Examples with GroupByQuery org.apache.druid.query.groupby.GroupByQuery used on opensource projects

Example 41 with GroupByQuery

use of org.apache.druid.query.groupby.GroupByQuery in project druid by druid-io.

the class RowBasedGrouperHelper method createResultRowBasedColumnSelectorFactory.

/**
 * Creates a {@link ColumnSelectorFactory} that can read rows which originate as results of the provided "query".
 *
 * @param query        a groupBy query
 * @param supplier     supplier of result rows from the query
 * @param finalization whether the column capabilities reported by this factory should reflect finalized types
 */
public static ColumnSelectorFactory createResultRowBasedColumnSelectorFactory(final GroupByQuery query, final Supplier<ResultRow> supplier, final RowSignature.Finalization finalization) {
    final RowSignature signature = query.getResultRowSignature(finalization);
    final RowAdapter<ResultRow> adapter = new RowAdapter<ResultRow>() {

        @Override
        public ToLongFunction<ResultRow> timestampFunction() {
            if (query.getResultRowHasTimestamp()) {
                return row -> row.getLong(0);
            } else {
                final long timestamp = query.getUniversalTimestamp().getMillis();
                return row -> timestamp;
            }
        }

        @Override
        public Function<ResultRow, Object> columnFunction(final String columnName) {
            final int columnIndex = signature.indexOf(columnName);
            if (columnIndex < 0) {
                return row -> null;
            } else {
                return row -> row.get(columnIndex);
            }
        }
    };
    // Decorate "signature" so that it returns hasMultipleValues = false. (groupBy does not return multiple values.)
    final ColumnInspector decoratedSignature = new ColumnInspector() {

        @Nullable
        @Override
        public ColumnCapabilities getColumnCapabilities(String column) {
            final ColumnCapabilities baseCapabilities = signature.getColumnCapabilities(column);
            if (baseCapabilities == null || baseCapabilities.hasMultipleValues().isFalse()) {
                return baseCapabilities;
            } else {
                return ColumnCapabilitiesImpl.copyOf(baseCapabilities).setHasMultipleValues(false);
            }
        }
    };
    return RowBasedColumnSelectorFactory.create(adapter, supplier::get, decoratedSignature, false);
}

Example 42 with GroupByQuery

use of org.apache.druid.query.groupby.GroupByQuery in project druid by druid-io.

the class GroupByMergingQueryRunnerV2 method run.

@Override
public Sequence<ResultRow> run(final QueryPlus<ResultRow> queryPlus, final ResponseContext responseContext) {
    final GroupByQuery query = (GroupByQuery) queryPlus.getQuery();
    final GroupByQueryConfig querySpecificConfig = config.withOverrides(query);
    // CTX_KEY_MERGE_RUNNERS_USING_CHAINED_EXECUTION is here because realtime servers use nested mergeRunners calls
    // (one for the entire query and one for each sink). We only want the outer call to actually do merging with a
    // merge buffer, otherwise the query will allocate too many merge buffers. This is potentially sub-optimal as it
    // will involve materializing the results for each sink before starting to feed them into the outer merge buffer.
    // I'm not sure of a better way to do this without tweaking how realtime servers do queries.
    final boolean forceChainedExecution = query.getContextBoolean(CTX_KEY_MERGE_RUNNERS_USING_CHAINED_EXECUTION, false);
    final QueryPlus<ResultRow> queryPlusForRunners = queryPlus.withQuery(query.withOverriddenContext(ImmutableMap.of(CTX_KEY_MERGE_RUNNERS_USING_CHAINED_EXECUTION, true))).withoutThreadUnsafeState();
    if (QueryContexts.isBySegment(query) || forceChainedExecution) {
        ChainedExecutionQueryRunner<ResultRow> runner = new ChainedExecutionQueryRunner<>(queryProcessingPool, queryWatcher, queryables);
        return runner.run(queryPlusForRunners, responseContext);
    }
    final boolean isSingleThreaded = querySpecificConfig.isSingleThreaded();
    final File temporaryStorageDirectory = new File(processingTmpDir, StringUtils.format("druid-groupBy-%s_%s", UUID.randomUUID(), query.getId()));
    final int priority = QueryContexts.getPriority(query);
    // Figure out timeoutAt time now, so we can apply the timeout to both the mergeBufferPool.take and the actual
    // query processing together.
    final long queryTimeout = QueryContexts.getTimeout(query);
    final boolean hasTimeout = QueryContexts.hasTimeout(query);
    final long timeoutAt = System.currentTimeMillis() + queryTimeout;
    return new BaseSequence<>(new BaseSequence.IteratorMaker<ResultRow, CloseableGrouperIterator<RowBasedKey, ResultRow>>() {

        @Override
        public CloseableGrouperIterator<RowBasedKey, ResultRow> make() {
            final Closer resources = Closer.create();
            try {
                final LimitedTemporaryStorage temporaryStorage = new LimitedTemporaryStorage(temporaryStorageDirectory, querySpecificConfig.getMaxOnDiskStorage());
                final ReferenceCountingResourceHolder<LimitedTemporaryStorage> temporaryStorageHolder = ReferenceCountingResourceHolder.fromCloseable(temporaryStorage);
                resources.register(temporaryStorageHolder);
                // If parallelCombine is enabled, we need two merge buffers for parallel aggregating and parallel combining
                final int numMergeBuffers = querySpecificConfig.getNumParallelCombineThreads() > 1 ? 2 : 1;
                final List<ReferenceCountingResourceHolder<ByteBuffer>> mergeBufferHolders = getMergeBuffersHolder(numMergeBuffers, hasTimeout, timeoutAt);
                resources.registerAll(mergeBufferHolders);
                final ReferenceCountingResourceHolder<ByteBuffer> mergeBufferHolder = mergeBufferHolders.get(0);
                final ReferenceCountingResourceHolder<ByteBuffer> combineBufferHolder = numMergeBuffers == 2 ? mergeBufferHolders.get(1) : null;
                Pair<Grouper<RowBasedKey>, Accumulator<AggregateResult, ResultRow>> pair = RowBasedGrouperHelper.createGrouperAccumulatorPair(query, null, config, Suppliers.ofInstance(mergeBufferHolder.get()), combineBufferHolder, concurrencyHint, temporaryStorage, spillMapper, // Passed as executor service
                queryProcessingPool, priority, hasTimeout, timeoutAt, mergeBufferSize);
                final Grouper<RowBasedKey> grouper = pair.lhs;
                final Accumulator<AggregateResult, ResultRow> accumulator = pair.rhs;
                grouper.init();
                final ReferenceCountingResourceHolder<Grouper<RowBasedKey>> grouperHolder = ReferenceCountingResourceHolder.fromCloseable(grouper);
                resources.register(grouperHolder);
                List<ListenableFuture<AggregateResult>> futures = Lists.newArrayList(Iterables.transform(queryables, new Function<QueryRunner<ResultRow>, ListenableFuture<AggregateResult>>() {

                    @Override
                    public ListenableFuture<AggregateResult> apply(final QueryRunner<ResultRow> input) {
                        if (input == null) {
                            throw new ISE("Null queryRunner! Looks to be some segment unmapping action happening");
                        }
                        ListenableFuture<AggregateResult> future = queryProcessingPool.submitRunnerTask(new AbstractPrioritizedQueryRunnerCallable<AggregateResult, ResultRow>(priority, input) {

                            @Override
                            public AggregateResult call() {
                                try (// These variables are used to close releasers automatically.
                                @SuppressWarnings("unused") Releaser bufferReleaser = mergeBufferHolder.increment();
                                    @SuppressWarnings("unused") Releaser grouperReleaser = grouperHolder.increment()) {
                                    // Return true if OK, false if resources were exhausted.
                                    return input.run(queryPlusForRunners, responseContext).accumulate(AggregateResult.ok(), accumulator);
                                } catch (QueryInterruptedException | QueryTimeoutException e) {
                                    throw e;
                                } catch (Exception e) {
                                    log.error(e, "Exception with one of the sequences!");
                                    throw new RuntimeException(e);
                                }
                            }
                        });
                        if (isSingleThreaded) {
                            waitForFutureCompletion(query, ImmutableList.of(future), hasTimeout, timeoutAt - System.currentTimeMillis());
                        }
                        return future;
                    }
                }));
                if (!isSingleThreaded) {
                    waitForFutureCompletion(query, futures, hasTimeout, timeoutAt - System.currentTimeMillis());
                }
                return RowBasedGrouperHelper.makeGrouperIterator(grouper, query, resources);
            } catch (Throwable t) {
                // Exception caught while setting up the iterator; release resources.
                try {
                    resources.close();
                } catch (Exception ex) {
                    t.addSuppressed(ex);
                }
                throw t;
            }
        }

        @Override
        public void cleanup(CloseableGrouperIterator<RowBasedKey, ResultRow> iterFromMake) {
            iterFromMake.close();
        }
    });
}

Example 43 with GroupByQuery

use of org.apache.druid.query.groupby.GroupByQuery in project druid by druid-io.

the class GroupByStrategyV1 method processSubqueryResult.

@Override
public Sequence<ResultRow> processSubqueryResult(GroupByQuery subquery, GroupByQuery query, GroupByQueryResource resource, Sequence<ResultRow> subqueryResult, boolean wasQueryPushedDown) {
    final Set<AggregatorFactory> aggs = new HashSet<>();
    // Nested group-bys work by first running the inner query and then materializing the results in an incremental
    // index which the outer query is then run against. To build the incremental index, we use the fieldNames from
    // the aggregators for the outer query to define the column names so that the index will match the query. If
    // there are multiple types of aggregators in the outer query referencing the same fieldName, we will try to build
    // multiple columns of the same name using different aggregator types and will fail. Here, we permit multiple
    // aggregators of the same type referencing the same fieldName (and skip creating identical columns for the
    // subsequent ones) and return an error if the aggregator types are different.
    final Set<String> dimensionNames = new HashSet<>();
    for (DimensionSpec dimension : subquery.getDimensions()) {
        dimensionNames.add(dimension.getOutputName());
    }
    for (AggregatorFactory aggregatorFactory : query.getAggregatorSpecs()) {
        for (final AggregatorFactory transferAgg : aggregatorFactory.getRequiredColumns()) {
            if (dimensionNames.contains(transferAgg.getName())) {
                // doesn't have this problem.
                continue;
            }
            if (Iterables.any(aggs, new Predicate<AggregatorFactory>() {

                @Override
                public boolean apply(AggregatorFactory agg) {
                    return agg.getName().equals(transferAgg.getName()) && !agg.equals(transferAgg);
                }
            })) {
                throw new IAE("Inner aggregator can currently only be referenced by a single type of outer aggregator" + " for '%s'", transferAgg.getName());
            }
            aggs.add(transferAgg);
        }
    }
    // We need the inner incremental index to have all the columns required by the outer query
    final GroupByQuery innerQuery = new GroupByQuery.Builder(subquery).setAggregatorSpecs(ImmutableList.copyOf(aggs)).setInterval(subquery.getIntervals()).setPostAggregatorSpecs(new ArrayList<>()).build();
    final GroupByQuery outerQuery = new GroupByQuery.Builder(query).setLimitSpec(query.getLimitSpec().merge(subquery.getLimitSpec())).build();
    final IncrementalIndex innerQueryResultIndex = GroupByQueryHelper.makeIncrementalIndex(innerQuery.withOverriddenContext(ImmutableMap.of(GroupByQueryHelper.CTX_KEY_SORT_RESULTS, true)), subquery, configSupplier.get(), subqueryResult);
    // Outer query might have multiple intervals, but they are expected to be non-overlapping and sorted which
    // is ensured by QuerySegmentSpec.
    // GroupByQueryEngine can only process one interval at a time, so we need to call it once per interval
    // and concatenate the results.
    final IncrementalIndex outerQueryResultIndex = GroupByQueryHelper.makeIncrementalIndex(outerQuery, null, configSupplier.get(), Sequences.concat(Sequences.map(Sequences.simple(outerQuery.getIntervals()), new Function<Interval, Sequence<ResultRow>>() {

        @Override
        public Sequence<ResultRow> apply(Interval interval) {
            return process(outerQuery.withQuerySegmentSpec(new MultipleIntervalSegmentSpec(ImmutableList.of(interval))), new IncrementalIndexStorageAdapter(innerQueryResultIndex));
        }
    })));
    innerQueryResultIndex.close();
    return Sequences.withBaggage(outerQuery.postProcess(GroupByQueryHelper.postAggregate(query, outerQueryResultIndex)), outerQueryResultIndex);
}

Example 44 with GroupByQuery

use of org.apache.druid.query.groupby.GroupByQuery in project druid by druid-io.

the class GroupByStrategyV2 method processSubqueryResult.

@Override
public Sequence<ResultRow> processSubqueryResult(GroupByQuery subquery, GroupByQuery query, GroupByQueryResource resource, Sequence<ResultRow> subqueryResult, boolean wasQueryPushedDown) {
    // Keep a reference to resultSupplier outside the "try" so we can close it if something goes wrong
    // while creating the sequence.
    GroupByRowProcessor.ResultSupplier resultSupplier = null;
    try {
        final GroupByQuery queryToRun;
        if (wasQueryPushedDown) {
            // If the query was pushed down, filters would have been applied downstream, so skip it here.
            queryToRun = query.withDimFilter(null).withQuerySegmentSpec(new MultipleIntervalSegmentSpec(Intervals.ONLY_ETERNITY));
        } else {
            queryToRun = query;
        }
        resultSupplier = GroupByRowProcessor.process(queryToRun, wasQueryPushedDown ? queryToRun : subquery, subqueryResult, configSupplier.get(), resource, spillMapper, processingConfig.getTmpDir(), processingConfig.intermediateComputeSizeBytes());
        final GroupByRowProcessor.ResultSupplier finalResultSupplier = resultSupplier;
        return Sequences.withBaggage(mergeResults((queryPlus, responseContext) -> finalResultSupplier.results(null), query, null), finalResultSupplier);
    } catch (Throwable e) {
        throw CloseableUtils.closeAndWrapInCatch(e, resultSupplier);
    }
}

Example 45 with GroupByQuery

use of org.apache.druid.query.groupby.GroupByQuery in project druid by druid-io.

the class DefaultLimitSpec method build.

@Override
public Function<Sequence<ResultRow>, Sequence<ResultRow>> build(final GroupByQuery query) {
    final List<DimensionSpec> dimensions = query.getDimensions();
    // Can avoid re-sorting if the natural ordering is good enough.
    boolean sortingNeeded = dimensions.size() < columns.size();
    final Set<String> aggAndPostAggNames = new HashSet<>();
    for (AggregatorFactory agg : query.getAggregatorSpecs()) {
        aggAndPostAggNames.add(agg.getName());
    }
    for (PostAggregator postAgg : query.getPostAggregatorSpecs()) {
        aggAndPostAggNames.add(postAgg.getName());
    }
    if (!sortingNeeded) {
        for (int i = 0; i < columns.size(); i++) {
            final OrderByColumnSpec columnSpec = columns.get(i);
            if (aggAndPostAggNames.contains(columnSpec.getDimension())) {
                sortingNeeded = true;
                break;
            }
            final ColumnType columnType = getOrderByType(columnSpec, dimensions);
            final StringComparator naturalComparator;
            if (columnType.is(ValueType.STRING)) {
                naturalComparator = StringComparators.LEXICOGRAPHIC;
            } else if (columnType.isNumeric()) {
                naturalComparator = StringComparators.NUMERIC;
            } else if (columnType.isArray()) {
                if (columnType.getElementType().isNumeric()) {
                    naturalComparator = StringComparators.NUMERIC;
                } else {
                    naturalComparator = StringComparators.LEXICOGRAPHIC;
                }
            } else {
                sortingNeeded = true;
                break;
            }
            if (columnSpec.getDirection() != OrderByColumnSpec.Direction.ASCENDING || !columnSpec.getDimensionComparator().equals(naturalComparator) || !columnSpec.getDimension().equals(dimensions.get(i).getOutputName())) {
                sortingNeeded = true;
                break;
            }
        }
    }
    if (!sortingNeeded) {
        // If granularity is ALL, sortByDimsFirst doesn't change the sorting order.
        sortingNeeded = !query.getGranularity().equals(Granularities.ALL) && query.getContextSortByDimsFirst();
    }
    if (!sortingNeeded) {
        String timestampField = query.getContextValue(GroupByQuery.CTX_TIMESTAMP_RESULT_FIELD);
        if (timestampField != null && !timestampField.isEmpty()) {
            int timestampResultFieldIndex = query.getContextValue(GroupByQuery.CTX_TIMESTAMP_RESULT_FIELD_INDEX);
            sortingNeeded = query.getContextSortByDimsFirst() ? timestampResultFieldIndex != query.getDimensions().size() - 1 : timestampResultFieldIndex != 0;
        }
    }
    final Function<Sequence<ResultRow>, Sequence<ResultRow>> sortAndLimitFn;
    if (sortingNeeded) {
        // Materialize the Comparator first for fast-fail error checking.
        final Ordering<ResultRow> ordering = makeComparator(query.getResultRowSignature(), query.getResultRowHasTimestamp(), query.getDimensions(), query.getAggregatorSpecs(), query.getPostAggregatorSpecs(), query.getContextSortByDimsFirst());
        // underlying data isn't changing. (Useful for query reproducibility and offset-based pagination.)
        if (isLimited()) {
            sortAndLimitFn = results -> new TopNSequence<>(results, ordering, limit + offset);
        } else {
            sortAndLimitFn = results -> Sequences.sort(results, ordering).limit(limit + offset);
        }
    } else {
        if (isLimited()) {
            sortAndLimitFn = results -> results.limit(limit + offset);
        } else {
            sortAndLimitFn = Functions.identity();
        }
    }
    // Finally, apply offset after sorting and limiting.
    if (isOffset()) {
        return results -> sortAndLimitFn.apply(results).skip(offset);
    } else {
        return sortAndLimitFn;
    }
}