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.persistence.RecoveryFailure;
  16 import akka.serialization.Serialization;
  17 import com.google.common.annotations.VisibleForTesting;
  18 import com.google.common.base.Optional;
  19 import com.google.common.base.Preconditions;
  20 import java.io.IOException;
  21 import java.util.Collections;
  22 import java.util.Map;
  23 import java.util.concurrent.TimeUnit;
  24 import javax.annotation.Nonnull;
  25 import org.opendaylight.controller.cluster.common.actor.CommonConfig;
  26 import org.opendaylight.controller.cluster.common.actor.MeteringBehavior;
  27 import org.opendaylight.controller.cluster.datastore.ShardCommitCoordinator.CohortEntry;
  28 import org.opendaylight.controller.cluster.datastore.exceptions.NoShardLeaderException;
  29 import org.opendaylight.controller.cluster.datastore.identifiers.ShardIdentifier;
  30 import org.opendaylight.controller.cluster.datastore.identifiers.ShardTransactionIdentifier;
  31 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardMBeanFactory;
  32 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardStats;
  33 import org.opendaylight.controller.cluster.datastore.messages.AbortTransaction;
  34 import org.opendaylight.controller.cluster.datastore.messages.ActorInitialized;
  35 import org.opendaylight.controller.cluster.datastore.messages.BatchedModifications;
  36 import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransaction;
  37 import org.opendaylight.controller.cluster.datastore.messages.CloseTransactionChain;
  38 import org.opendaylight.controller.cluster.datastore.messages.CommitTransaction;
  39 import org.opendaylight.controller.cluster.datastore.messages.CommitTransactionReply;
  40 import org.opendaylight.controller.cluster.datastore.messages.CreateTransaction;
  41 import org.opendaylight.controller.cluster.datastore.messages.CreateTransactionReply;
  42 import org.opendaylight.controller.cluster.datastore.messages.DatastoreSnapshot;
  43 import org.opendaylight.controller.cluster.datastore.messages.DatastoreSnapshot.ShardSnapshot;
  44 import org.opendaylight.controller.cluster.datastore.messages.ForwardedReadyTransaction;
  45 import org.opendaylight.controller.cluster.datastore.messages.GetShardDataTree;
  46 import org.opendaylight.controller.cluster.datastore.messages.PeerAddressResolved;
  47 import org.opendaylight.controller.cluster.datastore.messages.ReadyLocalTransaction;
  48 import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListener;
  49 import org.opendaylight.controller.cluster.datastore.messages.RegisterDataTreeChangeListener;
  50 import org.opendaylight.controller.cluster.datastore.messages.ShardLeaderStateChanged;
  51 import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext;
  52 import org.opendaylight.controller.cluster.datastore.modification.Modification;
  53 import org.opendaylight.controller.cluster.datastore.modification.MutableCompositeModification;
  54 import org.opendaylight.controller.cluster.datastore.utils.Dispatchers;
  55 import org.opendaylight.controller.cluster.datastore.utils.MessageTracker;
  56 import org.opendaylight.controller.cluster.notifications.LeaderStateChanged;
  57 import org.opendaylight.controller.cluster.notifications.RegisterRoleChangeListener;
  58 import org.opendaylight.controller.cluster.notifications.RoleChangeNotifier;
  59 import org.opendaylight.controller.cluster.raft.RaftActor;
  60 import org.opendaylight.controller.cluster.raft.RaftActorRecoveryCohort;
  61 import org.opendaylight.controller.cluster.raft.RaftActorSnapshotCohort;
  62 import org.opendaylight.controller.cluster.raft.RaftState;
  63 import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus;
  64 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  65 import org.opendaylight.controller.cluster.raft.messages.ServerRemoved;
  66 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationByteStringPayload;
  67 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationPayload;
  68 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTree;
  69 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidate;
  70 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException;
  71 import org.opendaylight.yangtools.yang.data.api.schema.tree.ModificationType;
  72 import org.opendaylight.yangtools.yang.data.api.schema.tree.TreeType;
  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     protected 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
 121     private ShardSnapshot restoreFromSnapshot;
 122
 123     private final ShardTransactionMessageRetrySupport messageRetrySupport;
 124
 125     protected Shard(AbstractBuilder<?, ?> builder) {
 126         super(builder.getId().toString(), builder.getPeerAddresses(),
 127                 Optional.of(builder.getDatastoreContext().getShardRaftConfig()), DataStoreVersions.CURRENT_VERSION);
 128
 129         this.name = builder.getId().toString();
 130         this.datastoreContext = builder.getDatastoreContext();
 131         this.restoreFromSnapshot = builder.getRestoreFromSnapshot();
 132
 133         setPersistence(datastoreContext.isPersistent());
 134
 135         LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
 136
 137         store = new ShardDataTree(builder.getSchemaContext(), builder.getTreeType());
 138
 139         shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
 140                 datastoreContext.getDataStoreMXBeanType());
 141         shardMBean.setShard(this);
 142
 143         if (isMetricsCaptureEnabled()) {
 144             getContext().become(new MeteringBehavior(this));
 145         }
 146
 147         commitCoordinator = new ShardCommitCoordinator(store,
 148                 datastoreContext.getShardCommitQueueExpiryTimeoutInMillis(),
 149                 datastoreContext.getShardTransactionCommitQueueCapacity(), LOG, this.name);
 150
 151         setTransactionCommitTimeout();
 152
 153         // create a notifier actor for each cluster member
 154         roleChangeNotifier = createRoleChangeNotifier(name.toString());
 155
 156         appendEntriesReplyTracker = new MessageTracker(AppendEntriesReply.class,
 157                 getRaftActorContext().getConfigParams().getIsolatedCheckIntervalInMillis());
 158
 159         transactionActorFactory = new ShardTransactionActorFactory(store, datastoreContext,
 160                 new Dispatchers(context().system().dispatchers()).getDispatcherPath(
 161                         Dispatchers.DispatcherType.Transaction), self(), getContext(), shardMBean);
 162
 163         snapshotCohort = new ShardSnapshotCohort(transactionActorFactory, store, LOG, this.name);
 164
 165         messageRetrySupport = new ShardTransactionMessageRetrySupport(this);
 166     }
 167
 168     private void setTransactionCommitTimeout() {
 169         transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
 170                 datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS) / 2;
 171     }
 172
 173     private Optional<ActorRef> createRoleChangeNotifier(String shardId) {
 174         ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
 175             RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
 176         return Optional.of(shardRoleChangeNotifier);
 177     }
 178
 179     @Override
 180     public void postStop() {
 181         LOG.info("Stopping Shard {}", persistenceId());
 182
 183         super.postStop();
 184
 185         messageRetrySupport.close();
 186
 187         if(txCommitTimeoutCheckSchedule != null) {
 188             txCommitTimeoutCheckSchedule.cancel();
 189         }
 190
 191         commitCoordinator.abortPendingTransactions("Transaction aborted due to shutdown.", this);
 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.isSerializedType(message)) {
 230                 handleCreateTransaction(message);
 231             } else if (BatchedModifications.class.isInstance(message)) {
 232                 handleBatchedModifications((BatchedModifications)message);
 233             } else if (message instanceof ForwardedReadyTransaction) {
 234                 handleForwardedReadyTransaction((ForwardedReadyTransaction) message);
 235             } else if (message instanceof ReadyLocalTransaction) {
 236                 handleReadyLocalTransaction((ReadyLocalTransaction)message);
 237             } else if (CanCommitTransaction.isSerializedType(message)) {
 238                 handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
 239             } else if (CommitTransaction.isSerializedType(message)) {
 240                 handleCommitTransaction(CommitTransaction.fromSerializable(message));
 241             } else if (AbortTransaction.isSerializedType(message)) {
 242                 handleAbortTransaction(AbortTransaction.fromSerializable(message));
 243             } else if (CloseTransactionChain.isSerializedType(message)) {
 244                 closeTransactionChain(CloseTransactionChain.fromSerializable(message));
 245             } else if (message instanceof RegisterChangeListener) {
 246                 changeSupport.onMessage((RegisterChangeListener) message, isLeader(), hasLeader());
 247             } else if (message instanceof RegisterDataTreeChangeListener) {
 248                 treeChangeSupport.onMessage((RegisterDataTreeChangeListener) message, isLeader(), hasLeader());
 249             } else if (message instanceof UpdateSchemaContext) {
 250                 updateSchemaContext((UpdateSchemaContext) message);
 251             } else if (message instanceof PeerAddressResolved) {
 252                 PeerAddressResolved resolved = (PeerAddressResolved) message;
 253                 setPeerAddress(resolved.getPeerId().toString(),
 254                         resolved.getPeerAddress());
 255             } else if (message.equals(TX_COMMIT_TIMEOUT_CHECK_MESSAGE)) {
 256                 commitCoordinator.checkForExpiredTransactions(transactionCommitTimeout, this);
 257             } else if(message instanceof DatastoreContext) {
 258                 onDatastoreContext((DatastoreContext)message);
 259             } else if(message instanceof RegisterRoleChangeListener){
 260                 roleChangeNotifier.get().forward(message, context());
 261             } else if (message instanceof FollowerInitialSyncUpStatus) {
 262                 shardMBean.setFollowerInitialSyncStatus(((FollowerInitialSyncUpStatus) message).isInitialSyncDone());
 263                 context().parent().tell(message, self());
 264             } else if(GET_SHARD_MBEAN_MESSAGE.equals(message)){
 265                 sender().tell(getShardMBean(), self());
 266             } else if(message instanceof GetShardDataTree) {
 267                 sender().tell(store.getDataTree(), self());
 268             } else if(message instanceof ServerRemoved){
 269                 context().parent().forward(message, context());
 270             } else if(ShardTransactionMessageRetrySupport.TIMER_MESSAGE_CLASS.isInstance(message)) {
 271                 messageRetrySupport.onTimerMessage(message);
 272             } else {
 273                 super.onReceiveCommand(message);
 274             }
 275         } finally {
 276             context.done();
 277         }
 278     }
 279
 280     private boolean hasLeader() {
 281         return getLeaderId() != null;
 282     }
 283
 284     public int getPendingTxCommitQueueSize() {
 285         return commitCoordinator.getQueueSize();
 286     }
 287
 288     public int getCohortCacheSize() {
 289         return commitCoordinator.getCohortCacheSize();
 290     }
 291
 292     @Override
 293     protected Optional<ActorRef> getRoleChangeNotifier() {
 294         return roleChangeNotifier;
 295     }
 296
 297     @Override
 298     protected LeaderStateChanged newLeaderStateChanged(String memberId, String leaderId, short leaderPayloadVersion) {
 299         return new ShardLeaderStateChanged(memberId, leaderId,
 300                 isLeader() ? Optional.<DataTree>of(store.getDataTree()) : Optional.<DataTree>absent(),
 301                 leaderPayloadVersion);
 302     }
 303
 304     protected void onDatastoreContext(DatastoreContext context) {
 305         datastoreContext = context;
 306
 307         commitCoordinator.setQueueCapacity(datastoreContext.getShardTransactionCommitQueueCapacity());
 308
 309         setTransactionCommitTimeout();
 310
 311         if(datastoreContext.isPersistent() && !persistence().isRecoveryApplicable()) {
 312             setPersistence(true);
 313         } else if(!datastoreContext.isPersistent() && persistence().isRecoveryApplicable()) {
 314             setPersistence(false);
 315         }
 316
 317         updateConfigParams(datastoreContext.getShardRaftConfig());
 318     }
 319
 320     private static boolean isEmptyCommit(final DataTreeCandidate candidate) {
 321         return ModificationType.UNMODIFIED.equals(candidate.getRootNode().getModificationType());
 322     }
 323
 324     void continueCommit(final CohortEntry cohortEntry) {
 325         final DataTreeCandidate candidate = cohortEntry.getCandidate();
 326
 327         // If we do not have any followers and we are not using persistence
 328         // or if cohortEntry has no modifications
 329         // we can apply modification to the state immediately
 330         if ((!hasFollowers() && !persistence().isRecoveryApplicable()) || isEmptyCommit(candidate)) {
 331             applyModificationToState(cohortEntry.getReplySender(), cohortEntry.getTransactionID(), candidate);
 332         } else {
 333             Shard.this.persistData(cohortEntry.getReplySender(), cohortEntry.getTransactionID(),
 334                     DataTreeCandidatePayload.create(candidate));
 335         }
 336     }
 337
 338     private void handleCommitTransaction(final CommitTransaction commit) {
 339         if(!commitCoordinator.handleCommit(commit.getTransactionID(), getSender(), this)) {
 340             shardMBean.incrementFailedTransactionsCount();
 341         }
 342     }
 343
 344     private void finishCommit(@Nonnull final ActorRef sender, @Nonnull final String transactionID, @Nonnull final CohortEntry cohortEntry) {
 345         LOG.debug("{}: Finishing commit for transaction {}", persistenceId(), cohortEntry.getTransactionID());
 346
 347         try {
 348             cohortEntry.commit();
 349
 350             sender.tell(CommitTransactionReply.instance(cohortEntry.getClientVersion()).toSerializable(), getSelf());
 351
 352             shardMBean.incrementCommittedTransactionCount();
 353             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 354
 355         } catch (Exception e) {
 356             sender.tell(new akka.actor.Status.Failure(e), getSelf());
 357
 358             LOG.error("{}, An exception occurred while committing transaction {}", persistenceId(),
 359                     transactionID, e);
 360             shardMBean.incrementFailedTransactionsCount();
 361         } finally {
 362             commitCoordinator.currentTransactionComplete(transactionID, true);
 363         }
 364     }
 365
 366     private void finishCommit(@Nonnull final ActorRef sender, final @Nonnull String transactionID) {
 367         // With persistence enabled, this method is called via applyState by the leader strategy
 368         // after the commit has been replicated to a majority of the followers.
 369
 370         CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
 371         if (cohortEntry == null) {
 372             // The transaction is no longer the current commit. This can happen if the transaction
 373             // was aborted prior, most likely due to timeout in the front-end. We need to finish
 374             // committing the transaction though since it was successfully persisted and replicated
 375             // however we can't use the original cohort b/c it was already preCommitted and may
 376             // conflict with the current commit or may have been aborted so we commit with a new
 377             // transaction.
 378             cohortEntry = commitCoordinator.getAndRemoveCohortEntry(transactionID);
 379             if(cohortEntry != null) {
 380                 try {
 381                     store.applyForeignCandidate(transactionID, cohortEntry.getCandidate());
 382                 } catch (DataValidationFailedException e) {
 383                     shardMBean.incrementFailedTransactionsCount();
 384                     LOG.error("{}: Failed to re-apply transaction {}", persistenceId(), transactionID, e);
 385                 }
 386
 387                 sender.tell(CommitTransactionReply.instance(cohortEntry.getClientVersion()).toSerializable(),
 388                         getSelf());
 389             } else {
 390                 // This really shouldn't happen - it likely means that persistence or replication
 391                 // took so long to complete such that the cohort entry was expired from the cache.
 392                 IllegalStateException ex = new IllegalStateException(
 393                         String.format("%s: Could not finish committing transaction %s - no CohortEntry found",
 394                                 persistenceId(), transactionID));
 395                 LOG.error(ex.getMessage());
 396                 sender.tell(new akka.actor.Status.Failure(ex), getSelf());
 397             }
 398         } else {
 399             finishCommit(sender, transactionID, cohortEntry);
 400         }
 401     }
 402
 403     private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
 404         LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionID());
 405         commitCoordinator.handleCanCommit(canCommit.getTransactionID(), getSender(), this);
 406     }
 407
 408     protected void handleBatchedModificationsLocal(BatchedModifications batched, ActorRef sender) {
 409         try {
 410             commitCoordinator.handleBatchedModifications(batched, sender, this);
 411         } catch (Exception e) {
 412             LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(),
 413                     batched.getTransactionID(), e);
 414             sender.tell(new akka.actor.Status.Failure(e), getSelf());
 415         }
 416     }
 417
 418     private void handleBatchedModifications(BatchedModifications batched) {
 419         // This message is sent to prepare the modifications transaction directly on the Shard as an
 420         // optimization to avoid the extra overhead of a separate ShardTransaction actor. On the last
 421         // BatchedModifications message, the caller sets the ready flag in the message indicating
 422         // modifications are complete. The reply contains the cohort actor path (this actor) for the caller
 423         // to initiate the 3-phase commit. This also avoids the overhead of sending an additional
 424         // ReadyTransaction message.
 425
 426         // If we're not the leader then forward to the leader. This is a safety measure - we shouldn't
 427         // normally get here if we're not the leader as the front-end (TransactionProxy) should determine
 428         // the primary/leader shard. However with timing and caching on the front-end, there's a small
 429         // window where it could have a stale leader during leadership transitions.
 430         //
 431         boolean isLeaderActive = isLeaderActive();
 432         if (isLeader() && isLeaderActive) {
 433             handleBatchedModificationsLocal(batched, getSender());
 434         } else {
 435             ActorSelection leader = getLeader();
 436             if (!isLeaderActive || leader == null) {
 437                 messageRetrySupport.addMessageToRetry(batched, getSender(),
 438                         "Could not commit transaction " + batched.getTransactionID());
 439             } else {
 440                 // TODO: what if this is not the first batch and leadership changed in between batched messages?
 441                 // We could check if the commitCoordinator already has a cached entry and forward all the previous
 442                 // batched modifications.
 443                 LOG.debug("{}: Forwarding BatchedModifications to leader {}", persistenceId(), leader);
 444                 leader.forward(batched, getContext());
 445             }
 446         }
 447     }
 448
 449     private boolean failIfIsolatedLeader(ActorRef sender) {
 450         if(isIsolatedLeader()) {
 451             sender.tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format(
 452                     "Shard %s was the leader but has lost contact with all of its followers. Either all" +
 453                     " other follower nodes are down or this node is isolated by a network partition.",
 454                     persistenceId()))), getSelf());
 455             return true;
 456         }
 457
 458         return false;
 459     }
 460
 461     protected boolean isIsolatedLeader() {
 462         return getRaftState() == RaftState.IsolatedLeader;
 463     }
 464
 465     private void handleReadyLocalTransaction(final ReadyLocalTransaction message) {
 466         LOG.debug("{}: handleReadyLocalTransaction for {}", persistenceId(), message.getTransactionID());
 467
 468         boolean isLeaderActive = isLeaderActive();
 469         if (isLeader() && isLeaderActive) {
 470             try {
 471                 commitCoordinator.handleReadyLocalTransaction(message, getSender(), this);
 472             } catch (Exception e) {
 473                 LOG.error("{}: Error handling ReadyLocalTransaction for Tx {}", persistenceId(),
 474                         message.getTransactionID(), e);
 475                 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 476             }
 477         } else {
 478             ActorSelection leader = getLeader();
 479             if (!isLeaderActive || leader == null) {
 480                 messageRetrySupport.addMessageToRetry(message, getSender(),
 481                         "Could not commit transaction " + message.getTransactionID());
 482             } else {
 483                 LOG.debug("{}: Forwarding ReadyLocalTransaction to leader {}", persistenceId(), leader);
 484                 message.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
 485                 leader.forward(message, getContext());
 486             }
 487         }
 488     }
 489
 490     private void handleForwardedReadyTransaction(ForwardedReadyTransaction forwardedReady) {
 491         LOG.debug("{}: handleForwardedReadyTransaction for {}", persistenceId(), forwardedReady.getTransactionID());
 492
 493         boolean isLeaderActive = isLeaderActive();
 494         if (isLeader() && isLeaderActive) {
 495             commitCoordinator.handleForwardedReadyTransaction(forwardedReady, getSender(), this);
 496         } else {
 497             ActorSelection leader = getLeader();
 498             if (!isLeaderActive || leader == null) {
 499                 messageRetrySupport.addMessageToRetry(forwardedReady, getSender(),
 500                         "Could not commit transaction " + forwardedReady.getTransactionID());
 501             } else {
 502                 LOG.debug("{}: Forwarding ForwardedReadyTransaction to leader {}", persistenceId(), leader);
 503
 504                 ReadyLocalTransaction readyLocal = new ReadyLocalTransaction(forwardedReady.getTransactionID(),
 505                         forwardedReady.getTransaction().getSnapshot(), forwardedReady.isDoImmediateCommit());
 506                 readyLocal.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
 507                 leader.forward(readyLocal, getContext());
 508             }
 509         }
 510     }
 511
 512     private void handleAbortTransaction(final AbortTransaction abort) {
 513         doAbortTransaction(abort.getTransactionID(), getSender());
 514     }
 515
 516     void doAbortTransaction(final String transactionID, final ActorRef sender) {
 517         commitCoordinator.handleAbort(transactionID, sender, this);
 518     }
 519
 520     private void handleCreateTransaction(final Object message) {
 521         if (isLeader()) {
 522             createTransaction(CreateTransaction.fromSerializable(message));
 523         } else if (getLeader() != null) {
 524             getLeader().forward(message, getContext());
 525         } else {
 526             getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(
 527                     "Could not create a shard transaction", persistenceId())), getSelf());
 528         }
 529     }
 530
 531     private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
 532         store.closeTransactionChain(closeTransactionChain.getTransactionChainId());
 533     }
 534
 535     private ActorRef createTypedTransactionActor(int transactionType,
 536             ShardTransactionIdentifier transactionId, String transactionChainId,
 537             short clientVersion ) {
 538
 539         return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType),
 540                 transactionId, transactionChainId, clientVersion);
 541     }
 542
 543     private void createTransaction(CreateTransaction createTransaction) {
 544         try {
 545             if(TransactionType.fromInt(createTransaction.getTransactionType()) != TransactionType.READ_ONLY &&
 546                     failIfIsolatedLeader(getSender())) {
 547                 return;
 548             }
 549
 550             ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
 551                 createTransaction.getTransactionId(), createTransaction.getTransactionChainId(),
 552                 createTransaction.getVersion());
 553
 554             getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
 555                     createTransaction.getTransactionId(), createTransaction.getVersion()).toSerializable(), getSelf());
 556         } catch (Exception e) {
 557             getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 558         }
 559     }
 560
 561     private ActorRef createTransaction(int transactionType, String remoteTransactionId,
 562             String transactionChainId, short clientVersion) {
 563
 564
 565         ShardTransactionIdentifier transactionId = new ShardTransactionIdentifier(remoteTransactionId);
 566
 567         if(LOG.isDebugEnabled()) {
 568             LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
 569         }
 570
 571         ActorRef transactionActor = createTypedTransactionActor(transactionType, transactionId,
 572                 transactionChainId, clientVersion);
 573
 574         return transactionActor;
 575     }
 576
 577     private void commitWithNewTransaction(final Modification modification) {
 578         ReadWriteShardDataTreeTransaction tx = store.newReadWriteTransaction(modification.toString(), null);
 579         modification.apply(tx.getSnapshot());
 580         try {
 581             snapshotCohort.syncCommitTransaction(tx);
 582             shardMBean.incrementCommittedTransactionCount();
 583             shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
 584         } catch (Exception e) {
 585             shardMBean.incrementFailedTransactionsCount();
 586             LOG.error("{}: Failed to commit", persistenceId(), e);
 587         }
 588     }
 589
 590     private void updateSchemaContext(final UpdateSchemaContext message) {
 591         updateSchemaContext(message.getSchemaContext());
 592     }
 593
 594     @VisibleForTesting
 595     void updateSchemaContext(final SchemaContext schemaContext) {
 596         store.updateSchemaContext(schemaContext);
 597     }
 598
 599     private boolean isMetricsCaptureEnabled() {
 600         CommonConfig config = new CommonConfig(getContext().system().settings().config());
 601         return config.isMetricCaptureEnabled();
 602     }
 603
 604     @Override
 605     @VisibleForTesting
 606     public RaftActorSnapshotCohort getRaftActorSnapshotCohort() {
 607         return snapshotCohort;
 608     }
 609
 610     @Override
 611     @Nonnull
 612     protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
 613         return new ShardRecoveryCoordinator(store, store.getSchemaContext(),
 614                 restoreFromSnapshot != null ? restoreFromSnapshot.getSnapshot() : null, persistenceId(), LOG);
 615     }
 616
 617     @Override
 618     protected void onRecoveryComplete() {
 619         restoreFromSnapshot = null;
 620
 621         //notify shard manager
 622         getContext().parent().tell(new ActorInitialized(), getSelf());
 623
 624         // Being paranoid here - this method should only be called once but just in case...
 625         if(txCommitTimeoutCheckSchedule == null) {
 626             // Schedule a message to be periodically sent to check if the current in-progress
 627             // transaction should be expired and aborted.
 628             FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
 629             txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
 630                     period, period, getSelf(),
 631                     TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
 632         }
 633     }
 634
 635     @Override
 636     protected void applyState(final ActorRef clientActor, final String identifier, final Object data) {
 637         if (data instanceof DataTreeCandidatePayload) {
 638             if (clientActor == null) {
 639                 // No clientActor indicates a replica coming from the leader
 640                 try {
 641                     store.applyForeignCandidate(identifier, ((DataTreeCandidatePayload)data).getCandidate());
 642                 } catch (DataValidationFailedException | IOException e) {
 643                     LOG.error("{}: Error applying replica {}", persistenceId(), identifier, e);
 644                 }
 645             } else {
 646                 // Replication consensus reached, proceed to commit
 647                 finishCommit(clientActor, identifier);
 648             }
 649         } else if (data instanceof CompositeModificationPayload) {
 650             Object modification = ((CompositeModificationPayload) data).getModification();
 651
 652             applyModificationToState(clientActor, identifier, modification);
 653         } else if(data instanceof CompositeModificationByteStringPayload ){
 654             Object modification = ((CompositeModificationByteStringPayload) data).getModification();
 655
 656             applyModificationToState(clientActor, identifier, modification);
 657         } else {
 658             LOG.error("{}: Unknown state received {} Class loader = {} CompositeNodeMod.ClassLoader = {}",
 659                     persistenceId(), data, data.getClass().getClassLoader(),
 660                     CompositeModificationPayload.class.getClassLoader());
 661         }
 662     }
 663
 664     private void applyModificationToState(ActorRef clientActor, String identifier, Object modification) {
 665         if(modification == null) {
 666             LOG.error(
 667                     "{}: modification is null - this is very unexpected, clientActor = {}, identifier = {}",
 668                     persistenceId(), identifier, clientActor != null ? clientActor.path().toString() : null);
 669         } else if(clientActor == null) {
 670             // There's no clientActor to which to send a commit reply so we must be applying
 671             // replicated state from the leader.
 672             commitWithNewTransaction(MutableCompositeModification.fromSerializable(modification));
 673         } else {
 674             // This must be the OK to commit after replication consensus.
 675             finishCommit(clientActor, identifier);
 676         }
 677     }
 678
 679     @Override
 680     protected void onStateChanged() {
 681         boolean isLeader = isLeader();
 682         boolean hasLeader = hasLeader();
 683         changeSupport.onLeadershipChange(isLeader, hasLeader);
 684         treeChangeSupport.onLeadershipChange(isLeader, hasLeader);
 685
 686         // If this actor is no longer the leader close all the transaction chains
 687         if (!isLeader) {
 688             if(LOG.isDebugEnabled()) {
 689                 LOG.debug(
 690                     "{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader",
 691                     persistenceId(), getId());
 692             }
 693
 694             store.closeAllTransactionChains();
 695
 696             commitCoordinator.abortPendingTransactions(
 697                     "The transacton was aborted due to inflight leadership change.", this);
 698         }
 699
 700         if(hasLeader && !isIsolatedLeader()) {
 701             messageRetrySupport.retryMessages();
 702         }
 703     }
 704
 705     @Override
 706     protected void onLeaderChanged(String oldLeader, String newLeader) {
 707         shardMBean.incrementLeadershipChangeCount();
 708
 709         if(hasLeader() && !isIsolatedLeader()) {
 710             messageRetrySupport.retryMessages();
 711         }
 712     }
 713
 714     @Override
 715     protected void pauseLeader(Runnable operation) {
 716         LOG.debug("{}: In pauseLeader, operation: {}", persistenceId(), operation);
 717         commitCoordinator.setRunOnPendingTransactionsComplete(operation);
 718     }
 719
 720     @Override
 721     public String persistenceId() {
 722         return this.name;
 723     }
 724
 725     @VisibleForTesting
 726     ShardCommitCoordinator getCommitCoordinator() {
 727         return commitCoordinator;
 728     }
 729
 730     public DatastoreContext getDatastoreContext() {
 731         return datastoreContext;
 732     }
 733
 734     @VisibleForTesting
 735     public ShardDataTree getDataStore() {
 736         return store;
 737     }
 738
 739     @VisibleForTesting
 740     ShardStats getShardMBean() {
 741         return shardMBean;
 742     }
 743
 744     public static Builder builder() {
 745         return new Builder();
 746     }
 747
 748     public static abstract class AbstractBuilder<T extends AbstractBuilder<T, S>, S extends Shard> {
 749         private final Class<S> shardClass;
 750         private ShardIdentifier id;
 751         private Map<String, String> peerAddresses = Collections.emptyMap();
 752         private DatastoreContext datastoreContext;
 753         private SchemaContext schemaContext;
 754         private DatastoreSnapshot.ShardSnapshot restoreFromSnapshot;
 755         private volatile boolean sealed;
 756
 757         protected AbstractBuilder(Class<S> shardClass) {
 758             this.shardClass = shardClass;
 759         }
 760
 761         protected void checkSealed() {
 762             Preconditions.checkState(!sealed, "Builder isalready sealed - further modifications are not allowed");
 763         }
 764
 765         @SuppressWarnings("unchecked")
 766         private T self() {
 767             return (T) this;
 768         }
 769
 770         public T id(ShardIdentifier id) {
 771             checkSealed();
 772             this.id = id;
 773             return self();
 774         }
 775
 776         public T peerAddresses(Map<String, String> peerAddresses) {
 777             checkSealed();
 778             this.peerAddresses = peerAddresses;
 779             return self();
 780         }
 781
 782         public T datastoreContext(DatastoreContext datastoreContext) {
 783             checkSealed();
 784             this.datastoreContext = datastoreContext;
 785             return self();
 786         }
 787
 788         public T schemaContext(SchemaContext schemaContext) {
 789             checkSealed();
 790             this.schemaContext = schemaContext;
 791             return self();
 792         }
 793
 794         public T restoreFromSnapshot(DatastoreSnapshot.ShardSnapshot restoreFromSnapshot) {
 795             checkSealed();
 796             this.restoreFromSnapshot = restoreFromSnapshot;
 797             return self();
 798         }
 799
 800         public ShardIdentifier getId() {
 801             return id;
 802         }
 803
 804         public Map<String, String> getPeerAddresses() {
 805             return peerAddresses;
 806         }
 807
 808         public DatastoreContext getDatastoreContext() {
 809             return datastoreContext;
 810         }
 811
 812         public SchemaContext getSchemaContext() {
 813             return schemaContext;
 814         }
 815
 816         public DatastoreSnapshot.ShardSnapshot getRestoreFromSnapshot() {
 817             return restoreFromSnapshot;
 818         }
 819
 820         public TreeType getTreeType() {
 821             switch (datastoreContext.getLogicalStoreType()) {
 822             case CONFIGURATION:
 823                 return TreeType.CONFIGURATION;
 824             case OPERATIONAL:
 825                 return TreeType.OPERATIONAL;
 826             }
 827
 828             throw new IllegalStateException("Unhandled logical store type " + datastoreContext.getLogicalStoreType());
 829         }
 830
 831         protected void verify() {
 832             Preconditions.checkNotNull(id, "id should not be null");
 833             Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
 834             Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
 835             Preconditions.checkNotNull(schemaContext, "schemaContext should not be null");
 836         }
 837
 838         public Props props() {
 839             sealed = true;
 840             verify();
 841             return Props.create(shardClass, this);
 842         }
 843     }
 844
 845     public static class Builder extends AbstractBuilder<Builder, Shard> {
 846         private Builder() {
 847             super(Shard.class);
 848         }
 849     }
 850 }