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.RegisterChangeListener;
  50 import org.opendaylight.controller.cluster.datastore.messages.RegisterDataTreeChangeListener;
  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.RegisterRoleChangeListener;
  58 import org.opendaylight.controller.cluster.notifications.RoleChangeNotifier;
  59 import org.opendaylight.controller.cluster.raft.RaftActor;
  60 import org.opendaylight.controller.cluster.raft.RaftActorRecoveryCohort;
  61 import org.opendaylight.controller.cluster.raft.RaftActorSnapshotCohort;
  62 import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus;
  63 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  64 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationByteStringPayload;
  65 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationPayload;
  66 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
  67 import scala.concurrent.duration.Duration;
  68 import scala.concurrent.duration.FiniteDuration;
  69
  70 /**
  71  * A Shard represents a portion of the logical data tree <br/>
  72  * <p>
  73  * Our Shard uses InMemoryDataTree as it's internal representation and delegates all requests it
  74  * </p>
  75  */
  76 public class Shard extends RaftActor {
  77
  78     private static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = "txCommitTimeoutCheck";
  79
  80     @VisibleForTesting
  81     static final String DEFAULT_NAME = "default";
  82
  83     // The state of this Shard
  84     private final ShardDataTree store;
  85
  86     /// The name of this shard
  87     private final String name;
  88
  89     private final ShardStats shardMBean;
  90
  91     private DatastoreContext datastoreContext;
  92
  93     private final ShardCommitCoordinator commitCoordinator;
  94
  95     private long transactionCommitTimeout;
  96
  97     private Cancellable txCommitTimeoutCheckSchedule;
  98
  99     private final Optional<ActorRef> roleChangeNotifier;
 100
 101     private final MessageTracker appendEntriesReplyTracker;
 102
 103     private final ShardTransactionActorFactory transactionActorFactory;
 104
 105     private final ShardSnapshotCohort snapshotCohort;
 106
 107     private final DataTreeChangeListenerSupport treeChangeSupport = new DataTreeChangeListenerSupport(this);
 108     private final DataChangeListenerSupport changeSupport = new DataChangeListenerSupport(this);
 109
 110     protected Shard(final ShardIdentifier name, final Map<String, String> peerAddresses,
 111             final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 112         super(name.toString(), new HashMap<>(peerAddresses), Optional.of(datastoreContext.getShardRaftConfig()));
 113
 114         this.name = name.toString();
 115         this.datastoreContext = datastoreContext;
 116
 117         setPersistence(datastoreContext.isPersistent());
 118
 119         LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
 120
 121         store = new ShardDataTree(schemaContext);
 122
 123         shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
 124                 datastoreContext.getDataStoreMXBeanType());
 125         shardMBean.setShardActor(getSelf());
 126
 127         if (isMetricsCaptureEnabled()) {
 128             getContext().become(new MeteringBehavior(this));
 129         }
 130
 131         commitCoordinator = new ShardCommitCoordinator(store,
 132                 TimeUnit.SECONDS.convert(5, TimeUnit.MINUTES),
 133                 datastoreContext.getShardTransactionCommitQueueCapacity(), self(), LOG, this.name);
 134
 135         setTransactionCommitTimeout();
 136
 137         // create a notifier actor for each cluster member
 138         roleChangeNotifier = createRoleChangeNotifier(name.toString());
 139
 140         appendEntriesReplyTracker = new MessageTracker(AppendEntriesReply.class,
 141                 getRaftActorContext().getConfigParams().getIsolatedCheckIntervalInMillis());
 142
 143         transactionActorFactory = new ShardTransactionActorFactory(store, datastoreContext,
 144                 new Dispatchers(context().system().dispatchers()).getDispatcherPath(
 145                         Dispatchers.DispatcherType.Transaction), self(), getContext(), shardMBean);
 146
 147         snapshotCohort = new ShardSnapshotCohort(transactionActorFactory, store, LOG, this.name);
 148     }
 149
 150     private void setTransactionCommitTimeout() {
 151         transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
 152                 datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS);
 153     }
 154
 155     public static Props props(final ShardIdentifier name,
 156         final Map<String, String> peerAddresses,
 157         final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 158         Preconditions.checkNotNull(name, "name should not be null");
 159         Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
 160         Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
 161         Preconditions.checkNotNull(schemaContext, "schemaContext should not be null");
 162
 163         return Props.create(new ShardCreator(name, peerAddresses, datastoreContext, schemaContext));
 164     }
 165
 166     private Optional<ActorRef> createRoleChangeNotifier(String shardId) {
 167         ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
 168             RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
 169         return Optional.of(shardRoleChangeNotifier);
 170     }
 171
 172     @Override
 173     public void postStop() {
 174         LOG.info("Stopping Shard {}", persistenceId());
 175
 176         super.postStop();
 177
 178         if(txCommitTimeoutCheckSchedule != null) {
 179             txCommitTimeoutCheckSchedule.cancel();
 180         }
 181
 182         shardMBean.unregisterMBean();
 183     }
 184
 185     @Override
 186     public void onReceiveRecover(final Object message) throws Exception {
 187         if(LOG.isDebugEnabled()) {
 188             LOG.debug("{}: onReceiveRecover: Received message {} from {}", persistenceId(),
 189                 message.getClass().toString(), getSender());
 190         }
 191
 192         if (message instanceof RecoveryFailure){
 193             LOG.error("{}: Recovery failed because of this cause",
 194                     persistenceId(), ((RecoveryFailure) message).cause());
 195
 196             // Even though recovery failed, we still need to finish our recovery, eg send the
 197             // ActorInitialized message and start the txCommitTimeoutCheckSchedule.
 198             onRecoveryComplete();
 199         } else {
 200             super.onReceiveRecover(message);
 201             if(LOG.isTraceEnabled()) {
 202                 appendEntriesReplyTracker.begin();
 203             }
 204         }
 205     }
 206
 207     @Override
 208     public void onReceiveCommand(final Object message) throws Exception {
 209
 210         MessageTracker.Context context = appendEntriesReplyTracker.received(message);
 211
 212         if(context.error().isPresent()){
 213             LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(),
 214                     context.error());
 215         }
 216
 217         try {
 218             if (CreateTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 219                 handleCreateTransaction(message);
 220             } else if (BatchedModifications.class.isInstance(message)) {
 221                 handleBatchedModifications((BatchedModifications)message);
 222             } else if (message instanceof ForwardedReadyTransaction) {
 223                 commitCoordinator.handleForwardedReadyTransaction((ForwardedReadyTransaction) message,
 224                         getSender(), this);
 225             } else if (CanCommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 226                 handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
 227             } else if (CommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 228                 handleCommitTransaction(CommitTransaction.fromSerializable(message));
 229             } else if (AbortTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 230                 handleAbortTransaction(AbortTransaction.fromSerializable(message));
 231             } else if (CloseTransactionChain.SERIALIZABLE_CLASS.isInstance(message)) {
 232                 closeTransactionChain(CloseTransactionChain.fromSerializable(message));
 233             } else if (message instanceof RegisterChangeListener) {
 234                 changeSupport.onMessage((RegisterChangeListener) message, isLeader());
 235             } else if (message instanceof RegisterDataTreeChangeListener) {
 236                 treeChangeSupport.onMessage((RegisterDataTreeChangeListener) message, isLeader());
 237             } else if (message instanceof UpdateSchemaContext) {
 238                 updateSchemaContext((UpdateSchemaContext) message);
 239             } else if (message instanceof PeerAddressResolved) {
 240                 PeerAddressResolved resolved = (PeerAddressResolved) message;
 241                 setPeerAddress(resolved.getPeerId().toString(),
 242                         resolved.getPeerAddress());
 243             } else if (message.equals(TX_COMMIT_TIMEOUT_CHECK_MESSAGE)) {
 244                 handleTransactionCommitTimeoutCheck();
 245             } else if(message instanceof DatastoreContext) {
 246                 onDatastoreContext((DatastoreContext)message);
 247             } else if(message instanceof RegisterRoleChangeListener){
 248                 roleChangeNotifier.get().forward(message, context());
 249             } else if (message instanceof FollowerInitialSyncUpStatus){
 250                 shardMBean.setFollowerInitialSyncStatus(((FollowerInitialSyncUpStatus) message).isInitialSyncDone());
 251                 context().parent().tell(message, self());
 252             } else {
 253                 super.onReceiveCommand(message);
 254             }
 255         } finally {
 256             context.done();
 257         }
 258     }
 259
 260     @Override
 261     protected Optional<ActorRef> getRoleChangeNotifier() {
 262         return roleChangeNotifier;
 263     }
 264
 265     private void onDatastoreContext(DatastoreContext context) {
 266         datastoreContext = context;
 267
 268         commitCoordinator.setQueueCapacity(datastoreContext.getShardTransactionCommitQueueCapacity());
 269
 270         setTransactionCommitTimeout();
 271
 272         if(datastoreContext.isPersistent() && !persistence().isRecoveryApplicable()) {
 273             setPersistence(true);
 274         } else if(!datastoreContext.isPersistent() && persistence().isRecoveryApplicable()) {
 275             setPersistence(false);
 276         }
 277
 278         updateConfigParams(datastoreContext.getShardRaftConfig());
 279     }
 280
 281     private void handleTransactionCommitTimeoutCheck() {
 282         CohortEntry cohortEntry = commitCoordinator.getCurrentCohortEntry();
 283         if(cohortEntry != null) {
 284             long elapsed = System.currentTimeMillis() - cohortEntry.getLastAccessTime();
 285             if(elapsed > transactionCommitTimeout) {
 286                 LOG.warn("{}: Current transaction {} has timed out after {} ms - aborting",
 287                         persistenceId(), cohortEntry.getTransactionID(), transactionCommitTimeout);
 288
 289                 doAbortTransaction(cohortEntry.getTransactionID(), null);
 290             }
 291         }
 292     }
 293
 294     void continueCommit(final CohortEntry cohortEntry) throws Exception {
 295         // If we do not have any followers and we are not using persistence
 296         // or if cohortEntry has no modifications
 297         // we can apply modification to the state immediately
 298         if((!hasFollowers() && !persistence().isRecoveryApplicable()) || (!cohortEntry.hasModifications())){
 299             applyModificationToState(getSender(), cohortEntry.getTransactionID(), cohortEntry.getModification());
 300         } else {
 301             Shard.this.persistData(getSender(), cohortEntry.getTransactionID(),
 302                     new ModificationPayload(cohortEntry.getModification()));
 303         }
 304     }
 305
 306     private void handleCommitTransaction(final CommitTransaction commit) {
 307         if(!commitCoordinator.handleCommit(commit.getTransactionID(), getSender(), this)) {
 308             shardMBean.incrementFailedTransactionsCount();
 309         }
 310     }
 311
 312     private void finishCommit(@Nonnull final ActorRef sender, final @Nonnull String transactionID) {
 313         // With persistence enabled, this method is called via applyState by the leader strategy
 314         // after the commit has been replicated to a majority of the followers.
 315
 316         CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
 317         if(cohortEntry == null) {
 318             // The transaction is no longer the current commit. This can happen if the transaction
 319             // was aborted prior, most likely due to timeout in the front-end. We need to finish
 320             // committing the transaction though since it was successfully persisted and replicated
 321             // however we can't use the original cohort b/c it was already preCommitted and may
 322             // conflict with the current commit or may have been aborted so we commit with a new
 323             // transaction.
 324             cohortEntry = commitCoordinator.getAndRemoveCohortEntry(transactionID);
 325             if(cohortEntry != null) {
 326                 commitWithNewTransaction(cohortEntry.getModification());
 327                 sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
 328             } else {
 329                 // This really shouldn't happen - it likely means that persistence or replication
 330                 // took so long to complete such that the cohort entry was expired from the cache.
 331                 IllegalStateException ex = new IllegalStateException(
 332                         String.format("%s: Could not finish committing transaction %s - no CohortEntry found",
 333                                 persistenceId(), transactionID));
 334                 LOG.error(ex.getMessage());
 335                 sender.tell(new akka.actor.Status.Failure(ex), getSelf());
 336             }
 337
 338             return;
 339         }
 340
 341         LOG.debug("{}: Finishing commit for transaction {}", persistenceId(), cohortEntry.getTransactionID());
 342
 343         try {
 344             // We block on the future here so we don't have to worry about possibly accessing our
 345             // state on a different thread outside of our dispatcher. Also, the data store
 346             // currently uses a same thread executor anyway.
 347             cohortEntry.getCohort().commit().get();
 348
 349             sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
 350
 351             shardMBean.incrementCommittedTransactionCount();
 352             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 353
 354         } catch (Exception e) {
 355             sender.tell(new akka.actor.Status.Failure(e), getSelf());
 356
 357             LOG.error("{}, An exception occurred while committing transaction {}", persistenceId(),
 358                     transactionID, e);
 359             shardMBean.incrementFailedTransactionsCount();
 360         } finally {
 361             commitCoordinator.currentTransactionComplete(transactionID, true);
 362         }
 363     }
 364
 365     private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
 366         LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionID());
 367         commitCoordinator.handleCanCommit(canCommit.getTransactionID(), getSender(), this);
 368     }
 369
 370     private void handleBatchedModifications(BatchedModifications batched) {
 371         // This message is sent to prepare the modificationsa transaction directly on the Shard as an
 372         // optimization to avoid the extra overhead of a separate ShardTransaction actor. On the last
 373         // BatchedModifications message, the caller sets the ready flag in the message indicating
 374         // modifications are complete. The reply contains the cohort actor path (this actor) for the caller
 375         // to initiate the 3-phase commit. This also avoids the overhead of sending an additional
 376         // ReadyTransaction message.
 377
 378         // If we're not the leader then forward to the leader. This is a safety measure - we shouldn't
 379         // normally get here if we're not the leader as the front-end (TransactionProxy) should determine
 380         // the primary/leader shard. However with timing and caching on the front-end, there's a small
 381         // window where it could have a stale leader during leadership transitions.
 382         //
 383         if(isLeader()) {
 384             try {
 385                 commitCoordinator.handleBatchedModifications(batched, getSender(), this);
 386             } catch (Exception e) {
 387                 LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(),
 388                         batched.getTransactionID(), e);
 389                 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 390             }
 391         } else {
 392             ActorSelection leader = getLeader();
 393             if(leader != null) {
 394                 // TODO: what if this is not the first batch and leadership changed in between batched messages?
 395                 // We could check if the commitCoordinator already has a cached entry and forward all the previous
 396                 // batched modifications.
 397                 LOG.debug("{}: Forwarding BatchedModifications to leader {}", persistenceId(), leader);
 398                 leader.forward(batched, getContext());
 399             } else {
 400                 // TODO: rather than throwing an immediate exception, we could schedule a timer to try again to make
 401                 // it more resilient in case we're in the process of electing a new leader.
 402                 getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format(
 403                     "Could not find the leader for shard %s. This typically happens" +
 404                     " when the system is coming up or recovering and a leader is being elected. Try again" +
 405                     " later.", persistenceId()))), getSelf());
 406             }
 407         }
 408     }
 409
 410     private void handleAbortTransaction(final AbortTransaction abort) {
 411         doAbortTransaction(abort.getTransactionID(), getSender());
 412     }
 413
 414     void doAbortTransaction(final String transactionID, final ActorRef sender) {
 415         final CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
 416         if(cohortEntry != null) {
 417             LOG.debug("{}: Aborting transaction {}", persistenceId(), transactionID);
 418
 419             // We don't remove the cached cohort entry here (ie pass false) in case the Tx was
 420             // aborted during replication in which case we may still commit locally if replication
 421             // succeeds.
 422             commitCoordinator.currentTransactionComplete(transactionID, false);
 423
 424             final ListenableFuture<Void> future = cohortEntry.getCohort().abort();
 425             final ActorRef self = getSelf();
 426
 427             Futures.addCallback(future, new FutureCallback<Void>() {
 428                 @Override
 429                 public void onSuccess(final Void v) {
 430                     shardMBean.incrementAbortTransactionsCount();
 431
 432                     if(sender != null) {
 433                         sender.tell(AbortTransactionReply.INSTANCE.toSerializable(), self);
 434                     }
 435                 }
 436
 437                 @Override
 438                 public void onFailure(final Throwable t) {
 439                     LOG.error("{}: An exception happened during abort", persistenceId(), t);
 440
 441                     if(sender != null) {
 442                         sender.tell(new akka.actor.Status.Failure(t), self);
 443                     }
 444                 }
 445             });
 446         }
 447     }
 448
 449     private void handleCreateTransaction(final Object message) {
 450         if (isLeader()) {
 451             createTransaction(CreateTransaction.fromSerializable(message));
 452         } else if (getLeader() != null) {
 453             getLeader().forward(message, getContext());
 454         } else {
 455             getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format(
 456                 "Could not find leader for shard %s so transaction cannot be created. This typically happens" +
 457                 " when the system is coming up or recovering and a leader is being elected. Try again" +
 458                 " later.", persistenceId()))), getSelf());
 459         }
 460     }
 461
 462     private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
 463         store.closeTransactionChain(closeTransactionChain.getTransactionChainId());
 464     }
 465
 466     private ActorRef createTypedTransactionActor(int transactionType,
 467             ShardTransactionIdentifier transactionId, String transactionChainId,
 468             short clientVersion ) {
 469
 470         return transactionActorFactory.newShardTransaction(TransactionProxy.TransactionType.fromInt(transactionType),
 471                 transactionId, transactionChainId, clientVersion);
 472     }
 473
 474     private void createTransaction(CreateTransaction createTransaction) {
 475         try {
 476             ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
 477                 createTransaction.getTransactionId(), createTransaction.getTransactionChainId(),
 478                 createTransaction.getVersion());
 479
 480             getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
 481                     createTransaction.getTransactionId()).toSerializable(), getSelf());
 482         } catch (Exception e) {
 483             getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 484         }
 485     }
 486
 487     private ActorRef createTransaction(int transactionType, String remoteTransactionId,
 488             String transactionChainId, short clientVersion) {
 489
 490
 491         ShardTransactionIdentifier transactionId = new ShardTransactionIdentifier(remoteTransactionId);
 492
 493         if(LOG.isDebugEnabled()) {
 494             LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
 495         }
 496
 497         ActorRef transactionActor = createTypedTransactionActor(transactionType, transactionId,
 498                 transactionChainId, clientVersion);
 499
 500         return transactionActor;
 501     }
 502
 503     private void commitWithNewTransaction(final Modification modification) {
 504         ReadWriteShardDataTreeTransaction tx = store.newReadWriteTransaction(modification.toString(), null);
 505         modification.apply(tx.getSnapshot());
 506         try {
 507             snapshotCohort.syncCommitTransaction(tx);
 508             shardMBean.incrementCommittedTransactionCount();
 509             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 510         } catch (Exception e) {
 511             shardMBean.incrementFailedTransactionsCount();
 512             LOG.error("{}: Failed to commit", persistenceId(), e);
 513         }
 514     }
 515
 516     private void updateSchemaContext(final UpdateSchemaContext message) {
 517         updateSchemaContext(message.getSchemaContext());
 518     }
 519
 520     @VisibleForTesting
 521     void updateSchemaContext(final SchemaContext schemaContext) {
 522         store.updateSchemaContext(schemaContext);
 523     }
 524
 525     private boolean isMetricsCaptureEnabled() {
 526         CommonConfig config = new CommonConfig(getContext().system().settings().config());
 527         return config.isMetricCaptureEnabled();
 528     }
 529
 530     @Override
 531     protected RaftActorSnapshotCohort getRaftActorSnapshotCohort() {
 532         return snapshotCohort;
 533     }
 534
 535     @Override
 536     @Nonnull
 537     protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
 538         return new ShardRecoveryCoordinator(store, persistenceId(), LOG);
 539     }
 540
 541     @Override
 542     protected void onRecoveryComplete() {
 543         //notify shard manager
 544         getContext().parent().tell(new ActorInitialized(), getSelf());
 545
 546         // Being paranoid here - this method should only be called once but just in case...
 547         if(txCommitTimeoutCheckSchedule == null) {
 548             // Schedule a message to be periodically sent to check if the current in-progress
 549             // transaction should be expired and aborted.
 550             FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
 551             txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
 552                     period, period, getSelf(),
 553                     TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
 554         }
 555     }
 556
 557     @Override
 558     protected void applyState(final ActorRef clientActor, final String identifier, final Object data) {
 559
 560         if(data instanceof ModificationPayload) {
 561             try {
 562                 applyModificationToState(clientActor, identifier, ((ModificationPayload) data).getModification());
 563             } catch (ClassNotFoundException | IOException e) {
 564                 LOG.error("{}: Error extracting ModificationPayload", persistenceId(), e);
 565             }
 566         }
 567         else if (data instanceof CompositeModificationPayload) {
 568             Object modification = ((CompositeModificationPayload) data).getModification();
 569
 570             applyModificationToState(clientActor, identifier, modification);
 571         } else if(data instanceof CompositeModificationByteStringPayload ){
 572             Object modification = ((CompositeModificationByteStringPayload) data).getModification();
 573
 574             applyModificationToState(clientActor, identifier, modification);
 575         } else {
 576             LOG.error("{}: Unknown state received {} Class loader = {} CompositeNodeMod.ClassLoader = {}",
 577                     persistenceId(), data, data.getClass().getClassLoader(),
 578                     CompositeModificationPayload.class.getClassLoader());
 579         }
 580     }
 581
 582     private void applyModificationToState(ActorRef clientActor, String identifier, Object modification) {
 583         if(modification == null) {
 584             LOG.error(
 585                     "{}: modification is null - this is very unexpected, clientActor = {}, identifier = {}",
 586                     persistenceId(), identifier, clientActor != null ? clientActor.path().toString() : null);
 587         } else if(clientActor == null) {
 588             // There's no clientActor to which to send a commit reply so we must be applying
 589             // replicated state from the leader.
 590             commitWithNewTransaction(MutableCompositeModification.fromSerializable(modification));
 591         } else {
 592             // This must be the OK to commit after replication consensus.
 593             finishCommit(clientActor, identifier);
 594         }
 595     }
 596
 597     @Override
 598     protected void onStateChanged() {
 599         boolean isLeader = isLeader();
 600         changeSupport.onLeadershipChange(isLeader);
 601         treeChangeSupport.onLeadershipChange(isLeader);
 602
 603         // If this actor is no longer the leader close all the transaction chains
 604         if (!isLeader) {
 605             if(LOG.isDebugEnabled()) {
 606                 LOG.debug(
 607                     "{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader",
 608                     persistenceId(), getId());
 609             }
 610
 611             store.closeAllTransactionChains();
 612         }
 613     }
 614
 615     @Override
 616     protected void onLeaderChanged(String oldLeader, String newLeader) {
 617         shardMBean.incrementLeadershipChangeCount();
 618     }
 619
 620     @Override
 621     public String persistenceId() {
 622         return this.name;
 623     }
 624
 625     @VisibleForTesting
 626     ShardCommitCoordinator getCommitCoordinator() {
 627         return commitCoordinator;
 628     }
 629
 630
 631     private static class ShardCreator implements Creator<Shard> {
 632
 633         private static final long serialVersionUID = 1L;
 634
 635         final ShardIdentifier name;
 636         final Map<String, String> peerAddresses;
 637         final DatastoreContext datastoreContext;
 638         final SchemaContext schemaContext;
 639
 640         ShardCreator(final ShardIdentifier name, final Map<String, String> peerAddresses,
 641                 final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 642             this.name = name;
 643             this.peerAddresses = peerAddresses;
 644             this.datastoreContext = datastoreContext;
 645             this.schemaContext = schemaContext;
 646         }
 647
 648         @Override
 649         public Shard create() throws Exception {
 650             return new Shard(name, peerAddresses, datastoreContext, schemaContext);
 651         }
 652     }
 653
 654     @VisibleForTesting
 655     public ShardDataTree getDataStore() {
 656         return store;
 657     }
 658
 659     @VisibleForTesting
 660     ShardStats getShardMBean() {
 661         return shardMBean;
 662     }
 663 }