use of org.elasticsearch.env.ShardLock in project elasticsearch by elastic.
the class IndicesService method processPendingDeletes.
/**
* Processes all pending deletes for the given index. This method will acquire all locks for the given index and will
* process all pending deletes for this index. Pending deletes might occur if the OS doesn't allow deletion of files because
* they are used by a different process ie. on Windows where files might still be open by a virus scanner. On a shared
* filesystem a replica might not have been closed when the primary is deleted causing problems on delete calls so we
* schedule there deletes later.
* @param index the index to process the pending deletes for
* @param timeout the timeout used for processing pending deletes
*/
@Override
public void processPendingDeletes(Index index, IndexSettings indexSettings, TimeValue timeout) throws IOException, InterruptedException, ShardLockObtainFailedException {
logger.debug("{} processing pending deletes", index);
final long startTimeNS = System.nanoTime();
final List<ShardLock> shardLocks = nodeEnv.lockAllForIndex(index, indexSettings, timeout.millis());
int numRemoved = 0;
try {
Map<ShardId, ShardLock> locks = new HashMap<>();
for (ShardLock lock : shardLocks) {
locks.put(lock.getShardId(), lock);
}
final List<PendingDelete> remove;
synchronized (pendingDeletes) {
remove = pendingDeletes.remove(index);
}
if (remove != null && remove.isEmpty() == false) {
numRemoved = remove.size();
// make sure we delete indices first
CollectionUtil.timSort(remove);
// ensure we retry after 10 sec
final long maxSleepTimeMs = 10 * 1000;
long sleepTime = 10;
do {
if (remove.isEmpty()) {
break;
}
Iterator<PendingDelete> iterator = remove.iterator();
while (iterator.hasNext()) {
PendingDelete delete = iterator.next();
if (delete.deleteIndex) {
assert delete.shardId == -1;
logger.debug("{} deleting index store reason [{}]", index, "pending delete");
try {
nodeEnv.deleteIndexDirectoryUnderLock(index, indexSettings);
iterator.remove();
} catch (IOException ex) {
logger.debug((Supplier<?>) () -> new ParameterizedMessage("{} retry pending delete", index), ex);
}
} else {
assert delete.shardId != -1;
ShardLock shardLock = locks.get(new ShardId(delete.index, delete.shardId));
if (shardLock != null) {
try {
deleteShardStore("pending delete", shardLock, delete.settings);
iterator.remove();
} catch (IOException ex) {
logger.debug((Supplier<?>) () -> new ParameterizedMessage("{} retry pending delete", shardLock.getShardId()), ex);
}
} else {
logger.warn("{} no shard lock for pending delete", delete.shardId);
iterator.remove();
}
}
}
if (remove.isEmpty() == false) {
logger.warn("{} still pending deletes present for shards {} - retrying", index, remove.toString());
Thread.sleep(sleepTime);
// increase the sleep time gradually
sleepTime = Math.min(maxSleepTimeMs, sleepTime * 2);
logger.debug("{} schedule pending delete retry after {} ms", index, sleepTime);
}
} while ((System.nanoTime() - startTimeNS) < timeout.nanos());
}
} finally {
IOUtils.close(shardLocks);
if (numRemoved > 0) {
int remainingUncompletedDeletes = numUncompletedDeletes.addAndGet(-numRemoved);
assert remainingUncompletedDeletes >= 0;
}
}
}
use of org.elasticsearch.env.ShardLock in project elasticsearch by elastic.
the class Store method tryOpenIndex.
/**
* Tries to open an index for the given location. This includes reading the
* segment infos and possible corruption markers. If the index can not
* be opened, an exception is thrown
*/
public static void tryOpenIndex(Path indexLocation, ShardId shardId, NodeEnvironment.ShardLocker shardLocker, Logger logger) throws IOException, ShardLockObtainFailedException {
try (ShardLock lock = shardLocker.lock(shardId, TimeUnit.SECONDS.toMillis(5));
Directory dir = new SimpleFSDirectory(indexLocation)) {
failIfCorrupted(dir, shardId);
SegmentInfos segInfo = Lucene.readSegmentInfos(dir);
logger.trace("{} loaded segment info [{}]", shardId, segInfo);
}
}
use of org.elasticsearch.env.ShardLock in project elasticsearch by elastic.
the class StoreTests method testOnCloseCallback.
public void testOnCloseCallback() throws IOException {
final ShardId shardId = new ShardId(new Index(randomRealisticUnicodeOfCodepointLengthBetween(1, 10), "_na_"), randomIntBetween(0, 100));
DirectoryService directoryService = new LuceneManagedDirectoryService(random());
final AtomicInteger count = new AtomicInteger(0);
final ShardLock lock = new DummyShardLock(shardId);
Store store = new Store(shardId, INDEX_SETTINGS, directoryService, lock, theLock -> {
assertEquals(shardId, theLock.getShardId());
assertEquals(lock, theLock);
count.incrementAndGet();
});
assertEquals(count.get(), 0);
final int iters = randomIntBetween(1, 10);
for (int i = 0; i < iters; i++) {
store.close();
}
assertEquals(count.get(), 1);
}
use of org.elasticsearch.env.ShardLock in project elasticsearch by elastic.
the class IndexShardIT method testLockTryingToDelete.
public void testLockTryingToDelete() throws Exception {
createIndex("test");
ensureGreen();
NodeEnvironment env = getInstanceFromNode(NodeEnvironment.class);
ClusterService cs = getInstanceFromNode(ClusterService.class);
final Index index = cs.state().metaData().index("test").getIndex();
Path[] shardPaths = env.availableShardPaths(new ShardId(index, 0));
logger.info("--> paths: [{}]", (Object) shardPaths);
// Should not be able to acquire the lock because it's already open
try {
NodeEnvironment.acquireFSLockForPaths(IndexSettingsModule.newIndexSettings("test", Settings.EMPTY), shardPaths);
fail("should not have been able to acquire the lock");
} catch (LockObtainFailedException e) {
assertTrue("msg: " + e.getMessage(), e.getMessage().contains("unable to acquire write.lock"));
}
// Test without the regular shard lock to assume we can acquire it
// (worst case, meaning that the shard lock could be acquired and
// we're green to delete the shard's directory)
ShardLock sLock = new DummyShardLock(new ShardId(index, 0));
try {
env.deleteShardDirectoryUnderLock(sLock, IndexSettingsModule.newIndexSettings("test", Settings.EMPTY));
fail("should not have been able to delete the directory");
} catch (LockObtainFailedException e) {
assertTrue("msg: " + e.getMessage(), e.getMessage().contains("unable to acquire write.lock"));
}
}
use of org.elasticsearch.env.ShardLock in project crate by crate.
the class IndexService method createShard.
public synchronized IndexShard createShard(final ShardRouting routing, final Consumer<ShardId> globalCheckpointSyncer, final RetentionLeaseSyncer retentionLeaseSyncer) throws IOException {
Objects.requireNonNull(retentionLeaseSyncer);
/*
* TODO: we execute this in parallel but it's a synced method. Yet, we might
* be able to serialize the execution via the cluster state in the future. for now we just
* keep it synced.
*/
if (closed.get()) {
throw new IllegalStateException("Can't create shard " + routing.shardId() + ", closed");
}
final Settings indexSettings = this.indexSettings.getSettings();
final ShardId shardId = routing.shardId();
boolean success = false;
Store store = null;
IndexShard indexShard = null;
ShardLock lock = null;
try {
lock = nodeEnv.shardLock(shardId, "starting shard", TimeUnit.SECONDS.toMillis(5));
eventListener.beforeIndexShardCreated(shardId, indexSettings);
ShardPath path;
try {
path = ShardPath.loadShardPath(logger, nodeEnv, shardId, this.indexSettings.customDataPath());
} catch (IllegalStateException ex) {
logger.warn("{} failed to load shard path, trying to remove leftover", shardId);
try {
ShardPath.deleteLeftoverShardDirectory(logger, nodeEnv, lock, this.indexSettings);
path = ShardPath.loadShardPath(logger, nodeEnv, shardId, this.indexSettings.customDataPath());
} catch (Exception inner) {
ex.addSuppressed(inner);
throw ex;
}
}
if (path == null) {
// TODO: we should, instead, hold a "bytes reserved" of how large we anticipate this shard will be, e.g. for a shard
// that's being relocated/replicated we know how large it will become once it's done copying:
// Count up how many shards are currently on each data path:
Map<Path, Integer> dataPathToShardCount = new HashMap<>();
for (IndexShard shard : this) {
Path dataPath = shard.shardPath().getRootStatePath();
Integer curCount = dataPathToShardCount.get(dataPath);
if (curCount == null) {
curCount = 0;
}
dataPathToShardCount.put(dataPath, curCount + 1);
}
path = ShardPath.selectNewPathForShard(nodeEnv, shardId, this.indexSettings, routing.getExpectedShardSize() == ShardRouting.UNAVAILABLE_EXPECTED_SHARD_SIZE ? getAvgShardSizeInBytes() : routing.getExpectedShardSize(), dataPathToShardCount);
logger.debug("{} creating using a new path [{}]", shardId, path);
} else {
logger.debug("{} creating using an existing path [{}]", shardId, path);
}
if (shards.containsKey(shardId.id())) {
throw new IllegalStateException(shardId + " already exists");
}
logger.debug("creating shard_id {}", shardId);
Directory directory = directoryFactory.newDirectory(this.indexSettings, path);
store = new Store(shardId, this.indexSettings, directory, lock, new StoreCloseListener(shardId, () -> eventListener.onStoreClosed(shardId)));
eventListener.onStoreCreated(shardId);
indexShard = new IndexShard(routing, this.indexSettings, path, store, indexCache, mapperService, engineFactory, eventListener, threadPool, bigArrays, indexingOperationListeners, () -> globalCheckpointSyncer.accept(shardId), retentionLeaseSyncer, circuitBreakerService);
eventListener.indexShardStateChanged(indexShard, null, indexShard.state(), "shard created");
eventListener.afterIndexShardCreated(indexShard);
shards = newMapBuilder(shards).put(shardId.id(), indexShard).immutableMap();
success = true;
return indexShard;
} catch (ShardLockObtainFailedException e) {
throw new IOException("failed to obtain in-memory shard lock", e);
} finally {
if (success == false) {
if (lock != null) {
IOUtils.closeWhileHandlingException(lock);
}
closeShard("initialization failed", shardId, indexShard, store, eventListener);
}
}
}
Aggregations