use of org.locationtech.geowave.core.store.query.filter.expression.FilterConstraints in project geowave by locationtech.
the class OptimalExpressionQuery method determineBestIndices.
@SuppressWarnings({ "rawtypes", "unchecked" })
public List<Pair<Index, List<InternalDataAdapter<?>>>> determineBestIndices(final BaseQueryOptions baseOptions, final InternalDataAdapter<?>[] adapters, final AdapterIndexMappingStore adapterIndexMappingStore, final IndexStore indexStore, final DataStatisticsStore statisticsStore) {
final Map<Index, List<InternalDataAdapter<?>>> bestIndices = Maps.newHashMap();
final Set<String> referencedFields = Sets.newHashSet();
filter.addReferencedFields(referencedFields);
for (final InternalDataAdapter<?> adapter : adapters) {
if (!adapterMatchesFilter(adapter, referencedFields)) {
continue;
}
final AdapterToIndexMapping[] adapterIndices = adapterIndexMappingStore.getIndicesForAdapter(adapter.getAdapterId());
final Map<Index, FilterConstraints<?>> indexConstraints = Maps.newHashMap();
Index bestIndex = null;
for (final AdapterToIndexMapping mapping : adapterIndices) {
if ((baseOptions.getIndexName() != null) && !baseOptions.getIndexName().equals(mapping.getIndexName())) {
continue;
}
final Index index = mapping.getIndex(indexStore);
if (indexFilter != null && !indexFilter.test(index)) {
continue;
}
if ((bestIndex == null) || ((bestIndex instanceof AttributeIndex) && !(index instanceof AttributeIndex))) {
bestIndex = index;
}
final Set<String> indexedFields = Sets.newHashSet();
final Class<? extends Comparable> filterClass;
if ((index instanceof CustomIndex) && (((CustomIndex<?, ?>) index).getCustomIndexStrategy() instanceof TextIndexStrategy)) {
final TextIndexStrategy<?> indexStrategy = (TextIndexStrategy<?>) ((CustomIndex<?, ?>) index).getCustomIndexStrategy();
if (!(indexStrategy.getEntryConverter() instanceof AdapterFieldTextIndexEntryConverter)) {
continue;
}
indexedFields.add(((AdapterFieldTextIndexEntryConverter<?>) indexStrategy.getEntryConverter()).getFieldName());
filterClass = String.class;
} else {
for (final IndexFieldMapper<?, ?> mapper : mapping.getIndexFieldMappers()) {
for (final String adapterField : mapper.getAdapterFields()) {
indexedFields.add(adapterField);
}
}
// Remove any fields that are part of the common index model, but not used in the index
// strategy. They shouldn't be considered when trying to find a best match. In the future
// it may be useful to consider an index that has extra common index dimensions that
// contain filtered fields over one that only matches indexed dimensions. For example, if
// I have a spatial index, and a spatial index that stores time, it should pick the one
// that stores time if I supply a temporal constraint, even though it isn't part of the
// index strategy.
final int modelDimensions = index.getIndexModel().getDimensions().length;
final int strategyDimensions = index.getIndexStrategy().getOrderedDimensionDefinitions().length;
for (int i = modelDimensions - 1; i >= strategyDimensions; i--) {
final IndexFieldMapper<?, ?> mapper = mapping.getMapperForIndexField(index.getIndexModel().getDimensions()[i].getFieldName());
for (final String adapterField : mapper.getAdapterFields()) {
indexedFields.remove(adapterField);
}
}
filterClass = Double.class;
}
if (referencedFields.containsAll(indexedFields)) {
final FilterConstraints<?> constraints = filter.getConstraints(filterClass, statisticsStore, adapter, mapping, index, indexedFields);
if (constraints.constrainsAllFields(indexedFields)) {
indexConstraints.put(index, constraints);
}
}
}
if (indexConstraints.size() == 1) {
final Entry<Index, FilterConstraints<?>> bestEntry = indexConstraints.entrySet().iterator().next();
bestIndex = bestEntry.getKey();
constraintCache.put(adapter.getTypeName(), bestEntry.getValue());
} else if (indexConstraints.size() > 1) {
// determine which constraint is the best
double bestCardinality = Double.MAX_VALUE;
Index bestConstrainedIndex = null;
for (final Entry<Index, FilterConstraints<?>> entry : indexConstraints.entrySet()) {
final QueryRanges ranges = entry.getValue().getQueryRanges(baseOptions, statisticsStore);
if (ranges.isEmpty()) {
continue;
}
// TODO: A future optimization would be to add a default numeric histogram for any numeric
// index dimensions and just use the index data ranges to determine cardinality rather
// than decomposing query ranges.
final StatisticId<RowRangeHistogramValue> statisticId = IndexStatistic.generateStatisticId(entry.getKey().getName(), RowRangeHistogramStatistic.STATS_TYPE, Statistic.INTERNAL_TAG);
final RowRangeHistogramStatistic histogram = (RowRangeHistogramStatistic) statisticsStore.getStatisticById(statisticId);
final double cardinality = DataStoreUtils.cardinality(statisticsStore, histogram, adapter, bestConstrainedIndex, ranges);
if ((bestConstrainedIndex == null) || (cardinality < bestCardinality)) {
bestConstrainedIndex = entry.getKey();
bestCardinality = cardinality;
}
}
if (bestConstrainedIndex != null) {
bestIndex = bestConstrainedIndex;
constraintCache.put(adapter.getTypeName(), indexConstraints.get(bestIndex));
}
}
if (bestIndex == null) {
continue;
}
if (!bestIndices.containsKey(bestIndex)) {
bestIndices.put(bestIndex, Lists.newArrayList());
}
bestIndices.get(bestIndex).add(adapter);
}
return bestIndices.entrySet().stream().map(e -> Pair.of(e.getKey(), e.getValue())).collect(Collectors.toList());
}
Aggregations