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