Search in sources :

Example 1 with SegmentPublisher

use of io.druid.segment.realtime.SegmentPublisher in project druid by druid-io.

the class RealtimePlumberSchoolTest method setUp.

@Before
public void setUp() throws Exception {
    tmpDir = Files.createTempDir();
    ObjectMapper jsonMapper = new DefaultObjectMapper();
    schema = new DataSchema("test", jsonMapper.convertValue(new StringInputRowParser(new JSONParseSpec(new TimestampSpec("timestamp", "auto", null), new DimensionsSpec(null, null, null), null, null), null), Map.class), new AggregatorFactory[] { new CountAggregatorFactory("rows") }, new UniformGranularitySpec(Granularities.HOUR, Granularities.NONE, null), jsonMapper);
    schema2 = new DataSchema("test", jsonMapper.convertValue(new StringInputRowParser(new JSONParseSpec(new TimestampSpec("timestamp", "auto", null), new DimensionsSpec(null, null, null), null, null), null), Map.class), new AggregatorFactory[] { new CountAggregatorFactory("rows") }, new UniformGranularitySpec(Granularities.YEAR, Granularities.NONE, null), jsonMapper);
    announcer = EasyMock.createMock(DataSegmentAnnouncer.class);
    announcer.announceSegment(EasyMock.<DataSegment>anyObject());
    EasyMock.expectLastCall().anyTimes();
    segmentPublisher = EasyMock.createNiceMock(SegmentPublisher.class);
    dataSegmentPusher = EasyMock.createNiceMock(DataSegmentPusher.class);
    handoffNotifierFactory = EasyMock.createNiceMock(SegmentHandoffNotifierFactory.class);
    handoffNotifier = EasyMock.createNiceMock(SegmentHandoffNotifier.class);
    EasyMock.expect(handoffNotifierFactory.createSegmentHandoffNotifier(EasyMock.anyString())).andReturn(handoffNotifier).anyTimes();
    EasyMock.expect(handoffNotifier.registerSegmentHandoffCallback(EasyMock.<SegmentDescriptor>anyObject(), EasyMock.<Executor>anyObject(), EasyMock.<Runnable>anyObject())).andReturn(true).anyTimes();
    emitter = EasyMock.createMock(ServiceEmitter.class);
    EasyMock.replay(announcer, segmentPublisher, dataSegmentPusher, handoffNotifierFactory, handoffNotifier, emitter);
    tuningConfig = new RealtimeTuningConfig(1, null, null, null, new IntervalStartVersioningPolicy(), rejectionPolicy, null, null, null, buildV9Directly, 0, 0, false, null);
    realtimePlumberSchool = new RealtimePlumberSchool(emitter, new DefaultQueryRunnerFactoryConglomerate(Maps.<Class<? extends Query>, QueryRunnerFactory>newHashMap()), dataSegmentPusher, announcer, segmentPublisher, handoffNotifierFactory, MoreExecutors.sameThreadExecutor(), TestHelper.getTestIndexMerger(), TestHelper.getTestIndexMergerV9(), TestHelper.getTestIndexIO(), MapCache.create(0), FireDepartmentTest.NO_CACHE_CONFIG, TestHelper.getObjectMapper());
    metrics = new FireDepartmentMetrics();
    plumber = (RealtimePlumber) realtimePlumberSchool.findPlumber(schema, tuningConfig, metrics);
}
Also used : ServiceEmitter(com.metamx.emitter.service.ServiceEmitter) DataSegmentPusher(io.druid.segment.loading.DataSegmentPusher) DataSegmentAnnouncer(io.druid.server.coordination.DataSegmentAnnouncer) DefaultQueryRunnerFactoryConglomerate(io.druid.query.DefaultQueryRunnerFactoryConglomerate) CountAggregatorFactory(io.druid.query.aggregation.CountAggregatorFactory) AggregatorFactory(io.druid.query.aggregation.AggregatorFactory) RealtimeTuningConfig(io.druid.segment.indexing.RealtimeTuningConfig) DataSchema(io.druid.segment.indexing.DataSchema) UniformGranularitySpec(io.druid.segment.indexing.granularity.UniformGranularitySpec) SegmentPublisher(io.druid.segment.realtime.SegmentPublisher) Executor(java.util.concurrent.Executor) FireDepartmentMetrics(io.druid.segment.realtime.FireDepartmentMetrics) CountAggregatorFactory(io.druid.query.aggregation.CountAggregatorFactory) SegmentDescriptor(io.druid.query.SegmentDescriptor) StringInputRowParser(io.druid.data.input.impl.StringInputRowParser) TimestampSpec(io.druid.data.input.impl.TimestampSpec) DimensionsSpec(io.druid.data.input.impl.DimensionsSpec) DefaultObjectMapper(io.druid.jackson.DefaultObjectMapper) JSONParseSpec(io.druid.data.input.impl.JSONParseSpec) DefaultObjectMapper(io.druid.jackson.DefaultObjectMapper) ObjectMapper(com.fasterxml.jackson.databind.ObjectMapper) Before(org.junit.Before)

Example 2 with SegmentPublisher

use of io.druid.segment.realtime.SegmentPublisher in project druid by druid-io.

the class RealtimeIndexTask method run.

@Override
public TaskStatus run(final TaskToolbox toolbox) throws Exception {
    runThread = Thread.currentThread();
    if (this.plumber != null) {
        throw new IllegalStateException("WTF?!? run with non-null plumber??!");
    }
    boolean normalExit = true;
    // It would be nice to get the PlumberSchool in the constructor.  Although that will need jackson injectables for
    // stuff like the ServerView, which seems kind of odd?  Perhaps revisit this when Guice has been introduced.
    final SegmentPublisher segmentPublisher = new TaskActionSegmentPublisher(this, toolbox);
    // NOTE: We talk to the coordinator in various places in the plumber and we could be more robust to issues
    // with the coordinator.  Right now, we'll block/throw in whatever thread triggered the coordinator behavior,
    // which will typically be either the main data processing loop or the persist thread.
    // Wrap default DataSegmentAnnouncer such that we unlock intervals as we unannounce segments
    final DataSegmentAnnouncer lockingSegmentAnnouncer = new DataSegmentAnnouncer() {

        @Override
        public void announceSegment(final DataSegment segment) throws IOException {
            // Side effect: Calling announceSegment causes a lock to be acquired
            toolbox.getTaskActionClient().submit(new LockAcquireAction(segment.getInterval()));
            toolbox.getSegmentAnnouncer().announceSegment(segment);
        }

        @Override
        public void unannounceSegment(final DataSegment segment) throws IOException {
            try {
                toolbox.getSegmentAnnouncer().unannounceSegment(segment);
            } finally {
                toolbox.getTaskActionClient().submit(new LockReleaseAction(segment.getInterval()));
            }
        }

        @Override
        public void announceSegments(Iterable<DataSegment> segments) throws IOException {
            // Side effect: Calling announceSegments causes locks to be acquired
            for (DataSegment segment : segments) {
                toolbox.getTaskActionClient().submit(new LockAcquireAction(segment.getInterval()));
            }
            toolbox.getSegmentAnnouncer().announceSegments(segments);
        }

        @Override
        public void unannounceSegments(Iterable<DataSegment> segments) throws IOException {
            try {
                toolbox.getSegmentAnnouncer().unannounceSegments(segments);
            } finally {
                for (DataSegment segment : segments) {
                    toolbox.getTaskActionClient().submit(new LockReleaseAction(segment.getInterval()));
                }
            }
        }

        @Override
        public boolean isAnnounced(DataSegment segment) {
            return toolbox.getSegmentAnnouncer().isAnnounced(segment);
        }
    };
    // NOTE: getVersion will block if there is lock contention, which will block plumber.getSink
    // NOTE: (and thus the firehose)
    // Shouldn't usually happen, since we don't expect people to submit tasks that intersect with the
    // realtime window, but if they do it can be problematic. If we decide to care, we can use more threads in
    // the plumber such that waiting for the coordinator doesn't block data processing.
    final VersioningPolicy versioningPolicy = new VersioningPolicy() {

        @Override
        public String getVersion(final Interval interval) {
            try {
                // Side effect: Calling getVersion causes a lock to be acquired
                final TaskLock myLock = toolbox.getTaskActionClient().submit(new LockAcquireAction(interval));
                return myLock.getVersion();
            } catch (IOException e) {
                throw Throwables.propagate(e);
            }
        }
    };
    DataSchema dataSchema = spec.getDataSchema();
    RealtimeIOConfig realtimeIOConfig = spec.getIOConfig();
    RealtimeTuningConfig tuningConfig = spec.getTuningConfig().withBasePersistDirectory(new File(toolbox.getTaskWorkDir(), "persist")).withVersioningPolicy(versioningPolicy);
    final FireDepartment fireDepartment = new FireDepartment(dataSchema, realtimeIOConfig, tuningConfig);
    this.metrics = fireDepartment.getMetrics();
    final RealtimeMetricsMonitor metricsMonitor = new RealtimeMetricsMonitor(ImmutableList.of(fireDepartment), ImmutableMap.of(DruidMetrics.TASK_ID, new String[] { getId() }));
    this.queryRunnerFactoryConglomerate = toolbox.getQueryRunnerFactoryConglomerate();
    // NOTE: This pusher selects path based purely on global configuration and the DataSegment, which means
    // NOTE: that redundant realtime tasks will upload to the same location. This can cause index.zip
    // NOTE: (partitionNum_index.zip for HDFS data storage) and descriptor.json (partitionNum_descriptor.json for
    // NOTE: HDFS data storage) to mismatch, or it can cause historical nodes to load different instances of
    // NOTE: the "same" segment.
    final PlumberSchool plumberSchool = new RealtimePlumberSchool(toolbox.getEmitter(), toolbox.getQueryRunnerFactoryConglomerate(), toolbox.getSegmentPusher(), lockingSegmentAnnouncer, segmentPublisher, toolbox.getSegmentHandoffNotifierFactory(), toolbox.getQueryExecutorService(), toolbox.getIndexMerger(), toolbox.getIndexMergerV9(), toolbox.getIndexIO(), toolbox.getCache(), toolbox.getCacheConfig(), toolbox.getObjectMapper());
    this.plumber = plumberSchool.findPlumber(dataSchema, tuningConfig, metrics);
    Supplier<Committer> committerSupplier = null;
    try {
        plumber.startJob();
        // Set up metrics emission
        toolbox.getMonitorScheduler().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);
        // Skip connecting firehose if we've been stopped before we got started.
        synchronized (this) {
            if (!gracefullyStopped) {
                firehose = firehoseFactory.connect(spec.getDataSchema().getParser());
                committerSupplier = Committers.supplierFromFirehose(firehose);
            }
        }
        // Time to read data!
        while (firehose != null && (!gracefullyStopped || firehoseDrainableByClosing) && firehose.hasMore()) {
            Plumbers.addNextRow(committerSupplier, firehose, plumber, tuningConfig.isReportParseExceptions(), metrics);
        }
    } catch (Throwable e) {
        normalExit = false;
        log.makeAlert(e, "Exception aborted realtime processing[%s]", dataSchema.getDataSource()).emit();
        throw e;
    } finally {
        if (normalExit) {
            try {
                // Persist if we had actually started.
                if (firehose != null) {
                    log.info("Persisting remaining data.");
                    final Committer committer = committerSupplier.get();
                    final CountDownLatch persistLatch = new CountDownLatch(1);
                    plumber.persist(new Committer() {

                        @Override
                        public Object getMetadata() {
                            return committer.getMetadata();
                        }

                        @Override
                        public void run() {
                            try {
                                committer.run();
                            } finally {
                                persistLatch.countDown();
                            }
                        }
                    });
                    persistLatch.await();
                }
                if (gracefullyStopped) {
                    log.info("Gracefully stopping.");
                } else {
                    log.info("Finishing the job.");
                    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) {
                        plumber.finishJob();
                    }
                }
            } catch (InterruptedException e) {
                log.debug(e, "Interrupted while finishing the job");
            } catch (Exception e) {
                log.makeAlert(e, "Failed to finish realtime task").emit();
                throw e;
            } finally {
                if (firehose != null) {
                    CloseQuietly.close(firehose);
                }
                toolbox.getMonitorScheduler().removeMonitor(metricsMonitor);
            }
        }
    }
    log.info("Job done!");
    return TaskStatus.success(getId());
}
Also used : RealtimeIOConfig(io.druid.segment.indexing.RealtimeIOConfig) DataSegmentAnnouncer(io.druid.server.coordination.DataSegmentAnnouncer) EventReceiverFirehoseFactory(io.druid.segment.realtime.firehose.EventReceiverFirehoseFactory) ClippedFirehoseFactory(io.druid.segment.realtime.firehose.ClippedFirehoseFactory) TimedShutoffFirehoseFactory(io.druid.segment.realtime.firehose.TimedShutoffFirehoseFactory) FirehoseFactory(io.druid.data.input.FirehoseFactory) DataSegment(io.druid.timeline.DataSegment) FireDepartment(io.druid.segment.realtime.FireDepartment) SegmentPublisher(io.druid.segment.realtime.SegmentPublisher) TaskLock(io.druid.indexing.common.TaskLock) LockAcquireAction(io.druid.indexing.common.actions.LockAcquireAction) IOException(java.io.IOException) RealtimePlumberSchool(io.druid.segment.realtime.plumber.RealtimePlumberSchool) PlumberSchool(io.druid.segment.realtime.plumber.PlumberSchool) RealtimeTuningConfig(io.druid.segment.indexing.RealtimeTuningConfig) CountDownLatch(java.util.concurrent.CountDownLatch) LockReleaseAction(io.druid.indexing.common.actions.LockReleaseAction) IOException(java.io.IOException) DataSchema(io.druid.segment.indexing.DataSchema) VersioningPolicy(io.druid.segment.realtime.plumber.VersioningPolicy) RealtimePlumberSchool(io.druid.segment.realtime.plumber.RealtimePlumberSchool) RealtimeMetricsMonitor(io.druid.segment.realtime.RealtimeMetricsMonitor) Committer(io.druid.data.input.Committer) File(java.io.File) Interval(org.joda.time.Interval)

Aggregations

DataSchema (io.druid.segment.indexing.DataSchema)2 RealtimeTuningConfig (io.druid.segment.indexing.RealtimeTuningConfig)2 SegmentPublisher (io.druid.segment.realtime.SegmentPublisher)2 DataSegmentAnnouncer (io.druid.server.coordination.DataSegmentAnnouncer)2 ObjectMapper (com.fasterxml.jackson.databind.ObjectMapper)1 ServiceEmitter (com.metamx.emitter.service.ServiceEmitter)1 Committer (io.druid.data.input.Committer)1 FirehoseFactory (io.druid.data.input.FirehoseFactory)1 DimensionsSpec (io.druid.data.input.impl.DimensionsSpec)1 JSONParseSpec (io.druid.data.input.impl.JSONParseSpec)1 StringInputRowParser (io.druid.data.input.impl.StringInputRowParser)1 TimestampSpec (io.druid.data.input.impl.TimestampSpec)1 TaskLock (io.druid.indexing.common.TaskLock)1 LockAcquireAction (io.druid.indexing.common.actions.LockAcquireAction)1 LockReleaseAction (io.druid.indexing.common.actions.LockReleaseAction)1 DefaultObjectMapper (io.druid.jackson.DefaultObjectMapper)1 DefaultQueryRunnerFactoryConglomerate (io.druid.query.DefaultQueryRunnerFactoryConglomerate)1 SegmentDescriptor (io.druid.query.SegmentDescriptor)1 AggregatorFactory (io.druid.query.aggregation.AggregatorFactory)1 CountAggregatorFactory (io.druid.query.aggregation.CountAggregatorFactory)1