use of voldemort.store.InsufficientOperationalNodesException in project voldemort by voldemort.
the class SlopStreamingTest method testSlopStreaming.
/*
* Does streaming puts with one node marked down. Then we drain all slops
* and ensure all keys end up on the right nodes
*/
@Test
public void testSlopStreaming() throws InterruptedException {
testEnv = new SlopStreamingTestEnvironment(1, false);
// 3-2-2 setup
testEnv.setPreferredWrite(2).setRequiredWrite(2);
testEnv.start();
Set<Integer> nodeIds = new HashSet<Integer>();
long startMs = System.currentTimeMillis();
long endMs = startMs + MAX_TOTAL_TIME_MS;
long totalPuts = 0;
long numRejectedPuts = 0;
long numAssertPuts = 0;
streamer = testEnv.makeSlopStreamingClient();
while (true) {
if (System.currentTimeMillis() > endMs) {
break;
}
// generate key
ByteArray key = new ByteArray(TestUtils.randomBytes(KEY_LENGTH));
byte[] value = TestUtils.randomBytes(VALUE_LENGTH);
Versioned<byte[]> outputValue = Versioned.value(value);
// put to nodes
try {
// if put does not throw exception
List<Node> routes = testEnv.routeRequest(key.get());
for (Node node : routes) {
numAssertPuts++;
nodeIds.add(node.getId());
Store<ByteArray, byte[], byte[]> realStore = testEnv.getRealStore(node.getId());
if (realStore instanceof InMemoryPutAssertionStorageEngine) {
((InMemoryPutAssertionStorageEngine<ByteArray, byte[], byte[]>) realStore).assertPut(key);
} else {
fail("realStore is not InMemoryPutAssertionStorageEngine. Test setup is wrong");
}
}
streamer.streamingPut(key, outputValue, testEnv.STORE_NAME);
} catch (InsufficientOperationalNodesException e) {
numRejectedPuts++;
if (logger.isDebugEnabled()) {
logger.debug("Key " + key + " is rejected for InsufficientOperationalNodes");
}
} finally {
totalPuts++;
}
}
streamer.commitToVoldemort();
streamer.closeStreamingSessions();
// bring all servers up
testEnv.warpUp();
try {
streamer.commitToVoldemort();
streamer.closeStreamingSessions();
} catch (Exception e) {
logger.warn(e);
}
// check
long numFailedAssertions = 0;
for (Integer nodeId : nodeIds) {
Store<ByteArray, byte[], byte[]> realStore = testEnv.getRealStore(nodeId);
if (realStore instanceof InMemoryPutAssertionStorageEngine) {
Set<ByteArray> keys = ((InMemoryPutAssertionStorageEngine<ByteArray, byte[], byte[]>) realStore).getFailedAssertions();
for (ByteArray key : keys) {
logger.error("key [" + key + "] is asserted but not recorded on node [" + nodeId + "]");
}
numFailedAssertions += keys.size();
} else {
fail("realStore is not InMemoryPutAssertionStorageEngine");
}
}
logger.info("Total Client Puts Rejected (InsufficientOperationalNodes): " + numRejectedPuts);
logger.info("Total Client Put Operations: " + totalPuts);
logger.info("Total Server Put Assertions: " + numAssertPuts);
logger.info("Total Server Put Lost: " + numFailedAssertions);
assertEquals(numFailedAssertions + " server puts are lost: " + numFailedAssertions, 0L, numFailedAssertions);
}
use of voldemort.store.InsufficientOperationalNodesException in project voldemort by voldemort.
the class DynamicTimeoutStoreClientTest method test.
/**
* Test the dynamic per call timeout. We do a regular put with the default
* configured timeout. We then do a put with a dynamic timeout of 200 ms
* which is less than the delay at the server side. After this we do a get
* with a dynamic timeout of 1500 ms which should succeed and return the
* value from the first put.
*/
@Test
public void test() {
long incorrectTimeout = 200;
long correctTimeout = 1500;
String key = "a";
String value = "First";
String newValue = "Second";
try {
this.dynamicTimeoutClient.put(new ByteArray(key.getBytes()), value.getBytes());
} catch (Exception e) {
fail("Error in regular put.");
}
long startTime = System.currentTimeMillis();
try {
this.dynamicTimeoutClient.putWithCustomTimeout(new CompositePutVoldemortRequest<ByteArray, byte[]>(new ByteArray(key.getBytes()), newValue.getBytes(), incorrectTimeout));
fail("Should not reach this point. The small (incorrect) timeout did not work.");
} catch (InsufficientOperationalNodesException ion) {
System.out.println("This failed as Expected.");
}
try {
List<Versioned<byte[]>> versionedValues = this.dynamicTimeoutClient.getWithCustomTimeout(new CompositeGetVoldemortRequest<ByteArray, byte[]>(new ByteArray(key.getBytes()), correctTimeout, true));
// We only expect one value in the response since resolve conflicts
// is set to true
assertTrue(versionedValues.size() == 1);
Versioned<byte[]> versionedValue = versionedValues.get(0);
long endTime = System.currentTimeMillis();
System.out.println("Total time taken = " + (endTime - startTime));
String response = new String(versionedValue.getValue());
if (!response.equals(value)) {
fail("The returned value does not match. Expected: " + value + " but Received: " + response);
}
} catch (Exception e) {
e.printStackTrace();
fail("The dynamic per call timeout did not work !");
}
}
use of voldemort.store.InsufficientOperationalNodesException in project voldemort by voldemort.
the class PerformParallelDeleteRequests method executeInternal.
private void executeInternal(final Pipeline pipeline) {
List<Node> nodes = pipelineData.getNodes();
final Map<Integer, Response<ByteArray, Object>> responses = new ConcurrentHashMap<Integer, Response<ByteArray, Object>>();
int attempts = nodes.size();
int blocks = Math.min(preferred, attempts);
final CountDownLatch attemptsLatch = new CountDownLatch(attempts);
final CountDownLatch blocksLatch = new CountDownLatch(blocks);
if (logger.isTraceEnabled())
logger.trace("Attempting " + attempts + " " + pipeline.getOperation().getSimpleName() + " operations in parallel");
long beginTime = System.nanoTime();
for (int i = 0; i < attempts; i++) {
final Node node = nodes.get(i);
pipelineData.incrementNodeIndex();
NonblockingStoreCallback callback = new NonblockingStoreCallback() {
public void requestComplete(Object result, long requestTime) {
if (logger.isTraceEnabled())
logger.trace(pipeline.getOperation().getSimpleName() + " response received (" + requestTime + " ms.) from node " + node.getId());
Response<ByteArray, Object> response = new Response<ByteArray, Object>(node, key, result, requestTime);
if (logger.isTraceEnabled()) {
logger.trace(attemptsLatch.getCount() + " attempts remaining. Will block " + " for " + blocksLatch.getCount() + " more ");
}
responses.put(node.getId(), response);
if (response.getValue() instanceof Exception && isOperationCompleted.get()) {
handleException(response, pipeline);
}
attemptsLatch.countDown();
blocksLatch.countDown();
}
};
if (logger.isTraceEnabled())
logger.info("Submitting " + pipeline.getOperation().getSimpleName() + " request on node " + node.getId());
NonblockingStore store = nonblockingStores.get(node.getId());
store.submitDeleteRequest(key, version, callback, timeoutMs);
}
try {
long ellapsedNs = System.nanoTime() - beginTime;
long remainingNs = (timeoutMs * Time.NS_PER_MS) - ellapsedNs;
if (remainingNs > 0) {
blocksLatch.await(remainingNs, TimeUnit.NANOSECONDS);
}
} catch (InterruptedException e) {
if (logger.isEnabledFor(Level.WARN))
logger.warn(e, e);
}
if (processResponses(responses, pipeline))
return;
// wait for more responses in case we did not have enough successful
// response to achieve the required count
boolean quorumSatisfied = true;
if (pipelineData.getSuccesses() < required) {
long ellapsedNs = System.nanoTime() - beginTime;
long remainingNs = (timeoutMs * Time.NS_PER_MS) - ellapsedNs;
if (remainingNs > 0) {
try {
attemptsLatch.await(remainingNs, TimeUnit.NANOSECONDS);
} catch (InterruptedException e) {
if (logger.isEnabledFor(Level.WARN))
logger.warn(e, e);
}
if (processResponses(responses, pipeline))
return;
}
if (pipelineData.getSuccesses() < required) {
pipelineData.setFatalError(new InsufficientOperationalNodesException(required + " " + pipeline.getOperation().getSimpleName() + "s required, but only " + pipelineData.getSuccesses() + " succeeded", pipelineData.getReplicationSet(), pipelineData.getNodes(), pipelineData.getFailedNodes(), pipelineData.getFailures()));
abortPipeline(pipeline);
quorumSatisfied = false;
}
}
if (quorumSatisfied) {
if (pipelineData.getZonesRequired() != null) {
int zonesSatisfied = pipelineData.getZoneResponses().size();
if (zonesSatisfied >= (pipelineData.getZonesRequired() + 1)) {
completePipeline(pipeline);
} else {
long timeMs = (System.nanoTime() - beginTime) / Time.NS_PER_MS;
if ((timeoutMs - timeMs) > 0) {
try {
attemptsLatch.await(timeoutMs - timeMs, TimeUnit.MILLISECONDS);
} catch (InterruptedException e) {
if (logger.isEnabledFor(Level.WARN))
logger.warn(e, e);
}
if (processResponses(responses, pipeline))
return;
}
if (pipelineData.getZoneResponses().size() >= (pipelineData.getZonesRequired() + 1)) {
completePipeline(pipeline);
} else {
pipelineData.setFatalError(new InsufficientZoneResponsesException((pipelineData.getZonesRequired() + 1) + " " + pipeline.getOperation().getSimpleName() + "s required zone, but only " + zonesSatisfied + " succeeded"));
abortPipeline(pipeline);
}
}
} else {
completePipeline(pipeline);
}
}
}
use of voldemort.store.InsufficientOperationalNodesException in project voldemort by voldemort.
the class PerformParallelPutRequests method execute.
@Override
public void execute(final Pipeline pipeline) {
final Node masterNode = pipelineData.getMaster();
final List<Node> nodes = pipelineData.getNodes();
final Versioned<byte[]> versionedCopy = pipelineData.getVersionedCopy();
final Integer numNodesTouchedInSerialPut = nodes.indexOf(masterNode) + 1;
numNodesPendingResponse = nodes.size() - numNodesTouchedInSerialPut;
if (logger.isDebugEnabled())
logger.debug("PUT {key:" + key + "} MasterNode={id:" + masterNode.getId() + "} totalNodesToAsyncPut=" + numNodesPendingResponse);
// initiate parallel puts
for (int i = numNodesTouchedInSerialPut; i < nodes.size(); i++) {
final Node node = nodes.get(i);
pipelineData.incrementNodeIndex();
NonblockingStoreCallback callback = new NonblockingStoreCallback() {
@Override
public void requestComplete(Object result, long requestTime) {
boolean responseHandledByMaster = false;
if (logger.isDebugEnabled())
logger.debug("PUT {key:" + key + "} response received from node={id:" + node.getId() + "} in " + requestTime + " ms)");
Response<ByteArray, Object> response;
response = new Response<ByteArray, Object>(node, key, result, requestTime);
if (logger.isDebugEnabled()) {
logger.debug("PUT {key:" + key + "} Parallel put thread trying to return result to main thread");
}
responseHandledByMaster = pipelineData.getSynchronizer().tryDelegateResponseHandling(response);
if (logger.isDebugEnabled()) {
logger.debug("PUT {key:" + key + "} Master thread accepted the response: " + responseHandledByMaster);
}
if (!responseHandledByMaster) {
if (logger.isDebugEnabled()) {
logger.debug("PUT {key:" + key + "} Master thread did not accept the response: will handle in worker thread");
}
if (PipelineRoutedStore.isSlopableFailure(response.getValue()) || response.getValue() instanceof QuotaExceededException) {
if (logger.isDebugEnabled())
logger.debug("PUT {key:" + key + "} failed on node={id:" + node.getId() + ",host:" + node.getHost() + "}");
if (isHintedHandoffEnabled()) {
boolean triedDelegateSlop = pipelineData.getSynchronizer().tryDelegateSlop(node);
if (logger.isDebugEnabled()) {
logger.debug("PUT {key:" + key + "} triedDelegateSlop: " + triedDelegateSlop);
}
if (!triedDelegateSlop) {
Slop slop = new Slop(pipelineData.getStoreName(), Slop.Operation.PUT, key, versionedCopy.getValue(), transforms, node.getId(), new Date());
pipelineData.addFailedNode(node);
if (logger.isDebugEnabled())
logger.debug("PUT {key:" + key + "} Start registering Slop(node:" + node.getId() + ",host:" + node.getHost() + ")");
hintedHandoff.sendHintParallel(node, versionedCopy.getVersion(), slop);
if (logger.isDebugEnabled())
logger.debug("PUT {key:" + key + "} Sent out request to register Slop(node: " + node.getId() + ",host:" + node.getHost() + ")");
}
}
} else {
// the exception is ignorable
if (logger.isDebugEnabled()) {
if (result instanceof Exception) {
logger.debug("PUT {key:" + key + "} will not send hint. Response is ignorable exception: " + result.getClass().toString());
} else {
logger.debug("PUT {key:" + key + "} will not send hint. Response is success");
}
}
}
if (result instanceof Exception && !(result instanceof ObsoleteVersionException)) {
if (response.getValue() instanceof InvalidMetadataException) {
pipelineData.reportException((InvalidMetadataException) response.getValue());
logger.warn("Received invalid metadata problem after a successful " + pipeline.getOperation().getSimpleName() + " call on node " + node.getId() + ", store '" + pipelineData.getStoreName() + "'");
} else if (response.getValue() instanceof QuotaExceededException) {
/**
* TODO Not sure if we need to count this
* Exception for stats or silently ignore and
* just log a warning. While
* QuotaExceededException thrown from other
* places mean the operation failed, this one
* does not fail the operation but instead
* stores slops. Introduce a new Exception in
* client side to just monitor how mamy Async
* writes fail on exceeding Quota?
*
*/
logger.warn("Received QuotaExceededException after a successful " + pipeline.getOperation().getSimpleName() + " call on node " + node.getId() + ", store '" + pipelineData.getStoreName() + "', master-node '" + masterNode.getId() + "'");
} else {
handleResponseError(response, pipeline, failureDetector);
}
}
}
}
};
if (logger.isTraceEnabled())
logger.trace("Submitting " + pipeline.getOperation().getSimpleName() + " request on node " + node.getId() + " for key " + key);
NonblockingStore store = nonblockingStores.get(node.getId());
store.submitPutRequest(key, versionedCopy, transforms, callback, timeoutMs);
}
try {
boolean preferredSatisfied = false;
while (true) {
long elapsedNs = System.nanoTime() - pipelineData.getStartTimeNs();
long remainingNs = (timeoutMs * Time.NS_PER_MS) - elapsedNs;
remainingNs = Math.max(0, remainingNs);
// preferred check
if (numResponsesGot >= preferred - 1) {
preferredSatisfied = true;
}
quorumSatisfied = isQuorumSatisfied();
zonesSatisfied = isZonesSatisfied();
if (quorumSatisfied && zonesSatisfied && preferredSatisfied || remainingNs <= 0 || numNodesPendingResponse <= 0) {
pipelineData.getSynchronizer().cutoffHandling();
break;
} else {
if (logger.isTraceEnabled()) {
logger.trace("PUT {key:" + key + "} trying to poll from queue");
}
Response<ByteArray, Object> response = pipelineData.getSynchronizer().responseQueuePoll(remainingNs, TimeUnit.NANOSECONDS);
processResponse(response, pipeline);
if (logger.isTraceEnabled()) {
logger.trace("PUT {key:" + key + "} tried to poll from queue. Null?: " + (response == null) + " numResponsesGot:" + numResponsesGot + " parallelResponseToWait: " + numNodesPendingResponse + "; preferred-1: " + (preferred - 1) + "; preferredOK: " + preferredSatisfied + " quorumOK: " + quorumSatisfied + "; zoneOK: " + zonesSatisfied);
}
}
}
// leftover)
while (!pipelineData.getSynchronizer().responseQueueIsEmpty()) {
Response<ByteArray, Object> response = pipelineData.getSynchronizer().responseQueuePoll(0, TimeUnit.NANOSECONDS);
processResponse(response, pipeline);
}
quorumSatisfied = isQuorumSatisfied();
zonesSatisfied = isZonesSatisfied();
if (quorumSatisfied && zonesSatisfied) {
if (logger.isDebugEnabled()) {
logger.debug("PUT {key:" + key + "} succeeded at parallel put stage");
}
pipelineData.getSynchronizer().disallowDelegateSlop();
pipeline.addEvent(completeEvent);
} else {
VoldemortException fatalError;
if (!quorumSatisfied) {
if (logger.isDebugEnabled()) {
logger.debug("PUT {key:" + key + "} failed due to insufficient nodes. required=" + required + " success=" + pipelineData.getSuccesses());
}
fatalError = new InsufficientOperationalNodesException(required + " " + pipeline.getOperation().getSimpleName() + "s required, but only " + pipelineData.getSuccesses() + " succeeded", pipelineData.getReplicationSet(), pipelineData.getNodes(), pipelineData.getFailedNodes(), pipelineData.getFailures());
pipelineData.setFatalError(fatalError);
} else if (!zonesSatisfied) {
if (logger.isDebugEnabled()) {
logger.debug("PUT {key:" + key + "} failed due to insufficient zones. required=" + pipelineData.getZonesRequired() + 1 + " success=" + pipelineData.getZoneResponses().size());
}
fatalError = new InsufficientZoneResponsesException((pipelineData.getZonesRequired() + 1) + " " + pipeline.getOperation().getSimpleName() + "s required zone, but only " + (pipelineData.getZoneResponses().size()) + " succeeded. Failing nodes : " + pipelineData.getFailedNodes());
pipelineData.setFatalError(fatalError);
}
pipeline.abort();
}
} catch (InterruptedException e) {
if (logger.isEnabledFor(Level.WARN))
logger.warn(e, e);
} catch (NoSuchElementException e) {
if (logger.isEnabledFor(Level.ERROR))
logger.error("Response Queue is empty. There may be a bug in PerformParallelPutRequest", e);
} finally {
if (logger.isDebugEnabled()) {
logger.debug("PUT {key:" + key + "} marking parallel put stage finished");
}
}
}
use of voldemort.store.InsufficientOperationalNodesException in project voldemort by voldemort.
the class PerformParallelRequests method execute.
public void execute(final Pipeline pipeline) {
List<Node> nodes = pipelineData.getNodes();
int attempts = Math.min(preferred, nodes.size());
final Map<Integer, Response<ByteArray, Object>> responses = new ConcurrentHashMap<Integer, Response<ByteArray, Object>>();
final CountDownLatch latch = new CountDownLatch(attempts);
if (logger.isTraceEnabled())
logger.trace("Attempting " + attempts + " " + pipeline.getOperation().getSimpleName() + " operations in parallel for key " + key);
final AtomicBoolean isResponseProcessed = new AtomicBoolean(false);
for (int i = 0; i < attempts; i++) {
final Node node = nodes.get(i);
pipelineData.incrementNodeIndex();
final long startMs = logger.isDebugEnabled() ? System.currentTimeMillis() : -1;
NonblockingStoreCallback callback = new NonblockingStoreCallback() {
public void requestComplete(Object result, long requestTime) {
if (logger.isTraceEnabled())
logger.trace(pipeline.getOperation().getSimpleName() + " response received (" + requestTime + " ms.) from node " + node.getId() + "for key " + key);
Response<ByteArray, Object> response = new Response<ByteArray, Object>(node, key, result, requestTime);
if (logger.isDebugEnabled())
logger.debug("Finished " + pipeline.getOperation().getSimpleName() + " for key " + ByteUtils.toHexString(key.get()) + " (keyRef: " + System.identityHashCode(key) + "); started at " + startMs + " took " + requestTime + " ms on node " + node.getId() + "(" + node.getHost() + ")");
responses.put(node.getId(), response);
latch.countDown();
// This reduces the window where an exception is lost
if (isResponseProcessed.get() && response.getValue() instanceof Exception) {
if (response.getValue() instanceof InvalidMetadataException) {
pipelineData.reportException((InvalidMetadataException) response.getValue());
logger.warn("Received invalid metadata problem after a successful " + pipeline.getOperation().getSimpleName() + " call on node " + node.getId() + ", store '" + pipelineData.getStoreName() + "' for key " + key);
} else {
handleResponseError(response, pipeline, failureDetector);
}
}
}
};
if (logger.isTraceEnabled())
logger.trace("Submitting " + pipeline.getOperation().getSimpleName() + " request on node " + node.getId() + " for key " + key);
NonblockingStore store = nonblockingStores.get(node.getId());
if (pipeline.getOperation() == Operation.GET)
store.submitGetRequest(key, transforms, callback, timeoutMs);
else if (pipeline.getOperation() == Operation.GET_VERSIONS)
store.submitGetVersionsRequest(key, callback, timeoutMs);
else
throw new IllegalStateException(getClass().getName() + " does not support pipeline operation " + pipeline.getOperation());
}
try {
latch.await(timeoutMs, TimeUnit.MILLISECONDS);
} catch (InterruptedException e) {
if (logger.isEnabledFor(Level.WARN))
logger.warn(e, e);
}
for (Response<ByteArray, Object> response : responses.values()) {
if (response.getValue() instanceof Exception) {
if (handleResponseError(response, pipeline, failureDetector))
return;
} else {
pipelineData.incrementSuccesses();
Response<ByteArray, V> rCast = Utils.uncheckedCast(response);
pipelineData.getResponses().add(rCast);
failureDetector.recordSuccess(response.getNode(), response.getRequestTime());
pipelineData.getZoneResponses().add(response.getNode().getZoneId());
}
}
isResponseProcessed.set(true);
if (logger.isDebugEnabled())
logger.debug("GET for key " + ByteUtils.toHexString(key.get()) + " (keyRef: " + System.identityHashCode(key) + "); successes: " + pipelineData.getSuccesses() + " preferred: " + preferred + " required: " + required);
if (pipelineData.getSuccesses() < required) {
if (insufficientSuccessesEvent != null) {
pipeline.addEvent(insufficientSuccessesEvent);
} else {
pipelineData.setFatalError(new InsufficientOperationalNodesException(required + " " + pipeline.getOperation().getSimpleName() + "s required, but only " + pipelineData.getSuccesses() + " succeeded", pipelineData.getReplicationSet(), pipelineData.getNodes(), pipelineData.getFailedNodes(), pipelineData.getFailures()));
pipeline.abort();
}
} else {
if (pipelineData.getZonesRequired() != null) {
int zonesSatisfied = pipelineData.getZoneResponses().size();
if (zonesSatisfied >= (pipelineData.getZonesRequired() + 1)) {
pipeline.addEvent(completeEvent);
} else {
if (logger.isDebugEnabled()) {
logger.debug("Operation " + pipeline.getOperation().getSimpleName() + "failed due to insufficient zone responses, required " + pipelineData.getZonesRequired() + " obtained " + zonesSatisfied + " " + pipelineData.getZoneResponses() + " for key " + key);
}
if (this.insufficientZonesEvent != null) {
pipeline.addEvent(this.insufficientZonesEvent);
} else {
pipelineData.setFatalError(new InsufficientZoneResponsesException((pipelineData.getZonesRequired() + 1) + " " + pipeline.getOperation().getSimpleName() + "s required zone, but only " + zonesSatisfied + " succeeded"));
}
}
} else {
pipeline.addEvent(completeEvent);
}
}
}
Aggregations