use of java.util.concurrent.CopyOnWriteArrayList in project elasticsearch by elastic.
the class RestClientMultipleHostsIntegTests method testAsyncRequests.
public void testAsyncRequests() throws Exception {
int numRequests = randomIntBetween(5, 20);
final CountDownLatch latch = new CountDownLatch(numRequests);
final List<TestResponse> responses = new CopyOnWriteArrayList<>();
for (int i = 0; i < numRequests; i++) {
final String method = RestClientTestUtil.randomHttpMethod(getRandom());
//we don't test status codes that are subject to retries as they interfere with hosts being stopped
final int statusCode = randomBoolean() ? randomOkStatusCode(getRandom()) : randomErrorNoRetryStatusCode(getRandom());
restClient.performRequestAsync(method, "/" + statusCode, new ResponseListener() {
@Override
public void onSuccess(Response response) {
responses.add(new TestResponse(method, statusCode, response));
latch.countDown();
}
@Override
public void onFailure(Exception exception) {
responses.add(new TestResponse(method, statusCode, exception));
latch.countDown();
}
});
}
assertTrue(latch.await(5, TimeUnit.SECONDS));
assertEquals(numRequests, responses.size());
for (TestResponse testResponse : responses) {
Response response = testResponse.getResponse();
assertEquals(testResponse.method, response.getRequestLine().getMethod());
assertEquals(testResponse.statusCode, response.getStatusLine().getStatusCode());
assertEquals((pathPrefix.length() > 0 ? pathPrefix : "") + "/" + testResponse.statusCode, response.getRequestLine().getUri());
}
}
use of java.util.concurrent.CopyOnWriteArrayList in project crate by crate.
the class BlobRecoverySourceHandler method phase1.
/**
* Perform phase1 of the recovery operations. Once this {@link SnapshotIndexCommit}
* snapshot has been performed no commit operations (files being fsync'd)
* are effectively allowed on this index until all recovery phases are done
* <p/>
* Phase1 examines the segment files on the target node and copies over the
* segments that are missing. Only segments that have the same size and
* checksum can be reused
*/
public void phase1(final SnapshotIndexCommit snapshot, final Translog.View translogView) {
cancellableThreads.checkForCancel();
// Total size of segment files that are recovered
long totalSize = 0;
// Total size of segment files that were able to be re-used
long existingTotalSize = 0;
final Store store = shard.store();
store.incRef();
try {
// CRATE CHANGE
if (blobRecoveryHandler != null) {
blobRecoveryHandler.phase1();
}
StopWatch stopWatch = new StopWatch().start();
final Store.MetadataSnapshot recoverySourceMetadata;
try {
recoverySourceMetadata = store.getMetadata(snapshot);
} catch (CorruptIndexException | IndexFormatTooOldException | IndexFormatTooNewException ex) {
shard.engine().failEngine("recovery", ex);
throw ex;
}
for (String name : snapshot.getFiles()) {
final StoreFileMetaData md = recoverySourceMetadata.get(name);
if (md == null) {
logger.info("Snapshot differs from actual index for file: {} meta: {}", name, recoverySourceMetadata.asMap());
throw new CorruptIndexException("Snapshot differs from actual index - maybe index was removed metadata has " + recoverySourceMetadata.asMap().size() + " files", name);
}
}
// Generate a "diff" of all the identical, different, and missing
// segment files on the target node, using the existing files on
// the source node
String recoverySourceSyncId = recoverySourceMetadata.getSyncId();
String recoveryTargetSyncId = request.metadataSnapshot().getSyncId();
final boolean recoverWithSyncId = recoverySourceSyncId != null && recoverySourceSyncId.equals(recoveryTargetSyncId);
if (recoverWithSyncId) {
final long numDocsTarget = request.metadataSnapshot().getNumDocs();
final long numDocsSource = recoverySourceMetadata.getNumDocs();
if (numDocsTarget != numDocsSource) {
throw new IllegalStateException("try to recover " + request.shardId() + " from primary shard with sync id but number of docs differ: " + numDocsTarget + " (" + request.sourceNode().getName() + ", primary) vs " + numDocsSource + "(" + request.targetNode().getName() + ")");
}
// we shortcut recovery here because we have nothing to copy. but we must still start the engine on the target.
// so we don't return here
logger.trace("[{}][{}] skipping [phase1] to {} - identical sync id [{}] found on both source and target", indexName, shardId, request.targetNode(), recoverySourceSyncId);
} else {
final Store.RecoveryDiff diff = recoverySourceMetadata.recoveryDiff(request.metadataSnapshot());
for (StoreFileMetaData md : diff.identical) {
response.phase1ExistingFileNames.add(md.name());
response.phase1ExistingFileSizes.add(md.length());
existingTotalSize += md.length();
if (logger.isTraceEnabled()) {
logger.trace("[{}][{}] recovery [phase1] to {}: not recovering [{}], exists in local store and has checksum [{}], size [{}]", indexName, shardId, request.targetNode(), md.name(), md.checksum(), md.length());
}
totalSize += md.length();
}
for (StoreFileMetaData md : Iterables.concat(diff.different, diff.missing)) {
if (request.metadataSnapshot().asMap().containsKey(md.name())) {
logger.trace("[{}][{}] recovery [phase1] to {}: recovering [{}], exists in local store, but is different: remote [{}], local [{}]", indexName, shardId, request.targetNode(), md.name(), request.metadataSnapshot().asMap().get(md.name()), md);
} else {
logger.trace("[{}][{}] recovery [phase1] to {}: recovering [{}], does not exists in remote", indexName, shardId, request.targetNode(), md.name());
}
response.phase1FileNames.add(md.name());
response.phase1FileSizes.add(md.length());
totalSize += md.length();
}
response.phase1TotalSize = totalSize;
response.phase1ExistingTotalSize = existingTotalSize;
logger.trace("[{}][{}] recovery [phase1] to {}: recovering_files [{}] with total_size [{}], reusing_files [{}] with total_size [{}]", indexName, shardId, request.targetNode(), response.phase1FileNames.size(), new ByteSizeValue(totalSize), response.phase1ExistingFileNames.size(), new ByteSizeValue(existingTotalSize));
cancellableThreads.execute(new Interruptable() {
@Override
public void run() throws InterruptedException {
RecoveryFilesInfoRequest recoveryInfoFilesRequest = new RecoveryFilesInfoRequest(request.recoveryId(), request.shardId(), response.phase1FileNames, response.phase1FileSizes, response.phase1ExistingFileNames, response.phase1ExistingFileSizes, translogView.totalOperations());
transportService.submitRequest(request.targetNode(), RecoveryTarget.Actions.FILES_INFO, recoveryInfoFilesRequest, TransportRequestOptions.builder().withTimeout(recoverySettings.internalActionTimeout()).build(), EmptyTransportResponseHandler.INSTANCE_SAME).txGet();
}
});
// This latch will be used to wait until all files have been transferred to the target node
final CountDownLatch latch = new CountDownLatch(response.phase1FileNames.size());
final CopyOnWriteArrayList<Throwable> exceptions = new CopyOnWriteArrayList<>();
final AtomicReference<Throwable> corruptedEngine = new AtomicReference<>();
int fileIndex = 0;
ThreadPoolExecutor pool;
// How many bytes we've copied since we last called RateLimiter.pause
final AtomicLong bytesSinceLastPause = new AtomicLong();
for (final String name : response.phase1FileNames) {
long fileSize = response.phase1FileSizes.get(fileIndex);
// separately.
if (fileSize > RecoverySettings.SMALL_FILE_CUTOFF_BYTES) {
pool = recoverySettings.concurrentStreamPool();
} else {
pool = recoverySettings.concurrentSmallFileStreamPool();
}
pool.execute(new AbstractRunnable() {
@Override
public void onFailure(Throwable t) {
// we either got rejected or the store can't be incremented / we are canceled
logger.debug("Failed to transfer file [" + name + "] on recovery");
}
@Override
public void onAfter() {
// Signify this file has completed by decrementing the latch
latch.countDown();
}
@Override
protected void doRun() {
cancellableThreads.checkForCancel();
store.incRef();
final StoreFileMetaData md = recoverySourceMetadata.get(name);
try (final IndexInput indexInput = store.directory().openInput(name, IOContext.READONCE)) {
// at least one!
final int BUFFER_SIZE = (int) Math.max(1, recoverySettings.fileChunkSize().getBytes());
final byte[] buf = new byte[BUFFER_SIZE];
boolean shouldCompressRequest = recoverySettings.compress();
if (CompressorFactory.isCompressed(indexInput)) {
shouldCompressRequest = false;
}
final long len = indexInput.length();
long readCount = 0;
final TransportRequestOptions requestOptions = TransportRequestOptions.builder().withCompress(shouldCompressRequest).withType(TransportRequestOptions.Type.RECOVERY).withTimeout(recoverySettings.internalActionTimeout()).build();
while (readCount < len) {
if (shard.state() == IndexShardState.CLOSED) {
// check if the shard got closed on us
throw new IndexShardClosedException(shard.shardId());
}
int toRead = readCount + BUFFER_SIZE > len ? (int) (len - readCount) : BUFFER_SIZE;
final long position = indexInput.getFilePointer();
// Pause using the rate limiter, if desired, to throttle the recovery
RateLimiter rl = recoverySettings.rateLimiter();
long throttleTimeInNanos = 0;
if (rl != null) {
long bytes = bytesSinceLastPause.addAndGet(toRead);
if (bytes > rl.getMinPauseCheckBytes()) {
// Time to pause
bytesSinceLastPause.addAndGet(-bytes);
throttleTimeInNanos = rl.pause(bytes);
shard.recoveryStats().addThrottleTime(throttleTimeInNanos);
}
}
indexInput.readBytes(buf, 0, toRead, false);
final BytesArray content = new BytesArray(buf, 0, toRead);
readCount += toRead;
final boolean lastChunk = readCount == len;
final RecoveryFileChunkRequest fileChunkRequest = new RecoveryFileChunkRequest(request.recoveryId(), request.shardId(), md, position, content, lastChunk, translogView.totalOperations(), throttleTimeInNanos);
cancellableThreads.execute(new Interruptable() {
@Override
public void run() throws InterruptedException {
// Actually send the file chunk to the target node, waiting for it to complete
transportService.submitRequest(request.targetNode(), RecoveryTarget.Actions.FILE_CHUNK, fileChunkRequest, requestOptions, EmptyTransportResponseHandler.INSTANCE_SAME).txGet();
}
});
}
} catch (Throwable e) {
final Throwable corruptIndexException;
if ((corruptIndexException = ExceptionsHelper.unwrapCorruption(e)) != null) {
if (store.checkIntegrityNoException(md) == false) {
// we are corrupted on the primary -- fail!
logger.warn("{} Corrupted file detected {} checksum mismatch", shard.shardId(), md);
if (corruptedEngine.compareAndSet(null, corruptIndexException) == false) {
// if we are not the first exception, add ourselves as suppressed to the main one:
corruptedEngine.get().addSuppressed(e);
}
} else {
// corruption has happened on the way to replica
RemoteTransportException exception = new RemoteTransportException("File corruption occurred on recovery but checksums are ok", null);
exception.addSuppressed(e);
// last exception first
exceptions.add(0, exception);
logger.warn("{} Remote file corruption on node {}, recovering {}. local checksum OK", corruptIndexException, shard.shardId(), request.targetNode(), md);
}
} else {
// last exceptions first
exceptions.add(0, e);
}
} finally {
store.decRef();
}
}
});
fileIndex++;
}
cancellableThreads.execute(new Interruptable() {
@Override
public void run() throws InterruptedException {
// Wait for all files that need to be transferred to finish transferring
latch.await();
}
});
if (corruptedEngine.get() != null) {
shard.engine().failEngine("recovery", corruptedEngine.get());
throw corruptedEngine.get();
} else {
ExceptionsHelper.rethrowAndSuppress(exceptions);
}
cancellableThreads.execute(new Interruptable() {
@Override
public void run() throws InterruptedException {
// are deleted
try {
transportService.submitRequest(request.targetNode(), RecoveryTarget.Actions.CLEAN_FILES, new RecoveryCleanFilesRequest(request.recoveryId(), shard.shardId(), recoverySourceMetadata, translogView.totalOperations()), TransportRequestOptions.builder().withTimeout(recoverySettings.internalActionTimeout()).build(), EmptyTransportResponseHandler.INSTANCE_SAME).txGet();
} catch (RemoteTransportException remoteException) {
final IOException corruptIndexException;
// - maybe due to old segments without checksums or length only checks
if ((corruptIndexException = ExceptionsHelper.unwrapCorruption(remoteException)) != null) {
try {
final Store.MetadataSnapshot recoverySourceMetadata = store.getMetadata(snapshot);
StoreFileMetaData[] metadata = Iterables.toArray(recoverySourceMetadata, StoreFileMetaData.class);
ArrayUtil.timSort(metadata, new Comparator<StoreFileMetaData>() {
@Override
public int compare(StoreFileMetaData o1, StoreFileMetaData o2) {
// check small files first
return Long.compare(o1.length(), o2.length());
}
});
for (StoreFileMetaData md : metadata) {
logger.debug("{} checking integrity for file {} after remove corruption exception", shard.shardId(), md);
if (store.checkIntegrityNoException(md) == false) {
// we are corrupted on the primary -- fail!
shard.engine().failEngine("recovery", corruptIndexException);
logger.warn("{} Corrupted file detected {} checksum mismatch", shard.shardId(), md);
throw corruptIndexException;
}
}
} catch (IOException ex) {
remoteException.addSuppressed(ex);
throw remoteException;
}
// corruption has happened on the way to replica
RemoteTransportException exception = new RemoteTransportException("File corruption occurred on recovery but checksums are ok", null);
exception.addSuppressed(remoteException);
logger.warn("{} Remote file corruption during finalization on node {}, recovering {}. local checksum OK", corruptIndexException, shard.shardId(), request.targetNode());
throw exception;
} else {
throw remoteException;
}
}
}
});
}
prepareTargetForTranslog(translogView);
logger.trace("[{}][{}] recovery [phase1] to {}: took [{}]", indexName, shardId, request.targetNode(), stopWatch.totalTime());
response.phase1Time = stopWatch.totalTime().millis();
} catch (Throwable e) {
throw new RecoverFilesRecoveryException(request.shardId(), response.phase1FileNames.size(), new ByteSizeValue(totalSize), e);
} finally {
store.decRef();
}
}
use of java.util.concurrent.CopyOnWriteArrayList in project elasticsearch by elastic.
the class ESIntegTestCase method indexRandom.
/**
* Indexes the given {@link IndexRequestBuilder} instances randomly. It shuffles the given builders and either
* indexes them in a blocking or async fashion. This is very useful to catch problems that relate to internal document
* ids or index segment creations. Some features might have bug when a given document is the first or the last in a
* segment or if only one document is in a segment etc. This method prevents issues like this by randomizing the index
* layout.
*
* @param forceRefresh if <tt>true</tt> all involved indices are refreshed once the documents are indexed.
* @param dummyDocuments if <tt>true</tt> some empty dummy documents may be randomly inserted into the document list and deleted once
* all documents are indexed. This is useful to produce deleted documents on the server side.
* @param maybeFlush if <tt>true</tt> this method may randomly execute full flushes after index operations.
* @param builders the documents to index.
*/
public void indexRandom(boolean forceRefresh, boolean dummyDocuments, boolean maybeFlush, List<IndexRequestBuilder> builders) throws InterruptedException, ExecutionException {
Random random = random();
Set<String> indicesSet = new HashSet<>();
for (IndexRequestBuilder builder : builders) {
indicesSet.add(builder.request().index());
}
Set<Tuple<String, String>> bogusIds = new HashSet<>();
if (random.nextBoolean() && !builders.isEmpty() && dummyDocuments) {
builders = new ArrayList<>(builders);
final String[] indices = indicesSet.toArray(new String[indicesSet.size()]);
// inject some bogus docs
final int numBogusDocs = scaledRandomIntBetween(1, builders.size() * 2);
final int unicodeLen = between(1, 10);
for (int i = 0; i < numBogusDocs; i++) {
String id = randomRealisticUnicodeOfLength(unicodeLen) + Integer.toString(dummmyDocIdGenerator.incrementAndGet());
String index = RandomPicks.randomFrom(random, indices);
bogusIds.add(new Tuple<>(index, id));
builders.add(client().prepareIndex(index, RANDOM_BOGUS_TYPE, id).setSource("{}", XContentType.JSON));
}
}
final String[] indices = indicesSet.toArray(new String[indicesSet.size()]);
Collections.shuffle(builders, random());
final CopyOnWriteArrayList<Tuple<IndexRequestBuilder, Exception>> errors = new CopyOnWriteArrayList<>();
List<CountDownLatch> inFlightAsyncOperations = new ArrayList<>();
// If you are indexing just a few documents then frequently do it one at a time. If many then frequently in bulk.
if (builders.size() < FREQUENT_BULK_THRESHOLD ? frequently() : builders.size() < ALWAYS_BULK_THRESHOLD ? rarely() : false) {
if (frequently()) {
logger.info("Index [{}] docs async: [{}] bulk: [{}]", builders.size(), true, false);
for (IndexRequestBuilder indexRequestBuilder : builders) {
indexRequestBuilder.execute(new PayloadLatchedActionListener<IndexResponse, IndexRequestBuilder>(indexRequestBuilder, newLatch(inFlightAsyncOperations), errors));
postIndexAsyncActions(indices, inFlightAsyncOperations, maybeFlush);
}
} else {
logger.info("Index [{}] docs async: [{}] bulk: [{}]", builders.size(), false, false);
for (IndexRequestBuilder indexRequestBuilder : builders) {
indexRequestBuilder.execute().actionGet();
postIndexAsyncActions(indices, inFlightAsyncOperations, maybeFlush);
}
}
} else {
List<List<IndexRequestBuilder>> partition = eagerPartition(builders, Math.min(MAX_BULK_INDEX_REQUEST_SIZE, Math.max(1, (int) (builders.size() * randomDouble()))));
logger.info("Index [{}] docs async: [{}] bulk: [{}] partitions [{}]", builders.size(), false, true, partition.size());
for (List<IndexRequestBuilder> segmented : partition) {
BulkRequestBuilder bulkBuilder = client().prepareBulk();
for (IndexRequestBuilder indexRequestBuilder : segmented) {
bulkBuilder.add(indexRequestBuilder);
}
BulkResponse actionGet = bulkBuilder.execute().actionGet();
assertThat(actionGet.hasFailures() ? actionGet.buildFailureMessage() : "", actionGet.hasFailures(), equalTo(false));
}
}
for (CountDownLatch operation : inFlightAsyncOperations) {
operation.await();
}
final List<Exception> actualErrors = new ArrayList<>();
for (Tuple<IndexRequestBuilder, Exception> tuple : errors) {
if (ExceptionsHelper.unwrapCause(tuple.v2()) instanceof EsRejectedExecutionException) {
// re-index if rejected
tuple.v1().execute().actionGet();
} else {
actualErrors.add(tuple.v2());
}
}
assertThat(actualErrors, emptyIterable());
if (!bogusIds.isEmpty()) {
// delete the bogus types again - it might trigger merges or at least holes in the segments and enforces deleted docs!
for (Tuple<String, String> doc : bogusIds) {
assertEquals("failed to delete a dummy doc [" + doc.v1() + "][" + doc.v2() + "]", DocWriteResponse.Result.DELETED, client().prepareDelete(doc.v1(), RANDOM_BOGUS_TYPE, doc.v2()).get().getResult());
}
}
if (forceRefresh) {
assertNoFailures(client().admin().indices().prepareRefresh(indices).setIndicesOptions(IndicesOptions.lenientExpandOpen()).execute().get());
}
}
use of java.util.concurrent.CopyOnWriteArrayList in project elasticsearch by elastic.
the class CacheTests method testDependentKeyDeadlock.
public void testDependentKeyDeadlock() throws BrokenBarrierException, InterruptedException {
class Key {
private final int key;
Key(int key) {
this.key = key;
}
@Override
public boolean equals(Object o) {
if (this == o)
return true;
if (o == null || getClass() != o.getClass())
return false;
Key key1 = (Key) o;
return key == key1.key;
}
@Override
public int hashCode() {
return key % 2;
}
}
int numberOfThreads = randomIntBetween(2, 32);
final Cache<Key, Integer> cache = CacheBuilder.<Key, Integer>builder().build();
CopyOnWriteArrayList<ExecutionException> failures = new CopyOnWriteArrayList<>();
CyclicBarrier barrier = new CyclicBarrier(1 + numberOfThreads);
CountDownLatch deadlockLatch = new CountDownLatch(numberOfThreads);
List<Thread> threads = new ArrayList<>();
for (int i = 0; i < numberOfThreads; i++) {
Thread thread = new Thread(() -> {
try {
try {
barrier.await();
} catch (BrokenBarrierException | InterruptedException e) {
throw new AssertionError(e);
}
Random random = new Random(random().nextLong());
for (int j = 0; j < numberOfEntries; j++) {
Key key = new Key(random.nextInt(numberOfEntries));
try {
cache.computeIfAbsent(key, k -> {
if (k.key == 0) {
return 0;
} else {
Integer value = cache.get(new Key(k.key / 2));
return value != null ? value : 0;
}
});
} catch (ExecutionException e) {
failures.add(e);
break;
}
}
} finally {
// successfully avoided deadlock, release the main thread
deadlockLatch.countDown();
}
});
threads.add(thread);
thread.start();
}
AtomicBoolean deadlock = new AtomicBoolean();
assert !deadlock.get();
// start a watchdog service
ScheduledExecutorService scheduler = Executors.newScheduledThreadPool(1);
scheduler.scheduleAtFixedRate(() -> {
Set<Long> ids = threads.stream().map(t -> t.getId()).collect(Collectors.toSet());
ThreadMXBean mxBean = ManagementFactory.getThreadMXBean();
long[] deadlockedThreads = mxBean.findDeadlockedThreads();
if (!deadlock.get() && deadlockedThreads != null) {
for (long deadlockedThread : deadlockedThreads) {
// ensure that we detected deadlock on our threads
if (ids.contains(deadlockedThread)) {
deadlock.set(true);
// release the main test thread to fail the test
for (int i = 0; i < numberOfThreads; i++) {
deadlockLatch.countDown();
}
break;
}
}
}
}, 1, 1, TimeUnit.SECONDS);
// everything is setup, release the hounds
barrier.await();
// wait for either deadlock to be detected or the threads to terminate
deadlockLatch.await();
// shutdown the watchdog service
scheduler.shutdown();
assertThat(failures, is(empty()));
assertFalse("deadlock", deadlock.get());
}
use of java.util.concurrent.CopyOnWriteArrayList in project elasticsearch by elastic.
the class CacheTests method testComputeIfAbsentCallsOnce.
public void testComputeIfAbsentCallsOnce() throws BrokenBarrierException, InterruptedException {
int numberOfThreads = randomIntBetween(2, 32);
final Cache<Integer, String> cache = CacheBuilder.<Integer, String>builder().build();
AtomicReferenceArray flags = new AtomicReferenceArray(numberOfEntries);
for (int j = 0; j < numberOfEntries; j++) {
flags.set(j, false);
}
CopyOnWriteArrayList<ExecutionException> failures = new CopyOnWriteArrayList<>();
CyclicBarrier barrier = new CyclicBarrier(1 + numberOfThreads);
for (int i = 0; i < numberOfThreads; i++) {
Thread thread = new Thread(() -> {
try {
barrier.await();
for (int j = 0; j < numberOfEntries; j++) {
try {
cache.computeIfAbsent(j, key -> {
assertTrue(flags.compareAndSet(key, false, true));
return Integer.toString(key);
});
} catch (ExecutionException e) {
failures.add(e);
break;
}
}
barrier.await();
} catch (BrokenBarrierException | InterruptedException e) {
throw new AssertionError(e);
}
});
thread.start();
}
// wait for all threads to be ready
barrier.await();
// wait for all threads to finish
barrier.await();
assertThat(failures, is(empty()));
}
Aggregations