use of org.apache.druid.indexing.common.TaskLock in project druid by druid-io.
the class TaskLockbox method unlock.
/**
* Release lock held for a task on a particular interval. Does nothing if the task does not currently
* hold the mentioned lock.
*
* @param task task to unlock
* @param interval interval to unlock
*/
public void unlock(final Task task, final Interval interval, @Nullable Integer partitionId) {
giant.lock();
try {
final String dataSource = task.getDataSource();
final NavigableMap<DateTime, SortedMap<Interval, List<TaskLockPosse>>> dsRunning = running.get(task.getDataSource());
if (dsRunning == null || dsRunning.isEmpty()) {
return;
}
final SortedMap<Interval, List<TaskLockPosse>> intervalToPosses = dsRunning.get(interval.getStart());
if (intervalToPosses == null || intervalToPosses.isEmpty()) {
return;
}
final List<TaskLockPosse> possesHolder = intervalToPosses.get(interval);
if (possesHolder == null || possesHolder.isEmpty()) {
return;
}
final List<TaskLockPosse> posses = possesHolder.stream().filter(posse -> posse.containsTask(task)).collect(Collectors.toList());
for (TaskLockPosse taskLockPosse : posses) {
final TaskLock taskLock = taskLockPosse.getTaskLock();
final boolean match = (partitionId == null && taskLock.getGranularity() == LockGranularity.TIME_CHUNK) || (partitionId != null && taskLock.getGranularity() == LockGranularity.SEGMENT && ((SegmentLock) taskLock).getPartitionId() == partitionId);
if (match) {
// Remove task from live list
log.info("Removing task[%s] from TaskLock[%s]", task.getId(), taskLock);
final boolean removed = taskLockPosse.removeTask(task);
if (taskLockPosse.isTasksEmpty()) {
log.info("TaskLock is now empty: %s", taskLock);
possesHolder.remove(taskLockPosse);
}
if (possesHolder.isEmpty()) {
intervalToPosses.remove(interval);
}
if (intervalToPosses.isEmpty()) {
dsRunning.remove(interval.getStart());
}
if (running.get(dataSource).size() == 0) {
running.remove(dataSource);
}
// Wake up blocking-lock waiters
lockReleaseCondition.signalAll();
// Remove lock from storage. If it cannot be removed, just ignore the failure.
try {
taskStorage.removeLock(task.getId(), taskLock);
} catch (Exception e) {
log.makeAlert(e, "Failed to clean up lock from storage").addData("task", task.getId()).addData("dataSource", taskLock.getDataSource()).addData("interval", taskLock.getInterval()).addData("version", taskLock.getVersion()).emit();
}
if (!removed) {
log.makeAlert("Lock release without acquire").addData("task", task.getId()).addData("interval", interval).emit();
}
}
}
} finally {
giant.unlock();
}
}
use of org.apache.druid.indexing.common.TaskLock in project druid by druid-io.
the class RangePartitionCachingLocalSegmentAllocatorTest method createTaskLock.
private static TaskLock createTaskLock(Interval interval) {
TaskLock taskLock = EasyMock.mock(TaskLock.class);
EasyMock.expect(taskLock.getInterval()).andStubReturn(interval);
EasyMock.expect(taskLock.getVersion()).andStubReturn(INTERVAL_TO_VERSION.get(interval));
EasyMock.replay(taskLock);
return taskLock;
}
use of org.apache.druid.indexing.common.TaskLock in project druid by druid-io.
the class AppenderatorDriverRealtimeIndexTask method run.
@Override
public TaskStatus run(final TaskToolbox toolbox) {
runThread = Thread.currentThread();
authorizerMapper = toolbox.getAuthorizerMapper();
rowIngestionMeters = toolbox.getRowIngestionMetersFactory().createRowIngestionMeters();
parseExceptionHandler = new ParseExceptionHandler(rowIngestionMeters, spec.getTuningConfig().isLogParseExceptions(), spec.getTuningConfig().getMaxParseExceptions(), spec.getTuningConfig().getMaxSavedParseExceptions());
setupTimeoutAlert();
DataSchema dataSchema = spec.getDataSchema();
RealtimeAppenderatorTuningConfig tuningConfig = spec.getTuningConfig().withBasePersistDirectory(toolbox.getPersistDir());
final FireDepartment fireDepartmentForMetrics = new FireDepartment(dataSchema, new RealtimeIOConfig(null, null), null);
final TaskRealtimeMetricsMonitor metricsMonitor = TaskRealtimeMetricsMonitorBuilder.build(this, fireDepartmentForMetrics, rowIngestionMeters);
this.metrics = fireDepartmentForMetrics.getMetrics();
final Supplier<Committer> committerSupplier = Committers.nilSupplier();
DiscoveryDruidNode discoveryDruidNode = createDiscoveryDruidNode(toolbox);
appenderator = newAppenderator(dataSchema, tuningConfig, metrics, toolbox);
TaskLockType lockType = getContextValue(Tasks.USE_SHARED_LOCK, false) ? TaskLockType.SHARED : TaskLockType.EXCLUSIVE;
StreamAppenderatorDriver driver = newDriver(dataSchema, appenderator, toolbox, metrics, lockType);
try {
log.debug("Found chat handler of class[%s]", toolbox.getChatHandlerProvider().getClass().getName());
toolbox.getChatHandlerProvider().register(getId(), this, false);
if (toolbox.getAppenderatorsManager().shouldTaskMakeNodeAnnouncements()) {
toolbox.getDataSegmentServerAnnouncer().announce();
toolbox.getDruidNodeAnnouncer().announce(discoveryDruidNode);
}
driver.startJob(segmentId -> {
try {
if (lockGranularity == LockGranularity.SEGMENT) {
return toolbox.getTaskActionClient().submit(new SegmentLockAcquireAction(TaskLockType.EXCLUSIVE, segmentId.getInterval(), segmentId.getVersion(), segmentId.getShardSpec().getPartitionNum(), 1000L)).isOk();
} else {
final TaskLock lock = toolbox.getTaskActionClient().submit(new TimeChunkLockAcquireAction(TaskLockType.EXCLUSIVE, segmentId.getInterval(), 1000L));
if (lock == null) {
return false;
}
if (lock.isRevoked()) {
throw new ISE(StringUtils.format("Lock for interval [%s] was revoked.", segmentId.getInterval()));
}
return true;
}
} catch (IOException e) {
throw new RuntimeException(e);
}
});
// Set up metrics emission
toolbox.addMonitor(metricsMonitor);
// Delay firehose connection to avoid claiming input resources while the plumber is starting up.
final FirehoseFactory firehoseFactory = spec.getIOConfig().getFirehoseFactory();
final boolean firehoseDrainableByClosing = isFirehoseDrainableByClosing(firehoseFactory);
int sequenceNumber = 0;
String sequenceName = makeSequenceName(getId(), sequenceNumber);
final TransactionalSegmentPublisher publisher = (mustBeNullOrEmptyOverwriteSegments, mustBeNullOrEmptyDropSegments, segments, commitMetadata) -> {
if (mustBeNullOrEmptyOverwriteSegments != null && !mustBeNullOrEmptyOverwriteSegments.isEmpty()) {
throw new ISE("Stream ingestion task unexpectedly attempted to overwrite segments: %s", SegmentUtils.commaSeparatedIdentifiers(mustBeNullOrEmptyOverwriteSegments));
}
if (mustBeNullOrEmptyDropSegments != null && !mustBeNullOrEmptyDropSegments.isEmpty()) {
throw new ISE("Stream ingestion task unexpectedly attempted to drop segments: %s", SegmentUtils.commaSeparatedIdentifiers(mustBeNullOrEmptyDropSegments));
}
final SegmentTransactionalInsertAction action = SegmentTransactionalInsertAction.appendAction(segments, null, null);
return toolbox.getTaskActionClient().submit(action);
};
// Skip connecting firehose if we've been stopped before we got started.
synchronized (this) {
if (!gracefullyStopped) {
firehose = firehoseFactory.connect(Preconditions.checkNotNull(spec.getDataSchema().getParser(), "inputRowParser"), toolbox.getIndexingTmpDir());
}
}
ingestionState = IngestionState.BUILD_SEGMENTS;
// Time to read data!
while (!gracefullyStopped && firehoseDrainableByClosing && firehose.hasMore()) {
try {
InputRow inputRow = firehose.nextRow();
if (inputRow == null) {
log.debug("Discarded null row, considering thrownAway.");
rowIngestionMeters.incrementThrownAway();
} else {
AppenderatorDriverAddResult addResult = driver.add(inputRow, sequenceName, committerSupplier);
if (addResult.isOk()) {
final boolean isPushRequired = addResult.isPushRequired(tuningConfig.getPartitionsSpec().getMaxRowsPerSegment(), tuningConfig.getPartitionsSpec().getMaxTotalRowsOr(DynamicPartitionsSpec.DEFAULT_MAX_TOTAL_ROWS));
if (isPushRequired) {
publishSegments(driver, publisher, committerSupplier, sequenceName);
sequenceNumber++;
sequenceName = makeSequenceName(getId(), sequenceNumber);
}
} else {
// If we allow continuing, then consider blacklisting the interval for a while to avoid constant checks.
throw new ISE("Could not allocate segment for row with timestamp[%s]", inputRow.getTimestamp());
}
}
} catch (ParseException e) {
handleParseException(e);
}
}
ingestionState = IngestionState.COMPLETED;
if (!gracefullyStopped) {
synchronized (this) {
if (gracefullyStopped) {
// Someone called stopGracefully after we checked the flag. That's okay, just stop now.
log.info("Gracefully stopping.");
} else {
finishingJob = true;
}
}
if (finishingJob) {
log.info("Finishing job...");
// Publish any remaining segments
publishSegments(driver, publisher, committerSupplier, sequenceName);
waitForSegmentPublishAndHandoff(tuningConfig.getPublishAndHandoffTimeout());
}
} else if (firehose != null) {
log.info("Task was gracefully stopped, will persist data before exiting");
persistAndWait(driver, committerSupplier.get());
}
} catch (Throwable e) {
log.makeAlert(e, "Exception aborted realtime processing[%s]", dataSchema.getDataSource()).emit();
errorMsg = Throwables.getStackTraceAsString(e);
toolbox.getTaskReportFileWriter().write(getId(), getTaskCompletionReports());
return TaskStatus.failure(getId(), errorMsg);
} finally {
toolbox.getChatHandlerProvider().unregister(getId());
CloseableUtils.closeAndSuppressExceptions(firehose, e -> log.warn("Failed to close Firehose"));
appenderator.close();
CloseableUtils.closeAndSuppressExceptions(driver, e -> log.warn("Failed to close AppenderatorDriver"));
toolbox.removeMonitor(metricsMonitor);
if (toolbox.getAppenderatorsManager().shouldTaskMakeNodeAnnouncements()) {
toolbox.getDataSegmentServerAnnouncer().unannounce();
toolbox.getDruidNodeAnnouncer().unannounce(discoveryDruidNode);
}
}
log.info("Job done!");
toolbox.getTaskReportFileWriter().write(getId(), getTaskCompletionReports());
return TaskStatus.success(getId());
}
use of org.apache.druid.indexing.common.TaskLock in project druid by druid-io.
the class AbstractBatchIndexTask method tryTimeChunkLock.
protected boolean tryTimeChunkLock(TaskActionClient client, List<Interval> intervals) throws IOException {
// The given intervals are first converted to align with segment granularity. This is because,
// when an overwriting task finds a version for a given input row, it expects the interval
// associated to each version to be equal or larger than the time bucket where the input row falls in.
// See ParallelIndexSupervisorTask.findVersion().
final Iterator<Interval> intervalIterator;
final Granularity segmentGranularity = getSegmentGranularity();
if (segmentGranularity == null) {
intervalIterator = JodaUtils.condenseIntervals(intervals).iterator();
} else {
IntervalsByGranularity intervalsByGranularity = new IntervalsByGranularity(intervals, segmentGranularity);
// the following is calling a condense that does not materialize the intervals:
intervalIterator = JodaUtils.condensedIntervalsIterator(intervalsByGranularity.granularityIntervalsIterator());
}
// Intervals are already condensed to avoid creating too many locks.
// Intervals are also sorted and thus it's safe to compare only the previous interval and current one for dedup.
Interval prev = null;
int locksAcquired = 0;
while (intervalIterator.hasNext()) {
final Interval cur = intervalIterator.next();
if (prev != null && cur.equals(prev)) {
continue;
}
if (maxAllowedLockCount >= 0 && locksAcquired >= maxAllowedLockCount) {
throw new MaxAllowedLocksExceededException(maxAllowedLockCount);
}
prev = cur;
final TaskLock lock = client.submit(new TimeChunkLockTryAcquireAction(TaskLockType.EXCLUSIVE, cur));
if (lock == null) {
return false;
}
if (lock.isRevoked()) {
throw new ISE(StringUtils.format("Lock for interval [%s] was revoked.", cur));
}
locksAcquired++;
}
return true;
}
use of org.apache.druid.indexing.common.TaskLock in project druid by druid-io.
the class TaskLocks method isLockCoversSegments.
public static boolean isLockCoversSegments(NavigableMap<DateTime, List<TaskLock>> taskLockMap, Collection<DataSegment> segments) {
return segments.stream().allMatch(segment -> {
final Entry<DateTime, List<TaskLock>> entry = taskLockMap.floorEntry(segment.getInterval().getStart());
if (entry == null) {
return false;
}
final List<TaskLock> locks = entry.getValue();
return locks.stream().anyMatch(lock -> {
if (lock.getGranularity() == LockGranularity.TIME_CHUNK) {
final TimeChunkLock timeChunkLock = (TimeChunkLock) lock;
return timeChunkLock.getInterval().contains(segment.getInterval()) && timeChunkLock.getDataSource().equals(segment.getDataSource()) && timeChunkLock.getVersion().compareTo(segment.getVersion()) >= 0;
} else {
final SegmentLock segmentLock = (SegmentLock) lock;
return segmentLock.getInterval().contains(segment.getInterval()) && segmentLock.getDataSource().equals(segment.getDataSource()) && segmentLock.getVersion().compareTo(segment.getVersion()) >= 0 && segmentLock.getPartitionId() == segment.getShardSpec().getPartitionNum();
}
});
});
}
Aggregations