/**
* We assume that the initial indexing has been done and a set of reference objects has been found
* and indexed in the separate directory. However further documents were added and they now need
* to get a ranked list of reference objects. So we (i) get all these new documents missing the
* field "ro-order" and (ii) add this field.
*
* @param indexPath the index to update
* @throws IOException
*/
public void updateIndex(String indexPath) throws IOException {
IndexReader reader = DirectoryReader.open(FSDirectory.open(new File(indexPath)));
int numDocs = reader.numDocs();
boolean hasDeletions = reader.hasDeletions();
int countUpdated = 0;
IndexReader readerRo = DirectoryReader.open(FSDirectory.open(new File(indexPath + "-ro")));
ImageSearcher searcher =
new GenericImageSearcher(numReferenceObjectsUsed, featureClass, featureFieldName);
Map<String, Analyzer> perField = new HashMap<String, Analyzer>(1);
perField.put("ro-order", new WhitespaceAnalyzer(LuceneUtils.LUCENE_VERSION));
PerFieldAnalyzerWrapper aWrapper =
new PerFieldAnalyzerWrapper(new SimpleAnalyzer(LuceneUtils.LUCENE_VERSION), perField);
IndexWriter iw =
new IndexWriter(
FSDirectory.open(new File(indexPath)),
new IndexWriterConfig(LuceneUtils.LUCENE_VERSION, aWrapper)
.setOpenMode(IndexWriterConfig.OpenMode.CREATE));
StringBuilder sb = new StringBuilder(256);
// Needed for check whether the document is deleted.
Bits liveDocs = MultiFields.getLiveDocs(reader);
for (int i = 0; i < numDocs; i++) {
if (reader.hasDeletions() && !liveDocs.get(i)) continue; // if it is deleted, just ignore it.
Document document = reader.document(i);
if (document.getField("ro-order") == null) { // if the field is not here we create it.
ImageSearchHits hits = searcher.search(document, readerRo);
sb.delete(0, sb.length());
for (int j = 0; j < numReferenceObjectsUsed; j++) {
sb.append(hits.doc(j).getValues("ro-id")[0]);
sb.append(' ');
}
// System.out.println(sb.toString());
document.add(new TextField("ro-order", sb.toString(), Field.Store.YES));
iw.updateDocument(
new Term(
DocumentBuilder.FIELD_NAME_IDENTIFIER,
document.getValues(DocumentBuilder.FIELD_NAME_IDENTIFIER)[0]),
document);
countUpdated++;
}
// progress report
progress.setNumDocsProcessed(progress.getNumDocsProcessed() + 1);
// debug:
System.out.println("countUpdated = " + countUpdated);
}
iw.commit();
iw.close();
}
protected DirectoryReader indexDocsWithLucene(TestDoc[] testDocs) throws IOException {
Map<String, Analyzer> mapping = new HashMap<String, Analyzer>();
for (TestFieldSetting field : testDocs[0].fieldSettings) {
if (field.storedPayloads) {
mapping.put(
field.name,
new Analyzer() {
@Override
protected TokenStreamComponents createComponents(String fieldName, Reader reader) {
Tokenizer tokenizer = new StandardTokenizer(Version.CURRENT.luceneVersion, reader);
TokenFilter filter = new LowerCaseFilter(Version.CURRENT.luceneVersion, tokenizer);
filter = new TypeAsPayloadTokenFilter(filter);
return new TokenStreamComponents(tokenizer, filter);
}
});
}
}
PerFieldAnalyzerWrapper wrapper =
new PerFieldAnalyzerWrapper(
new StandardAnalyzer(Version.CURRENT.luceneVersion, CharArraySet.EMPTY_SET), mapping);
Directory dir = new RAMDirectory();
IndexWriterConfig conf = new IndexWriterConfig(Version.CURRENT.luceneVersion, wrapper);
conf.setOpenMode(IndexWriterConfig.OpenMode.CREATE);
IndexWriter writer = new IndexWriter(dir, conf);
for (TestDoc doc : testDocs) {
Document d = new Document();
d.add(new Field("id", doc.id, StringField.TYPE_STORED));
for (int i = 0; i < doc.fieldContent.length; i++) {
FieldType type = new FieldType(TextField.TYPE_STORED);
TestFieldSetting fieldSetting = doc.fieldSettings[i];
type.setStoreTermVectorOffsets(fieldSetting.storedOffset);
type.setStoreTermVectorPayloads(fieldSetting.storedPayloads);
type.setStoreTermVectorPositions(
fieldSetting.storedPositions
|| fieldSetting.storedPayloads
|| fieldSetting.storedOffset);
type.setStoreTermVectors(true);
type.freeze();
d.add(new Field(fieldSetting.name, doc.fieldContent[i], type));
}
writer.updateDocument(new Term("id", doc.id), d);
writer.commit();
}
writer.close();
return DirectoryReader.open(dir);
}
public void testTermDocsEnum() throws Exception {
Directory dir = newDirectory();
IndexWriter w =
new IndexWriter(
dir, newIndexWriterConfig(TEST_VERSION_CURRENT, new MockAnalyzer(random())));
Document d = new Document();
d.add(newStringField("f", "j", Field.Store.NO));
w.addDocument(d);
w.commit();
w.addDocument(d);
IndexReader r = w.getReader();
w.close();
DocsEnum de = MultiFields.getTermDocsEnum(r, null, "f", new BytesRef("j"));
assertEquals(0, de.nextDoc());
assertEquals(1, de.nextDoc());
assertEquals(DocIdSetIterator.NO_MORE_DOCS, de.nextDoc());
r.close();
dir.close();
}
public void testSeparateEnums() throws Exception {
Directory dir = newDirectory();
IndexWriter w =
new IndexWriter(
dir, newIndexWriterConfig(TEST_VERSION_CURRENT, new MockAnalyzer(random())));
Document d = new Document();
d.add(newStringField("f", "j", Field.Store.NO));
w.addDocument(d);
w.commit();
w.addDocument(d);
IndexReader r = w.getReader();
w.close();
DocsEnum d1 = _TestUtil.docs(random(), r, "f", new BytesRef("j"), null, null, 0);
DocsEnum d2 = _TestUtil.docs(random(), r, "f", new BytesRef("j"), null, null, 0);
assertEquals(0, d1.nextDoc());
assertEquals(0, d2.nextDoc());
r.close();
dir.close();
}
public void testRandom() throws Exception {
int num = atLeast(2);
for (int iter = 0; iter < num; iter++) {
if (VERBOSE) {
System.out.println("TEST: iter=" + iter);
}
Directory dir = newDirectory();
IndexWriter w =
new IndexWriter(
dir,
newIndexWriterConfig(TEST_VERSION_CURRENT, new MockAnalyzer(random()))
.setMergePolicy(NoMergePolicy.COMPOUND_FILES));
_TestUtil.keepFullyDeletedSegments(w);
Map<BytesRef, List<Integer>> docs = new HashMap<BytesRef, List<Integer>>();
Set<Integer> deleted = new HashSet<Integer>();
List<BytesRef> terms = new ArrayList<BytesRef>();
int numDocs = _TestUtil.nextInt(random(), 1, 100 * RANDOM_MULTIPLIER);
Document doc = new Document();
Field f = newStringField("field", "", Field.Store.NO);
doc.add(f);
Field id = newStringField("id", "", Field.Store.NO);
doc.add(id);
boolean onlyUniqueTerms = random().nextBoolean();
if (VERBOSE) {
System.out.println("TEST: onlyUniqueTerms=" + onlyUniqueTerms + " numDocs=" + numDocs);
}
Set<BytesRef> uniqueTerms = new HashSet<BytesRef>();
for (int i = 0; i < numDocs; i++) {
if (!onlyUniqueTerms && random().nextBoolean() && terms.size() > 0) {
// re-use existing term
BytesRef term = terms.get(random().nextInt(terms.size()));
docs.get(term).add(i);
f.setStringValue(term.utf8ToString());
} else {
String s = _TestUtil.randomUnicodeString(random(), 10);
BytesRef term = new BytesRef(s);
if (!docs.containsKey(term)) {
docs.put(term, new ArrayList<Integer>());
}
docs.get(term).add(i);
terms.add(term);
uniqueTerms.add(term);
f.setStringValue(s);
}
id.setStringValue("" + i);
w.addDocument(doc);
if (random().nextInt(4) == 1) {
w.commit();
}
if (i > 0 && random().nextInt(20) == 1) {
int delID = random().nextInt(i);
deleted.add(delID);
w.deleteDocuments(new Term("id", "" + delID));
if (VERBOSE) {
System.out.println("TEST: delete " + delID);
}
}
}
if (VERBOSE) {
List<BytesRef> termsList = new ArrayList<BytesRef>(uniqueTerms);
Collections.sort(termsList, BytesRef.getUTF8SortedAsUTF16Comparator());
System.out.println("TEST: terms in UTF16 order:");
for (BytesRef b : termsList) {
System.out.println(" " + UnicodeUtil.toHexString(b.utf8ToString()) + " " + b);
for (int docID : docs.get(b)) {
if (deleted.contains(docID)) {
System.out.println(" " + docID + " (deleted)");
} else {
System.out.println(" " + docID);
}
}
}
}
IndexReader reader = w.getReader();
w.close();
if (VERBOSE) {
System.out.println("TEST: reader=" + reader);
}
Bits liveDocs = MultiFields.getLiveDocs(reader);
for (int delDoc : deleted) {
assertFalse(liveDocs.get(delDoc));
}
for (int i = 0; i < 100; i++) {
BytesRef term = terms.get(random().nextInt(terms.size()));
if (VERBOSE) {
System.out.println(
"TEST: seek term=" + UnicodeUtil.toHexString(term.utf8ToString()) + " " + term);
}
DocsEnum docsEnum = _TestUtil.docs(random(), reader, "field", term, liveDocs, null, 0);
assertNotNull(docsEnum);
for (int docID : docs.get(term)) {
if (!deleted.contains(docID)) {
assertEquals(docID, docsEnum.nextDoc());
}
}
assertEquals(DocIdSetIterator.NO_MORE_DOCS, docsEnum.nextDoc());
}
reader.close();
dir.close();
}
}
/**
* Assert that the {@code scoreType} operates as expected and parents are found in the expected
* order.
*
* <p>This will use the test index's parent/child types to create parents with multiple children.
* Each child will have a randomly generated scored stored in {@link #CHILD_SCORE_NAME}, which is
* used to score based on the {@code scoreType} by using a {@link MockScorer} to determine the
* expected scores.
*
* @param scoreType The score type to use within the query to score parents relative to their
* children.
* @throws IOException if any unexpected error occurs
*/
private void assertScoreType(ScoreType scoreType) throws IOException {
SearchContext context = SearchContext.current();
Directory directory = newDirectory();
IndexWriter writer =
new IndexWriter(directory, newIndexWriterConfig(new MockAnalyzer(random())));
// calculates the expected score per parent
MockScorer scorer = new MockScorer(scoreType);
scorer.scores = new FloatArrayList(10);
// number of parents to generate
int parentDocs = scaledRandomIntBetween(2, 10);
// unique child ID
int childDocId = 0;
// Parent ID to expected score
Map<String, Float> parentScores = new TreeMap<>();
// Add a few random parents to ensure that the children's score is appropriately taken into
// account
for (int parentDocId = 0; parentDocId < parentDocs; ++parentDocId) {
String parent = Integer.toString(parentDocId);
// Create the parent
Document parentDocument = new Document();
parentDocument.add(
new StringField(UidFieldMapper.NAME, Uid.createUid("parent", parent), Field.Store.YES));
parentDocument.add(new StringField(IdFieldMapper.NAME, parent, Field.Store.YES));
parentDocument.add(new StringField(TypeFieldMapper.NAME, "parent", Field.Store.NO));
// add the parent to the index
writer.addDocument(parentDocument);
int numChildDocs = scaledRandomIntBetween(1, 10);
// forget any parent's previous scores
scorer.scores.clear();
// associate children with the parent
for (int i = 0; i < numChildDocs; ++i) {
int childScore = random().nextInt(128);
Document childDocument = new Document();
childDocument.add(
new StringField(
UidFieldMapper.NAME,
Uid.createUid("child", Integer.toString(childDocId++)),
Field.Store.NO));
childDocument.add(new StringField(TypeFieldMapper.NAME, "child", Field.Store.NO));
// parent association:
childDocument.add(
new StringField(
ParentFieldMapper.NAME, Uid.createUid("parent", parent), Field.Store.NO));
childDocument.add(new DoubleField(CHILD_SCORE_NAME, childScore, Field.Store.NO));
// remember the score to be calculated
scorer.scores.add(childScore);
// add the associated child to the index
writer.addDocument(childDocument);
}
// this score that should be returned for this parent
parentScores.put(parent, scorer.score());
}
writer.commit();
IndexReader reader = DirectoryReader.open(writer, true);
IndexSearcher searcher = new IndexSearcher(reader);
// setup to read the parent/child map
Engine.Searcher engineSearcher =
new Engine.Searcher(ChildrenQueryTests.class.getSimpleName(), searcher);
((TestSearchContext) context).setSearcher(new ContextIndexSearcher(context, engineSearcher));
// child query that returns the score as the value of "childScore" for each child document, with
// the parent's score determined by the score type
QueryBuilder childQueryBuilder =
functionScoreQuery(typeFilter("child"))
.add(new FieldValueFactorFunctionBuilder(CHILD_SCORE_NAME));
QueryBuilder queryBuilder =
hasChildQuery("child", childQueryBuilder)
.scoreType(scoreType.name().toLowerCase(Locale.ENGLISH))
.setShortCircuitCutoff(parentDocs);
// Perform the search for the documents using the selected score type
TopDocs docs = searcher.search(parseQuery(queryBuilder), parentDocs);
assertThat("Expected all parents", docs.totalHits, is(parentDocs));
// score should be descending (just a sanity check)
float topScore = docs.scoreDocs[0].score;
// ensure each score is returned as expected
for (int i = 0; i < parentDocs; ++i) {
ScoreDoc scoreDoc = docs.scoreDocs[i];
// get the ID from the document to get its expected score; remove it so we cannot double-count
// it
float score = parentScores.remove(reader.document(scoreDoc.doc).get(IdFieldMapper.NAME));
// expect exact match
assertThat("Unexpected score", scoreDoc.score, is(score));
assertThat("Not descending", score, lessThanOrEqualTo(topScore));
// it had better keep descending
topScore = score;
}
reader.close();
writer.close();
directory.close();
}
/**
* Creates a set of reference objects and stores it in a new index (name "<indexPath>-ro"). Then
* creates ordered lists of reference object positions for each data item in the index with given
* feature. Finally a new index (name "<indexPath>-ms") is created where all the original
* documents as well as the new data are stored.
*
* @param indexPath the path to the original index
* @throws IOException
*/
public void createIndex(String indexPath) throws IOException {
IndexReader reader = DirectoryReader.open(FSDirectory.open(new File(indexPath)));
int numDocs = reader.numDocs();
if (numDocs < numReferenceObjects) {
throw new UnsupportedOperationException("Too few documents in index.");
}
// progress report
progress.setNumDocsAll(numDocs);
progress.setCurrentState(State.RoSelection);
boolean hasDeletions = reader.hasDeletions();
// init reference objects:
IndexWriter iw = LuceneUtils.createIndexWriter(indexPath + "-ro", true);
HashSet<Integer> referenceObjsIds = new HashSet<Integer>(numReferenceObjects);
double numDocsDouble = (double) numDocs;
while (referenceObjsIds.size() < numReferenceObjects) {
referenceObjsIds.add((int) (numDocsDouble * Math.random()));
}
int count = 0;
if (hasDeletions) {
System.err.println(
"WARNING: There are deleted docs in your index. You should "
+ "optimize your index before using this method.");
}
// progress report
progress.setCurrentState(State.RoIndexing);
// find them in the index and put them into a separate index:
for (int i : referenceObjsIds) {
count++;
Document document = reader.document(i);
document.add(new Field("ro-id", count + "", StringField.TYPE_STORED));
iw.addDocument(document);
}
iw.commit();
iw.close();
// progress report
progress.setCurrentState(State.Indexing);
// now find the reference objects for each entry ;)
IndexReader readerRo = DirectoryReader.open(FSDirectory.open(new File(indexPath + "-ro")));
ImageSearcher searcher =
new GenericImageSearcher(numReferenceObjectsUsed, featureClass, featureFieldName);
Map<String, Analyzer> analyzerPerField = new HashMap<String, Analyzer>();
analyzerPerField.put("ro-order", new WhitespaceAnalyzer(LuceneUtils.LUCENE_VERSION));
PerFieldAnalyzerWrapper aWrapper =
new PerFieldAnalyzerWrapper(
new SimpleAnalyzer(LuceneUtils.LUCENE_VERSION), analyzerPerField);
iw =
new IndexWriter(
FSDirectory.open(new File(indexPath)),
new IndexWriterConfig(LuceneUtils.LUCENE_VERSION, aWrapper)
.setOpenMode(IndexWriterConfig.OpenMode.CREATE));
StringBuilder sb = new StringBuilder(256);
// Needed for check whether the document is deleted.
Bits liveDocs = MultiFields.getLiveDocs(reader);
for (int i = 0; i < numDocs; i++) {
if (reader.hasDeletions() && !liveDocs.get(i)) continue; // if it is deleted, just ignore it.
Document document = reader.document(i);
ImageSearchHits hits = searcher.search(document, readerRo);
sb.delete(0, sb.length());
for (int j = 0; j < numReferenceObjectsUsed; j++) {
sb.append(hits.doc(j).getValues("ro-id")[0]);
sb.append(' ');
}
// System.out.println(sb.toString());
document.add(new TextField("ro-order", sb.toString(), Field.Store.YES));
iw.updateDocument(
new Term(
DocumentBuilder.FIELD_NAME_IDENTIFIER,
document.getValues(DocumentBuilder.FIELD_NAME_IDENTIFIER)[0]),
document);
// progress report
progress.setNumDocsProcessed(progress.getNumDocsProcessed() + 1);
}
iw.commit();
iw.close();
// progress report
progress.setCurrentState(State.Idle);
}
@Test
public void testRecoveryDiff() throws IOException, InterruptedException {
int numDocs = 2 + random().nextInt(100);
List<Document> docs = new ArrayList<>();
for (int i = 0; i < numDocs; i++) {
Document doc = new Document();
doc.add(
new StringField("id", "" + i, random().nextBoolean() ? Field.Store.YES : Field.Store.NO));
doc.add(
new TextField(
"body",
TestUtil.randomRealisticUnicodeString(random()),
random().nextBoolean() ? Field.Store.YES : Field.Store.NO));
doc.add(
new SortedDocValuesField(
"dv", new BytesRef(TestUtil.randomRealisticUnicodeString(random()))));
docs.add(doc);
}
long seed = random().nextLong();
Store.MetadataSnapshot first;
{
Random random = new Random(seed);
IndexWriterConfig iwc =
new IndexWriterConfig(new MockAnalyzer(random)).setCodec(actualDefaultCodec());
iwc.setMergePolicy(NoMergePolicy.INSTANCE);
iwc.setUseCompoundFile(random.nextBoolean());
iwc.setMaxThreadStates(1);
final ShardId shardId = new ShardId(new Index("index"), 1);
DirectoryService directoryService = new LuceneManagedDirectoryService(random);
Store store =
new Store(
shardId,
ImmutableSettings.EMPTY,
directoryService,
randomDistributor(random, directoryService),
new DummyShardLock(shardId));
IndexWriter writer = new IndexWriter(store.directory(), iwc);
final boolean lotsOfSegments = rarely(random);
for (Document d : docs) {
writer.addDocument(d);
if (lotsOfSegments && random.nextBoolean()) {
writer.commit();
} else if (rarely(random)) {
writer.commit();
}
}
writer.commit();
writer.close();
first = store.getMetadata();
assertDeleteContent(store, directoryService);
store.close();
}
long time = new Date().getTime();
while (time == new Date().getTime()) {
Thread.sleep(10); // bump the time
}
Store.MetadataSnapshot second;
Store store;
{
Random random = new Random(seed);
IndexWriterConfig iwc =
new IndexWriterConfig(new MockAnalyzer(random)).setCodec(actualDefaultCodec());
iwc.setMergePolicy(NoMergePolicy.INSTANCE);
iwc.setUseCompoundFile(random.nextBoolean());
iwc.setMaxThreadStates(1);
final ShardId shardId = new ShardId(new Index("index"), 1);
DirectoryService directoryService = new LuceneManagedDirectoryService(random);
store =
new Store(
shardId,
ImmutableSettings.EMPTY,
directoryService,
randomDistributor(random, directoryService),
new DummyShardLock(shardId));
IndexWriter writer = new IndexWriter(store.directory(), iwc);
final boolean lotsOfSegments = rarely(random);
for (Document d : docs) {
writer.addDocument(d);
if (lotsOfSegments && random.nextBoolean()) {
writer.commit();
} else if (rarely(random)) {
writer.commit();
}
}
writer.commit();
writer.close();
second = store.getMetadata();
}
Store.RecoveryDiff diff = first.recoveryDiff(second);
assertThat(first.size(), equalTo(second.size()));
for (StoreFileMetaData md : first) {
assertThat(second.get(md.name()), notNullValue());
// si files are different - containing timestamps etc
assertThat(second.get(md.name()).isSame(md), equalTo(false));
}
assertThat(diff.different.size(), equalTo(first.size()));
assertThat(
diff.identical.size(),
equalTo(0)); // in lucene 5 nothing is identical - we use random ids in file headers
assertThat(diff.missing, empty());
// check the self diff
Store.RecoveryDiff selfDiff = first.recoveryDiff(first);
assertThat(selfDiff.identical.size(), equalTo(first.size()));
assertThat(selfDiff.different, empty());
assertThat(selfDiff.missing, empty());
// lets add some deletes
Random random = new Random(seed);
IndexWriterConfig iwc =
new IndexWriterConfig(new MockAnalyzer(random)).setCodec(actualDefaultCodec());
iwc.setMergePolicy(NoMergePolicy.INSTANCE);
iwc.setUseCompoundFile(random.nextBoolean());
iwc.setMaxThreadStates(1);
iwc.setOpenMode(IndexWriterConfig.OpenMode.APPEND);
IndexWriter writer = new IndexWriter(store.directory(), iwc);
writer.deleteDocuments(new Term("id", Integer.toString(random().nextInt(numDocs))));
writer.commit();
writer.close();
Store.MetadataSnapshot metadata = store.getMetadata();
StoreFileMetaData delFile = null;
for (StoreFileMetaData md : metadata) {
if (md.name().endsWith(".liv")) {
delFile = md;
break;
}
}
Store.RecoveryDiff afterDeleteDiff = metadata.recoveryDiff(second);
if (delFile != null) {
assertThat(
afterDeleteDiff.identical.size(), equalTo(metadata.size() - 2)); // segments_N + del file
assertThat(afterDeleteDiff.different.size(), equalTo(0));
assertThat(afterDeleteDiff.missing.size(), equalTo(2));
} else {
// an entire segment must be missing (single doc segment got dropped)
assertThat(afterDeleteDiff.identical.size(), greaterThan(0));
assertThat(afterDeleteDiff.different.size(), equalTo(0));
assertThat(afterDeleteDiff.missing.size(), equalTo(1)); // the commit file is different
}
// check the self diff
selfDiff = metadata.recoveryDiff(metadata);
assertThat(selfDiff.identical.size(), equalTo(metadata.size()));
assertThat(selfDiff.different, empty());
assertThat(selfDiff.missing, empty());
// add a new commit
iwc = new IndexWriterConfig(new MockAnalyzer(random)).setCodec(actualDefaultCodec());
iwc.setMergePolicy(NoMergePolicy.INSTANCE);
iwc.setUseCompoundFile(
true); // force CFS - easier to test here since we know it will add 3 files
iwc.setMaxThreadStates(1);
iwc.setOpenMode(IndexWriterConfig.OpenMode.APPEND);
writer = new IndexWriter(store.directory(), iwc);
writer.addDocument(docs.get(0));
writer.close();
Store.MetadataSnapshot newCommitMetaData = store.getMetadata();
Store.RecoveryDiff newCommitDiff = newCommitMetaData.recoveryDiff(metadata);
if (delFile != null) {
assertThat(
newCommitDiff.identical.size(),
equalTo(
newCommitMetaData.size()
- 5)); // segments_N, del file, cfs, cfe, si for the new segment
assertThat(newCommitDiff.different.size(), equalTo(1)); // the del file must be different
assertThat(newCommitDiff.different.get(0).name(), endsWith(".liv"));
assertThat(
newCommitDiff.missing.size(), equalTo(4)); // segments_N,cfs, cfe, si for the new segment
} else {
assertThat(
newCommitDiff.identical.size(),
equalTo(newCommitMetaData.size() - 4)); // segments_N, cfs, cfe, si for the new segment
assertThat(newCommitDiff.different.size(), equalTo(0));
assertThat(
newCommitDiff.missing.size(),
equalTo(
4)); // an entire segment must be missing (single doc segment got dropped) plus the
// commit is different
}
store.deleteContent();
IOUtils.close(store);
}
@Test
public void testMixedChecksums() throws IOException {
final ShardId shardId = new ShardId(new Index("index"), 1);
DirectoryService directoryService = new LuceneManagedDirectoryService(random());
Store store =
new Store(
shardId,
ImmutableSettings.EMPTY,
directoryService,
randomDistributor(directoryService),
new DummyShardLock(shardId));
// this time random codec....
IndexWriter writer =
new IndexWriter(
store.directory(),
newIndexWriterConfig(random(), new MockAnalyzer(random()))
.setCodec(actualDefaultCodec()));
int docs = 1 + random().nextInt(100);
for (int i = 0; i < docs; i++) {
Document doc = new Document();
doc.add(
new TextField("id", "" + i, random().nextBoolean() ? Field.Store.YES : Field.Store.NO));
doc.add(
new TextField(
"body",
TestUtil.randomRealisticUnicodeString(random()),
random().nextBoolean() ? Field.Store.YES : Field.Store.NO));
doc.add(
new SortedDocValuesField(
"dv", new BytesRef(TestUtil.randomRealisticUnicodeString(random()))));
writer.addDocument(doc);
}
if (random().nextBoolean()) {
for (int i = 0; i < docs; i++) {
if (random().nextBoolean()) {
Document doc = new Document();
doc.add(
new TextField(
"id", "" + i, random().nextBoolean() ? Field.Store.YES : Field.Store.NO));
doc.add(
new TextField(
"body",
TestUtil.randomRealisticUnicodeString(random()),
random().nextBoolean() ? Field.Store.YES : Field.Store.NO));
writer.updateDocument(new Term("id", "" + i), doc);
}
}
}
if (random().nextBoolean()) {
DirectoryReader.open(writer, random().nextBoolean()).close(); // flush
}
Store.MetadataSnapshot metadata;
// check before we committed
try {
store.getMetadata();
fail("no index present - expected exception");
} catch (IndexNotFoundException ex) {
// expected
}
assertThat(store.getMetadataOrEmpty(), is(Store.MetadataSnapshot.EMPTY)); // nothing committed
writer.commit();
writer.close();
Store.LegacyChecksums checksums = new Store.LegacyChecksums();
metadata = store.getMetadata();
assertThat(metadata.asMap().isEmpty(), is(false));
for (StoreFileMetaData meta : metadata) {
try (IndexInput input = store.directory().openInput(meta.name(), IOContext.DEFAULT)) {
if (meta.checksum() == null) {
String checksum = null;
try {
CodecUtil.retrieveChecksum(input);
fail("expected a corrupt index - posting format has not checksums");
} catch (CorruptIndexException
| IndexFormatTooOldException
| IndexFormatTooNewException ex) {
try (ChecksumIndexInput checksumIndexInput =
store.directory().openChecksumInput(meta.name(), IOContext.DEFAULT)) {
checksumIndexInput.seek(meta.length());
checksum = Store.digestToString(checksumIndexInput.getChecksum());
}
// fine - it's a postings format without checksums
checksums.add(new StoreFileMetaData(meta.name(), meta.length(), checksum, null));
}
} else {
String checksum = Store.digestToString(CodecUtil.retrieveChecksum(input));
assertThat(
"File: " + meta.name() + " has a different checksum",
meta.checksum(),
equalTo(checksum));
assertThat(meta.hasLegacyChecksum(), equalTo(false));
assertThat(meta.writtenBy(), equalTo(Version.LATEST));
}
}
}
assertConsistent(store, metadata);
checksums.write(store);
metadata = store.getMetadata();
assertThat(metadata.asMap().isEmpty(), is(false));
for (StoreFileMetaData meta : metadata) {
assertThat(
"file: " + meta.name() + " has a null checksum", meta.checksum(), not(nullValue()));
if (meta.hasLegacyChecksum()) {
try (ChecksumIndexInput checksumIndexInput =
store.directory().openChecksumInput(meta.name(), IOContext.DEFAULT)) {
checksumIndexInput.seek(meta.length());
assertThat(
meta.checksum(), equalTo(Store.digestToString(checksumIndexInput.getChecksum())));
}
} else {
try (IndexInput input = store.directory().openInput(meta.name(), IOContext.DEFAULT)) {
String checksum = Store.digestToString(CodecUtil.retrieveChecksum(input));
assertThat(
"File: " + meta.name() + " has a different checksum",
meta.checksum(),
equalTo(checksum));
assertThat(meta.hasLegacyChecksum(), equalTo(false));
assertThat(meta.writtenBy(), equalTo(Version.LATEST));
}
}
}
assertConsistent(store, metadata);
TestUtil.checkIndex(store.directory());
assertDeleteContent(store, directoryService);
IOUtils.close(store);
}
@Test
public void testNewChecksums() throws IOException {
final ShardId shardId = new ShardId(new Index("index"), 1);
DirectoryService directoryService = new LuceneManagedDirectoryService(random());
Store store =
new Store(
shardId,
ImmutableSettings.EMPTY,
directoryService,
randomDistributor(directoryService),
new DummyShardLock(shardId));
// set default codec - all segments need checksums
IndexWriter writer =
new IndexWriter(
store.directory(),
newIndexWriterConfig(random(), new MockAnalyzer(random()))
.setCodec(actualDefaultCodec()));
int docs = 1 + random().nextInt(100);
for (int i = 0; i < docs; i++) {
Document doc = new Document();
doc.add(
new TextField("id", "" + i, random().nextBoolean() ? Field.Store.YES : Field.Store.NO));
doc.add(
new TextField(
"body",
TestUtil.randomRealisticUnicodeString(random()),
random().nextBoolean() ? Field.Store.YES : Field.Store.NO));
doc.add(
new SortedDocValuesField(
"dv", new BytesRef(TestUtil.randomRealisticUnicodeString(random()))));
writer.addDocument(doc);
}
if (random().nextBoolean()) {
for (int i = 0; i < docs; i++) {
if (random().nextBoolean()) {
Document doc = new Document();
doc.add(
new TextField(
"id", "" + i, random().nextBoolean() ? Field.Store.YES : Field.Store.NO));
doc.add(
new TextField(
"body",
TestUtil.randomRealisticUnicodeString(random()),
random().nextBoolean() ? Field.Store.YES : Field.Store.NO));
writer.updateDocument(new Term("id", "" + i), doc);
}
}
}
if (random().nextBoolean()) {
DirectoryReader.open(writer, random().nextBoolean()).close(); // flush
}
Store.MetadataSnapshot metadata;
// check before we committed
try {
store.getMetadata();
fail("no index present - expected exception");
} catch (IndexNotFoundException ex) {
// expected
}
assertThat(store.getMetadataOrEmpty(), is(Store.MetadataSnapshot.EMPTY)); // nothing committed
writer.commit();
writer.close();
metadata = store.getMetadata();
assertThat(metadata.asMap().isEmpty(), is(false));
for (StoreFileMetaData meta : metadata) {
try (IndexInput input = store.directory().openInput(meta.name(), IOContext.DEFAULT)) {
String checksum = Store.digestToString(CodecUtil.retrieveChecksum(input));
assertThat(
"File: " + meta.name() + " has a different checksum",
meta.checksum(),
equalTo(checksum));
assertThat(meta.hasLegacyChecksum(), equalTo(false));
assertThat(meta.writtenBy(), equalTo(Version.LATEST));
if (meta.name().endsWith(".si") || meta.name().startsWith("segments_")) {
assertThat(meta.hash().length, greaterThan(0));
}
}
}
assertConsistent(store, metadata);
TestUtil.checkIndex(store.directory());
assertDeleteContent(store, directoryService);
IOUtils.close(store);
}
@Test
public void testCleanupFromSnapshot() throws IOException {
final ShardId shardId = new ShardId(new Index("index"), 1);
DirectoryService directoryService = new LuceneManagedDirectoryService(random());
Store store =
new Store(
shardId,
ImmutableSettings.EMPTY,
directoryService,
randomDistributor(directoryService),
new DummyShardLock(shardId));
// this time random codec....
IndexWriterConfig indexWriterConfig =
newIndexWriterConfig(random(), new MockAnalyzer(random())).setCodec(actualDefaultCodec());
// we keep all commits and that allows us clean based on multiple snapshots
indexWriterConfig.setIndexDeletionPolicy(NoDeletionPolicy.INSTANCE);
IndexWriter writer = new IndexWriter(store.directory(), indexWriterConfig);
int docs = 1 + random().nextInt(100);
int numCommits = 0;
for (int i = 0; i < docs; i++) {
if (i > 0 && randomIntBetween(0, 10) == 0) {
writer.commit();
numCommits++;
}
Document doc = new Document();
doc.add(
new TextField("id", "" + i, random().nextBoolean() ? Field.Store.YES : Field.Store.NO));
doc.add(
new TextField(
"body",
TestUtil.randomRealisticUnicodeString(random()),
random().nextBoolean() ? Field.Store.YES : Field.Store.NO));
doc.add(
new SortedDocValuesField(
"dv", new BytesRef(TestUtil.randomRealisticUnicodeString(random()))));
writer.addDocument(doc);
}
if (numCommits < 1) {
writer.commit();
Document doc = new Document();
doc.add(
new TextField(
"id", "" + docs++, random().nextBoolean() ? Field.Store.YES : Field.Store.NO));
doc.add(
new TextField(
"body",
TestUtil.randomRealisticUnicodeString(random()),
random().nextBoolean() ? Field.Store.YES : Field.Store.NO));
doc.add(
new SortedDocValuesField(
"dv", new BytesRef(TestUtil.randomRealisticUnicodeString(random()))));
writer.addDocument(doc);
}
Store.MetadataSnapshot firstMeta = store.getMetadata();
if (random().nextBoolean()) {
for (int i = 0; i < docs; i++) {
if (random().nextBoolean()) {
Document doc = new Document();
doc.add(
new TextField(
"id", "" + i, random().nextBoolean() ? Field.Store.YES : Field.Store.NO));
doc.add(
new TextField(
"body",
TestUtil.randomRealisticUnicodeString(random()),
random().nextBoolean() ? Field.Store.YES : Field.Store.NO));
writer.updateDocument(new Term("id", "" + i), doc);
}
}
}
writer.commit();
writer.close();
Store.MetadataSnapshot secondMeta = store.getMetadata();
Store.LegacyChecksums checksums = new Store.LegacyChecksums();
Map<String, StoreFileMetaData> legacyMeta = new HashMap<>();
for (String file : store.directory().listAll()) {
if (file.equals("write.lock") || file.equals(IndexFileNames.OLD_SEGMENTS_GEN)) {
continue;
}
BytesRef hash = new BytesRef();
if (file.startsWith("segments")) {
hash = Store.MetadataSnapshot.hashFile(store.directory(), file);
}
StoreFileMetaData storeFileMetaData =
new StoreFileMetaData(
file, store.directory().fileLength(file), file + "checksum", null, hash);
legacyMeta.put(file, storeFileMetaData);
checksums.add(storeFileMetaData);
}
checksums.write(
store); // write one checksum file here - we expect it to survive all the cleanups
if (randomBoolean()) {
store.cleanupAndVerify("test", firstMeta);
String[] strings = store.directory().listAll();
int numChecksums = 0;
int numNotFound = 0;
for (String file : strings) {
assertTrue(firstMeta.contains(file) || Store.isChecksum(file));
if (Store.isChecksum(file)) {
numChecksums++;
} else if (secondMeta.contains(file) == false) {
numNotFound++;
}
}
assertTrue(
"at least one file must not be in here since we have two commits?", numNotFound > 0);
assertEquals(
"we wrote one checksum but it's gone now? - checksums are supposed to be kept",
numChecksums,
1);
} else {
store.cleanupAndVerify("test", secondMeta);
String[] strings = store.directory().listAll();
int numChecksums = 0;
int numNotFound = 0;
for (String file : strings) {
assertTrue(secondMeta.contains(file) || Store.isChecksum(file));
if (Store.isChecksum(file)) {
numChecksums++;
} else if (firstMeta.contains(file) == false) {
numNotFound++;
}
}
assertTrue(
"at least one file must not be in here since we have two commits?", numNotFound > 0);
assertEquals(
"we wrote one checksum but it's gone now? - checksums are supposed to be kept",
numChecksums,
1);
}
store.deleteContent();
IOUtils.close(store);
}
public void runTest(String testName) throws Exception {
failed.set(false);
addCount.set(0);
delCount.set(0);
packCount.set(0);
final long t0 = System.currentTimeMillis();
Random random = new Random(random().nextLong());
final LineFileDocs docs = new LineFileDocs(random, true);
final Path tempDir = createTempDir(testName);
dir = getDirectory(newMockFSDirectory(tempDir)); // some subclasses rely on this being MDW
if (dir instanceof BaseDirectoryWrapper) {
((BaseDirectoryWrapper) dir)
.setCheckIndexOnClose(false); // don't double-checkIndex, we do it ourselves.
}
MockAnalyzer analyzer = new MockAnalyzer(random());
analyzer.setMaxTokenLength(TestUtil.nextInt(random(), 1, IndexWriter.MAX_TERM_LENGTH));
final IndexWriterConfig conf = newIndexWriterConfig(analyzer).setCommitOnClose(false);
conf.setInfoStream(new FailOnNonBulkMergesInfoStream());
if (conf.getMergePolicy() instanceof MockRandomMergePolicy) {
((MockRandomMergePolicy) conf.getMergePolicy()).setDoNonBulkMerges(false);
}
if (LuceneTestCase.TEST_NIGHTLY) {
// newIWConfig makes smallish max seg size, which
// results in tons and tons of segments for this test
// when run nightly:
MergePolicy mp = conf.getMergePolicy();
if (mp instanceof TieredMergePolicy) {
((TieredMergePolicy) mp).setMaxMergedSegmentMB(5000.);
} else if (mp instanceof LogByteSizeMergePolicy) {
((LogByteSizeMergePolicy) mp).setMaxMergeMB(1000.);
} else if (mp instanceof LogMergePolicy) {
((LogMergePolicy) mp).setMaxMergeDocs(100000);
}
}
conf.setMergedSegmentWarmer(
new IndexWriter.IndexReaderWarmer() {
@Override
public void warm(LeafReader reader) throws IOException {
if (VERBOSE) {
System.out.println("TEST: now warm merged reader=" + reader);
}
warmed.put(((SegmentReader) reader).core, Boolean.TRUE);
final int maxDoc = reader.maxDoc();
final Bits liveDocs = reader.getLiveDocs();
int sum = 0;
final int inc = Math.max(1, maxDoc / 50);
for (int docID = 0; docID < maxDoc; docID += inc) {
if (liveDocs == null || liveDocs.get(docID)) {
final StoredDocument doc = reader.document(docID);
sum += doc.getFields().size();
}
}
IndexSearcher searcher = newSearcher(reader);
sum += searcher.search(new TermQuery(new Term("body", "united")), 10).totalHits;
if (VERBOSE) {
System.out.println("TEST: warm visited " + sum + " fields");
}
}
});
if (VERBOSE) {
conf.setInfoStream(
new PrintStreamInfoStream(System.out) {
@Override
public void message(String component, String message) {
if ("TP".equals(component)) {
return; // ignore test points!
}
super.message(component, message);
}
});
}
writer = new IndexWriter(dir, conf);
TestUtil.reduceOpenFiles(writer);
final ExecutorService es =
random().nextBoolean()
? null
: Executors.newCachedThreadPool(new NamedThreadFactory(testName));
doAfterWriter(es);
final int NUM_INDEX_THREADS = TestUtil.nextInt(random(), 2, 4);
final int RUN_TIME_SEC = LuceneTestCase.TEST_NIGHTLY ? 300 : RANDOM_MULTIPLIER;
final Set<String> delIDs = Collections.synchronizedSet(new HashSet<String>());
final Set<String> delPackIDs = Collections.synchronizedSet(new HashSet<String>());
final List<SubDocs> allSubDocs = Collections.synchronizedList(new ArrayList<SubDocs>());
final long stopTime = System.currentTimeMillis() + RUN_TIME_SEC * 1000;
final Thread[] indexThreads =
launchIndexingThreads(docs, NUM_INDEX_THREADS, stopTime, delIDs, delPackIDs, allSubDocs);
if (VERBOSE) {
System.out.println(
"TEST: DONE start "
+ NUM_INDEX_THREADS
+ " indexing threads ["
+ (System.currentTimeMillis() - t0)
+ " ms]");
}
// Let index build up a bit
Thread.sleep(100);
doSearching(es, stopTime);
if (VERBOSE) {
System.out.println("TEST: all searching done [" + (System.currentTimeMillis() - t0) + " ms]");
}
for (int thread = 0; thread < indexThreads.length; thread++) {
indexThreads[thread].join();
}
if (VERBOSE) {
System.out.println(
"TEST: done join indexing threads ["
+ (System.currentTimeMillis() - t0)
+ " ms]; addCount="
+ addCount
+ " delCount="
+ delCount);
}
final IndexSearcher s = getFinalSearcher();
if (VERBOSE) {
System.out.println("TEST: finalSearcher=" + s);
}
assertFalse(failed.get());
boolean doFail = false;
// Verify: make sure delIDs are in fact deleted:
for (String id : delIDs) {
final TopDocs hits = s.search(new TermQuery(new Term("docid", id)), 1);
if (hits.totalHits != 0) {
System.out.println(
"doc id="
+ id
+ " is supposed to be deleted, but got "
+ hits.totalHits
+ " hits; first docID="
+ hits.scoreDocs[0].doc);
doFail = true;
}
}
// Verify: make sure delPackIDs are in fact deleted:
for (String id : delPackIDs) {
final TopDocs hits = s.search(new TermQuery(new Term("packID", id)), 1);
if (hits.totalHits != 0) {
System.out.println(
"packID=" + id + " is supposed to be deleted, but got " + hits.totalHits + " matches");
doFail = true;
}
}
// Verify: make sure each group of sub-docs are still in docID order:
for (SubDocs subDocs : allSubDocs) {
TopDocs hits = s.search(new TermQuery(new Term("packID", subDocs.packID)), 20);
if (!subDocs.deleted) {
// We sort by relevance but the scores should be identical so sort falls back to by docID:
if (hits.totalHits != subDocs.subIDs.size()) {
System.out.println(
"packID="
+ subDocs.packID
+ ": expected "
+ subDocs.subIDs.size()
+ " hits but got "
+ hits.totalHits);
doFail = true;
} else {
int lastDocID = -1;
int startDocID = -1;
for (ScoreDoc scoreDoc : hits.scoreDocs) {
final int docID = scoreDoc.doc;
if (lastDocID != -1) {
assertEquals(1 + lastDocID, docID);
} else {
startDocID = docID;
}
lastDocID = docID;
final StoredDocument doc = s.doc(docID);
assertEquals(subDocs.packID, doc.get("packID"));
}
lastDocID = startDocID - 1;
for (String subID : subDocs.subIDs) {
hits = s.search(new TermQuery(new Term("docid", subID)), 1);
assertEquals(1, hits.totalHits);
final int docID = hits.scoreDocs[0].doc;
if (lastDocID != -1) {
assertEquals(1 + lastDocID, docID);
}
lastDocID = docID;
}
}
} else {
// Pack was deleted -- make sure its docs are
// deleted. We can't verify packID is deleted
// because we can re-use packID for update:
for (String subID : subDocs.subIDs) {
assertEquals(0, s.search(new TermQuery(new Term("docid", subID)), 1).totalHits);
}
}
}
// Verify: make sure all not-deleted docs are in fact
// not deleted:
final int endID = Integer.parseInt(docs.nextDoc().get("docid"));
docs.close();
for (int id = 0; id < endID; id++) {
String stringID = "" + id;
if (!delIDs.contains(stringID)) {
final TopDocs hits = s.search(new TermQuery(new Term("docid", stringID)), 1);
if (hits.totalHits != 1) {
System.out.println(
"doc id="
+ stringID
+ " is not supposed to be deleted, but got hitCount="
+ hits.totalHits
+ "; delIDs="
+ delIDs);
doFail = true;
}
}
}
assertFalse(doFail);
assertEquals(
"index=" + writer.segString() + " addCount=" + addCount + " delCount=" + delCount,
addCount.get() - delCount.get(),
s.getIndexReader().numDocs());
releaseSearcher(s);
writer.commit();
assertEquals(
"index=" + writer.segString() + " addCount=" + addCount + " delCount=" + delCount,
addCount.get() - delCount.get(),
writer.numDocs());
doClose();
try {
writer.commit();
} finally {
writer.close();
}
// Cannot close until after writer is closed because
// writer has merged segment warmer that uses IS to run
// searches, and that IS may be using this es!
if (es != null) {
es.shutdown();
es.awaitTermination(1, TimeUnit.SECONDS);
}
TestUtil.checkIndex(dir);
dir.close();
IOUtils.rm(tempDir);
if (VERBOSE) {
System.out.println("TEST: done [" + (System.currentTimeMillis() - t0) + " ms]");
}
}