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 package org.opendaylight.controller.cluster.raft;
  10
  11 import static com.google.common.base.Verify.verify;
  12 import static java.util.Objects.requireNonNull;
  13
  14 import akka.actor.ActorRef;
  15 import akka.actor.ActorSelection;
  16 import akka.actor.PoisonPill;
  17 import akka.actor.Status;
  18 import akka.persistence.JournalProtocol;
  19 import akka.persistence.SnapshotProtocol;
  20 import com.google.common.annotations.VisibleForTesting;
  21 import edu.umd.cs.findbugs.annotations.SuppressFBWarnings;
  22 import java.util.ArrayList;
  23 import java.util.Collection;
  24 import java.util.HashMap;
  25 import java.util.List;
  26 import java.util.Map;
  27 import java.util.Objects;
  28 import java.util.Optional;
  29 import java.util.concurrent.TimeUnit;
  30 import org.apache.commons.lang3.time.DurationFormatUtils;
  31 import org.eclipse.jdt.annotation.NonNull;
  32 import org.eclipse.jdt.annotation.Nullable;
  33 import org.opendaylight.controller.cluster.DataPersistenceProvider;
  34 import org.opendaylight.controller.cluster.DelegatingPersistentDataProvider;
  35 import org.opendaylight.controller.cluster.NonPersistentDataProvider;
  36 import org.opendaylight.controller.cluster.PersistentDataProvider;
  37 import org.opendaylight.controller.cluster.common.actor.AbstractUntypedPersistentActor;
  38 import org.opendaylight.controller.cluster.mgmt.api.FollowerInfo;
  39 import org.opendaylight.controller.cluster.notifications.LeaderStateChanged;
  40 import org.opendaylight.controller.cluster.notifications.RoleChanged;
  41 import org.opendaylight.controller.cluster.raft.base.messages.ApplyState;
  42 import org.opendaylight.controller.cluster.raft.base.messages.CheckConsensusReached;
  43 import org.opendaylight.controller.cluster.raft.base.messages.InitiateCaptureSnapshot;
  44 import org.opendaylight.controller.cluster.raft.base.messages.LeaderTransitioning;
  45 import org.opendaylight.controller.cluster.raft.base.messages.Replicate;
  46 import org.opendaylight.controller.cluster.raft.base.messages.SwitchBehavior;
  47 import org.opendaylight.controller.cluster.raft.behaviors.AbstractLeader;
  48 import org.opendaylight.controller.cluster.raft.behaviors.AbstractRaftActorBehavior;
  49 import org.opendaylight.controller.cluster.raft.behaviors.Follower;
  50 import org.opendaylight.controller.cluster.raft.behaviors.RaftActorBehavior;
  51 import org.opendaylight.controller.cluster.raft.client.messages.FindLeader;
  52 import org.opendaylight.controller.cluster.raft.client.messages.FindLeaderReply;
  53 import org.opendaylight.controller.cluster.raft.client.messages.GetOnDemandRaftState;
  54 import org.opendaylight.controller.cluster.raft.client.messages.OnDemandRaftState;
  55 import org.opendaylight.controller.cluster.raft.client.messages.Shutdown;
  56 import org.opendaylight.controller.cluster.raft.messages.Payload;
  57 import org.opendaylight.controller.cluster.raft.messages.RequestLeadership;
  58 import org.opendaylight.controller.cluster.raft.persisted.ApplyJournalEntries;
  59 import org.opendaylight.controller.cluster.raft.persisted.NoopPayload;
  60 import org.opendaylight.controller.cluster.raft.persisted.ServerConfigurationPayload;
  61 import org.opendaylight.controller.cluster.raft.persisted.SimpleReplicatedLogEntry;
  62 import org.opendaylight.yangtools.concepts.Identifier;
  63 import org.opendaylight.yangtools.concepts.Immutable;
  64
  65 /**
  66  * RaftActor encapsulates a state machine that needs to be kept synchronized
  67  * in a cluster. It implements the RAFT algorithm as described in the paper
  68  * <a href='https://ramcloud.stanford.edu/wiki/download/attachments/11370504/raft.pdf'>
  69  * In Search of an Understandable Consensus Algorithm</a>
  70  *
  71  * <p>
  72  * RaftActor has 3 states and each state has a certain behavior associated
  73  * with it. A Raft actor can behave as,
  74  * <ul>
  75  * <li> A Leader </li>
  76  * <li> A Follower (or) </li>
  77  * <li> A Candidate </li>
  78  * </ul>
  79  *
  80  * <p>
  81  * A RaftActor MUST be a Leader in order to accept requests from clients to
  82  * change the state of it's encapsulated state machine. Once a RaftActor becomes
  83  * a Leader it is also responsible for ensuring that all followers ultimately
  84  * have the same log and therefore the same state machine as itself.
  85  *
  86  * <p>
  87  * The current behavior of a RaftActor determines how election for leadership
  88  * is initiated and how peer RaftActors react to request for votes.
  89  *
  90  * <p>
  91  * Each RaftActor also needs to know the current election term. It uses this
  92  * information for a couple of things. One is to simply figure out who it
  93  * voted for in the last election. Another is to figure out if the message
  94  * it received to update it's state is stale.
  95  *
  96  * <p>
  97  * The RaftActor uses akka-persistence to store it's replicated log.
  98  * Furthermore through it's behaviors a Raft Actor determines
  99  * <ul>
 100  * <li> when a log entry should be persisted </li>
 101  * <li> when a log entry should be applied to the state machine (and) </li>
 102  * <li> when a snapshot should be saved </li>
 103  * </ul>
 104  */
 105 public abstract class RaftActor extends AbstractUntypedPersistentActor {
 106
 107     private static final long APPLY_STATE_DELAY_THRESHOLD_IN_NANOS = TimeUnit.MILLISECONDS.toNanos(50L); // 50 millis
 108
 109     /**
 110      * This context should NOT be passed directly to any other actor it is
 111      * only to be consumed by the RaftActorBehaviors.
 112      */
 113     private final RaftActorContextImpl context;
 114
 115     private final DelegatingPersistentDataProvider delegatingPersistenceProvider;
 116
 117     private final PersistentDataProvider persistentProvider;
 118
 119     private final BehaviorStateTracker behaviorStateTracker = new BehaviorStateTracker();
 120
 121     private RaftActorRecoverySupport raftRecovery;
 122
 123     private RaftActorSnapshotMessageSupport snapshotSupport;
 124
 125     private RaftActorServerConfigurationSupport serverConfigurationSupport;
 126
 127     private boolean shuttingDown;
 128
 129     @SuppressFBWarnings(value = "MC_OVERRIDABLE_METHOD_CALL_IN_CONSTRUCTOR", justification = "Akka class design")
 130     protected RaftActor(final String id, final Map<String, String> peerAddresses,
 131          final Optional<ConfigParams> configParams, final short payloadVersion) {
 132
 133         persistentProvider = new PersistentDataProvider(this);
 134         delegatingPersistenceProvider = new RaftActorDelegatingPersistentDataProvider(null, persistentProvider);
 135
 136         context = new RaftActorContextImpl(getSelf(), getContext(), id,
 137             new ElectionTermImpl(persistentProvider, id, LOG), -1, -1, peerAddresses,
 138             configParams.isPresent() ? configParams.get() : new DefaultConfigParamsImpl(),
 139             delegatingPersistenceProvider, this::handleApplyState, LOG, this::executeInSelf);
 140
 141         context.setPayloadVersion(payloadVersion);
 142         context.setReplicatedLog(ReplicatedLogImpl.newInstance(context));
 143     }
 144
 145     @Override
 146     public void preStart() throws Exception {
 147         LOG.info("Starting recovery for {} with journal batch size {}", persistenceId(),
 148                 context.getConfigParams().getJournalRecoveryLogBatchSize());
 149
 150         super.preStart();
 151
 152         snapshotSupport = newRaftActorSnapshotMessageSupport();
 153         serverConfigurationSupport = new RaftActorServerConfigurationSupport(this);
 154     }
 155
 156     @Override
 157     public void postStop() throws Exception {
 158         context.close();
 159         super.postStop();
 160     }
 161
 162     @Override
 163     protected void handleRecover(final Object message) {
 164         if (raftRecovery == null) {
 165             raftRecovery = newRaftActorRecoverySupport();
 166         }
 167
 168         boolean recoveryComplete = raftRecovery.handleRecoveryMessage(message, persistentProvider);
 169         if (recoveryComplete) {
 170             onRecoveryComplete();
 171
 172             initializeBehavior();
 173
 174             raftRecovery = null;
 175         }
 176     }
 177
 178     protected RaftActorRecoverySupport newRaftActorRecoverySupport() {
 179         return new RaftActorRecoverySupport(context, getRaftActorRecoveryCohort());
 180     }
 181
 182     @VisibleForTesting
 183     void initializeBehavior() {
 184         changeCurrentBehavior(new Follower(context));
 185     }
 186
 187     @VisibleForTesting
 188     @SuppressWarnings("checkstyle:IllegalCatch")
 189     protected void changeCurrentBehavior(final RaftActorBehavior newBehavior) {
 190         final RaftActorBehavior currentBehavior = getCurrentBehavior();
 191         if (currentBehavior != null) {
 192             try {
 193                 currentBehavior.close();
 194             } catch (Exception e) {
 195                 LOG.warn("{}: Error closing behavior {}", persistence(), currentBehavior, e);
 196             }
 197         }
 198
 199         final BehaviorState state = behaviorStateTracker.capture(currentBehavior);
 200         setCurrentBehavior(newBehavior);
 201         handleBehaviorChange(state, newBehavior);
 202     }
 203
 204     /**
 205      * Method exposed for subclasses to plug-in their logic. This method is invoked by {@link #handleCommand(Object)}
 206      * for messages which are not handled by this class. Subclasses overriding this class should fall back to this
 207      * implementation for messages which they do not handle
 208      *
 209      * @param message Incoming command message
 210      */
 211     protected void handleNonRaftCommand(final Object message) {
 212         unhandled(message);
 213     }
 214
 215     /**
 216      * Handles a message.
 217      *
 218      * @deprecated This method is not final for testing purposes. DO NOT OVERRIDE IT, override
 219      *             {@link #handleNonRaftCommand(Object)} instead.
 220      */
 221     @Deprecated
 222     @Override
 223     // FIXME: make this method final once our unit tests do not need to override it
 224     protected void handleCommand(final Object message) {
 225         if (serverConfigurationSupport.handleMessage(message, getSender())) {
 226             return;
 227         }
 228         if (snapshotSupport.handleSnapshotMessage(message, getSender())) {
 229             return;
 230         }
 231         if (message instanceof ApplyState applyState) {
 232             if (!hasFollowers()) {
 233                 // for single node, the capture should happen after the apply state
 234                 // as we delete messages from the persistent journal which have made it to the snapshot
 235                 // capturing the snapshot before applying makes the persistent journal and snapshot out of sync
 236                 // and recovery shows data missing
 237                 context.getReplicatedLog().captureSnapshotIfReady(applyState.getReplicatedLogEntry());
 238
 239                 context.getSnapshotManager().trimLog(context.getLastApplied());
 240             }
 241
 242             possiblyHandleBehaviorMessage(message);
 243         } else if (message instanceof ApplyJournalEntries applyEntries) {
 244             LOG.debug("{}: Persisting ApplyJournalEntries with index={}", persistenceId(), applyEntries.getToIndex());
 245
 246             persistence().persistAsync(applyEntries, NoopProcedure.instance());
 247         } else if (message instanceof FindLeader) {
 248             getSender().tell(new FindLeaderReply(getLeaderAddress()), getSelf());
 249         } else if (message instanceof GetOnDemandRaftState) {
 250             onGetOnDemandRaftStats();
 251         } else if (message instanceof InitiateCaptureSnapshot) {
 252             captureSnapshot();
 253         } else if (message instanceof SwitchBehavior switchBehavior) {
 254             switchBehavior(switchBehavior);
 255         } else if (message instanceof LeaderTransitioning leaderTransitioning) {
 256             onLeaderTransitioning(leaderTransitioning);
 257         } else if (message instanceof Shutdown) {
 258             onShutDown();
 259         } else if (message instanceof Runnable runnable) {
 260             runnable.run();
 261         } else if (message instanceof NoopPayload noopPayload) {
 262             persistData(null, null, noopPayload, false);
 263         } else if (message instanceof RequestLeadership requestLeadership) {
 264             onRequestLeadership(requestLeadership);
 265         } else if (!possiblyHandleBehaviorMessage(message)) {
 266             if (message instanceof JournalProtocol.Response response
 267                 && delegatingPersistenceProvider.handleJournalResponse(response)) {
 268                 LOG.debug("{}: handled a journal response", persistenceId());
 269             } else if (message instanceof SnapshotProtocol.Response response
 270                 && delegatingPersistenceProvider.handleSnapshotResponse(response)) {
 271                 LOG.debug("{}: handled a snapshot response", persistenceId());
 272             } else {
 273                 handleNonRaftCommand(message);
 274             }
 275         }
 276     }
 277
 278     private void onRequestLeadership(final RequestLeadership message) {
 279         LOG.debug("{}: onRequestLeadership {}", persistenceId(), message);
 280         if (!isLeader()) {
 281             // non-leader cannot satisfy leadership request
 282             LOG.warn("{}: onRequestLeadership {} was sent to non-leader."
 283                     + " Current behavior: {}. Sending failure response",
 284                     persistenceId(), message, getCurrentBehavior().state());
 285             message.getReplyTo().tell(new LeadershipTransferFailedException("Cannot transfer leader to "
 286                     + message.getRequestedFollowerId()
 287                     + ". RequestLeadership message was sent to non-leader " + persistenceId()), getSelf());
 288             return;
 289         }
 290
 291         final String requestedFollowerId = message.getRequestedFollowerId();
 292         final ActorRef replyTo = message.getReplyTo();
 293         initiateLeadershipTransfer(new RaftActorLeadershipTransferCohort.OnComplete() {
 294             @Override
 295             public void onSuccess(final ActorRef raftActorRef) {
 296                 // sanity check
 297                 if (!requestedFollowerId.equals(getLeaderId())) {
 298                     onFailure(raftActorRef);
 299                 }
 300
 301                 LOG.debug("{}: Leadership transferred successfully to {}", persistenceId(), requestedFollowerId);
 302                 replyTo.tell(new Status.Success(null), getSelf());
 303             }
 304
 305             @Override
 306             public void onFailure(final ActorRef raftActorRef) {
 307                 LOG.debug("{}: LeadershipTransfer request from {} failed", persistenceId(), requestedFollowerId);
 308                 replyTo.tell(new Status.Failure(
 309                         new LeadershipTransferFailedException(
 310                                 "Failed to transfer leadership to " + requestedFollowerId
 311                                         + ". Follower is not ready to become leader")),
 312                         getSelf());
 313             }
 314         }, message.getRequestedFollowerId(), RaftActorLeadershipTransferCohort.USE_DEFAULT_LEADER_TIMEOUT);
 315     }
 316
 317     private boolean possiblyHandleBehaviorMessage(final Object message) {
 318         final RaftActorBehavior currentBehavior = getCurrentBehavior();
 319         final BehaviorState state = behaviorStateTracker.capture(currentBehavior);
 320
 321         // A behavior indicates that it processed the change by returning a reference to the next behavior
 322         // to be used. A null return indicates it has not processed the message and we should be passing it to
 323         // the subclass for handling.
 324         final RaftActorBehavior nextBehavior = currentBehavior.handleMessage(getSender(), message);
 325         if (nextBehavior != null) {
 326             switchBehavior(state, nextBehavior);
 327             return true;
 328         }
 329
 330         return false;
 331     }
 332
 333     private void initiateLeadershipTransfer(final RaftActorLeadershipTransferCohort.OnComplete onComplete,
 334             final @Nullable String followerId, final long newLeaderTimeoutInMillis) {
 335         LOG.debug("{}: Initiating leader transfer", persistenceId());
 336
 337         RaftActorLeadershipTransferCohort leadershipTransferInProgress = context.getRaftActorLeadershipTransferCohort();
 338         if (leadershipTransferInProgress == null) {
 339             leadershipTransferInProgress = new RaftActorLeadershipTransferCohort(this, followerId);
 340             leadershipTransferInProgress.setNewLeaderTimeoutInMillis(newLeaderTimeoutInMillis);
 341             leadershipTransferInProgress.addOnComplete(new RaftActorLeadershipTransferCohort.OnComplete() {
 342                 @Override
 343                 public void onSuccess(final ActorRef raftActorRef) {
 344                     context.setRaftActorLeadershipTransferCohort(null);
 345                 }
 346
 347                 @Override
 348                 public void onFailure(final ActorRef raftActorRef) {
 349                     context.setRaftActorLeadershipTransferCohort(null);
 350                 }
 351             });
 352
 353             leadershipTransferInProgress.addOnComplete(onComplete);
 354
 355             context.setRaftActorLeadershipTransferCohort(leadershipTransferInProgress);
 356             leadershipTransferInProgress.init();
 357
 358         } else {
 359             LOG.debug("{}: prior leader transfer in progress - adding callback", persistenceId());
 360             leadershipTransferInProgress.addOnComplete(onComplete);
 361         }
 362     }
 363
 364     private void onShutDown() {
 365         LOG.debug("{}: onShutDown", persistenceId());
 366
 367         if (shuttingDown) {
 368             return;
 369         }
 370
 371         shuttingDown = true;
 372
 373         final RaftActorBehavior currentBehavior = context.getCurrentBehavior();
 374         switch (currentBehavior.state()) {
 375             case Leader:
 376             case PreLeader:
 377                 // Fall-through to more work
 378                 break;
 379             default:
 380                 // For non-leaders shutdown is a no-op
 381                 self().tell(PoisonPill.getInstance(), self());
 382                 return;
 383         }
 384
 385         if (context.hasFollowers()) {
 386             initiateLeadershipTransfer(new RaftActorLeadershipTransferCohort.OnComplete() {
 387                 @Override
 388                 public void onSuccess(final ActorRef raftActorRef) {
 389                     LOG.debug("{}: leader transfer succeeded - sending PoisonPill", persistenceId());
 390                     raftActorRef.tell(PoisonPill.getInstance(), raftActorRef);
 391                 }
 392
 393                 @Override
 394                 public void onFailure(final ActorRef raftActorRef) {
 395                     LOG.debug("{}: leader transfer failed - sending PoisonPill", persistenceId());
 396                     raftActorRef.tell(PoisonPill.getInstance(), raftActorRef);
 397                 }
 398             }, null, TimeUnit.MILLISECONDS.convert(2, TimeUnit.SECONDS));
 399         } else {
 400             pauseLeader(new TimedRunnable(context.getConfigParams().getElectionTimeOutInterval(), this) {
 401                 @Override
 402                 protected void doRun() {
 403                     self().tell(PoisonPill.getInstance(), self());
 404                 }
 405
 406                 @Override
 407                 protected void doCancel() {
 408                     self().tell(PoisonPill.getInstance(), self());
 409                 }
 410             });
 411         }
 412     }
 413
 414     private void onLeaderTransitioning(final LeaderTransitioning leaderTransitioning) {
 415         LOG.debug("{}: onLeaderTransitioning: {}", persistenceId(), leaderTransitioning);
 416         Optional<ActorRef> roleChangeNotifier = getRoleChangeNotifier();
 417         if (getRaftState() == RaftState.Follower && roleChangeNotifier.isPresent()
 418                 && leaderTransitioning.getLeaderId().equals(getCurrentBehavior().getLeaderId())) {
 419             roleChangeNotifier.get().tell(newLeaderStateChanged(getId(), null,
 420                 getCurrentBehavior().getLeaderPayloadVersion()), getSelf());
 421         }
 422     }
 423
 424     private void switchBehavior(final SwitchBehavior message) {
 425         if (!getRaftActorContext().getRaftPolicy().automaticElectionsEnabled()) {
 426             RaftState newState = message.getNewState();
 427             if (newState == RaftState.Leader || newState == RaftState.Follower) {
 428                 getRaftActorContext().getTermInformation().updateAndPersist(message.getNewTerm(), "");
 429                 switchBehavior(behaviorStateTracker.capture(getCurrentBehavior()),
 430                     AbstractRaftActorBehavior.createBehavior(context, message.getNewState()));
 431             } else {
 432                 LOG.warn("Switching to behavior : {} - not supported", newState);
 433             }
 434         }
 435     }
 436
 437     private void switchBehavior(final BehaviorState oldBehaviorState, final RaftActorBehavior nextBehavior) {
 438         setCurrentBehavior(nextBehavior);
 439         handleBehaviorChange(oldBehaviorState, nextBehavior);
 440     }
 441
 442     @VisibleForTesting
 443     RaftActorSnapshotMessageSupport newRaftActorSnapshotMessageSupport() {
 444         return new RaftActorSnapshotMessageSupport(context, getRaftActorSnapshotCohort());
 445     }
 446
 447     private void onGetOnDemandRaftStats() {
 448         // Debugging message to retrieve raft stats.
 449
 450         Map<String, String> peerAddresses = new HashMap<>();
 451         Map<String, Boolean> peerVotingStates = new HashMap<>();
 452         for (PeerInfo info: context.getPeers()) {
 453             peerVotingStates.put(info.getId(), info.isVoting());
 454             peerAddresses.put(info.getId(), info.getAddress() != null ? info.getAddress() : "");
 455         }
 456
 457         final RaftActorBehavior currentBehavior = context.getCurrentBehavior();
 458         OnDemandRaftState.AbstractBuilder<?, ?> builder = newOnDemandRaftStateBuilder()
 459                 .commitIndex(context.getCommitIndex())
 460                 .currentTerm(context.getTermInformation().getCurrentTerm())
 461                 .inMemoryJournalDataSize(replicatedLog().dataSize())
 462                 .inMemoryJournalLogSize(replicatedLog().size())
 463                 .isSnapshotCaptureInitiated(context.getSnapshotManager().isCapturing())
 464                 .lastApplied(context.getLastApplied())
 465                 .lastIndex(replicatedLog().lastIndex())
 466                 .lastTerm(replicatedLog().lastTerm())
 467                 .leader(getLeaderId())
 468                 .raftState(currentBehavior.state().toString())
 469                 .replicatedToAllIndex(currentBehavior.getReplicatedToAllIndex())
 470                 .snapshotIndex(replicatedLog().getSnapshotIndex())
 471                 .snapshotTerm(replicatedLog().getSnapshotTerm())
 472                 .votedFor(context.getTermInformation().getVotedFor())
 473                 .isVoting(context.isVotingMember())
 474                 .peerAddresses(peerAddresses)
 475                 .peerVotingStates(peerVotingStates)
 476                 .customRaftPolicyClassName(context.getConfigParams().getCustomRaftPolicyImplementationClass());
 477
 478         ReplicatedLogEntry lastLogEntry = replicatedLog().last();
 479         if (lastLogEntry != null) {
 480             builder.lastLogIndex(lastLogEntry.getIndex());
 481             builder.lastLogTerm(lastLogEntry.getTerm());
 482         }
 483
 484         if (getCurrentBehavior() instanceof AbstractLeader leader) {
 485             Collection<String> followerIds = leader.getFollowerIds();
 486             List<FollowerInfo> followerInfoList = new ArrayList<>(followerIds.size());
 487             for (String id : followerIds) {
 488                 final FollowerLogInformation info = leader.getFollower(id);
 489                 followerInfoList.add(new FollowerInfo(id, info.getNextIndex(), info.getMatchIndex(),
 490                         info.isFollowerActive(), DurationFormatUtils.formatDurationHMS(
 491                             TimeUnit.NANOSECONDS.toMillis(info.nanosSinceLastActivity())),
 492                         context.getPeerInfo(info.getId()).isVoting()));
 493             }
 494
 495             builder.followerInfoList(followerInfoList);
 496         }
 497
 498         sender().tell(builder.build(), self());
 499
 500     }
 501
 502     protected OnDemandRaftState.AbstractBuilder<?, ?> newOnDemandRaftStateBuilder() {
 503         return OnDemandRaftState.builder();
 504     }
 505
 506     private void handleBehaviorChange(final BehaviorState oldBehaviorState, final RaftActorBehavior currentBehavior) {
 507         RaftActorBehavior oldBehavior = oldBehaviorState.getBehavior();
 508
 509         if (oldBehavior != currentBehavior) {
 510             onStateChanged();
 511         }
 512
 513         String lastLeaderId = oldBehavior == null ? null : oldBehaviorState.getLastLeaderId();
 514         String lastValidLeaderId = oldBehavior == null ? null : oldBehaviorState.getLastValidLeaderId();
 515         String oldBehaviorStateName = oldBehavior == null ? null : oldBehavior.state().name();
 516
 517         // it can happen that the state has not changed but the leader has changed.
 518         Optional<ActorRef> roleChangeNotifier = getRoleChangeNotifier();
 519         if (!Objects.equals(lastLeaderId, currentBehavior.getLeaderId())
 520                 || oldBehaviorState.getLeaderPayloadVersion() != currentBehavior.getLeaderPayloadVersion()) {
 521             if (roleChangeNotifier.isPresent()) {
 522                 roleChangeNotifier.get().tell(newLeaderStateChanged(getId(), currentBehavior.getLeaderId(),
 523                         currentBehavior.getLeaderPayloadVersion()), getSelf());
 524             }
 525
 526             onLeaderChanged(lastValidLeaderId, currentBehavior.getLeaderId());
 527
 528             RaftActorLeadershipTransferCohort leadershipTransferInProgress =
 529                     context.getRaftActorLeadershipTransferCohort();
 530             if (leadershipTransferInProgress != null) {
 531                 leadershipTransferInProgress.onNewLeader(currentBehavior.getLeaderId());
 532             }
 533
 534             serverConfigurationSupport.onNewLeader(currentBehavior.getLeaderId());
 535         }
 536
 537         if (roleChangeNotifier.isPresent()
 538                 && (oldBehavior == null || oldBehavior.state() != currentBehavior.state())) {
 539             roleChangeNotifier.get().tell(new RoleChanged(getId(), oldBehaviorStateName ,
 540                     currentBehavior.state().name()), getSelf());
 541         }
 542     }
 543
 544     private void handleApplyState(final ApplyState applyState) {
 545         long startTime = System.nanoTime();
 546
 547         Payload payload = applyState.getReplicatedLogEntry().getData();
 548         if (LOG.isDebugEnabled()) {
 549             LOG.debug("{}: Applying state for log index {} data {}",
 550                 persistenceId(), applyState.getReplicatedLogEntry().getIndex(), payload);
 551         }
 552
 553         if (!(payload instanceof NoopPayload) && !(payload instanceof ServerConfigurationPayload)) {
 554             applyState(applyState.getClientActor(), applyState.getIdentifier(), payload);
 555         }
 556
 557         long elapsedTime = System.nanoTime() - startTime;
 558         if (elapsedTime >= APPLY_STATE_DELAY_THRESHOLD_IN_NANOS) {
 559             LOG.debug("ApplyState took more time than expected. Elapsed Time = {} ms ApplyState = {}",
 560                     TimeUnit.NANOSECONDS.toMillis(elapsedTime), applyState);
 561         }
 562
 563         // Send the ApplyState message back to self to handle further processing asynchronously.
 564         self().tell(applyState, self());
 565     }
 566
 567     protected LeaderStateChanged newLeaderStateChanged(final String memberId, final String leaderId,
 568             final short leaderPayloadVersion) {
 569         return new LeaderStateChanged(memberId, leaderId, leaderPayloadVersion);
 570     }
 571
 572     @Override
 573     public long snapshotSequenceNr() {
 574         // When we do a snapshot capture, we also capture and save the sequence-number of the persistent journal,
 575         // so that we can delete the persistent journal based on the saved sequence-number
 576         // However , when akka replays the journal during recovery, it replays it from the sequence number when the
 577         // snapshot was saved and not the number we saved. We would want to override it , by asking akka to use the
 578         // last-sequence number known to us.
 579         return context.getSnapshotManager().getLastSequenceNumber();
 580     }
 581
 582     /**
 583      * Persists the given Payload in the journal and replicates to any followers. After successful completion,
 584      * {@link #applyState(ActorRef, Identifier, Object)} is notified.
 585      *
 586      * @param clientActor optional ActorRef that is provided via the applyState callback
 587      * @param identifier the payload identifier
 588      * @param data the payload data to persist
 589      * @param batchHint if true, an attempt is made to delay immediate replication and batch the payload with
 590      *        subsequent payloads for efficiency. Otherwise the payload is immediately replicated.
 591      */
 592     protected final void persistData(final ActorRef clientActor, final Identifier identifier, final Payload data,
 593             final boolean batchHint) {
 594         ReplicatedLogEntry replicatedLogEntry = new SimpleReplicatedLogEntry(
 595             context.getReplicatedLog().lastIndex() + 1,
 596             context.getTermInformation().getCurrentTerm(), data);
 597         replicatedLogEntry.setPersistencePending(true);
 598
 599         LOG.debug("{}: Persist data {}", persistenceId(), replicatedLogEntry);
 600
 601         final RaftActorContext raftContext = getRaftActorContext();
 602
 603         boolean wasAppended = replicatedLog().appendAndPersist(replicatedLogEntry, persistedLogEntry -> {
 604             // Clear the persistence pending flag in the log entry.
 605             persistedLogEntry.setPersistencePending(false);
 606
 607             if (!hasFollowers()) {
 608                 // Increment the Commit Index and the Last Applied values
 609                 raftContext.setCommitIndex(persistedLogEntry.getIndex());
 610                 raftContext.setLastApplied(persistedLogEntry.getIndex());
 611
 612                 // Apply the state immediately.
 613                 handleApplyState(new ApplyState(clientActor, identifier, persistedLogEntry));
 614
 615                 // Send a ApplyJournalEntries message so that we write the fact that we applied
 616                 // the state to durable storage
 617                 self().tell(new ApplyJournalEntries(persistedLogEntry.getIndex()), self());
 618
 619             } else {
 620                 context.getReplicatedLog().captureSnapshotIfReady(replicatedLogEntry);
 621
 622                 // Local persistence is complete so send the CheckConsensusReached message to the behavior (which
 623                 // normally should still be the leader) to check if consensus has now been reached in conjunction with
 624                 // follower replication.
 625                 getCurrentBehavior().handleMessage(getSelf(), CheckConsensusReached.INSTANCE);
 626             }
 627         }, true);
 628
 629         if (wasAppended && hasFollowers()) {
 630             // Send log entry for replication.
 631             getCurrentBehavior().handleMessage(getSelf(), new Replicate(clientActor, identifier, replicatedLogEntry,
 632                     !batchHint));
 633         }
 634     }
 635
 636     private ReplicatedLog replicatedLog() {
 637         return context.getReplicatedLog();
 638     }
 639
 640     protected String getId() {
 641         return context.getId();
 642     }
 643
 644     @VisibleForTesting
 645     void setCurrentBehavior(final RaftActorBehavior behavior) {
 646         context.setCurrentBehavior(behavior);
 647     }
 648
 649     protected RaftActorBehavior getCurrentBehavior() {
 650         return context.getCurrentBehavior();
 651     }
 652
 653     /**
 654      * Derived actors can call the isLeader method to check if the current
 655      * RaftActor is the Leader or not.
 656      *
 657      * @return true it this RaftActor is a Leader false otherwise
 658      */
 659     protected boolean isLeader() {
 660         return context.getId().equals(getCurrentBehavior().getLeaderId());
 661     }
 662
 663     protected final boolean isLeaderActive() {
 664         return getRaftState() != RaftState.IsolatedLeader && getRaftState() != RaftState.PreLeader
 665                 && !shuttingDown && !isLeadershipTransferInProgress();
 666     }
 667
 668     protected boolean isLeadershipTransferInProgress() {
 669         RaftActorLeadershipTransferCohort leadershipTransferInProgress = context.getRaftActorLeadershipTransferCohort();
 670         return leadershipTransferInProgress != null && leadershipTransferInProgress.isTransferring();
 671     }
 672
 673     /**
 674      * Derived actor can call getLeader if they need a reference to the Leader.
 675      * This would be useful for example in forwarding a request to an actor
 676      * which is the leader
 677      *
 678      * @return A reference to the leader if known, null otherwise
 679      */
 680     public ActorSelection getLeader() {
 681         String leaderAddress = getLeaderAddress();
 682
 683         if (leaderAddress == null) {
 684             return null;
 685         }
 686
 687         return context.actorSelection(leaderAddress);
 688     }
 689
 690     /**
 691      * Returns the id of the current leader.
 692      *
 693      * @return the current leader's id
 694      */
 695     protected final String getLeaderId() {
 696         return getCurrentBehavior().getLeaderId();
 697     }
 698
 699     @VisibleForTesting
 700     protected final RaftState getRaftState() {
 701         return getCurrentBehavior().state();
 702     }
 703
 704     protected Long getCurrentTerm() {
 705         return context.getTermInformation().getCurrentTerm();
 706     }
 707
 708     protected RaftActorContext getRaftActorContext() {
 709         return context;
 710     }
 711
 712     protected void updateConfigParams(final ConfigParams configParams) {
 713
 714         // obtain the RaftPolicy for oldConfigParams and the updated one.
 715         String oldRaftPolicy = context.getConfigParams().getCustomRaftPolicyImplementationClass();
 716         String newRaftPolicy = configParams.getCustomRaftPolicyImplementationClass();
 717
 718         LOG.debug("{}: RaftPolicy used with prev.config {}, RaftPolicy used with newConfig {}", persistenceId(),
 719             oldRaftPolicy, newRaftPolicy);
 720         context.setConfigParams(configParams);
 721         if (!Objects.equals(oldRaftPolicy, newRaftPolicy)) {
 722             // The RaftPolicy was modified. If the current behavior is Follower then re-initialize to Follower
 723             // but transfer the previous leaderId so it doesn't immediately try to schedule an election. This
 724             // avoids potential disruption. Otherwise, switch to Follower normally.
 725             RaftActorBehavior behavior = getCurrentBehavior();
 726             if (behavior != null && behavior.state() == RaftState.Follower) {
 727                 String previousLeaderId = behavior.getLeaderId();
 728                 short previousLeaderPayloadVersion = behavior.getLeaderPayloadVersion();
 729
 730                 LOG.debug("{}: Re-initializing to Follower with previous leaderId {}", persistenceId(),
 731                         previousLeaderId);
 732
 733                 changeCurrentBehavior(new Follower(context, previousLeaderId, previousLeaderPayloadVersion));
 734             } else {
 735                 initializeBehavior();
 736             }
 737         }
 738     }
 739
 740     public final DataPersistenceProvider persistence() {
 741         return delegatingPersistenceProvider.getDelegate();
 742     }
 743
 744     public void setPersistence(final DataPersistenceProvider provider) {
 745         delegatingPersistenceProvider.setDelegate(provider);
 746     }
 747
 748     protected void setPersistence(final boolean persistent) {
 749         DataPersistenceProvider currentPersistence = persistence();
 750         if (persistent && (currentPersistence == null || !currentPersistence.isRecoveryApplicable())) {
 751             setPersistence(new PersistentDataProvider(this));
 752
 753             if (getCurrentBehavior() != null) {
 754                 LOG.info("{}: Persistence has been enabled - capturing snapshot", persistenceId());
 755                 captureSnapshot();
 756             }
 757         } else if (!persistent && (currentPersistence == null || currentPersistence.isRecoveryApplicable())) {
 758             setPersistence(new NonPersistentDataProvider(this) {
 759                 /*
 760                  * The way snapshotting works is,
 761                  * <ol>
 762                  * <li> RaftActor calls createSnapshot on the Shard
 763                  * <li> Shard sends a CaptureSnapshotReply and RaftActor then calls saveSnapshot
 764                  * <li> When saveSnapshot is invoked on the akka-persistence API it uses the SnapshotStore to save
 765                  * the snapshot. The SnapshotStore sends SaveSnapshotSuccess or SaveSnapshotFailure. When the
 766                  * RaftActor gets SaveSnapshot success it commits the snapshot to the in-memory journal. This
 767                  * commitSnapshot is mimicking what is done in SaveSnapshotSuccess.
 768                  * </ol>
 769                  */
 770                 @Override
 771                 public void saveSnapshot(final Object object) {
 772                     // Make saving Snapshot successful
 773                     // Committing the snapshot here would end up calling commit in the creating state which would
 774                     // be a state violation. That's why now we send a message to commit the snapshot.
 775                     self().tell(RaftActorSnapshotMessageSupport.COMMIT_SNAPSHOT, self());
 776                 }
 777             });
 778         }
 779     }
 780
 781     /**
 782      * setPeerAddress sets the address of a known peer at a later time.
 783      *
 784      * <p>
 785      * This is to account for situations where a we know that a peer
 786      * exists but we do not know an address up-front. This may also be used in
 787      * situations where a known peer starts off in a different location and we
 788      * need to change it's address
 789      *
 790      * <p>
 791      * Note that if the peerId does not match the list of peers passed to
 792      * this actor during construction an IllegalStateException will be thrown.
 793      */
 794     protected void setPeerAddress(final String peerId, final String peerAddress) {
 795         context.setPeerAddress(peerId, peerAddress);
 796     }
 797
 798     /**
 799      * The applyState method will be called by the RaftActor when some data
 800      * needs to be applied to the actor's state.
 801      *
 802      * @param clientActor A reference to the client who sent this message. This
 803      *                    is the same reference that was passed to persistData
 804      *                    by the derived actor. clientActor may be null when
 805      *                    the RaftActor is behaving as a follower or during
 806      *                    recovery.
 807      * @param identifier  The identifier of the persisted data. This is also
 808      *                    the same identifier that was passed to persistData by
 809      *                    the derived actor. identifier may be null when
 810      *                    the RaftActor is behaving as a follower or during
 811      *                    recovery
 812      * @param data        A piece of data that was persisted by the persistData call.
 813      *                    This should NEVER be null.
 814      */
 815     protected abstract void applyState(ActorRef clientActor, Identifier identifier, Object data);
 816
 817     /**
 818      * Returns the RaftActorRecoveryCohort to participate in persistence recovery.
 819      */
 820     protected abstract @NonNull RaftActorRecoveryCohort getRaftActorRecoveryCohort();
 821
 822     /**
 823      * This method is called when recovery is complete.
 824      */
 825     protected abstract void onRecoveryComplete();
 826
 827     /**
 828      * Returns the RaftActorSnapshotCohort to participate in snapshot captures.
 829      */
 830     protected abstract @NonNull RaftActorSnapshotCohort getRaftActorSnapshotCohort();
 831
 832     /**
 833      * This method will be called by the RaftActor when the state of the
 834      * RaftActor changes. The derived actor can then use methods like
 835      * isLeader or getLeader to do something useful
 836      */
 837     protected abstract void onStateChanged();
 838
 839     /**
 840      * Notifier Actor for this RaftActor to notify when a role change happens.
 841      *
 842      * @return ActorRef - ActorRef of the notifier or Optional.absent if none.
 843      */
 844     protected abstract Optional<ActorRef> getRoleChangeNotifier();
 845
 846     /**
 847      * This method is called on the leader when a voting change operation completes.
 848      */
 849     protected void onVotingStateChangeComplete() {
 850     }
 851
 852     /**
 853      * This method is called prior to operations such as leadership transfer and actor shutdown when the leader
 854      * must pause or stop its duties. This method allows derived classes to gracefully pause or finish current
 855      * work prior to performing the operation. On completion of any work, the run method must be called on the
 856      * given Runnable to proceed with the given operation. <b>Important:</b> the run method must be called on
 857      * this actor's thread dispatcher as as it modifies internal state.
 858      *
 859      * <p>
 860      * The default implementation immediately runs the operation.
 861      *
 862      * @param operation the operation to run
 863      */
 864     protected void pauseLeader(final Runnable operation) {
 865         operation.run();
 866     }
 867
 868     /**
 869      * This method is invoked when the actions hooked to the leader becoming paused failed to execute and the leader
 870      * should resume normal operations.
 871      *
 872      * <p>
 873      * Note this method can be invoked even before the operation supplied to {@link #pauseLeader(Runnable)} is invoked.
 874      */
 875     protected void unpauseLeader() {
 876
 877     }
 878
 879     protected void onLeaderChanged(final String oldLeader, final String newLeader) {
 880     }
 881
 882     private String getLeaderAddress() {
 883         if (isLeader()) {
 884             return getSelf().path().toString();
 885         }
 886         String leaderId = getLeaderId();
 887         if (leaderId == null) {
 888             return null;
 889         }
 890         String peerAddress = context.getPeerAddress(leaderId);
 891         LOG.debug("{}: getLeaderAddress leaderId = {} peerAddress = {}", persistenceId(), leaderId, peerAddress);
 892
 893         return peerAddress;
 894     }
 895
 896     protected boolean hasFollowers() {
 897         return getRaftActorContext().hasFollowers();
 898     }
 899
 900     private void captureSnapshot() {
 901         SnapshotManager snapshotManager = context.getSnapshotManager();
 902
 903         if (!snapshotManager.isCapturing()) {
 904             final long idx = getCurrentBehavior().getReplicatedToAllIndex();
 905             LOG.debug("Take a snapshot of current state. lastReplicatedLog is {} and replicatedToAllIndex is {}",
 906                 replicatedLog().last(), idx);
 907
 908             snapshotManager.captureWithForcedTrim(replicatedLog().last(), idx);
 909         }
 910     }
 911
 912     /**
 913      * Switch this member to non-voting status. This is a no-op for all behaviors except when we are the leader,
 914      * in which case we need to step down.
 915      */
 916     void becomeNonVoting() {
 917         if (isLeader()) {
 918             initiateLeadershipTransfer(new RaftActorLeadershipTransferCohort.OnComplete() {
 919                 @Override
 920                 public void onSuccess(final ActorRef raftActorRef) {
 921                     LOG.debug("{}: leader transfer succeeded after change to non-voting", persistenceId());
 922                     ensureFollowerState();
 923                 }
 924
 925                 @Override
 926                 public void onFailure(final ActorRef raftActorRef) {
 927                     LOG.debug("{}: leader transfer failed after change to non-voting", persistenceId());
 928                     ensureFollowerState();
 929                 }
 930
 931                 private void ensureFollowerState() {
 932                     // Whether or not leadership transfer succeeded, we have to step down as leader and
 933                     // switch to Follower so ensure that.
 934                     if (getRaftState() != RaftState.Follower) {
 935                         initializeBehavior();
 936                     }
 937                 }
 938             }, null, RaftActorLeadershipTransferCohort.USE_DEFAULT_LEADER_TIMEOUT);
 939         }
 940     }
 941
 942     /**
 943      * A point-in-time capture of {@link RaftActorBehavior} state critical for transitioning between behaviors.
 944      */
 945     private abstract static class BehaviorState implements Immutable {
 946         @Nullable abstract RaftActorBehavior getBehavior();
 947
 948         @Nullable abstract String getLastValidLeaderId();
 949
 950         @Nullable abstract String getLastLeaderId();
 951
 952         abstract short getLeaderPayloadVersion();
 953     }
 954
 955     /**
 956      * A {@link BehaviorState} corresponding to non-null {@link RaftActorBehavior} state.
 957      */
 958     private static final class SimpleBehaviorState extends BehaviorState {
 959         private final RaftActorBehavior behavior;
 960         private final String lastValidLeaderId;
 961         private final String lastLeaderId;
 962         private final short leaderPayloadVersion;
 963
 964         SimpleBehaviorState(final String lastValidLeaderId, final String lastLeaderId,
 965                 final RaftActorBehavior behavior) {
 966             this.lastValidLeaderId = lastValidLeaderId;
 967             this.lastLeaderId = lastLeaderId;
 968             this.behavior = requireNonNull(behavior);
 969             leaderPayloadVersion = behavior.getLeaderPayloadVersion();
 970         }
 971
 972         @Override
 973         RaftActorBehavior getBehavior() {
 974             return behavior;
 975         }
 976
 977         @Override
 978         String getLastValidLeaderId() {
 979             return lastValidLeaderId;
 980         }
 981
 982         @Override
 983         short getLeaderPayloadVersion() {
 984             return leaderPayloadVersion;
 985         }
 986
 987         @Override
 988         String getLastLeaderId() {
 989             return lastLeaderId;
 990         }
 991     }
 992
 993     /**
 994      * Class tracking behavior-related information, which we need to keep around and pass across behavior switches.
 995      * An instance is created for each RaftActor. It has two functions:
 996      * - it keeps track of the last leader ID we have encountered since we have been created
 997      * - it creates state capture needed to transition from one behavior to the next
 998      */
 999     private static final class BehaviorStateTracker {
1000         /**
1001          * A {@link BehaviorState} corresponding to null {@link RaftActorBehavior} state. Since null behavior is only
1002          * allowed before we receive the first message, we know the leader ID to be null.
1003          */
1004         private static final BehaviorState NULL_BEHAVIOR_STATE = new BehaviorState() {
1005             @Override
1006             RaftActorBehavior getBehavior() {
1007                 return null;
1008             }
1009
1010             @Override
1011             String getLastValidLeaderId() {
1012                 return null;
1013             }
1014
1015             @Override
1016             short getLeaderPayloadVersion() {
1017                 return -1;
1018             }
1019
1020             @Override
1021             String getLastLeaderId() {
1022                 return null;
1023             }
1024         };
1025
1026         private String lastValidLeaderId;
1027         private String lastLeaderId;
1028
1029         BehaviorState capture(final RaftActorBehavior behavior) {
1030             if (behavior == null) {
1031                 verify(lastValidLeaderId == null, "Null behavior with non-null last leader");
1032                 return NULL_BEHAVIOR_STATE;
1033             }
1034
1035             lastLeaderId = behavior.getLeaderId();
1036             if (lastLeaderId != null) {
1037                 lastValidLeaderId = lastLeaderId;
1038             }
1039
1040             return new SimpleBehaviorState(lastValidLeaderId, lastLeaderId, behavior);
1041         }
1042     }
1043 }