X-Git-Url: https://git.opendaylight.org/gerrit/gitweb?a=blobdiff_plain;f=opendaylight%2Fmd-sal%2Fsal-distributed-datastore%2Fsrc%2Fmain%2Fjava%2Forg%2Fopendaylight%2Fcontroller%2Fcluster%2Fdatastore%2FShard.java;h=a30b6f7516981411e589577f3fda9a9a0f9bc887;hb=e32959e0bbc326f47c30ed7347f9a9af26813f89;hp=87a0fb931edb171ca8e2bb7d4563ff1172de4912;hpb=83dfe301bf2a2b1eff6883a2af3282c95d5a752e;p=controller.git
diff --git a/opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/Shard.java b/opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/Shard.java
index 87a0fb931e..5fa9192ac0 100644
--- a/opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/Shard.java
+++ b/opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/Shard.java
@@ -5,737 +5,939 @@
* terms of the Eclipse Public License v1.0 which accompanies this distribution,
* and is available at http://www.eclipse.org/legal/epl-v10.html
*/
-
package org.opendaylight.controller.cluster.datastore;
+import static com.google.common.base.Preconditions.checkState;
+import static com.google.common.base.Verify.verify;
+import static java.util.Objects.requireNonNull;
+
import akka.actor.ActorRef;
import akka.actor.ActorSelection;
import akka.actor.Cancellable;
+import akka.actor.ExtendedActorSystem;
import akka.actor.Props;
-import akka.japi.Creator;
-import akka.persistence.RecoveryFailure;
+import akka.actor.Status;
+import akka.actor.Status.Failure;
+import akka.serialization.JavaSerializer;
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 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.ImmutableSet;
+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.List;
import java.util.Map;
-import java.util.concurrent.ExecutionException;
+import java.util.Optional;
+import java.util.OptionalLong;
import java.util.concurrent.TimeUnit;
-import javax.annotation.Nonnull;
-import org.opendaylight.controller.cluster.DataPersistenceProvider;
+import org.eclipse.jdt.annotation.NonNull;
+import org.eclipse.jdt.annotation.Nullable;
+import org.opendaylight.controller.cluster.access.ABIVersion;
+import org.opendaylight.controller.cluster.access.commands.ConnectClientRequest;
+import org.opendaylight.controller.cluster.access.commands.ConnectClientSuccess;
+import org.opendaylight.controller.cluster.access.commands.LocalHistoryRequest;
+import org.opendaylight.controller.cluster.access.commands.NotLeaderException;
+import org.opendaylight.controller.cluster.access.commands.OutOfSequenceEnvelopeException;
+import org.opendaylight.controller.cluster.access.commands.TransactionRequest;
+import org.opendaylight.controller.cluster.access.concepts.ClientIdentifier;
+import org.opendaylight.controller.cluster.access.concepts.FrontendIdentifier;
+import org.opendaylight.controller.cluster.access.concepts.LocalHistoryIdentifier;
+import org.opendaylight.controller.cluster.access.concepts.Request;
+import org.opendaylight.controller.cluster.access.concepts.RequestEnvelope;
+import org.opendaylight.controller.cluster.access.concepts.RequestException;
+import org.opendaylight.controller.cluster.access.concepts.RequestSuccess;
+import org.opendaylight.controller.cluster.access.concepts.RetiredGenerationException;
+import org.opendaylight.controller.cluster.access.concepts.RuntimeRequestException;
+import org.opendaylight.controller.cluster.access.concepts.SliceableMessage;
+import org.opendaylight.controller.cluster.access.concepts.TransactionIdentifier;
+import org.opendaylight.controller.cluster.access.concepts.UnsupportedRequestException;
import org.opendaylight.controller.cluster.common.actor.CommonConfig;
+import org.opendaylight.controller.cluster.common.actor.Dispatchers;
+import org.opendaylight.controller.cluster.common.actor.Dispatchers.DispatcherType;
+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.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.ShardDataTreeListenerInfoMXBeanImpl;
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.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.ForwardedReadyTransaction;
+import org.opendaylight.controller.cluster.datastore.messages.GetKnownClients;
+import org.opendaylight.controller.cluster.datastore.messages.GetKnownClientsReply;
+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.ReadyTransactionReply;
-import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListener;
-import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListenerReply;
+import org.opendaylight.controller.cluster.datastore.messages.ReadyLocalTransaction;
+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.SerializationUtils;
+import org.opendaylight.controller.cluster.datastore.persisted.DatastoreSnapshot;
+import org.opendaylight.controller.cluster.datastore.persisted.DatastoreSnapshot.ShardSnapshot;
+import org.opendaylight.controller.cluster.datastore.persisted.DisableTrackingPayload;
+import org.opendaylight.controller.cluster.messaging.MessageAssembler;
+import org.opendaylight.controller.cluster.messaging.MessageSlicer;
+import org.opendaylight.controller.cluster.messaging.SliceOptions;
+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.ReplicatedLogEntry;
-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.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.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.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.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.yangtools.concepts.Identifier;
+import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTree;
+import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException;
+import org.opendaylight.yangtools.yang.data.api.schema.tree.TreeType;
import org.opendaylight.yangtools.yang.model.api.SchemaContext;
-import scala.concurrent.duration.Duration;
+import org.opendaylight.yangtools.yang.model.api.SchemaContextProvider;
import scala.concurrent.duration.FiniteDuration;
/**
- * A Shard represents a portion of the logical data tree
+ * A Shard represents a portion of the logical data tree.
+ *
*
- * 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
*/
public class Shard extends RaftActor {
- private static final YangInstanceIdentifier DATASTORE_ROOT = YangInstanceIdentifier.builder().build();
-
- private static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = "txCommitTimeoutCheck";
+ @VisibleForTesting
+ static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = new Object() {
+ @Override
+ public String toString() {
+ return "txCommitTimeoutCheck";
+ }
+ };
@VisibleForTesting
- static final String DEFAULT_NAME = "default";
+ static final Object GET_SHARD_MBEAN_MESSAGE = new Object() {
+ @Override
+ public String toString() {
+ return "getShardMBeanMessage";
+ }
+ };
- // The state of this Shard
- private final InMemoryDOMDataStore store;
+ static final Object RESUME_NEXT_PENDING_TRANSACTION = new Object() {
+ @Override
+ public String toString() {
+ return "resumeNextPendingTransaction";
+ }
+ };
- /// The name of this shard
- private final ShardIdentifier name;
+ // 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 ShardStats shardMBean;
+ private static final Collection SUPPORTED_ABIVERSIONS;
+
+ static {
+ final ABIVersion[] values = ABIVersion.values();
+ final ABIVersion[] real = Arrays.copyOfRange(values, 1, values.length - 1);
+ SUPPORTED_ABIVERSIONS = ImmutableList.copyOf(real).reverse();
+ }
- private final List dataChangeListeners = Lists.newArrayList();
+ // FIXME: make this a dynamic property based on mailbox size and maximum number of clients
+ private static final int CLIENT_MAX_MESSAGES = 1000;
- private final List delayedListenerRegistrations =
- Lists.newArrayList();
+ // The state of this Shard
+ private final ShardDataTree store;
- private final DatastoreContext datastoreContext;
+ /// The name of this shard
+ private final String name;
- private final DataPersistenceProvider dataPersistenceProvider;
+ private final String shardName;
+
+ private final ShardStats shardMBean;
- private SchemaContext schemaContext;
+ private final ShardDataTreeListenerInfoMXBeanImpl listenerInfoMXBean;
- private int createSnapshotTransactionCounter;
+ private DatastoreContext datastoreContext;
private final ShardCommitCoordinator commitCoordinator;
- private final long transactionCommitTimeout;
+ private long transactionCommitTimeout;
private Cancellable txCommitTimeoutCheckSchedule;
private final Optional roleChangeNotifier;
- /**
- * Coordinates persistence recovery on startup.
- */
- private ShardRecoveryCoordinator recoveryCoordinator;
- private List currentLogRecoveryBatch;
+ private final MessageTracker appendEntriesReplyTracker;
- private final Map transactionChains = new HashMap<>();
+ private final ShardTransactionActorFactory transactionActorFactory;
- protected Shard(final ShardIdentifier name, final Map peerAddresses,
- final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
- super(name.toString(), mapPeerAddresses(peerAddresses),
- Optional.of(datastoreContext.getShardRaftConfig()));
+ private final ShardSnapshotCohort snapshotCohort;
- this.name = name;
- this.datastoreContext = datastoreContext;
- this.schemaContext = schemaContext;
- this.dataPersistenceProvider = (datastoreContext.isPersistent()) ? new PersistentDataProvider() : new NonPersistentRaftDataProvider();
+ private final DataTreeChangeListenerSupport treeChangeSupport = new DataTreeChangeListenerSupport(this);
- LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
+ private ShardSnapshot restoreFromSnapshot;
+
+ private final ShardTransactionMessageRetrySupport messageRetrySupport;
+
+ @VisibleForTesting
+ final FrontendMetadata frontendMetadata;
+
+ private Map knownFrontends = ImmutableMap.of();
+ private boolean paused;
+
+ private final MessageSlicer responseMessageSlicer;
+ private final Dispatchers dispatchers;
+
+ private final MessageAssembler requestMessageAssembler;
+
+ protected Shard(final AbstractBuilder, ?> builder) {
+ super(builder.getId().toString(), builder.getPeerAddresses(),
+ Optional.of(builder.getDatastoreContext().getShardRaftConfig()), DataStoreVersions.CURRENT_VERSION);
- store = InMemoryDOMDataStoreFactory.create(name.toString(), null,
- datastoreContext.getDataStoreProperties());
+ this.name = builder.getId().toString();
+ this.shardName = builder.getId().getShardName();
+ this.datastoreContext = builder.getDatastoreContext();
+ this.restoreFromSnapshot = builder.getRestoreFromSnapshot();
+ this.frontendMetadata = new FrontendMetadata(name);
- if(schemaContext != null) {
- store.onGlobalContextUpdated(schemaContext);
+ setPersistence(datastoreContext.isPersistent());
+
+ LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
+
+ ShardDataTreeChangeListenerPublisherActorProxy treeChangeListenerPublisher =
+ new ShardDataTreeChangeListenerPublisherActorProxy(getContext(), name + "-DTCL-publisher", name);
+ if (builder.getDataTree() != null) {
+ store = new ShardDataTree(this, builder.getSchemaContext(), builder.getDataTree(),
+ treeChangeListenerPublisher, name, frontendMetadata);
+ } else {
+ store = new ShardDataTree(this, builder.getSchemaContext(), builder.getTreeType(),
+ builder.getDatastoreContext().getStoreRoot(), treeChangeListenerPublisher, name, frontendMetadata);
}
- shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
- datastoreContext.getDataStoreMXBeanType());
- shardMBean.setNotificationManager(store.getDataChangeListenerNotificationManager());
+ shardMBean = ShardMBeanFactory.getShardStatsMBean(name, datastoreContext.getDataStoreMXBeanType(), this);
if (isMetricsCaptureEnabled()) {
getContext().become(new MeteringBehavior(this));
}
- commitCoordinator = new ShardCommitCoordinator(TimeUnit.SECONDS.convert(1, TimeUnit.MINUTES),
- datastoreContext.getShardTransactionCommitQueueCapacity(), LOG, name.toString());
+ commitCoordinator = new ShardCommitCoordinator(store, LOG, this.name);
- transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
- datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS);
+ setTransactionCommitTimeout();
// create a notifier actor for each cluster member
- roleChangeNotifier = createRoleChangeNotifier(name.toString());
- }
+ roleChangeNotifier = createRoleChangeNotifier(name);
- private static Map mapPeerAddresses(
- final Map peerAddresses) {
- Map map = new HashMap<>();
+ appendEntriesReplyTracker = new MessageTracker(AppendEntriesReply.class,
+ getRaftActorContext().getConfigParams().getIsolatedCheckIntervalInMillis());
- for (Map.Entry entry : peerAddresses
- .entrySet()) {
- map.put(entry.getKey().toString(), entry.getValue());
- }
+ dispatchers = new Dispatchers(context().system().dispatchers());
+ transactionActorFactory = new ShardTransactionActorFactory(store, datastoreContext,
+ dispatchers.getDispatcherPath(Dispatchers.DispatcherType.Transaction),
+ self(), getContext(), shardMBean, builder.getId().getShardName());
- return map;
- }
+ snapshotCohort = ShardSnapshotCohort.create(getContext(), builder.getId().getMemberName(), store, LOG,
+ this.name);
+
+ messageRetrySupport = new ShardTransactionMessageRetrySupport(this);
+
+ responseMessageSlicer = MessageSlicer.builder().logContext(this.name)
+ .messageSliceSize(datastoreContext.getMaximumMessageSliceSize())
+ .fileBackedStreamFactory(getRaftActorContext().getFileBackedOutputStreamFactory())
+ .expireStateAfterInactivity(2, TimeUnit.MINUTES).build();
- public static Props props(final ShardIdentifier name,
- final Map 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");
+ requestMessageAssembler = MessageAssembler.builder().logContext(this.name)
+ .fileBackedStreamFactory(getRaftActorContext().getFileBackedOutputStreamFactory())
+ .assembledMessageCallback((message, sender) -> self().tell(message, sender))
+ .expireStateAfterInactivity(datastoreContext.getRequestTimeout(), TimeUnit.NANOSECONDS).build();
- return Props.create(new ShardCreator(name, peerAddresses, datastoreContext, schemaContext));
+ listenerInfoMXBean = new ShardDataTreeListenerInfoMXBeanImpl(name, datastoreContext.getDataStoreMXBeanType(),
+ self());
+ listenerInfoMXBean.register();
}
- private Optional createRoleChangeNotifier(String shardId) {
+ private void setTransactionCommitTimeout() {
+ transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
+ datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS) / 2;
+ }
+
+ private Optional createRoleChangeNotifier(final String shardId) {
ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
- return Optional.of(shardRoleChangeNotifier);
+ return Optional.of(shardRoleChangeNotifier);
}
@Override
- public void postStop() {
+ public void postStop() throws Exception {
+ LOG.info("Stopping Shard {}", persistenceId());
+
super.postStop();
- if(txCommitTimeoutCheckSchedule != null) {
+ messageRetrySupport.close();
+
+ if (txCommitTimeoutCheckSchedule != null) {
txCommitTimeoutCheckSchedule.cancel();
}
+
+ commitCoordinator.abortPendingTransactions("Transaction aborted due to shutdown.", this);
+
+ shardMBean.unregisterMBean();
+ listenerInfoMXBean.unregister();
}
@Override
- public void onReceiveRecover(final Object message) throws Exception {
- if(LOG.isDebugEnabled()) {
- LOG.debug("{}: onReceiveRecover: Received message {} from {}", persistenceId(),
- message.getClass().toString(), getSender());
- }
+ protected void handleRecover(final Object message) {
+ LOG.debug("{}: onReceiveRecover: Received message {} from {}", persistenceId(), message.getClass(),
+ getSender());
- if (message instanceof RecoveryFailure){
- LOG.error("{}: Recovery failed because of this cause",
- persistenceId(), ((RecoveryFailure) message).cause());
-
- // Even though recovery failed, we still need to finish our recovery, eg send the
- // ActorInitialized message and start the txCommitTimeoutCheckSchedule.
- onRecoveryComplete();
- } else {
- super.onReceiveRecover(message);
+ super.handleRecover(message);
+ if (LOG.isTraceEnabled()) {
+ appendEntriesReplyTracker.begin();
}
}
@Override
- public void onReceiveCommand(final Object message) throws Exception {
- if(LOG.isDebugEnabled()) {
- LOG.debug("{}: onReceiveCommand: Received message {} from {}", persistenceId(), message, getSender());
- }
-
- if (message.getClass().equals(CreateTransaction.SERIALIZABLE_CLASS)) {
- handleCreateTransaction(message);
- } else if(message instanceof ForwardedReadyTransaction) {
- handleForwardedReadyTransaction((ForwardedReadyTransaction)message);
- } else if(message.getClass().equals(CanCommitTransaction.SERIALIZABLE_CLASS)) {
- handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
- } else if(message.getClass().equals(CommitTransaction.SERIALIZABLE_CLASS)) {
- handleCommitTransaction(CommitTransaction.fromSerializable(message));
- } else if(message.getClass().equals(AbortTransaction.SERIALIZABLE_CLASS)) {
- handleAbortTransaction(AbortTransaction.fromSerializable(message));
- } 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 PeerAddressResolved) {
- PeerAddressResolved resolved = (PeerAddressResolved) message;
- setPeerAddress(resolved.getPeerId().toString(),
- resolved.getPeerAddress());
- } else if(message.equals(TX_COMMIT_TIMEOUT_CHECK_MESSAGE)) {
- handleTransactionCommitTimeoutCheck();
- } else {
- super.onReceiveCommand(message);
+ protected void handleNonRaftCommand(final Object message) {
+ try (MessageTracker.Context context = appendEntriesReplyTracker.received(message)) {
+ final Optional maybeError = context.error();
+ if (maybeError.isPresent()) {
+ LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(),
+ maybeError.get());
+ }
+
+ store.resetTransactionBatch();
+
+ if (message instanceof RequestEnvelope) {
+ handleRequestEnvelope((RequestEnvelope)message);
+ } else if (MessageAssembler.isHandledMessage(message)) {
+ handleRequestAssemblerMessage(message);
+ } else if (message instanceof ConnectClientRequest) {
+ handleConnectClient((ConnectClientRequest)message);
+ } else 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 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(), resolved.getPeerAddress());
+ } else if (TX_COMMIT_TIMEOUT_CHECK_MESSAGE.equals(message)) {
+ commitTimeoutCheck();
+ } 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) {
+ store.processCohortRegistryCommand(getSender(),
+ (DataTreeCohortActorRegistry.CohortRegistryCommand) message);
+ } else if (message instanceof MakeLeaderLocal) {
+ onMakeLeaderLocal();
+ } else if (RESUME_NEXT_PENDING_TRANSACTION.equals(message)) {
+ store.resumeNextPendingTransaction();
+ } else if (GetKnownClients.INSTANCE.equals(message)) {
+ handleGetKnownClients();
+ } else if (!responseMessageSlicer.handleMessage(message)) {
+ super.handleNonRaftCommand(message);
+ }
}
}
- @Override
- protected Optional getRoleChangeNotifier() {
- return roleChangeNotifier;
+ private void handleRequestAssemblerMessage(final Object message) {
+ dispatchers.getDispatcher(DispatcherType.Serialization).execute(() -> {
+ JavaSerializer.currentSystem().value_$eq((ExtendedActorSystem) context().system());
+ requestMessageAssembler.handleMessage(message, self());
+ });
}
- private void handleTransactionCommitTimeoutCheck() {
- CohortEntry cohortEntry = commitCoordinator.getCurrentCohortEntry();
- if(cohortEntry != null) {
- long elapsed = System.currentTimeMillis() - cohortEntry.getLastAccessTime();
- if(elapsed > transactionCommitTimeout) {
- LOG.warn("{}: Current transaction {} has timed out after {} ms - aborting",
- persistenceId(), cohortEntry.getTransactionID(), transactionCommitTimeout);
-
- doAbortTransaction(cohortEntry.getTransactionID(), null);
+ @SuppressWarnings("checkstyle:IllegalCatch")
+ private void handleRequestEnvelope(final RequestEnvelope envelope) {
+ final long now = ticker().read();
+ try {
+ final RequestSuccess, ?> success = handleRequest(envelope, now);
+ if (success != null) {
+ final long executionTimeNanos = ticker().read() - now;
+ if (success instanceof SliceableMessage) {
+ dispatchers.getDispatcher(DispatcherType.Serialization).execute(() ->
+ responseMessageSlicer.slice(SliceOptions.builder().identifier(success.getTarget())
+ .message(envelope.newSuccessEnvelope(success, executionTimeNanos))
+ .sendTo(envelope.getMessage().getReplyTo()).replyTo(self())
+ .onFailureCallback(t -> LOG.warn("Error slicing response {}", success, t)).build()));
+ } else {
+ envelope.sendSuccess(success, executionTimeNanos);
+ }
}
+ } catch (RequestException e) {
+ LOG.debug("{}: request {} failed", persistenceId(), envelope, e);
+ envelope.sendFailure(e, ticker().read() - now);
+ } catch (Exception e) {
+ LOG.debug("{}: request {} caused failure", persistenceId(), envelope, e);
+ envelope.sendFailure(new RuntimeRequestException("Request failed to process", e),
+ ticker().read() - now);
}
}
- 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());
- shardMBean.incrementFailedTransactionsCount();
- getSender().tell(new akka.actor.Status.Failure(ex), getSelf());
- return;
+ private void commitTimeoutCheck() {
+ store.checkForExpiredTransactions(transactionCommitTimeout, this::updateAccess);
+ commitCoordinator.checkForExpiredTransactions(transactionCommitTimeout, this);
+ requestMessageAssembler.checkExpiredAssembledMessageState();
+ }
+
+ private OptionalLong updateAccess(final SimpleShardDataTreeCohort cohort) {
+ final FrontendIdentifier frontend = cohort.getIdentifier().getHistoryId().getClientId().getFrontendId();
+ final LeaderFrontendState state = knownFrontends.get(frontend);
+ if (state == null) {
+ // Not tell-based protocol, do nothing
+ return OptionalLong.empty();
}
- // 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.
+ if (isIsolatedLeader()) {
+ // We are isolated and no new request can come through until we emerge from it. We are still updating
+ // liveness of frontend when we see it attempting to communicate. Use the last access timer.
+ return OptionalLong.of(state.getLastSeenTicks());
+ }
- 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 we can
- // apply modification to the state immediately
- if(!hasFollowers() && !persistence().isRecoveryApplicable()){
- applyModificationToState(getSender(), transactionID, cohortEntry.getModification());
- } else {
- Shard.this.persistData(getSender(), transactionID,
- new ModificationPayload(cohortEntry.getModification()));
- }
- } catch (Exception e) {
- LOG.error("{} An exception occurred while preCommitting transaction {}",
- persistenceId(), cohortEntry.getTransactionID(), e);
- shardMBean.incrementFailedTransactionsCount();
- getSender().tell(new akka.actor.Status.Failure(e), getSelf());
- }
-
- cohortEntry.updateLastAccessTime();
- }
-
- private void finishCommit(@Nonnull final ActorRef sender, final @Nonnull String 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.
-
- 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) {
- commitWithNewTransaction(cohortEntry.getModification());
- sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
+ // If this frontend has freshly connected, give it some time to catch up before killing its transactions.
+ return OptionalLong.of(state.getLastConnectTicks());
+ }
+
+ private void disableTracking(final DisableTrackingPayload payload) {
+ final ClientIdentifier clientId = payload.getIdentifier();
+ LOG.debug("{}: disabling tracking of {}", persistenceId(), clientId);
+ frontendMetadata.disableTracking(clientId);
+
+ if (isLeader()) {
+ final FrontendIdentifier frontendId = clientId.getFrontendId();
+ final LeaderFrontendState frontend = knownFrontends.get(frontendId);
+ if (frontend != null) {
+ if (clientId.equals(frontend.getIdentifier())) {
+ if (!(frontend instanceof LeaderFrontendState.Disabled)) {
+ verify(knownFrontends.replace(frontendId, frontend,
+ new LeaderFrontendState.Disabled(persistenceId(), clientId, store)));
+ LOG.debug("{}: leader state for {} disabled", persistenceId(), clientId);
+ } else {
+ LOG.debug("{}: leader state {} is already disabled", persistenceId(), frontend);
+ }
+ } else {
+ LOG.debug("{}: leader state {} does not match {}", persistenceId(), frontend, clientId);
+ }
} 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());
+ LOG.debug("{}: leader state for {} not found", persistenceId(), clientId);
+ knownFrontends.put(frontendId, new LeaderFrontendState.Disabled(persistenceId(), clientId,
+ getDataStore()));
}
+ }
+ }
+ private void onMakeLeaderLocal() {
+ LOG.debug("{}: onMakeLeaderLocal received", persistenceId());
+ if (isLeader()) {
+ getSender().tell(new Status.Success(null), getSelf());
return;
}
- LOG.debug("{}: Finishing commit for transaction {}", persistenceId(), cohortEntry.getTransactionID());
+ final ActorSelection leader = getLeader();
- 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();
+ 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
- sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
+ // 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;
+ }
- shardMBean.incrementCommittedTransactionCount();
- shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
+ leader.tell(new RequestLeadership(getId(), getSender()), getSelf());
+ }
- } catch (Exception e) {
- sender.tell(new akka.actor.Status.Failure(e), getSelf());
+ // Acquire our frontend tracking handle and verify generation matches
+ private @Nullable LeaderFrontendState findFrontend(final ClientIdentifier clientId) throws RequestException {
+ final LeaderFrontendState existing = knownFrontends.get(clientId.getFrontendId());
+ if (existing != null) {
+ final int cmp = Long.compareUnsigned(existing.getIdentifier().getGeneration(), clientId.getGeneration());
+ if (cmp == 0) {
+ existing.touch();
+ return existing;
+ }
+ if (cmp > 0) {
+ LOG.debug("{}: rejecting request from outdated client {}", persistenceId(), clientId);
+ throw new RetiredGenerationException(clientId.getGeneration(),
+ existing.getIdentifier().getGeneration());
+ }
- LOG.error("{}, An exception occurred while committing transaction {}", persistenceId(),
- transactionID, e);
- shardMBean.incrementFailedTransactionsCount();
- } finally {
- commitCoordinator.currentTransactionComplete(transactionID, true);
+ LOG.info("{}: retiring state {}, outdated by request from client {}", persistenceId(), existing, clientId);
+ existing.retire();
+ knownFrontends.remove(clientId.getFrontendId());
+ } else {
+ LOG.debug("{}: client {} is not yet known", persistenceId(), clientId);
}
- }
- private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
- LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionID());
- commitCoordinator.handleCanCommit(canCommit, getSender(), self());
+ return null;
}
- private void handleForwardedReadyTransaction(ForwardedReadyTransaction ready) {
- LOG.debug("{}: Readying transaction {}, client version {}", persistenceId(),
- ready.getTransactionID(), ready.getTxnClientVersion());
-
- // 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(),
- ready.getModification());
-
- // 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.
- ActorRef replyActorPath = self();
- if(ready.getTxnClientVersion() < DataStoreVersions.HELIUM_1_VERSION) {
- LOG.debug("{}: Creating BackwardsCompatibleThreePhaseCommitCohort", persistenceId());
- replyActorPath = getContext().actorOf(BackwardsCompatibleThreePhaseCommitCohort.props(
- ready.getTransactionID()));
+ private LeaderFrontendState getFrontend(final ClientIdentifier clientId) throws RequestException {
+ final LeaderFrontendState ret = findFrontend(clientId);
+ if (ret != null) {
+ return ret;
}
- ReadyTransactionReply readyTransactionReply = new ReadyTransactionReply(
- Serialization.serializedActorPath(replyActorPath));
- getSender().tell(ready.isReturnSerialized() ? readyTransactionReply.toSerializable() :
- readyTransactionReply, getSelf());
+ // TODO: a dedicated exception would be better, but this is technically true, too
+ throw new OutOfSequenceEnvelopeException(0);
}
- private void handleAbortTransaction(final AbortTransaction abort) {
- doAbortTransaction(abort.getTransactionID(), getSender());
- }
+ private static @NonNull ABIVersion selectVersion(final ConnectClientRequest message) {
+ final Range clientRange = Range.closed(message.getMinVersion(), message.getMaxVersion());
+ for (ABIVersion v : SUPPORTED_ABIVERSIONS) {
+ if (clientRange.contains(v)) {
+ return v;
+ }
+ }
- void doAbortTransaction(final String transactionID, final ActorRef sender) {
- final CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
- if(cohortEntry != null) {
- LOG.debug("{}: Aborting transaction {}", persistenceId(), transactionID);
+ throw new IllegalArgumentException(String.format(
+ "No common version between backend versions %s and client versions %s", SUPPORTED_ABIVERSIONS,
+ clientRange));
+ }
- // 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);
+ @SuppressWarnings("checkstyle:IllegalCatch")
+ private void handleConnectClient(final ConnectClientRequest message) {
+ try {
+ final ClientIdentifier clientId = message.getTarget();
+ final LeaderFrontendState existing = findFrontend(clientId);
+ if (existing != null) {
+ existing.touch();
+ }
- final ListenableFuture future = cohortEntry.getCohort().abort();
- final ActorRef self = getSelf();
+ if (!isLeader() || !isLeaderActive()) {
+ LOG.info("{}: not currently leader, rejecting request {}. isLeader: {}, isLeaderActive: {},"
+ + "isLeadershipTransferInProgress: {}.",
+ persistenceId(), message, isLeader(), isLeaderActive(), isLeadershipTransferInProgress());
+ throw new NotLeaderException(getSelf());
+ }
- Futures.addCallback(future, new FutureCallback() {
- @Override
- public void onSuccess(final Void v) {
- shardMBean.incrementAbortTransactionsCount();
+ final ABIVersion selectedVersion = selectVersion(message);
+ final LeaderFrontendState frontend;
+ if (existing == null) {
+ frontend = new LeaderFrontendState.Enabled(persistenceId(), clientId, store);
+ knownFrontends.put(clientId.getFrontendId(), frontend);
+ LOG.debug("{}: created state {} for client {}", persistenceId(), frontend, clientId);
+ } else {
+ frontend = existing;
+ }
- if(sender != null) {
- sender.tell(AbortTransactionReply.INSTANCE.toSerializable(), self);
- }
- }
+ frontend.reconnect();
+ message.getReplyTo().tell(new ConnectClientSuccess(message.getTarget(), message.getSequence(), getSelf(),
+ ImmutableList.of(), store.getDataTree(), CLIENT_MAX_MESSAGES).toVersion(selectedVersion),
+ ActorRef.noSender());
+ } catch (RequestException | RuntimeException e) {
+ message.getReplyTo().tell(new Failure(e), ActorRef.noSender());
+ }
+ }
- @Override
- public void onFailure(final Throwable t) {
- LOG.error("{}: An exception happened during abort", persistenceId(), t);
+ private @Nullable RequestSuccess, ?> handleRequest(final RequestEnvelope envelope, final long now)
+ throws RequestException {
+ // We are not the leader, hence we want to fail-fast.
+ if (!isLeader() || paused || !isLeaderActive()) {
+ LOG.debug("{}: not currently active leader, rejecting request {}. isLeader: {}, isLeaderActive: {},"
+ + "isLeadershipTransferInProgress: {}, paused: {}",
+ persistenceId(), envelope, isLeader(), isLeaderActive(), isLeadershipTransferInProgress(), paused);
+ throw new NotLeaderException(getSelf());
+ }
- if(sender != null) {
- sender.tell(new akka.actor.Status.Failure(t), self);
- }
- }
- });
+ final Request, ?> request = envelope.getMessage();
+ if (request instanceof TransactionRequest) {
+ final TransactionRequest> txReq = (TransactionRequest>)request;
+ final ClientIdentifier clientId = txReq.getTarget().getHistoryId().getClientId();
+ return getFrontend(clientId).handleTransactionRequest(txReq, envelope, now);
+ } else if (request instanceof LocalHistoryRequest) {
+ final LocalHistoryRequest> lhReq = (LocalHistoryRequest>)request;
+ final ClientIdentifier clientId = lhReq.getTarget().getClientId();
+ return getFrontend(clientId).handleLocalHistoryRequest(lhReq, envelope, now);
+ } else {
+ LOG.warn("{}: rejecting unsupported request {}", persistenceId(), request);
+ throw new UnsupportedRequestException(request);
}
}
- private void handleCreateTransaction(final Object message) {
+ private void handleGetKnownClients() {
+ final ImmutableSet clients;
if (isLeader()) {
- createTransaction(CreateTransaction.fromSerializable(message));
- } else if (getLeader() != null) {
- getLeader().forward(message, getContext());
+ clients = knownFrontends.values().stream()
+ .map(LeaderFrontendState::getIdentifier)
+ .collect(ImmutableSet.toImmutableSet());
} 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());
+ clients = frontendMetadata.getClients();
}
+ sender().tell(new GetKnownClientsReply(clients), self());
}
- private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
- DOMStoreTransactionChain chain =
- transactionChains.remove(closeTransactionChain.getTransactionChainId());
-
- if(chain != null) {
- chain.close();
- }
+ private boolean hasLeader() {
+ return getLeaderId() != null;
}
- private ActorRef createTypedTransactionActor(int transactionType,
- ShardTransactionIdentifier transactionId, String transactionChainId,
- short clientVersion ) {
-
- DOMStoreTransactionFactory factory = store;
-
- if(!transactionChainId.isEmpty()) {
- factory = transactionChains.get(transactionChainId);
- if(factory == null){
- DOMStoreTransactionChain transactionChain = store.createTransactionChain();
- transactionChains.put(transactionChainId, transactionChain);
- factory = transactionChain;
- }
- }
-
- if(this.schemaContext == null) {
- throw new IllegalStateException("SchemaContext is not set");
- }
-
- if (transactionType == TransactionProxy.TransactionType.READ_ONLY.ordinal()) {
+ public int getPendingTxCommitQueueSize() {
+ return store.getQueueSize();
+ }
- shardMBean.incrementReadOnlyTransactionCount();
+ public int getCohortCacheSize() {
+ return commitCoordinator.getCohortCacheSize();
+ }
- return getContext().actorOf(
- ShardTransaction.props(factory.newReadOnlyTransaction(), getSelf(),
- schemaContext,datastoreContext, shardMBean,
- transactionId.getRemoteTransactionId(), clientVersion),
- transactionId.toString());
+ @Override
+ protected Optional getRoleChangeNotifier() {
+ return roleChangeNotifier;
+ }
- } else if (transactionType == TransactionProxy.TransactionType.READ_WRITE.ordinal()) {
+ String getShardName() {
+ return shardName;
+ }
- shardMBean.incrementReadWriteTransactionCount();
+ @Override
+ protected LeaderStateChanged newLeaderStateChanged(final String memberId, final String leaderId,
+ final short leaderPayloadVersion) {
+ return isLeader() ? new ShardLeaderStateChanged(memberId, leaderId, store.getDataTree(), leaderPayloadVersion)
+ : new ShardLeaderStateChanged(memberId, leaderId, leaderPayloadVersion);
+ }
- return getContext().actorOf(
- ShardTransaction.props(factory.newReadWriteTransaction(), getSelf(),
- schemaContext, datastoreContext, shardMBean,
- transactionId.getRemoteTransactionId(), clientVersion),
- transactionId.toString());
+ protected void onDatastoreContext(final DatastoreContext context) {
+ datastoreContext = context;
+ setTransactionCommitTimeout();
- } else if (transactionType == TransactionProxy.TransactionType.WRITE_ONLY.ordinal()) {
+ setPersistence(datastoreContext.isPersistent());
- shardMBean.incrementWriteOnlyTransactionCount();
+ updateConfigParams(datastoreContext.getShardRaftConfig());
+ }
- return getContext().actorOf(
- ShardTransaction.props(factory.newWriteOnlyTransaction(), getSelf(),
- schemaContext, datastoreContext, shardMBean,
- transactionId.getRemoteTransactionId(), clientVersion),
- transactionId.toString());
+ // applyState() will be invoked once consensus is reached on the payload
+ void persistPayload(final Identifier id, final Payload payload, final boolean batchHint) {
+ boolean canSkipPayload = !hasFollowers() && !persistence().isRecoveryApplicable();
+ if (canSkipPayload) {
+ applyState(self(), id, payload);
} else {
- throw new IllegalArgumentException(
- "Shard="+name + ":CreateTransaction message has unidentified transaction type="
- + transactionType);
+ // We are faking the sender
+ persistData(self(), id, payload, batchHint);
}
}
- private void createTransaction(CreateTransaction createTransaction) {
- try {
- ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
- createTransaction.getTransactionId(), createTransaction.getTransactionChainId(),
- createTransaction.getVersion());
-
- getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
- createTransaction.getTransactionId()).toSerializable(), getSelf());
- } catch (Exception e) {
- getSender().tell(new akka.actor.Status.Failure(e), getSelf());
+ private void handleCommitTransaction(final CommitTransaction commit) {
+ final TransactionIdentifier txId = commit.getTransactionId();
+ if (isLeader()) {
+ askProtocolEncountered(txId);
+ commitCoordinator.handleCommit(txId, getSender(), this);
+ } else {
+ ActorSelection leader = getLeader();
+ if (leader == null) {
+ messageRetrySupport.addMessageToRetry(commit, getSender(), "Could not commit transaction " + txId);
+ } else {
+ LOG.debug("{}: Forwarding CommitTransaction to leader {}", persistenceId(), leader);
+ leader.forward(commit, getContext());
+ }
}
}
- private ActorRef createTransaction(int transactionType, String remoteTransactionId,
- String transactionChainId, short clientVersion) {
-
- ShardTransactionIdentifier transactionId =
- ShardTransactionIdentifier.builder()
- .remoteTransactionId(remoteTransactionId)
- .build();
+ private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
+ final TransactionIdentifier txId = canCommit.getTransactionId();
+ LOG.debug("{}: Can committing transaction {}", persistenceId(), txId);
- if(LOG.isDebugEnabled()) {
- LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
+ if (isLeader()) {
+ askProtocolEncountered(txId);
+ commitCoordinator.handleCanCommit(txId, getSender(), this);
+ } else {
+ ActorSelection leader = getLeader();
+ if (leader == null) {
+ messageRetrySupport.addMessageToRetry(canCommit, getSender(),
+ "Could not canCommit transaction " + txId);
+ } else {
+ LOG.debug("{}: Forwarding CanCommitTransaction to leader {}", persistenceId(), leader);
+ leader.forward(canCommit, getContext());
+ }
}
+ }
- ActorRef transactionActor = createTypedTransactionActor(transactionType, transactionId,
- transactionChainId, clientVersion);
+ @SuppressWarnings("checkstyle:IllegalCatch")
+ protected void handleBatchedModificationsLocal(final BatchedModifications batched, final ActorRef sender) {
+ askProtocolEncountered(batched.getTransactionId());
- return transactionActor;
+ try {
+ commitCoordinator.handleBatchedModifications(batched, sender, this);
+ } catch (Exception e) {
+ LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(),
+ batched.getTransactionId(), e);
+ sender.tell(new Failure(e), getSelf());
+ }
}
- private void syncCommitTransaction(final DOMStoreWriteTransaction transaction)
- throws ExecutionException, InterruptedException {
- DOMStoreThreePhaseCommitCohort commitCohort = transaction.ready();
- commitCohort.preCommit().get();
- commitCohort.commit().get();
+ private void handleBatchedModifications(final 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 process BatchedModifications " + 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 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(final Modification modification) {
- DOMStoreWriteTransaction tx = store.newWriteOnlyTransaction();
- modification.apply(tx);
- try {
- syncCommitTransaction(tx);
- shardMBean.incrementCommittedTransactionCount();
- shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
- } catch (InterruptedException | ExecutionException e) {
- shardMBean.incrementFailedTransactionsCount();
- LOG.error("{}: Failed to commit", persistenceId(), e);
+ private boolean failIfIsolatedLeader(final ActorRef sender) {
+ if (isIsolatedLeader()) {
+ sender.tell(new 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;
}
- }
- private void updateSchemaContext(final UpdateSchemaContext message) {
- this.schemaContext = message.getSchemaContext();
- updateSchemaContext(message.getSchemaContext());
- store.onGlobalContextUpdated(message.getSchemaContext());
+ return false;
}
- @VisibleForTesting
- void updateSchemaContext(final SchemaContext schemaContext) {
- store.onGlobalContextUpdated(schemaContext);
+ protected boolean isIsolatedLeader() {
+ return getRaftState() == RaftState.IsolatedLeader;
}
- private void registerChangeListener(final RegisterChangeListener registerChangeListener) {
+ @SuppressWarnings("checkstyle:IllegalCatch")
+ private void handleReadyLocalTransaction(final ReadyLocalTransaction message) {
+ LOG.debug("{}: handleReadyLocalTransaction for {}", persistenceId(), message.getTransactionId());
+
+ boolean isLeaderActive = isLeaderActive();
+ if (isLeader() && isLeaderActive) {
+ try {
+ commitCoordinator.handleReadyLocalTransaction(message, getSender(), this);
+ } catch (Exception e) {
+ LOG.error("{}: Error handling ReadyLocalTransaction for Tx {}", persistenceId(),
+ message.getTransactionId(), e);
+ getSender().tell(new Failure(e), getSelf());
+ }
+ } else {
+ ActorSelection leader = getLeader();
+ if (!isLeaderActive || leader == null) {
+ messageRetrySupport.addMessageToRetry(message, getSender(),
+ "Could not process ready local transaction " + message.getTransactionId());
+ } else {
+ LOG.debug("{}: Forwarding ReadyLocalTransaction to leader {}", persistenceId(), leader);
+ message.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
+ leader.forward(message, getContext());
+ }
+ }
+ }
- LOG.debug("{}: registerDataChangeListener for {}", persistenceId(), registerChangeListener.getPath());
+ private void handleForwardedReadyTransaction(final ForwardedReadyTransaction forwardedReady) {
+ LOG.debug("{}: handleForwardedReadyTransaction for {}", persistenceId(), forwardedReady.getTransactionId());
- ListenerRegistration>> registration;
- if(isLeader()) {
- registration = doChangeListenerRegistration(registerChangeListener);
+ boolean isLeaderActive = isLeaderActive();
+ if (isLeader() && isLeaderActive) {
+ askProtocolEncountered(forwardedReady.getTransactionId());
+ commitCoordinator.handleForwardedReadyTransaction(forwardedReady, getSender(), this);
} else {
- LOG.debug("{}: Shard is not the leader - delaying registration", persistenceId());
+ ActorSelection leader = getLeader();
+ if (!isLeaderActive || leader == null) {
+ messageRetrySupport.addMessageToRetry(forwardedReady, getSender(),
+ "Could not process forwarded ready transaction " + forwardedReady.getTransactionId());
+ } else {
+ LOG.debug("{}: Forwarding ForwardedReadyTransaction to leader {}", persistenceId(), leader);
- DelayedListenerRegistration delayedReg =
- new DelayedListenerRegistration(registerChangeListener);
- delayedListenerRegistrations.add(delayedReg);
- registration = delayedReg;
+ ReadyLocalTransaction readyLocal = new ReadyLocalTransaction(forwardedReady.getTransactionId(),
+ forwardedReady.getTransaction().getSnapshot(), forwardedReady.isDoImmediateCommit(),
+ forwardedReady.getParticipatingShardNames());
+ readyLocal.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
+ leader.forward(readyLocal, getContext());
+ }
}
+ }
- ActorRef listenerRegistration = getContext().actorOf(
- DataChangeListenerRegistration.props(registration));
-
- LOG.debug("{}: registerDataChangeListener sending reply, listenerRegistrationPath = {} ",
- persistenceId(), listenerRegistration.path());
+ private void handleAbortTransaction(final AbortTransaction abort) {
+ final TransactionIdentifier transactionId = abort.getTransactionId();
+ askProtocolEncountered(transactionId);
+ doAbortTransaction(transactionId, getSender());
+ }
- getSender().tell(new RegisterChangeListenerReply(listenerRegistration.path()), getSelf());
+ void doAbortTransaction(final Identifier transactionID, final ActorRef sender) {
+ commitCoordinator.handleAbort(transactionID, sender, this);
}
- private ListenerRegistration>> doChangeListenerRegistration(
- final RegisterChangeListener registerChangeListener) {
+ private void handleCreateTransaction(final Object message) {
+ if (isLeader()) {
+ createTransaction(CreateTransaction.fromSerializable(message));
+ } else if (getLeader() != null) {
+ getLeader().forward(message, getContext());
+ } else {
+ getSender().tell(new Failure(new NoShardLeaderException(
+ "Could not create a shard transaction", persistenceId())), getSelf());
+ }
+ }
- ActorSelection dataChangeListenerPath = getContext().system().actorSelection(
- registerChangeListener.getDataChangeListenerPath());
+ private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
+ if (isLeader()) {
+ final LocalHistoryIdentifier id = closeTransactionChain.getIdentifier();
+ askProtocolEncountered(id.getClientId());
+ store.closeTransactionChain(id);
+ } else if (getLeader() != null) {
+ getLeader().forward(closeTransactionChain, getContext());
+ } else {
+ LOG.warn("{}: Could not close transaction {}", persistenceId(), closeTransactionChain.getIdentifier());
+ }
+ }
- // 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());
+ @SuppressWarnings("checkstyle:IllegalCatch")
+ private void createTransaction(final CreateTransaction createTransaction) {
+ askProtocolEncountered(createTransaction.getTransactionId());
- // 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);
+ try {
+ if (TransactionType.fromInt(createTransaction.getTransactionType()) != TransactionType.READ_ONLY
+ && failIfIsolatedLeader(getSender())) {
+ return;
+ }
- AsyncDataChangeListener> listener =
- new DataChangeListenerProxy(dataChangeListenerPath);
+ ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
+ createTransaction.getTransactionId());
- LOG.debug("{}: Registering for path {}", persistenceId(), registerChangeListener.getPath());
+ getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
+ createTransaction.getTransactionId(), createTransaction.getVersion()).toSerializable(), getSelf());
+ } catch (Exception e) {
+ getSender().tell(new Failure(e), getSelf());
+ }
+ }
- return store.registerChangeListener(registerChangeListener.getPath(), listener,
- registerChangeListener.getScope());
+ private ActorRef createTransaction(final int transactionType, final TransactionIdentifier transactionId) {
+ LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
+ return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType),
+ transactionId);
}
- private boolean isMetricsCaptureEnabled(){
- CommonConfig config = new CommonConfig(getContext().system().settings().config());
- return config.isMetricCaptureEnabled();
+ // Called on leader only
+ private void askProtocolEncountered(final TransactionIdentifier transactionId) {
+ askProtocolEncountered(transactionId.getHistoryId().getClientId());
}
- @Override
- protected
- void startLogRecoveryBatch(final int maxBatchSize) {
- currentLogRecoveryBatch = Lists.newArrayListWithCapacity(maxBatchSize);
+ // Called on leader only
+ private void askProtocolEncountered(final ClientIdentifier clientId) {
+ final FrontendIdentifier frontend = clientId.getFrontendId();
+ final LeaderFrontendState state = knownFrontends.get(frontend);
+ if (!(state instanceof LeaderFrontendState.Disabled)) {
+ LOG.debug("{}: encountered ask-based client {}, disabling transaction tracking", persistenceId(), clientId);
+ if (knownFrontends.isEmpty()) {
+ knownFrontends = new HashMap<>();
+ }
+ knownFrontends.put(frontend, new LeaderFrontendState.Disabled(persistenceId(), clientId, getDataStore()));
- if(LOG.isDebugEnabled()) {
- LOG.debug("{}: starting log recovery batch with max size {}", persistenceId(), maxBatchSize);
+ persistPayload(clientId, DisableTrackingPayload.create(clientId,
+ datastoreContext.getInitialPayloadSerializedBufferCapacity()), false);
}
}
- @Override
- protected void appendRecoveredLogEntry(final Payload data) {
- if(data instanceof ModificationPayload) {
- try {
- currentLogRecoveryBatch.add(((ModificationPayload) data).getModification());
- } catch (ClassNotFoundException | IOException e) {
- LOG.error("{}: Error extracting ModificationPayload", persistenceId(), e);
- }
- } else if (data instanceof CompositeModificationPayload) {
- currentLogRecoveryBatch.add(((CompositeModificationPayload) data).getModification());
- } else if (data instanceof CompositeModificationByteStringPayload) {
- currentLogRecoveryBatch.add(((CompositeModificationByteStringPayload) data).getModification());
- } else {
- LOG.error("{}: Unknown state received {} during recovery", persistenceId(), data);
- }
+ private void updateSchemaContext(final UpdateSchemaContext message) {
+ updateSchemaContext(message.getSchemaContext());
}
- @Override
- protected void applyRecoverySnapshot(final byte[] snapshotBytes) {
- if(recoveryCoordinator == null) {
- recoveryCoordinator = new ShardRecoveryCoordinator(persistenceId(), schemaContext,
- LOG, name.toString());
- }
+ @VisibleForTesting
+ void updateSchemaContext(final SchemaContext schemaContext) {
+ store.updateSchemaContext(schemaContext);
+ }
- recoveryCoordinator.submit(snapshotBytes, store.newWriteOnlyTransaction());
+ private boolean isMetricsCaptureEnabled() {
+ CommonConfig config = new CommonConfig(getContext().system().settings().config());
+ return config.isMetricCaptureEnabled();
+ }
- if(LOG.isDebugEnabled()) {
- LOG.debug("{}: submitted recovery sbapshot", persistenceId());
- }
+ @Override
+ @VisibleForTesting
+ public RaftActorSnapshotCohort getRaftActorSnapshotCohort() {
+ return snapshotCohort;
}
@Override
- protected void applyCurrentLogRecoveryBatch() {
- if(recoveryCoordinator == null) {
- recoveryCoordinator = new ShardRecoveryCoordinator(persistenceId(), schemaContext,
- LOG, name.toString());
+ protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
+ if (restoreFromSnapshot == null) {
+ return ShardRecoveryCoordinator.create(store, persistenceId(), LOG);
}
- recoveryCoordinator.submit(currentLogRecoveryBatch, store.newWriteOnlyTransaction());
-
- if(LOG.isDebugEnabled()) {
- LOG.debug("{}: submitted log recovery batch with size {}", persistenceId(),
- currentLogRecoveryBatch.size());
- }
+ return ShardRecoveryCoordinator.forSnapshot(store, persistenceId(), LOG, restoreFromSnapshot.getSnapshot());
}
@Override
protected void onRecoveryComplete() {
- if(recoveryCoordinator != null) {
- Collection txList = recoveryCoordinator.getTransactions();
-
- if(LOG.isDebugEnabled()) {
- LOG.debug("{}: recovery complete - committing {} Tx's", persistenceId(), txList.size());
- }
-
- for(DOMStoreWriteTransaction tx: txList) {
- try {
- syncCommitTransaction(tx);
- shardMBean.incrementCommittedTransactionCount();
- } catch (InterruptedException | ExecutionException e) {
- shardMBean.incrementFailedTransactionsCount();
- LOG.error("{}: Failed to commit", persistenceId(), e);
- }
- }
- }
-
- recoveryCoordinator = null;
- currentLogRecoveryBatch = null;
- updateJournalStats();
+ restoreFromSnapshot = null;
//notify shard manager
getContext().parent().tell(new ActorInitialized(), getSelf());
// Being paranoid here - this method should only be called once but just in case...
- if(txCommitTimeoutCheckSchedule == null) {
+ 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);
+ FiniteDuration period = FiniteDuration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
period, period, getSelf(),
TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
@@ -743,225 +945,289 @@ public class Shard extends RaftActor {
}
@Override
- protected void applyState(final ActorRef clientActor, final String identifier, final Object data) {
+ protected void applyState(final ActorRef clientActor, final Identifier identifier, final Object data) {
+ if (data instanceof Payload) {
+ if (data instanceof DisableTrackingPayload) {
+ disableTracking((DisableTrackingPayload) data);
+ return;
+ }
- if(data instanceof ModificationPayload) {
try {
- applyModificationToState(clientActor, identifier, ((ModificationPayload) data).getModification());
- } catch (ClassNotFoundException | IOException e) {
- LOG.error("{}: Error extracting ModificationPayload", persistenceId(), e);
+ store.applyReplicatedPayload(identifier, (Payload)data);
+ } catch (DataValidationFailedException | IOException e) {
+ LOG.error("{}: Error applying replica {}", persistenceId(), identifier, e);
}
- }
- else if (data instanceof CompositeModificationPayload) {
- Object modification = ((CompositeModificationPayload) data).getModification();
-
- applyModificationToState(clientActor, identifier, modification);
- } else if(data instanceof CompositeModificationByteStringPayload ){
- Object modification = ((CompositeModificationByteStringPayload) data).getModification();
-
- applyModificationToState(clientActor, identifier, modification);
} else {
- LOG.error("{}: Unknown state received {} Class loader = {} CompositeNodeMod.ClassLoader = {}",
- persistenceId(), data, data.getClass().getClassLoader(),
- CompositeModificationPayload.class.getClassLoader());
+ LOG.error("{}: Unknown state for {} received {}", persistenceId(), identifier, data);
}
-
- updateJournalStats();
-
}
- private void applyModificationToState(ActorRef clientActor, String 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.
- commitWithNewTransaction(MutableCompositeModification.fromSerializable(modification));
- } else {
- // This must be the OK to commit after replication consensus.
- finishCommit(clientActor, identifier);
- }
- }
+ @Override
+ protected void onStateChanged() {
+ boolean isLeader = isLeader();
+ boolean hasLeader = hasLeader();
+ treeChangeSupport.onLeadershipChange(isLeader, hasLeader);
- private void updateJournalStats() {
- ReplicatedLogEntry lastLogEntry = getLastLogEntry();
+ // If this actor is no longer the leader close all the transaction chains
+ if (!isLeader) {
+ if (LOG.isDebugEnabled()) {
+ LOG.debug(
+ "{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader",
+ persistenceId(), getId());
+ }
- if (lastLogEntry != null) {
- shardMBean.setLastLogIndex(lastLogEntry.getIndex());
- shardMBean.setLastLogTerm(lastLogEntry.getTerm());
+ paused = false;
+ store.purgeLeaderState();
}
- shardMBean.setCommitIndex(getCommitIndex());
- shardMBean.setLastApplied(getLastApplied());
- shardMBean.setInMemoryJournalDataSize(getRaftActorContext().getReplicatedLog().dataSize());
+ if (hasLeader && !isIsolatedLeader()) {
+ messageRetrySupport.retryMessages();
+ }
}
@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.
+ protected void onLeaderChanged(final String oldLeader, final String newLeader) {
+ shardMBean.incrementLeadershipChangeCount();
+ paused = false;
+
+ if (!isLeader()) {
+ if (!knownFrontends.isEmpty()) {
+ LOG.debug("{}: removing frontend state for {}", persistenceId(), knownFrontends.keySet());
+ knownFrontends = ImmutableMap.of();
+ }
- ActorRef createSnapshotTransaction = createTransaction(
- TransactionProxy.TransactionType.READ_ONLY.ordinal(),
- "createSnapshot" + ++createSnapshotTransactionCounter, "",
- DataStoreVersions.CURRENT_VERSION);
+ requestMessageAssembler.close();
- createSnapshotTransaction.tell(CreateSnapshot.INSTANCE, self());
- }
+ if (!hasLeader()) {
+ // No leader anywhere, nothing else to do
+ return;
+ }
- @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
+ // 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> messagesToForward = convertPendingTransactionsToMessages();
- LOG.info("{}: Applying snapshot", persistenceId());
- try {
- DOMStoreWriteTransaction transaction = store.newWriteOnlyTransaction();
+ if (!messagesToForward.isEmpty()) {
+ LOG.debug("{}: Forwarding {} pending transaction messages to leader {}", persistenceId(),
+ messagesToForward.size(), leader);
- NormalizedNode, ?> node = SerializationUtils.deserializeNormalizedNode(snapshotBytes);
+ for (Object message : messagesToForward) {
+ LOG.debug("{}: Forwarding pending transaction message {}", persistenceId(), message);
- // delete everything first
- transaction.delete(DATASTORE_ROOT);
+ leader.tell(message, self());
+ }
+ }
+ } else {
+ 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());
+ }
- // 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());
+ if (!isIsolatedLeader()) {
+ messageRetrySupport.retryMessages();
}
}
+ /**
+ * Clears all pending transactions and converts them to messages to be forwarded to a new leader.
+ *
+ * @return the converted messages
+ */
+ public Collection> convertPendingTransactionsToMessages() {
+ return commitCoordinator.convertPendingTransactionsToMessages(
+ datastoreContext.getShardBatchedModificationCount());
+ }
+
@Override
- protected void onStateChanged() {
- boolean isLeader = isLeader();
- for (ActorSelection dataChangeListener : dataChangeListeners) {
- dataChangeListener.tell(new EnableNotification(isLeader), getSelf());
+ protected void pauseLeader(final Runnable operation) {
+ LOG.debug("{}: In pauseLeader, operation: {}", persistenceId(), operation);
+ paused = true;
+
+ // Tell-based protocol can replay transaction state, so it is safe to blow it up when we are paused.
+ if (datastoreContext.isUseTellBasedProtocol()) {
+ knownFrontends.values().forEach(LeaderFrontendState::retire);
+ knownFrontends = ImmutableMap.of();
}
- if(isLeader) {
- for(DelayedListenerRegistration reg: delayedListenerRegistrations) {
- if(!reg.isClosed()) {
- reg.setDelegate(doChangeListenerRegistration(reg.getRegisterChangeListener()));
- }
- }
+ store.setRunOnPendingTransactionsComplete(operation);
+ }
- delayedListenerRegistrations.clear();
- }
+ @Override
+ protected void unpauseLeader() {
+ LOG.debug("{}: In unpauseLeader", persistenceId());
+ paused = false;
- shardMBean.setRaftState(getRaftState().name());
- shardMBean.setCurrentTerm(getCurrentTerm());
+ store.setRunOnPendingTransactionsComplete(null);
- // If this actor is no longer the leader close all the transaction chains
- if(!isLeader){
- for(Map.Entry entry : transactionChains.entrySet()){
- if(LOG.isDebugEnabled()) {
- LOG.debug(
- "{}: onStateChanged: Closing transaction chain {} because shard {} is no longer the leader",
- persistenceId(), entry.getKey(), getId());
- }
- entry.getValue().close();
- }
+ // Restore tell-based protocol state as if we were becoming the leader
+ knownFrontends = Verify.verifyNotNull(frontendMetadata.toLeaderState(this));
+ }
- transactionChains.clear();
- }
+ @Override
+ protected OnDemandRaftState.AbstractBuilder, ?> newOnDemandRaftStateBuilder() {
+ return OnDemandShardState.newBuilder().treeChangeListenerActors(treeChangeSupport.getListenerActors())
+ .commitCohortActors(store.getCohortActors());
}
@Override
- protected DataPersistenceProvider persistence() {
- return dataPersistenceProvider;
+ public String persistenceId() {
+ return this.name;
}
- @Override protected void onLeaderChanged(final String oldLeader, final String newLeader) {
- shardMBean.setLeader(newLeader);
+ @VisibleForTesting
+ ShardCommitCoordinator getCommitCoordinator() {
+ return commitCoordinator;
}
- @Override public String persistenceId() {
- return this.name.toString();
+ public DatastoreContext getDatastoreContext() {
+ return datastoreContext;
}
@VisibleForTesting
- DataPersistenceProvider getDataPersistenceProvider() {
- return dataPersistenceProvider;
+ public ShardDataTree getDataStore() {
+ return store;
}
- private static class ShardCreator implements Creator {
+ @VisibleForTesting
+ ShardStats getShardMBean() {
+ return shardMBean;
+ }
+
+ public static Builder builder() {
+ return new Builder();
+ }
- private static final long serialVersionUID = 1L;
+ public abstract static class AbstractBuilder, S extends Shard> {
+ private final Class extends S> shardClass;
+ private ShardIdentifier id;
+ private Map peerAddresses = Collections.emptyMap();
+ private DatastoreContext datastoreContext;
+ private SchemaContextProvider schemaContextProvider;
+ private DatastoreSnapshot.ShardSnapshot restoreFromSnapshot;
+ private DataTree dataTree;
- final ShardIdentifier name;
- final Map peerAddresses;
- final DatastoreContext datastoreContext;
- final SchemaContext schemaContext;
+ private volatile boolean sealed;
- ShardCreator(final ShardIdentifier name, final Map peerAddresses,
- final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
- this.name = name;
- this.peerAddresses = peerAddresses;
- this.datastoreContext = datastoreContext;
- this.schemaContext = schemaContext;
+ protected AbstractBuilder(final Class extends S> shardClass) {
+ this.shardClass = shardClass;
}
- @Override
- public Shard create() throws Exception {
- return new Shard(name, peerAddresses, datastoreContext, schemaContext);
+ protected void checkSealed() {
+ checkState(!sealed, "Builder isalready sealed - further modifications are not allowed");
}
- }
- @VisibleForTesting
- InMemoryDOMDataStore getDataStore() {
- return store;
- }
+ @SuppressWarnings("unchecked")
+ private T self() {
+ return (T) this;
+ }
- @VisibleForTesting
- ShardStats getShardMBean() {
- return shardMBean;
- }
+ public T id(final ShardIdentifier newId) {
+ checkSealed();
+ this.id = newId;
+ return self();
+ }
- private static class DelayedListenerRegistration implements
- ListenerRegistration>> {
+ public T peerAddresses(final Map newPeerAddresses) {
+ checkSealed();
+ this.peerAddresses = newPeerAddresses;
+ return self();
+ }
- private volatile boolean closed;
+ public T datastoreContext(final DatastoreContext newDatastoreContext) {
+ checkSealed();
+ this.datastoreContext = newDatastoreContext;
+ return self();
+ }
- private final RegisterChangeListener registerChangeListener;
+ public T schemaContextProvider(final SchemaContextProvider newSchemaContextProvider) {
+ checkSealed();
+ this.schemaContextProvider = requireNonNull(newSchemaContextProvider);
+ return self();
+ }
- private volatile ListenerRegistration>> delegate;
+ public T restoreFromSnapshot(final DatastoreSnapshot.ShardSnapshot newRestoreFromSnapshot) {
+ checkSealed();
+ this.restoreFromSnapshot = newRestoreFromSnapshot;
+ return self();
+ }
- DelayedListenerRegistration(final RegisterChangeListener registerChangeListener) {
- this.registerChangeListener = registerChangeListener;
+ public T dataTree(final DataTree newDataTree) {
+ checkSealed();
+ this.dataTree = newDataTree;
+ return self();
}
- void setDelegate( final ListenerRegistration>> registration) {
- this.delegate = registration;
+ public ShardIdentifier getId() {
+ return id;
}
- boolean isClosed() {
- return closed;
+ public Map getPeerAddresses() {
+ return peerAddresses;
}
- RegisterChangeListener getRegisterChangeListener() {
- return registerChangeListener;
+ public DatastoreContext getDatastoreContext() {
+ return datastoreContext;
}
- @Override
- public AsyncDataChangeListener> getInstance() {
- return delegate != null ? delegate.getInstance() : null;
+ public SchemaContext getSchemaContext() {
+ return Verify.verifyNotNull(schemaContextProvider.getSchemaContext());
}
- @Override
- public void close() {
- closed = true;
- if(delegate != null) {
- delegate.close();
+ public DatastoreSnapshot.ShardSnapshot getRestoreFromSnapshot() {
+ return restoreFromSnapshot;
+ }
+
+ public DataTree getDataTree() {
+ return dataTree;
+ }
+
+ public TreeType getTreeType() {
+ switch (datastoreContext.getLogicalStoreType()) {
+ case CONFIGURATION:
+ return TreeType.CONFIGURATION;
+ case OPERATIONAL:
+ return TreeType.OPERATIONAL;
+ default:
+ throw new IllegalStateException("Unhandled logical store type "
+ + datastoreContext.getLogicalStoreType());
}
}
+
+ protected void verify() {
+ requireNonNull(id, "id should not be null");
+ requireNonNull(peerAddresses, "peerAddresses should not be null");
+ requireNonNull(datastoreContext, "dataStoreContext should not be null");
+ requireNonNull(schemaContextProvider, "schemaContextProvider should not be null");
+ }
+
+ public Props props() {
+ sealed = true;
+ verify();
+ return Props.create(shardClass, this);
+ }
+ }
+
+ public static class Builder extends AbstractBuilder {
+ Builder() {
+ this(Shard.class);
+ }
+
+ Builder(final Class extends Shard> shardClass) {
+ super(shardClass);
+ }
+ }
+
+ Ticker ticker() {
+ return Ticker.systemTicker();
+ }
+
+ void scheduleNextPendingTransaction() {
+ self().tell(RESUME_NEXT_PENDING_TRANSACTION, ActorRef.noSender());
}
}