Search in sources :

Example 1 with Releaser

use of org.apache.druid.collections.Releaser 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();
        }
    });
}
Also used : Accumulator(org.apache.druid.java.util.common.guava.Accumulator) AbstractPrioritizedQueryRunnerCallable(org.apache.druid.query.AbstractPrioritizedQueryRunnerCallable) ChainedExecutionQueryRunner(org.apache.druid.query.ChainedExecutionQueryRunner) QueryTimeoutException(org.apache.druid.query.QueryTimeoutException) GroupByQuery(org.apache.druid.query.groupby.GroupByQuery) Releaser(org.apache.druid.collections.Releaser) ImmutableList(com.google.common.collect.ImmutableList) List(java.util.List) ISE(org.apache.druid.java.util.common.ISE) Pair(org.apache.druid.java.util.common.Pair) QueryInterruptedException(org.apache.druid.query.QueryInterruptedException) ResultRow(org.apache.druid.query.groupby.ResultRow) Closer(org.apache.druid.java.util.common.io.Closer) GroupByQueryConfig(org.apache.druid.query.groupby.GroupByQueryConfig) RowBasedKey(org.apache.druid.query.groupby.epinephelinae.RowBasedGrouperHelper.RowBasedKey) ByteBuffer(java.nio.ByteBuffer) BaseSequence(org.apache.druid.java.util.common.guava.BaseSequence) ChainedExecutionQueryRunner(org.apache.druid.query.ChainedExecutionQueryRunner) QueryRunner(org.apache.druid.query.QueryRunner) TimeoutException(java.util.concurrent.TimeoutException) CancellationException(java.util.concurrent.CancellationException) QueryInterruptedException(org.apache.druid.query.QueryInterruptedException) ExecutionException(java.util.concurrent.ExecutionException) QueryTimeoutException(org.apache.druid.query.QueryTimeoutException) ResourceLimitExceededException(org.apache.druid.query.ResourceLimitExceededException) ReferenceCountingResourceHolder(org.apache.druid.collections.ReferenceCountingResourceHolder) ListenableFuture(com.google.common.util.concurrent.ListenableFuture) File(java.io.File)

Example 2 with Releaser

use of org.apache.druid.collections.Releaser in project druid by druid-io.

the class ParallelCombiner method runCombiner.

private Pair<CloseableIterator<Entry<KeyType>>, Future> runCombiner(List<? extends CloseableIterator<Entry<KeyType>>> iterators, ByteBuffer combineBuffer, AggregatorFactory[] combiningFactories, List<String> dictionary) {
    final SettableColumnSelectorFactory settableColumnSelectorFactory = new SettableColumnSelectorFactory(combiningFactories);
    final StreamingMergeSortedGrouper<KeyType> grouper = new StreamingMergeSortedGrouper<>(Suppliers.ofInstance(combineBuffer), combineKeySerdeFactory.factorizeWithDictionary(dictionary), settableColumnSelectorFactory, combiningFactories, queryTimeoutAt);
    // init() must be called before iterator(), so cannot be called inside the below callable.
    grouper.init();
    final ListenableFuture future = executor.submit(new AbstractPrioritizedCallable<Void>(priority) {

        @Override
        public Void call() {
            try (CloseableIterator<Entry<KeyType>> mergedIterator = CloseableIterators.mergeSorted(iterators, keyObjComparator);
                // This variable is used to close releaser automatically.
                @SuppressWarnings("unused") final Releaser releaser = combineBufferHolder.increment()) {
                while (mergedIterator.hasNext()) {
                    final Entry<KeyType> next = mergedIterator.next();
                    settableColumnSelectorFactory.set(next.values);
                    // grouper always returns ok or throws an exception
                    grouper.aggregate(next.key);
                    settableColumnSelectorFactory.set(null);
                }
            } catch (IOException e) {
                throw new RuntimeException(e);
            }
            grouper.finish();
            return null;
        }
    });
    return new Pair<>(grouper.iterator(), future);
}
Also used : CloseableIterator(org.apache.druid.java.util.common.parsers.CloseableIterator) IOException(java.io.IOException) Entry(org.apache.druid.query.groupby.epinephelinae.Grouper.Entry) Releaser(org.apache.druid.collections.Releaser) ListenableFuture(com.google.common.util.concurrent.ListenableFuture) Pair(org.apache.druid.java.util.common.Pair)

Aggregations

ListenableFuture (com.google.common.util.concurrent.ListenableFuture)2 Releaser (org.apache.druid.collections.Releaser)2 Pair (org.apache.druid.java.util.common.Pair)2 ImmutableList (com.google.common.collect.ImmutableList)1 File (java.io.File)1 IOException (java.io.IOException)1 ByteBuffer (java.nio.ByteBuffer)1 List (java.util.List)1 CancellationException (java.util.concurrent.CancellationException)1 ExecutionException (java.util.concurrent.ExecutionException)1 TimeoutException (java.util.concurrent.TimeoutException)1 ReferenceCountingResourceHolder (org.apache.druid.collections.ReferenceCountingResourceHolder)1 ISE (org.apache.druid.java.util.common.ISE)1 Accumulator (org.apache.druid.java.util.common.guava.Accumulator)1 BaseSequence (org.apache.druid.java.util.common.guava.BaseSequence)1 Closer (org.apache.druid.java.util.common.io.Closer)1 CloseableIterator (org.apache.druid.java.util.common.parsers.CloseableIterator)1 AbstractPrioritizedQueryRunnerCallable (org.apache.druid.query.AbstractPrioritizedQueryRunnerCallable)1 ChainedExecutionQueryRunner (org.apache.druid.query.ChainedExecutionQueryRunner)1 QueryInterruptedException (org.apache.druid.query.QueryInterruptedException)1