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.Optional;
  21 import com.google.common.base.Stopwatch;
  22 import com.google.protobuf.ByteString;
  23 import java.io.Serializable;
  24 import java.util.Map;
  25 import java.util.concurrent.TimeUnit;
  26 import org.opendaylight.controller.cluster.DataPersistenceProvider;
  27 import org.opendaylight.controller.cluster.common.actor.AbstractUntypedPersistentActor;
  28 import org.opendaylight.controller.cluster.notifications.RoleChanged;
  29 import org.opendaylight.controller.cluster.raft.base.messages.ApplyJournalEntries;
  30 import org.opendaylight.controller.cluster.raft.base.messages.ApplyLogEntries;
  31 import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot;
  32 import org.opendaylight.controller.cluster.raft.base.messages.ApplyState;
  33 import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshot;
  34 import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshotReply;
  35 import org.opendaylight.controller.cluster.raft.base.messages.Replicate;
  36 import org.opendaylight.controller.cluster.raft.base.messages.SendInstallSnapshot;
  37 import org.opendaylight.controller.cluster.raft.behaviors.AbstractRaftActorBehavior;
  38 import org.opendaylight.controller.cluster.raft.behaviors.Follower;
  39 import org.opendaylight.controller.cluster.raft.behaviors.RaftActorBehavior;
  40 import org.opendaylight.controller.cluster.raft.client.messages.FindLeader;
  41 import org.opendaylight.controller.cluster.raft.client.messages.FindLeaderReply;
  42 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload;
  43 import org.slf4j.Logger;
  44 import org.slf4j.LoggerFactory;
  45
  46 /**
  47  * RaftActor encapsulates a state machine that needs to be kept synchronized
  48  * in a cluster. It implements the RAFT algorithm as described in the paper
  49  * <a href='https://ramcloud.stanford.edu/wiki/download/attachments/11370504/raft.pdf'>
  50  * In Search of an Understandable Consensus Algorithm</a>
  51  * <p/>
  52  * RaftActor has 3 states and each state has a certain behavior associated
  53  * with it. A Raft actor can behave as,
  54  * <ul>
  55  * <li> A Leader </li>
  56  * <li> A Follower (or) </li>
  57  * <li> A Candidate </li>
  58  * </ul>
  59  * <p/>
  60  * <p/>
  61  * A RaftActor MUST be a Leader in order to accept requests from clients to
  62  * change the state of it's encapsulated state machine. Once a RaftActor becomes
  63  * a Leader it is also responsible for ensuring that all followers ultimately
  64  * have the same log and therefore the same state machine as itself.
  65  * <p/>
  66  * <p/>
  67  * The current behavior of a RaftActor determines how election for leadership
  68  * is initiated and how peer RaftActors react to request for votes.
  69  * <p/>
  70  * <p/>
  71  * Each RaftActor also needs to know the current election term. It uses this
  72  * information for a couple of things. One is to simply figure out who it
  73  * voted for in the last election. Another is to figure out if the message
  74  * it received to update it's state is stale.
  75  * <p/>
  76  * <p/>
  77  * The RaftActor uses akka-persistence to store it's replicated log.
  78  * Furthermore through it's behaviors a Raft Actor determines
  79  * <p/>
  80  * <ul>
  81  * <li> when a log entry should be persisted </li>
  82  * <li> when a log entry should be applied to the state machine (and) </li>
  83  * <li> when a snapshot should be saved </li>
  84  * </ul>
  85  */
  86 public abstract class RaftActor extends AbstractUntypedPersistentActor {
  87
  88     private static final long APPLY_STATE_DELAY_THRESHOLD_IN_NANOS = TimeUnit.MILLISECONDS.toNanos(50L); // 50 millis
  89
  90     protected final Logger LOG = LoggerFactory.getLogger(getClass());
  91
  92     /**
  93      * The current state determines the current behavior of a RaftActor
  94      * A Raft Actor always starts off in the Follower State
  95      */
  96     private RaftActorBehavior currentBehavior;
  97
  98     /**
  99      * This context should NOT be passed directly to any other actor it is
 100      * only to be consumed by the RaftActorBehaviors
 101      */
 102     private final RaftActorContextImpl context;
 103
 104     /**
 105      * The in-memory journal
 106      */
 107     private ReplicatedLogImpl replicatedLog = new ReplicatedLogImpl();
 108
 109     private CaptureSnapshot captureSnapshot = null;
 110
 111     private Stopwatch recoveryTimer;
 112
 113     private int currentRecoveryBatchCount;
 114
 115     public RaftActor(String id, Map<String, String> peerAddresses) {
 116         this(id, peerAddresses, Optional.<ConfigParams>absent());
 117     }
 118
 119     public RaftActor(String id, Map<String, String> peerAddresses,
 120          Optional<ConfigParams> configParams) {
 121
 122         context = new RaftActorContextImpl(this.getSelf(),
 123             this.getContext(), id, new ElectionTermImpl(),
 124             -1, -1, replicatedLog, peerAddresses,
 125             (configParams.isPresent() ? configParams.get(): new DefaultConfigParamsImpl()),
 126             LOG);
 127     }
 128
 129     private void initRecoveryTimer() {
 130         if(recoveryTimer == null) {
 131             recoveryTimer = Stopwatch.createStarted();
 132         }
 133     }
 134
 135     @Override
 136     public void preStart() throws Exception {
 137         LOG.info("Starting recovery for {} with journal batch size {}", persistenceId(),
 138                 context.getConfigParams().getJournalRecoveryLogBatchSize());
 139
 140         super.preStart();
 141     }
 142
 143     @Override
 144     public void postStop() {
 145         if(currentBehavior != null) {
 146             try {
 147                 currentBehavior.close();
 148             } catch (Exception e) {
 149                 LOG.debug("{}: Error closing behavior {}", persistenceId(), currentBehavior.state());
 150             }
 151         }
 152
 153         super.postStop();
 154     }
 155
 156     @Override
 157     public void handleRecover(Object message) {
 158         if(persistence().isRecoveryApplicable()) {
 159             if (message instanceof SnapshotOffer) {
 160                 onRecoveredSnapshot((SnapshotOffer) message);
 161             } else if (message instanceof ReplicatedLogEntry) {
 162                 onRecoveredJournalLogEntry((ReplicatedLogEntry) message);
 163             } else if (message instanceof ApplyLogEntries) {
 164                 // Handle this message for backwards compatibility with pre-Lithium versions.
 165                 onRecoveredApplyLogEntries(((ApplyLogEntries) message).getToIndex());
 166             } else if (message instanceof ApplyJournalEntries) {
 167                 onRecoveredApplyLogEntries(((ApplyJournalEntries) message).getToIndex());
 168             } else if (message instanceof DeleteEntries) {
 169                 replicatedLog.removeFrom(((DeleteEntries) message).getFromIndex());
 170             } else if (message instanceof UpdateElectionTerm) {
 171                 context.getTermInformation().update(((UpdateElectionTerm) message).getCurrentTerm(),
 172                         ((UpdateElectionTerm) message).getVotedFor());
 173             } else if (message instanceof RecoveryCompleted) {
 174                 onRecoveryCompletedMessage();
 175             }
 176         } else {
 177             if (message instanceof RecoveryCompleted) {
 178                 // Delete all the messages from the akka journal so that we do not end up with consistency issues
 179                 // Note I am not using the dataPersistenceProvider and directly using the akka api here
 180                 deleteMessages(lastSequenceNr());
 181
 182                 // Delete all the akka snapshots as they will not be needed
 183                 deleteSnapshots(new SnapshotSelectionCriteria(scala.Long.MaxValue(), scala.Long.MaxValue()));
 184
 185                 onRecoveryComplete();
 186
 187                 initializeBehavior();
 188             }
 189         }
 190     }
 191
 192     private void onRecoveredSnapshot(SnapshotOffer offer) {
 193         if(LOG.isDebugEnabled()) {
 194             LOG.debug("{}: SnapshotOffer called..", persistenceId());
 195         }
 196
 197         initRecoveryTimer();
 198
 199         Snapshot snapshot = (Snapshot) offer.snapshot();
 200
 201         // Create a replicated log with the snapshot information
 202         // The replicated log can be used later on to retrieve this snapshot
 203         // when we need to install it on a peer
 204         replicatedLog = new ReplicatedLogImpl(snapshot);
 205
 206         context.setReplicatedLog(replicatedLog);
 207         context.setLastApplied(snapshot.getLastAppliedIndex());
 208         context.setCommitIndex(snapshot.getLastAppliedIndex());
 209
 210         Stopwatch timer = Stopwatch.createStarted();
 211
 212         // Apply the snapshot to the actors state
 213         applyRecoverySnapshot(snapshot.getState());
 214
 215         timer.stop();
 216         LOG.info("Recovery snapshot applied for {} in {}: snapshotIndex={}, snapshotTerm={}, journal-size=" +
 217                 replicatedLog.size(), persistenceId(), timer.toString(),
 218                 replicatedLog.getSnapshotIndex(), replicatedLog.getSnapshotTerm());
 219     }
 220
 221     private void onRecoveredJournalLogEntry(ReplicatedLogEntry logEntry) {
 222         if(LOG.isDebugEnabled()) {
 223             LOG.debug("{}: Received ReplicatedLogEntry for recovery: {}", persistenceId(), logEntry.getIndex());
 224         }
 225
 226         replicatedLog.append(logEntry);
 227     }
 228
 229     private void onRecoveredApplyLogEntries(long toIndex) {
 230         if(LOG.isDebugEnabled()) {
 231             LOG.debug("{}: Received ApplyLogEntries for recovery, applying to state: {} to {}",
 232                     persistenceId(), context.getLastApplied() + 1, toIndex);
 233         }
 234
 235         for (long i = context.getLastApplied() + 1; i <= toIndex; i++) {
 236             batchRecoveredLogEntry(replicatedLog.get(i));
 237         }
 238
 239         context.setLastApplied(toIndex);
 240         context.setCommitIndex(toIndex);
 241     }
 242
 243     private void batchRecoveredLogEntry(ReplicatedLogEntry logEntry) {
 244         initRecoveryTimer();
 245
 246         int batchSize = context.getConfigParams().getJournalRecoveryLogBatchSize();
 247         if(currentRecoveryBatchCount == 0) {
 248             startLogRecoveryBatch(batchSize);
 249         }
 250
 251         appendRecoveredLogEntry(logEntry.getData());
 252
 253         if(++currentRecoveryBatchCount >= batchSize) {
 254             endCurrentLogRecoveryBatch();
 255         }
 256     }
 257
 258     private void endCurrentLogRecoveryBatch() {
 259         applyCurrentLogRecoveryBatch();
 260         currentRecoveryBatchCount = 0;
 261     }
 262
 263     private void onRecoveryCompletedMessage() {
 264         if(currentRecoveryBatchCount > 0) {
 265             endCurrentLogRecoveryBatch();
 266         }
 267
 268         onRecoveryComplete();
 269
 270         String recoveryTime = "";
 271         if(recoveryTimer != null) {
 272             recoveryTimer.stop();
 273             recoveryTime = " in " + recoveryTimer.toString();
 274             recoveryTimer = null;
 275         }
 276
 277         LOG.info(
 278             "Recovery completed" + recoveryTime + " - Switching actor to Follower - " +
 279                 "Persistence Id =  " + persistenceId() +
 280                 " Last index in log={}, snapshotIndex={}, snapshotTerm={}, " +
 281                 "journal-size={}",
 282             replicatedLog.lastIndex(), replicatedLog.getSnapshotIndex(),
 283             replicatedLog.getSnapshotTerm(), replicatedLog.size());
 284
 285         initializeBehavior();
 286     }
 287
 288     protected void initializeBehavior(){
 289         changeCurrentBehavior(new Follower(context));
 290     }
 291
 292     protected void changeCurrentBehavior(RaftActorBehavior newBehavior){
 293         RaftActorBehavior oldBehavior = currentBehavior;
 294         currentBehavior = newBehavior;
 295         handleBehaviorChange(oldBehavior, currentBehavior);
 296     }
 297
 298     @Override public void handleCommand(Object message) {
 299         if (message instanceof ApplyState){
 300             ApplyState applyState = (ApplyState) message;
 301
 302             long elapsedTime = (System.nanoTime() - applyState.getStartTime());
 303             if(elapsedTime >= APPLY_STATE_DELAY_THRESHOLD_IN_NANOS){
 304                 LOG.warn("ApplyState took more time than expected. Elapsed Time = {} ms ApplyState = {}",
 305                         TimeUnit.NANOSECONDS.toMillis(elapsedTime), applyState);
 306             }
 307
 308             if(LOG.isDebugEnabled()) {
 309                 LOG.debug("{}: Applying state for log index {} data {}",
 310                     persistenceId(), applyState.getReplicatedLogEntry().getIndex(),
 311                     applyState.getReplicatedLogEntry().getData());
 312             }
 313
 314             applyState(applyState.getClientActor(), applyState.getIdentifier(),
 315                 applyState.getReplicatedLogEntry().getData());
 316
 317         } else if (message instanceof ApplyJournalEntries){
 318             ApplyJournalEntries applyEntries = (ApplyJournalEntries) message;
 319             if(LOG.isDebugEnabled()) {
 320                 LOG.debug("{}: Persisting ApplyLogEntries with index={}", persistenceId(), applyEntries.getToIndex());
 321             }
 322             persistence().persist(applyEntries, new Procedure<ApplyJournalEntries>() {
 323                 @Override
 324                 public void apply(ApplyJournalEntries param) throws Exception {
 325                 }
 326             });
 327
 328         } else if(message instanceof ApplySnapshot ) {
 329             Snapshot snapshot = ((ApplySnapshot) message).getSnapshot();
 330
 331             if(LOG.isDebugEnabled()) {
 332                 LOG.debug("{}: ApplySnapshot called on Follower Actor " +
 333                         "snapshotIndex:{}, snapshotTerm:{}", persistenceId(), snapshot.getLastAppliedIndex(),
 334                     snapshot.getLastAppliedTerm()
 335                 );
 336             }
 337
 338             applySnapshot(snapshot.getState());
 339
 340             //clears the followers log, sets the snapshot index to ensure adjusted-index works
 341             replicatedLog = new ReplicatedLogImpl(snapshot);
 342             context.setReplicatedLog(replicatedLog);
 343             context.setLastApplied(snapshot.getLastAppliedIndex());
 344
 345         } else if (message instanceof FindLeader) {
 346             getSender().tell(
 347                 new FindLeaderReply(getLeaderAddress()),
 348                 getSelf()
 349             );
 350
 351         } else if (message instanceof SaveSnapshotSuccess) {
 352             SaveSnapshotSuccess success = (SaveSnapshotSuccess) message;
 353             LOG.info("{}: SaveSnapshotSuccess received for snapshot", persistenceId());
 354
 355             long sequenceNumber = success.metadata().sequenceNr();
 356
 357             commitSnapshot(sequenceNumber);
 358
 359         } else if (message instanceof SaveSnapshotFailure) {
 360             SaveSnapshotFailure saveSnapshotFailure = (SaveSnapshotFailure) message;
 361
 362             LOG.error("{}: SaveSnapshotFailure received for snapshot Cause:",
 363                     persistenceId(), saveSnapshotFailure.cause());
 364
 365             context.getReplicatedLog().snapshotRollback();
 366
 367             LOG.info("{}: Replicated Log rollbacked. Snapshot will be attempted in the next cycle." +
 368                 "snapshotIndex:{}, snapshotTerm:{}, log-size:{}", persistenceId(),
 369                 context.getReplicatedLog().getSnapshotIndex(),
 370                 context.getReplicatedLog().getSnapshotTerm(),
 371                 context.getReplicatedLog().size());
 372
 373         } else if (message instanceof CaptureSnapshot) {
 374             LOG.debug("{}: CaptureSnapshot received by actor: {}", persistenceId(), message);
 375
 376             if(captureSnapshot == null) {
 377                 captureSnapshot = (CaptureSnapshot)message;
 378                 createSnapshot();
 379             }
 380
 381         } else if (message instanceof CaptureSnapshotReply){
 382             handleCaptureSnapshotReply(((CaptureSnapshotReply) message).getSnapshot());
 383
 384         } else {
 385             RaftActorBehavior oldBehavior = currentBehavior;
 386             currentBehavior = currentBehavior.handleMessage(getSender(), message);
 387
 388             handleBehaviorChange(oldBehavior, currentBehavior);
 389         }
 390     }
 391
 392     private void handleBehaviorChange(RaftActorBehavior oldBehavior, RaftActorBehavior currentBehavior) {
 393         if (oldBehavior != currentBehavior){
 394             onStateChanged();
 395         }
 396
 397         String oldBehaviorLeaderId = oldBehavior == null? null : oldBehavior.getLeaderId();
 398         String oldBehaviorState = oldBehavior == null? null : oldBehavior.state().name();
 399
 400         // it can happen that the state has not changed but the leader has changed.
 401         onLeaderChanged(oldBehaviorLeaderId, currentBehavior.getLeaderId());
 402
 403         if (getRoleChangeNotifier().isPresent() &&
 404                 (oldBehavior == null || (oldBehavior.state() != currentBehavior.state()))) {
 405             getRoleChangeNotifier().get().tell(
 406                     new RoleChanged(getId(), oldBehaviorState , currentBehavior.state().name()),
 407                     getSelf());
 408         }
 409     }
 410
 411     /**
 412      * When a derived RaftActor needs to persist something it must call
 413      * persistData.
 414      *
 415      * @param clientActor
 416      * @param identifier
 417      * @param data
 418      */
 419     protected void persistData(final ActorRef clientActor, final String identifier,
 420         final Payload data) {
 421
 422         ReplicatedLogEntry replicatedLogEntry = new ReplicatedLogImplEntry(
 423             context.getReplicatedLog().lastIndex() + 1,
 424             context.getTermInformation().getCurrentTerm(), data);
 425
 426         if(LOG.isDebugEnabled()) {
 427             LOG.debug("{}: Persist data {}", persistenceId(), replicatedLogEntry);
 428         }
 429
 430         final RaftActorContext raftContext = getRaftActorContext();
 431
 432         replicatedLog
 433                 .appendAndPersist(replicatedLogEntry, new Procedure<ReplicatedLogEntry>() {
 434                     @Override
 435                     public void apply(ReplicatedLogEntry replicatedLogEntry) throws Exception {
 436                         if(!hasFollowers()){
 437                             // Increment the Commit Index and the Last Applied values
 438                             raftContext.setCommitIndex(replicatedLogEntry.getIndex());
 439                             raftContext.setLastApplied(replicatedLogEntry.getIndex());
 440
 441                             // Apply the state immediately
 442                             applyState(clientActor, identifier, data);
 443
 444                             // Send a ApplyJournalEntries message so that we write the fact that we applied
 445                             // the state to durable storage
 446                             self().tell(new ApplyJournalEntries(replicatedLogEntry.getIndex()), self());
 447
 448                             // Check if the "real" snapshot capture has been initiated. If no then do the fake snapshot
 449                             if(!context.isSnapshotCaptureInitiated()){
 450                                 raftContext.getReplicatedLog().snapshotPreCommit(raftContext.getLastApplied(),
 451                                         raftContext.getTermInformation().getCurrentTerm());
 452                                 raftContext.getReplicatedLog().snapshotCommit();
 453                             } else {
 454                                 LOG.debug("{}: Skipping fake snapshotting for {} because real snapshotting is in progress",
 455                                         persistenceId(), getId());
 456                             }
 457                         } else if (clientActor != null) {
 458                             // Send message for replication
 459                             currentBehavior.handleMessage(getSelf(),
 460                                     new Replicate(clientActor, identifier,
 461                                             replicatedLogEntry)
 462                             );
 463                         }
 464
 465                     }
 466                 });    }
 467
 468     protected String getId() {
 469         return context.getId();
 470     }
 471
 472     /**
 473      * Derived actors can call the isLeader method to check if the current
 474      * RaftActor is the Leader or not
 475      *
 476      * @return true it this RaftActor is a Leader false otherwise
 477      */
 478     protected boolean isLeader() {
 479         return context.getId().equals(currentBehavior.getLeaderId());
 480     }
 481
 482     /**
 483      * Derived actor can call getLeader if they need a reference to the Leader.
 484      * This would be useful for example in forwarding a request to an actor
 485      * which is the leader
 486      *
 487      * @return A reference to the leader if known, null otherwise
 488      */
 489     protected ActorSelection getLeader(){
 490         String leaderAddress = getLeaderAddress();
 491
 492         if(leaderAddress == null){
 493             return null;
 494         }
 495
 496         return context.actorSelection(leaderAddress);
 497     }
 498
 499     /**
 500      *
 501      * @return the current leader's id
 502      */
 503     protected String getLeaderId(){
 504         return currentBehavior.getLeaderId();
 505     }
 506
 507     protected RaftState getRaftState() {
 508         return currentBehavior.state();
 509     }
 510
 511     protected ReplicatedLogEntry getLastLogEntry() {
 512         return replicatedLog.last();
 513     }
 514
 515     protected Long getCurrentTerm(){
 516         return context.getTermInformation().getCurrentTerm();
 517     }
 518
 519     protected Long getCommitIndex(){
 520         return context.getCommitIndex();
 521     }
 522
 523     protected Long getLastApplied(){
 524         return context.getLastApplied();
 525     }
 526
 527     protected RaftActorContext getRaftActorContext() {
 528         return context;
 529     }
 530
 531     protected void updateConfigParams(ConfigParams configParams) {
 532         context.setConfigParams(configParams);
 533     }
 534
 535     /**
 536      * setPeerAddress sets the address of a known peer at a later time.
 537      * <p>
 538      * This is to account for situations where a we know that a peer
 539      * exists but we do not know an address up-front. This may also be used in
 540      * situations where a known peer starts off in a different location and we
 541      * need to change it's address
 542      * <p>
 543      * Note that if the peerId does not match the list of peers passed to
 544      * this actor during construction an IllegalStateException will be thrown.
 545      *
 546      * @param peerId
 547      * @param peerAddress
 548      */
 549     protected void setPeerAddress(String peerId, String peerAddress){
 550         context.setPeerAddress(peerId, peerAddress);
 551     }
 552
 553     protected void commitSnapshot(long sequenceNumber) {
 554         context.getReplicatedLog().snapshotCommit();
 555
 556         // TODO: Not sure if we want to be this aggressive with trimming stuff
 557         trimPersistentData(sequenceNumber);
 558     }
 559
 560     /**
 561      * The applyState method will be called by the RaftActor when some data
 562      * needs to be applied to the actor's state
 563      *
 564      * @param clientActor A reference to the client who sent this message. This
 565      *                    is the same reference that was passed to persistData
 566      *                    by the derived actor. clientActor may be null when
 567      *                    the RaftActor is behaving as a follower or during
 568      *                    recovery.
 569      * @param identifier  The identifier of the persisted data. This is also
 570      *                    the same identifier that was passed to persistData by
 571      *                    the derived actor. identifier may be null when
 572      *                    the RaftActor is behaving as a follower or during
 573      *                    recovery
 574      * @param data        A piece of data that was persisted by the persistData call.
 575      *                    This should NEVER be null.
 576      */
 577     protected abstract void applyState(ActorRef clientActor, String identifier,
 578         Object data);
 579
 580     /**
 581      * This method is called during recovery at the start of a batch of state entries. Derived
 582      * classes should perform any initialization needed to start a batch.
 583      */
 584     protected abstract void startLogRecoveryBatch(int maxBatchSize);
 585
 586     /**
 587      * This method is called during recovery to append state data to the current batch. This method
 588      * is called 1 or more times after {@link #startLogRecoveryBatch}.
 589      *
 590      * @param data the state data
 591      */
 592     protected abstract void appendRecoveredLogEntry(Payload data);
 593
 594     /**
 595      * This method is called during recovery to reconstruct the state of the actor.
 596      *
 597      * @param snapshotBytes A snapshot of the state of the actor
 598      */
 599     protected abstract void applyRecoverySnapshot(byte[] snapshotBytes);
 600
 601     /**
 602      * This method is called during recovery at the end of a batch to apply the current batched
 603      * log entries. This method is called after {@link #appendRecoveredLogEntry}.
 604      */
 605     protected abstract void applyCurrentLogRecoveryBatch();
 606
 607     /**
 608      * This method is called when recovery is complete.
 609      */
 610     protected abstract void onRecoveryComplete();
 611
 612     /**
 613      * This method will be called by the RaftActor when a snapshot needs to be
 614      * created. The derived actor should respond with its current state.
 615      * <p/>
 616      * During recovery the state that is returned by the derived actor will
 617      * be passed back to it by calling the applySnapshot  method
 618      *
 619      * @return The current state of the actor
 620      */
 621     protected abstract void createSnapshot();
 622
 623     /**
 624      * This method can be called at any other point during normal
 625      * operations when the derived actor is out of sync with it's peers
 626      * and the only way to bring it in sync is by applying a snapshot
 627      *
 628      * @param snapshotBytes A snapshot of the state of the actor
 629      */
 630     protected abstract void applySnapshot(byte[] snapshotBytes);
 631
 632     /**
 633      * This method will be called by the RaftActor when the state of the
 634      * RaftActor changes. The derived actor can then use methods like
 635      * isLeader or getLeader to do something useful
 636      */
 637     protected abstract void onStateChanged();
 638
 639     protected abstract DataPersistenceProvider persistence();
 640
 641     /**
 642      * Notifier Actor for this RaftActor to notify when a role change happens
 643      * @return ActorRef - ActorRef of the notifier or Optional.absent if none.
 644      */
 645     protected abstract Optional<ActorRef> getRoleChangeNotifier();
 646
 647     protected void onLeaderChanged(String oldLeader, String newLeader){};
 648
 649     private void trimPersistentData(long sequenceNumber) {
 650         // Trim akka snapshots
 651         // FIXME : Not sure how exactly the SnapshotSelectionCriteria is applied
 652         // For now guessing that it is ANDed.
 653         persistence().deleteSnapshots(new SnapshotSelectionCriteria(
 654             sequenceNumber - context.getConfigParams().getSnapshotBatchCount(), 43200000));
 655
 656         // Trim akka journal
 657         persistence().deleteMessages(sequenceNumber);
 658     }
 659
 660     private String getLeaderAddress(){
 661         if(isLeader()){
 662             return getSelf().path().toString();
 663         }
 664         String leaderId = currentBehavior.getLeaderId();
 665         if (leaderId == null) {
 666             return null;
 667         }
 668         String peerAddress = context.getPeerAddress(leaderId);
 669         if(LOG.isDebugEnabled()) {
 670             LOG.debug("{}: getLeaderAddress leaderId = {} peerAddress = {}",
 671                     persistenceId(), leaderId, peerAddress);
 672         }
 673
 674         return peerAddress;
 675     }
 676
 677     private void handleCaptureSnapshotReply(byte[] snapshotBytes) {
 678         LOG.debug("{}: CaptureSnapshotReply received by actor: snapshot size {}", persistenceId(), snapshotBytes.length);
 679
 680         // create a snapshot object from the state provided and save it
 681         // when snapshot is saved async, SaveSnapshotSuccess is raised.
 682
 683         Snapshot sn = Snapshot.create(snapshotBytes,
 684             context.getReplicatedLog().getFrom(captureSnapshot.getLastAppliedIndex() + 1),
 685             captureSnapshot.getLastIndex(), captureSnapshot.getLastTerm(),
 686             captureSnapshot.getLastAppliedIndex(), captureSnapshot.getLastAppliedTerm());
 687
 688         persistence().saveSnapshot(sn);
 689
 690         LOG.info("{}: Persisting of snapshot done:{}", persistenceId(), sn.getLogMessage());
 691
 692         long dataThreshold = Runtime.getRuntime().totalMemory() *
 693                 getRaftActorContext().getConfigParams().getSnapshotDataThresholdPercentage() / 100;
 694         if (context.getReplicatedLog().dataSize() > dataThreshold) {
 695
 696             if(LOG.isDebugEnabled()) {
 697                 LOG.debug("{}: dataSize {} exceeds dataThreshold {} - doing snapshotPreCommit with index {}",
 698                         persistenceId(), context.getReplicatedLog().dataSize(), dataThreshold,
 699                         captureSnapshot.getLastAppliedIndex());
 700             }
 701
 702             // if memory is less, clear the log based on lastApplied.
 703             // this could/should only happen if one of the followers is down
 704             // as normally we keep removing from the log when its replicated to all.
 705             context.getReplicatedLog().snapshotPreCommit(captureSnapshot.getLastAppliedIndex(),
 706                     captureSnapshot.getLastAppliedTerm());
 707
 708             // Don't reset replicatedToAllIndex to -1 as this may prevent us from trimming the log after an
 709             // install snapshot to a follower.
 710             if(captureSnapshot.getReplicatedToAllIndex() >= 0) {
 711                 getCurrentBehavior().setReplicatedToAllIndex(captureSnapshot.getReplicatedToAllIndex());
 712             }
 713         } else if(captureSnapshot.getReplicatedToAllIndex() != -1){
 714             // clear the log based on replicatedToAllIndex
 715             context.getReplicatedLog().snapshotPreCommit(captureSnapshot.getReplicatedToAllIndex(),
 716                     captureSnapshot.getReplicatedToAllTerm());
 717
 718             getCurrentBehavior().setReplicatedToAllIndex(captureSnapshot.getReplicatedToAllIndex());
 719         } else {
 720             // The replicatedToAllIndex was not found in the log
 721             // This means that replicatedToAllIndex never moved beyond -1 or that it is already in the snapshot.
 722             // In this scenario we may need to save the snapshot to the akka persistence
 723             // snapshot for recovery but we do not need to do the replicated log trimming.
 724             context.getReplicatedLog().snapshotPreCommit(replicatedLog.getSnapshotIndex(),
 725                     replicatedLog.getSnapshotTerm());
 726         }
 727
 728
 729         LOG.info("{}: Removed in-memory snapshotted entries, adjusted snaphsotIndex: {} " +
 730             "and term: {}", persistenceId(), replicatedLog.getSnapshotIndex(),
 731             replicatedLog.getSnapshotTerm());
 732
 733         if (isLeader() && captureSnapshot.isInstallSnapshotInitiated()) {
 734             // this would be call straight to the leader and won't initiate in serialization
 735             currentBehavior.handleMessage(getSelf(), new SendInstallSnapshot(
 736                     ByteString.copyFrom(snapshotBytes)));
 737         }
 738
 739         captureSnapshot = null;
 740         context.setSnapshotCaptureInitiated(false);
 741     }
 742
 743     protected boolean hasFollowers(){
 744         return getRaftActorContext().getPeerAddresses().keySet().size() > 0;
 745     }
 746
 747     private class ReplicatedLogImpl extends AbstractReplicatedLogImpl {
 748
 749         private static final int DATA_SIZE_DIVIDER = 5;
 750         private long dataSizeSinceLastSnapshot = 0;
 751
 752         public ReplicatedLogImpl(Snapshot snapshot) {
 753             super(snapshot.getLastAppliedIndex(), snapshot.getLastAppliedTerm(),
 754                 snapshot.getUnAppliedEntries());
 755         }
 756
 757         public ReplicatedLogImpl() {
 758             super();
 759         }
 760
 761         @Override public void removeFromAndPersist(long logEntryIndex) {
 762             int adjustedIndex = adjustedIndex(logEntryIndex);
 763
 764             if (adjustedIndex < 0) {
 765                 return;
 766             }
 767
 768             // FIXME: Maybe this should be done after the command is saved
 769             journal.subList(adjustedIndex , journal.size()).clear();
 770
 771             persistence().persist(new DeleteEntries(adjustedIndex), new Procedure<DeleteEntries>() {
 772
 773                 @Override
 774                 public void apply(DeleteEntries param)
 775                         throws Exception {
 776                     //FIXME : Doing nothing for now
 777                     dataSize = 0;
 778                     for (ReplicatedLogEntry entry : journal) {
 779                         dataSize += entry.size();
 780                     }
 781                 }
 782             });
 783         }
 784
 785         @Override public void appendAndPersist(
 786             final ReplicatedLogEntry replicatedLogEntry) {
 787             appendAndPersist(replicatedLogEntry, null);
 788         }
 789
 790         public void appendAndPersist(
 791             final ReplicatedLogEntry replicatedLogEntry,
 792             final Procedure<ReplicatedLogEntry> callback)  {
 793
 794             if(LOG.isDebugEnabled()) {
 795                 LOG.debug("{}: Append log entry and persist {} ", persistenceId(), replicatedLogEntry);
 796             }
 797
 798             // FIXME : By adding the replicated log entry to the in-memory journal we are not truly ensuring durability of the logs
 799             journal.add(replicatedLogEntry);
 800
 801             // When persisting events with persist it is guaranteed that the
 802             // persistent actor will not receive further commands between the
 803             // persist call and the execution(s) of the associated event
 804             // handler. This also holds for multiple persist calls in context
 805             // of a single command.
 806             persistence().persist(replicatedLogEntry,
 807                 new Procedure<ReplicatedLogEntry>() {
 808                     @Override
 809                     public void apply(ReplicatedLogEntry evt) throws Exception {
 810                         int logEntrySize = replicatedLogEntry.size();
 811
 812                         dataSize += logEntrySize;
 813                         long dataSizeForCheck = dataSize;
 814
 815                         dataSizeSinceLastSnapshot += logEntrySize;
 816                         long journalSize = lastIndex() + 1;
 817
 818                         if(!hasFollowers()) {
 819                             // When we do not have followers we do not maintain an in-memory log
 820                             // due to this the journalSize will never become anything close to the
 821                             // snapshot batch count. In fact will mostly be 1.
 822                             // Similarly since the journal's dataSize depends on the entries in the
 823                             // journal the journal's dataSize will never reach a value close to the
 824                             // memory threshold.
 825                             // By maintaining the dataSize outside the journal we are tracking essentially
 826                             // what we have written to the disk however since we no longer are in
 827                             // need of doing a snapshot just for the sake of freeing up memory we adjust
 828                             // the real size of data by the DATA_SIZE_DIVIDER so that we do not snapshot as often
 829                             // as if we were maintaining a real snapshot
 830                             dataSizeForCheck = dataSizeSinceLastSnapshot / DATA_SIZE_DIVIDER;
 831                         }
 832
 833                         long dataThreshold = Runtime.getRuntime().totalMemory() *
 834                                 getRaftActorContext().getConfigParams().getSnapshotDataThresholdPercentage() / 100;
 835
 836                         // when a snaphsot is being taken, captureSnapshot != null
 837                         if (!context.isSnapshotCaptureInitiated() &&
 838                                 ( journalSize % context.getConfigParams().getSnapshotBatchCount() == 0 ||
 839                                         dataSizeForCheck > dataThreshold)) {
 840
 841                             dataSizeSinceLastSnapshot = 0;
 842
 843                             LOG.info("{}: Initiating Snapshot Capture, journalSize = {}, dataSizeForCheck = {}," +
 844                                 " dataThreshold = {}", persistenceId(), journalSize, dataSizeForCheck, dataThreshold);
 845
 846                             long lastAppliedIndex = -1;
 847                             long lastAppliedTerm = -1;
 848
 849                             ReplicatedLogEntry lastAppliedEntry = get(context.getLastApplied());
 850                             if (!hasFollowers()) {
 851                                 lastAppliedIndex = replicatedLogEntry.getIndex();
 852                                 lastAppliedTerm = replicatedLogEntry.getTerm();
 853                             } else if (lastAppliedEntry != null) {
 854                                 lastAppliedIndex = lastAppliedEntry.getIndex();
 855                                 lastAppliedTerm = lastAppliedEntry.getTerm();
 856                             }
 857
 858                             if(LOG.isDebugEnabled()) {
 859                                 LOG.debug("{}: Snapshot Capture logSize: {}", persistenceId(), journal.size());
 860                                 LOG.debug("{}: Snapshot Capture lastApplied:{} ",
 861                                         persistenceId(), context.getLastApplied());
 862                                 LOG.debug("{}: Snapshot Capture lastAppliedIndex:{}", persistenceId(),
 863                                         lastAppliedIndex);
 864                                 LOG.debug("{}: Snapshot Capture lastAppliedTerm:{}", persistenceId(),
 865                                         lastAppliedTerm);
 866                             }
 867
 868                             // send a CaptureSnapshot to self to make the expensive operation async.
 869                             long replicatedToAllIndex = getCurrentBehavior().getReplicatedToAllIndex();
 870                             ReplicatedLogEntry replicatedToAllEntry = context.getReplicatedLog().get(replicatedToAllIndex);
 871                             getSelf().tell(new CaptureSnapshot(lastIndex(), lastTerm(), lastAppliedIndex, lastAppliedTerm,
 872                                 (replicatedToAllEntry != null ? replicatedToAllEntry.getIndex() : -1),
 873                                 (replicatedToAllEntry != null ? replicatedToAllEntry.getTerm() : -1)),
 874                                 null);
 875                             context.setSnapshotCaptureInitiated(true);
 876                         }
 877                         if (callback != null){
 878                             callback.apply(replicatedLogEntry);
 879                         }
 880                     }
 881                 }
 882             );
 883         }
 884
 885     }
 886
 887     static class DeleteEntries implements Serializable {
 888         private static final long serialVersionUID = 1L;
 889         private final int fromIndex;
 890
 891         public DeleteEntries(int fromIndex) {
 892             this.fromIndex = fromIndex;
 893         }
 894
 895         public int getFromIndex() {
 896             return fromIndex;
 897         }
 898     }
 899
 900
 901     private class ElectionTermImpl implements ElectionTerm {
 902         /**
 903          * Identifier of the actor whose election term information this is
 904          */
 905         private long currentTerm = 0;
 906         private String votedFor = null;
 907
 908         @Override
 909         public long getCurrentTerm() {
 910             return currentTerm;
 911         }
 912
 913         @Override
 914         public String getVotedFor() {
 915             return votedFor;
 916         }
 917
 918         @Override public void update(long currentTerm, String votedFor) {
 919             if(LOG.isDebugEnabled()) {
 920                 LOG.debug("{}: Set currentTerm={}, votedFor={}", persistenceId(), currentTerm, votedFor);
 921             }
 922             this.currentTerm = currentTerm;
 923             this.votedFor = votedFor;
 924         }
 925
 926         @Override
 927         public void updateAndPersist(long currentTerm, String votedFor){
 928             update(currentTerm, votedFor);
 929             // FIXME : Maybe first persist then update the state
 930             persistence().persist(new UpdateElectionTerm(this.currentTerm, this.votedFor), new Procedure<UpdateElectionTerm>(){
 931
 932                 @Override public void apply(UpdateElectionTerm param)
 933                     throws Exception {
 934
 935                 }
 936             });
 937         }
 938     }
 939
 940     static class UpdateElectionTerm implements Serializable {
 941         private static final long serialVersionUID = 1L;
 942         private final long currentTerm;
 943         private final String votedFor;
 944
 945         public UpdateElectionTerm(long currentTerm, String votedFor) {
 946             this.currentTerm = currentTerm;
 947             this.votedFor = votedFor;
 948         }
 949
 950         public long getCurrentTerm() {
 951             return currentTerm;
 952         }
 953
 954         public String getVotedFor() {
 955             return votedFor;
 956         }
 957     }
 958
 959     protected class NonPersistentRaftDataProvider extends NonPersistentDataProvider {
 960
 961         public NonPersistentRaftDataProvider(){
 962
 963         }
 964
 965         /**
 966          * The way snapshotting works is,
 967          * <ol>
 968          * <li> RaftActor calls createSnapshot on the Shard
 969          * <li> Shard sends a CaptureSnapshotReply and RaftActor then calls saveSnapshot
 970          * <li> When saveSnapshot is invoked on the akka-persistence API it uses the SnapshotStore to save the snapshot.
 971          * The SnapshotStore sends SaveSnapshotSuccess or SaveSnapshotFailure. When the RaftActor gets SaveSnapshot
 972          * success it commits the snapshot to the in-memory journal. This commitSnapshot is mimicking what is done
 973          * in SaveSnapshotSuccess.
 974          * </ol>
 975          * @param o
 976          */
 977         @Override
 978         public void saveSnapshot(Object o) {
 979             // Make saving Snapshot successful
 980             commitSnapshot(-1L);
 981         }
 982     }
 983
 984     @VisibleForTesting
 985     void setCurrentBehavior(AbstractRaftActorBehavior behavior) {
 986         currentBehavior = behavior;
 987     }
 988
 989     protected RaftActorBehavior getCurrentBehavior() {
 990         return currentBehavior;
 991     }
 992
 993 }