FrankChen021 commented on code in PR #19772:
URL: https://github.com/apache/druid/pull/19772#discussion_r3690775099
##########
indexing-service/src/main/java/org/apache/druid/indexing/overlord/duty/UnusedSegmentsKiller.java:
##########
@@ -436,33 +509,27 @@ private EmbeddedKillTask(
{
super(
taskId,
- candidate.dataSource,
- candidate.interval,
- null,
- Map.of(Tasks.PRIORITY_KEY,
Tasks.DEFAULT_EMBEDDED_KILL_TASK_PRIORITY),
- null,
+ candidate.dataSource(),
+ candidate.interval(),
null,
+ Map.of(
+ Tasks.PRIORITY_KEY, Tasks.DEFAULT_EMBEDDED_KILL_TASK_PRIORITY,
+ Tasks.USE_CONCURRENT_LOCKS, useConcurrentLocks
Review Comment:
I don't think REPLACE alone closes this issue. Ingestion defaults to
`useConcurrentLocks=false` (`EXCLUSIVE`), and REPLACE remains incompatible with
EXCLUSIVE, SHARED, and REPLACE locks. Since skipped candidates are discarded
and the capped SQL subquery has neither ordering nor a cursor, the same
long-lived locked rows can be returned on every rebuild and keep rows beyond
`maxSegmentsToKill` from ever entering the queue. Please make the scan advance
past skipped candidates or otherwise guarantee progress.
Reviewed 29 of 29 changed files.
##########
indexing-service/src/main/java/org/apache/druid/indexing/overlord/SegmentStatusManager.java:
##########
@@ -0,0 +1,254 @@
+/*
+ * 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.indexing.overlord;
+
+import com.google.inject.Inject;
+import org.apache.druid.common.utils.IdUtils;
+import org.apache.druid.error.DruidException;
+import org.apache.druid.error.InternalServerError;
+import org.apache.druid.indexer.TaskStatus;
+import org.apache.druid.indexing.common.TaskLock;
+import org.apache.druid.indexing.common.TaskLockType;
+import org.apache.druid.indexing.common.TaskToolbox;
+import org.apache.druid.indexing.common.actions.TaskActionClient;
+import org.apache.druid.indexing.common.actions.TaskActionClientFactory;
+import org.apache.druid.indexing.common.actions.TimeChunkLockTryAcquireAction;
+import org.apache.druid.indexing.common.task.AbstractFixedIntervalTask;
+import org.apache.druid.indexing.common.task.Task;
+import org.apache.druid.indexing.common.task.TaskMetrics;
+import org.apache.druid.java.util.common.Intervals;
+import org.apache.druid.java.util.common.JodaUtils;
+import org.apache.druid.java.util.common.Stopwatch;
+import org.apache.druid.java.util.common.logger.Logger;
+import org.apache.druid.java.util.emitter.service.ServiceEmitter;
+import org.apache.druid.java.util.emitter.service.ServiceMetricEvent;
+import org.apache.druid.query.DruidMetrics;
+import org.apache.druid.timeline.SegmentId;
+import org.joda.time.Interval;
+
+import javax.annotation.Nullable;
+import java.util.List;
+import java.util.Map;
+import java.util.Set;
+
+/**
+ * Performs updates on segments to change their "used" status in the metadata
store.
+ * <p>
+ * Currently, this class exposes methods that are used to acquire an EXCLUSIVE
+ * lock while marking segments in an interval as "used". A similar restriction
+ * may be imposed on marking (non-overshadowed) segments as unused in the
future.
+ *
+ * @see #markAsUsedWithExclusiveLock(String, Interval, UpdateOperation)
+ * Reasons for using EXCLUSIVE locks
+ */
+public class SegmentStatusManager
+{
+ private static final Logger log = new Logger(SegmentStatusManager.class);
+ private static final String TASK_TYPE_MARK_USED = "markSegmentAsUsed";
+
+ private final ServiceEmitter emitter;
+ private final GlobalTaskLockbox taskLockbox;
+ private final TaskActionClientFactory taskActionClientFactory;
+ private final IndexerMetadataStorageCoordinator storageCoordinator;
+
+ @Inject
+ public SegmentStatusManager(
+ GlobalTaskLockbox taskLockbox,
+ IndexerMetadataStorageCoordinator storageCoordinator,
+ TaskActionClientFactory taskActionClientFactory,
+ ServiceEmitter emitter
+ )
+ {
+ this.emitter = emitter;
+ this.taskLockbox = taskLockbox;
+ this.storageCoordinator = storageCoordinator;
+ this.taskActionClientFactory = taskActionClientFactory;
+ }
+
+ /**
+ * Same as {@link
IndexerMetadataStorageCoordinator#markAllNonOvershadowedSegmentsAsUsed(String)}
+ * but with an EXCLUSIVE lock.
+ */
+ public int markAllNonOvershadowedSegmentsAsUsed(String dataSource)
+ {
+ return markAsUsedWithExclusiveLock(
+ dataSource,
+ Intervals.ETERNITY,
+ () ->
storageCoordinator.markAllNonOvershadowedSegmentsAsUsed(dataSource)
+ );
+ }
+
+ /**
+ * Same as {@link
IndexerMetadataStorageCoordinator#markNonOvershadowedSegmentsAsUsed(String,
Interval, List)}
+ * but with an EXCLUSIVE lock.
+ */
+ public int markNonOvershadowedSegmentsAsUsed(
+ String dataSource,
+ Interval interval,
+ @Nullable List<String> versions
+ )
+ {
+ return markAsUsedWithExclusiveLock(
+ dataSource,
+ interval,
+ () -> storageCoordinator.markNonOvershadowedSegmentsAsUsed(dataSource,
interval, versions)
+ );
+ }
+
+ /**
+ * Same as {@link
IndexerMetadataStorageCoordinator#markNonOvershadowedSegmentsAsUsed(String,
Set)}
+ * but with an EXCLUSIVE lock.
+ */
+ public int markNonOvershadowedSegmentsAsUsed(String dataSource,
Set<SegmentId> segmentIds)
+ {
+ return markAsUsedWithExclusiveLock(
+ dataSource,
+
JodaUtils.umbrellaInterval(segmentIds.stream().map(SegmentId::getInterval).toList()),
+ () -> storageCoordinator.markNonOvershadowedSegmentsAsUsed(dataSource,
segmentIds)
+ );
+ }
+
+ /**
+ * Same as {@link
IndexerMetadataStorageCoordinator#markSegmentAsUsed(SegmentId)}
+ * but with an EXCLUSIVE lock.
+ */
+ public int markSegmentAsUsed(SegmentId segmentId)
+ {
+ return markAsUsedWithExclusiveLock(
+ segmentId.getDataSource(),
+ segmentId.getInterval(),
+ () -> storageCoordinator.markSegmentAsUsed(segmentId) ? 1 : 0
+ );
+ }
+
+ /**
+ * Mark segments for a datasource-interval as used while holding an EXCLUSIVE
+ * lock. Most APIs mark non-overshadowed segments as used. So, as soon as
they
+ * are updated, they would become visible to other metadata operations. This
+ * may cause concurrent append jobs to see an inconsistent view of existing
+ * segments during their lifecycle. It may also cause a concurrent kill task
+ * to accidentally delete files of a segment that has just been marked as
used
+ * and then upgraded.
+ *
+ * @return Number of segments updated.
+ */
+ private int markAsUsedWithExclusiveLock(String dataSource, Interval
interval, UpdateOperation operation)
+ {
+ final Stopwatch taskRunTime = Stopwatch.createStarted();
+
+ final String taskId = IdUtils.newTaskId(TASK_TYPE_MARK_USED, dataSource,
interval);
+ final ExclusiveIntervalDummyTask dummyTask
+ = new ExclusiveIntervalDummyTask(taskId, dataSource, interval);
+
+ final TaskActionClient taskActionClient =
taskActionClientFactory.create(dummyTask);
+
+ final ServiceMetricEvent.Builder metricBuilder = new
ServiceMetricEvent.Builder();
+ metricBuilder.setDimension(DruidMetrics.INTERVAL, interval);
+ metricBuilder.setDimension(DruidMetrics.DATASOURCE, dataSource);
+
+ try {
+ // Acquire lock on the interval before performing the update operation
+ taskLockbox.add(dummyTask);
+ final boolean isReady = dummyTask.isReady(taskActionClient);
+ if (isReady) {
+ return operation.perform();
+ } else {
+ final String message =
+ "Could not acquire lock over interval[%s] of datasource[%s] since
other tasks are in progress."
+ + " Retry after the tasks have completed.";
+ throw DruidException.forPersona(DruidException.Persona.USER)
+ .ofCategory(DruidException.Category.CONFLICT)
+ .build(message, interval, dataSource);
+ }
+ }
+ catch (DruidException e) {
+ throw e;
+ }
+ catch (Exception e) {
+ throw InternalServerError.exception(
+ e,
+ "Could not mark segments in interval[%s] of datasource[%s] as used.",
+ interval, dataSource
+ );
+ }
+ finally {
+ cleanupLocksSilently(dummyTask);
+ emitter.emit(metricBuilder.setMetric(TaskMetrics.RUN_DURATION,
taskRunTime.millisElapsed()));
+ }
+ }
+
+ private void cleanupLocksSilently(Task task)
+ {
+ try {
+ taskLockbox.remove(task);
+ }
+ catch (Throwable t) {
+ log.error(t, "Error while cleaning up locks for embedded task[%s].",
task.getId());
+ }
+ }
+
+ /**
+ * Dummy task used to acquire an {@link TaskLockType#EXCLUSIVE} lock while
marking
+ * segments as used. The lock is acquired to ensure that the operation is
mutually
+ * exclusive with any other task on an overlapping interval and prevent data
+ * losses (due to kill tasks removing the files of an upgraded segment) or
+ * data inconsistencies (due to append tasks seeing multiple versions of used
+ * segments in an interval).
+ */
+ private static class ExclusiveIntervalDummyTask extends
AbstractFixedIntervalTask
+ {
+ public ExclusiveIntervalDummyTask(String id, String dataSource, Interval
interval)
+ {
+ super(id, dataSource, interval, Map.of());
Review Comment:
[P2] Prevent the lock probe from revoking active tasks
ExclusiveIntervalDummyTask has an empty context, so it requests its
EXCLUSIVE lock at priority 0. TaskLockbox.tryLock revokes incompatible locks
below the requester's priority before granting the lock. A mark-used API call
can therefore abort a negative-priority ingestion task and succeed instead of
returning the documented conflict response whenever another task is active. Use
minimum priority or a non-preempting acquisition path so any existing
incompatible lock makes this probe fail.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
