private IndexWriter createWriter() throws IOException {
IndexWriter indexWriter = null;
try {
// release locks when started
if (IndexWriter.isLocked(store.directory())) {
logger.warn("shard is locked, releasing lock");
IndexWriter.unlock(store.directory());
}
boolean create = !IndexReader.indexExists(store.directory());
indexWriter =
new IndexWriter(
store.directory(),
analysisService.defaultIndexAnalyzer(),
create,
deletionPolicy,
IndexWriter.MaxFieldLength.UNLIMITED);
indexWriter.setMergeScheduler(mergeScheduler.newMergeScheduler());
indexWriter.setMergePolicy(mergePolicyProvider.newMergePolicy(indexWriter));
indexWriter.setSimilarity(similarityService.defaultIndexSimilarity());
indexWriter.setRAMBufferSizeMB(indexingBufferSize.mbFrac());
indexWriter.setTermIndexInterval(termIndexInterval);
} catch (IOException e) {
safeClose(indexWriter);
throw e;
}
return indexWriter;
}
@Override
public void messageReceived(final RecoveryFileChunkRequest request, TransportChannel channel)
throws Exception {
try (RecoveriesCollection.StatusRef statusRef =
onGoingRecoveries.getStatusSafe(request.recoveryId(), request.shardId())) {
final RecoveryStatus recoveryStatus = statusRef.status();
final Store store = recoveryStatus.store();
recoveryStatus.state().getTranslog().totalOperations(request.totalTranslogOps());
final RecoveryState.Index indexState = recoveryStatus.state().getIndex();
if (request.sourceThrottleTimeInNanos() != RecoveryState.Index.UNKNOWN) {
indexState.addSourceThrottling(request.sourceThrottleTimeInNanos());
}
IndexOutput indexOutput;
if (request.position() == 0) {
indexOutput =
recoveryStatus.openAndPutIndexOutput(request.name(), request.metadata(), store);
} else {
indexOutput = recoveryStatus.getOpenIndexOutput(request.name());
}
BytesReference content = request.content();
if (!content.hasArray()) {
content = content.toBytesArray();
}
RateLimiter rl = recoverySettings.rateLimiter();
if (rl != null) {
long bytes = bytesSinceLastPause.addAndGet(content.length());
if (bytes > rl.getMinPauseCheckBytes()) {
// Time to pause
bytesSinceLastPause.addAndGet(-bytes);
long throttleTimeInNanos = rl.pause(bytes);
indexState.addTargetThrottling(throttleTimeInNanos);
recoveryStatus.indexShard().recoveryStats().addThrottleTime(throttleTimeInNanos);
}
}
indexOutput.writeBytes(content.array(), content.arrayOffset(), content.length());
indexState.addRecoveredBytesToFile(request.name(), content.length());
if (indexOutput.getFilePointer() >= request.length() || request.lastChunk()) {
try {
Store.verify(indexOutput);
} finally {
// we are done
indexOutput.close();
}
// write the checksum
recoveryStatus.legacyChecksums().add(request.metadata());
final String temporaryFileName = recoveryStatus.getTempNameForFile(request.name());
assert Arrays.asList(store.directory().listAll()).contains(temporaryFileName);
store.directory().sync(Collections.singleton(temporaryFileName));
IndexOutput remove = recoveryStatus.removeOpenIndexOutputs(request.name());
assert remove == null || remove == indexOutput; // remove maybe null if we got finished
}
}
channel.sendResponse(TransportResponse.Empty.INSTANCE);
}
public void assertDeleteContent(Store store, DirectoryService service) throws IOException {
store.deleteContent();
assertThat(
Arrays.toString(store.directory().listAll()),
store.directory().listAll().length,
equalTo(0));
assertThat(store.stats().sizeInBytes(), equalTo(0l));
for (Directory dir : service.build()) {
assertThat(dir.listAll().length, equalTo(0));
}
}
@Test
public void testCleanUpWithLegacyChecksums() throws IOException {
Map<String, StoreFileMetaData> metaDataMap = new HashMap<>();
metaDataMap.put(
"segments_1",
new StoreFileMetaData("segments_1", 50, null, null, new BytesRef(new byte[] {1})));
metaDataMap.put(
"_0_1.del", new StoreFileMetaData("_0_1.del", 42, "foobarbaz", null, new BytesRef()));
Store.MetadataSnapshot snapshot = new Store.MetadataSnapshot(metaDataMap);
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));
for (String file : metaDataMap.keySet()) {
try (IndexOutput output = store.directory().createOutput(file, IOContext.DEFAULT)) {
BytesRef bytesRef = new BytesRef(TestUtil.randomRealisticUnicodeString(random(), 10, 1024));
output.writeBytes(bytesRef.bytes, bytesRef.offset, bytesRef.length);
CodecUtil.writeFooter(output);
}
}
store.verifyAfterCleanup(snapshot, snapshot);
store.deleteContent();
IOUtils.close(store);
}
private long newTransactionLogId() throws IOException {
try {
return IndexWriters.rollbackSegmentInfos(indexWriter).getVersion();
} catch (Exception e) {
return IndexReader.getCurrentVersion(store.directory());
}
}
@Override
public void messageReceived(RecoveryCleanFilesRequest request, TransportChannel channel)
throws Exception {
try (RecoveriesCollection.StatusRef statusRef =
onGoingRecoveries.getStatusSafe(request.recoveryId(), request.shardId())) {
final RecoveryStatus recoveryStatus = statusRef.status();
recoveryStatus.state().getTranslog().totalOperations(request.totalTranslogOps());
// first, we go and move files that were created with the recovery id suffix to
// the actual names, its ok if we have a corrupted index here, since we have replicas
// to recover from in case of a full cluster shutdown just when this code executes...
recoveryStatus
.indexShard()
.deleteShardState(); // we have to delete it first since even if we fail to rename the
// shard might be invalid
recoveryStatus.renameAllTempFiles();
final Store store = recoveryStatus.store();
// now write checksums
recoveryStatus.legacyChecksums().write(store);
Store.MetadataSnapshot sourceMetaData = request.sourceMetaSnapshot();
try {
store.cleanupAndVerify("recovery CleanFilesRequestHandler", sourceMetaData);
} catch (CorruptIndexException
| IndexFormatTooNewException
| IndexFormatTooOldException ex) {
// this is a fatal exception at this stage.
// this means we transferred files from the remote that have not be checksummed and they
// are
// broken. We have to clean up this shard entirely, remove all files and bubble it up to
// the
// source shard since this index might be broken there as well? The Source can handle this
// and checks
// its content on disk if possible.
try {
try {
store.removeCorruptionMarker();
} finally {
Lucene.cleanLuceneIndex(store.directory()); // clean up and delete all files
}
} catch (Throwable e) {
logger.debug("Failed to clean lucene index", e);
ex.addSuppressed(e);
}
RecoveryFailedException rfe =
new RecoveryFailedException(
recoveryStatus.state(), "failed to clean after recovery", ex);
recoveryStatus.fail(rfe, true);
throw rfe;
} catch (Exception ex) {
RecoveryFailedException rfe =
new RecoveryFailedException(
recoveryStatus.state(), "failed to clean after recovery", ex);
recoveryStatus.fail(rfe, true);
throw rfe;
}
channel.sendResponse(TransportResponse.Empty.INSTANCE);
}
}
void sendFiles(
Store store,
StoreFileMetaData[] files,
Function<StoreFileMetaData, OutputStream> outputStreamFactory)
throws Exception {
store.incRef();
try {
ArrayUtil.timSort(
files, (a, b) -> Long.compare(a.length(), b.length())); // send smallest first
for (int i = 0; i < files.length; i++) {
final StoreFileMetaData md = files[i];
try (final IndexInput indexInput =
store.directory().openInput(md.name(), IOContext.READONCE)) {
// it's fine that we are only having the indexInput in the try/with block. The copy
// methods handles
// exceptions during close correctly and doesn't hide the original exception.
Streams.copy(
new InputStreamIndexInput(indexInput, md.length()), outputStreamFactory.apply(md));
} catch (Exception e) {
final IOException corruptIndexException;
if ((corruptIndexException = ExceptionsHelper.unwrapCorruption(e)) != null) {
if (store.checkIntegrityNoException(md)
== false) { // we are corrupted on the primary -- fail!
logger.warn("{} Corrupted file detected {} checksum mismatch", shardId, md);
failEngine(corruptIndexException);
throw corruptIndexException;
} else { // corruption has happened on the way to replica
RemoteTransportException exception =
new RemoteTransportException(
"File corruption occurred on recovery but " + "checksums are ok", null);
exception.addSuppressed(e);
logger.warn(
(org.apache.logging.log4j.util.Supplier<?>)
() ->
new ParameterizedMessage(
"{} Remote file corruption on node {}, recovering {}. local checksum OK",
shardId,
request.targetNode(),
md),
corruptIndexException);
throw exception;
}
} else {
throw e;
}
}
}
} finally {
store.decRef();
}
}
public static void assertConsistent(Store store, Store.MetadataSnapshot metadata)
throws IOException {
for (String file : store.directory().listAll()) {
if (!IndexWriter.WRITE_LOCK_NAME.equals(file)
&& !IndexFileNames.OLD_SEGMENTS_GEN.equals(file)
&& !Store.isChecksum(file)) {
assertTrue(
file
+ " is not in the map: "
+ metadata.asMap().size()
+ " vs. "
+ store.directory().listAll().length,
metadata.asMap().containsKey(file));
} else {
assertFalse(
file
+ " is not in the map: "
+ metadata.asMap().size()
+ " vs. "
+ store.directory().listAll().length,
metadata.asMap().containsKey(file));
}
}
}
public static void checkIndex(ESLogger logger, Store store, ShardId shardId) {
if (store.tryIncRef()) {
logger.info("start check index");
try {
Directory dir = store.directory();
if (!Lucene.indexExists(dir)) {
return;
}
if (IndexWriter.isLocked(dir)) {
ESTestCase.checkIndexFailed = true;
throw new IllegalStateException("IndexWriter is still open on shard " + shardId);
}
try (CheckIndex checkIndex = new CheckIndex(dir)) {
BytesStreamOutput os = new BytesStreamOutput();
PrintStream out = new PrintStream(os, false, StandardCharsets.UTF_8.name());
checkIndex.setInfoStream(out);
out.flush();
CheckIndex.Status status = checkIndex.checkIndex();
if (!status.clean) {
ESTestCase.checkIndexFailed = true;
logger.warn(
"check index [failure] index files={}\n{}",
Arrays.toString(dir.listAll()),
new String(os.bytes().toBytes(), StandardCharsets.UTF_8));
throw new IOException("index check failure");
} else {
if (logger.isDebugEnabled()) {
logger.debug(
"check index [success]\n{}",
new String(os.bytes().toBytes(), StandardCharsets.UTF_8));
}
}
}
} catch (Exception e) {
logger.warn("failed to check index", e);
} finally {
logger.info("end check index");
store.decRef();
}
}
}
@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 testSendFiles() throws Throwable {
Settings settings =
Settings.builder()
.put("indices.recovery.concurrent_streams", 1)
.put("indices.recovery.concurrent_small_file_streams", 1)
.build();
final RecoverySettings recoverySettings = new RecoverySettings(settings, service);
StartRecoveryRequest request =
new StartRecoveryRequest(
shardId,
new DiscoveryNode("b", DummyTransportAddress.INSTANCE, Version.CURRENT),
new DiscoveryNode("b", DummyTransportAddress.INSTANCE, Version.CURRENT),
null,
RecoveryState.Type.STORE,
randomLong());
Store store = newStore(createTempDir());
RecoverySourceHandler handler =
new RecoverySourceHandler(null, request, recoverySettings, null, logger);
Directory dir = store.directory();
RandomIndexWriter writer = new RandomIndexWriter(random(), dir, newIndexWriterConfig());
int numDocs = randomIntBetween(10, 100);
for (int i = 0; i < numDocs; i++) {
Document document = new Document();
document.add(new StringField("id", Integer.toString(i), Field.Store.YES));
document.add(
newField("field", randomUnicodeOfCodepointLengthBetween(1, 10), TextField.TYPE_STORED));
writer.addDocument(document);
}
writer.commit();
Store.MetadataSnapshot metadata = store.getMetadata();
List<StoreFileMetaData> metas = new ArrayList<>();
for (StoreFileMetaData md : metadata) {
metas.add(md);
}
Store targetStore = newStore(createTempDir());
handler.sendFiles(
store,
metas.toArray(new StoreFileMetaData[0]),
(md) -> {
try {
return new IndexOutputOutputStream(
targetStore.createVerifyingOutput(md.name(), md, IOContext.DEFAULT)) {
@Override
public void close() throws IOException {
super.close();
store
.directory()
.sync(Collections.singleton(md.name())); // sync otherwise MDW will mess with it
}
};
} catch (IOException e) {
throw new RuntimeException(e);
}
});
Store.MetadataSnapshot targetStoreMetadata = targetStore.getMetadata();
Store.RecoveryDiff recoveryDiff = targetStoreMetadata.recoveryDiff(metadata);
assertEquals(metas.size(), recoveryDiff.identical.size());
assertEquals(0, recoveryDiff.different.size());
assertEquals(0, recoveryDiff.missing.size());
IndexReader reader = DirectoryReader.open(targetStore.directory());
assertEquals(numDocs, reader.maxDoc());
IOUtils.close(reader, writer, store, targetStore);
}
public void testHandleExceptinoOnSendSendFiles() throws Throwable {
Settings settings =
Settings.builder()
.put("indices.recovery.concurrent_streams", 1)
.put("indices.recovery.concurrent_small_file_streams", 1)
.build();
final RecoverySettings recoverySettings = new RecoverySettings(settings, service);
StartRecoveryRequest request =
new StartRecoveryRequest(
shardId,
new DiscoveryNode("b", DummyTransportAddress.INSTANCE, Version.CURRENT),
new DiscoveryNode("b", DummyTransportAddress.INSTANCE, Version.CURRENT),
null,
RecoveryState.Type.STORE,
randomLong());
Path tempDir = createTempDir();
Store store = newStore(tempDir, false);
AtomicBoolean failedEngine = new AtomicBoolean(false);
RecoverySourceHandler handler =
new RecoverySourceHandler(null, request, recoverySettings, null, logger) {
@Override
protected void failEngine(IOException cause) {
assertFalse(failedEngine.get());
failedEngine.set(true);
}
};
Directory dir = store.directory();
RandomIndexWriter writer = new RandomIndexWriter(random(), dir, newIndexWriterConfig());
int numDocs = randomIntBetween(10, 100);
for (int i = 0; i < numDocs; i++) {
Document document = new Document();
document.add(new StringField("id", Integer.toString(i), Field.Store.YES));
document.add(
newField("field", randomUnicodeOfCodepointLengthBetween(1, 10), TextField.TYPE_STORED));
writer.addDocument(document);
}
writer.commit();
writer.close();
Store.MetadataSnapshot metadata = store.getMetadata();
List<StoreFileMetaData> metas = new ArrayList<>();
for (StoreFileMetaData md : metadata) {
metas.add(md);
}
final boolean throwCorruptedIndexException = randomBoolean();
Store targetStore = newStore(createTempDir(), false);
try {
handler.sendFiles(
store,
metas.toArray(new StoreFileMetaData[0]),
(md) -> {
if (throwCorruptedIndexException) {
throw new RuntimeException(new CorruptIndexException("foo", "bar"));
} else {
throw new RuntimeException("boom");
}
});
fail("exception index");
} catch (RuntimeException ex) {
assertNull(ExceptionsHelper.unwrapCorruption(ex));
if (throwCorruptedIndexException) {
assertEquals(
ex.getMessage(), "[File corruption occurred on recovery but checksums are ok]");
} else {
assertEquals(ex.getMessage(), "boom");
}
} catch (CorruptIndexException ex) {
fail("not expected here");
}
assertFalse(failedEngine.get());
IOUtils.close(store, targetStore);
}
public void testHandleCorruptedIndexOnSendSendFiles() throws Throwable {
Settings settings =
Settings.builder()
.put("indices.recovery.concurrent_streams", 1)
.put("indices.recovery.concurrent_small_file_streams", 1)
.build();
final RecoverySettings recoverySettings = new RecoverySettings(settings, service);
StartRecoveryRequest request =
new StartRecoveryRequest(
shardId,
new DiscoveryNode("b", DummyTransportAddress.INSTANCE, Version.CURRENT),
new DiscoveryNode("b", DummyTransportAddress.INSTANCE, Version.CURRENT),
null,
RecoveryState.Type.STORE,
randomLong());
Path tempDir = createTempDir();
Store store = newStore(tempDir, false);
AtomicBoolean failedEngine = new AtomicBoolean(false);
RecoverySourceHandler handler =
new RecoverySourceHandler(null, request, recoverySettings, null, logger) {
@Override
protected void failEngine(IOException cause) {
assertFalse(failedEngine.get());
failedEngine.set(true);
}
};
Directory dir = store.directory();
RandomIndexWriter writer = new RandomIndexWriter(random(), dir, newIndexWriterConfig());
int numDocs = randomIntBetween(10, 100);
for (int i = 0; i < numDocs; i++) {
Document document = new Document();
document.add(new StringField("id", Integer.toString(i), Field.Store.YES));
document.add(
newField("field", randomUnicodeOfCodepointLengthBetween(1, 10), TextField.TYPE_STORED));
writer.addDocument(document);
}
writer.commit();
writer.close();
Store.MetadataSnapshot metadata = store.getMetadata();
List<StoreFileMetaData> metas = new ArrayList<>();
for (StoreFileMetaData md : metadata) {
metas.add(md);
}
CorruptionUtils.corruptFile(
getRandom(),
FileSystemUtils.files(
tempDir,
(p) ->
(p.getFileName().toString().equals("write.lock")
|| p.getFileName().toString().startsWith("extra"))
== false));
Store targetStore = newStore(createTempDir(), false);
try {
handler.sendFiles(
store,
metas.toArray(new StoreFileMetaData[0]),
(md) -> {
try {
return new IndexOutputOutputStream(
targetStore.createVerifyingOutput(md.name(), md, IOContext.DEFAULT)) {
@Override
public void close() throws IOException {
super.close();
store
.directory()
.sync(
Collections.singleton(md.name())); // sync otherwise MDW will mess with it
}
};
} catch (IOException e) {
throw new RuntimeException(e);
}
});
fail("corrupted index");
} catch (IOException ex) {
assertNotNull(ExceptionsHelper.unwrapCorruption(ex));
}
assertTrue(failedEngine.get());
IOUtils.close(store, targetStore);
}
@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);
}
private void recoverIndex(CommitPoint commitPoint, ImmutableMap<String, BlobMetaData> blobs)
throws Exception {
int numberOfFiles = 0;
long totalSize = 0;
int numberOfReusedFiles = 0;
long reusedTotalSize = 0;
List<CommitPoint.FileInfo> filesToRecover = Lists.newArrayList();
for (CommitPoint.FileInfo fileInfo : commitPoint.indexFiles()) {
String fileName = fileInfo.physicalName();
StoreFileMetaData md = null;
try {
md = store.metaData(fileName);
} catch (Exception e) {
// no file
}
// we don't compute checksum for segments, so always recover them
if (!fileName.startsWith("segments") && md != null && fileInfo.isSame(md)) {
numberOfFiles++;
totalSize += md.length();
numberOfReusedFiles++;
reusedTotalSize += md.length();
if (logger.isTraceEnabled()) {
logger.trace(
"not_recovering [{}], exists in local store and is same", fileInfo.physicalName());
}
} else {
if (logger.isTraceEnabled()) {
if (md == null) {
logger.trace(
"recovering [{}], does not exists in local store", fileInfo.physicalName());
} else {
logger.trace(
"recovering [{}], exists in local store but is different", fileInfo.physicalName());
}
}
numberOfFiles++;
totalSize += fileInfo.length();
filesToRecover.add(fileInfo);
}
}
recoveryStatus.index().files(numberOfFiles, totalSize, numberOfReusedFiles, reusedTotalSize);
if (filesToRecover.isEmpty()) {
logger.trace("no files to recover, all exists within the local store");
}
if (logger.isTraceEnabled()) {
logger.trace(
"recovering_files [{}] with total_size [{}], reusing_files [{}] with reused_size [{}]",
numberOfFiles,
new ByteSizeValue(totalSize),
numberOfReusedFiles,
new ByteSizeValue(reusedTotalSize));
}
final CountDownLatch latch = new CountDownLatch(filesToRecover.size());
final CopyOnWriteArrayList<Throwable> failures = new CopyOnWriteArrayList<Throwable>();
for (final CommitPoint.FileInfo fileToRecover : filesToRecover) {
recoverFile(fileToRecover, blobs, latch, failures);
}
try {
latch.await();
} catch (InterruptedException e) {
throw new IndexShardGatewayRecoveryException(
shardId, "Interrupted while recovering index", e);
}
if (!failures.isEmpty()) {
throw new IndexShardGatewayRecoveryException(
shardId, "Failed to recover index", failures.get(0));
}
// read the gateway data persisted
long version = -1;
try {
if (IndexReader.indexExists(store.directory())) {
version = IndexReader.getCurrentVersion(store.directory());
}
} catch (IOException e) {
throw new IndexShardGatewayRecoveryException(
shardId(), "Failed to fetch index version after copying it over", e);
}
recoveryStatus.index().updateVersion(version);
/// now, go over and clean files that are in the store, but were not in the gateway
try {
for (String storeFile : store.directory().listAll()) {
if (!commitPoint.containPhysicalIndexFile(storeFile)) {
try {
store.directory().deleteFile(storeFile);
} catch (Exception e) {
// ignore
}
}
}
} catch (Exception e) {
// ignore
}
}
@Test
public void testRenameFile() throws IOException {
final ShardId shardId = new ShardId(new Index("index"), 1);
DirectoryService directoryService = new LuceneManagedDirectoryService(random(), false);
Store store =
new Store(
shardId,
ImmutableSettings.EMPTY,
directoryService,
randomDistributor(directoryService),
new DummyShardLock(shardId));
{
IndexOutput output = store.directory().createOutput("foo.bar", IOContext.DEFAULT);
int iters = scaledRandomIntBetween(10, 100);
for (int i = 0; i < iters; i++) {
BytesRef bytesRef = new BytesRef(TestUtil.randomRealisticUnicodeString(random(), 10, 1024));
output.writeBytes(bytesRef.bytes, bytesRef.offset, bytesRef.length);
}
CodecUtil.writeFooter(output);
output.close();
}
store.renameFile("foo.bar", "bar.foo");
assertThat(store.directory().listAll().length, is(1));
final long lastChecksum;
try (IndexInput input = store.directory().openInput("bar.foo", IOContext.DEFAULT)) {
lastChecksum = CodecUtil.checksumEntireFile(input);
}
try {
store.directory().openInput("foo.bar", IOContext.DEFAULT);
fail("file was renamed");
} catch (FileNotFoundException | NoSuchFileException ex) {
// expected
}
{
IndexOutput output = store.directory().createOutput("foo.bar", IOContext.DEFAULT);
int iters = scaledRandomIntBetween(10, 100);
for (int i = 0; i < iters; i++) {
BytesRef bytesRef = new BytesRef(TestUtil.randomRealisticUnicodeString(random(), 10, 1024));
output.writeBytes(bytesRef.bytes, bytesRef.offset, bytesRef.length);
}
CodecUtil.writeFooter(output);
output.close();
}
DistributorDirectory distributorDirectory =
DirectoryUtils.getLeaf(store.directory(), DistributorDirectory.class);
if (distributorDirectory != null
&& distributorDirectory.getDirectory("foo.bar")
!= distributorDirectory.getDirectory("bar.foo")) {
try {
store.renameFile("foo.bar", "bar.foo");
fail("target file already exists in a different directory");
} catch (IOException ex) {
// expected
}
try (IndexInput input = store.directory().openInput("bar.foo", IOContext.DEFAULT)) {
assertThat(lastChecksum, equalTo(CodecUtil.checksumEntireFile(input)));
}
assertThat(store.directory().listAll().length, is(2));
assertDeleteContent(store, directoryService);
IOUtils.close(store);
} else {
store.renameFile("foo.bar", "bar.foo");
assertThat(store.directory().listAll().length, is(1));
assertDeleteContent(store, directoryService);
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);
}
// IF THIS TEST FAILS ON UPGRADE GO LOOK AT THE
// OldSIMockingCodec!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
@Test
public void testWriteLegacyChecksums() 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
final boolean usesOldCodec = randomBoolean();
IndexWriter writer =
new IndexWriter(
store.directory(),
newIndexWriterConfig(random(), new MockAnalyzer(random()))
.setCodec(usesOldCodec ? new OldSIMockingCodec() : 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.close();
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);
metadata = store.getMetadata();
Map<String, StoreFileMetaData> stringStoreFileMetaDataMap = metadata.asMap();
assertThat(legacyMeta.size(), equalTo(stringStoreFileMetaDataMap.size()));
if (usesOldCodec) {
for (StoreFileMetaData meta : legacyMeta.values()) {
assertTrue(meta.toString(), stringStoreFileMetaDataMap.containsKey(meta.name()));
assertEquals(meta.name() + "checksum", meta.checksum());
assertTrue(
meta + " vs. " + stringStoreFileMetaDataMap.get(meta.name()),
stringStoreFileMetaDataMap.get(meta.name()).isSame(meta));
}
} else {
// even if we have a legacy checksum - if we use a new codec we should reuse
for (StoreFileMetaData meta : legacyMeta.values()) {
assertTrue(meta.toString(), stringStoreFileMetaDataMap.containsKey(meta.name()));
assertFalse(
meta + " vs. " + stringStoreFileMetaDataMap.get(meta.name()),
stringStoreFileMetaDataMap.get(meta.name()).isSame(meta));
StoreFileMetaData storeFileMetaData = metadata.get(meta.name());
try (IndexInput input =
store.openVerifyingInput(meta.name(), IOContext.DEFAULT, storeFileMetaData)) {
assertTrue(storeFileMetaData.toString(), input instanceof Store.VerifyingIndexInput);
input.seek(meta.length());
Store.verify(input);
}
}
}
assertDeleteContent(store, directoryService);
IOUtils.close(store);
}