import akka.actor.ActorSelection;
import akka.actor.Cancellable;
import akka.actor.Props;
-import akka.japi.Creator;
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.util.concurrent.FutureCallback;
-import com.google.common.util.concurrent.Futures;
-import com.google.common.util.concurrent.ListenableFuture;
import java.io.IOException;
-import java.util.HashMap;
+import java.util.Collections;
import java.util.Map;
import java.util.concurrent.TimeUnit;
import javax.annotation.Nonnull;
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;
-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.CanCommitTransaction;
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;
+import org.opendaylight.controller.cluster.datastore.messages.DatastoreSnapshot.ShardSnapshot;
import org.opendaylight.controller.cluster.datastore.messages.ForwardedReadyTransaction;
+import org.opendaylight.controller.cluster.datastore.messages.GetShardDataTree;
import org.opendaylight.controller.cluster.datastore.messages.PeerAddressResolved;
+import org.opendaylight.controller.cluster.datastore.messages.ReadyLocalTransaction;
import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListener;
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.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.raft.RaftActor;
import org.opendaylight.controller.cluster.raft.RaftActorRecoveryCohort;
import org.opendaylight.controller.cluster.raft.RaftActorSnapshotCohort;
+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.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.TreeType;
import org.opendaylight.yangtools.yang.model.api.SchemaContext;
import scala.concurrent.duration.Duration;
import scala.concurrent.duration.FiniteDuration;
*/
public class Shard extends RaftActor {
- private static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = "txCommitTimeoutCheck";
+ protected static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = "txCommitTimeoutCheck";
+
+ @VisibleForTesting
+ static final Object GET_SHARD_MBEAN_MESSAGE = "getShardMBeanMessage";
@VisibleForTesting
static final String DEFAULT_NAME = "default";
private final DataTreeChangeListenerSupport treeChangeSupport = new DataTreeChangeListenerSupport(this);
private final DataChangeListenerSupport changeSupport = new DataChangeListenerSupport(this);
- protected Shard(final ShardIdentifier name, final Map<String, String> peerAddresses,
- final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
- super(name.toString(), new HashMap<>(peerAddresses), Optional.of(datastoreContext.getShardRaftConfig()));
- this.name = name.toString();
- this.datastoreContext = datastoreContext;
+ private ShardSnapshot restoreFromSnapshot;
+
+ private final ShardTransactionMessageRetrySupport messageRetrySupport;
+
+ protected Shard(AbstractBuilder<?, ?> builder) {
+ super(builder.getId().toString(), builder.getPeerAddresses(),
+ Optional.of(builder.getDatastoreContext().getShardRaftConfig()), DataStoreVersions.CURRENT_VERSION);
+
+ this.name = builder.getId().toString();
+ this.datastoreContext = builder.getDatastoreContext();
+ this.restoreFromSnapshot = builder.getRestoreFromSnapshot();
setPersistence(datastoreContext.isPersistent());
LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
- store = new ShardDataTree(schemaContext);
+ store = new ShardDataTree(builder.getSchemaContext(), builder.getTreeType());
shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
datastoreContext.getDataStoreMXBeanType());
- shardMBean.setShardActor(getSelf());
+ shardMBean.setShard(this);
if (isMetricsCaptureEnabled()) {
getContext().become(new MeteringBehavior(this));
}
commitCoordinator = new ShardCommitCoordinator(store,
- TimeUnit.SECONDS.convert(5, TimeUnit.MINUTES),
- datastoreContext.getShardTransactionCommitQueueCapacity(), self(), LOG, this.name);
+ datastoreContext.getShardCommitQueueExpiryTimeoutInMillis(),
+ datastoreContext.getShardTransactionCommitQueueCapacity(), LOG, this.name);
setTransactionCommitTimeout();
Dispatchers.DispatcherType.Transaction), self(), getContext(), shardMBean);
snapshotCohort = new ShardSnapshotCohort(transactionActorFactory, store, LOG, this.name);
+
+ messageRetrySupport = new ShardTransactionMessageRetrySupport(this);
}
private void setTransactionCommitTimeout() {
transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
- datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS);
- }
-
- public static Props props(final ShardIdentifier name,
- final Map<String, String> peerAddresses,
- final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
- Preconditions.checkNotNull(name, "name should not be null");
- Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
- Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
- Preconditions.checkNotNull(schemaContext, "schemaContext should not be null");
-
- return Props.create(new ShardCreator(name, peerAddresses, datastoreContext, schemaContext));
+ datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS) / 2;
}
private Optional<ActorRef> createRoleChangeNotifier(String shardId) {
super.postStop();
+ messageRetrySupport.close();
+
if(txCommitTimeoutCheckSchedule != null) {
txCommitTimeoutCheckSchedule.cancel();
}
+ commitCoordinator.abortPendingTransactions("Transaction aborted due to shutdown.", this);
+
shardMBean.unregisterMBean();
}
if(context.error().isPresent()){
LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(),
- context.error());
+ context.error());
}
try {
- if (CreateTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
+ if (CreateTransaction.isSerializedType(message)) {
handleCreateTransaction(message);
} else if (BatchedModifications.class.isInstance(message)) {
handleBatchedModifications((BatchedModifications)message);
} else if (message instanceof ForwardedReadyTransaction) {
- commitCoordinator.handleForwardedReadyTransaction((ForwardedReadyTransaction) message,
- getSender(), this);
- } else if (CanCommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
+ handleForwardedReadyTransaction((ForwardedReadyTransaction) message);
+ } else if (message instanceof ReadyLocalTransaction) {
+ handleReadyLocalTransaction((ReadyLocalTransaction)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());
+ changeSupport.onMessage((RegisterChangeListener) message, isLeader(), hasLeader());
} else if (message instanceof RegisterDataTreeChangeListener) {
- treeChangeSupport.onMessage((RegisterDataTreeChangeListener) message, isLeader());
+ treeChangeSupport.onMessage((RegisterDataTreeChangeListener) message, isLeader(), hasLeader());
} else if (message instanceof UpdateSchemaContext) {
updateSchemaContext((UpdateSchemaContext) message);
} else if (message instanceof PeerAddressResolved) {
setPeerAddress(resolved.getPeerId().toString(),
resolved.getPeerAddress());
} else if (message.equals(TX_COMMIT_TIMEOUT_CHECK_MESSAGE)) {
- handleTransactionCommitTimeoutCheck();
+ commitCoordinator.checkForExpiredTransactions(transactionCommitTimeout, this);
} else if(message instanceof DatastoreContext) {
onDatastoreContext((DatastoreContext)message);
} else if(message instanceof RegisterRoleChangeListener){
roleChangeNotifier.get().forward(message, context());
- } else if (message instanceof FollowerInitialSyncUpStatus){
+ } else if (message instanceof FollowerInitialSyncUpStatus) {
shardMBean.setFollowerInitialSyncStatus(((FollowerInitialSyncUpStatus) message).isInitialSyncDone());
context().parent().tell(message, self());
+ } else if(GET_SHARD_MBEAN_MESSAGE.equals(message)){
+ sender().tell(getShardMBean(), self());
+ } else if(message instanceof GetShardDataTree) {
+ sender().tell(store.getDataTree(), self());
+ } else if(message instanceof ServerRemoved){
+ context().parent().forward(message, context());
+ } else if(ShardTransactionMessageRetrySupport.TIMER_MESSAGE_CLASS.isInstance(message)) {
+ messageRetrySupport.onTimerMessage(message);
} else {
super.onReceiveCommand(message);
}
}
}
+ private boolean hasLeader() {
+ return getLeaderId() != null;
+ }
+
+ public int getPendingTxCommitQueueSize() {
+ return commitCoordinator.getQueueSize();
+ }
+
+ public int getCohortCacheSize() {
+ return commitCoordinator.getCohortCacheSize();
+ }
+
@Override
protected Optional<ActorRef> getRoleChangeNotifier() {
return roleChangeNotifier;
}
@Override
- protected LeaderStateChanged newLeaderStateChanged(String memberId, String leaderId) {
+ protected LeaderStateChanged newLeaderStateChanged(String memberId, String leaderId, short leaderPayloadVersion) {
return new ShardLeaderStateChanged(memberId, leaderId,
- isLeader() ? Optional.<DataTree>of(store.getDataTree()) : Optional.<DataTree>absent());
+ isLeader() ? Optional.<DataTree>of(store.getDataTree()) : Optional.<DataTree>absent(),
+ leaderPayloadVersion);
}
- private void onDatastoreContext(DatastoreContext context) {
+ protected void onDatastoreContext(DatastoreContext context) {
datastoreContext = context;
commitCoordinator.setQueueCapacity(datastoreContext.getShardTransactionCommitQueueCapacity());
updateConfigParams(datastoreContext.getShardRaftConfig());
}
- private void handleTransactionCommitTimeoutCheck() {
- CohortEntry cohortEntry = commitCoordinator.getCurrentCohortEntry();
- if(cohortEntry != null) {
- long elapsed = System.currentTimeMillis() - cohortEntry.getLastAccessTime();
- if(elapsed > transactionCommitTimeout) {
- LOG.warn("{}: Current transaction {} has timed out after {} ms - aborting",
- persistenceId(), cohortEntry.getTransactionID(), transactionCommitTimeout);
-
- doAbortTransaction(cohortEntry.getTransactionID(), null);
- }
- }
- }
-
private static boolean isEmptyCommit(final DataTreeCandidate candidate) {
return ModificationType.UNMODIFIED.equals(candidate.getRootNode().getModificationType());
}
- void continueCommit(final CohortEntry cohortEntry) throws Exception {
- final DataTreeCandidate candidate = cohortEntry.getCohort().getCandidate();
+ 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(getSender(), cohortEntry.getTransactionID(), candidate);
+ applyModificationToState(cohortEntry.getReplySender(), cohortEntry.getTransactionID(), candidate);
} else {
- Shard.this.persistData(getSender(), cohortEntry.getTransactionID(),
- DataTreeCandidatePayload.create(candidate));
+ Shard.this.persistData(cohortEntry.getReplySender(), cohortEntry.getTransactionID(),
+ DataTreeCandidatePayload.create(candidate));
}
}
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
- // state on a different thread outside of our dispatcher. Also, the data store
- // currently uses a same thread executor anyway.
- cohortEntry.getCohort().commit().get();
+ cohortEntry.commit();
- sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
+ sender.tell(CommitTransactionReply.instance(cohortEntry.getClientVersion()).toSerializable(), getSelf());
shardMBean.incrementCommittedTransactionCount();
shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
cohortEntry = commitCoordinator.getAndRemoveCohortEntry(transactionID);
if(cohortEntry != null) {
try {
- store.applyForeignCandidate(transactionID, cohortEntry.getCohort().getCandidate());
+ 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());
+ sender.tell(CommitTransactionReply.instance(cohortEntry.getClientVersion()).toSerializable(),
+ getSelf());
} 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.
commitCoordinator.handleCanCommit(canCommit.getTransactionID(), getSender(), this);
}
+ protected void handleBatchedModificationsLocal(BatchedModifications batched, ActorRef sender) {
+ try {
+ commitCoordinator.handleBatchedModifications(batched, sender, this);
+ } catch (Exception e) {
+ LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(),
+ batched.getTransactionID(), e);
+ sender.tell(new akka.actor.Status.Failure(e), getSelf());
+ }
+ }
+
private void handleBatchedModifications(BatchedModifications batched) {
- // This message is sent to prepare the modificationsa transaction directly on the Shard as an
+ // 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
// modifications are complete. The reply contains the cohort actor path (this actor) for the caller
// 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 {
- commitCoordinator.handleBatchedModifications(batched, getSender(), this);
- } catch (Exception e) {
- LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(),
- batched.getTransactionID(), e);
- getSender().tell(new akka.actor.Status.Failure(e), getSelf());
- }
+ boolean isLeaderActive = isLeaderActive();
+ if (isLeader() && isLeaderActive) {
+ handleBatchedModificationsLocal(batched, getSender());
} else {
ActorSelection leader = getLeader();
- if(leader != null) {
+ if (!isLeaderActive || leader == null) {
+ 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());
+ }
+ }
+ }
+
+ private boolean failIfIsolatedLeader(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.",
+ persistenceId()))), getSelf());
+ return true;
+ }
+
+ return false;
+ }
+
+ protected boolean isIsolatedLeader() {
+ return getRaftState() == RaftState.IsolatedLeader;
+ }
+
+ private void handleReadyLocalTransaction(final ReadyLocalTransaction message) {
+ LOG.debug("{}: handleReadyLocalTransaction for {}", persistenceId(), message.getTransactionID());
+
+ boolean isLeaderActive = isLeaderActive();
+ if (isLeader() && isLeaderActive) {
+ try {
+ commitCoordinator.handleReadyLocalTransaction(message, getSender(), this);
+ } catch (Exception e) {
+ LOG.error("{}: Error handling ReadyLocalTransaction for Tx {}", persistenceId(),
+ message.getTransactionID(), e);
+ getSender().tell(new akka.actor.Status.Failure(e), getSelf());
+ }
+ } else {
+ ActorSelection leader = getLeader();
+ if (!isLeaderActive || leader == null) {
+ messageRetrySupport.addMessageToRetry(message, getSender(),
+ "Could not commit transaction " + message.getTransactionID());
} 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());
+ LOG.debug("{}: Forwarding ReadyLocalTransaction to leader {}", persistenceId(), leader);
+ message.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
+ leader.forward(message, getContext());
+ }
+ }
+ }
+
+ private void handleForwardedReadyTransaction(ForwardedReadyTransaction forwardedReady) {
+ LOG.debug("{}: handleForwardedReadyTransaction for {}", persistenceId(), forwardedReady.getTransactionID());
+
+ boolean isLeaderActive = isLeaderActive();
+ if (isLeader() && isLeaderActive) {
+ commitCoordinator.handleForwardedReadyTransaction(forwardedReady, getSender(), this);
+ } else {
+ ActorSelection leader = getLeader();
+ if (!isLeaderActive || leader == null) {
+ messageRetrySupport.addMessageToRetry(forwardedReady, getSender(),
+ "Could not commit transaction " + forwardedReady.getTransactionID());
+ } else {
+ LOG.debug("{}: Forwarding ForwardedReadyTransaction to leader {}", persistenceId(), leader);
+
+ ReadyLocalTransaction readyLocal = new ReadyLocalTransaction(forwardedReady.getTransactionID(),
+ forwardedReady.getTransaction().getSnapshot(), forwardedReady.isDoImmediateCommit());
+ readyLocal.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
+ leader.forward(readyLocal, getContext());
}
}
}
}
void doAbortTransaction(final String transactionID, final ActorRef sender) {
- final CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
- if(cohortEntry != null) {
- 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
- // succeeds.
- commitCoordinator.currentTransactionComplete(transactionID, false);
-
- final ListenableFuture<Void> future = cohortEntry.getCohort().abort();
- final ActorRef self = getSelf();
-
- Futures.addCallback(future, new FutureCallback<Void>() {
- @Override
- public void onSuccess(final Void v) {
- shardMBean.incrementAbortTransactionsCount();
-
- if(sender != null) {
- sender.tell(AbortTransactionReply.INSTANCE.toSerializable(), self);
- }
- }
-
- @Override
- public void onFailure(final Throwable t) {
- LOG.error("{}: An exception happened during abort", persistenceId(), t);
-
- if(sender != null) {
- sender.tell(new akka.actor.Status.Failure(t), self);
- }
- }
- });
- }
+ commitCoordinator.handleAbort(transactionID, sender, this);
}
private void handleCreateTransaction(final Object message) {
} else if (getLeader() != null) {
getLeader().forward(message, getContext());
} else {
- 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.", persistenceId()))), getSelf());
+ getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(
+ "Could not create a shard transaction", persistenceId())), getSelf());
}
}
}
private ActorRef createTypedTransactionActor(int transactionType,
- ShardTransactionIdentifier transactionId, String transactionChainId,
- short clientVersion ) {
+ ShardTransactionIdentifier transactionId, String transactionChainId) {
- return transactionActorFactory.newShardTransaction(TransactionProxy.TransactionType.fromInt(transactionType),
- transactionId, transactionChainId, clientVersion);
+ return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType),
+ transactionId, transactionChainId);
}
private void createTransaction(CreateTransaction createTransaction) {
try {
+ if(TransactionType.fromInt(createTransaction.getTransactionType()) != TransactionType.READ_ONLY &&
+ failIfIsolatedLeader(getSender())) {
+ return;
+ }
+
ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
- createTransaction.getTransactionId(), createTransaction.getTransactionChainId(),
- createTransaction.getVersion());
+ createTransaction.getTransactionId(), createTransaction.getTransactionChainId());
getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
- createTransaction.getTransactionId()).toSerializable(), getSelf());
+ createTransaction.getTransactionId(), createTransaction.getVersion()).toSerializable(), getSelf());
} catch (Exception e) {
getSender().tell(new akka.actor.Status.Failure(e), getSelf());
}
}
private ActorRef createTransaction(int transactionType, String remoteTransactionId,
- String transactionChainId, short clientVersion) {
+ String transactionChainId) {
ShardTransactionIdentifier transactionId = new ShardTransactionIdentifier(remoteTransactionId);
}
ActorRef transactionActor = createTypedTransactionActor(transactionType, transactionId,
- transactionChainId, clientVersion);
+ transactionChainId);
return transactionActor;
}
}
@Override
- protected RaftActorSnapshotCohort getRaftActorSnapshotCohort() {
+ @VisibleForTesting
+ public RaftActorSnapshotCohort getRaftActorSnapshotCohort() {
return snapshotCohort;
}
@Override
@Nonnull
protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
- return new ShardRecoveryCoordinator(store, persistenceId(), LOG);
+ return new ShardRecoveryCoordinator(store, store.getSchemaContext(),
+ restoreFromSnapshot != null ? restoreFromSnapshot.getSnapshot() : null, persistenceId(), LOG);
}
@Override
protected void onRecoveryComplete() {
+ restoreFromSnapshot = null;
+
//notify shard manager
getContext().parent().tell(new ActorInitialized(), getSelf());
// Replication consensus reached, proceed to commit
finishCommit(clientActor, identifier);
}
- } else if (data instanceof ModificationPayload) {
- try {
- applyModificationToState(clientActor, identifier, ((ModificationPayload) data).getModification());
- } catch (ClassNotFoundException | IOException e) {
- LOG.error("{}: Error extracting ModificationPayload", persistenceId(), e);
- }
} else if (data instanceof CompositeModificationPayload) {
Object modification = ((CompositeModificationPayload) data).getModification();
@Override
protected void onStateChanged() {
boolean isLeader = isLeader();
- changeSupport.onLeadershipChange(isLeader);
- treeChangeSupport.onLeadershipChange(isLeader);
+ boolean hasLeader = hasLeader();
+ changeSupport.onLeadershipChange(isLeader, hasLeader);
+ treeChangeSupport.onLeadershipChange(isLeader, hasLeader);
// If this actor is no longer the leader close all the transaction chains
if (!isLeader) {
}
store.closeAllTransactionChains();
+
+ commitCoordinator.abortPendingTransactions(
+ "The transacton was aborted due to inflight leadership change.", this);
+ }
+
+ if(hasLeader && !isIsolatedLeader()) {
+ messageRetrySupport.retryMessages();
}
}
@Override
protected void onLeaderChanged(String oldLeader, String newLeader) {
shardMBean.incrementLeadershipChangeCount();
+
+ if(hasLeader() && !isIsolatedLeader()) {
+ messageRetrySupport.retryMessages();
+ }
+ }
+
+ @Override
+ protected void pauseLeader(Runnable operation) {
+ LOG.debug("{}: In pauseLeader, operation: {}", persistenceId(), operation);
+ commitCoordinator.setRunOnPendingTransactionsComplete(operation);
}
@Override
return commitCoordinator;
}
+ public DatastoreContext getDatastoreContext() {
+ return datastoreContext;
+ }
+
+ @VisibleForTesting
+ public ShardDataTree getDataStore() {
+ return store;
+ }
+
+ @VisibleForTesting
+ ShardStats getShardMBean() {
+ return shardMBean;
+ }
+
+ public static Builder builder() {
+ return new Builder();
+ }
+
+ public static abstract class AbstractBuilder<T extends AbstractBuilder<T, S>, S extends Shard> {
+ private final Class<S> shardClass;
+ private ShardIdentifier id;
+ private Map<String, String> peerAddresses = Collections.emptyMap();
+ private DatastoreContext datastoreContext;
+ private SchemaContext schemaContext;
+ private DatastoreSnapshot.ShardSnapshot restoreFromSnapshot;
+ private volatile boolean sealed;
- private static class ShardCreator implements Creator<Shard> {
+ protected AbstractBuilder(Class<S> shardClass) {
+ this.shardClass = shardClass;
+ }
- private static final long serialVersionUID = 1L;
+ protected void checkSealed() {
+ Preconditions.checkState(!sealed, "Builder isalready sealed - further modifications are not allowed");
+ }
- final ShardIdentifier name;
- final Map<String, String> peerAddresses;
- final DatastoreContext datastoreContext;
- final SchemaContext schemaContext;
+ @SuppressWarnings("unchecked")
+ private T self() {
+ return (T) this;
+ }
- ShardCreator(final ShardIdentifier name, final Map<String, String> peerAddresses,
- final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
- this.name = name;
+ public T id(ShardIdentifier id) {
+ checkSealed();
+ this.id = id;
+ return self();
+ }
+
+ public T peerAddresses(Map<String, String> peerAddresses) {
+ checkSealed();
this.peerAddresses = peerAddresses;
+ return self();
+ }
+
+ public T datastoreContext(DatastoreContext datastoreContext) {
+ checkSealed();
this.datastoreContext = datastoreContext;
+ return self();
+ }
+
+ public T schemaContext(SchemaContext schemaContext) {
+ checkSealed();
this.schemaContext = schemaContext;
+ return self();
}
- @Override
- public Shard create() throws Exception {
- return new Shard(name, peerAddresses, datastoreContext, schemaContext);
+ public T restoreFromSnapshot(DatastoreSnapshot.ShardSnapshot restoreFromSnapshot) {
+ checkSealed();
+ this.restoreFromSnapshot = restoreFromSnapshot;
+ return self();
}
- }
- @VisibleForTesting
- public ShardDataTree getDataStore() {
- return store;
+ public ShardIdentifier getId() {
+ return id;
+ }
+
+ public Map<String, String> getPeerAddresses() {
+ return peerAddresses;
+ }
+
+ public DatastoreContext getDatastoreContext() {
+ return datastoreContext;
+ }
+
+ public SchemaContext getSchemaContext() {
+ return schemaContext;
+ }
+
+ public DatastoreSnapshot.ShardSnapshot getRestoreFromSnapshot() {
+ return restoreFromSnapshot;
+ }
+
+ public TreeType getTreeType() {
+ switch (datastoreContext.getLogicalStoreType()) {
+ case CONFIGURATION:
+ return TreeType.CONFIGURATION;
+ case OPERATIONAL:
+ return TreeType.OPERATIONAL;
+ }
+
+ throw new IllegalStateException("Unhandled logical store type " + datastoreContext.getLogicalStoreType());
+ }
+
+ protected void verify() {
+ Preconditions.checkNotNull(id, "id should not be null");
+ Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
+ Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
+ Preconditions.checkNotNull(schemaContext, "schemaContext should not be null");
+ }
+
+ public Props props() {
+ sealed = true;
+ verify();
+ return Props.create(shardClass, this);
+ }
}
- @VisibleForTesting
- ShardStats getShardMBean() {
- return shardMBean;
+ public static class Builder extends AbstractBuilder<Builder, Shard> {
+ private Builder() {
+ super(Shard.class);
+ }
}
}