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 java.io.IOException;
  22 import java.util.HashMap;
  23 import java.util.Map;
  24 import java.util.concurrent.TimeUnit;
  25 import javax.annotation.Nonnull;
  26 import org.opendaylight.controller.cluster.common.actor.CommonConfig;
  27 import org.opendaylight.controller.cluster.common.actor.MeteringBehavior;
  28 import org.opendaylight.controller.cluster.datastore.ShardCommitCoordinator.CohortEntry;
  29 import org.opendaylight.controller.cluster.datastore.exceptions.NoShardLeaderException;
  30 import org.opendaylight.controller.cluster.datastore.identifiers.ShardIdentifier;
  31 import org.opendaylight.controller.cluster.datastore.identifiers.ShardTransactionIdentifier;
  32 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardMBeanFactory;
  33 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardStats;
  34 import org.opendaylight.controller.cluster.datastore.messages.AbortTransaction;
  35 import org.opendaylight.controller.cluster.datastore.messages.ActorInitialized;
  36 import org.opendaylight.controller.cluster.datastore.messages.BatchedModifications;
  37 import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransaction;
  38 import org.opendaylight.controller.cluster.datastore.messages.CloseTransactionChain;
  39 import org.opendaylight.controller.cluster.datastore.messages.CommitTransaction;
  40 import org.opendaylight.controller.cluster.datastore.messages.CommitTransactionReply;
  41 import org.opendaylight.controller.cluster.datastore.messages.CreateTransaction;
  42 import org.opendaylight.controller.cluster.datastore.messages.CreateTransactionReply;
  43 import org.opendaylight.controller.cluster.datastore.messages.ForwardedReadyTransaction;
  44 import org.opendaylight.controller.cluster.datastore.messages.PeerAddressResolved;
  45 import org.opendaylight.controller.cluster.datastore.messages.ReadyLocalTransaction;
  46 import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListener;
  47 import org.opendaylight.controller.cluster.datastore.messages.RegisterDataTreeChangeListener;
  48 import org.opendaylight.controller.cluster.datastore.messages.ShardLeaderStateChanged;
  49 import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext;
  50 import org.opendaylight.controller.cluster.datastore.modification.Modification;
  51 import org.opendaylight.controller.cluster.datastore.modification.ModificationPayload;
  52 import org.opendaylight.controller.cluster.datastore.modification.MutableCompositeModification;
  53 import org.opendaylight.controller.cluster.datastore.utils.Dispatchers;
  54 import org.opendaylight.controller.cluster.datastore.utils.MessageTracker;
  55 import org.opendaylight.controller.cluster.notifications.LeaderStateChanged;
  56 import org.opendaylight.controller.cluster.notifications.RegisterRoleChangeListener;
  57 import org.opendaylight.controller.cluster.notifications.RoleChangeNotifier;
  58 import org.opendaylight.controller.cluster.raft.RaftActor;
  59 import org.opendaylight.controller.cluster.raft.RaftActorRecoveryCohort;
  60 import org.opendaylight.controller.cluster.raft.RaftActorSnapshotCohort;
  61 import org.opendaylight.controller.cluster.raft.RaftState;
  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.data.api.schema.tree.DataTree;
  67 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidate;
  68 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException;
  69 import org.opendaylight.yangtools.yang.data.api.schema.tree.ModificationType;
  70 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
  71 import scala.concurrent.duration.Duration;
  72 import scala.concurrent.duration.FiniteDuration;
  73
  74 /**
  75  * A Shard represents a portion of the logical data tree <br/>
  76  * <p>
  77  * Our Shard uses InMemoryDataTree as it's internal representation and delegates all requests it
  78  * </p>
  79  */
  80 public class Shard extends RaftActor {
  81
  82     protected static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = "txCommitTimeoutCheck";
  83
  84     @VisibleForTesting
  85     static final Object GET_SHARD_MBEAN_MESSAGE = "getShardMBeanMessage";
  86
  87     @VisibleForTesting
  88     static final String DEFAULT_NAME = "default";
  89
  90     // The state of this Shard
  91     private final ShardDataTree store;
  92
  93     /// The name of this shard
  94     private final String name;
  95
  96     private final ShardStats shardMBean;
  97
  98     private DatastoreContext datastoreContext;
  99
 100     private final ShardCommitCoordinator commitCoordinator;
 101
 102     private long transactionCommitTimeout;
 103
 104     private Cancellable txCommitTimeoutCheckSchedule;
 105
 106     private final Optional<ActorRef> roleChangeNotifier;
 107
 108     private final MessageTracker appendEntriesReplyTracker;
 109
 110     private final ShardTransactionActorFactory transactionActorFactory;
 111
 112     private final ShardSnapshotCohort snapshotCohort;
 113
 114     private final DataTreeChangeListenerSupport treeChangeSupport = new DataTreeChangeListenerSupport(this);
 115     private final DataChangeListenerSupport changeSupport = new DataChangeListenerSupport(this);
 116
 117     protected Shard(final ShardIdentifier name, final Map<String, String> peerAddresses,
 118             final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 119         super(name.toString(), new HashMap<>(peerAddresses), Optional.of(datastoreContext.getShardRaftConfig()),
 120                 DataStoreVersions.CURRENT_VERSION);
 121
 122         this.name = name.toString();
 123         this.datastoreContext = datastoreContext;
 124
 125         setPersistence(datastoreContext.isPersistent());
 126
 127         LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
 128
 129         store = new ShardDataTree(schemaContext);
 130
 131         shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
 132                 datastoreContext.getDataStoreMXBeanType());
 133         shardMBean.setShard(this);
 134
 135         if (isMetricsCaptureEnabled()) {
 136             getContext().become(new MeteringBehavior(this));
 137         }
 138
 139         commitCoordinator = new ShardCommitCoordinator(store,
 140                 datastoreContext.getShardCommitQueueExpiryTimeoutInMillis(),
 141                 datastoreContext.getShardTransactionCommitQueueCapacity(), self(), LOG, this.name);
 142
 143         setTransactionCommitTimeout();
 144
 145         // create a notifier actor for each cluster member
 146         roleChangeNotifier = createRoleChangeNotifier(name.toString());
 147
 148         appendEntriesReplyTracker = new MessageTracker(AppendEntriesReply.class,
 149                 getRaftActorContext().getConfigParams().getIsolatedCheckIntervalInMillis());
 150
 151         transactionActorFactory = new ShardTransactionActorFactory(store, datastoreContext,
 152                 new Dispatchers(context().system().dispatchers()).getDispatcherPath(
 153                         Dispatchers.DispatcherType.Transaction), self(), getContext(), shardMBean);
 154
 155         snapshotCohort = new ShardSnapshotCohort(transactionActorFactory, store, LOG, this.name);
 156
 157
 158     }
 159
 160     private void setTransactionCommitTimeout() {
 161         transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
 162                 datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS) / 2;
 163     }
 164
 165     public static Props props(final ShardIdentifier name, final Map<String, String> peerAddresses,
 166             final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 167         return Props.create(new ShardCreator(name, peerAddresses, datastoreContext, schemaContext));
 168     }
 169
 170     private Optional<ActorRef> createRoleChangeNotifier(String shardId) {
 171         ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
 172             RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
 173         return Optional.of(shardRoleChangeNotifier);
 174     }
 175
 176     @Override
 177     public void postStop() {
 178         LOG.info("Stopping Shard {}", persistenceId());
 179
 180         super.postStop();
 181
 182         if(txCommitTimeoutCheckSchedule != null) {
 183             txCommitTimeoutCheckSchedule.cancel();
 184         }
 185
 186         shardMBean.unregisterMBean();
 187     }
 188
 189     @Override
 190     public void onReceiveRecover(final Object message) throws Exception {
 191         if(LOG.isDebugEnabled()) {
 192             LOG.debug("{}: onReceiveRecover: Received message {} from {}", persistenceId(),
 193                 message.getClass().toString(), getSender());
 194         }
 195
 196         if (message instanceof RecoveryFailure){
 197             LOG.error("{}: Recovery failed because of this cause",
 198                     persistenceId(), ((RecoveryFailure) message).cause());
 199
 200             // Even though recovery failed, we still need to finish our recovery, eg send the
 201             // ActorInitialized message and start the txCommitTimeoutCheckSchedule.
 202             onRecoveryComplete();
 203         } else {
 204             super.onReceiveRecover(message);
 205             if(LOG.isTraceEnabled()) {
 206                 appendEntriesReplyTracker.begin();
 207             }
 208         }
 209     }
 210
 211     @Override
 212     public void onReceiveCommand(final Object message) throws Exception {
 213
 214         MessageTracker.Context context = appendEntriesReplyTracker.received(message);
 215
 216         if(context.error().isPresent()){
 217             LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(),
 218                 context.error());
 219         }
 220
 221         try {
 222             if (CreateTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 223                 handleCreateTransaction(message);
 224             } else if (BatchedModifications.class.isInstance(message)) {
 225                 handleBatchedModifications((BatchedModifications)message);
 226             } else if (message instanceof ForwardedReadyTransaction) {
 227                 commitCoordinator.handleForwardedReadyTransaction((ForwardedReadyTransaction) message,
 228                         getSender(), this);
 229             } else if (message instanceof ReadyLocalTransaction) {
 230                 handleReadyLocalTransaction((ReadyLocalTransaction)message);
 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(), hasLeader());
 241             } else if (message instanceof RegisterDataTreeChangeListener) {
 242                 treeChangeSupport.onMessage((RegisterDataTreeChangeListener) message, isLeader(), hasLeader());
 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 if(GET_SHARD_MBEAN_MESSAGE.equals(message)){
 259                 sender().tell(getShardMBean(), self());
 260             } else {
 261                 super.onReceiveCommand(message);
 262             }
 263         } finally {
 264             context.done();
 265         }
 266     }
 267
 268     private boolean hasLeader() {
 269         return getLeaderId() != null;
 270     }
 271
 272     public int getPendingTxCommitQueueSize() {
 273         return commitCoordinator.getQueueSize();
 274     }
 275
 276     @Override
 277     protected Optional<ActorRef> getRoleChangeNotifier() {
 278         return roleChangeNotifier;
 279     }
 280
 281     @Override
 282     protected LeaderStateChanged newLeaderStateChanged(String memberId, String leaderId, short leaderPayloadVersion) {
 283         return new ShardLeaderStateChanged(memberId, leaderId,
 284                 isLeader() ? Optional.<DataTree>of(store.getDataTree()) : Optional.<DataTree>absent(),
 285                 leaderPayloadVersion);
 286     }
 287
 288     protected void onDatastoreContext(DatastoreContext context) {
 289         datastoreContext = context;
 290
 291         commitCoordinator.setQueueCapacity(datastoreContext.getShardTransactionCommitQueueCapacity());
 292
 293         setTransactionCommitTimeout();
 294
 295         if(datastoreContext.isPersistent() && !persistence().isRecoveryApplicable()) {
 296             setPersistence(true);
 297         } else if(!datastoreContext.isPersistent() && persistence().isRecoveryApplicable()) {
 298             setPersistence(false);
 299         }
 300
 301         updateConfigParams(datastoreContext.getShardRaftConfig());
 302     }
 303
 304     private void handleTransactionCommitTimeoutCheck() {
 305         CohortEntry cohortEntry = commitCoordinator.getCurrentCohortEntry();
 306         if(cohortEntry != null) {
 307             if(cohortEntry.isExpired(transactionCommitTimeout)) {
 308                 LOG.warn("{}: Current transaction {} has timed out after {} ms - aborting",
 309                         persistenceId(), cohortEntry.getTransactionID(), transactionCommitTimeout);
 310
 311                 doAbortTransaction(cohortEntry.getTransactionID(), null);
 312             }
 313         }
 314
 315         commitCoordinator.cleanupExpiredCohortEntries();
 316     }
 317
 318     private static boolean isEmptyCommit(final DataTreeCandidate candidate) {
 319         return ModificationType.UNMODIFIED.equals(candidate.getRootNode().getModificationType());
 320     }
 321
 322     void continueCommit(final CohortEntry cohortEntry) throws Exception {
 323         final DataTreeCandidate candidate = cohortEntry.getCandidate();
 324
 325         // If we do not have any followers and we are not using persistence
 326         // or if cohortEntry has no modifications
 327         // we can apply modification to the state immediately
 328         if ((!hasFollowers() && !persistence().isRecoveryApplicable()) || isEmptyCommit(candidate)) {
 329             applyModificationToState(cohortEntry.getReplySender(), cohortEntry.getTransactionID(), candidate);
 330         } else {
 331             Shard.this.persistData(cohortEntry.getReplySender(), cohortEntry.getTransactionID(),
 332                 DataTreeCandidatePayload.create(candidate));
 333         }
 334     }
 335
 336     private void handleCommitTransaction(final CommitTransaction commit) {
 337         if(!commitCoordinator.handleCommit(commit.getTransactionID(), getSender(), this)) {
 338             shardMBean.incrementFailedTransactionsCount();
 339         }
 340     }
 341
 342     private void finishCommit(@Nonnull final ActorRef sender, @Nonnull final String transactionID, @Nonnull final CohortEntry cohortEntry) {
 343         LOG.debug("{}: Finishing commit for transaction {}", persistenceId(), cohortEntry.getTransactionID());
 344
 345         try {
 346             cohortEntry.commit();
 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.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(String errMessage, 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(errMessage, persistenceId())), getSelf());
 409     }
 410
 411     private void handleBatchedModifications(BatchedModifications batched) {
 412         // This message is sent to prepare the modifications transaction directly on the Shard as an
 413         // optimization to avoid the extra overhead of a separate ShardTransaction actor. On the last
 414         // BatchedModifications message, the caller sets the ready flag in the message indicating
 415         // modifications are complete. The reply contains the cohort actor path (this actor) for the caller
 416         // to initiate the 3-phase commit. This also avoids the overhead of sending an additional
 417         // ReadyTransaction message.
 418
 419         // If we're not the leader then forward to the leader. This is a safety measure - we shouldn't
 420         // normally get here if we're not the leader as the front-end (TransactionProxy) should determine
 421         // the primary/leader shard. However with timing and caching on the front-end, there's a small
 422         // window where it could have a stale leader during leadership transitions.
 423         //
 424         if(isLeader()) {
 425             failIfIsolatedLeader(getSender());
 426
 427             try {
 428                 commitCoordinator.handleBatchedModifications(batched, getSender(), this);
 429             } catch (Exception e) {
 430                 LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(),
 431                         batched.getTransactionID(), e);
 432                 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 433             }
 434         } else {
 435             ActorSelection leader = getLeader();
 436             if(leader != null) {
 437                 // TODO: what if this is not the first batch and leadership changed in between batched messages?
 438                 // We could check if the commitCoordinator already has a cached entry and forward all the previous
 439                 // batched modifications.
 440                 LOG.debug("{}: Forwarding BatchedModifications to leader {}", persistenceId(), leader);
 441                 leader.forward(batched, getContext());
 442             } else {
 443                 noLeaderError("Could not commit transaction " + batched.getTransactionID(), batched);
 444             }
 445         }
 446     }
 447
 448     private boolean failIfIsolatedLeader(ActorRef sender) {
 449         if(getRaftState() == RaftState.IsolatedLeader) {
 450             sender.tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format(
 451                     "Shard %s was the leader but has lost contact with all of its followers. Either all" +
 452                     " other follower nodes are down or this node is isolated by a network partition.",
 453                     persistenceId()))), getSelf());
 454             return true;
 455         }
 456
 457         return false;
 458     }
 459
 460     private void handleReadyLocalTransaction(final ReadyLocalTransaction message) {
 461         if (isLeader()) {
 462             failIfIsolatedLeader(getSender());
 463
 464             try {
 465                 commitCoordinator.handleReadyLocalTransaction(message, getSender(), this);
 466             } catch (Exception e) {
 467                 LOG.error("{}: Error handling ReadyLocalTransaction for Tx {}", persistenceId(),
 468                         message.getTransactionID(), e);
 469                 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 470             }
 471         } else {
 472             ActorSelection leader = getLeader();
 473             if (leader != null) {
 474                 LOG.debug("{}: Forwarding ReadyLocalTransaction to leader {}", persistenceId(), leader);
 475                 message.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
 476                 leader.forward(message, getContext());
 477             } else {
 478                 noLeaderError("Could not commit transaction " + message.getTransactionID(), message);
 479             }
 480         }
 481     }
 482
 483     private void handleAbortTransaction(final AbortTransaction abort) {
 484         doAbortTransaction(abort.getTransactionID(), getSender());
 485     }
 486
 487     void doAbortTransaction(final String transactionID, final ActorRef sender) {
 488         commitCoordinator.handleAbort(transactionID, sender, this);
 489     }
 490
 491     private void handleCreateTransaction(final Object message) {
 492         if (isLeader()) {
 493             createTransaction(CreateTransaction.fromSerializable(message));
 494         } else if (getLeader() != null) {
 495             getLeader().forward(message, getContext());
 496         } else {
 497             getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(
 498                     "Could not create a shard transaction", persistenceId())), getSelf());
 499         }
 500     }
 501
 502     private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
 503         store.closeTransactionChain(closeTransactionChain.getTransactionChainId());
 504     }
 505
 506     private ActorRef createTypedTransactionActor(int transactionType,
 507             ShardTransactionIdentifier transactionId, String transactionChainId,
 508             short clientVersion ) {
 509
 510         return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType),
 511                 transactionId, transactionChainId, clientVersion);
 512     }
 513
 514     private void createTransaction(CreateTransaction createTransaction) {
 515         try {
 516             if(TransactionType.fromInt(createTransaction.getTransactionType()) != TransactionType.READ_ONLY &&
 517                     failIfIsolatedLeader(getSender())) {
 518                 return;
 519             }
 520
 521             ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
 522                 createTransaction.getTransactionId(), createTransaction.getTransactionChainId(),
 523                 createTransaction.getVersion());
 524
 525             getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
 526                     createTransaction.getTransactionId()).toSerializable(), getSelf());
 527         } catch (Exception e) {
 528             getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 529         }
 530     }
 531
 532     private ActorRef createTransaction(int transactionType, String remoteTransactionId,
 533             String transactionChainId, short clientVersion) {
 534
 535
 536         ShardTransactionIdentifier transactionId = new ShardTransactionIdentifier(remoteTransactionId);
 537
 538         if(LOG.isDebugEnabled()) {
 539             LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
 540         }
 541
 542         ActorRef transactionActor = createTypedTransactionActor(transactionType, transactionId,
 543                 transactionChainId, clientVersion);
 544
 545         return transactionActor;
 546     }
 547
 548     private void commitWithNewTransaction(final Modification modification) {
 549         ReadWriteShardDataTreeTransaction tx = store.newReadWriteTransaction(modification.toString(), null);
 550         modification.apply(tx.getSnapshot());
 551         try {
 552             snapshotCohort.syncCommitTransaction(tx);
 553             shardMBean.incrementCommittedTransactionCount();
 554             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 555         } catch (Exception e) {
 556             shardMBean.incrementFailedTransactionsCount();
 557             LOG.error("{}: Failed to commit", persistenceId(), e);
 558         }
 559     }
 560
 561     private void updateSchemaContext(final UpdateSchemaContext message) {
 562         updateSchemaContext(message.getSchemaContext());
 563     }
 564
 565     @VisibleForTesting
 566     void updateSchemaContext(final SchemaContext schemaContext) {
 567         store.updateSchemaContext(schemaContext);
 568     }
 569
 570     private boolean isMetricsCaptureEnabled() {
 571         CommonConfig config = new CommonConfig(getContext().system().settings().config());
 572         return config.isMetricCaptureEnabled();
 573     }
 574
 575     @Override
 576     @VisibleForTesting
 577     public RaftActorSnapshotCohort getRaftActorSnapshotCohort() {
 578         return snapshotCohort;
 579     }
 580
 581     @Override
 582     @Nonnull
 583     protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
 584         return new ShardRecoveryCoordinator(store, store.getSchemaContext(), persistenceId(), LOG);
 585     }
 586
 587     @Override
 588     protected void onRecoveryComplete() {
 589         //notify shard manager
 590         getContext().parent().tell(new ActorInitialized(), getSelf());
 591
 592         // Being paranoid here - this method should only be called once but just in case...
 593         if(txCommitTimeoutCheckSchedule == null) {
 594             // Schedule a message to be periodically sent to check if the current in-progress
 595             // transaction should be expired and aborted.
 596             FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
 597             txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
 598                     period, period, getSelf(),
 599                     TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
 600         }
 601     }
 602
 603     @Override
 604     protected void applyState(final ActorRef clientActor, final String identifier, final Object data) {
 605         if (data instanceof DataTreeCandidatePayload) {
 606             if (clientActor == null) {
 607                 // No clientActor indicates a replica coming from the leader
 608                 try {
 609                     store.applyForeignCandidate(identifier, ((DataTreeCandidatePayload)data).getCandidate());
 610                 } catch (DataValidationFailedException | IOException e) {
 611                     LOG.error("{}: Error applying replica {}", persistenceId(), identifier, e);
 612                 }
 613             } else {
 614                 // Replication consensus reached, proceed to commit
 615                 finishCommit(clientActor, identifier);
 616             }
 617         } else if (data instanceof ModificationPayload) {
 618             try {
 619                 applyModificationToState(clientActor, identifier, ((ModificationPayload) data).getModification());
 620             } catch (ClassNotFoundException | IOException e) {
 621                 LOG.error("{}: Error extracting ModificationPayload", persistenceId(), e);
 622             }
 623         } else if (data instanceof CompositeModificationPayload) {
 624             Object modification = ((CompositeModificationPayload) data).getModification();
 625
 626             applyModificationToState(clientActor, identifier, modification);
 627         } else if(data instanceof CompositeModificationByteStringPayload ){
 628             Object modification = ((CompositeModificationByteStringPayload) data).getModification();
 629
 630             applyModificationToState(clientActor, identifier, modification);
 631         } else {
 632             LOG.error("{}: Unknown state received {} Class loader = {} CompositeNodeMod.ClassLoader = {}",
 633                     persistenceId(), data, data.getClass().getClassLoader(),
 634                     CompositeModificationPayload.class.getClassLoader());
 635         }
 636     }
 637
 638     private void applyModificationToState(ActorRef clientActor, String identifier, Object modification) {
 639         if(modification == null) {
 640             LOG.error(
 641                     "{}: modification is null - this is very unexpected, clientActor = {}, identifier = {}",
 642                     persistenceId(), identifier, clientActor != null ? clientActor.path().toString() : null);
 643         } else if(clientActor == null) {
 644             // There's no clientActor to which to send a commit reply so we must be applying
 645             // replicated state from the leader.
 646             commitWithNewTransaction(MutableCompositeModification.fromSerializable(modification));
 647         } else {
 648             // This must be the OK to commit after replication consensus.
 649             finishCommit(clientActor, identifier);
 650         }
 651     }
 652
 653     @Override
 654     protected void onStateChanged() {
 655         boolean isLeader = isLeader();
 656         boolean hasLeader = hasLeader();
 657         changeSupport.onLeadershipChange(isLeader, hasLeader);
 658         treeChangeSupport.onLeadershipChange(isLeader, hasLeader);
 659
 660         // If this actor is no longer the leader close all the transaction chains
 661         if (!isLeader) {
 662             if(LOG.isDebugEnabled()) {
 663                 LOG.debug(
 664                     "{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader",
 665                     persistenceId(), getId());
 666             }
 667
 668             store.closeAllTransactionChains();
 669         }
 670     }
 671
 672     @Override
 673     protected void onLeaderChanged(String oldLeader, String newLeader) {
 674         shardMBean.incrementLeadershipChangeCount();
 675     }
 676
 677     @Override
 678     public String persistenceId() {
 679         return this.name;
 680     }
 681
 682     @VisibleForTesting
 683     ShardCommitCoordinator getCommitCoordinator() {
 684         return commitCoordinator;
 685     }
 686
 687     protected abstract static class AbstractShardCreator implements Creator<Shard> {
 688         private static final long serialVersionUID = 1L;
 689
 690         protected final ShardIdentifier name;
 691         protected final Map<String, String> peerAddresses;
 692         protected final DatastoreContext datastoreContext;
 693         protected final SchemaContext schemaContext;
 694
 695         protected AbstractShardCreator(final ShardIdentifier name, final Map<String, String> peerAddresses,
 696                 final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 697             this.name = Preconditions.checkNotNull(name, "name should not be null");
 698             this.peerAddresses = Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
 699             this.datastoreContext = Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
 700             this.schemaContext = Preconditions.checkNotNull(schemaContext, "schemaContext should not be null");
 701         }
 702     }
 703
 704     private static class ShardCreator extends AbstractShardCreator {
 705         private static final long serialVersionUID = 1L;
 706
 707         ShardCreator(final ShardIdentifier name, final Map<String, String> peerAddresses,
 708                 final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 709             super(name, peerAddresses, datastoreContext, schemaContext);
 710         }
 711
 712         @Override
 713         public Shard create() throws Exception {
 714             return new Shard(name, peerAddresses, datastoreContext, schemaContext);
 715         }
 716     }
 717
 718     @VisibleForTesting
 719     public ShardDataTree getDataStore() {
 720         return store;
 721     }
 722
 723     @VisibleForTesting
 724     ShardStats getShardMBean() {
 725         return shardMBean;
 726     }
 727 }