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