opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/Shard.java

   1 /*
   2  * Copyright (c) 2014 Cisco Systems, Inc. and others.  All rights reserved.
   3  *
   4  * This program and the accompanying materials are made available under the
   5  * terms of the Eclipse Public License v1.0 which accompanies this distribution,
   6  * and is available at http://www.eclipse.org/legal/epl-v10.html
   7  */
   8
   9 package org.opendaylight.controller.cluster.datastore;
  10
  11 import akka.actor.ActorRef;
  12 import akka.actor.ActorSelection;
  13 import akka.actor.Cancellable;
  14 import akka.actor.Props;
  15 import akka.persistence.RecoveryFailure;
  16 import akka.serialization.Serialization;
  17 import com.google.common.annotations.VisibleForTesting;
  18 import com.google.common.base.Optional;
  19 import com.google.common.base.Preconditions;
  20 import java.io.IOException;
  21 import java.util.Collections;
  22 import java.util.Map;
  23 import java.util.concurrent.TimeUnit;
  24 import javax.annotation.Nonnull;
  25 import org.opendaylight.controller.cluster.common.actor.CommonConfig;
  26 import org.opendaylight.controller.cluster.common.actor.MeteringBehavior;
  27 import org.opendaylight.controller.cluster.datastore.ShardCommitCoordinator.CohortEntry;
  28 import org.opendaylight.controller.cluster.datastore.exceptions.NoShardLeaderException;
  29 import org.opendaylight.controller.cluster.datastore.identifiers.ShardIdentifier;
  30 import org.opendaylight.controller.cluster.datastore.identifiers.ShardTransactionIdentifier;
  31 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardMBeanFactory;
  32 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardStats;
  33 import org.opendaylight.controller.cluster.datastore.messages.AbortTransaction;
  34 import org.opendaylight.controller.cluster.datastore.messages.ActorInitialized;
  35 import org.opendaylight.controller.cluster.datastore.messages.BatchedModifications;
  36 import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransaction;
  37 import org.opendaylight.controller.cluster.datastore.messages.CloseTransactionChain;
  38 import org.opendaylight.controller.cluster.datastore.messages.CommitTransaction;
  39 import org.opendaylight.controller.cluster.datastore.messages.CommitTransactionReply;
  40 import org.opendaylight.controller.cluster.datastore.messages.CreateTransaction;
  41 import org.opendaylight.controller.cluster.datastore.messages.CreateTransactionReply;
  42 import org.opendaylight.controller.cluster.datastore.messages.DatastoreSnapshot;
  43 import org.opendaylight.controller.cluster.datastore.messages.DatastoreSnapshot.ShardSnapshot;
  44 import org.opendaylight.controller.cluster.datastore.messages.ForwardedReadyTransaction;
  45 import org.opendaylight.controller.cluster.datastore.messages.GetShardDataTree;
  46 import org.opendaylight.controller.cluster.datastore.messages.PeerAddressResolved;
  47 import org.opendaylight.controller.cluster.datastore.messages.ReadyLocalTransaction;
  48 import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListener;
  49 import org.opendaylight.controller.cluster.datastore.messages.RegisterDataTreeChangeListener;
  50 import org.opendaylight.controller.cluster.datastore.messages.ShardLeaderStateChanged;
  51 import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext;
  52 import org.opendaylight.controller.cluster.datastore.modification.Modification;
  53 import org.opendaylight.controller.cluster.datastore.modification.ModificationPayload;
  54 import org.opendaylight.controller.cluster.datastore.modification.MutableCompositeModification;
  55 import org.opendaylight.controller.cluster.datastore.utils.Dispatchers;
  56 import org.opendaylight.controller.cluster.datastore.utils.MessageTracker;
  57 import org.opendaylight.controller.cluster.notifications.LeaderStateChanged;
  58 import org.opendaylight.controller.cluster.notifications.RegisterRoleChangeListener;
  59 import org.opendaylight.controller.cluster.notifications.RoleChangeNotifier;
  60 import org.opendaylight.controller.cluster.raft.RaftActor;
  61 import org.opendaylight.controller.cluster.raft.RaftActorRecoveryCohort;
  62 import org.opendaylight.controller.cluster.raft.RaftActorSnapshotCohort;
  63 import org.opendaylight.controller.cluster.raft.RaftState;
  64 import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus;
  65 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  66 import org.opendaylight.controller.cluster.raft.messages.ServerRemoved;
  67 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationByteStringPayload;
  68 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationPayload;
  69 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTree;
  70 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidate;
  71 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException;
  72 import org.opendaylight.yangtools.yang.data.api.schema.tree.ModificationType;
  73 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
  74 import scala.concurrent.duration.Duration;
  75 import scala.concurrent.duration.FiniteDuration;
  76
  77 /**
  78  * A Shard represents a portion of the logical data tree <br/>
  79  * <p>
  80  * Our Shard uses InMemoryDataTree as it's internal representation and delegates all requests it
  81  * </p>
  82  */
  83 public class Shard extends RaftActor {
  84
  85     protected static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = "txCommitTimeoutCheck";
  86
  87     @VisibleForTesting
  88     static final Object GET_SHARD_MBEAN_MESSAGE = "getShardMBeanMessage";
  89
  90     @VisibleForTesting
  91     static final String DEFAULT_NAME = "default";
  92
  93     // The state of this Shard
  94     private final ShardDataTree store;
  95
  96     /// The name of this shard
  97     private final String name;
  98
  99     private final ShardStats shardMBean;
 100
 101     private DatastoreContext datastoreContext;
 102
 103     private final ShardCommitCoordinator commitCoordinator;
 104
 105     private long transactionCommitTimeout;
 106
 107     private Cancellable txCommitTimeoutCheckSchedule;
 108
 109     private final Optional<ActorRef> roleChangeNotifier;
 110
 111     private final MessageTracker appendEntriesReplyTracker;
 112
 113     private final ShardTransactionActorFactory transactionActorFactory;
 114
 115     private final ShardSnapshotCohort snapshotCohort;
 116
 117     private final DataTreeChangeListenerSupport treeChangeSupport = new DataTreeChangeListenerSupport(this);
 118     private final DataChangeListenerSupport changeSupport = new DataChangeListenerSupport(this);
 119
 120
 121     private ShardSnapshot restoreFromSnapshot;
 122
 123
 124
 125     protected Shard(AbstractBuilder<?, ?> builder) {
 126         super(builder.getId().toString(), builder.getPeerAddresses(),
 127                 Optional.of(builder.getDatastoreContext().getShardRaftConfig()), DataStoreVersions.CURRENT_VERSION);
 128
 129         this.name = builder.getId().toString();
 130         this.datastoreContext = builder.getDatastoreContext();
 131         this.restoreFromSnapshot = builder.getRestoreFromSnapshot();
 132
 133         setPersistence(datastoreContext.isPersistent());
 134
 135         LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
 136
 137         // FIXME: BUG-1014: pass down the proper TreeType
 138         store = new ShardDataTree(builder.getSchemaContext());
 139
 140         shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
 141                 datastoreContext.getDataStoreMXBeanType());
 142         shardMBean.setShard(this);
 143
 144         if (isMetricsCaptureEnabled()) {
 145             getContext().become(new MeteringBehavior(this));
 146         }
 147
 148         commitCoordinator = new ShardCommitCoordinator(store,
 149                 datastoreContext.getShardCommitQueueExpiryTimeoutInMillis(),
 150                 datastoreContext.getShardTransactionCommitQueueCapacity(), LOG, this.name);
 151
 152         setTransactionCommitTimeout();
 153
 154         // create a notifier actor for each cluster member
 155         roleChangeNotifier = createRoleChangeNotifier(name.toString());
 156
 157         appendEntriesReplyTracker = new MessageTracker(AppendEntriesReply.class,
 158                 getRaftActorContext().getConfigParams().getIsolatedCheckIntervalInMillis());
 159
 160         transactionActorFactory = new ShardTransactionActorFactory(store, datastoreContext,
 161                 new Dispatchers(context().system().dispatchers()).getDispatcherPath(
 162                         Dispatchers.DispatcherType.Transaction), self(), getContext(), shardMBean);
 163
 164         snapshotCohort = new ShardSnapshotCohort(transactionActorFactory, store, LOG, this.name);
 165
 166
 167
 168     }
 169
 170     private void setTransactionCommitTimeout() {
 171         transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
 172                 datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS) / 2;
 173     }
 174
 175     private Optional<ActorRef> createRoleChangeNotifier(String shardId) {
 176         ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
 177             RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
 178         return Optional.of(shardRoleChangeNotifier);
 179     }
 180
 181     @Override
 182     public void postStop() {
 183         LOG.info("Stopping Shard {}", persistenceId());
 184
 185         super.postStop();
 186
 187         if(txCommitTimeoutCheckSchedule != null) {
 188             txCommitTimeoutCheckSchedule.cancel();
 189         }
 190
 191         shardMBean.unregisterMBean();
 192     }
 193
 194     @Override
 195     public void onReceiveRecover(final Object message) throws Exception {
 196         if(LOG.isDebugEnabled()) {
 197             LOG.debug("{}: onReceiveRecover: Received message {} from {}", persistenceId(),
 198                 message.getClass().toString(), getSender());
 199         }
 200
 201         if (message instanceof RecoveryFailure){
 202             LOG.error("{}: Recovery failed because of this cause",
 203                     persistenceId(), ((RecoveryFailure) message).cause());
 204
 205             // Even though recovery failed, we still need to finish our recovery, eg send the
 206             // ActorInitialized message and start the txCommitTimeoutCheckSchedule.
 207             onRecoveryComplete();
 208         } else {
 209             super.onReceiveRecover(message);
 210             if(LOG.isTraceEnabled()) {
 211                 appendEntriesReplyTracker.begin();
 212             }
 213         }
 214     }
 215
 216     @Override
 217     public void onReceiveCommand(final Object message) throws Exception {
 218
 219         MessageTracker.Context context = appendEntriesReplyTracker.received(message);
 220
 221         if(context.error().isPresent()){
 222             LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(),
 223                 context.error());
 224         }
 225
 226         try {
 227             if (CreateTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 228                 handleCreateTransaction(message);
 229             } else if (BatchedModifications.class.isInstance(message)) {
 230                 handleBatchedModifications((BatchedModifications)message);
 231             } else if (message instanceof ForwardedReadyTransaction) {
 232                 commitCoordinator.handleForwardedReadyTransaction((ForwardedReadyTransaction) message,
 233                         getSender(), this);
 234             } else if (message instanceof ReadyLocalTransaction) {
 235                 handleReadyLocalTransaction((ReadyLocalTransaction)message);
 236             } else if (CanCommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 237                 handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
 238             } else if (CommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 239                 handleCommitTransaction(CommitTransaction.fromSerializable(message));
 240             } else if (AbortTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 241                 handleAbortTransaction(AbortTransaction.fromSerializable(message));
 242             } else if (CloseTransactionChain.SERIALIZABLE_CLASS.isInstance(message)) {
 243                 closeTransactionChain(CloseTransactionChain.fromSerializable(message));
 244             } else if (message instanceof RegisterChangeListener) {
 245                 changeSupport.onMessage((RegisterChangeListener) message, isLeader(), hasLeader());
 246             } else if (message instanceof RegisterDataTreeChangeListener) {
 247                 treeChangeSupport.onMessage((RegisterDataTreeChangeListener) message, isLeader(), hasLeader());
 248             } else if (message instanceof UpdateSchemaContext) {
 249                 updateSchemaContext((UpdateSchemaContext) message);
 250             } else if (message instanceof PeerAddressResolved) {
 251                 PeerAddressResolved resolved = (PeerAddressResolved) message;
 252                 setPeerAddress(resolved.getPeerId().toString(),
 253                         resolved.getPeerAddress());
 254             } else if (message.equals(TX_COMMIT_TIMEOUT_CHECK_MESSAGE)) {
 255                 handleTransactionCommitTimeoutCheck();
 256             } else if(message instanceof DatastoreContext) {
 257                 onDatastoreContext((DatastoreContext)message);
 258             } else if(message instanceof RegisterRoleChangeListener){
 259                 roleChangeNotifier.get().forward(message, context());
 260             } else if (message instanceof FollowerInitialSyncUpStatus) {
 261                 shardMBean.setFollowerInitialSyncStatus(((FollowerInitialSyncUpStatus) message).isInitialSyncDone());
 262                 context().parent().tell(message, self());
 263             } else if(GET_SHARD_MBEAN_MESSAGE.equals(message)){
 264                 sender().tell(getShardMBean(), self());
 265             } else if(message instanceof GetShardDataTree) {
 266                 sender().tell(store.getDataTree(), self());
 267             } else if(message instanceof ServerRemoved){
 268                 context().parent().forward(message, context());
 269             } else {
 270                 super.onReceiveCommand(message);
 271             }
 272         } finally {
 273             context.done();
 274         }
 275     }
 276
 277     private boolean hasLeader() {
 278         return getLeaderId() != null;
 279     }
 280
 281     public int getPendingTxCommitQueueSize() {
 282         return commitCoordinator.getQueueSize();
 283     }
 284
 285     @Override
 286     protected Optional<ActorRef> getRoleChangeNotifier() {
 287         return roleChangeNotifier;
 288     }
 289
 290     @Override
 291     protected LeaderStateChanged newLeaderStateChanged(String memberId, String leaderId, short leaderPayloadVersion) {
 292         return new ShardLeaderStateChanged(memberId, leaderId,
 293                 isLeader() ? Optional.<DataTree>of(store.getDataTree()) : Optional.<DataTree>absent(),
 294                 leaderPayloadVersion);
 295     }
 296
 297     protected void onDatastoreContext(DatastoreContext context) {
 298         datastoreContext = context;
 299
 300         commitCoordinator.setQueueCapacity(datastoreContext.getShardTransactionCommitQueueCapacity());
 301
 302         setTransactionCommitTimeout();
 303
 304         if(datastoreContext.isPersistent() && !persistence().isRecoveryApplicable()) {
 305             setPersistence(true);
 306         } else if(!datastoreContext.isPersistent() && persistence().isRecoveryApplicable()) {
 307             setPersistence(false);
 308         }
 309
 310         updateConfigParams(datastoreContext.getShardRaftConfig());
 311     }
 312
 313     private void handleTransactionCommitTimeoutCheck() {
 314         CohortEntry cohortEntry = commitCoordinator.getCurrentCohortEntry();
 315         if(cohortEntry != null) {
 316             if(cohortEntry.isExpired(transactionCommitTimeout)) {
 317                 LOG.warn("{}: Current transaction {} has timed out after {} ms - aborting",
 318                         persistenceId(), cohortEntry.getTransactionID(), transactionCommitTimeout);
 319
 320                 doAbortTransaction(cohortEntry.getTransactionID(), null);
 321             }
 322         }
 323
 324         commitCoordinator.cleanupExpiredCohortEntries();
 325     }
 326
 327     private static boolean isEmptyCommit(final DataTreeCandidate candidate) {
 328         return ModificationType.UNMODIFIED.equals(candidate.getRootNode().getModificationType());
 329     }
 330
 331     void continueCommit(final CohortEntry cohortEntry) {
 332         final DataTreeCandidate candidate = cohortEntry.getCandidate();
 333
 334         // If we do not have any followers and we are not using persistence
 335         // or if cohortEntry has no modifications
 336         // we can apply modification to the state immediately
 337         if ((!hasFollowers() && !persistence().isRecoveryApplicable()) || isEmptyCommit(candidate)) {
 338             applyModificationToState(cohortEntry.getReplySender(), cohortEntry.getTransactionID(), candidate);
 339         } else {
 340             Shard.this.persistData(cohortEntry.getReplySender(), cohortEntry.getTransactionID(),
 341                     DataTreeCandidatePayload.create(candidate));
 342         }
 343     }
 344
 345     private void handleCommitTransaction(final CommitTransaction commit) {
 346         if(!commitCoordinator.handleCommit(commit.getTransactionID(), getSender(), this)) {
 347             shardMBean.incrementFailedTransactionsCount();
 348         }
 349     }
 350
 351     private void finishCommit(@Nonnull final ActorRef sender, @Nonnull final String transactionID, @Nonnull final CohortEntry cohortEntry) {
 352         LOG.debug("{}: Finishing commit for transaction {}", persistenceId(), cohortEntry.getTransactionID());
 353
 354         try {
 355             cohortEntry.commit();
 356
 357             sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
 358
 359             shardMBean.incrementCommittedTransactionCount();
 360             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 361
 362         } catch (Exception e) {
 363             sender.tell(new akka.actor.Status.Failure(e), getSelf());
 364
 365             LOG.error("{}, An exception occurred while committing transaction {}", persistenceId(),
 366                     transactionID, e);
 367             shardMBean.incrementFailedTransactionsCount();
 368         } finally {
 369             commitCoordinator.currentTransactionComplete(transactionID, true);
 370         }
 371     }
 372
 373     private void finishCommit(@Nonnull final ActorRef sender, final @Nonnull String transactionID) {
 374         // With persistence enabled, this method is called via applyState by the leader strategy
 375         // after the commit has been replicated to a majority of the followers.
 376
 377         CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
 378         if (cohortEntry == null) {
 379             // The transaction is no longer the current commit. This can happen if the transaction
 380             // was aborted prior, most likely due to timeout in the front-end. We need to finish
 381             // committing the transaction though since it was successfully persisted and replicated
 382             // however we can't use the original cohort b/c it was already preCommitted and may
 383             // conflict with the current commit or may have been aborted so we commit with a new
 384             // transaction.
 385             cohortEntry = commitCoordinator.getAndRemoveCohortEntry(transactionID);
 386             if(cohortEntry != null) {
 387                 try {
 388                     store.applyForeignCandidate(transactionID, cohortEntry.getCandidate());
 389                 } catch (DataValidationFailedException e) {
 390                     shardMBean.incrementFailedTransactionsCount();
 391                     LOG.error("{}: Failed to re-apply transaction {}", persistenceId(), transactionID, e);
 392                 }
 393
 394                 sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
 395             } else {
 396                 // This really shouldn't happen - it likely means that persistence or replication
 397                 // took so long to complete such that the cohort entry was expired from the cache.
 398                 IllegalStateException ex = new IllegalStateException(
 399                         String.format("%s: Could not finish committing transaction %s - no CohortEntry found",
 400                                 persistenceId(), transactionID));
 401                 LOG.error(ex.getMessage());
 402                 sender.tell(new akka.actor.Status.Failure(ex), getSelf());
 403             }
 404         } else {
 405             finishCommit(sender, transactionID, cohortEntry);
 406         }
 407     }
 408
 409     private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
 410         LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionID());
 411         commitCoordinator.handleCanCommit(canCommit.getTransactionID(), getSender(), this);
 412     }
 413
 414     private void noLeaderError(String errMessage, Object message) {
 415         // TODO: rather than throwing an immediate exception, we could schedule a timer to try again to make
 416         // it more resilient in case we're in the process of electing a new leader.
 417         getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(errMessage, persistenceId())), getSelf());
 418     }
 419
 420     protected void handleBatchedModificationsLocal(BatchedModifications batched, ActorRef sender) {
 421         try {
 422             commitCoordinator.handleBatchedModifications(batched, sender, this);
 423         } catch (Exception e) {
 424             LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(),
 425                     batched.getTransactionID(), e);
 426             sender.tell(new akka.actor.Status.Failure(e), getSelf());
 427         }
 428     }
 429
 430     private void handleBatchedModifications(BatchedModifications batched) {
 431         // This message is sent to prepare the modifications transaction directly on the Shard as an
 432         // optimization to avoid the extra overhead of a separate ShardTransaction actor. On the last
 433         // BatchedModifications message, the caller sets the ready flag in the message indicating
 434         // modifications are complete. The reply contains the cohort actor path (this actor) for the caller
 435         // to initiate the 3-phase commit. This also avoids the overhead of sending an additional
 436         // ReadyTransaction message.
 437
 438         // If we're not the leader then forward to the leader. This is a safety measure - we shouldn't
 439         // normally get here if we're not the leader as the front-end (TransactionProxy) should determine
 440         // the primary/leader shard. However with timing and caching on the front-end, there's a small
 441         // window where it could have a stale leader during leadership transitions.
 442         //
 443         if(isLeader()) {
 444             failIfIsolatedLeader(getSender());
 445
 446             handleBatchedModificationsLocal(batched, getSender());
 447         } else {
 448             ActorSelection leader = getLeader();
 449             if(leader != null) {
 450                 // TODO: what if this is not the first batch and leadership changed in between batched messages?
 451                 // We could check if the commitCoordinator already has a cached entry and forward all the previous
 452                 // batched modifications.
 453                 LOG.debug("{}: Forwarding BatchedModifications to leader {}", persistenceId(), leader);
 454                 leader.forward(batched, getContext());
 455             } else {
 456                 noLeaderError("Could not commit transaction " + batched.getTransactionID(), batched);
 457             }
 458         }
 459     }
 460
 461     private boolean failIfIsolatedLeader(ActorRef sender) {
 462         if(isIsolatedLeader()) {
 463             sender.tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format(
 464                     "Shard %s was the leader but has lost contact with all of its followers. Either all" +
 465                     " other follower nodes are down or this node is isolated by a network partition.",
 466                     persistenceId()))), getSelf());
 467             return true;
 468         }
 469
 470         return false;
 471     }
 472
 473     protected boolean isIsolatedLeader() {
 474         return getRaftState() == RaftState.IsolatedLeader;
 475     }
 476
 477     private void handleReadyLocalTransaction(final ReadyLocalTransaction message) {
 478         if (isLeader()) {
 479             failIfIsolatedLeader(getSender());
 480
 481             try {
 482                 commitCoordinator.handleReadyLocalTransaction(message, getSender(), this);
 483             } catch (Exception e) {
 484                 LOG.error("{}: Error handling ReadyLocalTransaction for Tx {}", persistenceId(),
 485                         message.getTransactionID(), e);
 486                 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 487             }
 488         } else {
 489             ActorSelection leader = getLeader();
 490             if (leader != null) {
 491                 LOG.debug("{}: Forwarding ReadyLocalTransaction to leader {}", persistenceId(), leader);
 492                 message.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
 493                 leader.forward(message, getContext());
 494             } else {
 495                 noLeaderError("Could not commit transaction " + message.getTransactionID(), message);
 496             }
 497         }
 498     }
 499
 500     private void handleAbortTransaction(final AbortTransaction abort) {
 501         doAbortTransaction(abort.getTransactionID(), getSender());
 502     }
 503
 504     void doAbortTransaction(final String transactionID, final ActorRef sender) {
 505         commitCoordinator.handleAbort(transactionID, sender, this);
 506     }
 507
 508     private void handleCreateTransaction(final Object message) {
 509         if (isLeader()) {
 510             createTransaction(CreateTransaction.fromSerializable(message));
 511         } else if (getLeader() != null) {
 512             getLeader().forward(message, getContext());
 513         } else {
 514             getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(
 515                     "Could not create a shard transaction", persistenceId())), getSelf());
 516         }
 517     }
 518
 519     private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
 520         store.closeTransactionChain(closeTransactionChain.getTransactionChainId());
 521     }
 522
 523     private ActorRef createTypedTransactionActor(int transactionType,
 524             ShardTransactionIdentifier transactionId, String transactionChainId,
 525             short clientVersion ) {
 526
 527         return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType),
 528                 transactionId, transactionChainId, clientVersion);
 529     }
 530
 531     private void createTransaction(CreateTransaction createTransaction) {
 532         try {
 533             if(TransactionType.fromInt(createTransaction.getTransactionType()) != TransactionType.READ_ONLY &&
 534                     failIfIsolatedLeader(getSender())) {
 535                 return;
 536             }
 537
 538             ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
 539                 createTransaction.getTransactionId(), createTransaction.getTransactionChainId(),
 540                 createTransaction.getVersion());
 541
 542             getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
 543                     createTransaction.getTransactionId()).toSerializable(), getSelf());
 544         } catch (Exception e) {
 545             getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 546         }
 547     }
 548
 549     private ActorRef createTransaction(int transactionType, String remoteTransactionId,
 550             String transactionChainId, short clientVersion) {
 551
 552
 553         ShardTransactionIdentifier transactionId = new ShardTransactionIdentifier(remoteTransactionId);
 554
 555         if(LOG.isDebugEnabled()) {
 556             LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
 557         }
 558
 559         ActorRef transactionActor = createTypedTransactionActor(transactionType, transactionId,
 560                 transactionChainId, clientVersion);
 561
 562         return transactionActor;
 563     }
 564
 565     private void commitWithNewTransaction(final Modification modification) {
 566         ReadWriteShardDataTreeTransaction tx = store.newReadWriteTransaction(modification.toString(), null);
 567         modification.apply(tx.getSnapshot());
 568         try {
 569             snapshotCohort.syncCommitTransaction(tx);
 570             shardMBean.incrementCommittedTransactionCount();
 571             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 572         } catch (Exception e) {
 573             shardMBean.incrementFailedTransactionsCount();
 574             LOG.error("{}: Failed to commit", persistenceId(), e);
 575         }
 576     }
 577
 578     private void updateSchemaContext(final UpdateSchemaContext message) {
 579         updateSchemaContext(message.getSchemaContext());
 580     }
 581
 582     @VisibleForTesting
 583     void updateSchemaContext(final SchemaContext schemaContext) {
 584         store.updateSchemaContext(schemaContext);
 585     }
 586
 587     private boolean isMetricsCaptureEnabled() {
 588         CommonConfig config = new CommonConfig(getContext().system().settings().config());
 589         return config.isMetricCaptureEnabled();
 590     }
 591
 592     @Override
 593     @VisibleForTesting
 594     public RaftActorSnapshotCohort getRaftActorSnapshotCohort() {
 595         return snapshotCohort;
 596     }
 597
 598     @Override
 599     @Nonnull
 600     protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
 601         return new ShardRecoveryCoordinator(store, store.getSchemaContext(),
 602                 restoreFromSnapshot != null ? restoreFromSnapshot.getSnapshot() : null, persistenceId(), LOG);
 603     }
 604
 605     @Override
 606     protected void onRecoveryComplete() {
 607         restoreFromSnapshot = null;
 608
 609         //notify shard manager
 610         getContext().parent().tell(new ActorInitialized(), getSelf());
 611
 612         // Being paranoid here - this method should only be called once but just in case...
 613         if(txCommitTimeoutCheckSchedule == null) {
 614             // Schedule a message to be periodically sent to check if the current in-progress
 615             // transaction should be expired and aborted.
 616             FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
 617             txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
 618                     period, period, getSelf(),
 619                     TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
 620         }
 621     }
 622
 623     @Override
 624     protected void applyState(final ActorRef clientActor, final String identifier, final Object data) {
 625         if (data instanceof DataTreeCandidatePayload) {
 626             if (clientActor == null) {
 627                 // No clientActor indicates a replica coming from the leader
 628                 try {
 629                     store.applyForeignCandidate(identifier, ((DataTreeCandidatePayload)data).getCandidate());
 630                 } catch (DataValidationFailedException | IOException e) {
 631                     LOG.error("{}: Error applying replica {}", persistenceId(), identifier, e);
 632                 }
 633             } else {
 634                 // Replication consensus reached, proceed to commit
 635                 finishCommit(clientActor, identifier);
 636             }
 637         } else if (data instanceof ModificationPayload) {
 638             try {
 639                 applyModificationToState(clientActor, identifier, ((ModificationPayload) data).getModification());
 640             } catch (ClassNotFoundException | IOException e) {
 641                 LOG.error("{}: Error extracting ModificationPayload", persistenceId(), e);
 642             }
 643         } else if (data instanceof CompositeModificationPayload) {
 644             Object modification = ((CompositeModificationPayload) data).getModification();
 645
 646             applyModificationToState(clientActor, identifier, modification);
 647         } else if(data instanceof CompositeModificationByteStringPayload ){
 648             Object modification = ((CompositeModificationByteStringPayload) data).getModification();
 649
 650             applyModificationToState(clientActor, identifier, modification);
 651         } else {
 652             LOG.error("{}: Unknown state received {} Class loader = {} CompositeNodeMod.ClassLoader = {}",
 653                     persistenceId(), data, data.getClass().getClassLoader(),
 654                     CompositeModificationPayload.class.getClassLoader());
 655         }
 656     }
 657
 658     private void applyModificationToState(ActorRef clientActor, String identifier, Object modification) {
 659         if(modification == null) {
 660             LOG.error(
 661                     "{}: modification is null - this is very unexpected, clientActor = {}, identifier = {}",
 662                     persistenceId(), identifier, clientActor != null ? clientActor.path().toString() : null);
 663         } else if(clientActor == null) {
 664             // There's no clientActor to which to send a commit reply so we must be applying
 665             // replicated state from the leader.
 666             commitWithNewTransaction(MutableCompositeModification.fromSerializable(modification));
 667         } else {
 668             // This must be the OK to commit after replication consensus.
 669             finishCommit(clientActor, identifier);
 670         }
 671     }
 672
 673     @Override
 674     protected void onStateChanged() {
 675         boolean isLeader = isLeader();
 676         boolean hasLeader = hasLeader();
 677         changeSupport.onLeadershipChange(isLeader, hasLeader);
 678         treeChangeSupport.onLeadershipChange(isLeader, hasLeader);
 679
 680         // If this actor is no longer the leader close all the transaction chains
 681         if (!isLeader) {
 682             if(LOG.isDebugEnabled()) {
 683                 LOG.debug(
 684                     "{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader",
 685                     persistenceId(), getId());
 686             }
 687
 688             store.closeAllTransactionChains();
 689         }
 690     }
 691
 692     @Override
 693     protected void onLeaderChanged(String oldLeader, String newLeader) {
 694         shardMBean.incrementLeadershipChangeCount();
 695     }
 696
 697     @Override
 698     public String persistenceId() {
 699         return this.name;
 700     }
 701
 702     @VisibleForTesting
 703     ShardCommitCoordinator getCommitCoordinator() {
 704         return commitCoordinator;
 705     }
 706
 707     public DatastoreContext getDatastoreContext() {
 708         return datastoreContext;
 709     }
 710
 711     @VisibleForTesting
 712     public ShardDataTree getDataStore() {
 713         return store;
 714     }
 715
 716     @VisibleForTesting
 717     ShardStats getShardMBean() {
 718         return shardMBean;
 719     }
 720
 721     public static Builder builder() {
 722         return new Builder();
 723     }
 724
 725     public static abstract class AbstractBuilder<T extends AbstractBuilder<T, S>, S extends Shard> {
 726         private final Class<S> shardClass;
 727         private ShardIdentifier id;
 728         private Map<String, String> peerAddresses = Collections.emptyMap();
 729         private DatastoreContext datastoreContext;
 730         private SchemaContext schemaContext;
 731         private DatastoreSnapshot.ShardSnapshot restoreFromSnapshot;
 732         private volatile boolean sealed;
 733
 734         protected AbstractBuilder(Class<S> shardClass) {
 735             this.shardClass = shardClass;
 736         }
 737
 738         protected void checkSealed() {
 739             Preconditions.checkState(!sealed, "Builder isalready sealed - further modifications are not allowed");
 740         }
 741
 742         @SuppressWarnings("unchecked")
 743         private T self() {
 744             return (T) this;
 745         }
 746
 747         public T id(ShardIdentifier id) {
 748             checkSealed();
 749             this.id = id;
 750             return self();
 751         }
 752
 753         public T peerAddresses(Map<String, String> peerAddresses) {
 754             checkSealed();
 755             this.peerAddresses = peerAddresses;
 756             return self();
 757         }
 758
 759         public T datastoreContext(DatastoreContext datastoreContext) {
 760             checkSealed();
 761             this.datastoreContext = datastoreContext;
 762             return self();
 763         }
 764
 765         public T schemaContext(SchemaContext schemaContext) {
 766             checkSealed();
 767             this.schemaContext = schemaContext;
 768             return self();
 769         }
 770
 771         public T restoreFromSnapshot(DatastoreSnapshot.ShardSnapshot restoreFromSnapshot) {
 772             checkSealed();
 773             this.restoreFromSnapshot = restoreFromSnapshot;
 774             return self();
 775         }
 776
 777         public ShardIdentifier getId() {
 778             return id;
 779         }
 780
 781         public Map<String, String> getPeerAddresses() {
 782             return peerAddresses;
 783         }
 784
 785         public DatastoreContext getDatastoreContext() {
 786             return datastoreContext;
 787         }
 788
 789         public SchemaContext getSchemaContext() {
 790             return schemaContext;
 791         }
 792
 793         public DatastoreSnapshot.ShardSnapshot getRestoreFromSnapshot() {
 794             return restoreFromSnapshot;
 795         }
 796
 797         protected void verify() {
 798             Preconditions.checkNotNull(id, "id should not be null");
 799             Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
 800             Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
 801             Preconditions.checkNotNull(schemaContext, "schemaContext should not be null");
 802         }
 803
 804         public Props props() {
 805             sealed = true;
 806             verify();
 807             return Props.create(shardClass, this);
 808         }
 809     }
 810
 811     public static class Builder extends AbstractBuilder<Builder, Shard> {
 812         private Builder() {
 813             super(Shard.class);
 814         }
 815     }
 816 }