import akka.actor.ActorRef;
import akka.actor.ActorSelection;
-import akka.actor.PoisonPill;
+import akka.actor.Cancellable;
import akka.actor.Props;
-import akka.event.Logging;
-import akka.event.LoggingAdapter;
-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.CheckedFuture;
-import com.google.common.util.concurrent.FutureCallback;
-import com.google.common.util.concurrent.Futures;
-import com.google.common.util.concurrent.ListenableFuture;
-import com.google.protobuf.ByteString;
-import com.google.protobuf.InvalidProtocolBufferException;
+import java.io.IOException;
+import java.util.Collection;
+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.MessageTracker;
+import org.opendaylight.controller.cluster.common.actor.MessageTracker.Error;
import org.opendaylight.controller.cluster.common.actor.MeteringBehavior;
+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.ActorInitialized;
+import org.opendaylight.controller.cluster.datastore.messages.BatchedModifications;
+import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransaction;
import org.opendaylight.controller.cluster.datastore.messages.CloseTransactionChain;
+import org.opendaylight.controller.cluster.datastore.messages.CommitTransaction;
import org.opendaylight.controller.cluster.datastore.messages.CommitTransactionReply;
import org.opendaylight.controller.cluster.datastore.messages.CreateTransaction;
import org.opendaylight.controller.cluster.datastore.messages.CreateTransactionReply;
-import org.opendaylight.controller.cluster.datastore.messages.EnableNotification;
-import org.opendaylight.controller.cluster.datastore.messages.ForwardedCommitTransaction;
+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.ReadData;
-import org.opendaylight.controller.cluster.datastore.messages.ReadDataReply;
+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.MutableCompositeModification;
-import org.opendaylight.controller.cluster.datastore.node.NormalizedNodeToNodeCodec;
+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.RaftActor;
-import org.opendaylight.controller.cluster.raft.ReplicatedLogEntry;
-import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshotReply;
-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.common.api.data.ReadFailedException;
-import org.opendaylight.controller.md.sal.dom.store.impl.InMemoryDOMDataStore;
-import org.opendaylight.controller.md.sal.dom.store.impl.InMemoryDOMDataStoreFactory;
-import org.opendaylight.controller.protobuff.messages.common.NormalizedNodeMessages;
-import org.opendaylight.controller.sal.core.spi.data.DOMStoreReadTransaction;
-import org.opendaylight.controller.sal.core.spi.data.DOMStoreThreePhaseCommitCohort;
-import org.opendaylight.controller.sal.core.spi.data.DOMStoreTransactionChain;
-import org.opendaylight.controller.sal.core.spi.data.DOMStoreTransactionFactory;
-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.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.yangtools.concepts.Identifier;
+import org.opendaylight.yangtools.util.StringIdentifier;
+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 java.util.ArrayList;
-import java.util.Collection;
-import java.util.HashMap;
-import java.util.List;
-import java.util.Map;
-import java.util.concurrent.ExecutionException;
+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 {
+ @VisibleForTesting
+ static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = new Object() {
+ @Override
+ public String toString() {
+ return "txCommitTimeoutCheck";
+ }
+ };
+
+ @VisibleForTesting
+ static final Object GET_SHARD_MBEAN_MESSAGE = new Object() {
+ @Override
+ public String toString() {
+ return "getShardMBeanMessage";
+ }
+ };
+
+ // FIXME: shard names should be encapsulated in their own class and this should be exposed as a constant.
public static final String DEFAULT_NAME = "default";
// The state of this Shard
- private final InMemoryDOMDataStore store;
+ private final ShardDataTree store;
- private final Map<Object, DOMStoreThreePhaseCommitCohort>
- modificationToCohort = new HashMap<>();
+ /// The name of this shard
+ private final String name;
- private final LoggingAdapter LOG =
- Logging.getLogger(getContext().system(), this);
+ private final ShardStats shardMBean;
- // By default persistent will be true and can be turned off using the system
- // property shard.persistent
- private final boolean persistent;
+ private DatastoreContext datastoreContext;
- /// The name of this shard
- private final ShardIdentifier name;
+ private final ShardCommitCoordinator commitCoordinator;
- private final ShardStats shardMBean;
+ private long transactionCommitTimeout;
- private final List<ActorSelection> dataChangeListeners = new ArrayList<>();
+ private Cancellable txCommitTimeoutCheckSchedule;
- private final DatastoreContext datastoreContext;
+ private final Optional<ActorRef> roleChangeNotifier;
- private SchemaContext schemaContext;
+ private final MessageTracker appendEntriesReplyTracker;
- private ActorRef createSnapshotTransaction;
+ private final ShardTransactionActorFactory transactionActorFactory;
- /**
- * Coordinates persistence recovery on startup.
- */
- private ShardRecoveryCoordinator recoveryCoordinator;
- private List<Object> currentLogRecoveryBatch;
+ private final ShardSnapshotCohort snapshotCohort;
- private final Map<String, DOMStoreTransactionChain> transactionChains = new HashMap<>();
+ private final DataTreeChangeListenerSupport treeChangeSupport = new DataTreeChangeListenerSupport(this);
+ private final DataChangeListenerSupport changeSupport = new DataChangeListenerSupport(this);
- protected Shard(ShardIdentifier name, Map<ShardIdentifier, String> peerAddresses,
- DatastoreContext datastoreContext, SchemaContext schemaContext) {
- super(name.toString(), mapPeerAddresses(peerAddresses),
- Optional.of(datastoreContext.getShardRaftConfig()));
- this.name = name;
- this.datastoreContext = datastoreContext;
- this.schemaContext = schemaContext;
+ private ShardSnapshot restoreFromSnapshot;
- String setting = System.getProperty("shard.persistent");
+ private final ShardTransactionMessageRetrySupport messageRetrySupport;
- this.persistent = !"false".equals(setting);
+ protected Shard(AbstractBuilder<?, ?> builder) {
+ super(builder.getId().toString(), builder.getPeerAddresses(),
+ Optional.of(builder.getDatastoreContext().getShardRaftConfig()), DataStoreVersions.CURRENT_VERSION);
- LOG.info("Shard created : {} persistent : {}", name, persistent);
+ this.name = builder.getId().toString();
+ this.datastoreContext = builder.getDatastoreContext();
+ this.restoreFromSnapshot = builder.getRestoreFromSnapshot();
- store = InMemoryDOMDataStoreFactory.create(name.toString(), null,
- datastoreContext.getDataStoreProperties());
+ setPersistence(datastoreContext.isPersistent());
- if(schemaContext != null) {
- store.onGlobalContextUpdated(schemaContext);
- }
+ LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
+
+ store = new ShardDataTree(builder.getSchemaContext(), builder.getTreeType(),
+ new ShardDataTreeChangeListenerPublisherActorProxy(getContext(), name + "-DTCL-publisher"),
+ new ShardDataChangeListenerPublisherActorProxy(getContext(), name + "-DCL-publisher"), name);
shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
datastoreContext.getDataStoreMXBeanType());
- shardMBean.setDataStoreExecutor(store.getDomStoreExecutor());
- shardMBean.setNotificationManager(store.getDataChangeListenerNotificationManager());
+ shardMBean.setShard(this);
if (isMetricsCaptureEnabled()) {
getContext().become(new MeteringBehavior(this));
}
- }
- private static Map<String, String> mapPeerAddresses(
- Map<ShardIdentifier, String> peerAddresses) {
- Map<String, String> map = new HashMap<>();
+ commitCoordinator = new ShardCommitCoordinator(store,
+ datastoreContext.getShardCommitQueueExpiryTimeoutInMillis(),
+ datastoreContext.getShardTransactionCommitQueueCapacity(), LOG, this.name);
- for (Map.Entry<ShardIdentifier, String> entry : peerAddresses
- .entrySet()) {
- map.put(entry.getKey().toString(), entry.getValue());
- }
+ setTransactionCommitTimeout();
+
+ // create a notifier actor for each cluster member
+ roleChangeNotifier = createRoleChangeNotifier(name.toString());
+
+ appendEntriesReplyTracker = new MessageTracker(AppendEntriesReply.class,
+ getRaftActorContext().getConfigParams().getIsolatedCheckIntervalInMillis());
- return map;
+ 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);
+
+ messageRetrySupport = new ShardTransactionMessageRetrySupport(this);
}
- public static Props props(final ShardIdentifier name,
- final Map<ShardIdentifier, String> peerAddresses,
- DatastoreContext datastoreContext, 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");
+ private void setTransactionCommitTimeout() {
+ transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
+ datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS) / 2;
+ }
- return Props.create(new ShardCreator(name, peerAddresses, datastoreContext, schemaContext));
+ private Optional<ActorRef> createRoleChangeNotifier(String shardId) {
+ ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
+ RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
+ return Optional.of(shardRoleChangeNotifier);
}
- @Override public void onReceiveRecover(Object message) {
- if(LOG.isDebugEnabled()) {
- LOG.debug("onReceiveRecover: Received message {} from {}",
- message.getClass().toString(),
- getSender());
- }
+ @Override
+ public void postStop() {
+ LOG.info("Stopping Shard {}", persistenceId());
- if (message instanceof RecoveryFailure){
- LOG.error(((RecoveryFailure) message).cause(), "Recovery failed because of this cause");
- } else {
- super.onReceiveRecover(message);
- }
- }
-
- @Override public void onReceiveCommand(Object message) {
- if(LOG.isDebugEnabled()) {
- LOG.debug("onReceiveCommand: Received message {} from {}",
- message.getClass().toString(),
- getSender());
- }
-
- if(message.getClass().equals(ReadDataReply.SERIALIZABLE_CLASS)) {
- // This must be for install snapshot. Don't want to open this up and trigger
- // deSerialization
- self()
- .tell(new CaptureSnapshotReply(ReadDataReply.getNormalizedNodeByteString(message)),
- self());
-
- createSnapshotTransaction = null;
- // Send a PoisonPill instead of sending close transaction because we do not really need
- // a response
- getSender().tell(PoisonPill.getInstance(), self());
-
- } else if (message.getClass().equals(CloseTransactionChain.SERIALIZABLE_CLASS)){
- closeTransactionChain(CloseTransactionChain.fromSerializable(message));
- } else if (message instanceof RegisterChangeListener) {
- registerChangeListener((RegisterChangeListener) message);
- } else if (message instanceof UpdateSchemaContext) {
- updateSchemaContext((UpdateSchemaContext) message);
- } else if (message instanceof ForwardedCommitTransaction) {
- handleForwardedCommit((ForwardedCommitTransaction) message);
- } else if (message.getClass()
- .equals(CreateTransaction.SERIALIZABLE_CLASS)) {
- if (isLeader()) {
- createTransaction(CreateTransaction.fromSerializable(message));
- } else if (getLeader() != null) {
- getLeader().forward(message, getContext());
- } else {
- getSender().tell(new akka.actor.Status.Failure(new IllegalStateException(
- "Could not find shard leader so transaction cannot be created. This typically happens" +
- " when system is coming up or recovering and a leader is being elected. Try again" +
- " later.")), getSelf());
- }
- } else if (message instanceof PeerAddressResolved) {
- PeerAddressResolved resolved = (PeerAddressResolved) message;
- setPeerAddress(resolved.getPeerId().toString(),
- resolved.getPeerAddress());
- } else {
- super.onReceiveCommand(message);
+ super.postStop();
+
+ messageRetrySupport.close();
+
+ if(txCommitTimeoutCheckSchedule != null) {
+ txCommitTimeoutCheckSchedule.cancel();
}
+
+ commitCoordinator.abortPendingTransactions("Transaction aborted due to shutdown.", this);
+
+ shardMBean.unregisterMBean();
}
- private void closeTransactionChain(CloseTransactionChain closeTransactionChain) {
- DOMStoreTransactionChain chain =
- transactionChains.remove(closeTransactionChain.getTransactionChainId());
+ @Override
+ protected void handleRecover(final Object message) {
+ LOG.debug("{}: onReceiveRecover: Received message {} from {}", persistenceId(), message.getClass(),
+ getSender());
- if(chain != null) {
- chain.close();
+ super.handleRecover(message);
+ if (LOG.isTraceEnabled()) {
+ appendEntriesReplyTracker.begin();
}
}
- private ActorRef createTypedTransactionActor(
- int transactionType,
- ShardTransactionIdentifier transactionId,
- String transactionChainId ) {
-
- DOMStoreTransactionFactory factory = store;
+ @Override
+ protected void handleNonRaftCommand(final Object message) {
+ try (final MessageTracker.Context context = appendEntriesReplyTracker.received(message)) {
+ final Optional<Error> maybeError = context.error();
+ if (maybeError.isPresent()) {
+ LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(),
+ maybeError.get());
+ }
- if(!transactionChainId.isEmpty()) {
- factory = transactionChains.get(transactionChainId);
- if(factory == null){
- DOMStoreTransactionChain transactionChain = store.createTransactionChain();
- transactionChains.put(transactionChainId, transactionChain);
- factory = transactionChain;
+ if (CreateTransaction.isSerializedType(message)) {
+ handleCreateTransaction(message);
+ } else if (message instanceof BatchedModifications) {
+ handleBatchedModifications((BatchedModifications)message);
+ } else if (message instanceof ForwardedReadyTransaction) {
+ handleForwardedReadyTransaction((ForwardedReadyTransaction) message);
+ } else if (message instanceof ReadyLocalTransaction) {
+ handleReadyLocalTransaction((ReadyLocalTransaction)message);
+ } else if (CanCommitTransaction.isSerializedType(message)) {
+ handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
+ } else if (CommitTransaction.isSerializedType(message)) {
+ handleCommitTransaction(CommitTransaction.fromSerializable(message));
+ } else if (AbortTransaction.isSerializedType(message)) {
+ handleAbortTransaction(AbortTransaction.fromSerializable(message));
+ } else if (CloseTransactionChain.isSerializedType(message)) {
+ closeTransactionChain(CloseTransactionChain.fromSerializable(message));
+ } else if (message instanceof RegisterChangeListener) {
+ 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) {
+ PeerAddressResolved resolved = (PeerAddressResolved) message;
+ setPeerAddress(resolved.getPeerId().toString(),
+ resolved.getPeerAddress());
+ } else if (TX_COMMIT_TIMEOUT_CHECK_MESSAGE.equals(message)) {
+ 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) {
+ 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 if (message instanceof DataTreeCohortActorRegistry.CohortRegistryCommand) {
+ commitCoordinator.processCohortRegistryCommand(getSender(),
+ (DataTreeCohortActorRegistry.CohortRegistryCommand) message);
+ } else {
+ super.handleNonRaftCommand(message);
}
}
+ }
- if(this.schemaContext == null){
- throw new NullPointerException("schemaContext should not be null");
- }
+ private boolean hasLeader() {
+ return getLeaderId() != null;
+ }
- if (transactionType
- == TransactionProxy.TransactionType.READ_ONLY.ordinal()) {
+ public int getPendingTxCommitQueueSize() {
+ return commitCoordinator.getQueueSize();
+ }
- shardMBean.incrementReadOnlyTransactionCount();
+ public int getCohortCacheSize() {
+ return commitCoordinator.getCohortCacheSize();
+ }
- return getContext().actorOf(
- ShardTransaction.props(factory.newReadOnlyTransaction(), getSelf(),
- schemaContext,datastoreContext, shardMBean), transactionId.toString());
+ @Override
+ protected Optional<ActorRef> getRoleChangeNotifier() {
+ return roleChangeNotifier;
+ }
- } else if (transactionType
- == TransactionProxy.TransactionType.READ_WRITE.ordinal()) {
+ @Override
+ protected LeaderStateChanged newLeaderStateChanged(String memberId, String leaderId, short leaderPayloadVersion) {
+ return isLeader() ? new ShardLeaderStateChanged(memberId, leaderId, store.getDataTree(), leaderPayloadVersion)
+ : new ShardLeaderStateChanged(memberId, leaderId, leaderPayloadVersion);
+ }
- shardMBean.incrementReadWriteTransactionCount();
+ protected void onDatastoreContext(DatastoreContext context) {
+ datastoreContext = context;
- return getContext().actorOf(
- ShardTransaction.props(factory.newReadWriteTransaction(), getSelf(),
- schemaContext, datastoreContext, shardMBean), transactionId.toString());
+ commitCoordinator.setQueueCapacity(datastoreContext.getShardTransactionCommitQueueCapacity());
+ setTransactionCommitTimeout();
- } else if (transactionType
- == TransactionProxy.TransactionType.WRITE_ONLY.ordinal()) {
+ if(datastoreContext.isPersistent() && !persistence().isRecoveryApplicable()) {
+ setPersistence(true);
+ } else if(!datastoreContext.isPersistent() && persistence().isRecoveryApplicable()) {
+ setPersistence(false);
+ }
- shardMBean.incrementWriteOnlyTransactionCount();
+ updateConfigParams(datastoreContext.getShardRaftConfig());
+ }
- return getContext().actorOf(
- ShardTransaction.props(factory.newWriteOnlyTransaction(), getSelf(),
- schemaContext, datastoreContext, shardMBean), transactionId.toString());
+ 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 {
- throw new IllegalArgumentException(
- "Shard="+name + ":CreateTransaction message has unidentified transaction type="
- + transactionType);
+ persistData(cohortEntry.getReplySender(), cohortEntry.getTransactionID(),
+ DataTreeCandidatePayload.create(candidate));
}
}
- private void createTransaction(CreateTransaction createTransaction) {
- createTransaction(createTransaction.getTransactionType(),
- createTransaction.getTransactionId(), createTransaction.getTransactionChainId());
+ private void handleCommitTransaction(final CommitTransaction commit) {
+ if (isLeader()) {
+ if(!commitCoordinator.handleCommit(new StringIdentifier(commit.getTransactionID()), getSender(), this)) {
+ shardMBean.incrementFailedTransactionsCount();
+ }
+ } else {
+ ActorSelection leader = getLeader();
+ if (leader == null) {
+ messageRetrySupport.addMessageToRetry(commit, getSender(),
+ "Could not commit transaction " + commit.getTransactionID());
+ } else {
+ LOG.debug("{}: Forwarding CommitTransaction to leader {}", persistenceId(), leader);
+ leader.forward(commit, getContext());
+ }
+ }
}
- private ActorRef createTransaction(int transactionType, String remoteTransactionId, String transactionChainId) {
+ private void finishCommit(@Nonnull final ActorRef sender, @Nonnull final Identifier transactionID,
+ @Nonnull final CohortEntry cohortEntry) {
+ LOG.debug("{}: Finishing commit for transaction {}", persistenceId(), cohortEntry.getTransactionID());
- ShardTransactionIdentifier transactionId =
- ShardTransactionIdentifier.builder()
- .remoteTransactionId(remoteTransactionId)
- .build();
- if(LOG.isDebugEnabled()) {
- LOG.debug("Creating transaction : {} ", transactionId);
- }
- ActorRef transactionActor =
- createTypedTransactionActor(transactionType, transactionId, transactionChainId);
+ try {
+ try {
+ cohortEntry.commit();
+ } catch(ExecutionException e) {
+ // We may get a "store tree and candidate base differ" IllegalStateException from commit under
+ // certain edge case scenarios so we'll try to re-apply the candidate from scratch as a last
+ // resort. Eg, we're a follower and a tx payload is replicated but the leader goes down before
+ // applying it to the state. We then become the leader and a second tx is pre-committed and
+ // replicated. When consensus occurs, this will cause the first tx to be applied as a foreign
+ // candidate via applyState prior to the second tx. Since the second tx has already been
+ // pre-committed, when it gets here to commit it will get an IllegalStateException.
+
+ // FIXME - this is not an ideal way to handle this scenario. This is temporary - a cleaner
+ // solution will be forthcoming.
+ if(e.getCause() instanceof IllegalStateException) {
+ LOG.debug("{}: commit failed for transaction {} - retrying as foreign candidate", persistenceId(),
+ transactionID, e);
+ store.applyForeignCandidate(transactionID, cohortEntry.getCandidate());
+ } else {
+ throw e;
+ }
+ }
- getSender()
- .tell(new CreateTransactionReply(
- Serialization.serializedActorPath(transactionActor),
- remoteTransactionId).toSerializable(),
- getSelf());
+ sender.tell(CommitTransactionReply.instance(cohortEntry.getClientVersion()).toSerializable(), getSelf());
- return transactionActor;
- }
+ shardMBean.incrementCommittedTransactionCount();
+ shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
+
+ } catch (Exception e) {
+ sender.tell(new akka.actor.Status.Failure(e), getSelf());
- private void syncCommitTransaction(DOMStoreWriteTransaction transaction)
- throws ExecutionException, InterruptedException {
- DOMStoreThreePhaseCommitCohort commitCohort = transaction.ready();
- commitCohort.preCommit().get();
- commitCohort.commit().get();
+ LOG.error("{}, An exception occurred while committing transaction {}", persistenceId(),
+ transactionID, e);
+ shardMBean.incrementFailedTransactionsCount();
+ } finally {
+ commitCoordinator.currentTransactionComplete(transactionID, true);
+ }
}
+ private void finishCommit(@Nonnull final ActorRef sender, final @Nonnull Identifier transactionID) {
+ // With persistence enabled, this method is called via applyState by the leader strategy
+ // after the commit has been replicated to a majority of the followers.
- private void commit(final ActorRef sender, Object serialized) {
- Modification modification = MutableCompositeModification
- .fromSerializable(serialized, schemaContext);
- DOMStoreThreePhaseCommitCohort cohort =
- modificationToCohort.remove(serialized);
- if (cohort == null) {
- // If there's no cached cohort then we must be applying replicated state.
- commitWithNewTransaction(serialized);
- return;
- }
+ CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
+ if (cohortEntry == null) {
+ // The transaction is no longer the current commit. This can happen if the transaction
+ // was aborted prior, most likely due to timeout in the front-end. We need to finish
+ // committing the transaction though since it was successfully persisted and replicated
+ // however we can't use the original cohort b/c it was already preCommitted and may
+ // conflict with the current commit or may have been aborted so we commit with a new
+ // transaction.
+ cohortEntry = commitCoordinator.getAndRemoveCohortEntry(transactionID);
+ if(cohortEntry != null) {
+ try {
+ store.applyForeignCandidate(transactionID, cohortEntry.getCandidate());
+ } catch (DataValidationFailedException e) {
+ shardMBean.incrementFailedTransactionsCount();
+ LOG.error("{}: Failed to re-apply transaction {}", persistenceId(), transactionID, e);
+ }
- if(sender == null) {
- LOG.error("Commit failed. Sender cannot be null");
- return;
+ 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.
+ IllegalStateException ex = new IllegalStateException(
+ String.format("%s: Could not finish committing transaction %s - no CohortEntry found",
+ persistenceId(), transactionID));
+ LOG.error(ex.getMessage());
+ sender.tell(new akka.actor.Status.Failure(ex), getSelf());
+ }
+ } else {
+ finishCommit(sender, transactionID, cohortEntry);
}
+ }
- ListenableFuture<Void> future = cohort.commit();
+ private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
+ LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionID());
- Futures.addCallback(future, new FutureCallback<Void>() {
- @Override
- public void onSuccess(Void v) {
- sender.tell(new CommitTransactionReply().toSerializable(), getSelf());
- shardMBean.incrementCommittedTransactionCount();
- shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
+ if (isLeader()) {
+ commitCoordinator.handleCanCommit(new StringIdentifier(canCommit.getTransactionID()), getSender(), this);
+ } else {
+ ActorSelection leader = getLeader();
+ if (leader == null) {
+ messageRetrySupport.addMessageToRetry(canCommit, getSender(),
+ "Could not canCommit transaction " + canCommit.getTransactionID());
+ } else {
+ LOG.debug("{}: Forwarding CanCommitTransaction to leader {}", persistenceId(), leader);
+ leader.forward(canCommit, getContext());
}
+ }
+ }
- @Override
- public void onFailure(Throwable t) {
- LOG.error(t, "An exception happened during commit");
- shardMBean.incrementFailedTransactionsCount();
- sender.tell(new akka.actor.Status.Failure(t), getSelf());
+ protected void handleBatchedModificationsLocal(BatchedModifications batched, ActorRef sender) {
+ try {
+ commitCoordinator.handleBatchedModifications(batched, sender, this, store.getSchemaContext());
+ } 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 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
+ // to initiate the 3-phase commit. This also avoids the overhead of sending an additional
+ // ReadyTransaction message.
+
+ // If we're not the leader then forward to the leader. This is a safety measure - we shouldn't
+ // normally get here if we're not the leader as the front-end (TransactionProxy) should determine
+ // 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.
+ //
+ boolean isLeaderActive = isLeaderActive();
+ if (isLeader() && isLeaderActive) {
+ handleBatchedModificationsLocal(batched, getSender());
+ } else {
+ ActorSelection leader = getLeader();
+ if (!isLeaderActive || leader == null) {
+ messageRetrySupport.addMessageToRetry(batched, getSender(),
+ "Could not commit transaction " + batched.getTransactionID());
+ } else {
+ // If this is not the first batch and leadership changed in between batched messages,
+ // we need to reconstruct previous BatchedModifications from the transaction
+ // DataTreeModification, honoring the max batched modification count, and forward all the
+ // previous BatchedModifications to the new leader.
+ Collection<BatchedModifications> newModifications = commitCoordinator.createForwardedBatchedModifications(
+ batched, datastoreContext.getShardBatchedModificationCount());
+
+ LOG.debug("{}: Forwarding {} BatchedModifications to leader {}", persistenceId(),
+ newModifications.size(), leader);
+
+ for(BatchedModifications bm: newModifications) {
+ leader.forward(bm, getContext());
+ }
}
- });
-
+ }
}
- private void commitWithNewTransaction(Object modification) {
- DOMStoreWriteTransaction tx = store.newWriteOnlyTransaction();
- MutableCompositeModification.fromSerializable(modification, schemaContext).apply(tx);
- try {
- syncCommitTransaction(tx);
- shardMBean.incrementCommittedTransactionCount();
- } catch (InterruptedException | ExecutionException e) {
- shardMBean.incrementFailedTransactionsCount();
- LOG.error(e, "Failed to commit");
+ 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;
}
- private void handleForwardedCommit(ForwardedCommitTransaction message) {
- Object serializedModification =
- message.getModification().toSerializable();
+ protected boolean isIsolatedLeader() {
+ return getRaftState() == RaftState.IsolatedLeader;
+ }
- modificationToCohort
- .put(serializedModification, message.getCohort());
+ private void handleReadyLocalTransaction(final ReadyLocalTransaction message) {
+ LOG.debug("{}: handleReadyLocalTransaction for {}", persistenceId(), message.getTransactionID());
- if (persistent) {
- this.persistData(getSender(), "identifier",
- new CompositeModificationPayload(serializedModification));
+ boolean isLeaderActive = isLeaderActive();
+ if (isLeader() && isLeaderActive) {
+ try {
+ commitCoordinator.handleReadyLocalTransaction(message, getSender(), this, store.getSchemaContext());
+ } catch (Exception e) {
+ LOG.error("{}: Error handling ReadyLocalTransaction for Tx {}", persistenceId(),
+ message.getTransactionID(), e);
+ getSender().tell(new akka.actor.Status.Failure(e), getSelf());
+ }
} else {
- this.commit(getSender(), serializedModification);
+ ActorSelection leader = getLeader();
+ if (!isLeaderActive || leader == null) {
+ messageRetrySupport.addMessageToRetry(message, getSender(),
+ "Could not commit transaction " + message.getTransactionID());
+ } else {
+ LOG.debug("{}: Forwarding ReadyLocalTransaction to leader {}", persistenceId(), leader);
+ message.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
+ leader.forward(message, getContext());
+ }
}
}
- private void updateSchemaContext(UpdateSchemaContext message) {
- this.schemaContext = message.getSchemaContext();
- updateSchemaContext(message.getSchemaContext());
- store.onGlobalContextUpdated(message.getSchemaContext());
+ private void handleForwardedReadyTransaction(ForwardedReadyTransaction forwardedReady) {
+ LOG.debug("{}: handleForwardedReadyTransaction for {}", persistenceId(), forwardedReady.getTransactionID());
+
+ boolean isLeaderActive = isLeaderActive();
+ if (isLeader() && isLeaderActive) {
+ commitCoordinator.handleForwardedReadyTransaction(forwardedReady, getSender(), this,
+ store.getSchemaContext());
+ } 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());
+ }
+ }
}
- @VisibleForTesting void updateSchemaContext(SchemaContext schemaContext) {
- store.onGlobalContextUpdated(schemaContext);
+ private void handleAbortTransaction(final AbortTransaction abort) {
+ doAbortTransaction(abort.getTransactionID(), getSender());
}
- private void registerChangeListener(
- RegisterChangeListener registerChangeListener) {
+ void doAbortTransaction(final String transactionID, final ActorRef sender) {
+ commitCoordinator.handleAbort(new StringIdentifier(transactionID), sender, this);
+ }
- if(LOG.isDebugEnabled()) {
- LOG.debug("registerDataChangeListener for {}", registerChangeListener
- .getPath());
+ private void handleCreateTransaction(final Object message) {
+ if (isLeader()) {
+ createTransaction(CreateTransaction.fromSerializable(message));
+ } else if (getLeader() != null) {
+ getLeader().forward(message, getContext());
+ } else {
+ getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(
+ "Could not create a shard transaction", persistenceId())), getSelf());
}
+ }
+ private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
+ store.closeTransactionChain(closeTransactionChain.getTransactionChainId());
+ }
- 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(isLeader()), getSelf());
+ private void createTransaction(CreateTransaction createTransaction) {
+ try {
+ if(TransactionType.fromInt(createTransaction.getTransactionType()) != TransactionType.READ_ONLY &&
+ failIfIsolatedLeader(getSender())) {
+ return;
+ }
- // 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);
+ ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
+ createTransaction.getTransactionId(), createTransaction.getTransactionChainId());
- AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>>
- listener = new DataChangeListenerProxy(schemaContext, dataChangeListenerPath);
+ getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
+ createTransaction.getTransactionId(), createTransaction.getVersion()).toSerializable(), getSelf());
+ } catch (Exception e) {
+ getSender().tell(new akka.actor.Status.Failure(e), getSelf());
+ }
+ }
- ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>>>
- registration = store.registerChangeListener(registerChangeListener.getPath(),
- listener, registerChangeListener.getScope());
- ActorRef listenerRegistration =
- getContext().actorOf(
- DataChangeListenerRegistration.props(registration));
+ private ActorRef createTransaction(int transactionType, String transactionId, String transactionChainId) {
+ LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
+ return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType),
+ transactionId, transactionChainId);
+ }
- if(LOG.isDebugEnabled()) {
- LOG.debug(
- "registerDataChangeListener sending reply, listenerRegistrationPath = {} "
- , listenerRegistration.path().toString());
+ private void commitWithNewTransaction(final BatchedModifications modification) {
+ ReadWriteShardDataTreeTransaction tx = store.newReadWriteTransaction(modification.getTransactionID(),
+ modification.getTransactionChainID());
+ modification.apply(tx.getSnapshot());
+ try {
+ snapshotCohort.syncCommitTransaction(tx);
+ shardMBean.incrementCommittedTransactionCount();
+ shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
+ } catch (Exception e) {
+ shardMBean.incrementFailedTransactionsCount();
+ LOG.error("{}: Failed to commit", persistenceId(), e);
}
+ }
- getSender()
- .tell(new RegisterChangeListenerReply(listenerRegistration.path()),
- getSelf());
+ private void updateSchemaContext(final UpdateSchemaContext message) {
+ updateSchemaContext(message.getSchemaContext());
}
- private boolean isMetricsCaptureEnabled(){
+ @VisibleForTesting
+ void updateSchemaContext(final SchemaContext schemaContext) {
+ store.updateSchemaContext(schemaContext);
+ }
+
+ private boolean isMetricsCaptureEnabled() {
CommonConfig config = new CommonConfig(getContext().system().settings().config());
return config.isMetricCaptureEnabled();
}
@Override
- protected
- void startLogRecoveryBatch(int maxBatchSize) {
- currentLogRecoveryBatch = Lists.newArrayListWithCapacity(maxBatchSize);
-
- if(LOG.isDebugEnabled()) {
- LOG.debug("{} : starting log recovery batch with max size {}", persistenceId(), maxBatchSize);
- }
+ @VisibleForTesting
+ public RaftActorSnapshotCohort getRaftActorSnapshotCohort() {
+ return snapshotCohort;
}
@Override
- protected void appendRecoveredLogEntry(Payload data) {
- if (data instanceof CompositeModificationPayload) {
- currentLogRecoveryBatch.add(((CompositeModificationPayload) data).getModification());
- } else {
- LOG.error("Unknown state received {} during recovery", data);
- }
+ @Nonnull
+ protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
+ return new ShardRecoveryCoordinator(store, store.getSchemaContext(),
+ restoreFromSnapshot != null ? restoreFromSnapshot.getSnapshot() : null, persistenceId(), LOG);
}
@Override
- protected void applyRecoverySnapshot(ByteString snapshot) {
- if(recoveryCoordinator == null) {
- recoveryCoordinator = new ShardRecoveryCoordinator(persistenceId(), schemaContext);
- }
+ protected void onRecoveryComplete() {
+ restoreFromSnapshot = null;
- recoveryCoordinator.submit(snapshot, store.newWriteOnlyTransaction());
+ //notify shard manager
+ getContext().parent().tell(new ActorInitialized(), getSelf());
- if(LOG.isDebugEnabled()) {
- LOG.debug("{} : submitted recovery sbapshot", persistenceId());
+ // Being paranoid here - this method should only be called once but just in case...
+ if(txCommitTimeoutCheckSchedule == null) {
+ // Schedule a message to be periodically sent to check if the current in-progress
+ // transaction should be expired and aborted.
+ FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
+ txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
+ period, period, getSelf(),
+ TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
}
}
@Override
- protected void applyCurrentLogRecoveryBatch() {
- if(recoveryCoordinator == null) {
- recoveryCoordinator = new ShardRecoveryCoordinator(persistenceId(), schemaContext);
+ protected void applyState(final ActorRef clientActor, final Identifier identifier, final Object data) {
+ if (data instanceof DataTreeCandidatePayload) {
+ if (clientActor == null) {
+ // No clientActor indicates a replica coming from the leader
+ try {
+ store.applyForeignCandidate(identifier, ((DataTreeCandidatePayload)data).getCandidate());
+ } catch (DataValidationFailedException | IOException e) {
+ LOG.error("{}: Error applying replica {}", persistenceId(), identifier, e);
+ }
+ } else {
+ // Replication consensus reached, proceed to commit
+ finishCommit(clientActor, identifier);
+ }
+ } else {
+ LOG.error("{}: Unknown state received {} ClassLoader {}", persistenceId(), data,
+ data.getClass().getClassLoader());
}
+ }
- recoveryCoordinator.submit(currentLogRecoveryBatch, store.newWriteOnlyTransaction());
+ private void applyModificationToState(ActorRef clientActor, Identifier identifier, Object modification) {
+ if(modification == null) {
+ LOG.error(
+ "{}: modification is null - this is very unexpected, clientActor = {}, identifier = {}",
+ persistenceId(), identifier, clientActor != null ? clientActor.path().toString() : null);
+ } else if(clientActor == null) {
+ // There's no clientActor to which to send a commit reply so we must be applying
+ // replicated state from the leader.
- if(LOG.isDebugEnabled()) {
- LOG.debug("{} : submitted log recovery batch with size {}", persistenceId(),
- currentLogRecoveryBatch.size());
+ // The only implementation we know of is BatchedModifications, which also carries a transaction
+ // identifier -- which we really need that.
+ Preconditions.checkArgument(modification instanceof BatchedModifications);
+ commitWithNewTransaction((BatchedModifications)modification);
+ } else {
+ // This must be the OK to commit after replication consensus.
+ finishCommit(clientActor, identifier);
}
}
@Override
- protected void onRecoveryComplete() {
- if(recoveryCoordinator != null) {
- Collection<DOMStoreWriteTransaction> txList = recoveryCoordinator.getTransactions();
+ protected void onStateChanged() {
+ boolean isLeader = 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) {
if(LOG.isDebugEnabled()) {
- LOG.debug("{} : recovery complete - committing {} Tx's", persistenceId(), txList.size());
+ LOG.debug(
+ "{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader",
+ persistenceId(), getId());
}
- for(DOMStoreWriteTransaction tx: txList) {
- try {
- syncCommitTransaction(tx);
- shardMBean.incrementCommittedTransactionCount();
- } catch (InterruptedException | ExecutionException e) {
- shardMBean.incrementFailedTransactionsCount();
- LOG.error(e, "Failed to commit");
- }
- }
+ store.closeAllTransactionChains();
}
- recoveryCoordinator = null;
- currentLogRecoveryBatch = null;
- updateJournalStats();
-
- //notify shard manager
- getContext().parent().tell(new ActorInitialized(), getSelf());
+ if(hasLeader && !isIsolatedLeader()) {
+ messageRetrySupport.retryMessages();
+ }
}
@Override
- protected void applyState(ActorRef clientActor, String identifier, Object data) {
-
- if (data instanceof CompositeModificationPayload) {
- Object modification = ((CompositeModificationPayload) data).getModification();
-
- if (modification != null) {
- commit(clientActor, modification);
+ protected void onLeaderChanged(String oldLeader, String newLeader) {
+ shardMBean.incrementLeadershipChangeCount();
+
+ boolean hasLeader = hasLeader();
+ if(hasLeader && !isLeader()) {
+ // Another leader was elected. If we were the previous leader and had pending transactions, convert
+ // them to transaction messages and send to the new leader.
+ ActorSelection leader = getLeader();
+ if(leader != null) {
+ Collection<Object> messagesToForward = commitCoordinator.convertPendingTransactionsToMessages(
+ datastoreContext.getShardBatchedModificationCount());
+
+ if(!messagesToForward.isEmpty()) {
+ LOG.debug("{}: Forwarding {} pending transaction messages to leader {}", persistenceId(),
+ messagesToForward.size(), leader);
+
+ for(Object message: messagesToForward) {
+ leader.tell(message, self());
+ }
+ }
} else {
- LOG.error(
- "modification is null - this is very unexpected, clientActor = {}, identifier = {}",
- identifier, clientActor != null ? clientActor.path().toString() : null);
+ commitCoordinator.abortPendingTransactions(
+ "The transacton was aborted due to inflight leadership change and the leader address isn't available.",
+ this);
}
-
- } else {
- LOG.error("Unknown state received {} Class loader = {} CompositeNodeMod.ClassLoader = {}",
- data, data.getClass().getClassLoader(),
- CompositeModificationPayload.class.getClassLoader());
}
- updateJournalStats();
-
- }
-
- private void updateJournalStats() {
- ReplicatedLogEntry lastLogEntry = getLastLogEntry();
-
- if (lastLogEntry != null) {
- shardMBean.setLastLogIndex(lastLogEntry.getIndex());
- shardMBean.setLastLogTerm(lastLogEntry.getTerm());
+ if(hasLeader && !isIsolatedLeader()) {
+ messageRetrySupport.retryMessages();
}
-
- shardMBean.setCommitIndex(getCommitIndex());
- shardMBean.setLastApplied(getLastApplied());
}
@Override
- protected void createSnapshot() {
- if (createSnapshotTransaction == null) {
-
- // Create a transaction. We are really going to treat the transaction as a worker
- // so that this actor does not get block building the snapshot
- createSnapshotTransaction = createTransaction(
- TransactionProxy.TransactionType.READ_ONLY.ordinal(),
- "createSnapshot", "");
-
- createSnapshotTransaction.tell(
- new ReadData(YangInstanceIdentifier.builder().build()).toSerializable(), self());
-
- }
+ protected void pauseLeader(Runnable operation) {
+ LOG.debug("{}: In pauseLeader, operation: {}", persistenceId(), operation);
+ commitCoordinator.setRunOnPendingTransactionsComplete(operation);
}
- @VisibleForTesting
@Override
- protected void applySnapshot(ByteString snapshot) {
- // 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");
- try {
- DOMStoreWriteTransaction transaction = store.newWriteOnlyTransaction();
- NormalizedNodeMessages.Node serializedNode = NormalizedNodeMessages.Node.parseFrom(snapshot);
- NormalizedNode<?, ?> node = new NormalizedNodeToNodeCodec(schemaContext)
- .decode(serializedNode);
-
- // delete everything first
- transaction.delete(YangInstanceIdentifier.builder().build());
-
- // Add everything from the remote node back
- transaction.write(YangInstanceIdentifier.builder().build(), node);
- syncCommitTransaction(transaction);
- } catch (InvalidProtocolBufferException | InterruptedException | ExecutionException e) {
- LOG.error(e, "An exception occurred when applying snapshot");
- } finally {
- LOG.info("Done applying snapshot");
- }
+ public String persistenceId() {
+ return this.name;
}
- @Override protected void onStateChanged() {
- for (ActorSelection dataChangeListener : dataChangeListeners) {
- dataChangeListener
- .tell(new EnableNotification(isLeader()), getSelf());
- }
-
- shardMBean.setRaftState(getRaftState().name());
- shardMBean.setCurrentTerm(getCurrentTerm());
+ @VisibleForTesting
+ ShardCommitCoordinator getCommitCoordinator() {
+ return commitCoordinator;
+ }
- // If this actor is no longer the leader close all the transaction chains
- if(!isLeader()){
- for(Map.Entry<String, DOMStoreTransactionChain> entry : transactionChains.entrySet()){
- if(LOG.isDebugEnabled()) {
- LOG.debug(
- "onStateChanged: Closing transaction chain {} because shard {} is no longer the leader",
- entry.getKey(), getId());
- }
- entry.getValue().close();
- }
+ public DatastoreContext getDatastoreContext() {
+ return datastoreContext;
+ }
- transactionChains.clear();
- }
+ @VisibleForTesting
+ public ShardDataTree getDataStore() {
+ return store;
}
- @Override protected void onLeaderChanged(String oldLeader, String newLeader) {
- shardMBean.setLeader(newLeader);
+ @VisibleForTesting
+ ShardStats getShardMBean() {
+ return shardMBean;
}
- @Override public String persistenceId() {
- return this.name.toString();
+ public static Builder builder() {
+ return new Builder();
}
- private static class ShardCreator implements Creator<Shard> {
+ 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 final long serialVersionUID = 1L;
+ protected AbstractBuilder(Class<S> shardClass) {
+ this.shardClass = shardClass;
+ }
- final ShardIdentifier name;
- final Map<ShardIdentifier, String> peerAddresses;
- final DatastoreContext datastoreContext;
- final SchemaContext schemaContext;
+ protected void checkSealed() {
+ Preconditions.checkState(!sealed, "Builder isalready sealed - further modifications are not allowed");
+ }
- ShardCreator(ShardIdentifier name, Map<ShardIdentifier, String> peerAddresses,
- DatastoreContext datastoreContext, SchemaContext schemaContext) {
- this.name = name;
+ @SuppressWarnings("unchecked")
+ private T self() {
+ return (T) this;
+ }
+
+ 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
- NormalizedNode<?,?> readStore(YangInstanceIdentifier id)
- throws ExecutionException, InterruptedException {
- DOMStoreReadTransaction transaction = store.newReadOnlyTransaction();
+ public ShardIdentifier getId() {
+ return id;
+ }
+
+ public Map<String, String> getPeerAddresses() {
+ return peerAddresses;
+ }
- CheckedFuture<Optional<NormalizedNode<?, ?>>, ReadFailedException> future =
- transaction.read(id);
+ public DatastoreContext getDatastoreContext() {
+ return datastoreContext;
+ }
- Optional<NormalizedNode<?, ?>> optional = future.get();
- NormalizedNode<?, ?> node = optional.isPresent()? optional.get() : null;
+ public SchemaContext getSchemaContext() {
+ return schemaContext;
+ }
- transaction.close();
+ public DatastoreSnapshot.ShardSnapshot getRestoreFromSnapshot() {
+ return restoreFromSnapshot;
+ }
- return node;
- }
+ public TreeType getTreeType() {
+ switch (datastoreContext.getLogicalStoreType()) {
+ case CONFIGURATION:
+ return TreeType.CONFIGURATION;
+ case OPERATIONAL:
+ return TreeType.OPERATIONAL;
+ }
- @VisibleForTesting
- void writeToStore(YangInstanceIdentifier id, NormalizedNode<?,?> node)
- throws ExecutionException, InterruptedException {
- DOMStoreWriteTransaction transaction = store.newWriteOnlyTransaction();
+ throw new IllegalStateException("Unhandled logical store type " + datastoreContext.getLogicalStoreType());
+ }
- transaction.write(id, node);
+ 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");
+ }
- syncCommitTransaction(transaction);
+ 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);
+ }
}
}