private synchronized void scheduleSnapshotIfNeeded() {
if (!shardGateway.requiresSnapshot()) {
return;
}
if (!shardGateway.requiresSnapshotScheduling()) {
return;
}
if (!indexShard.routingEntry().primary()) {
// we only do snapshotting on the primary shard
return;
}
if (!indexShard.routingEntry().started()) {
// we only schedule when the cluster assumes we have started
return;
}
if (snapshotScheduleFuture != null) {
// we are already scheduling this one, ignore
return;
}
if (snapshotInterval.millis() != -1) {
// we need to schedule snapshot
if (logger.isDebugEnabled()) {
logger.debug("scheduling snapshot every [{}]", snapshotInterval);
}
snapshotScheduleFuture =
threadPool.schedule(snapshotInterval, ThreadPool.Names.SNAPSHOT, snapshotRunnable);
}
}
private Directory getStoreDirectory(String index, int shardId) {
Set<String> nodes = cluster().nodesInclude("test");
assertThat(nodes.isEmpty(), equalTo(false));
IndicesService indicesService =
cluster().getInstance(IndicesService.class, nodes.iterator().next());
InternalIndexShard indexShard =
(InternalIndexShard) (indicesService.indexService(index).shard(shardId));
return indexShard.store().directory();
}
@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)));
}
@Override
protected ShardTermlistResponse shardOperation(ShardTermlistRequest request)
throws ElasticSearchException {
synchronized (termlistMutex) {
InternalIndexShard indexShard =
(InternalIndexShard)
indicesService.indexServiceSafe(request.index()).shardSafe(request.shardId());
indexShard.store().directory();
Engine.Searcher searcher = indexShard.searcher();
try {
Set<String> set = new CompactHashSet();
Fields fields = MultiFields.getFields(searcher.reader());
if (fields != null) {
for (Iterator<String> it = fields.iterator(); it.hasNext(); ) {
String field = it.next();
if (field.charAt(0) == '_') {
continue;
}
if (request.getField() == null || field.equals(request.getField())) {
Terms terms = fields.terms(field);
if (terms != null) {
TermsEnum termsEnum = terms.iterator(null);
BytesRef text;
while ((text = termsEnum.next()) != null) {
set.add(text.utf8ToString());
System.out.println("field=" + field + "; text=" + text.utf8ToString());
}
}
}
}
}
return new ShardTermlistResponse(request.index(), request.shardId(), set);
} catch (IOException ex) {
throw new ElasticSearchException(ex.getMessage(), ex);
}
}
}
@Override
protected ShardStatus shardOperation(IndexShardStatusRequest request)
throws ElasticSearchException {
InternalIndexService indexService =
(InternalIndexService) indicesService.indexServiceSafe(request.index());
InternalIndexShard indexShard = (InternalIndexShard) indexService.shardSafe(request.shardId());
ShardStatus shardStatus = new ShardStatus(indexShard.routingEntry());
shardStatus.state = indexShard.state();
try {
shardStatus.storeSize = indexShard.store().estimateSize();
} catch (IOException e) {
// failure to get the store size...
}
if (indexShard.state() == IndexShardState.STARTED) {
// shardStatus.estimatedFlushableMemorySize =
// indexShard.estimateFlushableMemorySize();
shardStatus.translogId = indexShard.translog().currentId();
shardStatus.translogOperations = indexShard.translog().estimatedNumberOfOperations();
Engine.Searcher searcher = indexShard.searcher();
try {
shardStatus.docs = new DocsStatus();
shardStatus.docs.numDocs = searcher.reader().numDocs();
shardStatus.docs.maxDoc = searcher.reader().maxDoc();
shardStatus.docs.deletedDocs = searcher.reader().numDeletedDocs();
} finally {
searcher.release();
}
shardStatus.mergeStats = indexShard.mergeScheduler().stats();
shardStatus.refreshStats = indexShard.refreshStats();
shardStatus.flushStats = indexShard.flushStats();
}
if (request.recovery) {
// check on going recovery (from peer or gateway)
RecoveryStatus peerRecoveryStatus = indexShard.peerRecoveryStatus();
if (peerRecoveryStatus == null) {
peerRecoveryStatus = peerRecoveryTarget.peerRecoveryStatus(indexShard.shardId());
}
if (peerRecoveryStatus != null) {
PeerRecoveryStatus.Stage stage;
switch (peerRecoveryStatus.stage()) {
case INIT:
stage = PeerRecoveryStatus.Stage.INIT;
break;
case INDEX:
stage = PeerRecoveryStatus.Stage.INDEX;
break;
case TRANSLOG:
stage = PeerRecoveryStatus.Stage.TRANSLOG;
break;
case FINALIZE:
stage = PeerRecoveryStatus.Stage.FINALIZE;
break;
case DONE:
stage = PeerRecoveryStatus.Stage.DONE;
break;
default:
stage = PeerRecoveryStatus.Stage.INIT;
}
shardStatus.peerRecoveryStatus =
new PeerRecoveryStatus(
stage,
peerRecoveryStatus.startTime(),
peerRecoveryStatus.time(),
peerRecoveryStatus.phase1TotalSize(),
peerRecoveryStatus.phase1ExistingTotalSize(),
peerRecoveryStatus.currentFilesSize(),
peerRecoveryStatus.currentTranslogOperations());
}
IndexShardGatewayService gatewayService =
indexService.shardInjector(request.shardId()).getInstance(IndexShardGatewayService.class);
org.elasticsearch.index.gateway.RecoveryStatus gatewayRecoveryStatus =
gatewayService.recoveryStatus();
if (gatewayRecoveryStatus != null) {
GatewayRecoveryStatus.Stage stage;
switch (gatewayRecoveryStatus.stage()) {
case INIT:
stage = GatewayRecoveryStatus.Stage.INIT;
break;
case INDEX:
stage = GatewayRecoveryStatus.Stage.INDEX;
break;
case TRANSLOG:
stage = GatewayRecoveryStatus.Stage.TRANSLOG;
break;
case DONE:
stage = GatewayRecoveryStatus.Stage.DONE;
break;
default:
stage = GatewayRecoveryStatus.Stage.INIT;
}
shardStatus.gatewayRecoveryStatus =
new GatewayRecoveryStatus(
stage,
gatewayRecoveryStatus.startTime(),
gatewayRecoveryStatus.time(),
gatewayRecoveryStatus.index().totalSize(),
gatewayRecoveryStatus.index().reusedTotalSize(),
gatewayRecoveryStatus.index().currentFilesSize(),
gatewayRecoveryStatus.translog().currentTranslogOperations());
}
}
if (request.snapshot) {
IndexShardGatewayService gatewayService =
indexService.shardInjector(request.shardId()).getInstance(IndexShardGatewayService.class);
SnapshotStatus snapshotStatus = gatewayService.snapshotStatus();
if (snapshotStatus != null) {
GatewaySnapshotStatus.Stage stage;
switch (snapshotStatus.stage()) {
case DONE:
stage = GatewaySnapshotStatus.Stage.DONE;
break;
case FAILURE:
stage = GatewaySnapshotStatus.Stage.FAILURE;
break;
case TRANSLOG:
stage = GatewaySnapshotStatus.Stage.TRANSLOG;
break;
case FINALIZE:
stage = GatewaySnapshotStatus.Stage.FINALIZE;
break;
case INDEX:
stage = GatewaySnapshotStatus.Stage.INDEX;
break;
default:
stage = GatewaySnapshotStatus.Stage.NONE;
break;
}
shardStatus.gatewaySnapshotStatus =
new GatewaySnapshotStatus(
stage,
snapshotStatus.startTime(),
snapshotStatus.time(),
snapshotStatus.index().totalSize(),
snapshotStatus.translog().expectedNumberOfOperations());
}
}
return shardStatus;
}
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);
}
}
@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);
}
/** Snapshots the given shard into the gateway. */
public synchronized void snapshot(final String reason)
throws IndexShardGatewaySnapshotFailedException {
if (!indexShard.routingEntry().primary()) {
return;
// throw new IndexShardGatewaySnapshotNotAllowedException(shardId, "Snapshot not
// allowed on non primary shard");
}
if (indexShard.routingEntry().relocating()) {
// do not snapshot when in the process of relocation of primaries so we won't get conflicts
return;
}
if (indexShard.state() == IndexShardState.CREATED) {
// shard has just been created, ignore it and return
return;
}
if (indexShard.state() == IndexShardState.RECOVERING) {
// shard is recovering, don't snapshot
return;
}
if (snapshotLock == null) {
try {
snapshotLock = shardGateway.obtainSnapshotLock();
} catch (Exception e) {
logger.warn("failed to obtain snapshot lock, ignoring snapshot", e);
return;
}
}
try {
SnapshotStatus snapshotStatus =
indexShard.snapshot(
new Engine.SnapshotHandler<SnapshotStatus>() {
@Override
public SnapshotStatus snapshot(
SnapshotIndexCommit snapshotIndexCommit, Translog.Snapshot translogSnapshot)
throws EngineException {
if (lastIndexVersion != snapshotIndexCommit.getGeneration()
|| lastTranslogId != translogSnapshot.translogId()
|| lastTranslogLength < translogSnapshot.length()) {
logger.debug("snapshot ({}) to {} ...", reason, shardGateway);
SnapshotStatus snapshotStatus =
shardGateway.snapshot(
new IndexShardGateway.Snapshot(
snapshotIndexCommit,
translogSnapshot,
lastIndexVersion,
lastTranslogId,
lastTranslogLength,
lastTotalTranslogOperations));
lastIndexVersion = snapshotIndexCommit.getGeneration();
lastTranslogId = translogSnapshot.translogId();
lastTranslogLength = translogSnapshot.length();
lastTotalTranslogOperations = translogSnapshot.estimatedTotalOperations();
return snapshotStatus;
}
return null;
}
});
if (snapshotStatus != null) {
if (logger.isDebugEnabled()) {
StringBuilder sb = new StringBuilder();
sb.append("snapshot (")
.append(reason)
.append(") completed to ")
.append(shardGateway)
.append(", took [")
.append(TimeValue.timeValueMillis(snapshotStatus.time()))
.append("]\n");
sb.append(" index : version [")
.append(lastIndexVersion)
.append("], number_of_files [")
.append(snapshotStatus.index().numberOfFiles())
.append("] with total_size [")
.append(new ByteSizeValue(snapshotStatus.index().totalSize()))
.append("], took [")
.append(TimeValue.timeValueMillis(snapshotStatus.index().time()))
.append("]\n");
sb.append(" translog : id [")
.append(lastTranslogId)
.append("], number_of_operations [")
.append(snapshotStatus.translog().expectedNumberOfOperations())
.append("], took [")
.append(TimeValue.timeValueMillis(snapshotStatus.translog().time()))
.append("]");
logger.debug(sb.toString());
}
}
} catch (SnapshotFailedEngineException e) {
if (e.getCause() instanceof IllegalStateException) {
// ignore, that's fine, snapshot has not started yet
} else {
throw new IndexShardGatewaySnapshotFailedException(shardId, "Failed to snapshot", e);
}
} catch (IllegalIndexShardStateException e) {
// ignore, that's fine, snapshot has not started yet
} catch (IndexShardGatewaySnapshotFailedException e) {
throw e;
} catch (Exception e) {
throw new IndexShardGatewaySnapshotFailedException(shardId, "Failed to snapshot", e);
}
}
/** Recovers the state of the shard from the gateway. */
public void recover(final boolean indexShouldExists, final RecoveryListener listener)
throws IndexShardGatewayRecoveryException, IgnoreGatewayRecoveryException {
if (indexShard.state() == IndexShardState.CLOSED) {
// got closed on us, just ignore this recovery
listener.onIgnoreRecovery("shard closed");
return;
}
if (!indexShard.routingEntry().primary()) {
listener.onRecoveryFailed(
new IndexShardGatewayRecoveryException(
shardId, "Trying to recover when the shard is in backup state", null));
return;
}
try {
if (indexShard.routingEntry().restoreSource() != null) {
indexShard.recovering("from snapshot");
} else {
indexShard.recovering("from gateway");
}
} catch (IllegalIndexShardStateException e) {
// that's fine, since we might be called concurrently, just ignore this, we are already
// recovering
listener.onIgnoreRecovery("already in recovering process, " + e.getMessage());
return;
}
threadPool
.generic()
.execute(
new Runnable() {
@Override
public void run() {
recoveryStatus = new RecoveryStatus();
recoveryStatus.updateStage(RecoveryStatus.Stage.INIT);
try {
if (indexShard.routingEntry().restoreSource() != null) {
logger.debug(
"restoring from {} ...", indexShard.routingEntry().restoreSource());
snapshotService.restore(recoveryStatus);
} else {
logger.debug("starting recovery from {} ...", shardGateway);
shardGateway.recover(indexShouldExists, recoveryStatus);
}
lastIndexVersion = recoveryStatus.index().version();
lastTranslogId = -1;
lastTranslogLength = 0;
lastTotalTranslogOperations =
recoveryStatus.translog().currentTranslogOperations();
// start the shard if the gateway has not started it already. Note that if the
// gateway
// moved shard to POST_RECOVERY, it may have been started as well if:
// 1) master sent a new cluster state indicating shard is initializing
// 2) IndicesClusterStateService#applyInitializingShard will send a shard started
// event
// 3) Master will mark shard as started and this will be processed locally.
IndexShardState shardState = indexShard.state();
if (shardState != IndexShardState.POST_RECOVERY
&& shardState != IndexShardState.STARTED) {
indexShard.postRecovery("post recovery from gateway");
}
// refresh the shard
indexShard.refresh(new Engine.Refresh("post_gateway").force(true));
recoveryStatus.time(System.currentTimeMillis() - recoveryStatus.startTime());
recoveryStatus.updateStage(RecoveryStatus.Stage.DONE);
if (logger.isDebugEnabled()) {
logger.debug(
"recovery completed from [{}], took [{}]",
shardGateway,
timeValueMillis(recoveryStatus.time()));
} else if (logger.isTraceEnabled()) {
StringBuilder sb = new StringBuilder();
sb.append("recovery completed from ")
.append(shardGateway)
.append(", took [")
.append(timeValueMillis(recoveryStatus.time()))
.append("]\n");
sb.append(" index : files [")
.append(recoveryStatus.index().numberOfFiles())
.append("] with total_size [")
.append(new ByteSizeValue(recoveryStatus.index().totalSize()))
.append("], took[")
.append(TimeValue.timeValueMillis(recoveryStatus.index().time()))
.append("]\n");
sb.append(" : recovered_files [")
.append(recoveryStatus.index().numberOfRecoveredFiles())
.append("] with total_size [")
.append(new ByteSizeValue(recoveryStatus.index().recoveredTotalSize()))
.append("]\n");
sb.append(" : reusing_files [")
.append(recoveryStatus.index().numberOfReusedFiles())
.append("] with total_size [")
.append(new ByteSizeValue(recoveryStatus.index().reusedTotalSize()))
.append("]\n");
sb.append(" start : took [")
.append(TimeValue.timeValueMillis(recoveryStatus.start().time()))
.append("], check_index [")
.append(timeValueMillis(recoveryStatus.start().checkIndexTime()))
.append("]\n");
sb.append(" translog : number_of_operations [")
.append(recoveryStatus.translog().currentTranslogOperations())
.append("], took [")
.append(TimeValue.timeValueMillis(recoveryStatus.translog().time()))
.append("]");
logger.trace(sb.toString());
}
listener.onRecoveryDone();
scheduleSnapshotIfNeeded();
} catch (IndexShardGatewayRecoveryException e) {
if (indexShard.state() == IndexShardState.CLOSED) {
// got closed on us, just ignore this recovery
listener.onIgnoreRecovery("shard closed");
return;
}
if ((e.getCause() instanceof IndexShardClosedException)
|| (e.getCause() instanceof IndexShardNotStartedException)) {
// got closed on us, just ignore this recovery
listener.onIgnoreRecovery("shard closed");
return;
}
listener.onRecoveryFailed(e);
} catch (IndexShardClosedException e) {
listener.onIgnoreRecovery("shard closed");
} catch (IndexShardNotStartedException e) {
listener.onIgnoreRecovery("shard closed");
} catch (Exception e) {
if (indexShard.state() == IndexShardState.CLOSED) {
// got closed on us, just ignore this recovery
listener.onIgnoreRecovery("shard closed");
return;
}
listener.onRecoveryFailed(
new IndexShardGatewayRecoveryException(shardId, "failed recovery", e));
}
}
});
}
