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  * Copyright (c) 2015 Brocade Communications Systems, Inc. and others.  All rights reserved.
   4  *
   5  * This program and the accompanying materials are made available under the
   6  * terms of the Eclipse Public License v1.0 which accompanies this distribution,
   7  * and is available at http://www.eclipse.org/legal/epl-v10.html
   8  */
   9
  10 package org.opendaylight.controller.cluster.raft;
  11
  12 import akka.actor.ActorRef;
  13 import akka.actor.ActorSelection;
  14 import akka.japi.Procedure;
  15 import com.google.common.annotations.VisibleForTesting;
  16 import com.google.common.base.Objects;
  17 import com.google.common.base.Optional;
  18 import com.google.common.base.Supplier;
  19 import com.google.common.collect.Lists;
  20 import java.io.Serializable;
  21 import java.util.Collection;
  22 import java.util.HashMap;
  23 import java.util.List;
  24 import java.util.Map;
  25 import java.util.concurrent.TimeUnit;
  26 import javax.annotation.Nonnull;
  27 import org.apache.commons.lang3.time.DurationFormatUtils;
  28 import org.opendaylight.controller.cluster.DataPersistenceProvider;
  29 import org.opendaylight.controller.cluster.DelegatingPersistentDataProvider;
  30 import org.opendaylight.controller.cluster.NonPersistentDataProvider;
  31 import org.opendaylight.controller.cluster.PersistentDataProvider;
  32 import org.opendaylight.controller.cluster.common.actor.AbstractUntypedPersistentActor;
  33 import org.opendaylight.controller.cluster.notifications.LeaderStateChanged;
  34 import org.opendaylight.controller.cluster.notifications.RoleChanged;
  35 import org.opendaylight.controller.cluster.raft.base.messages.ApplyJournalEntries;
  36 import org.opendaylight.controller.cluster.raft.base.messages.ApplyState;
  37 import org.opendaylight.controller.cluster.raft.base.messages.InitiateCaptureSnapshot;
  38 import org.opendaylight.controller.cluster.raft.base.messages.LeaderTransitioning;
  39 import org.opendaylight.controller.cluster.raft.base.messages.Replicate;
  40 import org.opendaylight.controller.cluster.raft.base.messages.SwitchBehavior;
  41 import org.opendaylight.controller.cluster.raft.behaviors.AbstractLeader;
  42 import org.opendaylight.controller.cluster.raft.behaviors.DelegatingRaftActorBehavior;
  43 import org.opendaylight.controller.cluster.raft.behaviors.Follower;
  44 import org.opendaylight.controller.cluster.raft.behaviors.RaftActorBehavior;
  45 import org.opendaylight.controller.cluster.raft.client.messages.FindLeader;
  46 import org.opendaylight.controller.cluster.raft.client.messages.FindLeaderReply;
  47 import org.opendaylight.controller.cluster.raft.client.messages.FollowerInfo;
  48 import org.opendaylight.controller.cluster.raft.client.messages.GetOnDemandRaftState;
  49 import org.opendaylight.controller.cluster.raft.client.messages.OnDemandRaftState;
  50 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload;
  51 import org.slf4j.Logger;
  52 import org.slf4j.LoggerFactory;
  53
  54 /**
  55  * RaftActor encapsulates a state machine that needs to be kept synchronized
  56  * in a cluster. It implements the RAFT algorithm as described in the paper
  57  * <a href='https://ramcloud.stanford.edu/wiki/download/attachments/11370504/raft.pdf'>
  58  * In Search of an Understandable Consensus Algorithm</a>
  59  * <p/>
  60  * RaftActor has 3 states and each state has a certain behavior associated
  61  * with it. A Raft actor can behave as,
  62  * <ul>
  63  * <li> A Leader </li>
  64  * <li> A Follower (or) </li>
  65  * <li> A Candidate </li>
  66  * </ul>
  67  * <p/>
  68  * <p/>
  69  * A RaftActor MUST be a Leader in order to accept requests from clients to
  70  * change the state of it's encapsulated state machine. Once a RaftActor becomes
  71  * a Leader it is also responsible for ensuring that all followers ultimately
  72  * have the same log and therefore the same state machine as itself.
  73  * <p/>
  74  * <p/>
  75  * The current behavior of a RaftActor determines how election for leadership
  76  * is initiated and how peer RaftActors react to request for votes.
  77  * <p/>
  78  * <p/>
  79  * Each RaftActor also needs to know the current election term. It uses this
  80  * information for a couple of things. One is to simply figure out who it
  81  * voted for in the last election. Another is to figure out if the message
  82  * it received to update it's state is stale.
  83  * <p/>
  84  * <p/>
  85  * The RaftActor uses akka-persistence to store it's replicated log.
  86  * Furthermore through it's behaviors a Raft Actor determines
  87  * <p/>
  88  * <ul>
  89  * <li> when a log entry should be persisted </li>
  90  * <li> when a log entry should be applied to the state machine (and) </li>
  91  * <li> when a snapshot should be saved </li>
  92  * </ul>
  93  */
  94 public abstract class RaftActor extends AbstractUntypedPersistentActor {
  95
  96     private static final long APPLY_STATE_DELAY_THRESHOLD_IN_NANOS = TimeUnit.MILLISECONDS.toNanos(50L); // 50 millis
  97
  98     protected final Logger LOG = LoggerFactory.getLogger(getClass());
  99
 100     /**
 101      * The current state determines the current behavior of a RaftActor
 102      * A Raft Actor always starts off in the Follower State
 103      */
 104     private final DelegatingRaftActorBehavior currentBehavior = new DelegatingRaftActorBehavior();
 105
 106     /**
 107      * This context should NOT be passed directly to any other actor it is
 108      * only to be consumed by the RaftActorBehaviors
 109      */
 110     private final RaftActorContextImpl context;
 111
 112     private final DelegatingPersistentDataProvider delegatingPersistenceProvider;
 113
 114     private final PersistentDataProvider persistentProvider;
 115
 116     private RaftActorRecoverySupport raftRecovery;
 117
 118     private RaftActorSnapshotMessageSupport snapshotSupport;
 119
 120     private final BehaviorStateHolder reusableBehaviorStateHolder = new BehaviorStateHolder();
 121
 122     private final SwitchBehaviorSupplier reusableSwitchBehaviorSupplier = new SwitchBehaviorSupplier();
 123
 124     private RaftActorServerConfigurationSupport serverConfigurationSupport;
 125
 126     public RaftActor(String id, Map<String, String> peerAddresses,
 127          Optional<ConfigParams> configParams, short payloadVersion) {
 128
 129         persistentProvider = new PersistentDataProvider(this);
 130         delegatingPersistenceProvider = new RaftActorDelegatingPersistentDataProvider(null, persistentProvider);
 131
 132         context = new RaftActorContextImpl(this.getSelf(),
 133             this.getContext(), id, new ElectionTermImpl(persistentProvider, id, LOG),
 134             -1, -1, peerAddresses,
 135             (configParams.isPresent() ? configParams.get(): new DefaultConfigParamsImpl()),
 136             delegatingPersistenceProvider, LOG);
 137
 138         context.setPayloadVersion(payloadVersion);
 139         context.setReplicatedLog(ReplicatedLogImpl.newInstance(context, currentBehavior));
 140     }
 141
 142     @Override
 143     public void preStart() throws Exception {
 144         LOG.info("Starting recovery for {} with journal batch size {}", persistenceId(),
 145                 context.getConfigParams().getJournalRecoveryLogBatchSize());
 146
 147         super.preStart();
 148
 149         snapshotSupport = newRaftActorSnapshotMessageSupport();
 150         serverConfigurationSupport = new RaftActorServerConfigurationSupport(this);
 151     }
 152
 153     @Override
 154     public void postStop() {
 155         if(currentBehavior.getDelegate() != null) {
 156             try {
 157                 currentBehavior.close();
 158             } catch (Exception e) {
 159                 LOG.debug("{}: Error closing behavior {}", persistenceId(), currentBehavior.state());
 160             }
 161         }
 162
 163         super.postStop();
 164     }
 165
 166     @Override
 167     public void handleRecover(Object message) {
 168         if(raftRecovery == null) {
 169             raftRecovery = newRaftActorRecoverySupport();
 170         }
 171
 172         boolean recoveryComplete = raftRecovery.handleRecoveryMessage(message, persistentProvider);
 173         if(recoveryComplete) {
 174             onRecoveryComplete();
 175
 176             initializeBehavior();
 177
 178             raftRecovery = null;
 179
 180             if (context.getReplicatedLog().size() > 0) {
 181                 self().tell(new InitiateCaptureSnapshot(), self());
 182                 LOG.info("{}: Snapshot capture initiated after recovery", persistenceId());
 183             } else {
 184                 LOG.info("{}: Snapshot capture NOT initiated after recovery, journal empty", persistenceId());
 185             }
 186         }
 187     }
 188
 189     protected RaftActorRecoverySupport newRaftActorRecoverySupport() {
 190         return new RaftActorRecoverySupport(context, currentBehavior, getRaftActorRecoveryCohort());
 191     }
 192
 193     protected void initializeBehavior(){
 194         changeCurrentBehavior(new Follower(context));
 195     }
 196
 197     protected void changeCurrentBehavior(RaftActorBehavior newBehavior){
 198         reusableBehaviorStateHolder.init(getCurrentBehavior());
 199         setCurrentBehavior(newBehavior);
 200         handleBehaviorChange(reusableBehaviorStateHolder, getCurrentBehavior());
 201     }
 202
 203     @Override
 204     public void handleCommand(final Object message) {
 205         if(serverConfigurationSupport.handleMessage(message, getSender())) {
 206             return;
 207         } else if (message instanceof ApplyState){
 208             ApplyState applyState = (ApplyState) message;
 209
 210             long elapsedTime = (System.nanoTime() - applyState.getStartTime());
 211             if(elapsedTime >= APPLY_STATE_DELAY_THRESHOLD_IN_NANOS){
 212                 LOG.warn("ApplyState took more time than expected. Elapsed Time = {} ms ApplyState = {}",
 213                         TimeUnit.NANOSECONDS.toMillis(elapsedTime), applyState);
 214             }
 215
 216             if(LOG.isDebugEnabled()) {
 217                 LOG.debug("{}: Applying state for log index {} data {}",
 218                     persistenceId(), applyState.getReplicatedLogEntry().getIndex(),
 219                     applyState.getReplicatedLogEntry().getData());
 220             }
 221
 222             applyState(applyState.getClientActor(), applyState.getIdentifier(),
 223                 applyState.getReplicatedLogEntry().getData());
 224
 225             if (!hasFollowers()) {
 226                 // for single node, the capture should happen after the apply state
 227                 // as we delete messages from the persistent journal which have made it to the snapshot
 228                 // capturing the snapshot before applying makes the persistent journal and snapshot out of sync
 229                 // and recovery shows data missing
 230                 context.getReplicatedLog().captureSnapshotIfReady(applyState.getReplicatedLogEntry());
 231
 232                 context.getSnapshotManager().trimLog(context.getLastApplied(), currentBehavior);
 233             }
 234
 235         } else if (message instanceof ApplyJournalEntries){
 236             ApplyJournalEntries applyEntries = (ApplyJournalEntries) message;
 237             if(LOG.isDebugEnabled()) {
 238                 LOG.debug("{}: Persisting ApplyLogEntries with index={}", persistenceId(), applyEntries.getToIndex());
 239             }
 240
 241             persistence().persist(applyEntries, NoopProcedure.instance());
 242
 243         } else if (message instanceof FindLeader) {
 244             getSender().tell(
 245                 new FindLeaderReply(getLeaderAddress()),
 246                 getSelf()
 247             );
 248         } else if(message instanceof GetOnDemandRaftState) {
 249             onGetOnDemandRaftStats();
 250         } else if(message instanceof InitiateCaptureSnapshot) {
 251             captureSnapshot();
 252         } else if(message instanceof SwitchBehavior){
 253             switchBehavior(((SwitchBehavior) message));
 254         } else if(message instanceof LeaderTransitioning) {
 255             onLeaderTransitioning();
 256         } else if(!snapshotSupport.handleSnapshotMessage(message, getSender())) {
 257             switchBehavior(reusableSwitchBehaviorSupplier.handleMessage(getSender(), message));
 258         }
 259     }
 260
 261     private void onLeaderTransitioning() {
 262         LOG.debug("{}: onLeaderTransitioning", persistenceId());
 263         Optional<ActorRef> roleChangeNotifier = getRoleChangeNotifier();
 264         if(currentBehavior.state() == RaftState.Follower && roleChangeNotifier.isPresent()) {
 265             roleChangeNotifier.get().tell(newLeaderStateChanged(getId(), null,
 266                     currentBehavior.getLeaderPayloadVersion()), getSelf());
 267         }
 268     }
 269
 270     private void switchBehavior(SwitchBehavior message) {
 271         if(!getRaftActorContext().getRaftPolicy().automaticElectionsEnabled()) {
 272             RaftState newState = message.getNewState();
 273             if( newState == RaftState.Leader || newState == RaftState.Follower) {
 274                 switchBehavior(reusableSwitchBehaviorSupplier.handleMessage(getSender(), message));
 275                 getRaftActorContext().getTermInformation().updateAndPersist(message.getNewTerm(), "");
 276             } else {
 277                 LOG.warn("Switching to behavior : {} - not supported", newState);
 278             }
 279         }
 280     }
 281
 282     private void switchBehavior(Supplier<RaftActorBehavior> supplier){
 283         reusableBehaviorStateHolder.init(getCurrentBehavior());
 284
 285         setCurrentBehavior(supplier.get());
 286
 287         handleBehaviorChange(reusableBehaviorStateHolder, getCurrentBehavior());
 288     }
 289
 290     protected RaftActorSnapshotMessageSupport newRaftActorSnapshotMessageSupport() {
 291         return new RaftActorSnapshotMessageSupport(context, currentBehavior,
 292                 getRaftActorSnapshotCohort());
 293     }
 294
 295     private void onGetOnDemandRaftStats() {
 296         // Debugging message to retrieve raft stats.
 297
 298         Map<String, String> peerAddresses = new HashMap<>();
 299         for(String peerId: context.getPeerIds()) {
 300             peerAddresses.put(peerId, context.getPeerAddress(peerId));
 301         }
 302
 303         OnDemandRaftState.Builder builder = OnDemandRaftState.builder()
 304                 .commitIndex(context.getCommitIndex())
 305                 .currentTerm(context.getTermInformation().getCurrentTerm())
 306                 .inMemoryJournalDataSize(replicatedLog().dataSize())
 307                 .inMemoryJournalLogSize(replicatedLog().size())
 308                 .isSnapshotCaptureInitiated(context.getSnapshotManager().isCapturing())
 309                 .lastApplied(context.getLastApplied())
 310                 .lastIndex(replicatedLog().lastIndex())
 311                 .lastTerm(replicatedLog().lastTerm())
 312                 .leader(getLeaderId())
 313                 .raftState(currentBehavior.state().toString())
 314                 .replicatedToAllIndex(currentBehavior.getReplicatedToAllIndex())
 315                 .snapshotIndex(replicatedLog().getSnapshotIndex())
 316                 .snapshotTerm(replicatedLog().getSnapshotTerm())
 317                 .votedFor(context.getTermInformation().getVotedFor())
 318                 .peerAddresses(peerAddresses)
 319                 .customRaftPolicyClassName(context.getConfigParams().getCustomRaftPolicyImplementationClass());
 320
 321         ReplicatedLogEntry lastLogEntry = getLastLogEntry();
 322         if (lastLogEntry != null) {
 323             builder.lastLogIndex(lastLogEntry.getIndex());
 324             builder.lastLogTerm(lastLogEntry.getTerm());
 325         }
 326
 327         if(getCurrentBehavior() instanceof AbstractLeader) {
 328             AbstractLeader leader = (AbstractLeader)getCurrentBehavior();
 329             Collection<String> followerIds = leader.getFollowerIds();
 330             List<FollowerInfo> followerInfoList = Lists.newArrayListWithCapacity(followerIds.size());
 331             for(String id: followerIds) {
 332                 final FollowerLogInformation info = leader.getFollower(id);
 333                 followerInfoList.add(new FollowerInfo(id, info.getNextIndex(), info.getMatchIndex(),
 334                         info.isFollowerActive(), DurationFormatUtils.formatDurationHMS(info.timeSinceLastActivity())));
 335             }
 336
 337             builder.followerInfoList(followerInfoList);
 338         }
 339
 340         sender().tell(builder.build(), self());
 341
 342     }
 343
 344     private void handleBehaviorChange(BehaviorStateHolder oldBehaviorState, RaftActorBehavior currentBehavior) {
 345         RaftActorBehavior oldBehavior = oldBehaviorState.getBehavior();
 346
 347         if (oldBehavior != currentBehavior){
 348             onStateChanged();
 349         }
 350
 351         String lastValidLeaderId = oldBehavior == null ? null : oldBehaviorState.getLastValidLeaderId();
 352         String oldBehaviorStateName = oldBehavior == null ? null : oldBehavior.state().name();
 353
 354         // it can happen that the state has not changed but the leader has changed.
 355         Optional<ActorRef> roleChangeNotifier = getRoleChangeNotifier();
 356         if(!Objects.equal(lastValidLeaderId, currentBehavior.getLeaderId()) ||
 357            oldBehaviorState.getLeaderPayloadVersion() != currentBehavior.getLeaderPayloadVersion()) {
 358             if(roleChangeNotifier.isPresent()) {
 359                 roleChangeNotifier.get().tell(newLeaderStateChanged(getId(), currentBehavior.getLeaderId(),
 360                         currentBehavior.getLeaderPayloadVersion()), getSelf());
 361             }
 362
 363             onLeaderChanged(lastValidLeaderId, currentBehavior.getLeaderId());
 364         }
 365
 366         if (roleChangeNotifier.isPresent() &&
 367                 (oldBehavior == null || (oldBehavior.state() != currentBehavior.state()))) {
 368             roleChangeNotifier.get().tell(new RoleChanged(getId(), oldBehaviorStateName ,
 369                     currentBehavior.state().name()), getSelf());
 370         }
 371     }
 372
 373     protected LeaderStateChanged newLeaderStateChanged(String memberId, String leaderId, short leaderPayloadVersion) {
 374         return new LeaderStateChanged(memberId, leaderId, leaderPayloadVersion);
 375     }
 376
 377     @Override
 378     public long snapshotSequenceNr() {
 379         // When we do a snapshot capture, we also capture and save the sequence-number of the persistent journal,
 380         // so that we can delete the persistent journal based on the saved sequence-number
 381         // However , when akka replays the journal during recovery, it replays it from the sequence number when the snapshot
 382         // was saved and not the number we saved.
 383         // We would want to override it , by asking akka to use the last-sequence number known to us.
 384         return context.getSnapshotManager().getLastSequenceNumber();
 385     }
 386
 387     /**
 388      * When a derived RaftActor needs to persist something it must call
 389      * persistData.
 390      *
 391      * @param clientActor
 392      * @param identifier
 393      * @param data
 394      */
 395     protected void persistData(final ActorRef clientActor, final String identifier,
 396         final Payload data) {
 397
 398         ReplicatedLogEntry replicatedLogEntry = new ReplicatedLogImplEntry(
 399             context.getReplicatedLog().lastIndex() + 1,
 400             context.getTermInformation().getCurrentTerm(), data);
 401
 402         if(LOG.isDebugEnabled()) {
 403             LOG.debug("{}: Persist data {}", persistenceId(), replicatedLogEntry);
 404         }
 405
 406         final RaftActorContext raftContext = getRaftActorContext();
 407
 408         replicatedLog().appendAndPersist(replicatedLogEntry, new Procedure<ReplicatedLogEntry>() {
 409             @Override
 410             public void apply(ReplicatedLogEntry replicatedLogEntry) {
 411                 if (!hasFollowers()){
 412                     // Increment the Commit Index and the Last Applied values
 413                     raftContext.setCommitIndex(replicatedLogEntry.getIndex());
 414                     raftContext.setLastApplied(replicatedLogEntry.getIndex());
 415
 416                     // Apply the state immediately.
 417                     self().tell(new ApplyState(clientActor, identifier, replicatedLogEntry), self());
 418
 419                     // Send a ApplyJournalEntries message so that we write the fact that we applied
 420                     // the state to durable storage
 421                     self().tell(new ApplyJournalEntries(replicatedLogEntry.getIndex()), self());
 422
 423                 } else if (clientActor != null) {
 424                     context.getReplicatedLog().captureSnapshotIfReady(replicatedLogEntry);
 425
 426                     // Send message for replication
 427                     currentBehavior.handleMessage(getSelf(),
 428                             new Replicate(clientActor, identifier, replicatedLogEntry));
 429                 }
 430             }
 431         });
 432     }
 433
 434     private ReplicatedLog replicatedLog() {
 435         return context.getReplicatedLog();
 436     }
 437
 438     protected String getId() {
 439         return context.getId();
 440     }
 441
 442     @VisibleForTesting
 443     void setCurrentBehavior(RaftActorBehavior behavior) {
 444         currentBehavior.setDelegate(behavior);
 445     }
 446
 447     protected RaftActorBehavior getCurrentBehavior() {
 448         return currentBehavior.getDelegate();
 449     }
 450
 451     /**
 452      * Derived actors can call the isLeader method to check if the current
 453      * RaftActor is the Leader or not
 454      *
 455      * @return true it this RaftActor is a Leader false otherwise
 456      */
 457     protected boolean isLeader() {
 458         return context.getId().equals(currentBehavior.getLeaderId());
 459     }
 460
 461     /**
 462      * Derived actor can call getLeader if they need a reference to the Leader.
 463      * This would be useful for example in forwarding a request to an actor
 464      * which is the leader
 465      *
 466      * @return A reference to the leader if known, null otherwise
 467      */
 468     protected ActorSelection getLeader(){
 469         String leaderAddress = getLeaderAddress();
 470
 471         if(leaderAddress == null){
 472             return null;
 473         }
 474
 475         return context.actorSelection(leaderAddress);
 476     }
 477
 478     /**
 479      *
 480      * @return the current leader's id
 481      */
 482     protected String getLeaderId(){
 483         return currentBehavior.getLeaderId();
 484     }
 485
 486     protected RaftState getRaftState() {
 487         return currentBehavior.state();
 488     }
 489
 490     protected ReplicatedLogEntry getLastLogEntry() {
 491         return replicatedLog().last();
 492     }
 493
 494     protected Long getCurrentTerm(){
 495         return context.getTermInformation().getCurrentTerm();
 496     }
 497
 498     protected Long getCommitIndex(){
 499         return context.getCommitIndex();
 500     }
 501
 502     protected Long getLastApplied(){
 503         return context.getLastApplied();
 504     }
 505
 506     protected RaftActorContext getRaftActorContext() {
 507         return context;
 508     }
 509
 510     protected void updateConfigParams(ConfigParams configParams) {
 511
 512         // obtain the RaftPolicy for oldConfigParams and the updated one.
 513         String oldRaftPolicy = context.getConfigParams().
 514             getCustomRaftPolicyImplementationClass();
 515         String newRaftPolicy = configParams.
 516             getCustomRaftPolicyImplementationClass();
 517
 518         LOG.debug("{}: RaftPolicy used with prev.config {}, RaftPolicy used with newConfig {}", persistenceId(),
 519             oldRaftPolicy, newRaftPolicy);
 520         context.setConfigParams(configParams);
 521         if (!Objects.equal(oldRaftPolicy, newRaftPolicy)) {
 522             // The RaftPolicy was modified. If the current behavior is Follower then re-initialize to Follower
 523             // but transfer the previous leaderId so it doesn't immediately try to schedule an election. This
 524             // avoids potential disruption. Otherwise, switch to Follower normally.
 525             RaftActorBehavior behavior = currentBehavior.getDelegate();
 526             if(behavior instanceof Follower) {
 527                 String previousLeaderId = ((Follower)behavior).getLeaderId();
 528
 529                 LOG.debug("{}: Re-initializing to Follower with previous leaderId {}", persistenceId(), previousLeaderId);
 530
 531                 changeCurrentBehavior(new Follower(context, previousLeaderId));
 532             } else {
 533                 initializeBehavior();
 534             }
 535         }
 536     }
 537
 538     public final DataPersistenceProvider persistence() {
 539         return delegatingPersistenceProvider.getDelegate();
 540     }
 541
 542     public void setPersistence(DataPersistenceProvider provider) {
 543         delegatingPersistenceProvider.setDelegate(provider);
 544     }
 545
 546     protected void setPersistence(boolean persistent) {
 547         if(persistent) {
 548             setPersistence(new PersistentDataProvider(this));
 549         } else {
 550             setPersistence(new NonPersistentDataProvider() {
 551                 /**
 552                  * The way snapshotting works is,
 553                  * <ol>
 554                  * <li> RaftActor calls createSnapshot on the Shard
 555                  * <li> Shard sends a CaptureSnapshotReply and RaftActor then calls saveSnapshot
 556                  * <li> When saveSnapshot is invoked on the akka-persistence API it uses the SnapshotStore to save
 557                  * the snapshot. The SnapshotStore sends SaveSnapshotSuccess or SaveSnapshotFailure. When the
 558                  * RaftActor gets SaveSnapshot success it commits the snapshot to the in-memory journal. This
 559                  * commitSnapshot is mimicking what is done in SaveSnapshotSuccess.
 560                  * </ol>
 561                  */
 562                 @Override
 563                 public void saveSnapshot(Object o) {
 564                     // Make saving Snapshot successful
 565                     // Committing the snapshot here would end up calling commit in the creating state which would
 566                     // be a state violation. That's why now we send a message to commit the snapshot.
 567                     self().tell(RaftActorSnapshotMessageSupport.COMMIT_SNAPSHOT, self());
 568                 }
 569             });
 570         }
 571     }
 572
 573     /**
 574      * setPeerAddress sets the address of a known peer at a later time.
 575      * <p>
 576      * This is to account for situations where a we know that a peer
 577      * exists but we do not know an address up-front. This may also be used in
 578      * situations where a known peer starts off in a different location and we
 579      * need to change it's address
 580      * <p>
 581      * Note that if the peerId does not match the list of peers passed to
 582      * this actor during construction an IllegalStateException will be thrown.
 583      *
 584      * @param peerId
 585      * @param peerAddress
 586      */
 587     protected void setPeerAddress(String peerId, String peerAddress){
 588         context.setPeerAddress(peerId, peerAddress);
 589     }
 590
 591     /**
 592      * The applyState method will be called by the RaftActor when some data
 593      * needs to be applied to the actor's state
 594      *
 595      * @param clientActor A reference to the client who sent this message. This
 596      *                    is the same reference that was passed to persistData
 597      *                    by the derived actor. clientActor may be null when
 598      *                    the RaftActor is behaving as a follower or during
 599      *                    recovery.
 600      * @param identifier  The identifier of the persisted data. This is also
 601      *                    the same identifier that was passed to persistData by
 602      *                    the derived actor. identifier may be null when
 603      *                    the RaftActor is behaving as a follower or during
 604      *                    recovery
 605      * @param data        A piece of data that was persisted by the persistData call.
 606      *                    This should NEVER be null.
 607      */
 608     protected abstract void applyState(ActorRef clientActor, String identifier,
 609         Object data);
 610
 611     /**
 612      * Returns the RaftActorRecoveryCohort to participate in persistence recovery.
 613      */
 614     @Nonnull
 615     protected abstract RaftActorRecoveryCohort getRaftActorRecoveryCohort();
 616
 617     /**
 618      * This method is called when recovery is complete.
 619      */
 620     protected abstract void onRecoveryComplete();
 621
 622     /**
 623      * Returns the RaftActorSnapshotCohort to participate in persistence recovery.
 624      */
 625     @Nonnull
 626     protected abstract RaftActorSnapshotCohort getRaftActorSnapshotCohort();
 627
 628     /**
 629      * This method will be called by the RaftActor when the state of the
 630      * RaftActor changes. The derived actor can then use methods like
 631      * isLeader or getLeader to do something useful
 632      */
 633     protected abstract void onStateChanged();
 634
 635     /**
 636      * Notifier Actor for this RaftActor to notify when a role change happens
 637      * @return ActorRef - ActorRef of the notifier or Optional.absent if none.
 638      */
 639     protected abstract Optional<ActorRef> getRoleChangeNotifier();
 640
 641     protected void onLeaderChanged(String oldLeader, String newLeader){};
 642
 643     private String getLeaderAddress(){
 644         if(isLeader()){
 645             return getSelf().path().toString();
 646         }
 647         String leaderId = currentBehavior.getLeaderId();
 648         if (leaderId == null) {
 649             return null;
 650         }
 651         String peerAddress = context.getPeerAddress(leaderId);
 652         if(LOG.isDebugEnabled()) {
 653             LOG.debug("{}: getLeaderAddress leaderId = {} peerAddress = {}",
 654                     persistenceId(), leaderId, peerAddress);
 655         }
 656
 657         return peerAddress;
 658     }
 659
 660     protected boolean hasFollowers(){
 661         return getRaftActorContext().hasFollowers();
 662     }
 663
 664     private void captureSnapshot() {
 665         SnapshotManager snapshotManager = context.getSnapshotManager();
 666
 667         if(!snapshotManager.isCapturing()) {
 668             LOG.debug("Take a snapshot of current state. lastReplicatedLog is {} and replicatedToAllIndex is {}",
 669                 replicatedLog().last(), currentBehavior.getReplicatedToAllIndex());
 670
 671             snapshotManager.capture(replicatedLog().last(), currentBehavior.getReplicatedToAllIndex());
 672         }
 673     }
 674
 675     /**
 676      * @deprecated Deprecated in favor of {@link org.opendaylight.controller.cluster.raft.base.messages.DeleteEntries}
 677      *             whose type for fromIndex is long instead of int. This class was kept for backwards
 678      *             compatibility with Helium.
 679      */
 680     // Suppressing this warning as we can't set serialVersionUID to maintain backwards compatibility.
 681     @SuppressWarnings("serial")
 682     @Deprecated
 683     static class DeleteEntries implements Serializable {
 684         private final int fromIndex;
 685
 686         public DeleteEntries(int fromIndex) {
 687             this.fromIndex = fromIndex;
 688         }
 689
 690         public int getFromIndex() {
 691             return fromIndex;
 692         }
 693     }
 694
 695     /**
 696      * @deprecated Deprecated in favor of non-inner class {@link org.opendaylight.controller.cluster.raft.base.messages.UpdateElectionTerm}
 697      *             which has serialVersionUID set. This class was kept for backwards compatibility with Helium.
 698      */
 699     // Suppressing this warning as we can't set serialVersionUID to maintain backwards compatibility.
 700     @SuppressWarnings("serial")
 701     @Deprecated
 702     static class UpdateElectionTerm implements Serializable {
 703         private final long currentTerm;
 704         private final String votedFor;
 705
 706         public UpdateElectionTerm(long currentTerm, String votedFor) {
 707             this.currentTerm = currentTerm;
 708             this.votedFor = votedFor;
 709         }
 710
 711         public long getCurrentTerm() {
 712             return currentTerm;
 713         }
 714
 715         public String getVotedFor() {
 716             return votedFor;
 717         }
 718     }
 719
 720     private static class BehaviorStateHolder {
 721         private RaftActorBehavior behavior;
 722         private String lastValidLeaderId;
 723         private short leaderPayloadVersion;
 724
 725         void init(RaftActorBehavior behavior) {
 726             this.behavior = behavior;
 727             this.leaderPayloadVersion = behavior != null ? behavior.getLeaderPayloadVersion() : -1;
 728
 729             String behaviorLeaderId = behavior != null ? behavior.getLeaderId() : null;
 730             if(behaviorLeaderId != null) {
 731                 this.lastValidLeaderId = behaviorLeaderId;
 732             }
 733         }
 734
 735         RaftActorBehavior getBehavior() {
 736             return behavior;
 737         }
 738
 739         String getLastValidLeaderId() {
 740             return lastValidLeaderId;
 741         }
 742
 743         short getLeaderPayloadVersion() {
 744             return leaderPayloadVersion;
 745         }
 746     }
 747
 748     private class SwitchBehaviorSupplier implements Supplier<RaftActorBehavior> {
 749         private Object message;
 750         private ActorRef sender;
 751
 752         public SwitchBehaviorSupplier handleMessage(ActorRef sender, Object message){
 753             this.sender = sender;
 754             this.message = message;
 755             return this;
 756         }
 757
 758         @Override
 759         public RaftActorBehavior get() {
 760             if(this.message instanceof SwitchBehavior){
 761                 return ((SwitchBehavior) message).getNewState().createBehavior(getRaftActorContext());
 762             }
 763             return currentBehavior.handleMessage(sender, message);
 764         }
 765     }
 766 }