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.common.actor.AbstractUntypedPersistentActor;
  33 import org.opendaylight.controller.cluster.notifications.LeaderStateChanged;
  34 import org.opendaylight.controller.cluster.notifications.RoleChanged;
  35 import org.opendaylight.controller.cluster.raft.base.messages.ApplyJournalEntries;
  36 import org.opendaylight.controller.cluster.raft.base.messages.ApplyLogEntries;
  37 import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot;
  38 import org.opendaylight.controller.cluster.raft.base.messages.ApplyState;
  39 import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshot;
  40 import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshotReply;
  41 import org.opendaylight.controller.cluster.raft.base.messages.Replicate;
  42 import org.opendaylight.controller.cluster.raft.behaviors.AbstractLeader;
  43 import org.opendaylight.controller.cluster.raft.behaviors.AbstractRaftActorBehavior;
  44 import org.opendaylight.controller.cluster.raft.behaviors.Follower;
  45 import org.opendaylight.controller.cluster.raft.behaviors.RaftActorBehavior;
  46 import org.opendaylight.controller.cluster.raft.client.messages.FindLeader;
  47 import org.opendaylight.controller.cluster.raft.client.messages.FindLeaderReply;
  48 import org.opendaylight.controller.cluster.raft.client.messages.FollowerInfo;
  49 import org.opendaylight.controller.cluster.raft.client.messages.GetOnDemandRaftState;
  50 import org.opendaylight.controller.cluster.raft.client.messages.OnDemandRaftState;
  51 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload;
  52 import org.slf4j.Logger;
  53 import org.slf4j.LoggerFactory;
  54
  55 /**
  56  * RaftActor encapsulates a state machine that needs to be kept synchronized
  57  * in a cluster. It implements the RAFT algorithm as described in the paper
  58  * <a href='https://ramcloud.stanford.edu/wiki/download/attachments/11370504/raft.pdf'>
  59  * In Search of an Understandable Consensus Algorithm</a>
  60  * <p/>
  61  * RaftActor has 3 states and each state has a certain behavior associated
  62  * with it. A Raft actor can behave as,
  63  * <ul>
  64  * <li> A Leader </li>
  65  * <li> A Follower (or) </li>
  66  * <li> A Candidate </li>
  67  * </ul>
  68  * <p/>
  69  * <p/>
  70  * A RaftActor MUST be a Leader in order to accept requests from clients to
  71  * change the state of it's encapsulated state machine. Once a RaftActor becomes
  72  * a Leader it is also responsible for ensuring that all followers ultimately
  73  * have the same log and therefore the same state machine as itself.
  74  * <p/>
  75  * <p/>
  76  * The current behavior of a RaftActor determines how election for leadership
  77  * is initiated and how peer RaftActors react to request for votes.
  78  * <p/>
  79  * <p/>
  80  * Each RaftActor also needs to know the current election term. It uses this
  81  * information for a couple of things. One is to simply figure out who it
  82  * voted for in the last election. Another is to figure out if the message
  83  * it received to update it's state is stale.
  84  * <p/>
  85  * <p/>
  86  * The RaftActor uses akka-persistence to store it's replicated log.
  87  * Furthermore through it's behaviors a Raft Actor determines
  88  * <p/>
  89  * <ul>
  90  * <li> when a log entry should be persisted </li>
  91  * <li> when a log entry should be applied to the state machine (and) </li>
  92  * <li> when a snapshot should be saved </li>
  93  * </ul>
  94  */
  95 public abstract class RaftActor extends AbstractUntypedPersistentActor {
  96
  97     private static final long APPLY_STATE_DELAY_THRESHOLD_IN_NANOS = TimeUnit.MILLISECONDS.toNanos(50L); // 50 millis
  98
  99     private static final Procedure<ApplyJournalEntries> APPLY_JOURNAL_ENTRIES_PERSIST_CALLBACK =
 100             new Procedure<ApplyJournalEntries>() {
 101                 @Override
 102                 public void apply(ApplyJournalEntries param) throws Exception {
 103                 }
 104             };
 105     private static final String COMMIT_SNAPSHOT = "commit_snapshot";
 106
 107     protected final Logger LOG = LoggerFactory.getLogger(getClass());
 108
 109     /**
 110      * The current state determines the current behavior of a RaftActor
 111      * A Raft Actor always starts off in the Follower State
 112      */
 113     private RaftActorBehavior currentBehavior;
 114
 115     /**
 116      * This context should NOT be passed directly to any other actor it is
 117      * only to be consumed by the RaftActorBehaviors
 118      */
 119     private final RaftActorContextImpl context;
 120
 121     private final Procedure<Void> createSnapshotProcedure = new CreateSnapshotProcedure();
 122
 123     /**
 124      * The in-memory journal
 125      */
 126     private ReplicatedLogImpl replicatedLog = new ReplicatedLogImpl();
 127
 128     private Stopwatch recoveryTimer;
 129
 130     private int currentRecoveryBatchCount;
 131
 132     private final BehaviorStateHolder reusableBehaviorStateHolder = new BehaviorStateHolder();
 133
 134     public RaftActor(String id, Map<String, String> peerAddresses) {
 135         this(id, peerAddresses, Optional.<ConfigParams>absent());
 136     }
 137
 138     public RaftActor(String id, Map<String, String> peerAddresses,
 139          Optional<ConfigParams> configParams) {
 140
 141         context = new RaftActorContextImpl(this.getSelf(),
 142             this.getContext(), id, new ElectionTermImpl(),
 143             -1, -1, replicatedLog, peerAddresses,
 144             (configParams.isPresent() ? configParams.get(): new DefaultConfigParamsImpl()),
 145             LOG);
 146     }
 147
 148     private void initRecoveryTimer() {
 149         if(recoveryTimer == null) {
 150             recoveryTimer = Stopwatch.createStarted();
 151         }
 152     }
 153
 154     @Override
 155     public void preStart() throws Exception {
 156         LOG.info("Starting recovery for {} with journal batch size {}", persistenceId(),
 157                 context.getConfigParams().getJournalRecoveryLogBatchSize());
 158
 159         super.preStart();
 160     }
 161
 162     @Override
 163     public void postStop() {
 164         if(currentBehavior != null) {
 165             try {
 166                 currentBehavior.close();
 167             } catch (Exception e) {
 168                 LOG.debug("{}: Error closing behavior {}", persistenceId(), currentBehavior.state());
 169             }
 170         }
 171
 172         super.postStop();
 173     }
 174
 175     @Override
 176     public void handleRecover(Object message) {
 177         if(persistence().isRecoveryApplicable()) {
 178             if (message instanceof SnapshotOffer) {
 179                 onRecoveredSnapshot((SnapshotOffer) message);
 180             } else if (message instanceof ReplicatedLogEntry) {
 181                 onRecoveredJournalLogEntry((ReplicatedLogEntry) message);
 182             } else if (message instanceof ApplyLogEntries) {
 183                 // Handle this message for backwards compatibility with pre-Lithium versions.
 184                 onRecoveredApplyLogEntries(((ApplyLogEntries) message).getToIndex());
 185             } else if (message instanceof ApplyJournalEntries) {
 186                 onRecoveredApplyLogEntries(((ApplyJournalEntries) message).getToIndex());
 187             } else if (message instanceof DeleteEntries) {
 188                 replicatedLog.removeFrom(((DeleteEntries) message).getFromIndex());
 189             } else if (message instanceof UpdateElectionTerm) {
 190                 context.getTermInformation().update(((UpdateElectionTerm) message).getCurrentTerm(),
 191                         ((UpdateElectionTerm) message).getVotedFor());
 192             } else if (message instanceof RecoveryCompleted) {
 193                 onRecoveryCompletedMessage();
 194             }
 195         } else {
 196             if (message instanceof RecoveryCompleted) {
 197                 // Delete all the messages from the akka journal so that we do not end up with consistency issues
 198                 // Note I am not using the dataPersistenceProvider and directly using the akka api here
 199                 deleteMessages(lastSequenceNr());
 200
 201                 // Delete all the akka snapshots as they will not be needed
 202                 deleteSnapshots(new SnapshotSelectionCriteria(scala.Long.MaxValue(), scala.Long.MaxValue()));
 203
 204                 onRecoveryComplete();
 205
 206                 initializeBehavior();
 207             }
 208         }
 209     }
 210
 211     private void onRecoveredSnapshot(SnapshotOffer offer) {
 212         if(LOG.isDebugEnabled()) {
 213             LOG.debug("{}: SnapshotOffer called..", persistenceId());
 214         }
 215
 216         initRecoveryTimer();
 217
 218         Snapshot snapshot = (Snapshot) offer.snapshot();
 219
 220         // Create a replicated log with the snapshot information
 221         // The replicated log can be used later on to retrieve this snapshot
 222         // when we need to install it on a peer
 223         replicatedLog = new ReplicatedLogImpl(snapshot);
 224
 225         context.setReplicatedLog(replicatedLog);
 226         context.setLastApplied(snapshot.getLastAppliedIndex());
 227         context.setCommitIndex(snapshot.getLastAppliedIndex());
 228
 229         Stopwatch timer = Stopwatch.createStarted();
 230
 231         // Apply the snapshot to the actors state
 232         applyRecoverySnapshot(snapshot.getState());
 233
 234         timer.stop();
 235         LOG.info("Recovery snapshot applied for {} in {}: snapshotIndex={}, snapshotTerm={}, journal-size=" +
 236                 replicatedLog.size(), persistenceId(), timer.toString(),
 237                 replicatedLog.getSnapshotIndex(), replicatedLog.getSnapshotTerm());
 238     }
 239
 240     private void onRecoveredJournalLogEntry(ReplicatedLogEntry logEntry) {
 241         if(LOG.isDebugEnabled()) {
 242             LOG.debug("{}: Received ReplicatedLogEntry for recovery: {}", persistenceId(), logEntry.getIndex());
 243         }
 244
 245         replicatedLog.append(logEntry);
 246     }
 247
 248     private void onRecoveredApplyLogEntries(long toIndex) {
 249         if(LOG.isDebugEnabled()) {
 250             LOG.debug("{}: Received ApplyLogEntries for recovery, applying to state: {} to {}",
 251                     persistenceId(), context.getLastApplied() + 1, toIndex);
 252         }
 253
 254         for (long i = context.getLastApplied() + 1; i <= toIndex; i++) {
 255             batchRecoveredLogEntry(replicatedLog.get(i));
 256         }
 257
 258         context.setLastApplied(toIndex);
 259         context.setCommitIndex(toIndex);
 260     }
 261
 262     private void batchRecoveredLogEntry(ReplicatedLogEntry logEntry) {
 263         initRecoveryTimer();
 264
 265         int batchSize = context.getConfigParams().getJournalRecoveryLogBatchSize();
 266         if(currentRecoveryBatchCount == 0) {
 267             startLogRecoveryBatch(batchSize);
 268         }
 269
 270         appendRecoveredLogEntry(logEntry.getData());
 271
 272         if(++currentRecoveryBatchCount >= batchSize) {
 273             endCurrentLogRecoveryBatch();
 274         }
 275     }
 276
 277     private void endCurrentLogRecoveryBatch() {
 278         applyCurrentLogRecoveryBatch();
 279         currentRecoveryBatchCount = 0;
 280     }
 281
 282     private void onRecoveryCompletedMessage() {
 283         if(currentRecoveryBatchCount > 0) {
 284             endCurrentLogRecoveryBatch();
 285         }
 286
 287         onRecoveryComplete();
 288
 289         String recoveryTime = "";
 290         if(recoveryTimer != null) {
 291             recoveryTimer.stop();
 292             recoveryTime = " in " + recoveryTimer.toString();
 293             recoveryTimer = null;
 294         }
 295
 296         LOG.info(
 297             "Recovery completed" + recoveryTime + " - Switching actor to Follower - " +
 298                 "Persistence Id =  " + persistenceId() +
 299                 " Last index in log={}, snapshotIndex={}, snapshotTerm={}, " +
 300                 "journal-size={}",
 301             replicatedLog.lastIndex(), replicatedLog.getSnapshotIndex(),
 302             replicatedLog.getSnapshotTerm(), replicatedLog.size());
 303
 304         initializeBehavior();
 305     }
 306
 307     protected void initializeBehavior(){
 308         changeCurrentBehavior(new Follower(context));
 309     }
 310
 311     protected void changeCurrentBehavior(RaftActorBehavior newBehavior){
 312         reusableBehaviorStateHolder.init(currentBehavior);
 313         currentBehavior = newBehavior;
 314         handleBehaviorChange(reusableBehaviorStateHolder, currentBehavior);
 315     }
 316
 317     @Override public void handleCommand(Object message) {
 318         if (message instanceof ApplyState){
 319             ApplyState applyState = (ApplyState) message;
 320
 321             long elapsedTime = (System.nanoTime() - applyState.getStartTime());
 322             if(elapsedTime >= APPLY_STATE_DELAY_THRESHOLD_IN_NANOS){
 323                 LOG.warn("ApplyState took more time than expected. Elapsed Time = {} ms ApplyState = {}",
 324                         TimeUnit.NANOSECONDS.toMillis(elapsedTime), applyState);
 325             }
 326
 327             if(LOG.isDebugEnabled()) {
 328                 LOG.debug("{}: Applying state for log index {} data {}",
 329                     persistenceId(), applyState.getReplicatedLogEntry().getIndex(),
 330                     applyState.getReplicatedLogEntry().getData());
 331             }
 332
 333             applyState(applyState.getClientActor(), applyState.getIdentifier(),
 334                 applyState.getReplicatedLogEntry().getData());
 335
 336         } else if (message instanceof ApplyJournalEntries){
 337             ApplyJournalEntries applyEntries = (ApplyJournalEntries) message;
 338             if(LOG.isDebugEnabled()) {
 339                 LOG.debug("{}: Persisting ApplyLogEntries with index={}", persistenceId(), applyEntries.getToIndex());
 340             }
 341
 342             persistence().persist(applyEntries, APPLY_JOURNAL_ENTRIES_PERSIST_CALLBACK);
 343
 344         } else if(message instanceof ApplySnapshot ) {
 345             Snapshot snapshot = ((ApplySnapshot) message).getSnapshot();
 346
 347             if(LOG.isDebugEnabled()) {
 348                 LOG.debug("{}: ApplySnapshot called on Follower Actor " +
 349                         "snapshotIndex:{}, snapshotTerm:{}", persistenceId(), snapshot.getLastAppliedIndex(),
 350                     snapshot.getLastAppliedTerm()
 351                 );
 352             }
 353
 354             applySnapshot(snapshot.getState());
 355
 356             //clears the followers log, sets the snapshot index to ensure adjusted-index works
 357             replicatedLog = new ReplicatedLogImpl(snapshot);
 358             context.setReplicatedLog(replicatedLog);
 359             context.setLastApplied(snapshot.getLastAppliedIndex());
 360
 361         } else if (message instanceof FindLeader) {
 362             getSender().tell(
 363                 new FindLeaderReply(getLeaderAddress()),
 364                 getSelf()
 365             );
 366
 367         } else if (message instanceof SaveSnapshotSuccess) {
 368             SaveSnapshotSuccess success = (SaveSnapshotSuccess) message;
 369             LOG.info("{}: SaveSnapshotSuccess received for snapshot", persistenceId());
 370
 371             long sequenceNumber = success.metadata().sequenceNr();
 372
 373             commitSnapshot(sequenceNumber);
 374
 375         } else if (message instanceof SaveSnapshotFailure) {
 376             SaveSnapshotFailure saveSnapshotFailure = (SaveSnapshotFailure) message;
 377
 378             LOG.error("{}: SaveSnapshotFailure received for snapshot Cause:",
 379                     persistenceId(), saveSnapshotFailure.cause());
 380
 381             context.getSnapshotManager().rollback();
 382
 383         } else if (message instanceof CaptureSnapshot) {
 384             LOG.debug("{}: CaptureSnapshot received by actor: {}", persistenceId(), message);
 385
 386             context.getSnapshotManager().create(createSnapshotProcedure);
 387
 388         } else if (message instanceof CaptureSnapshotReply) {
 389             handleCaptureSnapshotReply(((CaptureSnapshotReply) message).getSnapshot());
 390         } else if(message instanceof GetOnDemandRaftState) {
 391             onGetOnDemandRaftStats();
 392         } else if (message.equals(COMMIT_SNAPSHOT)) {
 393             commitSnapshot(-1);
 394         } else {
 395             reusableBehaviorStateHolder.init(currentBehavior);
 396
 397             currentBehavior = currentBehavior.handleMessage(getSender(), message);
 398
 399             handleBehaviorChange(reusableBehaviorStateHolder, currentBehavior);
 400         }
 401     }
 402
 403     private void onGetOnDemandRaftStats() {
 404         // Debugging message to retrieve raft stats.
 405
 406         OnDemandRaftState.Builder builder = OnDemandRaftState.builder()
 407                 .commitIndex(context.getCommitIndex())
 408                 .currentTerm(context.getTermInformation().getCurrentTerm())
 409                 .inMemoryJournalDataSize(replicatedLog.dataSize())
 410                 .inMemoryJournalLogSize(replicatedLog.size())
 411                 .isSnapshotCaptureInitiated(context.getSnapshotManager().isCapturing())
 412                 .lastApplied(context.getLastApplied())
 413                 .lastIndex(replicatedLog.lastIndex())
 414                 .lastTerm(replicatedLog.lastTerm())
 415                 .leader(getLeaderId())
 416                 .raftState(currentBehavior.state().toString())
 417                 .replicatedToAllIndex(currentBehavior.getReplicatedToAllIndex())
 418                 .snapshotIndex(replicatedLog.getSnapshotIndex())
 419                 .snapshotTerm(replicatedLog.getSnapshotTerm())
 420                 .votedFor(context.getTermInformation().getVotedFor())
 421                 .peerAddresses(ImmutableMap.copyOf(context.getPeerAddresses()));
 422
 423         ReplicatedLogEntry lastLogEntry = getLastLogEntry();
 424         if (lastLogEntry != null) {
 425             builder.lastLogIndex(lastLogEntry.getIndex());
 426             builder.lastLogTerm(lastLogEntry.getTerm());
 427         }
 428
 429         if(currentBehavior instanceof AbstractLeader) {
 430             AbstractLeader leader = (AbstractLeader)currentBehavior;
 431             Collection<String> followerIds = leader.getFollowerIds();
 432             List<FollowerInfo> followerInfoList = Lists.newArrayListWithCapacity(followerIds.size());
 433             for(String id: followerIds) {
 434                 final FollowerLogInformation info = leader.getFollower(id);
 435                 followerInfoList.add(new FollowerInfo(id, info.getNextIndex(), info.getMatchIndex(),
 436                         info.isFollowerActive(), DurationFormatUtils.formatDurationHMS(info.timeSinceLastActivity())));
 437             }
 438
 439             builder.followerInfoList(followerInfoList);
 440         }
 441
 442         sender().tell(builder.build(), self());
 443
 444     }
 445
 446     private void handleBehaviorChange(BehaviorStateHolder oldBehaviorState, RaftActorBehavior currentBehavior) {
 447         RaftActorBehavior oldBehavior = oldBehaviorState.getBehavior();
 448
 449         if (oldBehavior != currentBehavior){
 450             onStateChanged();
 451         }
 452
 453         String oldBehaviorLeaderId = oldBehavior == null ? null : oldBehaviorState.getLeaderId();
 454         String oldBehaviorStateName = oldBehavior == null ? null : oldBehavior.state().name();
 455
 456         // it can happen that the state has not changed but the leader has changed.
 457         Optional<ActorRef> roleChangeNotifier = getRoleChangeNotifier();
 458         if(!Objects.equal(oldBehaviorLeaderId, currentBehavior.getLeaderId())) {
 459             if(roleChangeNotifier.isPresent()) {
 460                 roleChangeNotifier.get().tell(new LeaderStateChanged(getId(), currentBehavior.getLeaderId()), getSelf());
 461             }
 462
 463             onLeaderChanged(oldBehaviorLeaderId, currentBehavior.getLeaderId());
 464         }
 465
 466         if (roleChangeNotifier.isPresent() &&
 467                 (oldBehavior == null || (oldBehavior.state() != currentBehavior.state()))) {
 468             roleChangeNotifier.get().tell(new RoleChanged(getId(), oldBehaviorStateName ,
 469                     currentBehavior.state().name()), getSelf());
 470         }
 471     }
 472
 473     /**
 474      * When a derived RaftActor needs to persist something it must call
 475      * persistData.
 476      *
 477      * @param clientActor
 478      * @param identifier
 479      * @param data
 480      */
 481     protected void persistData(final ActorRef clientActor, final String identifier,
 482         final Payload data) {
 483
 484         ReplicatedLogEntry replicatedLogEntry = new ReplicatedLogImplEntry(
 485             context.getReplicatedLog().lastIndex() + 1,
 486             context.getTermInformation().getCurrentTerm(), data);
 487
 488         if(LOG.isDebugEnabled()) {
 489             LOG.debug("{}: Persist data {}", persistenceId(), replicatedLogEntry);
 490         }
 491
 492         final RaftActorContext raftContext = getRaftActorContext();
 493
 494         replicatedLog
 495                 .appendAndPersist(replicatedLogEntry, new Procedure<ReplicatedLogEntry>() {
 496                     @Override
 497                     public void apply(ReplicatedLogEntry replicatedLogEntry) throws Exception {
 498                         if(!hasFollowers()){
 499                             // Increment the Commit Index and the Last Applied values
 500                             raftContext.setCommitIndex(replicatedLogEntry.getIndex());
 501                             raftContext.setLastApplied(replicatedLogEntry.getIndex());
 502
 503                             // Apply the state immediately
 504                             applyState(clientActor, identifier, data);
 505
 506                             // Send a ApplyJournalEntries message so that we write the fact that we applied
 507                             // the state to durable storage
 508                             self().tell(new ApplyJournalEntries(replicatedLogEntry.getIndex()), self());
 509
 510                             context.getSnapshotManager().trimLog(context.getLastApplied(), currentBehavior);
 511
 512                         } else if (clientActor != null) {
 513                             // Send message for replication
 514                             currentBehavior.handleMessage(getSelf(),
 515                                     new Replicate(clientActor, identifier,
 516                                             replicatedLogEntry)
 517                             );
 518                         }
 519
 520                     }
 521                 });    }
 522
 523     protected String getId() {
 524         return context.getId();
 525     }
 526
 527     /**
 528      * Derived actors can call the isLeader method to check if the current
 529      * RaftActor is the Leader or not
 530      *
 531      * @return true it this RaftActor is a Leader false otherwise
 532      */
 533     protected boolean isLeader() {
 534         return context.getId().equals(currentBehavior.getLeaderId());
 535     }
 536
 537     /**
 538      * Derived actor can call getLeader if they need a reference to the Leader.
 539      * This would be useful for example in forwarding a request to an actor
 540      * which is the leader
 541      *
 542      * @return A reference to the leader if known, null otherwise
 543      */
 544     protected ActorSelection getLeader(){
 545         String leaderAddress = getLeaderAddress();
 546
 547         if(leaderAddress == null){
 548             return null;
 549         }
 550
 551         return context.actorSelection(leaderAddress);
 552     }
 553
 554     /**
 555      *
 556      * @return the current leader's id
 557      */
 558     protected String getLeaderId(){
 559         return currentBehavior.getLeaderId();
 560     }
 561
 562     protected RaftState getRaftState() {
 563         return currentBehavior.state();
 564     }
 565
 566     protected ReplicatedLogEntry getLastLogEntry() {
 567         return replicatedLog.last();
 568     }
 569
 570     protected Long getCurrentTerm(){
 571         return context.getTermInformation().getCurrentTerm();
 572     }
 573
 574     protected Long getCommitIndex(){
 575         return context.getCommitIndex();
 576     }
 577
 578     protected Long getLastApplied(){
 579         return context.getLastApplied();
 580     }
 581
 582     protected RaftActorContext getRaftActorContext() {
 583         return context;
 584     }
 585
 586     protected void updateConfigParams(ConfigParams configParams) {
 587         context.setConfigParams(configParams);
 588     }
 589
 590     /**
 591      * setPeerAddress sets the address of a known peer at a later time.
 592      * <p>
 593      * This is to account for situations where a we know that a peer
 594      * exists but we do not know an address up-front. This may also be used in
 595      * situations where a known peer starts off in a different location and we
 596      * need to change it's address
 597      * <p>
 598      * Note that if the peerId does not match the list of peers passed to
 599      * this actor during construction an IllegalStateException will be thrown.
 600      *
 601      * @param peerId
 602      * @param peerAddress
 603      */
 604     protected void setPeerAddress(String peerId, String peerAddress){
 605         context.setPeerAddress(peerId, peerAddress);
 606     }
 607
 608     protected void commitSnapshot(long sequenceNumber) {
 609         context.getSnapshotManager().commit(persistence(), sequenceNumber);
 610     }
 611
 612     /**
 613      * The applyState method will be called by the RaftActor when some data
 614      * needs to be applied to the actor's state
 615      *
 616      * @param clientActor A reference to the client who sent this message. This
 617      *                    is the same reference that was passed to persistData
 618      *                    by the derived actor. clientActor may be null when
 619      *                    the RaftActor is behaving as a follower or during
 620      *                    recovery.
 621      * @param identifier  The identifier of the persisted data. This is also
 622      *                    the same identifier that was passed to persistData by
 623      *                    the derived actor. identifier may be null when
 624      *                    the RaftActor is behaving as a follower or during
 625      *                    recovery
 626      * @param data        A piece of data that was persisted by the persistData call.
 627      *                    This should NEVER be null.
 628      */
 629     protected abstract void applyState(ActorRef clientActor, String identifier,
 630         Object data);
 631
 632     /**
 633      * This method is called during recovery at the start of a batch of state entries. Derived
 634      * classes should perform any initialization needed to start a batch.
 635      */
 636     protected abstract void startLogRecoveryBatch(int maxBatchSize);
 637
 638     /**
 639      * This method is called during recovery to append state data to the current batch. This method
 640      * is called 1 or more times after {@link #startLogRecoveryBatch}.
 641      *
 642      * @param data the state data
 643      */
 644     protected abstract void appendRecoveredLogEntry(Payload data);
 645
 646     /**
 647      * This method is called during recovery to reconstruct the state of the actor.
 648      *
 649      * @param snapshotBytes A snapshot of the state of the actor
 650      */
 651     protected abstract void applyRecoverySnapshot(byte[] snapshotBytes);
 652
 653     /**
 654      * This method is called during recovery at the end of a batch to apply the current batched
 655      * log entries. This method is called after {@link #appendRecoveredLogEntry}.
 656      */
 657     protected abstract void applyCurrentLogRecoveryBatch();
 658
 659     /**
 660      * This method is called when recovery is complete.
 661      */
 662     protected abstract void onRecoveryComplete();
 663
 664     /**
 665      * This method will be called by the RaftActor when a snapshot needs to be
 666      * created. The derived actor should respond with its current state.
 667      * <p/>
 668      * During recovery the state that is returned by the derived actor will
 669      * be passed back to it by calling the applySnapshot  method
 670      *
 671      * @return The current state of the actor
 672      */
 673     protected abstract void createSnapshot();
 674
 675     /**
 676      * This method can be called at any other point during normal
 677      * operations when the derived actor is out of sync with it's peers
 678      * and the only way to bring it in sync is by applying a snapshot
 679      *
 680      * @param snapshotBytes A snapshot of the state of the actor
 681      */
 682     protected abstract void applySnapshot(byte[] snapshotBytes);
 683
 684     /**
 685      * This method will be called by the RaftActor when the state of the
 686      * RaftActor changes. The derived actor can then use methods like
 687      * isLeader or getLeader to do something useful
 688      */
 689     protected abstract void onStateChanged();
 690
 691     protected abstract DataPersistenceProvider persistence();
 692
 693     /**
 694      * Notifier Actor for this RaftActor to notify when a role change happens
 695      * @return ActorRef - ActorRef of the notifier or Optional.absent if none.
 696      */
 697     protected abstract Optional<ActorRef> getRoleChangeNotifier();
 698
 699     protected void onLeaderChanged(String oldLeader, String newLeader){};
 700
 701     private String getLeaderAddress(){
 702         if(isLeader()){
 703             return getSelf().path().toString();
 704         }
 705         String leaderId = currentBehavior.getLeaderId();
 706         if (leaderId == null) {
 707             return null;
 708         }
 709         String peerAddress = context.getPeerAddress(leaderId);
 710         if(LOG.isDebugEnabled()) {
 711             LOG.debug("{}: getLeaderAddress leaderId = {} peerAddress = {}",
 712                     persistenceId(), leaderId, peerAddress);
 713         }
 714
 715         return peerAddress;
 716     }
 717
 718     private void handleCaptureSnapshotReply(byte[] snapshotBytes) {
 719         LOG.debug("{}: CaptureSnapshotReply received by actor: snapshot size {}", persistenceId(), snapshotBytes.length);
 720
 721         context.getSnapshotManager().persist(persistence(), snapshotBytes, currentBehavior, getTotalMemory());
 722     }
 723
 724     protected long getTotalMemory() {
 725         return Runtime.getRuntime().totalMemory();
 726     }
 727
 728     protected boolean hasFollowers(){
 729         return getRaftActorContext().getPeerAddresses().keySet().size() > 0;
 730     }
 731
 732     private class ReplicatedLogImpl extends AbstractReplicatedLogImpl {
 733         private static final int DATA_SIZE_DIVIDER = 5;
 734         private long dataSizeSinceLastSnapshot = 0L;
 735
 736
 737         public ReplicatedLogImpl(Snapshot snapshot) {
 738             super(snapshot.getLastAppliedIndex(), snapshot.getLastAppliedTerm(),
 739                 snapshot.getUnAppliedEntries());
 740         }
 741
 742         public ReplicatedLogImpl() {
 743             super();
 744         }
 745
 746         @Override public void removeFromAndPersist(long logEntryIndex) {
 747             int adjustedIndex = adjustedIndex(logEntryIndex);
 748
 749             if (adjustedIndex < 0) {
 750                 return;
 751             }
 752
 753             // FIXME: Maybe this should be done after the command is saved
 754             journal.subList(adjustedIndex , journal.size()).clear();
 755
 756             persistence().persist(new DeleteEntries(adjustedIndex), new Procedure<DeleteEntries>() {
 757
 758                 @Override
 759                 public void apply(DeleteEntries param)
 760                         throws Exception {
 761                     //FIXME : Doing nothing for now
 762                     dataSize = 0;
 763                     for (ReplicatedLogEntry entry : journal) {
 764                         dataSize += entry.size();
 765                     }
 766                 }
 767             });
 768         }
 769
 770         @Override public void appendAndPersist(
 771             final ReplicatedLogEntry replicatedLogEntry) {
 772             appendAndPersist(replicatedLogEntry, null);
 773         }
 774
 775         public void appendAndPersist(
 776             final ReplicatedLogEntry replicatedLogEntry,
 777             final Procedure<ReplicatedLogEntry> callback)  {
 778
 779             if(LOG.isDebugEnabled()) {
 780                 LOG.debug("{}: Append log entry and persist {} ", persistenceId(), replicatedLogEntry);
 781             }
 782
 783             // FIXME : By adding the replicated log entry to the in-memory journal we are not truly ensuring durability of the logs
 784             journal.add(replicatedLogEntry);
 785
 786             // When persisting events with persist it is guaranteed that the
 787             // persistent actor will not receive further commands between the
 788             // persist call and the execution(s) of the associated event
 789             // handler. This also holds for multiple persist calls in context
 790             // of a single command.
 791             persistence().persist(replicatedLogEntry,
 792                 new Procedure<ReplicatedLogEntry>() {
 793                     @Override
 794                     public void apply(ReplicatedLogEntry evt) throws Exception {
 795                         int logEntrySize = replicatedLogEntry.size();
 796
 797                         dataSize += logEntrySize;
 798                         long dataSizeForCheck = dataSize;
 799
 800                         dataSizeSinceLastSnapshot += logEntrySize;
 801
 802                         if (!hasFollowers()) {
 803                             // When we do not have followers we do not maintain an in-memory log
 804                             // due to this the journalSize will never become anything close to the
 805                             // snapshot batch count. In fact will mostly be 1.
 806                             // Similarly since the journal's dataSize depends on the entries in the
 807                             // journal the journal's dataSize will never reach a value close to the
 808                             // memory threshold.
 809                             // By maintaining the dataSize outside the journal we are tracking essentially
 810                             // what we have written to the disk however since we no longer are in
 811                             // need of doing a snapshot just for the sake of freeing up memory we adjust
 812                             // the real size of data by the DATA_SIZE_DIVIDER so that we do not snapshot as often
 813                             // as if we were maintaining a real snapshot
 814                             dataSizeForCheck = dataSizeSinceLastSnapshot / DATA_SIZE_DIVIDER;
 815                         }
 816                         long journalSize = replicatedLogEntry.getIndex() + 1;
 817                         long dataThreshold = getTotalMemory() *
 818                                 context.getConfigParams().getSnapshotDataThresholdPercentage() / 100;
 819
 820                         if ((journalSize % context.getConfigParams().getSnapshotBatchCount() == 0
 821                                 || dataSizeForCheck > dataThreshold)) {
 822
 823                             boolean started = context.getSnapshotManager().capture(replicatedLogEntry,
 824                                     currentBehavior.getReplicatedToAllIndex());
 825
 826                             if(started){
 827                                 dataSizeSinceLastSnapshot = 0;
 828                             }
 829
 830                         }
 831
 832                         if (callback != null){
 833                             callback.apply(replicatedLogEntry);
 834                         }
 835                     }
 836                 }
 837             );
 838         }
 839
 840     }
 841
 842     static class DeleteEntries implements Serializable {
 843         private static final long serialVersionUID = 1L;
 844         private final int fromIndex;
 845
 846         public DeleteEntries(int fromIndex) {
 847             this.fromIndex = fromIndex;
 848         }
 849
 850         public int getFromIndex() {
 851             return fromIndex;
 852         }
 853     }
 854
 855
 856     private class ElectionTermImpl implements ElectionTerm {
 857         /**
 858          * Identifier of the actor whose election term information this is
 859          */
 860         private long currentTerm = 0;
 861         private String votedFor = null;
 862
 863         @Override
 864         public long getCurrentTerm() {
 865             return currentTerm;
 866         }
 867
 868         @Override
 869         public String getVotedFor() {
 870             return votedFor;
 871         }
 872
 873         @Override public void update(long currentTerm, String votedFor) {
 874             if(LOG.isDebugEnabled()) {
 875                 LOG.debug("{}: Set currentTerm={}, votedFor={}", persistenceId(), currentTerm, votedFor);
 876             }
 877             this.currentTerm = currentTerm;
 878             this.votedFor = votedFor;
 879         }
 880
 881         @Override
 882         public void updateAndPersist(long currentTerm, String votedFor){
 883             update(currentTerm, votedFor);
 884             // FIXME : Maybe first persist then update the state
 885             persistence().persist(new UpdateElectionTerm(this.currentTerm, this.votedFor), new Procedure<UpdateElectionTerm>(){
 886
 887                 @Override public void apply(UpdateElectionTerm param)
 888                     throws Exception {
 889
 890                 }
 891             });
 892         }
 893     }
 894
 895     static class UpdateElectionTerm implements Serializable {
 896         private static final long serialVersionUID = 1L;
 897         private final long currentTerm;
 898         private final String votedFor;
 899
 900         public UpdateElectionTerm(long currentTerm, String votedFor) {
 901             this.currentTerm = currentTerm;
 902             this.votedFor = votedFor;
 903         }
 904
 905         public long getCurrentTerm() {
 906             return currentTerm;
 907         }
 908
 909         public String getVotedFor() {
 910             return votedFor;
 911         }
 912     }
 913
 914     protected class NonPersistentRaftDataProvider extends NonPersistentDataProvider {
 915
 916         public NonPersistentRaftDataProvider(){
 917
 918         }
 919
 920         /**
 921          * The way snapshotting works is,
 922          * <ol>
 923          * <li> RaftActor calls createSnapshot on the Shard
 924          * <li> Shard sends a CaptureSnapshotReply and RaftActor then calls saveSnapshot
 925          * <li> When saveSnapshot is invoked on the akka-persistence API it uses the SnapshotStore to save the snapshot.
 926          * The SnapshotStore sends SaveSnapshotSuccess or SaveSnapshotFailure. When the RaftActor gets SaveSnapshot
 927          * success it commits the snapshot to the in-memory journal. This commitSnapshot is mimicking what is done
 928          * in SaveSnapshotSuccess.
 929          * </ol>
 930          * @param o
 931          */
 932         @Override
 933         public void saveSnapshot(Object o) {
 934             // Make saving Snapshot successful
 935             // Committing the snapshot here would end up calling commit in the creating state which would
 936             // be a state violation. That's why now we send a message to commit the snapshot.
 937             self().tell(COMMIT_SNAPSHOT, self());
 938         }
 939     }
 940
 941
 942     private class CreateSnapshotProcedure implements Procedure<Void> {
 943
 944         @Override
 945         public void apply(Void aVoid) throws Exception {
 946             createSnapshot();
 947         }
 948     }
 949
 950     @VisibleForTesting
 951     void setCurrentBehavior(AbstractRaftActorBehavior behavior) {
 952         currentBehavior = behavior;
 953     }
 954
 955     protected RaftActorBehavior getCurrentBehavior() {
 956         return currentBehavior;
 957     }
 958
 959     private static class BehaviorStateHolder {
 960         private RaftActorBehavior behavior;
 961         private String leaderId;
 962
 963         void init(RaftActorBehavior behavior) {
 964             this.behavior = behavior;
 965             this.leaderId = behavior != null ? behavior.getLeaderId() : null;
 966         }
 967
 968         RaftActorBehavior getBehavior() {
 969             return behavior;
 970         }
 971
 972         String getLeaderId() {
 973             return leaderId;
 974         }
 975     }
 976 }