Examples with TransformBlock com.linkedin.pinot.core.operator.blocks.TransformBlock used on opensource projects

Example 1 with TransformBlock

use of com.linkedin.pinot.core.operator.blocks.TransformBlock in project pinot by linkedin.

the class AggregationOperator method getNextBlock.

@Override
public Block getNextBlock() {
    int numDocsScanned = 0;
    // Perform aggregation on all the blocks.
    AggregationExecutor aggregationExecutor = new DefaultAggregationExecutor(_aggregationFunctionContexts);
    aggregationExecutor.init();
    TransformBlock transformBlock;
    while ((transformBlock = (TransformBlock) _transformOperator.nextBlock()) != null) {
        numDocsScanned += transformBlock.getNumDocs();
        aggregationExecutor.aggregate(transformBlock);
    }
    aggregationExecutor.finish();
    // Create execution statistics.
    long numEntriesScannedInFilter = _transformOperator.getExecutionStatistics().getNumEntriesScannedInFilter();
    long numEntriesScannedPostFilter = numDocsScanned * _transformOperator.getNumProjectionColumns();
    _executionStatistics = new ExecutionStatistics(numDocsScanned, numEntriesScannedInFilter, numEntriesScannedPostFilter, _numTotalRawDocs);
    // Build intermediate result block based on aggregation result from the executor.
    return new IntermediateResultsBlock(_aggregationFunctionContexts, aggregationExecutor.getResult(), false);
}

Example 2 with TransformBlock

use of com.linkedin.pinot.core.operator.blocks.TransformBlock in project pinot by linkedin.

the class NoDictionaryGroupKeyGeneratorTest method testGroupKeyGenerator.

private void testGroupKeyGenerator(String[] groupByColumns, FieldSpec.DataType[] dataTypes) throws Exception {
    // Build the projection operator.
    MatchEntireSegmentOperator matchEntireSegmentOperator = new MatchEntireSegmentOperator(NUM_ROWS);
    BReusableFilteredDocIdSetOperator docIdSetOperator = new BReusableFilteredDocIdSetOperator(matchEntireSegmentOperator, NUM_ROWS, 10000);
    MProjectionOperator projectionOperator = new MProjectionOperator(_dataSourceMap, docIdSetOperator);
    TransformExpressionOperator transformOperator = new TransformExpressionOperator(projectionOperator, new ArrayList<TransformExpressionTree>());
    // Iterator over all projection blocks and generate group keys.
    TransformBlock transformBlock;
    int[] docIdToGroupKeys = new int[DocIdSetPlanNode.MAX_DOC_PER_CALL];
    GroupKeyGenerator groupKeyGenerator = null;
    while ((transformBlock = (TransformBlock) transformOperator.nextBlock()) != null) {
        if (groupKeyGenerator == null) {
            // Build the group key generator.
            groupKeyGenerator = (groupByColumns.length == 1) ? new NoDictionarySingleColumnGroupKeyGenerator(groupByColumns[0], dataTypes[0]) : new NoDictionaryMultiColumnGroupKeyGenerator(transformBlock, groupByColumns);
        }
        groupKeyGenerator.generateKeysForBlock(transformBlock, docIdToGroupKeys);
    }
    // Assert total number of group keys is as expected
    Assert.assertTrue(groupKeyGenerator != null);
    Set<String> expectedGroupKeys = getExpectedGroupKeys(_recordReader, groupByColumns);
    Assert.assertEquals(groupKeyGenerator.getCurrentGroupKeyUpperBound(), expectedGroupKeys.size(), "Number of group keys mis-match.");
    // Assert all group key values are as expected
    Iterator<GroupKeyGenerator.GroupKey> uniqueGroupKeys = groupKeyGenerator.getUniqueGroupKeys();
    while (uniqueGroupKeys.hasNext()) {
        GroupKeyGenerator.GroupKey groupKey = uniqueGroupKeys.next();
        String actual = groupKey.getStringKey();
        Assert.assertTrue(expectedGroupKeys.contains(actual), "Unexpected group key: " + actual);
    }
}

Example 3 with TransformBlock

use of com.linkedin.pinot.core.operator.blocks.TransformBlock in project pinot by linkedin.

the class DefaultGroupKeyGeneratorTest method setup.

@BeforeClass
private void setup() throws Exception {
    GenericRow[] segmentData = new GenericRow[NUM_ROWS];
    int value = _random.nextInt(MAX_STEP_LENGTH);
    // Generate random values for the segment.
    for (int i = 0; i < UNIQUE_ROWS; i++) {
        Map<String, Object> map = new HashMap<>();
        for (String singleValueColumn : SINGLE_VALUE_COLUMNS) {
            map.put(singleValueColumn, value);
            value += 1 + _random.nextInt(MAX_STEP_LENGTH);
        }
        for (String multiValueColumn : MULTI_VALUE_COLUMNS) {
            int numMultiValues = 1 + _random.nextInt(MAX_NUM_MULTI_VALUES);
            Integer[] values = new Integer[numMultiValues];
            for (int k = 0; k < numMultiValues; k++) {
                values[k] = value;
                value += 1 + _random.nextInt(MAX_STEP_LENGTH);
            }
            map.put(multiValueColumn, values);
        }
        GenericRow genericRow = new GenericRow();
        genericRow.init(map);
        segmentData[i] = genericRow;
    }
    for (int i = UNIQUE_ROWS; i < NUM_ROWS; i += UNIQUE_ROWS) {
        System.arraycopy(segmentData, 0, segmentData, i, UNIQUE_ROWS);
    }
    // Create an index segment with the random values.
    Schema schema = new Schema();
    for (String singleValueColumn : SINGLE_VALUE_COLUMNS) {
        DimensionFieldSpec dimensionFieldSpec = new DimensionFieldSpec(singleValueColumn, FieldSpec.DataType.INT, true);
        schema.addField(dimensionFieldSpec);
    }
    for (String multiValueColumn : MULTI_VALUE_COLUMNS) {
        DimensionFieldSpec dimensionFieldSpec = new DimensionFieldSpec(multiValueColumn, FieldSpec.DataType.INT, false);
        schema.addField(dimensionFieldSpec);
    }
    SegmentGeneratorConfig config = new SegmentGeneratorConfig(schema);
    FileUtils.deleteQuietly(new File(INDEX_DIR_PATH));
    config.setOutDir(INDEX_DIR_PATH);
    config.setSegmentName(SEGMENT_NAME);
    SegmentIndexCreationDriverImpl driver = new SegmentIndexCreationDriverImpl();
    driver.init(config, new TestDataRecordReader(schema, segmentData));
    driver.build();
    IndexSegment indexSegment = Loaders.IndexSegment.load(new File(INDEX_DIR_PATH, SEGMENT_NAME), ReadMode.heap);
    // Get a data fetcher for the index segment.
    Map<String, BaseOperator> dataSourceMap = new HashMap<>();
    Map<String, Block> blockMap = new HashMap<>();
    for (String column : indexSegment.getColumnNames()) {
        DataSource dataSource = indexSegment.getDataSource(column);
        dataSourceMap.put(column, dataSource);
        blockMap.put(column, dataSource.getNextBlock());
    }
    // Generate a random test doc id set.
    int num1 = _random.nextInt(50);
    int num2 = num1 + 1 + _random.nextInt(50);
    for (int i = 0; i < 20; i += 2) {
        _testDocIdSet[i] = num1 + 50 * i;
        _testDocIdSet[i + 1] = num2 + 50 * i;
    }
    DataFetcher dataFetcher = new DataFetcher(dataSourceMap);
    DocIdSetBlock docIdSetBlock = new DocIdSetBlock(_testDocIdSet, _testDocIdSet.length);
    ProjectionBlock projectionBlock = new ProjectionBlock(blockMap, new DataBlockCache(dataFetcher), docIdSetBlock);
    _transformBlock = new TransformBlock(projectionBlock, new HashMap<String, BlockValSet>());
}

Example 4 with TransformBlock

use of com.linkedin.pinot.core.operator.blocks.TransformBlock in project pinot by linkedin.

the class TransformExpressionOperator method getNextBlock.

@Override
public Block getNextBlock() {
    ProjectionBlock projectionBlock = _projectionOperator.getNextBlock();
    if (projectionBlock == null) {
        return null;
    }
    Map<String, BlockValSet> expressionResults = (_expressionEvaluator != null) ? _expressionEvaluator.evaluate(projectionBlock) : null;
    return new TransformBlock(projectionBlock, expressionResults);
}

Example 5 with TransformBlock

use of com.linkedin.pinot.core.operator.blocks.TransformBlock in project pinot by linkedin.

the class DefaultAggregationExecutorTest method testAggregation.

/**
   * Runs 'sum', 'min' & 'max' aggregation functions on the DefaultAggregationExecutor.
   * Asserts that the aggregation results returned by the executor are as expected.
   */
@Test
void testAggregation() {
    Map<String, BaseOperator> dataSourceMap = new HashMap<>();
    for (String column : _indexSegment.getColumnNames()) {
        dataSourceMap.put(column, _indexSegment.getDataSource(column));
    }
    int totalRawDocs = _indexSegment.getSegmentMetadata().getTotalRawDocs();
    MatchEntireSegmentOperator matchEntireSegmentOperator = new MatchEntireSegmentOperator(totalRawDocs);
    BReusableFilteredDocIdSetOperator docIdSetOperator = new BReusableFilteredDocIdSetOperator(matchEntireSegmentOperator, totalRawDocs, 10000);
    MProjectionOperator projectionOperator = new MProjectionOperator(dataSourceMap, docIdSetOperator);
    TransformExpressionOperator transformOperator = new TransformExpressionOperator(projectionOperator, Collections.<TransformExpressionTree>emptyList());
    TransformBlock transformBlock = (TransformBlock) transformOperator.nextBlock();
    int numAggFuncs = _aggregationInfoList.size();
    AggregationFunctionContext[] aggrFuncContextArray = new AggregationFunctionContext[numAggFuncs];
    AggregationFunctionInitializer aggFuncInitializer = new AggregationFunctionInitializer(_indexSegment.getSegmentMetadata());
    for (int i = 0; i < numAggFuncs; i++) {
        AggregationInfo aggregationInfo = _aggregationInfoList.get(i);
        aggrFuncContextArray[i] = AggregationFunctionContext.instantiate(aggregationInfo);
        aggrFuncContextArray[i].getAggregationFunction().accept(aggFuncInitializer);
    }
    AggregationExecutor aggregationExecutor = new DefaultAggregationExecutor(aggrFuncContextArray);
    aggregationExecutor.init();
    aggregationExecutor.aggregate(transformBlock);
    aggregationExecutor.finish();
    List<Object> result = aggregationExecutor.getResult();
    for (int i = 0; i < result.size(); i++) {
        double actual = (double) result.get(i);
        double expected = computeAggregation(AGGREGATION_FUNCTIONS[i], _inputData[i]);
        Assert.assertEquals(actual, expected, "Aggregation mis-match for function " + AGGREGATION_FUNCTIONS[i] + ", Expected: " + expected + " Actual: " + actual);
    }
}