use of org.apache.druid.query.groupby.orderby.NoopLimitSpec in project druid by druid-io.
the class GroupByStrategyV2 method processSubtotalsSpec.
@Override
public Sequence<ResultRow> processSubtotalsSpec(GroupByQuery query, GroupByQueryResource resource, Sequence<ResultRow> queryResult) {
// How it works?
// First we accumulate the result of top level base query aka queryResult arg inside a resultSupplierOne object.
// Next for each subtotalSpec
// If subtotalSpec is a prefix of top level dims then we iterate on rows in resultSupplierOne object which are still
// sorted by subtotalSpec, stream merge them and return.
//
// If subtotalSpec is not a prefix of top level dims then we create a resultSupplierTwo object filled with rows from
// resultSupplierOne object with only dims from subtotalSpec. Then we iterate on rows in resultSupplierTwo object which are
// of course sorted by subtotalSpec, stream merge them and return.
// Keep a reference to resultSupplier outside the "try" so we can close it if something goes wrong
// while creating the sequence.
GroupByRowProcessor.ResultSupplier resultSupplierOne = null;
try {
// baseSubtotalQuery is the original query with dimensions and aggregators rewritten to apply to the *results*
// rather than *inputs* of that query. It has its virtual columns and dim filter removed, because those only
// make sense when applied to inputs. Finally, it has subtotalsSpec removed, since we'll be computing them
// one-by-one soon enough.
GroupByQuery baseSubtotalQuery = query.withDimensionSpecs(query.getDimensions().stream().map(dimSpec -> new DefaultDimensionSpec(dimSpec.getOutputName(), dimSpec.getOutputName(), dimSpec.getOutputType())).collect(Collectors.toList())).withAggregatorSpecs(query.getAggregatorSpecs().stream().map(AggregatorFactory::getCombiningFactory).collect(Collectors.toList())).withVirtualColumns(VirtualColumns.EMPTY).withDimFilter(null).withSubtotalsSpec(null).withOverriddenContext(ImmutableMap.of(GroupByQuery.CTX_TIMESTAMP_RESULT_FIELD, ""));
resultSupplierOne = GroupByRowProcessor.process(baseSubtotalQuery, baseSubtotalQuery, queryResult, configSupplier.get(), resource, spillMapper, processingConfig.getTmpDir(), processingConfig.intermediateComputeSizeBytes());
List<String> queryDimNames = baseSubtotalQuery.getDimensions().stream().map(DimensionSpec::getOutputName).collect(Collectors.toList());
// Only needed to make LimitSpec.filterColumns(..) call later in case base query has a non default LimitSpec.
Set<String> aggsAndPostAggs = null;
if (!(baseSubtotalQuery.getLimitSpec() instanceof NoopLimitSpec)) {
aggsAndPostAggs = getAggregatorAndPostAggregatorNames(baseSubtotalQuery);
}
List<List<String>> subtotals = query.getSubtotalsSpec();
List<Sequence<ResultRow>> subtotalsResults = new ArrayList<>(subtotals.size());
// Iterate through each subtotalSpec, build results for it and add to subtotalsResults
for (List<String> subtotalSpec : subtotals) {
final ImmutableSet<String> dimsInSubtotalSpec = ImmutableSet.copyOf(subtotalSpec);
// Dimension spec including dimension name and output name
final List<DimensionSpec> subTotalDimensionSpec = new ArrayList<>(dimsInSubtotalSpec.size());
final List<DimensionSpec> dimensions = query.getDimensions();
for (DimensionSpec dimensionSpec : dimensions) {
if (dimsInSubtotalSpec.contains(dimensionSpec.getOutputName())) {
subTotalDimensionSpec.add(dimensionSpec);
}
}
// Create appropriate LimitSpec for subtotal query
LimitSpec subtotalQueryLimitSpec = NoopLimitSpec.instance();
if (!(baseSubtotalQuery.getLimitSpec() instanceof NoopLimitSpec)) {
Set<String> columns = new HashSet<>(aggsAndPostAggs);
columns.addAll(subtotalSpec);
subtotalQueryLimitSpec = baseSubtotalQuery.getLimitSpec().filterColumns(columns);
}
GroupByQuery subtotalQuery = baseSubtotalQuery.withLimitSpec(subtotalQueryLimitSpec);
final GroupByRowProcessor.ResultSupplier resultSupplierOneFinal = resultSupplierOne;
if (Utils.isPrefix(subtotalSpec, queryDimNames)) {
// Since subtotalSpec is a prefix of base query dimensions, so results from base query are also sorted
// by subtotalSpec as needed by stream merging.
subtotalsResults.add(processSubtotalsResultAndOptionallyClose(() -> resultSupplierOneFinal, subTotalDimensionSpec, subtotalQuery, false));
} else {
// Since subtotalSpec is not a prefix of base query dimensions, so results from base query are not sorted
// by subtotalSpec. So we first add the result of base query into another resultSupplier which are sorted
// by subtotalSpec and then stream merge them.
// Also note, we can't create the ResultSupplier eagerly here or as we don't want to eagerly allocate
// merge buffers for processing subtotal.
Supplier<GroupByRowProcessor.ResultSupplier> resultSupplierTwo = () -> GroupByRowProcessor.process(baseSubtotalQuery, subtotalQuery, resultSupplierOneFinal.results(subTotalDimensionSpec), configSupplier.get(), resource, spillMapper, processingConfig.getTmpDir(), processingConfig.intermediateComputeSizeBytes());
subtotalsResults.add(processSubtotalsResultAndOptionallyClose(resultSupplierTwo, subTotalDimensionSpec, subtotalQuery, true));
}
}
return Sequences.withBaggage(query.postProcess(Sequences.concat(subtotalsResults)), // this will close resources allocated by resultSupplierOne after sequence read
resultSupplierOne);
} catch (Throwable e) {
throw CloseableUtils.closeAndWrapInCatch(e, resultSupplierOne);
}
}
Aggregations