                  +- DataStreamScan(table=[[default_catalog, default_database, MyTable]], fields=[a, b, c, proctime, rowtime])
 ]]>
     </Resource>
   </TestCase>

flink-table/flink-table-planner/src/test/scala/org/apache/flink/table/planner/plan/stream/sql/DeduplicateTest.scala

@@ -272,4 +272,18 @@ class DeduplicateTest extends TableTestBase {
     util.verifyExecPlan(sqlQuery)
   }
 
+  @Test
+  def testRankConsumeChangelog(): Unit = {
+    val sqlQuery =
+      """
+        |SELECT *
+        |FROM (
+        |  SELECT *,
+        |    ROW_NUMBER() OVER (PARTITION BY a ORDER BY PROCTIME() ASC) as rowNum
+        |  FROM (SELECT a, count(b) as b FROM MyTable GROUP BY a)
+        |)
+        |WHERE rowNum = 1
+      """.stripMargin
+    util.verifyExecPlan(sqlQuery)
+  }
 }