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.serialization.Serialization;
  16 import com.google.common.annotations.VisibleForTesting;
  17 import com.google.common.base.Optional;
  18 import com.google.common.base.Preconditions;
  19 import java.io.IOException;
  20 import java.util.Collections;
  21 import java.util.Map;
  22 import java.util.concurrent.TimeUnit;
  23 import javax.annotation.Nonnull;
  24 import org.opendaylight.controller.cluster.common.actor.CommonConfig;
  25 import org.opendaylight.controller.cluster.common.actor.MeteringBehavior;
  26 import org.opendaylight.controller.cluster.datastore.ShardCommitCoordinator.CohortEntry;
  27 import org.opendaylight.controller.cluster.datastore.exceptions.NoShardLeaderException;
  28 import org.opendaylight.controller.cluster.datastore.identifiers.ShardIdentifier;
  29 import org.opendaylight.controller.cluster.datastore.identifiers.ShardTransactionIdentifier;
  30 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardMBeanFactory;
  31 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardStats;
  32 import org.opendaylight.controller.cluster.datastore.messages.AbortTransaction;
  33 import org.opendaylight.controller.cluster.datastore.messages.ActorInitialized;
  34 import org.opendaylight.controller.cluster.datastore.messages.BatchedModifications;
  35 import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransaction;
  36 import org.opendaylight.controller.cluster.datastore.messages.CloseTransactionChain;
  37 import org.opendaylight.controller.cluster.datastore.messages.CommitTransaction;
  38 import org.opendaylight.controller.cluster.datastore.messages.CommitTransactionReply;
  39 import org.opendaylight.controller.cluster.datastore.messages.CreateTransaction;
  40 import org.opendaylight.controller.cluster.datastore.messages.CreateTransactionReply;
  41 import org.opendaylight.controller.cluster.datastore.messages.DatastoreSnapshot;
  42 import org.opendaylight.controller.cluster.datastore.messages.DatastoreSnapshot.ShardSnapshot;
  43 import org.opendaylight.controller.cluster.datastore.messages.ForwardedReadyTransaction;
  44 import org.opendaylight.controller.cluster.datastore.messages.GetShardDataTree;
  45 import org.opendaylight.controller.cluster.datastore.messages.PeerAddressResolved;
  46 import org.opendaylight.controller.cluster.datastore.messages.ReadyLocalTransaction;
  47 import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListener;
  48 import org.opendaylight.controller.cluster.datastore.messages.RegisterDataTreeChangeListener;
  49 import org.opendaylight.controller.cluster.datastore.messages.ShardLeaderStateChanged;
  50 import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext;
  51 import org.opendaylight.controller.cluster.datastore.modification.Modification;
  52 import org.opendaylight.controller.cluster.datastore.modification.MutableCompositeModification;
  53 import org.opendaylight.controller.cluster.datastore.utils.Dispatchers;
  54 import org.opendaylight.controller.cluster.datastore.utils.MessageTracker;
  55 import org.opendaylight.controller.cluster.notifications.LeaderStateChanged;
  56 import org.opendaylight.controller.cluster.notifications.RegisterRoleChangeListener;
  57 import org.opendaylight.controller.cluster.notifications.RoleChangeNotifier;
  58 import org.opendaylight.controller.cluster.raft.RaftActor;
  59 import org.opendaylight.controller.cluster.raft.RaftActorRecoveryCohort;
  60 import org.opendaylight.controller.cluster.raft.RaftActorSnapshotCohort;
  61 import org.opendaylight.controller.cluster.raft.RaftState;
  62 import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus;
  63 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  64 import org.opendaylight.controller.cluster.raft.messages.ServerRemoved;
  65 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationByteStringPayload;
  66 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationPayload;
  67 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTree;
  68 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidate;
  69 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException;
  70 import org.opendaylight.yangtools.yang.data.api.schema.tree.ModificationType;
  71 import org.opendaylight.yangtools.yang.data.api.schema.tree.TreeType;
  72 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
  73 import scala.concurrent.duration.Duration;
  74 import scala.concurrent.duration.FiniteDuration;
  75
  76 /**
  77  * A Shard represents a portion of the logical data tree <br/>
  78  * <p>
  79  * Our Shard uses InMemoryDataTree as it's internal representation and delegates all requests it
  80  * </p>
  81  */
  82 public class Shard extends RaftActor {
  83
  84     protected static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = "txCommitTimeoutCheck";
  85
  86     @VisibleForTesting
  87     static final Object GET_SHARD_MBEAN_MESSAGE = "getShardMBeanMessage";
  88
  89     @VisibleForTesting
  90     static final String DEFAULT_NAME = "default";
  91
  92     // The state of this Shard
  93     private final ShardDataTree store;
  94
  95     /// The name of this shard
  96     private final String name;
  97
  98     private final ShardStats shardMBean;
  99
 100     private DatastoreContext datastoreContext;
 101
 102     private final ShardCommitCoordinator commitCoordinator;
 103
 104     private long transactionCommitTimeout;
 105
 106     private Cancellable txCommitTimeoutCheckSchedule;
 107
 108     private final Optional<ActorRef> roleChangeNotifier;
 109
 110     private final MessageTracker appendEntriesReplyTracker;
 111
 112     private final ShardTransactionActorFactory transactionActorFactory;
 113
 114     private final ShardSnapshotCohort snapshotCohort;
 115
 116     private final DataTreeChangeListenerSupport treeChangeSupport = new DataTreeChangeListenerSupport(this);
 117     private final DataChangeListenerSupport changeSupport = new DataChangeListenerSupport(this);
 118
 119
 120     private ShardSnapshot restoreFromSnapshot;
 121
 122     private final ShardTransactionMessageRetrySupport messageRetrySupport;
 123
 124     protected Shard(AbstractBuilder<?, ?> builder) {
 125         super(builder.getId().toString(), builder.getPeerAddresses(),
 126                 Optional.of(builder.getDatastoreContext().getShardRaftConfig()), DataStoreVersions.CURRENT_VERSION);
 127
 128         this.name = builder.getId().toString();
 129         this.datastoreContext = builder.getDatastoreContext();
 130         this.restoreFromSnapshot = builder.getRestoreFromSnapshot();
 131
 132         setPersistence(datastoreContext.isPersistent());
 133
 134         LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
 135
 136         store = new ShardDataTree(builder.getSchemaContext(), builder.getTreeType());
 137
 138         shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
 139                 datastoreContext.getDataStoreMXBeanType());
 140         shardMBean.setShard(this);
 141
 142         if (isMetricsCaptureEnabled()) {
 143             getContext().become(new MeteringBehavior(this));
 144         }
 145
 146         commitCoordinator = new ShardCommitCoordinator(store,
 147                 datastoreContext.getShardCommitQueueExpiryTimeoutInMillis(),
 148                 datastoreContext.getShardTransactionCommitQueueCapacity(), LOG, this.name);
 149
 150         setTransactionCommitTimeout();
 151
 152         // create a notifier actor for each cluster member
 153         roleChangeNotifier = createRoleChangeNotifier(name.toString());
 154
 155         appendEntriesReplyTracker = new MessageTracker(AppendEntriesReply.class,
 156                 getRaftActorContext().getConfigParams().getIsolatedCheckIntervalInMillis());
 157
 158         transactionActorFactory = new ShardTransactionActorFactory(store, datastoreContext,
 159                 new Dispatchers(context().system().dispatchers()).getDispatcherPath(
 160                         Dispatchers.DispatcherType.Transaction), self(), getContext(), shardMBean);
 161
 162         snapshotCohort = new ShardSnapshotCohort(transactionActorFactory, store, LOG, this.name);
 163
 164         messageRetrySupport = new ShardTransactionMessageRetrySupport(this);
 165     }
 166
 167     private void setTransactionCommitTimeout() {
 168         transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
 169                 datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS) / 2;
 170     }
 171
 172     private Optional<ActorRef> createRoleChangeNotifier(String shardId) {
 173         ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
 174             RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
 175         return Optional.of(shardRoleChangeNotifier);
 176     }
 177
 178     @Override
 179     public void postStop() {
 180         LOG.info("Stopping Shard {}", persistenceId());
 181
 182         super.postStop();
 183
 184         messageRetrySupport.close();
 185
 186         if(txCommitTimeoutCheckSchedule != null) {
 187             txCommitTimeoutCheckSchedule.cancel();
 188         }
 189
 190         commitCoordinator.abortPendingTransactions("Transaction aborted due to shutdown.", this);
 191
 192         shardMBean.unregisterMBean();
 193     }
 194
 195     @Override
 196     public void onReceiveRecover(final Object message) throws Exception {
 197         LOG.debug("{}: onReceiveRecover: Received message {} from {}", persistenceId(), message.getClass(),
 198             getSender());
 199
 200         super.onReceiveRecover(message);
 201         if (LOG.isTraceEnabled()) {
 202             appendEntriesReplyTracker.begin();
 203         }
 204     }
 205
 206     @Override
 207     public void onReceiveCommand(final Object message) throws Exception {
 208
 209         MessageTracker.Context context = appendEntriesReplyTracker.received(message);
 210
 211         if(context.error().isPresent()){
 212             LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(),
 213                 context.error());
 214         }
 215
 216         try {
 217             if (CreateTransaction.isSerializedType(message)) {
 218                 handleCreateTransaction(message);
 219             } else if (BatchedModifications.class.isInstance(message)) {
 220                 handleBatchedModifications((BatchedModifications)message);
 221             } else if (message instanceof ForwardedReadyTransaction) {
 222                 handleForwardedReadyTransaction((ForwardedReadyTransaction) message);
 223             } else if (message instanceof ReadyLocalTransaction) {
 224                 handleReadyLocalTransaction((ReadyLocalTransaction)message);
 225             } else if (CanCommitTransaction.isSerializedType(message)) {
 226                 handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
 227             } else if (CommitTransaction.isSerializedType(message)) {
 228                 handleCommitTransaction(CommitTransaction.fromSerializable(message));
 229             } else if (AbortTransaction.isSerializedType(message)) {
 230                 handleAbortTransaction(AbortTransaction.fromSerializable(message));
 231             } else if (CloseTransactionChain.isSerializedType(message)) {
 232                 closeTransactionChain(CloseTransactionChain.fromSerializable(message));
 233             } else if (message instanceof RegisterChangeListener) {
 234                 changeSupport.onMessage((RegisterChangeListener) message, isLeader(), hasLeader());
 235             } else if (message instanceof RegisterDataTreeChangeListener) {
 236                 treeChangeSupport.onMessage((RegisterDataTreeChangeListener) message, isLeader(), hasLeader());
 237             } else if (message instanceof UpdateSchemaContext) {
 238                 updateSchemaContext((UpdateSchemaContext) message);
 239             } else if (message instanceof PeerAddressResolved) {
 240                 PeerAddressResolved resolved = (PeerAddressResolved) message;
 241                 setPeerAddress(resolved.getPeerId().toString(),
 242                         resolved.getPeerAddress());
 243             } else if (message.equals(TX_COMMIT_TIMEOUT_CHECK_MESSAGE)) {
 244                 commitCoordinator.checkForExpiredTransactions(transactionCommitTimeout, this);
 245             } else if(message instanceof DatastoreContext) {
 246                 onDatastoreContext((DatastoreContext)message);
 247             } else if(message instanceof RegisterRoleChangeListener){
 248                 roleChangeNotifier.get().forward(message, context());
 249             } else if (message instanceof FollowerInitialSyncUpStatus) {
 250                 shardMBean.setFollowerInitialSyncStatus(((FollowerInitialSyncUpStatus) message).isInitialSyncDone());
 251                 context().parent().tell(message, self());
 252             } else if(GET_SHARD_MBEAN_MESSAGE.equals(message)){
 253                 sender().tell(getShardMBean(), self());
 254             } else if(message instanceof GetShardDataTree) {
 255                 sender().tell(store.getDataTree(), self());
 256             } else if(message instanceof ServerRemoved){
 257                 context().parent().forward(message, context());
 258             } else if(ShardTransactionMessageRetrySupport.TIMER_MESSAGE_CLASS.isInstance(message)) {
 259                 messageRetrySupport.onTimerMessage(message);
 260             } else {
 261                 super.onReceiveCommand(message);
 262             }
 263         } finally {
 264             context.done();
 265         }
 266     }
 267
 268     private boolean hasLeader() {
 269         return getLeaderId() != null;
 270     }
 271
 272     public int getPendingTxCommitQueueSize() {
 273         return commitCoordinator.getQueueSize();
 274     }
 275
 276     public int getCohortCacheSize() {
 277         return commitCoordinator.getCohortCacheSize();
 278     }
 279
 280     @Override
 281     protected Optional<ActorRef> getRoleChangeNotifier() {
 282         return roleChangeNotifier;
 283     }
 284
 285     @Override
 286     protected LeaderStateChanged newLeaderStateChanged(String memberId, String leaderId, short leaderPayloadVersion) {
 287         return new ShardLeaderStateChanged(memberId, leaderId,
 288                 isLeader() ? Optional.<DataTree>of(store.getDataTree()) : Optional.<DataTree>absent(),
 289                 leaderPayloadVersion);
 290     }
 291
 292     protected void onDatastoreContext(DatastoreContext context) {
 293         datastoreContext = context;
 294
 295         commitCoordinator.setQueueCapacity(datastoreContext.getShardTransactionCommitQueueCapacity());
 296
 297         setTransactionCommitTimeout();
 298
 299         if(datastoreContext.isPersistent() && !persistence().isRecoveryApplicable()) {
 300             setPersistence(true);
 301         } else if(!datastoreContext.isPersistent() && persistence().isRecoveryApplicable()) {
 302             setPersistence(false);
 303         }
 304
 305         updateConfigParams(datastoreContext.getShardRaftConfig());
 306     }
 307
 308     private static boolean isEmptyCommit(final DataTreeCandidate candidate) {
 309         return ModificationType.UNMODIFIED.equals(candidate.getRootNode().getModificationType());
 310     }
 311
 312     void continueCommit(final CohortEntry cohortEntry) {
 313         final DataTreeCandidate candidate = cohortEntry.getCandidate();
 314
 315         // If we do not have any followers and we are not using persistence
 316         // or if cohortEntry has no modifications
 317         // we can apply modification to the state immediately
 318         if ((!hasFollowers() && !persistence().isRecoveryApplicable()) || isEmptyCommit(candidate)) {
 319             applyModificationToState(cohortEntry.getReplySender(), cohortEntry.getTransactionID(), candidate);
 320         } else {
 321             Shard.this.persistData(cohortEntry.getReplySender(), cohortEntry.getTransactionID(),
 322                     DataTreeCandidatePayload.create(candidate));
 323         }
 324     }
 325
 326     private void handleCommitTransaction(final CommitTransaction commit) {
 327         if(!commitCoordinator.handleCommit(commit.getTransactionID(), getSender(), this)) {
 328             shardMBean.incrementFailedTransactionsCount();
 329         }
 330     }
 331
 332     private void finishCommit(@Nonnull final ActorRef sender, @Nonnull final String transactionID, @Nonnull final CohortEntry cohortEntry) {
 333         LOG.debug("{}: Finishing commit for transaction {}", persistenceId(), cohortEntry.getTransactionID());
 334
 335         try {
 336             cohortEntry.commit();
 337
 338             sender.tell(CommitTransactionReply.instance(cohortEntry.getClientVersion()).toSerializable(), getSelf());
 339
 340             shardMBean.incrementCommittedTransactionCount();
 341             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 342
 343         } catch (Exception e) {
 344             sender.tell(new akka.actor.Status.Failure(e), getSelf());
 345
 346             LOG.error("{}, An exception occurred while committing transaction {}", persistenceId(),
 347                     transactionID, e);
 348             shardMBean.incrementFailedTransactionsCount();
 349         } finally {
 350             commitCoordinator.currentTransactionComplete(transactionID, true);
 351         }
 352     }
 353
 354     private void finishCommit(@Nonnull final ActorRef sender, final @Nonnull String transactionID) {
 355         // With persistence enabled, this method is called via applyState by the leader strategy
 356         // after the commit has been replicated to a majority of the followers.
 357
 358         CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
 359         if (cohortEntry == null) {
 360             // The transaction is no longer the current commit. This can happen if the transaction
 361             // was aborted prior, most likely due to timeout in the front-end. We need to finish
 362             // committing the transaction though since it was successfully persisted and replicated
 363             // however we can't use the original cohort b/c it was already preCommitted and may
 364             // conflict with the current commit or may have been aborted so we commit with a new
 365             // transaction.
 366             cohortEntry = commitCoordinator.getAndRemoveCohortEntry(transactionID);
 367             if(cohortEntry != null) {
 368                 try {
 369                     store.applyForeignCandidate(transactionID, cohortEntry.getCandidate());
 370                 } catch (DataValidationFailedException e) {
 371                     shardMBean.incrementFailedTransactionsCount();
 372                     LOG.error("{}: Failed to re-apply transaction {}", persistenceId(), transactionID, e);
 373                 }
 374
 375                 sender.tell(CommitTransactionReply.instance(cohortEntry.getClientVersion()).toSerializable(),
 376                         getSelf());
 377             } else {
 378                 // This really shouldn't happen - it likely means that persistence or replication
 379                 // took so long to complete such that the cohort entry was expired from the cache.
 380                 IllegalStateException ex = new IllegalStateException(
 381                         String.format("%s: Could not finish committing transaction %s - no CohortEntry found",
 382                                 persistenceId(), transactionID));
 383                 LOG.error(ex.getMessage());
 384                 sender.tell(new akka.actor.Status.Failure(ex), getSelf());
 385             }
 386         } else {
 387             finishCommit(sender, transactionID, cohortEntry);
 388         }
 389     }
 390
 391     private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
 392         LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionID());
 393         commitCoordinator.handleCanCommit(canCommit.getTransactionID(), getSender(), this);
 394     }
 395
 396     protected void handleBatchedModificationsLocal(BatchedModifications batched, ActorRef sender) {
 397         try {
 398             commitCoordinator.handleBatchedModifications(batched, sender, this);
 399         } catch (Exception e) {
 400             LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(),
 401                     batched.getTransactionID(), e);
 402             sender.tell(new akka.actor.Status.Failure(e), getSelf());
 403         }
 404     }
 405
 406     private void handleBatchedModifications(BatchedModifications batched) {
 407         // This message is sent to prepare the modifications transaction directly on the Shard as an
 408         // optimization to avoid the extra overhead of a separate ShardTransaction actor. On the last
 409         // BatchedModifications message, the caller sets the ready flag in the message indicating
 410         // modifications are complete. The reply contains the cohort actor path (this actor) for the caller
 411         // to initiate the 3-phase commit. This also avoids the overhead of sending an additional
 412         // ReadyTransaction message.
 413
 414         // If we're not the leader then forward to the leader. This is a safety measure - we shouldn't
 415         // normally get here if we're not the leader as the front-end (TransactionProxy) should determine
 416         // the primary/leader shard. However with timing and caching on the front-end, there's a small
 417         // window where it could have a stale leader during leadership transitions.
 418         //
 419         boolean isLeaderActive = isLeaderActive();
 420         if (isLeader() && isLeaderActive) {
 421             handleBatchedModificationsLocal(batched, getSender());
 422         } else {
 423             ActorSelection leader = getLeader();
 424             if (!isLeaderActive || leader == null) {
 425                 messageRetrySupport.addMessageToRetry(batched, getSender(),
 426                         "Could not commit transaction " + batched.getTransactionID());
 427             } else {
 428                 // TODO: what if this is not the first batch and leadership changed in between batched messages?
 429                 // We could check if the commitCoordinator already has a cached entry and forward all the previous
 430                 // batched modifications.
 431                 LOG.debug("{}: Forwarding BatchedModifications to leader {}", persistenceId(), leader);
 432                 leader.forward(batched, getContext());
 433             }
 434         }
 435     }
 436
 437     private boolean failIfIsolatedLeader(ActorRef sender) {
 438         if(isIsolatedLeader()) {
 439             sender.tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format(
 440                     "Shard %s was the leader but has lost contact with all of its followers. Either all" +
 441                     " other follower nodes are down or this node is isolated by a network partition.",
 442                     persistenceId()))), getSelf());
 443             return true;
 444         }
 445
 446         return false;
 447     }
 448
 449     protected boolean isIsolatedLeader() {
 450         return getRaftState() == RaftState.IsolatedLeader;
 451     }
 452
 453     private void handleReadyLocalTransaction(final ReadyLocalTransaction message) {
 454         LOG.debug("{}: handleReadyLocalTransaction for {}", persistenceId(), message.getTransactionID());
 455
 456         boolean isLeaderActive = isLeaderActive();
 457         if (isLeader() && isLeaderActive) {
 458             try {
 459                 commitCoordinator.handleReadyLocalTransaction(message, getSender(), this);
 460             } catch (Exception e) {
 461                 LOG.error("{}: Error handling ReadyLocalTransaction for Tx {}", persistenceId(),
 462                         message.getTransactionID(), e);
 463                 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 464             }
 465         } else {
 466             ActorSelection leader = getLeader();
 467             if (!isLeaderActive || leader == null) {
 468                 messageRetrySupport.addMessageToRetry(message, getSender(),
 469                         "Could not commit transaction " + message.getTransactionID());
 470             } else {
 471                 LOG.debug("{}: Forwarding ReadyLocalTransaction to leader {}", persistenceId(), leader);
 472                 message.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
 473                 leader.forward(message, getContext());
 474             }
 475         }
 476     }
 477
 478     private void handleForwardedReadyTransaction(ForwardedReadyTransaction forwardedReady) {
 479         LOG.debug("{}: handleForwardedReadyTransaction for {}", persistenceId(), forwardedReady.getTransactionID());
 480
 481         boolean isLeaderActive = isLeaderActive();
 482         if (isLeader() && isLeaderActive) {
 483             commitCoordinator.handleForwardedReadyTransaction(forwardedReady, getSender(), this);
 484         } else {
 485             ActorSelection leader = getLeader();
 486             if (!isLeaderActive || leader == null) {
 487                 messageRetrySupport.addMessageToRetry(forwardedReady, getSender(),
 488                         "Could not commit transaction " + forwardedReady.getTransactionID());
 489             } else {
 490                 LOG.debug("{}: Forwarding ForwardedReadyTransaction to leader {}", persistenceId(), leader);
 491
 492                 ReadyLocalTransaction readyLocal = new ReadyLocalTransaction(forwardedReady.getTransactionID(),
 493                         forwardedReady.getTransaction().getSnapshot(), forwardedReady.isDoImmediateCommit());
 494                 readyLocal.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
 495                 leader.forward(readyLocal, getContext());
 496             }
 497         }
 498     }
 499
 500     private void handleAbortTransaction(final AbortTransaction abort) {
 501         doAbortTransaction(abort.getTransactionID(), getSender());
 502     }
 503
 504     void doAbortTransaction(final String transactionID, final ActorRef sender) {
 505         commitCoordinator.handleAbort(transactionID, sender, this);
 506     }
 507
 508     private void handleCreateTransaction(final Object message) {
 509         if (isLeader()) {
 510             createTransaction(CreateTransaction.fromSerializable(message));
 511         } else if (getLeader() != null) {
 512             getLeader().forward(message, getContext());
 513         } else {
 514             getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(
 515                     "Could not create a shard transaction", persistenceId())), getSelf());
 516         }
 517     }
 518
 519     private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
 520         store.closeTransactionChain(closeTransactionChain.getTransactionChainId());
 521     }
 522
 523     private ActorRef createTypedTransactionActor(int transactionType,
 524             ShardTransactionIdentifier transactionId, String transactionChainId) {
 525
 526         return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType),
 527                 transactionId, transactionChainId);
 528     }
 529
 530     private void createTransaction(CreateTransaction createTransaction) {
 531         try {
 532             if(TransactionType.fromInt(createTransaction.getTransactionType()) != TransactionType.READ_ONLY &&
 533                     failIfIsolatedLeader(getSender())) {
 534                 return;
 535             }
 536
 537             ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
 538                 createTransaction.getTransactionId(), createTransaction.getTransactionChainId());
 539
 540             getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
 541                     createTransaction.getTransactionId(), createTransaction.getVersion()).toSerializable(), getSelf());
 542         } catch (Exception e) {
 543             getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 544         }
 545     }
 546
 547     private ActorRef createTransaction(int transactionType, String remoteTransactionId,
 548             String transactionChainId) {
 549
 550
 551         ShardTransactionIdentifier transactionId = new ShardTransactionIdentifier(remoteTransactionId);
 552
 553         if(LOG.isDebugEnabled()) {
 554             LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
 555         }
 556
 557         ActorRef transactionActor = createTypedTransactionActor(transactionType, transactionId,
 558                 transactionChainId);
 559
 560         return transactionActor;
 561     }
 562
 563     private void commitWithNewTransaction(final Modification modification) {
 564         ReadWriteShardDataTreeTransaction tx = store.newReadWriteTransaction(modification.toString(), null);
 565         modification.apply(tx.getSnapshot());
 566         try {
 567             snapshotCohort.syncCommitTransaction(tx);
 568             shardMBean.incrementCommittedTransactionCount();
 569             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 570         } catch (Exception e) {
 571             shardMBean.incrementFailedTransactionsCount();
 572             LOG.error("{}: Failed to commit", persistenceId(), e);
 573         }
 574     }
 575
 576     private void updateSchemaContext(final UpdateSchemaContext message) {
 577         updateSchemaContext(message.getSchemaContext());
 578     }
 579
 580     @VisibleForTesting
 581     void updateSchemaContext(final SchemaContext schemaContext) {
 582         store.updateSchemaContext(schemaContext);
 583     }
 584
 585     private boolean isMetricsCaptureEnabled() {
 586         CommonConfig config = new CommonConfig(getContext().system().settings().config());
 587         return config.isMetricCaptureEnabled();
 588     }
 589
 590     @Override
 591     @VisibleForTesting
 592     public RaftActorSnapshotCohort getRaftActorSnapshotCohort() {
 593         return snapshotCohort;
 594     }
 595
 596     @Override
 597     @Nonnull
 598     protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
 599         return new ShardRecoveryCoordinator(store, store.getSchemaContext(),
 600                 restoreFromSnapshot != null ? restoreFromSnapshot.getSnapshot() : null, persistenceId(), LOG);
 601     }
 602
 603     @Override
 604     protected void onRecoveryComplete() {
 605         restoreFromSnapshot = null;
 606
 607         //notify shard manager
 608         getContext().parent().tell(new ActorInitialized(), getSelf());
 609
 610         // Being paranoid here - this method should only be called once but just in case...
 611         if(txCommitTimeoutCheckSchedule == null) {
 612             // Schedule a message to be periodically sent to check if the current in-progress
 613             // transaction should be expired and aborted.
 614             FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
 615             txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
 616                     period, period, getSelf(),
 617                     TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
 618         }
 619     }
 620
 621     @Override
 622     protected void applyState(final ActorRef clientActor, final String identifier, final Object data) {
 623         if (data instanceof DataTreeCandidatePayload) {
 624             if (clientActor == null) {
 625                 // No clientActor indicates a replica coming from the leader
 626                 try {
 627                     store.applyForeignCandidate(identifier, ((DataTreeCandidatePayload)data).getCandidate());
 628                 } catch (DataValidationFailedException | IOException e) {
 629                     LOG.error("{}: Error applying replica {}", persistenceId(), identifier, e);
 630                 }
 631             } else {
 632                 // Replication consensus reached, proceed to commit
 633                 finishCommit(clientActor, identifier);
 634             }
 635         } else if (data instanceof CompositeModificationPayload) {
 636             Object modification = ((CompositeModificationPayload) data).getModification();
 637
 638             applyModificationToState(clientActor, identifier, modification);
 639         } else if(data instanceof CompositeModificationByteStringPayload ){
 640             Object modification = ((CompositeModificationByteStringPayload) data).getModification();
 641
 642             applyModificationToState(clientActor, identifier, modification);
 643         } else {
 644             LOG.error("{}: Unknown state received {} Class loader = {} CompositeNodeMod.ClassLoader = {}",
 645                     persistenceId(), data, data.getClass().getClassLoader(),
 646                     CompositeModificationPayload.class.getClassLoader());
 647         }
 648     }
 649
 650     private void applyModificationToState(ActorRef clientActor, String identifier, Object modification) {
 651         if(modification == null) {
 652             LOG.error(
 653                     "{}: modification is null - this is very unexpected, clientActor = {}, identifier = {}",
 654                     persistenceId(), identifier, clientActor != null ? clientActor.path().toString() : null);
 655         } else if(clientActor == null) {
 656             // There's no clientActor to which to send a commit reply so we must be applying
 657             // replicated state from the leader.
 658             commitWithNewTransaction(MutableCompositeModification.fromSerializable(modification));
 659         } else {
 660             // This must be the OK to commit after replication consensus.
 661             finishCommit(clientActor, identifier);
 662         }
 663     }
 664
 665     @Override
 666     protected void onStateChanged() {
 667         boolean isLeader = isLeader();
 668         boolean hasLeader = hasLeader();
 669         changeSupport.onLeadershipChange(isLeader, hasLeader);
 670         treeChangeSupport.onLeadershipChange(isLeader, hasLeader);
 671
 672         // If this actor is no longer the leader close all the transaction chains
 673         if (!isLeader) {
 674             if(LOG.isDebugEnabled()) {
 675                 LOG.debug(
 676                     "{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader",
 677                     persistenceId(), getId());
 678             }
 679
 680             store.closeAllTransactionChains();
 681
 682             commitCoordinator.abortPendingTransactions(
 683                     "The transacton was aborted due to inflight leadership change.", this);
 684         }
 685
 686         if(hasLeader && !isIsolatedLeader()) {
 687             messageRetrySupport.retryMessages();
 688         }
 689     }
 690
 691     @Override
 692     protected void onLeaderChanged(String oldLeader, String newLeader) {
 693         shardMBean.incrementLeadershipChangeCount();
 694
 695         if(hasLeader() && !isIsolatedLeader()) {
 696             messageRetrySupport.retryMessages();
 697         }
 698     }
 699
 700     @Override
 701     protected void pauseLeader(Runnable operation) {
 702         LOG.debug("{}: In pauseLeader, operation: {}", persistenceId(), operation);
 703         commitCoordinator.setRunOnPendingTransactionsComplete(operation);
 704     }
 705
 706     @Override
 707     public String persistenceId() {
 708         return this.name;
 709     }
 710
 711     @VisibleForTesting
 712     ShardCommitCoordinator getCommitCoordinator() {
 713         return commitCoordinator;
 714     }
 715
 716     public DatastoreContext getDatastoreContext() {
 717         return datastoreContext;
 718     }
 719
 720     @VisibleForTesting
 721     public ShardDataTree getDataStore() {
 722         return store;
 723     }
 724
 725     @VisibleForTesting
 726     ShardStats getShardMBean() {
 727         return shardMBean;
 728     }
 729
 730     public static Builder builder() {
 731         return new Builder();
 732     }
 733
 734     public static abstract class AbstractBuilder<T extends AbstractBuilder<T, S>, S extends Shard> {
 735         private final Class<S> shardClass;
 736         private ShardIdentifier id;
 737         private Map<String, String> peerAddresses = Collections.emptyMap();
 738         private DatastoreContext datastoreContext;
 739         private SchemaContext schemaContext;
 740         private DatastoreSnapshot.ShardSnapshot restoreFromSnapshot;
 741         private volatile boolean sealed;
 742
 743         protected AbstractBuilder(Class<S> shardClass) {
 744             this.shardClass = shardClass;
 745         }
 746
 747         protected void checkSealed() {
 748             Preconditions.checkState(!sealed, "Builder isalready sealed - further modifications are not allowed");
 749         }
 750
 751         @SuppressWarnings("unchecked")
 752         private T self() {
 753             return (T) this;
 754         }
 755
 756         public T id(ShardIdentifier id) {
 757             checkSealed();
 758             this.id = id;
 759             return self();
 760         }
 761
 762         public T peerAddresses(Map<String, String> peerAddresses) {
 763             checkSealed();
 764             this.peerAddresses = peerAddresses;
 765             return self();
 766         }
 767
 768         public T datastoreContext(DatastoreContext datastoreContext) {
 769             checkSealed();
 770             this.datastoreContext = datastoreContext;
 771             return self();
 772         }
 773
 774         public T schemaContext(SchemaContext schemaContext) {
 775             checkSealed();
 776             this.schemaContext = schemaContext;
 777             return self();
 778         }
 779
 780         public T restoreFromSnapshot(DatastoreSnapshot.ShardSnapshot restoreFromSnapshot) {
 781             checkSealed();
 782             this.restoreFromSnapshot = restoreFromSnapshot;
 783             return self();
 784         }
 785
 786         public ShardIdentifier getId() {
 787             return id;
 788         }
 789
 790         public Map<String, String> getPeerAddresses() {
 791             return peerAddresses;
 792         }
 793
 794         public DatastoreContext getDatastoreContext() {
 795             return datastoreContext;
 796         }
 797
 798         public SchemaContext getSchemaContext() {
 799             return schemaContext;
 800         }
 801
 802         public DatastoreSnapshot.ShardSnapshot getRestoreFromSnapshot() {
 803             return restoreFromSnapshot;
 804         }
 805
 806         public TreeType getTreeType() {
 807             switch (datastoreContext.getLogicalStoreType()) {
 808             case CONFIGURATION:
 809                 return TreeType.CONFIGURATION;
 810             case OPERATIONAL:
 811                 return TreeType.OPERATIONAL;
 812             }
 813
 814             throw new IllegalStateException("Unhandled logical store type " + datastoreContext.getLogicalStoreType());
 815         }
 816
 817         protected void verify() {
 818             Preconditions.checkNotNull(id, "id should not be null");
 819             Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
 820             Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
 821             Preconditions.checkNotNull(schemaContext, "schemaContext should not be null");
 822         }
 823
 824         public Props props() {
 825             sealed = true;
 826             verify();
 827             return Props.create(shardClass, this);
 828         }
 829     }
 830
 831     public static class Builder extends AbstractBuilder<Builder, Shard> {
 832         private Builder() {
 833             super(Shard.class);
 834         }
 835     }
 836 }