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.BatchedModifications;
  47 import org.opendaylight.controller.cluster.datastore.messages.BatchedModificationsReply;
  48 import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransaction;
  49 import org.opendaylight.controller.cluster.datastore.messages.CloseTransactionChain;
  50 import org.opendaylight.controller.cluster.datastore.messages.CommitTransaction;
  51 import org.opendaylight.controller.cluster.datastore.messages.CommitTransactionReply;
  52 import org.opendaylight.controller.cluster.datastore.messages.CreateSnapshot;
  53 import org.opendaylight.controller.cluster.datastore.messages.CreateTransaction;
  54 import org.opendaylight.controller.cluster.datastore.messages.CreateTransactionReply;
  55 import org.opendaylight.controller.cluster.datastore.messages.EnableNotification;
  56 import org.opendaylight.controller.cluster.datastore.messages.ForwardedReadyTransaction;
  57 import org.opendaylight.controller.cluster.datastore.messages.PeerAddressResolved;
  58 import org.opendaylight.controller.cluster.datastore.messages.ReadyTransactionReply;
  59 import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListener;
  60 import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListenerReply;
  61 import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext;
  62 import org.opendaylight.controller.cluster.datastore.modification.Modification;
  63 import org.opendaylight.controller.cluster.datastore.modification.ModificationPayload;
  64 import org.opendaylight.controller.cluster.datastore.modification.MutableCompositeModification;
  65 import org.opendaylight.controller.cluster.datastore.utils.Dispatchers;
  66 import org.opendaylight.controller.cluster.datastore.utils.MessageTracker;
  67 import org.opendaylight.controller.cluster.datastore.utils.SerializationUtils;
  68 import org.opendaylight.controller.cluster.notifications.RegisterRoleChangeListener;
  69 import org.opendaylight.controller.cluster.notifications.RoleChangeNotifier;
  70 import org.opendaylight.controller.cluster.raft.RaftActor;
  71 import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus;
  72 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  73 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationByteStringPayload;
  74 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationPayload;
  75 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload;
  76 import org.opendaylight.controller.md.sal.common.api.data.AsyncDataChangeListener;
  77 import org.opendaylight.controller.md.sal.dom.store.impl.InMemoryDOMDataStore;
  78 import org.opendaylight.controller.md.sal.dom.store.impl.InMemoryDOMDataStoreFactory;
  79 import org.opendaylight.controller.sal.core.spi.data.DOMStoreThreePhaseCommitCohort;
  80 import org.opendaylight.controller.sal.core.spi.data.DOMStoreTransaction;
  81 import org.opendaylight.controller.sal.core.spi.data.DOMStoreWriteTransaction;
  82 import org.opendaylight.yangtools.concepts.ListenerRegistration;
  83 import org.opendaylight.yangtools.yang.data.api.YangInstanceIdentifier;
  84 import org.opendaylight.yangtools.yang.data.api.schema.NormalizedNode;
  85 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
  86 import scala.concurrent.duration.Duration;
  87 import scala.concurrent.duration.FiniteDuration;
  88
  89 /**
  90  * A Shard represents a portion of the logical data tree <br/>
  91  * <p>
  92  * Our Shard uses InMemoryDataStore as it's internal representation and delegates all requests it
  93  * </p>
  94  */
  95 public class Shard extends RaftActor {
  96
  97     private static final YangInstanceIdentifier DATASTORE_ROOT = YangInstanceIdentifier.builder().build();
  98
  99     private static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = "txCommitTimeoutCheck";
 100
 101     @VisibleForTesting
 102     static final String DEFAULT_NAME = "default";
 103
 104     // The state of this Shard
 105     private final InMemoryDOMDataStore store;
 106
 107     /// The name of this shard
 108     private final String name;
 109
 110     private final ShardStats shardMBean;
 111
 112     private final List<ActorSelection> dataChangeListeners =  Lists.newArrayList();
 113
 114     private final List<DelayedListenerRegistration> delayedListenerRegistrations =
 115                                                                        Lists.newArrayList();
 116
 117     private DatastoreContext datastoreContext;
 118
 119     private DataPersistenceProvider dataPersistenceProvider;
 120
 121     private SchemaContext schemaContext;
 122
 123     private int createSnapshotTransactionCounter;
 124
 125     private final ShardCommitCoordinator commitCoordinator;
 126
 127     private long transactionCommitTimeout;
 128
 129     private Cancellable txCommitTimeoutCheckSchedule;
 130
 131     private final Optional<ActorRef> roleChangeNotifier;
 132
 133     private final MessageTracker appendEntriesReplyTracker;
 134
 135     private final ReadyTransactionReply READY_TRANSACTION_REPLY = new ReadyTransactionReply(
 136             Serialization.serializedActorPath(getSelf()));
 137
 138
 139     /**
 140      * Coordinates persistence recovery on startup.
 141      */
 142     private ShardRecoveryCoordinator recoveryCoordinator;
 143     private List<Object> currentLogRecoveryBatch;
 144
 145     private final DOMTransactionFactory transactionFactory;
 146
 147     private final String txnDispatcherPath;
 148
 149     protected Shard(final ShardIdentifier name, final Map<String, String> peerAddresses,
 150             final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 151         super(name.toString(), new HashMap<>(peerAddresses), Optional.of(datastoreContext.getShardRaftConfig()));
 152
 153         this.name = name.toString();
 154         this.datastoreContext = datastoreContext;
 155         this.schemaContext = schemaContext;
 156         this.dataPersistenceProvider = (datastoreContext.isPersistent())
 157                 ? new PersistentDataProvider() : new NonPersistentRaftDataProvider();
 158         this.txnDispatcherPath = new Dispatchers(context().system().dispatchers())
 159                 .getDispatcherPath(Dispatchers.DispatcherType.Transaction);
 160
 161
 162         LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
 163
 164         store = InMemoryDOMDataStoreFactory.create(name.toString(), null,
 165                 datastoreContext.getDataStoreProperties());
 166
 167         if(schemaContext != null) {
 168             store.onGlobalContextUpdated(schemaContext);
 169         }
 170
 171         shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
 172                 datastoreContext.getDataStoreMXBeanType());
 173         shardMBean.setNotificationManager(store.getDataChangeListenerNotificationManager());
 174         shardMBean.setShardActor(getSelf());
 175
 176         if (isMetricsCaptureEnabled()) {
 177             getContext().become(new MeteringBehavior(this));
 178         }
 179
 180         transactionFactory = new DOMTransactionFactory(store, shardMBean, LOG, this.name);
 181
 182         commitCoordinator = new ShardCommitCoordinator(transactionFactory,
 183                 TimeUnit.SECONDS.convert(5, TimeUnit.MINUTES),
 184                 datastoreContext.getShardTransactionCommitQueueCapacity(), self(), LOG, this.name);
 185
 186         setTransactionCommitTimeout();
 187
 188         // create a notifier actor for each cluster member
 189         roleChangeNotifier = createRoleChangeNotifier(name.toString());
 190
 191         appendEntriesReplyTracker = new MessageTracker(AppendEntriesReply.class,
 192                 getRaftActorContext().getConfigParams().getIsolatedCheckIntervalInMillis());
 193     }
 194
 195     private void setTransactionCommitTimeout() {
 196         transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
 197                 datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS);
 198     }
 199
 200     public static Props props(final ShardIdentifier name,
 201         final Map<String, String> peerAddresses,
 202         final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 203         Preconditions.checkNotNull(name, "name should not be null");
 204         Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
 205         Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
 206         Preconditions.checkNotNull(schemaContext, "schemaContext should not be null");
 207
 208         return Props.create(new ShardCreator(name, peerAddresses, datastoreContext, schemaContext));
 209     }
 210
 211     private Optional<ActorRef> createRoleChangeNotifier(String shardId) {
 212         ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
 213             RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
 214         return Optional.of(shardRoleChangeNotifier);
 215     }
 216
 217     @Override
 218     public void postStop() {
 219         LOG.info("Stopping Shard {}", persistenceId());
 220
 221         super.postStop();
 222
 223         if(txCommitTimeoutCheckSchedule != null) {
 224             txCommitTimeoutCheckSchedule.cancel();
 225         }
 226
 227         shardMBean.unregisterMBean();
 228     }
 229
 230     @Override
 231     public void onReceiveRecover(final Object message) throws Exception {
 232         if(LOG.isDebugEnabled()) {
 233             LOG.debug("{}: onReceiveRecover: Received message {} from {}", persistenceId(),
 234                 message.getClass().toString(), getSender());
 235         }
 236
 237         if (message instanceof RecoveryFailure){
 238             LOG.error("{}: Recovery failed because of this cause",
 239                     persistenceId(), ((RecoveryFailure) message).cause());
 240
 241             // Even though recovery failed, we still need to finish our recovery, eg send the
 242             // ActorInitialized message and start the txCommitTimeoutCheckSchedule.
 243             onRecoveryComplete();
 244         } else {
 245             super.onReceiveRecover(message);
 246             if(LOG.isTraceEnabled()) {
 247                 appendEntriesReplyTracker.begin();
 248             }
 249         }
 250     }
 251
 252     @Override
 253     public void onReceiveCommand(final Object message) throws Exception {
 254
 255         MessageTracker.Context context = appendEntriesReplyTracker.received(message);
 256
 257         if(context.error().isPresent()){
 258             LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(),
 259                     context.error());
 260         }
 261
 262         try {
 263             if (CreateTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 264                 handleCreateTransaction(message);
 265             } else if (BatchedModifications.class.isInstance(message)) {
 266                 handleBatchedModifications((BatchedModifications)message);
 267             } else if (message instanceof ForwardedReadyTransaction) {
 268                 handleForwardedReadyTransaction((ForwardedReadyTransaction) message);
 269             } else if (CanCommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 270                 handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
 271             } else if (CommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 272                 handleCommitTransaction(CommitTransaction.fromSerializable(message));
 273             } else if (AbortTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 274                 handleAbortTransaction(AbortTransaction.fromSerializable(message));
 275             } else if (CloseTransactionChain.SERIALIZABLE_CLASS.isInstance(message)) {
 276                 closeTransactionChain(CloseTransactionChain.fromSerializable(message));
 277             } else if (message instanceof RegisterChangeListener) {
 278                 registerChangeListener((RegisterChangeListener) message);
 279             } else if (message instanceof UpdateSchemaContext) {
 280                 updateSchemaContext((UpdateSchemaContext) message);
 281             } else if (message instanceof PeerAddressResolved) {
 282                 PeerAddressResolved resolved = (PeerAddressResolved) message;
 283                 setPeerAddress(resolved.getPeerId().toString(),
 284                         resolved.getPeerAddress());
 285             } else if (message.equals(TX_COMMIT_TIMEOUT_CHECK_MESSAGE)) {
 286                 handleTransactionCommitTimeoutCheck();
 287             } else if(message instanceof DatastoreContext) {
 288                 onDatastoreContext((DatastoreContext)message);
 289             } else if(message instanceof RegisterRoleChangeListener){
 290                 roleChangeNotifier.get().forward(message, context());
 291             } else if (message instanceof FollowerInitialSyncUpStatus){
 292                 shardMBean.setFollowerInitialSyncStatus(((FollowerInitialSyncUpStatus) message).isInitialSyncDone());
 293                 context().parent().tell(message, self());
 294             } else {
 295                 super.onReceiveCommand(message);
 296             }
 297         } finally {
 298             context.done();
 299         }
 300     }
 301
 302     @Override
 303     protected Optional<ActorRef> getRoleChangeNotifier() {
 304         return roleChangeNotifier;
 305     }
 306
 307     private void onDatastoreContext(DatastoreContext context) {
 308         datastoreContext = context;
 309
 310         commitCoordinator.setQueueCapacity(datastoreContext.getShardTransactionCommitQueueCapacity());
 311
 312         setTransactionCommitTimeout();
 313
 314         if(datastoreContext.isPersistent() &&
 315                 dataPersistenceProvider instanceof NonPersistentRaftDataProvider) {
 316             dataPersistenceProvider = new PersistentDataProvider();
 317         } else if(!datastoreContext.isPersistent() &&
 318                 dataPersistenceProvider instanceof PersistentDataProvider) {
 319             dataPersistenceProvider = new NonPersistentRaftDataProvider();
 320         }
 321
 322         updateConfigParams(datastoreContext.getShardRaftConfig());
 323     }
 324
 325     private void handleTransactionCommitTimeoutCheck() {
 326         CohortEntry cohortEntry = commitCoordinator.getCurrentCohortEntry();
 327         if(cohortEntry != null) {
 328             long elapsed = System.currentTimeMillis() - cohortEntry.getLastAccessTime();
 329             if(elapsed > transactionCommitTimeout) {
 330                 LOG.warn("{}: Current transaction {} has timed out after {} ms - aborting",
 331                         persistenceId(), cohortEntry.getTransactionID(), transactionCommitTimeout);
 332
 333                 doAbortTransaction(cohortEntry.getTransactionID(), null);
 334             }
 335         }
 336     }
 337
 338     private void handleCommitTransaction(final CommitTransaction commit) {
 339         final String transactionID = commit.getTransactionID();
 340
 341         LOG.debug("{}: Committing transaction {}", persistenceId(), transactionID);
 342
 343         // Get the current in-progress cohort entry in the commitCoordinator if it corresponds to
 344         // this transaction.
 345         final CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
 346         if(cohortEntry == null) {
 347             // We're not the current Tx - the Tx was likely expired b/c it took too long in
 348             // between the canCommit and commit messages.
 349             IllegalStateException ex = new IllegalStateException(
 350                     String.format("%s: Cannot commit transaction %s - it is not the current transaction",
 351                             persistenceId(), transactionID));
 352             LOG.error(ex.getMessage());
 353             shardMBean.incrementFailedTransactionsCount();
 354             getSender().tell(new akka.actor.Status.Failure(ex), getSelf());
 355             return;
 356         }
 357
 358         // We perform the preCommit phase here atomically with the commit phase. This is an
 359         // optimization to eliminate the overhead of an extra preCommit message. We lose front-end
 360         // coordination of preCommit across shards in case of failure but preCommit should not
 361         // normally fail since we ensure only one concurrent 3-phase commit.
 362
 363         try {
 364             // We block on the future here so we don't have to worry about possibly accessing our
 365             // state on a different thread outside of our dispatcher. Also, the data store
 366             // currently uses a same thread executor anyway.
 367             cohortEntry.getCohort().preCommit().get();
 368
 369             // If we do not have any followers and we are not using persistence
 370             // or if cohortEntry has no modifications
 371             // we can apply modification to the state immediately
 372             if((!hasFollowers() && !persistence().isRecoveryApplicable()) || (!cohortEntry.hasModifications())){
 373                 applyModificationToState(getSender(), transactionID, cohortEntry.getModification());
 374             } else {
 375                 Shard.this.persistData(getSender(), transactionID,
 376                         new ModificationPayload(cohortEntry.getModification()));
 377             }
 378         } catch (Exception e) {
 379             LOG.error("{} An exception occurred while preCommitting transaction {}",
 380                     persistenceId(), cohortEntry.getTransactionID(), e);
 381             shardMBean.incrementFailedTransactionsCount();
 382             getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 383         }
 384
 385         cohortEntry.updateLastAccessTime();
 386     }
 387
 388     private void finishCommit(@Nonnull final ActorRef sender, final @Nonnull String transactionID) {
 389         // With persistence enabled, this method is called via applyState by the leader strategy
 390         // after the commit has been replicated to a majority of the followers.
 391
 392         CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
 393         if(cohortEntry == null) {
 394             // The transaction is no longer the current commit. This can happen if the transaction
 395             // was aborted prior, most likely due to timeout in the front-end. We need to finish
 396             // committing the transaction though since it was successfully persisted and replicated
 397             // however we can't use the original cohort b/c it was already preCommitted and may
 398             // conflict with the current commit or may have been aborted so we commit with a new
 399             // transaction.
 400             cohortEntry = commitCoordinator.getAndRemoveCohortEntry(transactionID);
 401             if(cohortEntry != null) {
 402                 commitWithNewTransaction(cohortEntry.getModification());
 403                 sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
 404             } else {
 405                 // This really shouldn't happen - it likely means that persistence or replication
 406                 // took so long to complete such that the cohort entry was expired from the cache.
 407                 IllegalStateException ex = new IllegalStateException(
 408                         String.format("%s: Could not finish committing transaction %s - no CohortEntry found",
 409                                 persistenceId(), transactionID));
 410                 LOG.error(ex.getMessage());
 411                 sender.tell(new akka.actor.Status.Failure(ex), getSelf());
 412             }
 413
 414             return;
 415         }
 416
 417         LOG.debug("{}: Finishing commit for transaction {}", persistenceId(), cohortEntry.getTransactionID());
 418
 419         try {
 420             // We block on the future here so we don't have to worry about possibly accessing our
 421             // state on a different thread outside of our dispatcher. Also, the data store
 422             // currently uses a same thread executor anyway.
 423             cohortEntry.getCohort().commit().get();
 424
 425             sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
 426
 427             shardMBean.incrementCommittedTransactionCount();
 428             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 429
 430         } catch (Exception e) {
 431             sender.tell(new akka.actor.Status.Failure(e), getSelf());
 432
 433             LOG.error("{}, An exception occurred while committing transaction {}", persistenceId(),
 434                     transactionID, e);
 435             shardMBean.incrementFailedTransactionsCount();
 436         } finally {
 437             commitCoordinator.currentTransactionComplete(transactionID, true);
 438         }
 439     }
 440
 441     private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
 442         LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionID());
 443         commitCoordinator.handleCanCommit(canCommit, getSender(), self());
 444     }
 445
 446     private void handleBatchedModifications(BatchedModifications batched) {
 447         // This message is sent to prepare the modificationsa transaction directly on the Shard as an
 448         // optimization to avoid the extra overhead of a separate ShardTransaction actor. On the last
 449         // BatchedModifications message, the caller sets the ready flag in the message indicating
 450         // modifications are complete. The reply contains the cohort actor path (this actor) for the caller
 451         // to initiate the 3-phase commit. This also avoids the overhead of sending an additional
 452         // ReadyTransaction message.
 453
 454         // If we're not the leader then forward to the leader. This is a safety measure - we shouldn't
 455         // normally get here if we're not the leader as the front-end (TransactionProxy) should determine
 456         // the primary/leader shard. However with timing and caching on the front-end, there's a small
 457         // window where it could have a stale leader during leadership transitions.
 458         //
 459         if(isLeader()) {
 460             try {
 461                 BatchedModificationsReply reply = commitCoordinator.handleTransactionModifications(batched);
 462                 sender().tell(reply, self());
 463             } catch (Exception e) {
 464                 LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(),
 465                         batched.getTransactionID(), e);
 466                 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 467             }
 468         } else {
 469             ActorSelection leader = getLeader();
 470             if(leader != null) {
 471                 // TODO: what if this is not the first batch and leadership changed in between batched messages?
 472                 // We could check if the commitCoordinator already has a cached entry and forward all the previous
 473                 // batched modifications.
 474                 LOG.debug("{}: Forwarding BatchedModifications to leader {}", persistenceId(), leader);
 475                 leader.forward(batched, getContext());
 476             } else {
 477                 // TODO: rather than throwing an immediate exception, we could schedule a timer to try again to make
 478                 // it more resilient in case we're in the process of electing a new leader.
 479                 getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format(
 480                     "Could not find the leader for shard %s. This typically happens" +
 481                     " when the system is coming up or recovering and a leader is being elected. Try again" +
 482                     " later.", persistenceId()))), getSelf());
 483             }
 484         }
 485     }
 486
 487     private void handleForwardedReadyTransaction(ForwardedReadyTransaction ready) {
 488         LOG.debug("{}: Readying transaction {}, client version {}", persistenceId(),
 489                 ready.getTransactionID(), ready.getTxnClientVersion());
 490
 491         // This message is forwarded by the ShardTransaction on ready. We cache the cohort in the
 492         // commitCoordinator in preparation for the subsequent three phase commit initiated by
 493         // the front-end.
 494         commitCoordinator.transactionReady(ready.getTransactionID(), ready.getCohort(),
 495                 (MutableCompositeModification) ready.getModification());
 496
 497         // Return our actor path as we'll handle the three phase commit, except if the Tx client
 498         // version < 1 (Helium-1 version). This means the Tx was initiated by a base Helium version
 499         // node. In that case, the subsequent 3-phase commit messages won't contain the
 500         // transactionId so to maintain backwards compatibility, we create a separate cohort actor
 501         // to provide the compatible behavior.
 502         if(ready.getTxnClientVersion() < DataStoreVersions.HELIUM_1_VERSION) {
 503             LOG.debug("{}: Creating BackwardsCompatibleThreePhaseCommitCohort", persistenceId());
 504             ActorRef replyActorPath = getContext().actorOf(BackwardsCompatibleThreePhaseCommitCohort.props(
 505                     ready.getTransactionID()));
 506
 507             ReadyTransactionReply readyTransactionReply =
 508                     new ReadyTransactionReply(Serialization.serializedActorPath(replyActorPath));
 509             getSender().tell(ready.isReturnSerialized() ? readyTransactionReply.toSerializable() :
 510                     readyTransactionReply, getSelf());
 511
 512         } else {
 513
 514             getSender().tell(ready.isReturnSerialized() ? READY_TRANSACTION_REPLY.toSerializable() :
 515                     READY_TRANSACTION_REPLY, getSelf());
 516         }
 517     }
 518
 519     private void handleAbortTransaction(final AbortTransaction abort) {
 520         doAbortTransaction(abort.getTransactionID(), getSender());
 521     }
 522
 523     void doAbortTransaction(final String transactionID, final ActorRef sender) {
 524         final CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
 525         if(cohortEntry != null) {
 526             LOG.debug("{}: Aborting transaction {}", persistenceId(), transactionID);
 527
 528             // We don't remove the cached cohort entry here (ie pass false) in case the Tx was
 529             // aborted during replication in which case we may still commit locally if replication
 530             // succeeds.
 531             commitCoordinator.currentTransactionComplete(transactionID, false);
 532
 533             final ListenableFuture<Void> future = cohortEntry.getCohort().abort();
 534             final ActorRef self = getSelf();
 535
 536             Futures.addCallback(future, new FutureCallback<Void>() {
 537                 @Override
 538                 public void onSuccess(final Void v) {
 539                     shardMBean.incrementAbortTransactionsCount();
 540
 541                     if(sender != null) {
 542                         sender.tell(AbortTransactionReply.INSTANCE.toSerializable(), self);
 543                     }
 544                 }
 545
 546                 @Override
 547                 public void onFailure(final Throwable t) {
 548                     LOG.error("{}: An exception happened during abort", persistenceId(), t);
 549
 550                     if(sender != null) {
 551                         sender.tell(new akka.actor.Status.Failure(t), self);
 552                     }
 553                 }
 554             });
 555         }
 556     }
 557
 558     private void handleCreateTransaction(final Object message) {
 559         if (isLeader()) {
 560             createTransaction(CreateTransaction.fromSerializable(message));
 561         } else if (getLeader() != null) {
 562             getLeader().forward(message, getContext());
 563         } else {
 564             getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format(
 565                 "Could not find leader for shard %s so transaction cannot be created. This typically happens" +
 566                 " when the system is coming up or recovering and a leader is being elected. Try again" +
 567                 " later.", persistenceId()))), getSelf());
 568         }
 569     }
 570
 571     private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
 572         transactionFactory.closeTransactionChain(closeTransactionChain.getTransactionChainId());
 573     }
 574
 575     private ActorRef createTypedTransactionActor(int transactionType,
 576             ShardTransactionIdentifier transactionId, String transactionChainId,
 577             short clientVersion ) {
 578
 579         DOMStoreTransaction transaction = transactionFactory.newTransaction(
 580                 TransactionProxy.TransactionType.fromInt(transactionType), transactionId.toString(),
 581                 transactionChainId);
 582
 583         return createShardTransaction(transaction, transactionId, clientVersion);
 584     }
 585
 586     private ActorRef createShardTransaction(DOMStoreTransaction transaction, ShardTransactionIdentifier transactionId,
 587                                             short clientVersion){
 588         return getContext().actorOf(
 589                 ShardTransaction.props(transaction, getSelf(),
 590                         schemaContext, datastoreContext, shardMBean,
 591                         transactionId.getRemoteTransactionId(), clientVersion)
 592                         .withDispatcher(txnDispatcherPath),
 593                 transactionId.toString());
 594
 595     }
 596
 597     private void createTransaction(CreateTransaction createTransaction) {
 598         try {
 599             ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
 600                 createTransaction.getTransactionId(), createTransaction.getTransactionChainId(),
 601                 createTransaction.getVersion());
 602
 603             getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
 604                     createTransaction.getTransactionId()).toSerializable(), getSelf());
 605         } catch (Exception e) {
 606             getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 607         }
 608     }
 609
 610     private ActorRef createTransaction(int transactionType, String remoteTransactionId,
 611             String transactionChainId, short clientVersion) {
 612
 613
 614         ShardTransactionIdentifier transactionId = new ShardTransactionIdentifier(remoteTransactionId);
 615
 616         if(LOG.isDebugEnabled()) {
 617             LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
 618         }
 619
 620         ActorRef transactionActor = createTypedTransactionActor(transactionType, transactionId,
 621                 transactionChainId, clientVersion);
 622
 623         return transactionActor;
 624     }
 625
 626     private void syncCommitTransaction(final DOMStoreWriteTransaction transaction)
 627         throws ExecutionException, InterruptedException {
 628         DOMStoreThreePhaseCommitCohort commitCohort = transaction.ready();
 629         commitCohort.preCommit().get();
 630         commitCohort.commit().get();
 631     }
 632
 633     private void commitWithNewTransaction(final Modification modification) {
 634         DOMStoreWriteTransaction tx = store.newWriteOnlyTransaction();
 635         modification.apply(tx);
 636         try {
 637             syncCommitTransaction(tx);
 638             shardMBean.incrementCommittedTransactionCount();
 639             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 640         } catch (InterruptedException | ExecutionException e) {
 641             shardMBean.incrementFailedTransactionsCount();
 642             LOG.error("{}: Failed to commit", persistenceId(), e);
 643         }
 644     }
 645
 646     private void updateSchemaContext(final UpdateSchemaContext message) {
 647         this.schemaContext = message.getSchemaContext();
 648         updateSchemaContext(message.getSchemaContext());
 649         store.onGlobalContextUpdated(message.getSchemaContext());
 650     }
 651
 652     @VisibleForTesting
 653     void updateSchemaContext(final SchemaContext schemaContext) {
 654         store.onGlobalContextUpdated(schemaContext);
 655     }
 656
 657     private void registerChangeListener(final RegisterChangeListener registerChangeListener) {
 658
 659         LOG.debug("{}: registerDataChangeListener for {}", persistenceId(), registerChangeListener.getPath());
 660
 661         ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
 662                                                      NormalizedNode<?, ?>>> registration;
 663         if(isLeader()) {
 664             registration = doChangeListenerRegistration(registerChangeListener);
 665         } else {
 666             LOG.debug("{}: Shard is not the leader - delaying registration", persistenceId());
 667
 668             DelayedListenerRegistration delayedReg =
 669                     new DelayedListenerRegistration(registerChangeListener);
 670             delayedListenerRegistrations.add(delayedReg);
 671             registration = delayedReg;
 672         }
 673
 674         ActorRef listenerRegistration = getContext().actorOf(
 675                 DataChangeListenerRegistration.props(registration));
 676
 677         LOG.debug("{}: registerDataChangeListener sending reply, listenerRegistrationPath = {} ",
 678                 persistenceId(), listenerRegistration.path());
 679
 680         getSender().tell(new RegisterChangeListenerReply(listenerRegistration.path()), getSelf());
 681     }
 682
 683     private ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
 684                                                NormalizedNode<?, ?>>> doChangeListenerRegistration(
 685             final RegisterChangeListener registerChangeListener) {
 686
 687         ActorSelection dataChangeListenerPath = getContext().system().actorSelection(
 688                 registerChangeListener.getDataChangeListenerPath());
 689
 690         // Notify the listener if notifications should be enabled or not
 691         // If this shard is the leader then it will enable notifications else
 692         // it will not
 693         dataChangeListenerPath.tell(new EnableNotification(true), getSelf());
 694
 695         // Now store a reference to the data change listener so it can be notified
 696         // at a later point if notifications should be enabled or disabled
 697         dataChangeListeners.add(dataChangeListenerPath);
 698
 699         AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>> listener =
 700                 new DataChangeListenerProxy(dataChangeListenerPath);
 701
 702         LOG.debug("{}: Registering for path {}", persistenceId(), registerChangeListener.getPath());
 703
 704         return store.registerChangeListener(registerChangeListener.getPath(), listener,
 705                 registerChangeListener.getScope());
 706     }
 707
 708     private boolean isMetricsCaptureEnabled(){
 709         CommonConfig config = new CommonConfig(getContext().system().settings().config());
 710         return config.isMetricCaptureEnabled();
 711     }
 712
 713     @Override
 714     protected
 715     void startLogRecoveryBatch(final int maxBatchSize) {
 716         currentLogRecoveryBatch = Lists.newArrayListWithCapacity(maxBatchSize);
 717
 718         if(LOG.isDebugEnabled()) {
 719             LOG.debug("{}: starting log recovery batch with max size {}", persistenceId(), maxBatchSize);
 720         }
 721     }
 722
 723     @Override
 724     protected void appendRecoveredLogEntry(final Payload data) {
 725         if(data instanceof ModificationPayload) {
 726             try {
 727                 currentLogRecoveryBatch.add(((ModificationPayload) data).getModification());
 728             } catch (ClassNotFoundException | IOException e) {
 729                 LOG.error("{}: Error extracting ModificationPayload", persistenceId(), e);
 730             }
 731         } else if (data instanceof CompositeModificationPayload) {
 732             currentLogRecoveryBatch.add(((CompositeModificationPayload) data).getModification());
 733         } else if (data instanceof CompositeModificationByteStringPayload) {
 734             currentLogRecoveryBatch.add(((CompositeModificationByteStringPayload) data).getModification());
 735         } else {
 736             LOG.error("{}: Unknown state received {} during recovery", persistenceId(), data);
 737         }
 738     }
 739
 740     @Override
 741     protected void applyRecoverySnapshot(final byte[] snapshotBytes) {
 742         if(recoveryCoordinator == null) {
 743             recoveryCoordinator = new ShardRecoveryCoordinator(persistenceId(), schemaContext,
 744                     LOG, name.toString());
 745         }
 746
 747         recoveryCoordinator.submit(snapshotBytes, store.newWriteOnlyTransaction());
 748
 749         if(LOG.isDebugEnabled()) {
 750             LOG.debug("{}: submitted recovery sbapshot", persistenceId());
 751         }
 752     }
 753
 754     @Override
 755     protected void applyCurrentLogRecoveryBatch() {
 756         if(recoveryCoordinator == null) {
 757             recoveryCoordinator = new ShardRecoveryCoordinator(persistenceId(), schemaContext,
 758                     LOG, name.toString());
 759         }
 760
 761         recoveryCoordinator.submit(currentLogRecoveryBatch, store.newWriteOnlyTransaction());
 762
 763         if(LOG.isDebugEnabled()) {
 764             LOG.debug("{}: submitted log recovery batch with size {}", persistenceId(),
 765                     currentLogRecoveryBatch.size());
 766         }
 767     }
 768
 769     @Override
 770     protected void onRecoveryComplete() {
 771         if(recoveryCoordinator != null) {
 772             Collection<DOMStoreWriteTransaction> txList = recoveryCoordinator.getTransactions();
 773
 774             if(LOG.isDebugEnabled()) {
 775                 LOG.debug("{}: recovery complete - committing {} Tx's", persistenceId(), txList.size());
 776             }
 777
 778             for(DOMStoreWriteTransaction tx: txList) {
 779                 try {
 780                     syncCommitTransaction(tx);
 781                     shardMBean.incrementCommittedTransactionCount();
 782                 } catch (InterruptedException | ExecutionException e) {
 783                     shardMBean.incrementFailedTransactionsCount();
 784                     LOG.error("{}: Failed to commit", persistenceId(), e);
 785                 }
 786             }
 787         }
 788
 789         recoveryCoordinator = null;
 790         currentLogRecoveryBatch = null;
 791
 792         //notify shard manager
 793         getContext().parent().tell(new ActorInitialized(), getSelf());
 794
 795         // Being paranoid here - this method should only be called once but just in case...
 796         if(txCommitTimeoutCheckSchedule == null) {
 797             // Schedule a message to be periodically sent to check if the current in-progress
 798             // transaction should be expired and aborted.
 799             FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
 800             txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
 801                     period, period, getSelf(),
 802                     TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
 803         }
 804     }
 805
 806     @Override
 807     protected void applyState(final ActorRef clientActor, final String identifier, final Object data) {
 808
 809         if(data instanceof ModificationPayload) {
 810             try {
 811                 applyModificationToState(clientActor, identifier, ((ModificationPayload) data).getModification());
 812             } catch (ClassNotFoundException | IOException e) {
 813                 LOG.error("{}: Error extracting ModificationPayload", persistenceId(), e);
 814             }
 815         }
 816         else if (data instanceof CompositeModificationPayload) {
 817             Object modification = ((CompositeModificationPayload) data).getModification();
 818
 819             applyModificationToState(clientActor, identifier, modification);
 820         } else if(data instanceof CompositeModificationByteStringPayload ){
 821             Object modification = ((CompositeModificationByteStringPayload) data).getModification();
 822
 823             applyModificationToState(clientActor, identifier, modification);
 824         } else {
 825             LOG.error("{}: Unknown state received {} Class loader = {} CompositeNodeMod.ClassLoader = {}",
 826                     persistenceId(), data, data.getClass().getClassLoader(),
 827                     CompositeModificationPayload.class.getClassLoader());
 828         }
 829     }
 830
 831     private void applyModificationToState(ActorRef clientActor, String identifier, Object modification) {
 832         if(modification == null) {
 833             LOG.error(
 834                     "{}: modification is null - this is very unexpected, clientActor = {}, identifier = {}",
 835                     persistenceId(), identifier, clientActor != null ? clientActor.path().toString() : null);
 836         } else if(clientActor == null) {
 837             // There's no clientActor to which to send a commit reply so we must be applying
 838             // replicated state from the leader.
 839             commitWithNewTransaction(MutableCompositeModification.fromSerializable(modification));
 840         } else {
 841             // This must be the OK to commit after replication consensus.
 842             finishCommit(clientActor, identifier);
 843         }
 844     }
 845
 846     @Override
 847     protected void createSnapshot() {
 848         // Create a transaction actor. We are really going to treat the transaction as a worker
 849         // so that this actor does not get block building the snapshot. THe transaction actor will
 850         // after processing the CreateSnapshot message.
 851
 852         ActorRef createSnapshotTransaction = createTransaction(
 853                 TransactionProxy.TransactionType.READ_ONLY.ordinal(),
 854                 "createSnapshot" + ++createSnapshotTransactionCounter, "",
 855                 DataStoreVersions.CURRENT_VERSION);
 856
 857         createSnapshotTransaction.tell(CreateSnapshot.INSTANCE, self());
 858     }
 859
 860     @VisibleForTesting
 861     @Override
 862     protected void applySnapshot(final byte[] snapshotBytes) {
 863         // Since this will be done only on Recovery or when this actor is a Follower
 864         // we can safely commit everything in here. We not need to worry about event notifications
 865         // as they would have already been disabled on the follower
 866
 867         LOG.info("{}: Applying snapshot", persistenceId());
 868         try {
 869             DOMStoreWriteTransaction transaction = store.newWriteOnlyTransaction();
 870
 871             NormalizedNode<?, ?> node = SerializationUtils.deserializeNormalizedNode(snapshotBytes);
 872
 873             // delete everything first
 874             transaction.delete(DATASTORE_ROOT);
 875
 876             // Add everything from the remote node back
 877             transaction.write(DATASTORE_ROOT, node);
 878             syncCommitTransaction(transaction);
 879         } catch (InterruptedException | ExecutionException e) {
 880             LOG.error("{}: An exception occurred when applying snapshot", persistenceId(), e);
 881         } finally {
 882             LOG.info("{}: Done applying snapshot", persistenceId());
 883         }
 884     }
 885
 886     @Override
 887     protected void onStateChanged() {
 888         boolean isLeader = isLeader();
 889         for (ActorSelection dataChangeListener : dataChangeListeners) {
 890             dataChangeListener.tell(new EnableNotification(isLeader), getSelf());
 891         }
 892
 893         if(isLeader) {
 894             for(DelayedListenerRegistration reg: delayedListenerRegistrations) {
 895                 if(!reg.isClosed()) {
 896                     reg.setDelegate(doChangeListenerRegistration(reg.getRegisterChangeListener()));
 897                 }
 898             }
 899
 900             delayedListenerRegistrations.clear();
 901         }
 902
 903         // If this actor is no longer the leader close all the transaction chains
 904         if(!isLeader) {
 905             if(LOG.isDebugEnabled()) {
 906                 LOG.debug(
 907                     "{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader",
 908                     persistenceId(), getId());
 909             }
 910
 911             transactionFactory.closeAllTransactionChains();
 912         }
 913     }
 914
 915     @Override
 916     protected DataPersistenceProvider persistence() {
 917         return dataPersistenceProvider;
 918     }
 919
 920     @Override public String persistenceId() {
 921         return this.name;
 922     }
 923
 924     @VisibleForTesting
 925     DataPersistenceProvider getDataPersistenceProvider() {
 926         return dataPersistenceProvider;
 927     }
 928
 929     @VisibleForTesting
 930     ShardCommitCoordinator getCommitCoordinator() {
 931         return commitCoordinator;
 932     }
 933
 934
 935     private static class ShardCreator implements Creator<Shard> {
 936
 937         private static final long serialVersionUID = 1L;
 938
 939         final ShardIdentifier name;
 940         final Map<String, String> peerAddresses;
 941         final DatastoreContext datastoreContext;
 942         final SchemaContext schemaContext;
 943
 944         ShardCreator(final ShardIdentifier name, final Map<String, String> peerAddresses,
 945                 final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 946             this.name = name;
 947             this.peerAddresses = peerAddresses;
 948             this.datastoreContext = datastoreContext;
 949             this.schemaContext = schemaContext;
 950         }
 951
 952         @Override
 953         public Shard create() throws Exception {
 954             return new Shard(name, peerAddresses, datastoreContext, schemaContext);
 955         }
 956     }
 957
 958     @VisibleForTesting
 959     public InMemoryDOMDataStore getDataStore() {
 960         return store;
 961     }
 962
 963     @VisibleForTesting
 964     ShardStats getShardMBean() {
 965         return shardMBean;
 966     }
 967
 968     private static class DelayedListenerRegistration implements
 969         ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>>> {
 970
 971         private volatile boolean closed;
 972
 973         private final RegisterChangeListener registerChangeListener;
 974
 975         private volatile ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
 976                                                              NormalizedNode<?, ?>>> delegate;
 977
 978         DelayedListenerRegistration(final RegisterChangeListener registerChangeListener) {
 979             this.registerChangeListener = registerChangeListener;
 980         }
 981
 982         void setDelegate( final ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
 983                                             NormalizedNode<?, ?>>> registration) {
 984             this.delegate = registration;
 985         }
 986
 987         boolean isClosed() {
 988             return closed;
 989         }
 990
 991         RegisterChangeListener getRegisterChangeListener() {
 992             return registerChangeListener;
 993         }
 994
 995         @Override
 996         public AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>> getInstance() {
 997             return delegate != null ? delegate.getInstance() : null;
 998         }
 999
1000         @Override
1001         public void close() {
1002             closed = true;
1003             if(delegate != null) {
1004                 delegate.close();
1005             }
1006         }
1007     }
1008 }