Examples with ThreadRenamingRunnable org.apache.druid.common.guava.ThreadRenamingRunnable used on opensource projects

Example 1 with ThreadRenamingRunnable

use of org.apache.druid.common.guava.ThreadRenamingRunnable in project druid by druid-io.

the class RealtimePlumber method persist.

@Override
public void persist(final Committer committer) {
    final List<Pair<FireHydrant, Interval>> indexesToPersist = new ArrayList<>();
    for (Sink sink : sinks.values()) {
        if (sink.swappable()) {
            indexesToPersist.add(Pair.of(sink.swap(), sink.getInterval()));
        }
    }
    log.info("Submitting persist runnable for dataSource[%s]", schema.getDataSource());
    final Stopwatch runExecStopwatch = Stopwatch.createStarted();
    final Stopwatch persistStopwatch = Stopwatch.createStarted();
    final Map<String, Object> metadataElems = committer.getMetadata() == null ? null : ImmutableMap.of(COMMIT_METADATA_KEY, committer.getMetadata(), COMMIT_METADATA_TIMESTAMP_KEY, System.currentTimeMillis());
    persistExecutor.execute(new ThreadRenamingRunnable(StringUtils.format("%s-incremental-persist", schema.getDataSource())) {

        @Override
        public void doRun() {
            /* Note:
            If plumber crashes after storing a subset of all the hydrants then we will lose data and next
            time we will start with the commitMetadata stored in those hydrants.
            option#1:
            maybe it makes sense to store the metadata outside the segments in a separate file. This is because the
            commit metadata isn't really associated with an individual segment-- it's associated with a set of segments
            that are persisted at the same time or maybe whole datasource. So storing it in the segments is asking for problems.
            Sort of like this:

            {
              "metadata" : {"foo": "bar"},
              "segments": [
                {"id": "datasource_2000_2001_2000_1", "hydrant": 10},
                {"id": "datasource_2001_2002_2001_1", "hydrant": 12},
              ]
            }
            When a realtime node crashes and starts back up, it would delete any hydrants numbered higher than the
            ones in the commit file.

            option#2
            We could also just include the set of segments for the same chunk of metadata in more metadata on each
            of the segments. we might also have to think about the hand-off in terms of the full set of segments being
            handed off instead of individual segments being handed off (that is, if one of the set succeeds in handing
            off and the others fail, the real-time would believe that it needs to re-ingest the data).
             */
            long persistThreadCpuTime = JvmUtils.safeGetThreadCpuTime();
            try {
                for (Pair<FireHydrant, Interval> pair : indexesToPersist) {
                    metrics.incrementRowOutputCount(persistHydrant(pair.lhs, schema, pair.rhs, metadataElems));
                }
                committer.run();
            } catch (Exception e) {
                metrics.incrementFailedPersists();
                throw e;
            } finally {
                metrics.incrementPersistCpuTime(JvmUtils.safeGetThreadCpuTime() - persistThreadCpuTime);
                metrics.incrementNumPersists();
                metrics.incrementPersistTimeMillis(persistStopwatch.elapsed(TimeUnit.MILLISECONDS));
                persistStopwatch.stop();
            }
        }
    });
    final long startDelay = runExecStopwatch.elapsed(TimeUnit.MILLISECONDS);
    metrics.incrementPersistBackPressureMillis(startDelay);
    if (startDelay > WARN_DELAY) {
        log.warn("Ingestion was throttled for [%,d] millis because persists were pending.", startDelay);
    }
    runExecStopwatch.stop();
    resetNextFlush();
}

Example 2 with ThreadRenamingRunnable

use of org.apache.druid.common.guava.ThreadRenamingRunnable in project druid by druid-io.

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();
        }
    });
}