@Override
public synchronized void close() throws Exception {
if (closed == false) {
closed = true;
for (IndexShard shard : this) {
shard.close("eol", false);
IOUtils.close(shard.store());
}
} else {
throw new AlreadyClosedException("too bad");
}
}
// NOTE: O(numShards) cost, but numShards should be smallish?
private long getAvgShardSizeInBytes() throws IOException {
long sum = 0;
int count = 0;
for (IndexShard indexShard : this) {
sum += indexShard.store().stats().sizeInBytes();
count++;
}
if (count == 0) {
return -1L;
} else {
return sum / count;
}
}
public synchronized void removeShard(int shardId, String reason) {
final ShardId sId = new ShardId(index(), shardId);
final IndexShard indexShard;
if (shards.containsKey(shardId) == false) {
return;
}
logger.debug("[{}] closing... (reason: [{}])", shardId, reason);
HashMap<Integer, IndexShard> newShards = new HashMap<>(shards);
indexShard = newShards.remove(shardId);
shards = unmodifiableMap(newShards);
closeShard(reason, sId, indexShard, indexShard.store(), indexShard.getIndexEventListener());
logger.debug("[{}] closed (reason: [{}])", shardId, reason);
}
private StoreFilesMetaData listStoreMetaData(ShardId shardId) throws IOException {
logger.trace("listing store meta data for {}", shardId);
long startTimeNS = System.nanoTime();
boolean exists = false;
try {
IndexService indexService = indicesService.indexService(shardId.index().name());
if (indexService != null) {
IndexShard indexShard = indexService.shard(shardId.id());
if (indexShard != null) {
final Store store = indexShard.store();
store.incRef();
try {
exists = true;
return new StoreFilesMetaData(true, shardId, store.getMetadataOrEmpty());
} finally {
store.decRef();
}
}
}
// try and see if we an list unallocated
IndexMetaData metaData = clusterService.state().metaData().index(shardId.index().name());
if (metaData == null) {
return new StoreFilesMetaData(false, shardId, Store.MetadataSnapshot.EMPTY);
}
String storeType = metaData.getSettings().get(IndexStoreModule.STORE_TYPE, "fs");
if (!storeType.contains("fs")) {
return new StoreFilesMetaData(false, shardId, Store.MetadataSnapshot.EMPTY);
}
final ShardPath shardPath =
ShardPath.loadShardPath(logger, nodeEnv, shardId, metaData.getSettings());
if (shardPath == null) {
return new StoreFilesMetaData(false, shardId, Store.MetadataSnapshot.EMPTY);
}
return new StoreFilesMetaData(
false, shardId, Store.readMetadataSnapshot(shardPath.resolveIndex(), logger));
} finally {
TimeValue took = new TimeValue(System.nanoTime() - startTimeNS, TimeUnit.NANOSECONDS);
if (exists) {
logger.debug("{} loaded store meta data (took [{}])", shardId, took);
} else {
logger.trace("{} didn't find any store meta data to load (took [{}])", shardId, took);
}
}
}
/**
* Perform phase1 of the recovery operations. Once this {@link IndexCommit} snapshot has been
* performed no commit operations (files being fsync'd) are effectively allowed on this index
* until all recovery phases are done
*
* <p>Phase1 examines the segment files on the target node and copies over the segments that are
* missing. Only segments that have the same size and checksum can be reused
*/
public void phase1(final IndexCommit snapshot, final Translog.View translogView) {
cancellableThreads.checkForCancel();
// Total size of segment files that are recovered
long totalSize = 0;
// Total size of segment files that were able to be re-used
long existingTotalSize = 0;
final Store store = shard.store();
store.incRef();
try {
StopWatch stopWatch = new StopWatch().start();
final Store.MetadataSnapshot recoverySourceMetadata;
try {
recoverySourceMetadata = store.getMetadata(snapshot);
} catch (CorruptIndexException | IndexFormatTooOldException | IndexFormatTooNewException ex) {
shard.failShard("recovery", ex);
throw ex;
}
for (String name : snapshot.getFileNames()) {
final StoreFileMetaData md = recoverySourceMetadata.get(name);
if (md == null) {
logger.info(
"Snapshot differs from actual index for file: {} meta: {}",
name,
recoverySourceMetadata.asMap());
throw new CorruptIndexException(
"Snapshot differs from actual index - maybe index was removed metadata has "
+ recoverySourceMetadata.asMap().size()
+ " files",
name);
}
}
// Generate a "diff" of all the identical, different, and missing
// segment files on the target node, using the existing files on
// the source node
String recoverySourceSyncId = recoverySourceMetadata.getSyncId();
String recoveryTargetSyncId = request.metadataSnapshot().getSyncId();
final boolean recoverWithSyncId =
recoverySourceSyncId != null && recoverySourceSyncId.equals(recoveryTargetSyncId);
if (recoverWithSyncId) {
final long numDocsTarget = request.metadataSnapshot().getNumDocs();
final long numDocsSource = recoverySourceMetadata.getNumDocs();
if (numDocsTarget != numDocsSource) {
throw new IllegalStateException(
"try to recover "
+ request.shardId()
+ " from primary shard with sync id but number "
+ "of docs differ: "
+ numDocsTarget
+ " ("
+ request.sourceNode().getName()
+ ", primary) vs "
+ numDocsSource
+ "("
+ request.targetNode().getName()
+ ")");
}
// we shortcut recovery here because we have nothing to copy. but we must still start the
// engine on the target.
// so we don't return here
logger.trace(
"[{}][{}] skipping [phase1] to {} - identical sync id [{}] found on both source and target",
indexName,
shardId,
request.targetNode(),
recoverySourceSyncId);
} else {
final Store.RecoveryDiff diff =
recoverySourceMetadata.recoveryDiff(request.metadataSnapshot());
for (StoreFileMetaData md : diff.identical) {
response.phase1ExistingFileNames.add(md.name());
response.phase1ExistingFileSizes.add(md.length());
existingTotalSize += md.length();
if (logger.isTraceEnabled()) {
logger.trace(
"[{}][{}] recovery [phase1] to {}: not recovering [{}], exists in local store and has checksum [{}],"
+ " size [{}]",
indexName,
shardId,
request.targetNode(),
md.name(),
md.checksum(),
md.length());
}
totalSize += md.length();
}
List<StoreFileMetaData> phase1Files =
new ArrayList<>(diff.different.size() + diff.missing.size());
phase1Files.addAll(diff.different);
phase1Files.addAll(diff.missing);
for (StoreFileMetaData md : phase1Files) {
if (request.metadataSnapshot().asMap().containsKey(md.name())) {
logger.trace(
"[{}][{}] recovery [phase1] to {}: recovering [{}], exists in local store, but is different: remote "
+ "[{}], local [{}]",
indexName,
shardId,
request.targetNode(),
md.name(),
request.metadataSnapshot().asMap().get(md.name()),
md);
} else {
logger.trace(
"[{}][{}] recovery [phase1] to {}: recovering [{}], does not exists in remote",
indexName,
shardId,
request.targetNode(),
md.name());
}
response.phase1FileNames.add(md.name());
response.phase1FileSizes.add(md.length());
totalSize += md.length();
}
response.phase1TotalSize = totalSize;
response.phase1ExistingTotalSize = existingTotalSize;
logger.trace(
"[{}][{}] recovery [phase1] to {}: recovering_files [{}] with total_size [{}], reusing_files [{}] with "
+ "total_size [{}]",
indexName,
shardId,
request.targetNode(),
response.phase1FileNames.size(),
new ByteSizeValue(totalSize),
response.phase1ExistingFileNames.size(),
new ByteSizeValue(existingTotalSize));
cancellableThreads.execute(
() ->
recoveryTarget.receiveFileInfo(
response.phase1FileNames,
response.phase1FileSizes,
response.phase1ExistingFileNames,
response.phase1ExistingFileSizes,
translogView.totalOperations()));
// How many bytes we've copied since we last called RateLimiter.pause
final Function<StoreFileMetaData, OutputStream> outputStreamFactories =
md ->
new BufferedOutputStream(
new RecoveryOutputStream(md, translogView), chunkSizeInBytes);
sendFiles(
store,
phase1Files.toArray(new StoreFileMetaData[phase1Files.size()]),
outputStreamFactories);
// Send the CLEAN_FILES request, which takes all of the files that
// were transferred and renames them from their temporary file
// names to the actual file names. It also writes checksums for
// the files after they have been renamed.
//
// Once the files have been renamed, any other files that are not
// related to this recovery (out of date segments, for example)
// are deleted
try {
cancellableThreads.executeIO(
() ->
recoveryTarget.cleanFiles(
translogView.totalOperations(), recoverySourceMetadata));
} catch (RemoteTransportException | IOException targetException) {
final IOException corruptIndexException;
// we realized that after the index was copied and we wanted to finalize the recovery
// the index was corrupted:
// - maybe due to a broken segments file on an empty index (transferred with no
// checksum)
// - maybe due to old segments without checksums or length only checks
if ((corruptIndexException = ExceptionsHelper.unwrapCorruption(targetException))
!= null) {
try {
final Store.MetadataSnapshot recoverySourceMetadata1 = store.getMetadata(snapshot);
StoreFileMetaData[] metadata =
StreamSupport.stream(recoverySourceMetadata1.spliterator(), false)
.toArray(size -> new StoreFileMetaData[size]);
ArrayUtil.timSort(
metadata,
(o1, o2) -> {
return Long.compare(o1.length(), o2.length()); // check small files first
});
for (StoreFileMetaData md : metadata) {
cancellableThreads.checkForCancel();
logger.debug(
"{} checking integrity for file {} after remove corruption exception",
shard.shardId(),
md);
if (store.checkIntegrityNoException(md)
== false) { // we are corrupted on the primary -- fail!
shard.failShard("recovery", corruptIndexException);
logger.warn(
"{} Corrupted file detected {} checksum mismatch", shard.shardId(), md);
throw corruptIndexException;
}
}
} catch (IOException ex) {
targetException.addSuppressed(ex);
throw targetException;
}
// corruption has happened on the way to replica
RemoteTransportException exception =
new RemoteTransportException(
"File corruption occurred on recovery but " + "checksums are ok", null);
exception.addSuppressed(targetException);
logger.warn(
(org.apache.logging.log4j.util.Supplier<?>)
() ->
new ParameterizedMessage(
"{} Remote file corruption during finalization of recovery on node {}. local checksum OK",
shard.shardId(),
request.targetNode()),
corruptIndexException);
throw exception;
} else {
throw targetException;
}
}
}
prepareTargetForTranslog(translogView.totalOperations());
logger.trace(
"[{}][{}] recovery [phase1] to {}: took [{}]",
indexName,
shardId,
request.targetNode(),
stopWatch.totalTime());
response.phase1Time = stopWatch.totalTime().millis();
} catch (Exception e) {
throw new RecoverFilesRecoveryException(
request.shardId(), response.phase1FileNames.size(), new ByteSizeValue(totalSize), e);
} finally {
store.decRef();
}
}
/**
* Perform phase1 of the recovery operations. Once this {@link SnapshotIndexCommit} snapshot has
* been performed no commit operations (files being fsync'd) are effectively allowed on this index
* until all recovery phases are done
*
* <p>Phase1 examines the segment files on the target node and copies over the segments that are
* missing. Only segments that have the same size and checksum can be reused
*
* <p>{@code InternalEngine#recover} is responsible for snapshotting the index and releasing the
* snapshot once all 3 phases of recovery are complete
*/
@Override
public void phase1(final SnapshotIndexCommit snapshot) throws ElasticsearchException {
cancellableThreads.checkForCancel();
// Total size of segment files that are recovered
long totalSize = 0;
// Total size of segment files that were able to be re-used
long existingTotalSize = 0;
final Store store = shard.store();
store.incRef();
try {
StopWatch stopWatch = new StopWatch().start();
final Store.MetadataSnapshot recoverySourceMetadata = store.getMetadata(snapshot);
for (String name : snapshot.getFiles()) {
final StoreFileMetaData md = recoverySourceMetadata.get(name);
if (md == null) {
logger.info(
"Snapshot differs from actual index for file: {} meta: {}",
name,
recoverySourceMetadata.asMap());
throw new CorruptIndexException(
"Snapshot differs from actual index - maybe index was removed metadata has "
+ recoverySourceMetadata.asMap().size()
+ " files");
}
}
String recoverySourceSyncId = recoverySourceMetadata.getSyncId();
String recoveryTargetSyncId = request.metadataSnapshot().getSyncId();
final boolean recoverWithSyncId =
recoverySourceSyncId != null && recoverySourceSyncId.equals(recoveryTargetSyncId);
if (recoverWithSyncId) {
final long numDocsTarget = request.metadataSnapshot().getNumDocs();
final long numDocsSource = recoverySourceMetadata.getNumDocs();
if (numDocsTarget != numDocsSource) {
throw new IllegalStateException(
"try to recover "
+ request.shardId()
+ " from primary shard with sync id but number of docs differ: "
+ numDocsTarget
+ " ("
+ request.sourceNode().getName()
+ ", primary) vs "
+ numDocsSource
+ "("
+ request.targetNode().getName()
+ ")");
}
// we shortcut recovery here because we have nothing to copy. but we must still start the
// engine on the target.
// so we don't return here
logger.trace(
"[{}][{}] skipping [phase1] to {} - identical sync id [{}] found on both source and target",
indexName,
shardId,
request.targetNode(),
recoverySourceSyncId);
} else {
// Generate a "diff" of all the identical, different, and missing
// segment files on the target node, using the existing files on
// the source node
final Store.RecoveryDiff diff =
recoverySourceMetadata.recoveryDiff(request.metadataSnapshot());
for (StoreFileMetaData md : diff.identical) {
response.phase1ExistingFileNames.add(md.name());
response.phase1ExistingFileSizes.add(md.length());
existingTotalSize += md.length();
if (logger.isTraceEnabled()) {
logger.trace(
"[{}][{}] recovery [phase1] to {}: not recovering [{}], exists in local store and has checksum [{}], size [{}]",
indexName,
shardId,
request.targetNode(),
md.name(),
md.checksum(),
md.length());
}
totalSize += md.length();
}
for (StoreFileMetaData md : Iterables.concat(diff.different, diff.missing)) {
if (request.metadataSnapshot().asMap().containsKey(md.name())) {
logger.trace(
"[{}][{}] recovery [phase1] to {}: recovering [{}], exists in local store, but is different: remote [{}], local [{}]",
indexName,
shardId,
request.targetNode(),
md.name(),
request.metadataSnapshot().get(md.name()),
md);
} else {
logger.trace(
"[{}][{}] recovery [phase1] to {}: recovering [{}], does not exists in remote",
indexName,
shardId,
request.targetNode(),
md.name());
}
response.phase1FileNames.add(md.name());
response.phase1FileSizes.add(md.length());
totalSize += md.length();
}
response.phase1TotalSize = totalSize;
response.phase1ExistingTotalSize = existingTotalSize;
logger.trace(
"[{}][{}] recovery [phase1] to {}: recovering_files [{}] with total_size [{}], reusing_files [{}] with total_size [{}]",
indexName,
shardId,
request.targetNode(),
response.phase1FileNames.size(),
new ByteSizeValue(totalSize),
response.phase1ExistingFileNames.size(),
new ByteSizeValue(existingTotalSize));
cancellableThreads.execute(
new Interruptable() {
@Override
public void run() throws InterruptedException {
RecoveryFilesInfoRequest recoveryInfoFilesRequest =
new RecoveryFilesInfoRequest(
request.recoveryId(),
request.shardId(),
response.phase1FileNames,
response.phase1FileSizes,
response.phase1ExistingFileNames,
response.phase1ExistingFileSizes,
shard.translog().estimatedNumberOfOperations(),
response.phase1TotalSize,
response.phase1ExistingTotalSize);
transportService
.submitRequest(
request.targetNode(),
RecoveryTarget.Actions.FILES_INFO,
recoveryInfoFilesRequest,
TransportRequestOptions.options()
.withTimeout(recoverySettings.internalActionTimeout()),
EmptyTransportResponseHandler.INSTANCE_SAME)
.txGet();
}
});
// This latch will be used to wait until all files have been transferred to the target node
final CountDownLatch latch = new CountDownLatch(response.phase1FileNames.size());
final CopyOnWriteArrayList<Throwable> exceptions = new CopyOnWriteArrayList<>();
final AtomicReference<Throwable> corruptedEngine = new AtomicReference<>();
int fileIndex = 0;
ThreadPoolExecutor pool;
// How many bytes we've copied since we last called RateLimiter.pause
final AtomicLong bytesSinceLastPause = new AtomicLong();
for (final String name : response.phase1FileNames) {
long fileSize = response.phase1FileSizes.get(fileIndex);
// Files are split into two categories, files that are "small"
// (under 5mb) and other files. Small files are transferred
// using a separate thread pool dedicated to small files.
//
// The idea behind this is that while we are transferring an
// older, large index, a user may create a new index, but that
// index will not be able to recover until the large index
// finishes, by using two different thread pools we can allow
// tiny files (like segments for a brand new index) to be
// recovered while ongoing large segment recoveries are
// happening. It also allows these pools to be configured
// separately.
if (fileSize > RecoverySettings.SMALL_FILE_CUTOFF_BYTES) {
pool = recoverySettings.concurrentStreamPool();
} else {
pool = recoverySettings.concurrentSmallFileStreamPool();
}
pool.execute(
new AbstractRunnable() {
@Override
public void onFailure(Throwable t) {
// we either got rejected or the store can't be incremented / we are canceled
logger.debug("Failed to transfer file [" + name + "] on recovery");
}
public void onAfter() {
// Signify this file has completed by decrementing the latch
latch.countDown();
}
@Override
protected void doRun() {
cancellableThreads.checkForCancel();
store.incRef();
final StoreFileMetaData md = recoverySourceMetadata.get(name);
try (final IndexInput indexInput =
store.directory().openInput(name, IOContext.READONCE)) {
final int BUFFER_SIZE = (int) recoverySettings.fileChunkSize().bytes();
final byte[] buf = new byte[BUFFER_SIZE];
boolean shouldCompressRequest = recoverySettings.compress();
if (CompressorFactory.isCompressed(indexInput)) {
shouldCompressRequest = false;
}
final long len = indexInput.length();
long readCount = 0;
final TransportRequestOptions requestOptions =
TransportRequestOptions.options()
.withCompress(shouldCompressRequest)
.withType(TransportRequestOptions.Type.RECOVERY)
.withTimeout(recoverySettings.internalActionTimeout());
while (readCount < len) {
if (shard.state()
== IndexShardState.CLOSED) { // check if the shard got closed on us
throw new IndexShardClosedException(shard.shardId());
}
int toRead =
readCount + BUFFER_SIZE > len ? (int) (len - readCount) : BUFFER_SIZE;
final long position = indexInput.getFilePointer();
// Pause using the rate limiter, if desired, to throttle the recovery
RateLimiter rl = recoverySettings.rateLimiter();
long throttleTimeInNanos = 0;
if (rl != null) {
long bytes = bytesSinceLastPause.addAndGet(toRead);
if (bytes > rl.getMinPauseCheckBytes()) {
// Time to pause
bytesSinceLastPause.addAndGet(-bytes);
throttleTimeInNanos = rl.pause(bytes);
shard.recoveryStats().addThrottleTime(throttleTimeInNanos);
}
}
indexInput.readBytes(buf, 0, toRead, false);
final BytesArray content = new BytesArray(buf, 0, toRead);
readCount += toRead;
final boolean lastChunk = readCount == len;
final RecoveryFileChunkRequest fileChunkRequest =
new RecoveryFileChunkRequest(
request.recoveryId(),
request.shardId(),
md,
position,
content,
lastChunk,
shard.translog().estimatedNumberOfOperations(),
throttleTimeInNanos);
cancellableThreads.execute(
new Interruptable() {
@Override
public void run() throws InterruptedException {
// Actually send the file chunk to the target node, waiting for it to
// complete
transportService
.submitRequest(
request.targetNode(),
RecoveryTarget.Actions.FILE_CHUNK,
fileChunkRequest,
requestOptions,
EmptyTransportResponseHandler.INSTANCE_SAME)
.txGet();
}
});
}
} catch (Throwable e) {
final Throwable corruptIndexException;
if ((corruptIndexException = ExceptionsHelper.unwrapCorruption(e)) != null) {
if (store.checkIntegrity(md)
== false) { // we are corrupted on the primary -- fail!
logger.warn(
"{} Corrupted file detected {} checksum mismatch", shard.shardId(), md);
if (corruptedEngine.compareAndSet(null, corruptIndexException) == false) {
// if we are not the first exception, add ourselves as suppressed to the
// main one:
corruptedEngine.get().addSuppressed(e);
}
} 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);
exceptions.add(0, exception); // last exception first
logger.warn(
"{} Remote file corruption on node {}, recovering {}. local checksum OK",
corruptIndexException,
shard.shardId(),
request.targetNode(),
md);
}
} else {
exceptions.add(0, e); // last exceptions first
}
} finally {
store.decRef();
}
}
});
fileIndex++;
}
cancellableThreads.execute(
new Interruptable() {
@Override
public void run() throws InterruptedException {
// Wait for all files that need to be transferred to finish transferring
latch.await();
}
});
if (corruptedEngine.get() != null) {
throw corruptedEngine.get();
} else {
ExceptionsHelper.rethrowAndSuppress(exceptions);
}
cancellableThreads.execute(
new Interruptable() {
@Override
public void run() throws InterruptedException {
// Send the CLEAN_FILES request, which takes all of the files that
// were transferred and renames them from their temporary file
// names to the actual file names. It also writes checksums for
// the files after they have been renamed.
//
// Once the files have been renamed, any other files that are not
// related to this recovery (out of date segments, for example)
// are deleted
try {
transportService
.submitRequest(
request.targetNode(),
RecoveryTarget.Actions.CLEAN_FILES,
new RecoveryCleanFilesRequest(
request.recoveryId(),
shard.shardId(),
recoverySourceMetadata,
shard.translog().estimatedNumberOfOperations()),
TransportRequestOptions.options()
.withTimeout(recoverySettings.internalActionTimeout()),
EmptyTransportResponseHandler.INSTANCE_SAME)
.txGet();
} catch (RemoteTransportException remoteException) {
final IOException corruptIndexException;
// we realized that after the index was copied and we wanted to finalize the
// recovery
// the index was corrupted:
// - maybe due to a broken segments file on an empty index (transferred with no
// checksum)
// - maybe due to old segments without checksums or length only checks
if ((corruptIndexException = ExceptionsHelper.unwrapCorruption(remoteException))
!= null) {
try {
final Store.MetadataSnapshot recoverySourceMetadata =
store.getMetadata(snapshot);
StoreFileMetaData[] metadata =
Iterables.toArray(recoverySourceMetadata, StoreFileMetaData.class);
ArrayUtil.timSort(
metadata,
new Comparator<StoreFileMetaData>() {
@Override
public int compare(StoreFileMetaData o1, StoreFileMetaData o2) {
return Long.compare(
o1.length(), o2.length()); // check small files first
}
});
for (StoreFileMetaData md : metadata) {
logger.debug(
"{} checking integrity for file {} after remove corruption exception",
shard.shardId(),
md);
if (store.checkIntegrity(md)
== false) { // we are corrupted on the primary -- fail!
logger.warn(
"{} Corrupted file detected {} checksum mismatch",
shard.shardId(),
md);
throw corruptIndexException;
}
}
} catch (IOException ex) {
remoteException.addSuppressed(ex);
throw remoteException;
}
// corruption has happened on the way to replica
RemoteTransportException exception =
new RemoteTransportException(
"File corruption occurred on recovery but checksums are ok", null);
exception.addSuppressed(remoteException);
logger.warn(
"{} Remote file corruption during finalization on node {}, recovering {}. local checksum OK",
corruptIndexException,
shard.shardId(),
request.targetNode());
throw exception;
} else {
throw remoteException;
}
}
}
});
}
stopWatch.stop();
logger.trace(
"[{}][{}] recovery [phase1] to {}: took [{}]",
indexName,
shardId,
request.targetNode(),
stopWatch.totalTime());
response.phase1Time = stopWatch.totalTime().millis();
} catch (Throwable e) {
throw new RecoverFilesRecoveryException(
request.shardId(), response.phase1FileNames.size(), new ByteSizeValue(totalSize), e);
} finally {
store.decRef();
}
}