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.japi.Creator;
  16 import akka.persistence.RecoveryFailure;
  17 import akka.serialization.Serialization;
  18 import com.google.common.annotations.VisibleForTesting;
  19 import com.google.common.base.Optional;
  20 import com.google.common.base.Preconditions;
  21 import com.google.common.util.concurrent.FutureCallback;
  22 import com.google.common.util.concurrent.Futures;
  23 import com.google.common.util.concurrent.ListenableFuture;
  24 import java.io.IOException;
  25 import java.util.HashMap;
  26 import java.util.Map;
  27 import java.util.concurrent.TimeUnit;
  28 import javax.annotation.Nonnull;
  29 import org.opendaylight.controller.cluster.common.actor.CommonConfig;
  30 import org.opendaylight.controller.cluster.common.actor.MeteringBehavior;
  31 import org.opendaylight.controller.cluster.datastore.ShardCommitCoordinator.CohortEntry;
  32 import org.opendaylight.controller.cluster.datastore.exceptions.NoShardLeaderException;
  33 import org.opendaylight.controller.cluster.datastore.identifiers.ShardIdentifier;
  34 import org.opendaylight.controller.cluster.datastore.identifiers.ShardTransactionIdentifier;
  35 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardMBeanFactory;
  36 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardStats;
  37 import org.opendaylight.controller.cluster.datastore.messages.AbortTransaction;
  38 import org.opendaylight.controller.cluster.datastore.messages.AbortTransactionReply;
  39 import org.opendaylight.controller.cluster.datastore.messages.ActorInitialized;
  40 import org.opendaylight.controller.cluster.datastore.messages.BatchedModifications;
  41 import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransaction;
  42 import org.opendaylight.controller.cluster.datastore.messages.CloseTransactionChain;
  43 import org.opendaylight.controller.cluster.datastore.messages.CommitTransaction;
  44 import org.opendaylight.controller.cluster.datastore.messages.CommitTransactionReply;
  45 import org.opendaylight.controller.cluster.datastore.messages.CreateTransaction;
  46 import org.opendaylight.controller.cluster.datastore.messages.CreateTransactionReply;
  47 import org.opendaylight.controller.cluster.datastore.messages.ForwardedReadyTransaction;
  48 import org.opendaylight.controller.cluster.datastore.messages.PeerAddressResolved;
  49 import org.opendaylight.controller.cluster.datastore.messages.ReadyLocalTransaction;
  50 import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListener;
  51 import org.opendaylight.controller.cluster.datastore.messages.RegisterDataTreeChangeListener;
  52 import org.opendaylight.controller.cluster.datastore.messages.ShardLeaderStateChanged;
  53 import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext;
  54 import org.opendaylight.controller.cluster.datastore.modification.Modification;
  55 import org.opendaylight.controller.cluster.datastore.modification.ModificationPayload;
  56 import org.opendaylight.controller.cluster.datastore.modification.MutableCompositeModification;
  57 import org.opendaylight.controller.cluster.datastore.utils.Dispatchers;
  58 import org.opendaylight.controller.cluster.datastore.utils.MessageTracker;
  59 import org.opendaylight.controller.cluster.notifications.LeaderStateChanged;
  60 import org.opendaylight.controller.cluster.notifications.RegisterRoleChangeListener;
  61 import org.opendaylight.controller.cluster.notifications.RoleChangeNotifier;
  62 import org.opendaylight.controller.cluster.raft.RaftActor;
  63 import org.opendaylight.controller.cluster.raft.RaftActorRecoveryCohort;
  64 import org.opendaylight.controller.cluster.raft.RaftActorSnapshotCohort;
  65 import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus;
  66 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  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     private 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     protected Shard(final ShardIdentifier name, final Map<String, String> peerAddresses,
 121             final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 122         super(name.toString(), new HashMap<>(peerAddresses), Optional.of(datastoreContext.getShardRaftConfig()),
 123                 DataStoreVersions.CURRENT_VERSION);
 124
 125         this.name = name.toString();
 126         this.datastoreContext = datastoreContext;
 127
 128         setPersistence(datastoreContext.isPersistent());
 129
 130         LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
 131
 132         store = new ShardDataTree(schemaContext);
 133
 134         shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
 135                 datastoreContext.getDataStoreMXBeanType());
 136         shardMBean.setShardActor(getSelf());
 137
 138         if (isMetricsCaptureEnabled()) {
 139             getContext().become(new MeteringBehavior(this));
 140         }
 141
 142         commitCoordinator = new ShardCommitCoordinator(store,
 143                 TimeUnit.SECONDS.convert(5, TimeUnit.MINUTES),
 144                 datastoreContext.getShardTransactionCommitQueueCapacity(), self(), LOG, this.name);
 145
 146         setTransactionCommitTimeout();
 147
 148         // create a notifier actor for each cluster member
 149         roleChangeNotifier = createRoleChangeNotifier(name.toString());
 150
 151         appendEntriesReplyTracker = new MessageTracker(AppendEntriesReply.class,
 152                 getRaftActorContext().getConfigParams().getIsolatedCheckIntervalInMillis());
 153
 154         transactionActorFactory = new ShardTransactionActorFactory(store, datastoreContext,
 155                 new Dispatchers(context().system().dispatchers()).getDispatcherPath(
 156                         Dispatchers.DispatcherType.Transaction), self(), getContext(), shardMBean);
 157
 158         snapshotCohort = new ShardSnapshotCohort(transactionActorFactory, store, LOG, this.name);
 159     }
 160
 161     private void setTransactionCommitTimeout() {
 162         transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
 163                 datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS);
 164     }
 165
 166     public static Props props(final ShardIdentifier name,
 167         final Map<String, String> peerAddresses,
 168         final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 169         Preconditions.checkNotNull(name, "name should not be null");
 170         Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
 171         Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
 172         Preconditions.checkNotNull(schemaContext, "schemaContext should not be null");
 173
 174         return Props.create(new ShardCreator(name, peerAddresses, datastoreContext, schemaContext));
 175     }
 176
 177     private Optional<ActorRef> createRoleChangeNotifier(String shardId) {
 178         ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
 179             RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
 180         return Optional.of(shardRoleChangeNotifier);
 181     }
 182
 183     @Override
 184     public void postStop() {
 185         LOG.info("Stopping Shard {}", persistenceId());
 186
 187         super.postStop();
 188
 189         if(txCommitTimeoutCheckSchedule != null) {
 190             txCommitTimeoutCheckSchedule.cancel();
 191         }
 192
 193         shardMBean.unregisterMBean();
 194     }
 195
 196     @Override
 197     public void onReceiveRecover(final Object message) throws Exception {
 198         if(LOG.isDebugEnabled()) {
 199             LOG.debug("{}: onReceiveRecover: Received message {} from {}", persistenceId(),
 200                 message.getClass().toString(), getSender());
 201         }
 202
 203         if (message instanceof RecoveryFailure){
 204             LOG.error("{}: Recovery failed because of this cause",
 205                     persistenceId(), ((RecoveryFailure) message).cause());
 206
 207             // Even though recovery failed, we still need to finish our recovery, eg send the
 208             // ActorInitialized message and start the txCommitTimeoutCheckSchedule.
 209             onRecoveryComplete();
 210         } else {
 211             super.onReceiveRecover(message);
 212             if(LOG.isTraceEnabled()) {
 213                 appendEntriesReplyTracker.begin();
 214             }
 215         }
 216     }
 217
 218     @Override
 219     public void onReceiveCommand(final Object message) throws Exception {
 220
 221         MessageTracker.Context context = appendEntriesReplyTracker.received(message);
 222
 223         if(context.error().isPresent()){
 224             LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(),
 225                     context.error());
 226         }
 227
 228         try {
 229             if (CreateTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 230                 handleCreateTransaction(message);
 231             } else if (BatchedModifications.class.isInstance(message)) {
 232                 handleBatchedModifications((BatchedModifications)message);
 233             } else if (message instanceof ForwardedReadyTransaction) {
 234                 commitCoordinator.handleForwardedReadyTransaction((ForwardedReadyTransaction) message,
 235                         getSender(), this);
 236             } else if (message instanceof ReadyLocalTransaction) {
 237                 handleReadyLocalTransaction((ReadyLocalTransaction)message);
 238             } else if (CanCommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 239                 handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
 240             } else if (CommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 241                 handleCommitTransaction(CommitTransaction.fromSerializable(message));
 242             } else if (AbortTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 243                 handleAbortTransaction(AbortTransaction.fromSerializable(message));
 244             } else if (CloseTransactionChain.SERIALIZABLE_CLASS.isInstance(message)) {
 245                 closeTransactionChain(CloseTransactionChain.fromSerializable(message));
 246             } else if (message instanceof RegisterChangeListener) {
 247                 changeSupport.onMessage((RegisterChangeListener) message, isLeader());
 248             } else if (message instanceof RegisterDataTreeChangeListener) {
 249                 treeChangeSupport.onMessage((RegisterDataTreeChangeListener) message, isLeader());
 250             } else if (message instanceof UpdateSchemaContext) {
 251                 updateSchemaContext((UpdateSchemaContext) message);
 252             } else if (message instanceof PeerAddressResolved) {
 253                 PeerAddressResolved resolved = (PeerAddressResolved) message;
 254                 setPeerAddress(resolved.getPeerId().toString(),
 255                         resolved.getPeerAddress());
 256             } else if (message.equals(TX_COMMIT_TIMEOUT_CHECK_MESSAGE)) {
 257                 handleTransactionCommitTimeoutCheck();
 258             } else if(message instanceof DatastoreContext) {
 259                 onDatastoreContext((DatastoreContext)message);
 260             } else if(message instanceof RegisterRoleChangeListener){
 261                 roleChangeNotifier.get().forward(message, context());
 262             } else if (message instanceof FollowerInitialSyncUpStatus) {
 263                 shardMBean.setFollowerInitialSyncStatus(((FollowerInitialSyncUpStatus) message).isInitialSyncDone());
 264                 context().parent().tell(message, self());
 265             } else if(GET_SHARD_MBEAN_MESSAGE.equals(message)){
 266                 sender().tell(getShardMBean(), self());
 267             } else {
 268                 super.onReceiveCommand(message);
 269             }
 270         } finally {
 271             context.done();
 272         }
 273     }
 274
 275     @Override
 276     protected Optional<ActorRef> getRoleChangeNotifier() {
 277         return roleChangeNotifier;
 278     }
 279
 280     @Override
 281     protected LeaderStateChanged newLeaderStateChanged(String memberId, String leaderId) {
 282         return new ShardLeaderStateChanged(memberId, leaderId,
 283                 isLeader() ? Optional.<DataTree>of(store.getDataTree()) : Optional.<DataTree>absent());
 284     }
 285
 286     private void onDatastoreContext(DatastoreContext context) {
 287         datastoreContext = context;
 288
 289         commitCoordinator.setQueueCapacity(datastoreContext.getShardTransactionCommitQueueCapacity());
 290
 291         setTransactionCommitTimeout();
 292
 293         if(datastoreContext.isPersistent() && !persistence().isRecoveryApplicable()) {
 294             setPersistence(true);
 295         } else if(!datastoreContext.isPersistent() && persistence().isRecoveryApplicable()) {
 296             setPersistence(false);
 297         }
 298
 299         updateConfigParams(datastoreContext.getShardRaftConfig());
 300     }
 301
 302     private void handleTransactionCommitTimeoutCheck() {
 303         CohortEntry cohortEntry = commitCoordinator.getCurrentCohortEntry();
 304         if(cohortEntry != null) {
 305             long elapsed = System.currentTimeMillis() - cohortEntry.getLastAccessTime();
 306             if(elapsed > transactionCommitTimeout) {
 307                 LOG.warn("{}: Current transaction {} has timed out after {} ms - aborting",
 308                         persistenceId(), cohortEntry.getTransactionID(), transactionCommitTimeout);
 309
 310                 doAbortTransaction(cohortEntry.getTransactionID(), null);
 311             }
 312         }
 313     }
 314
 315     private static boolean isEmptyCommit(final DataTreeCandidate candidate) {
 316         return ModificationType.UNMODIFIED.equals(candidate.getRootNode().getModificationType());
 317     }
 318
 319     void continueCommit(final CohortEntry cohortEntry) throws Exception {
 320         final DataTreeCandidate candidate = cohortEntry.getCohort().getCandidate();
 321
 322         // If we do not have any followers and we are not using persistence
 323         // or if cohortEntry has no modifications
 324         // we can apply modification to the state immediately
 325         if ((!hasFollowers() && !persistence().isRecoveryApplicable()) || isEmptyCommit(candidate)) {
 326             applyModificationToState(getSender(), cohortEntry.getTransactionID(), candidate);
 327         } else {
 328             Shard.this.persistData(getSender(), cohortEntry.getTransactionID(),
 329                 DataTreeCandidatePayload.create(candidate));
 330         }
 331     }
 332
 333     private void handleCommitTransaction(final CommitTransaction commit) {
 334         if(!commitCoordinator.handleCommit(commit.getTransactionID(), getSender(), this)) {
 335             shardMBean.incrementFailedTransactionsCount();
 336         }
 337     }
 338
 339     private void finishCommit(@Nonnull final ActorRef sender, @Nonnull final String transactionID, @Nonnull final CohortEntry cohortEntry) {
 340         LOG.debug("{}: Finishing commit for transaction {}", persistenceId(), cohortEntry.getTransactionID());
 341
 342         try {
 343             // We block on the future here so we don't have to worry about possibly accessing our
 344             // state on a different thread outside of our dispatcher. Also, the data store
 345             // currently uses a same thread executor anyway.
 346             cohortEntry.getCohort().commit().get();
 347
 348             sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
 349
 350             shardMBean.incrementCommittedTransactionCount();
 351             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 352
 353         } catch (Exception e) {
 354             sender.tell(new akka.actor.Status.Failure(e), getSelf());
 355
 356             LOG.error("{}, An exception occurred while committing transaction {}", persistenceId(),
 357                     transactionID, e);
 358             shardMBean.incrementFailedTransactionsCount();
 359         } finally {
 360             commitCoordinator.currentTransactionComplete(transactionID, true);
 361         }
 362     }
 363
 364     private void finishCommit(@Nonnull final ActorRef sender, final @Nonnull String transactionID) {
 365         // With persistence enabled, this method is called via applyState by the leader strategy
 366         // after the commit has been replicated to a majority of the followers.
 367
 368         CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
 369         if (cohortEntry == null) {
 370             // The transaction is no longer the current commit. This can happen if the transaction
 371             // was aborted prior, most likely due to timeout in the front-end. We need to finish
 372             // committing the transaction though since it was successfully persisted and replicated
 373             // however we can't use the original cohort b/c it was already preCommitted and may
 374             // conflict with the current commit or may have been aborted so we commit with a new
 375             // transaction.
 376             cohortEntry = commitCoordinator.getAndRemoveCohortEntry(transactionID);
 377             if(cohortEntry != null) {
 378                 try {
 379                     store.applyForeignCandidate(transactionID, cohortEntry.getCohort().getCandidate());
 380                 } catch (DataValidationFailedException e) {
 381                     shardMBean.incrementFailedTransactionsCount();
 382                     LOG.error("{}: Failed to re-apply transaction {}", persistenceId(), transactionID, e);
 383                 }
 384
 385                 sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
 386             } else {
 387                 // This really shouldn't happen - it likely means that persistence or replication
 388                 // took so long to complete such that the cohort entry was expired from the cache.
 389                 IllegalStateException ex = new IllegalStateException(
 390                         String.format("%s: Could not finish committing transaction %s - no CohortEntry found",
 391                                 persistenceId(), transactionID));
 392                 LOG.error(ex.getMessage());
 393                 sender.tell(new akka.actor.Status.Failure(ex), getSelf());
 394             }
 395         } else {
 396             finishCommit(sender, transactionID, cohortEntry);
 397         }
 398     }
 399
 400     private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
 401         LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionID());
 402         commitCoordinator.handleCanCommit(canCommit.getTransactionID(), getSender(), this);
 403     }
 404
 405     private void noLeaderError(Object message) {
 406         // TODO: rather than throwing an immediate exception, we could schedule a timer to try again to make
 407         // it more resilient in case we're in the process of electing a new leader.
 408         getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format(
 409             "Could not find the leader for shard %s. This typically happens" +
 410             " when the system is coming up or recovering and a leader is being elected. Try again" +
 411             " later.", persistenceId()))), getSelf());
 412     }
 413
 414     private void handleBatchedModifications(BatchedModifications batched) {
 415         // This message is sent to prepare the modificationsa transaction directly on the Shard as an
 416         // optimization to avoid the extra overhead of a separate ShardTransaction actor. On the last
 417         // BatchedModifications message, the caller sets the ready flag in the message indicating
 418         // modifications are complete. The reply contains the cohort actor path (this actor) for the caller
 419         // to initiate the 3-phase commit. This also avoids the overhead of sending an additional
 420         // ReadyTransaction message.
 421
 422         // If we're not the leader then forward to the leader. This is a safety measure - we shouldn't
 423         // normally get here if we're not the leader as the front-end (TransactionProxy) should determine
 424         // the primary/leader shard. However with timing and caching on the front-end, there's a small
 425         // window where it could have a stale leader during leadership transitions.
 426         //
 427         if(isLeader()) {
 428             try {
 429                 commitCoordinator.handleBatchedModifications(batched, getSender(), this);
 430             } catch (Exception e) {
 431                 LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(),
 432                         batched.getTransactionID(), e);
 433                 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 434             }
 435         } else {
 436             ActorSelection leader = getLeader();
 437             if(leader != null) {
 438                 // TODO: what if this is not the first batch and leadership changed in between batched messages?
 439                 // We could check if the commitCoordinator already has a cached entry and forward all the previous
 440                 // batched modifications.
 441                 LOG.debug("{}: Forwarding BatchedModifications to leader {}", persistenceId(), leader);
 442                 leader.forward(batched, getContext());
 443             } else {
 444                 noLeaderError(batched);
 445             }
 446         }
 447     }
 448
 449     private void handleReadyLocalTransaction(final ReadyLocalTransaction message) {
 450         if (isLeader()) {
 451             try {
 452                 commitCoordinator.handleReadyLocalTransaction(message, getSender(), this);
 453             } catch (Exception e) {
 454                 LOG.error("{}: Error handling LocalModifications for Tx {}", persistenceId(),
 455                         message.getTransactionID(), e);
 456                 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 457             }
 458         } else {
 459             ActorSelection leader = getLeader();
 460             if (leader != null) {
 461                 LOG.debug("{}: Forwarding LocalModifications to leader {}", persistenceId(), leader);
 462                 leader.forward(message, getContext());
 463             } else {
 464                 noLeaderError(message);
 465             }
 466         }
 467     }
 468
 469     private void handleAbortTransaction(final AbortTransaction abort) {
 470         doAbortTransaction(abort.getTransactionID(), getSender());
 471     }
 472
 473     void doAbortTransaction(final String transactionID, final ActorRef sender) {
 474         final CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
 475         if(cohortEntry != null) {
 476             LOG.debug("{}: Aborting transaction {}", persistenceId(), transactionID);
 477
 478             // We don't remove the cached cohort entry here (ie pass false) in case the Tx was
 479             // aborted during replication in which case we may still commit locally if replication
 480             // succeeds.
 481             commitCoordinator.currentTransactionComplete(transactionID, false);
 482
 483             final ListenableFuture<Void> future = cohortEntry.getCohort().abort();
 484             final ActorRef self = getSelf();
 485
 486             Futures.addCallback(future, new FutureCallback<Void>() {
 487                 @Override
 488                 public void onSuccess(final Void v) {
 489                     shardMBean.incrementAbortTransactionsCount();
 490
 491                     if(sender != null) {
 492                         sender.tell(AbortTransactionReply.INSTANCE.toSerializable(), self);
 493                     }
 494                 }
 495
 496                 @Override
 497                 public void onFailure(final Throwable t) {
 498                     LOG.error("{}: An exception happened during abort", persistenceId(), t);
 499
 500                     if(sender != null) {
 501                         sender.tell(new akka.actor.Status.Failure(t), self);
 502                     }
 503                 }
 504             });
 505         }
 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(String.format(
 515                 "Could not find leader for shard %s so transaction cannot be created. This typically happens" +
 516                 " when the system is coming up or recovering and a leader is being elected. Try again" +
 517                 " later.", persistenceId()))), getSelf());
 518         }
 519     }
 520
 521     private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
 522         store.closeTransactionChain(closeTransactionChain.getTransactionChainId());
 523     }
 524
 525     private ActorRef createTypedTransactionActor(int transactionType,
 526             ShardTransactionIdentifier transactionId, String transactionChainId,
 527             short clientVersion ) {
 528
 529         return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType),
 530                 transactionId, transactionChainId, clientVersion);
 531     }
 532
 533     private void createTransaction(CreateTransaction createTransaction) {
 534         try {
 535             ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
 536                 createTransaction.getTransactionId(), createTransaction.getTransactionChainId(),
 537                 createTransaction.getVersion());
 538
 539             getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
 540                     createTransaction.getTransactionId()).toSerializable(), getSelf());
 541         } catch (Exception e) {
 542             getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 543         }
 544     }
 545
 546     private ActorRef createTransaction(int transactionType, String remoteTransactionId,
 547             String transactionChainId, short clientVersion) {
 548
 549
 550         ShardTransactionIdentifier transactionId = new ShardTransactionIdentifier(remoteTransactionId);
 551
 552         if(LOG.isDebugEnabled()) {
 553             LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
 554         }
 555
 556         ActorRef transactionActor = createTypedTransactionActor(transactionType, transactionId,
 557                 transactionChainId, clientVersion);
 558
 559         return transactionActor;
 560     }
 561
 562     private void commitWithNewTransaction(final Modification modification) {
 563         ReadWriteShardDataTreeTransaction tx = store.newReadWriteTransaction(modification.toString(), null);
 564         modification.apply(tx.getSnapshot());
 565         try {
 566             snapshotCohort.syncCommitTransaction(tx);
 567             shardMBean.incrementCommittedTransactionCount();
 568             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 569         } catch (Exception e) {
 570             shardMBean.incrementFailedTransactionsCount();
 571             LOG.error("{}: Failed to commit", persistenceId(), e);
 572         }
 573     }
 574
 575     private void updateSchemaContext(final UpdateSchemaContext message) {
 576         updateSchemaContext(message.getSchemaContext());
 577     }
 578
 579     @VisibleForTesting
 580     void updateSchemaContext(final SchemaContext schemaContext) {
 581         store.updateSchemaContext(schemaContext);
 582     }
 583
 584     private boolean isMetricsCaptureEnabled() {
 585         CommonConfig config = new CommonConfig(getContext().system().settings().config());
 586         return config.isMetricCaptureEnabled();
 587     }
 588
 589     @Override
 590     protected RaftActorSnapshotCohort getRaftActorSnapshotCohort() {
 591         return snapshotCohort;
 592     }
 593
 594     @Override
 595     @Nonnull
 596     protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
 597         return new ShardRecoveryCoordinator(store, persistenceId(), LOG);
 598     }
 599
 600     @Override
 601     protected void onRecoveryComplete() {
 602         //notify shard manager
 603         getContext().parent().tell(new ActorInitialized(), getSelf());
 604
 605         // Being paranoid here - this method should only be called once but just in case...
 606         if(txCommitTimeoutCheckSchedule == null) {
 607             // Schedule a message to be periodically sent to check if the current in-progress
 608             // transaction should be expired and aborted.
 609             FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
 610             txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
 611                     period, period, getSelf(),
 612                     TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
 613         }
 614     }
 615
 616     @Override
 617     protected void applyState(final ActorRef clientActor, final String identifier, final Object data) {
 618         if (data instanceof DataTreeCandidatePayload) {
 619             if (clientActor == null) {
 620                 // No clientActor indicates a replica coming from the leader
 621                 try {
 622                     store.applyForeignCandidate(identifier, ((DataTreeCandidatePayload)data).getCandidate());
 623                 } catch (DataValidationFailedException | IOException e) {
 624                     LOG.error("{}: Error applying replica {}", persistenceId(), identifier, e);
 625                 }
 626             } else {
 627                 // Replication consensus reached, proceed to commit
 628                 finishCommit(clientActor, identifier);
 629             }
 630         } else if (data instanceof ModificationPayload) {
 631             try {
 632                 applyModificationToState(clientActor, identifier, ((ModificationPayload) data).getModification());
 633             } catch (ClassNotFoundException | IOException e) {
 634                 LOG.error("{}: Error extracting ModificationPayload", persistenceId(), e);
 635             }
 636         } else if (data instanceof CompositeModificationPayload) {
 637             Object modification = ((CompositeModificationPayload) data).getModification();
 638
 639             applyModificationToState(clientActor, identifier, modification);
 640         } else if(data instanceof CompositeModificationByteStringPayload ){
 641             Object modification = ((CompositeModificationByteStringPayload) data).getModification();
 642
 643             applyModificationToState(clientActor, identifier, modification);
 644         } else {
 645             LOG.error("{}: Unknown state received {} Class loader = {} CompositeNodeMod.ClassLoader = {}",
 646                     persistenceId(), data, data.getClass().getClassLoader(),
 647                     CompositeModificationPayload.class.getClassLoader());
 648         }
 649     }
 650
 651     private void applyModificationToState(ActorRef clientActor, String identifier, Object modification) {
 652         if(modification == null) {
 653             LOG.error(
 654                     "{}: modification is null - this is very unexpected, clientActor = {}, identifier = {}",
 655                     persistenceId(), identifier, clientActor != null ? clientActor.path().toString() : null);
 656         } else if(clientActor == null) {
 657             // There's no clientActor to which to send a commit reply so we must be applying
 658             // replicated state from the leader.
 659             commitWithNewTransaction(MutableCompositeModification.fromSerializable(modification));
 660         } else {
 661             // This must be the OK to commit after replication consensus.
 662             finishCommit(clientActor, identifier);
 663         }
 664     }
 665
 666     @Override
 667     protected void onStateChanged() {
 668         boolean isLeader = isLeader();
 669         changeSupport.onLeadershipChange(isLeader);
 670         treeChangeSupport.onLeadershipChange(isLeader);
 671
 672         // If this actor is no longer the leader close all the transaction chains
 673         if (!isLeader) {
 674             if(LOG.isDebugEnabled()) {
 675                 LOG.debug(
 676                     "{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader",
 677                     persistenceId(), getId());
 678             }
 679
 680             store.closeAllTransactionChains();
 681         }
 682     }
 683
 684     @Override
 685     protected void onLeaderChanged(String oldLeader, String newLeader) {
 686         shardMBean.incrementLeadershipChangeCount();
 687     }
 688
 689     @Override
 690     public String persistenceId() {
 691         return this.name;
 692     }
 693
 694     @VisibleForTesting
 695     ShardCommitCoordinator getCommitCoordinator() {
 696         return commitCoordinator;
 697     }
 698
 699
 700     private static class ShardCreator implements Creator<Shard> {
 701
 702         private static final long serialVersionUID = 1L;
 703
 704         final ShardIdentifier name;
 705         final Map<String, String> peerAddresses;
 706         final DatastoreContext datastoreContext;
 707         final SchemaContext schemaContext;
 708
 709         ShardCreator(final ShardIdentifier name, final Map<String, String> peerAddresses,
 710                 final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 711             this.name = name;
 712             this.peerAddresses = peerAddresses;
 713             this.datastoreContext = datastoreContext;
 714             this.schemaContext = schemaContext;
 715         }
 716
 717         @Override
 718         public Shard create() throws Exception {
 719             return new Shard(name, peerAddresses, datastoreContext, schemaContext);
 720         }
 721     }
 722
 723     @VisibleForTesting
 724     public ShardDataTree getDataStore() {
 725         return store;
 726     }
 727
 728     @VisibleForTesting
 729     ShardStats getShardMBean() {
 730         return shardMBean;
 731     }
 732 }