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     protected Shard(final AbstractBuilder<?, ?> builder) {
 133         super(builder.getId().toString(), builder.getPeerAddresses(),
 134                 Optional.of(builder.getDatastoreContext().getShardRaftConfig()), DataStoreVersions.CURRENT_VERSION);
 135
 136         this.name = builder.getId().toString();
 137         this.datastoreContext = builder.getDatastoreContext();
 138         this.restoreFromSnapshot = builder.getRestoreFromSnapshot();
 139
 140         setPersistence(datastoreContext.isPersistent());
 141
 142         LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
 143
 144         ShardDataTreeChangeListenerPublisherActorProxy treeChangeListenerPublisher =
 145                 new ShardDataTreeChangeListenerPublisherActorProxy(getContext(), name + "-DTCL-publisher");
 146         ShardDataChangeListenerPublisherActorProxy dataChangeListenerPublisher =
 147                 new ShardDataChangeListenerPublisherActorProxy(getContext(), name + "-DCL-publisher");
 148         if(builder.getDataTree() != null) {
 149             store = new ShardDataTree(this, builder.getSchemaContext(), builder.getDataTree(),
 150                     treeChangeListenerPublisher, dataChangeListenerPublisher, name);
 151         } else {
 152             store = new ShardDataTree(this, builder.getSchemaContext(), builder.getTreeType(),
 153                     treeChangeListenerPublisher, dataChangeListenerPublisher, name);
 154         }
 155
 156         shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
 157                 datastoreContext.getDataStoreMXBeanType());
 158         shardMBean.setShard(this);
 159
 160         if (isMetricsCaptureEnabled()) {
 161             getContext().become(new MeteringBehavior(this));
 162         }
 163
 164         commitCoordinator = new ShardCommitCoordinator(store, LOG, this.name);
 165
 166         setTransactionCommitTimeout();
 167
 168         // create a notifier actor for each cluster member
 169         roleChangeNotifier = createRoleChangeNotifier(name.toString());
 170
 171         appendEntriesReplyTracker = new MessageTracker(AppendEntriesReply.class,
 172                 getRaftActorContext().getConfigParams().getIsolatedCheckIntervalInMillis());
 173
 174         transactionActorFactory = new ShardTransactionActorFactory(store, datastoreContext,
 175                 new Dispatchers(context().system().dispatchers()).getDispatcherPath(
 176                         Dispatchers.DispatcherType.Transaction), self(), getContext(), shardMBean);
 177
 178         snapshotCohort = ShardSnapshotCohort.create(getContext(), builder.getId().getMemberName(), store, LOG,
 179             this.name);
 180
 181         messageRetrySupport = new ShardTransactionMessageRetrySupport(this);
 182     }
 183
 184     private void setTransactionCommitTimeout() {
 185         transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
 186                 datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS) / 2;
 187     }
 188
 189     private Optional<ActorRef> createRoleChangeNotifier(final String shardId) {
 190         ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
 191             RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
 192         return Optional.of(shardRoleChangeNotifier);
 193     }
 194
 195     @Override
 196     public void postStop() {
 197         LOG.info("Stopping Shard {}", persistenceId());
 198
 199         super.postStop();
 200
 201         messageRetrySupport.close();
 202
 203         if (txCommitTimeoutCheckSchedule != null) {
 204             txCommitTimeoutCheckSchedule.cancel();
 205         }
 206
 207         commitCoordinator.abortPendingTransactions("Transaction aborted due to shutdown.", this);
 208
 209         shardMBean.unregisterMBean();
 210     }
 211
 212     @Override
 213     protected void handleRecover(final Object message) {
 214         LOG.debug("{}: onReceiveRecover: Received message {} from {}", persistenceId(), message.getClass(),
 215             getSender());
 216
 217         super.handleRecover(message);
 218         if (LOG.isTraceEnabled()) {
 219             appendEntriesReplyTracker.begin();
 220         }
 221     }
 222
 223     @Override
 224     protected void handleNonRaftCommand(final Object message) {
 225         try (final MessageTracker.Context context = appendEntriesReplyTracker.received(message)) {
 226             final Optional<Error> maybeError = context.error();
 227             if (maybeError.isPresent()) {
 228                 LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(),
 229                     maybeError.get());
 230             }
 231
 232             if (CreateTransaction.isSerializedType(message)) {
 233                 handleCreateTransaction(message);
 234             } else if (message instanceof BatchedModifications) {
 235                 handleBatchedModifications((BatchedModifications)message);
 236             } else if (message instanceof ForwardedReadyTransaction) {
 237                 handleForwardedReadyTransaction((ForwardedReadyTransaction) message);
 238             } else if (message instanceof ReadyLocalTransaction) {
 239                 handleReadyLocalTransaction((ReadyLocalTransaction)message);
 240             } else if (CanCommitTransaction.isSerializedType(message)) {
 241                 handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
 242             } else if (CommitTransaction.isSerializedType(message)) {
 243                 handleCommitTransaction(CommitTransaction.fromSerializable(message));
 244             } else if (AbortTransaction.isSerializedType(message)) {
 245                 handleAbortTransaction(AbortTransaction.fromSerializable(message));
 246             } else if (CloseTransactionChain.isSerializedType(message)) {
 247                 closeTransactionChain(CloseTransactionChain.fromSerializable(message));
 248             } else if (message instanceof RegisterChangeListener) {
 249                 changeSupport.onMessage((RegisterChangeListener) message, isLeader(), hasLeader());
 250             } else if (message instanceof RegisterDataTreeChangeListener) {
 251                 treeChangeSupport.onMessage((RegisterDataTreeChangeListener) message, isLeader(), hasLeader());
 252             } else if (message instanceof UpdateSchemaContext) {
 253                 updateSchemaContext((UpdateSchemaContext) message);
 254             } else if (message instanceof PeerAddressResolved) {
 255                 PeerAddressResolved resolved = (PeerAddressResolved) message;
 256                 setPeerAddress(resolved.getPeerId().toString(),
 257                         resolved.getPeerAddress());
 258             } else if (TX_COMMIT_TIMEOUT_CHECK_MESSAGE.equals(message)) {
 259                 store.checkForExpiredTransactions(transactionCommitTimeout);
 260                 commitCoordinator.checkForExpiredTransactions(transactionCommitTimeout, this);
 261             } else if (message instanceof DatastoreContext) {
 262                 onDatastoreContext((DatastoreContext)message);
 263             } else if (message instanceof RegisterRoleChangeListener){
 264                 roleChangeNotifier.get().forward(message, context());
 265             } else if (message instanceof FollowerInitialSyncUpStatus) {
 266                 shardMBean.setFollowerInitialSyncStatus(((FollowerInitialSyncUpStatus) message).isInitialSyncDone());
 267                 context().parent().tell(message, self());
 268             } else if (GET_SHARD_MBEAN_MESSAGE.equals(message)){
 269                 sender().tell(getShardMBean(), self());
 270             } else if (message instanceof GetShardDataTree) {
 271                 sender().tell(store.getDataTree(), self());
 272             } else if (message instanceof ServerRemoved){
 273                 context().parent().forward(message, context());
 274             } else if (ShardTransactionMessageRetrySupport.TIMER_MESSAGE_CLASS.isInstance(message)) {
 275                 messageRetrySupport.onTimerMessage(message);
 276             } else if (message instanceof DataTreeCohortActorRegistry.CohortRegistryCommand) {
 277                 store.processCohortRegistryCommand(getSender(),
 278                         (DataTreeCohortActorRegistry.CohortRegistryCommand) message);
 279             } else {
 280                 super.handleNonRaftCommand(message);
 281             }
 282         }
 283     }
 284
 285     private boolean hasLeader() {
 286         return getLeaderId() != null;
 287     }
 288
 289     public int getPendingTxCommitQueueSize() {
 290         return store.getQueueSize();
 291     }
 292
 293     public int getCohortCacheSize() {
 294         return commitCoordinator.getCohortCacheSize();
 295     }
 296
 297     @Override
 298     protected Optional<ActorRef> getRoleChangeNotifier() {
 299         return roleChangeNotifier;
 300     }
 301
 302     @Override
 303     protected LeaderStateChanged newLeaderStateChanged(final String memberId, final String leaderId, final short leaderPayloadVersion) {
 304         return isLeader() ? new ShardLeaderStateChanged(memberId, leaderId, store.getDataTree(), leaderPayloadVersion)
 305                 : new ShardLeaderStateChanged(memberId, leaderId, leaderPayloadVersion);
 306     }
 307
 308     protected void onDatastoreContext(final DatastoreContext context) {
 309         datastoreContext = context;
 310
 311         setTransactionCommitTimeout();
 312
 313         setPersistence(datastoreContext.isPersistent());
 314
 315         updateConfigParams(datastoreContext.getShardRaftConfig());
 316     }
 317
 318     boolean canSkipPayload() {
 319         // If we do not have any followers and we are not using persistence we can apply modification to the state
 320         // immediately
 321         return !hasFollowers() && !persistence().isRecoveryApplicable();
 322     }
 323
 324     // applyState() will be invoked once consensus is reached on the payload
 325     void persistPayload(final TransactionIdentifier transactionId, final Payload payload) {
 326         // We are faking the sender
 327         persistData(self(), transactionId, payload);
 328     }
 329
 330     private void handleCommitTransaction(final CommitTransaction commit) {
 331         if (isLeader()) {
 332             commitCoordinator.handleCommit(commit.getTransactionID(), getSender(), this);
 333         } else {
 334             ActorSelection leader = getLeader();
 335             if (leader == null) {
 336                 messageRetrySupport.addMessageToRetry(commit, getSender(),
 337                         "Could not commit transaction " + commit.getTransactionID());
 338             } else {
 339                 LOG.debug("{}: Forwarding CommitTransaction to leader {}", persistenceId(), leader);
 340                 leader.forward(commit, getContext());
 341             }
 342         }
 343     }
 344
 345     private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
 346         LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionID());
 347
 348         if (isLeader()) {
 349         commitCoordinator.handleCanCommit(canCommit.getTransactionID(), getSender(), this);
 350         } else {
 351             ActorSelection leader = getLeader();
 352             if (leader == null) {
 353                 messageRetrySupport.addMessageToRetry(canCommit, getSender(),
 354                         "Could not canCommit transaction " + canCommit.getTransactionID());
 355             } else {
 356                 LOG.debug("{}: Forwarding CanCommitTransaction to leader {}", persistenceId(), leader);
 357                 leader.forward(canCommit, getContext());
 358             }
 359         }
 360     }
 361
 362     protected void handleBatchedModificationsLocal(final BatchedModifications batched, final ActorRef sender) {
 363         try {
 364             commitCoordinator.handleBatchedModifications(batched, sender, this);
 365         } catch (Exception e) {
 366             LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(),
 367                     batched.getTransactionID(), e);
 368             sender.tell(new akka.actor.Status.Failure(e), getSelf());
 369         }
 370     }
 371
 372     private void handleBatchedModifications(final BatchedModifications batched) {
 373         // This message is sent to prepare the modifications transaction directly on the Shard as an
 374         // optimization to avoid the extra overhead of a separate ShardTransaction actor. On the last
 375         // BatchedModifications message, the caller sets the ready flag in the message indicating
 376         // modifications are complete. The reply contains the cohort actor path (this actor) for the caller
 377         // to initiate the 3-phase commit. This also avoids the overhead of sending an additional
 378         // ReadyTransaction message.
 379
 380         // If we're not the leader then forward to the leader. This is a safety measure - we shouldn't
 381         // normally get here if we're not the leader as the front-end (TransactionProxy) should determine
 382         // the primary/leader shard. However with timing and caching on the front-end, there's a small
 383         // window where it could have a stale leader during leadership transitions.
 384         //
 385         boolean isLeaderActive = isLeaderActive();
 386         if (isLeader() && isLeaderActive) {
 387             handleBatchedModificationsLocal(batched, getSender());
 388         } else {
 389             ActorSelection leader = getLeader();
 390             if (!isLeaderActive || leader == null) {
 391                 messageRetrySupport.addMessageToRetry(batched, getSender(),
 392                         "Could not commit transaction " + batched.getTransactionID());
 393             } else {
 394                 // If this is not the first batch and leadership changed in between batched messages,
 395                 // we need to reconstruct previous BatchedModifications from the transaction
 396                 // DataTreeModification, honoring the max batched modification count, and forward all the
 397                 // previous BatchedModifications to the new leader.
 398                 Collection<BatchedModifications> newModifications = commitCoordinator.createForwardedBatchedModifications(
 399                         batched, datastoreContext.getShardBatchedModificationCount());
 400
 401                 LOG.debug("{}: Forwarding {} BatchedModifications to leader {}", persistenceId(),
 402                         newModifications.size(), leader);
 403
 404                 for (BatchedModifications bm : newModifications) {
 405                     leader.forward(bm, getContext());
 406                 }
 407             }
 408         }
 409     }
 410
 411     private boolean failIfIsolatedLeader(final ActorRef sender) {
 412         if (isIsolatedLeader()) {
 413             sender.tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format(
 414                     "Shard %s was the leader but has lost contact with all of its followers. Either all" +
 415                     " other follower nodes are down or this node is isolated by a network partition.",
 416                     persistenceId()))), getSelf());
 417             return true;
 418         }
 419
 420         return false;
 421     }
 422
 423     protected boolean isIsolatedLeader() {
 424         return getRaftState() == RaftState.IsolatedLeader;
 425     }
 426
 427     private void handleReadyLocalTransaction(final ReadyLocalTransaction message) {
 428         LOG.debug("{}: handleReadyLocalTransaction for {}", persistenceId(), message.getTransactionID());
 429
 430         boolean isLeaderActive = isLeaderActive();
 431         if (isLeader() && isLeaderActive) {
 432             try {
 433                 commitCoordinator.handleReadyLocalTransaction(message, getSender(), this);
 434             } catch (Exception e) {
 435                 LOG.error("{}: Error handling ReadyLocalTransaction for Tx {}", persistenceId(),
 436                         message.getTransactionID(), e);
 437                 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 438             }
 439         } else {
 440             ActorSelection leader = getLeader();
 441             if (!isLeaderActive || leader == null) {
 442                 messageRetrySupport.addMessageToRetry(message, getSender(),
 443                         "Could not commit transaction " + message.getTransactionID());
 444             } else {
 445                 LOG.debug("{}: Forwarding ReadyLocalTransaction to leader {}", persistenceId(), leader);
 446                 message.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
 447                 leader.forward(message, getContext());
 448             }
 449         }
 450     }
 451
 452     private void handleForwardedReadyTransaction(final ForwardedReadyTransaction forwardedReady) {
 453         LOG.debug("{}: handleForwardedReadyTransaction for {}", persistenceId(), forwardedReady.getTransactionID());
 454
 455         boolean isLeaderActive = isLeaderActive();
 456         if (isLeader() && isLeaderActive) {
 457             commitCoordinator.handleForwardedReadyTransaction(forwardedReady, getSender(), this);
 458         } else {
 459             ActorSelection leader = getLeader();
 460             if (!isLeaderActive || leader == null) {
 461                 messageRetrySupport.addMessageToRetry(forwardedReady, getSender(),
 462                         "Could not commit transaction " + forwardedReady.getTransactionID());
 463             } else {
 464                 LOG.debug("{}: Forwarding ForwardedReadyTransaction to leader {}", persistenceId(), leader);
 465
 466                 ReadyLocalTransaction readyLocal = new ReadyLocalTransaction(forwardedReady.getTransactionID(),
 467                         forwardedReady.getTransaction().getSnapshot(), forwardedReady.isDoImmediateCommit());
 468                 readyLocal.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
 469                 leader.forward(readyLocal, getContext());
 470             }
 471         }
 472     }
 473
 474     private void handleAbortTransaction(final AbortTransaction abort) {
 475         doAbortTransaction(abort.getTransactionID(), getSender());
 476     }
 477
 478     void doAbortTransaction(final TransactionIdentifier transactionID, final ActorRef sender) {
 479         commitCoordinator.handleAbort(transactionID, sender, this);
 480     }
 481
 482     private void handleCreateTransaction(final Object message) {
 483         if (isLeader()) {
 484             createTransaction(CreateTransaction.fromSerializable(message));
 485         } else if (getLeader() != null) {
 486             getLeader().forward(message, getContext());
 487         } else {
 488             getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(
 489                     "Could not create a shard transaction", persistenceId())), getSelf());
 490         }
 491     }
 492
 493     private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
 494         store.closeTransactionChain(closeTransactionChain.getIdentifier());
 495     }
 496
 497     private void createTransaction(final CreateTransaction createTransaction) {
 498         try {
 499             if (TransactionType.fromInt(createTransaction.getTransactionType()) != TransactionType.READ_ONLY &&
 500                     failIfIsolatedLeader(getSender())) {
 501                 return;
 502             }
 503
 504             ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
 505                 createTransaction.getTransactionId());
 506
 507             getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
 508                     createTransaction.getTransactionId(), createTransaction.getVersion()).toSerializable(), getSelf());
 509         } catch (Exception e) {
 510             getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 511         }
 512     }
 513
 514     private ActorRef createTransaction(final int transactionType, final TransactionIdentifier transactionId) {
 515         LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
 516         return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType),
 517             transactionId);
 518     }
 519
 520     private void updateSchemaContext(final UpdateSchemaContext message) {
 521         updateSchemaContext(message.getSchemaContext());
 522     }
 523
 524     @VisibleForTesting
 525     void updateSchemaContext(final SchemaContext schemaContext) {
 526         store.updateSchemaContext(schemaContext);
 527     }
 528
 529     private boolean isMetricsCaptureEnabled() {
 530         CommonConfig config = new CommonConfig(getContext().system().settings().config());
 531         return config.isMetricCaptureEnabled();
 532     }
 533
 534     @Override
 535     @VisibleForTesting
 536     public RaftActorSnapshotCohort getRaftActorSnapshotCohort() {
 537         return snapshotCohort;
 538     }
 539
 540     @Override
 541     @Nonnull
 542     protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
 543         return new ShardRecoveryCoordinator(store,
 544             restoreFromSnapshot != null ? restoreFromSnapshot.getSnapshot() : null, persistenceId(), LOG);
 545     }
 546
 547     @Override
 548     protected void onRecoveryComplete() {
 549         restoreFromSnapshot = null;
 550
 551         //notify shard manager
 552         getContext().parent().tell(new ActorInitialized(), getSelf());
 553
 554         // Being paranoid here - this method should only be called once but just in case...
 555         if (txCommitTimeoutCheckSchedule == null) {
 556             // Schedule a message to be periodically sent to check if the current in-progress
 557             // transaction should be expired and aborted.
 558             FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
 559             txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
 560                     period, period, getSelf(),
 561                     TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
 562         }
 563     }
 564
 565     @Override
 566     protected void applyState(final ActorRef clientActor, final Identifier identifier, final Object data) {
 567         if (data instanceof Payload) {
 568             try {
 569                 store.applyReplicatedPayload(identifier, (Payload)data);
 570             } catch (DataValidationFailedException | IOException e) {
 571                 LOG.error("{}: Error applying replica {}", persistenceId(), identifier, e);
 572             }
 573         } else {
 574             LOG.error("{}: Unknown state for {} received {}", persistenceId(), identifier, data);
 575         }
 576     }
 577
 578     @Override
 579     protected void onStateChanged() {
 580         boolean isLeader = isLeader();
 581         boolean hasLeader = hasLeader();
 582         changeSupport.onLeadershipChange(isLeader, hasLeader);
 583         treeChangeSupport.onLeadershipChange(isLeader, hasLeader);
 584
 585         // If this actor is no longer the leader close all the transaction chains
 586         if (!isLeader) {
 587             if (LOG.isDebugEnabled()) {
 588                 LOG.debug(
 589                     "{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader",
 590                     persistenceId(), getId());
 591             }
 592
 593             store.closeAllTransactionChains();
 594         }
 595
 596         if (hasLeader && !isIsolatedLeader()) {
 597             messageRetrySupport.retryMessages();
 598         }
 599     }
 600
 601     @Override
 602     protected void onLeaderChanged(final String oldLeader, final String newLeader) {
 603         shardMBean.incrementLeadershipChangeCount();
 604
 605         boolean hasLeader = hasLeader();
 606         if (hasLeader && !isLeader()) {
 607             // Another leader was elected. If we were the previous leader and had pending transactions, convert
 608             // them to transaction messages and send to the new leader.
 609             ActorSelection leader = getLeader();
 610             if (leader != null) {
 611                 Collection<?> messagesToForward = convertPendingTransactionsToMessages();
 612
 613                 if (!messagesToForward.isEmpty()) {
 614                     LOG.debug("{}: Forwarding {} pending transaction messages to leader {}", persistenceId(),
 615                             messagesToForward.size(), leader);
 616
 617                     for (Object message : messagesToForward) {
 618                         leader.tell(message, self());
 619                     }
 620                 }
 621             } else {
 622                 commitCoordinator.abortPendingTransactions(
 623                         "The transacton was aborted due to inflight leadership change and the leader address isn't available.",
 624                         this);
 625             }
 626         }
 627
 628         if (hasLeader && !isIsolatedLeader()) {
 629             messageRetrySupport.retryMessages();
 630         }
 631     }
 632
 633     /**
 634      * Clears all pending transactions and converts them to messages to be forwarded to a new leader.
 635      *
 636      * @return the converted messages
 637      */
 638     public Collection<?> convertPendingTransactionsToMessages() {
 639         return commitCoordinator.convertPendingTransactionsToMessages(datastoreContext.getShardBatchedModificationCount());
 640     }
 641
 642     @Override
 643     protected void pauseLeader(final Runnable operation) {
 644         LOG.debug("{}: In pauseLeader, operation: {}", persistenceId(), operation);
 645         store.setRunOnPendingTransactionsComplete(operation);
 646     }
 647
 648     @Override
 649     public String persistenceId() {
 650         return this.name;
 651     }
 652
 653     @VisibleForTesting
 654     ShardCommitCoordinator getCommitCoordinator() {
 655         return commitCoordinator;
 656     }
 657
 658     public DatastoreContext getDatastoreContext() {
 659         return datastoreContext;
 660     }
 661
 662     @VisibleForTesting
 663     public ShardDataTree getDataStore() {
 664         return store;
 665     }
 666
 667     @VisibleForTesting
 668     ShardStats getShardMBean() {
 669         return shardMBean;
 670     }
 671
 672     public static Builder builder() {
 673         return new Builder();
 674     }
 675
 676     public static abstract class AbstractBuilder<T extends AbstractBuilder<T, S>, S extends Shard> {
 677         private final Class<S> shardClass;
 678         private ShardIdentifier id;
 679         private Map<String, String> peerAddresses = Collections.emptyMap();
 680         private DatastoreContext datastoreContext;
 681         private SchemaContext schemaContext;
 682         private DatastoreSnapshot.ShardSnapshot restoreFromSnapshot;
 683         private TipProducingDataTree dataTree;
 684         private volatile boolean sealed;
 685
 686         protected AbstractBuilder(final Class<S> shardClass) {
 687             this.shardClass = shardClass;
 688         }
 689
 690         protected void checkSealed() {
 691             Preconditions.checkState(!sealed, "Builder isalready sealed - further modifications are not allowed");
 692         }
 693
 694         @SuppressWarnings("unchecked")
 695         private T self() {
 696             return (T) this;
 697         }
 698
 699         public T id(final ShardIdentifier id) {
 700             checkSealed();
 701             this.id = id;
 702             return self();
 703         }
 704
 705         public T peerAddresses(final Map<String, String> peerAddresses) {
 706             checkSealed();
 707             this.peerAddresses = peerAddresses;
 708             return self();
 709         }
 710
 711         public T datastoreContext(final DatastoreContext datastoreContext) {
 712             checkSealed();
 713             this.datastoreContext = datastoreContext;
 714             return self();
 715         }
 716
 717         public T schemaContext(final SchemaContext schemaContext) {
 718             checkSealed();
 719             this.schemaContext = schemaContext;
 720             return self();
 721         }
 722
 723         public T restoreFromSnapshot(final DatastoreSnapshot.ShardSnapshot restoreFromSnapshot) {
 724             checkSealed();
 725             this.restoreFromSnapshot = restoreFromSnapshot;
 726             return self();
 727         }
 728
 729         public T dataTree(final TipProducingDataTree dataTree) {
 730             checkSealed();
 731             this.dataTree = dataTree;
 732             return self();
 733         }
 734
 735         public ShardIdentifier getId() {
 736             return id;
 737         }
 738
 739         public Map<String, String> getPeerAddresses() {
 740             return peerAddresses;
 741         }
 742
 743         public DatastoreContext getDatastoreContext() {
 744             return datastoreContext;
 745         }
 746
 747         public SchemaContext getSchemaContext() {
 748             return schemaContext;
 749         }
 750
 751         public DatastoreSnapshot.ShardSnapshot getRestoreFromSnapshot() {
 752             return restoreFromSnapshot;
 753         }
 754
 755         public TipProducingDataTree getDataTree() {
 756             return dataTree;
 757         }
 758
 759         public TreeType getTreeType() {
 760             switch (datastoreContext.getLogicalStoreType()) {
 761             case CONFIGURATION:
 762                 return TreeType.CONFIGURATION;
 763             case OPERATIONAL:
 764                 return TreeType.OPERATIONAL;
 765             }
 766
 767             throw new IllegalStateException("Unhandled logical store type " + datastoreContext.getLogicalStoreType());
 768         }
 769
 770         protected void verify() {
 771             Preconditions.checkNotNull(id, "id should not be null");
 772             Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
 773             Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
 774             Preconditions.checkNotNull(schemaContext, "schemaContext should not be null");
 775         }
 776
 777         public Props props() {
 778             sealed = true;
 779             verify();
 780             return Props.create(shardClass, this);
 781         }
 782     }
 783
 784     public static class Builder extends AbstractBuilder<Builder, Shard> {
 785         private Builder() {
 786             super(Shard.class);
 787         }
 788     }
 789
 790     Ticker ticker() {
 791         return Ticker.systemTicker();
 792     }
 793 }