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

Example 6 with TuningConfig

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

the class ParallelIndexTuningConfigTest method testSerdeWithMaxRowsPerSegment.

@Test
public void testSerdeWithMaxRowsPerSegment() throws IOException {
    final ParallelIndexTuningConfig tuningConfig = new ParallelIndexTuningConfig(null, null, null, 10, 1000L, null, null, null, null, new DynamicPartitionsSpec(100, 100L), new IndexSpec(new RoaringBitmapSerdeFactory(true), CompressionStrategy.UNCOMPRESSED, CompressionStrategy.LZF, LongEncodingStrategy.LONGS), new IndexSpec(), 1, false, true, 10000L, OffHeapMemorySegmentWriteOutMediumFactory.instance(), null, 250, 100, 20L, new Duration(3600), 128, null, null, false, null, null, null, null, null);
    final byte[] json = mapper.writeValueAsBytes(tuningConfig);
    final ParallelIndexTuningConfig fromJson = (ParallelIndexTuningConfig) mapper.readValue(json, TuningConfig.class);
    Assert.assertEquals(fromJson, tuningConfig);
}

Example 7 with TuningConfig

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

the class ParallelIndexTuningConfigTest method testSerdeWithMaxNumConcurrentSubTasks.

@Test
public void testSerdeWithMaxNumConcurrentSubTasks() throws IOException {
    final int maxNumConcurrentSubTasks = 250;
    final ParallelIndexTuningConfig tuningConfig = new ParallelIndexTuningConfig(null, null, null, 10, 1000L, null, null, null, null, new DynamicPartitionsSpec(100, 100L), new IndexSpec(new RoaringBitmapSerdeFactory(true), CompressionStrategy.UNCOMPRESSED, CompressionStrategy.LZF, LongEncodingStrategy.LONGS), new IndexSpec(), 1, false, true, 10000L, OffHeapMemorySegmentWriteOutMediumFactory.instance(), null, maxNumConcurrentSubTasks, 100, 20L, new Duration(3600), 128, null, null, false, null, null, null, null, null);
    final byte[] json = mapper.writeValueAsBytes(tuningConfig);
    final ParallelIndexTuningConfig fromJson = (ParallelIndexTuningConfig) mapper.readValue(json, TuningConfig.class);
    Assert.assertEquals(fromJson, tuningConfig);
}

Example 8 with TuningConfig

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

the class IndexTaskSerdeTest method assertSerdeTuningConfig.

private static void assertSerdeTuningConfig(IndexTuningConfig tuningConfig) throws IOException {
    final byte[] json = MAPPER.writeValueAsBytes(tuningConfig);
    final IndexTuningConfig fromJson = (IndexTuningConfig) MAPPER.readValue(json, TuningConfig.class);
    Assert.assertEquals(tuningConfig, fromJson);
}

Example 9 with TuningConfig

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

the class AbstractBatchIndexTask method compactionStateAnnotateFunction.

public static Function<Set<DataSegment>, Set<DataSegment>> compactionStateAnnotateFunction(boolean storeCompactionState, TaskToolbox toolbox, IngestionSpec ingestionSpec) {
    if (storeCompactionState) {
        TuningConfig tuningConfig = ingestionSpec.getTuningConfig();
        GranularitySpec granularitySpec = ingestionSpec.getDataSchema().getGranularitySpec();
        // We do not need to store dimensionExclusions and spatialDimensions since auto compaction does not support them
        DimensionsSpec dimensionsSpec = ingestionSpec.getDataSchema().getDimensionsSpec() == null ? null : new DimensionsSpec(ingestionSpec.getDataSchema().getDimensionsSpec().getDimensions());
        // We only need to store filter since that is the only field auto compaction support
        Map<String, Object> transformSpec = ingestionSpec.getDataSchema().getTransformSpec() == null || TransformSpec.NONE.equals(ingestionSpec.getDataSchema().getTransformSpec()) ? null : new ClientCompactionTaskTransformSpec(ingestionSpec.getDataSchema().getTransformSpec().getFilter()).asMap(toolbox.getJsonMapper());
        List<Object> metricsSpec = ingestionSpec.getDataSchema().getAggregators() == null ? null : toolbox.getJsonMapper().convertValue(ingestionSpec.getDataSchema().getAggregators(), new TypeReference<List<Object>>() {
        });
        final CompactionState compactionState = new CompactionState(tuningConfig.getPartitionsSpec(), dimensionsSpec, metricsSpec, transformSpec, tuningConfig.getIndexSpec().asMap(toolbox.getJsonMapper()), granularitySpec.asMap(toolbox.getJsonMapper()));
        return segments -> segments.stream().map(s -> s.withLastCompactionState(compactionState)).collect(Collectors.toSet());
    } else {
        return Function.identity();
    }
}

Example 10 with TuningConfig

use of org.apache.druid.segment.indexing.TuningConfig in project druid by apache.

the class IndexTask method generateAndPublishSegments.

/**
 * This method reads input data row by row and adds the read row to a proper segment using {@link BaseAppenderatorDriver}.
 * If there is no segment for the row, a new one is created.  Segments can be published in the middle of reading inputs
 * if {@link DynamicPartitionsSpec} is used and one of below conditions are satisfied.
 *
 * <ul>
 * <li>
 * If the number of rows in a segment exceeds {@link DynamicPartitionsSpec#maxRowsPerSegment}
 * </li>
 * <li>
 * If the number of rows added to {@link BaseAppenderatorDriver} so far exceeds {@link DynamicPartitionsSpec#maxTotalRows}
 * </li>
 * </ul>
 * <p>
 * At the end of this method, all the remaining segments are published.
 *
 * @return the last {@link TaskStatus}
 */
private TaskStatus generateAndPublishSegments(final TaskToolbox toolbox, final DataSchema dataSchema, final InputSource inputSource, final File tmpDir, final PartitionAnalysis partitionAnalysis) throws IOException, InterruptedException {
    final FireDepartment fireDepartmentForMetrics = new FireDepartment(dataSchema, new RealtimeIOConfig(null, null), null);
    FireDepartmentMetrics buildSegmentsFireDepartmentMetrics = fireDepartmentForMetrics.getMetrics();
    if (toolbox.getMonitorScheduler() != null) {
        final TaskRealtimeMetricsMonitor metricsMonitor = TaskRealtimeMetricsMonitorBuilder.build(this, fireDepartmentForMetrics, buildSegmentsMeters);
        toolbox.getMonitorScheduler().addMonitor(metricsMonitor);
    }
    final PartitionsSpec partitionsSpec = partitionAnalysis.getPartitionsSpec();
    final IndexTuningConfig tuningConfig = ingestionSchema.getTuningConfig();
    final long pushTimeout = tuningConfig.getPushTimeout();
    final SegmentAllocatorForBatch segmentAllocator;
    final SequenceNameFunction sequenceNameFunction;
    switch(partitionsSpec.getType()) {
        case HASH:
        case RANGE:
            final SegmentAllocatorForBatch localSegmentAllocator = SegmentAllocators.forNonLinearPartitioning(toolbox, getDataSource(), baseSequenceName, dataSchema.getGranularitySpec(), null, (CompletePartitionAnalysis) partitionAnalysis);
            sequenceNameFunction = localSegmentAllocator.getSequenceNameFunction();
            segmentAllocator = localSegmentAllocator;
            break;
        case LINEAR:
            segmentAllocator = SegmentAllocators.forLinearPartitioning(toolbox, baseSequenceName, null, dataSchema, getTaskLockHelper(), ingestionSchema.getIOConfig().isAppendToExisting(), partitionAnalysis.getPartitionsSpec(), null);
            sequenceNameFunction = segmentAllocator.getSequenceNameFunction();
            break;
        default:
            throw new UOE("[%s] secondary partition type is not supported", partitionsSpec.getType());
    }
    Set<DataSegment> segmentsFoundForDrop = null;
    if (ingestionSchema.getIOConfig().isDropExisting()) {
        segmentsFoundForDrop = getUsedSegmentsWithinInterval(toolbox, getDataSource(), ingestionSchema.getDataSchema().getGranularitySpec().inputIntervals());
    }
    final TransactionalSegmentPublisher publisher = (segmentsToBeOverwritten, segmentsToDrop, segmentsToPublish, commitMetadata) -> toolbox.getTaskActionClient().submit(SegmentTransactionalInsertAction.overwriteAction(segmentsToBeOverwritten, segmentsToDrop, segmentsToPublish));
    String effectiveId = getContextValue(CompactionTask.CTX_KEY_APPENDERATOR_TRACKING_TASK_ID, null);
    if (effectiveId == null) {
        effectiveId = getId();
    }
    final Appenderator appenderator = BatchAppenderators.newAppenderator(effectiveId, toolbox.getAppenderatorsManager(), buildSegmentsFireDepartmentMetrics, toolbox, dataSchema, tuningConfig, buildSegmentsMeters, buildSegmentsParseExceptionHandler, isUseMaxMemoryEstimates());
    boolean exceptionOccurred = false;
    try (final BatchAppenderatorDriver driver = BatchAppenderators.newDriver(appenderator, toolbox, segmentAllocator)) {
        driver.startJob();
        InputSourceProcessor.process(dataSchema, driver, partitionsSpec, inputSource, inputSource.needsFormat() ? getInputFormat(ingestionSchema) : null, tmpDir, sequenceNameFunction, new DefaultIndexTaskInputRowIteratorBuilder(), buildSegmentsMeters, buildSegmentsParseExceptionHandler, pushTimeout);
        // If we use timeChunk lock, then we don't have to specify what segments will be overwritten because
        // it will just overwrite all segments overlapped with the new segments.
        final Set<DataSegment> inputSegments = getTaskLockHelper().isUseSegmentLock() ? getTaskLockHelper().getLockedExistingSegments() : null;
        final boolean storeCompactionState = getContextValue(Tasks.STORE_COMPACTION_STATE_KEY, Tasks.DEFAULT_STORE_COMPACTION_STATE);
        final Function<Set<DataSegment>, Set<DataSegment>> annotateFunction = compactionStateAnnotateFunction(storeCompactionState, toolbox, ingestionSchema);
        // Probably we can publish atomicUpdateGroup along with segments.
        final SegmentsAndCommitMetadata published = awaitPublish(driver.publishAll(inputSegments, segmentsFoundForDrop, publisher, annotateFunction), pushTimeout);
        appenderator.close();
        // for awaitSegmentAvailabilityTimeoutMillis
        if (tuningConfig.getAwaitSegmentAvailabilityTimeoutMillis() > 0 && published != null) {
            ingestionState = IngestionState.SEGMENT_AVAILABILITY_WAIT;
            ArrayList<DataSegment> segmentsToWaitFor = new ArrayList<>(published.getSegments());
            waitForSegmentAvailability(toolbox, segmentsToWaitFor, tuningConfig.getAwaitSegmentAvailabilityTimeoutMillis());
        }
        ingestionState = IngestionState.COMPLETED;
        if (published == null) {
            log.error("Failed to publish segments, aborting!");
            errorMsg = "Failed to publish segments.";
            toolbox.getTaskReportFileWriter().write(getId(), getTaskCompletionReports());
            return TaskStatus.failure(getId(), errorMsg);
        } else {
            log.info("Processed[%,d] events, unparseable[%,d], thrownAway[%,d].", buildSegmentsMeters.getProcessed(), buildSegmentsMeters.getUnparseable(), buildSegmentsMeters.getThrownAway());
            log.info("Published [%s] segments", published.getSegments().size());
            log.debugSegments(published.getSegments(), "Published segments");
            toolbox.getTaskReportFileWriter().write(getId(), getTaskCompletionReports());
            return TaskStatus.success(getId());
        }
    } catch (TimeoutException | ExecutionException e) {
        exceptionOccurred = true;
        throw new RuntimeException(e);
    } catch (Exception e) {
        exceptionOccurred = true;
        throw e;
    } finally {
        if (exceptionOccurred) {
            appenderator.closeNow();
        } else {
            appenderator.close();
        }
    }
}