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.Map;
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.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.protobuff.client.messages.CompositeModificationByteStringPayload;
*/
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";
}
void continueCommit(final CohortEntry cohortEntry) throws Exception {
- final DataTreeCandidate candidate = cohortEntry.getCohort().getCandidate();
+ final DataTreeCandidate candidate = cohortEntry.getCandidate();
// If we do not have any followers and we are not using persistence
// or if cohortEntry has no modifications
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());
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);
commitCoordinator.handleCanCommit(canCommit.getTransactionID(), getSender(), this);
}
- private void noLeaderError(Object message) {
+ 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(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());
+ getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(errMessage, persistenceId())), getSelf());
}
private void handleBatchedModifications(BatchedModifications batched) {
// window where it could have a stale leader during leadership transitions.
//
if(isLeader()) {
+ failIfIsolatedLeader(getSender());
+
try {
commitCoordinator.handleBatchedModifications(batched, getSender(), this);
} catch (Exception e) {
LOG.debug("{}: Forwarding BatchedModifications to leader {}", persistenceId(), leader);
leader.forward(batched, getContext());
} else {
- noLeaderError(batched);
+ noLeaderError("Could not commit transaction " + batched.getTransactionID(), batched);
}
}
}
+ private boolean failIfIsolatedLeader(ActorRef sender) {
+ if(getRaftState() == RaftState.IsolatedLeader) {
+ 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;
+ }
+
private void handleReadyLocalTransaction(final ReadyLocalTransaction message) {
if (isLeader()) {
+ failIfIsolatedLeader(getSender());
+
try {
commitCoordinator.handleReadyLocalTransaction(message, getSender(), this);
} catch (Exception e) {
message.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
leader.forward(message, getContext());
} else {
- noLeaderError(message);
+ 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 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());