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.PoisonPill;
  15 import akka.actor.Props;
  16 import akka.event.Logging;
  17 import akka.event.LoggingAdapter;
  18 import akka.japi.Creator;
  19 import akka.persistence.RecoveryFailure;
  20 import akka.serialization.Serialization;
  21 import com.google.common.annotations.VisibleForTesting;
  22 import com.google.common.base.Optional;
  23 import com.google.common.base.Preconditions;
  24 import com.google.common.collect.Lists;
  25 import com.google.common.util.concurrent.FutureCallback;
  26 import com.google.common.util.concurrent.Futures;
  27 import com.google.common.util.concurrent.ListenableFuture;
  28 import com.google.protobuf.ByteString;
  29 import com.google.protobuf.InvalidProtocolBufferException;
  30 import java.util.Collection;
  31 import java.util.HashMap;
  32 import java.util.List;
  33 import java.util.Map;
  34 import java.util.concurrent.ExecutionException;
  35 import java.util.concurrent.TimeUnit;
  36 import javax.annotation.Nonnull;
  37 import org.opendaylight.controller.cluster.DataPersistenceProvider;
  38 import org.opendaylight.controller.cluster.common.actor.CommonConfig;
  39 import org.opendaylight.controller.cluster.common.actor.MeteringBehavior;
  40 import org.opendaylight.controller.cluster.datastore.ShardCommitCoordinator.CohortEntry;
  41 import org.opendaylight.controller.cluster.datastore.compat.BackwardsCompatibleThreePhaseCommitCohort;
  42 import org.opendaylight.controller.cluster.datastore.exceptions.NoShardLeaderException;
  43 import org.opendaylight.controller.cluster.datastore.identifiers.ShardIdentifier;
  44 import org.opendaylight.controller.cluster.datastore.identifiers.ShardTransactionIdentifier;
  45 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardMBeanFactory;
  46 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardStats;
  47 import org.opendaylight.controller.cluster.datastore.messages.AbortTransaction;
  48 import org.opendaylight.controller.cluster.datastore.messages.AbortTransactionReply;
  49 import org.opendaylight.controller.cluster.datastore.messages.ActorInitialized;
  50 import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransaction;
  51 import org.opendaylight.controller.cluster.datastore.messages.CloseTransactionChain;
  52 import org.opendaylight.controller.cluster.datastore.messages.CommitTransaction;
  53 import org.opendaylight.controller.cluster.datastore.messages.CommitTransactionReply;
  54 import org.opendaylight.controller.cluster.datastore.messages.CreateTransaction;
  55 import org.opendaylight.controller.cluster.datastore.messages.CreateTransactionReply;
  56 import org.opendaylight.controller.cluster.datastore.messages.EnableNotification;
  57 import org.opendaylight.controller.cluster.datastore.messages.ForwardedReadyTransaction;
  58 import org.opendaylight.controller.cluster.datastore.messages.PeerAddressResolved;
  59 import org.opendaylight.controller.cluster.datastore.messages.ReadData;
  60 import org.opendaylight.controller.cluster.datastore.messages.ReadDataReply;
  61 import org.opendaylight.controller.cluster.datastore.messages.ReadyTransactionReply;
  62 import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListener;
  63 import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListenerReply;
  64 import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext;
  65 import org.opendaylight.controller.cluster.datastore.modification.Modification;
  66 import org.opendaylight.controller.cluster.datastore.modification.MutableCompositeModification;
  67 import org.opendaylight.controller.cluster.datastore.node.NormalizedNodeToNodeCodec;
  68 import org.opendaylight.controller.cluster.notifications.RoleChangeNotifier;
  69 import org.opendaylight.controller.cluster.raft.RaftActor;
  70 import org.opendaylight.controller.cluster.raft.ReplicatedLogEntry;
  71 import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshotReply;
  72 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationByteStringPayload;
  73 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationPayload;
  74 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload;
  75 import org.opendaylight.controller.md.sal.common.api.data.AsyncDataChangeListener;
  76 import org.opendaylight.controller.md.sal.dom.store.impl.InMemoryDOMDataStore;
  77 import org.opendaylight.controller.md.sal.dom.store.impl.InMemoryDOMDataStoreFactory;
  78 import org.opendaylight.controller.protobuff.messages.common.NormalizedNodeMessages;
  79 import org.opendaylight.controller.sal.core.spi.data.DOMStoreThreePhaseCommitCohort;
  80 import org.opendaylight.controller.sal.core.spi.data.DOMStoreTransactionChain;
  81 import org.opendaylight.controller.sal.core.spi.data.DOMStoreTransactionFactory;
  82 import org.opendaylight.controller.sal.core.spi.data.DOMStoreWriteTransaction;
  83 import org.opendaylight.yangtools.concepts.ListenerRegistration;
  84 import org.opendaylight.yangtools.yang.data.api.YangInstanceIdentifier;
  85 import org.opendaylight.yangtools.yang.data.api.schema.NormalizedNode;
  86 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
  87 import scala.concurrent.duration.Duration;
  88 import scala.concurrent.duration.FiniteDuration;
  89
  90 /**
  91  * A Shard represents a portion of the logical data tree <br/>
  92  * <p>
  93  * Our Shard uses InMemoryDataStore as it's internal representation and delegates all requests it
  94  * </p>
  95  */
  96 public class Shard extends RaftActor {
  97
  98     private static final Object COMMIT_TRANSACTION_REPLY = new CommitTransactionReply().toSerializable();
  99
 100     private static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = "txCommitTimeoutCheck";
 101
 102     public static final String DEFAULT_NAME = "default";
 103
 104     // The state of this Shard
 105     private final InMemoryDOMDataStore store;
 106
 107     private final LoggingAdapter LOG =
 108         Logging.getLogger(getContext().system(), this);
 109
 110     /// The name of this shard
 111     private final ShardIdentifier name;
 112
 113     private final ShardStats shardMBean;
 114
 115     private final List<ActorSelection> dataChangeListeners =  Lists.newArrayList();
 116
 117     private final List<DelayedListenerRegistration> delayedListenerRegistrations =
 118                                                                        Lists.newArrayList();
 119
 120     private final DatastoreContext datastoreContext;
 121
 122     private final DataPersistenceProvider dataPersistenceProvider;
 123
 124     private SchemaContext schemaContext;
 125
 126     private ActorRef createSnapshotTransaction;
 127
 128     private int createSnapshotTransactionCounter;
 129
 130     private final ShardCommitCoordinator commitCoordinator;
 131
 132     private final long transactionCommitTimeout;
 133
 134     private Cancellable txCommitTimeoutCheckSchedule;
 135
 136     private Optional<ActorRef> roleChangeNotifier;
 137
 138     /**
 139      * Coordinates persistence recovery on startup.
 140      */
 141     private ShardRecoveryCoordinator recoveryCoordinator;
 142     private List<Object> currentLogRecoveryBatch;
 143
 144     private final Map<String, DOMStoreTransactionChain> transactionChains = new HashMap<>();
 145
 146     protected Shard(final ShardIdentifier name, final Map<ShardIdentifier, String> peerAddresses,
 147             final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 148         super(name.toString(), mapPeerAddresses(peerAddresses),
 149                 Optional.of(datastoreContext.getShardRaftConfig()));
 150
 151         this.name = name;
 152         this.datastoreContext = datastoreContext;
 153         this.schemaContext = schemaContext;
 154         this.dataPersistenceProvider = (datastoreContext.isPersistent()) ? new PersistentDataProvider() : new NonPersistentRaftDataProvider();
 155
 156         LOG.info("Shard created : {} persistent : {}", name, datastoreContext.isPersistent());
 157
 158         store = InMemoryDOMDataStoreFactory.create(name.toString(), null,
 159                 datastoreContext.getDataStoreProperties());
 160
 161         if(schemaContext != null) {
 162             store.onGlobalContextUpdated(schemaContext);
 163         }
 164
 165         shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
 166                 datastoreContext.getDataStoreMXBeanType());
 167         shardMBean.setNotificationManager(store.getDataChangeListenerNotificationManager());
 168
 169         if (isMetricsCaptureEnabled()) {
 170             getContext().become(new MeteringBehavior(this));
 171         }
 172
 173         commitCoordinator = new ShardCommitCoordinator(TimeUnit.SECONDS.convert(1, TimeUnit.MINUTES),
 174                 datastoreContext.getShardTransactionCommitQueueCapacity());
 175
 176         transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
 177                 datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS);
 178
 179         // create a notifier actor for each cluster member
 180         roleChangeNotifier = createRoleChangeNotifier(name.toString());
 181     }
 182
 183     private static Map<String, String> mapPeerAddresses(
 184         final Map<ShardIdentifier, String> peerAddresses) {
 185         Map<String, String> map = new HashMap<>();
 186
 187         for (Map.Entry<ShardIdentifier, String> entry : peerAddresses
 188             .entrySet()) {
 189             map.put(entry.getKey().toString(), entry.getValue());
 190         }
 191
 192         return map;
 193     }
 194
 195     public static Props props(final ShardIdentifier name,
 196         final Map<ShardIdentifier, String> peerAddresses,
 197         final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 198         Preconditions.checkNotNull(name, "name should not be null");
 199         Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
 200         Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
 201         Preconditions.checkNotNull(schemaContext, "schemaContext should not be null");
 202
 203         return Props.create(new ShardCreator(name, peerAddresses, datastoreContext, schemaContext));
 204     }
 205
 206     private Optional<ActorRef> createRoleChangeNotifier(String shardId) {
 207         ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
 208             RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
 209         return Optional.<ActorRef>of(shardRoleChangeNotifier);
 210     }
 211
 212     @Override
 213     public void postStop() {
 214         super.postStop();
 215
 216         if(txCommitTimeoutCheckSchedule != null) {
 217             txCommitTimeoutCheckSchedule.cancel();
 218         }
 219     }
 220
 221     @Override
 222     public void onReceiveRecover(final Object message) throws Exception {
 223         if(LOG.isDebugEnabled()) {
 224             LOG.debug("onReceiveRecover: Received message {} from {}",
 225                 message.getClass().toString(),
 226                 getSender());
 227         }
 228
 229         if (message instanceof RecoveryFailure){
 230             LOG.error(((RecoveryFailure) message).cause(), "Recovery failed because of this cause");
 231
 232             // Even though recovery failed, we still need to finish our recovery, eg send the
 233             // ActorInitialized message and start the txCommitTimeoutCheckSchedule.
 234             onRecoveryComplete();
 235         } else {
 236             super.onReceiveRecover(message);
 237         }
 238     }
 239
 240     @Override
 241     public void onReceiveCommand(final Object message) throws Exception {
 242         if(LOG.isDebugEnabled()) {
 243             LOG.debug("onReceiveCommand: Received message {} from {}", message, getSender());
 244         }
 245
 246         if(message.getClass().equals(ReadDataReply.SERIALIZABLE_CLASS)) {
 247             handleReadDataReply(message);
 248         } else if (message.getClass().equals(CreateTransaction.SERIALIZABLE_CLASS)) {
 249             handleCreateTransaction(message);
 250         } else if(message instanceof ForwardedReadyTransaction) {
 251             handleForwardedReadyTransaction((ForwardedReadyTransaction)message);
 252         } else if(message.getClass().equals(CanCommitTransaction.SERIALIZABLE_CLASS)) {
 253             handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
 254         } else if(message.getClass().equals(CommitTransaction.SERIALIZABLE_CLASS)) {
 255             handleCommitTransaction(CommitTransaction.fromSerializable(message));
 256         } else if(message.getClass().equals(AbortTransaction.SERIALIZABLE_CLASS)) {
 257             handleAbortTransaction(AbortTransaction.fromSerializable(message));
 258         } else if (message.getClass().equals(CloseTransactionChain.SERIALIZABLE_CLASS)){
 259             closeTransactionChain(CloseTransactionChain.fromSerializable(message));
 260         } else if (message instanceof RegisterChangeListener) {
 261             registerChangeListener((RegisterChangeListener) message);
 262         } else if (message instanceof UpdateSchemaContext) {
 263             updateSchemaContext((UpdateSchemaContext) message);
 264         } else if (message instanceof PeerAddressResolved) {
 265             PeerAddressResolved resolved = (PeerAddressResolved) message;
 266             setPeerAddress(resolved.getPeerId().toString(),
 267                 resolved.getPeerAddress());
 268         } else if(message.equals(TX_COMMIT_TIMEOUT_CHECK_MESSAGE)) {
 269             handleTransactionCommitTimeoutCheck();
 270         } else {
 271             super.onReceiveCommand(message);
 272         }
 273     }
 274
 275     @Override
 276     protected Optional<ActorRef> getRoleChangeNotifier() {
 277         return roleChangeNotifier;
 278     }
 279
 280     private void handleTransactionCommitTimeoutCheck() {
 281         CohortEntry cohortEntry = commitCoordinator.getCurrentCohortEntry();
 282         if(cohortEntry != null) {
 283             long elapsed = System.currentTimeMillis() - cohortEntry.getLastAccessTime();
 284             if(elapsed > transactionCommitTimeout) {
 285                 LOG.warning("Current transaction {} has timed out after {} ms - aborting",
 286                         cohortEntry.getTransactionID(), transactionCommitTimeout);
 287
 288                 doAbortTransaction(cohortEntry.getTransactionID(), null);
 289             }
 290         }
 291     }
 292
 293     private void handleCommitTransaction(final CommitTransaction commit) {
 294         final String transactionID = commit.getTransactionID();
 295
 296         LOG.debug("Committing transaction {}", transactionID);
 297
 298         // Get the current in-progress cohort entry in the commitCoordinator if it corresponds to
 299         // this transaction.
 300         final CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
 301         if(cohortEntry == null) {
 302             // We're not the current Tx - the Tx was likely expired b/c it took too long in
 303             // between the canCommit and commit messages.
 304             IllegalStateException ex = new IllegalStateException(
 305                     String.format("Cannot commit transaction %s - it is not the current transaction",
 306                             transactionID));
 307             LOG.error(ex.getMessage());
 308             shardMBean.incrementFailedTransactionsCount();
 309             getSender().tell(new akka.actor.Status.Failure(ex), getSelf());
 310             return;
 311         }
 312
 313         // We perform the preCommit phase here atomically with the commit phase. This is an
 314         // optimization to eliminate the overhead of an extra preCommit message. We lose front-end
 315         // coordination of preCommit across shards in case of failure but preCommit should not
 316         // normally fail since we ensure only one concurrent 3-phase commit.
 317
 318         try {
 319             // We block on the future here so we don't have to worry about possibly accessing our
 320             // state on a different thread outside of our dispatcher. Also, the data store
 321             // currently uses a same thread executor anyway.
 322             cohortEntry.getCohort().preCommit().get();
 323
 324             // If we do not have any followers and we are not using persistence we can
 325             // apply modification to the state immediately
 326             if(!hasFollowers() && !persistence().isRecoveryApplicable()){
 327                 applyModificationToState(getSender(), transactionID, cohortEntry.getModification());
 328             } else {
 329                 Shard.this.persistData(getSender(), transactionID,
 330                         new CompositeModificationByteStringPayload(cohortEntry.getModification().toSerializable()));
 331             }
 332         } catch (InterruptedException | ExecutionException e) {
 333             LOG.error(e, "An exception occurred while preCommitting transaction {}",
 334                     cohortEntry.getTransactionID());
 335             shardMBean.incrementFailedTransactionsCount();
 336             getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 337         }
 338
 339         cohortEntry.updateLastAccessTime();
 340     }
 341
 342     private void finishCommit(@Nonnull final ActorRef sender, final @Nonnull String transactionID) {
 343         // With persistence enabled, this method is called via applyState by the leader strategy
 344         // after the commit has been replicated to a majority of the followers.
 345
 346         CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
 347         if(cohortEntry == null) {
 348             // The transaction is no longer the current commit. This can happen if the transaction
 349             // was aborted prior, most likely due to timeout in the front-end. We need to finish
 350             // committing the transaction though since it was successfully persisted and replicated
 351             // however we can't use the original cohort b/c it was already preCommitted and may
 352             // conflict with the current commit or may have been aborted so we commit with a new
 353             // transaction.
 354             cohortEntry = commitCoordinator.getAndRemoveCohortEntry(transactionID);
 355             if(cohortEntry != null) {
 356                 commitWithNewTransaction(cohortEntry.getModification());
 357                 sender.tell(COMMIT_TRANSACTION_REPLY, getSelf());
 358             } else {
 359                 // This really shouldn't happen - it likely means that persistence or replication
 360                 // took so long to complete such that the cohort entry was expired from the cache.
 361                 IllegalStateException ex = new IllegalStateException(
 362                         String.format("Could not finish committing transaction %s - no CohortEntry found",
 363                                 transactionID));
 364                 LOG.error(ex.getMessage());
 365                 sender.tell(new akka.actor.Status.Failure(ex), getSelf());
 366             }
 367
 368             return;
 369         }
 370
 371         LOG.debug("Finishing commit for transaction {}", cohortEntry.getTransactionID());
 372
 373         try {
 374             // We block on the future here so we don't have to worry about possibly accessing our
 375             // state on a different thread outside of our dispatcher. Also, the data store
 376             // currently uses a same thread executor anyway.
 377             cohortEntry.getCohort().commit().get();
 378
 379             sender.tell(COMMIT_TRANSACTION_REPLY, getSelf());
 380
 381             shardMBean.incrementCommittedTransactionCount();
 382             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 383
 384         } catch (InterruptedException | ExecutionException e) {
 385             sender.tell(new akka.actor.Status.Failure(e), getSelf());
 386
 387             LOG.error(e, "An exception occurred while committing transaction {}", transactionID);
 388             shardMBean.incrementFailedTransactionsCount();
 389         }
 390
 391         commitCoordinator.currentTransactionComplete(transactionID, true);
 392     }
 393
 394     private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
 395         LOG.debug("Can committing transaction {}", canCommit.getTransactionID());
 396         commitCoordinator.handleCanCommit(canCommit, getSender(), self());
 397     }
 398
 399     private void handleForwardedReadyTransaction(ForwardedReadyTransaction ready) {
 400         LOG.debug("Readying transaction {}, client version {}", ready.getTransactionID(),
 401                 ready.getTxnClientVersion());
 402
 403         // This message is forwarded by the ShardTransaction on ready. We cache the cohort in the
 404         // commitCoordinator in preparation for the subsequent three phase commit initiated by
 405         // the front-end.
 406         commitCoordinator.transactionReady(ready.getTransactionID(), ready.getCohort(),
 407                 ready.getModification());
 408
 409         // Return our actor path as we'll handle the three phase commit, except if the Tx client
 410         // version < 1 (Helium-1 version). This means the Tx was initiated by a base Helium version
 411         // node. In that case, the subsequent 3-phase commit messages won't contain the
 412         // transactionId so to maintain backwards compatibility, we create a separate cohort actor
 413         // to provide the compatible behavior.
 414         ActorRef replyActorPath = self();
 415         if(ready.getTxnClientVersion() < CreateTransaction.HELIUM_1_VERSION) {
 416             LOG.debug("Creating BackwardsCompatibleThreePhaseCommitCohort");
 417             replyActorPath = getContext().actorOf(BackwardsCompatibleThreePhaseCommitCohort.props(
 418                     ready.getTransactionID()));
 419         }
 420
 421         ReadyTransactionReply readyTransactionReply = new ReadyTransactionReply(
 422                 Serialization.serializedActorPath(replyActorPath));
 423         getSender().tell(ready.isReturnSerialized() ? readyTransactionReply.toSerializable() :
 424                 readyTransactionReply, getSelf());
 425     }
 426
 427     private void handleAbortTransaction(final AbortTransaction abort) {
 428         doAbortTransaction(abort.getTransactionID(), getSender());
 429     }
 430
 431     private void doAbortTransaction(final String transactionID, final ActorRef sender) {
 432         final CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
 433         if(cohortEntry != null) {
 434             LOG.debug("Aborting transaction {}", transactionID);
 435
 436             // We don't remove the cached cohort entry here (ie pass false) in case the Tx was
 437             // aborted during replication in which case we may still commit locally if replication
 438             // succeeds.
 439             commitCoordinator.currentTransactionComplete(transactionID, false);
 440
 441             final ListenableFuture<Void> future = cohortEntry.getCohort().abort();
 442             final ActorRef self = getSelf();
 443
 444             Futures.addCallback(future, new FutureCallback<Void>() {
 445                 @Override
 446                 public void onSuccess(final Void v) {
 447                     shardMBean.incrementAbortTransactionsCount();
 448
 449                     if(sender != null) {
 450                         sender.tell(new AbortTransactionReply().toSerializable(), self);
 451                     }
 452                 }
 453
 454                 @Override
 455                 public void onFailure(final Throwable t) {
 456                     LOG.error(t, "An exception happened during abort");
 457
 458                     if(sender != null) {
 459                         sender.tell(new akka.actor.Status.Failure(t), self);
 460                     }
 461                 }
 462             });
 463         }
 464     }
 465
 466     private void handleCreateTransaction(final Object message) {
 467         if (isLeader()) {
 468             createTransaction(CreateTransaction.fromSerializable(message));
 469         } else if (getLeader() != null) {
 470             getLeader().forward(message, getContext());
 471         } else {
 472             getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(
 473                 "Could not find shard leader so transaction cannot be created. This typically happens" +
 474                 " when the system is coming up or recovering and a leader is being elected. Try again" +
 475                 " later.")), getSelf());
 476         }
 477     }
 478
 479     private void handleReadDataReply(final Object message) {
 480         // This must be for install snapshot. Don't want to open this up and trigger
 481         // deSerialization
 482
 483         self().tell(new CaptureSnapshotReply(ReadDataReply.getNormalizedNodeByteString(message)),
 484                 self());
 485
 486         createSnapshotTransaction = null;
 487
 488         // Send a PoisonPill instead of sending close transaction because we do not really need
 489         // a response
 490         getSender().tell(PoisonPill.getInstance(), self());
 491     }
 492
 493     private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
 494         DOMStoreTransactionChain chain =
 495             transactionChains.remove(closeTransactionChain.getTransactionChainId());
 496
 497         if(chain != null) {
 498             chain.close();
 499         }
 500     }
 501
 502     private ActorRef createTypedTransactionActor(int transactionType,
 503             ShardTransactionIdentifier transactionId, String transactionChainId, int clientVersion ) {
 504
 505         DOMStoreTransactionFactory factory = store;
 506
 507         if(!transactionChainId.isEmpty()) {
 508             factory = transactionChains.get(transactionChainId);
 509             if(factory == null){
 510                 DOMStoreTransactionChain transactionChain = store.createTransactionChain();
 511                 transactionChains.put(transactionChainId, transactionChain);
 512                 factory = transactionChain;
 513             }
 514         }
 515
 516         if(this.schemaContext == null) {
 517             throw new IllegalStateException("SchemaContext is not set");
 518         }
 519
 520         if (transactionType == TransactionProxy.TransactionType.READ_ONLY.ordinal()) {
 521
 522             shardMBean.incrementReadOnlyTransactionCount();
 523
 524             return getContext().actorOf(
 525                 ShardTransaction.props(factory.newReadOnlyTransaction(), getSelf(),
 526                         schemaContext,datastoreContext, shardMBean,
 527                         transactionId.getRemoteTransactionId(), clientVersion),
 528                         transactionId.toString());
 529
 530         } else if (transactionType == TransactionProxy.TransactionType.READ_WRITE.ordinal()) {
 531
 532             shardMBean.incrementReadWriteTransactionCount();
 533
 534             return getContext().actorOf(
 535                 ShardTransaction.props(factory.newReadWriteTransaction(), getSelf(),
 536                         schemaContext, datastoreContext, shardMBean,
 537                         transactionId.getRemoteTransactionId(), clientVersion),
 538                         transactionId.toString());
 539
 540
 541         } else if (transactionType == TransactionProxy.TransactionType.WRITE_ONLY.ordinal()) {
 542
 543             shardMBean.incrementWriteOnlyTransactionCount();
 544
 545             return getContext().actorOf(
 546                 ShardTransaction.props(factory.newWriteOnlyTransaction(), getSelf(),
 547                         schemaContext, datastoreContext, shardMBean,
 548                         transactionId.getRemoteTransactionId(), clientVersion),
 549                         transactionId.toString());
 550         } else {
 551             throw new IllegalArgumentException(
 552                 "Shard="+name + ":CreateTransaction message has unidentified transaction type="
 553                     + transactionType);
 554         }
 555     }
 556
 557     private void createTransaction(CreateTransaction createTransaction) {
 558         try {
 559             ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
 560                 createTransaction.getTransactionId(), createTransaction.getTransactionChainId(),
 561                 createTransaction.getVersion());
 562
 563             getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
 564                     createTransaction.getTransactionId()).toSerializable(), getSelf());
 565         } catch (Exception e) {
 566             getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 567         }
 568     }
 569
 570     private ActorRef createTransaction(int transactionType, String remoteTransactionId,
 571             String transactionChainId, int clientVersion) {
 572
 573         ShardTransactionIdentifier transactionId =
 574             ShardTransactionIdentifier.builder()
 575                 .remoteTransactionId(remoteTransactionId)
 576                 .build();
 577
 578         if(LOG.isDebugEnabled()) {
 579             LOG.debug("Creating transaction : {} ", transactionId);
 580         }
 581
 582         ActorRef transactionActor = createTypedTransactionActor(transactionType, transactionId,
 583                 transactionChainId, clientVersion);
 584
 585         return transactionActor;
 586     }
 587
 588     private void syncCommitTransaction(final DOMStoreWriteTransaction transaction)
 589         throws ExecutionException, InterruptedException {
 590         DOMStoreThreePhaseCommitCohort commitCohort = transaction.ready();
 591         commitCohort.preCommit().get();
 592         commitCohort.commit().get();
 593     }
 594
 595     private void commitWithNewTransaction(final Modification modification) {
 596         DOMStoreWriteTransaction tx = store.newWriteOnlyTransaction();
 597         modification.apply(tx);
 598         try {
 599             syncCommitTransaction(tx);
 600             shardMBean.incrementCommittedTransactionCount();
 601             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 602         } catch (InterruptedException | ExecutionException e) {
 603             shardMBean.incrementFailedTransactionsCount();
 604             LOG.error(e, "Failed to commit");
 605         }
 606     }
 607
 608     private void updateSchemaContext(final UpdateSchemaContext message) {
 609         this.schemaContext = message.getSchemaContext();
 610         updateSchemaContext(message.getSchemaContext());
 611         store.onGlobalContextUpdated(message.getSchemaContext());
 612     }
 613
 614     @VisibleForTesting
 615     void updateSchemaContext(final SchemaContext schemaContext) {
 616         store.onGlobalContextUpdated(schemaContext);
 617     }
 618
 619     private void registerChangeListener(final RegisterChangeListener registerChangeListener) {
 620
 621         LOG.debug("registerDataChangeListener for {}", registerChangeListener.getPath());
 622
 623         ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
 624                                                      NormalizedNode<?, ?>>> registration;
 625         if(isLeader()) {
 626             registration = doChangeListenerRegistration(registerChangeListener);
 627         } else {
 628             LOG.debug("Shard is not the leader - delaying registration");
 629
 630             DelayedListenerRegistration delayedReg =
 631                     new DelayedListenerRegistration(registerChangeListener);
 632             delayedListenerRegistrations.add(delayedReg);
 633             registration = delayedReg;
 634         }
 635
 636         ActorRef listenerRegistration = getContext().actorOf(
 637                 DataChangeListenerRegistration.props(registration));
 638
 639         LOG.debug("registerDataChangeListener sending reply, listenerRegistrationPath = {} ",
 640                     listenerRegistration.path());
 641
 642         getSender().tell(new RegisterChangeListenerReply(listenerRegistration.path()), getSelf());
 643     }
 644
 645     private ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
 646                                                NormalizedNode<?, ?>>> doChangeListenerRegistration(
 647             final RegisterChangeListener registerChangeListener) {
 648
 649         ActorSelection dataChangeListenerPath = getContext().system().actorSelection(
 650                 registerChangeListener.getDataChangeListenerPath());
 651
 652         // Notify the listener if notifications should be enabled or not
 653         // If this shard is the leader then it will enable notifications else
 654         // it will not
 655         dataChangeListenerPath.tell(new EnableNotification(true), getSelf());
 656
 657         // Now store a reference to the data change listener so it can be notified
 658         // at a later point if notifications should be enabled or disabled
 659         dataChangeListeners.add(dataChangeListenerPath);
 660
 661         AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>> listener =
 662                 new DataChangeListenerProxy(schemaContext, dataChangeListenerPath);
 663
 664         LOG.debug("Registering for path {}", registerChangeListener.getPath());
 665
 666         return store.registerChangeListener(registerChangeListener.getPath(), listener,
 667                 registerChangeListener.getScope());
 668     }
 669
 670     private boolean isMetricsCaptureEnabled(){
 671         CommonConfig config = new CommonConfig(getContext().system().settings().config());
 672         return config.isMetricCaptureEnabled();
 673     }
 674
 675     @Override
 676     protected
 677     void startLogRecoveryBatch(final int maxBatchSize) {
 678         currentLogRecoveryBatch = Lists.newArrayListWithCapacity(maxBatchSize);
 679
 680         if(LOG.isDebugEnabled()) {
 681             LOG.debug("{} : starting log recovery batch with max size {}", persistenceId(), maxBatchSize);
 682         }
 683     }
 684
 685     @Override
 686     protected void appendRecoveredLogEntry(final Payload data) {
 687         if (data instanceof CompositeModificationPayload) {
 688             currentLogRecoveryBatch.add(((CompositeModificationPayload) data).getModification());
 689         } else if (data instanceof CompositeModificationByteStringPayload) {
 690             currentLogRecoveryBatch.add(((CompositeModificationByteStringPayload) data).getModification());
 691         } else {
 692             LOG.error("Unknown state received {} during recovery", data);
 693         }
 694     }
 695
 696     @Override
 697     protected void applyRecoverySnapshot(final ByteString snapshot) {
 698         if(recoveryCoordinator == null) {
 699             recoveryCoordinator = new ShardRecoveryCoordinator(persistenceId(), schemaContext);
 700         }
 701
 702         recoveryCoordinator.submit(snapshot, store.newWriteOnlyTransaction());
 703
 704         if(LOG.isDebugEnabled()) {
 705             LOG.debug("{} : submitted recovery sbapshot", persistenceId());
 706         }
 707     }
 708
 709     @Override
 710     protected void applyCurrentLogRecoveryBatch() {
 711         if(recoveryCoordinator == null) {
 712             recoveryCoordinator = new ShardRecoveryCoordinator(persistenceId(), schemaContext);
 713         }
 714
 715         recoveryCoordinator.submit(currentLogRecoveryBatch, store.newWriteOnlyTransaction());
 716
 717         if(LOG.isDebugEnabled()) {
 718             LOG.debug("{} : submitted log recovery batch with size {}", persistenceId(),
 719                     currentLogRecoveryBatch.size());
 720         }
 721     }
 722
 723     @Override
 724     protected void onRecoveryComplete() {
 725         if(recoveryCoordinator != null) {
 726             Collection<DOMStoreWriteTransaction> txList = recoveryCoordinator.getTransactions();
 727
 728             if(LOG.isDebugEnabled()) {
 729                 LOG.debug("{} : recovery complete - committing {} Tx's", persistenceId(), txList.size());
 730             }
 731
 732             for(DOMStoreWriteTransaction tx: txList) {
 733                 try {
 734                     syncCommitTransaction(tx);
 735                     shardMBean.incrementCommittedTransactionCount();
 736                 } catch (InterruptedException | ExecutionException e) {
 737                     shardMBean.incrementFailedTransactionsCount();
 738                     LOG.error(e, "Failed to commit");
 739                 }
 740             }
 741         }
 742
 743         recoveryCoordinator = null;
 744         currentLogRecoveryBatch = null;
 745         updateJournalStats();
 746
 747         //notify shard manager
 748         getContext().parent().tell(new ActorInitialized(), getSelf());
 749
 750         // Being paranoid here - this method should only be called once but just in case...
 751         if(txCommitTimeoutCheckSchedule == null) {
 752             // Schedule a message to be periodically sent to check if the current in-progress
 753             // transaction should be expired and aborted.
 754             FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
 755             txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
 756                     period, period, getSelf(),
 757                     TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
 758         }
 759     }
 760
 761     @Override
 762     protected void applyState(final ActorRef clientActor, final String identifier, final Object data) {
 763
 764         if (data instanceof CompositeModificationPayload) {
 765             Object modification = ((CompositeModificationPayload) data).getModification();
 766
 767             applyModificationToState(clientActor, identifier, modification);
 768         } else if(data instanceof CompositeModificationByteStringPayload ){
 769             Object modification = ((CompositeModificationByteStringPayload) data).getModification();
 770
 771             applyModificationToState(clientActor, identifier, modification);
 772
 773         } else {
 774             LOG.error("Unknown state received {} Class loader = {} CompositeNodeMod.ClassLoader = {}",
 775                     data, data.getClass().getClassLoader(),
 776                     CompositeModificationPayload.class.getClassLoader());
 777         }
 778
 779         updateJournalStats();
 780
 781     }
 782
 783     private void applyModificationToState(ActorRef clientActor, String identifier, Object modification) {
 784         if(modification == null) {
 785             LOG.error(
 786                     "modification is null - this is very unexpected, clientActor = {}, identifier = {}",
 787                     identifier, clientActor != null ? clientActor.path().toString() : null);
 788         } else if(clientActor == null) {
 789             // There's no clientActor to which to send a commit reply so we must be applying
 790             // replicated state from the leader.
 791             commitWithNewTransaction(MutableCompositeModification.fromSerializable(
 792                     modification, schemaContext));
 793         } else {
 794             // This must be the OK to commit after replication consensus.
 795             finishCommit(clientActor, identifier);
 796         }
 797     }
 798
 799     private void updateJournalStats() {
 800         ReplicatedLogEntry lastLogEntry = getLastLogEntry();
 801
 802         if (lastLogEntry != null) {
 803             shardMBean.setLastLogIndex(lastLogEntry.getIndex());
 804             shardMBean.setLastLogTerm(lastLogEntry.getTerm());
 805         }
 806
 807         shardMBean.setCommitIndex(getCommitIndex());
 808         shardMBean.setLastApplied(getLastApplied());
 809         shardMBean.setInMemoryJournalDataSize(getRaftActorContext().getReplicatedLog().dataSize());
 810     }
 811
 812     @Override
 813     protected void createSnapshot() {
 814         if (createSnapshotTransaction == null) {
 815
 816             // Create a transaction. We are really going to treat the transaction as a worker
 817             // so that this actor does not get block building the snapshot
 818             createSnapshotTransaction = createTransaction(
 819                 TransactionProxy.TransactionType.READ_ONLY.ordinal(),
 820                 "createSnapshot" + ++createSnapshotTransactionCounter, "",
 821                 CreateTransaction.CURRENT_VERSION);
 822
 823             createSnapshotTransaction.tell(
 824                 new ReadData(YangInstanceIdentifier.builder().build()).toSerializable(), self());
 825
 826         }
 827     }
 828
 829     @VisibleForTesting
 830     @Override
 831     protected void applySnapshot(final ByteString snapshot) {
 832         // Since this will be done only on Recovery or when this actor is a Follower
 833         // we can safely commit everything in here. We not need to worry about event notifications
 834         // as they would have already been disabled on the follower
 835
 836         LOG.info("Applying snapshot");
 837         try {
 838             DOMStoreWriteTransaction transaction = store.newWriteOnlyTransaction();
 839             NormalizedNodeMessages.Node serializedNode = NormalizedNodeMessages.Node.parseFrom(snapshot);
 840             NormalizedNode<?, ?> node = new NormalizedNodeToNodeCodec(schemaContext)
 841                 .decode(serializedNode);
 842
 843             // delete everything first
 844             transaction.delete(YangInstanceIdentifier.builder().build());
 845
 846             // Add everything from the remote node back
 847             transaction.write(YangInstanceIdentifier.builder().build(), node);
 848             syncCommitTransaction(transaction);
 849         } catch (InvalidProtocolBufferException | InterruptedException | ExecutionException e) {
 850             LOG.error(e, "An exception occurred when applying snapshot");
 851         } finally {
 852             LOG.info("Done applying snapshot");
 853         }
 854     }
 855
 856     @Override
 857     protected void onStateChanged() {
 858         boolean isLeader = isLeader();
 859         for (ActorSelection dataChangeListener : dataChangeListeners) {
 860             dataChangeListener.tell(new EnableNotification(isLeader), getSelf());
 861         }
 862
 863         if(isLeader) {
 864             for(DelayedListenerRegistration reg: delayedListenerRegistrations) {
 865                 if(!reg.isClosed()) {
 866                     reg.setDelegate(doChangeListenerRegistration(reg.getRegisterChangeListener()));
 867                 }
 868             }
 869
 870             delayedListenerRegistrations.clear();
 871         }
 872
 873         shardMBean.setRaftState(getRaftState().name());
 874         shardMBean.setCurrentTerm(getCurrentTerm());
 875
 876         // If this actor is no longer the leader close all the transaction chains
 877         if(!isLeader){
 878             for(Map.Entry<String, DOMStoreTransactionChain> entry : transactionChains.entrySet()){
 879                 if(LOG.isDebugEnabled()) {
 880                     LOG.debug(
 881                         "onStateChanged: Closing transaction chain {} because shard {} is no longer the leader",
 882                         entry.getKey(), getId());
 883                 }
 884                 entry.getValue().close();
 885             }
 886
 887             transactionChains.clear();
 888         }
 889     }
 890
 891     @Override
 892     protected DataPersistenceProvider persistence() {
 893         return dataPersistenceProvider;
 894     }
 895
 896     @Override protected void onLeaderChanged(final String oldLeader, final String newLeader) {
 897         shardMBean.setLeader(newLeader);
 898     }
 899
 900     @Override public String persistenceId() {
 901         return this.name.toString();
 902     }
 903
 904     @VisibleForTesting
 905     DataPersistenceProvider getDataPersistenceProvider() {
 906         return dataPersistenceProvider;
 907     }
 908
 909     private static class ShardCreator implements Creator<Shard> {
 910
 911         private static final long serialVersionUID = 1L;
 912
 913         final ShardIdentifier name;
 914         final Map<ShardIdentifier, String> peerAddresses;
 915         final DatastoreContext datastoreContext;
 916         final SchemaContext schemaContext;
 917
 918         ShardCreator(final ShardIdentifier name, final Map<ShardIdentifier, String> peerAddresses,
 919                 final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 920             this.name = name;
 921             this.peerAddresses = peerAddresses;
 922             this.datastoreContext = datastoreContext;
 923             this.schemaContext = schemaContext;
 924         }
 925
 926         @Override
 927         public Shard create() throws Exception {
 928             return new Shard(name, peerAddresses, datastoreContext, schemaContext);
 929         }
 930     }
 931
 932     @VisibleForTesting
 933     InMemoryDOMDataStore getDataStore() {
 934         return store;
 935     }
 936
 937     @VisibleForTesting
 938     ShardStats getShardMBean() {
 939         return shardMBean;
 940     }
 941
 942     private static class DelayedListenerRegistration implements
 943         ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>>> {
 944
 945         private volatile boolean closed;
 946
 947         private final RegisterChangeListener registerChangeListener;
 948
 949         private volatile ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
 950                                                              NormalizedNode<?, ?>>> delegate;
 951
 952         DelayedListenerRegistration(final RegisterChangeListener registerChangeListener) {
 953             this.registerChangeListener = registerChangeListener;
 954         }
 955
 956         void setDelegate( final ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
 957                                             NormalizedNode<?, ?>>> registration) {
 958             this.delegate = registration;
 959         }
 960
 961         boolean isClosed() {
 962             return closed;
 963         }
 964
 965         RegisterChangeListener getRegisterChangeListener() {
 966             return registerChangeListener;
 967         }
 968
 969         @Override
 970         public AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>> getInstance() {
 971             return delegate != null ? delegate.getInstance() : null;
 972         }
 973
 974         @Override
 975         public void close() {
 976             closed = true;
 977             if(delegate != null) {
 978                 delegate.close();
 979             }
 980         }
 981     }
 982 }