use of org.apache.asterix.common.transactions.IAppRuntimeContextProvider in project asterixdb by apache.
the class RecoveryManager method startRecoveryRedoPhase.
private synchronized void startRecoveryRedoPhase(Set<Integer> partitions, ILogReader logReader, long lowWaterMarkLSN, Set<Integer> winnerJobSet) throws IOException, ACIDException {
int redoCount = 0;
int jobId = -1;
long resourceId;
long maxDiskLastLsn;
long lsn = -1;
ILSMIndex index = null;
LocalResource localResource = null;
DatasetLocalResource localResourceMetadata = null;
boolean foundWinner = false;
JobEntityCommits jobEntityWinners = null;
IAppRuntimeContextProvider appRuntimeContext = txnSubsystem.getAsterixAppRuntimeContextProvider();
IDatasetLifecycleManager datasetLifecycleManager = appRuntimeContext.getDatasetLifecycleManager();
Map<Long, LocalResource> resourcesMap = localResourceRepository.loadAndGetAllResources();
Map<Long, Long> resourceId2MaxLSNMap = new HashMap<>();
TxnId tempKeyTxnId = new TxnId(-1, -1, -1, null, -1, false);
ILogRecord logRecord = null;
try {
logReader.initializeScan(lowWaterMarkLSN);
logRecord = logReader.next();
while (logRecord != null) {
if (IS_DEBUG_MODE) {
LOGGER.info(logRecord.getLogRecordForDisplay());
}
lsn = logRecord.getLSN();
jobId = logRecord.getJobId();
foundWinner = false;
switch(logRecord.getLogType()) {
case LogType.UPDATE:
if (partitions.contains(logRecord.getResourcePartition())) {
if (winnerJobSet.contains(jobId)) {
foundWinner = true;
} else if (jobId2WinnerEntitiesMap.containsKey(jobId)) {
jobEntityWinners = jobId2WinnerEntitiesMap.get(jobId);
tempKeyTxnId.setTxnId(jobId, logRecord.getDatasetId(), logRecord.getPKHashValue(), logRecord.getPKValue(), logRecord.getPKValueSize());
if (jobEntityWinners.containsEntityCommitForTxnId(lsn, tempKeyTxnId)) {
foundWinner = true;
}
}
if (foundWinner) {
resourceId = logRecord.getResourceId();
localResource = resourcesMap.get(resourceId);
/*******************************************************************
* [Notice]
* -> Issue
* Delete index may cause a problem during redo.
* The index operation to be redone couldn't be redone because the corresponding index
* may not exist in NC due to the possible index drop DDL operation.
* -> Approach
* Avoid the problem during redo.
* More specifically, the problem will be detected when the localResource of
* the corresponding index is retrieved, which will end up with 'null'.
* If null is returned, then just go and process the next
* log record.
*******************************************************************/
if (localResource == null) {
LOGGER.log(Level.WARNING, "resource was not found for resource id " + resourceId);
logRecord = logReader.next();
continue;
}
/*******************************************************************/
//get index instance from IndexLifeCycleManager
//if index is not registered into IndexLifeCycleManager,
//create the index using LocalMetadata stored in LocalResourceRepository
//get partition path in this node
localResourceMetadata = (DatasetLocalResource) localResource.getResource();
index = (ILSMIndex) datasetLifecycleManager.get(localResource.getPath());
if (index == null) {
//#. create index instance and register to indexLifeCycleManager
index = (ILSMIndex) localResourceMetadata.createInstance(serviceCtx);
datasetLifecycleManager.register(localResource.getPath(), index);
datasetLifecycleManager.open(localResource.getPath());
//#. get maxDiskLastLSN
ILSMIndex lsmIndex = index;
try {
maxDiskLastLsn = ((AbstractLSMIOOperationCallback) lsmIndex.getIOOperationCallback()).getComponentLSN(lsmIndex.getImmutableComponents());
} catch (HyracksDataException e) {
datasetLifecycleManager.close(localResource.getPath());
throw e;
}
//#. set resourceId and maxDiskLastLSN to the map
resourceId2MaxLSNMap.put(resourceId, maxDiskLastLsn);
} else {
maxDiskLastLsn = resourceId2MaxLSNMap.get(resourceId);
}
if (lsn > maxDiskLastLsn) {
redo(logRecord, datasetLifecycleManager);
redoCount++;
}
}
}
break;
case LogType.JOB_COMMIT:
case LogType.ENTITY_COMMIT:
case LogType.ABORT:
case LogType.FLUSH:
case LogType.WAIT:
case LogType.MARKER:
//do nothing
break;
default:
throw new ACIDException("Unsupported LogType: " + logRecord.getLogType());
}
logRecord = logReader.next();
}
LOGGER.info("Logs REDO phase completed. Redo logs count: " + redoCount);
} finally {
//close all indexes
Set<Long> resourceIdList = resourceId2MaxLSNMap.keySet();
for (long r : resourceIdList) {
datasetLifecycleManager.close(resourcesMap.get(r).getPath());
}
}
}
use of org.apache.asterix.common.transactions.IAppRuntimeContextProvider in project asterixdb by apache.
the class NCAppRuntimeContext method initialize.
@Override
public void initialize(boolean initialRun) throws IOException, ACIDException {
ioManager = getServiceContext().getIoManager();
threadExecutor = new ThreadExecutor(getServiceContext().getThreadFactory());
fileMapManager = new FileMapManager(ioManager);
ICacheMemoryAllocator allocator = new HeapBufferAllocator();
IPageCleanerPolicy pcp = new DelayPageCleanerPolicy(600000);
IPageReplacementStrategy prs = new ClockPageReplacementStrategy(allocator, storageProperties.getBufferCachePageSize(), storageProperties.getBufferCacheNumPages());
AsynchronousScheduler.INSTANCE.init(getServiceContext().getThreadFactory());
lsmIOScheduler = AsynchronousScheduler.INSTANCE;
metadataMergePolicyFactory = new PrefixMergePolicyFactory();
ILocalResourceRepositoryFactory persistentLocalResourceRepositoryFactory = new PersistentLocalResourceRepositoryFactory(ioManager, getServiceContext().getNodeId(), metadataProperties);
localResourceRepository = (PersistentLocalResourceRepository) persistentLocalResourceRepositoryFactory.createRepository();
IAppRuntimeContextProvider asterixAppRuntimeContextProvider = new AppRuntimeContextProviderForRecovery(this);
txnSubsystem = new TransactionSubsystem(getServiceContext(), getServiceContext().getNodeId(), asterixAppRuntimeContextProvider, txnProperties);
IRecoveryManager recoveryMgr = txnSubsystem.getRecoveryManager();
SystemState systemState = recoveryMgr.getSystemState();
if (initialRun || systemState == SystemState.PERMANENT_DATA_LOSS) {
//delete any storage data before the resource factory is initialized
localResourceRepository.deleteStorageData(true);
}
datasetLifecycleManager = new DatasetLifecycleManager(storageProperties, localResourceRepository, MetadataIndexImmutableProperties.FIRST_AVAILABLE_USER_DATASET_ID, txnSubsystem.getLogManager(), ioManager.getIODevices().size());
isShuttingdown = false;
activeManager = new ActiveManager(threadExecutor, getServiceContext().getNodeId(), activeProperties.getMemoryComponentGlobalBudget(), compilerProperties.getFrameSize());
if (replicationProperties.isParticipant(getServiceContext().getNodeId())) {
String nodeId = getServiceContext().getNodeId();
replicaResourcesManager = new ReplicaResourcesManager(localResourceRepository, metadataProperties);
replicationManager = new ReplicationManager(nodeId, replicationProperties, replicaResourcesManager, txnSubsystem.getLogManager(), asterixAppRuntimeContextProvider);
//pass replication manager to replication required object
//LogManager to replicate logs
txnSubsystem.getLogManager().setReplicationManager(replicationManager);
//PersistentLocalResourceRepository to replicate metadata files and delete backups on drop index
localResourceRepository.setReplicationManager(replicationManager);
/*
* add the partitions that will be replicated in this node as inactive partitions
*/
//get nodes which replicate to this node
Set<String> remotePrimaryReplicas = replicationProperties.getRemotePrimaryReplicasIds(nodeId);
for (String clientId : remotePrimaryReplicas) {
//get the partitions of each client
ClusterPartition[] clientPartitions = metadataProperties.getNodePartitions().get(clientId);
for (ClusterPartition partition : clientPartitions) {
localResourceRepository.addInactivePartition(partition.getPartitionId());
}
}
//initialize replication channel
replicationChannel = new ReplicationChannel(nodeId, replicationProperties, txnSubsystem.getLogManager(), replicaResourcesManager, replicationManager, getServiceContext(), asterixAppRuntimeContextProvider);
remoteRecoveryManager = new RemoteRecoveryManager(replicationManager, this, replicationProperties);
bufferCache = new BufferCache(ioManager, prs, pcp, fileMapManager, storageProperties.getBufferCacheMaxOpenFiles(), getServiceContext().getThreadFactory(), replicationManager);
} else {
bufferCache = new BufferCache(ioManager, prs, pcp, fileMapManager, storageProperties.getBufferCacheMaxOpenFiles(), getServiceContext().getThreadFactory());
}
/*
* The order of registration is important. The buffer cache must registered before recovery and transaction
* managers. Notes: registered components are stopped in reversed order
*/
ILifeCycleComponentManager lccm = getServiceContext().getLifeCycleComponentManager();
lccm.register((ILifeCycleComponent) bufferCache);
/*
* LogManager must be stopped after RecoveryManager, DatasetLifeCycleManager, and ReplicationManager
* to process any logs that might be generated during stopping these components
*/
lccm.register((ILifeCycleComponent) txnSubsystem.getLogManager());
/*
* ReplicationManager must be stopped after indexLifecycleManager and recovery manager
* so that any logs/files generated during closing datasets or checkpoints are sent to remote replicas
*/
if (replicationManager != null) {
lccm.register(replicationManager);
}
lccm.register((ILifeCycleComponent) txnSubsystem.getRecoveryManager());
/*
* Stopping indexLifecycleManager will flush and close all datasets.
*/
lccm.register((ILifeCycleComponent) datasetLifecycleManager);
lccm.register((ILifeCycleComponent) txnSubsystem.getTransactionManager());
lccm.register((ILifeCycleComponent) txnSubsystem.getLockManager());
lccm.register(txnSubsystem.getCheckpointManager());
}
Aggregations