Examples with PlumberSchool io.druid.segment.realtime.plumber.PlumberSchool used on opensource projects

Example 1 with PlumberSchool

use of io.druid.segment.realtime.plumber.PlumberSchool in project druid by druid-io.

the class RealtimeManagerTest method setUp.

@Before
public void setUp() throws Exception {
    final List<TestInputRowHolder> rows = Arrays.asList(makeRow(new DateTime("9000-01-01").getMillis()), makeRow(new ParseException("parse error")), null, makeRow(new DateTime().getMillis()));
    ObjectMapper jsonMapper = new DefaultObjectMapper();
    schema = new DataSchema("test", null, new AggregatorFactory[] { new CountAggregatorFactory("rows") }, new UniformGranularitySpec(Granularities.HOUR, Granularities.NONE, null), jsonMapper);
    schema2 = new DataSchema("testV2", null, new AggregatorFactory[] { new CountAggregatorFactory("rows") }, new UniformGranularitySpec(Granularities.HOUR, Granularities.NONE, null), jsonMapper);
    RealtimeIOConfig ioConfig = new RealtimeIOConfig(new FirehoseFactory() {

        @Override
        public Firehose connect(InputRowParser parser) throws IOException {
            return new TestFirehose(rows.iterator());
        }
    }, new PlumberSchool() {

        @Override
        public Plumber findPlumber(DataSchema schema, RealtimeTuningConfig config, FireDepartmentMetrics metrics) {
            return plumber;
        }
    }, null);
    RealtimeIOConfig ioConfig2 = new RealtimeIOConfig(null, new PlumberSchool() {

        @Override
        public Plumber findPlumber(DataSchema schema, RealtimeTuningConfig config, FireDepartmentMetrics metrics) {
            return plumber2;
        }
    }, new FirehoseFactoryV2() {

        @Override
        public FirehoseV2 connect(InputRowParser parser, Object arg1) throws IOException, ParseException {
            return new TestFirehoseV2(rows.iterator());
        }
    });
    RealtimeTuningConfig tuningConfig = new RealtimeTuningConfig(1, new Period("P1Y"), null, null, null, null, null, null, null, null, 0, 0, null, null);
    plumber = new TestPlumber(new Sink(new Interval("0/P5000Y"), schema, tuningConfig.getShardSpec(), new DateTime().toString(), tuningConfig.getMaxRowsInMemory(), tuningConfig.isReportParseExceptions()));
    realtimeManager = new RealtimeManager(Arrays.<FireDepartment>asList(new FireDepartment(schema, ioConfig, tuningConfig)), null);
    plumber2 = new TestPlumber(new Sink(new Interval("0/P5000Y"), schema2, tuningConfig.getShardSpec(), new DateTime().toString(), tuningConfig.getMaxRowsInMemory(), tuningConfig.isReportParseExceptions()));
    realtimeManager2 = new RealtimeManager(Arrays.<FireDepartment>asList(new FireDepartment(schema2, ioConfig2, tuningConfig)), null);
    tuningConfig_0 = new RealtimeTuningConfig(1, new Period("P1Y"), null, null, null, null, null, new LinearShardSpec(0), null, null, 0, 0, null, null);
    tuningConfig_1 = new RealtimeTuningConfig(1, new Period("P1Y"), null, null, null, null, null, new LinearShardSpec(1), null, null, 0, 0, null, null);
    schema3 = new DataSchema("testing", null, new AggregatorFactory[] { new CountAggregatorFactory("ignore") }, new UniformGranularitySpec(Granularities.HOUR, Granularities.NONE, null), jsonMapper);
    FireDepartment department_0 = new FireDepartment(schema3, ioConfig, tuningConfig_0);
    FireDepartment department_1 = new FireDepartment(schema3, ioConfig2, tuningConfig_1);
    QueryRunnerFactoryConglomerate conglomerate = new QueryRunnerFactoryConglomerate() {

        @Override
        public <T, QueryType extends Query<T>> QueryRunnerFactory<T, QueryType> findFactory(QueryType query) {
            return factory;
        }
    };
    chiefStartedLatch = new CountDownLatch(2);
    RealtimeManager.FireChief fireChief_0 = new RealtimeManager.FireChief(department_0, conglomerate) {

        @Override
        public void run() {
            super.initPlumber();
            chiefStartedLatch.countDown();
        }
    };
    RealtimeManager.FireChief fireChief_1 = new RealtimeManager.FireChief(department_1, conglomerate) {

        @Override
        public void run() {
            super.initPlumber();
            chiefStartedLatch.countDown();
        }
    };
    realtimeManager3 = new RealtimeManager(Arrays.asList(department_0, department_1), conglomerate, ImmutableMap.<String, Map<Integer, RealtimeManager.FireChief>>of("testing", ImmutableMap.of(0, fireChief_0, 1, fireChief_1)));
    startFireChiefWithPartitionNum(fireChief_0, 0);
    startFireChiefWithPartitionNum(fireChief_1, 1);
}

Example 2 with PlumberSchool

use of io.druid.segment.realtime.plumber.PlumberSchool in project druid by druid-io.

the class DruidJsonValidatorTest method testTaskValidator.

@Test
public void testTaskValidator() throws Exception {
    final ObjectMapper jsonMapper = new DefaultObjectMapper();
    for (final Module jacksonModule : new FirehoseModule().getJacksonModules()) {
        jsonMapper.registerModule(jacksonModule);
    }
    final RealtimeIndexTask task = new RealtimeIndexTask(null, new TaskResource("rofl", 2), new FireDepartment(new DataSchema("foo", null, new AggregatorFactory[0], new UniformGranularitySpec(Granularities.HOUR, Granularities.NONE, null), jsonMapper), new RealtimeIOConfig(new LocalFirehoseFactory(new File("lol"), "rofl", null), new PlumberSchool() {

        @Override
        public Plumber findPlumber(DataSchema schema, RealtimeTuningConfig config, FireDepartmentMetrics metrics) {
            return null;
        }
    }, null), new RealtimeTuningConfig(1, new Period("PT10M"), null, null, null, null, 1, NoneShardSpec.instance(), new IndexSpec(), null, 0, 0, true, null)), null);
    File tmp = temporaryFolder.newFile("test_task.json");
    jsonMapper.writeValue(tmp, task);
    parseCommand("validator", "-f", tmp.getAbsolutePath(), "-t", "task").run();
}

Example 3 with PlumberSchool

use of io.druid.segment.realtime.plumber.PlumberSchool in project druid by druid-io.

the class TaskAnnouncementTest method testBackwardsCompatibleSerde.

@Test
public void testBackwardsCompatibleSerde() throws Exception {
    final Task task = new RealtimeIndexTask("theid", new TaskResource("rofl", 2), new FireDepartment(new DataSchema("foo", null, new AggregatorFactory[0], null, new DefaultObjectMapper()), new RealtimeIOConfig(new LocalFirehoseFactory(new File("lol"), "rofl", null), new PlumberSchool() {

        @Override
        public Plumber findPlumber(DataSchema schema, RealtimeTuningConfig config, FireDepartmentMetrics metrics) {
            return null;
        }
    }, null), null), null);
    final TaskStatus status = TaskStatus.running(task.getId());
    final TaskAnnouncement announcement = TaskAnnouncement.create(task, status, TaskLocation.unknown());
    final String statusJson = jsonMapper.writeValueAsString(status);
    final String announcementJson = jsonMapper.writeValueAsString(announcement);
    final TaskStatus statusFromStatus = jsonMapper.readValue(statusJson, TaskStatus.class);
    final TaskStatus statusFromAnnouncement = jsonMapper.readValue(announcementJson, TaskStatus.class);
    final TaskAnnouncement announcementFromStatus = jsonMapper.readValue(statusJson, TaskAnnouncement.class);
    final TaskAnnouncement announcementFromAnnouncement = jsonMapper.readValue(announcementJson, TaskAnnouncement.class);
    Assert.assertEquals("theid", statusFromStatus.getId());
    Assert.assertEquals("theid", statusFromAnnouncement.getId());
    Assert.assertEquals("theid", announcementFromStatus.getTaskStatus().getId());
    Assert.assertEquals("theid", announcementFromAnnouncement.getTaskStatus().getId());
    Assert.assertEquals("theid", announcementFromStatus.getTaskResource().getAvailabilityGroup());
    Assert.assertEquals("rofl", announcementFromAnnouncement.getTaskResource().getAvailabilityGroup());
    Assert.assertEquals(1, announcementFromStatus.getTaskResource().getRequiredCapacity());
    Assert.assertEquals(2, announcementFromAnnouncement.getTaskResource().getRequiredCapacity());
}

Example 4 with PlumberSchool

use of io.druid.segment.realtime.plumber.PlumberSchool 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());
}

Example 5 with PlumberSchool

use of io.druid.segment.realtime.plumber.PlumberSchool in project druid by druid-io.

the class TaskSerdeTest method testRealtimeIndexTaskSerde.

@Test
public void testRealtimeIndexTaskSerde() throws Exception {
    final RealtimeIndexTask task = new RealtimeIndexTask(null, new TaskResource("rofl", 2), new FireDepartment(new DataSchema("foo", null, new AggregatorFactory[0], new UniformGranularitySpec(Granularities.HOUR, Granularities.NONE, null), jsonMapper), new RealtimeIOConfig(new LocalFirehoseFactory(new File("lol"), "rofl", null), new PlumberSchool() {

        @Override
        public Plumber findPlumber(DataSchema schema, RealtimeTuningConfig config, FireDepartmentMetrics metrics) {
            return null;
        }
    }, null), new RealtimeTuningConfig(1, new Period("PT10M"), null, null, null, null, 1, NoneShardSpec.instance(), indexSpec, null, 0, 0, true, null)), null);
    final String json = jsonMapper.writeValueAsString(task);
    // Just want to run the clock a bit to make sure the task id doesn't change
    Thread.sleep(100);
    final RealtimeIndexTask task2 = (RealtimeIndexTask) jsonMapper.readValue(json, Task.class);
    Assert.assertEquals("foo", task.getDataSource());
    Assert.assertEquals(2, task.getTaskResource().getRequiredCapacity());
    Assert.assertEquals("rofl", task.getTaskResource().getAvailabilityGroup());
    Assert.assertEquals(new Period("PT10M"), task.getRealtimeIngestionSchema().getTuningConfig().getWindowPeriod());
    Assert.assertEquals(Granularities.HOUR, task.getRealtimeIngestionSchema().getDataSchema().getGranularitySpec().getSegmentGranularity());
    Assert.assertTrue(task.getRealtimeIngestionSchema().getTuningConfig().isReportParseExceptions());
    Assert.assertEquals(task.getId(), task2.getId());
    Assert.assertEquals(task.getGroupId(), task2.getGroupId());
    Assert.assertEquals(task.getDataSource(), task2.getDataSource());
    Assert.assertEquals(task.getTaskResource().getRequiredCapacity(), task2.getTaskResource().getRequiredCapacity());
    Assert.assertEquals(task.getTaskResource().getAvailabilityGroup(), task2.getTaskResource().getAvailabilityGroup());
    Assert.assertEquals(task.getRealtimeIngestionSchema().getTuningConfig().getWindowPeriod(), task2.getRealtimeIngestionSchema().getTuningConfig().getWindowPeriod());
    Assert.assertEquals(task.getRealtimeIngestionSchema().getDataSchema().getGranularitySpec().getSegmentGranularity(), task2.getRealtimeIngestionSchema().getDataSchema().getGranularitySpec().getSegmentGranularity());
}