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

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