git.opendaylight Code Review - controller.git/blob

   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.serialization.Serialization;
  16 import com.google.common.annotations.VisibleForTesting;
  17 import com.google.common.base.Optional;
  18 import com.google.common.base.Preconditions;
  19 import com.google.common.base.Ticker;
  20 import java.io.IOException;
  21 import java.util.Collection;
  22 import java.util.Collections;
  23 import java.util.Map;
  24 import java.util.concurrent.TimeUnit;
  25 import javax.annotation.Nonnull;
  26 import org.opendaylight.controller.cluster.access.concepts.TransactionIdentifier;
  27 import org.opendaylight.controller.cluster.common.actor.CommonConfig;
  28 import org.opendaylight.controller.cluster.common.actor.MessageTracker;
  29 import org.opendaylight.controller.cluster.common.actor.MessageTracker.Error;
  30 import org.opendaylight.controller.cluster.common.actor.MeteringBehavior;
  31 import org.opendaylight.controller.cluster.datastore.exceptions.NoShardLeaderException;
  32 import org.opendaylight.controller.cluster.datastore.identifiers.ShardIdentifier;
  33 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardMBeanFactory;
  34 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardStats;
  35 import org.opendaylight.controller.cluster.datastore.messages.AbortTransaction;
  36 import org.opendaylight.controller.cluster.datastore.messages.ActorInitialized;
  37 import org.opendaylight.controller.cluster.datastore.messages.BatchedModifications;
  38 import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransaction;
  39 import org.opendaylight.controller.cluster.datastore.messages.CloseTransactionChain;
  40 import org.opendaylight.controller.cluster.datastore.messages.CommitTransaction;
  41 import org.opendaylight.controller.cluster.datastore.messages.CreateTransaction;
  42 import org.opendaylight.controller.cluster.datastore.messages.CreateTransactionReply;
  43 import org.opendaylight.controller.cluster.datastore.messages.DatastoreSnapshot;
  44 import org.opendaylight.controller.cluster.datastore.messages.DatastoreSnapshot.ShardSnapshot;
  45 import org.opendaylight.controller.cluster.datastore.messages.ForwardedReadyTransaction;
  46 import org.opendaylight.controller.cluster.datastore.messages.GetShardDataTree;
  47 import org.opendaylight.controller.cluster.datastore.messages.PeerAddressResolved;
  48 import org.opendaylight.controller.cluster.datastore.messages.ReadyLocalTransaction;
  49 import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListener;
  50 import org.opendaylight.controller.cluster.datastore.messages.RegisterDataTreeChangeListener;
  51 import org.opendaylight.controller.cluster.datastore.messages.ShardLeaderStateChanged;
  52 import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext;
  53 import org.opendaylight.controller.cluster.datastore.utils.Dispatchers;
  54 import org.opendaylight.controller.cluster.notifications.LeaderStateChanged;
  55 import org.opendaylight.controller.cluster.notifications.RegisterRoleChangeListener;
  56 import org.opendaylight.controller.cluster.notifications.RoleChangeNotifier;
  57 import org.opendaylight.controller.cluster.raft.RaftActor;
  58 import org.opendaylight.controller.cluster.raft.RaftActorRecoveryCohort;
  59 import org.opendaylight.controller.cluster.raft.RaftActorSnapshotCohort;
  60 import org.opendaylight.controller.cluster.raft.RaftState;
  61 import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus;
  62 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  63 import org.opendaylight.controller.cluster.raft.messages.ServerRemoved;
  64 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload;
  65 import org.opendaylight.yangtools.concepts.Identifier;
  66 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException;
  67 import org.opendaylight.yangtools.yang.data.api.schema.tree.TipProducingDataTree;
  68 import org.opendaylight.yangtools.yang.data.api.schema.tree.TreeType;
  69 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
  70 import scala.concurrent.duration.Duration;
  71 import scala.concurrent.duration.FiniteDuration;
  72
  73 /**
  74  * A Shard represents a portion of the logical data tree <br/>
  75  * <p>
  76  * Our Shard uses InMemoryDataTree as it's internal representation and delegates all requests it
  77  * </p>
  78  */
  79 public class Shard extends RaftActor {
  80
  81     @VisibleForTesting
  82     static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = new Object() {
  83         @Override
  84         public String toString() {
  85             return "txCommitTimeoutCheck";
  86         }
  87     };
  88
  89     @VisibleForTesting
  90     static final Object GET_SHARD_MBEAN_MESSAGE = new Object() {
  91         @Override
  92         public String toString() {
  93             return "getShardMBeanMessage";
  94         }
  95     };
  96
  97     // FIXME: shard names should be encapsulated in their own class and this should be exposed as a constant.
  98     public static final String DEFAULT_NAME = "default";
  99
 100     // The state of this Shard
 101     private final ShardDataTree store;
 102
 103     /// The name of this shard
 104     private final String name;
 105
 106     private final ShardStats shardMBean;
 107
 108     private DatastoreContext datastoreContext;
 109
 110     private final ShardCommitCoordinator commitCoordinator;
 111
 112     private long transactionCommitTimeout;
 113
 114     private Cancellable txCommitTimeoutCheckSchedule;
 115
 116     private final Optional<ActorRef> roleChangeNotifier;
 117
 118     private final MessageTracker appendEntriesReplyTracker;
 119
 120     private final ShardTransactionActorFactory transactionActorFactory;
 121
 122     private final ShardSnapshotCohort snapshotCohort;
 123
 124     private final DataTreeChangeListenerSupport treeChangeSupport = new DataTreeChangeListenerSupport(this);
 125     private final DataChangeListenerSupport changeSupport = new DataChangeListenerSupport(this);
 126
 127
 128     private ShardSnapshot restoreFromSnapshot;
 129
 130     private final ShardTransactionMessageRetrySupport messageRetrySupport;
 131
 132     private final FrontendMetadata frontendMetadata = new FrontendMetadata();
 133
 134     protected Shard(final AbstractBuilder<?, ?> builder) {
 135         super(builder.getId().toString(), builder.getPeerAddresses(),
 136                 Optional.of(builder.getDatastoreContext().getShardRaftConfig()), DataStoreVersions.CURRENT_VERSION);
 137
 138         this.name = builder.getId().toString();
 139         this.datastoreContext = builder.getDatastoreContext();
 140         this.restoreFromSnapshot = builder.getRestoreFromSnapshot();
 141
 142         setPersistence(datastoreContext.isPersistent());
 143
 144         LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
 145
 146         ShardDataTreeChangeListenerPublisherActorProxy treeChangeListenerPublisher =
 147                 new ShardDataTreeChangeListenerPublisherActorProxy(getContext(), name + "-DTCL-publisher");
 148         ShardDataChangeListenerPublisherActorProxy dataChangeListenerPublisher =
 149                 new ShardDataChangeListenerPublisherActorProxy(getContext(), name + "-DCL-publisher");
 150         if(builder.getDataTree() != null) {
 151             store = new ShardDataTree(this, builder.getSchemaContext(), builder.getDataTree(),
 152                     treeChangeListenerPublisher, dataChangeListenerPublisher, name);
 153         } else {
 154             store = new ShardDataTree(this, builder.getSchemaContext(), builder.getTreeType(),
 155                     treeChangeListenerPublisher, dataChangeListenerPublisher, name);
 156         }
 157
 158         shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
 159                 datastoreContext.getDataStoreMXBeanType());
 160         shardMBean.setShard(this);
 161
 162         if (isMetricsCaptureEnabled()) {
 163             getContext().become(new MeteringBehavior(this));
 164         }
 165
 166         commitCoordinator = new ShardCommitCoordinator(store, LOG, this.name);
 167
 168         setTransactionCommitTimeout();
 169
 170         // create a notifier actor for each cluster member
 171         roleChangeNotifier = createRoleChangeNotifier(name.toString());
 172
 173         appendEntriesReplyTracker = new MessageTracker(AppendEntriesReply.class,
 174                 getRaftActorContext().getConfigParams().getIsolatedCheckIntervalInMillis());
 175
 176         transactionActorFactory = new ShardTransactionActorFactory(store, datastoreContext,
 177                 new Dispatchers(context().system().dispatchers()).getDispatcherPath(
 178                         Dispatchers.DispatcherType.Transaction), self(), getContext(), shardMBean);
 179
 180         snapshotCohort = ShardSnapshotCohort.create(getContext(), builder.getId().getMemberName(), store, LOG,
 181             this.name);
 182
 183         messageRetrySupport = new ShardTransactionMessageRetrySupport(this);
 184     }
 185
 186     private void setTransactionCommitTimeout() {
 187         transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
 188                 datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS) / 2;
 189     }
 190
 191     private Optional<ActorRef> createRoleChangeNotifier(final String shardId) {
 192         ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
 193             RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
 194         return Optional.of(shardRoleChangeNotifier);
 195     }
 196
 197     @Override
 198     public void postStop() {
 199         LOG.info("Stopping Shard {}", persistenceId());
 200
 201         super.postStop();
 202
 203         messageRetrySupport.close();
 204
 205         if (txCommitTimeoutCheckSchedule != null) {
 206             txCommitTimeoutCheckSchedule.cancel();
 207         }
 208
 209         commitCoordinator.abortPendingTransactions("Transaction aborted due to shutdown.", this);
 210
 211         shardMBean.unregisterMBean();
 212     }
 213
 214     @Override
 215     protected void handleRecover(final Object message) {
 216         LOG.debug("{}: onReceiveRecover: Received message {} from {}", persistenceId(), message.getClass(),
 217             getSender());
 218
 219         super.handleRecover(message);
 220         if (LOG.isTraceEnabled()) {
 221             appendEntriesReplyTracker.begin();
 222         }
 223     }
 224
 225     @Override
 226     protected void handleNonRaftCommand(final Object message) {
 227         try (final MessageTracker.Context context = appendEntriesReplyTracker.received(message)) {
 228             final Optional<Error> maybeError = context.error();
 229             if (maybeError.isPresent()) {
 230                 LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(),
 231                     maybeError.get());
 232             }
 233
 234             if (CreateTransaction.isSerializedType(message)) {
 235                 handleCreateTransaction(message);
 236             } else if (message instanceof BatchedModifications) {
 237                 handleBatchedModifications((BatchedModifications)message);
 238             } else if (message instanceof ForwardedReadyTransaction) {
 239                 handleForwardedReadyTransaction((ForwardedReadyTransaction) message);
 240             } else if (message instanceof ReadyLocalTransaction) {
 241                 handleReadyLocalTransaction((ReadyLocalTransaction)message);
 242             } else if (CanCommitTransaction.isSerializedType(message)) {
 243                 handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
 244             } else if (CommitTransaction.isSerializedType(message)) {
 245                 handleCommitTransaction(CommitTransaction.fromSerializable(message));
 246             } else if (AbortTransaction.isSerializedType(message)) {
 247                 handleAbortTransaction(AbortTransaction.fromSerializable(message));
 248             } else if (CloseTransactionChain.isSerializedType(message)) {
 249                 closeTransactionChain(CloseTransactionChain.fromSerializable(message));
 250             } else if (message instanceof RegisterChangeListener) {
 251                 changeSupport.onMessage((RegisterChangeListener) message, isLeader(), hasLeader());
 252             } else if (message instanceof RegisterDataTreeChangeListener) {
 253                 treeChangeSupport.onMessage((RegisterDataTreeChangeListener) message, isLeader(), hasLeader());
 254             } else if (message instanceof UpdateSchemaContext) {
 255                 updateSchemaContext((UpdateSchemaContext) message);
 256             } else if (message instanceof PeerAddressResolved) {
 257                 PeerAddressResolved resolved = (PeerAddressResolved) message;
 258                 setPeerAddress(resolved.getPeerId().toString(),
 259                         resolved.getPeerAddress());
 260             } else if (TX_COMMIT_TIMEOUT_CHECK_MESSAGE.equals(message)) {
 261                 store.checkForExpiredTransactions(transactionCommitTimeout);
 262                 commitCoordinator.checkForExpiredTransactions(transactionCommitTimeout, this);
 263             } else if (message instanceof DatastoreContext) {
 264                 onDatastoreContext((DatastoreContext)message);
 265             } else if (message instanceof RegisterRoleChangeListener){
 266                 roleChangeNotifier.get().forward(message, context());
 267             } else if (message instanceof FollowerInitialSyncUpStatus) {
 268                 shardMBean.setFollowerInitialSyncStatus(((FollowerInitialSyncUpStatus) message).isInitialSyncDone());
 269                 context().parent().tell(message, self());
 270             } else if (GET_SHARD_MBEAN_MESSAGE.equals(message)){
 271                 sender().tell(getShardMBean(), self());
 272             } else if (message instanceof GetShardDataTree) {
 273                 sender().tell(store.getDataTree(), self());
 274             } else if (message instanceof ServerRemoved){
 275                 context().parent().forward(message, context());
 276             } else if (ShardTransactionMessageRetrySupport.TIMER_MESSAGE_CLASS.isInstance(message)) {
 277                 messageRetrySupport.onTimerMessage(message);
 278             } else if (message instanceof DataTreeCohortActorRegistry.CohortRegistryCommand) {
 279                 store.processCohortRegistryCommand(getSender(),
 280                         (DataTreeCohortActorRegistry.CohortRegistryCommand) message);
 281             } else {
 282                 super.handleNonRaftCommand(message);
 283             }
 284         }
 285     }
 286
 287     private boolean hasLeader() {
 288         return getLeaderId() != null;
 289     }
 290
 291     public int getPendingTxCommitQueueSize() {
 292         return store.getQueueSize();
 293     }
 294
 295     public int getCohortCacheSize() {
 296         return commitCoordinator.getCohortCacheSize();
 297     }
 298
 299     @Override
 300     protected Optional<ActorRef> getRoleChangeNotifier() {
 301         return roleChangeNotifier;
 302     }
 303
 304     @Override
 305     protected LeaderStateChanged newLeaderStateChanged(final String memberId, final String leaderId, final short leaderPayloadVersion) {
 306         return isLeader() ? new ShardLeaderStateChanged(memberId, leaderId, store.getDataTree(), leaderPayloadVersion)
 307                 : new ShardLeaderStateChanged(memberId, leaderId, leaderPayloadVersion);
 308     }
 309
 310     protected void onDatastoreContext(final DatastoreContext context) {
 311         datastoreContext = context;
 312
 313         setTransactionCommitTimeout();
 314
 315         setPersistence(datastoreContext.isPersistent());
 316
 317         updateConfigParams(datastoreContext.getShardRaftConfig());
 318     }
 319
 320     boolean canSkipPayload() {
 321         // If we do not have any followers and we are not using persistence we can apply modification to the state
 322         // immediately
 323         return !hasFollowers() && !persistence().isRecoveryApplicable();
 324     }
 325
 326     // applyState() will be invoked once consensus is reached on the payload
 327     void persistPayload(final TransactionIdentifier transactionId, final Payload payload) {
 328         // We are faking the sender
 329         persistData(self(), transactionId, payload);
 330     }
 331
 332     private void handleCommitTransaction(final CommitTransaction commit) {
 333         if (isLeader()) {
 334             commitCoordinator.handleCommit(commit.getTransactionID(), getSender(), this);
 335         } else {
 336             ActorSelection leader = getLeader();
 337             if (leader == null) {
 338                 messageRetrySupport.addMessageToRetry(commit, getSender(),
 339                         "Could not commit transaction " + commit.getTransactionID());
 340             } else {
 341                 LOG.debug("{}: Forwarding CommitTransaction to leader {}", persistenceId(), leader);
 342                 leader.forward(commit, getContext());
 343             }
 344         }
 345     }
 346
 347     private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
 348         LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionID());
 349
 350         if (isLeader()) {
 351         commitCoordinator.handleCanCommit(canCommit.getTransactionID(), getSender(), this);
 352         } else {
 353             ActorSelection leader = getLeader();
 354             if (leader == null) {
 355                 messageRetrySupport.addMessageToRetry(canCommit, getSender(),
 356                         "Could not canCommit transaction " + canCommit.getTransactionID());
 357             } else {
 358                 LOG.debug("{}: Forwarding CanCommitTransaction to leader {}", persistenceId(), leader);
 359                 leader.forward(canCommit, getContext());
 360             }
 361         }
 362     }
 363
 364     protected void handleBatchedModificationsLocal(final BatchedModifications batched, final ActorRef sender) {
 365         try {
 366             commitCoordinator.handleBatchedModifications(batched, sender, this);
 367         } catch (Exception e) {
 368             LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(),
 369                     batched.getTransactionID(), e);
 370             sender.tell(new akka.actor.Status.Failure(e), getSelf());
 371         }
 372     }
 373
 374     private void handleBatchedModifications(final BatchedModifications batched) {
 375         // This message is sent to prepare the modifications transaction directly on the Shard as an
 376         // optimization to avoid the extra overhead of a separate ShardTransaction actor. On the last
 377         // BatchedModifications message, the caller sets the ready flag in the message indicating
 378         // modifications are complete. The reply contains the cohort actor path (this actor) for the caller
 379         // to initiate the 3-phase commit. This also avoids the overhead of sending an additional
 380         // ReadyTransaction message.
 381
 382         // If we're not the leader then forward to the leader. This is a safety measure - we shouldn't
 383         // normally get here if we're not the leader as the front-end (TransactionProxy) should determine
 384         // the primary/leader shard. However with timing and caching on the front-end, there's a small
 385         // window where it could have a stale leader during leadership transitions.
 386         //
 387         boolean isLeaderActive = isLeaderActive();
 388         if (isLeader() && isLeaderActive) {
 389             handleBatchedModificationsLocal(batched, getSender());
 390         } else {
 391             ActorSelection leader = getLeader();
 392             if (!isLeaderActive || leader == null) {
 393                 messageRetrySupport.addMessageToRetry(batched, getSender(),
 394                         "Could not commit transaction " + batched.getTransactionID());
 395             } else {
 396                 // If this is not the first batch and leadership changed in between batched messages,
 397                 // we need to reconstruct previous BatchedModifications from the transaction
 398                 // DataTreeModification, honoring the max batched modification count, and forward all the
 399                 // previous BatchedModifications to the new leader.
 400                 Collection<BatchedModifications> newModifications = commitCoordinator.createForwardedBatchedModifications(
 401                         batched, datastoreContext.getShardBatchedModificationCount());
 402
 403                 LOG.debug("{}: Forwarding {} BatchedModifications to leader {}", persistenceId(),
 404                         newModifications.size(), leader);
 405
 406                 for (BatchedModifications bm : newModifications) {
 407                     leader.forward(bm, getContext());
 408                 }
 409             }
 410         }
 411     }
 412
 413     private boolean failIfIsolatedLeader(final ActorRef sender) {
 414         if (isIsolatedLeader()) {
 415             sender.tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format(
 416                     "Shard %s was the leader but has lost contact with all of its followers. Either all" +
 417                     " other follower nodes are down or this node is isolated by a network partition.",
 418                     persistenceId()))), getSelf());
 419             return true;
 420         }
 421
 422         return false;
 423     }
 424
 425     protected boolean isIsolatedLeader() {
 426         return getRaftState() == RaftState.IsolatedLeader;
 427     }
 428
 429     private void handleReadyLocalTransaction(final ReadyLocalTransaction message) {
 430         LOG.debug("{}: handleReadyLocalTransaction for {}", persistenceId(), message.getTransactionID());
 431
 432         boolean isLeaderActive = isLeaderActive();
 433         if (isLeader() && isLeaderActive) {
 434             try {
 435                 commitCoordinator.handleReadyLocalTransaction(message, getSender(), this);
 436             } catch (Exception e) {
 437                 LOG.error("{}: Error handling ReadyLocalTransaction for Tx {}", persistenceId(),
 438                         message.getTransactionID(), e);
 439                 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 440             }
 441         } else {
 442             ActorSelection leader = getLeader();
 443             if (!isLeaderActive || leader == null) {
 444                 messageRetrySupport.addMessageToRetry(message, getSender(),
 445                         "Could not commit transaction " + message.getTransactionID());
 446             } else {
 447                 LOG.debug("{}: Forwarding ReadyLocalTransaction to leader {}", persistenceId(), leader);
 448                 message.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
 449                 leader.forward(message, getContext());
 450             }
 451         }
 452     }
 453
 454     private void handleForwardedReadyTransaction(final ForwardedReadyTransaction forwardedReady) {
 455         LOG.debug("{}: handleForwardedReadyTransaction for {}", persistenceId(), forwardedReady.getTransactionID());
 456
 457         boolean isLeaderActive = isLeaderActive();
 458         if (isLeader() && isLeaderActive) {
 459             commitCoordinator.handleForwardedReadyTransaction(forwardedReady, getSender(), this);
 460         } else {
 461             ActorSelection leader = getLeader();
 462             if (!isLeaderActive || leader == null) {
 463                 messageRetrySupport.addMessageToRetry(forwardedReady, getSender(),
 464                         "Could not commit transaction " + forwardedReady.getTransactionID());
 465             } else {
 466                 LOG.debug("{}: Forwarding ForwardedReadyTransaction to leader {}", persistenceId(), leader);
 467
 468                 ReadyLocalTransaction readyLocal = new ReadyLocalTransaction(forwardedReady.getTransactionID(),
 469                         forwardedReady.getTransaction().getSnapshot(), forwardedReady.isDoImmediateCommit());
 470                 readyLocal.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
 471                 leader.forward(readyLocal, getContext());
 472             }
 473         }
 474     }
 475
 476     private void handleAbortTransaction(final AbortTransaction abort) {
 477         doAbortTransaction(abort.getTransactionID(), getSender());
 478     }
 479
 480     void doAbortTransaction(final TransactionIdentifier transactionID, final ActorRef sender) {
 481         commitCoordinator.handleAbort(transactionID, sender, this);
 482     }
 483
 484     private void handleCreateTransaction(final Object message) {
 485         if (isLeader()) {
 486             createTransaction(CreateTransaction.fromSerializable(message));
 487         } else if (getLeader() != null) {
 488             getLeader().forward(message, getContext());
 489         } else {
 490             getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(
 491                     "Could not create a shard transaction", persistenceId())), getSelf());
 492         }
 493     }
 494
 495     private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
 496         store.closeTransactionChain(closeTransactionChain.getIdentifier());
 497     }
 498
 499     private void createTransaction(final CreateTransaction createTransaction) {
 500         try {
 501             if (TransactionType.fromInt(createTransaction.getTransactionType()) != TransactionType.READ_ONLY &&
 502                     failIfIsolatedLeader(getSender())) {
 503                 return;
 504             }
 505
 506             ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
 507                 createTransaction.getTransactionId());
 508
 509             getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
 510                     createTransaction.getTransactionId(), createTransaction.getVersion()).toSerializable(), getSelf());
 511         } catch (Exception e) {
 512             getSender().tell(new akka.actor.Status.Failure(e), getSelf());
 513         }
 514     }
 515
 516     private ActorRef createTransaction(final int transactionType, final TransactionIdentifier transactionId) {
 517         LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
 518         return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType),
 519             transactionId);
 520     }
 521
 522     private void updateSchemaContext(final UpdateSchemaContext message) {
 523         updateSchemaContext(message.getSchemaContext());
 524     }
 525
 526     @VisibleForTesting
 527     void updateSchemaContext(final SchemaContext schemaContext) {
 528         store.updateSchemaContext(schemaContext);
 529     }
 530
 531     private boolean isMetricsCaptureEnabled() {
 532         CommonConfig config = new CommonConfig(getContext().system().settings().config());
 533         return config.isMetricCaptureEnabled();
 534     }
 535
 536     @Override
 537     @VisibleForTesting
 538     public RaftActorSnapshotCohort getRaftActorSnapshotCohort() {
 539         return snapshotCohort;
 540     }
 541
 542     @Override
 543     @Nonnull
 544     protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
 545         return new ShardRecoveryCoordinator(store,
 546             restoreFromSnapshot != null ? restoreFromSnapshot.getSnapshot() : null, persistenceId(), LOG);
 547     }
 548
 549     @Override
 550     protected void onRecoveryComplete() {
 551         restoreFromSnapshot = null;
 552
 553         //notify shard manager
 554         getContext().parent().tell(new ActorInitialized(), getSelf());
 555
 556         // Being paranoid here - this method should only be called once but just in case...
 557         if (txCommitTimeoutCheckSchedule == null) {
 558             // Schedule a message to be periodically sent to check if the current in-progress
 559             // transaction should be expired and aborted.
 560             FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
 561             txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
 562                     period, period, getSelf(),
 563                     TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
 564         }
 565     }
 566
 567     @Override
 568     protected void applyState(final ActorRef clientActor, final Identifier identifier, final Object data) {
 569         if (data instanceof Payload) {
 570             try {
 571                 store.applyReplicatedPayload(identifier, (Payload)data);
 572             } catch (DataValidationFailedException | IOException e) {
 573                 LOG.error("{}: Error applying replica {}", persistenceId(), identifier, e);
 574             }
 575         } else {
 576             LOG.error("{}: Unknown state for {} received {}", persistenceId(), identifier, data);
 577         }
 578     }
 579
 580     @Override
 581     protected void onStateChanged() {
 582         boolean isLeader = isLeader();
 583         boolean hasLeader = hasLeader();
 584         changeSupport.onLeadershipChange(isLeader, hasLeader);
 585         treeChangeSupport.onLeadershipChange(isLeader, hasLeader);
 586
 587         // If this actor is no longer the leader close all the transaction chains
 588         if (!isLeader) {
 589             if (LOG.isDebugEnabled()) {
 590                 LOG.debug(
 591                     "{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader",
 592                     persistenceId(), getId());
 593             }
 594
 595             store.closeAllTransactionChains();
 596         }
 597
 598         if (hasLeader && !isIsolatedLeader()) {
 599             messageRetrySupport.retryMessages();
 600         }
 601     }
 602
 603     @Override
 604     protected void onLeaderChanged(final String oldLeader, final String newLeader) {
 605         shardMBean.incrementLeadershipChangeCount();
 606
 607         boolean hasLeader = hasLeader();
 608         if (hasLeader && !isLeader()) {
 609             // Another leader was elected. If we were the previous leader and had pending transactions, convert
 610             // them to transaction messages and send to the new leader.
 611             ActorSelection leader = getLeader();
 612             if (leader != null) {
 613                 Collection<?> messagesToForward = commitCoordinator.convertPendingTransactionsToMessages(
 614                             datastoreContext.getShardBatchedModificationCount());
 615
 616                 if (!messagesToForward.isEmpty()) {
 617                     LOG.debug("{}: Forwarding {} pending transaction messages to leader {}", persistenceId(),
 618                             messagesToForward.size(), leader);
 619
 620                     for (Object message : messagesToForward) {
 621                         leader.tell(message, self());
 622                     }
 623                 }
 624             } else {
 625                 commitCoordinator.abortPendingTransactions(
 626                         "The transacton was aborted due to inflight leadership change and the leader address isn't available.",
 627                         this);
 628             }
 629         }
 630
 631         if (hasLeader && !isIsolatedLeader()) {
 632             messageRetrySupport.retryMessages();
 633         }
 634     }
 635
 636     @Override
 637     protected void pauseLeader(final Runnable operation) {
 638         LOG.debug("{}: In pauseLeader, operation: {}", persistenceId(), operation);
 639         store.setRunOnPendingTransactionsComplete(operation);
 640     }
 641
 642     @Override
 643     public String persistenceId() {
 644         return this.name;
 645     }
 646
 647     @VisibleForTesting
 648     ShardCommitCoordinator getCommitCoordinator() {
 649         return commitCoordinator;
 650     }
 651
 652     public DatastoreContext getDatastoreContext() {
 653         return datastoreContext;
 654     }
 655
 656     @VisibleForTesting
 657     public ShardDataTree getDataStore() {
 658         return store;
 659     }
 660
 661     @VisibleForTesting
 662     ShardStats getShardMBean() {
 663         return shardMBean;
 664     }
 665
 666     public static Builder builder() {
 667         return new Builder();
 668     }
 669
 670     public static abstract class AbstractBuilder<T extends AbstractBuilder<T, S>, S extends Shard> {
 671         private final Class<S> shardClass;
 672         private ShardIdentifier id;
 673         private Map<String, String> peerAddresses = Collections.emptyMap();
 674         private DatastoreContext datastoreContext;
 675         private SchemaContext schemaContext;
 676         private DatastoreSnapshot.ShardSnapshot restoreFromSnapshot;
 677         private TipProducingDataTree dataTree;
 678         private volatile boolean sealed;
 679
 680         protected AbstractBuilder(final Class<S> shardClass) {
 681             this.shardClass = shardClass;
 682         }
 683
 684         protected void checkSealed() {
 685             Preconditions.checkState(!sealed, "Builder isalready sealed - further modifications are not allowed");
 686         }
 687
 688         @SuppressWarnings("unchecked")
 689         private T self() {
 690             return (T) this;
 691         }
 692
 693         public T id(final ShardIdentifier id) {
 694             checkSealed();
 695             this.id = id;
 696             return self();
 697         }
 698
 699         public T peerAddresses(final Map<String, String> peerAddresses) {
 700             checkSealed();
 701             this.peerAddresses = peerAddresses;
 702             return self();
 703         }
 704
 705         public T datastoreContext(final DatastoreContext datastoreContext) {
 706             checkSealed();
 707             this.datastoreContext = datastoreContext;
 708             return self();
 709         }
 710
 711         public T schemaContext(final SchemaContext schemaContext) {
 712             checkSealed();
 713             this.schemaContext = schemaContext;
 714             return self();
 715         }
 716
 717         public T restoreFromSnapshot(final DatastoreSnapshot.ShardSnapshot restoreFromSnapshot) {
 718             checkSealed();
 719             this.restoreFromSnapshot = restoreFromSnapshot;
 720             return self();
 721         }
 722
 723         public T dataTree(final TipProducingDataTree dataTree) {
 724             checkSealed();
 725             this.dataTree = dataTree;
 726             return self();
 727         }
 728
 729         public ShardIdentifier getId() {
 730             return id;
 731         }
 732
 733         public Map<String, String> getPeerAddresses() {
 734             return peerAddresses;
 735         }
 736
 737         public DatastoreContext getDatastoreContext() {
 738             return datastoreContext;
 739         }
 740
 741         public SchemaContext getSchemaContext() {
 742             return schemaContext;
 743         }
 744
 745         public DatastoreSnapshot.ShardSnapshot getRestoreFromSnapshot() {
 746             return restoreFromSnapshot;
 747         }
 748
 749         public TipProducingDataTree getDataTree() {
 750             return dataTree;
 751         }
 752
 753         public TreeType getTreeType() {
 754             switch (datastoreContext.getLogicalStoreType()) {
 755             case CONFIGURATION:
 756                 return TreeType.CONFIGURATION;
 757             case OPERATIONAL:
 758                 return TreeType.OPERATIONAL;
 759             }
 760
 761             throw new IllegalStateException("Unhandled logical store type " + datastoreContext.getLogicalStoreType());
 762         }
 763
 764         protected void verify() {
 765             Preconditions.checkNotNull(id, "id should not be null");
 766             Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
 767             Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
 768             Preconditions.checkNotNull(schemaContext, "schemaContext should not be null");
 769         }
 770
 771         public Props props() {
 772             sealed = true;
 773             verify();
 774             return Props.create(shardClass, this);
 775         }
 776     }
 777
 778     public static class Builder extends AbstractBuilder<Builder, Shard> {
 779         private Builder() {
 780             super(Shard.class);
 781         }
 782     }
 783
 784     Ticker ticker() {
 785         return Ticker.systemTicker();
 786     }
 787 }