opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/behaviors/Follower.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 package org.opendaylight.controller.cluster.raft.behaviors;
   9
  10 import akka.actor.ActorRef;
  11 import akka.actor.ActorSelection;
  12 import akka.actor.Address;
  13 import akka.cluster.Cluster;
  14 import akka.cluster.ClusterEvent.CurrentClusterState;
  15 import akka.cluster.Member;
  16 import akka.cluster.MemberStatus;
  17 import com.google.common.annotations.VisibleForTesting;
  18 import com.google.common.base.Stopwatch;
  19 import java.io.IOException;
  20 import java.util.ArrayList;
  21 import java.util.List;
  22 import java.util.Optional;
  23 import java.util.Set;
  24 import java.util.concurrent.TimeUnit;
  25 import java.util.concurrent.atomic.AtomicBoolean;
  26 import java.util.function.Consumer;
  27 import org.eclipse.jdt.annotation.Nullable;
  28 import org.opendaylight.controller.cluster.messaging.MessageAssembler;
  29 import org.opendaylight.controller.cluster.raft.RaftActorContext;
  30 import org.opendaylight.controller.cluster.raft.RaftState;
  31 import org.opendaylight.controller.cluster.raft.ReplicatedLogEntry;
  32 import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot;
  33 import org.opendaylight.controller.cluster.raft.base.messages.ElectionTimeout;
  34 import org.opendaylight.controller.cluster.raft.base.messages.TimeoutNow;
  35 import org.opendaylight.controller.cluster.raft.messages.AppendEntries;
  36 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  37 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshot;
  38 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshotReply;
  39 import org.opendaylight.controller.cluster.raft.messages.RaftRPC;
  40 import org.opendaylight.controller.cluster.raft.messages.RequestVote;
  41 import org.opendaylight.controller.cluster.raft.messages.RequestVoteReply;
  42 import org.opendaylight.controller.cluster.raft.persisted.ServerConfigurationPayload;
  43 import org.opendaylight.controller.cluster.raft.persisted.Snapshot;
  44
  45 /**
  46  * The behavior of a RaftActor in the Follower raft state.
  47  * <ul>
  48  * <li> Respond to RPCs from candidates and leaders
  49  * <li> If election timeout elapses without receiving AppendEntries
  50  * RPC from current leader or granting vote to candidate:
  51  * convert to candidate
  52  * </ul>
  53  */
  54 public class Follower extends AbstractRaftActorBehavior {
  55     private static final long MAX_ELECTION_TIMEOUT_FACTOR = 18;
  56
  57     private final SyncStatusTracker initialSyncStatusTracker;
  58
  59     private final MessageAssembler appendEntriesMessageAssembler;
  60
  61     private final Stopwatch lastLeaderMessageTimer = Stopwatch.createStarted();
  62     private SnapshotTracker snapshotTracker = null;
  63     private String leaderId;
  64     private short leaderPayloadVersion;
  65
  66     public Follower(final RaftActorContext context) {
  67         this(context, null, (short)-1);
  68     }
  69
  70     public Follower(final RaftActorContext context, final String initialLeaderId,
  71             final short initialLeaderPayloadVersion) {
  72         super(context, RaftState.Follower);
  73         this.leaderId = initialLeaderId;
  74         this.leaderPayloadVersion = initialLeaderPayloadVersion;
  75
  76         initialSyncStatusTracker = new SyncStatusTracker(context.getActor(), getId(), context.getConfigParams()
  77             .getSyncIndexThreshold());
  78
  79         appendEntriesMessageAssembler = MessageAssembler.builder().logContext(logName())
  80                 .fileBackedStreamFactory(context.getFileBackedOutputStreamFactory())
  81                 .assembledMessageCallback((message, sender) -> handleMessage(sender, message)).build();
  82
  83         if (context.getPeerIds().isEmpty() && getLeaderId() == null) {
  84             actor().tell(TimeoutNow.INSTANCE, actor());
  85         } else {
  86             scheduleElection(electionDuration());
  87         }
  88     }
  89
  90     @Override
  91     public final String getLeaderId() {
  92         return leaderId;
  93     }
  94
  95     @VisibleForTesting
  96     protected final void setLeaderId(final @Nullable String leaderId) {
  97         this.leaderId = leaderId;
  98     }
  99
 100     @Override
 101     public short getLeaderPayloadVersion() {
 102         return leaderPayloadVersion;
 103     }
 104
 105     @VisibleForTesting
 106     protected final void setLeaderPayloadVersion(final short leaderPayloadVersion) {
 107         this.leaderPayloadVersion = leaderPayloadVersion;
 108     }
 109
 110     private void restartLastLeaderMessageTimer() {
 111         if (lastLeaderMessageTimer.isRunning()) {
 112             lastLeaderMessageTimer.reset();
 113         }
 114
 115         lastLeaderMessageTimer.start();
 116     }
 117
 118     private boolean isLogEntryPresent(final long index) {
 119         if (context.getReplicatedLog().isInSnapshot(index)) {
 120             return true;
 121         }
 122
 123         ReplicatedLogEntry entry = context.getReplicatedLog().get(index);
 124         return entry != null;
 125
 126     }
 127
 128     private void updateInitialSyncStatus(final long currentLeaderCommit, final String newLeaderId) {
 129         initialSyncStatusTracker.update(newLeaderId, currentLeaderCommit, context.getCommitIndex());
 130     }
 131
 132     @Override
 133     protected RaftActorBehavior handleAppendEntries(final ActorRef sender, final AppendEntries appendEntries) {
 134         int numLogEntries = appendEntries.getEntries().size();
 135         if (log.isTraceEnabled()) {
 136             log.trace("{}: handleAppendEntries: {}", logName(), appendEntries);
 137         } else if (log.isDebugEnabled() && numLogEntries > 0) {
 138             log.debug("{}: handleAppendEntries: {}", logName(), appendEntries);
 139         }
 140
 141         if (snapshotTracker != null && !snapshotTracker.getLeaderId().equals(appendEntries.getLeaderId())) {
 142             log.debug("{}: snapshot install is in progress but the prior snapshot leaderId {} does not match the "
 143                 + "AppendEntries leaderId {}", logName(), snapshotTracker.getLeaderId(), appendEntries.getLeaderId());
 144             closeSnapshotTracker();
 145         }
 146
 147         if (snapshotTracker != null || context.getSnapshotManager().isApplying()) {
 148             // if snapshot install is in progress, follower should just acknowledge append entries with a reply.
 149             AppendEntriesReply reply = new AppendEntriesReply(context.getId(), currentTerm(), true,
 150                     lastIndex(), lastTerm(), context.getPayloadVersion(), false, needsLeaderAddress(),
 151                     appendEntries.getLeaderRaftVersion());
 152
 153             log.debug("{}: snapshot install is in progress, replying immediately with {}", logName(), reply);
 154             sender.tell(reply, actor());
 155
 156             return this;
 157         }
 158
 159         // If we got here then we do appear to be talking to the leader
 160         leaderId = appendEntries.getLeaderId();
 161         leaderPayloadVersion = appendEntries.getPayloadVersion();
 162
 163         if (appendEntries.getLeaderAddress().isPresent()) {
 164             final String address = appendEntries.getLeaderAddress().get();
 165             log.debug("New leader address: {}", address);
 166
 167             context.setPeerAddress(leaderId, address);
 168             context.getConfigParams().getPeerAddressResolver().setResolved(leaderId, address);
 169         }
 170
 171         // First check if the logs are in sync or not
 172         if (isOutOfSync(appendEntries, sender)) {
 173             updateInitialSyncStatus(appendEntries.getLeaderCommit(), appendEntries.getLeaderId());
 174             return this;
 175         }
 176
 177         if (!processNewEntries(appendEntries, sender)) {
 178             updateInitialSyncStatus(appendEntries.getLeaderCommit(), appendEntries.getLeaderId());
 179             return this;
 180         }
 181
 182         long lastIndex = lastIndex();
 183         long prevCommitIndex = context.getCommitIndex();
 184
 185         // If leaderCommit > commitIndex, set commitIndex = min(leaderCommit, index of last new entry)
 186         if (appendEntries.getLeaderCommit() > prevCommitIndex) {
 187             context.setCommitIndex(Math.min(appendEntries.getLeaderCommit(), lastIndex));
 188         }
 189
 190         if (prevCommitIndex != context.getCommitIndex()) {
 191             log.debug("{}: Commit index set to {}", logName(), context.getCommitIndex());
 192         }
 193
 194         AppendEntriesReply reply = new AppendEntriesReply(context.getId(), currentTerm(), true,
 195                 lastIndex, lastTerm(), context.getPayloadVersion(), false, needsLeaderAddress(),
 196                 appendEntries.getLeaderRaftVersion());
 197
 198         if (log.isTraceEnabled()) {
 199             log.trace("{}: handleAppendEntries returning : {}", logName(), reply);
 200         } else if (log.isDebugEnabled() && numLogEntries > 0) {
 201             log.debug("{}: handleAppendEntries returning : {}", logName(), reply);
 202         }
 203
 204         // Reply to the leader before applying any previous state so as not to hold up leader consensus.
 205         sender.tell(reply, actor());
 206
 207         updateInitialSyncStatus(appendEntries.getLeaderCommit(), appendEntries.getLeaderId());
 208
 209         // If leaderCommit > lastApplied, increment lastApplied and apply log[lastApplied] to state machine (§5.3).
 210         // lastApplied can be equal to lastIndex.
 211         if (appendEntries.getLeaderCommit() > context.getLastApplied() && context.getLastApplied() < lastIndex) {
 212             if (log.isDebugEnabled()) {
 213                 log.debug("{}: applyLogToStateMachine, appendEntries.getLeaderCommit(): {}, "
 214                         + "context.getLastApplied(): {}, lastIndex(): {}", logName(),
 215                     appendEntries.getLeaderCommit(), context.getLastApplied(), lastIndex);
 216             }
 217
 218             applyLogToStateMachine(appendEntries.getLeaderCommit());
 219         }
 220
 221         if (!context.getSnapshotManager().isCapturing()) {
 222             super.performSnapshotWithoutCapture(appendEntries.getReplicatedToAllIndex());
 223         }
 224
 225         appendEntriesMessageAssembler.checkExpiredAssembledMessageState();
 226
 227         return this;
 228     }
 229
 230     private boolean processNewEntries(final AppendEntries appendEntries, final ActorRef sender) {
 231         int numLogEntries = appendEntries.getEntries().size();
 232         if (numLogEntries == 0) {
 233             return true;
 234         }
 235
 236         log.debug("{}: Number of entries to be appended = {}", logName(), numLogEntries);
 237
 238         long lastIndex = lastIndex();
 239         int addEntriesFrom = 0;
 240
 241         // First check for conflicting entries. If an existing entry conflicts with a new one (same index but different
 242         // term), delete the existing entry and all that follow it (§5.3)
 243         if (context.getReplicatedLog().size() > 0) {
 244             // Find the entry up until the one that is not in the follower's log
 245             for (int i = 0;i < numLogEntries; i++, addEntriesFrom++) {
 246                 ReplicatedLogEntry matchEntry = appendEntries.getEntries().get(i);
 247
 248                 if (!isLogEntryPresent(matchEntry.getIndex())) {
 249                     // newEntry not found in the log
 250                     break;
 251                 }
 252
 253                 long existingEntryTerm = getLogEntryTerm(matchEntry.getIndex());
 254
 255                 log.debug("{}: matchEntry {} is present: existingEntryTerm: {}", logName(), matchEntry,
 256                         existingEntryTerm);
 257
 258                 // existingEntryTerm == -1 means it's in the snapshot and not in the log. We don't know
 259                 // what the term was so we'll assume it matches.
 260                 if (existingEntryTerm == -1 || existingEntryTerm == matchEntry.getTerm()) {
 261                     continue;
 262                 }
 263
 264                 if (!context.getRaftPolicy().applyModificationToStateBeforeConsensus()) {
 265                     log.info("{}: Removing entries from log starting at {}, commitIndex: {}, lastApplied: {}",
 266                             logName(), matchEntry.getIndex(), context.getCommitIndex(), context.getLastApplied());
 267
 268                     // Entries do not match so remove all subsequent entries but only if the existing entries haven't
 269                     // been applied to the state yet.
 270                     if (matchEntry.getIndex() <= context.getLastApplied()
 271                             || !context.getReplicatedLog().removeFromAndPersist(matchEntry.getIndex())) {
 272                         // Could not remove the entries - this means the matchEntry index must be in the
 273                         // snapshot and not the log. In this case the prior entries are part of the state
 274                         // so we must send back a reply to force a snapshot to completely re-sync the
 275                         // follower's log and state.
 276
 277                         log.info("{}: Could not remove entries - sending reply to force snapshot", logName());
 278                         sender.tell(new AppendEntriesReply(context.getId(), currentTerm(), false, lastIndex,
 279                                 lastTerm(), context.getPayloadVersion(), true, needsLeaderAddress(),
 280                                 appendEntries.getLeaderRaftVersion()), actor());
 281                         return false;
 282                     }
 283
 284                     break;
 285                 } else {
 286                     sender.tell(new AppendEntriesReply(context.getId(), currentTerm(), false, lastIndex,
 287                             lastTerm(), context.getPayloadVersion(), true, needsLeaderAddress(),
 288                             appendEntries.getLeaderRaftVersion()), actor());
 289                     return false;
 290                 }
 291             }
 292         }
 293
 294         lastIndex = lastIndex();
 295         log.debug("{}: After cleanup, lastIndex: {}, entries to be added from: {}", logName(), lastIndex,
 296                 addEntriesFrom);
 297
 298         // When persistence successfully completes for each new log entry appended, we need to determine if we
 299         // should capture a snapshot to compact the persisted log. shouldCaptureSnapshot tracks whether or not
 300         // one of the log entries has exceeded the log size threshold whereby a snapshot should be taken. However
 301         // we don't initiate the snapshot at that log entry but rather after the last log entry has been persisted.
 302         // This is done because subsequent log entries after the one that tripped the threshold may have been
 303         // applied to the state already, as the persistence callback occurs async, and we want those entries
 304         // purged from the persisted log as well.
 305         final AtomicBoolean shouldCaptureSnapshot = new AtomicBoolean(false);
 306         final Consumer<ReplicatedLogEntry> appendAndPersistCallback = logEntry -> {
 307             final List<ReplicatedLogEntry> entries = appendEntries.getEntries();
 308             final ReplicatedLogEntry lastEntryToAppend = entries.get(entries.size() - 1);
 309             if (shouldCaptureSnapshot.get() && logEntry == lastEntryToAppend) {
 310                 context.getSnapshotManager().capture(context.getReplicatedLog().last(), getReplicatedToAllIndex());
 311             }
 312         };
 313
 314         // Append any new entries not already in the log
 315         for (int i = addEntriesFrom; i < numLogEntries; i++) {
 316             ReplicatedLogEntry entry = appendEntries.getEntries().get(i);
 317
 318             log.debug("{}: Append entry to log {}", logName(), entry.getData());
 319
 320             context.getReplicatedLog().appendAndPersist(entry, appendAndPersistCallback, false);
 321
 322             shouldCaptureSnapshot.compareAndSet(false,
 323                     context.getReplicatedLog().shouldCaptureSnapshot(entry.getIndex()));
 324
 325             if (entry.getData() instanceof ServerConfigurationPayload) {
 326                 context.updatePeerIds((ServerConfigurationPayload)entry.getData());
 327             }
 328         }
 329
 330         log.debug("{}: Log size is now {}", logName(), context.getReplicatedLog().size());
 331
 332         return true;
 333     }
 334
 335     private boolean isOutOfSync(final AppendEntries appendEntries, final ActorRef sender) {
 336
 337         final long lastIndex = lastIndex();
 338         if (lastIndex == -1 && appendEntries.getPrevLogIndex() != -1) {
 339
 340             // The follower's log is out of sync because the leader does have an entry at prevLogIndex and this
 341             // follower has no entries in it's log.
 342
 343             log.info("{}: The followers log is empty and the senders prevLogIndex is {}", logName(),
 344                 appendEntries.getPrevLogIndex());
 345
 346             sendOutOfSyncAppendEntriesReply(sender, false, appendEntries.getLeaderRaftVersion());
 347             return true;
 348         }
 349
 350         if (lastIndex > -1) {
 351             if (isLogEntryPresent(appendEntries.getPrevLogIndex())) {
 352                 final long leadersPrevLogTermInFollowersLogOrSnapshot =
 353                         getLogEntryOrSnapshotTerm(appendEntries.getPrevLogIndex());
 354                 if (leadersPrevLogTermInFollowersLogOrSnapshot != appendEntries.getPrevLogTerm()) {
 355
 356                     // The follower's log is out of sync because the Leader's prevLogIndex entry does exist
 357                     // in the follower's log or snapshot but it has a different term.
 358
 359                     log.info("{}: The prevLogIndex {} was found in the log but the term {} is not equal to the append "
 360                         + "entries prevLogTerm {} - lastIndex: {}, snapshotIndex: {}, snapshotTerm: {}", logName(),
 361                         appendEntries.getPrevLogIndex(), leadersPrevLogTermInFollowersLogOrSnapshot,
 362                         appendEntries.getPrevLogTerm(), lastIndex, context.getReplicatedLog().getSnapshotIndex(),
 363                         context.getReplicatedLog().getSnapshotTerm());
 364
 365                     sendOutOfSyncAppendEntriesReply(sender, false, appendEntries.getLeaderRaftVersion());
 366                     return true;
 367                 }
 368             } else if (appendEntries.getPrevLogIndex() != -1) {
 369
 370                 // The follower's log is out of sync because the Leader's prevLogIndex entry was not found in it's log
 371
 372                 log.info("{}: The log is not empty but the prevLogIndex {} was not found in it - lastIndex: {}, "
 373                         + "snapshotIndex: {}, snapshotTerm: {}", logName(), appendEntries.getPrevLogIndex(), lastIndex,
 374                         context.getReplicatedLog().getSnapshotIndex(), context.getReplicatedLog().getSnapshotTerm());
 375
 376                 sendOutOfSyncAppendEntriesReply(sender, false, appendEntries.getLeaderRaftVersion());
 377                 return true;
 378             }
 379         }
 380
 381         if (appendEntries.getPrevLogIndex() == -1 && appendEntries.getPrevLogTerm() == -1
 382                 && appendEntries.getReplicatedToAllIndex() != -1) {
 383             if (!isLogEntryPresent(appendEntries.getReplicatedToAllIndex())) {
 384                 // This append entry comes from a leader who has it's log aggressively trimmed and so does not have
 385                 // the previous entry in it's in-memory journal
 386
 387                 log.info("{}: Cannot append entries because the replicatedToAllIndex {} does not appear to be in the "
 388                         + "in-memory journal - lastIndex: {}, snapshotIndex: {}, snapshotTerm: {}", logName(),
 389                         appendEntries.getReplicatedToAllIndex(), lastIndex,
 390                         context.getReplicatedLog().getSnapshotIndex(), context.getReplicatedLog().getSnapshotTerm());
 391
 392                 sendOutOfSyncAppendEntriesReply(sender, false, appendEntries.getLeaderRaftVersion());
 393                 return true;
 394             }
 395
 396             final List<ReplicatedLogEntry> entries = appendEntries.getEntries();
 397             if (entries.size() > 0 && !isLogEntryPresent(entries.get(0).getIndex() - 1)) {
 398                 log.info("{}: Cannot append entries because the calculated previousIndex {} was not found in the "
 399                         + "in-memory journal - lastIndex: {}, snapshotIndex: {}, snapshotTerm: {}", logName(),
 400                         entries.get(0).getIndex() - 1, lastIndex, context.getReplicatedLog().getSnapshotIndex(),
 401                         context.getReplicatedLog().getSnapshotTerm());
 402
 403                 sendOutOfSyncAppendEntriesReply(sender, false, appendEntries.getLeaderRaftVersion());
 404                 return true;
 405             }
 406         }
 407
 408         return false;
 409     }
 410
 411     private void sendOutOfSyncAppendEntriesReply(final ActorRef sender, boolean forceInstallSnapshot,
 412             short leaderRaftVersion) {
 413         // We found that the log was out of sync so just send a negative reply.
 414         final AppendEntriesReply reply = new AppendEntriesReply(context.getId(), currentTerm(), false, lastIndex(),
 415                 lastTerm(), context.getPayloadVersion(), forceInstallSnapshot, needsLeaderAddress(),
 416                 leaderRaftVersion);
 417
 418         log.info("{}: Follower is out-of-sync so sending negative reply: {}", logName(), reply);
 419         sender.tell(reply, actor());
 420     }
 421
 422     private boolean needsLeaderAddress() {
 423         return context.getPeerAddress(leaderId) == null;
 424     }
 425
 426     @Override
 427     protected RaftActorBehavior handleAppendEntriesReply(final ActorRef sender,
 428         final AppendEntriesReply appendEntriesReply) {
 429         return this;
 430     }
 431
 432     @Override
 433     protected RaftActorBehavior handleRequestVoteReply(final ActorRef sender,
 434         final RequestVoteReply requestVoteReply) {
 435         return this;
 436     }
 437
 438     @Override
 439     public RaftActorBehavior handleMessage(final ActorRef sender, final Object message) {
 440         if (message instanceof ElectionTimeout || message instanceof TimeoutNow) {
 441             return handleElectionTimeout(message);
 442         }
 443
 444         if (appendEntriesMessageAssembler.handleMessage(message, actor())) {
 445             return this;
 446         }
 447
 448         if (!(message instanceof RaftRPC)) {
 449             // The rest of the processing requires the message to be a RaftRPC
 450             return null;
 451         }
 452
 453         final RaftRPC rpc = (RaftRPC) message;
 454         // If RPC request or response contains term T > currentTerm:
 455         // set currentTerm = T, convert to follower (§5.1)
 456         // This applies to all RPC messages and responses
 457         if (rpc.getTerm() > context.getTermInformation().getCurrentTerm()) {
 458             log.info("{}: Term {} in \"{}\" message is greater than follower's term {} - updating term",
 459                 logName(), rpc.getTerm(), rpc, context.getTermInformation().getCurrentTerm());
 460
 461             context.getTermInformation().updateAndPersist(rpc.getTerm(), null);
 462         }
 463
 464         if (rpc instanceof InstallSnapshot) {
 465             handleInstallSnapshot(sender, (InstallSnapshot) rpc);
 466             restartLastLeaderMessageTimer();
 467             scheduleElection(electionDuration());
 468             return this;
 469         }
 470
 471         if (!(rpc instanceof RequestVote) || canGrantVote((RequestVote) rpc)) {
 472             restartLastLeaderMessageTimer();
 473             scheduleElection(electionDuration());
 474         }
 475
 476         return super.handleMessage(sender, rpc);
 477     }
 478
 479     private RaftActorBehavior handleElectionTimeout(final Object message) {
 480         // If the message is ElectionTimeout, verify we haven't actually seen a message from the leader
 481         // during the election timeout interval. It may that the election timer expired b/c this actor
 482         // was busy and messages got delayed, in which case leader messages would be backed up in the
 483         // queue but would be processed before the ElectionTimeout message and thus would restart the
 484         // lastLeaderMessageTimer.
 485         long lastLeaderMessageInterval = lastLeaderMessageTimer.elapsed(TimeUnit.MILLISECONDS);
 486         long electionTimeoutInMillis = context.getConfigParams().getElectionTimeOutInterval().toMillis();
 487         boolean noLeaderMessageReceived = !lastLeaderMessageTimer.isRunning()
 488                 || lastLeaderMessageInterval >= electionTimeoutInMillis;
 489
 490         if (canStartElection()) {
 491             if (message instanceof TimeoutNow) {
 492                 log.debug("{}: Received TimeoutNow - switching to Candidate", logName());
 493                 return internalSwitchBehavior(RaftState.Candidate);
 494             } else if (noLeaderMessageReceived) {
 495                 // Check the cluster state to see if the leader is known to be up before we go to Candidate.
 496                 // However if we haven't heard from the leader in a long time even though the cluster state
 497                 // indicates it's up then something is wrong - leader might be stuck indefinitely - so switch
 498                 // to Candidate,
 499                 long maxElectionTimeout = electionTimeoutInMillis * MAX_ELECTION_TIMEOUT_FACTOR;
 500                 if (isLeaderAvailabilityKnown() && lastLeaderMessageInterval < maxElectionTimeout) {
 501                     log.debug("{}: Received ElectionTimeout but leader appears to be available", logName());
 502                     scheduleElection(electionDuration());
 503                 } else {
 504                     log.debug("{}: Received ElectionTimeout - switching to Candidate", logName());
 505                     return internalSwitchBehavior(RaftState.Candidate);
 506                 }
 507             } else {
 508                 log.debug("{}: Received ElectionTimeout but lastLeaderMessageInterval {} < election timeout {}",
 509                         logName(), lastLeaderMessageInterval, context.getConfigParams().getElectionTimeOutInterval());
 510                 scheduleElection(electionDuration());
 511             }
 512         } else if (message instanceof ElectionTimeout) {
 513             if (noLeaderMessageReceived) {
 514                 setLeaderId(null);
 515             }
 516
 517             scheduleElection(electionDuration());
 518         }
 519
 520         return this;
 521     }
 522
 523     private boolean isLeaderAvailabilityKnown() {
 524         if (leaderId == null) {
 525             return false;
 526         }
 527
 528         Optional<Cluster> cluster = context.getCluster();
 529         if (!cluster.isPresent()) {
 530             return false;
 531         }
 532
 533         ActorSelection leaderActor = context.getPeerActorSelection(leaderId);
 534         if (leaderActor == null) {
 535             return false;
 536         }
 537
 538         Address leaderAddress = leaderActor.anchorPath().address();
 539
 540         CurrentClusterState state = cluster.get().state();
 541         Set<Member> unreachable = state.getUnreachable();
 542
 543         log.debug("{}: Checking for leader {} in the cluster unreachable set {}", logName(), leaderAddress,
 544                 unreachable);
 545
 546         for (Member m: unreachable) {
 547             if (leaderAddress.equals(m.address())) {
 548                 log.info("{}: Leader {} is unreachable", logName(), leaderAddress);
 549                 return false;
 550             }
 551         }
 552
 553         for (Member m: state.getMembers()) {
 554             if (leaderAddress.equals(m.address())) {
 555                 if (m.status() == MemberStatus.up() || m.status() == MemberStatus.weaklyUp()) {
 556                     log.debug("{}: Leader {} cluster status is {} - leader is available", logName(),
 557                             leaderAddress, m.status());
 558                     return true;
 559                 } else {
 560                     log.debug("{}: Leader {} cluster status is {} - leader is unavailable", logName(),
 561                             leaderAddress, m.status());
 562                     return false;
 563                 }
 564             }
 565         }
 566
 567         log.debug("{}: Leader {} not found in the cluster member set", logName(), leaderAddress);
 568
 569         return false;
 570     }
 571
 572     private void handleInstallSnapshot(final ActorRef sender, final InstallSnapshot installSnapshot) {
 573
 574         log.debug("{}: handleInstallSnapshot: {}", logName(), installSnapshot);
 575
 576         leaderId = installSnapshot.getLeaderId();
 577
 578         if (snapshotTracker == null) {
 579             snapshotTracker = new SnapshotTracker(log, installSnapshot.getTotalChunks(), installSnapshot.getLeaderId(),
 580                     context);
 581         }
 582
 583         updateInitialSyncStatus(installSnapshot.getLastIncludedIndex(), installSnapshot.getLeaderId());
 584
 585         try {
 586             final InstallSnapshotReply reply = new InstallSnapshotReply(
 587                     currentTerm(), context.getId(), installSnapshot.getChunkIndex(), true);
 588
 589             if (snapshotTracker.addChunk(installSnapshot.getChunkIndex(), installSnapshot.getData(),
 590                     installSnapshot.getLastChunkHashCode())) {
 591
 592                 log.info("{}: Snapshot installed from leader: {}", logName(), installSnapshot.getLeaderId());
 593
 594                 Snapshot snapshot = Snapshot.create(
 595                         context.getSnapshotManager().convertSnapshot(snapshotTracker.getSnapshotBytes()),
 596                         new ArrayList<>(),
 597                         installSnapshot.getLastIncludedIndex(),
 598                         installSnapshot.getLastIncludedTerm(),
 599                         installSnapshot.getLastIncludedIndex(),
 600                         installSnapshot.getLastIncludedTerm(),
 601                         context.getTermInformation().getCurrentTerm(),
 602                         context.getTermInformation().getVotedFor(),
 603                         installSnapshot.getServerConfig().orNull());
 604
 605                 ApplySnapshot.Callback applySnapshotCallback = new ApplySnapshot.Callback() {
 606                     @Override
 607                     public void onSuccess() {
 608                         log.debug("{}: handleInstallSnapshot returning: {}", logName(), reply);
 609
 610                         sender.tell(reply, actor());
 611                     }
 612
 613                     @Override
 614                     public void onFailure() {
 615                         sender.tell(new InstallSnapshotReply(currentTerm(), context.getId(), -1, false), actor());
 616                     }
 617                 };
 618
 619                 actor().tell(new ApplySnapshot(snapshot, applySnapshotCallback), actor());
 620
 621                 closeSnapshotTracker();
 622             } else {
 623                 log.debug("{}: handleInstallSnapshot returning: {}", logName(), reply);
 624
 625                 sender.tell(reply, actor());
 626             }
 627         } catch (IOException e) {
 628             log.debug("{}: Exception in InstallSnapshot of follower", logName(), e);
 629
 630             sender.tell(new InstallSnapshotReply(currentTerm(), context.getId(),
 631                     -1, false), actor());
 632
 633             closeSnapshotTracker();
 634         }
 635     }
 636
 637     private void closeSnapshotTracker() {
 638         if (snapshotTracker != null) {
 639             snapshotTracker.close();
 640             snapshotTracker = null;
 641         }
 642     }
 643
 644     @Override
 645     public void close() {
 646         closeSnapshotTracker();
 647         stopElection();
 648         appendEntriesMessageAssembler.close();
 649     }
 650
 651     @VisibleForTesting
 652     SnapshotTracker getSnapshotTracker() {
 653         return snapshotTracker;
 654     }
 655 }