import akka.actor.ActorSelection;
import akka.actor.Cancellable;
import akka.actor.Props;
+import akka.actor.Status;
import akka.actor.Status.Failure;
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 com.google.common.base.Verify;
import com.google.common.collect.ImmutableList;
+import com.google.common.collect.ImmutableMap;
import com.google.common.collect.Range;
import java.io.IOException;
import java.util.Arrays;
import java.util.Collection;
import java.util.Collections;
-import java.util.HashMap;
import java.util.Map;
import java.util.concurrent.TimeUnit;
import javax.annotation.Nonnull;
import org.opendaylight.controller.cluster.datastore.messages.CreateTransactionReply;
import org.opendaylight.controller.cluster.datastore.messages.ForwardedReadyTransaction;
import org.opendaylight.controller.cluster.datastore.messages.GetShardDataTree;
+import org.opendaylight.controller.cluster.datastore.messages.MakeLeaderLocal;
import org.opendaylight.controller.cluster.datastore.messages.OnDemandShardState;
import org.opendaylight.controller.cluster.datastore.messages.PeerAddressResolved;
+import org.opendaylight.controller.cluster.datastore.messages.PersistAbortTransactionPayload;
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.persisted.AbortTransactionPayload;
import org.opendaylight.controller.cluster.datastore.persisted.DatastoreSnapshot;
import org.opendaylight.controller.cluster.datastore.persisted.DatastoreSnapshot.ShardSnapshot;
import org.opendaylight.controller.cluster.datastore.utils.Dispatchers;
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.LeadershipTransferFailedException;
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.base.messages.FollowerInitialSyncUpStatus;
import org.opendaylight.controller.cluster.raft.client.messages.OnDemandRaftState;
import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
+import org.opendaylight.controller.cluster.raft.messages.RequestLeadership;
import org.opendaylight.controller.cluster.raft.messages.ServerRemoved;
import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload;
import org.opendaylight.yangtools.concepts.Identifier;
}
};
+ static final Object RESUME_NEXT_PENDING_TRANSACTION = new Object() {
+ @Override
+ public String toString() {
+ return "resumeNextPendingTransaction";
+ }
+ };
+
// 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";
private final ShardTransactionMessageRetrySupport messageRetrySupport;
- private final FrontendMetadata frontendMetadata = new FrontendMetadata();
- private final Map<FrontendIdentifier, LeaderFrontendState> knownFrontends = new HashMap<>();
+ private final FrontendMetadata frontendMetadata;
+ private Map<FrontendIdentifier, LeaderFrontendState> knownFrontends = ImmutableMap.of();
protected Shard(final AbstractBuilder<?, ?> builder) {
super(builder.getId().toString(), builder.getPeerAddresses(),
this.name = builder.getId().toString();
this.datastoreContext = builder.getDatastoreContext();
this.restoreFromSnapshot = builder.getRestoreFromSnapshot();
+ this.frontendMetadata = new FrontendMetadata(name);
setPersistence(datastoreContext.isPersistent());
LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
ShardDataTreeChangeListenerPublisherActorProxy treeChangeListenerPublisher =
- new ShardDataTreeChangeListenerPublisherActorProxy(getContext(), name + "-DTCL-publisher");
+ new ShardDataTreeChangeListenerPublisherActorProxy(getContext(), name + "-DTCL-publisher", name);
ShardDataChangeListenerPublisherActorProxy dataChangeListenerPublisher =
- new ShardDataChangeListenerPublisherActorProxy(getContext(), name + "-DCL-publisher");
+ new ShardDataChangeListenerPublisherActorProxy(getContext(), name + "-DCL-publisher", name);
if (builder.getDataTree() != null) {
store = new ShardDataTree(this, builder.getSchemaContext(), builder.getDataTree(),
- treeChangeListenerPublisher, dataChangeListenerPublisher, name);
+ treeChangeListenerPublisher, dataChangeListenerPublisher, name, frontendMetadata);
} else {
store = new ShardDataTree(this, builder.getSchemaContext(), builder.getTreeType(),
builder.getDatastoreContext().getStoreRoot(), treeChangeListenerPublisher,
- dataChangeListenerPublisher, name);
+ dataChangeListenerPublisher, name, frontendMetadata);
}
shardMBean = ShardMBeanFactory.getShardStatsMBean(name, datastoreContext.getDataStoreMXBeanType(), this);
getRaftActorContext().getConfigParams().getIsolatedCheckIntervalInMillis());
transactionActorFactory = new ShardTransactionActorFactory(store, datastoreContext,
- new Dispatchers(context().system().dispatchers()).getDispatcherPath(
- Dispatchers.DispatcherType.Transaction), self(), getContext(), shardMBean);
+ new Dispatchers(context().system().dispatchers()).getDispatcherPath(Dispatchers.DispatcherType.Transaction),
+ self(), getContext(), shardMBean, builder.getId().getShardName());
snapshotCohort = ShardSnapshotCohort.create(getContext(), builder.getId().getMemberName(), store, LOG,
this.name);
maybeError.get());
}
+ store.resetTransactionBatch();
+
if (message instanceof RequestEnvelope) {
final long now = ticker().read();
final RequestEnvelope envelope = (RequestEnvelope)message;
} else if (message instanceof DataTreeCohortActorRegistry.CohortRegistryCommand) {
store.processCohortRegistryCommand(getSender(),
(DataTreeCohortActorRegistry.CohortRegistryCommand) message);
+ } else if (message instanceof PersistAbortTransactionPayload) {
+ final TransactionIdentifier txId = ((PersistAbortTransactionPayload) message).getTransactionId();
+ persistPayload(txId, AbortTransactionPayload.create(txId), true);
+ } else if (message instanceof MakeLeaderLocal) {
+ onMakeLeaderLocal();
+ } else if (RESUME_NEXT_PENDING_TRANSACTION.equals(message)) {
+ store.resumeNextPendingTransaction();
} else {
super.handleNonRaftCommand(message);
}
}
}
+ private void onMakeLeaderLocal() {
+ LOG.debug("{}: onMakeLeaderLocal received", persistenceId());
+ if (isLeader()) {
+ getSender().tell(new Status.Success(null), getSelf());
+ return;
+ }
+
+ final ActorSelection leader = getLeader();
+
+ if (leader == null) {
+ // Leader is not present. The cluster is most likely trying to
+ // elect a leader and we should let that run its normal course
+
+ // TODO we can wait for the election to complete and retry the
+ // request. We can also let the caller retry by sending a flag
+ // in the response indicating the request is "reTryable".
+ getSender().tell(new Failure(
+ new LeadershipTransferFailedException("We cannot initiate leadership transfer to local node. "
+ + "Currently there is no leader for " + persistenceId())),
+ getSelf());
+ return;
+ }
+
+ leader.tell(new RequestLeadership(getId(), getSender()), getSelf());
+ }
+
// Acquire our frontend tracking handle and verify generation matches
private LeaderFrontendState getFrontend(final ClientIdentifier clientId) throws RequestException {
final LeaderFrontendState existing = knownFrontends.get(clientId.getFrontendId());
final ClientIdentifier clientId = lhReq.getTarget().getClientId();
return getFrontend(clientId).handleLocalHistoryRequest(lhReq, envelope, now);
} else {
- LOG.debug("{}: rejecting unsupported request {}", persistenceId(), request);
+ LOG.warn("{}: rejecting unsupported request {}", persistenceId(), request);
throw new UnsupportedRequestException(request);
}
}
persistenceId(), getId());
}
- store.closeAllTransactionChains();
+ store.purgeLeaderState();
}
if (hasLeader && !isIsolatedLeader()) {
protected void onLeaderChanged(final String oldLeader, final String newLeader) {
shardMBean.incrementLeadershipChangeCount();
- boolean hasLeader = hasLeader();
- if (hasLeader && !isLeader()) {
+ final boolean hasLeader = hasLeader();
+ if (!hasLeader) {
+ // No leader implies we are not the leader, lose frontend state if we have any. This also places
+ // an explicit guard so the map will not get modified accidentally.
+ if (!knownFrontends.isEmpty()) {
+ LOG.debug("{}: removing frontend state for {}", persistenceId(), knownFrontends.keySet());
+ knownFrontends = ImmutableMap.of();
+ }
+ return;
+ }
+
+ if (!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();
commitCoordinator.abortPendingTransactions("The transacton was aborted due to inflight leadership "
+ "change and the leader address isn't available.", this);
}
+ } else {
+ // We have become the leader, we need to reconstruct frontend state
+ knownFrontends = Verify.verifyNotNull(frontendMetadata.toLeaderState(this));
+ LOG.debug("{}: became leader with frontend state for {}", persistenceId(), knownFrontends.keySet());
}
- if (hasLeader && !isIsolatedLeader()) {
+ if (!isIsolatedLeader()) {
messageRetrySupport.retryMessages();
}
}
Ticker ticker() {
return Ticker.systemTicker();
}
+
+ void scheduleNextPendingTransaction() {
+ self().tell(RESUME_NEXT_PENDING_TRANSACTION, ActorRef.noSender());
+ }
}