opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/RaftActor.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.raft;
  10
  11 import akka.actor.ActorRef;
  12 import akka.actor.ActorSelection;
  13 import akka.japi.Procedure;
  14 import akka.persistence.RecoveryCompleted;
  15 import akka.persistence.SaveSnapshotFailure;
  16 import akka.persistence.SaveSnapshotSuccess;
  17 import akka.persistence.SnapshotOffer;
  18 import akka.persistence.SnapshotSelectionCriteria;
  19 import com.google.common.annotations.VisibleForTesting;
  20 import com.google.common.base.Objects;
  21 import com.google.common.base.Optional;
  22 import com.google.common.base.Stopwatch;
  23 import com.google.common.collect.ImmutableMap;
  24 import com.google.common.collect.Lists;
  25 import java.io.Serializable;
  26 import java.util.Collection;
  27 import java.util.List;
  28 import java.util.Map;
  29 import java.util.concurrent.TimeUnit;
  30 import org.apache.commons.lang3.time.DurationFormatUtils;
  31 import org.opendaylight.controller.cluster.DataPersistenceProvider;
  32 import org.opendaylight.controller.cluster.DelegatingPersistentDataProvider;
  33 import org.opendaylight.controller.cluster.NonPersistentDataProvider;
  34 import org.opendaylight.controller.cluster.PersistentDataProvider;
  35 import org.opendaylight.controller.cluster.common.actor.AbstractUntypedPersistentActor;
  36 import org.opendaylight.controller.cluster.notifications.LeaderStateChanged;
  37 import org.opendaylight.controller.cluster.notifications.RoleChanged;
  38 import org.opendaylight.controller.cluster.raft.base.messages.ApplyJournalEntries;
  39 import org.opendaylight.controller.cluster.raft.base.messages.ApplyLogEntries;
  40 import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot;
  41 import org.opendaylight.controller.cluster.raft.base.messages.ApplyState;
  42 import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshot;
  43 import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshotReply;
  44 import org.opendaylight.controller.cluster.raft.base.messages.Replicate;
  45 import org.opendaylight.controller.cluster.raft.behaviors.AbstractLeader;
  46 import org.opendaylight.controller.cluster.raft.behaviors.AbstractRaftActorBehavior;
  47 import org.opendaylight.controller.cluster.raft.behaviors.Follower;
  48 import org.opendaylight.controller.cluster.raft.behaviors.RaftActorBehavior;
  49 import org.opendaylight.controller.cluster.raft.client.messages.FindLeader;
  50 import org.opendaylight.controller.cluster.raft.client.messages.FindLeaderReply;
  51 import org.opendaylight.controller.cluster.raft.client.messages.FollowerInfo;
  52 import org.opendaylight.controller.cluster.raft.client.messages.GetOnDemandRaftState;
  53 import org.opendaylight.controller.cluster.raft.client.messages.OnDemandRaftState;
  54 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload;
  55 import org.slf4j.Logger;
  56 import org.slf4j.LoggerFactory;
  57
  58 /**
  59  * RaftActor encapsulates a state machine that needs to be kept synchronized
  60  * in a cluster. It implements the RAFT algorithm as described in the paper
  61  * <a href='https://ramcloud.stanford.edu/wiki/download/attachments/11370504/raft.pdf'>
  62  * In Search of an Understandable Consensus Algorithm</a>
  63  * <p/>
  64  * RaftActor has 3 states and each state has a certain behavior associated
  65  * with it. A Raft actor can behave as,
  66  * <ul>
  67  * <li> A Leader </li>
  68  * <li> A Follower (or) </li>
  69  * <li> A Candidate </li>
  70  * </ul>
  71  * <p/>
  72  * <p/>
  73  * A RaftActor MUST be a Leader in order to accept requests from clients to
  74  * change the state of it's encapsulated state machine. Once a RaftActor becomes
  75  * a Leader it is also responsible for ensuring that all followers ultimately
  76  * have the same log and therefore the same state machine as itself.
  77  * <p/>
  78  * <p/>
  79  * The current behavior of a RaftActor determines how election for leadership
  80  * is initiated and how peer RaftActors react to request for votes.
  81  * <p/>
  82  * <p/>
  83  * Each RaftActor also needs to know the current election term. It uses this
  84  * information for a couple of things. One is to simply figure out who it
  85  * voted for in the last election. Another is to figure out if the message
  86  * it received to update it's state is stale.
  87  * <p/>
  88  * <p/>
  89  * The RaftActor uses akka-persistence to store it's replicated log.
  90  * Furthermore through it's behaviors a Raft Actor determines
  91  * <p/>
  92  * <ul>
  93  * <li> when a log entry should be persisted </li>
  94  * <li> when a log entry should be applied to the state machine (and) </li>
  95  * <li> when a snapshot should be saved </li>
  96  * </ul>
  97  */
  98 public abstract class RaftActor extends AbstractUntypedPersistentActor {
  99
 100     private static final long APPLY_STATE_DELAY_THRESHOLD_IN_NANOS = TimeUnit.MILLISECONDS.toNanos(50L); // 50 millis
 101
 102     private static final String COMMIT_SNAPSHOT = "commit_snapshot";
 103
 104     protected final Logger LOG = LoggerFactory.getLogger(getClass());
 105
 106     /**
 107      * The current state determines the current behavior of a RaftActor
 108      * A Raft Actor always starts off in the Follower State
 109      */
 110     private RaftActorBehavior currentBehavior;
 111
 112     /**
 113      * This context should NOT be passed directly to any other actor it is
 114      * only to be consumed by the RaftActorBehaviors
 115      */
 116     private final RaftActorContextImpl context;
 117
 118     private final DelegatingPersistentDataProvider delegatingPersistenceProvider = new DelegatingPersistentDataProvider(null);
 119
 120     private final Procedure<Void> createSnapshotProcedure = new CreateSnapshotProcedure();
 121
 122     /**
 123      * The in-memory journal
 124      */
 125     private ReplicatedLogImpl replicatedLog = new ReplicatedLogImpl();
 126
 127     private Stopwatch recoveryTimer;
 128
 129     private int currentRecoveryBatchCount;
 130
 131     private final BehaviorStateHolder reusableBehaviorStateHolder = new BehaviorStateHolder();
 132
 133     public RaftActor(String id, Map<String, String> peerAddresses) {
 134         this(id, peerAddresses, Optional.<ConfigParams>absent());
 135     }
 136
 137     public RaftActor(String id, Map<String, String> peerAddresses,
 138          Optional<ConfigParams> configParams) {
 139
 140         context = new RaftActorContextImpl(this.getSelf(),
 141             this.getContext(), id, new ElectionTermImpl(delegatingPersistenceProvider, id, LOG),
 142             -1, -1, replicatedLog, peerAddresses,
 143             (configParams.isPresent() ? configParams.get(): new DefaultConfigParamsImpl()),
 144             LOG);
 145     }
 146
 147     private void initRecoveryTimer() {
 148         if(recoveryTimer == null) {
 149             recoveryTimer = Stopwatch.createStarted();
 150         }
 151     }
 152
 153     @Override
 154     public void preStart() throws Exception {
 155         LOG.info("Starting recovery for {} with journal batch size {}", persistenceId(),
 156                 context.getConfigParams().getJournalRecoveryLogBatchSize());
 157
 158         super.preStart();
 159     }
 160
 161     @Override
 162     public void postStop() {
 163         if(currentBehavior != null) {
 164             try {
 165                 currentBehavior.close();
 166             } catch (Exception e) {
 167                 LOG.debug("{}: Error closing behavior {}", persistenceId(), currentBehavior.state());
 168             }
 169         }
 170
 171         super.postStop();
 172     }
 173
 174     @Override
 175     public void handleRecover(Object message) {
 176         if(persistence().isRecoveryApplicable()) {
 177             if (message instanceof SnapshotOffer) {
 178                 onRecoveredSnapshot((SnapshotOffer) message);
 179             } else if (message instanceof ReplicatedLogEntry) {
 180                 onRecoveredJournalLogEntry((ReplicatedLogEntry) message);
 181             } else if (message instanceof ApplyLogEntries) {
 182                 // Handle this message for backwards compatibility with pre-Lithium versions.
 183                 onRecoveredApplyLogEntries(((ApplyLogEntries) message).getToIndex());
 184             } else if (message instanceof ApplyJournalEntries) {
 185                 onRecoveredApplyLogEntries(((ApplyJournalEntries) message).getToIndex());
 186             } else if (message instanceof DeleteEntries) {
 187                 replicatedLog.removeFrom(((DeleteEntries) message).getFromIndex());
 188             } else if (message instanceof UpdateElectionTerm) {
 189                 context.getTermInformation().update(((UpdateElectionTerm) message).getCurrentTerm(),
 190                         ((UpdateElectionTerm) message).getVotedFor());
 191             } else if (message instanceof RecoveryCompleted) {
 192                 onRecoveryCompletedMessage();
 193             }
 194         } else {
 195             if (message instanceof RecoveryCompleted) {
 196                 // Delete all the messages from the akka journal so that we do not end up with consistency issues
 197                 // Note I am not using the dataPersistenceProvider and directly using the akka api here
 198                 deleteMessages(lastSequenceNr());
 199
 200                 // Delete all the akka snapshots as they will not be needed
 201                 deleteSnapshots(new SnapshotSelectionCriteria(scala.Long.MaxValue(), scala.Long.MaxValue()));
 202
 203                 onRecoveryComplete();
 204
 205                 initializeBehavior();
 206             }
 207         }
 208     }
 209
 210     private void onRecoveredSnapshot(SnapshotOffer offer) {
 211         if(LOG.isDebugEnabled()) {
 212             LOG.debug("{}: SnapshotOffer called..", persistenceId());
 213         }
 214
 215         initRecoveryTimer();
 216
 217         Snapshot snapshot = (Snapshot) offer.snapshot();
 218
 219         // Create a replicated log with the snapshot information
 220         // The replicated log can be used later on to retrieve this snapshot
 221         // when we need to install it on a peer
 222         replicatedLog = new ReplicatedLogImpl(snapshot);
 223
 224         context.setReplicatedLog(replicatedLog);
 225         context.setLastApplied(snapshot.getLastAppliedIndex());
 226         context.setCommitIndex(snapshot.getLastAppliedIndex());
 227
 228         Stopwatch timer = Stopwatch.createStarted();
 229
 230         // Apply the snapshot to the actors state
 231         applyRecoverySnapshot(snapshot.getState());
 232
 233         timer.stop();
 234         LOG.info("Recovery snapshot applied for {} in {}: snapshotIndex={}, snapshotTerm={}, journal-size=" +
 235                 replicatedLog.size(), persistenceId(), timer.toString(),
 236                 replicatedLog.getSnapshotIndex(), replicatedLog.getSnapshotTerm());
 237     }
 238
 239     private void onRecoveredJournalLogEntry(ReplicatedLogEntry logEntry) {
 240         if(LOG.isDebugEnabled()) {
 241             LOG.debug("{}: Received ReplicatedLogEntry for recovery: {}", persistenceId(), logEntry.getIndex());
 242         }
 243
 244         replicatedLog.append(logEntry);
 245     }
 246
 247     private void onRecoveredApplyLogEntries(long toIndex) {
 248         if(LOG.isDebugEnabled()) {
 249             LOG.debug("{}: Received ApplyLogEntries for recovery, applying to state: {} to {}",
 250                     persistenceId(), context.getLastApplied() + 1, toIndex);
 251         }
 252
 253         for (long i = context.getLastApplied() + 1; i <= toIndex; i++) {
 254             batchRecoveredLogEntry(replicatedLog.get(i));
 255         }
 256
 257         context.setLastApplied(toIndex);
 258         context.setCommitIndex(toIndex);
 259     }
 260
 261     private void batchRecoveredLogEntry(ReplicatedLogEntry logEntry) {
 262         initRecoveryTimer();
 263
 264         int batchSize = context.getConfigParams().getJournalRecoveryLogBatchSize();
 265         if(currentRecoveryBatchCount == 0) {
 266             startLogRecoveryBatch(batchSize);
 267         }
 268
 269         appendRecoveredLogEntry(logEntry.getData());
 270
 271         if(++currentRecoveryBatchCount >= batchSize) {
 272             endCurrentLogRecoveryBatch();
 273         }
 274     }
 275
 276     private void endCurrentLogRecoveryBatch() {
 277         applyCurrentLogRecoveryBatch();
 278         currentRecoveryBatchCount = 0;
 279     }
 280
 281     private void onRecoveryCompletedMessage() {
 282         if(currentRecoveryBatchCount > 0) {
 283             endCurrentLogRecoveryBatch();
 284         }
 285
 286         onRecoveryComplete();
 287
 288         String recoveryTime = "";
 289         if(recoveryTimer != null) {
 290             recoveryTimer.stop();
 291             recoveryTime = " in " + recoveryTimer.toString();
 292             recoveryTimer = null;
 293         }
 294
 295         LOG.info(
 296             "Recovery completed" + recoveryTime + " - Switching actor to Follower - " +
 297                 "Persistence Id =  " + persistenceId() +
 298                 " Last index in log={}, snapshotIndex={}, snapshotTerm={}, " +
 299                 "journal-size={}",
 300             replicatedLog.lastIndex(), replicatedLog.getSnapshotIndex(),
 301             replicatedLog.getSnapshotTerm(), replicatedLog.size());
 302
 303         initializeBehavior();
 304     }
 305
 306     protected void initializeBehavior(){
 307         changeCurrentBehavior(new Follower(context));
 308     }
 309
 310     protected void changeCurrentBehavior(RaftActorBehavior newBehavior){
 311         reusableBehaviorStateHolder.init(currentBehavior);
 312         currentBehavior = newBehavior;
 313         handleBehaviorChange(reusableBehaviorStateHolder, currentBehavior);
 314     }
 315
 316     @Override public void handleCommand(Object message) {
 317         if (message instanceof ApplyState){
 318             ApplyState applyState = (ApplyState) message;
 319
 320             long elapsedTime = (System.nanoTime() - applyState.getStartTime());
 321             if(elapsedTime >= APPLY_STATE_DELAY_THRESHOLD_IN_NANOS){
 322                 LOG.warn("ApplyState took more time than expected. Elapsed Time = {} ms ApplyState = {}",
 323                         TimeUnit.NANOSECONDS.toMillis(elapsedTime), applyState);
 324             }
 325
 326             if(LOG.isDebugEnabled()) {
 327                 LOG.debug("{}: Applying state for log index {} data {}",
 328                     persistenceId(), applyState.getReplicatedLogEntry().getIndex(),
 329                     applyState.getReplicatedLogEntry().getData());
 330             }
 331
 332             applyState(applyState.getClientActor(), applyState.getIdentifier(),
 333                 applyState.getReplicatedLogEntry().getData());
 334
 335         } else if (message instanceof ApplyJournalEntries){
 336             ApplyJournalEntries applyEntries = (ApplyJournalEntries) message;
 337             if(LOG.isDebugEnabled()) {
 338                 LOG.debug("{}: Persisting ApplyLogEntries with index={}", persistenceId(), applyEntries.getToIndex());
 339             }
 340
 341             persistence().persist(applyEntries, NoopProcedure.instance());
 342
 343         } else if(message instanceof ApplySnapshot ) {
 344             Snapshot snapshot = ((ApplySnapshot) message).getSnapshot();
 345
 346             if(LOG.isDebugEnabled()) {
 347                 LOG.debug("{}: ApplySnapshot called on Follower Actor " +
 348                         "snapshotIndex:{}, snapshotTerm:{}", persistenceId(), snapshot.getLastAppliedIndex(),
 349                     snapshot.getLastAppliedTerm()
 350                 );
 351             }
 352
 353             applySnapshot(snapshot.getState());
 354
 355             //clears the followers log, sets the snapshot index to ensure adjusted-index works
 356             replicatedLog = new ReplicatedLogImpl(snapshot);
 357             context.setReplicatedLog(replicatedLog);
 358             context.setLastApplied(snapshot.getLastAppliedIndex());
 359
 360         } else if (message instanceof FindLeader) {
 361             getSender().tell(
 362                 new FindLeaderReply(getLeaderAddress()),
 363                 getSelf()
 364             );
 365
 366         } else if (message instanceof SaveSnapshotSuccess) {
 367             SaveSnapshotSuccess success = (SaveSnapshotSuccess) message;
 368             LOG.info("{}: SaveSnapshotSuccess received for snapshot", persistenceId());
 369
 370             long sequenceNumber = success.metadata().sequenceNr();
 371
 372             commitSnapshot(sequenceNumber);
 373
 374         } else if (message instanceof SaveSnapshotFailure) {
 375             SaveSnapshotFailure saveSnapshotFailure = (SaveSnapshotFailure) message;
 376
 377             LOG.error("{}: SaveSnapshotFailure received for snapshot Cause:",
 378                     persistenceId(), saveSnapshotFailure.cause());
 379
 380             context.getSnapshotManager().rollback();
 381
 382         } else if (message instanceof CaptureSnapshot) {
 383             LOG.debug("{}: CaptureSnapshot received by actor: {}", persistenceId(), message);
 384
 385             context.getSnapshotManager().create(createSnapshotProcedure);
 386
 387         } else if (message instanceof CaptureSnapshotReply) {
 388             handleCaptureSnapshotReply(((CaptureSnapshotReply) message).getSnapshot());
 389         } else if(message instanceof GetOnDemandRaftState) {
 390             onGetOnDemandRaftStats();
 391         } else if (message.equals(COMMIT_SNAPSHOT)) {
 392             commitSnapshot(-1);
 393         } else {
 394             reusableBehaviorStateHolder.init(currentBehavior);
 395
 396             currentBehavior = currentBehavior.handleMessage(getSender(), message);
 397
 398             handleBehaviorChange(reusableBehaviorStateHolder, currentBehavior);
 399         }
 400     }
 401
 402     private void onGetOnDemandRaftStats() {
 403         // Debugging message to retrieve raft stats.
 404
 405         OnDemandRaftState.Builder builder = OnDemandRaftState.builder()
 406                 .commitIndex(context.getCommitIndex())
 407                 .currentTerm(context.getTermInformation().getCurrentTerm())
 408                 .inMemoryJournalDataSize(replicatedLog.dataSize())
 409                 .inMemoryJournalLogSize(replicatedLog.size())
 410                 .isSnapshotCaptureInitiated(context.getSnapshotManager().isCapturing())
 411                 .lastApplied(context.getLastApplied())
 412                 .lastIndex(replicatedLog.lastIndex())
 413                 .lastTerm(replicatedLog.lastTerm())
 414                 .leader(getLeaderId())
 415                 .raftState(currentBehavior.state().toString())
 416                 .replicatedToAllIndex(currentBehavior.getReplicatedToAllIndex())
 417                 .snapshotIndex(replicatedLog.getSnapshotIndex())
 418                 .snapshotTerm(replicatedLog.getSnapshotTerm())
 419                 .votedFor(context.getTermInformation().getVotedFor())
 420                 .peerAddresses(ImmutableMap.copyOf(context.getPeerAddresses()));
 421
 422         ReplicatedLogEntry lastLogEntry = getLastLogEntry();
 423         if (lastLogEntry != null) {
 424             builder.lastLogIndex(lastLogEntry.getIndex());
 425             builder.lastLogTerm(lastLogEntry.getTerm());
 426         }
 427
 428         if(currentBehavior instanceof AbstractLeader) {
 429             AbstractLeader leader = (AbstractLeader)currentBehavior;
 430             Collection<String> followerIds = leader.getFollowerIds();
 431             List<FollowerInfo> followerInfoList = Lists.newArrayListWithCapacity(followerIds.size());
 432             for(String id: followerIds) {
 433                 final FollowerLogInformation info = leader.getFollower(id);
 434                 followerInfoList.add(new FollowerInfo(id, info.getNextIndex(), info.getMatchIndex(),
 435                         info.isFollowerActive(), DurationFormatUtils.formatDurationHMS(info.timeSinceLastActivity())));
 436             }
 437
 438             builder.followerInfoList(followerInfoList);
 439         }
 440
 441         sender().tell(builder.build(), self());
 442
 443     }
 444
 445     private void handleBehaviorChange(BehaviorStateHolder oldBehaviorState, RaftActorBehavior currentBehavior) {
 446         RaftActorBehavior oldBehavior = oldBehaviorState.getBehavior();
 447
 448         if (oldBehavior != currentBehavior){
 449             onStateChanged();
 450         }
 451
 452         String oldBehaviorLeaderId = oldBehavior == null ? null : oldBehaviorState.getLeaderId();
 453         String oldBehaviorStateName = oldBehavior == null ? null : oldBehavior.state().name();
 454
 455         // it can happen that the state has not changed but the leader has changed.
 456         Optional<ActorRef> roleChangeNotifier = getRoleChangeNotifier();
 457         if(!Objects.equal(oldBehaviorLeaderId, currentBehavior.getLeaderId())) {
 458             if(roleChangeNotifier.isPresent()) {
 459                 roleChangeNotifier.get().tell(new LeaderStateChanged(getId(), currentBehavior.getLeaderId()), getSelf());
 460             }
 461
 462             onLeaderChanged(oldBehaviorLeaderId, currentBehavior.getLeaderId());
 463         }
 464
 465         if (roleChangeNotifier.isPresent() &&
 466                 (oldBehavior == null || (oldBehavior.state() != currentBehavior.state()))) {
 467             roleChangeNotifier.get().tell(new RoleChanged(getId(), oldBehaviorStateName ,
 468                     currentBehavior.state().name()), getSelf());
 469         }
 470     }
 471
 472     /**
 473      * When a derived RaftActor needs to persist something it must call
 474      * persistData.
 475      *
 476      * @param clientActor
 477      * @param identifier
 478      * @param data
 479      */
 480     protected void persistData(final ActorRef clientActor, final String identifier,
 481         final Payload data) {
 482
 483         ReplicatedLogEntry replicatedLogEntry = new ReplicatedLogImplEntry(
 484             context.getReplicatedLog().lastIndex() + 1,
 485             context.getTermInformation().getCurrentTerm(), data);
 486
 487         if(LOG.isDebugEnabled()) {
 488             LOG.debug("{}: Persist data {}", persistenceId(), replicatedLogEntry);
 489         }
 490
 491         final RaftActorContext raftContext = getRaftActorContext();
 492
 493         replicatedLog
 494                 .appendAndPersist(replicatedLogEntry, new Procedure<ReplicatedLogEntry>() {
 495                     @Override
 496                     public void apply(ReplicatedLogEntry replicatedLogEntry) throws Exception {
 497                         if(!hasFollowers()){
 498                             // Increment the Commit Index and the Last Applied values
 499                             raftContext.setCommitIndex(replicatedLogEntry.getIndex());
 500                             raftContext.setLastApplied(replicatedLogEntry.getIndex());
 501
 502                             // Apply the state immediately
 503                             applyState(clientActor, identifier, data);
 504
 505                             // Send a ApplyJournalEntries message so that we write the fact that we applied
 506                             // the state to durable storage
 507                             self().tell(new ApplyJournalEntries(replicatedLogEntry.getIndex()), self());
 508
 509                             context.getSnapshotManager().trimLog(context.getLastApplied(), currentBehavior);
 510
 511                         } else if (clientActor != null) {
 512                             // Send message for replication
 513                             currentBehavior.handleMessage(getSelf(),
 514                                     new Replicate(clientActor, identifier,
 515                                             replicatedLogEntry)
 516                             );
 517                         }
 518
 519                     }
 520                 });    }
 521
 522     protected String getId() {
 523         return context.getId();
 524     }
 525
 526     /**
 527      * Derived actors can call the isLeader method to check if the current
 528      * RaftActor is the Leader or not
 529      *
 530      * @return true it this RaftActor is a Leader false otherwise
 531      */
 532     protected boolean isLeader() {
 533         return context.getId().equals(currentBehavior.getLeaderId());
 534     }
 535
 536     /**
 537      * Derived actor can call getLeader if they need a reference to the Leader.
 538      * This would be useful for example in forwarding a request to an actor
 539      * which is the leader
 540      *
 541      * @return A reference to the leader if known, null otherwise
 542      */
 543     protected ActorSelection getLeader(){
 544         String leaderAddress = getLeaderAddress();
 545
 546         if(leaderAddress == null){
 547             return null;
 548         }
 549
 550         return context.actorSelection(leaderAddress);
 551     }
 552
 553     /**
 554      *
 555      * @return the current leader's id
 556      */
 557     protected String getLeaderId(){
 558         return currentBehavior.getLeaderId();
 559     }
 560
 561     protected RaftState getRaftState() {
 562         return currentBehavior.state();
 563     }
 564
 565     protected ReplicatedLogEntry getLastLogEntry() {
 566         return replicatedLog.last();
 567     }
 568
 569     protected Long getCurrentTerm(){
 570         return context.getTermInformation().getCurrentTerm();
 571     }
 572
 573     protected Long getCommitIndex(){
 574         return context.getCommitIndex();
 575     }
 576
 577     protected Long getLastApplied(){
 578         return context.getLastApplied();
 579     }
 580
 581     protected RaftActorContext getRaftActorContext() {
 582         return context;
 583     }
 584
 585     protected void updateConfigParams(ConfigParams configParams) {
 586         context.setConfigParams(configParams);
 587     }
 588
 589     public final DataPersistenceProvider persistence() {
 590         return delegatingPersistenceProvider.getDelegate();
 591     }
 592
 593     public void setPersistence(DataPersistenceProvider provider) {
 594         delegatingPersistenceProvider.setDelegate(provider);
 595     }
 596
 597     protected void setPersistence(boolean persistent) {
 598         if(persistent) {
 599             setPersistence(new PersistentDataProvider(this));
 600         } else {
 601             setPersistence(new NonPersistentDataProvider() {
 602                 /**
 603                  * The way snapshotting works is,
 604                  * <ol>
 605                  * <li> RaftActor calls createSnapshot on the Shard
 606                  * <li> Shard sends a CaptureSnapshotReply and RaftActor then calls saveSnapshot
 607                  * <li> When saveSnapshot is invoked on the akka-persistence API it uses the SnapshotStore to save
 608                  * the snapshot. The SnapshotStore sends SaveSnapshotSuccess or SaveSnapshotFailure. When the
 609                  * RaftActor gets SaveSnapshot success it commits the snapshot to the in-memory journal. This
 610                  * commitSnapshot is mimicking what is done in SaveSnapshotSuccess.
 611                  * </ol>
 612                  */
 613                 @Override
 614                 public void saveSnapshot(Object o) {
 615                     // Make saving Snapshot successful
 616                     // Committing the snapshot here would end up calling commit in the creating state which would
 617                     // be a state violation. That's why now we send a message to commit the snapshot.
 618                     self().tell(COMMIT_SNAPSHOT, self());
 619                 }
 620             });
 621         }
 622     }
 623
 624     /**
 625      * setPeerAddress sets the address of a known peer at a later time.
 626      * <p>
 627      * This is to account for situations where a we know that a peer
 628      * exists but we do not know an address up-front. This may also be used in
 629      * situations where a known peer starts off in a different location and we
 630      * need to change it's address
 631      * <p>
 632      * Note that if the peerId does not match the list of peers passed to
 633      * this actor during construction an IllegalStateException will be thrown.
 634      *
 635      * @param peerId
 636      * @param peerAddress
 637      */
 638     protected void setPeerAddress(String peerId, String peerAddress){
 639         context.setPeerAddress(peerId, peerAddress);
 640     }
 641
 642     protected void commitSnapshot(long sequenceNumber) {
 643         context.getSnapshotManager().commit(persistence(), sequenceNumber);
 644     }
 645
 646     /**
 647      * The applyState method will be called by the RaftActor when some data
 648      * needs to be applied to the actor's state
 649      *
 650      * @param clientActor A reference to the client who sent this message. This
 651      *                    is the same reference that was passed to persistData
 652      *                    by the derived actor. clientActor may be null when
 653      *                    the RaftActor is behaving as a follower or during
 654      *                    recovery.
 655      * @param identifier  The identifier of the persisted data. This is also
 656      *                    the same identifier that was passed to persistData by
 657      *                    the derived actor. identifier may be null when
 658      *                    the RaftActor is behaving as a follower or during
 659      *                    recovery
 660      * @param data        A piece of data that was persisted by the persistData call.
 661      *                    This should NEVER be null.
 662      */
 663     protected abstract void applyState(ActorRef clientActor, String identifier,
 664         Object data);
 665
 666     /**
 667      * This method is called during recovery at the start of a batch of state entries. Derived
 668      * classes should perform any initialization needed to start a batch.
 669      */
 670     protected abstract void startLogRecoveryBatch(int maxBatchSize);
 671
 672     /**
 673      * This method is called during recovery to append state data to the current batch. This method
 674      * is called 1 or more times after {@link #startLogRecoveryBatch}.
 675      *
 676      * @param data the state data
 677      */
 678     protected abstract void appendRecoveredLogEntry(Payload data);
 679
 680     /**
 681      * This method is called during recovery to reconstruct the state of the actor.
 682      *
 683      * @param snapshotBytes A snapshot of the state of the actor
 684      */
 685     protected abstract void applyRecoverySnapshot(byte[] snapshotBytes);
 686
 687     /**
 688      * This method is called during recovery at the end of a batch to apply the current batched
 689      * log entries. This method is called after {@link #appendRecoveredLogEntry}.
 690      */
 691     protected abstract void applyCurrentLogRecoveryBatch();
 692
 693     /**
 694      * This method is called when recovery is complete.
 695      */
 696     protected abstract void onRecoveryComplete();
 697
 698     /**
 699      * This method will be called by the RaftActor when a snapshot needs to be
 700      * created. The derived actor should respond with its current state.
 701      * <p/>
 702      * During recovery the state that is returned by the derived actor will
 703      * be passed back to it by calling the applySnapshot  method
 704      *
 705      * @return The current state of the actor
 706      */
 707     protected abstract void createSnapshot();
 708
 709     /**
 710      * This method can be called at any other point during normal
 711      * operations when the derived actor is out of sync with it's peers
 712      * and the only way to bring it in sync is by applying a snapshot
 713      *
 714      * @param snapshotBytes A snapshot of the state of the actor
 715      */
 716     protected abstract void applySnapshot(byte[] snapshotBytes);
 717
 718     /**
 719      * This method will be called by the RaftActor when the state of the
 720      * RaftActor changes. The derived actor can then use methods like
 721      * isLeader or getLeader to do something useful
 722      */
 723     protected abstract void onStateChanged();
 724
 725     /**
 726      * Notifier Actor for this RaftActor to notify when a role change happens
 727      * @return ActorRef - ActorRef of the notifier or Optional.absent if none.
 728      */
 729     protected abstract Optional<ActorRef> getRoleChangeNotifier();
 730
 731     protected void onLeaderChanged(String oldLeader, String newLeader){};
 732
 733     private String getLeaderAddress(){
 734         if(isLeader()){
 735             return getSelf().path().toString();
 736         }
 737         String leaderId = currentBehavior.getLeaderId();
 738         if (leaderId == null) {
 739             return null;
 740         }
 741         String peerAddress = context.getPeerAddress(leaderId);
 742         if(LOG.isDebugEnabled()) {
 743             LOG.debug("{}: getLeaderAddress leaderId = {} peerAddress = {}",
 744                     persistenceId(), leaderId, peerAddress);
 745         }
 746
 747         return peerAddress;
 748     }
 749
 750     private void handleCaptureSnapshotReply(byte[] snapshotBytes) {
 751         LOG.debug("{}: CaptureSnapshotReply received by actor: snapshot size {}", persistenceId(), snapshotBytes.length);
 752
 753         context.getSnapshotManager().persist(persistence(), snapshotBytes, currentBehavior, getTotalMemory());
 754     }
 755
 756     protected long getTotalMemory() {
 757         return Runtime.getRuntime().totalMemory();
 758     }
 759
 760     protected boolean hasFollowers(){
 761         return getRaftActorContext().getPeerAddresses().keySet().size() > 0;
 762     }
 763
 764     private class ReplicatedLogImpl extends AbstractReplicatedLogImpl {
 765         private static final int DATA_SIZE_DIVIDER = 5;
 766         private long dataSizeSinceLastSnapshot = 0L;
 767
 768
 769         public ReplicatedLogImpl(Snapshot snapshot) {
 770             super(snapshot.getLastAppliedIndex(), snapshot.getLastAppliedTerm(),
 771                 snapshot.getUnAppliedEntries());
 772         }
 773
 774         public ReplicatedLogImpl() {
 775             super();
 776         }
 777
 778         @Override public void removeFromAndPersist(long logEntryIndex) {
 779             int adjustedIndex = adjustedIndex(logEntryIndex);
 780
 781             if (adjustedIndex < 0) {
 782                 return;
 783             }
 784
 785             // FIXME: Maybe this should be done after the command is saved
 786             journal.subList(adjustedIndex , journal.size()).clear();
 787
 788             persistence().persist(new DeleteEntries(adjustedIndex), new Procedure<DeleteEntries>() {
 789
 790                 @Override
 791                 public void apply(DeleteEntries param)
 792                         throws Exception {
 793                     //FIXME : Doing nothing for now
 794                     dataSize = 0;
 795                     for (ReplicatedLogEntry entry : journal) {
 796                         dataSize += entry.size();
 797                     }
 798                 }
 799             });
 800         }
 801
 802         @Override public void appendAndPersist(
 803             final ReplicatedLogEntry replicatedLogEntry) {
 804             appendAndPersist(replicatedLogEntry, null);
 805         }
 806
 807         public void appendAndPersist(
 808             final ReplicatedLogEntry replicatedLogEntry,
 809             final Procedure<ReplicatedLogEntry> callback)  {
 810
 811             if(LOG.isDebugEnabled()) {
 812                 LOG.debug("{}: Append log entry and persist {} ", persistenceId(), replicatedLogEntry);
 813             }
 814
 815             // FIXME : By adding the replicated log entry to the in-memory journal we are not truly ensuring durability of the logs
 816             journal.add(replicatedLogEntry);
 817
 818             // When persisting events with persist it is guaranteed that the
 819             // persistent actor will not receive further commands between the
 820             // persist call and the execution(s) of the associated event
 821             // handler. This also holds for multiple persist calls in context
 822             // of a single command.
 823             persistence().persist(replicatedLogEntry,
 824                 new Procedure<ReplicatedLogEntry>() {
 825                     @Override
 826                     public void apply(ReplicatedLogEntry evt) throws Exception {
 827                         int logEntrySize = replicatedLogEntry.size();
 828
 829                         dataSize += logEntrySize;
 830                         long dataSizeForCheck = dataSize;
 831
 832                         dataSizeSinceLastSnapshot += logEntrySize;
 833
 834                         if (!hasFollowers()) {
 835                             // When we do not have followers we do not maintain an in-memory log
 836                             // due to this the journalSize will never become anything close to the
 837                             // snapshot batch count. In fact will mostly be 1.
 838                             // Similarly since the journal's dataSize depends on the entries in the
 839                             // journal the journal's dataSize will never reach a value close to the
 840                             // memory threshold.
 841                             // By maintaining the dataSize outside the journal we are tracking essentially
 842                             // what we have written to the disk however since we no longer are in
 843                             // need of doing a snapshot just for the sake of freeing up memory we adjust
 844                             // the real size of data by the DATA_SIZE_DIVIDER so that we do not snapshot as often
 845                             // as if we were maintaining a real snapshot
 846                             dataSizeForCheck = dataSizeSinceLastSnapshot / DATA_SIZE_DIVIDER;
 847                         }
 848                         long journalSize = replicatedLogEntry.getIndex() + 1;
 849                         long dataThreshold = getTotalMemory() *
 850                                 context.getConfigParams().getSnapshotDataThresholdPercentage() / 100;
 851
 852                         if ((journalSize % context.getConfigParams().getSnapshotBatchCount() == 0
 853                                 || dataSizeForCheck > dataThreshold)) {
 854
 855                             boolean started = context.getSnapshotManager().capture(replicatedLogEntry,
 856                                     currentBehavior.getReplicatedToAllIndex());
 857
 858                             if(started){
 859                                 dataSizeSinceLastSnapshot = 0;
 860                             }
 861
 862                         }
 863
 864                         if (callback != null){
 865                             callback.apply(replicatedLogEntry);
 866                         }
 867                     }
 868                 }
 869             );
 870         }
 871
 872     }
 873
 874     static class DeleteEntries implements Serializable {
 875         private static final long serialVersionUID = 1L;
 876         private final int fromIndex;
 877
 878         public DeleteEntries(int fromIndex) {
 879             this.fromIndex = fromIndex;
 880         }
 881
 882         public int getFromIndex() {
 883             return fromIndex;
 884         }
 885     }
 886
 887     static class UpdateElectionTerm implements Serializable {
 888         private static final long serialVersionUID = 1L;
 889         private final long currentTerm;
 890         private final String votedFor;
 891
 892         public UpdateElectionTerm(long currentTerm, String votedFor) {
 893             this.currentTerm = currentTerm;
 894             this.votedFor = votedFor;
 895         }
 896
 897         public long getCurrentTerm() {
 898             return currentTerm;
 899         }
 900
 901         public String getVotedFor() {
 902             return votedFor;
 903         }
 904     }
 905
 906     private class CreateSnapshotProcedure implements Procedure<Void> {
 907
 908         @Override
 909         public void apply(Void aVoid) throws Exception {
 910             createSnapshot();
 911         }
 912     }
 913
 914     @VisibleForTesting
 915     void setCurrentBehavior(AbstractRaftActorBehavior behavior) {
 916         currentBehavior = behavior;
 917     }
 918
 919     protected RaftActorBehavior getCurrentBehavior() {
 920         return currentBehavior;
 921     }
 922
 923     private static class BehaviorStateHolder {
 924         private RaftActorBehavior behavior;
 925         private String leaderId;
 926
 927         void init(RaftActorBehavior behavior) {
 928             this.behavior = behavior;
 929             this.leaderId = behavior != null ? behavior.getLeaderId() : null;
 930         }
 931
 932         RaftActorBehavior getBehavior() {
 933             return behavior;
 934         }
 935
 936         String getLeaderId() {
 937             return leaderId;
 938         }
 939     }
 940 }