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.collect.Lists;
-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.List;
+import java.util.Collections;
import java.util.Map;
-import java.util.concurrent.ExecutionException;
import java.util.concurrent.TimeUnit;
import javax.annotation.Nonnull;
import org.opendaylight.controller.cluster.common.actor.CommonConfig;
import org.opendaylight.controller.cluster.common.actor.MeteringBehavior;
import org.opendaylight.controller.cluster.datastore.ShardCommitCoordinator.CohortEntry;
-import org.opendaylight.controller.cluster.datastore.compat.BackwardsCompatibleThreePhaseCommitCohort;
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;
-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;
import org.opendaylight.controller.cluster.datastore.messages.CommitTransactionReply;
-import org.opendaylight.controller.cluster.datastore.messages.CreateSnapshot;
import org.opendaylight.controller.cluster.datastore.messages.CreateTransaction;
import org.opendaylight.controller.cluster.datastore.messages.CreateTransactionReply;
-import org.opendaylight.controller.cluster.datastore.messages.EnableNotification;
+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.ReadyTransactionReply;
+import org.opendaylight.controller.cluster.datastore.messages.ReadyLocalTransaction;
import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListener;
-import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListenerReply;
+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.datastore.utils.SerializationUtils;
+import org.opendaylight.controller.cluster.notifications.LeaderStateChanged;
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.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.controller.cluster.raft.protobuff.client.messages.Payload;
-import org.opendaylight.controller.md.sal.common.api.data.AsyncDataChangeListener;
-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.DOMStoreTransaction;
-import org.opendaylight.controller.sal.core.spi.data.DOMStoreWriteTransaction;
-import org.opendaylight.yangtools.concepts.ListenerRegistration;
-import org.opendaylight.yangtools.yang.data.api.YangInstanceIdentifier;
-import org.opendaylight.yangtools.yang.data.api.schema.NormalizedNode;
+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.model.api.SchemaContext;
import scala.concurrent.duration.Duration;
import scala.concurrent.duration.FiniteDuration;
/**
* A Shard represents a portion of the logical data tree <br/>
* <p>
- * Our Shard uses InMemoryDataStore as it's internal representation and delegates all requests it
+ * Our Shard uses InMemoryDataTree as it's internal representation and delegates all requests it
* </p>
*/
public class Shard extends RaftActor {
- private static final YangInstanceIdentifier DATASTORE_ROOT = YangInstanceIdentifier.builder().build();
+ protected static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = "txCommitTimeoutCheck";
- private 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";
// The state of this Shard
- private final InMemoryDOMDataStore store;
+ private final ShardDataTree store;
/// The name of this shard
private final String name;
private final ShardStats shardMBean;
- private final List<ActorSelection> dataChangeListeners = Lists.newArrayList();
-
- private final List<DelayedListenerRegistration> delayedListenerRegistrations =
- Lists.newArrayList();
-
private DatastoreContext datastoreContext;
- private SchemaContext schemaContext;
-
- private int createSnapshotTransactionCounter;
-
private final ShardCommitCoordinator commitCoordinator;
private long transactionCommitTimeout;
private final MessageTracker appendEntriesReplyTracker;
- private final ReadyTransactionReply READY_TRANSACTION_REPLY = new ReadyTransactionReply(
- Serialization.serializedActorPath(getSelf()));
+ private final ShardTransactionActorFactory transactionActorFactory;
+
+ private final ShardSnapshotCohort snapshotCohort;
+
+ private final DataTreeChangeListenerSupport treeChangeSupport = new DataTreeChangeListenerSupport(this);
+ private final DataChangeListenerSupport changeSupport = new DataChangeListenerSupport(this);
- /**
- * Coordinates persistence recovery on startup.
- */
- private ShardRecoveryCoordinator recoveryCoordinator;
+ private ShardSnapshot restoreFromSnapshot;
- private final DOMTransactionFactory transactionFactory;
- private final String txnDispatcherPath;
- 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()));
+ protected Shard(AbstractBuilder<?, ?> builder) {
+ super(builder.getId().toString(), builder.getPeerAddresses(),
+ Optional.of(builder.getDatastoreContext().getShardRaftConfig()), DataStoreVersions.CURRENT_VERSION);
- this.name = name.toString();
- this.datastoreContext = datastoreContext;
- this.schemaContext = schemaContext;
- this.txnDispatcherPath = new Dispatchers(context().system().dispatchers())
- .getDispatcherPath(Dispatchers.DispatcherType.Transaction);
+ 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 = InMemoryDOMDataStoreFactory.create(name.toString(), null,
- datastoreContext.getDataStoreProperties());
-
- if(schemaContext != null) {
- store.onGlobalContextUpdated(schemaContext);
- }
+ // FIXME: BUG-1014: pass down the proper TreeType
+ store = new ShardDataTree(builder.getSchemaContext());
shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
datastoreContext.getDataStoreMXBeanType());
- shardMBean.setNotificationManager(store.getDataChangeListenerNotificationManager());
- shardMBean.setShardActor(getSelf());
+ shardMBean.setShard(this);
if (isMetricsCaptureEnabled()) {
getContext().become(new MeteringBehavior(this));
}
- transactionFactory = new DOMTransactionFactory(store, shardMBean, LOG, this.name);
-
- commitCoordinator = new ShardCommitCoordinator(transactionFactory,
- TimeUnit.SECONDS.convert(5, TimeUnit.MINUTES),
- datastoreContext.getShardTransactionCommitQueueCapacity(), self(), LOG, this.name);
+ commitCoordinator = new ShardCommitCoordinator(store,
+ datastoreContext.getShardCommitQueueExpiryTimeoutInMillis(),
+ datastoreContext.getShardTransactionCommitQueueCapacity(), LOG, this.name);
setTransactionCommitTimeout();
appendEntriesReplyTracker = new MessageTracker(AppendEntriesReply.class,
getRaftActorContext().getConfigParams().getIsolatedCheckIntervalInMillis());
- recoveryCoordinator = new ShardRecoveryCoordinator(store, persistenceId(), LOG);
+ transactionActorFactory = new ShardTransactionActorFactory(store, datastoreContext,
+ new Dispatchers(context().system().dispatchers()).getDispatcherPath(
+ Dispatchers.DispatcherType.Transaction), self(), getContext(), shardMBean);
+
+ snapshotCohort = new ShardSnapshotCohort(transactionActorFactory, store, LOG, this.name);
+
+
+
}
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) {
if(context.error().isPresent()){
LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(),
- context.error());
+ context.error());
}
try {
} else if (BatchedModifications.class.isInstance(message)) {
handleBatchedModifications((BatchedModifications)message);
} else if (message instanceof ForwardedReadyTransaction) {
- handleForwardedReadyTransaction((ForwardedReadyTransaction) message);
+ commitCoordinator.handleForwardedReadyTransaction((ForwardedReadyTransaction) message,
+ getSender(), this);
+ } else if (message instanceof ReadyLocalTransaction) {
+ handleReadyLocalTransaction((ReadyLocalTransaction)message);
} else if (CanCommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
} else if (CommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
} else if (CloseTransactionChain.SERIALIZABLE_CLASS.isInstance(message)) {
closeTransactionChain(CloseTransactionChain.fromSerializable(message));
} else if (message instanceof RegisterChangeListener) {
- registerChangeListener((RegisterChangeListener) message);
+ changeSupport.onMessage((RegisterChangeListener) message, isLeader(), hasLeader());
+ } else if (message instanceof RegisterDataTreeChangeListener) {
+ treeChangeSupport.onMessage((RegisterDataTreeChangeListener) message, isLeader(), hasLeader());
} else if (message instanceof UpdateSchemaContext) {
updateSchemaContext((UpdateSchemaContext) message);
} else if (message instanceof PeerAddressResolved) {
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 {
super.onReceiveCommand(message);
}
}
}
+ private boolean hasLeader() {
+ return getLeaderId() != null;
+ }
+
+ public int getPendingTxCommitQueueSize() {
+ return commitCoordinator.getQueueSize();
+ }
+
@Override
protected Optional<ActorRef> getRoleChangeNotifier() {
return roleChangeNotifier;
}
- private void onDatastoreContext(DatastoreContext context) {
+ @Override
+ protected LeaderStateChanged newLeaderStateChanged(String memberId, String leaderId, short leaderPayloadVersion) {
+ return new ShardLeaderStateChanged(memberId, leaderId,
+ isLeader() ? Optional.<DataTree>of(store.getDataTree()) : Optional.<DataTree>absent(),
+ leaderPayloadVersion);
+ }
+
+ protected void onDatastoreContext(DatastoreContext context) {
datastoreContext = context;
commitCoordinator.setQueueCapacity(datastoreContext.getShardTransactionCommitQueueCapacity());
private void handleTransactionCommitTimeoutCheck() {
CohortEntry cohortEntry = commitCoordinator.getCurrentCohortEntry();
if(cohortEntry != null) {
- long elapsed = System.currentTimeMillis() - cohortEntry.getLastAccessTime();
- if(elapsed > transactionCommitTimeout) {
+ if(cohortEntry.isExpired(transactionCommitTimeout)) {
LOG.warn("{}: Current transaction {} has timed out after {} ms - aborting",
persistenceId(), cohortEntry.getTransactionID(), transactionCommitTimeout);
doAbortTransaction(cohortEntry.getTransactionID(), null);
}
}
+
+ commitCoordinator.cleanupExpiredCohortEntries();
+ }
+
+ private static boolean isEmptyCommit(final DataTreeCandidate candidate) {
+ return ModificationType.UNMODIFIED.equals(candidate.getRootNode().getModificationType());
+ }
+
+ 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));
+ }
}
private void handleCommitTransaction(final CommitTransaction commit) {
- final String transactionID = commit.getTransactionID();
-
- LOG.debug("{}: Committing transaction {}", persistenceId(), transactionID);
-
- // Get the current in-progress cohort entry in the commitCoordinator if it corresponds to
- // this transaction.
- final CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
- if(cohortEntry == null) {
- // 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("%s: Cannot commit transaction %s - it is not the current transaction",
- persistenceId(), transactionID));
- LOG.error(ex.getMessage());
+ if(!commitCoordinator.handleCommit(commit.getTransactionID(), getSender(), this)) {
shardMBean.incrementFailedTransactionsCount();
- getSender().tell(new akka.actor.Status.Failure(ex), getSelf());
- return;
}
+ }
- // We perform the preCommit phase here atomically with the commit phase. This is an
- // optimization to eliminate the overhead of an extra preCommit message. We lose front-end
- // coordination of preCommit across shards in case of failure but preCommit should not
- // normally fail since we ensure only one concurrent 3-phase commit.
+ 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 {
- // 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().preCommit().get();
-
- // 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()));
- }
+ cohortEntry.commit();
+
+ sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
+
+ shardMBean.incrementCommittedTransactionCount();
+ shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
+
} catch (Exception e) {
- LOG.error("{} An exception occurred while preCommitting transaction {}",
- persistenceId(), cohortEntry.getTransactionID(), e);
+ sender.tell(new akka.actor.Status.Failure(e), getSelf());
+
+ LOG.error("{}, An exception occurred while committing transaction {}", persistenceId(),
+ transactionID, e);
shardMBean.incrementFailedTransactionsCount();
- getSender().tell(new akka.actor.Status.Failure(e), getSelf());
+ } finally {
+ commitCoordinator.currentTransactionComplete(transactionID, true);
}
-
- cohortEntry.updateLastAccessTime();
}
private void finishCommit(@Nonnull final ActorRef sender, final @Nonnull String transactionID) {
// after the commit has been replicated to a majority of the followers.
CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
- if(cohortEntry == null) {
+ 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
// transaction.
cohortEntry = commitCoordinator.getAndRemoveCohortEntry(transactionID);
if(cohortEntry != null) {
- commitWithNewTransaction(cohortEntry.getModification());
+ 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());
} else {
// This really shouldn't happen - it likely means that persistence or replication
LOG.error(ex.getMessage());
sender.tell(new akka.actor.Status.Failure(ex), getSelf());
}
-
- return;
+ } else {
+ finishCommit(sender, transactionID, cohortEntry);
}
+ }
- 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();
-
- sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
+ private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
+ LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionID());
+ commitCoordinator.handleCanCommit(canCommit.getTransactionID(), getSender(), this);
+ }
- shardMBean.incrementCommittedTransactionCount();
- shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
+ private void noLeaderError(String errMessage, Object message) {
+ // 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(errMessage, persistenceId())), getSelf());
+ }
+ 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());
-
- LOG.error("{}, An exception occurred while committing transaction {}", persistenceId(),
- transactionID, e);
- shardMBean.incrementFailedTransactionsCount();
- } finally {
- commitCoordinator.currentTransactionComplete(transactionID, true);
}
}
- private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
- 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
+ // 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
// 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());
- }
+ failIfIsolatedLeader(getSender());
+
+ handleBatchedModificationsLocal(batched, getSender());
} else {
ActorSelection leader = getLeader();
if(leader != null) {
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());
+ noLeaderError("Could not commit transaction " + batched.getTransactionID(), batched);
}
}
}
- private void handleForwardedReadyTransaction(ForwardedReadyTransaction ready) {
- LOG.debug("{}: Readying transaction {}, client version {}", persistenceId(),
- ready.getTransactionID(), ready.getTxnClientVersion());
+ 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;
+ }
- // 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(),
- (MutableCompositeModification) ready.getModification());
+ return false;
+ }
- // 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.
- if(ready.getTxnClientVersion() < DataStoreVersions.HELIUM_1_VERSION) {
- LOG.debug("{}: Creating BackwardsCompatibleThreePhaseCommitCohort", persistenceId());
- ActorRef replyActorPath = getContext().actorOf(BackwardsCompatibleThreePhaseCommitCohort.props(
- ready.getTransactionID()));
+ protected boolean isIsolatedLeader() {
+ return getRaftState() == RaftState.IsolatedLeader;
+ }
- ReadyTransactionReply readyTransactionReply =
- new ReadyTransactionReply(Serialization.serializedActorPath(replyActorPath));
- getSender().tell(ready.isReturnSerialized() ? readyTransactionReply.toSerializable() :
- readyTransactionReply, getSelf());
+ private void handleReadyLocalTransaction(final ReadyLocalTransaction message) {
+ if (isLeader()) {
+ failIfIsolatedLeader(getSender());
+ 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 {
-
- getSender().tell(ready.isReturnSerialized() ? READY_TRANSACTION_REPLY.toSerializable() :
- READY_TRANSACTION_REPLY, getSelf());
+ ActorSelection leader = getLeader();
+ if (leader != null) {
+ LOG.debug("{}: Forwarding ReadyLocalTransaction to leader {}", persistenceId(), leader);
+ message.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
+ leader.forward(message, getContext());
+ } else {
+ noLeaderError("Could not commit transaction " + message.getTransactionID(), message);
+ }
}
}
}
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 void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
- transactionFactory.closeTransactionChain(closeTransactionChain.getTransactionChainId());
+ store.closeTransactionChain(closeTransactionChain.getTransactionChainId());
}
private ActorRef createTypedTransactionActor(int transactionType,
ShardTransactionIdentifier transactionId, String transactionChainId,
short clientVersion ) {
- DOMStoreTransaction transaction = transactionFactory.newTransaction(
- TransactionProxy.TransactionType.fromInt(transactionType), transactionId.toString(),
- transactionChainId);
-
- return createShardTransaction(transaction, transactionId, clientVersion);
- }
-
- private ActorRef createShardTransaction(DOMStoreTransaction transaction, ShardTransactionIdentifier transactionId,
- short clientVersion){
- return getContext().actorOf(
- ShardTransaction.props(transaction, getSelf(),
- schemaContext, datastoreContext, shardMBean,
- transactionId.getRemoteTransactionId(), clientVersion)
- .withDispatcher(txnDispatcherPath),
- transactionId.toString());
-
+ return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType),
+ transactionId, transactionChainId, clientVersion);
}
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());
return transactionActor;
}
- private void syncCommitTransaction(final DOMStoreWriteTransaction transaction)
- throws ExecutionException, InterruptedException {
- DOMStoreThreePhaseCommitCohort commitCohort = transaction.ready();
- commitCohort.preCommit().get();
- commitCohort.commit().get();
- }
-
private void commitWithNewTransaction(final Modification modification) {
- DOMStoreWriteTransaction tx = store.newWriteOnlyTransaction();
- modification.apply(tx);
+ ReadWriteShardDataTreeTransaction tx = store.newReadWriteTransaction(modification.toString(), null);
+ modification.apply(tx.getSnapshot());
try {
- syncCommitTransaction(tx);
+ snapshotCohort.syncCommitTransaction(tx);
shardMBean.incrementCommittedTransactionCount();
shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
- } catch (InterruptedException | ExecutionException e) {
+ } catch (Exception e) {
shardMBean.incrementFailedTransactionsCount();
LOG.error("{}: Failed to commit", persistenceId(), e);
}
}
private void updateSchemaContext(final UpdateSchemaContext message) {
- this.schemaContext = message.getSchemaContext();
updateSchemaContext(message.getSchemaContext());
- store.onGlobalContextUpdated(message.getSchemaContext());
}
@VisibleForTesting
void updateSchemaContext(final SchemaContext schemaContext) {
- store.onGlobalContextUpdated(schemaContext);
- }
-
- private void registerChangeListener(final RegisterChangeListener registerChangeListener) {
-
- LOG.debug("{}: registerDataChangeListener for {}", persistenceId(), registerChangeListener.getPath());
-
- ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
- NormalizedNode<?, ?>>> registration;
- if(isLeader()) {
- registration = doChangeListenerRegistration(registerChangeListener);
- } else {
- LOG.debug("{}: Shard is not the leader - delaying registration", persistenceId());
-
- DelayedListenerRegistration delayedReg =
- new DelayedListenerRegistration(registerChangeListener);
- delayedListenerRegistrations.add(delayedReg);
- registration = delayedReg;
- }
-
- ActorRef listenerRegistration = getContext().actorOf(
- DataChangeListenerRegistration.props(registration));
-
- LOG.debug("{}: registerDataChangeListener sending reply, listenerRegistrationPath = {} ",
- persistenceId(), listenerRegistration.path());
-
- getSender().tell(new RegisterChangeListenerReply(listenerRegistration), getSelf());
+ store.updateSchemaContext(schemaContext);
}
- private ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
- NormalizedNode<?, ?>>> doChangeListenerRegistration(
- final RegisterChangeListener registerChangeListener) {
-
- ActorSelection dataChangeListenerPath = getContext().system().actorSelection(
- registerChangeListener.getDataChangeListenerPath());
-
- // Notify the listener if notifications should be enabled or not
- // If this shard is the leader then it will enable notifications else
- // it will not
- dataChangeListenerPath.tell(new EnableNotification(true), getSelf());
-
- // Now store a reference to the data change listener so it can be notified
- // at a later point if notifications should be enabled or disabled
- dataChangeListeners.add(dataChangeListenerPath);
-
- AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>> listener =
- new DataChangeListenerProxy(dataChangeListenerPath);
-
- LOG.debug("{}: Registering for path {}", persistenceId(), registerChangeListener.getPath());
-
- return store.registerChangeListener(registerChangeListener.getPath(), listener,
- registerChangeListener.getScope());
- }
-
- private boolean isMetricsCaptureEnabled(){
+ private boolean isMetricsCaptureEnabled() {
CommonConfig config = new CommonConfig(getContext().system().settings().config());
return config.isMetricCaptureEnabled();
}
@Override
- protected
- void startLogRecoveryBatch(final int maxBatchSize) {
- recoveryCoordinator.startLogRecoveryBatch(maxBatchSize);
- }
-
- @Override
- protected void appendRecoveredLogEntry(final Payload data) {
- recoveryCoordinator.appendRecoveredLogPayload(data);
- }
-
- @Override
- protected void applyRecoverySnapshot(final byte[] snapshotBytes) {
- recoveryCoordinator.applyRecoveredSnapshot(snapshotBytes);
+ @VisibleForTesting
+ public RaftActorSnapshotCohort getRaftActorSnapshotCohort() {
+ return snapshotCohort;
}
@Override
- protected void applyCurrentLogRecoveryBatch() {
- recoveryCoordinator.applyCurrentLogRecoveryBatch();
+ @Nonnull
+ protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
+ return new ShardRecoveryCoordinator(store, store.getSchemaContext(),
+ restoreFromSnapshot != null ? restoreFromSnapshot.getSnapshot() : null, persistenceId(), LOG);
}
@Override
protected void onRecoveryComplete() {
- recoveryCoordinator = null;
+ restoreFromSnapshot = null;
//notify shard manager
getContext().parent().tell(new ActorInitialized(), getSelf());
@Override
protected void applyState(final ActorRef clientActor, final String identifier, final Object data) {
-
- if(data instanceof ModificationPayload) {
+ 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);
+ }
+ } 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) {
+ } else if (data instanceof CompositeModificationPayload) {
Object modification = ((CompositeModificationPayload) data).getModification();
applyModificationToState(clientActor, identifier, modification);
}
}
- @Override
- protected void createSnapshot() {
- // Create a transaction actor. We are really going to treat the transaction as a worker
- // so that this actor does not get block building the snapshot. THe transaction actor will
- // after processing the CreateSnapshot message.
-
- ActorRef createSnapshotTransaction = createTransaction(
- TransactionProxy.TransactionType.READ_ONLY.ordinal(),
- "createSnapshot" + ++createSnapshotTransactionCounter, "",
- DataStoreVersions.CURRENT_VERSION);
-
- createSnapshotTransaction.tell(CreateSnapshot.INSTANCE, self());
- }
-
- @VisibleForTesting
- @Override
- protected void applySnapshot(final byte[] snapshotBytes) {
- // Since this will be done only on Recovery or when this actor is a Follower
- // 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", persistenceId());
- try {
- DOMStoreWriteTransaction transaction = store.newWriteOnlyTransaction();
-
- NormalizedNode<?, ?> node = SerializationUtils.deserializeNormalizedNode(snapshotBytes);
-
- // delete everything first
- transaction.delete(DATASTORE_ROOT);
-
- // Add everything from the remote node back
- transaction.write(DATASTORE_ROOT, node);
- syncCommitTransaction(transaction);
- } catch (InterruptedException | ExecutionException e) {
- LOG.error("{}: An exception occurred when applying snapshot", persistenceId(), e);
- } finally {
- LOG.info("{}: Done applying snapshot", persistenceId());
- }
- }
-
@Override
protected void onStateChanged() {
boolean isLeader = isLeader();
- for (ActorSelection dataChangeListener : dataChangeListeners) {
- dataChangeListener.tell(new EnableNotification(isLeader), getSelf());
- }
-
- if(isLeader) {
- for(DelayedListenerRegistration reg: delayedListenerRegistrations) {
- if(!reg.isClosed()) {
- reg.setDelegate(doChangeListenerRegistration(reg.getRegisterChangeListener()));
- }
- }
-
- delayedListenerRegistrations.clear();
- }
+ 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) {
+ if (!isLeader) {
if(LOG.isDebugEnabled()) {
LOG.debug(
"{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader",
persistenceId(), getId());
}
- transactionFactory.closeAllTransactionChains();
+ store.closeAllTransactionChains();
}
}
+ @Override
+ protected void onLeaderChanged(String oldLeader, String newLeader) {
+ shardMBean.incrementLeadershipChangeCount();
+ }
+
@Override
public String persistenceId() {
return this.name;
return commitCoordinator;
}
-
- private static class ShardCreator implements Creator<Shard> {
-
- private static final long serialVersionUID = 1L;
-
- final ShardIdentifier name;
- final Map<String, String> peerAddresses;
- final DatastoreContext datastoreContext;
- final SchemaContext schemaContext;
-
- ShardCreator(final ShardIdentifier name, final Map<String, String> peerAddresses,
- final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
- this.name = name;
- this.peerAddresses = peerAddresses;
- this.datastoreContext = datastoreContext;
- this.schemaContext = schemaContext;
- }
-
- @Override
- public Shard create() throws Exception {
- return new Shard(name, peerAddresses, datastoreContext, schemaContext);
- }
+ public DatastoreContext getDatastoreContext() {
+ return datastoreContext;
}
@VisibleForTesting
- public InMemoryDOMDataStore getDataStore() {
+ public ShardDataTree getDataStore() {
return store;
}
return shardMBean;
}
- private static class DelayedListenerRegistration implements
- ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>>> {
+ public static Builder builder() {
+ return new Builder();
+ }
- private volatile boolean closed;
+ 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 final RegisterChangeListener registerChangeListener;
+ protected AbstractBuilder(Class<S> shardClass) {
+ this.shardClass = shardClass;
+ }
- private volatile ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
- NormalizedNode<?, ?>>> delegate;
+ protected void checkSealed() {
+ Preconditions.checkState(!sealed, "Builder isalready sealed - further modifications are not allowed");
+ }
- DelayedListenerRegistration(final RegisterChangeListener registerChangeListener) {
- this.registerChangeListener = registerChangeListener;
+ @SuppressWarnings("unchecked")
+ private T self() {
+ return (T) this;
}
- void setDelegate( final ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
- NormalizedNode<?, ?>>> registration) {
- this.delegate = registration;
+ public T id(ShardIdentifier id) {
+ checkSealed();
+ this.id = id;
+ return self();
}
- boolean isClosed() {
- return closed;
+ public T peerAddresses(Map<String, String> peerAddresses) {
+ checkSealed();
+ this.peerAddresses = peerAddresses;
+ return self();
}
- RegisterChangeListener getRegisterChangeListener() {
- return registerChangeListener;
+ public T datastoreContext(DatastoreContext datastoreContext) {
+ checkSealed();
+ this.datastoreContext = datastoreContext;
+ return self();
}
- @Override
- public AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>> getInstance() {
- return delegate != null ? delegate.getInstance() : null;
+ public T schemaContext(SchemaContext schemaContext) {
+ checkSealed();
+ this.schemaContext = schemaContext;
+ return self();
}
- @Override
- public void close() {
- closed = true;
- if(delegate != null) {
- delegate.close();
- }
+ public T restoreFromSnapshot(DatastoreSnapshot.ShardSnapshot restoreFromSnapshot) {
+ checkSealed();
+ this.restoreFromSnapshot = restoreFromSnapshot;
+ return self();
+ }
+
+ 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;
+ }
+
+ 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);
+ }
+ }
+
+ public static class Builder extends AbstractBuilder<Builder, Shard> {
+ private Builder() {
+ super(Shard.class);
}
}
}