import org.opendaylight.controller.cluster.datastore.messages.CommitTransactionReply;
import org.opendaylight.controller.cluster.datastore.messages.CreateTransaction;
import org.opendaylight.controller.cluster.datastore.messages.CreateTransactionReply;
+import org.opendaylight.controller.cluster.datastore.messages.DatastoreSnapshot;
+import org.opendaylight.controller.cluster.datastore.messages.DatastoreSnapshot.ShardSnapshot;
import org.opendaylight.controller.cluster.datastore.messages.ForwardedReadyTransaction;
import org.opendaylight.controller.cluster.datastore.messages.GetShardDataTree;
import org.opendaylight.controller.cluster.datastore.messages.PeerAddressResolved;
import org.opendaylight.controller.cluster.datastore.messages.ShardLeaderStateChanged;
import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext;
import org.opendaylight.controller.cluster.datastore.modification.Modification;
-import org.opendaylight.controller.cluster.datastore.modification.ModificationPayload;
import org.opendaylight.controller.cluster.datastore.modification.MutableCompositeModification;
import org.opendaylight.controller.cluster.datastore.utils.Dispatchers;
import org.opendaylight.controller.cluster.datastore.utils.MessageTracker;
import org.opendaylight.controller.cluster.raft.RaftState;
import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus;
import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
+import org.opendaylight.controller.cluster.raft.messages.ServerRemoved;
import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationByteStringPayload;
import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationPayload;
import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTree;
import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidate;
import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException;
import org.opendaylight.yangtools.yang.data.api.schema.tree.ModificationType;
+import org.opendaylight.yangtools.yang.data.api.schema.tree.TreeType;
import org.opendaylight.yangtools.yang.model.api.SchemaContext;
import scala.concurrent.duration.Duration;
import scala.concurrent.duration.FiniteDuration;
private final DataTreeChangeListenerSupport treeChangeSupport = new DataTreeChangeListenerSupport(this);
private final DataChangeListenerSupport changeSupport = new DataChangeListenerSupport(this);
+
+ private ShardSnapshot restoreFromSnapshot;
+
+ private final ShardTransactionMessageRetrySupport messageRetrySupport;
+
protected Shard(AbstractBuilder<?, ?> builder) {
super(builder.getId().toString(), builder.getPeerAddresses(),
Optional.of(builder.getDatastoreContext().getShardRaftConfig()), DataStoreVersions.CURRENT_VERSION);
this.name = builder.getId().toString();
this.datastoreContext = builder.getDatastoreContext();
+ this.restoreFromSnapshot = builder.getRestoreFromSnapshot();
setPersistence(datastoreContext.isPersistent());
LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
- store = new ShardDataTree(builder.getSchemaContext());
+ store = new ShardDataTree(builder.getSchemaContext(), builder.getTreeType());
shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
datastoreContext.getDataStoreMXBeanType());
commitCoordinator = new ShardCommitCoordinator(store,
datastoreContext.getShardCommitQueueExpiryTimeoutInMillis(),
- datastoreContext.getShardTransactionCommitQueueCapacity(), self(), LOG, this.name);
+ datastoreContext.getShardTransactionCommitQueueCapacity(), LOG, this.name);
setTransactionCommitTimeout();
snapshotCohort = new ShardSnapshotCohort(transactionActorFactory, store, LOG, this.name);
-
+ messageRetrySupport = new ShardTransactionMessageRetrySupport(this);
}
private void setTransactionCommitTimeout() {
super.postStop();
+ messageRetrySupport.close();
+
if(txCommitTimeoutCheckSchedule != null) {
txCommitTimeoutCheckSchedule.cancel();
}
+ commitCoordinator.abortPendingTransactions("Transaction aborted due to shutdown.", this);
+
shardMBean.unregisterMBean();
}
}
try {
- if (CreateTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
+ if (CreateTransaction.isSerializedType(message)) {
handleCreateTransaction(message);
} else if (BatchedModifications.class.isInstance(message)) {
handleBatchedModifications((BatchedModifications)message);
} else if (message instanceof ForwardedReadyTransaction) {
- commitCoordinator.handleForwardedReadyTransaction((ForwardedReadyTransaction) message,
- getSender(), this);
+ handleForwardedReadyTransaction((ForwardedReadyTransaction) message);
} else if (message instanceof ReadyLocalTransaction) {
handleReadyLocalTransaction((ReadyLocalTransaction)message);
- } else if (CanCommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
+ } else if (CanCommitTransaction.isSerializedType(message)) {
handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
- } else if (CommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
+ } else if (CommitTransaction.isSerializedType(message)) {
handleCommitTransaction(CommitTransaction.fromSerializable(message));
- } else if (AbortTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
+ } else if (AbortTransaction.isSerializedType(message)) {
handleAbortTransaction(AbortTransaction.fromSerializable(message));
- } else if (CloseTransactionChain.SERIALIZABLE_CLASS.isInstance(message)) {
+ } else if (CloseTransactionChain.isSerializedType(message)) {
closeTransactionChain(CloseTransactionChain.fromSerializable(message));
} else if (message instanceof RegisterChangeListener) {
changeSupport.onMessage((RegisterChangeListener) message, isLeader(), hasLeader());
setPeerAddress(resolved.getPeerId().toString(),
resolved.getPeerAddress());
} else if (message.equals(TX_COMMIT_TIMEOUT_CHECK_MESSAGE)) {
- handleTransactionCommitTimeoutCheck();
+ commitCoordinator.checkForExpiredTransactions(transactionCommitTimeout, this);
} else if(message instanceof DatastoreContext) {
onDatastoreContext((DatastoreContext)message);
} else if(message instanceof RegisterRoleChangeListener){
context().parent().tell(message, self());
} else if(GET_SHARD_MBEAN_MESSAGE.equals(message)){
sender().tell(getShardMBean(), self());
- } else if(message instanceof GetShardDataTree){
+ } else if(message instanceof GetShardDataTree) {
sender().tell(store.getDataTree(), self());
+ } else if(message instanceof ServerRemoved){
+ context().parent().forward(message, context());
+ } else if(ShardTransactionMessageRetrySupport.TIMER_MESSAGE_CLASS.isInstance(message)) {
+ messageRetrySupport.onTimerMessage(message);
} else {
super.onReceiveCommand(message);
}
return commitCoordinator.getQueueSize();
}
+ public int getCohortCacheSize() {
+ return commitCoordinator.getCohortCacheSize();
+ }
+
@Override
protected Optional<ActorRef> getRoleChangeNotifier() {
return roleChangeNotifier;
updateConfigParams(datastoreContext.getShardRaftConfig());
}
- private void handleTransactionCommitTimeoutCheck() {
- CohortEntry cohortEntry = commitCoordinator.getCurrentCohortEntry();
- if(cohortEntry != null) {
- 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());
}
applyModificationToState(cohortEntry.getReplySender(), cohortEntry.getTransactionID(), candidate);
} else {
Shard.this.persistData(cohortEntry.getReplySender(), cohortEntry.getTransactionID(),
- DataTreeCandidatePayload.create(candidate));
+ DataTreeCandidatePayload.create(candidate));
}
}
try {
cohortEntry.commit();
- sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
+ sender.tell(CommitTransactionReply.instance(cohortEntry.getClientVersion()).toSerializable(), getSelf());
shardMBean.incrementCommittedTransactionCount();
shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
LOG.error("{}: Failed to re-apply transaction {}", persistenceId(), transactionID, e);
}
- sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
+ sender.tell(CommitTransactionReply.instance(cohortEntry.getClientVersion()).toSerializable(),
+ getSelf());
} else {
// This really shouldn't happen - it likely means that persistence or replication
// took so long to complete such that the cohort entry was expired from the cache.
commitCoordinator.handleCanCommit(canCommit.getTransactionID(), getSender(), this);
}
- 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);
// the primary/leader shard. However with timing and caching on the front-end, there's a small
// window where it could have a stale leader during leadership transitions.
//
- if(isLeader()) {
- failIfIsolatedLeader(getSender());
-
+ boolean isLeaderActive = isLeaderActive();
+ if (isLeader() && isLeaderActive) {
handleBatchedModificationsLocal(batched, getSender());
} else {
ActorSelection leader = getLeader();
- if(leader != null) {
+ if (!isLeaderActive || leader == null) {
+ messageRetrySupport.addMessageToRetry(batched, getSender(),
+ "Could not commit transaction " + batched.getTransactionID());
+ } else {
// TODO: what if this is not the first batch and leadership changed in between batched messages?
// We could check if the commitCoordinator already has a cached entry and forward all the previous
// batched modifications.
LOG.debug("{}: Forwarding BatchedModifications to leader {}", persistenceId(), leader);
leader.forward(batched, getContext());
- } else {
- noLeaderError("Could not commit transaction " + batched.getTransactionID(), batched);
}
}
}
}
private void handleReadyLocalTransaction(final ReadyLocalTransaction message) {
- if (isLeader()) {
- failIfIsolatedLeader(getSender());
+ LOG.debug("{}: handleReadyLocalTransaction for {}", persistenceId(), message.getTransactionID());
+ boolean isLeaderActive = isLeaderActive();
+ if (isLeader() && isLeaderActive) {
try {
commitCoordinator.handleReadyLocalTransaction(message, getSender(), this);
} catch (Exception e) {
}
} else {
ActorSelection leader = getLeader();
- if (leader != null) {
+ 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 handleForwardedReadyTransaction(ForwardedReadyTransaction forwardedReady) {
+ LOG.debug("{}: handleForwardedReadyTransaction for {}", persistenceId(), forwardedReady.getTransactionID());
+
+ boolean isLeaderActive = isLeaderActive();
+ if (isLeader() && isLeaderActive) {
+ commitCoordinator.handleForwardedReadyTransaction(forwardedReady, getSender(), this);
+ } else {
+ ActorSelection leader = getLeader();
+ if (!isLeaderActive || leader == null) {
+ messageRetrySupport.addMessageToRetry(forwardedReady, getSender(),
+ "Could not commit transaction " + forwardedReady.getTransactionID());
} else {
- noLeaderError("Could not commit transaction " + message.getTransactionID(), message);
+ 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());
}
}
}
createTransaction.getVersion());
getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
- createTransaction.getTransactionId()).toSerializable(), getSelf());
+ createTransaction.getTransactionId(), createTransaction.getVersion()).toSerializable(), getSelf());
} catch (Exception e) {
getSender().tell(new akka.actor.Status.Failure(e), getSelf());
}
@Override
@Nonnull
protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
- return new ShardRecoveryCoordinator(store, store.getSchemaContext(), persistenceId(), LOG);
+ return new ShardRecoveryCoordinator(store, store.getSchemaContext(),
+ restoreFromSnapshot != null ? restoreFromSnapshot.getSnapshot() : null, persistenceId(), LOG);
}
@Override
protected void onRecoveryComplete() {
+ restoreFromSnapshot = null;
+
//notify shard manager
getContext().parent().tell(new ActorInitialized(), getSelf());
// Replication consensus reached, proceed to commit
finishCommit(clientActor, identifier);
}
- } else if (data instanceof ModificationPayload) {
- try {
- applyModificationToState(clientActor, identifier, ((ModificationPayload) data).getModification());
- } catch (ClassNotFoundException | IOException e) {
- LOG.error("{}: Error extracting ModificationPayload", persistenceId(), e);
- }
} else if (data instanceof CompositeModificationPayload) {
Object modification = ((CompositeModificationPayload) data).getModification();
}
store.closeAllTransactionChains();
+
+ commitCoordinator.abortPendingTransactions(
+ "The transacton was aborted due to inflight leadership change.", this);
+ }
+
+ if(hasLeader && !isIsolatedLeader()) {
+ messageRetrySupport.retryMessages();
}
}
@Override
protected void onLeaderChanged(String oldLeader, String newLeader) {
shardMBean.incrementLeadershipChangeCount();
+
+ if(hasLeader() && !isIsolatedLeader()) {
+ messageRetrySupport.retryMessages();
+ }
+ }
+
+ @Override
+ protected void pauseLeader(Runnable operation) {
+ LOG.debug("{}: In pauseLeader, operation: {}", persistenceId(), operation);
+ commitCoordinator.setRunOnPendingTransactionsComplete(operation);
}
@Override
private Map<String, String> peerAddresses = Collections.emptyMap();
private DatastoreContext datastoreContext;
private SchemaContext schemaContext;
+ private DatastoreSnapshot.ShardSnapshot restoreFromSnapshot;
private volatile boolean sealed;
protected AbstractBuilder(Class<S> shardClass) {
return self();
}
+ public T restoreFromSnapshot(DatastoreSnapshot.ShardSnapshot restoreFromSnapshot) {
+ checkSealed();
+ this.restoreFromSnapshot = restoreFromSnapshot;
+ return self();
+ }
+
public ShardIdentifier getId() {
return id;
}
return schemaContext;
}
+ public DatastoreSnapshot.ShardSnapshot getRestoreFromSnapshot() {
+ return restoreFromSnapshot;
+ }
+
+ public TreeType getTreeType() {
+ switch (datastoreContext.getLogicalStoreType()) {
+ case CONFIGURATION:
+ return TreeType.CONFIGURATION;
+ case OPERATIONAL:
+ return TreeType.OPERATIONAL;
+ }
+
+ throw new IllegalStateException("Unhandled logical store type " + datastoreContext.getLogicalStoreType());
+ }
+
protected void verify() {
Preconditions.checkNotNull(id, "id should not be null");
Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
Code Review / controller.git / blobdiff