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

Example 46 with PostAggregator

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

the class TopNQueryRunnerTest method testTopNOverHyperUniqueFinalizingPostAggregator.

@Test
public void testTopNOverHyperUniqueFinalizingPostAggregator() {
    TopNQuery query = new TopNQueryBuilder().dataSource(QueryRunnerTestHelper.dataSource).granularity(QueryRunnerTestHelper.allGran).dimension(QueryRunnerTestHelper.marketDimension).metric(QueryRunnerTestHelper.hyperUniqueFinalizingPostAggMetric).threshold(3).intervals(QueryRunnerTestHelper.fullOnInterval).aggregators(Arrays.<AggregatorFactory>asList(QueryRunnerTestHelper.qualityUniques)).postAggregators(Arrays.<PostAggregator>asList(new HyperUniqueFinalizingPostAggregator(QueryRunnerTestHelper.hyperUniqueFinalizingPostAggMetric, QueryRunnerTestHelper.uniqueMetric))).build();
    List<Result<TopNResultValue>> expectedResults = Arrays.asList(new Result<>(new DateTime("2011-01-12T00:00:00.000Z"), new TopNResultValue(Arrays.<Map<String, Object>>asList(ImmutableMap.<String, Object>builder().put("market", "spot").put(QueryRunnerTestHelper.uniqueMetric, QueryRunnerTestHelper.UNIQUES_9).put(QueryRunnerTestHelper.hyperUniqueFinalizingPostAggMetric, QueryRunnerTestHelper.UNIQUES_9).build(), ImmutableMap.<String, Object>builder().put("market", "total_market").put(QueryRunnerTestHelper.uniqueMetric, QueryRunnerTestHelper.UNIQUES_2).put(QueryRunnerTestHelper.hyperUniqueFinalizingPostAggMetric, QueryRunnerTestHelper.UNIQUES_2).build(), ImmutableMap.<String, Object>builder().put("market", "upfront").put(QueryRunnerTestHelper.uniqueMetric, QueryRunnerTestHelper.UNIQUES_2).put(QueryRunnerTestHelper.hyperUniqueFinalizingPostAggMetric, QueryRunnerTestHelper.UNIQUES_2).build()))));
    assertExpectedResults(expectedResults, query);
}

Example 47 with PostAggregator

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

the class TopNQueryTest method testQuerySerialization.

@Test
public void testQuerySerialization() throws IOException {
    Query query = new TopNQueryBuilder().dataSource(dataSource).granularity(allGran).dimension(marketDimension).metric(indexMetric).threshold(4).intervals(fullOnInterval).aggregators(Lists.<AggregatorFactory>newArrayList(Iterables.concat(commonAggregators, Lists.newArrayList(new DoubleMaxAggregatorFactory("maxIndex", "index"), new DoubleMinAggregatorFactory("minIndex", "index"))))).postAggregators(Arrays.<PostAggregator>asList(addRowsIndexConstant)).build();
    String json = jsonMapper.writeValueAsString(query);
    Query serdeQuery = jsonMapper.readValue(json, Query.class);
    Assert.assertEquals(query, serdeQuery);
}

Example 48 with PostAggregator

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

the class AppendTest method testRowFiltering.

@Test
public void testRowFiltering() {
    List<Result<TimeseriesResultValue>> expectedResults = Arrays.asList(new Result<TimeseriesResultValue>(new DateTime("2011-01-12T00:00:00.000Z"), new TimeseriesResultValue(ImmutableMap.<String, Object>builder().put("rows", 5L).put("index", 500.0D).put("addRowsIndexConstant", 506.0D).put("uniques", 0.0D).put("maxIndex", 100.0D).put("minIndex", 100.0D).build())));
    TimeseriesQuery query = Druids.newTimeseriesQueryBuilder().dataSource(dataSource).granularity(allGran).intervals(fullOnInterval).filters(marketDimension, "breakstuff").aggregators(Lists.<AggregatorFactory>newArrayList(Iterables.concat(commonAggregators, Lists.newArrayList(new DoubleMaxAggregatorFactory("maxIndex", "index"), new DoubleMinAggregatorFactory("minIndex", "index"))))).postAggregators(Arrays.<PostAggregator>asList(addRowsIndexConstant)).build();
    QueryRunner runner = TestQueryRunners.makeTimeSeriesQueryRunner(segment3);
    HashMap<String, Object> context = new HashMap<String, Object>();
    TestHelper.assertExpectedResults(expectedResults, runner.run(query, context));
}

Example 49 with PostAggregator

use of io.druid.query.aggregation.PostAggregator 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 50 with PostAggregator

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

the class GroupByRules method applyPostAggregation.

/**
   * Applies a projection to the aggregations of a druidRel, by potentially adding post-aggregators.
   *
   * @return new rel, or null if the projection cannot be applied
   */
private static DruidRel applyPostAggregation(final DruidRel druidRel, final Project postProject) {
    Preconditions.checkState(canApplyPostAggregation(druidRel), "Cannot applyPostAggregation");
    final List<String> rowOrder = druidRel.getQueryBuilder().getRowOrder();
    final Grouping grouping = druidRel.getQueryBuilder().getGrouping();
    final List<Aggregation> newAggregations = Lists.newArrayList(grouping.getAggregations());
    final List<PostAggregatorFactory> finalizingPostAggregatorFactories = Lists.newArrayList();
    final List<String> newRowOrder = Lists.newArrayList();
    // Build list of finalizingPostAggregatorFactories.
    final Map<String, Aggregation> aggregationMap = Maps.newHashMap();
    for (final Aggregation aggregation : grouping.getAggregations()) {
        aggregationMap.put(aggregation.getOutputName(), aggregation);
    }
    for (final String field : rowOrder) {
        final Aggregation aggregation = aggregationMap.get(field);
        finalizingPostAggregatorFactories.add(aggregation == null ? null : aggregation.getFinalizingPostAggregatorFactory());
    }
    // Walk through the postProject expressions.
    for (final RexNode projectExpression : postProject.getChildExps()) {
        if (projectExpression.isA(SqlKind.INPUT_REF)) {
            final RexInputRef ref = (RexInputRef) projectExpression;
            final String fieldName = rowOrder.get(ref.getIndex());
            newRowOrder.add(fieldName);
            finalizingPostAggregatorFactories.add(null);
        } else {
            // Attempt to convert to PostAggregator.
            final String postAggregatorName = aggOutputName(newAggregations.size());
            final PostAggregator postAggregator = Expressions.toPostAggregator(postAggregatorName, rowOrder, finalizingPostAggregatorFactories, projectExpression);
            if (postAggregator != null) {
                newAggregations.add(Aggregation.create(postAggregator));
                newRowOrder.add(postAggregator.getName());
                finalizingPostAggregatorFactories.add(null);
            } else {
                return null;
            }
        }
    }
    return druidRel.withQueryBuilder(druidRel.getQueryBuilder().withAdjustedGrouping(Grouping.create(grouping.getDimensions(), newAggregations), postProject.getRowType(), newRowOrder));
}