use of io.prestosql.orc.stream.OrcChunkLoader in project hetu-core by openlookeng.
the class StripeReader method createValueStreams.
private Map<StreamId, ValueInputStream<?>> createValueStreams(Map<StreamId, Stream> streams, Map<StreamId, OrcChunkLoader> streamsData, ColumnMetadata<ColumnEncoding> columnEncodings) {
ImmutableMap.Builder<StreamId, ValueInputStream<?>> valueStreams = ImmutableMap.builder();
for (Entry<StreamId, Stream> entry : streams.entrySet()) {
StreamId streamId = entry.getKey();
Stream stream = entry.getValue();
ColumnEncodingKind columnEncoding = columnEncodings.get(stream.getColumnId()).getColumnEncodingKind();
// skip index and empty streams
if (isIndexStream(stream) || stream.getLength() == 0) {
continue;
}
OrcChunkLoader chunkLoader = streamsData.get(streamId);
OrcTypeKind columnType = types.get(stream.getColumnId()).getOrcTypeKind();
valueStreams.put(streamId, ValueStreams.createValueStreams(streamId, chunkLoader, columnType, columnEncoding));
}
return valueStreams.build();
}
use of io.prestosql.orc.stream.OrcChunkLoader in project hetu-core by openlookeng.
the class StripeReader method readStripe.
public Stripe readStripe(StripeInformation stripe, AggregatedMemoryContext systemMemoryUsage) throws IOException {
// read the stripe footer
OrcStripeFooterCacheKey cacheKey = new OrcStripeFooterCacheKey();
cacheKey.setOrcDataSourceId(new OrcDataSourceIdWithTimeStamp(orcDataSource.getId(), orcDataSource.getLastModifiedTime()));
cacheKey.setStripeOffset(stripe.getOffset());
StripeFooter stripeFooter;
if (orcCacheProperties.isStripeFooterCacheEnabled()) {
try {
stripeFooter = orcCacheStore.getStripeFooterCache().get(cacheKey, () -> this.readStripeFooter(stripe, systemMemoryUsage));
} catch (UncheckedExecutionException | ExecutionException executionException) {
handleCacheLoadException(executionException);
log.debug(executionException.getCause(), "Error while caching ORC stripe footer. Falling back to default flow");
stripeFooter = readStripeFooter(stripe, systemMemoryUsage);
}
} else {
stripeFooter = readStripeFooter(stripe, systemMemoryUsage);
}
ColumnMetadata<ColumnEncoding> columnEncodings = stripeFooter.getColumnEncodings();
if (writeValidation.isPresent()) {
writeValidation.get().validateTimeZone(orcDataSource.getId(), stripeFooter.getTimeZone());
}
ZoneId fileTimeZone = stripeFooter.getTimeZone();
// get streams for selected columns
Map<StreamId, Stream> streams = new HashMap<>();
for (Stream stream : stripeFooter.getStreams()) {
if (includedOrcColumnIds.contains(stream.getColumnId()) && isSupportedStreamType(stream, types.get(stream.getColumnId()).getOrcTypeKind())) {
streams.put(new StreamId(stream), stream);
}
}
// handle stripes with more than one row group
boolean invalidCheckPoint = false;
if (stripe.getNumberOfRows() > rowsInRowGroup) {
// determine ranges of the stripe to read
Map<StreamId, DiskRange> diskRanges = getDiskRanges(stripeFooter.getStreams());
diskRanges = Maps.filterKeys(diskRanges, Predicates.in(streams.keySet()));
// read the file regions
Map<StreamId, OrcChunkLoader> streamsData = readDiskRanges(stripe.getOffset(), diskRanges, systemMemoryUsage);
// read the bloom filter for each column
Map<OrcColumnId, List<HashableBloomFilter>> bloomFilterIndexes = readBloomFilterIndexes(streams, streamsData, stripe);
// read the row index for each column
Map<StreamId, List<RowGroupIndex>> columnIndexes = readColumnIndexes(streams, streamsData, bloomFilterIndexes, stripe);
if (writeValidation.isPresent()) {
writeValidation.get().validateRowGroupStatistics(orcDataSource.getId(), stripe.getOffset(), columnIndexes);
}
// select the row groups matching the tuple domain
Set<Integer> selectedRowGroups = selectRowGroups(stripe, columnIndexes);
// if all row groups are skipped, return null
if (selectedRowGroups.isEmpty()) {
// set accounted memory usage to zero
systemMemoryUsage.close();
return null;
}
// value streams
Map<StreamId, ValueInputStream<?>> valueStreams = createValueStreams(streams, streamsData, columnEncodings);
// build the dictionary streams
InputStreamSources dictionaryStreamSources = createDictionaryStreamSources(streams, valueStreams, columnEncodings);
// build the row groups
try {
List<RowGroup> rowGroups = createRowGroups(stripe.getNumberOfRows(), streams, valueStreams, columnIndexes, selectedRowGroups, columnEncodings);
return new Stripe(stripe.getNumberOfRows(), fileTimeZone, columnEncodings, rowGroups, dictionaryStreamSources);
} catch (InvalidCheckpointException e) {
// The ORC file contains a corrupt checkpoint stream treat the stripe as a single row group.
invalidCheckPoint = true;
}
}
// stripe only has one row group
ImmutableMap.Builder<StreamId, DiskRange> diskRangesBuilder = ImmutableMap.builder();
for (Entry<StreamId, DiskRange> entry : getDiskRanges(stripeFooter.getStreams()).entrySet()) {
StreamId streamId = entry.getKey();
if (streams.containsKey(streamId)) {
diskRangesBuilder.put(entry);
}
}
ImmutableMap<StreamId, DiskRange> diskRanges = diskRangesBuilder.build();
// read the file regions
Map<StreamId, OrcChunkLoader> streamsData = readDiskRanges(stripe.getOffset(), diskRanges, systemMemoryUsage);
long minAverageRowBytes = 0;
for (Entry<StreamId, Stream> entry : streams.entrySet()) {
if (entry.getKey().getStreamKind() == ROW_INDEX) {
List<RowGroupIndex> rowGroupIndexes;
if (orcCacheProperties.isRowIndexCacheEnabled()) {
OrcRowIndexCacheKey indexCacheKey = new OrcRowIndexCacheKey();
indexCacheKey.setOrcDataSourceId(new OrcDataSourceIdWithTimeStamp(orcDataSource.getId(), orcDataSource.getLastModifiedTime()));
indexCacheKey.setStripeOffset(stripe.getOffset());
indexCacheKey.setStreamId(entry.getKey());
try {
rowGroupIndexes = orcCacheStore.getRowIndexCache().get(indexCacheKey, () -> metadataReader.readRowIndexes(hiveWriterVersion, new OrcInputStream(streamsData.get(entry.getKey()))));
} catch (UncheckedExecutionException | ExecutionException executionException) {
handleCacheLoadException(executionException);
log.debug(executionException.getCause(), "Error while caching row group indexes. Falling back to default flow");
rowGroupIndexes = metadataReader.readRowIndexes(hiveWriterVersion, new OrcInputStream(streamsData.get(entry.getKey())));
}
} else {
rowGroupIndexes = metadataReader.readRowIndexes(hiveWriterVersion, new OrcInputStream(streamsData.get(entry.getKey())));
}
checkState(rowGroupIndexes.size() == 1 || invalidCheckPoint, "expect a single row group or an invalid check point");
long totalBytes = 0;
long totalRows = 0;
for (RowGroupIndex rowGroupIndex : rowGroupIndexes) {
ColumnStatistics columnStatistics = rowGroupIndex.getColumnStatistics();
if (columnStatistics.hasMinAverageValueSizeInBytes()) {
totalBytes += columnStatistics.getMinAverageValueSizeInBytes() * columnStatistics.getNumberOfValues();
totalRows += columnStatistics.getNumberOfValues();
}
}
if (totalRows > 0) {
minAverageRowBytes += totalBytes / totalRows;
}
}
}
// value streams
Map<StreamId, ValueInputStream<?>> valueStreams = createValueStreams(streams, streamsData, columnEncodings);
// build the dictionary streams
InputStreamSources dictionaryStreamSources = createDictionaryStreamSources(streams, valueStreams, columnEncodings);
// build the row group
ImmutableMap.Builder<StreamId, InputStreamSource<?>> builder = ImmutableMap.builder();
for (Entry<StreamId, ValueInputStream<?>> entry : valueStreams.entrySet()) {
builder.put(entry.getKey(), new ValueInputStreamSource<>(entry.getValue()));
}
RowGroup rowGroup = new RowGroup(0, 0, stripe.getNumberOfRows(), minAverageRowBytes, new InputStreamSources(builder.build()));
return new Stripe(stripe.getNumberOfRows(), fileTimeZone, columnEncodings, ImmutableList.of(rowGroup), dictionaryStreamSources);
}
Aggregations