git.opendaylight Code Review - controller.git/blob

   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.RaftState;
  66 import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus;
  67 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  68 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationByteStringPayload;
  69 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationPayload;
  70 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTree;
  71 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidate;
  72 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException;
  73 import org.opendaylight.yangtools.yang.data.api.schema.tree.ModificationType;
  74 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
  75 import scala.concurrent.duration.Duration;
  76 import scala.concurrent.duration.FiniteDuration;
  77
  78 /**
  79  * A Shard represents a portion of the logical data tree <br/>
  80  * <p>
  81  * Our Shard uses InMemoryDataTree as it's internal representation and delegates all requests it
  82  * </p>
  83  */
  84 public class Shard extends RaftActor {
  85
  86     private static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = "txCommitTimeoutCheck";
  87
  88     @VisibleForTesting
  89     static final Object GET_SHARD_MBEAN_MESSAGE = "getShardMBeanMessage";
  90
  91     @VisibleForTesting
  92     static final String DEFAULT_NAME = "default";
  93
  94     // The state of this Shard
  95     private final ShardDataTree store;
  96
  97     /// The name of this shard
  98     private final String name;
  99
 100     private final ShardStats shardMBean;
 101
 102     private DatastoreContext datastoreContext;
 103
 104     private final ShardCommitCoordinator commitCoordinator;
 105
 106     private long transactionCommitTimeout;
 107
 108     private Cancellable txCommitTimeoutCheckSchedule;
 109
 110     private final Optional<ActorRef> roleChangeNotifier;
 111
 112     private final MessageTracker appendEntriesReplyTracker;
 113
 114     private final ShardTransactionActorFactory transactionActorFactory;
 115
 116     private final ShardSnapshotCohort snapshotCohort;
 117
 118     private final DataTreeChangeListenerSupport treeChangeSupport = new DataTreeChangeListenerSupport(this);
 119     private final DataChangeListenerSupport changeSupport = new DataChangeListenerSupport(this);
 120
 121     protected Shard(final ShardIdentifier name, final Map<String, String> peerAddresses,
 122             final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 123         super(name.toString(), new HashMap<>(peerAddresses), Optional.of(datastoreContext.getShardRaftConfig()),
 124                 DataStoreVersions.CURRENT_VERSION);
 125
 126         this.name = name.toString();
 127         this.datastoreContext = datastoreContext;
 128
 129         setPersistence(datastoreContext.isPersistent());
 130
 131         LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
 132
 133         store = new ShardDataTree(schemaContext);
 134
 135         shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
 136                 datastoreContext.getDataStoreMXBeanType());
 137         shardMBean.setShard(this);
 138
 139         if (isMetricsCaptureEnabled()) {
 140             getContext().become(new MeteringBehavior(this));
 141         }
 142
 143         commitCoordinator = new ShardCommitCoordinator(store,
 144                 datastoreContext.getShardCommitQueueExpiryTimeoutInMillis(),
 145                 datastoreContext.getShardTransactionCommitQueueCapacity(), self(), LOG, this.name);
 146
 147         setTransactionCommitTimeout();
 148
 149         // create a notifier actor for each cluster member
 150         roleChangeNotifier = createRoleChangeNotifier(name.toString());
 151
 152         appendEntriesReplyTracker = new MessageTracker(AppendEntriesReply.class,
 153                 getRaftActorContext().getConfigParams().getIsolatedCheckIntervalInMillis());
 154
 155         transactionActorFactory = new ShardTransactionActorFactory(store, datastoreContext,
 156                 new Dispatchers(context().system().dispatchers()).getDispatcherPath(
 157                         Dispatchers.DispatcherType.Transaction), self(), getContext(), shardMBean);
 158
 159         snapshotCohort = new ShardSnapshotCohort(transactionActorFactory, store, LOG, this.name);
 160
 161
 162     }
 163
 164     private void setTransactionCommitTimeout() {
 165         transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
 166                 datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS) / 2;
 167     }
 168
 169     public static Props props(final ShardIdentifier name,
 170         final Map<String, String> peerAddresses,
 171         final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 172         Preconditions.checkNotNull(name, "name should not be null");
 173         Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
 174         Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
 175         Preconditions.checkNotNull(schemaContext, "schemaContext should not be null");
 176
 177         return Props.create(new ShardCreator(name, peerAddresses, datastoreContext, schemaContext));
 178     }
 179
 180     private Optional<ActorRef> createRoleChangeNotifier(String shardId) {
 181         ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
 182             RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
 183         return Optional.of(shardRoleChangeNotifier);
 184     }
 185
 186     @Override
 187     public void postStop() {
 188         LOG.info("Stopping Shard {}", persistenceId());
 189
 190         super.postStop();
 191
 192         if(txCommitTimeoutCheckSchedule != null) {
 193             txCommitTimeoutCheckSchedule.cancel();
 194         }
 195
 196         shardMBean.unregisterMBean();
 197     }
 198
 199     @Override
 200     public void onReceiveRecover(final Object message) throws Exception {
 201         if(LOG.isDebugEnabled()) {
 202             LOG.debug("{}: onReceiveRecover: Received message {} from {}", persistenceId(),
 203                 message.getClass().toString(), getSender());
 204         }
 205
 206         if (message instanceof RecoveryFailure){
 207             LOG.error("{}: Recovery failed because of this cause",
 208                     persistenceId(), ((RecoveryFailure) message).cause());
 209
 210             // Even though recovery failed, we still need to finish our recovery, eg send the
 211             // ActorInitialized message and start the txCommitTimeoutCheckSchedule.
 212             onRecoveryComplete();
 213         } else {
 214             super.onReceiveRecover(message);
 215             if(LOG.isTraceEnabled()) {
 216                 appendEntriesReplyTracker.begin();
 217             }
 218         }
 219     }
 220
 221     @Override
 222     public void onReceiveCommand(final Object message) throws Exception {
 223
 224         MessageTracker.Context context = appendEntriesReplyTracker.received(message);
 225
 226         if(context.error().isPresent()){
 227             LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(),
 228                     context.error());
 229         }
 230
 231         try {
 232             if (CreateTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 233                 handleCreateTransaction(message);
 234             } else if (BatchedModifications.class.isInstance(message)) {
 235                 handleBatchedModifications((BatchedModifications)message);
 236             } else if (message instanceof ForwardedReadyTransaction) {
 237                 commitCoordinator.handleForwardedReadyTransaction((ForwardedReadyTransaction) message,
 238                         getSender(), this);
 239             } else if (message instanceof ReadyLocalTransaction) {
 240                 handleReadyLocalTransaction((ReadyLocalTransaction)message);
 241             } else if (CanCommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 242                 handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
 243             } else if (CommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 244                 handleCommitTransaction(CommitTransaction.fromSerializable(message));
 245             } else if (AbortTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 246                 handleAbortTransaction(AbortTransaction.fromSerializable(message));
 247             } else if (CloseTransactionChain.SERIALIZABLE_CLASS.isInstance(message)) {
 248                 closeTransactionChain(CloseTransactionChain.fromSerializable(message));
 249             } else if (message instanceof RegisterChangeListener) {
 250                 changeSupport.onMessage((RegisterChangeListener) message, isLeader());
 251             } else if (message instanceof RegisterDataTreeChangeListener) {
 252                 treeChangeSupport.onMessage((RegisterDataTreeChangeListener) message, isLeader());
 253             } else if (message instanceof UpdateSchemaContext) {
 254                 updateSchemaContext((UpdateSchemaContext) message);
 255             } else if (message instanceof PeerAddressResolved) {
 256                 PeerAddressResolved resolved = (PeerAddressResolved) message;
 257                 setPeerAddress(resolved.getPeerId().toString(),
 258                         resolved.getPeerAddress());
 259             } else if (message.equals(TX_COMMIT_TIMEOUT_CHECK_MESSAGE)) {
 260                 handleTransactionCommitTimeoutCheck();
 261             } else if(message instanceof DatastoreContext) {
 262                 onDatastoreContext((DatastoreContext)message);
 263             } else if(message instanceof RegisterRoleChangeListener){
 264                 roleChangeNotifier.get().forward(message, context());
 265             } else if (message instanceof FollowerInitialSyncUpStatus) {
 266                 shardMBean.setFollowerInitialSyncStatus(((FollowerInitialSyncUpStatus) message).isInitialSyncDone());
 267                 context().parent().tell(message, self());
 268             } else if(GET_SHARD_MBEAN_MESSAGE.equals(message)){
 269                 sender().tell(getShardMBean(), self());
 270             } else {
 271                 super.onReceiveCommand(message);
 272             }
 273         } finally {
 274             context.done();
 275         }
 276     }
 277
 278     public int getPendingTxCommitQueueSize() {
 279         return commitCoordinator.getQueueSize();
 280     }
 281
 282     @Override
 283     protected Optional<ActorRef> getRoleChangeNotifier() {
 284         return roleChangeNotifier;
 285     }
 286
 287     @Override
 288     protected LeaderStateChanged newLeaderStateChanged(String memberId, String leaderId, short leaderPayloadVersion) {
 289         return new ShardLeaderStateChanged(memberId, leaderId,
 290                 isLeader() ? Optional.<DataTree>of(store.getDataTree()) : Optional.<DataTree>absent(),
 291                 leaderPayloadVersion);
 292     }
 293
 294     private void onDatastoreContext(DatastoreContext context) {
 295         datastoreContext = context;
 296
 297         commitCoordinator.setQueueCapacity(datastoreContext.getShardTransactionCommitQueueCapacity());
 298
 299         setTransactionCommitTimeout();
 300
 301         if(datastoreContext.isPersistent() && !persistence().isRecoveryApplicable()) {
 302             setPersistence(true);
 303         } else if(!datastoreContext.isPersistent() && persistence().isRecoveryApplicable()) {
 304             setPersistence(false);
 305         }
 306
 307         updateConfigParams(datastoreContext.getShardRaftConfig());
 308     }
 309
 310     private void handleTransactionCommitTimeoutCheck() {
 311         CohortEntry cohortEntry = commitCoordinator.getCurrentCohortEntry();
 312         if(cohortEntry != null) {
 313             if(cohortEntry.isExpired(transactionCommitTimeout)) {
 314                 LOG.warn("{}: Current transaction {} has timed out after {} ms - aborting",
 315                         persistenceId(), cohortEntry.getTransactionID(), transactionCommitTimeout);
 316
 317                 doAbortTransaction(cohortEntry.getTransactionID(), null);
 318             }
 319         }
 320
 321         commitCoordinator.cleanupExpiredCohortEntries();
 322     }
 323
 324     private static boolean isEmptyCommit(final DataTreeCandidate candidate) {
 325         return ModificationType.UNMODIFIED.equals(candidate.getRootNode().getModificationType());
 326     }
 327
 328     void continueCommit(final CohortEntry cohortEntry) throws Exception {
 329         final DataTreeCandidate candidate = cohortEntry.getCohort().getCandidate();
 330
 331         // If we do not have any followers and we are not using persistence
 332         // or if cohortEntry has no modifications
 333         // we can apply modification to the state immediately
 334         if ((!hasFollowers() && !persistence().isRecoveryApplicable()) || isEmptyCommit(candidate)) {
 335             applyModificationToState(cohortEntry.getReplySender(), cohortEntry.getTransactionID(), candidate);
 336         } else {
 337             Shard.this.persistData(cohortEntry.getReplySender(), cohortEntry.getTransactionID(),
 338                 DataTreeCandidatePayload.create(candidate));
 339         }
 340     }
 341
 342     private void handleCommitTransaction(final CommitTransaction commit) {
 343         if(!commitCoordinator.handleCommit(commit.getTransactionID(), getSender(), this)) {
 344             shardMBean.incrementFailedTransactionsCount();
 345         }
 346     }
 347
 348     private void finishCommit(@Nonnull final ActorRef sender, @Nonnull final String transactionID, @Nonnull final CohortEntry cohortEntry) {
 349         LOG.debug("{}: Finishing commit for transaction {}", persistenceId(), cohortEntry.getTransactionID());
 350
 351         try {
 352             // We block on the future here so we don't have to worry about possibly accessing our
 353             // state on a different thread outside of our dispatcher. Also, the data store
 354             // currently uses a same thread executor anyway.
 355             cohortEntry.getCohort().commit().get();
 356
 357             sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
 358
 359             shardMBean.incrementCommittedTransactionCount();
 360             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 361
 362         } catch (Exception e) {
 363             sender.tell(new akka.actor.Status.Failure(e), getSelf());
 364
 365             LOG.error("{}, An exception occurred while committing transaction {}", persistenceId(),
 366                     transactionID, e);
 367             shardMBean.incrementFailedTransactionsCount();
 368         } finally {
 369             commitCoordinator.currentTransactionComplete(transactionID, true);
 370         }
 371     }
 372
 373     private void finishCommit(@Nonnull final ActorRef sender, final @Nonnull String transactionID) {
 374         // With persistence enabled, this method is called via applyState by the leader strategy
 375         // after the commit has been replicated to a majority of the followers.
 376
 377         CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
 378         if (cohortEntry == null) {
 379             // The transaction is no longer the current commit. This can happen if the transaction
 380             // was aborted prior, most likely due to timeout in the front-end. We need to finish
 381             // committing the transaction though since it was successfully persisted and replicated
 382             // however we can't use the original cohort b/c it was already preCommitted and may
 383             // conflict with the current commit or may have been aborted so we commit with a new
 384             // transaction.
 385             cohortEntry = commitCoordinator.getAndRemoveCohortEntry(transactionID);
 386             if(cohortEntry != null) {
 387                 try {
 388                     store.applyForeignCandidate(transactionID, cohortEntry.getCohort().getCandidate());
 389                 } catch (DataValidationFailedException e) {
 390                     shardMBean.incrementFailedTransactionsCount();
 391                     LOG.error("{}: Failed to re-apply transaction {}", persistenceId(), transactionID, e);
 392                 }
 393
 394                 sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
 395             } else {
 396                 // This really shouldn't happen - it likely means that persistence or replication
 397                 // took so long to complete such that the cohort entry was expired from the cache.
 398                 IllegalStateException ex = new IllegalStateException(
 399                         String.format("%s: Could not finish committing transaction %s - no CohortEntry found",
 400                                 persistenceId(), transactionID));
 401                 LOG.error(ex.getMessage());
 402                 sender.tell(new akka.actor.Status.Failure(ex), getSelf());
 403             }
 404         } else {
 405             finishCommit(sender, transactionID, cohortEntry);
 406         }
 407     }
 408
 409     private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
 410         LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionID());
 411         commitCoordinator.handleCanCommit(canCommit.getTransactionID(), getSender(), this);
 412     }
 413
 414     private void noLeaderError(String errMessage, Object message) {
 415         // TODO: rather than throwing an immediate exception, we could schedule a timer to try again to make
 416         // it more resilient in case we're in the process of electing a new leader.
 417         getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(errMessage, persistenceId())), getSelf());
 418     }
 419
 420     private void handleBatchedModifications(BatchedModifications batched) {
 421         // This message is sent to prepare the modifications transaction directly on the Shard as an
 422         // optimization to avoid the extra overhead of a separate ShardTransaction actor. On the last
 423         // BatchedModifications message, the caller sets the ready flag in the message indicating
 424         // modifications are complete. The reply contains the cohort actor path (this actor) for the caller
 425         // to initiate the 3-phase commit. This also avoids the overhead of sending an additional
 426         // ReadyTransaction message.
 427
 428         // If we're not the leader then forward to the leader. This is a safety measure - we shouldn't
 429         // normally get here if we're not the leader as the front-end (TransactionProxy) should determine
 430         // the primary/leader shard. However with timing and caching on the front-end, there's a small
 431         // window where it could have a stale leader during leadership transitions.
 432         //
 433         if(isLeader()) {
 434             failIfIsolatedLeader(getSender());
 435
 436             try {
 437                 commitCoordinator.handleBatchedModifications(batched, getSender(), this);
 438             } catch (Exception e) {
 439                 LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(),
 440                         batched.getTransactionID(), e);
 441                 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 442             }
 443         } else {
 444             ActorSelection leader = getLeader();
 445             if(leader != null) {
 446                 // TODO: what if this is not the first batch and leadership changed in between batched messages?
 447                 // We could check if the commitCoordinator already has a cached entry and forward all the previous
 448                 // batched modifications.
 449                 LOG.debug("{}: Forwarding BatchedModifications to leader {}", persistenceId(), leader);
 450                 leader.forward(batched, getContext());
 451             } else {
 452                 noLeaderError("Could not commit transaction " + batched.getTransactionID(), batched);
 453             }
 454         }
 455     }
 456
 457     private boolean failIfIsolatedLeader(ActorRef sender) {
 458         if(getRaftState() == RaftState.IsolatedLeader) {
 459             sender.tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format(
 460                     "Shard %s was the leader but has lost contact with all of its followers. Either all" +
 461                     " other follower nodes are down or this node is isolated by a network partition.",
 462                     persistenceId()))), getSelf());
 463             return true;
 464         }
 465
 466         return false;
 467     }
 468
 469     private void handleReadyLocalTransaction(final ReadyLocalTransaction message) {
 470         if (isLeader()) {
 471             failIfIsolatedLeader(getSender());
 472
 473             try {
 474                 commitCoordinator.handleReadyLocalTransaction(message, getSender(), this);
 475             } catch (Exception e) {
 476                 LOG.error("{}: Error handling ReadyLocalTransaction for Tx {}", persistenceId(),
 477                         message.getTransactionID(), e);
 478                 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 479             }
 480         } else {
 481             ActorSelection leader = getLeader();
 482             if (leader != null) {
 483                 LOG.debug("{}: Forwarding ReadyLocalTransaction to leader {}", persistenceId(), leader);
 484                 message.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
 485                 leader.forward(message, getContext());
 486             } else {
 487                 noLeaderError("Could not commit transaction " + message.getTransactionID(), message);
 488             }
 489         }
 490     }
 491
 492     private void handleAbortTransaction(final AbortTransaction abort) {
 493         doAbortTransaction(abort.getTransactionID(), getSender());
 494     }
 495
 496     void doAbortTransaction(final String transactionID, final ActorRef sender) {
 497         final CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
 498         if(cohortEntry != null) {
 499             LOG.debug("{}: Aborting transaction {}", persistenceId(), transactionID);
 500
 501             // We don't remove the cached cohort entry here (ie pass false) in case the Tx was
 502             // aborted during replication in which case we may still commit locally if replication
 503             // succeeds.
 504             commitCoordinator.currentTransactionComplete(transactionID, false);
 505
 506             final ListenableFuture<Void> future = cohortEntry.getCohort().abort();
 507             final ActorRef self = getSelf();
 508
 509             Futures.addCallback(future, new FutureCallback<Void>() {
 510                 @Override
 511                 public void onSuccess(final Void v) {
 512                     shardMBean.incrementAbortTransactionsCount();
 513
 514                     if(sender != null) {
 515                         sender.tell(AbortTransactionReply.INSTANCE.toSerializable(), self);
 516                     }
 517                 }
 518
 519                 @Override
 520                 public void onFailure(final Throwable t) {
 521                     LOG.error("{}: An exception happened during abort", persistenceId(), t);
 522
 523                     if(sender != null) {
 524                         sender.tell(new akka.actor.Status.Failure(t), self);
 525                     }
 526                 }
 527             });
 528         }
 529     }
 530
 531     private void handleCreateTransaction(final Object message) {
 532         if (isLeader()) {
 533             createTransaction(CreateTransaction.fromSerializable(message));
 534         } else if (getLeader() != null) {
 535             getLeader().forward(message, getContext());
 536         } else {
 537             getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(
 538                     "Could not create a shard transaction", persistenceId())), getSelf());
 539         }
 540     }
 541
 542     private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
 543         store.closeTransactionChain(closeTransactionChain.getTransactionChainId());
 544     }
 545
 546     private ActorRef createTypedTransactionActor(int transactionType,
 547             ShardTransactionIdentifier transactionId, String transactionChainId,
 548             short clientVersion ) {
 549
 550         return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType),
 551                 transactionId, transactionChainId, clientVersion);
 552     }
 553
 554     private void createTransaction(CreateTransaction createTransaction) {
 555         try {
 556             if(TransactionType.fromInt(createTransaction.getTransactionType()) != TransactionType.READ_ONLY &&
 557                     failIfIsolatedLeader(getSender())) {
 558                 return;
 559             }
 560
 561             ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
 562                 createTransaction.getTransactionId(), createTransaction.getTransactionChainId(),
 563                 createTransaction.getVersion());
 564
 565             getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
 566                     createTransaction.getTransactionId()).toSerializable(), getSelf());
 567         } catch (Exception e) {
 568             getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 569         }
 570     }
 571
 572     private ActorRef createTransaction(int transactionType, String remoteTransactionId,
 573             String transactionChainId, short clientVersion) {
 574
 575
 576         ShardTransactionIdentifier transactionId = new ShardTransactionIdentifier(remoteTransactionId);
 577
 578         if(LOG.isDebugEnabled()) {
 579             LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
 580         }
 581
 582         ActorRef transactionActor = createTypedTransactionActor(transactionType, transactionId,
 583                 transactionChainId, clientVersion);
 584
 585         return transactionActor;
 586     }
 587
 588     private void commitWithNewTransaction(final Modification modification) {
 589         ReadWriteShardDataTreeTransaction tx = store.newReadWriteTransaction(modification.toString(), null);
 590         modification.apply(tx.getSnapshot());
 591         try {
 592             snapshotCohort.syncCommitTransaction(tx);
 593             shardMBean.incrementCommittedTransactionCount();
 594             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 595         } catch (Exception e) {
 596             shardMBean.incrementFailedTransactionsCount();
 597             LOG.error("{}: Failed to commit", persistenceId(), e);
 598         }
 599     }
 600
 601     private void updateSchemaContext(final UpdateSchemaContext message) {
 602         updateSchemaContext(message.getSchemaContext());
 603     }
 604
 605     @VisibleForTesting
 606     void updateSchemaContext(final SchemaContext schemaContext) {
 607         store.updateSchemaContext(schemaContext);
 608     }
 609
 610     private boolean isMetricsCaptureEnabled() {
 611         CommonConfig config = new CommonConfig(getContext().system().settings().config());
 612         return config.isMetricCaptureEnabled();
 613     }
 614
 615     @Override
 616     @VisibleForTesting
 617     public RaftActorSnapshotCohort getRaftActorSnapshotCohort() {
 618         return snapshotCohort;
 619     }
 620
 621     @Override
 622     @Nonnull
 623     protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
 624         return new ShardRecoveryCoordinator(store, store.getSchemaContext(), persistenceId(), LOG);
 625     }
 626
 627     @Override
 628     protected void onRecoveryComplete() {
 629         //notify shard manager
 630         getContext().parent().tell(new ActorInitialized(), getSelf());
 631
 632         // Being paranoid here - this method should only be called once but just in case...
 633         if(txCommitTimeoutCheckSchedule == null) {
 634             // Schedule a message to be periodically sent to check if the current in-progress
 635             // transaction should be expired and aborted.
 636             FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
 637             txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
 638                     period, period, getSelf(),
 639                     TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
 640         }
 641     }
 642
 643     @Override
 644     protected void applyState(final ActorRef clientActor, final String identifier, final Object data) {
 645         if (data instanceof DataTreeCandidatePayload) {
 646             if (clientActor == null) {
 647                 // No clientActor indicates a replica coming from the leader
 648                 try {
 649                     store.applyForeignCandidate(identifier, ((DataTreeCandidatePayload)data).getCandidate());
 650                 } catch (DataValidationFailedException | IOException e) {
 651                     LOG.error("{}: Error applying replica {}", persistenceId(), identifier, e);
 652                 }
 653             } else {
 654                 // Replication consensus reached, proceed to commit
 655                 finishCommit(clientActor, identifier);
 656             }
 657         } else if (data instanceof ModificationPayload) {
 658             try {
 659                 applyModificationToState(clientActor, identifier, ((ModificationPayload) data).getModification());
 660             } catch (ClassNotFoundException | IOException e) {
 661                 LOG.error("{}: Error extracting ModificationPayload", persistenceId(), e);
 662             }
 663         } else if (data instanceof CompositeModificationPayload) {
 664             Object modification = ((CompositeModificationPayload) data).getModification();
 665
 666             applyModificationToState(clientActor, identifier, modification);
 667         } else if(data instanceof CompositeModificationByteStringPayload ){
 668             Object modification = ((CompositeModificationByteStringPayload) data).getModification();
 669
 670             applyModificationToState(clientActor, identifier, modification);
 671         } else {
 672             LOG.error("{}: Unknown state received {} Class loader = {} CompositeNodeMod.ClassLoader = {}",
 673                     persistenceId(), data, data.getClass().getClassLoader(),
 674                     CompositeModificationPayload.class.getClassLoader());
 675         }
 676     }
 677
 678     private void applyModificationToState(ActorRef clientActor, String identifier, Object modification) {
 679         if(modification == null) {
 680             LOG.error(
 681                     "{}: modification is null - this is very unexpected, clientActor = {}, identifier = {}",
 682                     persistenceId(), identifier, clientActor != null ? clientActor.path().toString() : null);
 683         } else if(clientActor == null) {
 684             // There's no clientActor to which to send a commit reply so we must be applying
 685             // replicated state from the leader.
 686             commitWithNewTransaction(MutableCompositeModification.fromSerializable(modification));
 687         } else {
 688             // This must be the OK to commit after replication consensus.
 689             finishCommit(clientActor, identifier);
 690         }
 691     }
 692
 693     @Override
 694     protected void onStateChanged() {
 695         boolean isLeader = isLeader();
 696         changeSupport.onLeadershipChange(isLeader);
 697         treeChangeSupport.onLeadershipChange(isLeader);
 698
 699         // If this actor is no longer the leader close all the transaction chains
 700         if (!isLeader) {
 701             if(LOG.isDebugEnabled()) {
 702                 LOG.debug(
 703                     "{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader",
 704                     persistenceId(), getId());
 705             }
 706
 707             store.closeAllTransactionChains();
 708         }
 709     }
 710
 711     @Override
 712     protected void onLeaderChanged(String oldLeader, String newLeader) {
 713         shardMBean.incrementLeadershipChangeCount();
 714     }
 715
 716     @Override
 717     public String persistenceId() {
 718         return this.name;
 719     }
 720
 721     @VisibleForTesting
 722     ShardCommitCoordinator getCommitCoordinator() {
 723         return commitCoordinator;
 724     }
 725
 726
 727     private static class ShardCreator implements Creator<Shard> {
 728
 729         private static final long serialVersionUID = 1L;
 730
 731         final ShardIdentifier name;
 732         final Map<String, String> peerAddresses;
 733         final DatastoreContext datastoreContext;
 734         final SchemaContext schemaContext;
 735
 736         ShardCreator(final ShardIdentifier name, final Map<String, String> peerAddresses,
 737                 final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 738             this.name = name;
 739             this.peerAddresses = peerAddresses;
 740             this.datastoreContext = datastoreContext;
 741             this.schemaContext = schemaContext;
 742         }
 743
 744         @Override
 745         public Shard create() throws Exception {
 746             return new Shard(name, peerAddresses, datastoreContext, schemaContext);
 747         }
 748     }
 749
 750     @VisibleForTesting
 751     public ShardDataTree getDataStore() {
 752         return store;
 753     }
 754
 755     @VisibleForTesting
 756     ShardStats getShardMBean() {
 757         return shardMBean;
 758     }
 759 }