Examples with MergeSequence io.druid.java.util.common.guava.MergeSequence used on opensource projects

Example 1 with MergeSequence

use of io.druid.java.util.common.guava.MergeSequence in project druid by druid-io.

the class CachingClusteredClient method run.

@Override
public Sequence<T> run(final Query<T> query, final Map<String, Object> responseContext) {
    final QueryToolChest<T, Query<T>> toolChest = warehouse.getToolChest(query);
    final CacheStrategy<T, Object, Query<T>> strategy = toolChest.getCacheStrategy(query);
    final Map<DruidServer, List<SegmentDescriptor>> serverSegments = Maps.newTreeMap();
    final List<Pair<Interval, byte[]>> cachedResults = Lists.newArrayList();
    final Map<String, CachePopulator> cachePopulatorMap = Maps.newHashMap();
    final boolean useCache = CacheUtil.useCacheOnBrokers(query, strategy, cacheConfig);
    final boolean populateCache = CacheUtil.populateCacheOnBrokers(query, strategy, cacheConfig);
    final boolean isBySegment = BaseQuery.getContextBySegment(query, false);
    final ImmutableMap.Builder<String, Object> contextBuilder = new ImmutableMap.Builder<>();
    final int priority = BaseQuery.getContextPriority(query, 0);
    contextBuilder.put("priority", priority);
    if (populateCache) {
        // prevent down-stream nodes from caching results as well if we are populating the cache
        contextBuilder.put(CacheConfig.POPULATE_CACHE, false);
        contextBuilder.put("bySegment", true);
    }
    TimelineLookup<String, ServerSelector> timeline = serverView.getTimeline(query.getDataSource());
    if (timeline == null) {
        return Sequences.empty();
    }
    // build set of segments to query
    Set<Pair<ServerSelector, SegmentDescriptor>> segments = Sets.newLinkedHashSet();
    List<TimelineObjectHolder<String, ServerSelector>> serversLookup = Lists.newLinkedList();
    // Note that enabling this leads to putting uncovered intervals information in the response headers
    // and might blow up in some cases https://github.com/druid-io/druid/issues/2108
    int uncoveredIntervalsLimit = BaseQuery.getContextUncoveredIntervalsLimit(query, 0);
    if (uncoveredIntervalsLimit > 0) {
        List<Interval> uncoveredIntervals = Lists.newArrayListWithCapacity(uncoveredIntervalsLimit);
        boolean uncoveredIntervalsOverflowed = false;
        for (Interval interval : query.getIntervals()) {
            Iterable<TimelineObjectHolder<String, ServerSelector>> lookup = timeline.lookup(interval);
            long startMillis = interval.getStartMillis();
            long endMillis = interval.getEndMillis();
            for (TimelineObjectHolder<String, ServerSelector> holder : lookup) {
                Interval holderInterval = holder.getInterval();
                long intervalStart = holderInterval.getStartMillis();
                if (!uncoveredIntervalsOverflowed && startMillis != intervalStart) {
                    if (uncoveredIntervalsLimit > uncoveredIntervals.size()) {
                        uncoveredIntervals.add(new Interval(startMillis, intervalStart));
                    } else {
                        uncoveredIntervalsOverflowed = true;
                    }
                }
                startMillis = holderInterval.getEndMillis();
                serversLookup.add(holder);
            }
            if (!uncoveredIntervalsOverflowed && startMillis < endMillis) {
                if (uncoveredIntervalsLimit > uncoveredIntervals.size()) {
                    uncoveredIntervals.add(new Interval(startMillis, endMillis));
                } else {
                    uncoveredIntervalsOverflowed = true;
                }
            }
        }
        if (!uncoveredIntervals.isEmpty()) {
            // This returns intervals for which NO segment is present.
            // Which is not necessarily an indication that the data doesn't exist or is
            // incomplete. The data could exist and just not be loaded yet.  In either
            // case, though, this query will not include any data from the identified intervals.
            responseContext.put("uncoveredIntervals", uncoveredIntervals);
            responseContext.put("uncoveredIntervalsOverflowed", uncoveredIntervalsOverflowed);
        }
    } else {
        for (Interval interval : query.getIntervals()) {
            Iterables.addAll(serversLookup, timeline.lookup(interval));
        }
    }
    // Let tool chest filter out unneeded segments
    final List<TimelineObjectHolder<String, ServerSelector>> filteredServersLookup = toolChest.filterSegments(query, serversLookup);
    Map<String, Optional<RangeSet<String>>> dimensionRangeCache = Maps.newHashMap();
    // Filter unneeded chunks based on partition dimension
    for (TimelineObjectHolder<String, ServerSelector> holder : filteredServersLookup) {
        final Set<PartitionChunk<ServerSelector>> filteredChunks = DimFilterUtils.filterShards(query.getFilter(), holder.getObject(), new Function<PartitionChunk<ServerSelector>, ShardSpec>() {

            @Override
            public ShardSpec apply(PartitionChunk<ServerSelector> input) {
                return input.getObject().getSegment().getShardSpec();
            }
        }, dimensionRangeCache);
        for (PartitionChunk<ServerSelector> chunk : filteredChunks) {
            ServerSelector selector = chunk.getObject();
            final SegmentDescriptor descriptor = new SegmentDescriptor(holder.getInterval(), holder.getVersion(), chunk.getChunkNumber());
            segments.add(Pair.of(selector, descriptor));
        }
    }
    final byte[] queryCacheKey;
    if (// implies strategy != null
    (populateCache || useCache) && // explicit bySegment queries are never cached
    !isBySegment) {
        queryCacheKey = strategy.computeCacheKey(query);
    } else {
        queryCacheKey = null;
    }
    if (query.getContext().get(QueryResource.HDR_IF_NONE_MATCH) != null) {
        String prevEtag = (String) query.getContext().get(QueryResource.HDR_IF_NONE_MATCH);
        //compute current Etag
        Hasher hasher = Hashing.sha1().newHasher();
        boolean hasOnlyHistoricalSegments = true;
        for (Pair<ServerSelector, SegmentDescriptor> p : segments) {
            if (!p.lhs.pick().getServer().isAssignable()) {
                hasOnlyHistoricalSegments = false;
                break;
            }
            hasher.putString(p.lhs.getSegment().getIdentifier(), Charsets.UTF_8);
        }
        if (hasOnlyHistoricalSegments) {
            hasher.putBytes(queryCacheKey == null ? strategy.computeCacheKey(query) : queryCacheKey);
            String currEtag = Base64.encodeBase64String(hasher.hash().asBytes());
            responseContext.put(QueryResource.HDR_ETAG, currEtag);
            if (prevEtag.equals(currEtag)) {
                return Sequences.empty();
            }
        }
    }
    if (queryCacheKey != null) {
        // cachKeys map must preserve segment ordering, in order for shards to always be combined in the same order
        Map<Pair<ServerSelector, SegmentDescriptor>, Cache.NamedKey> cacheKeys = Maps.newLinkedHashMap();
        for (Pair<ServerSelector, SegmentDescriptor> segment : segments) {
            final Cache.NamedKey segmentCacheKey = CacheUtil.computeSegmentCacheKey(segment.lhs.getSegment().getIdentifier(), segment.rhs, queryCacheKey);
            cacheKeys.put(segment, segmentCacheKey);
        }
        // Pull cached segments from cache and remove from set of segments to query
        final Map<Cache.NamedKey, byte[]> cachedValues;
        if (useCache) {
            cachedValues = cache.getBulk(Iterables.limit(cacheKeys.values(), cacheConfig.getCacheBulkMergeLimit()));
        } else {
            cachedValues = ImmutableMap.of();
        }
        for (Map.Entry<Pair<ServerSelector, SegmentDescriptor>, Cache.NamedKey> entry : cacheKeys.entrySet()) {
            Pair<ServerSelector, SegmentDescriptor> segment = entry.getKey();
            Cache.NamedKey segmentCacheKey = entry.getValue();
            final Interval segmentQueryInterval = segment.rhs.getInterval();
            final byte[] cachedValue = cachedValues.get(segmentCacheKey);
            if (cachedValue != null) {
                // remove cached segment from set of segments to query
                segments.remove(segment);
                cachedResults.add(Pair.of(segmentQueryInterval, cachedValue));
            } else if (populateCache) {
                // otherwise, if populating cache, add segment to list of segments to cache
                final String segmentIdentifier = segment.lhs.getSegment().getIdentifier();
                cachePopulatorMap.put(String.format("%s_%s", segmentIdentifier, segmentQueryInterval), new CachePopulator(cache, objectMapper, segmentCacheKey));
            }
        }
    }
    // Compile list of all segments not pulled from cache
    for (Pair<ServerSelector, SegmentDescriptor> segment : segments) {
        final QueryableDruidServer queryableDruidServer = segment.lhs.pick();
        if (queryableDruidServer == null) {
            log.makeAlert("No servers found for SegmentDescriptor[%s] for DataSource[%s]?! How can this be?!", segment.rhs, query.getDataSource()).emit();
        } else {
            final DruidServer server = queryableDruidServer.getServer();
            List<SegmentDescriptor> descriptors = serverSegments.get(server);
            if (descriptors == null) {
                descriptors = Lists.newArrayList();
                serverSegments.put(server, descriptors);
            }
            descriptors.add(segment.rhs);
        }
    }
    return new LazySequence<>(new Supplier<Sequence<T>>() {

        @Override
        public Sequence<T> get() {
            ArrayList<Sequence<T>> sequencesByInterval = Lists.newArrayList();
            addSequencesFromCache(sequencesByInterval);
            addSequencesFromServer(sequencesByInterval);
            return mergeCachedAndUncachedSequences(query, sequencesByInterval);
        }

        private void addSequencesFromCache(ArrayList<Sequence<T>> listOfSequences) {
            if (strategy == null) {
                return;
            }
            final Function<Object, T> pullFromCacheFunction = strategy.pullFromCache();
            final TypeReference<Object> cacheObjectClazz = strategy.getCacheObjectClazz();
            for (Pair<Interval, byte[]> cachedResultPair : cachedResults) {
                final byte[] cachedResult = cachedResultPair.rhs;
                Sequence<Object> cachedSequence = new BaseSequence<>(new BaseSequence.IteratorMaker<Object, Iterator<Object>>() {

                    @Override
                    public Iterator<Object> make() {
                        try {
                            if (cachedResult.length == 0) {
                                return Iterators.emptyIterator();
                            }
                            return objectMapper.readValues(objectMapper.getFactory().createParser(cachedResult), cacheObjectClazz);
                        } catch (IOException e) {
                            throw Throwables.propagate(e);
                        }
                    }

                    @Override
                    public void cleanup(Iterator<Object> iterFromMake) {
                    }
                });
                listOfSequences.add(Sequences.map(cachedSequence, pullFromCacheFunction));
            }
        }

        private void addSequencesFromServer(ArrayList<Sequence<T>> listOfSequences) {
            listOfSequences.ensureCapacity(listOfSequences.size() + serverSegments.size());
            final Query<T> rewrittenQuery = query.withOverriddenContext(contextBuilder.build());
            // The data gets handled as a Future and parsed in the long Sequence chain in the resultSeqToAdd setter.
            for (Map.Entry<DruidServer, List<SegmentDescriptor>> entry : serverSegments.entrySet()) {
                final DruidServer server = entry.getKey();
                final List<SegmentDescriptor> descriptors = entry.getValue();
                final QueryRunner clientQueryable = serverView.getQueryRunner(server);
                if (clientQueryable == null) {
                    log.error("WTF!? server[%s] doesn't have a client Queryable?", server);
                    continue;
                }
                final MultipleSpecificSegmentSpec segmentSpec = new MultipleSpecificSegmentSpec(descriptors);
                final Sequence<T> resultSeqToAdd;
                if (!server.isAssignable() || !populateCache || isBySegment) {
                    // Direct server queryable
                    if (!isBySegment) {
                        resultSeqToAdd = clientQueryable.run(query.withQuerySegmentSpec(segmentSpec), responseContext);
                    } else {
                        // bySegment queries need to be de-serialized, see DirectDruidClient.run()
                        @SuppressWarnings("unchecked") final Query<Result<BySegmentResultValueClass<T>>> bySegmentQuery = (Query<Result<BySegmentResultValueClass<T>>>) ((Query) query);
                        @SuppressWarnings("unchecked") final Sequence<Result<BySegmentResultValueClass<T>>> resultSequence = clientQueryable.run(bySegmentQuery.withQuerySegmentSpec(segmentSpec), responseContext);
                        resultSeqToAdd = (Sequence) Sequences.map(resultSequence, new Function<Result<BySegmentResultValueClass<T>>, Result<BySegmentResultValueClass<T>>>() {

                            @Override
                            public Result<BySegmentResultValueClass<T>> apply(Result<BySegmentResultValueClass<T>> input) {
                                final BySegmentResultValueClass<T> bySegmentValue = input.getValue();
                                return new Result<>(input.getTimestamp(), new BySegmentResultValueClass<T>(Lists.transform(bySegmentValue.getResults(), toolChest.makePreComputeManipulatorFn(query, MetricManipulatorFns.deserializing())), bySegmentValue.getSegmentId(), bySegmentValue.getInterval()));
                            }
                        });
                    }
                } else {
                    // Requires some manipulation on broker side
                    @SuppressWarnings("unchecked") final Sequence<Result<BySegmentResultValueClass<T>>> runningSequence = clientQueryable.run(rewrittenQuery.withQuerySegmentSpec(segmentSpec), responseContext);
                    resultSeqToAdd = new MergeSequence(query.getResultOrdering(), Sequences.<Result<BySegmentResultValueClass<T>>, Sequence<T>>map(runningSequence, new Function<Result<BySegmentResultValueClass<T>>, Sequence<T>>() {

                        private final Function<T, Object> cacheFn = strategy.prepareForCache();

                        // Acctually do something with the results
                        @Override
                        public Sequence<T> apply(Result<BySegmentResultValueClass<T>> input) {
                            final BySegmentResultValueClass<T> value = input.getValue();
                            final CachePopulator cachePopulator = cachePopulatorMap.get(String.format("%s_%s", value.getSegmentId(), value.getInterval()));
                            final Queue<ListenableFuture<Object>> cacheFutures = new ConcurrentLinkedQueue<>();
                            return Sequences.<T>withEffect(Sequences.<T, T>map(Sequences.<T, T>map(Sequences.<T>simple(value.getResults()), new Function<T, T>() {

                                @Override
                                public T apply(final T input) {
                                    if (cachePopulator != null) {
                                        // only compute cache data if populating cache
                                        cacheFutures.add(backgroundExecutorService.submit(new Callable<Object>() {

                                            @Override
                                            public Object call() {
                                                return cacheFn.apply(input);
                                            }
                                        }));
                                    }
                                    return input;
                                }
                            }), toolChest.makePreComputeManipulatorFn(// This casting is sub-optimal, but hasn't caused any major problems yet...
                            (Query) rewrittenQuery, MetricManipulatorFns.deserializing())), new Runnable() {

                                @Override
                                public void run() {
                                    if (cachePopulator != null) {
                                        Futures.addCallback(Futures.allAsList(cacheFutures), new FutureCallback<List<Object>>() {

                                            @Override
                                            public void onSuccess(List<Object> cacheData) {
                                                cachePopulator.populate(cacheData);
                                                // Help out GC by making sure all references are gone
                                                cacheFutures.clear();
                                            }

                                            @Override
                                            public void onFailure(Throwable throwable) {
                                                log.error(throwable, "Background caching failed");
                                            }
                                        }, backgroundExecutorService);
                                    }
                                }
                            }, MoreExecutors.sameThreadExecutor());
                        // End withEffect
                        }
                    }));
                }
                listOfSequences.add(resultSeqToAdd);
            }
        }
    });
}

Example 2 with MergeSequence

use of io.druid.java.util.common.guava.MergeSequence in project druid by druid-io.

the class RetryQueryRunner method run.

@Override
public Sequence<T> run(final Query<T> query, final Map<String, Object> context) {
    final List<Sequence<T>> listOfSequences = Lists.newArrayList();
    listOfSequences.add(baseRunner.run(query, context));
    return new YieldingSequenceBase<T>() {

        @Override
        public <OutType> Yielder<OutType> toYielder(OutType initValue, YieldingAccumulator<OutType, T> accumulator) {
            List<SegmentDescriptor> missingSegments = getMissingSegments(context);
            if (!missingSegments.isEmpty()) {
                for (int i = 0; i < config.getNumTries(); i++) {
                    log.info("[%,d] missing segments found. Retry attempt [%,d]", missingSegments.size(), i);
                    context.put(Result.MISSING_SEGMENTS_KEY, Lists.newArrayList());
                    final Query<T> retryQuery = query.withQuerySegmentSpec(new MultipleSpecificSegmentSpec(missingSegments));
                    Sequence<T> retrySequence = baseRunner.run(retryQuery, context);
                    listOfSequences.add(retrySequence);
                    missingSegments = getMissingSegments(context);
                    if (missingSegments.isEmpty()) {
                        break;
                    }
                }
                final List<SegmentDescriptor> finalMissingSegs = getMissingSegments(context);
                if (!config.isReturnPartialResults() && !finalMissingSegs.isEmpty()) {
                    throw new SegmentMissingException("No results found for segments[%s]", finalMissingSegs);
                }
                return new MergeSequence<>(query.getResultOrdering(), Sequences.simple(listOfSequences)).toYielder(initValue, accumulator);
            } else {
                return Iterables.getOnlyElement(listOfSequences).toYielder(initValue, accumulator);
            }
        }
    };
}

Example 3 with MergeSequence

use of io.druid.java.util.common.guava.MergeSequence in project druid by druid-io.

the class GroupByQueryRunnerFactoryTest method testMergeRunnersEnsureGroupMerging.

@Test
public void testMergeRunnersEnsureGroupMerging() throws Exception {
    GroupByQuery query = GroupByQuery.builder().setDataSource("xx").setQuerySegmentSpec(new LegacySegmentSpec("1970/3000")).setGranularity(Granularities.ALL).setDimensions(Lists.<DimensionSpec>newArrayList(new DefaultDimensionSpec("tags", "tags"))).setAggregatorSpecs(Arrays.asList(new AggregatorFactory[] { new CountAggregatorFactory("count") })).build();
    final QueryRunnerFactory factory = GroupByQueryRunnerTest.makeQueryRunnerFactory(new GroupByQueryConfig());
    QueryRunner mergedRunner = factory.getToolchest().mergeResults(new QueryRunner() {

        @Override
        public Sequence run(Query query, Map responseContext) {
            return factory.getToolchest().mergeResults(new QueryRunner() {

                @Override
                public Sequence run(Query query, Map responseContext) {
                    try {
                        return new MergeSequence(query.getResultOrdering(), Sequences.simple(Arrays.asList(factory.createRunner(createSegment()).run(query, responseContext), factory.createRunner(createSegment()).run(query, responseContext))));
                    } catch (Exception e) {
                        Throwables.propagate(e);
                        return null;
                    }
                }
            }).run(query, responseContext);
        }
    });
    Sequence<Row> result = mergedRunner.run(query, Maps.newHashMap());
    List<Row> expectedResults = Arrays.asList(GroupByQueryRunnerTestHelper.createExpectedRow("1970-01-01T00:00:00.000Z", "tags", "t1", "count", 2L), GroupByQueryRunnerTestHelper.createExpectedRow("1970-01-01T00:00:00.000Z", "tags", "t2", "count", 4L));
    TestHelper.assertExpectedObjects(expectedResults, Sequences.toList(result, new ArrayList<Row>()), "");
}

Example 4 with MergeSequence

use of io.druid.java.util.common.guava.MergeSequence in project druid by druid-io.

the class GroupByQueryRunnerTest method testMergedHavingSpec.

@Test
public void testMergedHavingSpec() {
    List<Row> expectedResults = Arrays.asList(GroupByQueryRunnerTestHelper.createExpectedRow("2011-04-01", "alias", "business", "rows", 2L, "idx", 217L), GroupByQueryRunnerTestHelper.createExpectedRow("2011-04-01", "alias", "mezzanine", "rows", 6L, "idx", 4420L), GroupByQueryRunnerTestHelper.createExpectedRow("2011-04-01", "alias", "premium", "rows", 6L, "idx", 4416L));
    GroupByQuery.Builder builder = GroupByQuery.builder().setDataSource(QueryRunnerTestHelper.dataSource).setInterval("2011-04-02/2011-04-04").setDimensions(Lists.<DimensionSpec>newArrayList(new DefaultDimensionSpec("quality", "alias"))).setAggregatorSpecs(Arrays.asList(QueryRunnerTestHelper.rowsCount, new LongSumAggregatorFactory("idx", "index"))).setGranularity(new PeriodGranularity(new Period("P1M"), null, null)).setHavingSpec(new OrHavingSpec(ImmutableList.<HavingSpec>of(new GreaterThanHavingSpec("rows", 2L), new EqualToHavingSpec("idx", 217L))));
    GroupByQuery fullQuery = builder.build();
    QueryRunner mergedRunner = factory.getToolchest().mergeResults(new QueryRunner<Row>() {

        @Override
        public Sequence<Row> run(Query<Row> query, Map<String, Object> responseContext) {
            // simulate two daily segments
            final Query query1 = query.withQuerySegmentSpec(new MultipleIntervalSegmentSpec(Lists.newArrayList(new Interval("2011-04-02/2011-04-03"))));
            final Query query2 = query.withQuerySegmentSpec(new MultipleIntervalSegmentSpec(Lists.newArrayList(new Interval("2011-04-03/2011-04-04"))));
            return new MergeSequence(query.getResultOrdering(), Sequences.simple(Arrays.asList(runner.run(query1, responseContext), runner.run(query2, responseContext))));
        }
    });
    Map<String, Object> context = Maps.newHashMap();
    TestHelper.assertExpectedObjects(expectedResults, mergedRunner.run(fullQuery, context), "merged");
}

Example 5 with MergeSequence

use of io.druid.java.util.common.guava.MergeSequence in project druid by druid-io.

the class GroupByQueryRunnerTest method doTestMergeResultsWithOrderBy.

private void doTestMergeResultsWithOrderBy(LimitSpec orderBySpec, List<Row> expectedResults) {
    GroupByQuery.Builder builder = GroupByQuery.builder().setDataSource(QueryRunnerTestHelper.dataSource).setInterval("2011-04-02/2011-04-04").setDimensions(Lists.<DimensionSpec>newArrayList(new DefaultDimensionSpec("quality", "alias"))).setAggregatorSpecs(Arrays.asList(QueryRunnerTestHelper.rowsCount, new LongSumAggregatorFactory("idx", "index"))).setGranularity(new PeriodGranularity(new Period("P1M"), null, null)).setLimitSpec(orderBySpec);
    final GroupByQuery fullQuery = builder.build();
    QueryRunner mergedRunner = factory.getToolchest().mergeResults(new QueryRunner<Row>() {

        @Override
        public Sequence<Row> run(Query<Row> query, Map<String, Object> responseContext) {
            // simulate two daily segments
            final Query query1 = query.withQuerySegmentSpec(new MultipleIntervalSegmentSpec(Lists.newArrayList(new Interval("2011-04-02/2011-04-03"))));
            final Query query2 = query.withQuerySegmentSpec(new MultipleIntervalSegmentSpec(Lists.newArrayList(new Interval("2011-04-03/2011-04-04"))));
            return new MergeSequence(query.getResultOrdering(), Sequences.simple(Arrays.asList(runner.run(query1, responseContext), runner.run(query2, responseContext))));
        }
    });
    Map<String, Object> context = Maps.newHashMap();
    TestHelper.assertExpectedObjects(expectedResults, mergedRunner.run(fullQuery, context), "merged");
}