opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/Shard.java

   1 /*
   2  * Copyright (c) 2014 Cisco Systems, Inc. and others.  All rights reserved.
   3  *
   4  * This program and the accompanying materials are made available under the
   5  * terms of the Eclipse Public License v1.0 which accompanies this distribution,
   6  * and is available at http://www.eclipse.org/legal/epl-v10.html
   7  */
   8
   9 package org.opendaylight.controller.cluster.datastore;
  10
  11 import akka.actor.ActorRef;
  12 import akka.actor.ActorSelection;
  13 import akka.actor.Cancellable;
  14 import akka.actor.Props;
  15 import akka.japi.Creator;
  16 import akka.persistence.RecoveryFailure;
  17 import akka.serialization.Serialization;
  18 import com.google.common.annotations.VisibleForTesting;
  19 import com.google.common.base.Optional;
  20 import com.google.common.base.Preconditions;
  21 import com.google.common.collect.Lists;
  22 import com.google.common.util.concurrent.FutureCallback;
  23 import com.google.common.util.concurrent.Futures;
  24 import com.google.common.util.concurrent.ListenableFuture;
  25 import java.io.IOException;
  26 import java.util.Collection;
  27 import java.util.HashMap;
  28 import java.util.List;
  29 import java.util.Map;
  30 import java.util.concurrent.ExecutionException;
  31 import java.util.concurrent.TimeUnit;
  32 import javax.annotation.Nonnull;
  33 import org.opendaylight.controller.cluster.DataPersistenceProvider;
  34 import org.opendaylight.controller.cluster.common.actor.CommonConfig;
  35 import org.opendaylight.controller.cluster.common.actor.MeteringBehavior;
  36 import org.opendaylight.controller.cluster.datastore.ShardCommitCoordinator.CohortEntry;
  37 import org.opendaylight.controller.cluster.datastore.compat.BackwardsCompatibleThreePhaseCommitCohort;
  38 import org.opendaylight.controller.cluster.datastore.exceptions.NoShardLeaderException;
  39 import org.opendaylight.controller.cluster.datastore.identifiers.ShardIdentifier;
  40 import org.opendaylight.controller.cluster.datastore.identifiers.ShardTransactionIdentifier;
  41 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardMBeanFactory;
  42 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardStats;
  43 import org.opendaylight.controller.cluster.datastore.messages.AbortTransaction;
  44 import org.opendaylight.controller.cluster.datastore.messages.AbortTransactionReply;
  45 import org.opendaylight.controller.cluster.datastore.messages.ActorInitialized;
  46 import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransaction;
  47 import org.opendaylight.controller.cluster.datastore.messages.CloseTransactionChain;
  48 import org.opendaylight.controller.cluster.datastore.messages.CommitTransaction;
  49 import org.opendaylight.controller.cluster.datastore.messages.CommitTransactionReply;
  50 import org.opendaylight.controller.cluster.datastore.messages.CreateSnapshot;
  51 import org.opendaylight.controller.cluster.datastore.messages.CreateTransaction;
  52 import org.opendaylight.controller.cluster.datastore.messages.CreateTransactionReply;
  53 import org.opendaylight.controller.cluster.datastore.messages.EnableNotification;
  54 import org.opendaylight.controller.cluster.datastore.messages.ForwardedReadyTransaction;
  55 import org.opendaylight.controller.cluster.datastore.messages.PeerAddressResolved;
  56 import org.opendaylight.controller.cluster.datastore.messages.ReadyTransactionReply;
  57 import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListener;
  58 import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListenerReply;
  59 import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext;
  60 import org.opendaylight.controller.cluster.datastore.modification.Modification;
  61 import org.opendaylight.controller.cluster.datastore.modification.ModificationPayload;
  62 import org.opendaylight.controller.cluster.datastore.modification.MutableCompositeModification;
  63 import org.opendaylight.controller.cluster.datastore.utils.SerializationUtils;
  64 import org.opendaylight.controller.cluster.notifications.RoleChangeNotifier;
  65 import org.opendaylight.controller.cluster.raft.RaftActor;
  66 import org.opendaylight.controller.cluster.raft.ReplicatedLogEntry;
  67 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationByteStringPayload;
  68 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationPayload;
  69 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload;
  70 import org.opendaylight.controller.md.sal.common.api.data.AsyncDataChangeListener;
  71 import org.opendaylight.controller.md.sal.dom.store.impl.InMemoryDOMDataStore;
  72 import org.opendaylight.controller.md.sal.dom.store.impl.InMemoryDOMDataStoreFactory;
  73 import org.opendaylight.controller.sal.core.spi.data.DOMStoreThreePhaseCommitCohort;
  74 import org.opendaylight.controller.sal.core.spi.data.DOMStoreTransactionChain;
  75 import org.opendaylight.controller.sal.core.spi.data.DOMStoreTransactionFactory;
  76 import org.opendaylight.controller.sal.core.spi.data.DOMStoreWriteTransaction;
  77 import org.opendaylight.yangtools.concepts.ListenerRegistration;
  78 import org.opendaylight.yangtools.yang.data.api.YangInstanceIdentifier;
  79 import org.opendaylight.yangtools.yang.data.api.schema.NormalizedNode;
  80 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
  81 import scala.concurrent.duration.Duration;
  82 import scala.concurrent.duration.FiniteDuration;
  83
  84 /**
  85  * A Shard represents a portion of the logical data tree <br/>
  86  * <p>
  87  * Our Shard uses InMemoryDataStore as it's internal representation and delegates all requests it
  88  * </p>
  89  */
  90 public class Shard extends RaftActor {
  91
  92     private static final YangInstanceIdentifier DATASTORE_ROOT = YangInstanceIdentifier.builder().build();
  93
  94     private static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = "txCommitTimeoutCheck";
  95
  96     @VisibleForTesting
  97     static final String DEFAULT_NAME = "default";
  98
  99     // The state of this Shard
 100     private final InMemoryDOMDataStore store;
 101
 102     /// The name of this shard
 103     private final ShardIdentifier name;
 104
 105     private final ShardStats shardMBean;
 106
 107     private final List<ActorSelection> dataChangeListeners =  Lists.newArrayList();
 108
 109     private final List<DelayedListenerRegistration> delayedListenerRegistrations =
 110                                                                        Lists.newArrayList();
 111
 112     private final DatastoreContext datastoreContext;
 113
 114     private final DataPersistenceProvider dataPersistenceProvider;
 115
 116     private SchemaContext schemaContext;
 117
 118     private int createSnapshotTransactionCounter;
 119
 120     private final ShardCommitCoordinator commitCoordinator;
 121
 122     private final long transactionCommitTimeout;
 123
 124     private Cancellable txCommitTimeoutCheckSchedule;
 125
 126     private final Optional<ActorRef> roleChangeNotifier;
 127
 128     /**
 129      * Coordinates persistence recovery on startup.
 130      */
 131     private ShardRecoveryCoordinator recoveryCoordinator;
 132     private List<Object> currentLogRecoveryBatch;
 133
 134     private final Map<String, DOMStoreTransactionChain> transactionChains = new HashMap<>();
 135
 136     protected Shard(final ShardIdentifier name, final Map<ShardIdentifier, String> peerAddresses,
 137             final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 138         super(name.toString(), mapPeerAddresses(peerAddresses),
 139                 Optional.of(datastoreContext.getShardRaftConfig()));
 140
 141         this.name = name;
 142         this.datastoreContext = datastoreContext;
 143         this.schemaContext = schemaContext;
 144         this.dataPersistenceProvider = (datastoreContext.isPersistent()) ? new PersistentDataProvider() : new NonPersistentRaftDataProvider();
 145
 146         LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
 147
 148         store = InMemoryDOMDataStoreFactory.create(name.toString(), null,
 149                 datastoreContext.getDataStoreProperties());
 150
 151         if(schemaContext != null) {
 152             store.onGlobalContextUpdated(schemaContext);
 153         }
 154
 155         shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
 156                 datastoreContext.getDataStoreMXBeanType());
 157         shardMBean.setNotificationManager(store.getDataChangeListenerNotificationManager());
 158
 159         if (isMetricsCaptureEnabled()) {
 160             getContext().become(new MeteringBehavior(this));
 161         }
 162
 163         commitCoordinator = new ShardCommitCoordinator(TimeUnit.SECONDS.convert(1, TimeUnit.MINUTES),
 164                 datastoreContext.getShardTransactionCommitQueueCapacity(), LOG, name.toString());
 165
 166         transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
 167                 datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS);
 168
 169         // create a notifier actor for each cluster member
 170         roleChangeNotifier = createRoleChangeNotifier(name.toString());
 171     }
 172
 173     private static Map<String, String> mapPeerAddresses(
 174         final Map<ShardIdentifier, String> peerAddresses) {
 175         Map<String, String> map = new HashMap<>();
 176
 177         for (Map.Entry<ShardIdentifier, String> entry : peerAddresses
 178             .entrySet()) {
 179             map.put(entry.getKey().toString(), entry.getValue());
 180         }
 181
 182         return map;
 183     }
 184
 185     public static Props props(final ShardIdentifier name,
 186         final Map<ShardIdentifier, String> peerAddresses,
 187         final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 188         Preconditions.checkNotNull(name, "name should not be null");
 189         Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
 190         Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
 191         Preconditions.checkNotNull(schemaContext, "schemaContext should not be null");
 192
 193         return Props.create(new ShardCreator(name, peerAddresses, datastoreContext, schemaContext));
 194     }
 195
 196     private Optional<ActorRef> createRoleChangeNotifier(String shardId) {
 197         ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
 198             RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
 199         return Optional.<ActorRef>of(shardRoleChangeNotifier);
 200     }
 201
 202     @Override
 203     public void postStop() {
 204         super.postStop();
 205
 206         if(txCommitTimeoutCheckSchedule != null) {
 207             txCommitTimeoutCheckSchedule.cancel();
 208         }
 209     }
 210
 211     @Override
 212     public void onReceiveRecover(final Object message) throws Exception {
 213         if(LOG.isDebugEnabled()) {
 214             LOG.debug("{}: onReceiveRecover: Received message {} from {}", persistenceId(),
 215                 message.getClass().toString(), getSender());
 216         }
 217
 218         if (message instanceof RecoveryFailure){
 219             LOG.error("{}: Recovery failed because of this cause",
 220                     persistenceId(), ((RecoveryFailure) message).cause());
 221
 222             // Even though recovery failed, we still need to finish our recovery, eg send the
 223             // ActorInitialized message and start the txCommitTimeoutCheckSchedule.
 224             onRecoveryComplete();
 225         } else {
 226             super.onReceiveRecover(message);
 227         }
 228     }
 229
 230     @Override
 231     public void onReceiveCommand(final Object message) throws Exception {
 232         if(LOG.isDebugEnabled()) {
 233             LOG.debug("{}: onReceiveCommand: Received message {} from {}", persistenceId(), message, getSender());
 234         }
 235
 236         if (message.getClass().equals(CreateTransaction.SERIALIZABLE_CLASS)) {
 237             handleCreateTransaction(message);
 238         } else if(message instanceof ForwardedReadyTransaction) {
 239             handleForwardedReadyTransaction((ForwardedReadyTransaction)message);
 240         } else if(message.getClass().equals(CanCommitTransaction.SERIALIZABLE_CLASS)) {
 241             handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
 242         } else if(message.getClass().equals(CommitTransaction.SERIALIZABLE_CLASS)) {
 243             handleCommitTransaction(CommitTransaction.fromSerializable(message));
 244         } else if(message.getClass().equals(AbortTransaction.SERIALIZABLE_CLASS)) {
 245             handleAbortTransaction(AbortTransaction.fromSerializable(message));
 246         } else if (message.getClass().equals(CloseTransactionChain.SERIALIZABLE_CLASS)){
 247             closeTransactionChain(CloseTransactionChain.fromSerializable(message));
 248         } else if (message instanceof RegisterChangeListener) {
 249             registerChangeListener((RegisterChangeListener) message);
 250         } else if (message instanceof UpdateSchemaContext) {
 251             updateSchemaContext((UpdateSchemaContext) message);
 252         } else if (message instanceof PeerAddressResolved) {
 253             PeerAddressResolved resolved = (PeerAddressResolved) message;
 254             setPeerAddress(resolved.getPeerId().toString(),
 255                 resolved.getPeerAddress());
 256         } else if(message.equals(TX_COMMIT_TIMEOUT_CHECK_MESSAGE)) {
 257             handleTransactionCommitTimeoutCheck();
 258         } else {
 259             super.onReceiveCommand(message);
 260         }
 261     }
 262
 263     @Override
 264     protected Optional<ActorRef> getRoleChangeNotifier() {
 265         return roleChangeNotifier;
 266     }
 267
 268     private void handleTransactionCommitTimeoutCheck() {
 269         CohortEntry cohortEntry = commitCoordinator.getCurrentCohortEntry();
 270         if(cohortEntry != null) {
 271             long elapsed = System.currentTimeMillis() - cohortEntry.getLastAccessTime();
 272             if(elapsed > transactionCommitTimeout) {
 273                 LOG.warn("{}: Current transaction {} has timed out after {} ms - aborting",
 274                         persistenceId(), cohortEntry.getTransactionID(), transactionCommitTimeout);
 275
 276                 doAbortTransaction(cohortEntry.getTransactionID(), null);
 277             }
 278         }
 279     }
 280
 281     private void handleCommitTransaction(final CommitTransaction commit) {
 282         final String transactionID = commit.getTransactionID();
 283
 284         LOG.debug("{}: Committing transaction {}", persistenceId(), transactionID);
 285
 286         // Get the current in-progress cohort entry in the commitCoordinator if it corresponds to
 287         // this transaction.
 288         final CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
 289         if(cohortEntry == null) {
 290             // We're not the current Tx - the Tx was likely expired b/c it took too long in
 291             // between the canCommit and commit messages.
 292             IllegalStateException ex = new IllegalStateException(
 293                     String.format("%s: Cannot commit transaction %s - it is not the current transaction",
 294                             persistenceId(), transactionID));
 295             LOG.error(ex.getMessage());
 296             shardMBean.incrementFailedTransactionsCount();
 297             getSender().tell(new akka.actor.Status.Failure(ex), getSelf());
 298             return;
 299         }
 300
 301         // We perform the preCommit phase here atomically with the commit phase. This is an
 302         // optimization to eliminate the overhead of an extra preCommit message. We lose front-end
 303         // coordination of preCommit across shards in case of failure but preCommit should not
 304         // normally fail since we ensure only one concurrent 3-phase commit.
 305
 306         try {
 307             // We block on the future here so we don't have to worry about possibly accessing our
 308             // state on a different thread outside of our dispatcher. Also, the data store
 309             // currently uses a same thread executor anyway.
 310             cohortEntry.getCohort().preCommit().get();
 311
 312             // If we do not have any followers and we are not using persistence we can
 313             // apply modification to the state immediately
 314             if(!hasFollowers() && !persistence().isRecoveryApplicable()){
 315                 applyModificationToState(getSender(), transactionID, cohortEntry.getModification());
 316             } else {
 317                 Shard.this.persistData(getSender(), transactionID,
 318                         new ModificationPayload(cohortEntry.getModification()));
 319             }
 320         } catch (Exception e) {
 321             LOG.error("{} An exception occurred while preCommitting transaction {}",
 322                     persistenceId(), cohortEntry.getTransactionID(), e);
 323             shardMBean.incrementFailedTransactionsCount();
 324             getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 325         }
 326
 327         cohortEntry.updateLastAccessTime();
 328     }
 329
 330     private void finishCommit(@Nonnull final ActorRef sender, final @Nonnull String transactionID) {
 331         // With persistence enabled, this method is called via applyState by the leader strategy
 332         // after the commit has been replicated to a majority of the followers.
 333
 334         CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
 335         if(cohortEntry == null) {
 336             // The transaction is no longer the current commit. This can happen if the transaction
 337             // was aborted prior, most likely due to timeout in the front-end. We need to finish
 338             // committing the transaction though since it was successfully persisted and replicated
 339             // however we can't use the original cohort b/c it was already preCommitted and may
 340             // conflict with the current commit or may have been aborted so we commit with a new
 341             // transaction.
 342             cohortEntry = commitCoordinator.getAndRemoveCohortEntry(transactionID);
 343             if(cohortEntry != null) {
 344                 commitWithNewTransaction(cohortEntry.getModification());
 345                 sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
 346             } else {
 347                 // This really shouldn't happen - it likely means that persistence or replication
 348                 // took so long to complete such that the cohort entry was expired from the cache.
 349                 IllegalStateException ex = new IllegalStateException(
 350                         String.format("%s: Could not finish committing transaction %s - no CohortEntry found",
 351                                 persistenceId(), transactionID));
 352                 LOG.error(ex.getMessage());
 353                 sender.tell(new akka.actor.Status.Failure(ex), getSelf());
 354             }
 355
 356             return;
 357         }
 358
 359         LOG.debug("{}: Finishing commit for transaction {}", persistenceId(), cohortEntry.getTransactionID());
 360
 361         try {
 362             // We block on the future here so we don't have to worry about possibly accessing our
 363             // state on a different thread outside of our dispatcher. Also, the data store
 364             // currently uses a same thread executor anyway.
 365             cohortEntry.getCohort().commit().get();
 366
 367             sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
 368
 369             shardMBean.incrementCommittedTransactionCount();
 370             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 371
 372         } catch (Exception e) {
 373             sender.tell(new akka.actor.Status.Failure(e), getSelf());
 374
 375             LOG.error("{}, An exception occurred while committing transaction {}", persistenceId(),
 376                     transactionID, e);
 377             shardMBean.incrementFailedTransactionsCount();
 378         } finally {
 379             commitCoordinator.currentTransactionComplete(transactionID, true);
 380         }
 381     }
 382
 383     private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
 384         LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionID());
 385         commitCoordinator.handleCanCommit(canCommit, getSender(), self());
 386     }
 387
 388     private void handleForwardedReadyTransaction(ForwardedReadyTransaction ready) {
 389         LOG.debug("{}: Readying transaction {}, client version {}", persistenceId(),
 390                 ready.getTransactionID(), ready.getTxnClientVersion());
 391
 392         // This message is forwarded by the ShardTransaction on ready. We cache the cohort in the
 393         // commitCoordinator in preparation for the subsequent three phase commit initiated by
 394         // the front-end.
 395         commitCoordinator.transactionReady(ready.getTransactionID(), ready.getCohort(),
 396                 ready.getModification());
 397
 398         // Return our actor path as we'll handle the three phase commit, except if the Tx client
 399         // version < 1 (Helium-1 version). This means the Tx was initiated by a base Helium version
 400         // node. In that case, the subsequent 3-phase commit messages won't contain the
 401         // transactionId so to maintain backwards compatibility, we create a separate cohort actor
 402         // to provide the compatible behavior.
 403         ActorRef replyActorPath = self();
 404         if(ready.getTxnClientVersion() < DataStoreVersions.HELIUM_1_VERSION) {
 405             LOG.debug("{}: Creating BackwardsCompatibleThreePhaseCommitCohort", persistenceId());
 406             replyActorPath = getContext().actorOf(BackwardsCompatibleThreePhaseCommitCohort.props(
 407                     ready.getTransactionID()));
 408         }
 409
 410         ReadyTransactionReply readyTransactionReply = new ReadyTransactionReply(
 411                 Serialization.serializedActorPath(replyActorPath));
 412         getSender().tell(ready.isReturnSerialized() ? readyTransactionReply.toSerializable() :
 413                 readyTransactionReply, getSelf());
 414     }
 415
 416     private void handleAbortTransaction(final AbortTransaction abort) {
 417         doAbortTransaction(abort.getTransactionID(), getSender());
 418     }
 419
 420     void doAbortTransaction(final String transactionID, final ActorRef sender) {
 421         final CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
 422         if(cohortEntry != null) {
 423             LOG.debug("{}: Aborting transaction {}", persistenceId(), transactionID);
 424
 425             // We don't remove the cached cohort entry here (ie pass false) in case the Tx was
 426             // aborted during replication in which case we may still commit locally if replication
 427             // succeeds.
 428             commitCoordinator.currentTransactionComplete(transactionID, false);
 429
 430             final ListenableFuture<Void> future = cohortEntry.getCohort().abort();
 431             final ActorRef self = getSelf();
 432
 433             Futures.addCallback(future, new FutureCallback<Void>() {
 434                 @Override
 435                 public void onSuccess(final Void v) {
 436                     shardMBean.incrementAbortTransactionsCount();
 437
 438                     if(sender != null) {
 439                         sender.tell(AbortTransactionReply.INSTANCE.toSerializable(), self);
 440                     }
 441                 }
 442
 443                 @Override
 444                 public void onFailure(final Throwable t) {
 445                     LOG.error("{}: An exception happened during abort", persistenceId(), t);
 446
 447                     if(sender != null) {
 448                         sender.tell(new akka.actor.Status.Failure(t), self);
 449                     }
 450                 }
 451             });
 452         }
 453     }
 454
 455     private void handleCreateTransaction(final Object message) {
 456         if (isLeader()) {
 457             createTransaction(CreateTransaction.fromSerializable(message));
 458         } else if (getLeader() != null) {
 459             getLeader().forward(message, getContext());
 460         } else {
 461             getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format(
 462                 "Could not find leader for shard %s so transaction cannot be created. This typically happens" +
 463                 " when the system is coming up or recovering and a leader is being elected. Try again" +
 464                 " later.", persistenceId()))), getSelf());
 465         }
 466     }
 467
 468     private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
 469         DOMStoreTransactionChain chain =
 470             transactionChains.remove(closeTransactionChain.getTransactionChainId());
 471
 472         if(chain != null) {
 473             chain.close();
 474         }
 475     }
 476
 477     private ActorRef createTypedTransactionActor(int transactionType,
 478             ShardTransactionIdentifier transactionId, String transactionChainId,
 479             short clientVersion ) {
 480
 481         DOMStoreTransactionFactory factory = store;
 482
 483         if(!transactionChainId.isEmpty()) {
 484             factory = transactionChains.get(transactionChainId);
 485             if(factory == null){
 486                 DOMStoreTransactionChain transactionChain = store.createTransactionChain();
 487                 transactionChains.put(transactionChainId, transactionChain);
 488                 factory = transactionChain;
 489             }
 490         }
 491
 492         if(this.schemaContext == null) {
 493             throw new IllegalStateException("SchemaContext is not set");
 494         }
 495
 496         if (transactionType == TransactionProxy.TransactionType.READ_ONLY.ordinal()) {
 497
 498             shardMBean.incrementReadOnlyTransactionCount();
 499
 500             return getContext().actorOf(
 501                 ShardTransaction.props(factory.newReadOnlyTransaction(), getSelf(),
 502                         schemaContext,datastoreContext, shardMBean,
 503                         transactionId.getRemoteTransactionId(), clientVersion),
 504                         transactionId.toString());
 505
 506         } else if (transactionType == TransactionProxy.TransactionType.READ_WRITE.ordinal()) {
 507
 508             shardMBean.incrementReadWriteTransactionCount();
 509
 510             return getContext().actorOf(
 511                 ShardTransaction.props(factory.newReadWriteTransaction(), getSelf(),
 512                         schemaContext, datastoreContext, shardMBean,
 513                         transactionId.getRemoteTransactionId(), clientVersion),
 514                         transactionId.toString());
 515
 516
 517         } else if (transactionType == TransactionProxy.TransactionType.WRITE_ONLY.ordinal()) {
 518
 519             shardMBean.incrementWriteOnlyTransactionCount();
 520
 521             return getContext().actorOf(
 522                 ShardTransaction.props(factory.newWriteOnlyTransaction(), getSelf(),
 523                         schemaContext, datastoreContext, shardMBean,
 524                         transactionId.getRemoteTransactionId(), clientVersion),
 525                         transactionId.toString());
 526         } else {
 527             throw new IllegalArgumentException(
 528                 "Shard="+name + ":CreateTransaction message has unidentified transaction type="
 529                     + transactionType);
 530         }
 531     }
 532
 533     private void createTransaction(CreateTransaction createTransaction) {
 534         try {
 535             ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
 536                 createTransaction.getTransactionId(), createTransaction.getTransactionChainId(),
 537                 createTransaction.getVersion());
 538
 539             getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
 540                     createTransaction.getTransactionId()).toSerializable(), getSelf());
 541         } catch (Exception e) {
 542             getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 543         }
 544     }
 545
 546     private ActorRef createTransaction(int transactionType, String remoteTransactionId,
 547             String transactionChainId, short clientVersion) {
 548
 549         ShardTransactionIdentifier transactionId =
 550             ShardTransactionIdentifier.builder()
 551                 .remoteTransactionId(remoteTransactionId)
 552                 .build();
 553
 554         if(LOG.isDebugEnabled()) {
 555             LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
 556         }
 557
 558         ActorRef transactionActor = createTypedTransactionActor(transactionType, transactionId,
 559                 transactionChainId, clientVersion);
 560
 561         return transactionActor;
 562     }
 563
 564     private void syncCommitTransaction(final DOMStoreWriteTransaction transaction)
 565         throws ExecutionException, InterruptedException {
 566         DOMStoreThreePhaseCommitCohort commitCohort = transaction.ready();
 567         commitCohort.preCommit().get();
 568         commitCohort.commit().get();
 569     }
 570
 571     private void commitWithNewTransaction(final Modification modification) {
 572         DOMStoreWriteTransaction tx = store.newWriteOnlyTransaction();
 573         modification.apply(tx);
 574         try {
 575             syncCommitTransaction(tx);
 576             shardMBean.incrementCommittedTransactionCount();
 577             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 578         } catch (InterruptedException | ExecutionException e) {
 579             shardMBean.incrementFailedTransactionsCount();
 580             LOG.error("{}: Failed to commit", persistenceId(), e);
 581         }
 582     }
 583
 584     private void updateSchemaContext(final UpdateSchemaContext message) {
 585         this.schemaContext = message.getSchemaContext();
 586         updateSchemaContext(message.getSchemaContext());
 587         store.onGlobalContextUpdated(message.getSchemaContext());
 588     }
 589
 590     @VisibleForTesting
 591     void updateSchemaContext(final SchemaContext schemaContext) {
 592         store.onGlobalContextUpdated(schemaContext);
 593     }
 594
 595     private void registerChangeListener(final RegisterChangeListener registerChangeListener) {
 596
 597         LOG.debug("{}: registerDataChangeListener for {}", persistenceId(), registerChangeListener.getPath());
 598
 599         ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
 600                                                      NormalizedNode<?, ?>>> registration;
 601         if(isLeader()) {
 602             registration = doChangeListenerRegistration(registerChangeListener);
 603         } else {
 604             LOG.debug("{}: Shard is not the leader - delaying registration", persistenceId());
 605
 606             DelayedListenerRegistration delayedReg =
 607                     new DelayedListenerRegistration(registerChangeListener);
 608             delayedListenerRegistrations.add(delayedReg);
 609             registration = delayedReg;
 610         }
 611
 612         ActorRef listenerRegistration = getContext().actorOf(
 613                 DataChangeListenerRegistration.props(registration));
 614
 615         LOG.debug("{}: registerDataChangeListener sending reply, listenerRegistrationPath = {} ",
 616                 persistenceId(), listenerRegistration.path());
 617
 618         getSender().tell(new RegisterChangeListenerReply(listenerRegistration.path()), getSelf());
 619     }
 620
 621     private ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
 622                                                NormalizedNode<?, ?>>> doChangeListenerRegistration(
 623             final RegisterChangeListener registerChangeListener) {
 624
 625         ActorSelection dataChangeListenerPath = getContext().system().actorSelection(
 626                 registerChangeListener.getDataChangeListenerPath());
 627
 628         // Notify the listener if notifications should be enabled or not
 629         // If this shard is the leader then it will enable notifications else
 630         // it will not
 631         dataChangeListenerPath.tell(new EnableNotification(true), getSelf());
 632
 633         // Now store a reference to the data change listener so it can be notified
 634         // at a later point if notifications should be enabled or disabled
 635         dataChangeListeners.add(dataChangeListenerPath);
 636
 637         AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>> listener =
 638                 new DataChangeListenerProxy(dataChangeListenerPath);
 639
 640         LOG.debug("{}: Registering for path {}", persistenceId(), registerChangeListener.getPath());
 641
 642         return store.registerChangeListener(registerChangeListener.getPath(), listener,
 643                 registerChangeListener.getScope());
 644     }
 645
 646     private boolean isMetricsCaptureEnabled(){
 647         CommonConfig config = new CommonConfig(getContext().system().settings().config());
 648         return config.isMetricCaptureEnabled();
 649     }
 650
 651     @Override
 652     protected
 653     void startLogRecoveryBatch(final int maxBatchSize) {
 654         currentLogRecoveryBatch = Lists.newArrayListWithCapacity(maxBatchSize);
 655
 656         if(LOG.isDebugEnabled()) {
 657             LOG.debug("{}: starting log recovery batch with max size {}", persistenceId(), maxBatchSize);
 658         }
 659     }
 660
 661     @Override
 662     protected void appendRecoveredLogEntry(final Payload data) {
 663         if(data instanceof ModificationPayload) {
 664             try {
 665                 currentLogRecoveryBatch.add(((ModificationPayload) data).getModification());
 666             } catch (ClassNotFoundException | IOException e) {
 667                 LOG.error("{}: Error extracting ModificationPayload", persistenceId(), e);
 668             }
 669         } else if (data instanceof CompositeModificationPayload) {
 670             currentLogRecoveryBatch.add(((CompositeModificationPayload) data).getModification());
 671         } else if (data instanceof CompositeModificationByteStringPayload) {
 672             currentLogRecoveryBatch.add(((CompositeModificationByteStringPayload) data).getModification());
 673         } else {
 674             LOG.error("{}: Unknown state received {} during recovery", persistenceId(), data);
 675         }
 676     }
 677
 678     @Override
 679     protected void applyRecoverySnapshot(final byte[] snapshotBytes) {
 680         if(recoveryCoordinator == null) {
 681             recoveryCoordinator = new ShardRecoveryCoordinator(persistenceId(), schemaContext,
 682                     LOG, name.toString());
 683         }
 684
 685         recoveryCoordinator.submit(snapshotBytes, store.newWriteOnlyTransaction());
 686
 687         if(LOG.isDebugEnabled()) {
 688             LOG.debug("{}: submitted recovery sbapshot", persistenceId());
 689         }
 690     }
 691
 692     @Override
 693     protected void applyCurrentLogRecoveryBatch() {
 694         if(recoveryCoordinator == null) {
 695             recoveryCoordinator = new ShardRecoveryCoordinator(persistenceId(), schemaContext,
 696                     LOG, name.toString());
 697         }
 698
 699         recoveryCoordinator.submit(currentLogRecoveryBatch, store.newWriteOnlyTransaction());
 700
 701         if(LOG.isDebugEnabled()) {
 702             LOG.debug("{}: submitted log recovery batch with size {}", persistenceId(),
 703                     currentLogRecoveryBatch.size());
 704         }
 705     }
 706
 707     @Override
 708     protected void onRecoveryComplete() {
 709         if(recoveryCoordinator != null) {
 710             Collection<DOMStoreWriteTransaction> txList = recoveryCoordinator.getTransactions();
 711
 712             if(LOG.isDebugEnabled()) {
 713                 LOG.debug("{}: recovery complete - committing {} Tx's", persistenceId(), txList.size());
 714             }
 715
 716             for(DOMStoreWriteTransaction tx: txList) {
 717                 try {
 718                     syncCommitTransaction(tx);
 719                     shardMBean.incrementCommittedTransactionCount();
 720                 } catch (InterruptedException | ExecutionException e) {
 721                     shardMBean.incrementFailedTransactionsCount();
 722                     LOG.error("{}: Failed to commit", persistenceId(), e);
 723                 }
 724             }
 725         }
 726
 727         recoveryCoordinator = null;
 728         currentLogRecoveryBatch = null;
 729         updateJournalStats();
 730
 731         //notify shard manager
 732         getContext().parent().tell(new ActorInitialized(), getSelf());
 733
 734         // Being paranoid here - this method should only be called once but just in case...
 735         if(txCommitTimeoutCheckSchedule == null) {
 736             // Schedule a message to be periodically sent to check if the current in-progress
 737             // transaction should be expired and aborted.
 738             FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
 739             txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
 740                     period, period, getSelf(),
 741                     TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
 742         }
 743     }
 744
 745     @Override
 746     protected void applyState(final ActorRef clientActor, final String identifier, final Object data) {
 747
 748         if(data instanceof ModificationPayload) {
 749             try {
 750                 applyModificationToState(clientActor, identifier, ((ModificationPayload) data).getModification());
 751             } catch (ClassNotFoundException | IOException e) {
 752                 LOG.error("{}: Error extracting ModificationPayload", persistenceId(), e);
 753             }
 754         }
 755         else if (data instanceof CompositeModificationPayload) {
 756             Object modification = ((CompositeModificationPayload) data).getModification();
 757
 758             applyModificationToState(clientActor, identifier, modification);
 759         } else if(data instanceof CompositeModificationByteStringPayload ){
 760             Object modification = ((CompositeModificationByteStringPayload) data).getModification();
 761
 762             applyModificationToState(clientActor, identifier, modification);
 763         } else {
 764             LOG.error("{}: Unknown state received {} Class loader = {} CompositeNodeMod.ClassLoader = {}",
 765                     persistenceId(), data, data.getClass().getClassLoader(),
 766                     CompositeModificationPayload.class.getClassLoader());
 767         }
 768
 769         updateJournalStats();
 770
 771     }
 772
 773     private void applyModificationToState(ActorRef clientActor, String identifier, Object modification) {
 774         if(modification == null) {
 775             LOG.error(
 776                     "{}: modification is null - this is very unexpected, clientActor = {}, identifier = {}",
 777                     persistenceId(), identifier, clientActor != null ? clientActor.path().toString() : null);
 778         } else if(clientActor == null) {
 779             // There's no clientActor to which to send a commit reply so we must be applying
 780             // replicated state from the leader.
 781             commitWithNewTransaction(MutableCompositeModification.fromSerializable(modification));
 782         } else {
 783             // This must be the OK to commit after replication consensus.
 784             finishCommit(clientActor, identifier);
 785         }
 786     }
 787
 788     private void updateJournalStats() {
 789         ReplicatedLogEntry lastLogEntry = getLastLogEntry();
 790
 791         if (lastLogEntry != null) {
 792             shardMBean.setLastLogIndex(lastLogEntry.getIndex());
 793             shardMBean.setLastLogTerm(lastLogEntry.getTerm());
 794         }
 795
 796         shardMBean.setCommitIndex(getCommitIndex());
 797         shardMBean.setLastApplied(getLastApplied());
 798         shardMBean.setInMemoryJournalDataSize(getRaftActorContext().getReplicatedLog().dataSize());
 799     }
 800
 801     @Override
 802     protected void createSnapshot() {
 803         // Create a transaction actor. We are really going to treat the transaction as a worker
 804         // so that this actor does not get block building the snapshot. THe transaction actor will
 805         // after processing the CreateSnapshot message.
 806
 807         ActorRef createSnapshotTransaction = createTransaction(
 808                 TransactionProxy.TransactionType.READ_ONLY.ordinal(),
 809                 "createSnapshot" + ++createSnapshotTransactionCounter, "",
 810                 DataStoreVersions.CURRENT_VERSION);
 811
 812         createSnapshotTransaction.tell(CreateSnapshot.INSTANCE, self());
 813     }
 814
 815     @VisibleForTesting
 816     @Override
 817     protected void applySnapshot(final byte[] snapshotBytes) {
 818         // Since this will be done only on Recovery or when this actor is a Follower
 819         // we can safely commit everything in here. We not need to worry about event notifications
 820         // as they would have already been disabled on the follower
 821
 822         LOG.info("{}: Applying snapshot", persistenceId());
 823         try {
 824             DOMStoreWriteTransaction transaction = store.newWriteOnlyTransaction();
 825
 826             NormalizedNode<?, ?> node = SerializationUtils.deserializeNormalizedNode(snapshotBytes);
 827
 828             // delete everything first
 829             transaction.delete(DATASTORE_ROOT);
 830
 831             // Add everything from the remote node back
 832             transaction.write(DATASTORE_ROOT, node);
 833             syncCommitTransaction(transaction);
 834         } catch (InterruptedException | ExecutionException e) {
 835             LOG.error("{}: An exception occurred when applying snapshot", persistenceId(), e);
 836         } finally {
 837             LOG.info("{}: Done applying snapshot", persistenceId());
 838         }
 839     }
 840
 841     @Override
 842     protected void onStateChanged() {
 843         boolean isLeader = isLeader();
 844         for (ActorSelection dataChangeListener : dataChangeListeners) {
 845             dataChangeListener.tell(new EnableNotification(isLeader), getSelf());
 846         }
 847
 848         if(isLeader) {
 849             for(DelayedListenerRegistration reg: delayedListenerRegistrations) {
 850                 if(!reg.isClosed()) {
 851                     reg.setDelegate(doChangeListenerRegistration(reg.getRegisterChangeListener()));
 852                 }
 853             }
 854
 855             delayedListenerRegistrations.clear();
 856         }
 857
 858         shardMBean.setRaftState(getRaftState().name());
 859         shardMBean.setCurrentTerm(getCurrentTerm());
 860
 861         // If this actor is no longer the leader close all the transaction chains
 862         if(!isLeader){
 863             for(Map.Entry<String, DOMStoreTransactionChain> entry : transactionChains.entrySet()){
 864                 if(LOG.isDebugEnabled()) {
 865                     LOG.debug(
 866                         "{}: onStateChanged: Closing transaction chain {} because shard {} is no longer the leader",
 867                         persistenceId(), entry.getKey(), getId());
 868                 }
 869                 entry.getValue().close();
 870             }
 871
 872             transactionChains.clear();
 873         }
 874     }
 875
 876     @Override
 877     protected DataPersistenceProvider persistence() {
 878         return dataPersistenceProvider;
 879     }
 880
 881     @Override protected void onLeaderChanged(final String oldLeader, final String newLeader) {
 882         shardMBean.setLeader(newLeader);
 883     }
 884
 885     @Override public String persistenceId() {
 886         return this.name.toString();
 887     }
 888
 889     @VisibleForTesting
 890     DataPersistenceProvider getDataPersistenceProvider() {
 891         return dataPersistenceProvider;
 892     }
 893
 894     private static class ShardCreator implements Creator<Shard> {
 895
 896         private static final long serialVersionUID = 1L;
 897
 898         final ShardIdentifier name;
 899         final Map<ShardIdentifier, String> peerAddresses;
 900         final DatastoreContext datastoreContext;
 901         final SchemaContext schemaContext;
 902
 903         ShardCreator(final ShardIdentifier name, final Map<ShardIdentifier, String> peerAddresses,
 904                 final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 905             this.name = name;
 906             this.peerAddresses = peerAddresses;
 907             this.datastoreContext = datastoreContext;
 908             this.schemaContext = schemaContext;
 909         }
 910
 911         @Override
 912         public Shard create() throws Exception {
 913             return new Shard(name, peerAddresses, datastoreContext, schemaContext);
 914         }
 915     }
 916
 917     @VisibleForTesting
 918     InMemoryDOMDataStore getDataStore() {
 919         return store;
 920     }
 921
 922     @VisibleForTesting
 923     ShardStats getShardMBean() {
 924         return shardMBean;
 925     }
 926
 927     private static class DelayedListenerRegistration implements
 928         ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>>> {
 929
 930         private volatile boolean closed;
 931
 932         private final RegisterChangeListener registerChangeListener;
 933
 934         private volatile ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
 935                                                              NormalizedNode<?, ?>>> delegate;
 936
 937         DelayedListenerRegistration(final RegisterChangeListener registerChangeListener) {
 938             this.registerChangeListener = registerChangeListener;
 939         }
 940
 941         void setDelegate( final ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
 942                                             NormalizedNode<?, ?>>> registration) {
 943             this.delegate = registration;
 944         }
 945
 946         boolean isClosed() {
 947             return closed;
 948         }
 949
 950         RegisterChangeListener getRegisterChangeListener() {
 951             return registerChangeListener;
 952         }
 953
 954         @Override
 955         public AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>> getInstance() {
 956             return delegate != null ? delegate.getInstance() : null;
 957         }
 958
 959         @Override
 960         public void close() {
 961             closed = true;
 962             if(delegate != null) {
 963                 delegate.close();
 964             }
 965         }
 966     }
 967 }