use of io.druid.segment.QueryableIndex in project druid by druid-io.
the class DruidSchemaTest method setUp.
@Before
public void setUp() throws Exception {
Calcites.setSystemProperties();
final File tmpDir = temporaryFolder.newFolder();
final QueryableIndex index1 = IndexBuilder.create().tmpDir(new File(tmpDir, "1")).indexMerger(TestHelper.getTestIndexMergerV9()).schema(new IncrementalIndexSchema.Builder().withMetrics(new AggregatorFactory[] { new CountAggregatorFactory("cnt"), new DoubleSumAggregatorFactory("m1", "m1"), new HyperUniquesAggregatorFactory("unique_dim1", "dim1") }).withRollup(false).build()).rows(ROWS1).buildMMappedIndex();
final QueryableIndex index2 = IndexBuilder.create().tmpDir(new File(tmpDir, "2")).indexMerger(TestHelper.getTestIndexMergerV9()).schema(new IncrementalIndexSchema.Builder().withMetrics(new AggregatorFactory[] { new LongSumAggregatorFactory("m1", "m1") }).withRollup(false).build()).rows(ROWS2).buildMMappedIndex();
walker = new SpecificSegmentsQuerySegmentWalker(CalciteTests.queryRunnerFactoryConglomerate()).add(DataSegment.builder().dataSource(CalciteTests.DATASOURCE1).interval(new Interval("2000/P1Y")).version("1").shardSpec(new LinearShardSpec(0)).build(), index1).add(DataSegment.builder().dataSource(CalciteTests.DATASOURCE1).interval(new Interval("2001/P1Y")).version("1").shardSpec(new LinearShardSpec(0)).build(), index2).add(DataSegment.builder().dataSource(CalciteTests.DATASOURCE2).interval(index2.getDataInterval()).version("1").shardSpec(new LinearShardSpec(0)).build(), index2);
schema = new DruidSchema(walker, new TestServerInventoryView(walker.getSegments()), PLANNER_CONFIG_DEFAULT);
schema.start();
schema.awaitInitialization();
}
use of io.druid.segment.QueryableIndex in project druid by druid-io.
the class CalciteTests method createMockWalker.
public static SpecificSegmentsQuerySegmentWalker createMockWalker(final File tmpDir) {
final QueryableIndex index1 = IndexBuilder.create().tmpDir(new File(tmpDir, "1")).indexMerger(TestHelper.getTestIndexMergerV9()).schema(INDEX_SCHEMA).rows(ROWS1).buildMMappedIndex();
final QueryableIndex index2 = IndexBuilder.create().tmpDir(new File(tmpDir, "2")).indexMerger(TestHelper.getTestIndexMergerV9()).schema(INDEX_SCHEMA).rows(ROWS2).buildMMappedIndex();
return new SpecificSegmentsQuerySegmentWalker(queryRunnerFactoryConglomerate()).add(DataSegment.builder().dataSource(DATASOURCE1).interval(index1.getDataInterval()).version("1").shardSpec(new LinearShardSpec(0)).build(), index1).add(DataSegment.builder().dataSource(DATASOURCE2).interval(index2.getDataInterval()).version("1").shardSpec(new LinearShardSpec(0)).build(), index2);
}
use of io.druid.segment.QueryableIndex in project druid by druid-io.
the class SegmentAnalyzer method analyze.
public Map<String, ColumnAnalysis> analyze(Segment segment) {
Preconditions.checkNotNull(segment, "segment");
// index is null for incremental-index-based segments, but storageAdapter is always available
final QueryableIndex index = segment.asQueryableIndex();
final StorageAdapter storageAdapter = segment.asStorageAdapter();
// get length and column names from storageAdapter
final int length = storageAdapter.getNumRows();
final Set<String> columnNames = Sets.newHashSet();
Iterables.addAll(columnNames, storageAdapter.getAvailableDimensions());
Iterables.addAll(columnNames, storageAdapter.getAvailableMetrics());
Map<String, ColumnAnalysis> columns = Maps.newTreeMap();
for (String columnName : columnNames) {
final Column column = index == null ? null : index.getColumn(columnName);
final ColumnCapabilities capabilities = column != null ? column.getCapabilities() : storageAdapter.getColumnCapabilities(columnName);
final ColumnAnalysis analysis;
final ValueType type = capabilities.getType();
switch(type) {
case LONG:
analysis = analyzeNumericColumn(capabilities, length, Longs.BYTES);
break;
case FLOAT:
analysis = analyzeNumericColumn(capabilities, length, NUM_BYTES_IN_TEXT_FLOAT);
break;
case STRING:
if (index != null) {
analysis = analyzeStringColumn(capabilities, column);
} else {
analysis = analyzeStringColumn(capabilities, storageAdapter, columnName);
}
break;
case COMPLEX:
analysis = analyzeComplexColumn(capabilities, column, storageAdapter.getColumnTypeName(columnName));
break;
default:
log.warn("Unknown column type[%s].", type);
analysis = ColumnAnalysis.error(String.format("unknown_type_%s", type));
}
columns.put(columnName, analysis);
}
// Add time column too
ColumnCapabilities timeCapabilities = storageAdapter.getColumnCapabilities(Column.TIME_COLUMN_NAME);
if (timeCapabilities == null) {
timeCapabilities = new ColumnCapabilitiesImpl().setType(ValueType.LONG).setHasMultipleValues(false);
}
columns.put(Column.TIME_COLUMN_NAME, analyzeNumericColumn(timeCapabilities, length, NUM_BYTES_IN_TIMESTAMP));
return columns;
}
use of io.druid.segment.QueryableIndex 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 = String.format("%s-%s-persist-n-merge", schema.getDataSource(), new DateTime(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.wtf("Merged target[%s] exists?!", 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 = VMUtils.safeGetThreadCpuTime();
mergeStopwatch = Stopwatch.createStarted();
List<QueryableIndex> indexes = Lists.newArrayList();
for (FireHydrant fireHydrant : sink) {
Segment segment = fireHydrant.getSegment();
final QueryableIndex queryableIndex = segment.asQueryableIndex();
log.info("Adding hydrant[%s]", fireHydrant);
indexes.add(queryableIndex);
}
final File mergedFile = indexMerger.mergeQueryableIndex(indexes, schema.getGranularitySpec().isRollup(), schema.getAggregators(), mergedTarget, config.getIndexSpec());
// emit merge metrics before publishing segment
metrics.incrementMergeCpuTime(VMUtils.safeGetThreadCpuTime() - mergeThreadCpuTime);
metrics.incrementMergeTimeMillis(mergeStopwatch.elapsed(TimeUnit.MILLISECONDS));
QueryableIndex index = indexIO.loadIndex(mergedFile);
log.info("Pushing [%s] to deep storage", sink.getSegment().getIdentifier());
DataSegment segment = dataSegmentPusher.push(mergedFile, sink.getSegment().withDimensions(Lists.newArrayList(index.getAvailableDimensions())));
log.info("Inserting [%s] to the metadata store", sink.getSegment().getIdentifier());
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();
}
});
}
use of io.druid.segment.QueryableIndex in project druid by druid-io.
the class RealtimePlumber method bootstrapSinksFromDisk.
protected Object bootstrapSinksFromDisk() {
final VersioningPolicy versioningPolicy = config.getVersioningPolicy();
File baseDir = computeBaseDir(schema);
if (baseDir == null || !baseDir.exists()) {
return null;
}
File[] files = baseDir.listFiles();
if (files == null) {
return null;
}
Object metadata = null;
long latestCommitTime = 0;
for (File sinkDir : files) {
final Interval sinkInterval = new Interval(sinkDir.getName().replace("_", "/"));
//final File[] sinkFiles = sinkDir.listFiles();
// To avoid reading and listing of "merged" dir
final File[] sinkFiles = sinkDir.listFiles(new FilenameFilter() {
@Override
public boolean accept(File dir, String fileName) {
return !(Ints.tryParse(fileName) == null);
}
});
Arrays.sort(sinkFiles, new Comparator<File>() {
@Override
public int compare(File o1, File o2) {
try {
return Ints.compare(Integer.parseInt(o1.getName()), Integer.parseInt(o2.getName()));
} catch (NumberFormatException e) {
log.error(e, "Couldn't compare as numbers? [%s][%s]", o1, o2);
return o1.compareTo(o2);
}
}
});
boolean isCorrupted = false;
List<FireHydrant> hydrants = Lists.newArrayList();
for (File segmentDir : sinkFiles) {
log.info("Loading previously persisted segment at [%s]", segmentDir);
// If 100% sure that this is not needed, this check can be removed.
if (Ints.tryParse(segmentDir.getName()) == null) {
continue;
}
QueryableIndex queryableIndex = null;
try {
queryableIndex = indexIO.loadIndex(segmentDir);
} catch (IOException e) {
log.error(e, "Problem loading segmentDir from disk.");
isCorrupted = true;
}
if (isCorrupted) {
try {
File corruptSegmentDir = computeCorruptedFileDumpDir(segmentDir, schema);
log.info("Renaming %s to %s", segmentDir.getAbsolutePath(), corruptSegmentDir.getAbsolutePath());
FileUtils.copyDirectory(segmentDir, corruptSegmentDir);
FileUtils.deleteDirectory(segmentDir);
} catch (Exception e1) {
log.error(e1, "Failed to rename %s", segmentDir.getAbsolutePath());
}
//at some point.
continue;
}
Metadata segmentMetadata = queryableIndex.getMetadata();
if (segmentMetadata != null) {
Object timestampObj = segmentMetadata.get(COMMIT_METADATA_TIMESTAMP_KEY);
if (timestampObj != null) {
long timestamp = ((Long) timestampObj).longValue();
if (timestamp > latestCommitTime) {
log.info("Found metaData [%s] with latestCommitTime [%s] greater than previous recorded [%s]", queryableIndex.getMetadata(), timestamp, latestCommitTime);
latestCommitTime = timestamp;
metadata = queryableIndex.getMetadata().get(COMMIT_METADATA_KEY);
}
}
}
hydrants.add(new FireHydrant(new QueryableIndexSegment(DataSegment.makeDataSegmentIdentifier(schema.getDataSource(), sinkInterval.getStart(), sinkInterval.getEnd(), versioningPolicy.getVersion(sinkInterval), config.getShardSpec()), queryableIndex), Integer.parseInt(segmentDir.getName())));
}
if (hydrants.isEmpty()) {
// Probably encountered a corrupt sink directory
log.warn("Found persisted segment directory with no intermediate segments present at %s, skipping sink creation.", sinkDir.getAbsolutePath());
continue;
}
final Sink currSink = new Sink(sinkInterval, schema, config.getShardSpec(), versioningPolicy.getVersion(sinkInterval), config.getMaxRowsInMemory(), config.isReportParseExceptions(), hydrants);
addSink(currSink);
}
return metadata;
}
Aggregations