public CompletionFieldsConsumer(SegmentWriteState state) throws IOException {
this.delegatesFieldsConsumer = delegatePostingsFormat.fieldsConsumer(state);
String suggestFSTFile =
IndexFileNames.segmentFileName(state.segmentInfo.name, state.segmentSuffix, EXTENSION);
IndexOutput output = null;
boolean success = false;
try {
output = state.directory.createOutput(suggestFSTFile, state.context);
CodecUtil.writeIndexHeader(
output,
CODEC_NAME,
SUGGEST_VERSION_CURRENT,
state.segmentInfo.getId(),
state.segmentSuffix);
/*
* we write the delegate postings format name so we can load it
* without getting an instance in the ctor
*/
output.writeString(delegatePostingsFormat.getName());
output.writeString(writeProvider.getName());
this.suggestFieldsConsumer = writeProvider.consumer(output);
success = true;
} finally {
if (!success) {
IOUtils.closeWhileHandlingException(output);
}
}
}
@Test
@Override
public void testIndexWritingAndFinding() throws IOException, InterruptedException {
verifyBoth(cache0, cache1);
IndexOutput indexOutput = dirA.createOutput(filename, IOContext.DEFAULT);
indexOutput.writeString("no need to write, nobody ever will read this");
indexOutput.flush();
indexOutput.close();
assertFileExistsHavingRLCount(filename, 1, true);
IndexInput firstOpenOnB = dirB.openInput(filename, IOContext.DEFAULT);
assertFileExistsHavingRLCount(filename, 2, true);
dirA.deleteFile(filename);
assertFileExistsHavingRLCount(filename, 1, false);
// Lucene does use clone() - lock implementation ignores it as a clone is
// cast on locked segments and released before the close on the parent object
IndexInput cloneOfFirstOpenOnB = (IndexInput) firstOpenOnB.clone();
assertFileExistsHavingRLCount(filename, 1, false);
cloneOfFirstOpenOnB.close();
assertFileExistsHavingRLCount(filename, 1, false);
IndexInput firstOpenOnA = dirA.openInput(filename, IOContext.DEFAULT);
assertFileExistsHavingRLCount(filename, 2, false);
IndexInput secondOpenOnA = dirA.openInput(filename, IOContext.DEFAULT);
assertFileExistsHavingRLCount(filename, 2, false);
firstOpenOnA.close();
assertFileExistsHavingRLCount(filename, 2, false);
secondOpenOnA.close();
assertFileExistsHavingRLCount(filename, 1, false);
firstOpenOnB.close();
assertFileNotExists(filename);
dirA.close();
dirB.close();
verifyBoth(cache0, cache1);
}
@Override
public void write(
Directory directory,
SegmentInfo segmentInfo,
String segmentSuffix,
FieldInfos infos,
IOContext context)
throws IOException {
final String fileName =
IndexFileNames.segmentFileName(
segmentInfo.name, segmentSuffix, Lucene46FieldInfosFormat.EXTENSION);
try (IndexOutput output = directory.createOutput(fileName, context)) {
CodecUtil.writeHeader(
output, Lucene46FieldInfosFormat.CODEC_NAME, Lucene46FieldInfosFormat.FORMAT_CURRENT);
output.writeVInt(infos.size());
for (FieldInfo fi : infos) {
IndexOptions indexOptions = fi.getIndexOptions();
byte bits = 0x0;
if (fi.hasVectors()) bits |= Lucene46FieldInfosFormat.STORE_TERMVECTOR;
if (fi.omitsNorms()) bits |= Lucene46FieldInfosFormat.OMIT_NORMS;
if (fi.hasPayloads()) bits |= Lucene46FieldInfosFormat.STORE_PAYLOADS;
if (fi.getIndexOptions() != IndexOptions.NONE) {
bits |= Lucene46FieldInfosFormat.IS_INDEXED;
assert indexOptions.compareTo(IndexOptions.DOCS_AND_FREQS_AND_POSITIONS) >= 0
|| !fi.hasPayloads();
if (indexOptions == IndexOptions.DOCS) {
bits |= Lucene46FieldInfosFormat.OMIT_TERM_FREQ_AND_POSITIONS;
} else if (indexOptions == IndexOptions.DOCS_AND_FREQS_AND_POSITIONS_AND_OFFSETS) {
bits |= Lucene46FieldInfosFormat.STORE_OFFSETS_IN_POSTINGS;
} else if (indexOptions == IndexOptions.DOCS_AND_FREQS) {
bits |= Lucene46FieldInfosFormat.OMIT_POSITIONS;
}
}
output.writeString(fi.name);
output.writeVInt(fi.number);
output.writeByte(bits);
// pack the DV types in one byte
final byte dv = docValuesByte(fi.getDocValuesType());
final byte nrm = docValuesByte(fi.hasNorms() ? DocValuesType.NUMERIC : DocValuesType.NONE);
assert (dv & (~0xF)) == 0 && (nrm & (~0x0F)) == 0;
byte val = (byte) (0xff & ((nrm << 4) | dv));
output.writeByte(val);
output.writeLong(fi.getDocValuesGen());
output.writeStringStringMap(fi.attributes());
}
CodecUtil.writeFooter(output);
}
}
@Override
public void close() throws IOException {
delegateFieldsConsumer.close();
// Now we are done accumulating values for these fields
List<Entry<FieldInfo, FuzzySet>> nonSaturatedBlooms =
new ArrayList<Map.Entry<FieldInfo, FuzzySet>>();
for (Entry<FieldInfo, FuzzySet> entry : bloomFilters.entrySet()) {
FuzzySet bloomFilter = entry.getValue();
if (!bloomFilterFactory.isSaturated(bloomFilter, entry.getKey())) {
nonSaturatedBlooms.add(entry);
}
}
String bloomFileName =
IndexFileNames.segmentFileName(
state.segmentInfo.name, state.segmentSuffix, BLOOM_EXTENSION);
IndexOutput bloomOutput = null;
try {
bloomOutput = state.directory.createOutput(bloomFileName, state.context);
CodecUtil.writeHeader(bloomOutput, BLOOM_CODEC_NAME, BLOOM_CODEC_VERSION);
// remember the name of the postings format we will delegate to
bloomOutput.writeString(delegatePostingsFormat.getName());
// First field in the output file is the number of fields+blooms saved
bloomOutput.writeInt(nonSaturatedBlooms.size());
for (Entry<FieldInfo, FuzzySet> entry : nonSaturatedBlooms) {
FieldInfo fieldInfo = entry.getKey();
FuzzySet bloomFilter = entry.getValue();
bloomOutput.writeInt(fieldInfo.number);
saveAppropriatelySizedBloomFilter(bloomOutput, bloomFilter, fieldInfo);
}
} finally {
IOUtils.close(bloomOutput);
}
// We are done with large bitsets so no need to keep them hanging around
bloomFilters.clear();
}
@Override
public void writeString(String s) throws IOException {
delegate.writeString(s);
}
private void write(Directory directory) throws IOException {
long nextGeneration = getNextPendingGeneration();
String segmentFileName =
IndexFileNames.fileNameFromGeneration(IndexFileNames.PENDING_SEGMENTS, "", nextGeneration);
// Always advance the generation on write:
generation = nextGeneration;
IndexOutput segnOutput = null;
boolean success = false;
try {
segnOutput = directory.createOutput(segmentFileName, IOContext.DEFAULT);
CodecUtil.writeIndexHeader(
segnOutput,
"segments",
VERSION_CURRENT,
StringHelper.randomId(),
Long.toString(nextGeneration, Character.MAX_RADIX));
segnOutput.writeVInt(Version.LATEST.major);
segnOutput.writeVInt(Version.LATEST.minor);
segnOutput.writeVInt(Version.LATEST.bugfix);
segnOutput.writeLong(version);
segnOutput.writeInt(counter); // write counter
segnOutput.writeInt(size());
if (size() > 0) {
Version minSegmentVersion = null;
// We do a separate loop up front so we can write the minSegmentVersion before
// any SegmentInfo; this makes it cleaner to throw IndexFormatTooOldExc at read time:
for (SegmentCommitInfo siPerCommit : this) {
Version segmentVersion = siPerCommit.info.getVersion();
if (minSegmentVersion == null || segmentVersion.onOrAfter(minSegmentVersion) == false) {
minSegmentVersion = segmentVersion;
}
}
segnOutput.writeVInt(minSegmentVersion.major);
segnOutput.writeVInt(minSegmentVersion.minor);
segnOutput.writeVInt(minSegmentVersion.bugfix);
}
// write infos
for (SegmentCommitInfo siPerCommit : this) {
SegmentInfo si = siPerCommit.info;
segnOutput.writeString(si.name);
byte segmentID[] = si.getId();
// TODO: remove this in lucene 6, we don't need to include 4.x segments in commits anymore
if (segmentID == null) {
segnOutput.writeByte((byte) 0);
} else {
if (segmentID.length != StringHelper.ID_LENGTH) {
throw new IllegalStateException(
"cannot write segment: invalid id segment="
+ si.name
+ "id="
+ StringHelper.idToString(segmentID));
}
segnOutput.writeByte((byte) 1);
segnOutput.writeBytes(segmentID, segmentID.length);
}
segnOutput.writeString(si.getCodec().getName());
segnOutput.writeLong(siPerCommit.getDelGen());
int delCount = siPerCommit.getDelCount();
if (delCount < 0 || delCount > si.maxDoc()) {
throw new IllegalStateException(
"cannot write segment: invalid maxDoc segment="
+ si.name
+ " maxDoc="
+ si.maxDoc()
+ " delCount="
+ delCount);
}
segnOutput.writeInt(delCount);
segnOutput.writeLong(siPerCommit.getFieldInfosGen());
segnOutput.writeLong(siPerCommit.getDocValuesGen());
segnOutput.writeSetOfStrings(siPerCommit.getFieldInfosFiles());
final Map<Integer, Set<String>> dvUpdatesFiles = siPerCommit.getDocValuesUpdatesFiles();
segnOutput.writeInt(dvUpdatesFiles.size());
for (Entry<Integer, Set<String>> e : dvUpdatesFiles.entrySet()) {
segnOutput.writeInt(e.getKey());
segnOutput.writeSetOfStrings(e.getValue());
}
}
segnOutput.writeMapOfStrings(userData);
CodecUtil.writeFooter(segnOutput);
segnOutput.close();
directory.sync(Collections.singleton(segmentFileName));
success = true;
} finally {
if (success) {
pendingCommit = true;
} else {
// We hit an exception above; try to close the file
// but suppress any exception:
IOUtils.closeWhileHandlingException(segnOutput);
// Try not to leave a truncated segments_N file in
// the index:
IOUtils.deleteFilesIgnoringExceptions(directory, segmentFileName);
}
}
}
