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.ShardLeaderStateChanged;
  52 import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext;
  53 import org.opendaylight.controller.cluster.datastore.modification.Modification;
  54 import org.opendaylight.controller.cluster.datastore.modification.ModificationPayload;
  55 import org.opendaylight.controller.cluster.datastore.modification.MutableCompositeModification;
  56 import org.opendaylight.controller.cluster.datastore.utils.Dispatchers;
  57 import org.opendaylight.controller.cluster.datastore.utils.MessageTracker;
  58 import org.opendaylight.controller.cluster.notifications.LeaderStateChanged;
  59 import org.opendaylight.controller.cluster.notifications.RegisterRoleChangeListener;
  60 import org.opendaylight.controller.cluster.notifications.RoleChangeNotifier;
  61 import org.opendaylight.controller.cluster.raft.RaftActor;
  62 import org.opendaylight.controller.cluster.raft.RaftActorRecoveryCohort;
  63 import org.opendaylight.controller.cluster.raft.RaftActorSnapshotCohort;
  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.protobuff.client.messages.CompositeModificationByteStringPayload;
  67 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationPayload;
  68 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTree;
  69 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidate;
  70 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException;
  71 import org.opendaylight.yangtools.yang.data.api.schema.tree.ModificationType;
  72 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
  73 import scala.concurrent.duration.Duration;
  74 import scala.concurrent.duration.FiniteDuration;
  75
  76 /**
  77  * A Shard represents a portion of the logical data tree <br/>
  78  * <p>
  79  * Our Shard uses InMemoryDataTree as it's internal representation and delegates all requests it
  80  * </p>
  81  */
  82 public class Shard extends RaftActor {
  83
  84     private static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = "txCommitTimeoutCheck";
  85
  86     @VisibleForTesting
  87     static final String DEFAULT_NAME = "default";
  88
  89     // The state of this Shard
  90     private final ShardDataTree store;
  91
  92     /// The name of this shard
  93     private final String name;
  94
  95     private final ShardStats shardMBean;
  96
  97     private DatastoreContext datastoreContext;
  98
  99     private final ShardCommitCoordinator commitCoordinator;
 100
 101     private long transactionCommitTimeout;
 102
 103     private Cancellable txCommitTimeoutCheckSchedule;
 104
 105     private final Optional<ActorRef> roleChangeNotifier;
 106
 107     private final MessageTracker appendEntriesReplyTracker;
 108
 109     private final ShardTransactionActorFactory transactionActorFactory;
 110
 111     private final ShardSnapshotCohort snapshotCohort;
 112
 113     private final DataTreeChangeListenerSupport treeChangeSupport = new DataTreeChangeListenerSupport(this);
 114     private final DataChangeListenerSupport changeSupport = new DataChangeListenerSupport(this);
 115
 116     protected Shard(final ShardIdentifier name, final Map<String, String> peerAddresses,
 117             final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 118         super(name.toString(), new HashMap<>(peerAddresses), Optional.of(datastoreContext.getShardRaftConfig()));
 119
 120         this.name = name.toString();
 121         this.datastoreContext = datastoreContext;
 122
 123         setPersistence(datastoreContext.isPersistent());
 124
 125         LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
 126
 127         store = new ShardDataTree(schemaContext);
 128
 129         shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
 130                 datastoreContext.getDataStoreMXBeanType());
 131         shardMBean.setShardActor(getSelf());
 132
 133         if (isMetricsCaptureEnabled()) {
 134             getContext().become(new MeteringBehavior(this));
 135         }
 136
 137         commitCoordinator = new ShardCommitCoordinator(store,
 138                 TimeUnit.SECONDS.convert(5, TimeUnit.MINUTES),
 139                 datastoreContext.getShardTransactionCommitQueueCapacity(), self(), LOG, this.name);
 140
 141         setTransactionCommitTimeout();
 142
 143         // create a notifier actor for each cluster member
 144         roleChangeNotifier = createRoleChangeNotifier(name.toString());
 145
 146         appendEntriesReplyTracker = new MessageTracker(AppendEntriesReply.class,
 147                 getRaftActorContext().getConfigParams().getIsolatedCheckIntervalInMillis());
 148
 149         transactionActorFactory = new ShardTransactionActorFactory(store, datastoreContext,
 150                 new Dispatchers(context().system().dispatchers()).getDispatcherPath(
 151                         Dispatchers.DispatcherType.Transaction), self(), getContext(), shardMBean);
 152
 153         snapshotCohort = new ShardSnapshotCohort(transactionActorFactory, store, LOG, this.name);
 154     }
 155
 156     private void setTransactionCommitTimeout() {
 157         transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
 158                 datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS);
 159     }
 160
 161     public static Props props(final ShardIdentifier name,
 162         final Map<String, String> peerAddresses,
 163         final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 164         Preconditions.checkNotNull(name, "name should not be null");
 165         Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
 166         Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
 167         Preconditions.checkNotNull(schemaContext, "schemaContext should not be null");
 168
 169         return Props.create(new ShardCreator(name, peerAddresses, datastoreContext, schemaContext));
 170     }
 171
 172     private Optional<ActorRef> createRoleChangeNotifier(String shardId) {
 173         ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
 174             RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
 175         return Optional.of(shardRoleChangeNotifier);
 176     }
 177
 178     @Override
 179     public void postStop() {
 180         LOG.info("Stopping Shard {}", persistenceId());
 181
 182         super.postStop();
 183
 184         if(txCommitTimeoutCheckSchedule != null) {
 185             txCommitTimeoutCheckSchedule.cancel();
 186         }
 187
 188         shardMBean.unregisterMBean();
 189     }
 190
 191     @Override
 192     public void onReceiveRecover(final Object message) throws Exception {
 193         if(LOG.isDebugEnabled()) {
 194             LOG.debug("{}: onReceiveRecover: Received message {} from {}", persistenceId(),
 195                 message.getClass().toString(), getSender());
 196         }
 197
 198         if (message instanceof RecoveryFailure){
 199             LOG.error("{}: Recovery failed because of this cause",
 200                     persistenceId(), ((RecoveryFailure) message).cause());
 201
 202             // Even though recovery failed, we still need to finish our recovery, eg send the
 203             // ActorInitialized message and start the txCommitTimeoutCheckSchedule.
 204             onRecoveryComplete();
 205         } else {
 206             super.onReceiveRecover(message);
 207             if(LOG.isTraceEnabled()) {
 208                 appendEntriesReplyTracker.begin();
 209             }
 210         }
 211     }
 212
 213     @Override
 214     public void onReceiveCommand(final Object message) throws Exception {
 215
 216         MessageTracker.Context context = appendEntriesReplyTracker.received(message);
 217
 218         if(context.error().isPresent()){
 219             LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(),
 220                     context.error());
 221         }
 222
 223         try {
 224             if (CreateTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 225                 handleCreateTransaction(message);
 226             } else if (BatchedModifications.class.isInstance(message)) {
 227                 handleBatchedModifications((BatchedModifications)message);
 228             } else if (message instanceof ForwardedReadyTransaction) {
 229                 commitCoordinator.handleForwardedReadyTransaction((ForwardedReadyTransaction) message,
 230                         getSender(), this);
 231             } else if (CanCommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 232                 handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
 233             } else if (CommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 234                 handleCommitTransaction(CommitTransaction.fromSerializable(message));
 235             } else if (AbortTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 236                 handleAbortTransaction(AbortTransaction.fromSerializable(message));
 237             } else if (CloseTransactionChain.SERIALIZABLE_CLASS.isInstance(message)) {
 238                 closeTransactionChain(CloseTransactionChain.fromSerializable(message));
 239             } else if (message instanceof RegisterChangeListener) {
 240                 changeSupport.onMessage((RegisterChangeListener) message, isLeader());
 241             } else if (message instanceof RegisterDataTreeChangeListener) {
 242                 treeChangeSupport.onMessage((RegisterDataTreeChangeListener) message, isLeader());
 243             } else if (message instanceof UpdateSchemaContext) {
 244                 updateSchemaContext((UpdateSchemaContext) message);
 245             } else if (message instanceof PeerAddressResolved) {
 246                 PeerAddressResolved resolved = (PeerAddressResolved) message;
 247                 setPeerAddress(resolved.getPeerId().toString(),
 248                         resolved.getPeerAddress());
 249             } else if (message.equals(TX_COMMIT_TIMEOUT_CHECK_MESSAGE)) {
 250                 handleTransactionCommitTimeoutCheck();
 251             } else if(message instanceof DatastoreContext) {
 252                 onDatastoreContext((DatastoreContext)message);
 253             } else if(message instanceof RegisterRoleChangeListener){
 254                 roleChangeNotifier.get().forward(message, context());
 255             } else if (message instanceof FollowerInitialSyncUpStatus){
 256                 shardMBean.setFollowerInitialSyncStatus(((FollowerInitialSyncUpStatus) message).isInitialSyncDone());
 257                 context().parent().tell(message, self());
 258             } else {
 259                 super.onReceiveCommand(message);
 260             }
 261         } finally {
 262             context.done();
 263         }
 264     }
 265
 266     @Override
 267     protected Optional<ActorRef> getRoleChangeNotifier() {
 268         return roleChangeNotifier;
 269     }
 270
 271     @Override
 272     protected LeaderStateChanged newLeaderStateChanged(String memberId, String leaderId) {
 273         return new ShardLeaderStateChanged(memberId, leaderId,
 274                 isLeader() ? Optional.<DataTree>of(store.getDataTree()) : Optional.<DataTree>absent());
 275     }
 276
 277     private void onDatastoreContext(DatastoreContext context) {
 278         datastoreContext = context;
 279
 280         commitCoordinator.setQueueCapacity(datastoreContext.getShardTransactionCommitQueueCapacity());
 281
 282         setTransactionCommitTimeout();
 283
 284         if(datastoreContext.isPersistent() && !persistence().isRecoveryApplicable()) {
 285             setPersistence(true);
 286         } else if(!datastoreContext.isPersistent() && persistence().isRecoveryApplicable()) {
 287             setPersistence(false);
 288         }
 289
 290         updateConfigParams(datastoreContext.getShardRaftConfig());
 291     }
 292
 293     private void handleTransactionCommitTimeoutCheck() {
 294         CohortEntry cohortEntry = commitCoordinator.getCurrentCohortEntry();
 295         if(cohortEntry != null) {
 296             long elapsed = System.currentTimeMillis() - cohortEntry.getLastAccessTime();
 297             if(elapsed > transactionCommitTimeout) {
 298                 LOG.warn("{}: Current transaction {} has timed out after {} ms - aborting",
 299                         persistenceId(), cohortEntry.getTransactionID(), transactionCommitTimeout);
 300
 301                 doAbortTransaction(cohortEntry.getTransactionID(), null);
 302             }
 303         }
 304     }
 305
 306     private static boolean isEmptyCommit(final DataTreeCandidate candidate) {
 307         return ModificationType.UNMODIFIED.equals(candidate.getRootNode().getModificationType());
 308     }
 309
 310     void continueCommit(final CohortEntry cohortEntry) throws Exception {
 311         final DataTreeCandidate candidate = cohortEntry.getCohort().getCandidate();
 312
 313         // If we do not have any followers and we are not using persistence
 314         // or if cohortEntry has no modifications
 315         // we can apply modification to the state immediately
 316         if ((!hasFollowers() && !persistence().isRecoveryApplicable()) || isEmptyCommit(candidate)) {
 317             applyModificationToState(getSender(), cohortEntry.getTransactionID(), candidate);
 318         } else {
 319             Shard.this.persistData(getSender(), cohortEntry.getTransactionID(),
 320                 DataTreeCandidatePayload.create(candidate));
 321         }
 322     }
 323
 324     private void handleCommitTransaction(final CommitTransaction commit) {
 325         if(!commitCoordinator.handleCommit(commit.getTransactionID(), getSender(), this)) {
 326             shardMBean.incrementFailedTransactionsCount();
 327         }
 328     }
 329
 330     private void finishCommit(@Nonnull final ActorRef sender, @Nonnull final String transactionID, @Nonnull final CohortEntry cohortEntry) {
 331         LOG.debug("{}: Finishing commit for transaction {}", persistenceId(), cohortEntry.getTransactionID());
 332
 333         try {
 334             // We block on the future here so we don't have to worry about possibly accessing our
 335             // state on a different thread outside of our dispatcher. Also, the data store
 336             // currently uses a same thread executor anyway.
 337             cohortEntry.getCohort().commit().get();
 338
 339             sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
 340
 341             shardMBean.incrementCommittedTransactionCount();
 342             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 343
 344         } catch (Exception e) {
 345             sender.tell(new akka.actor.Status.Failure(e), getSelf());
 346
 347             LOG.error("{}, An exception occurred while committing transaction {}", persistenceId(),
 348                     transactionID, e);
 349             shardMBean.incrementFailedTransactionsCount();
 350         } finally {
 351             commitCoordinator.currentTransactionComplete(transactionID, true);
 352         }
 353     }
 354
 355     private void finishCommit(@Nonnull final ActorRef sender, final @Nonnull String transactionID) {
 356         // With persistence enabled, this method is called via applyState by the leader strategy
 357         // after the commit has been replicated to a majority of the followers.
 358
 359         CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
 360         if (cohortEntry == null) {
 361             // The transaction is no longer the current commit. This can happen if the transaction
 362             // was aborted prior, most likely due to timeout in the front-end. We need to finish
 363             // committing the transaction though since it was successfully persisted and replicated
 364             // however we can't use the original cohort b/c it was already preCommitted and may
 365             // conflict with the current commit or may have been aborted so we commit with a new
 366             // transaction.
 367             cohortEntry = commitCoordinator.getAndRemoveCohortEntry(transactionID);
 368             if(cohortEntry != null) {
 369                 try {
 370                     store.applyForeignCandidate(transactionID, cohortEntry.getCohort().getCandidate());
 371                 } catch (DataValidationFailedException e) {
 372                     shardMBean.incrementFailedTransactionsCount();
 373                     LOG.error("{}: Failed to re-apply transaction {}", persistenceId(), transactionID, e);
 374                 }
 375
 376                 sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
 377             } else {
 378                 // This really shouldn't happen - it likely means that persistence or replication
 379                 // took so long to complete such that the cohort entry was expired from the cache.
 380                 IllegalStateException ex = new IllegalStateException(
 381                         String.format("%s: Could not finish committing transaction %s - no CohortEntry found",
 382                                 persistenceId(), transactionID));
 383                 LOG.error(ex.getMessage());
 384                 sender.tell(new akka.actor.Status.Failure(ex), getSelf());
 385             }
 386         } else {
 387             finishCommit(sender, transactionID, cohortEntry);
 388         }
 389     }
 390
 391     private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
 392         LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionID());
 393         commitCoordinator.handleCanCommit(canCommit.getTransactionID(), getSender(), this);
 394     }
 395
 396     private void handleBatchedModifications(BatchedModifications batched) {
 397         // This message is sent to prepare the modificationsa transaction directly on the Shard as an
 398         // optimization to avoid the extra overhead of a separate ShardTransaction actor. On the last
 399         // BatchedModifications message, the caller sets the ready flag in the message indicating
 400         // modifications are complete. The reply contains the cohort actor path (this actor) for the caller
 401         // to initiate the 3-phase commit. This also avoids the overhead of sending an additional
 402         // ReadyTransaction message.
 403
 404         // If we're not the leader then forward to the leader. This is a safety measure - we shouldn't
 405         // normally get here if we're not the leader as the front-end (TransactionProxy) should determine
 406         // the primary/leader shard. However with timing and caching on the front-end, there's a small
 407         // window where it could have a stale leader during leadership transitions.
 408         //
 409         if(isLeader()) {
 410             try {
 411                 commitCoordinator.handleBatchedModifications(batched, getSender(), this);
 412             } catch (Exception e) {
 413                 LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(),
 414                         batched.getTransactionID(), e);
 415                 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 416             }
 417         } else {
 418             ActorSelection leader = getLeader();
 419             if(leader != null) {
 420                 // TODO: what if this is not the first batch and leadership changed in between batched messages?
 421                 // We could check if the commitCoordinator already has a cached entry and forward all the previous
 422                 // batched modifications.
 423                 LOG.debug("{}: Forwarding BatchedModifications to leader {}", persistenceId(), leader);
 424                 leader.forward(batched, getContext());
 425             } else {
 426                 // TODO: rather than throwing an immediate exception, we could schedule a timer to try again to make
 427                 // it more resilient in case we're in the process of electing a new leader.
 428                 getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format(
 429                     "Could not find the leader for shard %s. This typically happens" +
 430                     " when the system is coming up or recovering and a leader is being elected. Try again" +
 431                     " later.", persistenceId()))), getSelf());
 432             }
 433         }
 434     }
 435
 436     private void handleAbortTransaction(final AbortTransaction abort) {
 437         doAbortTransaction(abort.getTransactionID(), getSender());
 438     }
 439
 440     void doAbortTransaction(final String transactionID, final ActorRef sender) {
 441         final CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
 442         if(cohortEntry != null) {
 443             LOG.debug("{}: Aborting transaction {}", persistenceId(), transactionID);
 444
 445             // We don't remove the cached cohort entry here (ie pass false) in case the Tx was
 446             // aborted during replication in which case we may still commit locally if replication
 447             // succeeds.
 448             commitCoordinator.currentTransactionComplete(transactionID, false);
 449
 450             final ListenableFuture<Void> future = cohortEntry.getCohort().abort();
 451             final ActorRef self = getSelf();
 452
 453             Futures.addCallback(future, new FutureCallback<Void>() {
 454                 @Override
 455                 public void onSuccess(final Void v) {
 456                     shardMBean.incrementAbortTransactionsCount();
 457
 458                     if(sender != null) {
 459                         sender.tell(AbortTransactionReply.INSTANCE.toSerializable(), self);
 460                     }
 461                 }
 462
 463                 @Override
 464                 public void onFailure(final Throwable t) {
 465                     LOG.error("{}: An exception happened during abort", persistenceId(), t);
 466
 467                     if(sender != null) {
 468                         sender.tell(new akka.actor.Status.Failure(t), self);
 469                     }
 470                 }
 471             });
 472         }
 473     }
 474
 475     private void handleCreateTransaction(final Object message) {
 476         if (isLeader()) {
 477             createTransaction(CreateTransaction.fromSerializable(message));
 478         } else if (getLeader() != null) {
 479             getLeader().forward(message, getContext());
 480         } else {
 481             getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format(
 482                 "Could not find leader for shard %s so transaction cannot be created. This typically happens" +
 483                 " when the system is coming up or recovering and a leader is being elected. Try again" +
 484                 " later.", persistenceId()))), getSelf());
 485         }
 486     }
 487
 488     private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
 489         store.closeTransactionChain(closeTransactionChain.getTransactionChainId());
 490     }
 491
 492     private ActorRef createTypedTransactionActor(int transactionType,
 493             ShardTransactionIdentifier transactionId, String transactionChainId,
 494             short clientVersion ) {
 495
 496         return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType),
 497                 transactionId, transactionChainId, clientVersion);
 498     }
 499
 500     private void createTransaction(CreateTransaction createTransaction) {
 501         try {
 502             ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
 503                 createTransaction.getTransactionId(), createTransaction.getTransactionChainId(),
 504                 createTransaction.getVersion());
 505
 506             getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
 507                     createTransaction.getTransactionId()).toSerializable(), getSelf());
 508         } catch (Exception e) {
 509             getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 510         }
 511     }
 512
 513     private ActorRef createTransaction(int transactionType, String remoteTransactionId,
 514             String transactionChainId, short clientVersion) {
 515
 516
 517         ShardTransactionIdentifier transactionId = new ShardTransactionIdentifier(remoteTransactionId);
 518
 519         if(LOG.isDebugEnabled()) {
 520             LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
 521         }
 522
 523         ActorRef transactionActor = createTypedTransactionActor(transactionType, transactionId,
 524                 transactionChainId, clientVersion);
 525
 526         return transactionActor;
 527     }
 528
 529     private void commitWithNewTransaction(final Modification modification) {
 530         ReadWriteShardDataTreeTransaction tx = store.newReadWriteTransaction(modification.toString(), null);
 531         modification.apply(tx.getSnapshot());
 532         try {
 533             snapshotCohort.syncCommitTransaction(tx);
 534             shardMBean.incrementCommittedTransactionCount();
 535             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 536         } catch (Exception e) {
 537             shardMBean.incrementFailedTransactionsCount();
 538             LOG.error("{}: Failed to commit", persistenceId(), e);
 539         }
 540     }
 541
 542     private void updateSchemaContext(final UpdateSchemaContext message) {
 543         updateSchemaContext(message.getSchemaContext());
 544     }
 545
 546     @VisibleForTesting
 547     void updateSchemaContext(final SchemaContext schemaContext) {
 548         store.updateSchemaContext(schemaContext);
 549     }
 550
 551     private boolean isMetricsCaptureEnabled() {
 552         CommonConfig config = new CommonConfig(getContext().system().settings().config());
 553         return config.isMetricCaptureEnabled();
 554     }
 555
 556     @Override
 557     protected RaftActorSnapshotCohort getRaftActorSnapshotCohort() {
 558         return snapshotCohort;
 559     }
 560
 561     @Override
 562     @Nonnull
 563     protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
 564         return new ShardRecoveryCoordinator(store, persistenceId(), LOG);
 565     }
 566
 567     @Override
 568     protected void onRecoveryComplete() {
 569         //notify shard manager
 570         getContext().parent().tell(new ActorInitialized(), getSelf());
 571
 572         // Being paranoid here - this method should only be called once but just in case...
 573         if(txCommitTimeoutCheckSchedule == null) {
 574             // Schedule a message to be periodically sent to check if the current in-progress
 575             // transaction should be expired and aborted.
 576             FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
 577             txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
 578                     period, period, getSelf(),
 579                     TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
 580         }
 581     }
 582
 583     @Override
 584     protected void applyState(final ActorRef clientActor, final String identifier, final Object data) {
 585         if (data instanceof DataTreeCandidatePayload) {
 586             if (clientActor == null) {
 587                 // No clientActor indicates a replica coming from the leader
 588                 try {
 589                     store.applyForeignCandidate(identifier, ((DataTreeCandidatePayload)data).getCandidate());
 590                 } catch (DataValidationFailedException | IOException e) {
 591                     LOG.error("{}: Error applying replica {}", persistenceId(), identifier, e);
 592                 }
 593             } else {
 594                 // Replication consensus reached, proceed to commit
 595                 finishCommit(clientActor, identifier);
 596             }
 597         } else if (data instanceof ModificationPayload) {
 598             try {
 599                 applyModificationToState(clientActor, identifier, ((ModificationPayload) data).getModification());
 600             } catch (ClassNotFoundException | IOException e) {
 601                 LOG.error("{}: Error extracting ModificationPayload", persistenceId(), e);
 602             }
 603         } else if (data instanceof CompositeModificationPayload) {
 604             Object modification = ((CompositeModificationPayload) data).getModification();
 605
 606             applyModificationToState(clientActor, identifier, modification);
 607         } else if(data instanceof CompositeModificationByteStringPayload ){
 608             Object modification = ((CompositeModificationByteStringPayload) data).getModification();
 609
 610             applyModificationToState(clientActor, identifier, modification);
 611         } else {
 612             LOG.error("{}: Unknown state received {} Class loader = {} CompositeNodeMod.ClassLoader = {}",
 613                     persistenceId(), data, data.getClass().getClassLoader(),
 614                     CompositeModificationPayload.class.getClassLoader());
 615         }
 616     }
 617
 618     private void applyModificationToState(ActorRef clientActor, String identifier, Object modification) {
 619         if(modification == null) {
 620             LOG.error(
 621                     "{}: modification is null - this is very unexpected, clientActor = {}, identifier = {}",
 622                     persistenceId(), identifier, clientActor != null ? clientActor.path().toString() : null);
 623         } else if(clientActor == null) {
 624             // There's no clientActor to which to send a commit reply so we must be applying
 625             // replicated state from the leader.
 626             commitWithNewTransaction(MutableCompositeModification.fromSerializable(modification));
 627         } else {
 628             // This must be the OK to commit after replication consensus.
 629             finishCommit(clientActor, identifier);
 630         }
 631     }
 632
 633     @Override
 634     protected void onStateChanged() {
 635         boolean isLeader = isLeader();
 636         changeSupport.onLeadershipChange(isLeader);
 637         treeChangeSupport.onLeadershipChange(isLeader);
 638
 639         // If this actor is no longer the leader close all the transaction chains
 640         if (!isLeader) {
 641             if(LOG.isDebugEnabled()) {
 642                 LOG.debug(
 643                     "{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader",
 644                     persistenceId(), getId());
 645             }
 646
 647             store.closeAllTransactionChains();
 648         }
 649     }
 650
 651     @Override
 652     protected void onLeaderChanged(String oldLeader, String newLeader) {
 653         shardMBean.incrementLeadershipChangeCount();
 654     }
 655
 656     @Override
 657     public String persistenceId() {
 658         return this.name;
 659     }
 660
 661     @VisibleForTesting
 662     ShardCommitCoordinator getCommitCoordinator() {
 663         return commitCoordinator;
 664     }
 665
 666
 667     private static class ShardCreator implements Creator<Shard> {
 668
 669         private static final long serialVersionUID = 1L;
 670
 671         final ShardIdentifier name;
 672         final Map<String, String> peerAddresses;
 673         final DatastoreContext datastoreContext;
 674         final SchemaContext schemaContext;
 675
 676         ShardCreator(final ShardIdentifier name, final Map<String, String> peerAddresses,
 677                 final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 678             this.name = name;
 679             this.peerAddresses = peerAddresses;
 680             this.datastoreContext = datastoreContext;
 681             this.schemaContext = schemaContext;
 682         }
 683
 684         @Override
 685         public Shard create() throws Exception {
 686             return new Shard(name, peerAddresses, datastoreContext, schemaContext);
 687         }
 688     }
 689
 690     @VisibleForTesting
 691     public ShardDataTree getDataStore() {
 692         return store;
 693     }
 694
 695     @VisibleForTesting
 696     ShardStats getShardMBean() {
 697         return shardMBean;
 698     }
 699 }