Examples with SegmentsAndMetadata io.druid.segment.realtime.appenderator.SegmentsAndMetadata used on opensource projects

Example 1 with SegmentsAndMetadata

use of io.druid.segment.realtime.appenderator.SegmentsAndMetadata in project druid by druid-io.

the class KafkaIndexTask method run.

@Override
public TaskStatus run(final TaskToolbox toolbox) throws Exception {
    log.info("Starting up!");
    startTime = DateTime.now();
    mapper = toolbox.getObjectMapper();
    status = Status.STARTING;
    if (chatHandlerProvider.isPresent()) {
        log.info("Found chat handler of class[%s]", chatHandlerProvider.get().getClass().getName());
        chatHandlerProvider.get().register(getId(), this, false);
    } else {
        log.warn("No chat handler detected");
    }
    runThread = Thread.currentThread();
    // Set up FireDepartmentMetrics
    final FireDepartment fireDepartmentForMetrics = new FireDepartment(dataSchema, new RealtimeIOConfig(null, null, null), null);
    fireDepartmentMetrics = fireDepartmentForMetrics.getMetrics();
    toolbox.getMonitorScheduler().addMonitor(new RealtimeMetricsMonitor(ImmutableList.of(fireDepartmentForMetrics), ImmutableMap.of(DruidMetrics.TASK_ID, new String[] { getId() })));
    try (final Appenderator appenderator0 = newAppenderator(fireDepartmentMetrics, toolbox);
        final FiniteAppenderatorDriver driver = newDriver(appenderator0, toolbox, fireDepartmentMetrics);
        final KafkaConsumer<byte[], byte[]> consumer = newConsumer()) {
        appenderator = appenderator0;
        final String topic = ioConfig.getStartPartitions().getTopic();
        // Start up, set up initial offsets.
        final Object restoredMetadata = driver.startJob();
        if (restoredMetadata == null) {
            nextOffsets.putAll(ioConfig.getStartPartitions().getPartitionOffsetMap());
        } else {
            final Map<String, Object> restoredMetadataMap = (Map) restoredMetadata;
            final KafkaPartitions restoredNextPartitions = toolbox.getObjectMapper().convertValue(restoredMetadataMap.get(METADATA_NEXT_PARTITIONS), KafkaPartitions.class);
            nextOffsets.putAll(restoredNextPartitions.getPartitionOffsetMap());
            // Sanity checks.
            if (!restoredNextPartitions.getTopic().equals(ioConfig.getStartPartitions().getTopic())) {
                throw new ISE("WTF?! Restored topic[%s] but expected topic[%s]", restoredNextPartitions.getTopic(), ioConfig.getStartPartitions().getTopic());
            }
            if (!nextOffsets.keySet().equals(ioConfig.getStartPartitions().getPartitionOffsetMap().keySet())) {
                throw new ISE("WTF?! Restored partitions[%s] but expected partitions[%s]", nextOffsets.keySet(), ioConfig.getStartPartitions().getPartitionOffsetMap().keySet());
            }
        }
        // Set up sequenceNames.
        final Map<Integer, String> sequenceNames = Maps.newHashMap();
        for (Integer partitionNum : nextOffsets.keySet()) {
            sequenceNames.put(partitionNum, String.format("%s_%s", ioConfig.getBaseSequenceName(), partitionNum));
        }
        // Set up committer.
        final Supplier<Committer> committerSupplier = new Supplier<Committer>() {

            @Override
            public Committer get() {
                final Map<Integer, Long> snapshot = ImmutableMap.copyOf(nextOffsets);
                return new Committer() {

                    @Override
                    public Object getMetadata() {
                        return ImmutableMap.of(METADATA_NEXT_PARTITIONS, new KafkaPartitions(ioConfig.getStartPartitions().getTopic(), snapshot));
                    }

                    @Override
                    public void run() {
                    // Do nothing.
                    }
                };
            }
        };
        Set<Integer> assignment = assignPartitionsAndSeekToNext(consumer, topic);
        // Main loop.
        // Could eventually support leader/follower mode (for keeping replicas more in sync)
        boolean stillReading = !assignment.isEmpty();
        status = Status.READING;
        try {
            while (stillReading) {
                if (possiblyPause(assignment)) {
                    // The partition assignments may have changed while paused by a call to setEndOffsets() so reassign
                    // partitions upon resuming. This is safe even if the end offsets have not been modified.
                    assignment = assignPartitionsAndSeekToNext(consumer, topic);
                    if (assignment.isEmpty()) {
                        log.info("All partitions have been fully read");
                        publishOnStop = true;
                        stopRequested = true;
                    }
                }
                if (stopRequested) {
                    break;
                }
                // The retrying business is because the KafkaConsumer throws OffsetOutOfRangeException if the seeked-to
                // offset is not present in the topic-partition. This can happen if we're asking a task to read from data
                // that has not been written yet (which is totally legitimate). So let's wait for it to show up.
                ConsumerRecords<byte[], byte[]> records = ConsumerRecords.empty();
                try {
                    records = consumer.poll(POLL_TIMEOUT);
                } catch (OffsetOutOfRangeException e) {
                    log.warn("OffsetOutOfRangeException with message [%s]", e.getMessage());
                    possiblyResetOffsetsOrWait(e.offsetOutOfRangePartitions(), consumer, toolbox);
                    stillReading = ioConfig.isPauseAfterRead() || !assignment.isEmpty();
                }
                for (ConsumerRecord<byte[], byte[]> record : records) {
                    if (log.isTraceEnabled()) {
                        log.trace("Got topic[%s] partition[%d] offset[%,d].", record.topic(), record.partition(), record.offset());
                    }
                    if (record.offset() < endOffsets.get(record.partition())) {
                        if (record.offset() != nextOffsets.get(record.partition())) {
                            throw new ISE("WTF?! Got offset[%,d] after offset[%,d] in partition[%d].", record.offset(), nextOffsets.get(record.partition()), record.partition());
                        }
                        try {
                            final byte[] valueBytes = record.value();
                            if (valueBytes == null) {
                                throw new ParseException("null value");
                            }
                            final InputRow row = Preconditions.checkNotNull(parser.parse(ByteBuffer.wrap(valueBytes)), "row");
                            if (!ioConfig.getMinimumMessageTime().isPresent() || !ioConfig.getMinimumMessageTime().get().isAfter(row.getTimestamp())) {
                                final SegmentIdentifier identifier = driver.add(row, sequenceNames.get(record.partition()), committerSupplier);
                                if (identifier == null) {
                                    // If we allow continuing, then consider blacklisting the interval for a while to avoid constant checks.
                                    throw new ISE("Could not allocate segment for row with timestamp[%s]", row.getTimestamp());
                                }
                                fireDepartmentMetrics.incrementProcessed();
                            } else {
                                fireDepartmentMetrics.incrementThrownAway();
                            }
                        } catch (ParseException e) {
                            if (tuningConfig.isReportParseExceptions()) {
                                throw e;
                            } else {
                                log.debug(e, "Dropping unparseable row from partition[%d] offset[%,d].", record.partition(), record.offset());
                                fireDepartmentMetrics.incrementUnparseable();
                            }
                        }
                        nextOffsets.put(record.partition(), record.offset() + 1);
                    }
                    if (nextOffsets.get(record.partition()).equals(endOffsets.get(record.partition())) && assignment.remove(record.partition())) {
                        log.info("Finished reading topic[%s], partition[%,d].", record.topic(), record.partition());
                        assignPartitions(consumer, topic, assignment);
                        stillReading = ioConfig.isPauseAfterRead() || !assignment.isEmpty();
                    }
                }
            }
        } finally {
            // persist pending data
            driver.persist(committerSupplier.get());
        }
        synchronized (statusLock) {
            if (stopRequested && !publishOnStop) {
                throw new InterruptedException("Stopping without publishing");
            }
            status = Status.PUBLISHING;
        }
        final TransactionalSegmentPublisher publisher = new TransactionalSegmentPublisher() {

            @Override
            public boolean publishSegments(Set<DataSegment> segments, Object commitMetadata) throws IOException {
                final KafkaPartitions finalPartitions = toolbox.getObjectMapper().convertValue(((Map) commitMetadata).get(METADATA_NEXT_PARTITIONS), KafkaPartitions.class);
                // Sanity check, we should only be publishing things that match our desired end state.
                if (!endOffsets.equals(finalPartitions.getPartitionOffsetMap())) {
                    throw new ISE("WTF?! Driver attempted to publish invalid metadata[%s].", commitMetadata);
                }
                final SegmentTransactionalInsertAction action;
                if (ioConfig.isUseTransaction()) {
                    action = new SegmentTransactionalInsertAction(segments, new KafkaDataSourceMetadata(ioConfig.getStartPartitions()), new KafkaDataSourceMetadata(finalPartitions));
                } else {
                    action = new SegmentTransactionalInsertAction(segments, null, null);
                }
                log.info("Publishing with isTransaction[%s].", ioConfig.isUseTransaction());
                return toolbox.getTaskActionClient().submit(action).isSuccess();
            }
        };
        final SegmentsAndMetadata published = driver.finish(publisher, committerSupplier.get());
        if (published == null) {
            throw new ISE("Transaction failure publishing segments, aborting");
        } else {
            log.info("Published segments[%s] with metadata[%s].", Joiner.on(", ").join(Iterables.transform(published.getSegments(), new Function<DataSegment, String>() {

                @Override
                public String apply(DataSegment input) {
                    return input.getIdentifier();
                }
            })), published.getCommitMetadata());
        }
    } catch (InterruptedException | RejectedExecutionException e) {
        // handle the InterruptedException that gets wrapped in a RejectedExecutionException
        if (e instanceof RejectedExecutionException && (e.getCause() == null || !(e.getCause() instanceof InterruptedException))) {
            throw e;
        }
        // if we were interrupted because we were asked to stop, handle the exception and return success, else rethrow
        if (!stopRequested) {
            Thread.currentThread().interrupt();
            throw e;
        }
        log.info("The task was asked to stop before completing");
    } finally {
        if (chatHandlerProvider.isPresent()) {
            chatHandlerProvider.get().unregister(getId());
        }
    }
    return success();
}

Example 2 with SegmentsAndMetadata

use of io.druid.segment.realtime.appenderator.SegmentsAndMetadata in project druid by druid-io.

the class IndexTask method generateAndPublishSegments.

private boolean generateAndPublishSegments(final TaskToolbox toolbox, final DataSchema dataSchema, final Map<Interval, List<ShardSpec>> shardSpecs, final String version, final FirehoseFactory firehoseFactory) throws IOException, InterruptedException {
    final GranularitySpec granularitySpec = dataSchema.getGranularitySpec();
    final FireDepartment fireDepartmentForMetrics = new FireDepartment(dataSchema, new RealtimeIOConfig(null, null, null), null);
    final FireDepartmentMetrics fireDepartmentMetrics = fireDepartmentForMetrics.getMetrics();
    final Map<String, ShardSpec> sequenceNameToShardSpecMap = Maps.newHashMap();
    if (toolbox.getMonitorScheduler() != null) {
        toolbox.getMonitorScheduler().addMonitor(new RealtimeMetricsMonitor(ImmutableList.of(fireDepartmentForMetrics), ImmutableMap.of(DruidMetrics.TASK_ID, new String[] { getId() })));
    }
    final SegmentAllocator segmentAllocator;
    if (ingestionSchema.getIOConfig().isAppendToExisting()) {
        segmentAllocator = new ActionBasedSegmentAllocator(toolbox.getTaskActionClient(), dataSchema);
    } else {
        segmentAllocator = new SegmentAllocator() {

            @Override
            public SegmentIdentifier allocate(DateTime timestamp, String sequenceName, String previousSegmentId) throws IOException {
                Optional<Interval> interval = granularitySpec.bucketInterval(timestamp);
                if (!interval.isPresent()) {
                    throw new ISE("Could not find interval for timestamp [%s]", timestamp);
                }
                ShardSpec shardSpec = sequenceNameToShardSpecMap.get(sequenceName);
                if (shardSpec == null) {
                    throw new ISE("Could not find ShardSpec for sequenceName [%s]", sequenceName);
                }
                return new SegmentIdentifier(getDataSource(), interval.get(), version, shardSpec);
            }
        };
    }
    try (final Appenderator appenderator = newAppenderator(fireDepartmentMetrics, toolbox, dataSchema);
        final FiniteAppenderatorDriver driver = newDriver(appenderator, toolbox, segmentAllocator, fireDepartmentMetrics);
        final Firehose firehose = firehoseFactory.connect(dataSchema.getParser())) {
        final Supplier<Committer> committerSupplier = Committers.supplierFromFirehose(firehose);
        final Map<Interval, ShardSpecLookup> shardSpecLookups = Maps.newHashMap();
        if (driver.startJob() != null) {
            driver.clear();
        }
        try {
            while (firehose.hasMore()) {
                try {
                    final InputRow inputRow = firehose.nextRow();
                    final Optional<Interval> optInterval = granularitySpec.bucketInterval(inputRow.getTimestamp());
                    if (!optInterval.isPresent()) {
                        fireDepartmentMetrics.incrementThrownAway();
                        continue;
                    }
                    final Interval interval = optInterval.get();
                    if (!shardSpecLookups.containsKey(interval)) {
                        final List<ShardSpec> intervalShardSpecs = shardSpecs.get(interval);
                        if (intervalShardSpecs == null || intervalShardSpecs.isEmpty()) {
                            throw new ISE("Failed to get shardSpec for interval[%s]", interval);
                        }
                        shardSpecLookups.put(interval, intervalShardSpecs.get(0).getLookup(intervalShardSpecs));
                    }
                    final ShardSpec shardSpec = shardSpecLookups.get(interval).getShardSpec(inputRow.getTimestampFromEpoch(), inputRow);
                    final String sequenceName = String.format("index_%s_%s_%d", interval, version, shardSpec.getPartitionNum());
                    if (!sequenceNameToShardSpecMap.containsKey(sequenceName)) {
                        final ShardSpec shardSpecForPublishing = ingestionSchema.getTuningConfig().isForceExtendableShardSpecs() || ingestionSchema.getIOConfig().isAppendToExisting() ? new NumberedShardSpec(shardSpec.getPartitionNum(), shardSpecs.get(interval).size()) : shardSpec;
                        sequenceNameToShardSpecMap.put(sequenceName, shardSpecForPublishing);
                    }
                    final SegmentIdentifier identifier = driver.add(inputRow, sequenceName, committerSupplier);
                    if (identifier == null) {
                        throw new ISE("Could not allocate segment for row with timestamp[%s]", inputRow.getTimestamp());
                    }
                    fireDepartmentMetrics.incrementProcessed();
                } catch (ParseException e) {
                    if (ingestionSchema.getTuningConfig().isReportParseExceptions()) {
                        throw e;
                    } else {
                        fireDepartmentMetrics.incrementUnparseable();
                    }
                }
            }
        } finally {
            driver.persist(committerSupplier.get());
        }
        final TransactionalSegmentPublisher publisher = new TransactionalSegmentPublisher() {

            @Override
            public boolean publishSegments(Set<DataSegment> segments, Object commitMetadata) throws IOException {
                final SegmentTransactionalInsertAction action = new SegmentTransactionalInsertAction(segments, null, null);
                return toolbox.getTaskActionClient().submit(action).isSuccess();
            }
        };
        final SegmentsAndMetadata published = driver.finish(publisher, committerSupplier.get());
        if (published == null) {
            log.error("Failed to publish segments, aborting!");
            return false;
        } else {
            log.info("Published segments[%s]", Joiner.on(", ").join(Iterables.transform(published.getSegments(), new Function<DataSegment, String>() {

                @Override
                public String apply(DataSegment input) {
                    return input.getIdentifier();
                }
            })));
            return true;
        }
    }
}

Example 3 with SegmentsAndMetadata

use of io.druid.segment.realtime.appenderator.SegmentsAndMetadata in project hive by apache.

the class DruidRecordWriter method pushSegments.

private void pushSegments(List<SegmentIdentifier> segmentsToPush) {
    try {
        SegmentsAndMetadata segmentsAndMetadata = appenderator.push(segmentsToPush, committerSupplier.get()).get();
        final HashSet<String> pushedSegmentIdentifierHashSet = new HashSet<>();
        for (DataSegment pushedSegment : segmentsAndMetadata.getSegments()) {
            pushedSegmentIdentifierHashSet.add(SegmentIdentifier.fromDataSegment(pushedSegment).getIdentifierAsString());
            final Path segmentDescriptorOutputPath = DruidStorageHandlerUtils.makeSegmentDescriptorOutputPath(pushedSegment, segmentsDescriptorDir);
            DruidStorageHandlerUtils.writeSegmentDescriptor(fileSystem, pushedSegment, segmentDescriptorOutputPath);
            LOG.info(String.format("Pushed the segment [%s] and persisted the descriptor located at [%s]", pushedSegment, segmentDescriptorOutputPath));
        }
        final HashSet<String> toPushSegmentsHashSet = new HashSet(FluentIterable.from(segmentsToPush).transform(new Function<SegmentIdentifier, String>() {

            @Nullable
            @Override
            public String apply(@Nullable SegmentIdentifier input) {
                return input.getIdentifierAsString();
            }
        }).toList());
        if (!pushedSegmentIdentifierHashSet.equals(toPushSegmentsHashSet)) {
            throw new IllegalStateException(String.format("was asked to publish [%s] but was able to publish only [%s]", Joiner.on(", ").join(toPushSegmentsHashSet), Joiner.on(", ").join(pushedSegmentIdentifierHashSet)));
        }
        for (SegmentIdentifier dataSegmentId : segmentsToPush) {
            LOG.info("Dropping segment {}", dataSegmentId.toString());
            appenderator.drop(dataSegmentId).get();
        }
        LOG.info(String.format("Published [%,d] segments.", segmentsToPush.size()));
    } catch (InterruptedException e) {
        LOG.error(String.format("got interrupted, failed to push  [%,d] segments.", segmentsToPush.size()), e);
        Thread.currentThread().interrupt();
    } catch (IOException | ExecutionException e) {
        LOG.error(String.format("Failed to push  [%,d] segments.", segmentsToPush.size()), e);
        Throwables.propagate(e);
    }
}