Examples with StreamPartition org.apache.druid.indexing.seekablestream.common.StreamPartition used on opensource projects

Example 11 with StreamPartition

use of org.apache.druid.indexing.seekablestream.common.StreamPartition in project druid by druid-io.

the class SeekableStreamIndexTaskRunner method runInternal.

private TaskStatus runInternal(TaskToolbox toolbox) throws Exception {
    startTime = DateTimes.nowUtc();
    status = Status.STARTING;
    setToolbox(toolbox);
    authorizerMapper = toolbox.getAuthorizerMapper();
    rowIngestionMeters = toolbox.getRowIngestionMetersFactory().createRowIngestionMeters();
    parseExceptionHandler = new ParseExceptionHandler(rowIngestionMeters, tuningConfig.isLogParseExceptions(), tuningConfig.getMaxParseExceptions(), tuningConfig.getMaxSavedParseExceptions());
    // Now we can initialize StreamChunkReader with the given toolbox.
    final StreamChunkParser parser = new StreamChunkParser<RecordType>(this.parser, inputFormat, inputRowSchema, task.getDataSchema().getTransformSpec(), toolbox.getIndexingTmpDir(), row -> row != null && task.withinMinMaxRecordTime(row), rowIngestionMeters, parseExceptionHandler);
    initializeSequences();
    log.debug("Found chat handler of class[%s]", toolbox.getChatHandlerProvider().getClass().getName());
    toolbox.getChatHandlerProvider().register(task.getId(), this, false);
    runThread = Thread.currentThread();
    // Set up FireDepartmentMetrics
    final FireDepartment fireDepartmentForMetrics = new FireDepartment(task.getDataSchema(), new RealtimeIOConfig(null, null), null);
    this.fireDepartmentMetrics = fireDepartmentForMetrics.getMetrics();
    toolbox.addMonitor(TaskRealtimeMetricsMonitorBuilder.build(task, fireDepartmentForMetrics, rowIngestionMeters));
    final String lookupTier = task.getContextValue(RealtimeIndexTask.CTX_KEY_LOOKUP_TIER);
    final LookupNodeService lookupNodeService = lookupTier == null ? toolbox.getLookupNodeService() : new LookupNodeService(lookupTier);
    final DiscoveryDruidNode discoveryDruidNode = new DiscoveryDruidNode(toolbox.getDruidNode(), NodeRole.PEON, ImmutableMap.of(toolbox.getDataNodeService().getName(), toolbox.getDataNodeService(), lookupNodeService.getName(), lookupNodeService));
    Throwable caughtExceptionOuter = null;
    // milliseconds waited for created segments to be handed off
    long handoffWaitMs = 0L;
    try (final RecordSupplier<PartitionIdType, SequenceOffsetType, RecordType> recordSupplier = task.newTaskRecordSupplier()) {
        if (toolbox.getAppenderatorsManager().shouldTaskMakeNodeAnnouncements()) {
            toolbox.getDataSegmentServerAnnouncer().announce();
            toolbox.getDruidNodeAnnouncer().announce(discoveryDruidNode);
        }
        appenderator = task.newAppenderator(toolbox, fireDepartmentMetrics, rowIngestionMeters, parseExceptionHandler);
        driver = task.newDriver(appenderator, toolbox, fireDepartmentMetrics);
        // Start up, set up initial sequences.
        final Object restoredMetadata = driver.startJob(segmentId -> {
            try {
                if (lockGranularityToUse == LockGranularity.SEGMENT) {
                    return toolbox.getTaskActionClient().submit(new SegmentLockAcquireAction(TaskLockType.EXCLUSIVE, segmentId.getInterval(), segmentId.getVersion(), segmentId.getShardSpec().getPartitionNum(), 1000L)).isOk();
                } else {
                    final TaskLock lock = toolbox.getTaskActionClient().submit(new TimeChunkLockAcquireAction(TaskLockType.EXCLUSIVE, segmentId.getInterval(), 1000L));
                    if (lock == null) {
                        return false;
                    }
                    if (lock.isRevoked()) {
                        throw new ISE(StringUtils.format("Lock for interval [%s] was revoked.", segmentId.getInterval()));
                    }
                    return true;
                }
            } catch (IOException e) {
                throw new RuntimeException(e);
            }
        });
        if (restoredMetadata == null) {
            // no persist has happened so far
            // so either this is a brand new task or replacement of a failed task
            Preconditions.checkState(sequences.get(0).startOffsets.entrySet().stream().allMatch(partitionOffsetEntry -> createSequenceNumber(partitionOffsetEntry.getValue()).compareTo(createSequenceNumber(ioConfig.getStartSequenceNumbers().getPartitionSequenceNumberMap().get(partitionOffsetEntry.getKey()))) >= 0), "Sequence sequences are not compatible with start sequences of task");
            currOffsets.putAll(sequences.get(0).startOffsets);
        } else {
            @SuppressWarnings("unchecked") final Map<String, Object> restoredMetadataMap = (Map) restoredMetadata;
            final SeekableStreamEndSequenceNumbers<PartitionIdType, SequenceOffsetType> restoredNextPartitions = deserializePartitionsFromMetadata(toolbox.getJsonMapper(), restoredMetadataMap.get(METADATA_NEXT_PARTITIONS));
            currOffsets.putAll(restoredNextPartitions.getPartitionSequenceNumberMap());
            // Sanity checks.
            if (!restoredNextPartitions.getStream().equals(ioConfig.getStartSequenceNumbers().getStream())) {
                throw new ISE("Restored stream[%s] but expected stream[%s]", restoredNextPartitions.getStream(), ioConfig.getStartSequenceNumbers().getStream());
            }
            if (!currOffsets.keySet().equals(ioConfig.getStartSequenceNumbers().getPartitionSequenceNumberMap().keySet())) {
                throw new ISE("Restored partitions[%s] but expected partitions[%s]", currOffsets.keySet(), ioConfig.getStartSequenceNumbers().getPartitionSequenceNumberMap().keySet());
            }
            // which is super rare
            if (sequences.size() == 0 || getLastSequenceMetadata().isCheckpointed()) {
                this.endOffsets.putAll(sequences.size() == 0 ? currOffsets : getLastSequenceMetadata().getEndOffsets());
            }
        }
        log.info("Initialized sequences: %s", sequences.stream().map(SequenceMetadata::toString).collect(Collectors.joining(", ")));
        // Filter out partitions with END_OF_SHARD markers since these partitions have already been fully read. This
        // should have been done by the supervisor already so this is defensive.
        int numPreFilterPartitions = currOffsets.size();
        if (currOffsets.entrySet().removeIf(x -> isEndOfShard(x.getValue()))) {
            log.info("Removed [%d] partitions from assignment which have already been closed.", numPreFilterPartitions - currOffsets.size());
        }
        // When end offsets are exclusive, we never skip the start record.
        if (!isEndOffsetExclusive()) {
            for (Map.Entry<PartitionIdType, SequenceOffsetType> entry : currOffsets.entrySet()) {
                final boolean isAtStart = entry.getValue().equals(ioConfig.getStartSequenceNumbers().getPartitionSequenceNumberMap().get(entry.getKey()));
                if (!isAtStart || ioConfig.getStartSequenceNumbers().getExclusivePartitions().contains(entry.getKey())) {
                    lastReadOffsets.put(entry.getKey(), entry.getValue());
                }
            }
        }
        // Set up committer.
        final Supplier<Committer> committerSupplier = () -> {
            final Map<PartitionIdType, SequenceOffsetType> snapshot = ImmutableMap.copyOf(currOffsets);
            lastPersistedOffsets.clear();
            lastPersistedOffsets.putAll(snapshot);
            return new Committer() {

                @Override
                public Object getMetadata() {
                    return ImmutableMap.of(METADATA_NEXT_PARTITIONS, new SeekableStreamEndSequenceNumbers<>(stream, snapshot));
                }

                @Override
                public void run() {
                // Do nothing.
                }
            };
        };
        // restart publishing of sequences (if any)
        maybePersistAndPublishSequences(committerSupplier);
        Set<StreamPartition<PartitionIdType>> assignment = assignPartitions(recordSupplier);
        possiblyResetDataSourceMetadata(toolbox, recordSupplier, assignment);
        seekToStartingSequence(recordSupplier, assignment);
        ingestionState = IngestionState.BUILD_SEGMENTS;
        // Main loop.
        // Could eventually support leader/follower mode (for keeping replicas more in sync)
        boolean stillReading = !assignment.isEmpty();
        status = Status.READING;
        Throwable caughtExceptionInner = null;
        try {
            while (stillReading) {
                if (possiblyPause()) {
                    // The partition assignments may have changed while paused by a call to setEndOffsets() so reassign
                    // partitions upon resuming. Don't call "seekToStartingSequence" after "assignPartitions", because there's
                    // no need to re-seek here. All we're going to be doing is dropping partitions.
                    assignment = assignPartitions(recordSupplier);
                    possiblyResetDataSourceMetadata(toolbox, recordSupplier, assignment);
                    if (assignment.isEmpty()) {
                        log.debug("All partitions have been fully read.");
                        publishOnStop.set(true);
                        stopRequested.set(true);
                    }
                }
                // if stop is requested or task's end sequence is set by call to setEndOffsets method with finish set to true
                if (stopRequested.get() || sequences.size() == 0 || getLastSequenceMetadata().isCheckpointed()) {
                    status = Status.PUBLISHING;
                }
                if (stopRequested.get()) {
                    break;
                }
                if (backgroundThreadException != null) {
                    throw new RuntimeException(backgroundThreadException);
                }
                checkPublishAndHandoffFailure();
                maybePersistAndPublishSequences(committerSupplier);
                // calling getRecord() ensures that exceptions specific to kafka/kinesis like OffsetOutOfRangeException
                // are handled in the subclasses.
                List<OrderedPartitionableRecord<PartitionIdType, SequenceOffsetType, RecordType>> records = getRecords(recordSupplier, toolbox);
                // note: getRecords() also updates assignment
                stillReading = !assignment.isEmpty();
                SequenceMetadata<PartitionIdType, SequenceOffsetType> sequenceToCheckpoint = null;
                for (OrderedPartitionableRecord<PartitionIdType, SequenceOffsetType, RecordType> record : records) {
                    final boolean shouldProcess = verifyRecordInRange(record.getPartitionId(), record.getSequenceNumber());
                    log.trace("Got stream[%s] partition[%s] sequenceNumber[%s], shouldProcess[%s].", record.getStream(), record.getPartitionId(), record.getSequenceNumber(), shouldProcess);
                    if (shouldProcess) {
                        final List<InputRow> rows = parser.parse(record.getData(), isEndOfShard(record.getSequenceNumber()));
                        boolean isPersistRequired = false;
                        final SequenceMetadata<PartitionIdType, SequenceOffsetType> sequenceToUse = sequences.stream().filter(sequenceMetadata -> sequenceMetadata.canHandle(this, record)).findFirst().orElse(null);
                        if (sequenceToUse == null) {
                            throw new ISE("Cannot find any valid sequence for record with partition [%s] and sequenceNumber [%s]. Current sequences: %s", record.getPartitionId(), record.getSequenceNumber(), sequences);
                        }
                        for (InputRow row : rows) {
                            final AppenderatorDriverAddResult addResult = driver.add(row, sequenceToUse.getSequenceName(), committerSupplier, true, // of rows are indexed
                            false);
                            if (addResult.isOk()) {
                                // If the number of rows in the segment exceeds the threshold after adding a row,
                                // move the segment out from the active segments of BaseAppenderatorDriver to make a new segment.
                                final boolean isPushRequired = addResult.isPushRequired(tuningConfig.getPartitionsSpec().getMaxRowsPerSegment(), tuningConfig.getPartitionsSpec().getMaxTotalRowsOr(DynamicPartitionsSpec.DEFAULT_MAX_TOTAL_ROWS));
                                if (isPushRequired && !sequenceToUse.isCheckpointed()) {
                                    sequenceToCheckpoint = sequenceToUse;
                                }
                                isPersistRequired |= addResult.isPersistRequired();
                            } else {
                                // 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());
                            }
                        }
                        if (isPersistRequired) {
                            Futures.addCallback(driver.persistAsync(committerSupplier.get()), new FutureCallback<Object>() {

                                @Override
                                public void onSuccess(@Nullable Object result) {
                                    log.debug("Persist completed with metadata: %s", result);
                                }

                                @Override
                                public void onFailure(Throwable t) {
                                    log.error("Persist failed, dying");
                                    backgroundThreadException = t;
                                }
                            });
                        }
                        // in kafka, we can easily get the next offset by adding 1, but for kinesis, there's no way
                        // to get the next sequence number without having to make an expensive api call. So the behavior
                        // here for kafka is to +1 while for kinesis we simply save the current sequence number
                        lastReadOffsets.put(record.getPartitionId(), record.getSequenceNumber());
                        currOffsets.put(record.getPartitionId(), getNextStartOffset(record.getSequenceNumber()));
                    }
                    // Use record.getSequenceNumber() in the moreToRead check, since currOffsets might not have been
                    // updated if we were skipping records for being beyond the end.
                    final boolean moreToReadAfterThisRecord = isMoreToReadAfterReadingRecord(record.getSequenceNumber(), endOffsets.get(record.getPartitionId()));
                    if (!moreToReadAfterThisRecord && assignment.remove(record.getStreamPartition())) {
                        log.info("Finished reading stream[%s], partition[%s].", record.getStream(), record.getPartitionId());
                        recordSupplier.assign(assignment);
                        stillReading = !assignment.isEmpty();
                    }
                }
                if (!stillReading) {
                    // We let the fireDepartmentMetrics know that all messages have been read. This way, some metrics such as
                    // high message gap need not be reported
                    fireDepartmentMetrics.markProcessingDone();
                }
                if (System.currentTimeMillis() > nextCheckpointTime) {
                    sequenceToCheckpoint = getLastSequenceMetadata();
                }
                if (sequenceToCheckpoint != null && stillReading) {
                    Preconditions.checkArgument(getLastSequenceMetadata().getSequenceName().equals(sequenceToCheckpoint.getSequenceName()), "Cannot checkpoint a sequence [%s] which is not the latest one, sequences %s", sequenceToCheckpoint, sequences);
                    requestPause();
                    final CheckPointDataSourceMetadataAction checkpointAction = new CheckPointDataSourceMetadataAction(task.getDataSource(), ioConfig.getTaskGroupId(), null, createDataSourceMetadata(new SeekableStreamStartSequenceNumbers<>(stream, sequenceToCheckpoint.getStartOffsets(), sequenceToCheckpoint.getExclusiveStartPartitions())));
                    if (!toolbox.getTaskActionClient().submit(checkpointAction)) {
                        throw new ISE("Checkpoint request with sequences [%s] failed, dying", currOffsets);
                    }
                }
            }
            ingestionState = IngestionState.COMPLETED;
        } catch (Exception e) {
            // (1) catch all exceptions while reading from kafka
            caughtExceptionInner = e;
            log.error(e, "Encountered exception in run() before persisting.");
            throw e;
        } finally {
            try {
                // persist pending data
                driver.persist(committerSupplier.get());
            } catch (Exception e) {
                if (caughtExceptionInner != null) {
                    caughtExceptionInner.addSuppressed(e);
                } else {
                    throw e;
                }
            }
        }
        synchronized (statusLock) {
            if (stopRequested.get() && !publishOnStop.get()) {
                throw new InterruptedException("Stopping without publishing");
            }
            status = Status.PUBLISHING;
        }
        // We need to copy sequences here, because the success callback in publishAndRegisterHandoff removes items from
        // the sequence list. If a publish finishes before we finish iterating through the sequence list, we can
        // end up skipping some sequences.
        List<SequenceMetadata<PartitionIdType, SequenceOffsetType>> sequencesSnapshot = new ArrayList<>(sequences);
        for (int i = 0; i < sequencesSnapshot.size(); i++) {
            final SequenceMetadata<PartitionIdType, SequenceOffsetType> sequenceMetadata = sequencesSnapshot.get(i);
            if (!publishingSequences.contains(sequenceMetadata.getSequenceName())) {
                final boolean isLast = i == (sequencesSnapshot.size() - 1);
                if (isLast) {
                    // Shorten endOffsets of the last sequence to match currOffsets.
                    sequenceMetadata.setEndOffsets(currOffsets);
                }
                // Update assignments of the sequence, which should clear them. (This will be checked later, when the
                // Committer is built.)
                sequenceMetadata.updateAssignments(currOffsets, this::isMoreToReadAfterReadingRecord);
                publishingSequences.add(sequenceMetadata.getSequenceName());
                // persist already done in finally, so directly add to publishQueue
                publishAndRegisterHandoff(sequenceMetadata);
            }
        }
        if (backgroundThreadException != null) {
            throw new RuntimeException(backgroundThreadException);
        }
        // Wait for publish futures to complete.
        Futures.allAsList(publishWaitList).get();
        // Wait for handoff futures to complete.
        // Note that every publishing task (created by calling AppenderatorDriver.publish()) has a corresponding
        // handoffFuture. handoffFuture can throw an exception if 1) the corresponding publishFuture failed or 2) it
        // failed to persist sequences. It might also return null if handoff failed, but was recoverable.
        // See publishAndRegisterHandoff() for details.
        List<SegmentsAndCommitMetadata> handedOffList = Collections.emptyList();
        if (tuningConfig.getHandoffConditionTimeout() == 0) {
            handedOffList = Futures.allAsList(handOffWaitList).get();
        } else {
            final long start = System.nanoTime();
            try {
                handedOffList = Futures.allAsList(handOffWaitList).get(tuningConfig.getHandoffConditionTimeout(), TimeUnit.MILLISECONDS);
            } catch (TimeoutException e) {
                // Handoff timeout is not an indexing failure, but coordination failure. We simply ignore timeout exception
                // here.
                log.makeAlert("Timeout waiting for handoff").addData("taskId", task.getId()).addData("handoffConditionTimeout", tuningConfig.getHandoffConditionTimeout()).emit();
            } finally {
                handoffWaitMs = TimeUnit.NANOSECONDS.toMillis(System.nanoTime() - start);
            }
        }
        for (SegmentsAndCommitMetadata handedOff : handedOffList) {
            log.info("Handoff complete for segments: %s", String.join(", ", Lists.transform(handedOff.getSegments(), DataSegment::toString)));
        }
        appenderator.close();
    } catch (InterruptedException | RejectedExecutionException e) {
        // (2) catch InterruptedException and RejectedExecutionException thrown for the whole ingestion steps including
        // the final publishing.
        caughtExceptionOuter = e;
        try {
            Futures.allAsList(publishWaitList).cancel(true);
            Futures.allAsList(handOffWaitList).cancel(true);
            if (appenderator != null) {
                appenderator.closeNow();
            }
        } catch (Exception e2) {
            e.addSuppressed(e2);
        }
        // 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.get()) {
            Thread.currentThread().interrupt();
            throw e;
        }
    } catch (Exception e) {
        // (3) catch all other exceptions thrown for the whole ingestion steps including the final publishing.
        caughtExceptionOuter = e;
        try {
            Futures.allAsList(publishWaitList).cancel(true);
            Futures.allAsList(handOffWaitList).cancel(true);
            if (appenderator != null) {
                appenderator.closeNow();
            }
        } catch (Exception e2) {
            e.addSuppressed(e2);
        }
        throw e;
    } finally {
        try {
            if (driver != null) {
                driver.close();
            }
            toolbox.getChatHandlerProvider().unregister(task.getId());
            if (toolbox.getAppenderatorsManager().shouldTaskMakeNodeAnnouncements()) {
                toolbox.getDruidNodeAnnouncer().unannounce(discoveryDruidNode);
                toolbox.getDataSegmentServerAnnouncer().unannounce();
            }
        } catch (Throwable e) {
            if (caughtExceptionOuter != null) {
                caughtExceptionOuter.addSuppressed(e);
            } else {
                throw e;
            }
        }
    }
    toolbox.getTaskReportFileWriter().write(task.getId(), getTaskCompletionReports(null, handoffWaitMs));
    return TaskStatus.success(task.getId());
}

Example 12 with StreamPartition

use of org.apache.druid.indexing.seekablestream.common.StreamPartition in project druid by druid-io.

the class SeekableStreamIndexTaskRunner method sendResetRequestAndWait.

protected void sendResetRequestAndWait(Map<StreamPartition<PartitionIdType>, SequenceOffsetType> outOfRangePartitions, TaskToolbox taskToolbox) throws IOException {
    Map<PartitionIdType, SequenceOffsetType> partitionOffsetMap = CollectionUtils.mapKeys(outOfRangePartitions, StreamPartition::getPartitionId);
    boolean result = taskToolbox.getTaskActionClient().submit(new ResetDataSourceMetadataAction(task.getDataSource(), createDataSourceMetadata(new SeekableStreamEndSequenceNumbers<>(ioConfig.getStartSequenceNumbers().getStream(), partitionOffsetMap))));
    if (result) {
        log.makeAlert("Offsets were reset automatically, potential data duplication or loss").addData("task", task.getId()).addData("dataSource", task.getDataSource()).addData("partitions", partitionOffsetMap.keySet()).emit();
        requestPause();
    } else {
        log.makeAlert("Failed to send offset reset request").addData("task", task.getId()).addData("dataSource", task.getDataSource()).addData("partitions", ImmutableSet.copyOf(partitionOffsetMap.keySet())).emit();
    }
}