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.util.concurrent.FutureCallback;
  22 import com.google.common.util.concurrent.Futures;
  23 import com.google.common.util.concurrent.ListenableFuture;
  24 import java.io.IOException;
  25 import java.util.HashMap;
  26 import java.util.Map;
  27 import java.util.concurrent.TimeUnit;
  28 import javax.annotation.Nonnull;
  29 import org.opendaylight.controller.cluster.common.actor.CommonConfig;
  30 import org.opendaylight.controller.cluster.common.actor.MeteringBehavior;
  31 import org.opendaylight.controller.cluster.datastore.ShardCommitCoordinator.CohortEntry;
  32 import org.opendaylight.controller.cluster.datastore.exceptions.NoShardLeaderException;
  33 import org.opendaylight.controller.cluster.datastore.identifiers.ShardIdentifier;
  34 import org.opendaylight.controller.cluster.datastore.identifiers.ShardTransactionIdentifier;
  35 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardMBeanFactory;
  36 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardStats;
  37 import org.opendaylight.controller.cluster.datastore.messages.AbortTransaction;
  38 import org.opendaylight.controller.cluster.datastore.messages.AbortTransactionReply;
  39 import org.opendaylight.controller.cluster.datastore.messages.ActorInitialized;
  40 import org.opendaylight.controller.cluster.datastore.messages.BatchedModifications;
  41 import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransaction;
  42 import org.opendaylight.controller.cluster.datastore.messages.CloseTransactionChain;
  43 import org.opendaylight.controller.cluster.datastore.messages.CommitTransaction;
  44 import org.opendaylight.controller.cluster.datastore.messages.CommitTransactionReply;
  45 import org.opendaylight.controller.cluster.datastore.messages.CreateTransaction;
  46 import org.opendaylight.controller.cluster.datastore.messages.CreateTransactionReply;
  47 import org.opendaylight.controller.cluster.datastore.messages.ForwardedReadyTransaction;
  48 import org.opendaylight.controller.cluster.datastore.messages.PeerAddressResolved;
  49 import org.opendaylight.controller.cluster.datastore.messages.ReadyLocalTransaction;
  50 import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListener;
  51 import org.opendaylight.controller.cluster.datastore.messages.RegisterDataTreeChangeListener;
  52 import org.opendaylight.controller.cluster.datastore.messages.ShardLeaderStateChanged;
  53 import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext;
  54 import org.opendaylight.controller.cluster.datastore.modification.Modification;
  55 import org.opendaylight.controller.cluster.datastore.modification.ModificationPayload;
  56 import org.opendaylight.controller.cluster.datastore.modification.MutableCompositeModification;
  57 import org.opendaylight.controller.cluster.datastore.utils.Dispatchers;
  58 import org.opendaylight.controller.cluster.datastore.utils.MessageTracker;
  59 import org.opendaylight.controller.cluster.notifications.LeaderStateChanged;
  60 import org.opendaylight.controller.cluster.notifications.RegisterRoleChangeListener;
  61 import org.opendaylight.controller.cluster.notifications.RoleChangeNotifier;
  62 import org.opendaylight.controller.cluster.raft.RaftActor;
  63 import org.opendaylight.controller.cluster.raft.RaftActorRecoveryCohort;
  64 import org.opendaylight.controller.cluster.raft.RaftActorSnapshotCohort;
  65 import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus;
  66 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  67 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationByteStringPayload;
  68 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationPayload;
  69 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTree;
  70 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidate;
  71 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException;
  72 import org.opendaylight.yangtools.yang.data.api.schema.tree.ModificationType;
  73 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
  74 import scala.concurrent.duration.Duration;
  75 import scala.concurrent.duration.FiniteDuration;
  76
  77 /**
  78  * A Shard represents a portion of the logical data tree <br/>
  79  * <p>
  80  * Our Shard uses InMemoryDataTree as it's internal representation and delegates all requests it
  81  * </p>
  82  */
  83 public class Shard extends RaftActor {
  84
  85     private static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = "txCommitTimeoutCheck";
  86
  87     @VisibleForTesting
  88     static final String DEFAULT_NAME = "default";
  89
  90     // The state of this Shard
  91     private final ShardDataTree store;
  92
  93     /// The name of this shard
  94     private final String name;
  95
  96     private final ShardStats shardMBean;
  97
  98     private DatastoreContext datastoreContext;
  99
 100     private final ShardCommitCoordinator commitCoordinator;
 101
 102     private long transactionCommitTimeout;
 103
 104     private Cancellable txCommitTimeoutCheckSchedule;
 105
 106     private final Optional<ActorRef> roleChangeNotifier;
 107
 108     private final MessageTracker appendEntriesReplyTracker;
 109
 110     private final ShardTransactionActorFactory transactionActorFactory;
 111
 112     private final ShardSnapshotCohort snapshotCohort;
 113
 114     private final DataTreeChangeListenerSupport treeChangeSupport = new DataTreeChangeListenerSupport(this);
 115     private final DataChangeListenerSupport changeSupport = new DataChangeListenerSupport(this);
 116
 117     protected Shard(final ShardIdentifier name, final Map<String, String> peerAddresses,
 118             final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 119         super(name.toString(), new HashMap<>(peerAddresses), Optional.of(datastoreContext.getShardRaftConfig()),
 120                 DataStoreVersions.CURRENT_VERSION);
 121
 122         this.name = name.toString();
 123         this.datastoreContext = datastoreContext;
 124
 125         setPersistence(datastoreContext.isPersistent());
 126
 127         LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
 128
 129         store = new ShardDataTree(schemaContext);
 130
 131         shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
 132                 datastoreContext.getDataStoreMXBeanType());
 133         shardMBean.setShardActor(getSelf());
 134
 135         if (isMetricsCaptureEnabled()) {
 136             getContext().become(new MeteringBehavior(this));
 137         }
 138
 139         commitCoordinator = new ShardCommitCoordinator(store,
 140                 TimeUnit.SECONDS.convert(5, TimeUnit.MINUTES),
 141                 datastoreContext.getShardTransactionCommitQueueCapacity(), self(), LOG, this.name);
 142
 143         setTransactionCommitTimeout();
 144
 145         // create a notifier actor for each cluster member
 146         roleChangeNotifier = createRoleChangeNotifier(name.toString());
 147
 148         appendEntriesReplyTracker = new MessageTracker(AppendEntriesReply.class,
 149                 getRaftActorContext().getConfigParams().getIsolatedCheckIntervalInMillis());
 150
 151         transactionActorFactory = new ShardTransactionActorFactory(store, datastoreContext,
 152                 new Dispatchers(context().system().dispatchers()).getDispatcherPath(
 153                         Dispatchers.DispatcherType.Transaction), self(), getContext(), shardMBean);
 154
 155         snapshotCohort = new ShardSnapshotCohort(transactionActorFactory, store, LOG, this.name);
 156     }
 157
 158     private void setTransactionCommitTimeout() {
 159         transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
 160                 datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS);
 161     }
 162
 163     public static Props props(final ShardIdentifier name,
 164         final Map<String, String> peerAddresses,
 165         final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 166         Preconditions.checkNotNull(name, "name should not be null");
 167         Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
 168         Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
 169         Preconditions.checkNotNull(schemaContext, "schemaContext should not be null");
 170
 171         return Props.create(new ShardCreator(name, peerAddresses, datastoreContext, schemaContext));
 172     }
 173
 174     private Optional<ActorRef> createRoleChangeNotifier(String shardId) {
 175         ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
 176             RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
 177         return Optional.of(shardRoleChangeNotifier);
 178     }
 179
 180     @Override
 181     public void postStop() {
 182         LOG.info("Stopping Shard {}", persistenceId());
 183
 184         super.postStop();
 185
 186         if(txCommitTimeoutCheckSchedule != null) {
 187             txCommitTimeoutCheckSchedule.cancel();
 188         }
 189
 190         shardMBean.unregisterMBean();
 191     }
 192
 193     @Override
 194     public void onReceiveRecover(final Object message) throws Exception {
 195         if(LOG.isDebugEnabled()) {
 196             LOG.debug("{}: onReceiveRecover: Received message {} from {}", persistenceId(),
 197                 message.getClass().toString(), getSender());
 198         }
 199
 200         if (message instanceof RecoveryFailure){
 201             LOG.error("{}: Recovery failed because of this cause",
 202                     persistenceId(), ((RecoveryFailure) message).cause());
 203
 204             // Even though recovery failed, we still need to finish our recovery, eg send the
 205             // ActorInitialized message and start the txCommitTimeoutCheckSchedule.
 206             onRecoveryComplete();
 207         } else {
 208             super.onReceiveRecover(message);
 209             if(LOG.isTraceEnabled()) {
 210                 appendEntriesReplyTracker.begin();
 211             }
 212         }
 213     }
 214
 215     @Override
 216     public void onReceiveCommand(final Object message) throws Exception {
 217
 218         MessageTracker.Context context = appendEntriesReplyTracker.received(message);
 219
 220         if(context.error().isPresent()){
 221             LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(),
 222                     context.error());
 223         }
 224
 225         try {
 226             if (CreateTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 227                 handleCreateTransaction(message);
 228             } else if (BatchedModifications.class.isInstance(message)) {
 229                 handleBatchedModifications((BatchedModifications)message);
 230             } else if (message instanceof ForwardedReadyTransaction) {
 231                 commitCoordinator.handleForwardedReadyTransaction((ForwardedReadyTransaction) message,
 232                         getSender(), this);
 233             } else if (message instanceof ReadyLocalTransaction) {
 234                 handleReadyLocalTransaction((ReadyLocalTransaction)message);
 235             } else if (CanCommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 236                 handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
 237             } else if (CommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 238                 handleCommitTransaction(CommitTransaction.fromSerializable(message));
 239             } else if (AbortTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 240                 handleAbortTransaction(AbortTransaction.fromSerializable(message));
 241             } else if (CloseTransactionChain.SERIALIZABLE_CLASS.isInstance(message)) {
 242                 closeTransactionChain(CloseTransactionChain.fromSerializable(message));
 243             } else if (message instanceof RegisterChangeListener) {
 244                 changeSupport.onMessage((RegisterChangeListener) message, isLeader());
 245             } else if (message instanceof RegisterDataTreeChangeListener) {
 246                 treeChangeSupport.onMessage((RegisterDataTreeChangeListener) message, isLeader());
 247             } else if (message instanceof UpdateSchemaContext) {
 248                 updateSchemaContext((UpdateSchemaContext) message);
 249             } else if (message instanceof PeerAddressResolved) {
 250                 PeerAddressResolved resolved = (PeerAddressResolved) message;
 251                 setPeerAddress(resolved.getPeerId().toString(),
 252                         resolved.getPeerAddress());
 253             } else if (message.equals(TX_COMMIT_TIMEOUT_CHECK_MESSAGE)) {
 254                 handleTransactionCommitTimeoutCheck();
 255             } else if(message instanceof DatastoreContext) {
 256                 onDatastoreContext((DatastoreContext)message);
 257             } else if(message instanceof RegisterRoleChangeListener){
 258                 roleChangeNotifier.get().forward(message, context());
 259             } else if (message instanceof FollowerInitialSyncUpStatus){
 260                 shardMBean.setFollowerInitialSyncStatus(((FollowerInitialSyncUpStatus) message).isInitialSyncDone());
 261                 context().parent().tell(message, self());
 262             } else {
 263                 super.onReceiveCommand(message);
 264             }
 265         } finally {
 266             context.done();
 267         }
 268     }
 269
 270     @Override
 271     protected Optional<ActorRef> getRoleChangeNotifier() {
 272         return roleChangeNotifier;
 273     }
 274
 275     @Override
 276     protected LeaderStateChanged newLeaderStateChanged(String memberId, String leaderId) {
 277         return new ShardLeaderStateChanged(memberId, leaderId,
 278                 isLeader() ? Optional.<DataTree>of(store.getDataTree()) : Optional.<DataTree>absent());
 279     }
 280
 281     private void onDatastoreContext(DatastoreContext context) {
 282         datastoreContext = context;
 283
 284         commitCoordinator.setQueueCapacity(datastoreContext.getShardTransactionCommitQueueCapacity());
 285
 286         setTransactionCommitTimeout();
 287
 288         if(datastoreContext.isPersistent() && !persistence().isRecoveryApplicable()) {
 289             setPersistence(true);
 290         } else if(!datastoreContext.isPersistent() && persistence().isRecoveryApplicable()) {
 291             setPersistence(false);
 292         }
 293
 294         updateConfigParams(datastoreContext.getShardRaftConfig());
 295     }
 296
 297     private void handleTransactionCommitTimeoutCheck() {
 298         CohortEntry cohortEntry = commitCoordinator.getCurrentCohortEntry();
 299         if(cohortEntry != null) {
 300             long elapsed = System.currentTimeMillis() - cohortEntry.getLastAccessTime();
 301             if(elapsed > transactionCommitTimeout) {
 302                 LOG.warn("{}: Current transaction {} has timed out after {} ms - aborting",
 303                         persistenceId(), cohortEntry.getTransactionID(), transactionCommitTimeout);
 304
 305                 doAbortTransaction(cohortEntry.getTransactionID(), null);
 306             }
 307         }
 308     }
 309
 310     private static boolean isEmptyCommit(final DataTreeCandidate candidate) {
 311         return ModificationType.UNMODIFIED.equals(candidate.getRootNode().getModificationType());
 312     }
 313
 314     void continueCommit(final CohortEntry cohortEntry) throws Exception {
 315         final DataTreeCandidate candidate = cohortEntry.getCohort().getCandidate();
 316
 317         // If we do not have any followers and we are not using persistence
 318         // or if cohortEntry has no modifications
 319         // we can apply modification to the state immediately
 320         if ((!hasFollowers() && !persistence().isRecoveryApplicable()) || isEmptyCommit(candidate)) {
 321             applyModificationToState(getSender(), cohortEntry.getTransactionID(), candidate);
 322         } else {
 323             Shard.this.persistData(getSender(), cohortEntry.getTransactionID(),
 324                 DataTreeCandidatePayload.create(candidate));
 325         }
 326     }
 327
 328     private void handleCommitTransaction(final CommitTransaction commit) {
 329         if(!commitCoordinator.handleCommit(commit.getTransactionID(), getSender(), this)) {
 330             shardMBean.incrementFailedTransactionsCount();
 331         }
 332     }
 333
 334     private void finishCommit(@Nonnull final ActorRef sender, @Nonnull final String transactionID, @Nonnull final CohortEntry cohortEntry) {
 335         LOG.debug("{}: Finishing commit for transaction {}", persistenceId(), cohortEntry.getTransactionID());
 336
 337         try {
 338             // We block on the future here so we don't have to worry about possibly accessing our
 339             // state on a different thread outside of our dispatcher. Also, the data store
 340             // currently uses a same thread executor anyway.
 341             cohortEntry.getCohort().commit().get();
 342
 343             sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
 344
 345             shardMBean.incrementCommittedTransactionCount();
 346             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 347
 348         } catch (Exception e) {
 349             sender.tell(new akka.actor.Status.Failure(e), getSelf());
 350
 351             LOG.error("{}, An exception occurred while committing transaction {}", persistenceId(),
 352                     transactionID, e);
 353             shardMBean.incrementFailedTransactionsCount();
 354         } finally {
 355             commitCoordinator.currentTransactionComplete(transactionID, true);
 356         }
 357     }
 358
 359     private void finishCommit(@Nonnull final ActorRef sender, final @Nonnull String transactionID) {
 360         // With persistence enabled, this method is called via applyState by the leader strategy
 361         // after the commit has been replicated to a majority of the followers.
 362
 363         CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
 364         if (cohortEntry == null) {
 365             // The transaction is no longer the current commit. This can happen if the transaction
 366             // was aborted prior, most likely due to timeout in the front-end. We need to finish
 367             // committing the transaction though since it was successfully persisted and replicated
 368             // however we can't use the original cohort b/c it was already preCommitted and may
 369             // conflict with the current commit or may have been aborted so we commit with a new
 370             // transaction.
 371             cohortEntry = commitCoordinator.getAndRemoveCohortEntry(transactionID);
 372             if(cohortEntry != null) {
 373                 try {
 374                     store.applyForeignCandidate(transactionID, cohortEntry.getCohort().getCandidate());
 375                 } catch (DataValidationFailedException e) {
 376                     shardMBean.incrementFailedTransactionsCount();
 377                     LOG.error("{}: Failed to re-apply transaction {}", persistenceId(), transactionID, e);
 378                 }
 379
 380                 sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
 381             } else {
 382                 // This really shouldn't happen - it likely means that persistence or replication
 383                 // took so long to complete such that the cohort entry was expired from the cache.
 384                 IllegalStateException ex = new IllegalStateException(
 385                         String.format("%s: Could not finish committing transaction %s - no CohortEntry found",
 386                                 persistenceId(), transactionID));
 387                 LOG.error(ex.getMessage());
 388                 sender.tell(new akka.actor.Status.Failure(ex), getSelf());
 389             }
 390         } else {
 391             finishCommit(sender, transactionID, cohortEntry);
 392         }
 393     }
 394
 395     private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
 396         LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionID());
 397         commitCoordinator.handleCanCommit(canCommit.getTransactionID(), getSender(), this);
 398     }
 399
 400     private void noLeaderError(Object message) {
 401         // TODO: rather than throwing an immediate exception, we could schedule a timer to try again to make
 402         // it more resilient in case we're in the process of electing a new leader.
 403         getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format(
 404             "Could not find the leader for shard %s. This typically happens" +
 405             " when the system is coming up or recovering and a leader is being elected. Try again" +
 406             " later.", persistenceId()))), getSelf());
 407     }
 408
 409     private void handleBatchedModifications(BatchedModifications batched) {
 410         // This message is sent to prepare the modificationsa transaction directly on the Shard as an
 411         // optimization to avoid the extra overhead of a separate ShardTransaction actor. On the last
 412         // BatchedModifications message, the caller sets the ready flag in the message indicating
 413         // modifications are complete. The reply contains the cohort actor path (this actor) for the caller
 414         // to initiate the 3-phase commit. This also avoids the overhead of sending an additional
 415         // ReadyTransaction message.
 416
 417         // If we're not the leader then forward to the leader. This is a safety measure - we shouldn't
 418         // normally get here if we're not the leader as the front-end (TransactionProxy) should determine
 419         // the primary/leader shard. However with timing and caching on the front-end, there's a small
 420         // window where it could have a stale leader during leadership transitions.
 421         //
 422         if(isLeader()) {
 423             try {
 424                 commitCoordinator.handleBatchedModifications(batched, getSender(), this);
 425             } catch (Exception e) {
 426                 LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(),
 427                         batched.getTransactionID(), e);
 428                 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 429             }
 430         } else {
 431             ActorSelection leader = getLeader();
 432             if(leader != null) {
 433                 // TODO: what if this is not the first batch and leadership changed in between batched messages?
 434                 // We could check if the commitCoordinator already has a cached entry and forward all the previous
 435                 // batched modifications.
 436                 LOG.debug("{}: Forwarding BatchedModifications to leader {}", persistenceId(), leader);
 437                 leader.forward(batched, getContext());
 438             } else {
 439                 noLeaderError(batched);
 440             }
 441         }
 442     }
 443
 444     private void handleReadyLocalTransaction(final ReadyLocalTransaction message) {
 445         if (isLeader()) {
 446             try {
 447                 commitCoordinator.handleReadyLocalTransaction(message, getSender(), this);
 448             } catch (Exception e) {
 449                 LOG.error("{}: Error handling LocalModifications for Tx {}", persistenceId(),
 450                         message.getTransactionID(), e);
 451                 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 452             }
 453         } else {
 454             ActorSelection leader = getLeader();
 455             if (leader != null) {
 456                 LOG.debug("{}: Forwarding LocalModifications to leader {}", persistenceId(), leader);
 457                 leader.forward(message, getContext());
 458             } else {
 459                 noLeaderError(message);
 460             }
 461         }
 462     }
 463
 464     private void handleAbortTransaction(final AbortTransaction abort) {
 465         doAbortTransaction(abort.getTransactionID(), getSender());
 466     }
 467
 468     void doAbortTransaction(final String transactionID, final ActorRef sender) {
 469         final CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
 470         if(cohortEntry != null) {
 471             LOG.debug("{}: Aborting transaction {}", persistenceId(), transactionID);
 472
 473             // We don't remove the cached cohort entry here (ie pass false) in case the Tx was
 474             // aborted during replication in which case we may still commit locally if replication
 475             // succeeds.
 476             commitCoordinator.currentTransactionComplete(transactionID, false);
 477
 478             final ListenableFuture<Void> future = cohortEntry.getCohort().abort();
 479             final ActorRef self = getSelf();
 480
 481             Futures.addCallback(future, new FutureCallback<Void>() {
 482                 @Override
 483                 public void onSuccess(final Void v) {
 484                     shardMBean.incrementAbortTransactionsCount();
 485
 486                     if(sender != null) {
 487                         sender.tell(AbortTransactionReply.INSTANCE.toSerializable(), self);
 488                     }
 489                 }
 490
 491                 @Override
 492                 public void onFailure(final Throwable t) {
 493                     LOG.error("{}: An exception happened during abort", persistenceId(), t);
 494
 495                     if(sender != null) {
 496                         sender.tell(new akka.actor.Status.Failure(t), self);
 497                     }
 498                 }
 499             });
 500         }
 501     }
 502
 503     private void handleCreateTransaction(final Object message) {
 504         if (isLeader()) {
 505             createTransaction(CreateTransaction.fromSerializable(message));
 506         } else if (getLeader() != null) {
 507             getLeader().forward(message, getContext());
 508         } else {
 509             getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format(
 510                 "Could not find leader for shard %s so transaction cannot be created. This typically happens" +
 511                 " when the system is coming up or recovering and a leader is being elected. Try again" +
 512                 " later.", persistenceId()))), getSelf());
 513         }
 514     }
 515
 516     private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
 517         store.closeTransactionChain(closeTransactionChain.getTransactionChainId());
 518     }
 519
 520     private ActorRef createTypedTransactionActor(int transactionType,
 521             ShardTransactionIdentifier transactionId, String transactionChainId,
 522             short clientVersion ) {
 523
 524         return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType),
 525                 transactionId, transactionChainId, clientVersion);
 526     }
 527
 528     private void createTransaction(CreateTransaction createTransaction) {
 529         try {
 530             ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
 531                 createTransaction.getTransactionId(), createTransaction.getTransactionChainId(),
 532                 createTransaction.getVersion());
 533
 534             getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
 535                     createTransaction.getTransactionId()).toSerializable(), getSelf());
 536         } catch (Exception e) {
 537             getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 538         }
 539     }
 540
 541     private ActorRef createTransaction(int transactionType, String remoteTransactionId,
 542             String transactionChainId, short clientVersion) {
 543
 544
 545         ShardTransactionIdentifier transactionId = new ShardTransactionIdentifier(remoteTransactionId);
 546
 547         if(LOG.isDebugEnabled()) {
 548             LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
 549         }
 550
 551         ActorRef transactionActor = createTypedTransactionActor(transactionType, transactionId,
 552                 transactionChainId, clientVersion);
 553
 554         return transactionActor;
 555     }
 556
 557     private void commitWithNewTransaction(final Modification modification) {
 558         ReadWriteShardDataTreeTransaction tx = store.newReadWriteTransaction(modification.toString(), null);
 559         modification.apply(tx.getSnapshot());
 560         try {
 561             snapshotCohort.syncCommitTransaction(tx);
 562             shardMBean.incrementCommittedTransactionCount();
 563             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 564         } catch (Exception e) {
 565             shardMBean.incrementFailedTransactionsCount();
 566             LOG.error("{}: Failed to commit", persistenceId(), e);
 567         }
 568     }
 569
 570     private void updateSchemaContext(final UpdateSchemaContext message) {
 571         updateSchemaContext(message.getSchemaContext());
 572     }
 573
 574     @VisibleForTesting
 575     void updateSchemaContext(final SchemaContext schemaContext) {
 576         store.updateSchemaContext(schemaContext);
 577     }
 578
 579     private boolean isMetricsCaptureEnabled() {
 580         CommonConfig config = new CommonConfig(getContext().system().settings().config());
 581         return config.isMetricCaptureEnabled();
 582     }
 583
 584     @Override
 585     protected RaftActorSnapshotCohort getRaftActorSnapshotCohort() {
 586         return snapshotCohort;
 587     }
 588
 589     @Override
 590     @Nonnull
 591     protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
 592         return new ShardRecoveryCoordinator(store, persistenceId(), LOG);
 593     }
 594
 595     @Override
 596     protected void onRecoveryComplete() {
 597         //notify shard manager
 598         getContext().parent().tell(new ActorInitialized(), getSelf());
 599
 600         // Being paranoid here - this method should only be called once but just in case...
 601         if(txCommitTimeoutCheckSchedule == null) {
 602             // Schedule a message to be periodically sent to check if the current in-progress
 603             // transaction should be expired and aborted.
 604             FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
 605             txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
 606                     period, period, getSelf(),
 607                     TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
 608         }
 609     }
 610
 611     @Override
 612     protected void applyState(final ActorRef clientActor, final String identifier, final Object data) {
 613         if (data instanceof DataTreeCandidatePayload) {
 614             if (clientActor == null) {
 615                 // No clientActor indicates a replica coming from the leader
 616                 try {
 617                     store.applyForeignCandidate(identifier, ((DataTreeCandidatePayload)data).getCandidate());
 618                 } catch (DataValidationFailedException | IOException e) {
 619                     LOG.error("{}: Error applying replica {}", persistenceId(), identifier, e);
 620                 }
 621             } else {
 622                 // Replication consensus reached, proceed to commit
 623                 finishCommit(clientActor, identifier);
 624             }
 625         } else if (data instanceof ModificationPayload) {
 626             try {
 627                 applyModificationToState(clientActor, identifier, ((ModificationPayload) data).getModification());
 628             } catch (ClassNotFoundException | IOException e) {
 629                 LOG.error("{}: Error extracting ModificationPayload", persistenceId(), e);
 630             }
 631         } else if (data instanceof CompositeModificationPayload) {
 632             Object modification = ((CompositeModificationPayload) data).getModification();
 633
 634             applyModificationToState(clientActor, identifier, modification);
 635         } else if(data instanceof CompositeModificationByteStringPayload ){
 636             Object modification = ((CompositeModificationByteStringPayload) data).getModification();
 637
 638             applyModificationToState(clientActor, identifier, modification);
 639         } else {
 640             LOG.error("{}: Unknown state received {} Class loader = {} CompositeNodeMod.ClassLoader = {}",
 641                     persistenceId(), data, data.getClass().getClassLoader(),
 642                     CompositeModificationPayload.class.getClassLoader());
 643         }
 644     }
 645
 646     private void applyModificationToState(ActorRef clientActor, String identifier, Object modification) {
 647         if(modification == null) {
 648             LOG.error(
 649                     "{}: modification is null - this is very unexpected, clientActor = {}, identifier = {}",
 650                     persistenceId(), identifier, clientActor != null ? clientActor.path().toString() : null);
 651         } else if(clientActor == null) {
 652             // There's no clientActor to which to send a commit reply so we must be applying
 653             // replicated state from the leader.
 654             commitWithNewTransaction(MutableCompositeModification.fromSerializable(modification));
 655         } else {
 656             // This must be the OK to commit after replication consensus.
 657             finishCommit(clientActor, identifier);
 658         }
 659     }
 660
 661     @Override
 662     protected void onStateChanged() {
 663         boolean isLeader = isLeader();
 664         changeSupport.onLeadershipChange(isLeader);
 665         treeChangeSupport.onLeadershipChange(isLeader);
 666
 667         // If this actor is no longer the leader close all the transaction chains
 668         if (!isLeader) {
 669             if(LOG.isDebugEnabled()) {
 670                 LOG.debug(
 671                     "{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader",
 672                     persistenceId(), getId());
 673             }
 674
 675             store.closeAllTransactionChains();
 676         }
 677     }
 678
 679     @Override
 680     protected void onLeaderChanged(String oldLeader, String newLeader) {
 681         shardMBean.incrementLeadershipChangeCount();
 682     }
 683
 684     @Override
 685     public String persistenceId() {
 686         return this.name;
 687     }
 688
 689     @VisibleForTesting
 690     ShardCommitCoordinator getCommitCoordinator() {
 691         return commitCoordinator;
 692     }
 693
 694
 695     private static class ShardCreator implements Creator<Shard> {
 696
 697         private static final long serialVersionUID = 1L;
 698
 699         final ShardIdentifier name;
 700         final Map<String, String> peerAddresses;
 701         final DatastoreContext datastoreContext;
 702         final SchemaContext schemaContext;
 703
 704         ShardCreator(final ShardIdentifier name, final Map<String, String> peerAddresses,
 705                 final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 706             this.name = name;
 707             this.peerAddresses = peerAddresses;
 708             this.datastoreContext = datastoreContext;
 709             this.schemaContext = schemaContext;
 710         }
 711
 712         @Override
 713         public Shard create() throws Exception {
 714             return new Shard(name, peerAddresses, datastoreContext, schemaContext);
 715         }
 716     }
 717
 718     @VisibleForTesting
 719     public ShardDataTree getDataStore() {
 720         return store;
 721     }
 722
 723     @VisibleForTesting
 724     ShardStats getShardMBean() {
 725         return shardMBean;
 726     }
 727 }