Examples with DataSchema org.apache.druid.segment.indexing.DataSchema used on opensource projects

Example 11 with DataSchema

use of org.apache.druid.segment.indexing.DataSchema in project druid by druid-io.

the class CompactionTask method createIngestionSchema.

/**
 * Generate {@link ParallelIndexIngestionSpec} from input segments.
 *
 * @return an empty list if input segments don't exist. Otherwise, a generated ingestionSpec.
 */
@VisibleForTesting
static List<ParallelIndexIngestionSpec> createIngestionSchema(final TaskToolbox toolbox, final LockGranularity lockGranularityInUse, final SegmentProvider segmentProvider, final PartitionConfigurationManager partitionConfigurationManager, @Nullable final DimensionsSpec dimensionsSpec, @Nullable final ClientCompactionTaskTransformSpec transformSpec, @Nullable final AggregatorFactory[] metricsSpec, @Nullable final ClientCompactionTaskGranularitySpec granularitySpec, final CoordinatorClient coordinatorClient, final SegmentCacheManagerFactory segmentCacheManagerFactory, final RetryPolicyFactory retryPolicyFactory, final boolean dropExisting) throws IOException, SegmentLoadingException {
    NonnullPair<Map<DataSegment, File>, List<TimelineObjectHolder<String, DataSegment>>> pair = prepareSegments(toolbox, segmentProvider, lockGranularityInUse);
    final Map<DataSegment, File> segmentFileMap = pair.lhs;
    final List<TimelineObjectHolder<String, DataSegment>> timelineSegments = pair.rhs;
    if (timelineSegments.size() == 0) {
        return Collections.emptyList();
    }
    // find metadata for interval
    // queryableIndexAndSegments is sorted by the interval of the dataSegment
    final List<NonnullPair<QueryableIndex, DataSegment>> queryableIndexAndSegments = loadSegments(timelineSegments, segmentFileMap, toolbox.getIndexIO());
    final CompactionTuningConfig compactionTuningConfig = partitionConfigurationManager.computeTuningConfig();
    if (granularitySpec == null || granularitySpec.getSegmentGranularity() == null) {
        // original granularity
        final Map<Interval, List<NonnullPair<QueryableIndex, DataSegment>>> intervalToSegments = new TreeMap<>(Comparators.intervalsByStartThenEnd());
        queryableIndexAndSegments.forEach(p -> intervalToSegments.computeIfAbsent(p.rhs.getInterval(), k -> new ArrayList<>()).add(p));
        // unify overlapping intervals to ensure overlapping segments compacting in the same indexSpec
        List<NonnullPair<Interval, List<NonnullPair<QueryableIndex, DataSegment>>>> intervalToSegmentsUnified = new ArrayList<>();
        Interval union = null;
        List<NonnullPair<QueryableIndex, DataSegment>> segments = new ArrayList<>();
        for (Entry<Interval, List<NonnullPair<QueryableIndex, DataSegment>>> entry : intervalToSegments.entrySet()) {
            Interval cur = entry.getKey();
            if (union == null) {
                union = cur;
                segments.addAll(entry.getValue());
            } else if (union.overlaps(cur)) {
                union = Intervals.utc(union.getStartMillis(), Math.max(union.getEndMillis(), cur.getEndMillis()));
                segments.addAll(entry.getValue());
            } else {
                intervalToSegmentsUnified.add(new NonnullPair<>(union, segments));
                union = cur;
                segments = new ArrayList<>(entry.getValue());
            }
        }
        intervalToSegmentsUnified.add(new NonnullPair<>(union, segments));
        final List<ParallelIndexIngestionSpec> specs = new ArrayList<>(intervalToSegmentsUnified.size());
        for (NonnullPair<Interval, List<NonnullPair<QueryableIndex, DataSegment>>> entry : intervalToSegmentsUnified) {
            final Interval interval = entry.lhs;
            final List<NonnullPair<QueryableIndex, DataSegment>> segmentsToCompact = entry.rhs;
            // If granularitySpec is not null, then set segmentGranularity. Otherwise,
            // creates new granularitySpec and set segmentGranularity
            Granularity segmentGranularityToUse = GranularityType.fromPeriod(interval.toPeriod()).getDefaultGranularity();
            final DataSchema dataSchema = createDataSchema(segmentProvider.dataSource, segmentsToCompact, dimensionsSpec, transformSpec, metricsSpec, granularitySpec == null ? new ClientCompactionTaskGranularitySpec(segmentGranularityToUse, null, null) : granularitySpec.withSegmentGranularity(segmentGranularityToUse));
            specs.add(new ParallelIndexIngestionSpec(dataSchema, createIoConfig(toolbox, dataSchema, interval, coordinatorClient, segmentCacheManagerFactory, retryPolicyFactory, dropExisting), compactionTuningConfig));
        }
        return specs;
    } else {
        // given segment granularity
        final DataSchema dataSchema = createDataSchema(segmentProvider.dataSource, queryableIndexAndSegments, dimensionsSpec, transformSpec, metricsSpec, granularitySpec);
        return Collections.singletonList(new ParallelIndexIngestionSpec(dataSchema, createIoConfig(toolbox, dataSchema, segmentProvider.interval, coordinatorClient, segmentCacheManagerFactory, retryPolicyFactory, dropExisting), compactionTuningConfig));
    }
}

Example 12 with DataSchema

use of org.apache.druid.segment.indexing.DataSchema in project druid by druid-io.

the class CompactionTask method createDataSchema.

private static DataSchema createDataSchema(String dataSource, List<NonnullPair<QueryableIndex, DataSegment>> queryableIndexAndSegments, @Nullable DimensionsSpec dimensionsSpec, @Nullable ClientCompactionTaskTransformSpec transformSpec, @Nullable AggregatorFactory[] metricsSpec, @Nonnull ClientCompactionTaskGranularitySpec granularitySpec) {
    // check index metadata &
    // Decide which values to propagate (i.e. carry over) for rollup & queryGranularity
    final SettableSupplier<Boolean> rollup = new SettableSupplier<>();
    final SettableSupplier<Granularity> queryGranularity = new SettableSupplier<>();
    decideRollupAndQueryGranularityCarryOver(rollup, queryGranularity, queryableIndexAndSegments);
    final Interval totalInterval = JodaUtils.umbrellaInterval(queryableIndexAndSegments.stream().map(p -> p.rhs.getInterval()).collect(Collectors.toList()));
    final Granularity queryGranularityToUse;
    if (granularitySpec.getQueryGranularity() == null) {
        queryGranularityToUse = queryGranularity.get();
        log.info("Generate compaction task spec with segments original query granularity [%s]", queryGranularityToUse);
    } else {
        queryGranularityToUse = granularitySpec.getQueryGranularity();
        log.info("Generate compaction task spec with new query granularity overrided from input [%s]", queryGranularityToUse);
    }
    final GranularitySpec uniformGranularitySpec = new UniformGranularitySpec(Preconditions.checkNotNull(granularitySpec.getSegmentGranularity()), queryGranularityToUse, granularitySpec.isRollup() == null ? rollup.get() : granularitySpec.isRollup(), Collections.singletonList(totalInterval));
    // find unique dimensions
    final DimensionsSpec finalDimensionsSpec = dimensionsSpec == null ? createDimensionsSpec(queryableIndexAndSegments) : dimensionsSpec;
    final AggregatorFactory[] finalMetricsSpec = metricsSpec == null ? createMetricsSpec(queryableIndexAndSegments) : metricsSpec;
    return new DataSchema(dataSource, new TimestampSpec(ColumnHolder.TIME_COLUMN_NAME, "millis", null), finalDimensionsSpec, finalMetricsSpec, uniformGranularitySpec, transformSpec == null ? null : new TransformSpec(transformSpec.getFilter(), null));
}

Example 13 with DataSchema

use of org.apache.druid.segment.indexing.DataSchema 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("Plumber must be null");
    }
    setupTimeoutAlert();
    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(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 long lockTimeoutMs = getContextValue(Tasks.LOCK_TIMEOUT_KEY, Tasks.DEFAULT_LOCK_TIMEOUT_MILLIS);
    // Note: if lockTimeoutMs is larger than ServerConfig.maxIdleTime, http timeout error can occur while waiting for a
    // lock to be acquired.
    final DataSegmentAnnouncer lockingSegmentAnnouncer = new DataSegmentAnnouncer() {

        @Override
        public void announceSegment(final DataSegment segment) throws IOException {
            // Side effect: Calling announceSegment causes a lock to be acquired
            final TaskLock lock = Preconditions.checkNotNull(toolbox.getTaskActionClient().submit(new TimeChunkLockAcquireAction(TaskLockType.EXCLUSIVE, segment.getInterval(), lockTimeoutMs)), "Cannot acquire a lock for interval[%s]", segment.getInterval());
            if (lock.isRevoked()) {
                throw new ISE(StringUtils.format("Lock for interval [%s] was revoked.", 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) {
                final TaskLock lock = Preconditions.checkNotNull(toolbox.getTaskActionClient().submit(new TimeChunkLockAcquireAction(TaskLockType.EXCLUSIVE, segment.getInterval(), lockTimeoutMs)), "Cannot acquire a lock for interval[%s]", segment.getInterval());
                if (lock.isRevoked()) {
                    throw new ISE(StringUtils.format("Lock for interval [%s] was revoked.", 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()));
                }
            }
        }
    };
    // 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 TimeChunkLockAcquireAction action = new TimeChunkLockAcquireAction(TaskLockType.EXCLUSIVE, interval, lockTimeoutMs);
                final TaskLock lock = Preconditions.checkNotNull(toolbox.getTaskActionClient().submit(action), "Cannot acquire a lock for interval[%s]", interval);
                if (lock.isRevoked()) {
                    throw new ISE(StringUtils.format("Lock for interval [%s] was revoked.", interval));
                }
                return lock.getVersion();
            } catch (IOException e) {
                throw new RuntimeException(e);
            }
        }
    };
    DataSchema dataSchema = spec.getDataSchema();
    RealtimeIOConfig realtimeIOConfig = spec.getIOConfig();
    RealtimeTuningConfig tuningConfig = spec.getTuningConfig().withBasePersistDirectory(toolbox.getPersistDir()).withVersioningPolicy(versioningPolicy);
    final FireDepartment fireDepartment = new FireDepartment(dataSchema, realtimeIOConfig, tuningConfig);
    this.metrics = fireDepartment.getMetrics();
    final RealtimeMetricsMonitor metricsMonitor = TaskRealtimeMetricsMonitorBuilder.build(this, fireDepartment);
    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) to mismatch, or it can cause historical nodes to load
    // NOTE: different instances of the "same" segment.
    final PlumberSchool plumberSchool = new RealtimePlumberSchool(toolbox.getEmitter(), toolbox.getQueryRunnerFactoryConglomerate(), toolbox.getSegmentPusher(), lockingSegmentAnnouncer, segmentPublisher, toolbox.getSegmentHandoffNotifierFactory(), toolbox.getQueryProcessingPool(), toolbox.getJoinableFactory(), toolbox.getIndexMergerV9(), toolbox.getIndexIO(), toolbox.getCache(), toolbox.getCacheConfig(), toolbox.getCachePopulatorStats(), toolbox.getJsonMapper());
    this.plumber = plumberSchool.findPlumber(dataSchema, tuningConfig, metrics);
    final Supplier<Committer> committerSupplier = Committers.nilSupplier();
    LookupNodeService lookupNodeService = getContextValue(CTX_KEY_LOOKUP_TIER) == null ? toolbox.getLookupNodeService() : new LookupNodeService((String) getContextValue(CTX_KEY_LOOKUP_TIER));
    DiscoveryDruidNode discoveryDruidNode = new DiscoveryDruidNode(toolbox.getDruidNode(), NodeRole.PEON, ImmutableMap.of(toolbox.getDataNodeService().getName(), toolbox.getDataNodeService(), lookupNodeService.getName(), lookupNodeService));
    try {
        toolbox.getDataSegmentServerAnnouncer().announce();
        toolbox.getDruidNodeAnnouncer().announce(discoveryDruidNode);
        plumber.startJob();
        // 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);
        // 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());
            }
        }
        // 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) {
                    CloseableUtils.closeAndSuppressExceptions(firehose, e -> log.warn("Failed to close Firehose"));
                }
                toolbox.removeMonitor(metricsMonitor);
            }
        }
        toolbox.getDataSegmentServerAnnouncer().unannounce();
        toolbox.getDruidNodeAnnouncer().unannounce(discoveryDruidNode);
    }
    log.info("Job done!");
    return TaskStatus.success(getId());
}

Example 14 with DataSchema

use of org.apache.druid.segment.indexing.DataSchema in project druid by druid-io.

the class PartialSegmentGenerateTask method generateSegments.

private List<DataSegment> generateSegments(final TaskToolbox toolbox, final ParallelIndexSupervisorTaskClient taskClient, final InputSource inputSource, final File tmpDir) throws IOException, InterruptedException, ExecutionException, TimeoutException {
    final DataSchema dataSchema = ingestionSchema.getDataSchema();
    final FireDepartment fireDepartmentForMetrics = new FireDepartment(dataSchema, new RealtimeIOConfig(null, null), null);
    final FireDepartmentMetrics fireDepartmentMetrics = fireDepartmentForMetrics.getMetrics();
    final RowIngestionMeters buildSegmentsMeters = toolbox.getRowIngestionMetersFactory().createRowIngestionMeters();
    toolbox.addMonitor(new RealtimeMetricsMonitor(Collections.singletonList(fireDepartmentForMetrics), Collections.singletonMap(DruidMetrics.TASK_ID, new String[] { getId() })));
    final ParallelIndexTuningConfig tuningConfig = ingestionSchema.getTuningConfig();
    final PartitionsSpec partitionsSpec = tuningConfig.getGivenOrDefaultPartitionsSpec();
    final long pushTimeout = tuningConfig.getPushTimeout();
    final SegmentAllocatorForBatch segmentAllocator = createSegmentAllocator(toolbox, taskClient);
    final SequenceNameFunction sequenceNameFunction = segmentAllocator.getSequenceNameFunction();
    final ParseExceptionHandler parseExceptionHandler = new ParseExceptionHandler(buildSegmentsMeters, tuningConfig.isLogParseExceptions(), tuningConfig.getMaxParseExceptions(), tuningConfig.getMaxSavedParseExceptions());
    final boolean useMaxMemoryEstimates = getContextValue(Tasks.USE_MAX_MEMORY_ESTIMATES, Tasks.DEFAULT_USE_MAX_MEMORY_ESTIMATES);
    final Appenderator appenderator = BatchAppenderators.newAppenderator(getId(), toolbox.getAppenderatorsManager(), fireDepartmentMetrics, toolbox, dataSchema, tuningConfig, new ShuffleDataSegmentPusher(supervisorTaskId, getId(), toolbox.getIntermediaryDataManager()), buildSegmentsMeters, parseExceptionHandler, useMaxMemoryEstimates);
    boolean exceptionOccurred = false;
    try (final BatchAppenderatorDriver driver = BatchAppenderators.newDriver(appenderator, toolbox, segmentAllocator)) {
        driver.startJob();
        final SegmentsAndCommitMetadata pushed = InputSourceProcessor.process(dataSchema, driver, partitionsSpec, inputSource, inputSource.needsFormat() ? ParallelIndexSupervisorTask.getInputFormat(ingestionSchema) : null, tmpDir, sequenceNameFunction, inputRowIteratorBuilder, buildSegmentsMeters, parseExceptionHandler, pushTimeout);
        return pushed.getSegments();
    } catch (Exception e) {
        exceptionOccurred = true;
        throw e;
    } finally {
        if (exceptionOccurred) {
            appenderator.closeNow();
        } else {
            appenderator.close();
        }
    }
}

Example 15 with DataSchema

use of org.apache.druid.segment.indexing.DataSchema in project druid by druid-io.

the class AppenderatorDriverRealtimeIndexTaskTest method makeRealtimeTask.

private AppenderatorDriverRealtimeIndexTask makeRealtimeTask(final String taskId, final TransformSpec transformSpec, final boolean reportParseExceptions, final long handoffTimeout, final Boolean logParseExceptions, final Integer maxParseExceptions, final Integer maxSavedParseExceptions, final Integer maxRowsPerSegment, final Long maxTotalRows) {
    DataSchema dataSchema = new DataSchema("test_ds", TestHelper.makeJsonMapper().convertValue(new MapInputRowParser(new TimeAndDimsParseSpec(new TimestampSpec("t", "auto", null), new DimensionsSpec(ImmutableList.of(new StringDimensionSchema("dim1"), new StringDimensionSchema("dim2"), new StringDimensionSchema("dim1t"), new LongDimensionSchema("dimLong"), new FloatDimensionSchema("dimFloat"))))), JacksonUtils.TYPE_REFERENCE_MAP_STRING_OBJECT), new AggregatorFactory[] { new CountAggregatorFactory("rows"), new LongSumAggregatorFactory("met1", "met1") }, new UniformGranularitySpec(Granularities.DAY, Granularities.NONE, null), transformSpec, OBJECT_MAPPER);
    RealtimeIOConfig realtimeIOConfig = new RealtimeIOConfig(new TestFirehoseFactory(), null);
    RealtimeAppenderatorTuningConfig tuningConfig = new RealtimeAppenderatorTuningConfig(null, 1000, null, null, maxRowsPerSegment, maxTotalRows, null, null, null, null, null, null, reportParseExceptions, handoffTimeout, null, null, logParseExceptions, maxParseExceptions, maxSavedParseExceptions);
    return new AppenderatorDriverRealtimeIndexTask(taskId, null, new RealtimeAppenderatorIngestionSpec(dataSchema, realtimeIOConfig, tuningConfig), null) {

        @Override
        protected boolean isFirehoseDrainableByClosing(FirehoseFactory firehoseFactory) {
            return true;
        }
    };
}