Search in sources :

Example 6 with DimFilter

use of io.druid.query.filter.DimFilter in project druid by druid-io.

the class DumpSegment method runDump.

private void runDump(final Injector injector, final QueryableIndex index) throws IOException {
    final ObjectMapper objectMapper = injector.getInstance(Key.get(ObjectMapper.class, Json.class));
    final QueryableIndexStorageAdapter adapter = new QueryableIndexStorageAdapter(index);
    final List<String> columnNames = getColumnsToInclude(index);
    final DimFilter filter = filterJson != null ? objectMapper.readValue(filterJson, DimFilter.class) : null;
    final Sequence<Cursor> cursors = adapter.makeCursors(Filters.toFilter(filter), index.getDataInterval().withChronology(ISOChronology.getInstanceUTC()), VirtualColumns.EMPTY, Granularities.ALL, false);
    withOutputStream(new Function<OutputStream, Object>() {

        @Override
        public Object apply(final OutputStream out) {
            final Sequence<Object> sequence = Sequences.map(cursors, new Function<Cursor, Object>() {

                @Override
                public Object apply(Cursor cursor) {
                    final List<ObjectColumnSelector> selectors = Lists.newArrayList();
                    for (String columnName : columnNames) {
                        selectors.add(makeSelector(columnName, index.getColumn(columnName), cursor));
                    }
                    while (!cursor.isDone()) {
                        final Map<String, Object> row = Maps.newLinkedHashMap();
                        for (int i = 0; i < columnNames.size(); i++) {
                            final String columnName = columnNames.get(i);
                            final Object value = selectors.get(i).get();
                            if (timeISO8601 && columnNames.get(i).equals(Column.TIME_COLUMN_NAME)) {
                                row.put(columnName, new DateTime(value, DateTimeZone.UTC).toString());
                            } else {
                                row.put(columnName, value);
                            }
                        }
                        try {
                            out.write(objectMapper.writeValueAsBytes(row));
                            out.write('\n');
                        } catch (IOException e) {
                            throw Throwables.propagate(e);
                        }
                        cursor.advance();
                    }
                    return null;
                }
            });
            evaluateSequenceForSideEffects(sequence);
            return null;
        }
    });
}
Also used : OutputStream(java.io.OutputStream) FileOutputStream(java.io.FileOutputStream) QueryableIndexStorageAdapter(io.druid.segment.QueryableIndexStorageAdapter) Json(io.druid.guice.annotations.Json) IOException(java.io.IOException) Sequence(io.druid.java.util.common.guava.Sequence) Cursor(io.druid.segment.Cursor) DateTime(org.joda.time.DateTime) Function(com.google.common.base.Function) DimFilter(io.druid.query.filter.DimFilter) ObjectMapper(com.fasterxml.jackson.databind.ObjectMapper) ObjectColumnSelector(io.druid.segment.ObjectColumnSelector)

Example 7 with DimFilter

use of io.druid.query.filter.DimFilter in project druid by druid-io.

the class DruidFilterRule method onMatch.

@Override
public void onMatch(RelOptRuleCall call) {
    final Filter filter = call.rel(0);
    final DruidRel druidRel = call.rel(1);
    if (druidRel.getQueryBuilder().getFilter() != null || druidRel.getQueryBuilder().getSelectProjection() != null || druidRel.getQueryBuilder().getGrouping() != null) {
        return;
    }
    final DimFilter dimFilter = Expressions.toFilter(operatorTable, druidRel.getPlannerContext(), druidRel.getSourceRowSignature(), filter.getCondition());
    if (dimFilter != null) {
        call.transformTo(druidRel.withQueryBuilder(druidRel.getQueryBuilder().withFilter(dimFilter)));
    }
}
Also used : DruidRel(io.druid.sql.calcite.rel.DruidRel) Filter(org.apache.calcite.rel.core.Filter) DimFilter(io.druid.query.filter.DimFilter) DimFilter(io.druid.query.filter.DimFilter)

Example 8 with DimFilter

use of io.druid.query.filter.DimFilter 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;
    }
}
Also used : RexLiteral(org.apache.calcite.rex.RexLiteral) ArithmeticPostAggregator(io.druid.query.aggregation.post.ArithmeticPostAggregator) DoubleMaxAggregatorFactory(io.druid.query.aggregation.DoubleMaxAggregatorFactory) SqlTypeName(org.apache.calcite.sql.type.SqlTypeName) LongSumAggregatorFactory(io.druid.query.aggregation.LongSumAggregatorFactory) DoubleMinAggregatorFactory(io.druid.query.aggregation.DoubleMinAggregatorFactory) LongMinAggregatorFactory(io.druid.query.aggregation.LongMinAggregatorFactory) RexCall(org.apache.calcite.rex.RexCall) Aggregation(io.druid.sql.calcite.aggregation.Aggregation) ISE(io.druid.java.util.common.ISE) LongMaxAggregatorFactory(io.druid.query.aggregation.LongMaxAggregatorFactory) NotDimFilter(io.druid.query.filter.NotDimFilter) FieldAccessPostAggregator(io.druid.query.aggregation.post.FieldAccessPostAggregator) DoubleSumAggregatorFactory(io.druid.query.aggregation.DoubleSumAggregatorFactory) PostAggregator(io.druid.query.aggregation.PostAggregator) FieldAccessPostAggregator(io.druid.query.aggregation.post.FieldAccessPostAggregator) ArithmeticPostAggregator(io.druid.query.aggregation.post.ArithmeticPostAggregator) SqlKind(org.apache.calcite.sql.SqlKind) CountAggregatorFactory(io.druid.query.aggregation.CountAggregatorFactory) DoubleMaxAggregatorFactory(io.druid.query.aggregation.DoubleMaxAggregatorFactory) LongMaxAggregatorFactory(io.druid.query.aggregation.LongMaxAggregatorFactory) DoubleSumAggregatorFactory(io.druid.query.aggregation.DoubleSumAggregatorFactory) AggregatorFactory(io.druid.query.aggregation.AggregatorFactory) LongMinAggregatorFactory(io.druid.query.aggregation.LongMinAggregatorFactory) PostAggregatorFactory(io.druid.sql.calcite.aggregation.PostAggregatorFactory) DoubleMinAggregatorFactory(io.druid.query.aggregation.DoubleMinAggregatorFactory) LongSumAggregatorFactory(io.druid.query.aggregation.LongSumAggregatorFactory) CountAggregatorFactory(io.druid.query.aggregation.CountAggregatorFactory) SqlAggregator(io.druid.sql.calcite.aggregation.SqlAggregator) ApproxCountDistinctSqlAggregator(io.druid.sql.calcite.aggregation.ApproxCountDistinctSqlAggregator) DimFilter(io.druid.query.filter.DimFilter) NotDimFilter(io.druid.query.filter.NotDimFilter) AndDimFilter(io.druid.query.filter.AndDimFilter) RexNode(org.apache.calcite.rex.RexNode)

Example 9 with DimFilter

use of io.druid.query.filter.DimFilter in project druid by druid-io.

the class Expressions method toFilter.

/**
   * Translates "condition" to a Druid filter, or returns null if we cannot translate the condition.
   *
   * @param plannerContext planner context
   * @param rowSignature   row signature of the dataSource to be filtered
   * @param expression     Calcite row expression
   */
public static DimFilter toFilter(final DruidOperatorTable operatorTable, final PlannerContext plannerContext, final RowSignature rowSignature, final RexNode expression) {
    if (expression.getKind() == SqlKind.AND || expression.getKind() == SqlKind.OR || expression.getKind() == SqlKind.NOT) {
        final List<DimFilter> filters = Lists.newArrayList();
        for (final RexNode rexNode : ((RexCall) expression).getOperands()) {
            final DimFilter nextFilter = toFilter(operatorTable, plannerContext, rowSignature, rexNode);
            if (nextFilter == null) {
                return null;
            }
            filters.add(nextFilter);
        }
        if (expression.getKind() == SqlKind.AND) {
            return new AndDimFilter(filters);
        } else if (expression.getKind() == SqlKind.OR) {
            return new OrDimFilter(filters);
        } else {
            assert expression.getKind() == SqlKind.NOT;
            return new NotDimFilter(Iterables.getOnlyElement(filters));
        }
    } else {
        // Handle filter conditions on everything else.
        return toLeafFilter(operatorTable, plannerContext, rowSignature, expression);
    }
}
Also used : RexCall(org.apache.calcite.rex.RexCall) NotDimFilter(io.druid.query.filter.NotDimFilter) AndDimFilter(io.druid.query.filter.AndDimFilter) OrDimFilter(io.druid.query.filter.OrDimFilter) LikeDimFilter(io.druid.query.filter.LikeDimFilter) OrDimFilter(io.druid.query.filter.OrDimFilter) DimFilter(io.druid.query.filter.DimFilter) NotDimFilter(io.druid.query.filter.NotDimFilter) AndDimFilter(io.druid.query.filter.AndDimFilter) RexNode(org.apache.calcite.rex.RexNode)

Example 10 with DimFilter

use of io.druid.query.filter.DimFilter in project druid by druid-io.

the class CombineAndSimplifyBounds method process.

@Override
public DimFilter process(DimFilter filter) {
    if (filter instanceof AndDimFilter) {
        final List<DimFilter> children = getAndFilterChildren((AndDimFilter) filter);
        final DimFilter one = doSimplifyAnd(children);
        final DimFilter two = negate(doSimplifyOr(negateAll(children)));
        return computeCost(one) <= computeCost(two) ? one : two;
    } else if (filter instanceof OrDimFilter) {
        final List<DimFilter> children = getOrFilterChildren((OrDimFilter) filter);
        final DimFilter one = doSimplifyOr(children);
        final DimFilter two = negate(doSimplifyAnd(negateAll(children)));
        return computeCost(one) <= computeCost(two) ? one : two;
    } else if (filter instanceof NotDimFilter) {
        final DimFilter field = ((NotDimFilter) filter).getField();
        final DimFilter candidate;
        if (field instanceof OrDimFilter) {
            candidate = doSimplifyAnd(negateAll(getOrFilterChildren((OrDimFilter) field)));
        } else if (field instanceof AndDimFilter) {
            candidate = doSimplifyOr(negateAll(getAndFilterChildren((AndDimFilter) field)));
        } else {
            candidate = negate(field);
        }
        return computeCost(filter) <= computeCost(candidate) ? filter : candidate;
    } else {
        return filter;
    }
}
Also used : NotDimFilter(io.druid.query.filter.NotDimFilter) AndDimFilter(io.druid.query.filter.AndDimFilter) OrDimFilter(io.druid.query.filter.OrDimFilter) ArrayList(java.util.ArrayList) List(java.util.List) OrDimFilter(io.druid.query.filter.OrDimFilter) DimFilter(io.druid.query.filter.DimFilter) NotDimFilter(io.druid.query.filter.NotDimFilter) AndDimFilter(io.druid.query.filter.AndDimFilter) BoundDimFilter(io.druid.query.filter.BoundDimFilter)

Aggregations

DimFilter (io.druid.query.filter.DimFilter)56 AndDimFilter (io.druid.query.filter.AndDimFilter)45 SelectorDimFilter (io.druid.query.filter.SelectorDimFilter)39 Test (org.junit.Test)32 BoundDimFilter (io.druid.query.filter.BoundDimFilter)30 OrDimFilter (io.druid.query.filter.OrDimFilter)27 InDimFilter (io.druid.query.filter.InDimFilter)24 NotDimFilter (io.druid.query.filter.NotDimFilter)19 RegexDimFilter (io.druid.query.filter.RegexDimFilter)18 Result (io.druid.query.Result)14 ExtractionDimFilter (io.druid.query.filter.ExtractionDimFilter)14 DateTime (org.joda.time.DateTime)14 JavaScriptDimFilter (io.druid.query.filter.JavaScriptDimFilter)11 SearchQueryDimFilter (io.druid.query.filter.SearchQueryDimFilter)11 List (java.util.List)11 DefaultDimensionSpec (io.druid.query.dimension.DefaultDimensionSpec)10 ArrayList (java.util.ArrayList)9 LongSumAggregatorFactory (io.druid.query.aggregation.LongSumAggregatorFactory)8 DimensionSpec (io.druid.query.dimension.DimensionSpec)7 Row (io.druid.data.input.Row)6