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 Object GET_SHARD_MBEAN_MESSAGE = "getShardMBeanMessage";
  89
  90     @VisibleForTesting
  91     static final String DEFAULT_NAME = "default";
  92
  93     // The state of this Shard
  94     private final ShardDataTree store;
  95
  96     /// The name of this shard
  97     private final String name;
  98
  99     private final ShardStats shardMBean;
 100
 101     private DatastoreContext datastoreContext;
 102
 103     private final ShardCommitCoordinator commitCoordinator;
 104
 105     private long transactionCommitTimeout;
 106
 107     private Cancellable txCommitTimeoutCheckSchedule;
 108
 109     private final Optional<ActorRef> roleChangeNotifier;
 110
 111     private final MessageTracker appendEntriesReplyTracker;
 112
 113     private final ShardTransactionActorFactory transactionActorFactory;
 114
 115     private final ShardSnapshotCohort snapshotCohort;
 116
 117     private final DataTreeChangeListenerSupport treeChangeSupport = new DataTreeChangeListenerSupport(this);
 118     private final DataChangeListenerSupport changeSupport = new DataChangeListenerSupport(this);
 119
 120     protected Shard(final ShardIdentifier name, final Map<String, String> peerAddresses,
 121             final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 122         super(name.toString(), new HashMap<>(peerAddresses), Optional.of(datastoreContext.getShardRaftConfig()),
 123                 DataStoreVersions.CURRENT_VERSION);
 124
 125         this.name = name.toString();
 126         this.datastoreContext = datastoreContext;
 127
 128         setPersistence(datastoreContext.isPersistent());
 129
 130         LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
 131
 132         store = new ShardDataTree(schemaContext);
 133
 134         shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
 135                 datastoreContext.getDataStoreMXBeanType());
 136         shardMBean.setShardActor(getSelf());
 137
 138         if (isMetricsCaptureEnabled()) {
 139             getContext().become(new MeteringBehavior(this));
 140         }
 141
 142         commitCoordinator = new ShardCommitCoordinator(store,
 143                 datastoreContext.getShardCommitQueueExpiryTimeoutInMillis(),
 144                 datastoreContext.getShardTransactionCommitQueueCapacity(), self(), LOG, this.name);
 145
 146         setTransactionCommitTimeout();
 147
 148         // create a notifier actor for each cluster member
 149         roleChangeNotifier = createRoleChangeNotifier(name.toString());
 150
 151         appendEntriesReplyTracker = new MessageTracker(AppendEntriesReply.class,
 152                 getRaftActorContext().getConfigParams().getIsolatedCheckIntervalInMillis());
 153
 154         transactionActorFactory = new ShardTransactionActorFactory(store, datastoreContext,
 155                 new Dispatchers(context().system().dispatchers()).getDispatcherPath(
 156                         Dispatchers.DispatcherType.Transaction), self(), getContext(), shardMBean);
 157
 158         snapshotCohort = new ShardSnapshotCohort(transactionActorFactory, store, LOG, this.name);
 159     }
 160
 161     private void setTransactionCommitTimeout() {
 162         transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
 163                 datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS) / 2;
 164     }
 165
 166     public static Props props(final ShardIdentifier name,
 167         final Map<String, String> peerAddresses,
 168         final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 169         Preconditions.checkNotNull(name, "name should not be null");
 170         Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
 171         Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
 172         Preconditions.checkNotNull(schemaContext, "schemaContext should not be null");
 173
 174         return Props.create(new ShardCreator(name, peerAddresses, datastoreContext, schemaContext));
 175     }
 176
 177     private Optional<ActorRef> createRoleChangeNotifier(String shardId) {
 178         ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
 179             RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
 180         return Optional.of(shardRoleChangeNotifier);
 181     }
 182
 183     @Override
 184     public void postStop() {
 185         LOG.info("Stopping Shard {}", persistenceId());
 186
 187         super.postStop();
 188
 189         if(txCommitTimeoutCheckSchedule != null) {
 190             txCommitTimeoutCheckSchedule.cancel();
 191         }
 192
 193         shardMBean.unregisterMBean();
 194     }
 195
 196     @Override
 197     public void onReceiveRecover(final Object message) throws Exception {
 198         if(LOG.isDebugEnabled()) {
 199             LOG.debug("{}: onReceiveRecover: Received message {} from {}", persistenceId(),
 200                 message.getClass().toString(), getSender());
 201         }
 202
 203         if (message instanceof RecoveryFailure){
 204             LOG.error("{}: Recovery failed because of this cause",
 205                     persistenceId(), ((RecoveryFailure) message).cause());
 206
 207             // Even though recovery failed, we still need to finish our recovery, eg send the
 208             // ActorInitialized message and start the txCommitTimeoutCheckSchedule.
 209             onRecoveryComplete();
 210         } else {
 211             super.onReceiveRecover(message);
 212             if(LOG.isTraceEnabled()) {
 213                 appendEntriesReplyTracker.begin();
 214             }
 215         }
 216     }
 217
 218     @Override
 219     public void onReceiveCommand(final Object message) throws Exception {
 220
 221         MessageTracker.Context context = appendEntriesReplyTracker.received(message);
 222
 223         if(context.error().isPresent()){
 224             LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(),
 225                     context.error());
 226         }
 227
 228         try {
 229             if (CreateTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 230                 handleCreateTransaction(message);
 231             } else if (BatchedModifications.class.isInstance(message)) {
 232                 handleBatchedModifications((BatchedModifications)message);
 233             } else if (message instanceof ForwardedReadyTransaction) {
 234                 commitCoordinator.handleForwardedReadyTransaction((ForwardedReadyTransaction) message,
 235                         getSender(), this);
 236             } else if (message instanceof ReadyLocalTransaction) {
 237                 handleReadyLocalTransaction((ReadyLocalTransaction)message);
 238             } else if (CanCommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 239                 handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
 240             } else if (CommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 241                 handleCommitTransaction(CommitTransaction.fromSerializable(message));
 242             } else if (AbortTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
 243                 handleAbortTransaction(AbortTransaction.fromSerializable(message));
 244             } else if (CloseTransactionChain.SERIALIZABLE_CLASS.isInstance(message)) {
 245                 closeTransactionChain(CloseTransactionChain.fromSerializable(message));
 246             } else if (message instanceof RegisterChangeListener) {
 247                 changeSupport.onMessage((RegisterChangeListener) message, isLeader());
 248             } else if (message instanceof RegisterDataTreeChangeListener) {
 249                 treeChangeSupport.onMessage((RegisterDataTreeChangeListener) message, isLeader());
 250             } else if (message instanceof UpdateSchemaContext) {
 251                 updateSchemaContext((UpdateSchemaContext) message);
 252             } else if (message instanceof PeerAddressResolved) {
 253                 PeerAddressResolved resolved = (PeerAddressResolved) message;
 254                 setPeerAddress(resolved.getPeerId().toString(),
 255                         resolved.getPeerAddress());
 256             } else if (message.equals(TX_COMMIT_TIMEOUT_CHECK_MESSAGE)) {
 257                 handleTransactionCommitTimeoutCheck();
 258             } else if(message instanceof DatastoreContext) {
 259                 onDatastoreContext((DatastoreContext)message);
 260             } else if(message instanceof RegisterRoleChangeListener){
 261                 roleChangeNotifier.get().forward(message, context());
 262             } else if (message instanceof FollowerInitialSyncUpStatus) {
 263                 shardMBean.setFollowerInitialSyncStatus(((FollowerInitialSyncUpStatus) message).isInitialSyncDone());
 264                 context().parent().tell(message, self());
 265             } else if(GET_SHARD_MBEAN_MESSAGE.equals(message)){
 266                 sender().tell(getShardMBean(), self());
 267             } else {
 268                 super.onReceiveCommand(message);
 269             }
 270         } finally {
 271             context.done();
 272         }
 273     }
 274
 275     @Override
 276     protected Optional<ActorRef> getRoleChangeNotifier() {
 277         return roleChangeNotifier;
 278     }
 279
 280     @Override
 281     protected LeaderStateChanged newLeaderStateChanged(String memberId, String leaderId) {
 282         return new ShardLeaderStateChanged(memberId, leaderId,
 283                 isLeader() ? Optional.<DataTree>of(store.getDataTree()) : Optional.<DataTree>absent());
 284     }
 285
 286     private void onDatastoreContext(DatastoreContext context) {
 287         datastoreContext = context;
 288
 289         commitCoordinator.setQueueCapacity(datastoreContext.getShardTransactionCommitQueueCapacity());
 290
 291         setTransactionCommitTimeout();
 292
 293         if(datastoreContext.isPersistent() && !persistence().isRecoveryApplicable()) {
 294             setPersistence(true);
 295         } else if(!datastoreContext.isPersistent() && persistence().isRecoveryApplicable()) {
 296             setPersistence(false);
 297         }
 298
 299         updateConfigParams(datastoreContext.getShardRaftConfig());
 300     }
 301
 302     private void handleTransactionCommitTimeoutCheck() {
 303         CohortEntry cohortEntry = commitCoordinator.getCurrentCohortEntry();
 304         if(cohortEntry != null) {
 305             if(cohortEntry.isExpired(transactionCommitTimeout)) {
 306                 LOG.warn("{}: Current transaction {} has timed out after {} ms - aborting",
 307                         persistenceId(), cohortEntry.getTransactionID(), transactionCommitTimeout);
 308
 309                 doAbortTransaction(cohortEntry.getTransactionID(), null);
 310             }
 311         }
 312
 313         commitCoordinator.cleanupExpiredCohortEntries();
 314     }
 315
 316     private static boolean isEmptyCommit(final DataTreeCandidate candidate) {
 317         return ModificationType.UNMODIFIED.equals(candidate.getRootNode().getModificationType());
 318     }
 319
 320     void continueCommit(final CohortEntry cohortEntry) throws Exception {
 321         final DataTreeCandidate candidate = cohortEntry.getCohort().getCandidate();
 322
 323         // If we do not have any followers and we are not using persistence
 324         // or if cohortEntry has no modifications
 325         // we can apply modification to the state immediately
 326         if ((!hasFollowers() && !persistence().isRecoveryApplicable()) || isEmptyCommit(candidate)) {
 327             applyModificationToState(cohortEntry.getReplySender(), cohortEntry.getTransactionID(), candidate);
 328         } else {
 329             Shard.this.persistData(cohortEntry.getReplySender(), cohortEntry.getTransactionID(),
 330                 DataTreeCandidatePayload.create(candidate));
 331         }
 332     }
 333
 334     private void handleCommitTransaction(final CommitTransaction commit) {
 335         if(!commitCoordinator.handleCommit(commit.getTransactionID(), getSender(), this)) {
 336             shardMBean.incrementFailedTransactionsCount();
 337         }
 338     }
 339
 340     private void finishCommit(@Nonnull final ActorRef sender, @Nonnull final String transactionID, @Nonnull final CohortEntry cohortEntry) {
 341         LOG.debug("{}: Finishing commit for transaction {}", persistenceId(), cohortEntry.getTransactionID());
 342
 343         try {
 344             // We block on the future here so we don't have to worry about possibly accessing our
 345             // state on a different thread outside of our dispatcher. Also, the data store
 346             // currently uses a same thread executor anyway.
 347             cohortEntry.getCohort().commit().get();
 348
 349             sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
 350
 351             shardMBean.incrementCommittedTransactionCount();
 352             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 353
 354         } catch (Exception e) {
 355             sender.tell(new akka.actor.Status.Failure(e), getSelf());
 356
 357             LOG.error("{}, An exception occurred while committing transaction {}", persistenceId(),
 358                     transactionID, e);
 359             shardMBean.incrementFailedTransactionsCount();
 360         } finally {
 361             commitCoordinator.currentTransactionComplete(transactionID, true);
 362         }
 363     }
 364
 365     private void finishCommit(@Nonnull final ActorRef sender, final @Nonnull String transactionID) {
 366         // With persistence enabled, this method is called via applyState by the leader strategy
 367         // after the commit has been replicated to a majority of the followers.
 368
 369         CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
 370         if (cohortEntry == null) {
 371             // The transaction is no longer the current commit. This can happen if the transaction
 372             // was aborted prior, most likely due to timeout in the front-end. We need to finish
 373             // committing the transaction though since it was successfully persisted and replicated
 374             // however we can't use the original cohort b/c it was already preCommitted and may
 375             // conflict with the current commit or may have been aborted so we commit with a new
 376             // transaction.
 377             cohortEntry = commitCoordinator.getAndRemoveCohortEntry(transactionID);
 378             if(cohortEntry != null) {
 379                 try {
 380                     store.applyForeignCandidate(transactionID, cohortEntry.getCohort().getCandidate());
 381                 } catch (DataValidationFailedException e) {
 382                     shardMBean.incrementFailedTransactionsCount();
 383                     LOG.error("{}: Failed to re-apply transaction {}", persistenceId(), transactionID, e);
 384                 }
 385
 386                 sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
 387             } else {
 388                 // This really shouldn't happen - it likely means that persistence or replication
 389                 // took so long to complete such that the cohort entry was expired from the cache.
 390                 IllegalStateException ex = new IllegalStateException(
 391                         String.format("%s: Could not finish committing transaction %s - no CohortEntry found",
 392                                 persistenceId(), transactionID));
 393                 LOG.error(ex.getMessage());
 394                 sender.tell(new akka.actor.Status.Failure(ex), getSelf());
 395             }
 396         } else {
 397             finishCommit(sender, transactionID, cohortEntry);
 398         }
 399     }
 400
 401     private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
 402         LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionID());
 403         commitCoordinator.handleCanCommit(canCommit.getTransactionID(), getSender(), this);
 404     }
 405
 406     private void noLeaderError(Object message) {
 407         // TODO: rather than throwing an immediate exception, we could schedule a timer to try again to make
 408         // it more resilient in case we're in the process of electing a new leader.
 409         getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format(
 410             "Could not find the leader for shard %s. This typically happens" +
 411             " when the system is coming up or recovering and a leader is being elected. Try again" +
 412             " later.", persistenceId()))), getSelf());
 413     }
 414
 415     private void handleBatchedModifications(BatchedModifications batched) {
 416         // This message is sent to prepare the modifications transaction directly on the Shard as an
 417         // optimization to avoid the extra overhead of a separate ShardTransaction actor. On the last
 418         // BatchedModifications message, the caller sets the ready flag in the message indicating
 419         // modifications are complete. The reply contains the cohort actor path (this actor) for the caller
 420         // to initiate the 3-phase commit. This also avoids the overhead of sending an additional
 421         // ReadyTransaction message.
 422
 423         // If we're not the leader then forward to the leader. This is a safety measure - we shouldn't
 424         // normally get here if we're not the leader as the front-end (TransactionProxy) should determine
 425         // the primary/leader shard. However with timing and caching on the front-end, there's a small
 426         // window where it could have a stale leader during leadership transitions.
 427         //
 428         if(isLeader()) {
 429             try {
 430                 commitCoordinator.handleBatchedModifications(batched, getSender(), this);
 431             } catch (Exception e) {
 432                 LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(),
 433                         batched.getTransactionID(), e);
 434                 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 435             }
 436         } else {
 437             ActorSelection leader = getLeader();
 438             if(leader != null) {
 439                 // TODO: what if this is not the first batch and leadership changed in between batched messages?
 440                 // We could check if the commitCoordinator already has a cached entry and forward all the previous
 441                 // batched modifications.
 442                 LOG.debug("{}: Forwarding BatchedModifications to leader {}", persistenceId(), leader);
 443                 leader.forward(batched, getContext());
 444             } else {
 445                 noLeaderError(batched);
 446             }
 447         }
 448     }
 449
 450     private void handleReadyLocalTransaction(final ReadyLocalTransaction message) {
 451         if (isLeader()) {
 452             try {
 453                 commitCoordinator.handleReadyLocalTransaction(message, getSender(), this);
 454             } catch (Exception e) {
 455                 LOG.error("{}: Error handling LocalModifications for Tx {}", persistenceId(),
 456                         message.getTransactionID(), e);
 457                 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 458             }
 459         } else {
 460             ActorSelection leader = getLeader();
 461             if (leader != null) {
 462                 LOG.debug("{}: Forwarding LocalModifications to leader {}", persistenceId(), leader);
 463                 leader.forward(message, getContext());
 464             } else {
 465                 noLeaderError(message);
 466             }
 467         }
 468     }
 469
 470     private void handleAbortTransaction(final AbortTransaction abort) {
 471         doAbortTransaction(abort.getTransactionID(), getSender());
 472     }
 473
 474     void doAbortTransaction(final String transactionID, final ActorRef sender) {
 475         final CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
 476         if(cohortEntry != null) {
 477             LOG.debug("{}: Aborting transaction {}", persistenceId(), transactionID);
 478
 479             // We don't remove the cached cohort entry here (ie pass false) in case the Tx was
 480             // aborted during replication in which case we may still commit locally if replication
 481             // succeeds.
 482             commitCoordinator.currentTransactionComplete(transactionID, false);
 483
 484             final ListenableFuture<Void> future = cohortEntry.getCohort().abort();
 485             final ActorRef self = getSelf();
 486
 487             Futures.addCallback(future, new FutureCallback<Void>() {
 488                 @Override
 489                 public void onSuccess(final Void v) {
 490                     shardMBean.incrementAbortTransactionsCount();
 491
 492                     if(sender != null) {
 493                         sender.tell(AbortTransactionReply.INSTANCE.toSerializable(), self);
 494                     }
 495                 }
 496
 497                 @Override
 498                 public void onFailure(final Throwable t) {
 499                     LOG.error("{}: An exception happened during abort", persistenceId(), t);
 500
 501                     if(sender != null) {
 502                         sender.tell(new akka.actor.Status.Failure(t), self);
 503                     }
 504                 }
 505             });
 506         }
 507     }
 508
 509     private void handleCreateTransaction(final Object message) {
 510         if (isLeader()) {
 511             createTransaction(CreateTransaction.fromSerializable(message));
 512         } else if (getLeader() != null) {
 513             getLeader().forward(message, getContext());
 514         } else {
 515             getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format(
 516                 "Could not find leader for shard %s so transaction cannot be created. This typically happens" +
 517                 " when the system is coming up or recovering and a leader is being elected. Try again" +
 518                 " later.", persistenceId()))), getSelf());
 519         }
 520     }
 521
 522     private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
 523         store.closeTransactionChain(closeTransactionChain.getTransactionChainId());
 524     }
 525
 526     private ActorRef createTypedTransactionActor(int transactionType,
 527             ShardTransactionIdentifier transactionId, String transactionChainId,
 528             short clientVersion ) {
 529
 530         return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType),
 531                 transactionId, transactionChainId, clientVersion);
 532     }
 533
 534     private void createTransaction(CreateTransaction createTransaction) {
 535         try {
 536             ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
 537                 createTransaction.getTransactionId(), createTransaction.getTransactionChainId(),
 538                 createTransaction.getVersion());
 539
 540             getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
 541                     createTransaction.getTransactionId()).toSerializable(), getSelf());
 542         } catch (Exception e) {
 543             getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 544         }
 545     }
 546
 547     private ActorRef createTransaction(int transactionType, String remoteTransactionId,
 548             String transactionChainId, short clientVersion) {
 549
 550
 551         ShardTransactionIdentifier transactionId = new ShardTransactionIdentifier(remoteTransactionId);
 552
 553         if(LOG.isDebugEnabled()) {
 554             LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
 555         }
 556
 557         ActorRef transactionActor = createTypedTransactionActor(transactionType, transactionId,
 558                 transactionChainId, clientVersion);
 559
 560         return transactionActor;
 561     }
 562
 563     private void commitWithNewTransaction(final Modification modification) {
 564         ReadWriteShardDataTreeTransaction tx = store.newReadWriteTransaction(modification.toString(), null);
 565         modification.apply(tx.getSnapshot());
 566         try {
 567             snapshotCohort.syncCommitTransaction(tx);
 568             shardMBean.incrementCommittedTransactionCount();
 569             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 570         } catch (Exception e) {
 571             shardMBean.incrementFailedTransactionsCount();
 572             LOG.error("{}: Failed to commit", persistenceId(), e);
 573         }
 574     }
 575
 576     private void updateSchemaContext(final UpdateSchemaContext message) {
 577         updateSchemaContext(message.getSchemaContext());
 578     }
 579
 580     @VisibleForTesting
 581     void updateSchemaContext(final SchemaContext schemaContext) {
 582         store.updateSchemaContext(schemaContext);
 583     }
 584
 585     private boolean isMetricsCaptureEnabled() {
 586         CommonConfig config = new CommonConfig(getContext().system().settings().config());
 587         return config.isMetricCaptureEnabled();
 588     }
 589
 590     @Override
 591     protected RaftActorSnapshotCohort getRaftActorSnapshotCohort() {
 592         return snapshotCohort;
 593     }
 594
 595     @Override
 596     @Nonnull
 597     protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
 598         return new ShardRecoveryCoordinator(store, persistenceId(), LOG);
 599     }
 600
 601     @Override
 602     protected void onRecoveryComplete() {
 603         store.recoveryDone();
 604         //notify shard manager
 605         getContext().parent().tell(new ActorInitialized(), getSelf());
 606
 607         // Being paranoid here - this method should only be called once but just in case...
 608         if(txCommitTimeoutCheckSchedule == null) {
 609             // Schedule a message to be periodically sent to check if the current in-progress
 610             // transaction should be expired and aborted.
 611             FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
 612             txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
 613                     period, period, getSelf(),
 614                     TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
 615         }
 616     }
 617
 618     @Override
 619     protected void applyState(final ActorRef clientActor, final String identifier, final Object data) {
 620         if (data instanceof DataTreeCandidatePayload) {
 621             if (clientActor == null) {
 622                 // No clientActor indicates a replica coming from the leader
 623                 try {
 624                     store.applyForeignCandidate(identifier, ((DataTreeCandidatePayload)data).getCandidate());
 625                 } catch (DataValidationFailedException | IOException e) {
 626                     LOG.error("{}: Error applying replica {}", persistenceId(), identifier, e);
 627                 }
 628             } else {
 629                 // Replication consensus reached, proceed to commit
 630                 finishCommit(clientActor, identifier);
 631             }
 632         } else if (data instanceof ModificationPayload) {
 633             try {
 634                 applyModificationToState(clientActor, identifier, ((ModificationPayload) data).getModification());
 635             } catch (ClassNotFoundException | IOException e) {
 636                 LOG.error("{}: Error extracting ModificationPayload", persistenceId(), e);
 637             }
 638         } else if (data instanceof CompositeModificationPayload) {
 639             Object modification = ((CompositeModificationPayload) data).getModification();
 640
 641             applyModificationToState(clientActor, identifier, modification);
 642         } else if(data instanceof CompositeModificationByteStringPayload ){
 643             Object modification = ((CompositeModificationByteStringPayload) data).getModification();
 644
 645             applyModificationToState(clientActor, identifier, modification);
 646         } else {
 647             LOG.error("{}: Unknown state received {} Class loader = {} CompositeNodeMod.ClassLoader = {}",
 648                     persistenceId(), data, data.getClass().getClassLoader(),
 649                     CompositeModificationPayload.class.getClassLoader());
 650         }
 651     }
 652
 653     private void applyModificationToState(ActorRef clientActor, String identifier, Object modification) {
 654         if(modification == null) {
 655             LOG.error(
 656                     "{}: modification is null - this is very unexpected, clientActor = {}, identifier = {}",
 657                     persistenceId(), identifier, clientActor != null ? clientActor.path().toString() : null);
 658         } else if(clientActor == null) {
 659             // There's no clientActor to which to send a commit reply so we must be applying
 660             // replicated state from the leader.
 661             commitWithNewTransaction(MutableCompositeModification.fromSerializable(modification));
 662         } else {
 663             // This must be the OK to commit after replication consensus.
 664             finishCommit(clientActor, identifier);
 665         }
 666     }
 667
 668     @Override
 669     protected void onStateChanged() {
 670         boolean isLeader = isLeader();
 671         changeSupport.onLeadershipChange(isLeader);
 672         treeChangeSupport.onLeadershipChange(isLeader);
 673
 674         // If this actor is no longer the leader close all the transaction chains
 675         if (!isLeader) {
 676             if(LOG.isDebugEnabled()) {
 677                 LOG.debug(
 678                     "{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader",
 679                     persistenceId(), getId());
 680             }
 681
 682             store.closeAllTransactionChains();
 683         }
 684     }
 685
 686     @Override
 687     protected void onLeaderChanged(String oldLeader, String newLeader) {
 688         shardMBean.incrementLeadershipChangeCount();
 689     }
 690
 691     @Override
 692     public String persistenceId() {
 693         return this.name;
 694     }
 695
 696     @VisibleForTesting
 697     ShardCommitCoordinator getCommitCoordinator() {
 698         return commitCoordinator;
 699     }
 700
 701
 702     private static class ShardCreator implements Creator<Shard> {
 703
 704         private static final long serialVersionUID = 1L;
 705
 706         final ShardIdentifier name;
 707         final Map<String, String> peerAddresses;
 708         final DatastoreContext datastoreContext;
 709         final SchemaContext schemaContext;
 710
 711         ShardCreator(final ShardIdentifier name, final Map<String, String> peerAddresses,
 712                 final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
 713             this.name = name;
 714             this.peerAddresses = peerAddresses;
 715             this.datastoreContext = datastoreContext;
 716             this.schemaContext = schemaContext;
 717         }
 718
 719         @Override
 720         public Shard create() throws Exception {
 721             return new Shard(name, peerAddresses, datastoreContext, schemaContext);
 722         }
 723     }
 724
 725     @VisibleForTesting
 726     public ShardDataTree getDataStore() {
 727         return store;
 728     }
 729
 730     @VisibleForTesting
 731     ShardStats getShardMBean() {
 732         return shardMBean;
 733     }
 734 }