Examples with DynamicPartitionsSpec org.apache.druid.indexer.partitions.DynamicPartitionsSpec used on opensource projects

Example 36 with DynamicPartitionsSpec

use of org.apache.druid.indexer.partitions.DynamicPartitionsSpec in project druid by druid-io.

the class SinglePhaseSubTask method generateAndPushSegments.

/**
 * 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 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 true if generated segments are successfully published, otherwise false
 */
private Set<DataSegment> generateAndPushSegments(final TaskToolbox toolbox, final ParallelIndexSupervisorTaskClient taskClient, final InputSource inputSource, final File tmpDir) throws IOException, InterruptedException {
    final DataSchema dataSchema = ingestionSchema.getDataSchema();
    final GranularitySpec granularitySpec = dataSchema.getGranularitySpec();
    final FireDepartment fireDepartmentForMetrics = new FireDepartment(dataSchema, new RealtimeIOConfig(null, null), null);
    final FireDepartmentMetrics fireDepartmentMetrics = fireDepartmentForMetrics.getMetrics();
    toolbox.addMonitor(new RealtimeMetricsMonitor(Collections.singletonList(fireDepartmentForMetrics), Collections.singletonMap(DruidMetrics.TASK_ID, new String[] { getId() })));
    final ParallelIndexTuningConfig tuningConfig = ingestionSchema.getTuningConfig();
    final DynamicPartitionsSpec partitionsSpec = (DynamicPartitionsSpec) tuningConfig.getGivenOrDefaultPartitionsSpec();
    final long pushTimeout = tuningConfig.getPushTimeout();
    final boolean explicitIntervals = !granularitySpec.inputIntervals().isEmpty();
    final boolean useLineageBasedSegmentAllocation = getContextValue(SinglePhaseParallelIndexTaskRunner.CTX_USE_LINEAGE_BASED_SEGMENT_ALLOCATION_KEY, SinglePhaseParallelIndexTaskRunner.LEGACY_DEFAULT_USE_LINEAGE_BASED_SEGMENT_ALLOCATION);
    // subtaskSpecId is used as the sequenceName, so that retry tasks for the same spec
    // can allocate the same set of segments.
    final String sequenceName = useLineageBasedSegmentAllocation ? Preconditions.checkNotNull(subtaskSpecId, "subtaskSpecId") : getId();
    final SegmentAllocatorForBatch segmentAllocator = SegmentAllocators.forLinearPartitioning(toolbox, sequenceName, new SupervisorTaskAccess(getSupervisorTaskId(), taskClient), getIngestionSchema().getDataSchema(), getTaskLockHelper(), ingestionSchema.getIOConfig().isAppendToExisting(), partitionsSpec, useLineageBasedSegmentAllocation);
    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, rowIngestionMeters, parseExceptionHandler, useMaxMemoryEstimates);
    boolean exceptionOccurred = false;
    try (final BatchAppenderatorDriver driver = BatchAppenderators.newDriver(appenderator, toolbox, segmentAllocator);
        final CloseableIterator<InputRow> inputRowIterator = AbstractBatchIndexTask.inputSourceReader(tmpDir, dataSchema, inputSource, inputSource.needsFormat() ? ParallelIndexSupervisorTask.getInputFormat(ingestionSchema) : null, inputRow -> {
            if (inputRow == null) {
                return false;
            }
            if (explicitIntervals) {
                final Optional<Interval> optInterval = granularitySpec.bucketInterval(inputRow.getTimestamp());
                return optInterval.isPresent();
            }
            return true;
        }, rowIngestionMeters, parseExceptionHandler)) {
        driver.startJob();
        final Set<DataSegment> pushedSegments = new HashSet<>();
        while (inputRowIterator.hasNext()) {
            final InputRow inputRow = inputRowIterator.next();
            // Segments are created as needed, using a single sequence name. They may be allocated from the overlord
            // (in append mode) or may be created on our own authority (in overwrite mode).
            final AppenderatorDriverAddResult addResult = driver.add(inputRow, sequenceName);
            if (addResult.isOk()) {
                final boolean isPushRequired = addResult.isPushRequired(partitionsSpec.getMaxRowsPerSegment(), partitionsSpec.getMaxTotalRowsOr(DynamicPartitionsSpec.DEFAULT_MAX_TOTAL_ROWS));
                if (isPushRequired) {
                    // There can be some segments waiting for being published even though any rows won't be added to them.
                    // If those segments are not published here, the available space in appenderator will be kept to be small
                    // which makes the size of segments smaller.
                    final SegmentsAndCommitMetadata pushed = driver.pushAllAndClear(pushTimeout);
                    pushedSegments.addAll(pushed.getSegments());
                    LOG.info("Pushed [%s] segments", pushed.getSegments().size());
                    LOG.infoSegments(pushed.getSegments(), "Pushed segments");
                }
            } else {
                throw new ISE("Failed to add a row with timestamp[%s]", inputRow.getTimestamp());
            }
            fireDepartmentMetrics.incrementProcessed();
        }
        final SegmentsAndCommitMetadata pushed = driver.pushAllAndClear(pushTimeout);
        pushedSegments.addAll(pushed.getSegments());
        LOG.info("Pushed [%s] segments", pushed.getSegments().size());
        LOG.infoSegments(pushed.getSegments(), "Pushed segments");
        appenderator.close();
        return pushedSegments;
    } 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();
        }
    }
}

Example 37 with DynamicPartitionsSpec

use of org.apache.druid.indexer.partitions.DynamicPartitionsSpec in project druid by druid-io.

the class CompactionTuningConfigTest method testSerdeWithNullAwaitSegmentAvailabilityTimeoutMillis.

@Test
public void testSerdeWithNullAwaitSegmentAvailabilityTimeoutMillis() {
    final CompactionTask.CompactionTuningConfig tuningConfig = new CompactionTask.CompactionTuningConfig(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);
    Assert.assertEquals(0L, tuningConfig.getAwaitSegmentAvailabilityTimeoutMillis());
}

Example 38 with DynamicPartitionsSpec

use of org.apache.druid.indexer.partitions.DynamicPartitionsSpec in project druid by druid-io.

the class IndexTaskTest method testMultipleParseExceptionsFailure.

@Test
public void testMultipleParseExceptionsFailure() throws Exception {
    final File tmpDir = temporaryFolder.newFolder();
    final File tmpFile = File.createTempFile("druid", "index", tmpDir);
    try (BufferedWriter writer = Files.newWriter(tmpFile, StandardCharsets.UTF_8)) {
        writer.write("time,dim,dimLong,dimFloat,val\n");
        // unparseable
        writer.write("unparseable,a,2,3.0,1\n");
        // valid row
        writer.write("2014-01-01T00:00:10Z,a,2,3.0,1\n");
        // unparseable
        writer.write("9.0,a,2,3.0,1\n");
        // thrown away
        writer.write("3014-03-01T00:00:10Z,outsideofinterval,2,3.0,1\n");
        // unparseable
        writer.write("99999999999-01-01T00:00:10Z,b,2,3.0,1\n");
    }
    // Allow up to 3 parse exceptions, and save up to 2 parse exceptions
    final IndexTuningConfig tuningConfig = new IndexTuningConfig(null, null, null, null, null, null, null, null, null, null, new DynamicPartitionsSpec(2, null), INDEX_SPEC, null, null, false, false, null, null, null, true, 2, 5, null, null);
    final TimestampSpec timestampSpec = new TimestampSpec("time", "auto", null);
    final DimensionsSpec dimensionsSpec = new DimensionsSpec(Arrays.asList(new StringDimensionSchema("dim"), new LongDimensionSchema("dimLong"), new FloatDimensionSchema("dimFloat")));
    final List<String> columns = Arrays.asList("time", "dim", "dimLong", "dimFloat", "val");
    final IndexIngestionSpec ingestionSpec;
    List<String> expectedMessages;
    if (useInputFormatApi) {
        ingestionSpec = createIngestionSpec(jsonMapper, tmpDir, timestampSpec, dimensionsSpec, new CsvInputFormat(columns, null, null, true, 0), null, null, tuningConfig, false, false);
        expectedMessages = Arrays.asList(StringUtils.format("Timestamp[99999999999-01-01T00:00:10Z] is unparseable! Event: {time=99999999999-01-01T00:00:10Z, dim=b, dimLong=2, dimFloat=3.0, val=1} (Path: %s, Record: 3, Line: 6)", tmpFile.toURI()), StringUtils.format("Timestamp[9.0] is unparseable! Event: {time=9.0, dim=a, dimLong=2, dimFloat=3.0, val=1} (Path: %s, Record: 2, Line: 4)", tmpFile.toURI()), StringUtils.format("Timestamp[unparseable] is unparseable! Event: {time=unparseable, dim=a, dimLong=2, dimFloat=3.0, val=1} (Path: %s, Record: 1, Line: 2)", tmpFile.toURI()));
    } else {
        ingestionSpec = createIngestionSpec(jsonMapper, tmpDir, new CSVParseSpec(timestampSpec, dimensionsSpec, null, columns, true, 0), null, null, tuningConfig, false, false);
        expectedMessages = Arrays.asList("Timestamp[99999999999-01-01T00:00:10Z] is unparseable! Event: {time=99999999999-01-01T00:00:10Z, dim=b, dimLong=2, dimFloat=3.0, val=1}", "Timestamp[9.0] is unparseable! Event: {time=9.0, dim=a, dimLong=2, dimFloat=3.0, val=1}", "Timestamp[unparseable] is unparseable! Event: {time=unparseable, dim=a, dimLong=2, dimFloat=3.0, val=1}");
    }
    IndexTask indexTask = new IndexTask(null, null, ingestionSpec, null);
    TaskStatus status = runTask(indexTask).lhs;
    Assert.assertEquals(TaskState.FAILED, status.getStatusCode());
    checkTaskStatusErrorMsgForParseExceptionsExceeded(status);
    IngestionStatsAndErrorsTaskReportData reportData = getTaskReportData();
    Map<String, Object> expectedMetrics = ImmutableMap.of(RowIngestionMeters.DETERMINE_PARTITIONS, ImmutableMap.of(RowIngestionMeters.PROCESSED_WITH_ERROR, 0, RowIngestionMeters.PROCESSED, 0, RowIngestionMeters.UNPARSEABLE, 0, RowIngestionMeters.THROWN_AWAY, 0), RowIngestionMeters.BUILD_SEGMENTS, ImmutableMap.of(RowIngestionMeters.PROCESSED_WITH_ERROR, 0, RowIngestionMeters.PROCESSED, 1, RowIngestionMeters.UNPARSEABLE, 3, RowIngestionMeters.THROWN_AWAY, useInputFormatApi ? 1 : 2));
    Assert.assertEquals(expectedMetrics, reportData.getRowStats());
    List<LinkedHashMap> parseExceptionReports = (List<LinkedHashMap>) reportData.getUnparseableEvents().get(RowIngestionMeters.BUILD_SEGMENTS);
    List<String> actualMessages = parseExceptionReports.stream().map((r) -> {
        return ((List<String>) r.get("details")).get(0);
    }).collect(Collectors.toList());
    Assert.assertEquals(expectedMessages, actualMessages);
    List<String> expectedInputs = Arrays.asList("{time=99999999999-01-01T00:00:10Z, dim=b, dimLong=2, dimFloat=3.0, val=1}", "{time=9.0, dim=a, dimLong=2, dimFloat=3.0, val=1}", "{time=unparseable, dim=a, dimLong=2, dimFloat=3.0, val=1}");
    List<String> actualInputs = parseExceptionReports.stream().map((r) -> {
        return (String) r.get("input");
    }).collect(Collectors.toList());
    Assert.assertEquals(expectedInputs, actualInputs);
}

Example 39 with DynamicPartitionsSpec

use of org.apache.druid.indexer.partitions.DynamicPartitionsSpec in project druid by druid-io.

the class CompactionTaskRunTest method getDefaultCompactionState.

public static CompactionState getDefaultCompactionState(Granularity segmentGranularity, Granularity queryGranularity, List<Interval> intervals) throws JsonProcessingException {
    ObjectMapper mapper = new DefaultObjectMapper();
    // Expected compaction state to exist after compaction as we store compaction state by default
    Map<String, String> expectedLongSumMetric = new HashMap<>();
    expectedLongSumMetric.put("type", "longSum");
    expectedLongSumMetric.put("name", "val");
    expectedLongSumMetric.put("fieldName", "val");
    expectedLongSumMetric.put("expression", null);
    return new CompactionState(new DynamicPartitionsSpec(5000000, Long.MAX_VALUE), new DimensionsSpec(DimensionsSpec.getDefaultSchemas(ImmutableList.of("ts", "dim"))), ImmutableList.of(expectedLongSumMetric), null, mapper.readValue(mapper.writeValueAsString(new IndexSpec()), Map.class), mapper.readValue(mapper.writeValueAsString(new UniformGranularitySpec(segmentGranularity, queryGranularity, true, intervals)), Map.class));
}

Example 40 with DynamicPartitionsSpec

use of org.apache.druid.indexer.partitions.DynamicPartitionsSpec in project druid by druid-io.

the class CompactionTaskRunTest method testCompactionWithNewMetricInMetricsSpec.

@Test
public void testCompactionWithNewMetricInMetricsSpec() throws Exception {
    runIndexTask();
    final Builder builder = new Builder(DATA_SOURCE, segmentCacheManagerFactory, RETRY_POLICY_FACTORY);
    // day segmentGranularity
    final CompactionTask compactionTask = builder.interval(Intervals.of("2014-01-01/2014-01-02")).granularitySpec(new ClientCompactionTaskGranularitySpec(Granularities.DAY, null, null)).metricsSpec(new AggregatorFactory[] { new CountAggregatorFactory("cnt"), new LongSumAggregatorFactory("val", "val") }).build();
    Pair<TaskStatus, List<DataSegment>> resultPair = runTask(compactionTask);
    Assert.assertTrue(resultPair.lhs.isSuccess());
    List<DataSegment> segments = resultPair.rhs;
    Assert.assertEquals(1, segments.size());
    Assert.assertEquals(Intervals.of("2014-01-01/2014-01-02"), segments.get(0).getInterval());
    Assert.assertEquals(new NumberedShardSpec(0, 1), segments.get(0).getShardSpec());
    ObjectMapper mapper = new DefaultObjectMapper();
    Map<String, String> expectedCountMetric = new HashMap<>();
    expectedCountMetric.put("type", "count");
    expectedCountMetric.put("name", "cnt");
    Map<String, String> expectedLongSumMetric = new HashMap<>();
    expectedLongSumMetric.put("type", "longSum");
    expectedLongSumMetric.put("name", "val");
    expectedLongSumMetric.put("fieldName", "val");
    expectedLongSumMetric.put("expression", null);
    CompactionState expectedCompactionState = new CompactionState(new DynamicPartitionsSpec(5000000, Long.MAX_VALUE), new DimensionsSpec(DimensionsSpec.getDefaultSchemas(ImmutableList.of("ts", "dim"))), ImmutableList.of(expectedCountMetric, expectedLongSumMetric), getObjectMapper().readValue(getObjectMapper().writeValueAsString(compactionTask.getTransformSpec()), Map.class), mapper.readValue(mapper.writeValueAsString(new IndexSpec()), Map.class), mapper.readValue(mapper.writeValueAsString(new UniformGranularitySpec(Granularities.DAY, Granularities.MINUTE, true, ImmutableList.of(Intervals.of("2014-01-01T00:00:00/2014-01-01T03:00:00")))), Map.class));
    Assert.assertEquals(expectedCompactionState, segments.get(0).getLastCompactionState());
}