MSQ: Allow for worker gaps. (apache#17277)

In a Dart query, all Historicals are given worker IDs, but not all of them
are going to actually be started or receive work orders. This can create gaps
in the set of workers. For example, workers 1 and 3 could have work assigned
while workers 0 and 2 do not.

This patch updates ControllerStageTracker and WorkerInputs to handle such
gaps, by using the set of actual worker numbers, rather than 0..workerCount,
in various places.

(cherry picked from commit 06bbdb3)

Author: gianm

extensions-core/multi-stage-query/src/main/java/org/apache/druid/msq/input/stage/ReadablePartition.java

@@ -59,6 +59,18 @@ public static ReadablePartition striped(final int stageNumber, final int numWork
     return new ReadablePartition(stageNumber, workerNumbers, partitionNumber);
   }
 
+  /**
+   * Returns an output partition that is striped across a set of {@code workerNumbers}.
+   */
+  public static ReadablePartition striped(
+      final int stageNumber,
+      final IntSortedSet workerNumbers,
+      final int partitionNumber
+  )
+  {
+    return new ReadablePartition(stageNumber, workerNumbers, partitionNumber);
+  }
+
   /**
    * Returns an output partition that has been collected onto a single worker.
    */

extensions-core/multi-stage-query/src/main/java/org/apache/druid/msq/input/stage/ReadablePartitions.java

@@ -24,6 +24,7 @@
 import it.unimi.dsi.fastutil.ints.Int2IntAVLTreeMap;
 import it.unimi.dsi.fastutil.ints.Int2IntSortedMap;
 import it.unimi.dsi.fastutil.ints.IntAVLTreeSet;
+import it.unimi.dsi.fastutil.ints.IntSortedSet;
 
 import java.util.Collections;
 import java.util.List;
@@ -39,6 +40,7 @@
 @JsonSubTypes(value = {
     @JsonSubTypes.Type(name = "collected", value = CollectedReadablePartitions.class),
     @JsonSubTypes.Type(name = "striped", value = StripedReadablePartitions.class),
+    @JsonSubTypes.Type(name = "sparseStriped", value = SparseStripedReadablePartitions.class),
     @JsonSubTypes.Type(name = "combined", value = CombinedReadablePartitions.class)
 })
 public interface ReadablePartitions extends Iterable<ReadablePartition>
@@ -59,7 +61,7 @@ static ReadablePartitions empty()
   /**
    * Combines various sets of partitions into a single set.
    */
-  static CombinedReadablePartitions combine(List<ReadablePartitions> readablePartitions)
+  static ReadablePartitions combine(List<ReadablePartitions> readablePartitions)
   {
     return new CombinedReadablePartitions(readablePartitions);
   }
@@ -68,7 +70,7 @@ static CombinedReadablePartitions combine(List<ReadablePartitions> readableParti
    * Returns a set of {@code numPartitions} partitions striped across {@code numWorkers} workers: each worker contains
    * a "stripe" of each partition.
    */
-  static StripedReadablePartitions striped(
+  static ReadablePartitions striped(
       final int stageNumber,
       final int numWorkers,
       final int numPartitions
@@ -82,11 +84,36 @@ static StripedReadablePartitions striped(
     return new StripedReadablePartitions(stageNumber, numWorkers, partitionNumbers);
   }
 
+  /**
+   * Returns a set of {@code numPartitions} partitions striped across {@code workers}: each worker contains
+   * a "stripe" of each partition.
+   */
+  static ReadablePartitions striped(
+      final int stageNumber,
+      final IntSortedSet workers,
+      final int numPartitions
+  )
+  {
+    final IntAVLTreeSet partitionNumbers = new IntAVLTreeSet();
+    for (int i = 0; i < numPartitions; i++) {
+      partitionNumbers.add(i);
+    }
+
+    if (workers.lastInt() == workers.size() - 1) {
+      // Dense worker set. Use StripedReadablePartitions for compactness (send a single number rather than the
+      // entire worker set) and for backwards compatibility (older workers cannot understand
+      // SparseStripedReadablePartitions).
+      return new StripedReadablePartitions(stageNumber, workers.size(), partitionNumbers);
+    } else {
+      return new SparseStripedReadablePartitions(stageNumber, workers, partitionNumbers);
+    }
+  }
+
   /**
    * Returns a set of partitions that have been collected onto specific workers: each partition is on exactly
    * one worker.
    */
-  static CollectedReadablePartitions collected(
+  static ReadablePartitions collected(
       final int stageNumber,
       final Map<Integer, Integer> partitionToWorkerMap
   )

extensions-core/multi-stage-query/src/main/java/org/apache/druid/msq/input/stage/SparseStripedReadablePartitions.java

@@ -0,0 +1,142 @@
+/*
+ * 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.druid.msq.input.stage;
+
+import com.fasterxml.jackson.annotation.JsonCreator;
+import com.fasterxml.jackson.annotation.JsonProperty;
+import com.google.common.collect.Iterators;
+import it.unimi.dsi.fastutil.ints.IntAVLTreeSet;
+import it.unimi.dsi.fastutil.ints.IntSortedSet;
+import org.apache.druid.msq.input.SlicerUtils;
+
+import java.util.ArrayList;
+import java.util.Iterator;
+import java.util.List;
+import java.util.Objects;
+import java.util.Set;
+
+/**
+ * Set of partitions striped across a sparse set of {@code workers}. Each worker contains a "stripe" of each partition.
+ *
+ * @see StripedReadablePartitions dense version, where workers from [0..N) are all used.
+ */
+public class SparseStripedReadablePartitions implements ReadablePartitions
+{
+  private final int stageNumber;
+  private final IntSortedSet workers;
+  private final IntSortedSet partitionNumbers;
+
+  /**
+   * Constructor. Most callers should use {@link ReadablePartitions#striped(int, int, int)} instead, which takes
+   * a partition count rather than a set of partition numbers.
+   */
+  public SparseStripedReadablePartitions(
+      final int stageNumber,
+      final IntSortedSet workers,
+      final IntSortedSet partitionNumbers
+  )
+  {
+    this.stageNumber = stageNumber;
+    this.workers = workers;
+    this.partitionNumbers = partitionNumbers;
+  }
+
+  @JsonCreator
+  private SparseStripedReadablePartitions(
+      @JsonProperty("stageNumber") final int stageNumber,
+      @JsonProperty("workers") final Set<Integer> workers,
+      @JsonProperty("partitionNumbers") final Set<Integer> partitionNumbers
+  )
+  {
+    this(stageNumber, new IntAVLTreeSet(workers), new IntAVLTreeSet(partitionNumbers));
+  }
+
+  @Override
+  public Iterator<ReadablePartition> iterator()
+  {
+    return Iterators.transform(
+        partitionNumbers.iterator(),
+        partitionNumber -> ReadablePartition.striped(stageNumber, workers, partitionNumber)
+    );
+  }
+
+  @Override
+  public List<ReadablePartitions> split(final int maxNumSplits)
+  {
+    final List<ReadablePartitions> retVal = new ArrayList<>();
+
+    for (List<Integer> entries : SlicerUtils.makeSlicesStatic(partitionNumbers.iterator(), maxNumSplits)) {
+      if (!entries.isEmpty()) {
+        retVal.add(new SparseStripedReadablePartitions(stageNumber, workers, new IntAVLTreeSet(entries)));
+      }
+    }
+
+    return retVal;
+  }
+
+  @JsonProperty
+  int getStageNumber()
+  {
+    return stageNumber;
+  }
+
+  @JsonProperty
+  IntSortedSet getWorkers()
+  {
+    return workers;
+  }
+
+  @JsonProperty
+  IntSortedSet getPartitionNumbers()
+  {
+    return partitionNumbers;
+  }
+
+  @Override
+  public boolean equals(Object o)
+  {
+    if (this == o) {
+      return true;
+    }
+    if (o == null || getClass() != o.getClass()) {
+      return false;
+    }
+    SparseStripedReadablePartitions that = (SparseStripedReadablePartitions) o;
+    return stageNumber == that.stageNumber
+           && Objects.equals(workers, that.workers)
+           && Objects.equals(partitionNumbers, that.partitionNumbers);
+  }
+
+  @Override
+  public int hashCode()
+  {
+    return Objects.hash(stageNumber, workers, partitionNumbers);
+  }
+
+  @Override
+  public String toString()
+  {
+    return "StripedReadablePartitions{" +
+           "stageNumber=" + stageNumber +
+           ", workers=" + workers +
+           ", partitionNumbers=" + partitionNumbers +
+           '}';
+  }
+}

extensions-core/multi-stage-query/src/main/java/org/apache/druid/msq/kernel/controller/ControllerStageTracker.java

@@ -403,7 +403,7 @@ void addPartialKeyInformationForWorker(
       throw new ISE("Stage does not gather result key statistics");
     }
 
-    if (workerNumber < 0 || workerNumber >= workerCount) {
+    if (!workerInputs.workers().contains(workerNumber)) {
       throw new IAE("Invalid workerNumber [%s]", workerNumber);
     }
 
@@ -522,7 +522,7 @@ void mergeClusterByStatisticsCollectorForTimeChunk(
       throw new ISE("Stage does not gather result key statistics");
     }
 
-    if (workerNumber < 0 || workerNumber >= workerCount) {
+    if (!workerInputs.workers().contains(workerNumber)) {
       throw new IAE("Invalid workerNumber [%s]", workerNumber);
     }
 
@@ -656,7 +656,7 @@ void mergeClusterByStatisticsCollectorForAllTimeChunks(
       throw new ISE("Stage does not gather result key statistics");
     }
 
-    if (workerNumber < 0 || workerNumber >= workerCount) {
+    if (!workerInputs.workers().contains(workerNumber)) {
       throw new IAE("Invalid workerNumber [%s]", workerNumber);
     }
 
@@ -763,7 +763,7 @@ void setClusterByPartitionBoundaries(ClusterByPartitions clusterByPartitions)
     this.resultPartitionBoundaries = clusterByPartitions;
     this.resultPartitions = ReadablePartitions.striped(
         stageDef.getStageNumber(),
-        workerCount,
+        workerInputs.workers(),
         clusterByPartitions.size()
     );
 
@@ -788,7 +788,7 @@ void setDoneReadingInputForWorker(final int workerNumber)
       throw DruidException.defensive("Cannot setDoneReadingInput for stage[%s], it is not sorting", stageDef.getId());
     }
 
-    if (workerNumber < 0 || workerNumber >= workerCount) {
+    if (!workerInputs.workers().contains(workerNumber)) {
       throw new IAE("Invalid workerNumber[%s] for stage[%s]", workerNumber, stageDef.getId());
     }
 
@@ -830,7 +830,7 @@ void setDoneReadingInputForWorker(final int workerNumber)
   @SuppressWarnings("unchecked")
   boolean setResultsCompleteForWorker(final int workerNumber, final Object resultObject)
   {
-    if (workerNumber < 0 || workerNumber >= workerCount) {
+    if (!workerInputs.workers().contains(workerNumber)) {
       throw new IAE("Invalid workerNumber [%s]", workerNumber);
     }
 
@@ -947,14 +947,18 @@ private void generateResultPartitionsAndBoundariesWithoutKeyStatistics()
         resultPartitionBoundaries = maybeResultPartitionBoundaries.valueOrThrow();
         resultPartitions = ReadablePartitions.striped(
             stageNumber,
-            workerCount,
+            workerInputs.workers(),
             resultPartitionBoundaries.size()
         );
-      } else if (shuffleSpec.kind() == ShuffleKind.MIX) {
-        resultPartitionBoundaries = ClusterByPartitions.oneUniversalPartition();
-        resultPartitions = ReadablePartitions.striped(stageNumber, workerCount, shuffleSpec.partitionCount());
       } else {
-        resultPartitions = ReadablePartitions.striped(stageNumber, workerCount, shuffleSpec.partitionCount());
+        if (shuffleSpec.kind() == ShuffleKind.MIX) {
+          resultPartitionBoundaries = ClusterByPartitions.oneUniversalPartition();
+        }
+        resultPartitions = ReadablePartitions.striped(
+            stageNumber,
+            workerInputs.workers(),
+            shuffleSpec.partitionCount()
+        );
       }
     } else {
       // No reshuffling: retain partitioning from nonbroadcast inputs.