Examples with AggregatorFactory io.druid.query.aggregation.AggregatorFactory used on opensource projects

Example 56 with AggregatorFactory

use of io.druid.query.aggregation.AggregatorFactory in project druid by druid-io.

the class GroupByRules method translateAggregateCall.

/**
   * Translate an AggregateCall to Druid equivalents.
   *
   * @return translated aggregation, or null if translation failed.
   */
private static Aggregation translateAggregateCall(final PlannerContext plannerContext, final RowSignature sourceRowSignature, final Project project, final AggregateCall call, final DruidOperatorTable operatorTable, final List<Aggregation> existingAggregations, final int aggNumber, final boolean approximateCountDistinct) {
    final List<DimFilter> filters = Lists.newArrayList();
    final List<String> rowOrder = sourceRowSignature.getRowOrder();
    final String name = aggOutputName(aggNumber);
    final SqlKind kind = call.getAggregation().getKind();
    final SqlTypeName outputType = call.getType().getSqlTypeName();
    if (call.filterArg >= 0) {
        // AGG(xxx) FILTER(WHERE yyy)
        if (project == null) {
            // We need some kind of projection to support filtered aggregations.
            return null;
        }
        final RexNode expression = project.getChildExps().get(call.filterArg);
        final DimFilter filter = Expressions.toFilter(operatorTable, plannerContext, sourceRowSignature, expression);
        if (filter == null) {
            return null;
        }
        filters.add(filter);
    }
    if (kind == SqlKind.COUNT && call.getArgList().isEmpty()) {
        // COUNT(*)
        return Aggregation.create(new CountAggregatorFactory(name)).filter(makeFilter(filters, sourceRowSignature));
    } else if (kind == SqlKind.COUNT && call.isDistinct()) {
        // COUNT(DISTINCT x)
        return approximateCountDistinct ? APPROX_COUNT_DISTINCT.toDruidAggregation(name, sourceRowSignature, operatorTable, plannerContext, existingAggregations, project, call, makeFilter(filters, sourceRowSignature)) : null;
    } else if (kind == SqlKind.COUNT || kind == SqlKind.SUM || kind == SqlKind.SUM0 || kind == SqlKind.MIN || kind == SqlKind.MAX || kind == SqlKind.AVG) {
        // Built-in agg, not distinct, not COUNT(*)
        boolean forceCount = false;
        final FieldOrExpression input;
        final int inputField = Iterables.getOnlyElement(call.getArgList());
        final RexNode rexNode = Expressions.fromFieldAccess(sourceRowSignature, project, inputField);
        final FieldOrExpression foe = FieldOrExpression.fromRexNode(operatorTable, plannerContext, rowOrder, rexNode);
        if (foe != null) {
            input = foe;
        } else if (rexNode.getKind() == SqlKind.CASE && ((RexCall) rexNode).getOperands().size() == 3) {
            // Possibly a CASE-style filtered aggregation. Styles supported:
            // A: SUM(CASE WHEN x = 'foo' THEN cnt END) => operands (x = 'foo', cnt, null)
            // B: SUM(CASE WHEN x = 'foo' THEN 1 ELSE 0 END) => operands (x = 'foo', 1, 0)
            // C: COUNT(CASE WHEN x = 'foo' THEN 'dummy' END) => operands (x = 'foo', 'dummy', null)
            // If the null and non-null args are switched, "flip" is set, which negates the filter.
            final RexCall caseCall = (RexCall) rexNode;
            final boolean flip = RexLiteral.isNullLiteral(caseCall.getOperands().get(1)) && !RexLiteral.isNullLiteral(caseCall.getOperands().get(2));
            final RexNode arg1 = caseCall.getOperands().get(flip ? 2 : 1);
            final RexNode arg2 = caseCall.getOperands().get(flip ? 1 : 2);
            // Operand 1: Filter
            final DimFilter filter = Expressions.toFilter(operatorTable, plannerContext, sourceRowSignature, caseCall.getOperands().get(0));
            if (filter == null) {
                return null;
            } else {
                filters.add(flip ? new NotDimFilter(filter) : filter);
            }
            if (call.getAggregation().getKind() == SqlKind.COUNT && arg1 instanceof RexLiteral && !RexLiteral.isNullLiteral(arg1) && RexLiteral.isNullLiteral(arg2)) {
                // Case C
                forceCount = true;
                input = null;
            } else if (call.getAggregation().getKind() == SqlKind.SUM && arg1 instanceof RexLiteral && ((Number) RexLiteral.value(arg1)).intValue() == 1 && arg2 instanceof RexLiteral && ((Number) RexLiteral.value(arg2)).intValue() == 0) {
                // Case B
                forceCount = true;
                input = null;
            } else if (RexLiteral.isNullLiteral(arg2)) {
                // Maybe case A
                input = FieldOrExpression.fromRexNode(operatorTable, plannerContext, rowOrder, arg1);
                if (input == null) {
                    return null;
                }
            } else {
                // Can't translate CASE into a filter.
                return null;
            }
        } else {
            // Can't translate operand.
            return null;
        }
        if (!forceCount) {
            Preconditions.checkNotNull(input, "WTF?! input was null for non-COUNT aggregation");
        }
        if (forceCount || kind == SqlKind.COUNT) {
            // COUNT(x)
            return Aggregation.create(new CountAggregatorFactory(name)).filter(makeFilter(filters, sourceRowSignature));
        } else {
            // Built-in aggregator that is not COUNT.
            final Aggregation retVal;
            final String fieldName = input.getFieldName();
            final String expression = input.getExpression();
            final boolean isLong = SqlTypeName.INT_TYPES.contains(outputType) || SqlTypeName.TIMESTAMP == outputType || SqlTypeName.DATE == outputType;
            if (kind == SqlKind.SUM || kind == SqlKind.SUM0) {
                retVal = isLong ? Aggregation.create(new LongSumAggregatorFactory(name, fieldName, expression)) : Aggregation.create(new DoubleSumAggregatorFactory(name, fieldName, expression));
            } else if (kind == SqlKind.MIN) {
                retVal = isLong ? Aggregation.create(new LongMinAggregatorFactory(name, fieldName, expression)) : Aggregation.create(new DoubleMinAggregatorFactory(name, fieldName, expression));
            } else if (kind == SqlKind.MAX) {
                retVal = isLong ? Aggregation.create(new LongMaxAggregatorFactory(name, fieldName, expression)) : Aggregation.create(new DoubleMaxAggregatorFactory(name, fieldName, expression));
            } else if (kind == SqlKind.AVG) {
                final String sumName = aggInternalName(aggNumber, "sum");
                final String countName = aggInternalName(aggNumber, "count");
                final AggregatorFactory sum = isLong ? new LongSumAggregatorFactory(sumName, fieldName, expression) : new DoubleSumAggregatorFactory(sumName, fieldName, expression);
                final AggregatorFactory count = new CountAggregatorFactory(countName);
                retVal = Aggregation.create(ImmutableList.of(sum, count), new ArithmeticPostAggregator(name, "quotient", ImmutableList.<PostAggregator>of(new FieldAccessPostAggregator(null, sumName), new FieldAccessPostAggregator(null, countName))));
            } else {
                // Not reached.
                throw new ISE("WTF?! Kind[%s] got into the built-in aggregator path somehow?!", kind);
            }
            return retVal.filter(makeFilter(filters, sourceRowSignature));
        }
    } else {
        // Not a built-in aggregator, check operator table.
        final SqlAggregator sqlAggregator = operatorTable.lookupAggregator(call.getAggregation().getName());
        return sqlAggregator != null ? sqlAggregator.toDruidAggregation(name, sourceRowSignature, operatorTable, plannerContext, existingAggregations, project, call, makeFilter(filters, sourceRowSignature)) : null;
    }
}

Example 57 with AggregatorFactory

use of io.druid.query.aggregation.AggregatorFactory in project druid by druid-io.

the class ApproxCountDistinctSqlAggregator method toDruidAggregation.

@Override
public Aggregation toDruidAggregation(final String name, final RowSignature rowSignature, final DruidOperatorTable operatorTable, final PlannerContext plannerContext, final List<Aggregation> existingAggregations, final Project project, final AggregateCall aggregateCall, final DimFilter filter) {
    final RexNode rexNode = Expressions.fromFieldAccess(rowSignature, project, Iterables.getOnlyElement(aggregateCall.getArgList()));
    final RowExtraction rex = Expressions.toRowExtraction(operatorTable, plannerContext, rowSignature.getRowOrder(), rexNode);
    if (rex == null) {
        return null;
    }
    final AggregatorFactory aggregatorFactory;
    if (rowSignature.getColumnType(rex.getColumn()) == ValueType.COMPLEX) {
        aggregatorFactory = new HyperUniquesAggregatorFactory(name, rex.getColumn());
    } else {
        final SqlTypeName sqlTypeName = rexNode.getType().getSqlTypeName();
        final ValueType outputType = Calcites.getValueTypeForSqlTypeName(sqlTypeName);
        if (outputType == null) {
            throw new ISE("Cannot translate sqlTypeName[%s] to Druid type for field[%s]", sqlTypeName, name);
        }
        final DimensionSpec dimensionSpec = rex.toDimensionSpec(rowSignature, null, ValueType.STRING);
        if (dimensionSpec == null) {
            return null;
        }
        aggregatorFactory = new CardinalityAggregatorFactory(name, ImmutableList.of(dimensionSpec), false);
    }
    return Aggregation.createFinalizable(ImmutableList.<AggregatorFactory>of(aggregatorFactory), null, new PostAggregatorFactory() {

        @Override
        public PostAggregator factorize(String outputName) {
            return new HyperUniqueFinalizingPostAggregator(outputName, name);
        }
    }).filter(filter);
}

Example 58 with AggregatorFactory

use of io.druid.query.aggregation.AggregatorFactory in project druid by druid-io.

the class DruidParquetReadSupport method getPartialReadSchema.

private MessageType getPartialReadSchema(InitContext context) {
    MessageType fullSchema = context.getFileSchema();
    String name = fullSchema.getName();
    HadoopDruidIndexerConfig config = HadoopDruidIndexerConfig.fromConfiguration(context.getConfiguration());
    String tsField = config.getParser().getParseSpec().getTimestampSpec().getTimestampColumn();
    List<DimensionSchema> dimensionSchema = config.getParser().getParseSpec().getDimensionsSpec().getDimensions();
    Set<String> dimensions = Sets.newHashSet();
    for (DimensionSchema dim : dimensionSchema) {
        dimensions.add(dim.getName());
    }
    Set<String> metricsFields = Sets.newHashSet();
    for (AggregatorFactory agg : config.getSchema().getDataSchema().getAggregators()) {
        metricsFields.addAll(agg.requiredFields());
    }
    List<Type> partialFields = Lists.newArrayList();
    for (Type type : fullSchema.getFields()) {
        if (tsField.equals(type.getName()) || metricsFields.contains(type.getName()) || dimensions.size() > 0 && dimensions.contains(type.getName()) || dimensions.size() == 0) {
            partialFields.add(type);
        }
    }
    return new MessageType(name, partialFields);
}

Example 59 with AggregatorFactory

use of io.druid.query.aggregation.AggregatorFactory in project druid by druid-io.

the class IncrementalIndexStorageAdapterTest method testSingleValueTopN.

@Test
public void testSingleValueTopN() throws IOException {
    IncrementalIndex index = indexCreator.createIndex();
    DateTime t = DateTime.now();
    index.add(new MapBasedInputRow(t.minus(1).getMillis(), Lists.newArrayList("sally"), ImmutableMap.<String, Object>of("sally", "bo")));
    TopNQueryEngine engine = new TopNQueryEngine(new StupidPool<ByteBuffer>("TopNQueryEngine-bufferPool", new Supplier<ByteBuffer>() {

        @Override
        public ByteBuffer get() {
            return ByteBuffer.allocate(50000);
        }
    }));
    final Iterable<Result<TopNResultValue>> results = Sequences.toList(engine.query(new TopNQueryBuilder().dataSource("test").granularity(Granularities.ALL).intervals(Lists.newArrayList(new Interval(0, new DateTime().getMillis()))).dimension("sally").metric("cnt").threshold(10).aggregators(Lists.<AggregatorFactory>newArrayList(new LongSumAggregatorFactory("cnt", "cnt"))).build(), new IncrementalIndexStorageAdapter(index)), Lists.<Result<TopNResultValue>>newLinkedList());
    Assert.assertEquals(1, Iterables.size(results));
    Assert.assertEquals(1, results.iterator().next().getValue().getValue().size());
}

Example 60 with AggregatorFactory

use of io.druid.query.aggregation.AggregatorFactory in project druid by druid-io.

the class SpatialFilterTest method testSpatialQueryWithOtherSpatialDim.

@Test
public void testSpatialQueryWithOtherSpatialDim() {
    TimeseriesQuery query = Druids.newTimeseriesQueryBuilder().dataSource("test").granularity(Granularities.ALL).intervals(Arrays.asList(new Interval("2013-01-01/2013-01-07"))).filters(new SpatialDimFilter("spatialIsRad", new RadiusBound(new float[] { 0.0f, 0.0f }, 5))).aggregators(Arrays.<AggregatorFactory>asList(new CountAggregatorFactory("rows"), new LongSumAggregatorFactory("val", "val"))).build();
    List<Result<TimeseriesResultValue>> expectedResults = Arrays.asList(new Result<TimeseriesResultValue>(new DateTime("2013-01-01T00:00:00.000Z"), new TimeseriesResultValue(ImmutableMap.<String, Object>builder().put("rows", 1L).put("val", 13L).build())));
    try {
        TimeseriesQueryRunnerFactory factory = new TimeseriesQueryRunnerFactory(new TimeseriesQueryQueryToolChest(QueryRunnerTestHelper.NoopIntervalChunkingQueryRunnerDecorator()), new TimeseriesQueryEngine(), QueryRunnerTestHelper.NOOP_QUERYWATCHER);
        QueryRunner runner = new FinalizeResultsQueryRunner(factory.createRunner(segment), factory.getToolchest());
        TestHelper.assertExpectedResults(expectedResults, runner.run(query, Maps.newHashMap()));
    } catch (Exception e) {
        throw Throwables.propagate(e);
    }
}