feat: MSQ supervisors using mostFragmentedFirst policy and minor compaction can scale taskCount down for minor compactions

docs/data-management/automatic-compaction.md

@@ -163,6 +163,15 @@ The MSQ task engine is available as a compaction engine if you configure auto-co
 
 You can use [MSQ task engine context parameters](../multi-stage-query/reference.md#context-parameters) in `spec.taskContext` when configuring your datasource for automatic compaction, such as setting the maximum number of tasks using the `spec.taskContext.maxNumTasks` parameter. Some of the MSQ task engine context parameters overlap with automatic compaction parameters. When these settings overlap, set one or the other.
 
+#### Scaling task count for minor compactions
+
+[Minor compactions](../api-reference/automatic-compaction-api.md#compaction-policy-mostfragmentedfirst) typically rewrite a small subset of segments and do less work than full compactions. To avoid spawning a full-sized MSQ task topology for that lighter workload, opt in by setting `spec.taskContext.minorCompactionTaskPercent` to a value between 1 and 100. The percent is applied to `maxNumTasks` to derive the task count for minor compaction tasks; the result is floored at 2 (the MSQ minimum of one controller and one worker).
+
+| Parameter | Description | Default value |
+|---|---|---|
+| `minorCompactionTaskPercent` | Percent (1-100) used to scale `maxNumTasks` for MSQ minor compactions. Has no effect on full compactions or on the native compaction engine. By default, minor compactions use the same task count as full compactions; set this parameter to opt into scaling. A starting value in the range of 40 to 50 is reasonable for most workloads. | 100 (no scaling) |
+
+For example, with `maxNumTasks` set to 5 and `minorCompactionTaskPercent` set to 40, a minor compaction task launches with `maxNumTasks` of 2 (40% of 5, rounded, then floored at 2), while a full compaction task on the same datasource still launches with `maxNumTasks` of 5.
 
 #### MSQ task engine limitations
 

embedded-tests/src/test/java/org/apache/druid/testing/embedded/compact/CompactionSupervisorTest.java

@@ -19,8 +19,12 @@
 
 package org.apache.druid.testing.embedded.compact;
 
+import com.google.common.collect.ImmutableList;
 import org.apache.druid.catalog.guice.CatalogClientModule;
 import org.apache.druid.catalog.guice.CatalogCoordinatorModule;
+import org.apache.druid.client.indexing.ClientCompactionTaskQuery;
+import org.apache.druid.client.indexing.ClientMSQContext;
+import org.apache.druid.client.indexing.TaskPayloadResponse;
 import org.apache.druid.common.utils.IdUtils;
 import org.apache.druid.data.input.StringTuple;
 import org.apache.druid.data.input.impl.DimensionsSpec;
@@ -29,6 +33,7 @@
 import org.apache.druid.data.input.impl.StringDimensionSchema;
 import org.apache.druid.data.input.impl.TimestampSpec;
 import org.apache.druid.indexer.CompactionEngine;
+import org.apache.druid.indexer.TaskStatusPlus;
 import org.apache.druid.indexer.partitions.DimensionRangePartitionsSpec;
 import org.apache.druid.indexer.partitions.DynamicPartitionsSpec;
 import org.apache.druid.indexer.partitions.HashedPartitionsSpec;
@@ -50,6 +55,7 @@
 import org.apache.druid.java.util.common.StringUtils;
 import org.apache.druid.java.util.common.granularity.Granularities;
 import org.apache.druid.java.util.common.granularity.Granularity;
+import org.apache.druid.java.util.common.parsers.CloseableIterator;
 import org.apache.druid.java.util.emitter.service.ServiceMetricEvent;
 import org.apache.druid.query.DruidMetrics;
 import org.apache.druid.query.aggregation.CountAggregatorFactory;
@@ -113,8 +119,10 @@
 import java.nio.charset.StandardCharsets;
 import java.time.Duration;
 import java.util.ArrayList;
+import java.util.HashMap;
 import java.util.List;
 import java.util.Map;
+import java.util.function.Predicate;
 import java.util.stream.Collectors;
 
 /**
@@ -296,6 +304,86 @@ public void test_minorCompactionWithMSQ(MostFragmentedIntervalFirstPolicy policy
     Assertions.assertEquals(4000, getTotalRowCount());
   }
 
+  @Test
+  public void test_minorCompactionWithMSQ_minorCompactionTaskPercentScalesWorkers()
+  {
+    final MostFragmentedIntervalFirstPolicy policy =
+        new MostFragmentedIntervalFirstPolicy(2, new HumanReadableBytes("1KiB"), null, 80, null, null);
+    configureCompaction(CompactionEngine.MSQ, policy);
+
+    ingest1kRecords();
+    ingest1kRecords();
+
+    overlord.latchableEmitter().waitForNextEvent(event -> event.hasMetricName("segment/metadataCache/sync/time"));
+    cluster.callApi().waitForAllSegmentsToBeAvailable(dataSource, coordinator, broker);
+    Assertions.assertEquals(2, getNumSegmentsWith(Granularities.DAY));
+
+    // maxNumTasks=3 (controller + 2 workers) for full; minorCompactionTaskPercent=50
+    // scales minor compactions to maxNumTasks=2 (controller + 1 worker).
+    final InlineSchemaDataSourceCompactionConfig dayConfig =
+        InlineSchemaDataSourceCompactionConfig
+            .builder()
+            .forDataSource(dataSource)
+            .withSkipOffsetFromLatest(Period.seconds(0))
+            .withGranularitySpec(new UserCompactionTaskGranularityConfig(Granularities.DAY, null, false))
+            .withDimensionsSpec(new UserCompactionTaskDimensionsConfig(
+                WikipediaStreamEventStreamGenerator.dimensions()
+                                                   .stream()
+                                                   .map(StringDimensionSchema::new)
+                                                   .collect(Collectors.toUnmodifiableList())))
+            .withTaskContext(buildScalingTaskContext(3, 50))
+            .withIoConfig(new UserCompactionTaskIOConfig(true))
+            .withTuningConfig(
+                UserCompactionTaskQueryTuningConfig
+                    .builder()
+                    .partitionsSpec(new DimensionRangePartitionsSpec(null, 10_000, List.of("page"), false))
+                    .build()
+            )
+            .build();
+
+    // First run: full compaction (no compacted segments yet).
+    runCompactionWithSpec(dayConfig);
+    waitForAllCompactionTasksToFinish();
+    pauseCompaction(dayConfig);
+
+    overlord.latchableEmitter().waitForNextEvent(event -> event.hasMetricName("segment/metadataCache/sync/time"));
+    cluster.callApi().waitForAllSegmentsToBeAvailable(dataSource, coordinator, broker);
+    Assertions.assertEquals(1, getNumSegmentsWith(Granularities.DAY));
+
+    // Capture the full task before the minor task arrives (its ID does not contain "-minor").
+    final TaskStatusPlus fullTask = findMostRecentCompactionTask(taskId -> !taskId.contains("-minor"));
+    Assertions.assertNotNull(fullTask, "Expected a full compaction task for datasource[" + dataSource + "]");
+    assertCompactionTaskMaxNumTasks(fullTask, 3);
+
+    // Ingest more so the next run sees a mix of compacted + uncompacted segments.
+    ingest1kRecords();
+    ingest1kRecords();
+
+    overlord.latchableEmitter().waitForNextEvent(event -> event.hasMetricName("segment/metadataCache/sync/time"));
+    cluster.callApi().waitForAllSegmentsToBeAvailable(dataSource, coordinator, broker);
+    Assertions.assertEquals(3, getNumSegmentsWith(Granularities.DAY));
+
+    final long totalUsed = overlord.latchableEmitter().getMetricValues(
+        "segment/metadataCache/used/count",
+        Map.of(DruidMetrics.DATASOURCE, dataSource)
+    ).stream().reduce((first, second) -> second).orElse(0).longValue();
+
+    // Second run: minor compaction (uncompacted bytes ratio is below the 80% threshold).
+    runCompactionWithSpec(dayConfig);
+    waitForAllCompactionTasksToFinish();
+
+    overlord.latchableEmitter().waitForEvent(
+        event -> event.hasMetricName("segment/metadataCache/used/count")
+                      .hasDimension(DruidMetrics.DATASOURCE, dataSource)
+                      .hasValueMatching(Matchers.greaterThan(totalUsed)));
+
+    Assertions.assertEquals(2, getNumSegmentsWith(Granularities.DAY));
+
+    final TaskStatusPlus minorTask = findMostRecentCompactionTask(taskId -> taskId.contains("-minor"));
+    Assertions.assertNotNull(minorTask, "Expected a minor compaction task for datasource[" + dataSource + "]");
+    assertCompactionTaskMaxNumTasks(minorTask, 2);
+  }
+
   @MethodSource("getEngine")
   @ParameterizedTest(name = "compactionEngine={0}")
   public void test_compaction_withPersistLastCompactionStateFalse_storesOnlyFingerprint(CompactionEngine compactionEngine)
@@ -1208,6 +1296,50 @@ private int getNumSegmentsWith(Granularity granularity)
         .count();
   }
 
+  private static Map<String, Object> buildScalingTaskContext(int maxNumTasks, int minorCompactionTaskPercent)
+  {
+    final Map<String, Object> context = new HashMap<>();
+    context.put("useConcurrentLocks", true);
+    context.put(ClientMSQContext.CTX_MAX_NUM_TASKS, maxNumTasks);
+    context.put(ClientMSQContext.CTX_MINOR_COMPACTION_TASK_PERCENT, minorCompactionTaskPercent);
+    return context;
+  }
+
+  /**
+   * Returns the most recent completed compaction task whose id matches the given filter,
+   * or null if none is found.
+   */
+  @Nullable
+  private TaskStatusPlus findMostRecentCompactionTask(Predicate<String> taskIdFilter)
+  {
+    final List<TaskStatusPlus> tasks = ImmutableList.copyOf(
+        (CloseableIterator<TaskStatusPlus>) cluster.callApi().onLeaderOverlord(
+            o -> o.taskStatuses("complete", dataSource, 100)
+        )
+    );
+    TaskStatusPlus match = null;
+    for (TaskStatusPlus task : tasks) {
+      if ("compact".equals(task.getType()) && taskIdFilter.test(task.getId())) {
+        match = task;
+      }
+    }
+    return match;
+  }
+
+  /**
+   * Asserts that the submitted compaction task carries the expected {@code maxNumTasks} in its context.
+   */
+  private void assertCompactionTaskMaxNumTasks(TaskStatusPlus task, int expectedMaxNumTasks)
+  {
+    final TaskPayloadResponse payload = cluster.callApi().onLeaderOverlord(o -> o.taskPayload(task.getId()));
+    final ClientCompactionTaskQuery query = (ClientCompactionTaskQuery) payload.getPayload();
+    Assertions.assertEquals(
+        expectedMaxNumTasks,
+        ((Number) query.getContext().get(ClientMSQContext.CTX_MAX_NUM_TASKS)).intValue(),
+        "maxNumTasks in compaction task[" + task.getId() + "] context"
+    );
+  }
+
   private void runIngestionAtGranularity(
       String granularity,
       String inlineDataCsv

server/src/main/java/org/apache/druid/client/indexing/ClientMSQContext.java

@@ -32,4 +32,11 @@ public class ClientMSQContext
    * Limit to ensure that an MSQ compaction task doesn't take up all task slots in a cluster.
    */
   public static final int MAX_TASK_SLOTS_FOR_MSQ_COMPACTION_TASK = 5;
+
+  /**
+   * Compaction-only context key (1-100) used to scale {@link #CTX_MAX_NUM_TASKS} for MSQ minor compactions.
+   * The scaled value is floored at {@link #DEFAULT_MAX_NUM_TASKS}. Has no effect on full compactions or on the native
+   * compaction engine.
+   */
+  public static final String CTX_MINOR_COMPACTION_TASK_PERCENT = "minorCompactionTaskPercent";
 }

server/src/main/java/org/apache/druid/server/coordinator/duty/CompactSegments.java

@@ -32,15 +32,18 @@
 import org.apache.druid.client.indexing.ClientCompactionTaskGranularitySpec;
 import org.apache.druid.client.indexing.ClientCompactionTaskQuery;
 import org.apache.druid.client.indexing.ClientCompactionTaskQueryTuningConfig;
+import org.apache.druid.client.indexing.ClientMSQContext;
 import org.apache.druid.client.indexing.ClientMinorCompactionInputSpec;
 import org.apache.druid.common.guava.FutureUtils;
 import org.apache.druid.common.utils.IdUtils;
 import org.apache.druid.data.input.impl.AggregateProjectionSpec;
 import org.apache.druid.error.DruidException;
+import org.apache.druid.error.InvalidInput;
 import org.apache.druid.indexer.CompactionEngine;
 import org.apache.druid.java.util.common.granularity.Granularity;
 import org.apache.druid.java.util.common.granularity.GranularityType;
 import org.apache.druid.java.util.common.logger.Logger;
+import org.apache.druid.query.QueryContext;
 import org.apache.druid.query.SegmentDescriptor;
 import org.apache.druid.query.aggregation.AggregatorFactory;
 import org.apache.druid.rpc.indexing.OverlordClient;
@@ -294,6 +297,17 @@ private int submitCompactionTasks(
     return numSubmittedTasks;
   }
 
+  /**
+   * Default percent applied to {@code maxNumTasks} for MSQ minor compactions
+   * when the supervisor's {@code taskContext} does not specify
+   * {@link ClientMSQContext#CTX_MINOR_COMPACTION_TASK_PERCENT}. The default of
+   * 100 disables scaling, matching the opt-in shape of the minor compaction
+   * feature itself (which is gated by non-zero
+   * {@code minUncompactedBytesPercentForFullCompaction} or
+   * {@code minUncompactedRowsPercentForFullCompaction} on the policy).
+   */
+  public static final int DEFAULT_MINOR_COMPACTION_TASK_PERCENT = 100;
+
   /**
    * Creates a {@link ClientCompactionTaskQuery} which can be submitted to an
    * {@link OverlordClient} to start a compaction task.
@@ -448,6 +462,52 @@ public Map<String, AutoCompactionSnapshot> getAutoCompactionSnapshot()
     return autoCompactionSnapshotPerDataSource.get();
   }
 
+  /**
+   * Reduces {@code maxNumTasks} on the task context for MSQ minor compactions
+   * by the percent specified under
+   * {@link ClientMSQContext#CTX_MINOR_COMPACTION_TASK_PERCENT} in the same
+   * context, defaulting to {@link #DEFAULT_MINOR_COMPACTION_TASK_PERCENT} when
+   * absent. The scaled value is floored at {@link ClientMSQContext#DEFAULT_MAX_NUM_TASKS}
+   * (the MSQ minimum of 1 controller + 1 worker). No-op when the engine is
+   * native, the mode is full, or the percent is 100.
+   */
+  private static void maybeScaleMaxNumTasksForMinorCompaction(
+      Map<String, Object> context,
+      CompactionMode compactionMode,
+      ClientCompactionRunnerInfo compactionRunner
+  )
+  {
+    if (compactionMode != CompactionMode.UNCOMPACTED_SEGMENTS_ONLY
+        || !CompactionEngine.MSQ.equals(compactionRunner.getType())) {
+      return;
+    }
+
+    final int percent = QueryContext.of(context).getInt(
+        ClientMSQContext.CTX_MINOR_COMPACTION_TASK_PERCENT,
+        DEFAULT_MINOR_COMPACTION_TASK_PERCENT
+    );
+    if (percent < 1 || percent > 100) {
+      throw InvalidInput.exception(
+          "'%s'[%d] must be between 1 and 100",
+          ClientMSQContext.CTX_MINOR_COMPACTION_TASK_PERCENT,
+          percent
+      );
+    }
+    if (percent == 100) {
+      return;
+    }
+
+    final int originalMaxNumTasks = QueryContext.of(context).getInt(
+        ClientMSQContext.CTX_MAX_NUM_TASKS,
+        ClientMSQContext.DEFAULT_MAX_NUM_TASKS
+    );
+    final int scaledMaxNumTasks = Math.max(
+        ClientMSQContext.DEFAULT_MAX_NUM_TASKS,
+        (int) Math.round(originalMaxNumTasks * percent / 100.0)
+    );
+    context.put(ClientMSQContext.CTX_MAX_NUM_TASKS, scaledMaxNumTasks);
+  }
+
   private static ClientCompactionTaskQuery compactSegments(
       CompactionCandidate entry,
       Eligibility eligibility,
@@ -480,6 +540,8 @@ private static ClientCompactionTaskQuery compactSegments(
       context.put(COMPACTION_POLICY_RESULT, eligibility.getReason());
     }
 
+    maybeScaleMaxNumTasksForMinorCompaction(context, compactionMode, compactionRunner);
+
     String taskIdPrefix = compactionMode == CompactionMode.UNCOMPACTED_SEGMENTS_ONLY
         ? TASK_ID_PREFIX + "-minor"
         : TASK_ID_PREFIX;