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
}
}
private void doSnapshot(final Snapshot snapshot) throws IndexShardGatewaySnapshotFailedException {
ImmutableMap<String, BlobMetaData> blobs;
try {
blobs = blobContainer.listBlobs();
} catch (IOException e) {
throw new IndexShardGatewaySnapshotFailedException(shardId, "failed to list blobs", e);
}
long generation = findLatestFileNameGeneration(blobs);
CommitPoints commitPoints = buildCommitPoints(blobs);
currentSnapshotStatus.index().startTime(System.currentTimeMillis());
currentSnapshotStatus.updateStage(SnapshotStatus.Stage.INDEX);
final SnapshotIndexCommit snapshotIndexCommit = snapshot.indexCommit();
final Translog.Snapshot translogSnapshot = snapshot.translogSnapshot();
final CountDownLatch indexLatch = new CountDownLatch(snapshotIndexCommit.getFiles().length);
final CopyOnWriteArrayList<Throwable> failures = new CopyOnWriteArrayList<Throwable>();
final List<CommitPoint.FileInfo> indexCommitPointFiles = Lists.newArrayList();
int indexNumberOfFiles = 0;
long indexTotalFilesSize = 0;
for (final String fileName : snapshotIndexCommit.getFiles()) {
StoreFileMetaData md;
try {
md = store.metaData(fileName);
} catch (IOException e) {
throw new IndexShardGatewaySnapshotFailedException(
shardId, "Failed to get store file metadata", e);
}
boolean snapshotRequired = false;
if (snapshot.indexChanged() && fileName.equals(snapshotIndexCommit.getSegmentsFileName())) {
snapshotRequired = true; // we want to always snapshot the segment file if the index changed
}
CommitPoint.FileInfo fileInfo = commitPoints.findPhysicalIndexFile(fileName);
if (fileInfo == null
|| !fileInfo.isSame(md)
|| !commitPointFileExistsInBlobs(fileInfo, blobs)) {
// commit point file does not exists in any commit point, or has different length, or does
// not fully exists in the listed blobs
snapshotRequired = true;
}
if (snapshotRequired) {
indexNumberOfFiles++;
indexTotalFilesSize += md.length();
// create a new FileInfo
try {
CommitPoint.FileInfo snapshotFileInfo =
new CommitPoint.FileInfo(
fileNameFromGeneration(++generation), fileName, md.length(), md.checksum());
indexCommitPointFiles.add(snapshotFileInfo);
snapshotFile(snapshotIndexCommit.getDirectory(), snapshotFileInfo, indexLatch, failures);
} catch (IOException e) {
failures.add(e);
indexLatch.countDown();
}
} else {
indexCommitPointFiles.add(fileInfo);
indexLatch.countDown();
}
}
currentSnapshotStatus.index().files(indexNumberOfFiles, indexTotalFilesSize);
try {
indexLatch.await();
} catch (InterruptedException e) {
failures.add(e);
}
if (!failures.isEmpty()) {
throw new IndexShardGatewaySnapshotFailedException(
shardId(), "Failed to perform snapshot (index files)", failures.get(failures.size() - 1));
}
currentSnapshotStatus
.index()
.time(System.currentTimeMillis() - currentSnapshotStatus.index().startTime());
currentSnapshotStatus.updateStage(SnapshotStatus.Stage.TRANSLOG);
currentSnapshotStatus.translog().startTime(System.currentTimeMillis());
// Note, we assume the snapshot is always started from "base 0". We need to seek forward if we
// want to lastTranslogPosition if we want the delta
List<CommitPoint.FileInfo> translogCommitPointFiles = Lists.newArrayList();
int expectedNumberOfOperations = 0;
boolean snapshotRequired = false;
if (snapshot.newTranslogCreated()) {
if (translogSnapshot.lengthInBytes() > 0) {
snapshotRequired = true;
expectedNumberOfOperations = translogSnapshot.estimatedTotalOperations();
}
} else {
// if we have a commit point, check that we have all the files listed in it in the blob store
if (!commitPoints.commits().isEmpty()) {
CommitPoint commitPoint = commitPoints.commits().get(0);
boolean allTranslogFilesExists = true;
for (CommitPoint.FileInfo fileInfo : commitPoint.translogFiles()) {
if (!commitPointFileExistsInBlobs(fileInfo, blobs)) {
allTranslogFilesExists = false;
break;
}
}
// if everything exists, we can seek forward in case there are new operations, otherwise, we
// copy over all again...
if (allTranslogFilesExists) {
translogCommitPointFiles.addAll(commitPoint.translogFiles());
if (snapshot.sameTranslogNewOperations()) {
translogSnapshot.seekForward(snapshot.lastTranslogLength());
if (translogSnapshot.lengthInBytes() > 0) {
snapshotRequired = true;
expectedNumberOfOperations =
translogSnapshot.estimatedTotalOperations()
- snapshot.lastTotalTranslogOperations();
}
} // else (no operations, nothing to snapshot)
} else {
// a full translog snapshot is required
if (translogSnapshot.lengthInBytes() > 0) {
expectedNumberOfOperations = translogSnapshot.estimatedTotalOperations();
snapshotRequired = true;
}
}
} else {
// no commit point, snapshot all the translog
if (translogSnapshot.lengthInBytes() > 0) {
expectedNumberOfOperations = translogSnapshot.estimatedTotalOperations();
snapshotRequired = true;
}
}
}
currentSnapshotStatus.translog().expectedNumberOfOperations(expectedNumberOfOperations);
if (snapshotRequired) {
CommitPoint.FileInfo addedTranslogFileInfo =
new CommitPoint.FileInfo(
fileNameFromGeneration(++generation),
"translog-" + translogSnapshot.translogId(),
translogSnapshot.lengthInBytes(),
null /* no need for checksum in translog */);
translogCommitPointFiles.add(addedTranslogFileInfo);
try {
snapshotTranslog(translogSnapshot, addedTranslogFileInfo);
} catch (Exception e) {
throw new IndexShardGatewaySnapshotFailedException(
shardId, "Failed to snapshot translog", e);
}
}
currentSnapshotStatus
.translog()
.time(System.currentTimeMillis() - currentSnapshotStatus.translog().startTime());
// now create and write the commit point
currentSnapshotStatus.updateStage(SnapshotStatus.Stage.FINALIZE);
long version = 0;
if (!commitPoints.commits().isEmpty()) {
version = commitPoints.commits().iterator().next().version() + 1;
}
String commitPointName = "commit-" + Long.toString(version, Character.MAX_RADIX);
CommitPoint commitPoint =
new CommitPoint(
version,
commitPointName,
CommitPoint.Type.GENERATED,
indexCommitPointFiles,
translogCommitPointFiles);
try {
byte[] commitPointData = CommitPoints.toXContent(commitPoint);
blobContainer.writeBlob(
commitPointName, new FastByteArrayInputStream(commitPointData), commitPointData.length);
} catch (Exception e) {
throw new IndexShardGatewaySnapshotFailedException(
shardId, "Failed to write commit point", e);
}
// delete all files that are not referenced by any commit point
// build a new CommitPoint, that includes this one and all the saved ones
List<CommitPoint> newCommitPointsList = Lists.newArrayList();
newCommitPointsList.add(commitPoint);
for (CommitPoint point : commitPoints) {
if (point.type() == CommitPoint.Type.SAVED) {
newCommitPointsList.add(point);
}
}
CommitPoints newCommitPoints = new CommitPoints(newCommitPointsList);
// first, go over and delete all the commit points
for (String blobName : blobs.keySet()) {
if (!blobName.startsWith("commit-")) {
continue;
}
long checkedVersion =
Long.parseLong(blobName.substring("commit-".length()), Character.MAX_RADIX);
if (!newCommitPoints.hasVersion(checkedVersion)) {
try {
blobContainer.deleteBlob(blobName);
} catch (IOException e) {
// ignore
}
}
}
// now go over all the blobs, and if they don't exists in a commit point, delete them
for (String blobName : blobs.keySet()) {
String name = blobName;
if (!name.startsWith("__")) {
continue;
}
if (blobName.contains(".part")) {
name = blobName.substring(0, blobName.indexOf(".part"));
}
if (newCommitPoints.findNameFile(name) == null) {
try {
blobContainer.deleteBlob(blobName);
} catch (IOException e) {
// ignore, will delete it laters
}
}
}
}