X-Git-Url: https://git.opendaylight.org/gerrit/gitweb?p=controller.git;a=blobdiff_plain;f=opendaylight%2Fmd-sal%2Fsal-distributed-datastore%2Fsrc%2Fmain%2Fjava%2Forg%2Fopendaylight%2Fcontroller%2Fcluster%2Fdatastore%2FShard.java;h=a30b6f7516981411e589577f3fda9a9a0f9bc887;hp=9cd758ba30fdb94e85cd1703d99a8e0c55a50a17;hb=4349b034606957d3e876b82b14a292e6739a986a;hpb=2eac38c80e07108767d43609eff41881d9d6f361 diff --git a/opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/Shard.java b/opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/Shard.java index 9cd758ba30..a30b6f7516 100644 --- a/opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/Shard.java +++ b/opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/Shard.java @@ -12,8 +12,6 @@ import akka.actor.ActorRef; import akka.actor.ActorSelection; import akka.actor.Cancellable; import akka.actor.Props; -import akka.event.Logging; -import akka.event.LoggingAdapter; import akka.japi.Creator; import akka.persistence.RecoveryFailure; import akka.serialization.Serialization; @@ -45,6 +43,8 @@ import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardStats import org.opendaylight.controller.cluster.datastore.messages.AbortTransaction; import org.opendaylight.controller.cluster.datastore.messages.AbortTransactionReply; import org.opendaylight.controller.cluster.datastore.messages.ActorInitialized; +import org.opendaylight.controller.cluster.datastore.messages.BatchedModifications; +import org.opendaylight.controller.cluster.datastore.messages.BatchedModificationsReply; import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransaction; import org.opendaylight.controller.cluster.datastore.messages.CloseTransactionChain; import org.opendaylight.controller.cluster.datastore.messages.CommitTransaction; @@ -62,10 +62,14 @@ import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContex import org.opendaylight.controller.cluster.datastore.modification.Modification; import org.opendaylight.controller.cluster.datastore.modification.ModificationPayload; import org.opendaylight.controller.cluster.datastore.modification.MutableCompositeModification; +import org.opendaylight.controller.cluster.datastore.utils.Dispatchers; +import org.opendaylight.controller.cluster.datastore.utils.MessageTracker; import org.opendaylight.controller.cluster.datastore.utils.SerializationUtils; +import org.opendaylight.controller.cluster.notifications.RegisterRoleChangeListener; import org.opendaylight.controller.cluster.notifications.RoleChangeNotifier; import org.opendaylight.controller.cluster.raft.RaftActor; -import org.opendaylight.controller.cluster.raft.ReplicatedLogEntry; +import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus; +import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply; import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationByteStringPayload; import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationPayload; import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload; @@ -73,8 +77,7 @@ import org.opendaylight.controller.md.sal.common.api.data.AsyncDataChangeListene import org.opendaylight.controller.md.sal.dom.store.impl.InMemoryDOMDataStore; import org.opendaylight.controller.md.sal.dom.store.impl.InMemoryDOMDataStoreFactory; import org.opendaylight.controller.sal.core.spi.data.DOMStoreThreePhaseCommitCohort; -import org.opendaylight.controller.sal.core.spi.data.DOMStoreTransactionChain; -import org.opendaylight.controller.sal.core.spi.data.DOMStoreTransactionFactory; +import org.opendaylight.controller.sal.core.spi.data.DOMStoreTransaction; import org.opendaylight.controller.sal.core.spi.data.DOMStoreWriteTransaction; import org.opendaylight.yangtools.concepts.ListenerRegistration; import org.opendaylight.yangtools.yang.data.api.YangInstanceIdentifier; @@ -101,11 +104,8 @@ public class Shard extends RaftActor { // The state of this Shard private final InMemoryDOMDataStore store; - private final LoggingAdapter LOG = - Logging.getLogger(getContext().system(), this); - /// The name of this shard - private final ShardIdentifier name; + private final String name; private final ShardStats shardMBean; @@ -114,9 +114,9 @@ public class Shard extends RaftActor { private final List delayedListenerRegistrations = Lists.newArrayList(); - private final DatastoreContext datastoreContext; + private DatastoreContext datastoreContext; - private final DataPersistenceProvider dataPersistenceProvider; + private DataPersistenceProvider dataPersistenceProvider; private SchemaContext schemaContext; @@ -124,31 +124,42 @@ public class Shard extends RaftActor { private final ShardCommitCoordinator commitCoordinator; - private final long transactionCommitTimeout; + private long transactionCommitTimeout; private Cancellable txCommitTimeoutCheckSchedule; private final Optional roleChangeNotifier; + private final MessageTracker appendEntriesReplyTracker; + + private final ReadyTransactionReply READY_TRANSACTION_REPLY = new ReadyTransactionReply( + Serialization.serializedActorPath(getSelf())); + + /** * Coordinates persistence recovery on startup. */ private ShardRecoveryCoordinator recoveryCoordinator; private List currentLogRecoveryBatch; - private final Map transactionChains = new HashMap<>(); + private final DOMTransactionFactory transactionFactory; + + private final String txnDispatcherPath; - protected Shard(final ShardIdentifier name, final Map peerAddresses, + protected Shard(final ShardIdentifier name, final Map peerAddresses, final DatastoreContext datastoreContext, final SchemaContext schemaContext) { - super(name.toString(), mapPeerAddresses(peerAddresses), - Optional.of(datastoreContext.getShardRaftConfig())); + super(name.toString(), new HashMap<>(peerAddresses), Optional.of(datastoreContext.getShardRaftConfig())); - this.name = name; + this.name = name.toString(); this.datastoreContext = datastoreContext; this.schemaContext = schemaContext; - this.dataPersistenceProvider = (datastoreContext.isPersistent()) ? new PersistentDataProvider() : new NonPersistentRaftDataProvider(); + this.dataPersistenceProvider = (datastoreContext.isPersistent()) + ? new PersistentDataProvider() : new NonPersistentRaftDataProvider(); + this.txnDispatcherPath = new Dispatchers(context().system().dispatchers()) + .getDispatcherPath(Dispatchers.DispatcherType.Transaction); + - LOG.info("Shard created : {} persistent : {}", name, datastoreContext.isPersistent()); + LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent()); store = InMemoryDOMDataStoreFactory.create(name.toString(), null, datastoreContext.getDataStoreProperties()); @@ -160,35 +171,34 @@ public class Shard extends RaftActor { shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(), datastoreContext.getDataStoreMXBeanType()); shardMBean.setNotificationManager(store.getDataChangeListenerNotificationManager()); + shardMBean.setShardActor(getSelf()); if (isMetricsCaptureEnabled()) { getContext().become(new MeteringBehavior(this)); } - commitCoordinator = new ShardCommitCoordinator(TimeUnit.SECONDS.convert(1, TimeUnit.MINUTES), - datastoreContext.getShardTransactionCommitQueueCapacity()); + transactionFactory = new DOMTransactionFactory(store, shardMBean, LOG, this.name); - transactionCommitTimeout = TimeUnit.MILLISECONDS.convert( - datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS); + commitCoordinator = new ShardCommitCoordinator(transactionFactory, + TimeUnit.SECONDS.convert(5, TimeUnit.MINUTES), + datastoreContext.getShardTransactionCommitQueueCapacity(), self(), LOG, this.name); + + setTransactionCommitTimeout(); // create a notifier actor for each cluster member roleChangeNotifier = createRoleChangeNotifier(name.toString()); - } - - private static Map mapPeerAddresses( - final Map peerAddresses) { - Map map = new HashMap<>(); - for (Map.Entry entry : peerAddresses - .entrySet()) { - map.put(entry.getKey().toString(), entry.getValue()); - } + appendEntriesReplyTracker = new MessageTracker(AppendEntriesReply.class, + getRaftActorContext().getConfigParams().getIsolatedCheckIntervalInMillis()); + } - return map; + private void setTransactionCommitTimeout() { + transactionCommitTimeout = TimeUnit.MILLISECONDS.convert( + datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS); } public static Props props(final ShardIdentifier name, - final Map peerAddresses, + final Map peerAddresses, final DatastoreContext datastoreContext, final SchemaContext schemaContext) { Preconditions.checkNotNull(name, "name should not be null"); Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null"); @@ -201,67 +211,91 @@ public class Shard extends RaftActor { private Optional createRoleChangeNotifier(String shardId) { ActorRef shardRoleChangeNotifier = this.getContext().actorOf( RoleChangeNotifier.getProps(shardId), shardId + "-notifier"); - return Optional.of(shardRoleChangeNotifier); + return Optional.of(shardRoleChangeNotifier); } @Override public void postStop() { + LOG.info("Stopping Shard {}", persistenceId()); + super.postStop(); if(txCommitTimeoutCheckSchedule != null) { txCommitTimeoutCheckSchedule.cancel(); } + + shardMBean.unregisterMBean(); } @Override public void onReceiveRecover(final Object message) throws Exception { if(LOG.isDebugEnabled()) { - LOG.debug("onReceiveRecover: Received message {} from {}", - message.getClass().toString(), - getSender()); + LOG.debug("{}: onReceiveRecover: Received message {} from {}", persistenceId(), + message.getClass().toString(), getSender()); } if (message instanceof RecoveryFailure){ - LOG.error(((RecoveryFailure) message).cause(), "Recovery failed because of this cause"); + LOG.error("{}: Recovery failed because of this cause", + persistenceId(), ((RecoveryFailure) message).cause()); // Even though recovery failed, we still need to finish our recovery, eg send the // ActorInitialized message and start the txCommitTimeoutCheckSchedule. onRecoveryComplete(); } else { super.onReceiveRecover(message); + if(LOG.isTraceEnabled()) { + appendEntriesReplyTracker.begin(); + } } } @Override public void onReceiveCommand(final Object message) throws Exception { - if(LOG.isDebugEnabled()) { - LOG.debug("onReceiveCommand: Received message {} from {}", message, getSender()); - } - - if (message.getClass().equals(CreateTransaction.SERIALIZABLE_CLASS)) { - handleCreateTransaction(message); - } else if(message instanceof ForwardedReadyTransaction) { - handleForwardedReadyTransaction((ForwardedReadyTransaction)message); - } else if(message.getClass().equals(CanCommitTransaction.SERIALIZABLE_CLASS)) { - handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message)); - } else if(message.getClass().equals(CommitTransaction.SERIALIZABLE_CLASS)) { - handleCommitTransaction(CommitTransaction.fromSerializable(message)); - } else if(message.getClass().equals(AbortTransaction.SERIALIZABLE_CLASS)) { - handleAbortTransaction(AbortTransaction.fromSerializable(message)); - } else if (message.getClass().equals(CloseTransactionChain.SERIALIZABLE_CLASS)){ - closeTransactionChain(CloseTransactionChain.fromSerializable(message)); - } else if (message instanceof RegisterChangeListener) { - registerChangeListener((RegisterChangeListener) message); - } else if (message instanceof UpdateSchemaContext) { - updateSchemaContext((UpdateSchemaContext) message); - } else if (message instanceof PeerAddressResolved) { - PeerAddressResolved resolved = (PeerAddressResolved) message; - setPeerAddress(resolved.getPeerId().toString(), - resolved.getPeerAddress()); - } else if(message.equals(TX_COMMIT_TIMEOUT_CHECK_MESSAGE)) { - handleTransactionCommitTimeoutCheck(); - } else { - super.onReceiveCommand(message); + + MessageTracker.Context context = appendEntriesReplyTracker.received(message); + + if(context.error().isPresent()){ + LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(), + context.error()); + } + + try { + if (CreateTransaction.SERIALIZABLE_CLASS.isInstance(message)) { + handleCreateTransaction(message); + } else if (BatchedModifications.class.isInstance(message)) { + handleBatchedModifications((BatchedModifications)message); + } else if (message instanceof ForwardedReadyTransaction) { + handleForwardedReadyTransaction((ForwardedReadyTransaction) message); + } else if (CanCommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) { + handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message)); + } else if (CommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) { + handleCommitTransaction(CommitTransaction.fromSerializable(message)); + } else if (AbortTransaction.SERIALIZABLE_CLASS.isInstance(message)) { + handleAbortTransaction(AbortTransaction.fromSerializable(message)); + } else if (CloseTransactionChain.SERIALIZABLE_CLASS.isInstance(message)) { + closeTransactionChain(CloseTransactionChain.fromSerializable(message)); + } else if (message instanceof RegisterChangeListener) { + registerChangeListener((RegisterChangeListener) message); + } else if (message instanceof UpdateSchemaContext) { + updateSchemaContext((UpdateSchemaContext) message); + } else if (message instanceof PeerAddressResolved) { + PeerAddressResolved resolved = (PeerAddressResolved) message; + setPeerAddress(resolved.getPeerId().toString(), + resolved.getPeerAddress()); + } else if (message.equals(TX_COMMIT_TIMEOUT_CHECK_MESSAGE)) { + handleTransactionCommitTimeoutCheck(); + } else if(message instanceof DatastoreContext) { + onDatastoreContext((DatastoreContext)message); + } else if(message instanceof RegisterRoleChangeListener){ + roleChangeNotifier.get().forward(message, context()); + } else if (message instanceof FollowerInitialSyncUpStatus){ + shardMBean.setFollowerInitialSyncStatus(((FollowerInitialSyncUpStatus) message).isInitialSyncDone()); + context().parent().tell(message, self()); + } else { + super.onReceiveCommand(message); + } + } finally { + context.done(); } } @@ -270,13 +304,31 @@ public class Shard extends RaftActor { return roleChangeNotifier; } + private void onDatastoreContext(DatastoreContext context) { + datastoreContext = context; + + commitCoordinator.setQueueCapacity(datastoreContext.getShardTransactionCommitQueueCapacity()); + + setTransactionCommitTimeout(); + + if(datastoreContext.isPersistent() && + dataPersistenceProvider instanceof NonPersistentRaftDataProvider) { + dataPersistenceProvider = new PersistentDataProvider(); + } else if(!datastoreContext.isPersistent() && + dataPersistenceProvider instanceof PersistentDataProvider) { + dataPersistenceProvider = new NonPersistentRaftDataProvider(); + } + + updateConfigParams(datastoreContext.getShardRaftConfig()); + } + private void handleTransactionCommitTimeoutCheck() { CohortEntry cohortEntry = commitCoordinator.getCurrentCohortEntry(); if(cohortEntry != null) { long elapsed = System.currentTimeMillis() - cohortEntry.getLastAccessTime(); if(elapsed > transactionCommitTimeout) { - LOG.warning("Current transaction {} has timed out after {} ms - aborting", - cohortEntry.getTransactionID(), transactionCommitTimeout); + LOG.warn("{}: Current transaction {} has timed out after {} ms - aborting", + persistenceId(), cohortEntry.getTransactionID(), transactionCommitTimeout); doAbortTransaction(cohortEntry.getTransactionID(), null); } @@ -286,7 +338,7 @@ public class Shard extends RaftActor { private void handleCommitTransaction(final CommitTransaction commit) { final String transactionID = commit.getTransactionID(); - LOG.debug("Committing transaction {}", transactionID); + LOG.debug("{}: Committing transaction {}", persistenceId(), transactionID); // Get the current in-progress cohort entry in the commitCoordinator if it corresponds to // this transaction. @@ -295,8 +347,8 @@ public class Shard extends RaftActor { // We're not the current Tx - the Tx was likely expired b/c it took too long in // between the canCommit and commit messages. IllegalStateException ex = new IllegalStateException( - String.format("Cannot commit transaction %s - it is not the current transaction", - transactionID)); + String.format("%s: Cannot commit transaction %s - it is not the current transaction", + persistenceId(), transactionID)); LOG.error(ex.getMessage()); shardMBean.incrementFailedTransactionsCount(); getSender().tell(new akka.actor.Status.Failure(ex), getSelf()); @@ -314,17 +366,18 @@ public class Shard extends RaftActor { // currently uses a same thread executor anyway. cohortEntry.getCohort().preCommit().get(); - // If we do not have any followers and we are not using persistence we can - // apply modification to the state immediately - if(!hasFollowers() && !persistence().isRecoveryApplicable()){ + // If we do not have any followers and we are not using persistence + // or if cohortEntry has no modifications + // we can apply modification to the state immediately + if((!hasFollowers() && !persistence().isRecoveryApplicable()) || (!cohortEntry.hasModifications())){ applyModificationToState(getSender(), transactionID, cohortEntry.getModification()); } else { Shard.this.persistData(getSender(), transactionID, new ModificationPayload(cohortEntry.getModification())); } - } catch (InterruptedException | ExecutionException | IOException e) { - LOG.error(e, "An exception occurred while preCommitting transaction {}", - cohortEntry.getTransactionID()); + } catch (Exception e) { + LOG.error("{} An exception occurred while preCommitting transaction {}", + persistenceId(), cohortEntry.getTransactionID(), e); shardMBean.incrementFailedTransactionsCount(); getSender().tell(new akka.actor.Status.Failure(e), getSelf()); } @@ -352,8 +405,8 @@ public class Shard extends RaftActor { // This really shouldn't happen - it likely means that persistence or replication // took so long to complete such that the cohort entry was expired from the cache. IllegalStateException ex = new IllegalStateException( - String.format("Could not finish committing transaction %s - no CohortEntry found", - transactionID)); + String.format("%s: Could not finish committing transaction %s - no CohortEntry found", + persistenceId(), transactionID)); LOG.error(ex.getMessage()); sender.tell(new akka.actor.Status.Failure(ex), getSelf()); } @@ -361,7 +414,7 @@ public class Shard extends RaftActor { return; } - LOG.debug("Finishing commit for transaction {}", cohortEntry.getTransactionID()); + LOG.debug("{}: Finishing commit for transaction {}", persistenceId(), cohortEntry.getTransactionID()); try { // We block on the future here so we don't have to worry about possibly accessing our @@ -374,47 +427,93 @@ public class Shard extends RaftActor { shardMBean.incrementCommittedTransactionCount(); shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis()); - } catch (InterruptedException | ExecutionException e) { + } catch (Exception e) { sender.tell(new akka.actor.Status.Failure(e), getSelf()); - LOG.error(e, "An exception occurred while committing transaction {}", transactionID); + LOG.error("{}, An exception occurred while committing transaction {}", persistenceId(), + transactionID, e); shardMBean.incrementFailedTransactionsCount(); + } finally { + commitCoordinator.currentTransactionComplete(transactionID, true); } - - commitCoordinator.currentTransactionComplete(transactionID, true); } private void handleCanCommitTransaction(final CanCommitTransaction canCommit) { - LOG.debug("Can committing transaction {}", canCommit.getTransactionID()); + LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionID()); commitCoordinator.handleCanCommit(canCommit, getSender(), self()); } + private void handleBatchedModifications(BatchedModifications batched) { + // This message is sent to prepare the modificationsa transaction directly on the Shard as an + // optimization to avoid the extra overhead of a separate ShardTransaction actor. On the last + // BatchedModifications message, the caller sets the ready flag in the message indicating + // modifications are complete. The reply contains the cohort actor path (this actor) for the caller + // to initiate the 3-phase commit. This also avoids the overhead of sending an additional + // ReadyTransaction message. + + // If we're not the leader then forward to the leader. This is a safety measure - we shouldn't + // normally get here if we're not the leader as the front-end (TransactionProxy) should determine + // the primary/leader shard. However with timing and caching on the front-end, there's a small + // window where it could have a stale leader during leadership transitions. + // + if(isLeader()) { + try { + BatchedModificationsReply reply = commitCoordinator.handleTransactionModifications(batched); + sender().tell(reply, self()); + } catch (Exception e) { + LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(), + batched.getTransactionID(), e); + getSender().tell(new akka.actor.Status.Failure(e), getSelf()); + } + } else { + ActorSelection leader = getLeader(); + if(leader != null) { + // TODO: what if this is not the first batch and leadership changed in between batched messages? + // We could check if the commitCoordinator already has a cached entry and forward all the previous + // batched modifications. + LOG.debug("{}: Forwarding BatchedModifications to leader {}", persistenceId(), leader); + leader.forward(batched, getContext()); + } else { + // TODO: rather than throwing an immediate exception, we could schedule a timer to try again to make + // it more resilient in case we're in the process of electing a new leader. + getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format( + "Could not find the leader for shard %s. This typically happens" + + " when the system is coming up or recovering and a leader is being elected. Try again" + + " later.", persistenceId()))), getSelf()); + } + } + } + private void handleForwardedReadyTransaction(ForwardedReadyTransaction ready) { - LOG.debug("Readying transaction {}, client version {}", ready.getTransactionID(), - ready.getTxnClientVersion()); + LOG.debug("{}: Readying transaction {}, client version {}", persistenceId(), + ready.getTransactionID(), ready.getTxnClientVersion()); // This message is forwarded by the ShardTransaction on ready. We cache the cohort in the // commitCoordinator in preparation for the subsequent three phase commit initiated by // the front-end. commitCoordinator.transactionReady(ready.getTransactionID(), ready.getCohort(), - ready.getModification()); + (MutableCompositeModification) ready.getModification()); // Return our actor path as we'll handle the three phase commit, except if the Tx client // version < 1 (Helium-1 version). This means the Tx was initiated by a base Helium version // node. In that case, the subsequent 3-phase commit messages won't contain the // transactionId so to maintain backwards compatibility, we create a separate cohort actor // to provide the compatible behavior. - ActorRef replyActorPath = self(); if(ready.getTxnClientVersion() < DataStoreVersions.HELIUM_1_VERSION) { - LOG.debug("Creating BackwardsCompatibleThreePhaseCommitCohort"); - replyActorPath = getContext().actorOf(BackwardsCompatibleThreePhaseCommitCohort.props( + LOG.debug("{}: Creating BackwardsCompatibleThreePhaseCommitCohort", persistenceId()); + ActorRef replyActorPath = getContext().actorOf(BackwardsCompatibleThreePhaseCommitCohort.props( ready.getTransactionID())); - } - ReadyTransactionReply readyTransactionReply = new ReadyTransactionReply( - Serialization.serializedActorPath(replyActorPath)); - getSender().tell(ready.isReturnSerialized() ? readyTransactionReply.toSerializable() : - readyTransactionReply, getSelf()); + ReadyTransactionReply readyTransactionReply = + new ReadyTransactionReply(Serialization.serializedActorPath(replyActorPath)); + getSender().tell(ready.isReturnSerialized() ? readyTransactionReply.toSerializable() : + readyTransactionReply, getSelf()); + + } else { + + getSender().tell(ready.isReturnSerialized() ? READY_TRANSACTION_REPLY.toSerializable() : + READY_TRANSACTION_REPLY, getSelf()); + } } private void handleAbortTransaction(final AbortTransaction abort) { @@ -424,7 +523,7 @@ public class Shard extends RaftActor { void doAbortTransaction(final String transactionID, final ActorRef sender) { final CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID); if(cohortEntry != null) { - LOG.debug("Aborting transaction {}", transactionID); + LOG.debug("{}: Aborting transaction {}", persistenceId(), transactionID); // We don't remove the cached cohort entry here (ie pass false) in case the Tx was // aborted during replication in which case we may still commit locally if replication @@ -446,7 +545,7 @@ public class Shard extends RaftActor { @Override public void onFailure(final Throwable t) { - LOG.error(t, "An exception happened during abort"); + LOG.error("{}: An exception happened during abort", persistenceId(), t); if(sender != null) { sender.tell(new akka.actor.Status.Failure(t), self); @@ -462,76 +561,37 @@ public class Shard extends RaftActor { } else if (getLeader() != null) { getLeader().forward(message, getContext()); } else { - getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException( - "Could not find shard leader so transaction cannot be created. This typically happens" + + getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format( + "Could not find leader for shard %s so transaction cannot be created. This typically happens" + " when the system is coming up or recovering and a leader is being elected. Try again" + - " later.")), getSelf()); + " later.", persistenceId()))), getSelf()); } } private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) { - DOMStoreTransactionChain chain = - transactionChains.remove(closeTransactionChain.getTransactionChainId()); - - if(chain != null) { - chain.close(); - } + transactionFactory.closeTransactionChain(closeTransactionChain.getTransactionChainId()); } private ActorRef createTypedTransactionActor(int transactionType, ShardTransactionIdentifier transactionId, String transactionChainId, short clientVersion ) { - DOMStoreTransactionFactory factory = store; + DOMStoreTransaction transaction = transactionFactory.newTransaction( + TransactionProxy.TransactionType.fromInt(transactionType), transactionId.toString(), + transactionChainId); - if(!transactionChainId.isEmpty()) { - factory = transactionChains.get(transactionChainId); - if(factory == null){ - DOMStoreTransactionChain transactionChain = store.createTransactionChain(); - transactionChains.put(transactionChainId, transactionChain); - factory = transactionChain; - } - } - - if(this.schemaContext == null) { - throw new IllegalStateException("SchemaContext is not set"); - } - - if (transactionType == TransactionProxy.TransactionType.READ_ONLY.ordinal()) { - - shardMBean.incrementReadOnlyTransactionCount(); - - return getContext().actorOf( - ShardTransaction.props(factory.newReadOnlyTransaction(), getSelf(), - schemaContext,datastoreContext, shardMBean, - transactionId.getRemoteTransactionId(), clientVersion), - transactionId.toString()); - - } else if (transactionType == TransactionProxy.TransactionType.READ_WRITE.ordinal()) { - - shardMBean.incrementReadWriteTransactionCount(); + return createShardTransaction(transaction, transactionId, clientVersion); + } - return getContext().actorOf( - ShardTransaction.props(factory.newReadWriteTransaction(), getSelf(), + private ActorRef createShardTransaction(DOMStoreTransaction transaction, ShardTransactionIdentifier transactionId, + short clientVersion){ + return getContext().actorOf( + ShardTransaction.props(transaction, getSelf(), schemaContext, datastoreContext, shardMBean, - transactionId.getRemoteTransactionId(), clientVersion), - transactionId.toString()); - + transactionId.getRemoteTransactionId(), clientVersion) + .withDispatcher(txnDispatcherPath), + transactionId.toString()); - } else if (transactionType == TransactionProxy.TransactionType.WRITE_ONLY.ordinal()) { - - shardMBean.incrementWriteOnlyTransactionCount(); - - return getContext().actorOf( - ShardTransaction.props(factory.newWriteOnlyTransaction(), getSelf(), - schemaContext, datastoreContext, shardMBean, - transactionId.getRemoteTransactionId(), clientVersion), - transactionId.toString()); - } else { - throw new IllegalArgumentException( - "Shard="+name + ":CreateTransaction message has unidentified transaction type=" - + transactionType); - } } private void createTransaction(CreateTransaction createTransaction) { @@ -550,13 +610,11 @@ public class Shard extends RaftActor { private ActorRef createTransaction(int transactionType, String remoteTransactionId, String transactionChainId, short clientVersion) { - ShardTransactionIdentifier transactionId = - ShardTransactionIdentifier.builder() - .remoteTransactionId(remoteTransactionId) - .build(); + + ShardTransactionIdentifier transactionId = new ShardTransactionIdentifier(remoteTransactionId); if(LOG.isDebugEnabled()) { - LOG.debug("Creating transaction : {} ", transactionId); + LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId); } ActorRef transactionActor = createTypedTransactionActor(transactionType, transactionId, @@ -581,7 +639,7 @@ public class Shard extends RaftActor { shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis()); } catch (InterruptedException | ExecutionException e) { shardMBean.incrementFailedTransactionsCount(); - LOG.error(e, "Failed to commit"); + LOG.error("{}: Failed to commit", persistenceId(), e); } } @@ -598,14 +656,14 @@ public class Shard extends RaftActor { private void registerChangeListener(final RegisterChangeListener registerChangeListener) { - LOG.debug("registerDataChangeListener for {}", registerChangeListener.getPath()); + LOG.debug("{}: registerDataChangeListener for {}", persistenceId(), registerChangeListener.getPath()); ListenerRegistration>> registration; if(isLeader()) { registration = doChangeListenerRegistration(registerChangeListener); } else { - LOG.debug("Shard is not the leader - delaying registration"); + LOG.debug("{}: Shard is not the leader - delaying registration", persistenceId()); DelayedListenerRegistration delayedReg = new DelayedListenerRegistration(registerChangeListener); @@ -616,10 +674,10 @@ public class Shard extends RaftActor { ActorRef listenerRegistration = getContext().actorOf( DataChangeListenerRegistration.props(registration)); - LOG.debug("registerDataChangeListener sending reply, listenerRegistrationPath = {} ", - listenerRegistration.path()); + LOG.debug("{}: registerDataChangeListener sending reply, listenerRegistrationPath = {} ", + persistenceId(), listenerRegistration.path()); - getSender().tell(new RegisterChangeListenerReply(listenerRegistration.path()), getSelf()); + getSender().tell(new RegisterChangeListenerReply(listenerRegistration), getSelf()); } private ListenerRegistration> listener = new DataChangeListenerProxy(dataChangeListenerPath); - LOG.debug("Registering for path {}", registerChangeListener.getPath()); + LOG.debug("{}: Registering for path {}", persistenceId(), registerChangeListener.getPath()); return store.registerChangeListener(registerChangeListener.getPath(), listener, registerChangeListener.getScope()); @@ -658,7 +716,7 @@ public class Shard extends RaftActor { currentLogRecoveryBatch = Lists.newArrayListWithCapacity(maxBatchSize); if(LOG.isDebugEnabled()) { - LOG.debug("{} : starting log recovery batch with max size {}", persistenceId(), maxBatchSize); + LOG.debug("{}: starting log recovery batch with max size {}", persistenceId(), maxBatchSize); } } @@ -668,40 +726,42 @@ public class Shard extends RaftActor { try { currentLogRecoveryBatch.add(((ModificationPayload) data).getModification()); } catch (ClassNotFoundException | IOException e) { - LOG.error(e, "Error extracting ModificationPayload"); + LOG.error("{}: Error extracting ModificationPayload", persistenceId(), e); } } else if (data instanceof CompositeModificationPayload) { currentLogRecoveryBatch.add(((CompositeModificationPayload) data).getModification()); } else if (data instanceof CompositeModificationByteStringPayload) { currentLogRecoveryBatch.add(((CompositeModificationByteStringPayload) data).getModification()); } else { - LOG.error("Unknown state received {} during recovery", data); + LOG.error("{}: Unknown state received {} during recovery", persistenceId(), data); } } @Override protected void applyRecoverySnapshot(final byte[] snapshotBytes) { if(recoveryCoordinator == null) { - recoveryCoordinator = new ShardRecoveryCoordinator(persistenceId(), schemaContext); + recoveryCoordinator = new ShardRecoveryCoordinator(persistenceId(), schemaContext, + LOG, name.toString()); } recoveryCoordinator.submit(snapshotBytes, store.newWriteOnlyTransaction()); if(LOG.isDebugEnabled()) { - LOG.debug("{} : submitted recovery sbapshot", persistenceId()); + LOG.debug("{}: submitted recovery sbapshot", persistenceId()); } } @Override protected void applyCurrentLogRecoveryBatch() { if(recoveryCoordinator == null) { - recoveryCoordinator = new ShardRecoveryCoordinator(persistenceId(), schemaContext); + recoveryCoordinator = new ShardRecoveryCoordinator(persistenceId(), schemaContext, + LOG, name.toString()); } recoveryCoordinator.submit(currentLogRecoveryBatch, store.newWriteOnlyTransaction()); if(LOG.isDebugEnabled()) { - LOG.debug("{} : submitted log recovery batch with size {}", persistenceId(), + LOG.debug("{}: submitted log recovery batch with size {}", persistenceId(), currentLogRecoveryBatch.size()); } } @@ -712,7 +772,7 @@ public class Shard extends RaftActor { Collection txList = recoveryCoordinator.getTransactions(); if(LOG.isDebugEnabled()) { - LOG.debug("{} : recovery complete - committing {} Tx's", persistenceId(), txList.size()); + LOG.debug("{}: recovery complete - committing {} Tx's", persistenceId(), txList.size()); } for(DOMStoreWriteTransaction tx: txList) { @@ -721,14 +781,13 @@ public class Shard extends RaftActor { shardMBean.incrementCommittedTransactionCount(); } catch (InterruptedException | ExecutionException e) { shardMBean.incrementFailedTransactionsCount(); - LOG.error(e, "Failed to commit"); + LOG.error("{}: Failed to commit", persistenceId(), e); } } } recoveryCoordinator = null; currentLogRecoveryBatch = null; - updateJournalStats(); //notify shard manager getContext().parent().tell(new ActorInitialized(), getSelf()); @@ -751,7 +810,7 @@ public class Shard extends RaftActor { try { applyModificationToState(clientActor, identifier, ((ModificationPayload) data).getModification()); } catch (ClassNotFoundException | IOException e) { - LOG.error(e, "Error extracting ModificationPayload"); + LOG.error("{}: Error extracting ModificationPayload", persistenceId(), e); } } else if (data instanceof CompositeModificationPayload) { @@ -763,20 +822,17 @@ public class Shard extends RaftActor { applyModificationToState(clientActor, identifier, modification); } else { - LOG.error("Unknown state received {} Class loader = {} CompositeNodeMod.ClassLoader = {}", - data, data.getClass().getClassLoader(), + LOG.error("{}: Unknown state received {} Class loader = {} CompositeNodeMod.ClassLoader = {}", + persistenceId(), data, data.getClass().getClassLoader(), CompositeModificationPayload.class.getClassLoader()); } - - updateJournalStats(); - } private void applyModificationToState(ActorRef clientActor, String identifier, Object modification) { if(modification == null) { LOG.error( - "modification is null - this is very unexpected, clientActor = {}, identifier = {}", - identifier, clientActor != null ? clientActor.path().toString() : null); + "{}: modification is null - this is very unexpected, clientActor = {}, identifier = {}", + persistenceId(), identifier, clientActor != null ? clientActor.path().toString() : null); } else if(clientActor == null) { // There's no clientActor to which to send a commit reply so we must be applying // replicated state from the leader. @@ -787,19 +843,6 @@ public class Shard extends RaftActor { } } - private void updateJournalStats() { - ReplicatedLogEntry lastLogEntry = getLastLogEntry(); - - if (lastLogEntry != null) { - shardMBean.setLastLogIndex(lastLogEntry.getIndex()); - shardMBean.setLastLogTerm(lastLogEntry.getTerm()); - } - - shardMBean.setCommitIndex(getCommitIndex()); - shardMBean.setLastApplied(getLastApplied()); - shardMBean.setInMemoryJournalDataSize(getRaftActorContext().getReplicatedLog().dataSize()); - } - @Override protected void createSnapshot() { // Create a transaction actor. We are really going to treat the transaction as a worker @@ -821,7 +864,7 @@ public class Shard extends RaftActor { // we can safely commit everything in here. We not need to worry about event notifications // as they would have already been disabled on the follower - LOG.info("Applying snapshot"); + LOG.info("{}: Applying snapshot", persistenceId()); try { DOMStoreWriteTransaction transaction = store.newWriteOnlyTransaction(); @@ -834,9 +877,9 @@ public class Shard extends RaftActor { transaction.write(DATASTORE_ROOT, node); syncCommitTransaction(transaction); } catch (InterruptedException | ExecutionException e) { - LOG.error(e, "An exception occurred when applying snapshot"); + LOG.error("{}: An exception occurred when applying snapshot", persistenceId(), e); } finally { - LOG.info("Done applying snapshot"); + LOG.info("{}: Done applying snapshot", persistenceId()); } } @@ -857,21 +900,15 @@ public class Shard extends RaftActor { delayedListenerRegistrations.clear(); } - shardMBean.setRaftState(getRaftState().name()); - shardMBean.setCurrentTerm(getCurrentTerm()); - // If this actor is no longer the leader close all the transaction chains - if(!isLeader){ - for(Map.Entry entry : transactionChains.entrySet()){ - if(LOG.isDebugEnabled()) { - LOG.debug( - "onStateChanged: Closing transaction chain {} because shard {} is no longer the leader", - entry.getKey(), getId()); - } - entry.getValue().close(); + if(!isLeader) { + if(LOG.isDebugEnabled()) { + LOG.debug( + "{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader", + persistenceId(), getId()); } - transactionChains.clear(); + transactionFactory.closeAllTransactionChains(); } } @@ -880,12 +917,8 @@ public class Shard extends RaftActor { return dataPersistenceProvider; } - @Override protected void onLeaderChanged(final String oldLeader, final String newLeader) { - shardMBean.setLeader(newLeader); - } - @Override public String persistenceId() { - return this.name.toString(); + return this.name; } @VisibleForTesting @@ -893,16 +926,22 @@ public class Shard extends RaftActor { return dataPersistenceProvider; } + @VisibleForTesting + ShardCommitCoordinator getCommitCoordinator() { + return commitCoordinator; + } + + private static class ShardCreator implements Creator { private static final long serialVersionUID = 1L; final ShardIdentifier name; - final Map peerAddresses; + final Map peerAddresses; final DatastoreContext datastoreContext; final SchemaContext schemaContext; - ShardCreator(final ShardIdentifier name, final Map peerAddresses, + ShardCreator(final ShardIdentifier name, final Map peerAddresses, final DatastoreContext datastoreContext, final SchemaContext schemaContext) { this.name = name; this.peerAddresses = peerAddresses; @@ -917,7 +956,7 @@ public class Shard extends RaftActor { } @VisibleForTesting - InMemoryDOMDataStore getDataStore() { + public InMemoryDOMDataStore getDataStore() { return store; }