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