Examples with Function com.google.common.base.Function used on opensource projects

Example 51 with Function

use of com.google.common.base.Function 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 52 with Function

use of com.google.common.base.Function in project druid by druid-io.

the class CachingQueryRunner method run.

@Override
public Sequence<T> run(Query<T> query, Map<String, Object> responseContext) {
    final CacheStrategy strategy = toolChest.getCacheStrategy(query);
    final boolean populateCache = CacheUtil.populateCacheOnDataNodes(query, strategy, cacheConfig);
    final boolean useCache = CacheUtil.useCacheOnDataNodes(query, strategy, cacheConfig);
    final Cache.NamedKey key;
    if (strategy != null && (useCache || populateCache)) {
        key = CacheUtil.computeSegmentCacheKey(segmentIdentifier, segmentDescriptor, strategy.computeCacheKey(query));
    } else {
        key = null;
    }
    if (useCache) {
        final Function cacheFn = strategy.pullFromCache();
        final byte[] cachedResult = cache.get(key);
        if (cachedResult != null) {
            final TypeReference cacheObjectClazz = strategy.getCacheObjectClazz();
            return Sequences.map(new BaseSequence<>(new BaseSequence.IteratorMaker<T, Iterator<T>>() {

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

                @Override
                public void cleanup(Iterator<T> iterFromMake) {
                }
            }), cacheFn);
        }
    }
    final Collection<ListenableFuture<?>> cacheFutures = Collections.synchronizedList(Lists.<ListenableFuture<?>>newLinkedList());
    if (populateCache) {
        final Function cacheFn = strategy.prepareForCache();
        return Sequences.withEffect(Sequences.map(base.run(query, responseContext), new Function<T, T>() {

            @Override
            public T apply(final T input) {
                final SettableFuture<Object> future = SettableFuture.create();
                cacheFutures.add(future);
                backgroundExecutorService.submit(new Runnable() {

                    @Override
                    public void run() {
                        try {
                            future.set(cacheFn.apply(input));
                        } catch (Exception e) {
                            // if there is exception, should setException to quit the caching processing
                            future.setException(e);
                        }
                    }
                });
                return input;
            }
        }), new Runnable() {

            @Override
            public void run() {
                try {
                    CacheUtil.populate(cache, mapper, key, Futures.allAsList(cacheFutures).get());
                } catch (Exception e) {
                    log.error(e, "Error while getting future for cache task");
                    throw Throwables.propagate(e);
                }
            }
        }, backgroundExecutorService);
    } else {
        return base.run(query, responseContext);
    }
}

Example 53 with Function

use of com.google.common.base.Function in project druid by druid-io.

the class AppenderatorPlumber method mergeAndPush.

private void mergeAndPush() {
    final Granularity segmentGranularity = schema.getGranularitySpec().getSegmentGranularity();
    final Period windowPeriod = config.getWindowPeriod();
    final long windowMillis = windowPeriod.toStandardDuration().getMillis();
    log.info("Starting merge and push.");
    DateTime minTimestampAsDate = segmentGranularity.bucketStart(new DateTime(Math.max(windowMillis, rejectionPolicy.getCurrMaxTime().getMillis()) - windowMillis));
    long minTimestamp = minTimestampAsDate.getMillis();
    final List<SegmentIdentifier> appenderatorSegments = appenderator.getSegments();
    final List<SegmentIdentifier> segmentsToPush = Lists.newArrayList();
    if (shuttingDown) {
        log.info("Found [%,d] segments. Attempting to hand off all of them.", appenderatorSegments.size());
        segmentsToPush.addAll(appenderatorSegments);
    } else {
        log.info("Found [%,d] segments. Attempting to hand off segments that start before [%s].", appenderatorSegments.size(), minTimestampAsDate);
        for (SegmentIdentifier segment : appenderatorSegments) {
            final Long intervalStart = segment.getInterval().getStartMillis();
            if (intervalStart < minTimestamp) {
                log.info("Adding entry [%s] for merge and push.", segment);
                segmentsToPush.add(segment);
            } else {
                log.info("Skipping persist and merge for entry [%s] : Start time [%s] >= [%s] min timestamp required in this run. Segment will be picked up in a future run.", segment, new DateTime(intervalStart), minTimestampAsDate);
            }
        }
    }
    log.info("Found [%,d] segments to persist and merge", segmentsToPush.size());
    final Function<Throwable, Void> errorHandler = new Function<Throwable, Void>() {

        @Override
        public Void apply(Throwable throwable) {
            final List<String> segmentIdentifierStrings = Lists.transform(segmentsToPush, new Function<SegmentIdentifier, String>() {

                @Override
                public String apply(SegmentIdentifier input) {
                    return input.getIdentifierAsString();
                }
            });
            log.makeAlert(throwable, "Failed to publish merged indexes[%s]", schema.getDataSource()).addData("segments", segmentIdentifierStrings).emit();
            if (shuttingDown) {
                // We're trying to shut down, and these segments failed to push. Let's just get rid of them.
                // This call will also delete possibly-partially-written files, so we don't need to do it explicitly.
                cleanShutdown = false;
                for (SegmentIdentifier identifier : segmentsToPush) {
                    dropSegment(identifier);
                }
            }
            return null;
        }
    };
    // WARNING: Committers.nil() here means that on-disk data can get out of sync with committing.
    Futures.addCallback(appenderator.push(segmentsToPush, Committers.nil()), new FutureCallback<SegmentsAndMetadata>() {

        @Override
        public void onSuccess(SegmentsAndMetadata result) {
            // Immediately publish after pushing
            for (DataSegment pushedSegment : result.getSegments()) {
                try {
                    segmentPublisher.publishSegment(pushedSegment);
                } catch (Exception e) {
                    errorHandler.apply(e);
                }
            }
            log.info("Published [%,d] sinks.", segmentsToPush.size());
        }

        @Override
        public void onFailure(Throwable e) {
            log.warn(e, "Failed to push [%,d] segments.", segmentsToPush.size());
            errorHandler.apply(e);
        }
    });
}

Example 54 with Function

use of com.google.common.base.Function in project druid by druid-io.

the class RealtimeManager method getQueryRunnerForSegments.

@Override
public <T> QueryRunner<T> getQueryRunnerForSegments(final Query<T> query, final Iterable<SegmentDescriptor> specs) {
    final QueryRunnerFactory<T, Query<T>> factory = conglomerate.findFactory(query);
    final Map<Integer, FireChief> partitionChiefs = chiefs.get(Iterables.getOnlyElement(query.getDataSource().getNames()));
    return partitionChiefs == null ? new NoopQueryRunner<T>() : factory.getToolchest().mergeResults(factory.mergeRunners(MoreExecutors.sameThreadExecutor(), Iterables.transform(specs, new Function<SegmentDescriptor, QueryRunner<T>>() {

        @Override
        public QueryRunner<T> apply(SegmentDescriptor spec) {
            final FireChief retVal = partitionChiefs.get(spec.getPartitionNumber());
            return retVal == null ? new NoopQueryRunner<T>() : retVal.getQueryRunner(query.withQuerySegmentSpec(new SpecificSegmentSpec(spec)));
        }
    })));
}

Example 55 with Function

use of com.google.common.base.Function in project druid by druid-io.

the class AppenderatorImpl method bootstrapSinksFromDisk.

/**
   * Populate "sinks" and "sinkTimeline" with committed segments, and announce them with the segmentAnnouncer.
   *
   * @return persisted commit metadata
   */
private Object bootstrapSinksFromDisk() {
    Preconditions.checkState(sinks.isEmpty(), "Already bootstrapped?!");
    final File baseDir = tuningConfig.getBasePersistDirectory();
    if (!baseDir.exists()) {
        return null;
    }
    final File[] files = baseDir.listFiles();
    if (files == null) {
        return null;
    }
    final File commitFile = computeCommitFile();
    final Committed committed;
    try {
        if (commitFile.exists()) {
            committed = objectMapper.readValue(commitFile, Committed.class);
        } else {
            committed = Committed.nil();
        }
    } catch (Exception e) {
        throw new ISE(e, "Failed to read commitFile: %s", commitFile);
    }
    log.info("Loading sinks from[%s]: %s", baseDir, committed.getHydrants().keySet());
    for (File sinkDir : files) {
        final File identifierFile = new File(sinkDir, IDENTIFIER_FILE_NAME);
        if (!identifierFile.isFile()) {
            // No identifier in this sinkDir; it must not actually be a sink directory. Skip it.
            continue;
        }
        try {
            final SegmentIdentifier identifier = objectMapper.readValue(new File(sinkDir, "identifier.json"), SegmentIdentifier.class);
            final int committedHydrants = committed.getCommittedHydrants(identifier.getIdentifierAsString());
            if (committedHydrants <= 0) {
                log.info("Removing uncommitted sink at [%s]", sinkDir);
                FileUtils.deleteDirectory(sinkDir);
                continue;
            }
            // To avoid reading and listing of "merged" dir and other special files
            final File[] sinkFiles = sinkDir.listFiles(new FilenameFilter() {

                @Override
                public boolean accept(File dir, String fileName) {
                    return !(Ints.tryParse(fileName) == null);
                }
            });
            Arrays.sort(sinkFiles, new Comparator<File>() {

                @Override
                public int compare(File o1, File o2) {
                    return Ints.compare(Integer.parseInt(o1.getName()), Integer.parseInt(o2.getName()));
                }
            });
            List<FireHydrant> hydrants = Lists.newArrayList();
            for (File hydrantDir : sinkFiles) {
                final int hydrantNumber = Integer.parseInt(hydrantDir.getName());
                if (hydrantNumber >= committedHydrants) {
                    log.info("Removing uncommitted segment at [%s]", hydrantDir);
                    FileUtils.deleteDirectory(hydrantDir);
                } else {
                    log.info("Loading previously persisted segment at [%s]", hydrantDir);
                    if (hydrantNumber != hydrants.size()) {
                        throw new ISE("Missing hydrant [%,d] in sinkDir [%s].", hydrants.size(), sinkDir);
                    }
                    hydrants.add(new FireHydrant(new QueryableIndexSegment(identifier.getIdentifierAsString(), indexIO.loadIndex(hydrantDir)), hydrantNumber));
                }
            }
            // Make sure we loaded enough hydrants.
            if (committedHydrants != hydrants.size()) {
                throw new ISE("Missing hydrant [%,d] in sinkDir [%s].", hydrants.size(), sinkDir);
            }
            Sink currSink = new Sink(identifier.getInterval(), schema, identifier.getShardSpec(), identifier.getVersion(), tuningConfig.getMaxRowsInMemory(), tuningConfig.isReportParseExceptions(), hydrants);
            sinks.put(identifier, currSink);
            sinkTimeline.add(currSink.getInterval(), currSink.getVersion(), identifier.getShardSpec().createChunk(currSink));
            segmentAnnouncer.announceSegment(currSink.getSegment());
        } catch (IOException e) {
            log.makeAlert(e, "Problem loading sink[%s] from disk.", schema.getDataSource()).addData("sinkDir", sinkDir).emit();
        }
    }
    // Make sure we loaded all committed sinks.
    final Set<String> loadedSinks = Sets.newHashSet(Iterables.transform(sinks.keySet(), new Function<SegmentIdentifier, String>() {

        @Override
        public String apply(SegmentIdentifier input) {
            return input.getIdentifierAsString();
        }
    }));
    final Set<String> missingSinks = Sets.difference(committed.getHydrants().keySet(), loadedSinks);
    if (!missingSinks.isEmpty()) {
        throw new ISE("Missing committed sinks [%s]", Joiner.on(", ").join(missingSinks));
    }
    return committed.getMetadata();
}