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
}
}
@Override
public void recover(RecoveryStatus recoveryStatus) throws IndexShardGatewayRecoveryException {
this.recoveryStatus = recoveryStatus;
final ImmutableMap<String, BlobMetaData> blobs;
try {
blobs = blobContainer.listBlobs();
} catch (IOException e) {
throw new IndexShardGatewayRecoveryException(shardId, "Failed to list content of gateway", e);
}
List<CommitPoint> commitPointsList = Lists.newArrayList();
boolean atLeastOneCommitPointExists = false;
for (String name : blobs.keySet()) {
if (name.startsWith("commit-")) {
atLeastOneCommitPointExists = true;
try {
commitPointsList.add(CommitPoints.fromXContent(blobContainer.readBlobFully(name)));
} catch (Exception e) {
logger.warn("failed to read commit point [{}]", e, name);
}
}
}
if (atLeastOneCommitPointExists && commitPointsList.isEmpty()) {
// no commit point managed to load, bail so we won't corrupt the index, will require manual
// intervention
throw new IndexShardGatewayRecoveryException(
shardId, "Commit points exists but none could be loaded", null);
}
CommitPoints commitPoints = new CommitPoints(commitPointsList);
if (commitPoints.commits().isEmpty()) {
// no commit points, clean the store just so we won't recover wrong files
try {
indexShard.store().deleteContent();
} catch (IOException e) {
logger.warn("failed to clean store before starting shard", e);
}
recoveryStatus.index().startTime(System.currentTimeMillis());
recoveryStatus.index().time(System.currentTimeMillis() - recoveryStatus.index().startTime());
recoveryStatus.translog().startTime(System.currentTimeMillis());
recoveryStatus
.translog()
.time(System.currentTimeMillis() - recoveryStatus.index().startTime());
return;
}
for (CommitPoint commitPoint : commitPoints) {
if (!commitPointExistsInBlobs(commitPoint, blobs)) {
logger.warn(
"listed commit_point [{}]/[{}], but not all files exists, ignoring",
commitPoint.name(),
commitPoint.version());
continue;
}
try {
recoveryStatus.index().startTime(System.currentTimeMillis());
recoveryStatus.updateStage(RecoveryStatus.Stage.INDEX);
recoverIndex(commitPoint, blobs);
recoveryStatus
.index()
.time(System.currentTimeMillis() - recoveryStatus.index().startTime());
recoveryStatus.translog().startTime(System.currentTimeMillis());
recoveryStatus.updateStage(RecoveryStatus.Stage.TRANSLOG);
recoverTranslog(commitPoint, blobs);
recoveryStatus
.translog()
.time(System.currentTimeMillis() - recoveryStatus.index().startTime());
return;
} catch (Exception e) {
throw new IndexShardGatewayRecoveryException(
shardId,
"failed to recover commit_point ["
+ commitPoint.name()
+ "]/["
+ commitPoint.version()
+ "]",
e);
}
}
throw new IndexShardGatewayRecoveryException(
shardId, "No commit point data is available in gateway", null);
}
