@Override
public RecoveryStatus recover() throws IndexShardGatewayRecoveryException {
// in the none case, we simply start the shard
indexShard.start();
return new RecoveryStatus(
new RecoveryStatus.Index(-1, 0, new SizeValue(0, SizeUnit.BYTES)),
new RecoveryStatus.Translog(-1, 0, new SizeValue(0, SizeUnit.BYTES)));
}
private void recoverTranslog(CommitPoint commitPoint, ImmutableMap<String, BlobMetaData> blobs)
throws IndexShardGatewayRecoveryException {
if (commitPoint.translogFiles().isEmpty()) {
// no translog files, bail
indexShard.start("post recovery from gateway, no translog");
return;
}
try {
indexShard.performRecoveryPrepareForTranslog();
final AtomicReference<Throwable> failure = new AtomicReference<Throwable>();
final CountDownLatch latch = new CountDownLatch(1);
final Iterator<CommitPoint.FileInfo> transIt = commitPoint.translogFiles().iterator();
blobContainer.readBlob(
transIt.next().name(),
new BlobContainer.ReadBlobListener() {
FastByteArrayOutputStream bos = new FastByteArrayOutputStream();
boolean ignore = false;
@Override
public synchronized void onPartial(byte[] data, int offset, int size)
throws IOException {
if (ignore) {
return;
}
bos.write(data, offset, size);
// if we don't have enough to read the header size of the first translog, bail and
// wait for the next one
if (bos.size() < 4) {
return;
}
BytesStreamInput si = new BytesStreamInput(bos.unsafeByteArray(), 0, bos.size());
int position;
while (true) {
try {
position = si.position();
if (position + 4 > bos.size()) {
break;
}
int opSize = si.readInt();
int curPos = si.position();
if ((si.position() + opSize) > bos.size()) {
break;
}
Translog.Operation operation = TranslogStreams.readTranslogOperation(si);
if ((si.position() - curPos) != opSize) {
logger.warn(
"mismatch in size, expected [{}], got [{}]",
opSize,
si.position() - curPos);
}
recoveryStatus.translog().addTranslogOperations(1);
indexShard.performRecoveryOperation(operation);
if (si.position() >= bos.size()) {
position = si.position();
break;
}
} catch (Exception e) {
logger.warn(
"failed to retrieve translog after [{}] operations, ignoring the rest, considered corrupted",
e,
recoveryStatus.translog().currentTranslogOperations());
ignore = true;
latch.countDown();
return;
}
}
FastByteArrayOutputStream newBos = new FastByteArrayOutputStream();
int leftOver = bos.size() - position;
if (leftOver > 0) {
newBos.write(bos.unsafeByteArray(), position, leftOver);
}
bos = newBos;
}
@Override
public synchronized void onCompleted() {
if (ignore) {
return;
}
if (!transIt.hasNext()) {
latch.countDown();
return;
}
blobContainer.readBlob(transIt.next().name(), this);
}
@Override
public void onFailure(Throwable t) {
failure.set(t);
latch.countDown();
}
});
latch.await();
if (failure.get() != null) {
throw failure.get();
}
indexShard.performRecoveryFinalization(true);
} catch (Throwable e) {
throw new IndexShardGatewayRecoveryException(shardId, "Failed to recover translog", e);
}
}
