use of com.linkedin.pinot.common.request.FilterOperator in project pinot by linkedin.
the class RangeMergeOptimizer method optimizeRanges.
/**
* Recursive method that performs the actual optimization of merging range predicates.
*
* @param current Current node being visited in the DFS of the filter query tree.
* @param timeColumn Name of time column
* @return Returns the optimized filter query tree
*/
@Nonnull
private static FilterQueryTree optimizeRanges(@Nonnull FilterQueryTree current, @Nullable String timeColumn) {
if (timeColumn == null) {
return current;
}
List<FilterQueryTree> children = current.getChildren();
if (children == null || children.isEmpty()) {
return current;
}
// For OR, we optimize all its children, but do not propagate up.
FilterOperator operator = current.getOperator();
if (operator == FilterOperator.OR) {
int length = children.size();
for (int i = 0; i < length; i++) {
children.set(i, optimizeRanges(children.get(i), timeColumn));
}
return current;
}
// After this point, since the node has children, it can only be an 'AND' node (only OR/AND supported).
assert operator == FilterOperator.AND;
List<FilterQueryTree> newChildren = new ArrayList<>();
List<String> intersect = null;
for (FilterQueryTree child : children) {
FilterQueryTree newChild = optimizeRanges(child, timeColumn);
if (newChild.getOperator() == FilterOperator.RANGE && newChild.getColumn().equals(timeColumn)) {
List<String> value = newChild.getValue();
intersect = (intersect == null) ? value : intersectRanges(intersect, value);
} else {
newChildren.add(newChild);
}
}
if (newChildren.isEmpty()) {
return new FilterQueryTree(timeColumn, intersect, FilterOperator.RANGE, null);
} else {
if (intersect != null) {
newChildren.add(new FilterQueryTree(timeColumn, intersect, FilterOperator.RANGE, null));
}
return new FilterQueryTree(null, null, FilterOperator.AND, newChildren);
}
}
use of com.linkedin.pinot.common.request.FilterOperator in project pinot by linkedin.
the class Predicate method newPredicate.
public static Predicate newPredicate(FilterQueryTree filterQueryTree) {
assert (filterQueryTree.getChildren() == null) || filterQueryTree.getChildren().isEmpty();
final FilterOperator filterType = filterQueryTree.getOperator();
final String column = filterQueryTree.getColumn();
final List<String> value = filterQueryTree.getValue();
Predicate predicate = null;
switch(filterType) {
case EQUALITY:
predicate = new EqPredicate(column, value);
break;
case RANGE:
predicate = new RangePredicate(column, value);
break;
case REGEX:
predicate = new RegexPredicate(column, value);
break;
case NOT:
predicate = new NEqPredicate(column, value);
break;
case NOT_IN:
predicate = new NotInPredicate(column, value);
break;
case IN:
predicate = new InPredicate(column, value);
break;
default:
throw new UnsupportedOperationException("Unsupported filterType:" + filterType);
}
return predicate;
}
use of com.linkedin.pinot.common.request.FilterOperator in project pinot by linkedin.
the class ColumnValueSegmentPruner method pruneSegment.
/**
* Helper method to determine if a segment can be pruned based on the column min/max value in segment metadata and
* the predicates on time column. The algorithm is as follows:
*
* <ul>
* <li> For leaf node: Returns true if there is a predicate on the column and apply the predicate would result in
* filtering out all docs of the segment, false otherwise. </li>
* <li> For non-leaf AND node: True if any of its children returned true, false otherwise. </li>
* <li> For non-leaf OR node: True if all its children returned true, false otherwise. </li>
* </ul>
*
* @param filterQueryTree Filter tree for the query.
* @param columnMetadataMap Map from column name to column metadata.
* @return True if segment can be pruned out, false otherwise.
*/
@SuppressWarnings("unchecked")
public static boolean pruneSegment(@Nonnull FilterQueryTree filterQueryTree, @Nonnull Map<String, ColumnMetadata> columnMetadataMap) {
FilterOperator filterOperator = filterQueryTree.getOperator();
List<FilterQueryTree> children = filterQueryTree.getChildren();
if (children == null || children.isEmpty()) {
// Skip operator other than EQUALITY and RANGE
if ((filterOperator != FilterOperator.EQUALITY) && (filterOperator != FilterOperator.RANGE)) {
return false;
}
ColumnMetadata columnMetadata = columnMetadataMap.get(filterQueryTree.getColumn());
if (columnMetadata == null) {
// Should not reach here after DataSchemaSegmentPruner
return true;
}
Comparable minValue = columnMetadata.getMinValue();
Comparable maxValue = columnMetadata.getMaxValue();
if (filterOperator == FilterOperator.EQUALITY) {
// Doesn't have min/max value set in metadata
if ((minValue == null) || (maxValue == null)) {
return false;
}
// Check if the value is in the min/max range
FieldSpec.DataType dataType = columnMetadata.getDataType();
Comparable value = getValue(filterQueryTree.getValue().get(0), dataType);
return (value.compareTo(minValue) < 0) || (value.compareTo(maxValue) > 0);
} else {
// RANGE
// Get lower/upper boundary value
FieldSpec.DataType dataType = columnMetadata.getDataType();
RangePredicate rangePredicate = new RangePredicate(null, filterQueryTree.getValue());
String lowerBoundary = rangePredicate.getLowerBoundary();
boolean includeLowerBoundary = rangePredicate.includeLowerBoundary();
Comparable lowerBoundaryValue = null;
if (!lowerBoundary.equals(RangePredicate.UNBOUNDED)) {
lowerBoundaryValue = getValue(lowerBoundary, dataType);
}
String upperBoundary = rangePredicate.getUpperBoundary();
boolean includeUpperBoundary = rangePredicate.includeUpperBoundary();
Comparable upperBoundaryValue = null;
if (!upperBoundary.equals(RangePredicate.UNBOUNDED)) {
upperBoundaryValue = getValue(upperBoundary, dataType);
}
// Check if the range is valid
if ((lowerBoundaryValue != null) && (upperBoundaryValue != null)) {
if (includeLowerBoundary && includeUpperBoundary) {
if (lowerBoundaryValue.compareTo(upperBoundaryValue) > 0) {
return true;
}
} else {
if (lowerBoundaryValue.compareTo(upperBoundaryValue) >= 0) {
return true;
}
}
}
// Doesn't have min/max value set in metadata
if ((minValue == null) || (maxValue == null)) {
return false;
}
if (lowerBoundaryValue != null) {
if (includeLowerBoundary) {
if (lowerBoundaryValue.compareTo(maxValue) > 0) {
return true;
}
} else {
if (lowerBoundaryValue.compareTo(maxValue) >= 0) {
return true;
}
}
}
if (upperBoundaryValue != null) {
if (includeUpperBoundary) {
if (upperBoundaryValue.compareTo(minValue) < 0) {
return true;
}
} else {
if (upperBoundaryValue.compareTo(minValue) <= 0) {
return true;
}
}
}
return false;
}
} else {
switch(filterOperator) {
case AND:
for (FilterQueryTree child : children) {
if (pruneSegment(child, columnMetadataMap)) {
return true;
}
}
return false;
case OR:
for (FilterQueryTree child : children) {
if (!pruneSegment(child, columnMetadataMap)) {
return false;
}
}
return true;
default:
throw new IllegalStateException("Unsupported filter operator: " + filterOperator);
}
}
}
use of com.linkedin.pinot.common.request.FilterOperator in project pinot by linkedin.
the class SegmentQueryProcessor method filterDocIds.
private List<Integer> filterDocIds(FilterQueryTree filterQueryTree, List<Integer> inputDocIds) {
// If no filter predicate, return the input without filtering.
if (filterQueryTree == null) {
List<Integer> allDocs = new ArrayList<>(_totalDocs);
for (int i = 0; i < _totalDocs; ++i) {
allDocs.add(i);
}
return allDocs;
}
final List<FilterQueryTree> childFilters = filterQueryTree.getChildren();
final boolean isLeaf = (childFilters == null) || childFilters.isEmpty();
if (isLeaf) {
FilterOperator filterType = filterQueryTree.getOperator();
String column = filterQueryTree.getColumn();
final List<String> value = filterQueryTree.getValue();
return getMatchingDocIds(inputDocIds, filterType, column, value);
}
List<Integer> result = filterDocIds(childFilters.get(0), inputDocIds);
final FilterOperator operator = filterQueryTree.getOperator();
for (int i = 1; i < childFilters.size(); ++i) {
// List<Integer> childResult = operator.equals(FilterOperator.AND) ? filterDocIds(childFilters.get(i), result)
// : filterDocIds(childFilters.get(i), inputDocIds);
List<Integer> childResult = filterDocIds(childFilters.get(i), inputDocIds);
result = combine(result, childResult, operator);
}
return result;
}
use of com.linkedin.pinot.common.request.FilterOperator in project pinot by linkedin.
the class FilterPlanNode method constructPhysicalOperator.
/**
* Helper method to build the operator tree from the filter query tree.
* @param filterQueryTree
* @param segment Index segment
* @param optimizeAlwaysFalse Optimize isResultEmpty predicates
* @return Filter Operator created
*/
@VisibleForTesting
public static BaseFilterOperator constructPhysicalOperator(FilterQueryTree filterQueryTree, IndexSegment segment, boolean optimizeAlwaysFalse) {
BaseFilterOperator ret;
if (null == filterQueryTree) {
return new MatchEntireSegmentOperator(segment.getSegmentMetadata().getTotalRawDocs());
}
final List<FilterQueryTree> childFilters = filterQueryTree.getChildren();
final boolean isLeaf = (childFilters == null) || childFilters.isEmpty();
if (!isLeaf) {
int numChildrenAlwaysFalse = 0;
int numChildren = childFilters.size();
List<BaseFilterOperator> operators = new ArrayList<>();
final FilterOperator filterType = filterQueryTree.getOperator();
for (final FilterQueryTree query : childFilters) {
BaseFilterOperator childOperator = constructPhysicalOperator(query, segment, optimizeAlwaysFalse);
// Count number of always false children.
if (optimizeAlwaysFalse && childOperator.isResultEmpty()) {
numChildrenAlwaysFalse++;
// Early bailout for 'AND' as soon as one of the children always evaluates to false.
if (filterType == FilterOperator.AND) {
break;
}
}
operators.add(childOperator);
}
ret = buildNonLeafOperator(filterType, operators, numChildrenAlwaysFalse, numChildren, optimizeAlwaysFalse);
} else {
final FilterOperator filterType = filterQueryTree.getOperator();
final String column = filterQueryTree.getColumn();
Predicate predicate = Predicate.newPredicate(filterQueryTree);
DataSource ds;
ds = segment.getDataSource(column);
DataSourceMetadata dataSourceMetadata = ds.getDataSourceMetadata();
BaseFilterOperator baseFilterOperator;
int startDocId = 0;
//end is inclusive
int endDocId = segment.getSegmentMetadata().getTotalRawDocs() - 1;
if (dataSourceMetadata.hasInvertedIndex()) {
// range evaluation based on inv index is inefficient, so do this only if is NOT range.
if (!filterType.equals(FilterOperator.RANGE)) {
if (dataSourceMetadata.isSingleValue() && dataSourceMetadata.isSorted()) {
// if the column is sorted use sorted inverted index based implementation
baseFilterOperator = new SortedInvertedIndexBasedFilterOperator(predicate, ds, startDocId, endDocId);
} else {
baseFilterOperator = new BitmapBasedFilterOperator(predicate, ds, startDocId, endDocId);
}
} else {
baseFilterOperator = new ScanBasedFilterOperator(predicate, ds, startDocId, endDocId);
}
} else {
baseFilterOperator = new ScanBasedFilterOperator(predicate, ds, startDocId, endDocId);
}
ret = baseFilterOperator;
}
// If operator evaluates to false, then just return an empty operator.
if (ret.isResultEmpty()) {
ret = new EmptyFilterOperator();
}
return ret;
}
Aggregations