Examples with FireHydrant org.apache.druid.segment.realtime.FireHydrant used on opensource projects

Example 11 with FireHydrant

use of org.apache.druid.segment.realtime.FireHydrant in project druid by apache.

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 Committed committed;
    File commitFile = null;
    try {
        commitLock.lock();
        commitFile = computeCommitFile();
        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);
    } finally {
        commitLock.unlock();
    }
    int rowsSoFar = 0;
    if (committed.equals(Committed.nil())) {
        log.debug("No previously committed metadata.");
    } else {
        log.info("Loading partially-persisted segments[%s] from[%s] with commit metadata: %s", String.join(", ", committed.getHydrants().keySet()), baseDir, committed.getMetadata());
    }
    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 SegmentIdWithShardSpec identifier = objectMapper.readValue(new File(sinkDir, "identifier.json"), SegmentIdWithShardSpec.class);
            final int committedHydrants = committed.getCommittedHydrants(identifier.toString());
            if (committedHydrants <= 0) {
                log.info("Removing uncommitted segment at [%s].", sinkDir);
                FileUtils.deleteDirectory(sinkDir);
                continue;
            }
            // To avoid reading and listing of "merged" dir and other special files
            final File[] sinkFiles = sinkDir.listFiles((dir, fileName) -> !(Ints.tryParse(fileName) == null));
            Arrays.sort(sinkFiles, (o1, o2) -> Ints.compare(Integer.parseInt(o1.getName()), Integer.parseInt(o2.getName())));
            List<FireHydrant> hydrants = new ArrayList<>();
            for (File hydrantDir : sinkFiles) {
                final int hydrantNumber = Integer.parseInt(hydrantDir.getName());
                if (hydrantNumber >= committedHydrants) {
                    log.info("Removing uncommitted partial segment at [%s]", hydrantDir);
                    FileUtils.deleteDirectory(hydrantDir);
                } else {
                    log.debug("Loading previously persisted partial 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(indexIO.loadIndex(hydrantDir), identifier.asSegmentId()), 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.getAppendableIndexSpec(), tuningConfig.getMaxRowsInMemory(), maxBytesTuningConfig, useMaxMemoryEstimates, null, hydrants);
            rowsSoFar += currSink.getNumRows();
            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(), SegmentIdWithShardSpec::toString));
    final Set<String> missingSinks = Sets.difference(committed.getHydrants().keySet(), loadedSinks);
    if (!missingSinks.isEmpty()) {
        throw new ISE("Missing committed sinks [%s]", Joiner.on(", ").join(missingSinks));
    }
    totalRows.set(rowsSoFar);
    return committed.getMetadata();
}

Example 12 with FireHydrant

use of org.apache.druid.segment.realtime.FireHydrant in project druid by apache.

the class AppenderatorImpl method mergeAndPush.

/**
 * Merge segment, push to deep storage. Should only be used on segments that have been fully persisted. Must only
 * be run in the single-threaded pushExecutor.
 *
 * @param identifier    sink identifier
 * @param sink          sink to push
 * @param useUniquePath true if the segment should be written to a path with a unique identifier
 *
 * @return segment descriptor, or null if the sink is no longer valid
 */
@Nullable
private DataSegment mergeAndPush(final SegmentIdWithShardSpec identifier, final Sink sink, final boolean useUniquePath) {
    // noinspection ObjectEquality
    if (sinks.get(identifier) != sink) {
        log.warn("Sink for segment[%s] no longer valid, bailing out of mergeAndPush.", identifier);
        return null;
    }
    // Use a descriptor file to indicate that pushing has completed.
    final File persistDir = computePersistDir(identifier);
    final File mergedTarget = new File(persistDir, "merged");
    final File descriptorFile = computeDescriptorFile(identifier);
    // Sanity checks
    for (FireHydrant hydrant : sink) {
        if (sink.isWritable()) {
            throw new ISE("Expected sink to be no longer writable before mergeAndPush for segment[%s].", identifier);
        }
        synchronized (hydrant) {
            if (!hydrant.hasSwapped()) {
                throw new ISE("Expected sink to be fully persisted before mergeAndPush for segment[%s].", identifier);
            }
        }
    }
    try {
        if (descriptorFile.exists()) {
            if (useUniquePath) {
                // Don't reuse the descriptor, because the caller asked for a unique path. Leave the old one as-is, since
                // it might serve some unknown purpose.
                log.debug("Segment[%s] already pushed, but we want a unique path, so will push again with a new path.", identifier);
            } else {
                log.info("Segment[%s] already pushed, skipping.", identifier);
                return objectMapper.readValue(descriptorFile, DataSegment.class);
            }
        }
        removeDirectory(mergedTarget);
        if (mergedTarget.exists()) {
            throw new ISE("Merged target[%s] exists after removing?!", mergedTarget);
        }
        final File mergedFile;
        final long mergeFinishTime;
        final long startTime = System.nanoTime();
        List<QueryableIndex> indexes = new ArrayList<>();
        Closer closer = Closer.create();
        try {
            for (FireHydrant fireHydrant : sink) {
                // if batch, swap/persist did not memory map the incremental index, we need it mapped now:
                if (!isOpenSegments()) {
                    // sanity
                    Pair<File, SegmentId> persistedMetadata = persistedHydrantMetadata.get(fireHydrant);
                    if (persistedMetadata == null) {
                        throw new ISE("Persisted metadata for batch hydrant [%s] is null!", fireHydrant);
                    }
                    File persistedFile = persistedMetadata.lhs;
                    SegmentId persistedSegmentId = persistedMetadata.rhs;
                    // sanity:
                    if (persistedFile == null) {
                        throw new ISE("Persisted file for batch hydrant [%s] is null!", fireHydrant);
                    } else if (persistedSegmentId == null) {
                        throw new ISE("Persisted segmentId for batch hydrant in file [%s] is null!", persistedFile.getPath());
                    }
                    fireHydrant.swapSegment(new QueryableIndexSegment(indexIO.loadIndex(persistedFile), persistedSegmentId));
                }
                Pair<ReferenceCountingSegment, Closeable> segmentAndCloseable = fireHydrant.getAndIncrementSegment();
                final QueryableIndex queryableIndex = segmentAndCloseable.lhs.asQueryableIndex();
                log.debug("Segment[%s] adding hydrant[%s]", identifier, fireHydrant);
                indexes.add(queryableIndex);
                closer.register(segmentAndCloseable.rhs);
            }
            mergedFile = indexMerger.mergeQueryableIndex(indexes, schema.getGranularitySpec().isRollup(), schema.getAggregators(), schema.getDimensionsSpec(), mergedTarget, tuningConfig.getIndexSpec(), tuningConfig.getIndexSpecForIntermediatePersists(), new BaseProgressIndicator(), tuningConfig.getSegmentWriteOutMediumFactory(), tuningConfig.getMaxColumnsToMerge());
            mergeFinishTime = System.nanoTime();
            log.debug("Segment[%s] built in %,dms.", identifier, (mergeFinishTime - startTime) / 1000000);
        } catch (Throwable t) {
            throw closer.rethrow(t);
        } finally {
            closer.close();
        }
        final DataSegment segmentToPush = sink.getSegment().withDimensions(IndexMerger.getMergedDimensionsFromQueryableIndexes(indexes, schema.getDimensionsSpec()));
        // Retry pushing segments because uploading to deep storage might fail especially for cloud storage types
        final DataSegment segment = RetryUtils.retry(// semantics.
        () -> dataSegmentPusher.push(mergedFile, segmentToPush, useUniquePath), exception -> exception instanceof Exception, 5);
        if (!isOpenSegments()) {
            // can generate OOMs during merge if enough of them are held back...
            for (FireHydrant fireHydrant : sink) {
                fireHydrant.swapSegment(null);
            }
        }
        final long pushFinishTime = System.nanoTime();
        objectMapper.writeValue(descriptorFile, segment);
        log.info("Segment[%s] of %,d bytes " + "built from %d incremental persist(s) in %,dms; " + "pushed to deep storage in %,dms. " + "Load spec is: %s", identifier, segment.getSize(), indexes.size(), (mergeFinishTime - startTime) / 1000000, (pushFinishTime - mergeFinishTime) / 1000000, objectMapper.writeValueAsString(segment.getLoadSpec()));
        return segment;
    } catch (Exception e) {
        metrics.incrementFailedHandoffs();
        log.warn(e, "Failed to push merged index for segment[%s].", identifier);
        throw new RuntimeException(e);
    }
}

Example 13 with FireHydrant

use of org.apache.druid.segment.realtime.FireHydrant in project druid by apache.

the class RealtimePlumber method persistAndMerge.

// Submits persist-n-merge task for a Sink to the mergeExecutor
private void persistAndMerge(final long truncatedTime, final Sink sink) {
    final String threadName = StringUtils.format("%s-%s-persist-n-merge", schema.getDataSource(), DateTimes.utc(truncatedTime));
    mergeExecutor.execute(new ThreadRenamingRunnable(threadName) {

        final Interval interval = sink.getInterval();

        Stopwatch mergeStopwatch = null;

        @Override
        public void doRun() {
            try {
                // Bail out if this sink has been abandoned by a previously-executed task.
                if (sinks.get(truncatedTime) != sink) {
                    log.info("Sink[%s] was abandoned, bailing out of persist-n-merge.", sink);
                    return;
                }
                // Use a file to indicate that pushing has completed.
                final File persistDir = computePersistDir(schema, interval);
                final File mergedTarget = new File(persistDir, "merged");
                final File isPushedMarker = new File(persistDir, "isPushedMarker");
                if (!isPushedMarker.exists()) {
                    removeSegment(sink, mergedTarget);
                    if (mergedTarget.exists()) {
                        log.warn("Merged target[%s] still exists after attempt to delete it; skipping push.", mergedTarget);
                        return;
                    }
                } else {
                    log.info("Already pushed sink[%s]", sink);
                    return;
                }
                /*
            Note: it the plumber crashes after persisting a subset of hydrants then might duplicate data as these
            hydrants will be read but older commitMetadata will be used. fixing this possibly needs structural
            changes to plumber.
             */
                for (FireHydrant hydrant : sink) {
                    synchronized (hydrant) {
                        if (!hydrant.hasSwapped()) {
                            log.info("Hydrant[%s] hasn't swapped yet, swapping. Sink[%s]", hydrant, sink);
                            final int rowCount = persistHydrant(hydrant, schema, interval, null);
                            metrics.incrementRowOutputCount(rowCount);
                        }
                    }
                }
                final long mergeThreadCpuTime = JvmUtils.safeGetThreadCpuTime();
                mergeStopwatch = Stopwatch.createStarted();
                final File mergedFile;
                List<QueryableIndex> indexes = new ArrayList<>();
                Closer closer = Closer.create();
                try {
                    for (FireHydrant fireHydrant : sink) {
                        Pair<ReferenceCountingSegment, Closeable> segmentAndCloseable = fireHydrant.getAndIncrementSegment();
                        final QueryableIndex queryableIndex = segmentAndCloseable.lhs.asQueryableIndex();
                        log.info("Adding hydrant[%s]", fireHydrant);
                        indexes.add(queryableIndex);
                        closer.register(segmentAndCloseable.rhs);
                    }
                    mergedFile = indexMerger.mergeQueryableIndex(indexes, schema.getGranularitySpec().isRollup(), schema.getAggregators(), null, mergedTarget, config.getIndexSpec(), config.getIndexSpecForIntermediatePersists(), new BaseProgressIndicator(), config.getSegmentWriteOutMediumFactory(), -1);
                } catch (Throwable t) {
                    throw closer.rethrow(t);
                } finally {
                    closer.close();
                }
                // emit merge metrics before publishing segment
                metrics.incrementMergeCpuTime(JvmUtils.safeGetThreadCpuTime() - mergeThreadCpuTime);
                metrics.incrementMergeTimeMillis(mergeStopwatch.elapsed(TimeUnit.MILLISECONDS));
                log.info("Pushing [%s] to deep storage", sink.getSegment().getId());
                DataSegment segment = dataSegmentPusher.push(mergedFile, sink.getSegment().withDimensions(IndexMerger.getMergedDimensionsFromQueryableIndexes(indexes, schema.getDimensionsSpec())), false);
                log.info("Inserting [%s] to the metadata store", sink.getSegment().getId());
                segmentPublisher.publishSegment(segment);
                if (!isPushedMarker.createNewFile()) {
                    log.makeAlert("Failed to create marker file for [%s]", schema.getDataSource()).addData("interval", sink.getInterval()).addData("partitionNum", segment.getShardSpec().getPartitionNum()).addData("marker", isPushedMarker).emit();
                }
            } catch (Exception e) {
                metrics.incrementFailedHandoffs();
                log.makeAlert(e, "Failed to persist merged index[%s]", schema.getDataSource()).addData("interval", interval).emit();
                if (shuttingDown) {
                    // We're trying to shut down, and this segment failed to push. Let's just get rid of it.
                    // This call will also delete possibly-partially-written files, so we don't need to do it explicitly.
                    cleanShutdown = false;
                    abandonSegment(truncatedTime, sink);
                }
            } finally {
                if (mergeStopwatch != null) {
                    mergeStopwatch.stop();
                }
            }
        }
    });
    handoffNotifier.registerSegmentHandoffCallback(new SegmentDescriptor(sink.getInterval(), sink.getVersion(), config.getShardSpec().getPartitionNum()), mergeExecutor, new Runnable() {

        @Override
        public void run() {
            abandonSegment(sink.getInterval().getStartMillis(), sink);
            metrics.incrementHandOffCount();
        }
    });
}

Example 14 with FireHydrant

use of org.apache.druid.segment.realtime.FireHydrant in project druid by apache.

the class RealtimePlumberSchoolTest method testDimOrderInheritanceHelper.

private void testDimOrderInheritanceHelper(final Object commitMetadata) throws Exception {
    List<List<String>> expectedDims = ImmutableList.of(ImmutableList.of("dimD"), ImmutableList.of("dimC"), ImmutableList.of("dimA"), ImmutableList.of("dimB"), ImmutableList.of("dimE"), ImmutableList.of("dimD", "dimC", "dimA", "dimB", "dimE"));
    QueryableIndex qindex;
    FireHydrant hydrant;
    Map<Long, Sink> sinks;
    RealtimePlumber plumber = (RealtimePlumber) realtimePlumberSchool.findPlumber(schema2, tuningConfig, metrics);
    Assert.assertNull(plumber.startJob());
    final CountDownLatch doneSignal = new CountDownLatch(1);
    final Committer committer = new Committer() {

        @Override
        public Object getMetadata() {
            return commitMetadata;
        }

        @Override
        public void run() {
            doneSignal.countDown();
        }
    };
    plumber.add(getTestInputRowFull("1970-01-01", ImmutableList.of("dimD"), ImmutableList.of("1")), Suppliers.ofInstance(committer));
    plumber.add(getTestInputRowFull("1970-01-01", ImmutableList.of("dimC"), ImmutableList.of("1")), Suppliers.ofInstance(committer));
    plumber.add(getTestInputRowFull("1970-01-01", ImmutableList.of("dimA"), ImmutableList.of("1")), Suppliers.ofInstance(committer));
    plumber.add(getTestInputRowFull("1970-01-01", ImmutableList.of("dimB"), ImmutableList.of("1")), Suppliers.ofInstance(committer));
    plumber.add(getTestInputRowFull("1970-01-01", ImmutableList.of("dimE"), ImmutableList.of("1")), Suppliers.ofInstance(committer));
    plumber.add(getTestInputRowFull("1970-01-01", ImmutableList.of("dimA", "dimB", "dimC", "dimD", "dimE"), ImmutableList.of("1")), Suppliers.ofInstance(committer));
    plumber.persist(committer);
    doneSignal.await();
    plumber.getSinks().clear();
    plumber.finishJob();
    RealtimePlumber restoredPlumber = (RealtimePlumber) realtimePlumberSchool.findPlumber(schema2, tuningConfig, metrics);
    restoredPlumber.bootstrapSinksFromDisk();
    sinks = restoredPlumber.getSinks();
    Assert.assertEquals(1, sinks.size());
    List<FireHydrant> hydrants = Lists.newArrayList(sinks.get(0L));
    for (int i = 0; i < hydrants.size(); i++) {
        hydrant = hydrants.get(i);
        ReferenceCountingSegment segment = hydrant.getIncrementedSegment();
        try {
            qindex = segment.asQueryableIndex();
            Assert.assertEquals(i, hydrant.getCount());
            Assert.assertEquals(expectedDims.get(i), ImmutableList.copyOf(qindex.getAvailableDimensions()));
        } finally {
            segment.decrement();
        }
    }
}

Example 15 with FireHydrant

use of org.apache.druid.segment.realtime.FireHydrant in project druid by apache.

the class StreamAppenderator 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 Committed committed;
    File commitFile = null;
    try {
        commitLock.lock();
        commitFile = computeCommitFile();
        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);
    } finally {
        commitLock.unlock();
    }
    int rowsSoFar = 0;
    if (committed.equals(Committed.nil())) {
        log.debug("No previously committed metadata.");
    } else {
        log.info("Loading partially-persisted segments[%s] from[%s] with commit metadata: %s", String.join(", ", committed.getHydrants().keySet()), baseDir, committed.getMetadata());
    }
    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 SegmentIdWithShardSpec identifier = objectMapper.readValue(new File(sinkDir, "identifier.json"), SegmentIdWithShardSpec.class);
            final int committedHydrants = committed.getCommittedHydrants(identifier.toString());
            if (committedHydrants <= 0) {
                log.info("Removing uncommitted segment at [%s].", sinkDir);
                FileUtils.deleteDirectory(sinkDir);
                continue;
            }
            // To avoid reading and listing of "merged" dir and other special files
            final File[] sinkFiles = sinkDir.listFiles((dir, fileName) -> !(Ints.tryParse(fileName) == null));
            Arrays.sort(sinkFiles, (o1, o2) -> Ints.compare(Integer.parseInt(o1.getName()), Integer.parseInt(o2.getName())));
            List<FireHydrant> hydrants = new ArrayList<>();
            for (File hydrantDir : sinkFiles) {
                final int hydrantNumber = Integer.parseInt(hydrantDir.getName());
                if (hydrantNumber >= committedHydrants) {
                    log.info("Removing uncommitted partial segment at [%s]", hydrantDir);
                    FileUtils.deleteDirectory(hydrantDir);
                } else {
                    log.debug("Loading previously persisted partial 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(indexIO.loadIndex(hydrantDir), identifier.asSegmentId()), 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.getAppendableIndexSpec(), tuningConfig.getMaxRowsInMemory(), maxBytesTuningConfig, useMaxMemoryEstimates, null, hydrants);
            rowsSoFar += currSink.getNumRows();
            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(), SegmentIdWithShardSpec::toString));
    final Set<String> missingSinks = Sets.difference(committed.getHydrants().keySet(), loadedSinks);
    if (!missingSinks.isEmpty()) {
        throw new ISE("Missing committed sinks [%s]", Joiner.on(", ").join(missingSinks));
    }
    totalRows.set(rowsSoFar);
    return committed.getMetadata();
}