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=d52d3a3bf20e576fce0556e866a53cf74d7c6a4d;hp=3cab853f9adeaa069de9e9ce852ae94d32a94666;hb=925cb4a228d0fda99c7bfeb432eb25285a223887;hpb=3e80db38f7f579505173c29c42f800983d7ca6c1 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 3cab853f9a..d52d3a3bf2 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 @@ -12,22 +12,24 @@ import akka.actor.ActorRef; import akka.actor.ActorSelection; import akka.actor.Cancellable; import akka.actor.Props; -import akka.persistence.RecoveryFailure; import akka.serialization.Serialization; import com.google.common.annotations.VisibleForTesting; import com.google.common.base.Optional; import com.google.common.base.Preconditions; +import com.google.common.base.Ticker; import java.io.IOException; +import java.util.Collection; import java.util.Collections; import java.util.Map; import java.util.concurrent.TimeUnit; import javax.annotation.Nonnull; +import org.opendaylight.controller.cluster.access.concepts.TransactionIdentifier; import org.opendaylight.controller.cluster.common.actor.CommonConfig; +import org.opendaylight.controller.cluster.common.actor.MessageTracker; +import org.opendaylight.controller.cluster.common.actor.MessageTracker.Error; import org.opendaylight.controller.cluster.common.actor.MeteringBehavior; -import org.opendaylight.controller.cluster.datastore.ShardCommitCoordinator.CohortEntry; import org.opendaylight.controller.cluster.datastore.exceptions.NoShardLeaderException; import org.opendaylight.controller.cluster.datastore.identifiers.ShardIdentifier; -import org.opendaylight.controller.cluster.datastore.identifiers.ShardTransactionIdentifier; import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardMBeanFactory; import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardStats; import org.opendaylight.controller.cluster.datastore.messages.AbortTransaction; @@ -36,7 +38,6 @@ import org.opendaylight.controller.cluster.datastore.messages.BatchedModificatio import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransaction; import org.opendaylight.controller.cluster.datastore.messages.CloseTransactionChain; import org.opendaylight.controller.cluster.datastore.messages.CommitTransaction; -import org.opendaylight.controller.cluster.datastore.messages.CommitTransactionReply; import org.opendaylight.controller.cluster.datastore.messages.CreateTransaction; import org.opendaylight.controller.cluster.datastore.messages.CreateTransactionReply; import org.opendaylight.controller.cluster.datastore.messages.DatastoreSnapshot; @@ -49,10 +50,7 @@ import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeList import org.opendaylight.controller.cluster.datastore.messages.RegisterDataTreeChangeListener; import org.opendaylight.controller.cluster.datastore.messages.ShardLeaderStateChanged; import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext; -import org.opendaylight.controller.cluster.datastore.modification.Modification; -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.notifications.LeaderStateChanged; import org.opendaylight.controller.cluster.notifications.RegisterRoleChangeListener; import org.opendaylight.controller.cluster.notifications.RoleChangeNotifier; @@ -63,32 +61,40 @@ import org.opendaylight.controller.cluster.raft.RaftState; import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus; import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply; import org.opendaylight.controller.cluster.raft.messages.ServerRemoved; -import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationByteStringPayload; -import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationPayload; -import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTree; -import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidate; +import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload; +import org.opendaylight.yangtools.concepts.Identifier; import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException; -import org.opendaylight.yangtools.yang.data.api.schema.tree.ModificationType; +import org.opendaylight.yangtools.yang.data.api.schema.tree.TipProducingDataTree; import org.opendaylight.yangtools.yang.data.api.schema.tree.TreeType; import org.opendaylight.yangtools.yang.model.api.SchemaContext; import scala.concurrent.duration.Duration; import scala.concurrent.duration.FiniteDuration; /** - * A Shard represents a portion of the logical data tree
- *

+ * A Shard represents a portion of the logical data tree. + *

* Our Shard uses InMemoryDataTree as it's internal representation and delegates all requests it - *

*/ public class Shard extends RaftActor { - protected static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = "txCommitTimeoutCheck"; - @VisibleForTesting - static final Object GET_SHARD_MBEAN_MESSAGE = "getShardMBeanMessage"; + static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = new Object() { + @Override + public String toString() { + return "txCommitTimeoutCheck"; + } + }; @VisibleForTesting - static final String DEFAULT_NAME = "default"; + static final Object GET_SHARD_MBEAN_MESSAGE = new Object() { + @Override + public String toString() { + return "getShardMBeanMessage"; + } + }; + + // FIXME: shard names should be encapsulated in their own class and this should be exposed as a constant. + public static final String DEFAULT_NAME = "default"; // The state of this Shard private final ShardDataTree store; @@ -122,7 +128,9 @@ public class Shard extends RaftActor { private final ShardTransactionMessageRetrySupport messageRetrySupport; - protected Shard(AbstractBuilder builder) { + private final FrontendMetadata frontendMetadata = new FrontendMetadata(); + + protected Shard(final AbstractBuilder builder) { super(builder.getId().toString(), builder.getPeerAddresses(), Optional.of(builder.getDatastoreContext().getShardRaftConfig()), DataStoreVersions.CURRENT_VERSION); @@ -134,7 +142,17 @@ public class Shard extends RaftActor { LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent()); - store = new ShardDataTree(builder.getSchemaContext(), builder.getTreeType()); + ShardDataTreeChangeListenerPublisherActorProxy treeChangeListenerPublisher = + new ShardDataTreeChangeListenerPublisherActorProxy(getContext(), name + "-DTCL-publisher"); + ShardDataChangeListenerPublisherActorProxy dataChangeListenerPublisher = + new ShardDataChangeListenerPublisherActorProxy(getContext(), name + "-DCL-publisher"); + if (builder.getDataTree() != null) { + store = new ShardDataTree(this, builder.getSchemaContext(), builder.getDataTree(), + treeChangeListenerPublisher, dataChangeListenerPublisher, name); + } else { + store = new ShardDataTree(this, builder.getSchemaContext(), builder.getTreeType(), + treeChangeListenerPublisher, dataChangeListenerPublisher, name); + } shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(), datastoreContext.getDataStoreMXBeanType()); @@ -144,9 +162,7 @@ public class Shard extends RaftActor { getContext().become(new MeteringBehavior(this)); } - commitCoordinator = new ShardCommitCoordinator(store, - datastoreContext.getShardCommitQueueExpiryTimeoutInMillis(), - datastoreContext.getShardTransactionCommitQueueCapacity(), LOG, this.name); + commitCoordinator = new ShardCommitCoordinator(store, LOG, this.name); setTransactionCommitTimeout(); @@ -160,7 +176,8 @@ public class Shard extends RaftActor { new Dispatchers(context().system().dispatchers()).getDispatcherPath( Dispatchers.DispatcherType.Transaction), self(), getContext(), shardMBean); - snapshotCohort = new ShardSnapshotCohort(transactionActorFactory, store, LOG, this.name); + snapshotCohort = ShardSnapshotCohort.create(getContext(), builder.getId().getMemberName(), store, LOG, + this.name); messageRetrySupport = new ShardTransactionMessageRetrySupport(this); } @@ -170,7 +187,7 @@ public class Shard extends RaftActor { datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS) / 2; } - private Optional createRoleChangeNotifier(String shardId) { + private Optional createRoleChangeNotifier(final String shardId) { ActorRef shardRoleChangeNotifier = this.getContext().actorOf( RoleChangeNotifier.getProps(shardId), shardId + "-notifier"); return Optional.of(shardRoleChangeNotifier); @@ -184,7 +201,7 @@ public class Shard extends RaftActor { messageRetrySupport.close(); - if(txCommitTimeoutCheckSchedule != null) { + if (txCommitTimeoutCheckSchedule != null) { txCommitTimeoutCheckSchedule.cancel(); } @@ -194,53 +211,40 @@ public class Shard extends RaftActor { } @Override - public void onReceiveRecover(final Object message) throws Exception { - if(LOG.isDebugEnabled()) { - LOG.debug("{}: onReceiveRecover: Received message {} from {}", persistenceId(), - message.getClass().toString(), getSender()); - } - - if (message instanceof RecoveryFailure){ - LOG.error("{}: Recovery failed because of this cause", - persistenceId(), ((RecoveryFailure) message).cause()); + protected void handleRecover(final Object message) { + LOG.debug("{}: onReceiveRecover: Received message {} from {}", persistenceId(), message.getClass(), + getSender()); - // 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(); - } + super.handleRecover(message); + if (LOG.isTraceEnabled()) { + appendEntriesReplyTracker.begin(); } } @Override - public void onReceiveCommand(final Object message) throws Exception { - - MessageTracker.Context context = appendEntriesReplyTracker.received(message); - - if(context.error().isPresent()){ - LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(), - context.error()); - } + protected void handleNonRaftCommand(final Object message) { + try (final MessageTracker.Context context = appendEntriesReplyTracker.received(message)) { + final Optional maybeError = context.error(); + if (maybeError.isPresent()) { + LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(), + maybeError.get()); + } - try { if (CreateTransaction.isSerializedType(message)) { handleCreateTransaction(message); - } else if (BatchedModifications.class.isInstance(message)) { + } else if (message instanceof BatchedModifications) { handleBatchedModifications((BatchedModifications)message); } else if (message instanceof ForwardedReadyTransaction) { handleForwardedReadyTransaction((ForwardedReadyTransaction) message); } else if (message instanceof ReadyLocalTransaction) { handleReadyLocalTransaction((ReadyLocalTransaction)message); - } else if (CanCommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) { + } else if (CanCommitTransaction.isSerializedType(message)) { handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message)); - } else if (CommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) { + } else if (CommitTransaction.isSerializedType(message)) { handleCommitTransaction(CommitTransaction.fromSerializable(message)); - } else if (AbortTransaction.SERIALIZABLE_CLASS.isInstance(message)) { + } else if (AbortTransaction.isSerializedType(message)) { handleAbortTransaction(AbortTransaction.fromSerializable(message)); - } else if (CloseTransactionChain.SERIALIZABLE_CLASS.isInstance(message)) { + } else if (CloseTransactionChain.isSerializedType(message)) { closeTransactionChain(CloseTransactionChain.fromSerializable(message)); } else if (message instanceof RegisterChangeListener) { changeSupport.onMessage((RegisterChangeListener) message, isLeader(), hasLeader()); @@ -252,28 +256,30 @@ public class Shard extends RaftActor { PeerAddressResolved resolved = (PeerAddressResolved) message; setPeerAddress(resolved.getPeerId().toString(), resolved.getPeerAddress()); - } else if (message.equals(TX_COMMIT_TIMEOUT_CHECK_MESSAGE)) { + } else if (TX_COMMIT_TIMEOUT_CHECK_MESSAGE.equals(message)) { + store.checkForExpiredTransactions(transactionCommitTimeout); commitCoordinator.checkForExpiredTransactions(transactionCommitTimeout, this); - } else if(message instanceof DatastoreContext) { + } else if (message instanceof DatastoreContext) { onDatastoreContext((DatastoreContext)message); - } else if(message instanceof RegisterRoleChangeListener){ + } 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 if(GET_SHARD_MBEAN_MESSAGE.equals(message)){ + } else if (GET_SHARD_MBEAN_MESSAGE.equals(message)) { sender().tell(getShardMBean(), self()); - } else if(message instanceof GetShardDataTree) { + } else if (message instanceof GetShardDataTree) { sender().tell(store.getDataTree(), self()); - } else if(message instanceof ServerRemoved){ + } else if (message instanceof ServerRemoved) { context().parent().forward(message, context()); - } else if(ShardTransactionMessageRetrySupport.TIMER_MESSAGE_CLASS.isInstance(message)) { + } else if (ShardTransactionMessageRetrySupport.TIMER_MESSAGE_CLASS.isInstance(message)) { messageRetrySupport.onTimerMessage(message); + } else if (message instanceof DataTreeCohortActorRegistry.CohortRegistryCommand) { + store.processCohortRegistryCommand(getSender(), + (DataTreeCohortActorRegistry.CohortRegistryCommand) message); } else { - super.onReceiveCommand(message); + super.handleNonRaftCommand(message); } - } finally { - context.done(); } } @@ -282,7 +288,7 @@ public class Shard extends RaftActor { } public int getPendingTxCommitQueueSize() { - return commitCoordinator.getQueueSize(); + return store.getQueueSize(); } public int getCohortCacheSize() { @@ -295,116 +301,68 @@ public class Shard extends RaftActor { } @Override - protected LeaderStateChanged newLeaderStateChanged(String memberId, String leaderId, short leaderPayloadVersion) { - return new ShardLeaderStateChanged(memberId, leaderId, - isLeader() ? Optional.of(store.getDataTree()) : Optional.absent(), - leaderPayloadVersion); + protected LeaderStateChanged newLeaderStateChanged(final String memberId, final String leaderId, + final short leaderPayloadVersion) { + return isLeader() ? new ShardLeaderStateChanged(memberId, leaderId, store.getDataTree(), leaderPayloadVersion) + : new ShardLeaderStateChanged(memberId, leaderId, leaderPayloadVersion); } - protected void onDatastoreContext(DatastoreContext context) { + protected void onDatastoreContext(final DatastoreContext context) { datastoreContext = context; - commitCoordinator.setQueueCapacity(datastoreContext.getShardTransactionCommitQueueCapacity()); - setTransactionCommitTimeout(); - if(datastoreContext.isPersistent() && !persistence().isRecoveryApplicable()) { - setPersistence(true); - } else if(!datastoreContext.isPersistent() && persistence().isRecoveryApplicable()) { - setPersistence(false); - } + setPersistence(datastoreContext.isPersistent()); updateConfigParams(datastoreContext.getShardRaftConfig()); } - private static boolean isEmptyCommit(final DataTreeCandidate candidate) { - return ModificationType.UNMODIFIED.equals(candidate.getRootNode().getModificationType()); + boolean canSkipPayload() { + // If we do not have any followers and we are not using persistence we can apply modification to the state + // immediately + return !hasFollowers() && !persistence().isRecoveryApplicable(); } - void continueCommit(final CohortEntry cohortEntry) { - final DataTreeCandidate candidate = cohortEntry.getCandidate(); - - // 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()) || isEmptyCommit(candidate)) { - applyModificationToState(cohortEntry.getReplySender(), cohortEntry.getTransactionID(), candidate); - } else { - Shard.this.persistData(cohortEntry.getReplySender(), cohortEntry.getTransactionID(), - DataTreeCandidatePayload.create(candidate)); - } + // applyState() will be invoked once consensus is reached on the payload + void persistPayload(final TransactionIdentifier transactionId, final Payload payload) { + // We are faking the sender + persistData(self(), transactionId, payload); } private void handleCommitTransaction(final CommitTransaction commit) { - if(!commitCoordinator.handleCommit(commit.getTransactionID(), getSender(), this)) { - shardMBean.incrementFailedTransactionsCount(); - } - } - - private void finishCommit(@Nonnull final ActorRef sender, @Nonnull final String transactionID, @Nonnull final CohortEntry cohortEntry) { - LOG.debug("{}: Finishing commit for transaction {}", persistenceId(), cohortEntry.getTransactionID()); - - try { - cohortEntry.commit(); - - sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf()); - - shardMBean.incrementCommittedTransactionCount(); - shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis()); - - } catch (Exception e) { - sender.tell(new akka.actor.Status.Failure(e), getSelf()); - - LOG.error("{}, An exception occurred while committing transaction {}", persistenceId(), - transactionID, e); - shardMBean.incrementFailedTransactionsCount(); - } finally { - commitCoordinator.currentTransactionComplete(transactionID, true); - } - } - - private void finishCommit(@Nonnull final ActorRef sender, final @Nonnull String transactionID) { - // With persistence enabled, this method is called via applyState by the leader strategy - // after the commit has been replicated to a majority of the followers. - - CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID); - if (cohortEntry == null) { - // The transaction is no longer the current commit. This can happen if the transaction - // was aborted prior, most likely due to timeout in the front-end. We need to finish - // committing the transaction though since it was successfully persisted and replicated - // however we can't use the original cohort b/c it was already preCommitted and may - // conflict with the current commit or may have been aborted so we commit with a new - // transaction. - cohortEntry = commitCoordinator.getAndRemoveCohortEntry(transactionID); - if(cohortEntry != null) { - try { - store.applyForeignCandidate(transactionID, cohortEntry.getCandidate()); - } catch (DataValidationFailedException e) { - shardMBean.incrementFailedTransactionsCount(); - LOG.error("{}: Failed to re-apply transaction {}", persistenceId(), transactionID, e); - } - - sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf()); + if (isLeader()) { + commitCoordinator.handleCommit(commit.getTransactionID(), getSender(), this); + } else { + ActorSelection leader = getLeader(); + if (leader == null) { + messageRetrySupport.addMessageToRetry(commit, getSender(), + "Could not commit transaction " + commit.getTransactionID()); } else { - // 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("%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()); + LOG.debug("{}: Forwarding CommitTransaction to leader {}", persistenceId(), leader); + leader.forward(commit, getContext()); } - } else { - finishCommit(sender, transactionID, cohortEntry); } } private void handleCanCommitTransaction(final CanCommitTransaction canCommit) { LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionID()); - commitCoordinator.handleCanCommit(canCommit.getTransactionID(), getSender(), this); + + if (isLeader()) { + commitCoordinator.handleCanCommit(canCommit.getTransactionID(), getSender(), this); + } else { + ActorSelection leader = getLeader(); + if (leader == null) { + messageRetrySupport.addMessageToRetry(canCommit, getSender(), + "Could not canCommit transaction " + canCommit.getTransactionID()); + } else { + LOG.debug("{}: Forwarding CanCommitTransaction to leader {}", persistenceId(), leader); + leader.forward(canCommit, getContext()); + } + } } - protected void handleBatchedModificationsLocal(BatchedModifications batched, ActorRef sender) { + @SuppressWarnings("checkstyle:IllegalCatch") + protected void handleBatchedModificationsLocal(final BatchedModifications batched, final ActorRef sender) { try { commitCoordinator.handleBatchedModifications(batched, sender, this); } catch (Exception e) { @@ -414,7 +372,7 @@ public class Shard extends RaftActor { } } - private void handleBatchedModifications(BatchedModifications batched) { + private void handleBatchedModifications(final BatchedModifications batched) { // This message is sent to prepare the modifications 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 @@ -436,20 +394,29 @@ public class Shard extends RaftActor { messageRetrySupport.addMessageToRetry(batched, getSender(), "Could not commit transaction " + batched.getTransactionID()); } else { - // 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()); + // If this is not the first batch and leadership changed in between batched messages, + // we need to reconstruct previous BatchedModifications from the transaction + // DataTreeModification, honoring the max batched modification count, and forward all the + // previous BatchedModifications to the new leader. + Collection newModifications = commitCoordinator + .createForwardedBatchedModifications(batched, + datastoreContext.getShardBatchedModificationCount()); + + LOG.debug("{}: Forwarding {} BatchedModifications to leader {}", persistenceId(), + newModifications.size(), leader); + + for (BatchedModifications bm : newModifications) { + leader.forward(bm, getContext()); + } } } } - private boolean failIfIsolatedLeader(ActorRef sender) { - if(isIsolatedLeader()) { + private boolean failIfIsolatedLeader(final ActorRef sender) { + if (isIsolatedLeader()) { sender.tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format( - "Shard %s was the leader but has lost contact with all of its followers. Either all" + - " other follower nodes are down or this node is isolated by a network partition.", + "Shard %s was the leader but has lost contact with all of its followers. Either all" + + " other follower nodes are down or this node is isolated by a network partition.", persistenceId()))), getSelf()); return true; } @@ -461,6 +428,7 @@ public class Shard extends RaftActor { return getRaftState() == RaftState.IsolatedLeader; } + @SuppressWarnings("checkstyle:IllegalCatch") private void handleReadyLocalTransaction(final ReadyLocalTransaction message) { LOG.debug("{}: handleReadyLocalTransaction for {}", persistenceId(), message.getTransactionID()); @@ -486,7 +454,7 @@ public class Shard extends RaftActor { } } - private void handleForwardedReadyTransaction(ForwardedReadyTransaction forwardedReady) { + private void handleForwardedReadyTransaction(final ForwardedReadyTransaction forwardedReady) { LOG.debug("{}: handleForwardedReadyTransaction for {}", persistenceId(), forwardedReady.getTransactionID()); boolean isLeaderActive = isLeaderActive(); @@ -512,7 +480,7 @@ public class Shard extends RaftActor { doAbortTransaction(abort.getTransactionID(), getSender()); } - void doAbortTransaction(final String transactionID, final ActorRef sender) { + void doAbortTransaction(final TransactionIdentifier transactionID, final ActorRef sender) { commitCoordinator.handleAbort(transactionID, sender, this); } @@ -528,27 +496,19 @@ public class Shard extends RaftActor { } private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) { - store.closeTransactionChain(closeTransactionChain.getTransactionChainId()); + store.closeTransactionChain(closeTransactionChain.getIdentifier()); } - private ActorRef createTypedTransactionActor(int transactionType, - ShardTransactionIdentifier transactionId, String transactionChainId, - short clientVersion ) { - - return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType), - transactionId, transactionChainId, clientVersion); - } - - private void createTransaction(CreateTransaction createTransaction) { + @SuppressWarnings("checkstyle:IllegalCatch") + private void createTransaction(final CreateTransaction createTransaction) { try { - if(TransactionType.fromInt(createTransaction.getTransactionType()) != TransactionType.READ_ONLY && - failIfIsolatedLeader(getSender())) { + if (TransactionType.fromInt(createTransaction.getTransactionType()) != TransactionType.READ_ONLY + && failIfIsolatedLeader(getSender())) { return; } ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(), - createTransaction.getTransactionId(), createTransaction.getTransactionChainId(), - createTransaction.getVersion()); + createTransaction.getTransactionId()); getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor), createTransaction.getTransactionId(), createTransaction.getVersion()).toSerializable(), getSelf()); @@ -557,33 +517,10 @@ public class Shard extends RaftActor { } } - private ActorRef createTransaction(int transactionType, String remoteTransactionId, - String transactionChainId, short clientVersion) { - - - ShardTransactionIdentifier transactionId = new ShardTransactionIdentifier(remoteTransactionId); - - if(LOG.isDebugEnabled()) { - LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId); - } - - ActorRef transactionActor = createTypedTransactionActor(transactionType, transactionId, - transactionChainId, clientVersion); - - return transactionActor; - } - - private void commitWithNewTransaction(final Modification modification) { - ReadWriteShardDataTreeTransaction tx = store.newReadWriteTransaction(modification.toString(), null); - modification.apply(tx.getSnapshot()); - try { - snapshotCohort.syncCommitTransaction(tx); - shardMBean.incrementCommittedTransactionCount(); - shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis()); - } catch (Exception e) { - shardMBean.incrementFailedTransactionsCount(); - LOG.error("{}: Failed to commit", persistenceId(), e); - } + private ActorRef createTransaction(final int transactionType, final TransactionIdentifier transactionId) { + LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId); + return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType), + transactionId); } private void updateSchemaContext(final UpdateSchemaContext message) { @@ -609,8 +546,8 @@ public class Shard extends RaftActor { @Override @Nonnull protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() { - return new ShardRecoveryCoordinator(store, store.getSchemaContext(), - restoreFromSnapshot != null ? restoreFromSnapshot.getSnapshot() : null, persistenceId(), LOG); + return new ShardRecoveryCoordinator(store, + restoreFromSnapshot != null ? restoreFromSnapshot.getSnapshot() : null, persistenceId(), LOG); } @Override @@ -621,7 +558,7 @@ public class Shard extends RaftActor { getContext().parent().tell(new ActorInitialized(), getSelf()); // Being paranoid here - this method should only be called once but just in case... - if(txCommitTimeoutCheckSchedule == null) { + if (txCommitTimeoutCheckSchedule == null) { // Schedule a message to be periodically sent to check if the current in-progress // transaction should be expired and aborted. FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS); @@ -632,46 +569,15 @@ public class Shard extends RaftActor { } @Override - protected void applyState(final ActorRef clientActor, final String identifier, final Object data) { - if (data instanceof DataTreeCandidatePayload) { - if (clientActor == null) { - // No clientActor indicates a replica coming from the leader - try { - store.applyForeignCandidate(identifier, ((DataTreeCandidatePayload)data).getCandidate()); - } catch (DataValidationFailedException | IOException e) { - LOG.error("{}: Error applying replica {}", persistenceId(), identifier, e); - } - } else { - // Replication consensus reached, proceed to commit - finishCommit(clientActor, identifier); + protected void applyState(final ActorRef clientActor, final Identifier identifier, final Object data) { + if (data instanceof Payload) { + try { + store.applyReplicatedPayload(identifier, (Payload)data); + } catch (DataValidationFailedException | IOException e) { + LOG.error("{}: Error applying replica {}", persistenceId(), identifier, e); } - } else if (data instanceof CompositeModificationPayload) { - Object modification = ((CompositeModificationPayload) data).getModification(); - - applyModificationToState(clientActor, identifier, modification); - } else if(data instanceof CompositeModificationByteStringPayload ){ - Object modification = ((CompositeModificationByteStringPayload) data).getModification(); - - applyModificationToState(clientActor, identifier, modification); - } else { - LOG.error("{}: Unknown state received {} Class loader = {} CompositeNodeMod.ClassLoader = {}", - persistenceId(), data, data.getClass().getClassLoader(), - CompositeModificationPayload.class.getClassLoader()); - } - } - - private void applyModificationToState(ActorRef clientActor, String identifier, Object modification) { - if(modification == null) { - LOG.error( - "{}: 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. - commitWithNewTransaction(MutableCompositeModification.fromSerializable(modification)); } else { - // This must be the OK to commit after replication consensus. - finishCommit(clientActor, identifier); + LOG.error("{}: Unknown state for {} received {}", persistenceId(), identifier, data); } } @@ -684,36 +590,65 @@ public class Shard extends RaftActor { // If this actor is no longer the leader close all the transaction chains if (!isLeader) { - if(LOG.isDebugEnabled()) { + if (LOG.isDebugEnabled()) { LOG.debug( "{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader", persistenceId(), getId()); } store.closeAllTransactionChains(); - - commitCoordinator.abortPendingTransactions( - "The transacton was aborted due to inflight leadership change.", this); } - if(hasLeader && !isIsolatedLeader()) { + if (hasLeader && !isIsolatedLeader()) { messageRetrySupport.retryMessages(); } } @Override - protected void onLeaderChanged(String oldLeader, String newLeader) { + protected void onLeaderChanged(final String oldLeader, final String newLeader) { shardMBean.incrementLeadershipChangeCount(); - if(hasLeader() && !isIsolatedLeader()) { + boolean hasLeader = hasLeader(); + if (hasLeader && !isLeader()) { + // Another leader was elected. If we were the previous leader and had pending transactions, convert + // them to transaction messages and send to the new leader. + ActorSelection leader = getLeader(); + if (leader != null) { + Collection messagesToForward = convertPendingTransactionsToMessages(); + + if (!messagesToForward.isEmpty()) { + LOG.debug("{}: Forwarding {} pending transaction messages to leader {}", persistenceId(), + messagesToForward.size(), leader); + + for (Object message : messagesToForward) { + leader.tell(message, self()); + } + } + } else { + commitCoordinator.abortPendingTransactions("The transacton was aborted due to inflight leadership " + + "change and the leader address isn't available.", this); + } + } + + if (hasLeader && !isIsolatedLeader()) { messageRetrySupport.retryMessages(); } } + /** + * Clears all pending transactions and converts them to messages to be forwarded to a new leader. + * + * @return the converted messages + */ + public Collection convertPendingTransactionsToMessages() { + return commitCoordinator.convertPendingTransactionsToMessages( + datastoreContext.getShardBatchedModificationCount()); + } + @Override - protected void pauseLeader(Runnable operation) { + protected void pauseLeader(final Runnable operation) { LOG.debug("{}: In pauseLeader, operation: {}", persistenceId(), operation); - commitCoordinator.setRunOnPendingTransactionsComplete(operation); + store.setRunOnPendingTransactionsComplete(operation); } @Override @@ -744,16 +679,17 @@ public class Shard extends RaftActor { return new Builder(); } - public static abstract class AbstractBuilder, S extends Shard> { + public abstract static class AbstractBuilder, S extends Shard> { private final Class shardClass; private ShardIdentifier id; private Map peerAddresses = Collections.emptyMap(); private DatastoreContext datastoreContext; private SchemaContext schemaContext; private DatastoreSnapshot.ShardSnapshot restoreFromSnapshot; + private TipProducingDataTree dataTree; private volatile boolean sealed; - protected AbstractBuilder(Class shardClass) { + protected AbstractBuilder(final Class shardClass) { this.shardClass = shardClass; } @@ -766,33 +702,39 @@ public class Shard extends RaftActor { return (T) this; } - public T id(ShardIdentifier id) { + public T id(final ShardIdentifier newId) { + checkSealed(); + this.id = newId; + return self(); + } + + public T peerAddresses(final Map newPeerAddresses) { checkSealed(); - this.id = id; + this.peerAddresses = newPeerAddresses; return self(); } - public T peerAddresses(Map peerAddresses) { + public T datastoreContext(final DatastoreContext newDatastoreContext) { checkSealed(); - this.peerAddresses = peerAddresses; + this.datastoreContext = newDatastoreContext; return self(); } - public T datastoreContext(DatastoreContext datastoreContext) { + public T schemaContext(final SchemaContext newSchemaContext) { checkSealed(); - this.datastoreContext = datastoreContext; + this.schemaContext = newSchemaContext; return self(); } - public T schemaContext(SchemaContext schemaContext) { + public T restoreFromSnapshot(final DatastoreSnapshot.ShardSnapshot newRestoreFromSnapshot) { checkSealed(); - this.schemaContext = schemaContext; + this.restoreFromSnapshot = newRestoreFromSnapshot; return self(); } - public T restoreFromSnapshot(DatastoreSnapshot.ShardSnapshot restoreFromSnapshot) { + public T dataTree(final TipProducingDataTree newDataTree) { checkSealed(); - this.restoreFromSnapshot = restoreFromSnapshot; + this.dataTree = newDataTree; return self(); } @@ -816,15 +758,20 @@ public class Shard extends RaftActor { return restoreFromSnapshot; } + public TipProducingDataTree getDataTree() { + return dataTree; + } + public TreeType getTreeType() { switch (datastoreContext.getLogicalStoreType()) { - case CONFIGURATION: - return TreeType.CONFIGURATION; - case OPERATIONAL: - return TreeType.OPERATIONAL; + case CONFIGURATION: + return TreeType.CONFIGURATION; + case OPERATIONAL: + return TreeType.OPERATIONAL; + default: + throw new IllegalStateException("Unhandled logical store type " + + datastoreContext.getLogicalStoreType()); } - - throw new IllegalStateException("Unhandled logical store type " + datastoreContext.getLogicalStoreType()); } protected void verify() { @@ -846,4 +793,8 @@ public class Shard extends RaftActor { super(Shard.class); } } + + Ticker ticker() { + return Ticker.systemTicker(); + } }