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.event.Logging;
  14 import akka.event.LoggingAdapter;
  15 import akka.japi.Procedure;
  16 import akka.persistence.RecoveryCompleted;
  17 import akka.persistence.SaveSnapshotFailure;
  18 import akka.persistence.SaveSnapshotSuccess;
  19 import akka.persistence.SnapshotOffer;
  20 import akka.persistence.SnapshotSelectionCriteria;
  21 import com.google.common.base.Optional;
  22 import com.google.common.base.Stopwatch;
  23 import com.google.protobuf.ByteString;
  24 import org.opendaylight.controller.cluster.DataPersistenceProvider;
  25 import org.opendaylight.controller.cluster.common.actor.AbstractUntypedPersistentActor;
  26 import org.opendaylight.controller.cluster.raft.base.messages.ApplyLogEntries;
  27 import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot;
  28 import org.opendaylight.controller.cluster.raft.base.messages.ApplyState;
  29 import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshot;
  30 import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshotReply;
  31 import org.opendaylight.controller.cluster.raft.base.messages.Replicate;
  32 import org.opendaylight.controller.cluster.raft.base.messages.SendHeartBeat;
  33 import org.opendaylight.controller.cluster.raft.behaviors.Follower;
  34 import org.opendaylight.controller.cluster.raft.behaviors.RaftActorBehavior;
  35 import org.opendaylight.controller.cluster.raft.client.messages.AddRaftPeer;
  36 import org.opendaylight.controller.cluster.raft.client.messages.FindLeader;
  37 import org.opendaylight.controller.cluster.raft.client.messages.FindLeaderReply;
  38 import org.opendaylight.controller.cluster.raft.client.messages.RemoveRaftPeer;
  39 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  40 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload;
  41 import org.opendaylight.controller.protobuff.messages.cluster.raft.AppendEntriesMessages;
  42
  43 import java.io.Serializable;
  44 import java.util.Map;
  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     protected final LoggingAdapter LOG =
  88         Logging.getLogger(getContext().system(), this);
  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 volatile boolean hasSnapshotCaptureInitiated = false;
 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 = new Stopwatch();
 132             recoveryTimer.start();
 133         }
 134     }
 135
 136     @Override
 137     public void preStart() throws Exception {
 138         LOG.info("Starting recovery for {} with journal batch size {}", persistenceId(),
 139                 context.getConfigParams().getJournalRecoveryLogBatchSize());
 140
 141         super.preStart();
 142     }
 143
 144     @Override
 145     public void handleRecover(Object message) {
 146         if(persistence().isRecoveryApplicable()) {
 147             if (message instanceof SnapshotOffer) {
 148                 onRecoveredSnapshot((SnapshotOffer) message);
 149             } else if (message instanceof ReplicatedLogEntry) {
 150                 onRecoveredJournalLogEntry((ReplicatedLogEntry) message);
 151             } else if (message instanceof ApplyLogEntries) {
 152                 onRecoveredApplyLogEntries((ApplyLogEntries) message);
 153             } else if (message instanceof DeleteEntries) {
 154                 replicatedLog.removeFrom(((DeleteEntries) message).getFromIndex());
 155             } else if (message instanceof UpdateElectionTerm) {
 156                 context.getTermInformation().update(((UpdateElectionTerm) message).getCurrentTerm(),
 157                         ((UpdateElectionTerm) message).getVotedFor());
 158             } else if (message instanceof RecoveryCompleted) {
 159                 onRecoveryCompletedMessage();
 160             }
 161         } else {
 162             if (message instanceof RecoveryCompleted) {
 163                 // Delete all the messages from the akka journal so that we do not end up with consistency issues
 164                 // Note I am not using the dataPersistenceProvider and directly using the akka api here
 165                 deleteMessages(lastSequenceNr());
 166
 167                 // Delete all the akka snapshots as they will not be needed
 168                 deleteSnapshots(new SnapshotSelectionCriteria(scala.Long.MaxValue(), scala.Long.MaxValue()));
 169
 170                 onRecoveryComplete();
 171                 currentBehavior = new Follower(context);
 172                 onStateChanged();
 173             }
 174         }
 175     }
 176
 177     private void onRecoveredSnapshot(SnapshotOffer offer) {
 178         if(LOG.isDebugEnabled()) {
 179             LOG.debug("SnapshotOffer called..");
 180         }
 181
 182         initRecoveryTimer();
 183
 184         Snapshot snapshot = (Snapshot) offer.snapshot();
 185
 186         // Create a replicated log with the snapshot information
 187         // The replicated log can be used later on to retrieve this snapshot
 188         // when we need to install it on a peer
 189         replicatedLog = new ReplicatedLogImpl(snapshot);
 190
 191         context.setReplicatedLog(replicatedLog);
 192         context.setLastApplied(snapshot.getLastAppliedIndex());
 193         context.setCommitIndex(snapshot.getLastAppliedIndex());
 194
 195         Stopwatch timer = new Stopwatch();
 196         timer.start();
 197
 198         // Apply the snapshot to the actors state
 199         applyRecoverySnapshot(ByteString.copyFrom(snapshot.getState()));
 200
 201         timer.stop();
 202         LOG.info("Recovery snapshot applied for {} in {}: snapshotIndex={}, snapshotTerm={}, journal-size=" +
 203                 replicatedLog.size(), persistenceId(), timer.toString(),
 204                 replicatedLog.snapshotIndex, replicatedLog.snapshotTerm);
 205     }
 206
 207     private void onRecoveredJournalLogEntry(ReplicatedLogEntry logEntry) {
 208         if(LOG.isDebugEnabled()) {
 209             LOG.debug("Received ReplicatedLogEntry for recovery: {}", logEntry.getIndex());
 210         }
 211
 212         replicatedLog.append(logEntry);
 213     }
 214
 215     private void onRecoveredApplyLogEntries(ApplyLogEntries ale) {
 216         if(LOG.isDebugEnabled()) {
 217             LOG.debug("Received ApplyLogEntries for recovery, applying to state: {} to {}",
 218                     context.getLastApplied() + 1, ale.getToIndex());
 219         }
 220
 221         for (long i = context.getLastApplied() + 1; i <= ale.getToIndex(); i++) {
 222             batchRecoveredLogEntry(replicatedLog.get(i));
 223         }
 224
 225         context.setLastApplied(ale.getToIndex());
 226         context.setCommitIndex(ale.getToIndex());
 227     }
 228
 229     private void batchRecoveredLogEntry(ReplicatedLogEntry logEntry) {
 230         initRecoveryTimer();
 231
 232         int batchSize = context.getConfigParams().getJournalRecoveryLogBatchSize();
 233         if(currentRecoveryBatchCount == 0) {
 234             startLogRecoveryBatch(batchSize);
 235         }
 236
 237         appendRecoveredLogEntry(logEntry.getData());
 238
 239         if(++currentRecoveryBatchCount >= batchSize) {
 240             endCurrentLogRecoveryBatch();
 241         }
 242     }
 243
 244     private void endCurrentLogRecoveryBatch() {
 245         applyCurrentLogRecoveryBatch();
 246         currentRecoveryBatchCount = 0;
 247     }
 248
 249     private void onRecoveryCompletedMessage() {
 250         if(currentRecoveryBatchCount > 0) {
 251             endCurrentLogRecoveryBatch();
 252         }
 253
 254         onRecoveryComplete();
 255
 256         String recoveryTime = "";
 257         if(recoveryTimer != null) {
 258             recoveryTimer.stop();
 259             recoveryTime = " in " + recoveryTimer.toString();
 260             recoveryTimer = null;
 261         }
 262
 263         LOG.info(
 264             "Recovery completed" + recoveryTime + " - Switching actor to Follower - " +
 265                 "Persistence Id =  " + persistenceId() +
 266                 " Last index in log={}, snapshotIndex={}, snapshotTerm={}, " +
 267                 "journal-size={}",
 268             replicatedLog.lastIndex(), replicatedLog.snapshotIndex,
 269             replicatedLog.snapshotTerm, replicatedLog.size());
 270
 271         currentBehavior = new Follower(context);
 272         onStateChanged();
 273     }
 274
 275     @Override public void handleCommand(Object message) {
 276         if (message instanceof ApplyState){
 277             ApplyState applyState = (ApplyState) message;
 278
 279             if(LOG.isDebugEnabled()) {
 280                 LOG.debug("Applying state for log index {} data {}",
 281                     applyState.getReplicatedLogEntry().getIndex(),
 282                     applyState.getReplicatedLogEntry().getData());
 283             }
 284
 285             applyState(applyState.getClientActor(), applyState.getIdentifier(),
 286                 applyState.getReplicatedLogEntry().getData());
 287
 288         } else if (message instanceof ApplyLogEntries){
 289             ApplyLogEntries ale = (ApplyLogEntries) message;
 290             if(LOG.isDebugEnabled()) {
 291                 LOG.debug("Persisting ApplyLogEntries with index={}", ale.getToIndex());
 292             }
 293             persistence().persist(new ApplyLogEntries(ale.getToIndex()), new Procedure<ApplyLogEntries>() {
 294                 @Override
 295                 public void apply(ApplyLogEntries param) throws Exception {
 296                 }
 297             });
 298
 299         } else if(message instanceof ApplySnapshot ) {
 300             Snapshot snapshot = ((ApplySnapshot) message).getSnapshot();
 301
 302             if(LOG.isDebugEnabled()) {
 303                 LOG.debug("ApplySnapshot called on Follower Actor " +
 304                         "snapshotIndex:{}, snapshotTerm:{}", snapshot.getLastAppliedIndex(),
 305                     snapshot.getLastAppliedTerm()
 306                 );
 307             }
 308             applySnapshot(ByteString.copyFrom(snapshot.getState()));
 309
 310             //clears the followers log, sets the snapshot index to ensure adjusted-index works
 311             replicatedLog = new ReplicatedLogImpl(snapshot);
 312             context.setReplicatedLog(replicatedLog);
 313             context.setLastApplied(snapshot.getLastAppliedIndex());
 314
 315         } else if (message instanceof FindLeader) {
 316             getSender().tell(
 317                 new FindLeaderReply(getLeaderAddress()),
 318                 getSelf()
 319             );
 320
 321         } else if (message instanceof SaveSnapshotSuccess) {
 322             SaveSnapshotSuccess success = (SaveSnapshotSuccess) message;
 323             LOG.info("SaveSnapshotSuccess received for snapshot");
 324
 325             long sequenceNumber = success.metadata().sequenceNr();
 326
 327             commitSnapshot(sequenceNumber);
 328
 329         } else if (message instanceof SaveSnapshotFailure) {
 330             SaveSnapshotFailure saveSnapshotFailure = (SaveSnapshotFailure) message;
 331
 332             LOG.info("saveSnapshotFailure.metadata():{}", saveSnapshotFailure.metadata().toString());
 333             LOG.error(saveSnapshotFailure.cause(), "SaveSnapshotFailure received for snapshot Cause:");
 334
 335             context.getReplicatedLog().snapshotRollback();
 336
 337             LOG.info("Replicated Log rollbacked. Snapshot will be attempted in the next cycle." +
 338                 "snapshotIndex:{}, snapshotTerm:{}, log-size:{}",
 339                 context.getReplicatedLog().getSnapshotIndex(),
 340                 context.getReplicatedLog().getSnapshotTerm(),
 341                 context.getReplicatedLog().size());
 342
 343         } else if (message instanceof AddRaftPeer){
 344
 345             // FIXME : Do not add raft peers like this.
 346             // When adding a new Peer we have to ensure that the a majority of
 347             // the peers know about the new Peer. Doing it this way may cause
 348             // a situation where multiple Leaders may emerge
 349             AddRaftPeer arp = (AddRaftPeer)message;
 350            context.addToPeers(arp.getName(), arp.getAddress());
 351
 352         } else if (message instanceof RemoveRaftPeer){
 353
 354             RemoveRaftPeer rrp = (RemoveRaftPeer)message;
 355             context.removePeer(rrp.getName());
 356
 357         } else if (message instanceof CaptureSnapshot) {
 358             LOG.info("CaptureSnapshot received by actor");
 359             CaptureSnapshot cs = (CaptureSnapshot)message;
 360             captureSnapshot = cs;
 361             createSnapshot();
 362
 363         } else if (message instanceof CaptureSnapshotReply){
 364             LOG.info("CaptureSnapshotReply received by actor");
 365             CaptureSnapshotReply csr = (CaptureSnapshotReply) message;
 366
 367             ByteString stateInBytes = csr.getSnapshot();
 368             LOG.info("CaptureSnapshotReply stateInBytes size:{}", stateInBytes.size());
 369             handleCaptureSnapshotReply(stateInBytes);
 370
 371         } else {
 372             if (!(message instanceof AppendEntriesMessages.AppendEntries)
 373                 && !(message instanceof AppendEntriesReply) && !(message instanceof SendHeartBeat)) {
 374                 if(LOG.isDebugEnabled()) {
 375                     LOG.debug("onReceiveCommand: message: {}", message.getClass());
 376                 }
 377             }
 378
 379             RaftActorBehavior oldBehavior = currentBehavior;
 380             currentBehavior = currentBehavior.handleMessage(getSender(), message);
 381
 382             if(oldBehavior != currentBehavior){
 383                 onStateChanged();
 384             }
 385
 386             onLeaderChanged(oldBehavior.getLeaderId(), currentBehavior.getLeaderId());
 387         }
 388     }
 389
 390     public java.util.Set<String> getPeers() {
 391         return context.getPeerAddresses().keySet();
 392     }
 393
 394     protected String getReplicatedLogState() {
 395         return "snapshotIndex=" + context.getReplicatedLog().getSnapshotIndex()
 396             + ", snapshotTerm=" + context.getReplicatedLog().getSnapshotTerm()
 397             + ", im-mem journal size=" + context.getReplicatedLog().size();
 398     }
 399
 400
 401     /**
 402      * When a derived RaftActor needs to persist something it must call
 403      * persistData.
 404      *
 405      * @param clientActor
 406      * @param identifier
 407      * @param data
 408      */
 409     protected void persistData(ActorRef clientActor, String identifier,
 410         Payload data) {
 411
 412         ReplicatedLogEntry replicatedLogEntry = new ReplicatedLogImplEntry(
 413             context.getReplicatedLog().lastIndex() + 1,
 414             context.getTermInformation().getCurrentTerm(), data);
 415
 416         if(LOG.isDebugEnabled()) {
 417             LOG.debug("Persist data {}", replicatedLogEntry);
 418         }
 419
 420         replicatedLog
 421             .appendAndPersist(clientActor, identifier, replicatedLogEntry);
 422     }
 423
 424     protected String getId() {
 425         return context.getId();
 426     }
 427
 428     /**
 429      * Derived actors can call the isLeader method to check if the current
 430      * RaftActor is the Leader or not
 431      *
 432      * @return true it this RaftActor is a Leader false otherwise
 433      */
 434     protected boolean isLeader() {
 435         return context.getId().equals(currentBehavior.getLeaderId());
 436     }
 437
 438     /**
 439      * Derived actor can call getLeader if they need a reference to the Leader.
 440      * This would be useful for example in forwarding a request to an actor
 441      * which is the leader
 442      *
 443      * @return A reference to the leader if known, null otherwise
 444      */
 445     protected ActorSelection getLeader(){
 446         String leaderAddress = getLeaderAddress();
 447
 448         if(leaderAddress == null){
 449             return null;
 450         }
 451
 452         return context.actorSelection(leaderAddress);
 453     }
 454
 455     /**
 456      *
 457      * @return the current leader's id
 458      */
 459     protected String getLeaderId(){
 460         return currentBehavior.getLeaderId();
 461     }
 462
 463     protected RaftState getRaftState() {
 464         return currentBehavior.state();
 465     }
 466
 467     protected ReplicatedLogEntry getLastLogEntry() {
 468         return replicatedLog.last();
 469     }
 470
 471     protected Long getCurrentTerm(){
 472         return context.getTermInformation().getCurrentTerm();
 473     }
 474
 475     protected Long getCommitIndex(){
 476         return context.getCommitIndex();
 477     }
 478
 479     protected Long getLastApplied(){
 480         return context.getLastApplied();
 481     }
 482
 483     protected RaftActorContext getRaftActorContext() {
 484         return context;
 485     }
 486
 487     /**
 488      * setPeerAddress sets the address of a known peer at a later time.
 489      * <p>
 490      * This is to account for situations where a we know that a peer
 491      * exists but we do not know an address up-front. This may also be used in
 492      * situations where a known peer starts off in a different location and we
 493      * need to change it's address
 494      * <p>
 495      * Note that if the peerId does not match the list of peers passed to
 496      * this actor during construction an IllegalStateException will be thrown.
 497      *
 498      * @param peerId
 499      * @param peerAddress
 500      */
 501     protected void setPeerAddress(String peerId, String peerAddress){
 502         context.setPeerAddress(peerId, peerAddress);
 503     }
 504
 505     protected void commitSnapshot(long sequenceNumber) {
 506         context.getReplicatedLog().snapshotCommit();
 507
 508         // TODO: Not sure if we want to be this aggressive with trimming stuff
 509         trimPersistentData(sequenceNumber);
 510     }
 511
 512     /**
 513      * The applyState method will be called by the RaftActor when some data
 514      * needs to be applied to the actor's state
 515      *
 516      * @param clientActor A reference to the client who sent this message. This
 517      *                    is the same reference that was passed to persistData
 518      *                    by the derived actor. clientActor may be null when
 519      *                    the RaftActor is behaving as a follower or during
 520      *                    recovery.
 521      * @param identifier  The identifier of the persisted data. This is also
 522      *                    the same identifier that was passed to persistData by
 523      *                    the derived actor. identifier may be null when
 524      *                    the RaftActor is behaving as a follower or during
 525      *                    recovery
 526      * @param data        A piece of data that was persisted by the persistData call.
 527      *                    This should NEVER be null.
 528      */
 529     protected abstract void applyState(ActorRef clientActor, String identifier,
 530         Object data);
 531
 532     /**
 533      * This method is called during recovery at the start of a batch of state entries. Derived
 534      * classes should perform any initialization needed to start a batch.
 535      */
 536     protected abstract void startLogRecoveryBatch(int maxBatchSize);
 537
 538     /**
 539      * This method is called during recovery to append state data to the current batch. This method
 540      * is called 1 or more times after {@link #startLogRecoveryBatch}.
 541      *
 542      * @param data the state data
 543      */
 544     protected abstract void appendRecoveredLogEntry(Payload data);
 545
 546     /**
 547      * This method is called during recovery to reconstruct the state of the actor.
 548      *
 549      * @param snapshot A snapshot of the state of the actor
 550      */
 551     protected abstract void applyRecoverySnapshot(ByteString snapshot);
 552
 553     /**
 554      * This method is called during recovery at the end of a batch to apply the current batched
 555      * log entries. This method is called after {@link #appendRecoveredLogEntry}.
 556      */
 557     protected abstract void applyCurrentLogRecoveryBatch();
 558
 559     /**
 560      * This method is called when recovery is complete.
 561      */
 562     protected abstract void onRecoveryComplete();
 563
 564     /**
 565      * This method will be called by the RaftActor when a snapshot needs to be
 566      * created. The derived actor should respond with its current state.
 567      * <p/>
 568      * During recovery the state that is returned by the derived actor will
 569      * be passed back to it by calling the applySnapshot  method
 570      *
 571      * @return The current state of the actor
 572      */
 573     protected abstract void createSnapshot();
 574
 575     /**
 576      * This method can be called at any other point during normal
 577      * operations when the derived actor is out of sync with it's peers
 578      * and the only way to bring it in sync is by applying a snapshot
 579      *
 580      * @param snapshot A snapshot of the state of the actor
 581      */
 582     protected abstract void applySnapshot(ByteString snapshot);
 583
 584     /**
 585      * This method will be called by the RaftActor when the state of the
 586      * RaftActor changes. The derived actor can then use methods like
 587      * isLeader or getLeader to do something useful
 588      */
 589     protected abstract void onStateChanged();
 590
 591     protected abstract DataPersistenceProvider persistence();
 592
 593     protected void onLeaderChanged(String oldLeader, String newLeader){};
 594
 595     private void trimPersistentData(long sequenceNumber) {
 596         // Trim akka snapshots
 597         // FIXME : Not sure how exactly the SnapshotSelectionCriteria is applied
 598         // For now guessing that it is ANDed.
 599         persistence().deleteSnapshots(new SnapshotSelectionCriteria(
 600             sequenceNumber - context.getConfigParams().getSnapshotBatchCount(), 43200000));
 601
 602         // Trim akka journal
 603         persistence().deleteMessages(sequenceNumber);
 604     }
 605
 606     private String getLeaderAddress(){
 607         if(isLeader()){
 608             return getSelf().path().toString();
 609         }
 610         String leaderId = currentBehavior.getLeaderId();
 611         if (leaderId == null) {
 612             return null;
 613         }
 614         String peerAddress = context.getPeerAddress(leaderId);
 615         if(LOG.isDebugEnabled()) {
 616             LOG.debug("getLeaderAddress leaderId = {} peerAddress = {}",
 617                     leaderId, peerAddress);
 618         }
 619
 620         return peerAddress;
 621     }
 622
 623     private void handleCaptureSnapshotReply(ByteString stateInBytes) {
 624         // create a snapshot object from the state provided and save it
 625         // when snapshot is saved async, SaveSnapshotSuccess is raised.
 626
 627         Snapshot sn = Snapshot.create(stateInBytes.toByteArray(),
 628             context.getReplicatedLog().getFrom(captureSnapshot.getLastAppliedIndex() + 1),
 629             captureSnapshot.getLastIndex(), captureSnapshot.getLastTerm(),
 630             captureSnapshot.getLastAppliedIndex(), captureSnapshot.getLastAppliedTerm());
 631
 632         persistence().saveSnapshot(sn);
 633
 634         LOG.info("Persisting of snapshot done:{}", sn.getLogMessage());
 635
 636         //be greedy and remove entries from in-mem journal which are in the snapshot
 637         // and update snapshotIndex and snapshotTerm without waiting for the success,
 638
 639         context.getReplicatedLog().snapshotPreCommit(stateInBytes,
 640             captureSnapshot.getLastAppliedIndex(),
 641             captureSnapshot.getLastAppliedTerm());
 642
 643         LOG.info("Removed in-memory snapshotted entries, adjusted snaphsotIndex:{} " +
 644             "and term:{}", captureSnapshot.getLastAppliedIndex(),
 645             captureSnapshot.getLastAppliedTerm());
 646
 647         captureSnapshot = null;
 648         hasSnapshotCaptureInitiated = false;
 649     }
 650
 651
 652     private class ReplicatedLogImpl extends AbstractReplicatedLogImpl {
 653
 654         public ReplicatedLogImpl(Snapshot snapshot) {
 655             super(ByteString.copyFrom(snapshot.getState()),
 656                 snapshot.getLastAppliedIndex(), snapshot.getLastAppliedTerm(),
 657                 snapshot.getUnAppliedEntries());
 658         }
 659
 660         public ReplicatedLogImpl() {
 661             super();
 662         }
 663
 664         @Override public void removeFromAndPersist(long logEntryIndex) {
 665             int adjustedIndex = adjustedIndex(logEntryIndex);
 666
 667             if (adjustedIndex < 0) {
 668                 return;
 669             }
 670
 671             // FIXME: Maybe this should be done after the command is saved
 672             journal.subList(adjustedIndex , journal.size()).clear();
 673
 674             persistence().persist(new DeleteEntries(adjustedIndex), new Procedure<DeleteEntries>(){
 675
 676                 @Override public void apply(DeleteEntries param)
 677                     throws Exception {
 678                     //FIXME : Doing nothing for now
 679                 }
 680             });
 681         }
 682
 683         @Override public void appendAndPersist(
 684             final ReplicatedLogEntry replicatedLogEntry) {
 685             appendAndPersist(null, null, replicatedLogEntry);
 686         }
 687
 688         public void appendAndPersist(final ActorRef clientActor,
 689             final String identifier,
 690             final ReplicatedLogEntry replicatedLogEntry) {
 691
 692             if(LOG.isDebugEnabled()) {
 693                 LOG.debug("Append log entry and persist {} ", replicatedLogEntry);
 694             }
 695
 696             // FIXME : By adding the replicated log entry to the in-memory journal we are not truly ensuring durability of the logs
 697             journal.add(replicatedLogEntry);
 698
 699             // When persisting events with persist it is guaranteed that the
 700             // persistent actor will not receive further commands between the
 701             // persist call and the execution(s) of the associated event
 702             // handler. This also holds for multiple persist calls in context
 703             // of a single command.
 704             persistence().persist(replicatedLogEntry,
 705                 new Procedure<ReplicatedLogEntry>() {
 706                     @Override
 707                     public void apply(ReplicatedLogEntry evt) throws Exception {
 708                         // when a snaphsot is being taken, captureSnapshot != null
 709                         if (hasSnapshotCaptureInitiated == false &&
 710                             journal.size() % context.getConfigParams().getSnapshotBatchCount() == 0) {
 711
 712                             LOG.info("Initiating Snapshot Capture..");
 713                             long lastAppliedIndex = -1;
 714                             long lastAppliedTerm = -1;
 715
 716                             ReplicatedLogEntry lastAppliedEntry = get(context.getLastApplied());
 717                             if (lastAppliedEntry != null) {
 718                                 lastAppliedIndex = lastAppliedEntry.getIndex();
 719                                 lastAppliedTerm = lastAppliedEntry.getTerm();
 720                             }
 721
 722                             if(LOG.isDebugEnabled()) {
 723                                 LOG.debug("Snapshot Capture logSize: {}", journal.size());
 724                                 LOG.debug("Snapshot Capture lastApplied:{} ",
 725                                     context.getLastApplied());
 726                                 LOG.debug("Snapshot Capture lastAppliedIndex:{}", lastAppliedIndex);
 727                                 LOG.debug("Snapshot Capture lastAppliedTerm:{}", lastAppliedTerm);
 728                             }
 729
 730                             // send a CaptureSnapshot to self to make the expensive operation async.
 731                             getSelf().tell(new CaptureSnapshot(
 732                                 lastIndex(), lastTerm(), lastAppliedIndex, lastAppliedTerm),
 733                                 null);
 734                             hasSnapshotCaptureInitiated = true;
 735                         }
 736                         // Send message for replication
 737                         if (clientActor != null) {
 738                             currentBehavior.handleMessage(getSelf(),
 739                                 new Replicate(clientActor, identifier,
 740                                     replicatedLogEntry)
 741                             );
 742                         }
 743                     }
 744                 }
 745             );
 746         }
 747
 748     }
 749
 750     static class DeleteEntries implements Serializable {
 751         private final int fromIndex;
 752
 753
 754         public DeleteEntries(int fromIndex) {
 755             this.fromIndex = fromIndex;
 756         }
 757
 758         public int getFromIndex() {
 759             return fromIndex;
 760         }
 761     }
 762
 763
 764     private class ElectionTermImpl implements ElectionTerm {
 765         /**
 766          * Identifier of the actor whose election term information this is
 767          */
 768         private long currentTerm = 0;
 769         private String votedFor = null;
 770
 771         @Override
 772         public long getCurrentTerm() {
 773             return currentTerm;
 774         }
 775
 776         @Override
 777         public String getVotedFor() {
 778             return votedFor;
 779         }
 780
 781         @Override public void update(long currentTerm, String votedFor) {
 782             if(LOG.isDebugEnabled()) {
 783                 LOG.debug("Set currentTerm={}, votedFor={}", currentTerm, votedFor);
 784             }
 785             this.currentTerm = currentTerm;
 786             this.votedFor = votedFor;
 787         }
 788
 789         @Override
 790         public void updateAndPersist(long currentTerm, String votedFor){
 791             update(currentTerm, votedFor);
 792             // FIXME : Maybe first persist then update the state
 793             persistence().persist(new UpdateElectionTerm(this.currentTerm, this.votedFor), new Procedure<UpdateElectionTerm>(){
 794
 795                 @Override public void apply(UpdateElectionTerm param)
 796                     throws Exception {
 797
 798                 }
 799             });
 800         }
 801     }
 802
 803     static class UpdateElectionTerm implements Serializable {
 804         private final long currentTerm;
 805         private final String votedFor;
 806
 807         public UpdateElectionTerm(long currentTerm, String votedFor) {
 808             this.currentTerm = currentTerm;
 809             this.votedFor = votedFor;
 810         }
 811
 812         public long getCurrentTerm() {
 813             return currentTerm;
 814         }
 815
 816         public String getVotedFor() {
 817             return votedFor;
 818         }
 819     }
 820
 821     protected class NonPersistentRaftDataProvider extends NonPersistentDataProvider {
 822
 823         public NonPersistentRaftDataProvider(){
 824
 825         }
 826
 827         /**
 828          * The way snapshotting works is,
 829          * <ol>
 830          * <li> RaftActor calls createSnapshot on the Shard
 831          * <li> Shard sends a CaptureSnapshotReply and RaftActor then calls saveSnapshot
 832          * <li> When saveSnapshot is invoked on the akka-persistence API it uses the SnapshotStore to save the snapshot.
 833          * The SnapshotStore sends SaveSnapshotSuccess or SaveSnapshotFailure. When the RaftActor gets SaveSnapshot
 834          * success it commits the snapshot to the in-memory journal. This commitSnapshot is mimicking what is done
 835          * in SaveSnapshotSuccess.
 836          * </ol>
 837          * @param o
 838          */
 839         @Override
 840         public void saveSnapshot(Object o) {
 841             // Make saving Snapshot successful
 842             commitSnapshot(-1L);
 843         }
 844     }
 845
 846 }