use of org.apache.druid.segment.realtime.FireHydrant in project druid by druid-io.
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();
}
use of org.apache.druid.segment.realtime.FireHydrant in project druid by druid-io.
the class StreamAppenderator 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) {
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);
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);
}
}
use of org.apache.druid.segment.realtime.FireHydrant in project druid by druid-io.
the class RealtimePlumber method abandonSegment.
/**
* Unannounces a given sink and removes all local references to it. It is important that this is only called
* from the single-threaded mergeExecutor, since otherwise chaos may ensue if merged segments are deleted while
* being created.
*
* @param truncatedTime sink key
* @param sink sink to unannounce
*/
protected void abandonSegment(final long truncatedTime, final Sink sink) {
if (sinks.containsKey(truncatedTime)) {
try {
segmentAnnouncer.unannounceSegment(sink.getSegment());
removeSegment(sink, computePersistDir(schema, sink.getInterval()));
log.info("Removing sinkKey %d for segment %s", truncatedTime, sink.getSegment().getId());
sinks.remove(truncatedTime);
metrics.setSinkCount(sinks.size());
sinkTimeline.remove(sink.getInterval(), sink.getVersion(), new SingleElementPartitionChunk<>(sink));
for (FireHydrant hydrant : sink) {
cache.close(SinkQuerySegmentWalker.makeHydrantCacheIdentifier(hydrant));
hydrant.swapSegment(null);
}
synchronized (handoffCondition) {
handoffCondition.notifyAll();
}
} catch (Exception e) {
log.makeAlert(e, "Unable to abandon old segment for dataSource[%s]", schema.getDataSource()).addData("interval", sink.getInterval()).emit();
}
}
}
use of org.apache.druid.segment.realtime.FireHydrant in project druid by druid-io.
the class Sink method makeNewCurrIndex.
private FireHydrant makeNewCurrIndex(long minTimestamp, DataSchema schema) {
final IncrementalIndexSchema indexSchema = new IncrementalIndexSchema.Builder().withMinTimestamp(minTimestamp).withTimestampSpec(schema.getTimestampSpec()).withQueryGranularity(schema.getGranularitySpec().getQueryGranularity()).withDimensionsSpec(schema.getDimensionsSpec()).withMetrics(schema.getAggregators()).withRollup(schema.getGranularitySpec().isRollup()).build();
// Build the incremental-index according to the spec that was chosen by the user
final IncrementalIndex newIndex = appendableIndexSpec.builder().setIndexSchema(indexSchema).setMaxRowCount(maxRowsInMemory).setMaxBytesInMemory(maxBytesInMemory).setUseMaxMemoryEstimates(useMaxMemoryEstimates).build();
final FireHydrant old;
synchronized (hydrantLock) {
if (writable) {
old = currHydrant;
int newCount = 0;
int numHydrants = hydrants.size();
if (numHydrants > 0) {
FireHydrant lastHydrant = hydrants.get(numHydrants - 1);
newCount = lastHydrant.getCount() + 1;
if (!indexSchema.getDimensionsSpec().hasCustomDimensions()) {
Map<String, ColumnCapabilities> oldCapabilities;
if (lastHydrant.hasSwapped()) {
oldCapabilities = new HashMap<>();
ReferenceCountingSegment segment = lastHydrant.getIncrementedSegment();
try {
QueryableIndex oldIndex = segment.asQueryableIndex();
for (String dim : oldIndex.getAvailableDimensions()) {
dimOrder.add(dim);
oldCapabilities.put(dim, oldIndex.getColumnHolder(dim).getCapabilities());
}
} finally {
segment.decrement();
}
} else {
IncrementalIndex oldIndex = lastHydrant.getIndex();
dimOrder.addAll(oldIndex.getDimensionOrder());
oldCapabilities = oldIndex.getColumnCapabilities();
}
newIndex.loadDimensionIterable(dimOrder, oldCapabilities);
}
}
currHydrant = new FireHydrant(newIndex, newCount, getSegment().getId());
if (old != null) {
numRowsExcludingCurrIndex.addAndGet(old.getIndex().size());
}
hydrants.add(currHydrant);
} else {
// Oops, someone called finishWriting while we were making this new index.
newIndex.close();
throw new ISE("finishWriting() called during swap");
}
}
return old;
}
use of org.apache.druid.segment.realtime.FireHydrant in project druid by druid-io.
the class RealtimePlumberSchoolTest method testPersistHydrantGapsHelper.
private void testPersistHydrantGapsHelper(final Object commitMetadata) throws Exception {
Interval testInterval = new Interval(DateTimes.of("1970-01-01"), DateTimes.of("1971-01-01"));
RealtimePlumber plumber2 = (RealtimePlumber) realtimePlumberSchool.findPlumber(schema2, tuningConfig, metrics);
Sink sink = new Sink(testInterval, schema2, tuningConfig.getShardSpec(), DateTimes.of("2014-12-01T12:34:56.789").toString(), tuningConfig.getAppendableIndexSpec(), tuningConfig.getMaxRowsInMemory(), tuningConfig.getMaxBytesInMemoryOrDefault(), true, tuningConfig.getDedupColumn());
plumber2.getSinks().put(0L, sink);
Assert.assertNull(plumber2.startJob());
final CountDownLatch doneSignal = new CountDownLatch(1);
final Committer committer = new Committer() {
@Override
public Object getMetadata() {
return commitMetadata;
}
@Override
public void run() {
doneSignal.countDown();
}
};
plumber2.add(getTestInputRow("1970-01-01"), Suppliers.ofInstance(committer));
plumber2.add(getTestInputRow("1970-02-01"), Suppliers.ofInstance(committer));
plumber2.add(getTestInputRow("1970-03-01"), Suppliers.ofInstance(committer));
plumber2.add(getTestInputRow("1970-04-01"), Suppliers.ofInstance(committer));
plumber2.add(getTestInputRow("1970-05-01"), Suppliers.ofInstance(committer));
plumber2.persist(committer);
doneSignal.await();
plumber2.getSinks().clear();
plumber2.finishJob();
File persistDir = plumber2.computePersistDir(schema2, testInterval);
/* Check that all hydrants were persisted */
for (int i = 0; i < 5; i++) {
Assert.assertTrue(new File(persistDir, String.valueOf(i)).exists());
}
/* Create some gaps in the persisted hydrants and reload */
FileUtils.deleteDirectory(new File(persistDir, "1"));
FileUtils.deleteDirectory(new File(persistDir, "3"));
RealtimePlumber restoredPlumber = (RealtimePlumber) realtimePlumberSchool.findPlumber(schema2, tuningConfig, metrics);
restoredPlumber.bootstrapSinksFromDisk();
Map<Long, Sink> sinks = restoredPlumber.getSinks();
Assert.assertEquals(1, sinks.size());
List<FireHydrant> hydrants = Lists.newArrayList(sinks.get(new Long(0)));
DateTime startTime = DateTimes.of("1970-01-01T00:00:00.000Z");
Interval expectedInterval = new Interval(startTime, DateTimes.of("1971-01-01T00:00:00.000Z"));
Assert.assertEquals(0, hydrants.get(0).getCount());
Assert.assertEquals(expectedInterval, hydrants.get(0).getSegmentDataInterval());
Assert.assertEquals(2, hydrants.get(1).getCount());
Assert.assertEquals(expectedInterval, hydrants.get(1).getSegmentDataInterval());
Assert.assertEquals(4, hydrants.get(2).getCount());
Assert.assertEquals(expectedInterval, hydrants.get(2).getSegmentDataInterval());
/* Delete all the hydrants and reload, no sink should be created */
FileUtils.deleteDirectory(new File(persistDir, "0"));
FileUtils.deleteDirectory(new File(persistDir, "2"));
FileUtils.deleteDirectory(new File(persistDir, "4"));
RealtimePlumber restoredPlumber2 = (RealtimePlumber) realtimePlumberSchool.findPlumber(schema2, tuningConfig, metrics);
restoredPlumber2.bootstrapSinksFromDisk();
Assert.assertEquals(0, restoredPlumber2.getSinks().size());
}
Aggregations