private DataSetStats calculateStats(StorageEngine<ByteArray, byte[], byte[]> store) {
DataSetStats stats = new DataSetStats();
ClosableIterator<Pair<ByteArray, Versioned<byte[]>>> iter = store.entries();
try {
int count = 0;
while (iter.hasNext()) {
Pair<ByteArray, Versioned<byte[]>> pair = iter.next();
VectorClock clock = (VectorClock) pair.getSecond().getVersion();
stats.countEntry(
pair.getFirst().length(), pair.getSecond().getValue().length + clock.sizeInBytes());
if (count % 10000 == 0) logger.debug("Processing key " + count);
count++;
}
} finally {
iter.close();
}
return stats;
}
public VAdminProto.DeletePartitionEntriesResponse handleDeletePartitionEntries(
VAdminProto.DeletePartitionEntriesRequest request) {
VAdminProto.DeletePartitionEntriesResponse.Builder response =
VAdminProto.DeletePartitionEntriesResponse.newBuilder();
ClosableIterator<Pair<ByteArray, Versioned<byte[]>>> iterator = null;
try {
String storeName = request.getStore();
List<Integer> partitions = request.getPartitionsList();
StorageEngine<ByteArray, byte[]> storageEngine = getStorageEngine(storeRepository, storeName);
VoldemortFilter filter =
(request.hasFilter())
? getFilterFromRequest(request.getFilter(), voldemortConfig, networkClassLoader)
: new DefaultVoldemortFilter();
RoutingStrategy routingStrategy = metadataStore.getRoutingStrategy(storageEngine.getName());
EventThrottler throttler = new EventThrottler(voldemortConfig.getStreamMaxReadBytesPerSec());
iterator = storageEngine.entries();
int deleteSuccess = 0;
while (iterator.hasNext()) {
Pair<ByteArray, Versioned<byte[]>> entry = iterator.next();
ByteArray key = entry.getFirst();
Versioned<byte[]> value = entry.getSecond();
throttler.maybeThrottle(key.length() + valueSize(value));
if (checkKeyBelongsToDeletePartition(key.get(), partitions, routingStrategy)
&& filter.accept(key, value)) {
if (storageEngine.delete(key, value.getVersion())) deleteSuccess++;
}
}
response.setCount(deleteSuccess);
} catch (VoldemortException e) {
response.setError(ProtoUtils.encodeError(errorCodeMapper, e));
logger.error(
"handleDeletePartitionEntries failed for request(" + request.toString() + ")", e);
} finally {
if (null != iterator) iterator.close();
}
return response.build();
}
public void run() {
// don't try to run slop pusher job when rebalancing
if (metadataStore
.getServerState()
.equals(MetadataStore.VoldemortState.REBALANCING_MASTER_SERVER)) {
logger.error("Cannot run slop pusher job since Voldemort server is rebalancing");
return;
}
boolean terminatedEarly = false;
Date startTime = new Date();
logger.info("Started streaming slop pusher job at " + startTime);
SlopStorageEngine slopStorageEngine = storeRepo.getSlopStore();
ClosableIterator<Pair<ByteArray, Versioned<Slop>>> iterator = null;
if (adminClient == null) {
adminClient =
new AdminClient(
cluster,
new AdminClientConfig()
.setMaxThreads(cluster.getNumberOfNodes())
.setMaxConnectionsPerNode(1));
}
if (voldemortConfig.getSlopZonesDownToTerminate() > 0) {
// Populating the zone mapping for early termination
zoneMapping.clear();
for (Node n : cluster.getNodes()) {
if (failureDetector.isAvailable(n)) {
Set<Integer> nodes = zoneMapping.get(n.getZoneId());
if (nodes == null) {
nodes = Sets.newHashSet();
zoneMapping.put(n.getZoneId(), nodes);
}
nodes.add(n.getId());
}
}
// Check how many zones are down
int zonesDown = 0;
for (Zone zone : cluster.getZones()) {
if (zoneMapping.get(zone.getId()) == null || zoneMapping.get(zone.getId()).size() == 0)
zonesDown++;
}
// Terminate early
if (voldemortConfig.getSlopZonesDownToTerminate() <= zoneMapping.size()
&& zonesDown >= voldemortConfig.getSlopZonesDownToTerminate()) {
logger.info(
"Completed streaming slop pusher job at "
+ startTime
+ " early because "
+ zonesDown
+ " zones are down");
stopAdminClient();
return;
}
}
// Clearing the statistics
AtomicLong attemptedPushes = new AtomicLong(0);
for (Node node : cluster.getNodes()) {
attemptedByNode.put(node.getId(), 0L);
succeededByNode.put(node.getId(), 0L);
}
acquireRepairPermit();
try {
StorageEngine<ByteArray, Slop, byte[]> slopStore = slopStorageEngine.asSlopStore();
iterator = slopStore.entries();
while (iterator.hasNext()) {
Pair<ByteArray, Versioned<Slop>> keyAndVal;
try {
keyAndVal = iterator.next();
Versioned<Slop> versioned = keyAndVal.getSecond();
// Retrieve the node
int nodeId = versioned.getValue().getNodeId();
Node node = cluster.getNodeById(nodeId);
attemptedPushes.incrementAndGet();
Long attempted = attemptedByNode.get(nodeId);
attemptedByNode.put(nodeId, attempted + 1L);
if (attemptedPushes.get() % 10000 == 0)
logger.info("Attempted pushing " + attemptedPushes + " slops");
if (logger.isTraceEnabled())
logger.trace(
"Pushing slop for "
+ versioned.getValue().getNodeId()
+ " and store "
+ versioned.getValue().getStoreName());
if (failureDetector.isAvailable(node)) {
SynchronousQueue<Versioned<Slop>> slopQueue = slopQueues.get(nodeId);
if (slopQueue == null) {
// No previous slop queue, add one
slopQueue = new SynchronousQueue<Versioned<Slop>>();
slopQueues.put(nodeId, slopQueue);
consumerResults.add(
consumerExecutor.submit(new SlopConsumer(nodeId, slopQueue, slopStorageEngine)));
}
boolean offered =
slopQueue.offer(
versioned, voldemortConfig.getClientRoutingTimeoutMs(), TimeUnit.MILLISECONDS);
if (!offered) {
if (logger.isDebugEnabled())
logger.debug(
"No consumer appeared for slop in "
+ voldemortConfig.getClientConnectionTimeoutMs()
+ " ms");
}
readThrottler.maybeThrottle(nBytesRead(keyAndVal));
} else {
logger.trace(node + " declared down, won't push slop");
}
} catch (RejectedExecutionException e) {
throw new VoldemortException("Ran out of threads in executor", e);
}
}
} catch (InterruptedException e) {
logger.warn("Interrupted exception", e);
terminatedEarly = true;
} catch (Exception e) {
logger.error(e, e);
terminatedEarly = true;
} finally {
try {
if (iterator != null) iterator.close();
} catch (Exception e) {
logger.warn("Failed to close iterator cleanly as database might be closed", e);
}
// Adding the poison pill
for (SynchronousQueue<Versioned<Slop>> slopQueue : slopQueues.values()) {
try {
slopQueue.put(END);
} catch (InterruptedException e) {
logger.warn("Error putting poison pill", e);
}
}
for (Future result : consumerResults) {
try {
result.get();
} catch (Exception e) {
logger.warn("Exception in consumer", e);
}
}
// Only if exception didn't take place do we update the counts
if (!terminatedEarly) {
Map<Integer, Long> outstanding =
Maps.newHashMapWithExpectedSize(cluster.getNumberOfNodes());
for (int nodeId : succeededByNode.keySet()) {
logger.info(
"Slops to node "
+ nodeId
+ " - Succeeded - "
+ succeededByNode.get(nodeId)
+ " - Attempted - "
+ attemptedByNode.get(nodeId));
outstanding.put(nodeId, attemptedByNode.get(nodeId) - succeededByNode.get(nodeId));
}
slopStorageEngine.resetStats(outstanding);
logger.info("Completed streaming slop pusher job which started at " + startTime);
} else {
for (int nodeId : succeededByNode.keySet()) {
logger.info(
"Slops to node "
+ nodeId
+ " - Succeeded - "
+ succeededByNode.get(nodeId)
+ " - Attempted - "
+ attemptedByNode.get(nodeId));
}
logger.info("Completed early streaming slop pusher job which started at " + startTime);
}
// Shut down admin client as not to waste connections
consumerResults.clear();
slopQueues.clear();
stopAdminClient();
this.repairPermits.release();
}
}