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