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