mirror of https://github.com/apache/druid.git
modify equality and typed in filter behavior for numeric match values on string columns (#16593)
* fix equality and typed in filter behavior for numeric match values on string columns changes: * EqualityFilter and TypedInfilter numeric match values against string columns will now cast strings to numeric values instead of converting the numeric values directly to string for pure string equality, which is consistent with the casts which are eaten in the SQL layer, as well as classic druid behavior * added tests to cover numeric equality matching. Double match values in particular would fail to match the string values since `1.0` would become `'1.0'` which does not match `'1'`.
This commit is contained in:
parent
7c6f2b1e20
commit
09e0eefdc3
|
@ -31,6 +31,7 @@ import com.google.common.collect.TreeRangeSet;
|
|||
import org.apache.druid.error.InvalidInput;
|
||||
import org.apache.druid.java.util.common.IAE;
|
||||
import org.apache.druid.math.expr.ExprEval;
|
||||
import org.apache.druid.math.expr.ExprType;
|
||||
import org.apache.druid.math.expr.ExpressionType;
|
||||
import org.apache.druid.query.cache.CacheKeyBuilder;
|
||||
import org.apache.druid.query.filter.vector.VectorValueMatcher;
|
||||
|
@ -43,12 +44,14 @@ import org.apache.druid.segment.ColumnInspector;
|
|||
import org.apache.druid.segment.ColumnProcessorFactory;
|
||||
import org.apache.druid.segment.ColumnProcessors;
|
||||
import org.apache.druid.segment.ColumnSelectorFactory;
|
||||
import org.apache.druid.segment.DimensionHandlerUtils;
|
||||
import org.apache.druid.segment.DimensionSelector;
|
||||
import org.apache.druid.segment.column.ColumnCapabilities;
|
||||
import org.apache.druid.segment.column.ColumnIndexSupplier;
|
||||
import org.apache.druid.segment.column.ColumnType;
|
||||
import org.apache.druid.segment.column.TypeSignature;
|
||||
import org.apache.druid.segment.column.TypeStrategy;
|
||||
import org.apache.druid.segment.column.Types;
|
||||
import org.apache.druid.segment.column.ValueType;
|
||||
import org.apache.druid.segment.filter.Filters;
|
||||
import org.apache.druid.segment.filter.PredicateValueMatcherFactory;
|
||||
|
@ -244,8 +247,9 @@ public class EqualityFilter extends AbstractOptimizableDimFilter implements Filt
|
|||
public VectorValueMatcher makeVectorMatcher(VectorColumnSelectorFactory factory)
|
||||
{
|
||||
final ColumnCapabilities capabilities = factory.getColumnCapabilities(column);
|
||||
|
||||
if (matchValueType.isPrimitive() && (capabilities == null || capabilities.isPrimitive())) {
|
||||
final boolean primitiveMatch = matchValueType.isPrimitive() && (capabilities == null || capabilities.isPrimitive());
|
||||
if (primitiveMatch && useSimpleEquality(capabilities, matchValueType)) {
|
||||
// if possible, use simplified value matcher instead of predicate
|
||||
return ColumnProcessors.makeVectorProcessor(
|
||||
column,
|
||||
VectorValueMatcherColumnProcessorFactory.instance(),
|
||||
|
@ -298,6 +302,20 @@ public class EqualityFilter extends AbstractOptimizableDimFilter implements Filt
|
|||
);
|
||||
}
|
||||
|
||||
/**
|
||||
* Can the match value type be cast directly to column type for equality comparison? For non-numeric match types, we
|
||||
* just use exact string equality regardless of the column type. For numeric match value types against string columns,
|
||||
* we instead cast the string to the match value type number for matching equality.
|
||||
*/
|
||||
public static boolean useSimpleEquality(TypeSignature<ValueType> columnType, ColumnType matchValueType)
|
||||
{
|
||||
if (Types.is(columnType, ValueType.STRING)) {
|
||||
return !matchValueType.isNumeric();
|
||||
}
|
||||
return true;
|
||||
}
|
||||
|
||||
@Nullable
|
||||
public static BitmapColumnIndex getEqualityIndex(
|
||||
String column,
|
||||
ExprEval<?> matchValueEval,
|
||||
|
@ -311,20 +329,22 @@ public class EqualityFilter extends AbstractOptimizableDimFilter implements Filt
|
|||
return new AllUnknownBitmapColumnIndex(selector);
|
||||
}
|
||||
|
||||
final ValueIndexes valueIndexes = indexSupplier.as(ValueIndexes.class);
|
||||
if (valueIndexes != null) {
|
||||
// matchValueEval.value() cannot be null here due to check in the constructor
|
||||
//noinspection DataFlowIssue
|
||||
return valueIndexes.forValue(matchValueEval.value(), matchValueType);
|
||||
}
|
||||
if (useSimpleEquality(selector.getColumnCapabilities(column), matchValueType)) {
|
||||
final ValueIndexes valueIndexes = indexSupplier.as(ValueIndexes.class);
|
||||
if (valueIndexes != null) {
|
||||
// matchValueEval.value() cannot be null here due to check in the constructor
|
||||
//noinspection DataFlowIssue
|
||||
return valueIndexes.forValue(matchValueEval.value(), matchValueType);
|
||||
}
|
||||
if (matchValueType.isPrimitive()) {
|
||||
final StringValueSetIndexes stringValueSetIndexes = indexSupplier.as(StringValueSetIndexes.class);
|
||||
if (stringValueSetIndexes != null) {
|
||||
|
||||
if (matchValueType.isPrimitive()) {
|
||||
final StringValueSetIndexes stringValueSetIndexes = indexSupplier.as(StringValueSetIndexes.class);
|
||||
if (stringValueSetIndexes != null) {
|
||||
|
||||
return stringValueSetIndexes.forValue(matchValueEval.asString());
|
||||
return stringValueSetIndexes.forValue(matchValueEval.asString());
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// fall back to predicate based index if it is available
|
||||
final DruidPredicateIndexes predicateIndexes = indexSupplier.as(DruidPredicateIndexes.class);
|
||||
if (predicateIndexes != null) {
|
||||
|
@ -408,11 +428,38 @@ public class EqualityFilter extends AbstractOptimizableDimFilter implements Filt
|
|||
private Supplier<DruidObjectPredicate<String>> makeStringPredicateSupplier()
|
||||
{
|
||||
return Suppliers.memoize(() -> {
|
||||
final ExprEval<?> castForComparison = ExprEval.castForEqualityComparison(matchValue, ExpressionType.STRING);
|
||||
if (castForComparison == null) {
|
||||
return DruidObjectPredicate.alwaysFalseWithNullUnknown();
|
||||
// when matching strings to numeric match values, use numeric comparator to implicitly cast the string to number
|
||||
if (matchValue.type().isNumeric()) {
|
||||
if (matchValue.type().is(ExprType.LONG)) {
|
||||
return value -> {
|
||||
if (value == null) {
|
||||
return DruidPredicateMatch.UNKNOWN;
|
||||
}
|
||||
final Long l = DimensionHandlerUtils.convertObjectToLong(value);
|
||||
if (l == null) {
|
||||
return DruidPredicateMatch.FALSE;
|
||||
}
|
||||
return DruidPredicateMatch.of(matchValue.asLong() == l);
|
||||
};
|
||||
} else {
|
||||
return value -> {
|
||||
if (value == null) {
|
||||
return DruidPredicateMatch.UNKNOWN;
|
||||
}
|
||||
final Double d = DimensionHandlerUtils.convertObjectToDouble(value);
|
||||
if (d == null) {
|
||||
return DruidPredicateMatch.FALSE;
|
||||
}
|
||||
return DruidPredicateMatch.of(matchValue.asDouble() == d);
|
||||
};
|
||||
}
|
||||
} else {
|
||||
final ExprEval<?> castForComparison = ExprEval.castForEqualityComparison(matchValue, ExpressionType.STRING);
|
||||
if (castForComparison == null) {
|
||||
return DruidObjectPredicate.alwaysFalseWithNullUnknown();
|
||||
}
|
||||
return DruidObjectPredicate.equalTo(castForComparison.asString());
|
||||
}
|
||||
return DruidObjectPredicate.equalTo(castForComparison.asString());
|
||||
});
|
||||
}
|
||||
|
||||
|
@ -548,6 +595,10 @@ public class EqualityFilter extends AbstractOptimizableDimFilter implements Filt
|
|||
@Override
|
||||
public ValueMatcher makeDimensionProcessor(DimensionSelector selector, boolean multiValue)
|
||||
{
|
||||
// use the predicate matcher when matching numeric values since it casts the strings to numeric types
|
||||
if (matchValue.type().isNumeric()) {
|
||||
return predicateMatcherFactory.makeDimensionProcessor(selector, multiValue);
|
||||
}
|
||||
final ExprEval<?> castForComparison = ExprEval.castForEqualityComparison(matchValue, ExpressionType.STRING);
|
||||
if (castForComparison == null) {
|
||||
return ValueMatchers.makeAlwaysFalseWithNullUnknownDimensionMatcher(selector, multiValue);
|
||||
|
|
|
@ -314,8 +314,7 @@ public class RangeFilter extends AbstractOptimizableDimFilter implements Filter
|
|||
final String upper = hasUpperBound() ? upperEval.asString() : null;
|
||||
return rangeIndexes.forRange(lower, lowerOpen, upper, upperOpen);
|
||||
}
|
||||
}
|
||||
if (matchValueType.isNumeric()) {
|
||||
} else if (matchValueType.isNumeric()) {
|
||||
final NumericRangeIndexes rangeIndexes = indexSupplier.as(NumericRangeIndexes.class);
|
||||
if (rangeIndexes != null) {
|
||||
final Number lower = (Number) lowerEval.value();
|
||||
|
|
|
@ -36,15 +36,21 @@ import com.google.common.collect.Sets;
|
|||
import com.google.common.collect.TreeRangeSet;
|
||||
import com.google.common.hash.Hasher;
|
||||
import com.google.common.hash.Hashing;
|
||||
import com.google.common.primitives.Doubles;
|
||||
import it.unimi.dsi.fastutil.doubles.DoubleOpenHashSet;
|
||||
import it.unimi.dsi.fastutil.doubles.DoubleSet;
|
||||
import it.unimi.dsi.fastutil.floats.FloatOpenHashSet;
|
||||
import it.unimi.dsi.fastutil.longs.LongOpenHashSet;
|
||||
import it.unimi.dsi.fastutil.longs.LongSet;
|
||||
import it.unimi.dsi.fastutil.objects.ObjectArrays;
|
||||
import org.apache.druid.common.config.NullHandling;
|
||||
import org.apache.druid.common.guava.GuavaUtils;
|
||||
import org.apache.druid.error.InvalidInput;
|
||||
import org.apache.druid.java.util.common.IAE;
|
||||
import org.apache.druid.java.util.common.StringUtils;
|
||||
import org.apache.druid.math.expr.Evals;
|
||||
import org.apache.druid.math.expr.ExprEval;
|
||||
import org.apache.druid.math.expr.ExpressionType;
|
||||
import org.apache.druid.query.cache.CacheKeyBuilder;
|
||||
import org.apache.druid.query.filter.vector.VectorValueMatcher;
|
||||
import org.apache.druid.query.filter.vector.VectorValueMatcherColumnProcessorFactory;
|
||||
|
@ -301,9 +307,11 @@ public class TypedInFilter extends AbstractOptimizableDimFilter implements Filte
|
|||
}
|
||||
}
|
||||
|
||||
final ValueSetIndexes valueSetIndexes = indexSupplier.as(ValueSetIndexes.class);
|
||||
if (valueSetIndexes != null) {
|
||||
return valueSetIndexes.forSortedValues(sortedMatchValues.get(), matchValueType);
|
||||
if (EqualityFilter.useSimpleEquality(selector.getColumnCapabilities(column), matchValueType)) {
|
||||
final ValueSetIndexes valueSetIndexes = indexSupplier.as(ValueSetIndexes.class);
|
||||
if (valueSetIndexes != null) {
|
||||
return valueSetIndexes.forSortedValues(sortedMatchValues.get(), matchValueType);
|
||||
}
|
||||
}
|
||||
|
||||
return Filters.makePredicateIndex(
|
||||
|
@ -452,20 +460,20 @@ public class TypedInFilter extends AbstractOptimizableDimFilter implements Filte
|
|||
}
|
||||
|
||||
@Nullable
|
||||
private static Object coerceValue(@Nullable Object o, ColumnType matchValueType)
|
||||
private static <T> T coerceValue(@Nullable Object o, ColumnType matchValueType)
|
||||
{
|
||||
if (o == null) {
|
||||
return null;
|
||||
}
|
||||
switch (matchValueType.getType()) {
|
||||
case STRING:
|
||||
return DimensionHandlerUtils.convertObjectToString(o);
|
||||
return (T) DimensionHandlerUtils.convertObjectToString(o);
|
||||
case LONG:
|
||||
return DimensionHandlerUtils.convertObjectToLong(o);
|
||||
return (T) DimensionHandlerUtils.convertObjectToLong(o);
|
||||
case FLOAT:
|
||||
return DimensionHandlerUtils.convertObjectToFloat(o);
|
||||
return (T) DimensionHandlerUtils.convertObjectToFloat(o);
|
||||
case DOUBLE:
|
||||
return DimensionHandlerUtils.convertObjectToDouble(o);
|
||||
return (T) DimensionHandlerUtils.convertObjectToDouble(o);
|
||||
default:
|
||||
throw InvalidInput.exception("Unsupported matchValueType[%s]", matchValueType);
|
||||
}
|
||||
|
@ -540,11 +548,51 @@ public class TypedInFilter extends AbstractOptimizableDimFilter implements Filte
|
|||
final int index = Collections.binarySearch(sortedValues, value, comparator);
|
||||
return DruidPredicateMatch.of(index >= 0);
|
||||
};
|
||||
} else if (matchValueType.is(ValueType.LONG)) {
|
||||
final LongSet valueSet = new LongOpenHashSet(sortedValues.size());
|
||||
for (Object o : sortedValues) {
|
||||
final Long l = DimensionHandlerUtils.convertObjectToLong(o);
|
||||
if (l != null) {
|
||||
valueSet.add(l.longValue());
|
||||
}
|
||||
}
|
||||
return value -> {
|
||||
if (value == null) {
|
||||
return containsNull ? DruidPredicateMatch.TRUE : DruidPredicateMatch.UNKNOWN;
|
||||
}
|
||||
final Long castValue = GuavaUtils.tryParseLong(value);
|
||||
if (castValue == null) {
|
||||
return DruidPredicateMatch.FALSE;
|
||||
}
|
||||
return DruidPredicateMatch.of(valueSet.contains(castValue));
|
||||
};
|
||||
} else if (matchValueType.isNumeric()) {
|
||||
// double or float
|
||||
final DoubleSet valueSet = new DoubleOpenHashSet(sortedValues.size());
|
||||
for (Object o : sortedValues) {
|
||||
Double d = DimensionHandlerUtils.convertObjectToDouble(o);
|
||||
if (d != null) {
|
||||
valueSet.add(d.doubleValue());
|
||||
}
|
||||
}
|
||||
return value -> {
|
||||
if (value == null) {
|
||||
return containsNull ? DruidPredicateMatch.TRUE : DruidPredicateMatch.UNKNOWN;
|
||||
}
|
||||
|
||||
final Double d = Doubles.tryParse(value);
|
||||
if (d == null) {
|
||||
return DruidPredicateMatch.FALSE;
|
||||
}
|
||||
return DruidPredicateMatch.of(valueSet.contains(d));
|
||||
};
|
||||
}
|
||||
|
||||
// convert set to strings
|
||||
final ExpressionType matchExpressionType = ExpressionType.fromColumnTypeStrict(matchValueType);
|
||||
final Set<String> stringSet = Sets.newHashSetWithExpectedSize(sortedValues.size());
|
||||
for (Object o : sortedValues) {
|
||||
stringSet.add(Evals.asString(o));
|
||||
stringSet.add(ExprEval.ofType(matchExpressionType, o).castTo(ExpressionType.STRING).asString());
|
||||
}
|
||||
return value -> {
|
||||
if (value == null) {
|
||||
|
|
|
@ -230,7 +230,7 @@ public final class IndexedUtf8ValueIndexes<TDictionary extends Indexed<ByteBuffe
|
|||
final Object minValueInColumn = dictionary.get(0);
|
||||
final int position = Collections.binarySearch(
|
||||
sortedValues,
|
||||
StringUtils.fromUtf8((ByteBuffer) minValueInColumn),
|
||||
StringUtils.fromUtf8Nullable((ByteBuffer) minValueInColumn),
|
||||
matchValueType.getNullableStrategy()
|
||||
);
|
||||
tailSet = baseSet.subList(position >= 0 ? position : -(position + 1), baseSet.size());
|
||||
|
|
|
@ -108,6 +108,12 @@ public class EqualityFilterTests
|
|||
NotDimFilter.of(new EqualityFilter("dim0", ColumnType.LONG, 1L, null)),
|
||||
ImmutableList.of("0", "2", "3", "4", "5")
|
||||
);
|
||||
|
||||
assertFilterMatches(new EqualityFilter("dim0", ColumnType.DOUBLE, 1, null), ImmutableList.of("1"));
|
||||
assertFilterMatches(
|
||||
NotDimFilter.of(new EqualityFilter("dim0", ColumnType.DOUBLE, 1, null)),
|
||||
ImmutableList.of("0", "2", "3", "4", "5")
|
||||
);
|
||||
}
|
||||
|
||||
@Test
|
||||
|
|
|
@ -138,6 +138,29 @@ public class InFilterTests
|
|||
NotDimFilter.of(inFilter("dim0", ColumnType.STRING, Arrays.asList("e", "x"))),
|
||||
ImmutableList.of("a", "b", "c", "d", "f")
|
||||
);
|
||||
|
||||
if (NullHandling.sqlCompatible()) {
|
||||
assertTypedFilterMatches(
|
||||
inFilter("dim1", ColumnType.LONG, Arrays.asList(2L, 10L)),
|
||||
ImmutableList.of("b", "c")
|
||||
);
|
||||
|
||||
assertTypedFilterMatches(
|
||||
inFilter("dim1", ColumnType.DOUBLE, Arrays.asList(2.0, 10.0)),
|
||||
ImmutableList.of("b", "c")
|
||||
);
|
||||
} else {
|
||||
// in default value mode, we actually end up using a classic InDimFilter, it does not match numbers well
|
||||
assertTypedFilterMatches(
|
||||
inFilter("dim1", ColumnType.LONG, Arrays.asList(2L, 10L)),
|
||||
ImmutableList.of("b", "c")
|
||||
);
|
||||
|
||||
assertTypedFilterMatches(
|
||||
inFilter("dim1", ColumnType.DOUBLE, Arrays.asList(2.0, 10.0)),
|
||||
ImmutableList.of()
|
||||
);
|
||||
}
|
||||
}
|
||||
@Test
|
||||
public void testSingleValueStringColumnWithNulls()
|
||||
|
|
Loading…
Reference in New Issue